Escolar Documentos
Profissional Documentos
Cultura Documentos
This is a special issue published in “Abstract and Applied Analysis.” All articles are open access articles distributed under the Creative
Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original
work is properly cited.
Editorial Board
Ravi P. Agarwal, USA Jean M. Combes, France Jaume Giné, Spain
Bashir Ahmad, Saudi Arabia Monica Conti, Italy Valery Y. Glizer, Israel
M. O. Ahmedou, Germany Diego Córdoba, Spain Laurent Gosse, Italy
Nicholas D. Alikakos, Greece Juan C. Cortés, Spain Jean P. Gossez, Belgium
Debora Amadori, Italy Graziano Crasta, Italy Jose L. Gracia, Spain
Pablo Amster, Argentina Bernard Dacorogna, Switzerland Maurizio Grasselli, Italy
Douglas R. Anderson, USA Vladimir Danilov, Russia Qian Guo, China
Jan Andres, Czech Republic Mohammad T. Darvishi, Iran Yuxia Guo, China
Giovanni Anello, Italy Luis F. Pinheiro de Castro, Portugal Chaitan P. Gupta, USA
Stanislav Antontsev, Portugal T. Diagana, USA Uno Hämarik, Estonia
Mohamed Kamal Aouf, Egypt Jesús I. Dı́az, Spain Ferenc Hartung, Hungary
Narcisa C. Apreutesei, Romania Josef Diblı́k, Czech Republic Behnam Hashemi, Iran
Natig M. Atakishiyev, Mexico Fasma Diele, Italy Norimichi Hirano, Japan
Ferhan M. Atici, USA Tomas Dominguez, Spain Jiaxin Hu, China
Ivan G. Avramidi, USA Alexander I. Domoshnitsky, Israel Zhongyi Huang, China
Soohyun Bae, Korea Marco Donatelli, Italy Chengming Huang, China
Chuanzhi Bai, China Bo-Qing Dong, China Gennaro Infante, Italy
Zhanbing Bai, China Ondr̆ej Došlý, Czech Republic Ivan Ivanov, Bulgaria
Dumitru Baleanu, Turkey Wei-Shih Du, Taiwan Hossein Jafari, Iran
Józef Banaś, Poland Luiz Duarte, Brazil Jaan Janno, Estonia
Gerassimos Barbatis, Greece Roman Dwilewicz, USA Aref Jeribi, Tunisia
Martino Bardi, Italy Paul W. Eloe, USA Un C. Ji, Korea
Roberto Barrio, Spain Ahmed El-Sayed, Egypt Zhongxiao Jia, China
Feyzi Başar, Turkey Luca Esposito, Italy L. Jódar, Spain
Abdelghani Bellouquid, Morocco Jose A. Ezquerro, Spain Jong Soo Jung, Republic of Korea
Daniele Bertaccini, Italy Khalil Ezzinbi, Morocco Henrik Kalisch, Norway
Lucio Boccardo, Italy Dashan Fan, USA Hamid Reza Karimi, Norway
Igor Boglaev, New Zealand Angelo Favini, Italy Satyanad Kichenassamy, France
Martin J. Bohner, USA Marcia Federson, Brazil Tero Kilpeläinen, Finland
Geraldo Botelho, Brazil Julian F. Bonder, Argentina Sung Guen Kim, Republic of Korea
Elena Braverman, Canada S. Filippas, Equatorial Guinea Ljubisa Kocinac, Serbia
Romeo Brunetti, Italy Alberto Fiorenza, Italy Andrei Korobeinikov, Spain
Janusz Brzdek, Poland Tore Flȧtten, Norway Pekka Koskela, Finland
Detlev Buchholz, Germany Ilaria Fragala, Italy Victor Kovtunenko, Austria
Sun-Sig Byun, Korea Bruno Franchi, Italy Ren-Jieh Kuo, Taiwan
Fabio M. Camilli, Italy Xianlong Fu, China Pavel Kurasov, Sweden
Jinde Cao, China Massimo Furi, Italy Miroslaw Lachowicz, Poland
Anna Capietto, Italy Giovanni P. Galdi, USA Kunquan Lan, Canada
Jianqing Chen, China Isaac Garcia, Spain Ruediger Landes, USA
Wing-Sum Cheung, Hong Kong Jesús Garcı́a Falset, Spain Irena Lasiecka, USA
Michel Chipot, Switzerland José A. Garcı́a-Rodrı́guez, Spain Matti Lassas, Finland
Changbum Chun, Korea Leszek Gasinski, Poland Chun-Kong Law, Taiwan
Soon Y. Chung, Korea György Gát, Hungary Ming-Yi Lee, Taiwan
Jaeyoung Chung, Korea Vladimir Georgiev, Italy Gongbao Li, China
Silvia Cingolani, Italy Lorenzo Giacomelli, Italy Elena Litsyn, Israel
Yansheng Liu, China Somyot Plubtieng, Thailand Gabriel Turinici, France
Shengqiang Liu, China Milan Pokorny, Czech Republic Milan Tvrdy, Czech Republic
Carlos Lizama, Chile Sergio Polidoro, Italy Mehmet Unal, Turkey
Milton C. Lopes Filho, Brazil Ziemowit Popowicz, Poland Csaba Varga, Romania
Julian López-Gómez, Spain Maria M. Porzio, Italy Carlos Vazquez, Spain
Guozhen Lu, USA Enrico Priola, Italy Gianmaria Verzini, Italy
Jinhu Lü, China Vladimir S. Rabinovich, Mexico Jesus Vigo-Aguiar, Spain
Grzegorz Lukaszewicz, Poland Irena Rachu̇nková, Czech Republic Qing-Wen Wang, China
Wanbiao Ma, China Maria Alessandra Ragusa, Italy Yushun Wang, China
Shiwang Ma, China Simeon Reich, Israel Shawn X. Wang, Canada
Eberhard Malkowsky, Turkey Abdelaziz Rhandi, Italy Jing Ping Wang, UK
Salvatore A. Marano, Italy Hassan Riahi, Malaysia Youyu Wang, China
Cristina Marcelli, Italy Juan P. Rincón-Zapatero, Spain Peixuan Weng, China
Paolo Marcellini, Italy Luigi Rodino, Italy Noemi Wolanski, Argentina
Jesús Marı́n-Solano, Spain Yuriy V. Rogovchenko, Norway Ngai-Ching Wong, Taiwan
Jose M. Martell, Spain Julio D. Rossi, Argentina Patricia J. Y. Wong, Singapore
Mieczysław Mastyło, Poland Wolfgang Ruess, Germany Yong Hong Wu, Australia
Ming Mei, Canada Bernhard Ruf, Italy Zili Wu, China
Taras Mel’nyk, Ukraine Marco Sabatini, Italy Shanhe Wu, China
Anna Mercaldo, Italy Satit Saejung, Thailand Tie-cheng Xia, China
Changxing Miao, China Stefan G. Samko, Portugal Xu Xian, China
Stanislaw Migorski, Poland Martin Schechter, USA Yanni Xiao, China
Mihai Mihǎilescu, Romania Javier Segura, Spain Gonang Xie, China
Feliz Minhós, Portugal Sigmund Selberg, Norway Fuding Xie, China
Dumitru Motreanu, France Valery Serov, Finland Naihua Xiu, China
Roberta Musina, Italy N. Shahzad, Saudi Arabia Daoyi Xu, China
G. M. N’Guérékata, USA Andrey Shishkov, Ukraine Zhenya Yan, China
Maria Grazia Naso, Italy Stefan Siegmund, Germany Xiaodong Yan, USA
Sylvia Novo, Spain Abdel-Maksoud A. Soliman, Egypt Norio Yoshida, Japan
Micah Osilike, Nigeria Pierpaolo Soravia, Italy Beong In Yun, Korea
Mitsuharu Ôtani, Japan Marco Squassina, Italy Vjacheslav Yurko, Russia
Turgut Öziş, Turkey Svatoslav Staněk, Czech Republic Agacik Zafer, Turkey
Filomena Pacella, Italy Stevo Stevic, Serbia Sergey V. Zelik, UK
Nikolaos S. Papageorgiou, Greece Antonio Suárez, Spain Jianming Zhan, China
Sehie Park, Korea Wenchang Sun, China Meirong Zhang, China
Alberto Parmeggiani, Italy Robert Szalai, UK Weinian Zhang, China
Kailash C. Patidar, South Africa Sanyi Tang, China Chengjian Zhang, China
Kevin R. Payne, Italy Chun-Lei Tang, China Zengqin Zhao, China
Ademir Fernando Pazoto, Brazil Youshan Tao, China Sining Zheng, China
Josip E. Pečarić, Croatia Gabriella Tarantello, Italy Tianshou Zhou, China
Shuangjie Peng, China Nasser-eddine Tatar, Saudi Arabia Yong Zhou, China
Sergei V. Pereverzyev, Austria Gerd Teschke, Germany Qiji J. Zhu, USA
Maria Eugenia Perez, Spain Bevan Thompson, Australia Chun-Gang Zhu, China
Josefina Perles, Spain Sergey Tikhonov, Spain Malisa R. Zizovic, Serbia
Allan Peterson, USA Claudia Timofte, Romania Wenming Zou, China
Andrew Pickering, Spain Thanh Tran, Australia
Cristina Pignotti, Italy Juan J. Trujillo, Spain
Contents
Fractional and Time-Scales Differential Equations, Dumitru Baleanu, Ali H. Bhrawy, Delfim F. M. Torres,
and Soheil Salahshour
Volume 2014, Article ID 365250, 2 pages
Fractional Cauchy Problem with Riemann-Liouville Fractional Delta Derivative on Time Scales,
Jiang Zhu and Ying Zhu
Volume 2013, Article ID 401596, 19 pages
On Solutions to Fractional Discrete Systems with Sequential ℎ-Differences, Małgorzata Wyrwas,
Dorota Mozyrska, and Ewa Girejko
Volume 2013, Article ID 475350, 11 pages
Numerical Solution of Fuzzy Fractional Pharmacokinetics Model Arising from Drug Assimilation into
the Bloodstream, Ali Ahmadian, Norazak Senu, Farhad Larki, Soheil Salahshour, Mohamed Suleiman,
and Md. Shabiul Islam
Volume 2013, Article ID 304739, 17 pages
Numerical Solutions of Fractional Fokker-Planck Equations Using Iterative Laplace Transform Method,
Limei Yan
Volume 2013, Article ID 465160, 7 pages
The Second Noether Theorem on Time Scales, Agnieszka B. Malinowska and Natália Martins
Volume 2013, Article ID 675127, 14 pages
A Jacobi Collocation Method for Solving Nonlinear Burgers-Type Equations, E. H. Doha, D. Baleanu,
A. H. Bhrawy, and M. A. Abdelkawy
Volume 2013, Article ID 760542, 12 pages
Mappings for Special Functions on Cantor Sets and Special Integral Transforms via Local Fractional
Operators, Yang Zhao, Dumitru Baleanu, Mihaela Cristina Baleanu, De-Fu Cheng, and Xiao-Jun Yang
Volume 2013, Article ID 316978, 6 pages
New Wavelets Collocation Method for Solving Second-Order Multipoint Boundary Value Problems
Using Chebyshev Polynomials of Third and Fourth Kinds, W. M. Abd-Elhameed, E. H. Doha,
and Y. H. Youssri
Volume 2013, Article ID 542839, 9 pages
Existence Results for a Class of Fractional Differential Equations with Periodic Boundary Value
Conditions and with Delay, Hadi Karami, Azizollah Babakhani, and Dumitru Baleanu
Volume 2013, Article ID 176180, 8 pages
Application of Fuzzy Fractional Kinetic Equations to Modelling of the Acid Hydrolysis Reaction,
Ferial Ghaemi, Robiah Yunus, Ali Ahmadian, Soheil Salahshour, Mohamed Suleiman,
and Shanti Faridah Saleh
Volume 2013, Article ID 610314, 19 pages
A Modified Generalized Laguerre Spectral Method for Fractional Differential Equations on the Half
Line, D. Baleanu, A. H. Bhrawy, and T. M. Taha
Volume 2013, Article ID 413529, 12 pages
Oscillation Criteria for Fourth-Order Nonlinear Dynamic Equations on Time Scales, Xin Wu,
Taixiang Sun, Hongjian Xi, and Changhong Chen
Volume 2013, Article ID 740568, 11 pages
The Bernstein Operational Matrices for Solving the Fractional Quadratic Riccati Differential Equations
with the Riemann-Liouville Derivative, Dumitru Baleanu, Mohsen Alipour, and Hossein Jafari
Volume 2013, Article ID 461970, 7 pages
Fractional Dynamics of Genetic Algorithms Using Hexagonal Space Tessellation, J. A. Tenreiro Machado
Volume 2013, Article ID 739464, 7 pages
An Operational Matrix Based on Legendre Polynomials for Solving Fuzzy Fractional-Order Differential
Equations, Ali Ahmadian, Mohamed Suleiman, and Soheil Salahshour
Volume 2013, Article ID 505903, 29 pages
Two Efficient Generalized Laguerre Spectral Algorithms for Fractional Initial Value Problems,
D. Baleanu, A. H. Bhrawy, and T. M. Taha
Volume 2013, Article ID 546502, 10 pages
Local Fractional Series Expansion Method for Solving Wave and Diffusion Equations on Cantor Sets,
Ai-Min Yang, Xiao-Jun Yang, and Zheng-Biao Li
Volume 2013, Article ID 351057, 5 pages
Editorial
Fractional and Time-Scales Differential Equations
Copyright © 2014 Dumitru Baleanu et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
The theory and applications of fractional differential equa- expansion method for wave and diffusion equations and
tions (FDEs) are gaining more relevance since they are used some mappings for special functions on Cantor sets were
extensively in the modeling of various processes in physics, presented. Application of fuzzy fractional kinetic equations
chemistry, engineering, and other areas of science. As it is to modeling of the acid hydrolysis reaction, the solutions
known, the behavior of dynamics of most complex systems of linear fractional differential equations with uncertainty,
of the real world phenomena has memory. Therefore, the and an operational matrix based on Legendre polynomials
modeling of dynamics of these types of systems by FDEs has for solving fuzzy fractional-order differential equations were
more advantages than the classical ones, in which such effects also investigated. A class of fractional-order differential
are neglected. On the other hand, the time-scales formalism models of biological systems with memory to model the
unifies the theories of difference and differential equations. interaction of immune system with tumor cells and with
Accordingly, the time-scales analysis constitutes a good tool HIV infection of CD4+ T-cells was reported. Fractional-
to study both discrete and continuous systems. order total variation image restoration based on primal-dual
Advanced analytical and numerical techniques and com- algorithm and the fractional dynamics of genetic algorithms
putational methods are of important interest for classical, using hexagonal space tessellation were pointed out. The
fractional, fuzzy fractional, and time-scales differential equa- positive solution using bifurcation techniques for boundary
tions. value problems of fractional differential equations was a
The papers of this special issue contain some new algo- contribution of our special issue. The existence results for a
rithms and techniques designed to investigate classical, frac- class of fractional differential equations with boundary value
tional, fractal, fuzzy fractional, and time-scales differential conditions and with delay, a modified generalized Laguerre
equations of general interest. New insights of existence and spectral methods for fractional differential equations on the
uniqueness theorems of some differential equations were also half line, a Jacobi collocation method for solving nonlin-
presented. ear Burgers-type equations, and new wavelets collocation
In the following we summarize briefly the content of method for solving second-order multipoint boundary value
the special issue. The second Noether theorem, the local problems using Chebyshev polynomials of third and fourth
observability of systems, and the fractional Cauchy problem kinds were shown. The investigation of the nonlinear frac-
within Riemann-Liouville fractional delta derivative on time tional Jaulent-Miodek and Whitham-Broer-Kaup equations
scales were reported and the solutions of fractional discrete within Sumudu transform, the Bernstein operational matri-
systems with sequential h-differences were obtained. Besides, ces applied for solving the fractional quadratic Riccati dif-
the oscillation criteria for fourth-order nonlinear dynamic ferential equations, and the approximate solutions of Fisher’s
equations on time scales were depicted. Local fractional series type equations with variable coefficients are topics covered by
2 Abstract and Applied Analysis
Research Article
Fractional Cauchy Problem with Riemann-Liouville
Fractional Delta Derivative on Time Scales
Copyright © 2013 J. Zhu and Y. Zhu. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The Δ-power function and fractional Δ-integrals and fractional Δ-differential are defined, and then the definitions and properties
of Δ-Mittag-Leffler function are given. The properties of fractional Δ-integrals and fractional Δ-differential on time scales are
discussed in detail. After that, the existence of the solution and the dependency of the solution upon the initial value for Cauchy
type problem with fractional Δ-derivative are studied. Also the explicit solutions to homogeneous fractional Δ-differential equations
and nonhomogeneous fractional Δ-differential equations are derived by using Laplace transform method.
𝑛
𝑛 𝑛−𝑖 Definition 11 (see [12]). If 𝑝 ∈ R, then one defines the Δ-
𝐿𝑦 = 0, where 𝐿𝑦 = 𝑦Δ + ∑𝑝𝑖 𝑦Δ , (3) exponential function by
𝑖=1
𝑛 𝑒𝑝 (𝑡, 𝑠)
to be for each fixed 𝑠 ∈ T 𝑘 the solution of the initial value
𝑡
problem 1
= exp (∫ Log (1 + 𝜇 (𝜏) 𝑝 (𝜏)) Δ𝜏) for 𝑠, 𝑡 ∈ T.
𝑠 𝜇 (𝜏)
Δ𝑖
𝐿𝑦 = 0, 𝑦 (𝜎 (𝑠) , 𝑠) = 0, 0 ≤ 𝑖 ≤ 𝑛 − 2, (14)
(4)
𝑛−1
𝑦Δ (𝜎 (𝑠) , 𝑠) = 1. Definition 12 (see [12]). If 𝑝 ∈ R, then the first order linear
dynamic equation
Remark 6 (see [12]). Note that
𝑦Δ = 𝑝 (𝑡) 𝑦 (15)
𝑦 (𝑡, 𝑠) := ℎ𝑛−1 (𝑡, 𝜎 (𝑠)) (5)
is called regressive.
𝑛
is the Cauchy function of 𝑦Δ = 0.
Theorem 13 (see [12]). Suppose that (15) is regressive and fix
Theorem 7 (variation of constants [12]). Let 𝑓 ∈ 𝐶rd ; then the 𝑡0 ∈ T. Then 𝑒𝑝 (⋅, 𝑡0 ) is a solution of the initial value problem
solution of the initial value problem
𝑦Δ = 𝑝 (𝑡) 𝑦, 𝑦 (𝑡0 ) = 1 (16)
Δ𝑖
𝐿𝑦 = 𝑓 (𝑡) , 𝑦 (𝑡0 ) = 0, 0 ≤ 𝑖 ≤ 𝑛 − 1, (6) on T.
Abstract and Applied Analysis 3
Theorem 14 (see [12]). If 𝑝 ∈ R, then Theorem 22 (see [18]). If 𝑓 and 𝑔 are infinitely often Δ-
differentiable, then for all 𝑘 ∈ 𝑁0
𝑒𝑝 (𝜎 (𝑡) , 𝑠) := 𝑒𝑝𝜎 (𝑡, 𝑠) = (1 + 𝜇 (𝑡) 𝑝 (𝑡)) 𝑒𝑝 (𝑡, 𝑠) . (17)
𝑘−1
Δ𝑘 𝑘 ] 𝑘−1−]
(𝑓 ∗ 𝑔) = 𝑓Δ ∗ 𝑔 + ∑ 𝑓Δ (𝑡0 ) 𝑔Δ
Definition 15 (see [12]). Assume that 𝑥 : T → R is regulated. ]=0
Then the Δ-Laplace transform of 𝑥 is defined by
𝑘
𝑘−1 ] 𝑘−1−]
∞
𝜎
= 𝑓 ∗ 𝑔 Δ + ∑ 𝑓 Δ 𝑔Δ (𝑡0 ) , (24)
L {𝑥} (𝑧, 𝑡0 ) = ∫ 𝑥 (𝑡) 𝑒⊖𝑧 (𝑡, 𝑡0 ) Δ𝑡 (18) ]=0
𝑡0
𝑘−1
Δ𝑘 ] 𝑘−1−]
for 𝑧 ∈ D{𝑥}, where D{𝑥} consists of all complex numbers (𝑓 ∗ 𝑔) (𝑡0 ) = ∑ 𝑓Δ (𝑡0 ) 𝑔Δ (𝑡0 ) .
𝑧 ∈ C for which the improper integral exists. ]=0
and fix 𝑡0 ∈ T. For a given 𝑓 : [𝑡0 , ∞)T → C, the solution of for those regressive 𝑧 ∈ C satisfying
the shifting problem
𝑖
lim {𝑥Δ (𝑡) 𝑒⊖𝑧 (𝑡, 𝑡0 )} = 0, 𝑖 = 0, 1, . . . , 𝑘 − 1. (27)
Δ𝑡 Δ𝑠 𝑡→∞
𝑢 (𝑡, 𝜎 (𝑠)) = −𝑢 (𝑡, 𝑠) , 𝑡, 𝑠 ∈ T, 𝑡 ≥ 𝑠 ≥ 𝑡0 ,
(19) Theorem 25 (see [12]). Assume that ℎ𝑘 (𝑡, 𝑡0 ), 𝑘 ∈ N0 are
𝑢 (𝑡, 𝑡0 ) = 𝑓 (𝑡) , 𝑡 ∈ T, 𝑡 ≥ 𝑡0 defined as in Definition 4. Then
ℎ𝛼 (𝑡, 𝑠) = ℎ̂
Δ
(𝑓 ∗ 𝑔) = 𝑓 ∗ 𝑔Δ + 𝑓𝑔 (𝑡0 ) . (23) 𝛼 (⋅, 𝑡0 ) (𝑡, 𝑠) (𝑡 ≥ 𝑠 ≥ 𝑡0 ) . (31)
4 Abstract and Applied Analysis
(ℎ𝛼 (⋅, 𝑡0 ) ∗ ℎ𝛽 (⋅, 𝑡0 )) (𝑡) Example 29 (see [3]). When T = R, the fractional Δ-integral
of order 𝛼 is defined by
𝑡 (35)
= ∫ ℎ̂
𝑥
𝛼 (⋅, 𝑡0 ) (𝑡, 𝜎 (𝜏)) ℎ𝛽 (𝜏, 𝑠) Δ𝜏 = ℎ𝛼+𝛽+1 (𝑡, 𝑡0 ) . 𝛼 1 1
(𝐼𝑎+ 𝑓) (𝑥) := ∫ 𝑓 (𝑡) 𝑑𝑡 (𝑥 > 𝑎) . (43)
𝑡0 Γ (𝛼) 𝑎 (𝑥 − 𝑡)1−𝛼
Example 30. When T = Z, Consider the following.
Moreover, if we take 𝛼 = 0, then
𝑡
(1) The 𝑛th integral of 𝑓 is defined by
(1 ∗ ℎ𝛽 ) (𝑡, 𝑡0 ) = ∫ ℎ𝛽 (𝜏, 𝑠) Δ𝜏 = ℎ𝛽+1 (𝑡, 𝑡0 ) . (36)
−𝑛 1 𝑡−𝑛
𝑡0
𝑎 Δ 𝑥 (𝑡) = ∑(𝑡 − 𝜎 (𝑠))(𝑛−1) 𝑥 (𝑠) . (44)
Γ (𝑛) 𝑠=𝑎
That is,
Δ Here 𝑡 − 𝑛 ≥ 𝑎, 𝜎(𝑠) = 𝑠 + 1.
ℎ𝛽+1 (𝑡, 𝑡0 ) = ℎ𝛽 (𝑡, 𝑡0 ) . (37)
Note that power function (𝑡 − 𝜎(𝑠))(𝑛−1) /Γ(𝑛) vanishes
Now, we will give the definitions of fractional Δ-integral at 𝑠 ≥ 𝑡 − 𝑛 + 1. So
and Δ-derivative which are the main context in this section.
−𝑛 1 𝑡−1
𝑎 Δ 𝑥 (𝑡) = ∑(𝑡 − 𝜎 (𝑠))(𝑛−1) 𝑥 (𝑠)
Definition 28. Let Ω be a finite interval on a time scale T, Γ (𝑛) 𝑠=𝑎
𝑡0 , 𝑡 ∈ Ω. For 𝛼 ≥ 0 and for a function 𝑓 : T → R, the
Riemann-Liouville fractional Δ-integral of order 𝛼 is defined 1 𝑡−𝑛
= ∑(𝑡 − 𝜎 (𝑠))(𝑛−1) 𝑥 (𝑠) , for 𝑡 − 𝑛 ≥ 𝑎.
0
by 𝐼Δ,𝑡 𝑓(𝑡) = 𝑓(𝑡) and Γ (𝑛) 𝑠=𝑎
0
(45)
𝛼
(𝐼Δ,𝑡 𝑓) (𝑡) = (ℎ𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓) (𝑡)
0
(2) The 𝛾th fractional sum of 𝑓 is defined by
𝑡
̂
= ∫ ℎ𝛼−1 (⋅, 𝑡0 ) (𝑡, 𝜎 (𝜏)) 𝑓 (𝜏) Δ𝜏 −𝛾 1 𝑡−1
𝑡0 (38)
𝑎Δ 𝑥 (𝑡) = ∑(𝑡 − 𝜎 (𝑠))(𝛾−1) 𝑥 (𝑠) ,
Γ (𝛾) 𝑠=𝑎 (46)
𝑡
= ∫ ℎ𝛼−1 (𝑡, 𝜎 (𝜏)) 𝑓 (𝜏) Δ𝜏, for 𝑡 ≥ 𝑎 − [−𝛾] .
𝑡0
𝑛
Throughout this paper, we denote 𝑓Δ = 𝐷Δ𝑛 𝑓 = 𝐷Δ,𝑠 𝑛
𝑓, By differentiating 𝑘 times with respect to 𝜆 on both sides
𝛼 −𝛼 𝛼
𝑛 ∈ N, and, for 𝛼 < 0, 𝐷Δ,𝑠 𝑓(𝑡) means 𝐼Δ,𝑠 𝑓(𝑡) and 𝐼Δ,𝑠 𝑓(𝑡) of the formula in Theorem 35, we get the following result:
−𝛼
means 𝐷Δ,𝑠 𝑓(𝑡).
𝜕𝑘 𝑘!𝑧𝛼−𝛽
Finally, we present the definition of Δ-Mittag-Leffler L{ 𝑘 Δ 𝐹𝛼,𝛽 (𝜆, 𝑡, 𝑡0 )} (𝑧, 𝑡0 ) = . (51)
function which is an important tool for solving fractional 𝜕𝜆 (𝑧𝛼 − 𝜆)𝑘+1
difference equation.
4. Properties of Fractional Δ-Integral and
Definition 32. Δ-Mittag-Leffler function is defined by Δ-Derivative on Time Scales
∞
In this section, we mainly give the properties of fractional Δ-
Δ 𝐹𝛼,𝛽 (𝜆, 𝑡, 𝑡0 ) = ∑𝜆𝑗 ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 ) , (48) integral and Δ-derivative on time scales which are needed in
𝑗=0
the following sections.
provided that the right series is convergent, where 𝛼, 𝛽 > 0, Theorem 36. Let 𝛼 > 0, 𝑚 = [𝑎] + 1, and 𝛽 ∈ R. Then
𝜆 ∈ R.
𝛼
(1) 𝐼Δ,𝑡 ℎ (𝑡, 𝑡0 ) = ℎ𝛽+𝛼−1 (𝑡, 𝑡0 ) ,
0 𝛽−1
Example 33. When 0 < 𝜆 < 1, for any 𝑗, |ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 )| < (52)
𝛼
𝑀, we can obtain that the series ∑∞ 𝑗 (2) 𝐷Δ,𝑡 ℎ (𝑡, 𝑡0 ) = ℎ𝛽−𝛼−1 (𝑡, 𝑡0 ) .
𝑗=0 𝜆 ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 ) is 0 𝛽−1
Proof. According to the definition of Laplace transform, it is In particular, if 𝛽 = 1, 𝛼 > 0, then the Riemann-Liouville
obtained that fractional Δ-derivatives of a constant are, in general, not equal
to zero:
L {Δ 𝐹𝛼,𝛽 (𝜆, 𝑡, 𝑡0 )} (𝑧, 𝑡0 ) 𝛼
𝐷Δ,𝑡 1 = ℎ−𝛼 (𝑡, 𝑡0 ) (0 < 𝛼 < 1) . (56)
0
∞
𝜎 On the other hand, for 𝑗 = 1, 2, . . . , 𝑚,
=∫ 𝐹𝛼,𝛽 (𝜆, 𝑡, 𝑡0 ) ⋅ 𝑒⊖𝑧 (𝑡, 𝑡0 ) Δ𝑡
𝑡0 Δ 𝛼
𝐷Δ,𝑡 ℎ (𝑡, 𝑠) = 0.
0 𝛼−𝑗
(57)
∞ ∞
= ∫ ∑𝜆𝑗 ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 ) ⋅ 𝑒⊖𝑧
𝜎
(𝑡, 𝑡0 ) Δ𝑡 In fact,
𝑡0 𝑗=0
𝛼
𝐷Δ,𝑡 ℎ (𝑡, 𝑠) = 𝐷Δ𝑚 𝐼Δ,𝑡
0 𝛼−𝑗
𝑚−𝛼
0
ℎ𝛼−𝑗 (𝑡, 𝑠)
∞ ∞ (58)
= ∑ 𝜆𝑗 ∫ ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 ) ⋅ 𝑒⊖𝑧
𝜎
(𝑡, 𝑡0 ) Δ𝑡 = 𝐷Δ𝑚 ℎ𝑚−𝑗 (𝑡, 𝑠) = 0.
𝑗=0 𝑡0 (50)
From Theorem 36, we derive the following result in [3] when
∞
𝑗 T = R.
= ∑ 𝜆 L {ℎ𝑗𝛼+𝛽−1 (𝑡, 𝑡0 )} (𝑧, 𝑡0 )
𝑗=0
Corollary 37 (see [3]). If 𝛼 ≥ 0 and 𝛽 > 0, then
∞ ∞
1 𝑗 Γ (𝛽)
= ∑ 𝜆𝑗 = 𝑧−𝛽 ∑𝜆𝑗 (𝑧−𝛼 ) 𝛼
(𝐼𝑎+ (𝑡 − 𝑎)𝛽−1 ) (𝑥) = (𝑥 − 𝑎)𝛽+𝛼−1 (𝛼 > 0) ,
𝑗=0 𝑧𝑗𝛼+𝛽 𝑗=0 Γ (𝛽 + 𝛼)
𝑧𝛼−𝛽 𝛼 Γ (𝛽)
= (|𝜆| < |𝑧|𝛼 ) . (𝐷𝑎+ (𝑡 − 𝑎)𝛽−1 ) (𝑥) = (𝑥 − 𝑎)𝛽−𝛼−1 (𝛼 ≥ 0) .
𝑧𝛼 − 𝜆 Γ (𝛽 − 𝛼)
(59)
6 Abstract and Applied Analysis
In particular, if 𝛽 = 1 and 𝛼 ≥ 0, then the Riemann-Liouville Lemma 40. (1) For 𝛼 > 0, 𝑛 = [𝛼]+1, let 𝑓 be a function which
𝑛 𝑛
fractional derivatives of a constant are, in general, not equal to is 𝑛 times Δ-differentiable on T 𝑘 with 𝑓Δ rd-continuous over
zero: T, and it is valid that
𝛼 (𝑥 − 𝑎)−𝛼 𝛼
(𝐷𝑎+ 1) (𝑥) = (0 < 𝛼 < 1) . (60) 𝐼Δ,𝑡 𝑓 (𝑡)
Γ (1 − 𝛼) 0
𝑛−1
On the other hand, for 𝑗 = 1, 2, . . . , [𝛼] + 1, 𝑘 𝑛
= ∑ ℎ𝑘+𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) + (ℎ𝑛+𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
𝛼
(𝐷𝑎+ (𝑡 − 𝑎)𝛼−𝑗 ) (𝑥) = 0. (61) 𝑘=0
(70)
As to the fractional sum and difference, we have the
following result, which is an improvement of Lemma 3.1 in (2) For 𝛼 ≥ 0, 𝑛 = [𝛼] + 1, let 𝑓 be a function which is
𝑛 𝑛
[1]. 𝑛 times Δ-differentiable on T 𝑘 with 𝑓Δ rd-continuous over T
𝛼
and 𝐷Δ,𝑡 𝑓 exists almost on T, and it is valid that
Corollary 38. Let 𝜇 ∈ R \ {. . . , −2, −1}. Then 0
𝛼
(1) Δ−] (𝜇)
= 𝜇(−]) (𝑡 − 𝑎)(𝑢+𝜐) , 𝐷Δ,𝑡 𝑓 (𝑡)
𝑎 (𝑡 − 𝑎) 𝑓𝑜𝑟 𝑡 ≥ 𝑎 − [−]] , 0
𝑛−1 (71)
(2) Δ]𝑎 (𝑡 − 𝑎)(𝜇) = 𝜇(]) (𝑡 − 𝑎)(𝑢−𝜐) , 𝑓𝑜𝑟 𝑡 ≥ 𝑎 + 1. 𝑘
= ∑ ℎ𝑘−𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) + (ℎ𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
𝑛
(62) 𝑘=0
𝑛 𝑘
𝜔 (𝑡) := ∑ ℎ𝑘 (𝑡, 𝛼) 𝑓Δ (𝛼) . (67) Besides,
𝑘=0
𝛼
𝑛+1 𝐷Δ,𝑡 𝑓 (𝑡) = 𝐷Δ𝑚 𝐼Δ,𝑡
𝑚−𝛼
𝑓 (𝑡)
By the properties of ℎ𝑘 (𝑡, 𝛼), 𝜔Δ (𝑡) = 0. We have moreover 0 0
that 𝑛−1 𝑘
𝑚
𝑛 𝑘
= 𝐷Δ𝑚 [ ∑ ℎ𝑘+𝑚−𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )
𝜔Δ (𝑡) = ∑ ℎ𝑘−𝑚 (𝑡, 𝛼) 𝑓Δ (𝛼) , (68) 𝑘=0
𝑘=𝑚
𝑛
so that + (ℎ𝑚+𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) ]
𝑚
𝑛 𝑘 𝑚
𝜔Δ (𝛼) = ∑ ℎ𝑘−𝑚 (𝛼, 𝛼) 𝑓Δ (𝛼) = 𝑓Δ (𝛼) , (69) 𝑛−1 𝑘 𝑛
𝑘=𝑚 = ∑ ℎ𝑘−𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )+(ℎ𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) ,
for 0 ≤ 𝑚 ≤ 𝑛. We consequently have that 𝜔 also solves (66), 𝑘=0
whence 𝑢 ≡ 𝜔 by uniqueness. (74)
Abstract and Applied Analysis 7
𝑛
𝑚 𝑚−𝑘 Δ𝑘 Δ−𝜇 𝑘−𝜇
𝑎 𝑓 (𝑡) = Δ 𝑎 𝑓 (𝑡) , 𝑓𝑜𝑟 𝑡 ≥ 𝑎 − [−𝜇] ,
= (ℎ𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) + ∑ ℎ𝑚+𝑛−𝛼−1−𝑘 (𝑡0 , 𝑡0 ) 𝑔Δ (80)
𝑘=0
Δ𝑘 Δ𝜇𝑎 𝑓 (𝑡) = Δ𝑘+𝜇
𝑎 𝑓 (𝑡) , 𝑓𝑜𝑟 𝑡 ≥ 𝑎 + 1.
𝑛
= (ℎ𝑛−𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
(75) Theorem 45. For 𝛼 > 0, 𝑛 is a positive integer; if 𝑓 is Δ-
differentiable and the highest order derivative is rd-continuous
over T, then it is valid that
When T = R, there is the following corollary.
𝑛−1 𝑘
𝑛+𝛼 𝛼 𝑛
Corollary 41 (see [3]). Let 𝛼 ≥ 0 and 𝑛 = [𝛼] + 1. If (1) 𝐷Δ,𝑡0
𝑓 (𝑡) = 𝐷Δ,𝑡0
𝐷Δ,𝑡0
𝑓 (𝑡) + ∑ ℎ𝑘−𝛼−𝑛 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) ,
𝑦(𝑥) ∈ 𝐴𝐶𝑛 [𝑎, 𝑏], then the fractional derivative 𝐷𝑎+
𝛼
𝑦 exists 𝑘=0
𝑘=0
Γ (1 + 𝑘 − 𝛼) (81)
(76)
𝑥 𝑦(𝑛)
1 (𝑡) 𝑑𝑡
+ ∫ . Proof. (1) In the proof of Theorem 43(1), if we take 𝑠 = 𝑛 + 𝑚,
Γ (𝑛 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼−𝑛+1
then we have
𝛼 𝛽
Theorem 42. For 𝛼 > 0 and 𝛽 > 0, then (𝐼Δ,𝑡 𝐼 𝑓)(𝑡) =
0 Δ,𝑡0
𝛼+𝛽 𝐷Δ𝑛 𝐷Δ,𝑡
𝛼 𝑛+𝛼
𝑓 (𝑡) = 𝐷Δ,𝑡 𝑓 (𝑡)
(𝐼Δ,𝑡0 𝑓)(𝑡). 0 0
𝑛+𝑚−1 𝑘
Proof. According to Definition 28, Theorem 20 and 27, = ∑ ℎ𝑘−𝛼−𝑛 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) (82)
𝛼 𝛽 𝑘=0
(𝐼Δ,𝑡 𝐼 𝑓) (𝑡)
0 Δ,𝑡0
= (ℎ𝛼−1 (⋅, 𝑡0 ) ∗ (ℎ𝛽−1 (⋅, 𝑡0 ) ∗ 𝑓)) (𝑡) 𝑠
+ (ℎ𝑠−1−𝛼−𝑛 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
= ((ℎ𝛼−1 (⋅, 𝑡0 ) ∗ ℎ𝛽−1 (⋅, 𝑡0 )) ∗ 𝑓) (𝑡)
𝛼+𝛽
= (ℎ𝛼+𝛽−1 (⋅, 𝑡0 ) ∗ 𝑓) (𝑡) = (𝐼Δ,𝑡0 𝑓) (𝑡) . 𝑚+𝑛
As 𝐷Δ𝑛 𝑓(𝑡) is 𝑚 times Δ-differentiable on T 𝑘 , we have
(77)
𝑚−1 𝑛+𝑘
𝛼
𝐷Δ,𝑡0
(𝐷Δ𝑛 𝑓 (𝑡)) = ∑ ℎ𝑘−𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )
Theorem 43. For 𝛼 > 0, 𝑛 is a positive integer; if 𝑓 is Δ- 𝑘=0 (83)
differentiable and the highest order derivative is rd-continuous 𝑠
over T, then it is valid that + (ℎ−𝛼+𝑚−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) .
(1) 𝐷Δ𝑛 𝐷Δ,𝑡
𝛼
0
𝑛+𝛼
𝑓 (𝑡) = 𝐷Δ,𝑡0
𝑓 (𝑡) ,
(78) By (82) and (83), if 𝑓 is at least 𝑛 times Δ-differentiable with
(2) 𝐷Δ𝑛 𝐼Δ,𝑡
𝛼
0
𝛼−𝑛
𝑓 (𝑡) = 𝐼Δ,𝑡0
𝑓 (𝑡) . the highest order derivative rd-continuous over T, then we
Proof. (1) Suppose that 𝑓 is a function which is 𝑠 times have
𝑠 𝑠
Δ-differentiable on T 𝑘 with 𝑓Δ rd-continuous over T. By
Lemma 40(2), 𝐷Δ𝑛 𝐷Δ,𝑡
𝛼
0
𝛼
𝑓 (𝑡) = 𝐷Δ,𝑡0
(𝐷Δ𝑛 𝑓 (𝑡))
𝐷Δ𝑛 𝐷Δ,𝑡
𝛼
𝑓 (𝑡) 𝑛−1 𝑘
(84)
0
𝑠−1
+ ∑ ℎ−𝛼−𝑛+𝑘 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) .
𝑘 𝑠
𝑘=0
= 𝐷Δ𝑛 [ ∑ ℎ𝑘−𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) + (ℎ𝑠−1−𝛼 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡)]
𝑘=0
𝑠−1 Thus
𝑘 𝑠
= ∑ ℎ𝑘−𝛼−𝑛 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) + (ℎ𝑠−1−𝛼−𝑛 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡)
𝑘=0 𝐷Δ𝑛 𝐷Δ,𝑡
𝛼
0
𝛼
𝑓 (𝑡) = 𝐷Δ,𝑡0
(𝐷Δ𝑛 𝑓 (𝑡)) (85)
𝑛+𝛼
= 𝐷Δ,𝑡0
𝑓 (𝑡) .
(79) is valid if and only if
By a similar way, we can get (2).
𝑘
(2) Similarly, we have Theorem 48. Let 𝑓 be Δ-differentiable and let its highest order
derivative be rd-continuous over T. When 𝛼, 𝛽 > 0, 𝑀 = [𝛽] +
𝑚−1 𝑘+𝑛
1, one has the following:
𝛼
𝐼Δ,𝑡 𝐷Δ𝑛 𝑓 (𝑡) = ∑ ℎ𝑘+𝛼 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 )
0 𝛽𝛼 𝛼−𝛽
𝑘=0 (1) (𝐷Δ,𝑡0 𝐼Δ,𝑡0
𝑓) (𝑡) = (𝐼Δ,𝑡0 ) 𝑓 (𝑡) ,
𝑚+𝑛
+ (ℎ𝛼+𝑚−1 (⋅, 𝑡0 ) ∗ 𝑓Δ ) (𝑡) , 𝛼 𝛽 𝛽−𝛼
(2) (𝐼Δ,𝑡 𝐷Δ,𝑡0 𝑓) (𝑡) = (𝐷Δ,𝑡0 ) 𝑓 (𝑡)
0 (93)
𝐷Δ𝑛 𝐼Δ,𝑡
𝛼
0
𝛼−𝑛
𝑓 (𝑡) = 𝐼Δ,𝑡0
𝑓 (𝑡) (87) 𝑀
𝛽−𝑘
𝑛+𝑚−1 − ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝐷Δ,𝑡0 𝑓 (𝑡0 ) .
𝑘
= ∑ ℎ𝑘+𝛼−𝑛 (𝑡, 𝑡0 ) 𝑓Δ (𝑡0 ) 𝑘=1
In addition,
Provided that 𝑓 is at least 𝑛 times Δ-differentiable with the
𝛼 𝛽 𝛼 𝑀−𝛽
highest order derivative rd-continuous over T. 𝐼Δ,𝑡0
𝐷Δ,𝑡0 𝑓 (𝑡) = 𝐼Δ,𝑡0
𝐷Δ𝑀𝐼Δ,𝑡0 𝑓 (𝑡)
Thus
𝑀−𝛽
= 𝐷Δ𝑀𝐼Δ,𝑡
𝛼
𝐼
0 Δ,𝑡0
𝑓 (𝑡)
𝐷Δ𝑛 𝐼Δ,𝑡
𝛼
0
𝑓 (𝑡) = 𝛼
𝐼Δ,𝑡0
𝐷Δ𝑛 𝑓 (𝑡) (89)
𝑀−1
𝑘+𝛽−𝑀 (95)
is valid if and only if − ∑ ℎ𝛼−𝑀+𝑘 (𝑡, 𝑡0 ) 𝐷Δ,𝑡0 𝑓 (𝑡0 )
𝑘=0
𝑘 𝑀
𝑓Δ (𝑡0 ) = 0, 𝑘 = 0, . . . , 𝑛 − 1. (90) 𝛼−𝛽 𝛽−𝑘
= 𝐼Δ,𝑡0 𝑓 (𝑡) − ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝐷Δ,𝑡0 𝑓 (𝑡0 ) .
𝑘=1
In particular, there are corollaries for T = R and for T = As a direct corollary of Theorem 48, we get Lemma 2.5 in
Z. [3].
Corollary 46 (see [3]). Let 𝛼 > 0 and 𝛽 > 0 be such that Corollary 49 (see [3]). Let 𝑅(𝛼) > 0, 𝑛 = [𝑅(𝛼)] + 1, and
𝑛−𝛼
𝑛 − 1 < 𝛼 ≤ 𝑛, 𝑚 − 1 < 𝛽 ≤ 𝑚 (𝑛, 𝑚 ∈ N) and 𝛼 + 𝛽 < 𝑛, let 𝑓𝑛−𝛼 (𝑥) = (𝐼𝑎+ 𝑓)(𝑥). If 𝑓(𝑥) ∈ 𝐿 1 (𝑎, 𝑏) and 𝑓𝑛−𝛼 (𝑥) ∈
𝑛
and let 𝑓 ∈ 𝐿 1 (𝑎, 𝑏) and 𝑓𝑚−𝛼 ∈ 𝐴𝐶𝑚 ([𝑎, 𝑏]). Then one has 𝐴𝐶 [𝑎, 𝑏], then
the following index rule: 𝑛 (𝑛−𝑗)
𝛼 𝛼 𝑓𝑛−𝛼 (𝑎)
(𝐼𝑎+ 𝐷𝑎+ 𝑓) (𝑥) = 𝑓 (𝑥) − ∑ (𝑥 − 𝑎)𝛼−𝑗 (96)
𝛼
(𝐷𝑎+ 𝛽
𝐷𝑎+ 𝛼+𝛽
𝑓) (𝑥) = (𝐷𝑎+ 𝑓) (𝑥) 𝑗=1 Γ (𝛼 − 𝑗 + 1)
Next, we will give the Laplace transform of fractional Here 𝐿 Δ [𝑎, 𝑏) := 𝐿 Δ,1 [𝑎, 𝑏) is the space of Δ-Lebesgue
integral and derivative on time scales. summable functions in a finite interval T̃ := [𝑎, 𝑏)T .
In the following, we prove that Cauchy type problem and
Theorem 51. (1) Let 𝛼 > 0 and 𝑓 : T → R be locally Δ- nonlinear Volterra integral equation are equivalent in the
integrable. For 𝑠, 𝑡 ∈ T with 𝑡 > 𝑡0 , one has sense that if 𝑦(𝑡) ∈ 𝐿 Δ [𝑎, 𝑏) satisfies one of these relations,
𝛼 1 then it also satisfies the other.
L {𝐼Δ,𝑡 𝑓 (𝑡)} (𝑧, 𝑡0 ) = L {𝑓 (𝑡)} (𝑧, 𝑡0 ) . (98)
0
𝑧𝛼 𝑚
Theorem 52. Let 𝛼 > 0, 𝑚 = −[−𝛼], T̃ := [𝑎, 𝑏]T , 𝑡0 , 𝑡 ∈ T̃ 𝑘 .
(2) Let 𝛼 > 0 and 𝑓 : T → R be locally Δ-integrable. For
𝑚
Let 𝐺 be an open set in R and let 𝑓 : T̃ × 𝐺 → R be a function
𝑡0 , 𝑡 ∈ T 𝑘 with 𝑡 > 𝑠, one has such that 𝑓(𝑡, 𝑦) ∈ 𝐿 Δ [𝑎, 𝑏) for any 𝑦 ∈ 𝐺. If 𝑦(𝑡) ∈ 𝐿 Δ [𝑎, 𝑏),
𝛼
then Cauchy type problem (102) and (103) is equivalent to
L {𝐷Δ,𝑡0
𝑓 (𝑡)} (𝑧, 𝑡0 ) 𝑚
𝛼
𝑚 (99) 𝑦 (𝑡) = ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝑏𝑘 + 𝐼Δ,𝑡0
𝑓 (𝑡, 𝑦 (𝑡)) . (105)
𝛼−𝑗
= 𝑧𝛼 L {𝑓 (𝑡)} (𝑧, 𝑡0 ) − ∑𝑧𝑗−1 𝐷Δ,𝑡0 𝑓 (𝑡0 ) . 𝑘=1
𝛼
𝑗=1 Proof. First we prove the necessity. We apply 𝐼Δ,𝑡0
to both sides
Proof. (1) According to Definition 28 and convolution theo- of (102) and get by Theorem 48
rem, we have 𝑚
𝛼 𝛼 𝛼−𝑘
𝐼Δ,𝑡 𝐷Δ,𝑡 𝑦 (𝑡) = 𝑦 (𝑡) − ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝐷Δ,𝑡 𝑦 (𝑡0 )
𝛼 0 0 0
L {𝐼Δ,𝑡0
𝑓 (𝑡)} (𝑧, 𝑡0 ) 𝑘=1
= (ℎ𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓 (⋅, ⋅)) (𝑡) . (106)
= L {(ℎ𝛼−1 (⋅, 𝑡0 ) ∗ 𝑓) (𝑡)} (𝑧, 𝑡0 )
𝛼
(100) = 𝐼Δ,𝑡 𝑓 (𝑡, 𝑦 (𝑡)) .
= L {ℎ𝛼−1 (𝑡, 𝑡0 )} (𝑧, 𝑡0 ) L {𝑓 (𝑡)} (𝑧, 𝑡0 ) 0
Thus
1 𝑚
= 𝛼 L {𝑓 (𝑡)} (𝑧, 𝑡0 ) . 𝛼
𝑧 𝑦 (𝑡) = ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝑏𝑘 + 𝐼Δ,𝑡0
𝑓 (𝑡, 𝑦 (𝑡)) . (107)
𝑘=1
(2) By Definition 31 and (26) and taking the Laplace
𝛼
transform of fractional integral into account, we get Now we prove the sufficiency. Applying the operator 𝐷Δ,𝑡0
to
both sides of (105) and by (57) and Theorem 48(1), we have
𝛼
L {𝐷Δ,𝑡0
𝑓 (𝑡)} (𝑧, 𝑡0 ) 𝑚
𝛼 𝛼
𝑚
𝐷Δ,𝑡 𝑦 (𝑡) = ∑ 𝐷Δ,𝑡 ℎ (𝑡, 𝑡0 ) 𝑏𝑘
0 𝛼−𝑘
= L {𝐷Δ,𝑡 𝐼𝑚−𝛼 𝑓 (𝑡)} (𝑧, 𝑡0 )
0 Δ,𝑡0
0
𝑘=1
𝛼 (108)
𝑚−1 + 𝐷Δ,𝑡 𝐼𝛼 𝑓 (𝑡, 𝑦 (𝑡))
0 Δ,𝑡0
= 𝑧𝑚 L {𝐼Δ,𝑡
𝑚−𝛼
0
𝑓 (𝑡)} (𝑧, 𝑡0 ) − ∑ 𝑧𝑚−𝑗−1 𝐷𝑗 𝐼Δ,𝑡
𝑚−𝛼
0
𝑓 (𝑡0 )
𝑗=0 = 𝑓 (𝑡, 𝑦 (𝑡)) .
𝑚−1 Now we show that the relations in (103) also hold. For this,
1 𝑗−𝑚+𝛼
= 𝑧𝑚 L {𝑓 (𝑡)} (𝑧, 𝑡0 ) − ∑ 𝑧𝑚−𝑗−1 𝐷Δ,𝑡0 𝑓 (𝑡0 ) 𝛼−𝑘
we apply the operators 𝐷Δ,𝑡 (𝑘 = 1, . . . , 𝑚) to both sides of
𝑧𝑚−𝛼 𝑗=0
0
(105):
𝑚 𝛼−𝑘
𝛼−𝑗 𝐷Δ,𝑡 𝑦 (𝑡)
= 𝑧𝛼 L {𝑓 (𝑡)} (𝑧, 𝑡0 ) − ∑𝑧𝑗−1 𝐷Δ,𝑡0 𝑓 (𝑡0 ) . 0
𝑗=1 𝑚
𝛼−𝑘 𝛼−𝑘 𝛼
(101) = ∑ 𝐷Δ,𝑡 ℎ (𝑡, 𝑡0 ) 𝑏𝑗 + 𝐷Δ,𝑡
0 𝛼−𝑗
𝐼 𝑓 (𝑡, 𝑦 (𝑡))
0 Δ,𝑡0
𝑗=1
𝑚 (109)
𝑚−𝑘 𝑚−𝛼 𝑘
5. Cauchy Type Problem with = ∑ 𝐷Δ,𝑡 𝐼 ℎ𝛼−𝑗 (𝑡, 𝑡0 ) 𝑏𝑗 + 𝐼Δ,𝑡
0 Δ,𝑡0 0
𝑓 (𝑡, 𝑦 (𝑡))
𝑗=1
Riemann-Liouville Fractional Derivative
𝑚
𝑚−𝑘 𝑘
In this section, we consider Cauchy type problem with = ∑ 𝐷Δ,𝑡0
ℎ𝑚−𝑗 (𝑡, 𝑡0 ) 𝑏𝑗 + 𝐼Δ,𝑡0
𝑓 (𝑡, 𝑦 (𝑡)) .
Riemann-Liouville fractional derivative 𝑗=1
𝛼
𝐷Δ,𝑡0
𝑦 (𝑡) = 𝑓 (𝑡, 𝑦 (𝑡)) (𝛼 > 0) , (102) Since
{ 0, 𝑘 < 𝑗,
𝛼−𝑘
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 𝑏𝑘 (𝑘 = 1, . . . , 𝑚 = − [−𝛼]) . (103) 𝑚−𝑘 {
0 𝐷Δ,𝑡 ℎ𝑚−𝑗 (𝑡, 𝑡0 ) = { 1, 𝑘 = 𝑗,
0 {
{ℎ𝑘−𝑗 (𝑡, 𝑡0 ) , 𝑘 > 𝑗,
In the space 𝐿𝛼Δ [𝑎, 𝑏) defined for 𝛼 > 0 by (110)
𝛼
where ≤ (𝐼Δ,𝑡0
𝐴𝑀𝐴𝑙−1 ℎ𝑙𝛼 (⋅, 𝑡0 )) (𝑡) (123)
𝑚
𝑦0 (𝑡) = ∑ ℎ𝛼−𝑘 (𝑡, 𝑡0 ) 𝑏𝑘 . (113) ≤ 𝑀𝐴𝑙 (𝐼Δ,𝑡
𝛼
ℎ (⋅, 𝑡0 )) (𝑡)
0 𝑙𝛼
𝑘=1
Theorem 55. Let the condition of Theorem 54 be valid and let Suppose that
𝑓(𝑡, 𝑦, 𝑦1 , . . . , 𝑦𝑙 ) satisfy the Lipschitzian condition (128). Then
there exists a unique solution 𝑦(𝑡) to the generalized Cauchy
𝑧𝑚−1 (𝑡) − 𝑦𝑚−1 (𝑡)
type problem.
𝑚−1 (136)
≤ 𝜂 − 𝜂 ∑ 𝐴𝑗 ℎ𝑗𝛼+𝛼−1 (𝑡, 𝑡0 ) .
6. The Dependency of the Solution upon 𝑗=0
the Initial Value
We consider fractional differential initial value problem Then
again:
𝛼
𝑧𝑚 (𝑡) − 𝑦𝑚 (𝑡)
𝐷Δ,𝑡 𝑦 (𝑡) = 𝑓 (𝑡, 𝑦 (𝑡)) ,
≤ 𝜂 − 𝜂 ℎ𝛼−1 (𝑡, 𝑡0 )
0
𝛼
(129)
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 𝜂,
𝛼
0
𝑗=0 (139)
≤ 𝜂 − 𝜂 ℎ𝛼−1 (𝑡, 𝑡0 ) + (𝐼Δ,𝑡
𝛼
𝐴 𝑧0 (𝜏) − 𝑦0 (𝜏)) (𝑡)
0
≤ 𝜂 − 𝜂 Δ 𝐹𝛼,𝛼 (𝐴, 𝑏, 𝑡0 ) .
= 𝜂 − 𝜂 ℎ𝛼−1 (𝑡, 𝑡0 ) + 𝜂 − 𝜂 𝐴 (𝐼Δ,𝑡
𝛼
ℎ (𝜏, 𝑡0 )) (𝑡)
0 𝛼−1
= 𝜂 − 𝜂 ℎ𝛼−1 (𝑡, 𝑡0 ) + 𝜂 − 𝜂 𝐴ℎ2𝛼−1 (𝑡, 𝑡0 )
As a special case, when fractional equation is linear, we
= 𝜂 − 𝜂 [ℎ𝛼−1 (𝑡, 𝑡0 ) + 𝐴ℎ2𝛼−1 (𝑡, 𝑡0 )] . can obtain its explicit solutions and we will explain it in next
(135) section.
12 Abstract and Applied Analysis
𝑐1
7. Homogeneous Equations with 𝑐2
𝛼
∑ 𝑐𝑗 𝑦𝑗 (𝑡) = 0, (149)
L {𝐷Δ,𝑡0
𝑦 (𝑡)} (𝑧, 𝑡0 ) = 𝑧𝛼 L {𝑦 (𝑡)} (𝑧, 𝑡0 ) 𝑗=1
(144) 𝛼−𝑘
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 0 (𝑘, 𝑗 = 1, . . . , 𝑙; 𝑘 ≠
𝑗) ,
0 𝑗
(151)
Lemma 58. The solutions 𝑦1 (𝑡), 𝑦2 (𝑡), . . . , 𝑦𝑛 (𝑡) are linearly 𝛼−𝑘
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 1 (𝑘 = 1, . . . , 𝑙) .
independent if and only if 𝑊𝛼 (𝑡∗ ) ≠
0 at some point 𝑡∗ ∈ [𝑎, 𝑏]. 0 𝑘
𝑛 𝑐 2
𝛼−𝑘
((𝐷Δ,𝑡 𝑦 ) (𝑡))𝑘,𝑗=1 ( .. ) ≡ 0 (145) where 𝑑𝑗 (𝑗 = 1, . . . , 𝑙) are given by (142).
0 𝑗
. Formula (49) with 𝛽 = 𝛼 + 1 − 𝑗 yields
𝑐𝑛
𝑧𝑗−1
L { Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 )} (𝑧, 𝑡0 ) = (|𝜆| < |𝑧|𝛼 ) .
holds, and thus 𝑊𝛼 (𝑡) ≡ 0 which leads to a contradiction. 𝑧𝛼 − 𝜆
Therefore, if 𝑊𝛼 (𝑡∗ ) ≠ 0 at some point 𝑡∗ ∈ Ω, then (153)
𝑦1 (𝑡), 𝑦2 (𝑡), . . . , 𝑦𝑛 (𝑡) are linearly independent. Now we prove
the necessity. Suppose, to the contrary, for 𝑡 ∈ Ω, 𝑊𝛼 (𝑡) = 0. Thus, from (152), we derive the following solution to (143):
Consider 𝑙
𝛼−𝑘 𝑛 𝑦 (𝑡) = ∑ 𝑑𝑗 𝑦𝑗 (𝑡) , 𝑦𝑗 (𝑡) = Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 ) . (154)
((𝐷Δ,𝑡 𝑦 ) (𝑡∗ ))𝑘,𝑗=1 𝐶 = 0,
0 𝑗
(146) 𝑗=1
Abstract and Applied Analysis 13
It is easily verified that the functions 𝑦𝑗 (𝑡) are solutions to Corollary 60. Consider that
(143):
𝛼
𝐷Δ,𝑡 𝑦 (𝑡) − 𝜆𝑦 (𝑡) = 0 (𝑡 > 𝑡0 ; 0 < 𝛼 ≤ 1; 𝜆 ∈ R) (161)
𝛼 0
𝐷Δ,𝑡0
[ Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 )]
(155) has its solution given by
= 𝜆 Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 ) (𝑗 = 1, . . . , 𝑙) .
𝑦 (𝑡) = Δ 𝐹𝛼,𝛼 (𝜆, 𝑡, 𝑡0 ) , (162)
In fact,
∞ while
𝛼 𝛼 𝑘
𝐷Δ,𝑡 [ Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 )] = 𝐷Δ,𝑡 [ ∑ 𝜆 ℎ𝑘𝛼+𝛼−𝑗 (𝑡, 𝑡0 )] 𝛼
0 0
𝑘=0
𝐷Δ,𝑡0
𝑦 (𝑡) − 𝜆𝑦 (𝑡) = 0 (𝑡 > 𝑡0 ; 1 < 𝛼 ≤ 2; 𝜆 ∈ R) (163)
∞
has the fundamental system of solutions given by
= ∑ 𝜆𝑘 ℎ𝑘𝛼−𝑗 (𝑡, 𝑡0 )
𝑘=0
𝑦1 (𝑡) = Δ 𝐹𝛼,𝛼 (𝜆, 𝑡, 𝑡0 ) , 𝑦2 (𝑡) = Δ 𝐹𝛼,𝛼−1 (𝜆, 𝑡, 𝑡0 ) .
∞ (164)
= ∑ 𝜆𝑘+1 ℎ(𝑘+1)𝛼−𝑗 (𝑡, 𝑡0 )
𝑘=−1 Next we derive the explicit solutions to (140) with 𝑚 = 2
0 of the form
= 𝜆 ℎ−𝑗 (𝑡, 𝑡0 )
𝛼 𝛽
∞ 𝐷Δ,𝑡0
𝑦 (𝑡) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡) − 𝜇𝑦 (𝑡) = 0
𝑘+1 (165)
+ ∑𝜆 ℎ(𝑘+1)𝛼−𝑗 (𝑡, 𝑡0 )
(𝑡 > 𝑡0 ; 𝑙 − 1 < 𝛼 ≤ 𝑙; 𝑙 ∈ N; 0 < 𝛽 < 𝛼)
𝑘=0
∞ with 𝜆, 𝜇 ∈ R.
= 𝜆 ∑ 𝜆𝑘 ℎ𝑘𝛼+𝛼−𝑗 (𝑡, 𝑡0 )
𝑘=0 Theorem 61. Let 𝑙 − 1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N), 0 < 𝛽 < 𝛼, and
= 𝜆 Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 ) . 𝜆, 𝜇 ∈ R. Then the functions
∞
(156) 𝜇𝑘 𝜕𝑘
𝑦𝑗 (𝑡) = ∑ 𝐹 (𝜆, 𝑡, 𝑡0 ) (166)
Moreover, 𝑘=0
𝑘! 𝜕𝜆𝑘 Δ 𝛼−𝛽,𝛼+𝑘𝛽+1−𝑗
∞
yield the fundamental system of solutions to (165) provided that
𝛼−𝑘
𝐷Δ,𝑡 𝛼−𝑘
𝑦 (𝑡) = 𝐷Δ,𝑡
0 𝑗 0
[∑𝜆𝑠 ̂ℎ𝑠𝛼+𝛼−𝑗 (𝑡, 𝑡0 )] the series in (166) are convergent. Moreover, if 𝛼 + 1 − 𝑙 > 𝛽 >
𝑠=0
(157) 𝑙 − 1, then 𝑦𝑗 (𝑡), 𝑗 = 1, 2, . . . , 𝑙, in (166) satisfy (151).
∞
𝑠
= ∑𝜆 ℎ𝑠𝛼+𝑘−𝑗 (𝑡, 𝑡0 ) . Proof. Let 𝑚 − 1 < 𝛽 ≤ 𝑚 (𝑚 ≤ 𝑙; 𝑚 ∈ N). Applying
𝑠=0 the Laplace transform to (165) and using (141) as in (152), we
It follows from (157) that obtain
𝑙
𝛼−𝑘
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 0 (𝑘, 𝑗 = 1, . . . , 𝑙; 𝑘 > 𝑗) , 𝑧𝑗−1
0 𝑗 L {𝑦 (𝑡)} (𝑧, 𝑡0 ) = ∑ 𝑑𝑗 , (167)
(158) 𝑗=1 𝑧𝛼 − 𝜆𝑧𝛽 − 𝜇
𝛼−𝑘
𝐷Δ,𝑡 𝑦
0 𝑘
(𝑡0 ) = 1 (𝑘 = 1, . . . , 𝑙) .
𝛼−𝑗 𝛽−𝑗
where 𝑑𝑗 = 𝐷Δ,𝑡0 𝑦(𝑡0 ) − 𝜆𝐷Δ,𝑡0 𝑦(𝑡0 ) (𝑗 = 1, . . . , 𝑚), 𝑑𝑗 =
If 𝑘 < 𝑗, then 𝛼−𝑗
𝐷Δ,𝑡0 𝑦(𝑡0 ) (𝑗 = 𝑚 + 1, . . . , 𝑙).
∞
𝛼−𝑘
𝐷Δ,𝑡 𝑦 (𝑡) = ∑𝜆𝑠 ℎ𝑠𝛼+𝑘−𝑗 (𝑡, 𝑡0 ) For 𝑧 ∈ C and |𝜇𝑧−𝛽 /(𝑧𝛼−𝛽 − 𝜆)| < 1, we have
0 𝑗
𝑠=1
(159) 1 𝑧−𝛽 1
∞
𝛼 𝛽
= 𝛼−𝛽
⋅
= ∑𝜆𝑠+1 ℎ𝑠𝛼+𝛼+𝑘−𝑗 (𝑡, 𝑡0 ) , 𝑧 − 𝜆𝑧 − 𝜇 𝑧 − 𝜆 1 − 𝜇𝑧 / (𝑧𝛼−𝛽 − 𝜆)
−𝛽
∞ (168)
𝑠=0
𝜇𝑘 𝑧−𝛽−𝑘𝛽
=∑ ,
and since 𝛼 + 𝑘 − 𝑗 ≥ 𝛼 + 1 − 𝑙 > 0 for any 𝑘, 𝑗 = 1, . . . , 𝑙, the 𝛼−𝛽 − 𝜆)𝑘+1
𝑘=0 (𝑧
following relations hold:
𝛼−𝑘
and hence (167) has the following representation:
𝐷Δ,𝑡 𝑦 (𝑡0 ) = 0
0 𝑗
(𝑘, 𝑗 = 1, . . . , 𝑙; 𝑘 < 𝑗) . (160)
𝑙 ∞
𝑧𝑗−1−𝛽−𝑘𝛽
By (158) and (160), 𝑊𝛼 (𝑡0 ) = 1. Then 𝑦𝑗 (𝑡), 𝑗 = 1, . . . , 𝑙, L {𝑦 (𝑡)} (𝑧, 𝑡0 ) = ∑𝑑𝑗 ∑ 𝜇𝑘 𝑘+1
. (169)
yield the fundamental system of solutions to (143). 𝑗=1 𝑘=0 (𝑧𝛼−𝛽 − 𝜆)
14 Abstract and Applied Analysis
By (51), for 𝑧 ∈ C and |𝜆𝑧𝛽−𝛼 | < 1, we have for 𝑠 = 0, 𝑞, 𝑗 = 1, . . . , 𝑙, 𝑘 ∈ N+ . Besides, we also have 𝑠(𝛼 −
𝛽) + 𝑘𝛽 + 𝑞 − 𝑗 ≥ (𝛼 − 𝛽) + 𝛽 + 1 − 𝑙 = 𝛼 + 1 − 𝑙 > 0 for
𝑧𝑗−1−𝛽−𝑘𝛽 𝑧(𝛼−𝛽)−(𝛼+𝑘𝛽+1−𝑗) 𝛼−𝑞
𝑞, 𝑗 = 1, . . . , 𝑙, 𝑠, 𝑘 ∈ N+ . These imply that 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) = 0.
𝑘+1
= 𝑘+1
(𝑧𝛼−𝛽 − 𝜆) (𝑧𝛼−𝛽 − 𝜆) Thus the relations in (151) are valid. The proof is finished.
∞
𝜇𝑘 𝜕𝑘 𝛼−𝑞 ∞ 𝑠 ̂ (172)
=∑ 𝐷 [∑𝜆 ℎ (𝑡, 𝑡0 )] 𝑚−2
𝑘=0
𝑘! 𝜕𝜆𝑘 Δ,𝑡0 𝑠=0 𝑠(𝛼−𝛽)+𝛼+𝑘𝛽−𝑗 𝛼
𝐷Δ,𝑡
𝛽 𝛼
𝑦 (𝑡) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡) − ∑ 𝐴 𝑘 𝐷Δ,𝑡𝑘 0 𝑦 (𝑡) = 0
0
𝑘=0
∞
𝜇𝑘 𝜕𝑘 ∞ 𝑠 ̂ (176)
=∑ ∑𝜆 ℎ (𝑡, 𝑡0 ) . (𝑡 > 𝑡0 ; 𝑚 ∈ N \ {1, 2} ; 0 = 𝛼0 < 𝛼1
𝑘=0
𝑘! 𝜕𝜆𝑘 𝑠=0 𝑠(𝛼−𝛽)+𝑘𝛽+𝑞−𝑗
𝛼−𝑞 𝛼−𝑞 < ⋅ ⋅ ⋅ < 𝛼𝑚−2 < 𝛽 < 𝛼; 𝜆, 𝐴 0 , . . . , 𝐴 𝑚−2 ∈ R) .
For 𝑞 > 𝑗, 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) = 0, and for 𝑞 = 𝑗, 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) =
1. Thus we have 𝑊𝛼 (𝑡0 ) = 1. It follows from Lemma 58
that the functions 𝑦𝑗 (𝑡), 𝑗 = 1, 2, . . . , 𝑙 in (166) are linearly Theorem 63. Let 𝑚 ∈ N\{1, 2}, 𝑙−1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N), let 𝛽 and
independent solutions, and then they yield the fundamental 𝛼1 , . . . , 𝛼𝑚−2 be such that 𝛼 > 𝛽 > 𝛼𝑚−2 > ⋅ ⋅ ⋅ > 𝛼1 > 𝛼0 = 0,
system of solutions to (165). Furthermore, if 𝑞 < 𝑗, then we and let 𝜆, 𝐴 0 , . . . , 𝐴 𝑚−2 ∈ R. Then the functions
rewrite (172) as follows:
𝑠=1 𝑚−2
𝑘 𝜕𝑛
∞ 𝑘 𝑘 × [ ∏ (𝐴 ] ) ] ] 𝐹 𝑚−2 (𝜆, 𝑡, 𝑡0 )
𝜇 𝜕 ̂ ]=0 𝜕𝜆𝑛 Δ 𝛼−𝛽,𝛼+1−𝑗+∑]=0 (𝛽−𝛼] )𝑘]
+∑ ℎ (𝑡, 𝑡0 )
𝑘=1
𝑘! 𝜕𝜆𝑘 𝑘𝛽+𝑞−𝑗 (177)
∞
𝜇𝑘 𝜕𝑘 ∞ 𝑠 ̂
+∑ ∑𝜆 ℎ (𝑡, 𝑡0 )
𝑘=1
𝑘! 𝜕𝜆𝑘 𝑠=1 𝑠(𝛼−𝛽)+𝑘𝛽+𝑞−𝑗 (173) yield the fundamental system of solutions to (176) provided that
the series in (177) are convergent. The inner sum is taken over
∞ all 𝑘0 , . . . , 𝑘𝑚−2 ∈ N0 such that 𝑘0 + ⋅ ⋅ ⋅ + 𝑘𝑚−2 = 𝑛. Moreover,
= ∑𝜆𝑠 ̂ℎ𝑠(𝛼−𝛽)+𝑞−𝑗 (𝑡, 𝑡0 ) if 𝛼 + 1 − 𝑙 > 𝛽 > 𝛼𝑚−2 + 𝑙 − 1, then 𝑦𝑗 (𝑡), 𝑗 = 1, 2, . . . , 𝑙, in
𝑠=1 (177) satisfy (151).
∞
𝜇𝑘 𝜕𝑘 ̂ Proof. Let 𝑙𝑚−1 − 1 < 𝛽 ≤ 𝑙𝑚−1 , 𝑙𝑘 − 1 < 𝛼𝑘 < 𝑙𝑘 (𝑘 = 1, . . . , 𝑚 −
+∑ ℎ (𝑡, 𝑡0 )
𝑘=1
𝑘! 𝜕𝜆𝑘 𝑘𝛽+𝑞−𝑗 2; 0 ≤ 𝑙1 ≤ ⋅ ⋅ ⋅ ≤ 𝑙𝑚−1 ≤ 𝑙). Applying the Laplace transform to
(176) and using (141) as in (167), we obtain
∞
𝜇𝑘 𝜕𝑘 ∞ 𝑠 ̂
+∑ ∑𝜆 ℎ (𝑡, 𝑡0 ) .
𝑘=1
𝑘! 𝜕𝜆𝑘 𝑠=1 𝑠(𝛼−𝛽)+𝑘𝛽+𝑞−𝑗
𝑙
𝑧𝑗−1
If 𝛼 + 1 − 𝑙 > 𝛽 > 𝑙 − 1, then 𝑠(𝛼 − 𝛽) + 𝑞 − 𝑗 ≥ (𝛼 − 𝛽) + 1 − 𝑙 > 0 L {𝑦 (𝑡)} (𝑧, 𝑡0 ) = ∑𝑑𝑗 , (178)
for 𝑘 = 0, 𝑞, 𝑗 = 1, . . . , 𝑙, 𝑠 ∈ N+ , and 𝑘𝛽 + 𝑞 − 𝑗 ≥ 𝛽 + 1 − 𝑙 > 0 𝑗=1 𝑧𝛼 − 𝜆𝑧𝛽 − ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧
𝛼𝑘
Abstract and Applied Analysis 15
where From (178), (180), and (182), we derive the solution to (176),
𝛼−𝑗 𝛽−𝑗 as
𝑑𝑗 = 𝐷Δ,𝑡0 𝑦 (𝑡0 ) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡0 )
𝑙
𝑚−2
𝛼 −𝑗 𝑦 (𝑡) = ∑ 𝑑𝑗 𝑦𝑗 (𝑡) , (183)
− ∑ 𝐴 𝑘 𝐷Δ,𝑡𝑘 0 𝑦 (𝑡0 ) (𝑗 = 1, . . . , 𝑙1 ) , 𝑗=1
𝑘=1
𝛼−𝑗
𝑑𝑗 = 𝐷Δ,𝑡0 𝑦 (𝑡0 ) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡0 )
𝛽−𝑗 which shows that arbitrary solution 𝑦(𝑡) can be expressed by
𝑦𝑗 (𝑡), 𝑗 = 1, . . . , 𝑙, where 𝑦𝑗 (𝑡) (𝑗 = 1, . . . , 𝑙) are given by (177).
𝑚−2 For 𝑞, 𝑗 = 1, . . . , 𝑙, the direct evaluation yields
𝛼 −𝑗
− ∑ 𝐴 𝑘 𝐷Δ,𝑡𝑘 0 𝑦 (𝑡0 ) (𝑗 = 𝑙1 + 1 . . . , 𝑙2 ) ,
𝛼−𝑞
𝑘=2 𝐷Δ,𝑡0 𝑦𝑗 (𝑡)
..
. ∞ 𝑚−2
𝛼−𝑞 { 1 𝑘
= 𝐷Δ,𝑡0 { ∑ ( ∑ ) [ ∏ (𝐴 ] ) ] ]
𝛼−𝑗
𝑑𝑗 = 𝐷Δ,𝑡0 𝑦 (𝑡0 ) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡0 )
𝛽−𝑗
(𝑗 = 𝑙𝑚−2 + 1, . . . , 𝑙𝑚−1 ) , 𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
{𝑛=0 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛 ]=0
𝛼−𝑗
𝑑𝑗 = 𝐷Δ,𝑡0 𝑦 (𝑡0 ) (𝑗 = 𝑙𝑚−1 + 1, . . . , 𝑙) . 𝜕𝑛 }
× 𝑛
𝐹𝛼−𝛽,𝛼+1−𝑗+∑𝑚−2 (𝛽−𝛼] )𝑘] (𝜆; 𝑡, 𝑡0 ) }
(179) 𝜕𝜆 Δ ]=0
}
Here ∑𝑛𝑘=𝑚 𝐴 𝑘 := 0 (𝑚 > 𝑛). For 𝑧 ∈ C and
∞ 𝑚−2
| ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧
𝛼𝑘 −𝛽
/(𝑧𝛼−𝛽 − 𝜆)| < 1, we have 1 𝑘
= ∑( ∑ ) [ ∏ (𝐴 ] ) ] ]
1 𝑛=0 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 ! ]=0
𝑧𝛼 − 𝜆𝑧𝛽 − ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧𝛼𝑘 𝜕 𝑛 ∞
𝜆𝑠 ̂ℎ𝑠(𝛼−𝛽)+𝛼−𝑗+∑𝑚−2 (𝛽−𝛼] )𝑘] (𝑡, 𝑡0 )]
𝛼−𝑞
× 𝐷Δ,𝑡 [ ∑
𝜕𝜆𝑛 0 ]=0
𝑧−𝛽 1 𝑠=0
= ⋅
𝑧𝛼−𝛽 − 𝜆 (1 − ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧
𝛼𝑘 −𝛽 / (𝑧𝛼−𝛽 − 𝜆)) ∞
1 𝑚−2
𝑘
= ∑( ∑ ) [ ∏ (𝐴 ] ) ] ]
∞ 𝑚−2 𝑛 𝑛=0 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 ! ]=0
𝑧−𝛽
=∑ 𝑛+1
( ∑ 𝐴 𝑘 𝑧𝛼𝑘 −𝛽 ) (180)
𝑛=0 (𝑧
𝛼−𝛽 − 𝜆) 𝑘=0 𝜕𝑛 ∞ 𝑠 ̂
× ∑𝜆 ℎ 𝑚−2 (𝑡, 𝑡0 ) .
𝜕𝜆𝑛 𝑠=0 𝑠(𝛼−𝛽)+∑]=0 (𝛽−𝛼] )𝑘] +𝑞−𝑗
∞
𝑛!
= ∑( ∑ ) (184)
𝑛=0 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
𝛼−𝑞 𝛼−𝑞
𝑚−2 −𝛽−∑𝑚−2
For 𝑞 > 𝑗, 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) = 0, and for 𝑞 = 𝑗, 𝐷Δ,𝑡0 𝑦𝑗 (𝑡0 ) =
𝑧 ]=0 (𝛽−𝛼] )𝑘]
𝑘 1. Thus we have 𝑊𝛼 (𝑡0 ) = 1. It follows from Lemma 58
× [ ∏ (𝐴 ] ) ] ] 𝑛+1
,
]=0 (𝑧𝛼−𝛽 − 𝜆) that the functions 𝑦𝑗 (𝑡), 𝑗 = 1, 2, . . . , 𝑙 in (177) are linearly
independent solutions and then they yield the fundamental
if we also take into account the following relation:
system of solutions to (176). Furthermore, if 𝑞 < 𝑗, then we
𝑛
(𝑥0 + ⋅ ⋅ ⋅ + 𝑥𝑚−2 ) rewrite (184) as follows:
𝑚−2 (181) 𝛼−𝑞
𝑛! 𝐷Δ,𝑡0 𝑦𝑗 (𝑡)
=( ∑ ) ∏ 𝑥𝑘] ,
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 ! ]=0 𝜐
= 𝐷Δ,𝑡0 ̂ℎ𝛼−𝑗 (𝑡, 𝑡0 )
𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛 𝛼−𝑞
𝑗−1−𝛽−∑𝑚−2
]=0 (𝛽−𝛼] )𝑘]
𝑧 ∞
𝑛+1 1
(𝑧𝛼−𝛽 − 𝜆) +∑( ∑ )
𝑛=1 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
(𝛼−𝛽)−(𝛼+1−𝑗+∑𝑚−2
]=0 (𝛽−𝛼] )𝑘] )
𝑧 𝑚−2
= 𝑛+1 𝑘 𝜕𝑛 ̂
(𝑧𝛼−𝛽 − 𝜆) × [ ∏ (𝐴 ] ) ] ] ℎ 𝑚−2 (𝑡, 𝑡0 )
]=0 𝜕𝜆𝑛 ∑]=0 (𝛽−𝛼] )𝑘] +𝑞−𝑗
1 𝜕𝑛
= LΔ,𝑡0 { 𝑛 𝐹𝛼−𝛽,𝛼+1−𝑗+∑𝑚−2 (𝛽−𝛼] )𝑘] (𝜆; 𝑡, 𝑡0 )} (𝑧) . ∞
𝑛! 𝜕𝜆 ∇ ]=0 1
+∑( ∑ )
(182) 𝑛=1 𝑘0 +⋅⋅⋅+𝑘𝑚−2 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
16 Abstract and Applied Analysis
𝑚−2
𝑘 𝜕𝑛 ∞ 𝑠 ̂ Using the inverse Laplace transform L−1 Δ from here, we
× [ ∏ (𝐴 ] ) ] ] ∑𝜆 ℎ 𝑚−2 (𝑡, 𝑡0 )
]=0 𝜕𝜆𝑛 𝑠=1 𝑠(𝛼−𝛽)+∑]=0 (𝛽−𝛼] )𝑘] +𝑞−𝑗 obtain a particular solution to (186) in the form
∞ L {𝑓 (𝑡)} (𝑧, 𝑡0 )
= ∑𝜆𝑠 ̂ℎ𝑠(𝛼−𝛽)+𝑞−𝑗 (𝑡, 𝑡0 ) 𝑦 (𝑡) = L−1 [ ] (𝑡) . (189)
𝐴 0 + ∑𝑚
𝑘=1 𝐴 𝑘 𝑧
𝛼𝑘
𝑠=1
Using the Laplace convolution formula
∞
1
+∑( ∑ ) L {𝑓 ∗ 𝑔} (𝑧, 𝑡0 ) = L {𝑓} (𝑧, 𝑡0 ) L {𝑔} (𝑧, 𝑡0 ) , (190)
𝑛=1 𝑘 +⋅⋅⋅+𝑘 =𝑛 𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
0 𝑚−2
Applying the Laplace transform to (186) and taking (187) into 𝑦 (𝑡) = (𝑓 ∗ 𝐺𝛼,𝛽;𝜆,𝜇 (𝑠)) (𝑡) , (197)
account, we have ∞
𝜇𝑘 𝜕𝑘
𝐺𝛼,𝛽;𝜆,𝜇 (𝑡) = ∑ 𝐹 (𝜆, 𝑡, 𝑡0 ) (198)
𝑚
𝛼𝑘 𝑘=0
𝑘! 𝜕𝜆𝑘 Δ 𝛼−𝛽,𝛼+𝑘𝛽
[𝐴 0 + ∑ 𝐴 𝑘 𝑧 ] L {𝑦 (𝑡)} (𝑧, 𝑡0 ) = L {𝑓 (𝑡)} (𝑧, 𝑡0 ) .
𝑘=1 provided that the series in (198) and the integral in (197) are
(188) convergent.
Abstract and Applied Analysis 17
Proof. Equation (196) is the same as (186) with 𝑚 = 2, 𝛼2 = 𝛼, Proof. Equation (203) is the same equation as (186) with 𝛼𝑚 =
𝛼1 = 𝛽, 𝐴 2 = 1, 𝐴 1 = −𝜆, and 𝐴 0 = −𝜇, and (191) is given by 𝛼, 𝛼𝑚−1 = 𝛽, 𝐴 𝑚 = 1, . . . , 𝐴 𝑚−1 = −𝜆, and with −𝐴 𝑘 instead
of 𝐴 𝑘 for 𝑘 = 0, . . . , 𝑚 − 2. Since 𝛼0 = 0, (191) takes the form
1
𝐺𝛼,𝛽 (𝑡) = L−1 { } (𝑡) . (199) 1
𝑧𝛼 − 𝜆𝑧𝛽 − 𝜇 𝐺𝛼1 ,...,𝛼𝑚−2 ,𝛽,𝛼;𝜆 (𝑡) = L−1 { } (𝑡) .
𝑧𝛼 − 𝜆𝛼𝛽 − ∑𝑚−2
𝑘=0 𝐴 𝑘 𝑧
𝛼𝑘
𝑚−2
As in the case of ordinary differential equations, a general
𝛼 𝛽 𝛼 solution to the nonhomogeneous equation (186) is a sum
𝐷Δ,𝑡0
𝑦 (𝑡) − 𝜆𝐷Δ,𝑡0 𝑦 (𝑡) − ∑ 𝐴 𝑘 𝐷Δ,𝑡𝑘 0 𝑦 (𝑡) − 𝐴 0 𝑦 (𝑡)
𝑘=1 (203) of a particular solution to this equation and of the general
solution to the corresponding homogeneous equation (140).
= ̂ℎ𝑟 (𝑡, 𝑡0 ) (𝑡 > 𝑡0 ) Therefore, the results established in Section 7 and in Section 8
can be used to derive general solutions to the nonhomo-
with 𝑚 ∈ N \ {1, 2}, 0 < 𝛼1 < ⋅ ⋅ ⋅ < 𝛼𝑚−2 < 𝛽 < 𝛼, and geneous equations (193), (196), and (203). The following
𝜆, 𝐴 0 , . . . , 𝐴 𝑚−2 ∈ R. statements can thus be derived from Theorems 59, 64, 61, 65
and Theorems 63 and 66, respectively.
Theorem 66. Let 𝑚 ∈ N \ {1, 2}, 0 = 𝛼0 < 𝛼1 < ⋅ ⋅ ⋅ < 𝛼𝑚−2 <
𝛽 < 𝛼, and let 𝜆, 𝐴 0 , . . . , 𝐴 𝑚−2 ∈ R. Then (203) is solvable, Theorem 67. Let 𝑙 − 1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N), 𝜆 ∈ R. Then (193) is
and its particular solution has the form solvable, and its general solution is given by
𝑙
𝑦 (𝑡) = (𝑓 ∗ 𝐺𝛼1 ,...,𝛼𝑚−2 ,𝛽,𝛼;𝜆 (𝑠)) (𝑡) , (204)
𝑦 (𝑡) = ∑ 𝑐𝑗Δ 𝐹𝛼,𝛼+1−𝑗 (𝜆, 𝑡, 𝑡0 ) + 𝑓 (𝑡) , (209)
𝑗=1
𝐺𝛼1 ,...,𝛼𝑚−2 ,𝛽,𝛼;𝜆 (𝑡)
where 𝑐𝑗 (𝑗 = 1, . . . , 𝑙) are arbitrary real constants.
∞
1
=∑( ∑ ) Theorem 68. Let 𝑙 − 1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N), 0 < 𝛽 < 𝛼, 𝜆, 𝜇 ∈ R.
𝑛=0 𝑘 +⋅⋅⋅+𝑘 =𝑛
𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
0 𝑚−2 Then (196) is solvable, and its general solution has the form
𝑚−2
𝑘 𝜕𝑛 𝑙 ∞
𝜇𝑛 𝜕𝑛
× [ ∏ (𝐴 𝜐 ) 𝜐 ] 𝐹 𝑚−2 (𝜆, 𝑡, 𝑡0 ) 𝑦 (𝑡) = ∑𝑐𝑗 ∑ 𝐹 (𝜆, 𝑡, 𝑡0 )
𝜕𝜆𝑛 Δ 𝛼−𝛽,𝛼+∑𝜐=0 (𝛽−𝛼𝜐 )𝑘𝜐 𝑛 Δ 𝛼−𝛽,𝛼+𝑛𝛽+1−𝑗
𝑛=0 𝑛! 𝜕𝜆
𝜐=0
𝑗=1 (210)
(205)
+ (𝑓 ∗ 𝐺𝛼,𝛽,𝜆,𝜇 (𝑠)) (𝑡) ,
provided that the series (205) and integral in (204) are
convergent. The inner sum is taken over all 𝑘0 , . . . , 𝑘𝑚−2 such where 𝐺𝛼,𝛽,𝜆,𝜇 (𝑡) is given by (198) and 𝑐𝑗 (𝑗 = 1, . . . , 𝑙) are
that 𝑘0 + ⋅ ⋅ ⋅ + 𝑘𝑚−2 = 𝑛. arbitrary real constants.
18 Abstract and Applied Analysis
Theorem 69. Let 𝑚 ∈ N \ {1, 2}, 𝑙 − 1 < 𝛼 ≤ 𝑙 (𝑙 ∈ N), let 𝛽 Conflict of Interests
and 𝛼1 , 𝛼2 , . . . , 𝛼𝑚−2 be such that 𝛼 > 𝛽 > 𝛼𝑚−2 > ⋅ ⋅ ⋅ > 𝛼1 >
𝛼0 = 0 and 𝛼 − 𝑙 + 1 ≥ 𝛽, and let 𝜆, 𝐴 0 , . . . , 𝐴 𝑚−2 ∈ R. Then The authors declare that there is no conflict of interests
(203) is solvable, and its general solution is given by regarding the publication of this paper.
Authors’ Contribution
𝑙 ∞
1
𝑦 (𝑡) = ∑𝑐𝑗 ∑ ( ∑ ) All authors contributed equally and significantly to the
𝑗=1 𝑛=0 𝑘 +⋅⋅⋅+𝑘 =𝑟 𝑘0 ! ⋅ ⋅ ⋅ 𝑘𝑚−2 !
0 𝑚−2 writing of this paper. All authors read and approved the final
𝑚−2 paper.
𝑘
× [ ∏ (𝐴 𝜐 ) 𝜐 ]
𝜐=0
(211)
Acknowledgments
𝑛
𝜕 First, the authors are very grateful to the referees for their
⋅ 𝐹 𝑚−2 (𝜆, 𝑡, 𝑡0 )
𝜕𝜆𝑛 Δ 𝛼−𝛽,𝛼+1−𝑗+∑𝜐=0 (𝛽−𝛼𝜐 )𝑘𝜐 careful reading of the paper, and lots of valuable comments
and suggestions, which greatly improve this manuscript.
+ (𝑓 ∗ 𝐺𝛼1 ,...,𝛼𝑚−2, 𝛽,𝛼;𝜆 (𝑠)) (𝑡) ,
Next, this work was supported by the National Natural Sci-
ence Foundation of China (11171286) and by Jiangsu Province
where 𝐺𝛼1 ,...,𝛼𝑚−2, 𝛽,𝛼;𝜆 (𝑡) is given by (205) and 𝑐𝑗 (𝑗 = 1, . . . , 𝑙) Colleges and Universities Graduate Scientific Research Inno-
vative Program (CXZZ12-0974).
are arbitrary real constant.
References
9. Conclusions
[1] M. Holm, “Sum and difference compositions in discrete frac-
In this paper, we first give a generalized definition of frac- tional calculus,” CUBO A Mathematical Journal, vol. 13, no. 3,
tional Δ-power function on general time scale by inversion pp. 153–184, 2011.
of Laplace transform and shift transform. The fractional Δ- [2] F. M. Atici and P. W. Eloe, “A transform method in discrete frac-
power function on a time scale is an important basis of tional calculus,” International Journal of Difference Equations,
fractional Δ-integral and fractional Δ-differential on time vol. 2, pp. 165–176, 2007.
scales. Then, based on the fractional Δ-power function, [3] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory and
Applications of Fractional Differentia Equation, Elsevier B.V.,
we give a new definition of Riemann-Liouville fractional
London, UK, 2006.
Δ-integral and Riemann-Liouville fractional Δ-derivative
[4] J. Cheng, Theory of Fractional Difference Equation, Xiamen
on time scales. Some of properties of Riemann-Liouville University Press, Fujian, China, 2011.
fractional Δ-integral and Riemann-Liouville fractional Δ- [5] C. Kou, H. Zhou, and Y. Yan, “Existence of solutions of initial
derivative on time scales are studied in detail. On this value problems for nonlinear fractional differential equations
basis, equivalencies of Cauchy type problem with Riemann- on the half-axis,” Nonlinear Analysis, Theory, Methods and
Liouville fractional Δ-derivative and nonlinear Volterra inte- Applications, vol. 74, no. 17, pp. 5975–5986, 2011.
gral equation are obtained. By employing Laplace transform, [6] R. P. Agarwal, V. Lakshmikantham, and J. J. Nieto, “On the
we derive explicit solutions to homogeneous and nonho- concept of solution for fractional differential equations with
mogeneous equations of Riemann-Liouville fractional Δ- uncertainty,” Nonlinear Analysis, Theory, Methods and Applica-
derivative with constant coefficient. We give the conditions tions, vol. 72, no. 6, pp. 2859–2862, 2009.
when the solutions of linear fractional differential equation [7] Y. Liu, W. Zhang, and X. Liu, “A sufficient condition for
will be linearly independent and when these linearly inde- the existence of a positive solution for a nonlinear fractional
pendent solutions form the fundamental system of solutions. differential equation with the Riemann-Liouville derivative,”
On the other hand, we know that continuous fractional Applied Mathematics Letters, vol. 25, no. 11, pp. 1986–1992, 2012.
differential equation theory and discrete fractional difference [8] V. Lakshmikantham and A. S. Vatsala, “Basic theory of frac-
equation theory have been studied by many authors, the tional differential equations,” Nonlinear Analysis, Theory, Meth-
ods and Applications, vol. 69, no. 8, pp. 2677–2682, 2008.
existence and uniqueness of the solution to the boundary
[9] K. Diethelm, The Analysis of Fractional Differential Equations,
value problems for fractional differential equations have
Springer, Berlin, Germany, 2010.
been studied a lot by many methods involving partial order
[10] R. A. C. Ferreira and D. M. Torres, “Fractional h-
method, fixed point method, lower and upper solutions
difference equations arising from the calculus of variations,”
method, transform method, and so on. For example, for the http://arxiv.org/abs/1101.5904v1 .
recent developments about continuous fractional differential [11] C. S. Goodrich, “Existence and uniqueness of solutions to
equations and discrete fractional difference equations, one a fractional difference equation with nonlocal conditions,”
can refer to [3, 5–11] and the references therein. However, the Computers and Mathematics with Applications, vol. 61, no. 2, pp.
fractional differential equation theory on time scales is still an 191–202, 2011.
open problem. Therefore, the authors believe that the present [12] M. Bohner and A. Peterson, Dynamic Equations on Time Scales,
work will potentiate further research in the study of fractional An Introduction with Applications, Birkhäauser, Boston, Mass,
differential equation theory on time scales in the future. USA, 2001.
Abstract and Applied Analysis 19
Research Article
On Solutions to Fractional Discrete Systems with
Sequential ℎ-Differences
Copyright © 2013 Małgorzata Wyrwas et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
We study the subject of a behaviour of the solutions of systems with sequential fractional ℎ-differences. We give formulas for
the unique solutions to initial value problems for systems in linear and semilinear cases. Moreover, the sufficient condition that
guaranties the positivity of considered systems is presented.
Section 3 presents systems with sequential fractional differ- The next definition with another notation was stated in
ences with results on uniqueness of solutions. We include [17]. Here we use more suitable summations.
semilinear systems in Section 4. The Section 5 concerns pos-
itivity of considered systems. Finally, the illustrative example Definition 4. For a function 𝑥 ∈ F(ℎN)𝑎 the fractional ℎ-sum
is presented. of order 𝛼 > 0 is given by
ℎ 𝑛
2. Preliminaries (𝑎 Δ−𝛼
ℎ 𝑥) (𝑡) := ∑ (𝑡 − 𝜎 (𝑎 + 𝑘ℎ))(𝛼−1)
ℎ 𝑥 (𝑎 + 𝑘ℎ) , (6)
Γ (𝛼) 𝑘=0
Let us denote by F𝐷 the set of all real valued functions
defined on 𝐷. Let ℎ > 0, 𝛼 > 0 and put (ℎN)𝑎 := {𝑎, 𝑎 + where 𝑡 = 𝑎 + (𝛼 + 𝑛)ℎ, 𝑛 ∈ N0 . Moreover, we define
ℎ, 𝑎 + 2ℎ, . . .}, where 𝑎 ∈ R. Let (𝑎 Δ0ℎ 𝑥)(𝑡) := 𝑥(𝑡).
It is important to notice that the operator 𝑎 Δ−𝛼
ℎ changes
R𝑁 𝑁
+ = {𝑥 ∈ R : 𝑥𝑖 ≥ 0, 1 ≤ 𝑖 ≤ 𝑁} , 𝑁 ∈ N1 . (1) the domains of functions.
Then the operator 𝜎 : (ℎN)𝑎 → (ℎN)𝑎 is defined by 𝜎(𝑡) := Remark 5. Note that 𝑎 Δ−𝛼
ℎ : F(ℎN)𝑎 → F(ℎN)𝑎+𝛼ℎ .
𝑡 + ℎ. The next two definitions of ℎ-difference operator were According to the definition of ℎ-factorial function the
originally given in [16, 17]. formula given in Definition 4 can be rewritten as
𝑛
Definition 1. For a function 𝑥 ∈ F(ℎN)𝑎 the forward ℎ- Γ (𝛼 + 𝑛 − 𝑘)
(𝑎 Δ−𝛼 𝛼
ℎ 𝑥) (𝑡) = ℎ ∑ 𝑥 (𝑎 + 𝑘ℎ)
difference operator is defined as 𝑘=0
Γ (𝛼) Γ (𝑛 − 𝑘 + 1)
(7)
𝑛
𝑥 (𝜎 (𝑡)) − 𝑥 (𝑡) 𝑛−𝑘+𝛼−1
(Δ ℎ 𝑥) (𝑡) := , 𝑡 = 𝑎 + 𝑛ℎ, 𝑛 ∈ N0 , (2) = ℎ𝛼 ∑ ( ) 𝑥 (𝑎 + 𝑘ℎ)
ℎ 𝑛−𝑘
𝑘=0
while the ℎ-difference sum is given by
for 𝑡 = 𝑎 + (𝛼 + 𝑛)ℎ, 𝑛 ∈ N0 . Observe that (𝑎 Δ−𝛼
ℎ 𝑥)(𝑎 + 𝛼ℎ) =
𝑛 ℎ𝛼 𝑥(𝑎) and for 𝛼 = 1 we have again (3).
(𝑎 Δ−1
ℎ 𝑥) (𝑡) := ℎ ∑ 𝑥 (𝑎 + 𝑘ℎ) , (3)
𝑘=0 Remark 6. In [7] one can find the following form of the
fractional ℎ-sum of order 𝛼 > 0:
where 𝑡 = 𝑎 + (𝑛 + 1)ℎ, 𝑛 ∈ N0 and (𝑎 Δ−1
ℎ 𝑥)(𝑎) := 0.
ℎ𝛼 𝑡−𝛼ℎ 𝑡 − 𝜎 (𝑘) (𝛼−1)
(𝑎 Δ−𝛼
ℎ 𝑥) (𝑡) = ∑( ) 𝑥 (𝑘) (8)
Definition 2. For arbitrary 𝛼 ∈ R the ℎ-factorial function is Γ (𝛼) 𝑘=𝑎 ℎ ℎ=1
defined by
that can be useful in implementation.
Γ ((𝑡/ℎ) + 1)
𝑡ℎ(𝛼) := ℎ𝛼 , (4)
Γ ((𝑡/ℎ) + 1 − 𝛼) The following definition can be found in [33] for ℎ = 1 or
in [10] for an arbitrary ℎ > 0.
where Γ is the Euler gamma function, that is, Γ(𝑧) =
+∞
∫0 𝑥𝑧−1 𝑒−𝑥 d𝑥 for all 𝑧 ∈ C such that Re 𝑧 > 0, (𝑡/ℎ) ∉ Z− := Definition 7. Let 𝛼 ∈ (0, 1]. The Caputo ℎ-difference operator
𝛼
{−1, −2, −3, . . .}, and we use the convention that division at a 𝑎 Δ ℎ,∗ 𝑥
of order 𝛼 for a function 𝑥 ∈ F(ℎN)𝑎 is defined by
pole yields zero.
(𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑡) := (𝑎 Δ−(1−𝛼)
ℎ (Δ ℎ 𝑥)) (𝑡) , 𝑡 ∈ (ℎN)𝑎+(1−𝛼)ℎ .
Notice that if we use the general binomial coefficient (9)
( 𝑎𝑏 ) := (Γ(𝑎 + 1))/(Γ(𝑏 + 1)Γ(𝑎 − 𝑏 + 1)), then (4) can be
𝛼
rewritten as Remark 8. Note that 𝑎 Δ ℎ,∗ : F(ℎN)𝑎 → F(ℎN)𝑎+(1−𝛼)ℎ , where
𝑡 𝛼 ∈ (0, 1].
𝑡ℎ(𝛼) 𝛼 ℎ
= ℎ Γ (𝛼 + 1) ( ) . (5) We need the power rule formulas in the sequel. Firstly,
𝛼
we easily notice that for 𝑝 ≠0 the well-defined ℎ-factorial
functions have the following property:
In the sequel we need the following technical properties.
(𝑝) (𝑝−1)
Δ ℎ (𝑡 − 𝑎)ℎ = 𝑝(𝑡 − 𝑎)ℎ . (10)
Proposition 3 (see [12]). Let 𝛼 ∈ R.
𝑗−𝛼−1
More properties of ℎ-factorial functions can be found in [10].
(1) For 𝑗 ∈ N0 one has (−1)𝑗 ( 𝛼𝑗 ) = ( 𝑗 ), where 𝛼, 𝑗 In our consideration the crucial role plays the power rule
are such that both sides are well defined; formula presented in [16], that is,
𝑗−𝛼−1
(2) For 𝑛 ∈ N0 one has ∑𝑛𝑗=0 ( 𝑗 ) = ( 𝑛−𝛼
𝑛 ); Γ (𝜇 + 1) (𝜇+𝛼)
(𝑎 Δ−𝛼
ℎ 𝜓) (𝑡) = (𝑡 − 𝑎 + 𝜇ℎ)ℎ , (11)
(3) For 𝑘 ∈ N1 one has ( 𝛼−1
𝑘 ) + ( 𝛼−1
𝑘−1 ) = ( 𝛼𝑘 ). Γ (𝜇 + 𝛼 + 1)
Abstract and Applied Analysis 3
(𝜇)
where 𝜓(𝑟) = (𝑟 − 𝑎 + 𝜇ℎ)ℎ , 𝑟 ∈ (ℎN)𝑎 , 𝑡 ∈ (ℎN)𝑎+𝛼ℎ . Note Let 𝜇 := 𝑘𝛼 + 𝑠𝛽. For 𝑟 ∈ (ℎN)𝑙ℎ we define the following
(𝜇)
that using the general binomial coefficient one can write (11) ℎ-factorial function 𝜓(𝑟) := (𝑟 + 𝜇ℎ)ℎ . Since
as
1
𝜑𝑘,𝑠 ((𝑛 − 𝑙) ℎ) =
𝑛 + 𝛼 + 𝜇 𝜇+𝛼 Γ (𝑘𝛼 + 𝑠𝛽 + 1)
(𝑎 Δ−𝛼
ℎ 𝜓) (𝑡) = Γ (𝜇 + 1) ( )ℎ . (12)
𝑛
(𝑘𝛼+𝑠𝛽)
⋅ ((𝑛 − 𝑙) ℎ + 𝑘𝛼ℎ + 𝑠𝛽ℎ)ℎ (17)
If 𝜓 ≡ 1, then we have for 𝜇 = 0, 𝑎 = (1−𝛼)ℎ and 𝑡 = 𝑛ℎ+𝑎+𝛼ℎ 1
= 𝜓 (𝑛ℎ − 𝑙ℎ)
Γ (𝜇 + 1)
1
(𝑎 Δ−𝛼
ℎ 1) (𝑡) = (𝑡 − 𝑎)(𝛼)
Γ (𝛼 + 1) ℎ for 𝑛 ≥ 𝑙 and 𝜑𝑘,𝑠 ((𝑚 − 𝑙)ℎ) = 0 for 𝑚 < 𝑙, by (11) we get
(13)
Γ (𝑛 + 𝛼 + 1) 𝑛+𝛼 𝛼 1
= ℎ𝛼 = ( )ℎ . (0 Δ−𝛼 −𝛼
ℎ 𝜑𝑘,𝑠 ) (𝑡) = (𝑙ℎ Δ ℎ 𝜑𝑘,𝑠 ) (𝑡) = ( Δ−𝛼 𝜓) (𝑡)
Γ (𝛼 + 1) Γ (𝑛 + 1) 𝑛 Γ (𝜇 + 1) 0 ℎ
Let us define special functions that we use in the next section 1 Γ (𝜇 + 1) (𝜇+𝛼)
= (𝑡 + 𝜇ℎ)ℎ (18)
to write the formula for solutions. Γ (𝜇 + 1) Γ (𝜇 + 𝛼 + 1)
𝑛 − (𝑙 + 1) + (𝑘 + 1) 𝛼 + 𝑠𝛽 (𝑘+1)𝛼+𝑠𝛽
(a) 𝜑0,0 (𝑛ℎ) = 1; =( )ℎ
𝑛 − (𝑙 + 1)
𝛼 −𝛼
(b) 𝜑1,0 (𝑛ℎ) = ( 𝑛+𝛼
𝑛 ) ℎ = (0 Δ ℎ 1)(𝑛ℎ+𝛼ℎ) and the values = 𝜑𝑘+1,𝑠 ((𝑛 − 𝑙 − 1) ℎ) .
𝜑1,0 ((𝑛 − 1)ℎ) = ( 𝑛−1 ) ℎ𝛼 = (0 Δ−𝛼
𝑛+𝛼−1
ℎ 1)((𝑛 − 1)ℎ + 𝛼ℎ) (19)
are neglected for 𝑛 = 0;
𝑛
summations for the case when an order is from the interval 𝛼−1
= ℎ1−𝛼 ∑ (−1)𝑗 ( )
(0, 1]. 𝑗
𝑗=0
𝛼−1
⋅ 𝑥 (𝑎 + ℎ) − (−1)𝑛 ( ) 𝑥 (𝑎) } .
Note that 𝑦 : (ℎN)𝑏 → R𝑁. Then we apply the next difference 𝑛
operator of order 𝛽 on the new function 𝑦 and consider here (27)
an initial value problem stated by the system
(𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) = 𝑦 (𝑏 + 𝑛ℎ) , (23) Since by Proposition 3 the following relation ( 𝛼−1 𝛼−1
𝑘 ) + ( 𝑘−1 ) =
𝛼
( 𝑘 ) holds, one gets
𝛽
(𝑏 Δ ℎ,∗ 𝑦) (𝑛ℎ) = 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) , (24)
𝛼
where 𝑓 : (ℎN)0 × R𝑁 → R𝑁, with initial values 𝑦 (𝑏 + 𝑛ℎ) = ℎ−𝛼 [𝑥 (𝑎 + (𝑛 + 1) ℎ) − ( ) 𝑥 (𝑎 + 𝑛ℎ)
1
(𝑎 Δ𝛼ℎ,∗ 𝑥) (0) = 𝑥0 , 𝛼
(25) + ( ) 𝑥 (𝑎 + (𝑛 − 1) ℎ)
2
𝑥 (𝑎) = 𝑥𝑎 , (26)
𝛼
− ( ) 𝑥 (𝑎 + (𝑛 − 2) ℎ)
3
where 𝑥𝑎 , 𝑥0 are constant vectors from R𝑁.
Solutions of the state equations (23) and (24) of the 𝛼
+ ( ) 𝑥 (𝑎 + (𝑛 − 3) ℎ)
sequential fractional discrete-time system can be computed 4
in the recursive way. From Definition 7 and by Proposition 3,
𝛼
(23) can be written as + ⋅ ⋅ ⋅ + (−1)𝑛−1 ( ) 𝑥 (𝑎 + 2ℎ)
𝑛−1
(28)
𝑦 (𝑏 + 𝑛ℎ) 𝛼
+ (−1)𝑛 ( ) 𝑥 (𝑎 + ℎ)
𝑛
= (𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) = (𝑎 Δ−(1−𝛼)
ℎ (Δ ℎ 𝑥)) (𝑛ℎ)
𝛼−1
−(−1)𝑛 ( ) 𝑥 (𝑎)]
ℎ 𝑛 𝑛
= ∑ (𝑛ℎ − 𝑎 − 𝑘ℎ − ℎ)(−𝛼)
ℎ (Δ ℎ 𝑥) (𝑎 + 𝑘ℎ)
Γ (1 − 𝛼) 𝑘=0
𝑛
𝛼
𝑛 = ℎ−𝛼 [∑ (−1)𝑗 ( ) 𝑥 (𝑎 + (𝑛 − 𝑗 + 1) ℎ)
ℎ Γ (𝑛 − 𝑘 − 𝛼 + 1) 𝑗
= ∑ ℎ−𝛼 (Δ ℎ 𝑥) (𝑎 + 𝑘ℎ) [𝑗=0
Γ (1 − 𝛼) 𝑘=0 Γ (𝑛 − 𝑘 + 1)
𝛼−1
𝑛
𝑛−𝑘−𝛼 +(−1)𝑛+1 ( ) 𝑥 (𝑎) ] .
= ℎ1−𝛼 ∑ ( ) (Δ ℎ 𝑥) (𝑎 + 𝑘ℎ) 𝑛
𝑛−𝑘 ]
𝑘=0
Abstract and Applied Analysis 5
𝑛
Repeating the same computation for (24) one gets 𝛼+𝛽
= ∑ (−1)𝑘 ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑘
𝛽 −(1−𝛽) 𝑘=0
𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) = (𝑏 Δ ℎ,∗ 𝑦) (𝑛ℎ) = (𝑏 Δ ℎ (Δ ℎ 𝑦)) (𝑛ℎ)
𝛼+𝛽 𝛽 𝛽−1
𝑛 + (−1)𝑛+1 [( )−( )+( )] 𝑥 (𝑎 + ℎ)
−𝛽 𝛽 𝑖 𝑛+1 𝑛+1 𝑛
= ℎ ∑(−1) ( ) 𝑦 (𝑏 + (𝑛 − 𝑖 + 1) ℎ)
𝑖
𝑖=0 𝑛
𝛼−1 𝛽−1
+ [∑(−1)𝑛−𝑖 ( ) + (−1)𝑛 ( )] 𝑥 (𝑎) .
𝛽−1 𝑛−𝑖+1 𝑛
+ ℎ−𝛽 (−1)𝑛+1 ( ) 𝑦 (𝑏) . 𝑖=0
𝑛
(31)
(29)
𝑛−𝑖+1
𝛼 𝛼+𝛽 𝛽 𝛽−1
⋅ [ℎ−𝛼 ∑ (−1)𝑗 ( ) + (−1)𝑛+1 [( )−( ) +( )] 𝑥 (𝑎 + ℎ)
𝑗 𝑛+1 𝑛+1 𝑛
𝑗=0
[
𝛼−2 𝛽−1
⋅ 𝑥 (𝑎 + (𝑛 − 𝑖 − 𝑗 + 2) ℎ) + [1 + (−1)𝑛 ( ) + (−1)𝑛 ⋅ ( )] 𝑥 (𝑎) .
𝑛+1 𝑛
(32)
𝛼−1
−𝛼
+ℎ (−1) 𝑛−𝑖+2
( ) 𝑥 (𝑎) ]
𝑛−𝑖+1
] Consequently, since ℎ𝛼 𝑥0 = 𝑥(𝑎 + ℎ) − 𝑥(𝑎), 𝑥(𝑎 + 2ℎ) =
𝛽−1 ℎ𝛼+𝛽 ⋅ 𝑓(0, 𝑥(𝑎)) − 𝛼𝑥(𝑎) + (1 + 𝛼) ⋅ 𝑥(𝑎 + ℎ) and for 𝑛 ≥ 1
+ ℎ−𝛽 (−1)𝑛+1 ( ) ⋅ 𝑥0
𝑛
𝑥 (𝑎 + (𝑛 + 2) ℎ)
𝑛 𝑛−𝑖+1
−𝛼−𝛽 [∑ ∑ (−1) 𝑖+𝑗 𝛽 𝛼
=ℎ ( )( ) = ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ))
𝑖 𝑗
[𝑖=0 𝑗=0
𝑛
𝛼+𝛽
⋅ 𝑥 (𝑎 + (𝑛 − 𝑖 − 𝑗 + 2) ℎ) + ∑ (−1)𝑘+1 ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑘
𝑘=1
𝑛
𝑛−𝑖+2 𝛼−1
+ ∑(−1) ( ) 𝑥 (𝑎) 𝛼+𝛽 𝛽 𝛽−1
𝑛−𝑖+1 + (−1)𝑛 [( )−( )+( )] 𝑥 (𝑎 + ℎ)
𝑖=0 𝑛+1 𝑛+1 𝑛
Then using the Chu-Vandermonde identity, that is, Using Proposition 3 we get for 𝑛 ≥ 1
∑𝑘𝑖=0 ( 𝛽𝑖 ) ( 𝑘−𝑖
𝛼
) = ( 𝛼+𝛽
𝑘
), 𝛼, 𝛽 ∈ R, 𝑘 ∈ Z, one gets
𝑥 (𝑎 + (𝑛 + 2) ℎ)
𝛼+𝛽
ℎ ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ))
= ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ))
𝑛+1 𝑘
𝛽 𝛼 𝑛
= ∑ (−1)𝑘 ∑ ( ) ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ) 𝛼+𝛽
𝑖 𝑘−𝑖 + ∑ (−1)𝑘+1 ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑘=0 𝑖=0 𝑘
𝑘=1
𝑛
𝛼−1 𝛽−1 𝑛−𝛽 𝑛−𝛽 𝑛−𝛼−𝛽
+ [∑(−1)𝑛−𝑖 ( ) + (−1)𝑛 ( )] 𝑥 (𝑎) + [( )+( )−( )] 𝑥 (𝑎 + ℎ)
𝑛−𝑖+1 𝑛 𝑛 𝑛+1 𝑛+1
𝑖=0
= ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) with initial conditions (25) and (26), that is, (𝑎 Δ𝛼ℎ,∗ 𝑥)(0) =
𝑛 𝑥0 and 𝑥(𝑎) = 𝑥𝑎 , 𝑥0 , 𝑥𝑎 ∈ R𝑁, is given by the following
𝛼+𝛽
+ ∑ (−1)𝑘+1 ( ) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ) formula:
𝑘
𝑘=1 𝑛
(34) for 𝑛 ∈ N0 .
Therefore the solution of (23) and (24) is given recursively
Proof. Notice that for 𝑛 = 0 we get 𝑥(𝑎+0⋅ℎ) = 𝐴0 (𝜑0,0 (0)𝑥𝑎 +
by the following formula:
𝜑1,0 (−ℎ)𝑥0 ). Since 𝜑0,0 (0) = 1 and 𝜑1,0 (−ℎ) = 0, we get 𝑥(𝑎 +
𝑥 (𝑎 + ℎ) = ℎ𝛼 𝑥0 + 𝑥 (𝑎) , 0 ⋅ ℎ) = 𝑥𝑎 .
For 𝑛 > 0 let us define the sequence {𝑥𝑚 }𝑚≥1 in the
𝑥 (𝑎 + 2ℎ) = ℎ𝛼+𝛽 ⋅ 𝑓 (0, 𝑥 (𝑎)) − 𝛼𝑥 (𝑎) following way:
where 𝑔𝑚 (𝑛ℎ) = (0 Δ ℎ 𝑓̃𝑚 )(𝑏 + 𝑛ℎ) and 𝑓̃𝑚 (𝑛ℎ) = 𝐴𝑥𝑚 (𝑎 + 𝑛ℎ)
−𝛽
= ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ))
(35) with 𝑥0 (𝑎 + 𝑛ℎ) = 𝑥𝑎 .
𝑛
+ ∑ (−1)𝑘+1 (
𝛼+𝛽
) 𝑥 (𝑎 + (𝑛 − 𝑘 + 2) ℎ) We calculate the first step. As 𝑓̃0 (𝑛ℎ) = 𝐴𝑥0 (𝑎 + 𝑛ℎ) =
𝑘 𝐴𝑥𝑎 , then 𝑔0 (𝑛ℎ) = 𝐴𝑥𝑎 (0 Δ−𝛽 1)(𝑏 + 𝑛ℎ) = 𝐴𝑥𝑎 𝜑0,1 ((𝑛 − 1)ℎ).
𝑘=1
Going further,
𝑛−𝛽+1 𝑛−𝛼−𝛽
+ [( )−( )] 𝑥 (𝑎 + ℎ)
𝑛+1 𝑛+1 𝑥1 (𝑎 + 𝑛ℎ) = 𝑥𝑎 𝜑0,0 (𝑛ℎ) + 𝑥0 𝜑1,0 ((𝑛 − 1) ℎ)
(43)
𝑛−𝛼+2 𝑛−𝛽 + (0 Δ−𝛼
ℎ 𝑔0 ) (𝑎 + 𝑛ℎ) ,
+ [( )−1−( )] 𝑥 (𝑎) .
𝑛+1 𝑛
which could be written as
Another possibility of computing the solution of (23) and
(24) is to use Proposition 13 twice and then, for 𝑛 ≥ 1, we get: 𝑥1 (𝑎 + 𝑛ℎ) = 𝑥𝑎 𝜑0,0 (𝑛ℎ) + 𝑥0 𝜑1,0 ((𝑛 − 1) ℎ)
(44)
−𝛽 𝛽 + 𝐴𝑥𝑎 𝜑1,1 ((𝑛 − 2) ℎ)
(0 Δ ℎ (𝑏 Δ ℎ,∗ 𝑦)) (𝑏 + 𝑛ℎ) = 𝑦 (𝑏 + 𝑛ℎ) − 𝑦 (𝑏)
and, using Proposition 11, we get
= (𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) − 𝑥0 , (36)
𝑥2 (𝑎 + 𝑛ℎ) = 𝑥𝑎 𝜑0,0 (𝑛ℎ) + 𝑥0 𝜑1,0 ((𝑛 − 1) ℎ)
(0 Δ−𝛼 𝛼
ℎ ( 𝑎 Δ ℎ,∗ 𝑥)) (𝑎 + 𝑛ℎ) = 𝑥 (𝑎 + 𝑛ℎ) − 𝑥𝑎 .
+ 𝐴𝑥𝑎 𝜑1,1 ((𝑛 − 2) ℎ)
Hence (45)
+ 𝐴𝑥0 𝜑2,1 ((𝑛 − 3) ℎ)
−𝛽 ̃
(𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) = 𝑥0 + (0 Δ ℎ 𝑓) (𝑏 + 𝑛ℎ) , (37)
+ 𝐴2 𝑥𝑎 𝜑2,2 ((𝑛 − 4) ℎ) .
̃
where 𝑓(𝑛ℎ) := 𝑓(𝑛ℎ, 𝑥(𝑎 + 𝑛ℎ)). Nextly,
Taking 𝑚 tending to +∞ we get formula (41) as the
𝑥 (𝑛ℎ + 𝑎) = 𝑥𝑎 + 𝑥0 (0 Δ−𝛼 −𝛼
ℎ 1) (𝑎 + 𝑛ℎ) + (0 Δ ℎ 𝑔) (𝑎 + 𝑛ℎ) ,
solution of (39) and (40) with initial conditions (25) and
(38) (26).
̃ + 𝑛ℎ).
where 𝑔(𝑛ℎ) = (0 Δ ℎ 𝑓)(𝑏
−𝛽
3.1. Semilinear Sequential Systems. Firstly we state a technical
Firstly we prove the formula for the unique solution in lemma and notations.
linear case of (23) and (24): 𝑓(𝑛ℎ, 𝑥(𝑛ℎ + 𝑎)) = 𝐴𝑥(𝑎 + 𝑛ℎ),
where 𝐴 is a constant square matrix of degree 𝑛. Lemma 15. Let 𝛾 : (ℎN)0 → R and 𝛼 > 0. Let
(0 Δ−𝑘𝛼
ℎ 𝛾)(𝑘𝛼ℎ + 𝑛ℎ) = 𝛾1 (𝑘𝛼ℎ + 𝑛ℎ) and 𝛾 ̃1 (𝑛ℎ) := 𝛾1 (𝑘𝛼ℎ +
Theorem 14. The solution to the system 𝑛ℎ) for 𝑘 ∈ N1 . Then for 𝑘 ∈ N1 one gets
(𝑎 Δ𝛼ℎ,∗ 𝑥) (𝑛ℎ) = 𝑦 (𝑏 + 𝑛ℎ) , (39) (0 Δ−𝛼 ̃1 ) (𝑡) = (0 Δ−(𝑘+1)𝛼
ℎ 𝛾 ℎ 𝛾) (𝑘𝛼ℎ + 𝑡) , (46)
𝛽
(𝑏 Δ ℎ,∗ 𝑦) (𝑛ℎ) = 𝐴𝑥 (𝑎 + 𝑛ℎ) (40) where 𝑡 = 𝛼ℎ + 𝑛ℎ.
Abstract and Applied Analysis 7
Proof. First let us consider the case 𝑘 = 1. Then from Using the power rule formula for 𝜇 = 0 and by Remark 10 we
Proposition 12 we can write can write the recursive formula for the solution to nonlinear
problem given by (23) and (24) and conditions (25) and (26):
(𝛼ℎ Δ−𝛼 −𝛼 −2𝛼
ℎ (0 Δ ℎ 𝛾)) (𝑡) = (0 Δ ℎ 𝛾) (𝑡) , (47)
where 𝑡 = 2𝛼ℎ + 𝑛ℎ, 𝑛 ∈ N0 . 𝑥 (𝑎 + 𝑛ℎ) = 𝑥𝑎 + 𝑥0 𝜑1,0 ((𝑛 − 1) ℎ)
Let 𝛾1 (𝛼ℎ + 𝑛ℎ) = (0 Δ−𝛼 ̃1 (𝑛ℎ) := 𝛾1 (𝛼ℎ +
ℎ 𝛾)(𝛼ℎ + 𝑛ℎ) and 𝛾 𝑛−1
𝑛ℎ). Then
+ ∑ 𝜑̃1,1 (𝑛ℎ − ℎ − 𝜎 (𝑟ℎ)) 𝑓 (𝑟ℎ, 𝑥 (𝑎 + 𝑟ℎ)) .
(0 Δ−𝛼
ℎ 𝛾̃1 ) (𝛼ℎ + 𝑛ℎ) 𝑟=0
(53)
ℎ 𝑛
= ∑(𝑛ℎ + 𝛼ℎ − 𝜎 (𝑟ℎ))(𝛼−1) 𝛾̃1 (𝑟ℎ)
Γ (𝛼) 𝑟=0 ℎ The given formula (53) also works for 𝑛 = 0 as 𝜑̃1,1 (−2ℎ) = 0.
Then 𝑥(𝑎 + 0ℎ) = 𝑥𝑎 . We can check the next steps:
ℎ 𝑛+𝛼 (48)
= ∑ (𝑛ℎ + 2𝛼ℎ − 𝜎 (𝑠ℎ))(𝛼−1)
ℎ 𝛾1 (𝑠ℎ)
Γ (𝛼) 𝑠=𝛼 𝑥 (𝑎 + ℎ) = 𝑥𝑎 + 𝑥0 𝜑1,0 (0) + 𝜑̃1,1 (−ℎ) 𝑓 (0, 𝑥 (𝑎))
= (𝛼ℎ Δ−𝛼
ℎ 𝛾1 ) (2𝛼ℎ + 𝑛ℎ) = 𝑥𝑎 + 𝑥0 ℎ𝛼 ,
= (0 Δ−2𝛼
ℎ 𝛾) (2𝛼ℎ + 𝑛ℎ) .
𝑥 (𝑎 + 2ℎ) = 𝑥𝑎 + 𝑥0 𝜑1,0 (ℎ) + 𝜑̃1,1 (0ℎ) 𝑓 (0, 𝑥 (𝑎)) (54)
Equation (46) for 𝑘 > 1 follows inductively. + 𝜑̃1,1 (−ℎ) 𝑓 (ℎ, 𝑥 (𝑎 + ℎ))
𝑛−1 for 𝑛 ∈ N0 .
+ ∑ 𝜑̃1,1 (𝑛ℎ − ℎ − 𝜎 (𝑟ℎ)) 𝑓 (𝑟ℎ, 𝑥 (𝑎 + 𝑟ℎ)) .
𝑟=0 Proof. Note that 𝜑0,0 (0) = 1, 𝜑1,0 (−ℎ) = 0 and 𝜑𝑘+1,𝑘+1 (−ℎ)
(52) = 0 for 𝑘 ≥ 0, so we get 𝑥(𝑎 + 0ℎ) = 𝑥𝑎 .
8 Abstract and Applied Analysis
For 𝑛 > 0 based on the proof for linear case we can write Moreover,
the solution formula as follows:
𝑛 𝑥𝑖 (𝑎 + ℎ) ≥ 𝑥𝑎𝑖 = 𝑥𝑖 (𝑎) , (63)
𝑘
𝑥 (𝑎 + 𝑛ℎ) = ∑ 𝐴 𝜑𝑘,𝑘 ((𝑛 − 2𝑘) ℎ) 𝑥𝑎
𝑘=0 where 𝑥(𝑎 + ℎ) = [𝑥1 (𝑎 + ℎ), . . . , 𝑥𝑁(𝑎 + ℎ)]𝑇 and 𝑥(𝑎) =
𝑛 [𝑥1 (𝑎), . . . , 𝑥𝑁(𝑎)]𝑇 = [𝑥𝑎1 , . . . , 𝑥𝑎𝑁]𝑇 . Let 𝑥𝑖 (𝑎 + 𝑛ℎ) and
+ ∑ 𝐴𝑘 𝜑𝑘+1,𝑘 ((𝑛 − (2𝑘 + 1)) ℎ) 𝑥0 (59) 𝑓𝑖 (𝑛ℎ, 𝑥(𝑎 + 𝑛ℎ)) denote the 𝑖th coordinates of the vectors
𝑘=0 𝑥(𝑎 + 𝑛ℎ) and 𝑓(𝑛ℎ, 𝑥(𝑎 + 𝑛ℎ)), respectively. Then since 𝑥𝑖 (𝑎 +
𝑛
ℎ) ≥ 𝑥𝑖 (𝑎) ≥ 0 for 𝑖 = 1, . . . , 𝑁,
+ ∑ 𝐴𝑘 (0 Δ−𝜏
ℎ 𝛾) ((𝑛 − 1 + 𝜏) ℎ) ,
𝑘=0 𝑥𝑖 (𝑎 + 2ℎ) = ℎ𝛼+𝛽 ⋅ 𝑓𝑖 (0, 𝑥 (𝑎))
where 𝜏 = (𝑘 + 1)(𝛼 + 𝛽). Then taking into account formulas − 𝛼𝑥𝑖 (𝑎) + (1 + 𝛼) ⋅ 𝑥𝑖 (𝑎 + ℎ)
(50) and (51) we get the form (58) as the solution of (56) and
(57) with initial conditions (25) and (26). ≥ ℎ𝛼+𝛽 ⋅ 𝑓𝑖 (0, 𝑥 (𝑎)) − 𝛼𝑥𝑖 (𝑎) + (1 + 𝛼) ⋅ 𝑥𝑖 (𝑎)
Similarly as in [32] we will use the recursive formula (35) For 𝑛 = 1 we have
to show the positivity of the considered systems. Observe that
the systems considered in this paper are of the sequential 𝛼+𝛽
𝑥𝑖 (𝑎 + 3ℎ) = ℎ𝛼+𝛽 ⋅ 𝑓𝑖 (ℎ, 𝑥 (𝑎 + ℎ)) − ( )
type while in [32] the sequential systems are not studied. 2
Moreover, the Grünwald-Letnikov operator with the step
2−𝛽
equal to one is used in [32], whereas we study the systems with ⋅ 𝑥𝑖 (𝑎 + ℎ) + (𝛼 + 𝛽) ⋅ 𝑥𝑖 (𝑎 + 2ℎ) + ( )
the ℎ-differences of Caputo type. So in our case the steps are 2
equal to ℎ. 3−𝛼
From [32] we have the following lemma. ⋅ 𝑥 (𝑎 + ℎ) + [( ) − 1 − (1 − 𝛽)] 𝑥𝑖 (𝑎) .
2
(66)
Lemma 18. If 0 < 𝛼 < 1, then (−1)𝑖+1 ( 𝛼𝑖 ) > 0, 𝑖 = 1, 2, 3, . . ..
Since ( 3−𝛽
3
) − ( 2−𝛼−𝛽
3
) > 0 for 0 < 𝛼, 𝛽 ≤ 1, 𝑥𝑖 (𝑎 + ℎ) ≥
𝑥𝑖 (𝑎) ≥ 0 for 𝑖 = 1, . . . , 𝑁, and (61) holds for 𝑛 = 2, using (60)
we have 0.4
3−𝛽 4−𝛼
𝑥𝑖 (𝑎 + 4ℎ) ≥ [( )+( )
3 3
(69) 0.3
2−𝛼−𝛽 3
−( ) − ( )] 𝑥𝑖 (𝑎) ≥ 0
3 3
for 0 < 𝛼, 𝛽 ≤ 1.
0.2
Assume that 𝑥(𝑎 + 𝑘ℎ) ∈ R𝑁 + for 𝑘 = 1, 2, 3, . . . , 𝑛 + 1.
Using the properties of gamma function one can show that
( 𝑛−𝛽+1
𝑛+1
) ≥ ( 𝑛−𝛼−𝛽
𝑛+1
) for 𝑛 ≥ 3. Then applying 𝑥(𝑎 + ℎ) ≥ 𝑥(𝑎)
to (35) we get for 𝑛 ≥ 3 0.1
𝑥𝑖 (𝑎 + (𝑛 + 2) ℎ)
𝛼+𝛽
= ℎ𝛼+𝛽 ⋅ 𝑓𝑖 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) − ( ) 𝑥𝑖 (𝑎 + 𝑛ℎ)
2 0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16
+ (𝛼 + 𝛽) 𝑥𝑖 (𝑎 + (𝑛 + 1) ℎ) Figure 1: Consider the following system (𝑎 Δ0.4 ℎ,∗ 𝑥)(𝑛ℎ) = 𝑦(𝑏 + 𝑛ℎ),
(𝑏 Δ0.5
ℎ,∗ 𝑦)(𝑛ℎ) = 𝐴𝑥(𝑎 + 𝑛ℎ), where 𝐴 is given in Example 22, with
𝑛
𝛼+𝛽 initial conditions: 𝑥0 = (0, 1), 𝑥𝑎 = (0, 0.1), ℎ0.9 𝐴 − ( 0.9 2×2
2 ) 𝐼2 ∈ R+ .
+ ∑ (−1)𝑘+1 ( ) 𝑥𝑖 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑘
𝑘=3
+ [(
𝑛−𝛽+1
)−(
𝑛−𝛼−𝛽
)] 𝑥𝑖 (𝑎 + ℎ) using the induction principle we get that 𝑥(𝑎 + 𝑛ℎ) ∈ R𝑁
+ for
𝑛+1 𝑛+1 all 𝑛 ≥ 1.
𝑛−𝛼+2 𝑛−𝛽 Corollary 20. Let 0 < 𝛼 + 𝛽 ≤ 1 and 𝐼𝑁 denote the identity
+ [( )−1−( )] 𝑥𝑖 (𝑎)
𝑛+1 𝑛 matrix. If 𝑥0 , 𝑥𝑎 ∈ R𝑁 𝑁
+ and for all 𝑛 ≥ 1, 𝑥 ∈ R+
𝛼+𝛽
≥ ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) − ( ) 𝑥𝑖 (𝑎 + 𝑛ℎ) ℎ𝛼+𝛽 𝐴 − (
𝛼+𝛽
) 𝐼𝑁 ∈ R𝑁×𝑁 ,
2 2 + (71)
+ (𝛼 + 𝛽) 𝑥 (𝑎 + (𝑛 + 1) ℎ)
then 𝑥(𝑎 + 𝑛ℎ) ∈ R𝑁
+ for all 𝑛 ≥ 1 that is, (39) and (40) is
𝑛 positive.
𝛼+𝛽
+ ∑ (−1)𝑘+1 ( ) 𝑥𝑖 (𝑎 + (𝑛 − 𝑘 + 2) ℎ) (70)
𝑘
𝑘=3 Remark 21. In [32] the sufficient condition concerning the
positivity of the linear discrete systems with Grünwald-
𝑛−𝛽+1 𝑛−𝛼−𝛽
+ [(
𝑛+1
)−(
𝑛+1
) Letnikov operator is as follows: 𝐴 + 𝛼𝐼𝑁 ∈ R𝑁×𝑁 + . In our
case since we have systems with the sequential fractional ℎ-
𝑛−𝛼+2 𝑛−𝛽 difference, in our condition ℎ and both orders 𝛼 and 𝛽 appear.
+( )−1−( )] 𝑥𝑖 (𝑎) Note that taking ℎ = 1 the sufficient condition (71) has the
𝑛+1 𝑛
form 𝐴 − ( 𝛼+𝛽2
) 𝐼𝑁 ∈ R𝑁×𝑁
+ .
𝛼+𝛽
= ℎ𝛼+𝛽 ⋅ 𝑓 (𝑛ℎ, 𝑥 (𝑎 + 𝑛ℎ)) − ( ) 𝑥𝑖 (𝑎 + 𝑛ℎ)
2 Let us now consider some examples that illustrate the
solution of the considered systems.
+ (𝛼 + 𝛽) 𝑥 (𝑎 + (𝑛 + 1) ℎ)
𝑛 Example 22. Let 𝑁 = 2, 𝛼 = 0.4, 𝛽 = 0.5 and ℎ = 0.01. Then
𝑘+1 𝛼+𝛽 let us take 𝑎 = −0.006 and 𝑏 = −0.005 and consider the linear
+ ∑ (−1) ( ) 𝑥𝑖 (𝑎 + (𝑛 − 𝑘 + 2) ℎ)
𝑘 system with sequential difference in the following form:
𝑘=3
Research Article
Numerical Solution of Fuzzy Fractional Pharmacokinetics Model
Arising from Drug Assimilation into the Bloodstream
Copyright © 2013 Ali Ahmadian et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We propose a Jacobi tau method for solving a fuzzy fractional pharmacokinetics. This problem can model the concentration of the
drug in the blood as time increases. The proposed approach is based on the Jacobi tau (JT) method. To illustrate the reliability of
the method, some special cases of the equations are solved as test examples. The method reduces the solution of the problem to the
solution of a system of algebraic equations. Error analysis included the fractional derivative error estimation, and the upper bound
of the absolute errors is introduced for this method.
composition, the blood composition, and blood flow to mainly two: first and foremost, PKPD models incorporating
the tissue is utilized to develop a partitioning model [8– fractional calculus have not been proposed; second is compu-
10]. Möhler et al. [11] proposed a model that describes cell tational: while the analytical solution of fractional differential
growth and detailed investigations on the metabolism and equations is available in special cases, it turns out that even
the kinetics of the influenza infection cycle which allow the simplest PKPD models that can be constructed using
for the optimization of influenza virus vaccine production. fractional calculus do not allow analytical solutions [23, 24].
In addition to these simple models, more complex math- In this paper, we propose new category of PKPD mod-
ematical manipulations called mathematical models have els incorporating fuzzy fractional calculus and investigate
also been used to describe pharmacokinetics [12]. There their behavior using purposely written algorithm. The main
are many classifications of pharmacokinetics models. In a purpose of this paper is to attract the attention of the
specific classification for pharmacokinetics modeling, two field into these possibly interesting families of PKPD mod-
general approaches can be considered: compartment-based els. In particular, we will focus on compartment models,
modeling [13] and noncompartment-based modeling [14]. direct concentration-response relationships describing mod-
Both types of modeling take advantage of the quantitative els relating drug concentration (in plasma or biophase) to a
structure-pharmacokinetics relationships that are described pharmacodynamic effect.
by empirical mathematical algorithms. They can be used to The concept of fractional or noninteger order derivation
estimate the activity of a compound based on its chemical and integration can be traced back to the genesis of integer
structure in a numeric format [15]. order calculus itself [25–27]. Due to its tremendous scope and
In this study, our focus is to find the approximate solution applications in several disciplines, a considerable attention
of the fuzzy fractional model of the compartment model for has been given to exact and numerical solutions of fractional
the flow of antihistamine in the blood. To this end, we use differential equations which is an extremely difficult task.
the operational matrix of the Caputo fractional derivative of Moreover, the solution techniques and their reliability are
the Jacobi tau approximation based on the Jacobi polynomials rather more important aspects. Several methods have been
to derive the fuzzy fractional approximate solution under proposed by many researchers to solve the fractional order
Hukuhara differentiability (H-differentiability). To the best differential equations such as Taylor series [28], variational
of our knowledge, there are no results on the JT method for iteration method [29–31], adomian decomposition method
solving pharmacokinetics equations arising in mathematical [32, 33], fractional differential transform method (FDTM)
physics. This partially motivated our interest in such model [34, 35], and homotopy analysis method [36, 37].
and solution method. As it is known, the spectral method is one of the flexible
methods of discretization for most types of differential
1.1. Compartment Model. The compartment model frame- equations [38–40]. Historically, spectral method has been
work is an extremely natural and valuable means with which relegated to fractional calculus, but in few years, it has been
to formulate models for processes which have inputs and successfully applied for the fractional equation models based
outputs over time [16]. Compartment model itself can be on the different types of orthogonal polynomials such as
classified in one-compartment model and two-compartment Block pulse functions [41, 42], Legendre polynomials [43–
model. In the former one, the body is depicted as a kinetically 46], Chebyshev polynomials [47–49], Laguerre polynomials
homogeneous unit, while in the latter one the body resolves [50–53], and Bernstein polynomials [54–56]. Doha et al.
into a central compartment and a peripheral compartment [57] introduced the shifted Jacobi operational matrix of
[17]. According to the one-compartment model, the drug fractional derivative which is based on Jacobi tau method for
instantaneously distributes throughout the body, and at the solving numerically linear multiterm fractional differential
same time it equilibrates between tissues. Thus, the drug equations with initial or boundary conditions. Subsequently,
concentration time profile shows a monophasic response. Kazem [58] generalized Jacobi integral operational matrix to
On the other hand, in two-compartment model the drug fractional calculus. Afterwards, Doha et al. [59] proposed a
does not achieve instantaneous equilibration between the two direct solution techniques for solving the linear multiorder
compartments. Although, in this model, two compartments fractional initial value problems with constant and variable
have no anatomical or physiological meaning, it is usually coefficients using shifted Jacobi tau method and quadrature
surmised that the central compartment is composed of tissues shifted Jacobi tau method, respectively. In this paper, we
that are highly perfused such as heart, lungs, kidneys, liver, intend to extend the application of the Jacobi polynomi-
and brain while the peripheral compartment comprises less als to solve fuzzy fractional pharmacokinetics model of
well-perfused tissues such as muscle, fat, and skin [17]. order [0, 1].
On the other hand, a considerable attention has been
1.2. Methods Have Been Proposed to Solve Fractional Pharma- made to the fractional differential equations in the sense
cokinetics Models. While an increasing number of fractional of fuzzy setting theory. Agarwal et al. [60] is among the
order integrals and differential equations applications have pioneers who presented the concept of solutions for fractional
been reported in the physics [18, 19], signal processing differential equations with uncertainty. Thereafter, the exis-
[20], engineering, and bioengineering literatures [21, 22], tence and uniqueness of the solution of the fuzzy fractional
little attention has been paid to this class of models in the differential equations (FFDEs) under the Riemann-Liouville
pharmacokinetics-pharmacodynamic (PKPD) literature. The and Caputo’s fuzzy fractional differentiability were investi-
reasons for the lack of application to pharmacodynamics are gated in the literature [61–64]. However, the application of
Abstract and Applied Analysis 3
the numerical methods for solving FFDEs is unknown and In this paper, the set of all fuzzy numbers is denoted
traceless in the literature of the fuzzy fractional calculus, and by RF .
only a few number of researches have been reported for the
approximate solution of FFDEs [64–69]. Definition 2. Let 𝑢 ∈ RF and 𝑟 ∈ [0, 1]. The 𝑟-cut of 𝑢 is
In the present paper, we intend to introduce new families the crisp set [𝑢]𝑟 that contains all elements with membership
of PKPD models based on the application of fractional degree in 𝑢 greater than or equal to 𝑟; that is
calculus to PKPD models. The aim of the paper is not to
[𝑢]𝑟 = {𝑥 ∈ R | 𝑢 (𝑥) ≥ 𝑟} . (3)
claim the superiority of fractional dynamics models with
respect to standard ones, but it is simply to define the new For a fuzzy number 𝑢, its 𝑟-cuts are closed intervals in R, and
families and provide some insights into their qualitative we denote them by
behavior. The main purpose is to apply the fuzzy logic in
differential equations of fractional order which has been used [𝑢]𝑟 = [𝑢1𝑟 , 𝑢2𝑟 ] . (4)
as an effective tool for considering uncertainty in modeling
the processes; FFDEs can also offer a more comprehensive According to Zadeh’s extension principle, the operation
account of the process or phenomenon, specifically for of addition on RF is defined as follows:
analyzing the behavior of the PKPD models. Furthermore, we (𝑢 + V) (𝑥) = sup min {𝑢 (𝑦) , V (𝑥 − 𝑦)} , 𝑥 ∈ R,
suggest possible applications and stimulate further research, 𝑦∈R
(5)
which might, or might not, demonstrate the applicability and
importance of spectral methods by using of the orthogonal and scalar multiplication of a fuzzy number is given by
polynomials for finding the approximate solutions of the 𝑥
PKPD models based on the fuzzy fractional calculus. {𝑢 ( ) , 𝑘 > 0,
(𝑘 ⊙ 𝑢) (𝑥) = { 𝑘 (6)
The structure of this paper is as follows. In the next ̃
section, we briefly recall the mathematical foundations of {0, 𝑘 = 0,
fractional calculus, required definitions of fuzzy setting the- where 0̃ ∈ RF .
ory, and summarize the properties of Jacobi polynomials.
In Section 3, we then provide illustrations of the govern- Definition 3 (see [70]). The distance 𝐷(𝑢, V) between two
ing fraction equation. In Section 4, the proposed method fuzzy numbers 𝑢 and V is defined as
is explained for numerical solution of the derived FFDE.
Section 5 is devoted to the numerical solution of the problem 𝐷 (𝑢, V) = sup 𝑑𝐻 ([𝑢]𝑟 , [V]𝑟 ) , (7)
with different dose of the drug, and the error analysis is 𝑟∈[0,1]
𝑢 (𝜆𝑥 + (1 − 𝜆) 𝑦 ≥ min {𝑢 (𝑥) , 𝑢 (𝑦)}) ; (1) 𝐷∗ (𝑓, 𝑔) := sup 𝐷 (𝑓 (𝑥) , 𝑔 (𝑥)) . (9)
𝑥∈[𝑎,𝑏]
(iii) 𝑢 is upper semicontinuous: for any 𝑥0 ∈ R, it holds Remark 5 (see [73]). Let 𝑓 : [𝑎, 𝑏] → RF be fuzzy contin-
that uous. Then from property (iv) of Hausdorff distance, we can
define
𝑢 (𝑥0 ) ≥ lim± 𝑢 (𝑥) ;
̃ = sup max {𝑓1𝑟 (𝑥) , 𝑓2𝑟 (𝑥)} , ∀𝑥 ∈ [𝑎, 𝑏] .
𝑥 → 𝑥0 (2)
𝐷 (𝑓 (𝑥) , 0)
𝑟∈[0,1]
0 (10)
(iv) [𝑢] = supp(𝑢) is a compact subset of R.
4 Abstract and Applied Analysis
Definition 6 (see [74]). Let 𝑥, 𝑦 ∈ RF . If there exists 𝑧 ∈ In this part, we firstly give some basic definitions and
RF such that 𝑥 = 𝑦 ⊕ 𝑧, then 𝑧 is called the H-difference some properties of fractional calculus [27]. Afterwards, the
of 𝑥 and 𝑦, and it is denoted by 𝑥 ⊖ 𝑦. extension of the fractional differentiability in the sense of
fuzzy concept is provided, and some relevant properties
In this paper, the sign “⊖” always stands for H-difference, which are used in the rest of the paper are given [61, 63, 64].
and note that 𝑥 ⊕ 𝑦 ≠𝑥 + (−𝑦). Also throughout the paper, Let 𝑚 be the smallest integer that exceeds V, then
it is assumed that the Hukuhara difference and generalized Caputo’s fractional derivative operator of order V > 0 is
Hukuhara differentiability exist. defined as
Theorem 7 (see [75]). Let 𝐹 : (𝑎, 𝑏) → RF be a function 𝑐 𝐽𝑚−V 𝐷𝑚 𝑓 (𝑥) , if 𝑚 − 1 < V < 𝑚,
and denote [𝐹(𝑡)]𝑟 = [𝑓𝑟 (𝑡), 𝑔𝑟 (𝑡)], for each 𝑟 ∈ [0, 1]. Then, 𝐷V 𝑓 (𝑥) = { (17)
𝐷𝑚 𝑓 (𝑥) , if V = 𝑚, 𝑚 ∈ N,
(1) if 𝐹 is (1)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are dif- where
ferentiable functions and
𝑥
1
𝑟 𝐽V 𝑓 (𝑥) = ∫ (𝑥 − 𝑡)V−1 𝑓 (𝑡) 𝑑𝑡, V > 0, 𝑥 > 0. (18)
[𝐹 (𝑡)] = [𝑓𝑟 (𝑡) , 𝑔𝑟 (𝑡)] ; (11) Γ (V) 0
For the Caputo derivative, we have
(2) if 𝐹 is (2)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are dif-
𝑐
ferentiable functions and 𝐷V 𝐶 = 0, (𝐶 is a constant) ,
𝑟 𝑐
[𝐹 (𝑡)] = [𝑔𝑟 (𝑡) , 𝑓𝑟 (𝑡)] . (12) 𝐷V 𝑥 𝛽
(i) 𝐿R𝑝F (𝑎, 𝑏), 1 ≤ 𝑝 ≤ ∞ is the set of all fuzzy-valued For the sake of simplicity, we say that the fuzzy-valued
measurable functions 𝑓 on [𝑎, 𝑏] where ||𝑓||𝑝 = function 𝑓 is 𝑐 [(1) − V]-differentiable if it is differentiable as
1 1/𝑝 in Definition 13 case (i), and 𝑓 is 𝑐 [(2) − V]-differentiable if
(∫0 (𝑑(𝑓(𝑡), 0))𝑝 𝑑𝑡) . it is differentiable as in Definition 13 case (ii), and so on for
RF the other cases.
(ii) 𝐶 [𝑎, 𝑏] is a space of fuzzy-valued functions which
are continuous on [𝑎, 𝑏].
Theorem 14 (see [63]). Let 0 < V ≤ 1 and 𝑓 ∈ 𝐴𝐶RF [𝑎, 𝑏];
(iii) 𝐶𝑛RF [𝑎, 𝑏] indicates the set of all fuzzy-valued func- then the fuzzy Caputo fractional derivative exists almost
tions which are continuous up to order 𝑛. everywhere on (𝑎, 𝑏), and for all 0 ≤ 𝑟 ≤ 1 one has
(iv) 𝐴𝐶RF [𝑎, 𝑏] denotes the set of all fuzzy-valued func-
𝑥 𝑓𝑟 (𝑡) 𝑑𝑡
tions which are absolutely continuous. 1
(𝑐 𝐷𝑎+
V
𝑓) (𝑥; 𝑟) = [ ∫ − ,
Note that one can easily find these notations in the crisp Γ (1 − V) 𝑎 (𝑥 − 𝑡)V
context in [25, 27] and references therein. 𝑥 𝑓𝑟 (𝑡) 𝑑𝑡
1 (30)
RF RF ∫ + ]
Definition 11 (see [63]). Let 𝑓 ∈ 𝐶 [𝑎, 𝑏] ∩ 𝐿 [𝑎, 𝑏]. The Γ (1 − V) 𝑎 (𝑥 − 𝑡)V
Riemann-Liouville integral of fuzzy-valued function 𝑓 is
1−V
defined as = [(𝐼𝑎+ 𝐷𝑓−𝑟 ) (𝑥) , (𝐼𝑎+
1−V
𝐷𝑓+𝑟 ) (𝑥)] ,
𝑥 𝑓
1 (𝑡) 𝑑𝑡
(𝑅𝐿 𝐼𝑎+
V
𝑓) (𝑥) = ∫ , when 𝑓 is (1)-differentiable, and
Γ (V) 𝑎 (𝑥 − 𝑡)1−V (24)
𝑥 > 𝑎, 0 < V ≤ 1. (𝑐 𝐷𝑎+
V
𝑓) (𝑥; 𝑟)
𝑥 𝑓𝑟 (𝑡) 𝑑𝑡 𝑥 𝑓𝑟 (𝑡) 𝑑𝑡
Definition 12 (see [63]). Let 𝑓 ∈ 𝐶RF [𝑎, 𝑏] ∩ 𝐿RF [𝑎, 𝑏]. Then, 1 1
𝑓 is said to be Caputo’s H-differentiable at 𝑥 when =[ ∫ + , ∫ −
,]
Γ (1 − V) 𝑎 (𝑥 − 𝑡)V Γ (1 − V) 𝑎 (𝑥 − 𝑡)V
(i) (𝑐 𝐷𝑎+
V
𝑓) (𝑥) = (𝑅𝐿 𝐷𝑎+
𝛽
[𝑓 (𝑡) ⊖ 𝑓 (𝑎)]) (𝑥) , 1−V
= [(𝐼𝑎+ 𝐷𝑓+𝑟 ) (𝑥) , (𝐼𝑎+
1−V
𝐷𝑓−𝑟 ) (𝑥)] ,
(25)
(ii) (𝑐 𝐷𝑎+
V
𝑓) (𝑥) = (𝑅𝐿 𝐷𝑎+
𝛽
[−𝑓 (𝑎) ⊖ (−𝑓 (𝑡))]) (𝑥) . (31)
Definition 13 (see [63]). Let 𝑓 : 𝐿RF [𝑎,𝑏] ∩ 𝐶RF [𝑎, 𝑏] and 𝑥0 ∈ when f is (2)-differentiable.
𝑥
(𝑎, 𝑏) and Φ(𝑥) = (1/Γ(1 − V)) ∫𝑎 (𝑓(𝑡)/(𝑥 − 𝑡)V )𝑑𝑡. We say
Theorem 15 (fuzzy generalized Taylor’s formula [72]).
that 𝑓(𝑥) is fuzzy Caputo fractional differentiable of order
Let 𝑓(𝑥) ∈ 𝐴𝐶RF [𝑎,𝑏] (0, 𝑏], and suppose that 𝑐 𝐷𝑘𝛼 𝑓(𝑥) ∈
0 < V ≤ 1 at 𝑥0 , if there exists an element (𝑐 𝐷𝑎V + 𝑓)(𝑥0 ) ∈
𝐶RF [𝑎,𝑏] (0, 𝑏] for 𝑘 = 0, 1, . . . , 𝑛 + 1 where 0 < 𝛼 < 1, 0 ≤
𝐶RF [𝑎,𝑏] [𝑎, 𝑏] such that for all 0 ≤ 𝑟 ≤ 1, ℎ > 0,
𝑥0 ≤ 𝑥 and 𝑥 ∈ (0, 𝑏]. Then, one has
Φ (𝑥0 + ℎ) ⊖ Φ (𝑥0 )
(i) (𝑐 𝐷𝑎V + 𝑓) (𝑥0 ) = lim+ 𝑟 𝑟
ℎ→0 ℎ [𝑓 (𝑥)] = [𝑓𝑟 (𝑥) , 𝑓 (𝑥)] ,
(26)
Φ (𝑥0 ) ⊖ Φ (𝑥0 − ℎ) 𝑛
𝑥𝑖𝛼
= lim+ , 𝑓𝑟 (𝑥) = ∑ 𝑐 𝑖𝛼 𝑟
𝐷 𝑓 (0+ )
ℎ→0 ℎ Γ (𝑖𝛼 + 1)
𝑖=0
or
𝑐
Φ (𝑥0 ) ⊖ Φ (𝑥0 + ℎ) 𝐷(𝑛+1)𝛼 𝑓𝑟 (𝑥0 ) (𝑛+1)𝛼
(ii) (𝑐 𝐷𝑎V + 𝑓) (𝑥0 ) = lim+ + 𝑥 , (32)
ℎ→0 −ℎ Γ (𝑛𝛼 + 𝛼 + 1)
(27)
𝑛
Φ (𝑥0 − ℎ) ⊖ Φ (𝑥0 ) 𝑟 𝑥𝑖𝛼 𝑐 𝑖𝛼 𝑟
= lim+ , 𝑓 (𝑥) = ∑ 𝐷 𝑓 (0+ )
ℎ→0 −ℎ 𝑖=0 Γ (𝑖𝛼 + 1)
or
𝑐
Φ (𝑥0 + ℎ) ⊖ Φ (𝑥0 ) 𝐷(𝑛+1)𝛼 𝑓𝑟 (𝑥0 ) (𝑛+1)𝛼
(iii) ( 𝑐
𝐷𝑎V + 𝑓) (𝑥0 ) = lim+ + 𝑥 ,
ℎ→0 ℎ Γ (𝑛𝛼 + 𝛼 + 1)
(28) 𝑟 𝑟
Φ (𝑥0 − ℎ) ⊖ Φ (𝑥0 ) where 𝑐 𝐷𝛼 𝑓𝑟 (0) = 𝑐 𝐷𝛼 𝑓𝑟 (𝑥)|𝑥=0 , 𝑐 𝐷𝛼 𝑓 (0) = 𝑐 𝐷𝛼 𝑓 (𝑥)|𝑥=0 .
= lim+ ,
ℎ→0 −ℎ
or 2.2. Jacobi Polynomials. The well-known Jacobi polynomials
𝑐 Φ (𝑥0 ) ⊖ Φ (𝑥0 + ℎ) associated with the parameters (𝛼 > −1, 𝛽 > −1)
(iv) ( 𝐷𝑎V + 𝑓) (𝑥0 ) = lim+ (see, e.g., Luke [79] and Szegö [80]), are a sequence of
ℎ→0 −ℎ
(29) (𝛼,𝛽)
polynomials 𝑃𝑖 (𝑡) (𝑖 = 0, 1, . . .), each, respectively, of
Φ (𝑥0 ) ⊖ Φ (𝑥0 − ℎ) degree 𝑖. For using these polynomials on (0, 𝐿), we present
= lim+ .
ℎ→0 ℎ the shifted Jacobi polynomials by implementing the change
6 Abstract and Applied Analysis
of variable 𝑡 = (2𝑥/𝐿 − 1). Let the shifted Jacobi polyno- where the coefficients 𝑎𝑗 are
(𝛼,𝛽) (𝛼,𝛽)
mials 𝑃𝑖 (2𝑥/𝐿 − 1) be denoted by 𝑃𝐿,𝑖 (𝑥), satisfying the
orthogonality relation
1 𝐿 (𝛼,𝛽) (𝛼,𝛽)
𝐿 𝑎𝑗 = ∫ 𝑃 (𝑥) 𝑢 (𝑥) 𝑤𝐿 (𝑥) 𝑑𝑥, 𝑗 = 0, 1, . . . (39)
(𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) ℎ𝑗 0 𝐿,𝑗
∫ 𝑃𝐿,𝑗 (𝑥) 𝑃𝐿,𝑘 (𝑥) 𝑤𝐿 (𝑥) 𝑑𝑥 = ℎ𝑘 , (33)
0
where 𝑤𝐿
(𝛼,𝛽)
(𝑥) = 𝑥𝛽 (𝐿 − 𝑥)𝛼 and For a given particular problem, only the first (𝑁 + 1)-terms
shifted Jacobi polynomials are used. Therefore, we have
{ 𝐿𝛼+𝛽+1 Γ (𝑘 + 𝛼 + 1) Γ (𝑘 + 𝛽 + 1)
{ 𝑖 = 𝑗,
ℎ𝑘 = { (2𝑘 + 𝛼 + 𝛽 + 1) 𝑘!Γ (𝑘 + 𝛼 + 𝛽 + 1) (34) 𝑁
{ (𝛼,𝛽)
𝑢𝑁 (𝑥) ≃ ∑ 𝑎𝑗 𝑃𝐿,𝑗 (𝑥) .
{0, 𝑖 ≠𝑗.
𝑗=0
(40)
(𝛼,𝛽)
The shifted Jacobi polynomial 𝑃𝐿,𝑖 (𝑥) of degree 𝑖 has the
form Theorem 16 (see [59]). The Caputo fractional derivative of
(𝛼,𝛽) order V of the shifted Jacobi polynomials of degree 𝑖 is obtained
𝑃𝐿,𝑖 (𝑥)
from
𝑖
Γ (𝑖 + 𝛽 + 1) Γ (𝑖 + 𝑘 + 𝛼 + 𝛽 + 1)
= ∑ (−1)𝑖−𝑘 𝑥𝑘 , ∞
Γ (𝑘 + 𝛽 + 1) Γ (𝑖 + 𝛼 + 𝛽 + 1) (𝑖 − 𝑘)!𝑘!𝐿𝑘 (𝛼,𝛽) (𝛼,𝛽)
𝑘=0 𝐷V 𝑃𝐿,𝑖 (𝑥) = ∑ 𝑆V (𝑖, 𝑗, 𝛼, 𝛽) 𝑃𝐿,𝑖 (𝑥) ,
(35) 𝑗=0 (41)
where 𝑖 = ⌈V⌉ , ⌈V⌉ + 1, . . . ,
(𝛼,𝛽) Γ (𝑖 + 𝛽 + 1)
𝑃𝐿,𝑖 (0) = (−1)𝑖 ,
Γ (𝛽 + 1) 𝑖! where
(36)
(𝛼,𝛽) Γ (𝑖 + 𝛼 + 1)
𝑃𝐿,𝑖 (𝐿) = . 𝑆V (𝑖, 𝑗, 𝛼, 𝛽)
Γ (𝛼 + 1) 𝑖!
𝑖
Also we can state the shifted Jacobi polynomial by the
following recurrence relation: = ∑ (−1)𝑖−𝑘 𝐿𝛼+𝛽−V+1 Γ (𝑗 + 𝛽 + 1)
𝑘=⌈V⌉
(𝛼,𝛽)
𝑃𝐿,𝑖 (𝑥) × Γ (𝑖 + 𝛽 + 1) Γ (𝑖 + 𝑘 + 𝛼 + 𝛽 + 1)
2𝑥
= (𝛼 + 𝛽 + 2𝑖 − 1) {(𝛼2 − 𝛽2 + ( − 1)) × (ℎ𝑗 Γ (𝑗 + 𝑘 + 𝛼 + 𝛽 + 1) Γ (𝑘 + 𝛽 + 1)
𝐿
−1
× Γ (𝑖 + 𝛼 + 𝛽 + 1) Γ (𝑘 − V + 1) (𝑖 − 𝑘)!)
× (𝛼 + 𝛽 + 2𝑖) (𝛼 + 𝛽 + 2𝑖 − 2) }
𝑗
× (2𝑖 (𝛼 + 𝛽 + 𝑖) (𝛼 + 𝛽 + 2𝑖 − 2))
−1 × ∑(−1)𝑗−𝑙 Γ (𝑗 + 𝑙 + 𝛼 + 𝛽 + 1) Γ (𝛼 + 1)
𝑙=0
(𝛼,𝛽)
× 𝑃𝐿,𝑖−1 (𝑥) × Γ (𝑙 + 𝑘 + 𝛽 − V + 1)
(𝛼 + 𝑖 − 1) (𝛽 + 𝑖 − 1) (𝛼 + 𝛽 + 2𝑖) −1
− × (Γ (𝑙 + 𝛽 + 1) Γ (𝑙 + 𝑘 + 𝛼 + 𝛽 − V + 2) (𝑗 − 𝑙)!𝑙!) .
𝑖 (𝛼 + 𝛽 + 𝑖) (𝛼 + 𝛽 + 2𝑖 − 2) (42)
(𝛼,𝛽)
× 𝑃𝐿,𝑖−2 (𝑥) 𝑖 = 2, 3, . . . ,
(37) 3. Pharmacokinetics Model Equation
(𝛼,𝛽) (𝛼,𝛽) 3.1. Drug Assimilation into the Blood. The drug dissolves in
where 𝑃𝐿,0 (𝑥) = 1 and 𝑃𝐿,1 (𝑥) = ((𝛼 + 𝛽 + 2)/2)(2𝑥/𝐿 −
1) + (𝛼 − 𝛽)/2. the gastrointestinal tract (GI), and each ingredient is diffused
We notice that a function 𝑢(𝑥), square integrable into the bloodstream. They are carried to the locations in
in (0, 𝐿), can be expanded in terms of shifted Jacobi which they act and are removed from the blood by the
polynomials as kidneys and liver. Generally, the problem of drug assimilation
into the body can be considered as a two-compartment
+∞ model: GI-tract and the bloodstream [16]. Different com-
(𝛼,𝛽)
𝑢 (𝑥) = ∑ 𝑎𝑗 𝑃𝐿,𝑗 (𝑥) , (38) partments and the input/output of the model are depicted in
𝑗=0 Figure 1.
Abstract and Applied Analysis 7
Drug intake Digestion Tissue Insert this formula for 𝑥(𝑡) into the second rate equation,
GI tract Blood
which becomes
𝑑𝑦 (𝑡)
Figure 1: Schematic of input/output compartment for drug assimi- + 𝑘2 𝑦 (𝑡) = 𝑘1 𝐴𝑒−𝑘1 𝑡 𝑦 (0) = 0. (47)
𝑑𝑡
lation.
In this paper, we are concerned with fractional time deriva-
tives regarding (17) and (18). Having defined 𝑐 𝐷V , we can
For each compartment by applying the balance law we can now turn to fractional differential equations and systems
obtain of fractional differential equations, which will be used to
specify PKPD models and will need to be solved over an
rate of change of interval [0, 𝑡], in accordance with appropriate initial condi-
{ } tions.
drug in GI tract
A typical feature of differential equations (both classical
rate of drug rate drug leaves and fractional) is the need to specify additional conditions
={ }−{ }
intake GI-tract in order to produce a unique solution. For the case of
(43) Caputo fractional differential equations, these additional
rate of change of conditions are just the initial conditions which are simi-
{ }
drug in blood larly required by classical ODEs. In contrast, for Riemann-
Liouville fractional differential equations, these additional
rate drug rate drug leaves
={ }−{ }. conditions constitute certain fractional derivatives (and/or
enters blood blood
integrals) of the unknown solution at the initial point 𝑡 =
0 [81], which are functions of 𝑡. These initial conditions are
In this study, we consider a case of a single cold pill. Also, there
not physical; furthermore, it is not clear how such quantities
is no ingestion of the drug except that which occurs initially.
are to be measured from experiment, say, so that they can
be appropriately assigned in an analysis [82]. If for no other
3.2. Case of a Single Cold Pill. Let us consider 𝑥(𝑡) to be reason, the need to solve fractional differential equations
the amount of drug in the GI-tract at time 𝑡 and 𝑦(𝑡), the is justification enough for choosing Caputo’s definition for
amount in the bloodstream at time 𝑡. In the GI-tract we fractional differentiation over the more commonly used (at
suppose that the pill is to be swallowed, and so after this event least in mathematical analysis) definition of Liouville and
(over subsequent time) we have nothing more entering the Riemann, and this is the operator that we choose to use in
GI-tract. The pill dissolves and diffuses into the bloodstream the following.
from the GI-tract. So, the GI-tract is only an output term. We start by representing drug concentration in the
Assuming that the output rate is proportional to the amount effect compartment by the (Caputo) fractional differential
of drug in the bloodstream, which is in fact proportional to equation:
the GI-tract drug concentration, then [16],
𝑐
𝐷V 𝑦 (𝑡) + 𝑘2 𝑦 (𝑡) = 𝑘1 𝐴𝑒−𝑘1 𝑡 𝑦 (0) = 0. (48)
𝑑𝑦
= −𝑘1 𝑥, 𝑥 (0) = 𝑥0 , (44) In the standard direct action model, the effect at
𝑑𝑡 time 𝑡, 𝑌(𝑡), is expressed by an arbitrary (memory-less) func-
where 𝑥0 is the amount of a drug in the pill and 𝑘1 is a tion of drug concentration in the effect site at time 𝑡, 𝐺(𝑦(𝑡));
positive coefficient. We suppose that 𝑦(0) = 0, which means however to generate a wider class of relationships, we assume
that the initial amount of the drug in the bloodstream is zero. that the effect at time 𝑡 ∈ [0, 1] is related to the fuzzy Caputo
As the drug diffuses from the GI-tract, the level increases, and fractional derivative of 𝑦(𝑡). So, we have
as the kidneys and liver remove it, the level of drug gradually 𝑐
𝐷V 𝑦 (𝑡) + 𝑘2 𝑦 (𝑡) = 𝑘1 𝐴𝑒−𝑘1 𝑡 𝑦 (0; 𝑟) = [𝑦𝑟 , 𝑦𝑟0 ] , (49)
decreases. Thus, 0
RF RF
𝑑𝑦 in which 𝑦(𝑥) : 𝐿 [0, 1] ∩ 𝐶 [0, 1] is a continuous fuzzy-
= 𝑘1 𝑥 − 𝑘2 𝑦, 𝑦 (0) = 0, (45) valued function and 𝑐 𝐷0V+ denotes the fuzzy Caputo frac-
𝑑𝑡
tional derivative of order V ∈ [0, 1].
with 𝑘2 another positive constant. Decongestant and an
antihistamine are the component of the cold pill, and the Remark 17. In this paper, the drug concentration is modeled
coefficient of proportionality (𝑘1 and 𝑘2 ) is different for the by an oscillation-relaxation fuzzy fractional differential equa-
different component drugs in the pill. tion. So the right hand side of (49) is determined based on the
In order to obtain the growth and decay of antihistamine model parameters. It should mention here that the proposed
levels in the GI tract and bloodstream, rearrange the first rate method can be extended easily for solving other types of
FFDEs with more complicated right hand side functions. For
equation, multiply by the integrating factor 𝑒𝑘1 𝑡 , integrate,
more details, one can refer to (33) in [65] and [57–59].
and then use the initial data to obtain
We have assumed that 𝑘1 ≠𝑘2 , an assumption that is
𝑋 = 𝐴𝑒(−𝑘1 𝑡) . (46) justified by the pharmaceutical data. For the “average” person,
8 Abstract and Applied Analysis
a pharmaceutical company estimates that the values of the Remark 18. In the remainder of paper, for more simplicity, we
rate constants for the antihistamine in the cold pills are 𝑘1 = (𝛼,𝛽) (𝛼,𝛽)
consider 𝑃𝑖 (𝑥) instead of 𝑃1,𝑖 (𝑥) and 𝑤(𝛼,𝛽) (𝑥) = (1 −
0.6931 (hour)−1 and 𝑘2 = 0.0231 (hour)−1 . It can be observed (𝛼,𝛽)
𝑥)𝛼 ⊙ 𝑥𝛽 instead of 𝑤1 (𝑥) = (1 − 𝑥)𝛼 ⊙ 𝑥𝛽 .
from (46) that level of antihistamine in the bloodstream
increases as the time increases and saturate in a maximum Remark 19. Practically, only the first (𝑁 + 1)-terms shifted
value of antihistamine; however, (47) can conclude that as Jacobi polynomials are taken into consideration. So, we have
the time increases the amount of antihistamine in GI-tract
declines and reaches a minimum value. We now consider 𝑁
(𝛼,𝛽)
two different cases of patients who are not average. We 𝑢 (𝑥) ≃ 𝑢𝑁 (𝑥) = ∑ ∗ 𝑎𝑖 ⊙ 𝑃𝑖 (𝑥) = a𝑇 ⊙ Φ𝑁 (𝑥) , (52)
often define 𝑘2 as clearance coefficient of medication from 𝑖=0
the blood. The value of 𝑘2 often varies from old and sick
patients than young and healthy cases. This means that the where the fuzzy shifted Jacobi coefficient vector 𝐴𝑇 and
level of medication in the blood may become and then shifted Jacobi polynomials vector Φ𝑁+1 (𝑥) are presented by
remain excessively high with a standard dosage for the cases
that 𝑘2 is much lower, which is normally observed in old and a𝑇 = [𝑎0 , 𝑎1 , . . . , 𝑎𝑁] ,
sick cases. (53)
We investigate sensitivity of the medication over a 24- (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) 𝑇
Φ𝑁 (𝑥) = [𝑃0 (𝑥) , 𝑃1 (𝑥) , . . . , 𝑃𝑁 (𝑥)] .
hour period by keeping the value of 𝑘1 fixed at some value
such as 1.386 but setting 𝑘2 = 0.01386, 0.06386, 0.1386,
0.6386, 1.386 (𝐴 = 1). Different values of 𝑘2 correspond to We can state the fuzzy approximate function (52) in the
people of different ages and states of health. For all cases parametric form as follows.
as it is expected that, for the first few hours, the amount
of antihistamine in the bloodstream increases and then Definition 20 (see [65]). Let 𝑢(𝑥) ∈ 𝐿R𝑝F [0, 1] ∩ 𝐶RF [0, 1];
declines gradually. However, for the cases with lower value of the approximation of fuzzy-valued function 𝑢(𝑥) in the para-
clearance coefficient (old and sick), the maximum value of the metric form is
medication in bloodstream was much higher than the cases 𝑁 𝑁
with high 𝑘2 , and it did not decline for remaining times. This 𝑢𝑟 (𝑥) ≃ 𝑢𝑁
𝑟 𝑟
(𝑥) = [∑𝑎𝑖,− 𝑃𝑖
(𝛼,𝛽) 𝑟
(𝑥) , ∑𝑎𝑖,+ 𝑃𝑖
(𝛼,𝛽)
(𝑥)] ,
means that the level of medication in the bloodstream stays at 𝑖=0 𝑖=0
high level for a long time, and it could not be absorbed from
the blood. The same trend is observed for the case that 𝑘2 is 0 ≤ 𝑟 ≤ 1.
constant (𝑘2 = 0.0231) and the value of 𝑘1 varies (𝑘1 = (54)
0.06931, 0.11, 0.3, 0.6931, 1.0, and 1.5). In this case, the 𝑘1 is
a constant for the GI tract which is analogous to the 𝑘2 for We call that Span{𝑃𝑛(𝛼,𝛽) (𝑥) : 𝑛 ≥ 0} forms a complete
the bloodstream. RF
orthogonal system in 𝐿 2,𝑤 (𝛼,𝛽) (0, 1). Hence, we set
𝑘V
Hence, the variational formulation of (56) according to Proof. Since 𝐷0+ 𝑓(𝑡) ∈ 𝐶(0, 𝑏], 𝑘 = 0, 1, . . . , 𝑁 + 1, 𝑓 can be
Relation (14) in [65], by means of a typical tau method like expanded to the fractional Taylor series
in the crisp context [29] and (52), is equivalent to
𝑁
𝑡𝑖V
𝑁 𝑓𝑇𝑁𝑖 (𝑡) = ∑ 𝑖V
[𝐷0+ 𝑓 (𝑡)]𝑡=0+ , (63)
∑𝑎𝑗 ⊙ [(𝐷(V) 𝑃𝑗
(𝛼,𝛽)
(𝑥) , 𝑃𝑘
(𝛼,𝛽)
(𝑥)) 𝑖=0 Γ (𝑖V + 1)
𝑤(𝛼,𝛽)
𝑗=0
and its reminder term is
(𝛼,𝛽) (𝛼,𝛽)
+ 𝑘2 (𝑃𝑗 (𝑥) , 𝑃𝑘 (𝑥)) ]
𝑤(𝛼,𝛽)
(58) V 𝑡(𝑁+1)V
−𝑘1 𝑡 (𝛼,𝛽) 𝑅𝑁 (𝑡, 0) = [𝐷(𝑁+1)V 𝑓 (𝑡)]𝑡=𝜉 ,
= (𝑘1 𝐴𝑒 , 𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) , 𝑘 = 0, 1, . . . , 𝑁 − 1, Γ ((𝑁 + 1) V + 1) 0+ (64)
𝑁 0 ⩽ 𝜉 ⩽ 𝑡, ∀𝑡 ∈ (0, 𝑏] .
(𝛼,𝛽)
∑𝑎𝑗 ⊙ 𝑃𝑗 (0) = 𝑑0 .
𝑗=0 Therefore,
Denoting
𝑓 (𝑡) − 𝑓𝑇𝑁𝑖 (𝑡) = 𝑅𝑁
V
(𝑡, 0) . (65)
𝐴 = (𝑎𝑘𝑗 )0<𝑘,𝑗<𝑁, 𝐶 = (𝑐𝑘𝑗 )0<𝑘,𝑗<𝑁, (59) From (65) and triangle inequality, we obtain
we investigate that (58) is equivalent to the matrix system 𝑓 (𝑡) − 𝑦𝑁,V (𝑡) = 𝑓 (𝑡) − 𝑦𝑁,V (𝑡) + 𝑓𝑇 (𝑡) − 𝑓𝑇 (𝑡)
𝑁𝑖 𝑁𝑖
(𝐴 + 𝑘2 𝐶) ⊙ a = f.
(60) ⩽ 𝑓 (𝑡) − 𝑓𝑇𝑁𝑖 (𝑡) + 𝑓𝑇𝑁𝑖 (𝑡) − 𝑦𝑁,V (𝑡)
V
The elements of the matrices mentioned above are deter- = 𝑅𝑁 (𝑡, 0) + 𝑓𝑇𝑁𝑖 (𝑡) − 𝑦𝑁,V (𝑡) .
mined in the next theorem.
(66)
(𝛼,𝛽)
Theorem 21 (see [59]). Let us denote 𝑎𝑘𝑗 = (𝐷(V) 𝑃𝑗 (𝑥),
(𝛼,𝛽) Therefore, an upper bound of the absolute errors is obtained
𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) (0 ⩽ 𝑘 ⩽ 𝑁 − 1, 0 ⩽ 𝑗 ⩽ 𝑁), 𝑎𝑘𝑗 = for the method in the crisp cases.
(𝛼,𝛽) (𝛼,𝛽)
𝐷𝑘−𝑁𝑃𝑗 (0) (𝑘 = 𝑁, 0 ⩽ 𝑗 ⩽ 𝑁) and 𝑐𝑘𝑗 = (𝑃𝑗 (𝑥),
(𝛼,𝛽)
𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) (0 ⩽ 𝑘 ⩽ 𝑁 − 1, 0 ⩽ 𝑗 ⩽ 𝑁); then the nonzero Now, we provide an upper bound for the absolute errors
elements of 𝑎𝑘𝑗 and 𝑐𝑘𝑗 are given as of the fuzzy approximate solution by using the proposed
method.
𝑎𝑘𝑗 = ℎ𝑘 𝑆V (𝑗, 𝑘, 𝛼, 𝛽) , 0 ⩽ 𝑘 ⩽ 𝑁 − 1, 1 ⩽ 𝑗 ⩽ 𝑁, 𝑎𝑘𝑗
Theorem 23. Let 𝑢(𝑡) ∈ 𝐿R𝑝F [0, 1] ∩ 𝐶RF [0, 1] be the fuzzy
𝑗−𝑘+𝑁
(−1) Γ (𝑗 + 𝛽 + 1) (𝑗 + 𝛼 + 𝛽 + 1)𝑘−𝑁 exact solution of (49), and 𝑢𝑁(𝑡) is the best fuzzy Jacobi
= ,
Γ (𝑗 − 𝑘 + 𝑁 + 1) Γ (𝑘 − 𝑁 + 1 + 𝛽) approximate function (52), and suppose that 𝑐 𝐷𝑘V 𝑢(𝑡) ∈
𝐶RF [0, 𝑏] for 𝑘 = 0, 1, . . . , 𝑁 + 1 where 0 < V < 1 and 𝑡 ∈
𝑘 = 𝑁, 0 ⩽ 𝑗 ⩽ 𝑁, 𝑐𝑘𝑗 [0, 𝑏]. Then, one has
= ℎ𝑘 , 0 ⩽ 𝑘 = 𝑗 ⩽ 𝑁 − 1. 𝑟
[𝑢 (𝑡)]𝑟 = [𝑢𝑟 (𝑡) , 𝑢𝑟 (𝑡)] , [𝑢𝑁 (𝑡)] = [𝑢𝑟𝑁 (𝑡) , 𝑢𝑟𝑁 (𝑡)] ,
(61)
V 𝑁𝑖
𝑢 (𝑡) − 𝑢𝑁 (𝑡) ⩽ 𝑅𝑁 (𝑡, 0) + 𝑢𝑇 (𝑡) − 𝑢𝑁 (𝑡) ,
4.1. Error Analysis. In this section, error analysis of the
method will be presented for the FFDEs. Firstly, an upper 𝑢 (𝑡) − 𝑢𝑁 (𝑡) ⩽ 𝑅V (𝑡, 0) + 𝑢𝑁𝑖 (𝑡) − 𝑢𝑁 (𝑡) ,
𝑁 𝑇
bound of the absolute errors will be given for the technique (67)
by using generalized Taylor formula. Secondly, an error
bound will be introduced for the approximation of the fuzzy where regarding Theorem 15, one has
fractional Caputo’s derivative using Jacobi polynomials.
𝑟 𝑁𝑖
Lemma 22. Let 𝑦𝑁,V (𝑡) and 𝑓(𝑡) be the Jacobi approximate [𝑓𝑇𝑁𝑖 (𝑡)] = [𝑓𝑁𝑖 (𝑡; 𝑟) , 𝑓𝑇 (𝑡; 𝑟)]
𝑇
solution (40) and the exact solution of (48), respectively.
𝑁
𝑘V
If 𝐷0+ 𝑓(𝑡) ∈ 𝐶(0, 𝑏], 𝑘 = 0, 1, . . . , 𝑁 + 1, then 𝑡𝑖V 𝑖V
= [∑ [𝐷0+ 𝑓 (𝑡)] ,
Γ (𝑖V + 1) 𝑡=0+
V 𝑁𝑖 𝑖=0
𝑓 (𝑡) − 𝑦𝑁,V (𝑡) ⩽ 𝑅𝑁 (𝑡, 0) + 𝑓𝑇 (𝑡) − 𝑦𝑁,V (𝑡) , (62)
𝑁
𝑡𝑖V 𝑖V
∑ [𝐷0+ 𝑓 (𝑡)]𝑡=0+ ] ,
where 𝑓𝑇𝑁𝑖 and 𝑅𝑁
V
are defined according to Theorem 10. 𝑖=0 Γ (𝑖V + 1)
10 Abstract and Applied Analysis
V 𝑟 V,𝑟
[𝑅𝑁 (𝑡, 0)] = [𝑅V,𝑟
𝑁 (𝑡, 0) , 𝑅𝑁 (𝑡, 0)] assumption, namely, 𝑘2 = 0.0231 and Dose 𝐴 = 1, is as
follows:
𝑐 (𝑁+1)V 𝑟
𝐷 𝑓 (0+) (𝑁+1)V 𝑐
=[ 𝑡 , 𝐷V 𝑦 (𝑡) + 0.0231𝑦 (𝑡) = 0.6931𝑒−0.6931𝑡 ,
Γ (𝑁V + V + 1) (71)
[ 𝑦 (0; 𝑟) = [−1 + 𝑟, 1 − 𝑟] .
𝑐 𝑟
𝐷(𝑁+1)V 𝑓 (0+) (𝑁+1)V ] By using Theorem 7 for the above equation under 𝑐 [(1)−V]
𝑡 .
Γ (𝑁V + V + 1) differentiability, we have the following systems:
]
𝑐 V
(68) 𝐷 𝑦 (𝑡; 𝑟) + 0.0231𝑦 (𝑡; 𝑟)
In this section, the fuzzy fractional PKPD model (49) is × (0.6931𝑒−0.6931𝑥 ) 𝑑𝑥, 0 < 𝑟 ≤ 1.
solved for different values of 𝑘1 and 𝑘2 by using the JT
method presented in Section 4. We have performed all We seek the fuzzy approximate solutions by applying the
numerical computations with a computer program written in technique described in Section 4 with 𝑁 = 2 and as
MATLAB. Also, absolute errors between fuzzy approximate
solution [𝑦𝑁]𝑟 = [𝑦𝑟 , 𝑦𝑟𝑁] and the corresponding exact solu- 𝑦2 (𝑡) = 𝑎0 ⊙ 𝑃0
(𝛼,𝛽) (𝛼,𝛽)
(𝑡) + 𝑎1 ⊙ 𝑃1 (𝑡) + 𝑎2 ⊙ 𝑃0
(𝛼,𝛽)
(𝑡) .
𝑁
tions 𝑦(𝑡; 𝑟) = [𝑦(𝑡; 𝑟), 𝑦(𝑡; 𝑟)], that is, [𝑁𝑒 ]𝑟 = [|𝑦𝑟 − 𝑦𝑟 |, (74)
𝑁
|𝑦𝑟𝑁 − 𝑦𝑟 |], are considered.
Now, we recall the FFDE (49) as follows: Here, we have
𝑐 0 0 0
𝐷V 𝑦 (𝑡) + 𝑘2 𝑦 (𝑡) = 𝑘1 𝐴𝑒−𝑘1 𝑡
(70) 𝐷0.85 = ( 1.8639 0.3901 −0.1755) , (75)
𝑦 (0; 𝑟) = [−1 + 𝑟, 1 − 𝑟] , −0.3901 4.5267 0.8696
in which 𝑦(𝑥) : 𝐿RF [0, 1] ∩ 𝐶RF [0, 1] is a continuous fuzzy- where 𝛼 = 𝛽 = 0. Regarding (54), we have
valued function and 𝑐 𝐷0V+ denotes the fuzzy Caputo frac-
𝑦𝑟 (𝑡) = 𝑎𝑟0 + 𝑎𝑟1 (2𝑥 − 1) + 𝑎𝑟2 (6𝑥2 − 6𝑥 + 1)
tional derivative of order V ∈ [0, 1]. 2
Two cases are considered for solving by the proposed (76)
technique. In the first case, we assume that 𝑘2 is unchange- 𝑦𝑟2 (𝑡) = 𝑎𝑟0 + 𝑎𝑟1 (2𝑥 − 1) + 𝑎𝑟2 (6𝑥2 − 6𝑥 + 1) .
able and different values of 𝑘1 are substituted in (49) to get
By solving the fuzzy linear system (60) in the parametric
the fuzzy approximate solution. Conversely, in the second
form, one can get fuzzy unknown coefficients in (76) as
case we try to solve the problem such that 𝑘1 has an invari-
able value and 𝑘2 is varied. The problem is analyzed for both 𝑎0 = −0.5757, 𝑎1 = 0.2624, 𝑎2 = −0.0619,
cases in details. (77)
𝑎0 = 1.2008, 𝑎1 = 0.2408, 𝑎2 = −0.0600,
Case I. Let us consider from Section 3 that 𝑘2 = 0.0231,
but let 𝑘1 vary (e.g., 0.6931, 0.11, and 0.3). So (49), with the where we assume that 𝑟-cut = 0.1 in (77).
Abstract and Applied Analysis 11
Table 1: The results of the proposed method for Case I with V = 0.85, 𝛼 = 𝛽 = 0, and 𝑁 = 8.
10−3.4
10−3.5
Absolute error (Ne )
10−3.6
10−3.7
10−3.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts
𝛼=𝛽=0 𝛼 = 0, 𝛽 = 0.5
𝛼 = 𝛽 = 0.5 𝛼 = 0.5, 𝛽 = 0
Figure 2: The absolute errors for different 𝛼 and 𝛽 with 𝑁 = 9, 𝑘1 = 0.3 and V = 0.75, Case I.
10−5
Absolute error (Ne )
10−6
10−7
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts
N=4 N=9
N=7 N = 11
Figure 3: The absolute errors for different values 𝑁 with V = 0.95, 𝑘1 = 0.11 and 𝛼 = 𝛽 = 0, Case I.
12 Abstract and Applied Analysis
Table 2: The fuzzy coefficients for Case II with V = 0.75, 𝛼 = 𝛽 = 0.5, 𝑁 = 2, and 𝑘2 = 0.01386.
𝑟 𝑎0 𝑎1 𝑎2 𝑎0 𝑎1 𝑎2
0 −0.42793 0.24218 −0.11135 1.55482 0.23234 −0.11003
0.1 −0.32879 0.24169 −0.11128 1.45568 0.23283 −0.11009
0.2 −0.22965 0.24120 −0.11121 1.35654 0.23332 −0.11016
0.3 −0.13052 0.24071 −0.11115 1.25741 0.23382 −0.11022
0.4 −0.03138 0.24021 −0.11108 1.15827 0.23431 −0.11029
0.5 0.06775 0.23972 −0.11102 1.05913 0.23480 −0.11036
0.6 0.16689 0.23923 −0.11095 0.95999 0.23529 −0.11042
0.7 0.26603 0.23874 −0.11088 0.86085 0.23578 −0.11049
0.8 0.36516 0.23824 −0.11082 0.76172 0.23628 −0.11055
0.9 0.46430 0.23775 −0.11075 0.66258 0.23677 −0.11062
1 0.56344 0.23726 −0.11069 0.56344 0.23726 −0.11069
Table 3: The results of the proposed method for Case II with V = 0.95, 𝛼 = 𝛽 = 0.5, and 𝑁 = 7.
k1 = 0.6931
Approximate solution
−2
1
0.8 1
0.5 0.6
0.2 0.4
x 0 0
r-cuts
k1 = 0.11 k1 = 0.3
Approximate solution
Approximate solution
2 2
0 0
−2 −2
1 1
1 1
0.5 0.5 0.5 0.5
x 0 0 r-cuts x 0 0 r-cuts
Figure 4: The fuzzy approximate solution of Case I, for different value of 𝑘1 , 𝛼 = 𝛽 = 0.5, 𝑁 = 8.
Abstract and Applied Analysis 13
10−4
(79)
Absolute error (Ne )
10−3
By applying the JT method in Section 4, we can get the
fuzzy unknown coefficients {𝑎𝑗 }𝑁
𝑗=0 by using (58) which is in
the parametric form as follows:
10−4 𝑁
(𝛼,𝛽) (𝛼,𝛽)
∑𝑎𝑗 [(𝐷(V) 𝑃𝑗 (𝑥) , 𝑃𝑘 (𝑥))
𝑤(𝛼,𝛽)
𝑗=0
(𝛼,𝛽) (𝛼,𝛽)
10−5 + 0.01386(𝑃𝑗 (𝑥) , 𝑃𝑘 (𝑥)) ]
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 𝑤(𝛼,𝛽)
r-cuts
(𝛼,𝛽)
= (1.386𝐴𝑒−1.386𝑡 , 𝑃𝑘 (𝑥))𝑤(𝛼,𝛽) ,
m=4 m=9
m=7 m = 11
𝑘 = 0, 1, . . . , 𝑁 − 1,
Figure 6: The absolute errors for different values 𝑁 with V =
0.85, 𝑘2 = 0.1386, and 𝛼 = 𝛽 = 0, Case II. 𝑁
(𝛼,𝛽)
∑ 𝑎𝑗 𝑃𝑗 (0) = −1 + 𝑟,
𝑗=0
k2 = 0.01386
Approximate solution
2
−2
1
0.8 1
0.5 0.6
0.2 0.4
x 0 0
r-cuts
k2 = 0.06386 k2 = 0.1386
Approximate solution
Approximate solution
2 2
0 0
−2 −2
1 1
1 1
0.5 0.5 0.5 0.5
x 0 0 r-cuts x 0 0 r-cuts
Figure 7: The fuzzy approximate solution of Case II, for different value of 𝑘2 , 𝛼 = 0, 𝛽 = 0.5, and 𝑁 = 9.
where 𝐷(V) is given as drug stays in the bloodstream for longer time. Since the
constant value of 𝑘2 appears with negative sign in the main
0 0 0 differential equation as we expected, the value of 𝑁𝑒𝑟 which is
𝐷0.75 = ( 2.6929 0.5524 −0.1755) , (81) directly proportional to the exact solutions is decreased with
−1.2429 4.2241 1.1048 increasing the value of 𝑘2 .
with 𝛼 = 𝛽 = 0.5 and 𝑁 = 2. By substituting the above
matric in (80) we can reach the fuzzy linear algebraic system
(60) that can solve easily to determine the fuzzy coefficients
as shown in Table 2. 6. Conclusion
We compared JT method results for different values Fuzzy theory provides a suitable way to objectively account
of 𝑘2 and the outcomes are tabulated in Table 3. The absolute for parameter uncertainty in models. Fuzzy logic approaches
errors of the proposed method for this case are exhibited appear promising in preclinical applications and might be
in Figure 5 with four choices of 𝛼 and 𝛽. Clearly, the best useful in drug discovery and design. Considerable progress
approximation is achieved when the values of 𝛼 = 0 and 𝛽 = has been made in the last few years in the development
0. Additionally, we compared the approximate solutions of computational approaches for prediction of drug absorp-
obtained by the present method at 𝑁 = 4, 7, 9, and 11 which tion, distribution, metabolism, and excretion. Whilst several
is shown in Figure 6. From Figure 6, one can conclude approaches have been developed in pharmacokinetics, most
that with increasing the number of Jacobi polynomials, the of these approaches have not yet been adequately used in the
absolute errors are decreasing dramatically. Finally, Figure 7 complex process such as prediction of metabolism, and they
shows the numerical results for different values of 𝑘2 at V = require further improvement.
0.75. It can be seen that for all the values of 𝑘2 the approxi- In summary in this research, a tau method based on
mate solution is the fuzzy number. the Jacobi operational matrix was utilized to numerically
In Table 3, the value of absolute error for 𝑟-cut varied solve the PKPD equation, arising from drug assimilation
from 0 to 1 for different value of 𝑘2 is calculated. At a into the bloodstream. The comparison of the results shows
constant 𝑟-cut value, a descending trend is observed with that the present method is a powerful mathematical tool for
increasing the value of 𝑘2 . This trend is opposed to the trend finding the numerical solutions of a generalized linear fuzzy
observed in Table 1 for different value of 𝑘1 . 𝑘2 is clearance fractional PKPD equation.
constant and it is a coefficient for 𝑦(𝑡) which is the amount Although we concentrated on applying our algorithm to
of drug in the bloodstream in (44). We already explained solve fuzzy fractional PKPD equation, we show that such
that by decreasing the value of 𝑘2 ,which happens in the algorithm can be applied to solve other types of fractional
case of old and sick patient, the drug absorption from the equations models in science and engineering fields. Our
bloodstream by kidney and liver also decreases and the algorithm for the fuzzy fractional PKPD equations is efficient
Abstract and Applied Analysis 15
and numerically stable. Numerical results are presented [14] P. Veng-Pedersen, “Noncompartmentally-based pharmacoki-
which exhibit the high accuracy of the proposed algorithm. netic modeling,” Advanced Drug Delivery Reviews, vol. 48, no.
By using the obtained operational matrix, the following 2-3, pp. 265–300, 2001.
objectives were achieved. [15] D. E. Mager, “Quantitative structure-pharmacokinetic/pharma-
codynamic relationships,” Advanced Drug Delivery Reviews, vol.
(i) The given fuzzy fractional differential equation was 58, no. 12-13, pp. 1326–1356, 2006.
converted into a fuzzy algebraic system of equations, [16] B. Barnes and G. R. Fulford, Mathematical Modelling with Case
simplifying the solution procedure. Studies: A Differential Equation Approach Using Maple and
(ii) The method is computer oriented; thus, solving fuzzy MATLAB, CRC Press, 2011.
fractional differential equations of different fractional [17] S. Dhillon and K. Gill, “Basic pharmacokinetics,” in Clinical
orders V becomes a matter of changing V only. Pharmacokinetics, pp. 1–44, Pharmaceutical Press, 2006.
[18] L. Gaul, P. Klein, and S. Kemple, “Damping description
involving fractional operators,” Mechanical Systems and Signal
References Processing, vol. 5, no. 2, pp. 81–88, 1991.
[1] B. G. Katzung, S. B. Masters, and A. J. Trevor, Basic & Clinical [19] W. G. Glockle and T. F. Nonnenmacher, “A fractional calculus
Pharmacology, McGraw-Hill, 2004. approach to self-similar protein dynamics,” Biophysical Journal,
vol. 68, no. 1, pp. 46–53, 1995.
[2] H. Peng and B. Cheung, “A review on pharmacokinetic model-
ing and the effects of environmental stressors on pharmacoki- [20] R. J. Marks II and M. W. Hall, “Differintegral interpolation from
netics for operational medicine: operational pharmacokinetics,” a bandlimited signal’s samples,” IEEE Transactions on Acoustics,
Tech. Rep. ADA509469, DTIC, 2009. Speech, and Signal Processing, vol. 29, no. 4, pp. 872–877, 1981.
[3] A. Suzuki, W. I. Higuchi, and N. F. Ho, “Theoretical model [21] A. Atangana and D. Baleanu, “Nonlinear fractional Jaulent-
studies of drug absorption and transport in the gastrointestinal Miodek and Whitham-Broer-Kaup equations within Sumudu
tract. I,” Journal of Pharmaceutical Sciences, vol. 59, no. 5, pp. transform,” Abstract and Applied Analysis, vol. 2013, Article ID
644–651, 1970. 160681, 8 pages, 2013.
[4] H. Boxenbaum, “Absorption potential and its variants,” Phar- [22] B. Mehdinejadiani, A. A. Naseri, H. Jafari, A. Ghanbarzadeh,
maceutical Research, vol. 16, no. 12, p. 1893, 1999. and D. Baleanu, “A mathematical model for simulation of
a water table profile between two parallel subsurface drains
[5] A. Boobis, U. Gundert-Remy, P. Kremers, P. Macheras, and
using fractional derivatives,” Computers & Mathematics with
O. Pelkonen, “In silico prediction of ADME and pharmacoki-
Applications, vol. 66, no. 5, pp. 785–794, 2013.
netics: report of an expert meeting organised by COST B15,”
European Journal of Pharmaceutical Sciences, vol. 17, no. 4-5, pp. [23] C. Csajka and D. Verotta, “Pharmacokinetic-pharmacodynamic
183–193, 2002. modelling: history and perspectives,” Journal of Pharmacokinet-
[6] P. Poulin, K. Schoenlein, and F. P. Theil, “Prediction of adipose ics and Pharmacodynamics, vol. 33, no. 3, pp. 227–279, 2006.
tissue: plasma partition coefcients for structurally unrelated [24] D. Verotta, “Fractional dynamics pharmacokinetics-pharmaco-
drugs,” Journal of Pharmaceutical Sciences, vol. 90, pp. 436–447, dynamic models,” Journal of Pharmacokinetics and Pharmaco-
2001. dynamics, vol. 37, no. 3, pp. 257–276, 2010.
[7] F. Lombardo, R. S. Obach, M. Y. Shalaeva, and F. Gao, “Predic- [25] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
tion of volume of distribution values in humans for neutral and Calculus: Models and Numerical Methods, vol. 3 of Series on
basic drugs using physicochemical measurements and plasma Complexity, Nonlinearity and Chaos, World Scientific Publish-
protein binding data,” Journal of Medicinal Chemistry, vol. 45, ing, Hackensack, NJ, USA, 2012.
no. 13, pp. 2867–2876, 2002. [26] D. Baleanu, Z. B. G. Güvenç, and J. A. Tenreiro Machado, New
[8] P. Poulin and F. P. Theil, “A priori prediction of tissue: Trends in Nanotechnology and Fractional Calculus Applications,
plasma partition coefcients of drugs to facilitate the use of Springer, New York, NY, USA, 2010.
physiologically-based pharmacokinetic models in drug discov- [27] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
ery,” Journal of Pharmaceutical Sciences, vol. 89, pp. 16–35, 1999. ematics in Science and Engineering, Academic Press, San Diego,
[9] P. Poulin and F.-P. Theil, “Prediction of pharmacokinetics prior Calif, USA, 1999.
to in vivo studies. 1. Mechanism-based prediction of volume of [28] M. Gülsu, Y. Öztürk, and A. Anapalı, “Numerical approach
distribution,” Journal of Pharmaceutical Sciences, vol. 91, no. 1, for solving fractional relaxation-oscillation equation,” Applied
pp. 129–156, 2002. Mathematical Modelling, vol. 37, no. 8, pp. 5927–5937, 2013.
[10] P. Poulin and F.-P. Theil, “Prediction of pharmacokinetics prior [29] M. Dehghan and A. Saadatmandi, “A tau method for the one-
to in vivo studies. II. Generic physiologically based pharma- dimensional parabolic inverse problem subject to temperature
cokinetic models of drug disposition,” Journal of Pharmaceu- overspecification,” Computers and Mathematics with Applica-
tical Sciences, vol. 91, no. 5, pp. 1358–1370, 2002. tions, vol. 52, no. 6-7, pp. 933–940, 2006.
[11] L. Möhler, D. Flockerzi, H. Sann, and U. Reichl, “Mathematical [30] H. Jafari, H. Tajadodi, and D. Baleanu, “A modified variational
model of influenza a virus production in large-scale microcar- iteration method for solving fractional Riccati differential equa-
rier culture,” Biotechnology and Bioengineering, vol. 90, no. 1, pp. tion by Adomian polynomials,” Fractional Calculus and Applied
46–58, 2005. Analysis, vol. 16, no. 1, pp. 109–122, 2013.
[12] P. L. Bonate, Pharmacokinetic-Pharmacodynamic Modeling and [31] G.-C. Wu and D. Baleanu, “Variational iteration method for
Simulation, Springer, 2011. fractional calculus—a universal approach by Laplace trans-
[13] M. Holz and A. Fahr, “Compartment modeling,” Advanced Drug form,” Advances in Difference Equations, vol. 2013, article 18,
Delivery Reviews, vol. 48, no. 2-3, pp. 249–264, 2001. 2013.
16 Abstract and Applied Analysis
[32] Y. Hu, Y. Luo, and Z. Lu, “Analytical solution of the linear [49] A. H. Bhrawy and A. S. Alofi, “The operational matrix of frac-
fractional differential equation by Adomian decomposition tional integration for shifted Chebyshev polynomials,” Applied
method,” Journal of Computational and Applied Mathematics, Mathematics Letters, vol. 26, no. 1, pp. 25–31, 2013.
vol. 215, no. 1, pp. 220–229, 2008. [50] D. Baleanu, A. H. Bhrawy, and T. M. Taha, “Two efficient
[33] S. Saha Ray and R. K. Bera, “Analytical solution of the Bagley generalized Laguerre spectral algorithms for fractional initial
Torvik equation by Adomian decomposition method,” Applied value problems,” Abstract and Applied Analysis, vol. 2013, Article
Mathematics and Computation, vol. 168, no. 1, pp. 398–410, 2005. ID 546502, 10 pages, 2013.
[34] A. Arikoglu and I. Ozkol, “Solution of fractional differential
[51] D. Baleanu, A. H. Bhrawy, and T. M. Taha, “A modified
equations by using differential transform method,” Chaos,
generalized Laguerre spectral method for fractional differential
Solitons & Fractals, vol. 34, no. 5, pp. 1473–1481, 2007.
equations on the half line,” Abstract and Applied Analysis, vol.
[35] P. Darania and A. Ebadian, “A method for the numerical solu- 2013, Article ID 413529, 12 pages, 2013.
tion of the integro-differential equations,” Applied Mathematics
and Computation, vol. 188, no. 1, pp. 657–668, 2007. [52] A. H. Bhrawy and T. M. Taha, “An operational matrix of
fractional integration of the Laguerre polynomials and its
[36] I. Hashim, O. Abdulaziz, and S. Momani, “Homotopy analysis
application on a semi-infinite interval,” Mathematical Sciences,
method for fractional IVPs,” Communications in Nonlinear
vol. 6, article 41, 2012.
Science and Numerical Simulation, vol. 14, no. 3, pp. 674–684,
2009. [53] A. H. Bhrawy, M. M. Alghamdi, and T. M. Taha, “A new
[37] H. M. Jaradat, F. Awawdeh, and E. A. Rawashdeh, “An analytical modified generalized Laguerre operational matrix of fractional
scheme for multi-order fractional differential equations,” Tam- integration for solving fractional differential equations on the
sui Oxford Journal of Mathematical Sciences, vol. 26, no. 3, pp. half line,” Advances in Difference Equations, vol. 2012, article 179,
305–320, 2010. 2012.
[38] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, [54] D. Rostamy, M. Alipour, H. Jafari, and D. Baleanu, “Solving
Spectral Methods in Fluid Dynamics, Springer Series in Com- multi-term orders fractional differential equations by opera-
putational Physics, Springer, New York, NY, USA, 1988. tional matrices of BPs with convergence analysis,” Romanian
[39] B. Fornberg, A Practical Guide to Pseudospectral Methods, Reports in Physics, vol. 65, pp. 334–349, 2013.
Cambridge University Press, Cambridge, Mass, USA, 1998. [55] S. Yüzbaşı, “Numerical solutions of fractional Riccati type
[40] E. L. Ortiz and H. Samara, “Numerical solution of differential differential equations by means of the Bernstein polynomials,”
eigenvalue problems with an operational approach to the Tau Applied Mathematics and Computation, vol. 219, no. 11, pp.
method,” Computing, vol. 31, no. 2, pp. 95–103, 1983. 6328–6343, 2013.
[41] Y. Li and N. Sun, “Numerical solution of fractional differential [56] D. Baleanu, M. Alipour, and H. Jafari, “The Bernstein oper-
equations using the generalized block pulse operational matrix,” ational matrices for solving the fractional quadratic Riccati
Computers and Mathematics with Applications, vol. 62, no. 3, pp. differential equations with the Riemann-Liouville derivative,”
1046–1054, 2011. Abstract and Applied Analysis, vol. 2013, Article ID 461970, 7
[42] M. Yi, J. Huang, and J. Wei, “Block pulse operational matrix pages, 2013.
method for solving fractional partial differential equation,”
Applied Mathematics and Computation, vol. 221, pp. 121–131, [57] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new
2013. Jacobi operational matrix: an application for solving fractional
differential equations,” Applied Mathematical Modelling, vol. 36,
[43] A. Saadatmandi and M. Dehghan, “A new operational matrix
no. 10, pp. 4931–4943, 2012.
for solving fractional-order differential equations,” Computers
and Mathematics with Applications, vol. 59, no. 3, pp. 1326–1336, [58] S. Kazem, “An integral operational matrix based on Jacobi
2010. polynomials for solving fractional-order differential equations,”
[44] S. Kazem, S. Abbasbandy, and S. Kumar, “Fractional-order Applied Mathematical Modelling, vol. 37, no. 3, pp. 1126–1136,
Legendre functions for solving fractional-order differential 2013.
equations,” Applied Mathematical Modelling, vol. 37, no. 7, pp. [59] E. H. Doha, A. H. Bhrawy, D. Baleanu, and S. S. Ezz-Eldien, “On
5498–5510, 2013. shifted Jacobi spectral approximations for solving fractional
[45] A. H. Bhrawy and M. M. Al-Shomrani, “A shifted Legen- differential equations,” Applied Mathematics and Computation,
dre spectral method for fractional-order multipoint boundary vol. 219, no. 15, pp. 8042–8056, 2013.
value problems,” Advances in Difference Equations, vol. 2012, [60] R. P. Agarwal, V. Lakshmikantham, and J. J. Nieto, “On the
article 8, 2012. concept of solution for fractional differential equations with
[46] A. H. Bhrawy, A. S. Alofi, and S. S. Ezz-Eldien, “A quadrature uncertainty,” Nonlinear Analysis: Theory, Methods & Applica-
tau method for fractional differential equations with variable tions, vol. 72, no. 6, pp. 2859–2862, 2010.
coefficients,” Applied Mathematics Letters, vol. 24, no. 12, pp.
[61] T. Allahviranloo, S. Salahshour, and S. Abbasbandy, “Explicit
2146–2152, 2011.
solutions of fractional differential equations with uncertainty,”
[47] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “Efficient Soft Computing, vol. 16, no. 2, pp. 297–302, 2012.
Chebyshev spectral methods for solving multi-term fractional
orders differential equations,” Applied Mathematical Modelling, [62] T. Allahviranloo, Z. Gouyandeh, and A. Armand, “Fuzzy
vol. 35, no. 12, pp. 5662–5672, 2011. fractionaldifferential equations under generalized fuzzy Caputo
[48] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A Chebyshev derivative,” Journal of Intelligent and Fuzzy Systems. In press.
spectral method based on operational matrix for initial and [63] S. Salahshour, T. Allahviranloo, S. Abbasbandy, and D. Baleanu,
boundary value problems of fractional order,” Computers and “Existence and uniqueness results for fractional differential
Mathematics with Applications, vol. 62, no. 5, pp. 2364–2373, equations with uncertainty,” Advances in Difference Equations,
2011. vol. 2012, article 112, 2012.
Abstract and Applied Analysis 17
Research Article
Numerical Solutions of Fractional Fokker-Planck Equations
Using Iterative Laplace Transform Method
Limei Yan
School of Mathematical Sciences, Dezhou University, Dezhou 253023, China
Copyright © 2013 Limei Yan. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
A relatively new iterative Laplace transform method, which combines two methods; the iterative method and the Laplace transform
method, is applied to obtain the numerical solutions of fractional Fokker-Planck equations. The method gives numerical solutions in
the form of convergent series with easily computable components, requiring no linearization or small perturbation. The numerical
results show that the approach is easy to implement and straightforward when applied to space-time fractional Fokker-Planck
equations. The method provides a promising tool for solving space-time fractional partial differential equations.
where 𝐴(𝑥), 𝐵(𝑥) > 0 are called diffusion coefficient and drift partial differential equation with initial conditions of the
coefficient, respectively. If 𝐴(𝑥), 𝐵(𝑥) > 0 depend on 𝑢(𝑥, 𝑡) form
and the time 𝑡, then (1) becomes the following generalized
nonlinear form [25]: 𝐷𝑡𝛼 𝑢 = A (𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .) , 𝑚 − 1 < 𝛼 ≤ 𝑚,
2 (8)
𝜕𝑢 𝜕 𝜕 𝑛 − 1 < 𝛽 ≤ 𝑛, 𝑚, 𝑛 ∈ 𝑁,
= [− 𝐴 (𝑥, 𝑡, 𝑢) + 2 𝐵 (𝑥, 𝑡, 𝑢)] 𝑢 (𝑥, 𝑡) . (3)
𝜕𝑡 𝜕𝑥 𝜕𝑥
which is the generalized fractional form of (3). Here where A(𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .) is a linear or nonlinear operator
𝐷𝑡𝛼 (⋅), 𝐷𝑥𝛽 (⋅), 𝐷𝑥2𝛽 (⋅) are the Caputo fractional derivative with of 𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . ., and 𝑢 = 𝑢(𝑥, 𝑡) is the unknown function
respect to 𝑡 and 𝑥 defined in Section 2. When 𝛼 = 𝛽 = 1, (4) that will be determined later.
reduces to (3). Taking Laplace transfer of both sides of (8) results in
The rest of this paper is organized as follows. In Section 2,
we review some basic definitions of Caputo fractional deriva-
𝑚−1
tive and Laplace transform. In Section 3, we describe the
𝑠𝛼 L [𝑢 (𝑥, 𝑡)] − ∑ 𝑠𝛼−1−𝑘 𝑢(𝑘) (𝑥, 0)
iterative Laplace transform method for solving fractional
𝑘=0 (10)
partial differential equations. In Section 4, we give three
applications of the method to Fokker-Planck equations. In = L [A (𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .)] .
Section 5, some conclusions and discussions are given.
Equivalently,
2. Preliminaries
Definition 1. The Caputo fractional derivative [27, 28] of 𝑚−1
function 𝑢(𝑥, 𝑡) is defined as L [𝑢 (𝑥, 𝑡)] = ∑ 𝑠−1−𝑘 𝑢(𝑘) (𝑥, 0)
𝑘=0 (11)
𝑡
1 𝑚−𝛼−1 (𝑚)
𝐷𝑡𝛼 𝑢 (𝑥, 𝑡) = ∫ (𝑡 − 𝜂) 𝑢 (𝑥, 𝜂) 𝑑𝜂, +𝑠 −𝛼
L [A (𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .)] .
Γ (𝑚 − 𝛼) 0 (5)
𝑚 − 1 < 𝛼 ≤ 𝑚, 𝑚 ∈ 𝑁,
Operating with Laplace inverse (denoted by L−1
where Γ(⋅) denotes the gamma function. throughout the present paper) on both sides of (11) gives
where the terms 𝑢𝑛 are to be recursively computed. The Example 1. Consider the Fokker-Planck equation in the case
linear or nonlinear operator B(𝑢, 𝐷𝑥𝛽 𝑢, 𝐷𝑥2𝛽 𝑢, . . .) can be that [30]
decomposed as follows:
𝜕𝑢 𝜕2 𝑢
∞ ∞ ∞ 𝐷𝑡𝛼 𝑢 = + , 0 < 𝛼 ≤ 1, (19)
𝜕𝑥 𝜕𝑥2
B ( ∑ 𝑢𝑛 , 𝐷𝑥𝛽 ∑ 𝑢𝑛 , 𝐷𝑥2𝛽 ∑ 𝑢𝑛 , . . .)
𝑛=0 𝑛=0 𝑛=0 subject to
= B (𝑢0 , 𝐷𝑥𝛽 𝑢0 , 𝐷𝑥2𝛽 𝑢0 , . . .) 𝑢 (𝑥, 0) = 𝑥. (20)
∞ 𝑗 𝑗 𝑗 (15) Taking Laplace transform on both sides of (19) gives
+ ∑ B ( ∑ 𝑢𝑘 , 𝐷𝑥𝛽 ∑ 𝑢𝑘 , 𝐷𝑥2𝛽 ∑ 𝑢𝑘 , . . .)
𝑗=1 𝑘=0 𝑘=0 𝑘=0 𝜕𝑢 𝜕2 𝑢
𝑠𝛼 L [𝑢 (𝑥, 𝑡)] − 𝑠𝛼−1 𝑢 (𝑥, 0) = L [ + ], (21)
∞ 𝑗−1 𝑗−1 𝑗−1 𝜕𝑥 𝜕𝑥2
− ∑ B ( ∑ 𝑢𝑘 , 𝐷𝑥𝛽 ∑ 𝑢𝑘 , 𝐷𝑥2𝛽 ∑ 𝑢𝑘 , . . .) .
𝑗=1 𝑘=0 𝑘=0 𝑘=0 𝑥 1 𝜕𝑢 𝜕2 𝑢
L [𝑢 (𝑥, 𝑡)] = + 𝛼L[ + ]. (22)
𝑠 𝑠 𝜕𝑥 𝜕𝑥2
Substituting (14) and (15) into (13) yields
∞ 𝑚−1
Operating with Laplace inverse on both sides of (22)
−1 −1−𝑘 (𝑘) results in
∑ 𝑢𝑛 = L [ ∑ 𝑠 𝑢 (𝑥, 0)]
𝑛=0 𝑘=0
1 𝜕𝑢 𝜕2 𝑢
𝑢 (𝑥, 𝑡) = 𝑥 + L−1 [ L [ + ]] . (23)
+ B (𝑢0 , 𝐷𝑥𝛽 𝑢0 , 𝐷𝑥2𝛽 𝑢0 , . . .) 𝑠𝛼 𝜕𝑥 𝜕𝑥2
𝑢𝑥 𝑢𝑥2
The iterative Laplace transform method, described in 𝐷𝑡𝛼 𝑢 = −𝐷𝑥𝛽 ( ) + 𝐷𝑥2𝛽 ( ),
Section 3, will be applied to solve three special cases of 6 12 (27)
space-time fractional Fokker-Planck equations with initial
conditions. 𝑡 > 0, 𝑥 > 0, 0 < 𝛼, 𝛽 ≤ 1,
4 Abstract and Applied Analysis
𝑢 (𝑥, 𝑡)
𝑢 (𝑥, 0) = 𝑥2 . (28) 1 2 𝑡𝛼
= 𝑥2 + (− 𝑥3−𝛽 + 𝑥4−2𝛽 ) ⋅
Γ (4 − 𝛽) Γ (5 − 2𝛽) Γ (1 + 𝛼)
4−𝛽 5 − 3𝛽
Taking Laplace transform on both sides of (27) gives +[ 𝑥4−2𝛽 − 𝑥5−3𝛽
6Γ (5 − 2𝛽) 3Γ (6 − 3𝛽)
𝑡 (𝑡/2)2
𝑢 (𝑥, 𝑡) = 𝑥2 (1 + + + ⋅⋅⋅) (34)
Operating with Laplace inverse on both sides of (30), we 2 2!
obtain the following Laplace equation:
and in a closed form by
subject to
𝑢0 (𝑥, 𝑡) = 𝑢 (𝑥, 0) = 𝑥2 ,
𝑢 (𝑥, 0) = 𝑥2 . (37)
𝑢1 (𝑥, 𝑡)
Taking Laplace transform on both sides of (35) gives
1 𝑢𝑥 𝑢 𝑥2
= L [ 𝛼 L [−𝐷𝑥𝛽 ( 0 ) + 𝐷𝑥2𝛽 ( 0 )]]
−1
𝑠𝛼 L [𝑢 (𝑥, 𝑡)] − 𝑠𝛼−1 𝑢 (𝑥, 0)
𝑠 6 12
4𝑢2 𝑥𝑢 (38)
1 2 𝑡𝛼 = L [−𝐷𝑥𝛽 ( − ) + 𝐷𝑥2𝛽 𝑢2 ] ,
= (− 𝑥3−𝛽 + 𝑥4−2𝛽 ) ⋅ , 𝑥 3
Γ (4 − 𝛽) Γ (5 − 2𝛽) Γ (1 + 𝛼)
Table 1: Several approximate values and exact solutions for (36) and (37) when 𝛼 = 𝛽 = 1.
2 ..
1 4(𝑢0 + 𝑢1 ) 𝑥 (𝑢0 + 𝑢1 ) .
= L−1 [ 𝛼
L [−𝐷𝑥𝛽 ( − )
𝑠 𝑥 3 (41)
2
The solution in series form is then given by
+𝐷𝑥2𝛽 (𝑢0 + 𝑢1 ) ]]
𝑢 (𝑥, 𝑡) = 𝑢0 (𝑥, 𝑡) + 𝑢1 (𝑥, 𝑡) + 𝑢2 (𝑥, 𝑡) + ⋅ ⋅ ⋅ . (42)
1 𝑥𝑢 4𝑢02 If we take 𝛼 = 𝛽 = 1, the first few components of the solution
− L−1 [ 𝛼
L [−𝐷𝑥𝛽 ( − 0 ) + 𝐷𝑥2𝛽 𝑢02 ]]
𝑠 𝑥 3 are as follows:
Γ (1 + 2𝛼) Γ (8 − 4𝛽) 𝑢0 (𝑥, 𝑡) = 𝑥2 ,
=−
Γ2 (1 + 𝛼) Γ (1 + 3𝛼) Γ (8 − 5𝛽) Γ (5 − 2𝛽)
𝑢1 (𝑥, 𝑡) = 𝑥2 𝑡,
2304 1056
⋅( + ) 𝑡3𝛼 𝑥7−5𝛽 𝑥2 𝑡2 (43)
Γ (5 − 2𝛽) Γ (4 − 𝛽) 𝑢2 (𝑥, 𝑡) = ,
2
1 184Γ (6 − 2𝛽) 44Γ (6 − 𝛽)
− ( + ) ..
Γ (1 + 2𝛼) Γ (6 − 3𝛽) Γ (5 − 2𝛽) Γ (4 − 𝛽) .
⋅ 𝑡2𝛼 𝑥5−3𝛽 For this special case, the exact solution of (36) and (37) is
therefore given by
Γ (1 + 2𝛼)
+
Γ2 (1 + 𝛼) Γ (1 + 3𝛼) Γ (7 − 4𝛽) Γ (4 − 𝛽) 𝑢 (𝑥, 𝑡) = 𝑥2 𝑒𝑡 (44)
484Γ (7 − 2𝛽) 4224Γ (7 − 3𝛽) which is exactly the result obtained by homotopy perturba-
×( + ) ⋅ 𝑡3𝛼 𝑥6−4𝛽
Γ (4 − 𝛽) Γ (5 − 2𝛽) tion transformation method in [12].
6 Abstract and Applied Analysis
Table 1 shows the numerical solutions for (36) and (37) [7] J. L. Wu, “A wavelet operational method for solving fractional
by using iterative Laplace transform method, homotopy partial differential equations numerically,” Applied Mathematics
perturbation transform method, Adomian decomposition and Computation, vol. 214, no. 1, pp. 31–40, 2009.
method, and the exact solution as 𝛼 = 𝛽 = 1. It should be [8] Ü. Lepik, “Solving fractional integral equations by the Haar
pointed out that only three terms of these methods are used wavelet method,” Applied Mathematics and Computation, vol.
to evaluate the numerical solutions in Table 1. It is obvious 214, no. 2, pp. 468–478, 2009.
that the iterative Laplace transform method used in the [9] V. S. Ertürk and S. Momani, “Solving systems of fractional
present paper has the same convergence as the convergence differential equations using differential transform method,”
of homotopy perturbation transform method and Adomian Journal of Computational and Applied Mathematics, vol. 215, no.
decomposition method for solving this fractional nonlinear 1, pp. 142–151, 2008.
Fokker-Planck equation. Therefore, iterative Laplace trans- [10] Y. Li and N. Sun, “Numerical solution of fractional differential
form method is an effective method for solving fractional equations using the generalized block pulse operational matrix,”
partial differential equations just as homotopy perturbation Computers & Mathematics with Applications, vol. 62, no. 3, pp.
transform method and Adomian decomposition method. 1046–1054, 2011.
[11] A. Saadatmandi and M. Dehghan, “A new operational matrix
for solving fractional-order differential equations,” Computers
5. Conclusions & Mathematics with Applications, vol. 59, no. 3, pp. 1326–1336,
2010.
With the aid of the symbolic computation system Math-
[12] Y. Liu, “Approximate solutions of fractional nonlinear equa-
ematica, the iterative Laplace transform method is first tions using homotopy perturbation transformation method,”
successfully applied to solve fractional Fokker-Planck equa- Abstract and Applied Analysis, vol. 2012, Article ID 752869, 14
tions. The results obtained by the iterative Laplace transform pages, 2012.
method are the same as those obtained by homotopy per- [13] Y. Liu, “Variational homotopy perturbation method for solv-
turbation transform method and Adomian decomposition ing fractional initial boundary value problems,” Abstract and
method. The method finds the solutions without unneces- Applied Analysis, vol. 2012, Article ID 727031, 10 pages, 2012.
sary linearization, small perturbation and other restrictive [14] N. H. Sweilam, M. M. Khader, and R. F. Al-Bar, “Numeri-
assumptions. Therefore, the method considerably reduces the cal studies for a multi-order fractional differential equation,”
computational work to a great extent. It is worth mentioning Physics Letters A, vol. 371, no. 1-2, pp. 26–33, 2007.
that the method can also be applied to solve other nonlinear [15] S. Das, “Analytical solution of a fractional diffusion equation by
fractional differential equations with initial value conditions. variational iteration method,” Computers & Mathematics with
Applications, vol. 57, no. 3, pp. 483–487, 2009.
Acknowledgments [16] H. Jafari, C. M. Khalique, and M. Nazari, “Application of the
Laplace decomposition method for solving linear and nonlin-
The author would like to express her sincere thanks to ear fractional diffusion-wave equations,” Applied Mathematics
the referees for their valuable suggestions which lead to an Letters, vol. 24, no. 11, pp. 1799–1805, 2011.
improved version. This research is supported by Excellent [17] M. Y. Ongun, “The Laplace adomian decomposition method for
Young Scientist Foundation of Shandong Province under solving a model for HIV infection of CD4+ cells,” Mathematical
Grant no. BS2013HZ026. and Computer Modelling, vol. 53, no. 5-6, pp. 597–603, 2011.
[18] V. Daftardar-Gejji and H. Jafari, “An iterative method for solving
nonlinear functional equations,” Journal of Mathematical Anal-
References ysis and Applications, vol. 316, no. 2, pp. 753–763, 2006.
[1] J. Sabatier, O. P. Agrawal, and J. A. Tenreiro Machado, Advances [19] H. Jafari, Iterative Methods for solving system of fractional
in Fractional Calculus: Theoretical Developments and Applica- differential equations [Ph.D. thesis], Pune University, 2006.
tions in Physics and Engineering, Springer, 2007. [20] S. Bhalekar and V. Daftardar-Gejji, “Solving evolution equations
[2] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional using a new iterative method,” Numerical Methods for Partial
Calculus, vol. 3 of Series on Complexity, Nonlinearity and Chaos, Differential Equations, vol. 26, no. 4, pp. 906–916, 2010.
World Scientific, Singapore, 2012. [21] V. Daftardar-Gejji and S. Bhalekar, “Solving fractional boundary
[3] Y. Liu and B. Xin, “Numerical solutions of a fractional predator- value problems with Dirichlet boundary conditions using a new
prey system,” Advances in Difference Equations, vol. 2011, Article iterative method,” Computers & Mathematics with Applications,
ID 190475, 11 pages, 2011. vol. 59, no. 5, pp. 1801–1809, 2010.
[4] J. Ma and Y. Liu, “Exact solutions for a generalized nonlinear [22] H. Jafari, M. Nazari, D. Baleanu, and C. M. Khalique, “A new
fractional Fokker-Planck equation,” Nonlinear Analysis: Real approach for solving a system of fractional partial differential
World Applications, vol. 11, no. 1, pp. 515–521, 2010. equations,” Computers & Mathematics with Applications, vol. 66,
[5] M. Rehman and R. Ali Khan, “The Legendre wavelet method no. 5, pp. 838–843, 2013.
for solving fractional differential equations,” Communications in [23] H. Jafari and S. Seifi, “Solving a system of nonlinear fractional
Nonlinear Science and Numerical Simulation, vol. 16, no. 11, pp. partial differential equations using homotopy analysis method,”
4163–4173, 2011. Communications in Nonlinear Science and Numerical Simula-
[6] Y. Li, “Solving a nonlinear fractional differential equation using tion, vol. 14, no. 5, pp. 1962–1969, 2009.
Chebyshev wavelets,” Communications in Nonlinear Science and [24] H. Risken, The Fokker-Planck Equation, Methods of Solution and
Numerical Simulation, vol. 15, no. 9, pp. 2284–2292, 2010. Applications, Springer, Berlin, Germany, 1989.
Abstract and Applied Analysis 7
Research Article
The Second Noether Theorem on Time Scales
Copyright © 2013 A. B. Malinowska and N. Martins. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
We extend the second Noether theorem to variational problems on time scales. As corollaries we obtain the classical second Noether
theorem, the second Noether theorem for the ℎ-calculus and the second Noether theorem for the 𝑞-calculus.
theorem is valid for an infinite number of time scales. As 𝑎 < 𝑏. We consider the following optimization problem on
we will see, in the particular case where the time scale T T:
is R, we get from our result the classical second Noether
𝑏
theorem; when T = Z, we obtain the analogue of the second
Noether theorem for the difference calculus of variations; L [𝑦] = ∫ 𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) Δ𝑡 → extremize, (1)
𝑎
when T = 𝑞N0 (for some 𝑞 > 1), we obtain a new result:
the second Noether theorem for the 𝑞-calculus (quantum where the set of admissible functions are
calculus). For more on the theory of quantum calculus and
quantum calculus of variation, we refer to [11–21]. D = {𝑦 | 𝑦 : [𝑎, 𝑏] ∩ T → R𝑛 , 𝑦 ∈ 𝐶rd
1
([𝑎, 𝑏] ∩ T, R𝑛 ) ,
The theory of time scales was introduced in 1988 by
Hilger in his Ph.D. thesis [22] as a means of unifying theories 𝑦 (𝑎) = 𝛼, 𝑦 (𝑏) = 𝛽} ,
of differential calculus and difference calculus into a single (2)
theory. With a short time this unification aspect has been
supplemented by the extension and generalization features.
The time scale calculus allows to consider more complex time for some 𝛼, 𝛽 ∈ R𝑛 , and where 𝜎 is the forward jump operator,
domains, such as ℎZ, T = 𝑞N0 , or hybrid domains. The study 𝑦Δ is the delta-derivative of 𝑦, and, for 𝑖 ∈ N,
of the calculus of variations in the context of time scales had
𝑖
its beginning only in 2004 with the paper [23] of Bohner 𝐶rd ([𝑎, 𝑏] ∩ T, R𝑛 ) := {𝑦 | 𝑦 : [𝑎, 𝑏] ∩ T → R𝑛 ,
(see also [24]). Since then, the variational calculus on time
𝑖 𝑖
scales advanced fairly quickly, as can be verified with the large 𝑦Δ is rd-continuous on [𝑎, 𝑏]𝜅 } .
number of published papers on the subject [25–35]. Noether’s
first theorem has been extended to the variational calculus (3)
on time scales using several approaches [4, 9, 36], while the
𝑖
second Noether theorem on times scales is still not available As usual, 𝑦𝜎 (𝑡) denotes 𝑦(𝜎(𝑡)) and [𝑎, 𝑏]𝜅 := [𝑎, 𝜌𝑖 (𝑏)],
in the literature. So, there is evidently a need for a time scale where 𝜌 is the backward jump operator. By extremize, we
analogue of Noether’s second theorem. mean maximize or minimize.
The paper is organized as follows. In Section 2, we In what follows, all intervals are time scales intervals; that
review some preliminaries about single-variable variational is, we simply write [𝑎, 𝑏] to denote the set [𝑎, 𝑏] ∩ T. Let 𝑦 :=
calculus on time scales; for example, we recall the Euler- (𝑦1 , . . . , 𝑦𝑛 ) and denote by 𝜕𝐿/𝜕𝑦𝑘 the partial derivative of 𝐿
Lagrange equation for a delta variational problem. Our main with respect to 𝑦𝑘 .
results are stated in Section 3. Namely, in Subsection 3.1, we
1
prove Noether’s second theorem for variational problems Definition 1. We say that 𝑦∗ ∈ 𝐶rd ([𝑎, 𝑏], R𝑛 ) is a local mini-
involving a single delta integral (with and without trans- mizer (resp., local maximizer) for problem (1) if there exists
formation of time) and in Section 3.2 we prove Noether’s 𝛿 > 0 such that
second theorem for variational problems involving multiple
delta integrals (without transformation of time). Section 4 L [𝑦∗ ] ≤ L [𝑦] (resp., L [𝑦∗ ] ≥ L [𝑦]) , (4)
provides a concrete example of application of our results.
Finally, in Section 5, we present some concluding remarks. 1
for all 𝑦 ∈ 𝐶rd ([𝑎, 𝑏], R𝑛 ) satisfying the boundary conditions
𝑦(𝑎) = 𝛼, 𝑦(𝑏) = 𝛽, and
2. Preliminaries
𝜎
𝑦 − 𝑦∗ := sup 𝑦 (𝑡) − 𝑦∗ (𝑡)
𝜎
In this paper, we assume the reader to be familiar with the 𝜅
𝑡∈[𝑎,𝑏]
calculus on time scales. For a good introduction to the theory
(5)
+ sup 𝑦Δ (𝑡) − 𝑦∗Δ (𝑡) < 𝛿,
of time scales, we refer to the well-known books in this field
[37, 38]. The first developments on time scale calculus were 𝑡∈[𝑎,𝑏] 𝜅
by 𝜙(𝜖) := L[𝑦 + 𝜖𝜂] and 𝑓(𝑡, 𝜖) := 𝐿(𝑡, 𝑦𝜎 (𝑡) + 𝜖𝜂𝜎 (𝑡), 𝑦Δ (𝑡) + 3.1. Noether’s Second Theorem: Single Delta Integral Case. In
𝜖𝜂Δ (𝑡)). If 𝜕𝑓/𝜕𝜖 is continuous in 𝜖, uniformly in 𝑡, then this subsection, we suppose that the time scale T satisfies
condition (H) and that 𝐿 satisfies the assumption of Lemma 4,
0 0
𝑏 𝑛
𝜕𝐿 for every 𝑦 and 𝜂. As usual, 𝜂Δ and 𝜂𝜎 denote 𝜂. Let 𝑚 be
𝜙 ̇(0) = ∫ ∑ ( (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) 𝜂𝑘𝜎 (𝑡)
𝑎 𝑘=1 𝜕𝑦𝑘𝜎 a fixed natural number. We also assume that the time scale
(6) interval [𝑎, 𝑏] has, at least, 2𝑚 + 1 points.
𝜕𝐿 We begin with some technical results that will be useful
+ Δ (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) 𝜂𝑘Δ (𝑡)) Δ𝑡. in the proofs of Theorems 16 and 22.
𝜕𝑦𝑘
Lemma 9 (higher-order fundamental lemma of the calculus
Next, we present the following result that is a fundamental
of variations). Let T be a time scale satisfying condition (H)
tool in the calculus of variations on time scales.
and 𝑓0 , 𝑓1 , . . . , 𝑓𝑚 ∈ 𝐶rd ([𝑎, 𝑏], R). If
Theorem 5 (Euler-Lagrange equation on time scales [23]). If 𝜌𝑚−1 (𝑏) 𝑚 𝑚−𝑖
Δ𝑖
𝑦∗ is a weak local extremizer for problem (1) and 𝐿 satisfies ∫ (∑𝑓𝑖 (𝑡) 𝜂𝜎 (𝑡)) Δ𝑡 = 0, (9)
the assumption of Lemma 4, for every 𝑦 and 𝜂, then the 𝑎 𝑖=0
components of 𝑦∗ satisfy the 𝑛 Euler-Lagrange equations: 2𝑚
for all 𝜂 ∈ 𝐶rd ([𝑎, 𝑏], R) such that
Δ 𝜕𝐿 𝜕𝐿 𝜂 (𝜌𝑚−1 (𝑏)) = 0,
Δ
(𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) = 𝜎 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) , 𝜂 (𝑎) = 0,
Δ𝑡 𝜕𝑦𝑘 𝜕𝑦𝑘 (7)
..
𝑘 = 1, . . . , 𝑛 . (10)
𝑚−1 𝑚−1
for 𝑡 ∈ [𝑎, 𝑏]𝜅 . 𝜂Δ (𝑎) = 0, 𝜂Δ (𝜌𝑚−1 (𝑏)) = 0,
Lemma 7 (see [27]). Let T be a time scale satisfying condition 𝜂𝜎 (𝑎) = 0, 𝑖 = 0, 1, . . . , 𝑚. (13)
(𝐻). If 𝑓 : T → R is two times delta differentiable, then Proof. If 𝑎 is right-dense, the result is trivial. Suppose that 𝑎
2 is right-scattered. Since 𝜂Δ (𝑎) = (𝜂𝜎 (𝑎) − 𝜂(𝑎))/(𝜎(𝑎) − 𝑎) = 0
𝑓𝜎Δ (𝑡) = 𝑏1 𝑓Δ𝜎 (𝑡) , 𝑡 ∈ T𝜅 . (8)
and 𝜂(𝑎) = 0, we conclude that 𝜂𝜎 (𝑎) = 0. Since 𝜂Δ (𝑎) =
2
Lemma 12. Assume that the time scale T satisfies condition Theorem 15 (necessary condition of invariance). If func-
2(𝑚−1) tional L is invariant under transformations (16), then
(H) and 𝜂 ∈ 𝐶rd ([𝑎, 𝑏], R) is such that,
𝑖 𝜎
𝜂Δ (𝑎) = 0, 𝑖 = 0, 1, . . . , 𝑚 − 1. (14) 𝑛
𝜕𝐿 𝑏 𝑟
∑ ∫ ( 𝜎 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) ⋅ ( ∑ 𝑇𝑘𝑗 (𝑝𝑗 )) (𝑡)
𝑘=1 𝑎
𝜕𝑦𝑘 𝑗=1
Then,
Δ
𝑟
𝜎Δ𝑚−2 𝜎2 Δ𝑚−3 𝜎3 Δ𝑚−4 𝜎𝑚−2 Δ 𝜕𝐿
𝜂 (𝑎) = 𝜂 (𝑎) = 𝜂 (𝑎) = . . . = 𝜂 (𝑎) = 0. + Δ (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) ⋅ ( ∑ 𝑇𝑘𝑗 (𝑝𝑗 )) (𝑡)) Δ𝑡 = 0.
(15) 𝜕𝑦𝑘 𝑗=1
(19)
Proof. If 𝑎 is right-dense, the result is trivial. Suppose that 𝑎
𝑚−1 𝑚−2 𝑚−2
is right-scattered. Since 𝜂Δ (𝑎) = (𝜂Δ 𝜎 (𝑎) − 𝜂Δ (𝑎))/ Proof. Using the definition of invariance and noting that the
𝑚−2 𝑚−2
(𝜎(𝑎) − 𝑎) = 0 and 𝜂Δ (𝑎) = 0, then 𝜂Δ 𝜎 (𝑎) = 0. Using family of transformations (16) depends upon arbitrary func-
𝑚−2
Lemma 7, we get 𝜂𝜎Δ (𝑎) = 0 (or use Lemma 8). Note that tions 𝑝1 , 𝑝2 , . . . , 𝑝𝑟 , we conclude that, for any real number 𝜀,
𝑚−2 𝑚−3 𝑚−3
𝜂𝜎Δ (𝑎) = (𝜂𝜎Δ 𝜎 (𝑎) − 𝜂𝜎Δ (𝑎))/(𝜎(𝑎) − 𝑎) = 0. Lemma 8
𝑚−3 𝑚−3 𝑏
shows that 𝜂𝜎Δ (𝑎) = 0; hence, 𝜂𝜎Δ 𝜎 (𝑎) = 0. Using ∫ 𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) Δ𝑡
2 𝑚−3 2 𝑚−4
Lemma 7, we get 𝜂𝜎 Δ (𝑎) = 0. Next, we prove that 𝜂𝜎 Δ 𝑎
𝑚−3 𝑚−4 𝑚−4
(𝑎) = 0. Since 𝜂𝜎Δ (𝑎) = (𝜂𝜎Δ 𝜎 (𝑎) − 𝜂𝜎Δ (𝑎))/ (𝜎(𝑎) − 𝑏 𝑟
𝜎
𝑚−4
𝑎) = 0 and 𝜂𝜎Δ (𝑎) = 0 (by Lemma 8), then 𝜂𝜎Δ 𝜎 (𝑎) =
𝑚−4
=∫ 𝐿 (𝑡, 𝑦1𝜎 (𝑡) + 𝜀( ∑ 𝑇 (𝑝𝑗 )) (𝑡) , . . . , 𝑦𝑛𝜎 (𝑡)
1𝑗
𝑗=1
Let 𝑦 := (𝑦1 , 𝑦2 , . . . , 𝑦𝑛 ). Firstly, we will prove the
second Noether theorem without transformation of time. For Δ
𝑟
that consider the following transformations that depend on 𝑛𝑗
+𝜀( ∑ 𝑇 (𝑝𝑗 )) (𝑡)) Δ𝑡.
arbitrary functions 𝑝1 , 𝑝2 , . . . , 𝑝𝑟 and their delta-derivatives 𝑗=1
up to order 𝑚:
(20)
𝑡=𝑡
Differentiating with respect to 𝜀 (use Lemma 4) and taking
𝑟 (16)
𝑘𝑗 𝜀 = 0, we get equality (19).
𝑦𝑘 (𝑡) = 𝑦𝑘 (𝑡) + ∑ 𝑇 (𝑝𝑗 ) (𝑡) , 𝑘 = 1, 2, . . . , 𝑛,
𝑗=1
2𝑚
where, for each 𝑗 = 1, 2, . . . , 𝑟, 𝑝𝑗 ∈ 𝐶rd ([𝑎, 𝜎𝑚 (𝑏)], R), Define
𝑚 𝑚−(𝑖+1)
Δ𝑖
𝜕𝐿 Δ 𝜕𝐿
𝑇𝑘𝑗 (𝑝𝑗 ) := ∑𝑔𝑖𝑗𝑘 𝑝𝑗𝜎 , (17) 𝐸𝑘 (𝐿) := − , 𝑘 = 1, 2, . . . , 𝑛. (21)
𝜕𝑦𝑘𝜎 Δ𝑡 𝜕𝑦𝑘Δ
𝑖=0
that is, 𝑦Δ (𝑡) = Δ ℎ [𝑦](𝑡) = (𝑦(𝑡+ℎ)−𝑦(𝑡))/ℎ), then there exist Moreover, we assume that the map
the following identities:
𝑟
𝑛 𝑚 𝑡 → 𝛼 (𝑡) := 𝑡 + ∑ 𝐻𝑗 (𝑝𝑗 ) (𝑡) (41)
∑ ∑(−1)𝑖 [𝑔𝑖𝑗𝑘 (𝑡 + ℎ) 𝑗=1
𝑘=1 𝑖=0
1
Δ𝑖ℎ (36) is a strictly increasing 𝐶rd function and its image is again a
⋅𝐸𝑘 (𝐿) (𝑡, 𝑦 (𝑡 + ℎ) , Δ ℎ [𝑦] (𝑡)) ] = 0, time scale, T. We denote the forward shift operator relative
𝑡 ∈ [𝑎, 𝑏 − 𝑚ℎ] , 𝑗 = 1, 2, . . . , 𝑟. to T by 𝜎 and the delta derivative by Δ. We remark that the
following holds [40]:
For ℎ = 1, we obtain the analogue of Noether’s second
𝜎 ∘ 𝛼 = 𝛼 ∘ 𝜎. (42)
theorem for the difference calculus of variations recently
proved in [39]. In the case T = 𝑞N0 , we obtain the new result. Definition 20. Functional L is invariant under transforma-
1
tions (39) if, and only if, for all 𝑦 ∈ 𝐶rd ([𝑎, 𝑏], R), we have
Corollary 19. Let 𝑞 > 1, 𝐿 : 𝑞N0 × R𝑛 × R𝑛 → R and 𝑎, 𝑏 ∈
𝑞N0 , 𝑎 < 𝑏. If functional L defined by 𝑏 𝑏
∫ 𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) Δ𝑡 = ∫ 𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) Δ𝑡.
𝑏/𝑞 𝑎 𝑎
L [𝑦] = ∑𝐿 (𝑡, 𝑦 (𝑞𝑡) , Δ 𝑞 [𝑦] (𝑡)) (37) (43)
𝑡=𝑎
We recall the following results that will be very useful in
is invariant under transformations (16) (where 𝜎 denotes in this the proof of Theorem 22.
context the function 𝜎(𝑡) = 𝑞𝑡 and Δ denotes the 𝑞-derivative,
that is, 𝑦Δ (𝑡) = Δ 𝑞 [𝑦](𝑡) = (𝑦(𝑞𝑡) − 𝑦(𝑡))/(𝑞 − 1)𝑡), then there Theorem 21 (see [37]). Assume that ] : T → R is strictly
exist the following identities: increasing and T̃ := ](T) is a time scale.
̃
(1) Chain rule: let 𝜔 : T̃ → R. If ]Δ (𝑡) and 𝜔Δ (](𝑡)) exist
𝑛
1 𝑖(𝑖+1)/2 𝑘
𝑚
𝑖
∑ ∑(−1) ( ) [𝑔𝑖𝑗 (𝑞𝑡) for all 𝑡 ∈ T 𝜅 , then
𝑘=1 𝑖=0
𝑞
̃
Δ𝑖𝑞 (𝜔 ∘ ])Δ = (𝜔Δ ∘ ]) ]Δ . (44)
⋅𝐸𝑘 (𝐿) (𝑡, 𝑦 (𝑞𝑡) , Δ 𝑞 [𝑦] (𝑡))] = 0,
Let 𝑟 ≠
0 and define Hence, using (50) and (52), we get
̃ (𝑡, 𝑠, 𝑦, 𝑟, V) := 𝐿 (𝑠 − 𝜇 (𝑡) 𝑟, 𝑦, V ) 𝑟.
𝐿 (47)
̃ [𝑠 (⋅) , 𝑦 (⋅)] = L
L ̃ [𝐻 (𝑠 (⋅) , 𝑦 (⋅)) , 𝑇 (𝑠 (⋅) , 𝑦 (⋅))] . (53)
𝑟
̃ is invariant on
This means that L
1
Note that, for 𝑠(𝑡) = 𝑡 and any 𝑦 ∈ 𝐶rd ([𝑎, 𝑏], R𝑛 ), we have
̃ = {(𝑠, 𝑦) | 𝑠 (𝑡) = 𝑡 ∧ 𝑦 ∈ 𝐶1 ([𝑎, 𝑏] , R𝑛 )}
𝑈 (54)
rd
𝜎 Δ̃ (𝑡, 𝑠 (𝑡) , 𝑦 (𝑡) , 𝑠 (𝑡) , 𝑦 (𝑡)) .
𝐿 (𝑡, 𝑦 (𝑡) , 𝑦 (𝑡)) = 𝐿 𝜎 𝜎 Δ Δ
under the group of state transformations
(48)
(𝑠, 𝑦) = (𝐻 (𝑠, 𝑦) , 𝑇 (𝑠, 𝑦)) (55)
Therefore, for 𝑠(𝑡) = 𝑡,
in the sense of Definition 13.
𝑏
𝜎 Δ Using Theorem 16, we can conclude that there exist the
L [𝑦] := ∫ 𝐿 (𝑡, 𝑦 (𝑡) , 𝑦 (𝑡)) Δ𝑡
𝑎 following 𝑟 identities (𝑗 = 1, 2, . . . , 𝑟)
𝑏 𝑛 𝑚
1 𝑖(𝑖+1)/2
𝑖
𝜎 Δ
=∫ 𝐿 ̃ [𝑠, 𝑦] .
̃ (𝑡, 𝑠𝜎 (𝑡) , 𝑦𝜎 (𝑡) , 𝑠Δ (𝑡) , 𝑦Δ (𝑡)) Δ𝑡 := L ∑ ∑(−1)𝑖 ( ) ̃
((𝑔𝑖𝑗𝑘 ) 𝐸𝑘 (𝐿))
𝑎 𝑘=1 𝑖=0
𝑏1
(49) (56)
𝑚
1 𝑖(𝑖+1)/2
𝑖
𝜎 Δ
Note that, for 𝑠(𝑡) = 𝑡, + ∑(−1) ( ) 𝑖 ̃
((𝑓𝑖𝑗 ) 𝐸𝑠 (𝐿)) ≡ 0,
𝑖=0 𝑏1
̃ [𝑠 (⋅) , 𝑦 (⋅)]
L
̃ := (𝜕𝐿/𝜕𝑠
where we denote 𝐸𝑠 (𝐿) ̃ 𝜎 ) − (Δ/Δ𝑡)(𝜕𝐿/𝜕𝑠
̃ Δ ).
= L [𝑦 (⋅)] Note that, for 𝑠(𝑡) = 𝑡,
𝑏 ̃
𝜕𝐿
𝜎
= ∫ 𝐿 (𝑡, 𝑦 (𝑡) , 𝑦 (𝑡)) Δ𝑡 Δ (𝑡, 𝑠𝜎 (𝑡) , 𝑦𝜎 (𝑡) , 𝑠Δ (𝑡) , 𝑦Δ (𝑡))
𝑎
𝜕𝑠𝜎
𝛼(𝑏) 𝜕𝐿 𝜎 𝑦Δ (𝑡)
=∫ Δ
𝐿 (𝑡, (𝑦 ∘ 𝜎) (𝑡) , 𝑦 (𝑡)) Δ𝑡 = (𝑠 (𝑡) − 𝜇 (𝑡) 𝑠Δ (𝑡) , 𝑦𝜎 (𝑡) , Δ ) 𝑠Δ (𝑡) ,
𝛼(𝑎)
𝜕𝑡 𝑠 (𝑡)
𝑏 𝜕𝐿̃
= ∫ 𝐿 (𝛼 (𝑡) , (𝑦 ∘ 𝜎 ∘ 𝛼) (𝑡) , 𝑦Δ (𝛼 (𝑡))) 𝛼Δ (𝑡) Δ𝑡 (𝑡, 𝑠𝜎 (𝑡) , 𝑦𝜎 (𝑡) , 𝑠Δ (𝑡) , 𝑦Δ (𝑡))
𝑎
𝜕𝑠Δ
𝑏 (𝑦 ∘ 𝛼) (𝑡)
Δ 𝑦Δ (𝑡)
= 𝐿 (𝑠𝜎 (𝑡) − 𝜇 (𝑡) 𝑠Δ (𝑡) , 𝑦𝜎 (𝑡) , )
= ∫ 𝐿 (𝛼 (𝑡) , (𝑦 ∘ 𝛼 ∘ 𝜎) (𝑡) , Δ (𝑡)
) 𝛼Δ (𝑡) Δ𝑡 𝑠Δ (𝑡)
𝑎 𝛼
Δ
𝑛
𝑦𝑘Δ (𝑡) 𝜕𝐿 𝜎 Δ 𝜎 𝑦Δ (𝑡)
𝑏
𝜎 (𝑦 ∘ 𝛼) (𝑡) −∑ (𝑠 (𝑡) − 𝜇 (𝑡) 𝑠 (𝑡) , 𝑦 (𝑡) , )
= ∫ 𝐿 (𝛼𝜎 (𝑡)−𝜇 (𝑡) 𝛼Δ (𝑡) , (𝑦 ∘ 𝛼) (𝑡) , ) 𝑘=1
𝑠Δ (𝑡) 𝜕𝑦𝑘Δ 𝑠Δ (𝑡)
𝑎 𝛼Δ (𝑡)
𝜕𝐿 𝜎 𝑦Δ (𝑡)
⋅ 𝛼Δ (𝑡) Δ𝑡 − (𝑠 (𝑡) − 𝜇 (𝑡) 𝑠Δ (𝑡) , 𝑦𝜎 (𝑡) , Δ )
𝜕𝑡 𝑠 (𝑡)
𝑏
̃ (𝑡, 𝛼𝜎 (𝑡) , (𝑦 ∘ 𝛼)𝜎 (𝑡) , 𝛼Δ (𝑡) , (𝑦 ∘ 𝛼)Δ (𝑡)) Δ𝑡
=∫ 𝐿 ⋅ 𝜇 (𝑡) 𝑠Δ (𝑡) .
𝑎
(57)
̃ [𝛼 (⋅) , (𝑦 ∘ 𝛼) (⋅)] .
=L Hence, for 𝑠(𝑡) = 𝑡,
(50)
̃ (𝑡, 𝑠𝜎 (𝑡) , 𝑦𝜎 (𝑡) , 𝑠Δ (𝑡) , 𝑦Δ (𝑡))
𝐸𝑠 (𝐿)
1 2 𝑛
Let 𝐻(𝑡, 𝑦(𝑡)) := 𝛼(𝑡) and 𝑇 = (𝑇 , 𝑇 , . . . , 𝑇 ) where
𝜕𝐿
𝑟 = (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡))
𝜕𝑡
𝑇𝑘 (𝑡, 𝑦 (𝑡)) := 𝑦𝑘 (𝑡) + ∑ 𝑇𝑘𝑗 (𝑝𝑗 ) (𝑡) , 𝑘 = 1, 2, . . . , 𝑛.
𝑗=1
Δ
(51) − (𝐿 (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡))
Δ𝑡 (58)
Then, for 𝑠(𝑡) = 𝑡, 𝑛
𝜕𝐿
− ∑ 𝑦𝑘Δ (𝑡) (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡))
(𝛼 (𝑡) , (𝑦 ∘ 𝛼) (𝑡)) = (𝑡, 𝑦 (𝑡)) 𝜕𝑦𝑘Δ
𝑘=1
Also note that, for 𝑠(𝑡) = 𝑡 and 𝑘 = 1, 2, . . . , 𝑛, 3.2. Noether’s Second Theorem: Multiple Delta Integral Case.
In this subsection, we extend the second Noether theorem
̃ (𝑡, 𝑠𝜎 (𝑡) , 𝑦𝜎 (𝑡) , 𝑠Δ (𝑡) , 𝑦Δ (𝑡))
𝐸𝑘 (𝐿) (without transformation of time) to multiple integral vari-
(59) ational problems in the time scale setting. For simplicity
= 𝐸𝑘 (𝐿) (𝑡, 𝑦𝜎 (𝑡) , 𝑦Δ (𝑡)) . of presentation, we prove the result for the case of two
independent variables and transformations that depend on an
Substituting the above equalities into (56), we conclude the arbitrary function and its first-order partial delta derivatives.
desired result: Clearly, our result can be generalized for 𝑛 independent
variables and 𝑟 arbitrary functions and their higher-order
𝑛 𝑚
1 𝑖(𝑖+1)/2 [
𝑖
𝜎 Δ partial delta derivatives.
∑ ∑(−1)𝑖 ( ) ((𝑔𝑖𝑗𝑘 ) 𝐸𝑘 (𝐿))
𝑘=1 𝑖=0
𝑏1 For the convenience of the reader, we recall notions and
[ results that are needed in the sequel. A general introduction to
Δ𝑖 differential calculus and integration theory for multivariable
𝜕𝐿 Δ
𝜎 𝜕𝐿 𝜕𝐿
+((𝑓𝑖𝑗 ) ( − (𝐿 − 𝑦𝑘Δ Δ − 𝜇 ))) ] ≡ 0 functions on time scales is presented, respectively, in [41]
𝜕𝑡 Δ𝑡 𝜕𝑦𝑘 𝜕𝑡 (see also [42]) and [43]. For the double integral calculus of
] variations on time scales, we refer the reader to [44].
(60)
Let T1 and T2 be two given time scales. For 𝑖 = 1, 2,
for 𝑗 = 1, 2, . . . , 𝑟. denote by 𝜎𝑖 and Δ 𝑖 the forward jump operator and the delta
(1)
derivative on T𝑖 , respectively. Let 𝐶rd denote the set of all
Remark 23. Define continuous functions defined on T1 × T2 for which both the
𝑛
Δ 1 -partial derivative and the Δ 2 -partial derivative exist and
𝜕𝐿 Δ 𝜕𝐿 𝜕𝐿 are of class 𝐶rd (for a definition see [44]).
𝐸𝑘𝑠 (𝐿) := − (𝐿 − ∑ 𝑦𝑘Δ Δ − 𝜇 ) , (61)
𝜕𝑡 Δ𝑡 𝑘=1 𝜕𝑦𝑘 𝜕𝑡 Let Ω ⊆ T1 × T2 be an 𝜔-type set and let Γ be its positively
fence (see [44]). Denote
for 𝑘 = 1, 2, . . . , 𝑛. Then, 𝐸𝑘𝑠 (𝐿) = 0 are the second Euler-
Lagrange equations for problem (1) [36]. Therefore, expres- Ω∘ := {(𝑥, 𝑦) ∈ Ω : (𝜎1 (𝑥) , 𝜎2 (𝑦)) ∈ Ω} . (65)
sion (46) provides “dependencies” between two types of the Let a function 𝐿(𝑥, 𝑦, 𝑢, 𝑝, 𝑞), where (𝑥, 𝑦) ∈ Ω ∪ Γ and
Euler-Lagrange expressions. (𝑢, 𝑝, 𝑞) ∈ R3𝑛 , be given. We will suppose that 𝐿 is continuous,
together with its partial delta derivatives of first and second
Note that if T = R, Noether’s identity (46) simplifies
order with respect to 𝑥, 𝑦 and partial usual derivatives of the
because
first and second order with respect to 𝑢, 𝑝, 𝑞. In what follows,
𝜕𝐿 Δ 𝑛
𝜕𝐿 𝜕𝐿 𝑢Δ 1 and 𝑢Δ 2 denote, respectively, 𝜕𝑢/Δ 1 𝑥 and 𝜕𝑢/Δ 2 𝑦.
− (𝐿 − ∑ 𝑦𝑘Δ Δ − 𝜇 ) Consider the following optimization problem:
𝜕𝑡 Δ𝑡 𝑘=1 𝜕𝑦𝑘 𝜕𝑡
(62)
𝑛 𝑛 L [𝑢] = ∫ ∫ 𝐿 (𝑥, 𝑦, 𝑢 (𝜎1 (𝑥) , 𝜎2 (𝑦)) , 𝑢Δ 1 (𝑥, 𝜎2 (𝑦)) ,
𝜕𝐿 𝑑 𝜕𝐿 Ω
= − (𝐿 − ∑ 𝑦𝑘̇ ) = − ∑ 𝑦𝑘̇𝐸𝑘 (𝐿) ,
𝜕𝑡 𝑑𝑡 𝜕𝑦𝑘̇
𝑘=1 𝑘=1 𝑢Δ 2 (𝜎1 (𝑥) , 𝑦)) Δ1 𝑥Δ 2 𝑦 → extremize,
and we obtain the following corollary. (66)
Corollary 24 (classical Noether’s second theorem, cf. [1]). If where the set of admissible functions are
functional L defined by D = {𝑢 | 𝑢 : Ω ∪ Γ → R𝑛 , 𝑢 ∈ 𝐶rd
(1)
, 𝑢 = 𝑔 on Γ} , (67)
𝑏
where 𝑔 is a fixed function defined and continuous on the
L [𝑦] = ∫ 𝐿 (𝑡, 𝑦 (𝑡) , 𝑦̇(𝑡)) 𝑑𝑡 (63)
𝑎 fence Γ of Ω.
As noticed in [44], for the variational problem (66) being
is invariant under transformations (39) (where 𝜎 denotes in well posed, we have to assume that there exists at least one
this context the identity function and Δ denotes the usual admissible function 𝑢0 ∈ D because it is possible to choose a
derivative), then there exist the following identities: continuous function 𝑔 such that no function 𝑢 is admissible.
𝑛 𝑚 Note that if there exists an admissible function 𝑢0 , then the set
(𝑖) (𝑖)
∑ ∑(−1)𝑖 [(𝑔𝑖𝑗𝑘 𝐸𝑘 (𝐿)) − (𝑓𝑖𝑗 ⋅ 𝑦𝑘̇𝐸𝑘 (𝐿)) ] ≡ 0, D contains a set of functions of the form 𝑢 = 𝑢0 + 𝜂, where
(1)
𝑘=1 𝑖=0 (64) 𝜂 : Ω ∪ Γ → R𝑛 is 𝐶rd and 𝜂 = 0 on Γ. Any such 𝜂 is called
an admissible variation for problem (66).
𝑗 = 1, 2, . . . , 𝑟.
Definition 25. We say that 𝑢∗ ∈ D is a local minimizer (resp.,
In the case T = ℎZ (for some ℎ > 0), we obtain from local maximizer) for problem (66) if there exists 𝛿 > 0 such
Theorem 22 the second Noether theorem for the ℎ-calculus; that
whereas if T = 𝑞N0 (for some 𝑞 > 1), we get the second
Noether theorem for the 𝑞-calculus. L [𝑢∗ ] ≤ L [𝑢] (resp., L [𝑢∗ ] ≥ L [𝑢]) , (68)
Abstract and Applied Analysis 9
for all 𝑢 ∈ D with ℎ > 0), 𝑞N0 (for some 𝑞 > 1), and many other interesting time
scales satisfy property (74).
𝑢 − 𝑢∗ := sup 𝑢 (𝑥, 𝑦) − 𝑢∗ (𝑥, 𝑦) Let 𝑢(𝑥, 𝑦) = (𝑢1 (𝑥, 𝑦), 𝑢2 (𝑥, 𝑦), . . . , 𝑢𝑛 (𝑥, 𝑦)) and con-
(𝑥,𝑦)∈Ω∪Γ
sider the following transformations that depend on an arbi-
trary continuous function 𝑝 and the partial delta derivatives
+ sup 𝑢Δ 1 (𝑥, 𝜎2 (𝑦)) − 𝑢∗Δ 1 (𝑥, 𝜎2 (𝑦))
(𝑥,𝑦)∈Ω of 𝑝:
+ sup 𝑢Δ 2 (𝜎1 (𝑥) , 𝑦) − 𝑢∗Δ 2 (𝜎1 (𝑥) , 𝑦) < 𝛿, 𝑥=𝑥
(𝑥,𝑦)∈Ω 𝑦=𝑦 (75)
(69)
𝑢𝑘 (𝑥, 𝑦) = 𝑢𝑘 (𝑥, 𝑦) + 𝑇𝑘 (𝑝) (𝑥, 𝑦) ,
where | ⋅ | denotes a norm in R𝑛 .
We recall the following results which will play an impor- where, for each 𝑘 = 1, 2, . . . , 𝑛,
tant role in the proofs of our results.
𝑇𝑘 (𝑝) (𝑥, 𝑦) := 𝑎0𝑘 (𝑥, 𝑦) 𝑝 (𝑥, 𝑦)
Theorem 26 (Green’s theorem [45]). If the functions 𝑀 and 𝜕
𝑁 are continuous and have continuous partial delta derivatives + 𝑎1𝑘 (𝑥, 𝑦) 𝑝 (𝜌1 (𝑥) , 𝑦)
𝜕𝑀/Δ 2 𝑦 and 𝜕𝑁/Δ 1 𝑥 on Ω ∪ Γ, then Δ 1𝑥 (76)
𝜕
𝜕𝑁 𝜕𝑀 + 𝑎2𝑘 (𝑥, 𝑦) 𝑝 (𝑥, 𝜌2 (𝑦)) ,
∫∫ ( − ) Δ 1 𝑥Δ 2 𝑦 = ∫ 𝑀𝑑∗ 𝑥 + 𝑁𝑑∗ 𝑦, Δ 2𝑦
Ω Δ 1𝑥 Δ 2𝑦 Γ
(70)
𝑎0 , 𝑎1 , and 𝑎2 are 𝐶1 functions and we assume that 𝑝 has
where the “star line integrals” on the right side in (70) denote continuous partial delta derivatives of the first and second
the sum of line delta integrals taken over the line segment order.
constituents of Γ directed to the right or upwards and line nabla
integrals taken over the line segment constituents of Γ directed Definition 29. Functional L is invariant under transforma-
to the left or downwards. tions (75) if, and only if, for all 𝑢 ∈ D, we have
Lemma 27 (fundamental lemma of the double variational ∫ ∫ 𝐿 (𝑥, 𝑦, 𝑢 (𝜎1 (𝑥) , 𝜎2 (𝑦)) , 𝑢Δ 1 (𝑥, 𝜎2 (𝑦)) ,
calculus [44]). If 𝑀 is continuous on Ω ∪ Γ with Ω
The following lemmas will be used in the proof of By Theorem 30, we obtain
Theorem 34. 𝑛
𝜕𝐿 𝜕 𝜕𝐿
∑ ∫∫ ( ⋅ 𝑇𝑘 (𝑝𝜎 ) − ⋅ 𝑇𝑘 (𝑝𝜎 )
Lemma 31. If L is invariant under transformations (75), then 𝑘=1 Ω 𝜕𝑢𝑘𝜎 Δ 1 𝑥 𝜕𝑢Δ 1
𝑘
𝑛 (86)
∑ ∫ ∫ 𝐸̂𝑘 (𝐿) ⋅ 𝑇𝑘 (𝑝𝜎 ) Δ 1 𝑥Δ 2 𝑦 = 0. (81) −
𝜕 𝜕𝐿
⋅ 𝑇𝑘 (𝑝𝜎 )) Δ 1 𝑥Δ 2 𝑦 = 0,
𝑘=1 Ω Δ 2 𝑦 𝜕𝑢Δ 2
𝑘
Proof. Fix 𝑖 ∈ {1, 2, . . . , 𝑛}. Observe that
which proves that ∑𝑛𝑘=1 ∫ ∫Ω 𝐸̂𝑘 (𝐿) ⋅ 𝑇𝑘 (𝑝𝜎 )Δ 1 𝑥Δ 2 𝑦 = 0.
𝜕𝐿 𝜕 𝑖 𝜎2 𝜕𝐿 𝜕 𝑖 𝜎1 Lemma 32. For each 𝑘 = 1, 2, . . . , 𝑛,
∫∫ ( Δ
𝑇 (𝑝 ) + Δ 𝑇 (𝑝 )) Δ 1 𝑥Δ 2 𝑦
Ω 𝜕𝑢𝑖 1 Δ 1𝑥 𝜕𝑢𝑖 2 Δ 2 𝑦
∫ ∫ 𝑞 ⋅ 𝑇𝑘 (𝑝𝜎 ) Δ 1 𝑥Δ 2 𝑦
Ω
𝜕 𝜕𝐿
= ∫∫ [ ( ⋅ 𝑇𝑖 (𝑝𝜎2 ))
Δ 1 𝑥 𝜕𝑢Δ 1 𝜕 𝜕
Ω
𝑖 = ∫ ∫ (𝑞𝑎0𝑘 − (𝑞𝑎1𝑘 ) − (𝑞𝑎2𝑘 )) ⋅ 𝑝𝜎 Δ 1 𝑥Δ 2 𝑦
Ω Δ 1𝑥 Δ 2𝑦
𝜕 𝜕𝐿 (87)
+ ( ⋅ 𝑇𝑖 (𝑝𝜎1 ))] Δ 1 𝑥Δ 2 𝑦
Δ 2 𝑦 𝜕𝑢Δ 2 holds.
𝑖
Using Green’s theorem, we can conclude that Corollary 35 (classical Noether’s second theorem for double
integrals problems, cf. [1]). Let Ω ⊆ R2 be an 𝜔-type set and
𝜕 let Γ be its positive fence. Let 𝐿(𝑥, 𝑦, 𝑢, 𝑝, 𝑞) be a function of
∫∫ [ (𝑞𝑎1𝑘 (𝑥, 𝑦) 𝑝 (𝑥, 𝜎2 (𝑦)))
Ω Δ 1𝑥 class 𝐶2 , (𝑥, 𝑦) ∈ Ω ∪ Γ, 𝑢 = (𝑢1 , 𝑢2 , . . . , 𝑢𝑛 ). If functional L
defined by
𝜕
+ (𝑞𝑎2𝑘 (𝑥, 𝑦) 𝑝 (𝜎1 (𝑥) , 𝑦))] Δ 1 𝑥Δ 2 𝑦 = 0.
Δ 2𝑦 𝜕𝑢 𝜕𝑢
(90) L [𝑦] = ∫ ∫ 𝐿 (𝑥, 𝑦, 𝑢 (𝑥, 𝑦) , (𝑥, 𝑦) , (𝑥, 𝑦)) 𝑑𝑥 𝑑𝑦
Ω 𝜕𝑥 𝜕𝑦
(96)
Hence,
is invariant under transformations (75) (where 𝜌1 and 𝜌2
∫ ∫ 𝑞 ⋅ 𝑇𝑘 (𝑝𝜎 ) Δ 1 𝑥Δ 2 𝑦 denote in this context the identity function, and Δ 1 and Δ 2
Ω denote the usual derivative), then
𝜕
= ∫ ∫ [𝑞𝑎0𝑘 ⋅ 𝑝𝜎 − (𝑞𝑎1𝑘 ) ⋅ 𝑝𝜎 (91) 𝑛
Ω Δ 1𝑥 ̃ 𝑘 (𝐸̂𝑘 (𝐿)) ≡ 0 𝑜𝑛 Ω,
∑𝑇 (97)
𝑘=1
𝜕
− (𝑞𝑎2𝑘 ) ⋅ 𝑝𝜎 ] Δ 1 𝑥Δ 2 𝑦
Δ 2𝑦 where
𝜕𝐿 𝜕 𝜕𝐿 𝜕 𝜕𝐿
proving the desired result. 𝐸̂𝑘 (𝐿) := − − , 𝑘 = 1, 2, . . . , 𝑛 (98)
𝜕𝑢𝑘 𝜕𝑥 𝜕𝑝𝑘 𝜕𝑦 𝜕𝑞𝑘
Remark 33. Lemma 32 shows that we can define an adjoint
̃ 𝑘 , by
operator of 𝑇𝑘 , 𝑇 ̃ 𝑘 is the adjoint operator of 𝑇𝑘 .
and 𝑇
𝑛 𝜕𝐴 2
(𝑡 , 𝜎 (𝑡 ) , 𝜎2 (𝑡2 ) , 𝜎3 (𝑡3 )) ,
̃ 𝑘 (𝐸̂𝑘 (𝐿)) ≡ 0
∑𝑇 on Ω , ∘
(95) Δ 0 𝑡0 0 1 1
𝑘=1
𝜕𝐴 3
(𝑡 , 𝜎 (𝑡 ) , 𝜎2 (𝑡2 ) , 𝜎3 (𝑡3 ))) ,
proving the desired result. Δ 0 𝑡0 0 1 1
12 Abstract and Applied Analysis
𝜕𝐴 3 order and that 𝐴 0 and the vector field A satisfy the so-called
curl A (𝑡) := ( (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝑡2 , 𝜎3 (𝑡3 ))
Δ 2 𝑡2 0 0 Lorentz conditions on time scales:
𝜕𝐴 2 𝜕𝐴 0
− (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝜎2 (𝑡2 ) , 𝑡3 ) , div A|(𝑡0 ,𝜎1 (𝑡1 ),𝜎2 (𝑡2 ),𝜎3 (𝑡3 )) =
Δ 3 𝑡3 0 0 Δ 0 𝑡0 (𝑡0 ,𝜎1 (𝑡1 ),𝜎2 (𝑡2 ),𝜎3 (𝑡3 ))
𝜕𝐴 1 𝜕𝐴 0
(𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝜎2 (𝑡2 ) , 𝑡3 ) div A|(𝜎0 (𝑡0 ),𝑡1 ,𝜎2 (𝑡2 ),𝜎3 (𝑡3 )) =
Δ 3 𝑡3 0 0 Δ 0 𝑡0 (𝜎0 (𝑡0 ),𝑡1 ,𝜎2 (𝑡2 ),𝜎3 (𝑡3 ))
(106)
−
𝜕𝐴 3
(𝜎 (𝑡 ) , 𝑡 , 𝜎 (𝑡 ) , 𝜎3 (𝑡3 )) , 𝜕𝐴 0
Δ 1 𝑡1 0 0 1 2 2 div A|(𝜎0 (𝑡0 ),𝜎1 (𝑡1 ),𝑡2 ,𝜎3 (𝑡3 )) =
Δ 0 𝑡0 (𝜎0 (𝑡0 ),𝜎1 (𝑡1 ),𝑡2 ,𝜎3 (𝑡3 ))
𝜕𝐴 2
(𝜎 (𝑡 ) , 𝑡 , 𝜎 (𝑡 ) , 𝜎3 (𝑡3 )) 𝜕𝐴 0
Δ 1 𝑡1 0 0 1 2 2 div A|(𝜎0 (𝑡0 ),𝜎1 (𝑡1 ),𝜎2 (𝑡2 ),𝑡3 ) = ,
Δ 0 𝑡0 (𝜎0 (𝑡0 ),𝜎1 (𝑡1 ),𝜎2 (𝑡2 ),𝑡3 )
𝜕𝐴 1
− (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝑡2 , 𝜎3 (𝑡3 ))) . where div A denotes the divergence of a vector field A, that is,
Δ 2 𝑡2 0 0
(99) 𝜕𝐴 1 𝜕𝐴 2 𝜕𝐴 3
div A := + + , (107)
Δ 1 𝑡1 Δ 2 𝑡2 Δ 3 𝑡3
We will consider the following Lagrangian function:
then the Euler-Lagrange expressions can be written in the
1 𝜕A 2 1 following way:
𝐿 = ∇𝐴 0 − − ‖curl A‖
2
(100)
2 Δ 0 𝑡0 2
𝜕2 𝐴 𝑘
𝐸̂𝑘 (𝐿) = (𝑡 , 𝜎 (𝑡 ) , 𝜎2 (𝑡2 ) , 𝜎3 (𝑡3 ))
that is the time scale version of the Lagrangian density for the Δ 0 𝑡02 0 1 1 (108)
electromagnetic field (see, e.g., [46]). 2
It can be proved that the functional − ∇ 𝐴 𝑘 (𝑡0 , 𝑡1 , 𝑡2 , 𝑡3 ) , 𝑘 = 0, 1, 2, 3,
where
L = ∫ ⋅ ⋅ ⋅ ∫ 𝐿 Δ 0Δ 1Δ 2Δ 3 (101)
Ω 𝜕2 𝐴 𝑘
∇2 𝐴 𝑘 (𝑡) := (𝜎 (𝑡 ) , 𝑡 , 𝜎 (𝑡 ) , 𝜎3 (𝑡3 ))
is invariant under the gauge transformations: Δ 1 𝑡12 0 0 1 2 2
𝜕 𝜌𝑘 𝜕2 𝐴 𝑘
𝐴𝑘 = 𝐴 𝑘 + 𝑝 , 𝑘 = 0, 1, 2, 3, + (𝜎 (𝑡 ) , 𝜎1 (𝑡1 ) , 𝑡2 , 𝜎3 (𝑡3 )) (109)
Δ 𝑘 𝑡𝑘
(102) Δ 2 𝑡22 0 0
̃ 𝑘 (𝑞) = − 𝜕 𝑞.
𝑇 (104)
Δ 𝑘 𝑡𝑘 5. Concluding Remarks
We proved that the important Noether’s second theorem is
Hence, from the second Noether theorem (Theorem 34), we
valid not only for the continuous and discrete calculus but
get
also for the quantum calculus. Moreover, in our opinion, the
3 proofs presented in this paper are elegant and clear to follow.
𝜕 ̂
∑ 𝐸 (𝐿) ≡ 0 on Ω∘ , (105) The question of obtaining Noether’s second theorem for
Δ 𝑡 𝑘
𝑘=0 𝑘 𝑘 multiple integrals with transformation of time in the time
scale setting remains an interesting open question. To the
where 𝐸̂𝑘 (𝐿), 𝑘 = 0, 1, 2, 3 are the Euler-Lagrange expressions best of the authors’ knowledge, to extend the second Noether
associated to functional L. theorem to multiple integrals with transformation of time,
If we suppose that, for each 𝑘 = 0, 1, 2, 3, 𝐴 𝑘 has substitution in the multiple integral is a fundamental tool and
continuous partial delta derivatives of the first and second this result is not yet available in the literature.
Abstract and Applied Analysis 13
For other generalizations of the second Noether theorem, [12] G. Bangerezako, “Variational 𝑞-calculus,” Journal of Mathemati-
we refer the reader to [47] (in the context of the fractional cal Analysis and Applications, vol. 289, no. 2, pp. 650–665, 2004.
calculus of variations) and [48] (in the context of the optimal [13] J. Baoguo, L. Erbe, and A. Peterson, “Oscillation of a family of 𝑞-
control). difference equations,” Applied Mathematics Letters, vol. 22, no.
6, pp. 871–875, 2009.
[14] A. M. C. Brito da Cruz, N. Martins, and D. F. M. Torres, “Higher-
Acknowledgments order Hahn’s quantum variational calculus,” Nonlinear Analysis:
Theory, Methods & Applications, vol. 75, no. 3, pp. 1147–1157,
This work was supported by FEDER funds through COM- 2012.
PETE-Operational Programme Factors of Competitiveness
[15] A. M. C. Brito da Cruz and N. Martins, “The 𝑞-symmetric vari-
(“Programa Operacional Factores de Competitividade”) and ational calculus,” Computers & Mathematics with Applications,
by Portuguese funds through the Center for Research and vol. 64, no. 7, pp. 2241–2250, 2012.
Development in Mathematics and Applications (University [16] A. M. C. Brito da Cruz, N. Martins, and D. F. M. Torres, “Hahn’s
of Aveiro) and the Portuguese Foundation for Science and symmetric quantum variational calculus,” Numerical Algebra,
Technology (“FCT-Fundação para a Ciência e a Tecnologia”), Control and Optimization, vol. 3, no. 1, pp. 77–94, 2013.
within Project PEst-C/MAT/UI4106/2011 with COMPETE [17] J. Cresson, G. S. F. Frederico, and D. F. M. Torres, “Constants of
no. FCOMP-01-0124-FEDER-022690. Agnieszka B. Malino- motion for non-differentiable quantum variational problems,”
wska was also supported by Bialystok University of Technol- Topological Methods in Nonlinear Analysis, vol. 33, no. 2, pp. 217–
ogy Grant S/WI/02/2011. 231, 2009.
[18] A. B. Malinowska and N. Martins, “Generalized transversality
conditions for the Hahn quantum variational calculus,” Opti-
References mization, vol. 62, no. 3, pp. 323–344, 2013.
[1] E. Noether, “Invariante variationsprobleme,” Nachrichten von [19] A. B. Malinowska and D. F. M. Torres, “The Hahn quantum
der Gesellschaft der Wissenschaften zu Göttingen, pp. 235–257, variational calculus,” Journal of Optimization Theory and Appli-
1918. cations, vol. 147, no. 3, pp. 419–442, 2010.
[2] Y. Kosmann-Schwarzbach, The Noether Theorems: Invariance [20] N. Martins and D. F. M. Torres, “Higher-order infinite horizon
and Conservation Laws in the Twentieth Century, Sources and variational problems in discrete quantum calculus,” Computers
Studies in the History of Mathematics and Physical Sciences, & Mathematics with Applications, vol. 64, no. 7, pp. 2166–2175,
Springer, New York, NY, USA, 2011. 2012.
[3] T. M. Atanacković, S. Konjik, S. Pilipović, and S. Simić, “Vari- [21] V. Kac and P. Cheung, Quantum Calculus, Springer, New York,
ational problems with fractional derivatives: invariance condi- NY, USA, 2002.
tions and Noether’s theorem,” Nonlinear Analysis: Theory, Meth- [22] S. Hilger, Ein maÿkettenkalkül mit anwendung auf zentrums-
ods & Applications, vol. 71, no. 5-6, pp. 1504–1517, 2009. mannigfaltigkeiten [Ph.D. thesis], Universität Würzburg, Wür-
zburg, Germany, 1988.
[4] Z. Bartosiewicz and D. F. M. Torres, “Noether’s theorem on time
scales,” Journal of Mathematical Analysis and Applications, vol. [23] M. Bohner, “Calculus of variations on time scales,” Dynamic
342, no. 2, pp. 1220–1226, 2008. Systems and Applications, vol. 13, no. 3-4, pp. 339–349, 2004.
[5] G. S. F. Frederico and D. F. M. Torres, “A formulation of Noe- [24] R. Hilscher and V. Zeidan, “Calculus of variations on time sca-
1
ther’s theorem for fractional problems of the calculus of vari- les: weak local piecewise 𝐶𝑟𝑑 solutions with variable endpoints,”
ations,” Journal of Mathematical Analysis and Applications, vol. Journal of Mathematical Analysis and Applications, vol. 289, no.
334, no. 2, pp. 834–846, 2007. 1, pp. 143–166, 2004.
[25] R. Almeida and D. F. M. Torres, “Isoperimetric problems on
[6] G. S. F. Frederico and D. F. M. Torres, “Noether’s symmetry
time scales with nabla derivatives,” Journal of Vibration and
theorem for variational and optimal control problems with time
Control, vol. 15, no. 6, pp. 951–958, 2009.
delay,” Numerical Algebra, Control and Optimization, vol. 2, no.
3, pp. 619–630, 2012. [26] M. J. Bohner, R. A. C. Ferreira, and D. F. M. Torres, “Integral
inequalities and their applications to the calculus of variations
[7] A. B. Malinowska, “A formulation of the fractional Noether- on time scales,” Mathematical Inequalities & Applications, vol.
type theorem for multidimensional Lagrangians,” Applied 13, no. 3, pp. 511–522, 2010.
Mathematics Letters, vol. 25, no. 11, pp. 1941–1946, 2012.
[27] R. A. C. Ferreira and D. F. M. Torres, “Higher-order calculus of
[8] A. B. Malinowska and D. F. M. Torres, Introduction to the Frac- variations on time scales,” in Mathematical Control Theory and
tional Calculus of Variations, Imperial College Press, London, Finance, pp. 149–159, Springer, Berlin, Germany, 2008.
UK, 2012.
[28] R. A. C. Ferreira and D. F. M. Torres, “Isoperimetric problems of
[9] N. Martins and D. F. M. Torres, “Noether’s symmetry theorem the calculus of variations on time scales,” in Nonlinear Analysis
for nabla problems of the calculus of variations,” Applied Math- and Optimization II. Optimization, vol. 514 of Contemporary
ematics Letters, vol. 23, no. 12, pp. 1432–1438, 2010. Mathematics, pp. 123–131, American Mathematical Society,
[10] J. D. Logan, “On variational problems which admit an infinite Providence, RI, USA, 2010.
continuous group,” Yokohama Mathematical Journal, vol. 22, pp. [29] E. Girejko, A. B. Malinowska, and D. F. M. Torres, “The con-
31–42, 1974. tingent epiderivative and the calculus of variations on time
[11] K. A. Aldwoah, A. B. Malinowska, and D. F. M. Torres, “The scales,” Optimization, vol. 61, no. 3, pp. 251–264, 2012.
power quantum calculus and variational problems,” Dynamics [30] A. B. Malinowska, N. Martins, and D. F. M. Torres, “Transversal-
of Continuous, Discrete and Impulsive Systems B, vol. 19, no. 1-2, ity conditions for infinite horizon variational problems on time
pp. 93–116, 2012. scales,” Optimization Letters, vol. 5, no. 1, pp. 41–53, 2011.
14 Abstract and Applied Analysis
Research Article
On Solutions of Linear Fractional Differential
Equations with Uncertainty
Copyright © 2013 T. Allahviranloo et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The solutions of linear fuzzy fractional differential equations (FFDEs) under the Caputo differentiability have been investigated.
To this end, the fuzzy Laplace transform was used to obtain the solutions of FFDEs. Then, some new results regarding the relation
between some types of differentiability have been obtained. Finally, some applicable examples are solved in order to show the ability
of the proposed method.
the damped harmonic oscillator, relaxation equation, all in According to Zadeh’s extension principle, operation of
the fuzzy fractional versions by using fuzzy laplace trans- addition on E is defined by
forms; so we should first implement analytical method to
solve it; then numerical methods can be applied. To this end, (𝑢 ⊕ V) (𝑥) = sup min {𝑢 (𝑦) , V (𝑥 − 𝑦)} , 𝑥 ∈ R, (1)
we adopt fractional Green’s functions to solve FFDEs by using 𝑦∈R
(1) 𝑢(𝑟) is a bounded nondecreasing left continuous 2.1. Caputo’s H-Differentiability. In this section, the concept
function in (0, 1]; and right continuous at 0; of the fuzzy Caputo derivatives has been reviewed [33–
35]. Also, we denote by 𝐶F [𝑎, 𝑏] a space of all fuzzy-valued
(2) 𝑢(𝑟) is a bounded nonincreasing left continuous functions which are continuous on [𝑎, 𝑏]. Also, we denote the
function in (0, 1]; and right continuous at 0; space of all Lebesgue integrable fuzzy-value functions on the
(3) 𝑢(𝑟) ≤ 𝑢(𝑟), 0 ≤ 𝑟 ≤ 1. bounded interval [𝑎, 𝑏] ⊂ R by 𝐿F [𝑎, 𝑏], and we denote the
space of fuzzy-value functions 𝑓(𝑥) which have continuous
Moreover, we also can present the 𝑟-cut representation of H-derivative up to order 𝑛 − 1 on [𝑎, 𝑏] such that 𝑓(𝑛−1) (𝑥) ∈
fuzzy number as [𝑢]𝑟 = [𝑢(𝑟), 𝑢(𝑟)] for all 0 ≤ 𝑟 ≤ 1. 𝐴𝐶F ([𝑎, 𝑏]) by 𝐴𝐶(𝑛)F ([𝑎, 𝑏]).
Abstract and Applied Analysis 3
Definition 4. Let 𝑓 ∈ 𝐶F [𝑎, 𝑏] ∩ 𝐿F [𝑎, 𝑏]; the fuzzy Riemann- Theorem 7. Let 𝑓(𝑥) ∈ 𝐶(𝑛)F [𝑎, 𝑏] ∩ 𝐿F [𝑎, 𝑏], 𝑥0 ∈ (𝑎, 𝑏),
Liouville integral of fuzzy-valued function 𝑓 is defined as (𝛽 > 0), and (𝑛 = [𝛽] + 1) such that for all 0 ≤ 𝑟 ≤ 1, then
follows: (i) if 𝑓(𝑥) is a 𝑅𝐿 [(𝑖) − 𝛽]-differentiable fuzzy-valued
𝛽 1 𝑥 𝑓
(𝑡) 𝑑𝑡 function, then
(𝐼𝑎+ 𝑓) (𝑥) = ∫ , 𝑥 > 𝑎, 0 < 𝛽 < 1.
Γ (𝛽) 𝑎 (𝑥 − 𝑡)1−𝛽
(5) 𝑅𝐿 𝛽 𝑅𝐿 𝛽 𝛽
( 𝐷𝑎+ 𝑓) (𝑥0 ; 𝑟) = [ 𝐷𝑎+ 𝑓 (𝑥0 , 𝑟) ,𝑅𝐿 𝐷𝑎+ 𝑓 (𝑥0 , 𝑟)] ;
Since 𝑓(𝑥; 𝑟) = [𝑓(𝑥; 𝑟), 𝑓(𝑥; 𝑟)], for all 0 ≤ 𝑟 ≤ 1, then one (10)
can indicate the fuzzy Riemann-Liouville integral of fuzzy-
valued function 𝑓 based on the lower and upper functions as
follows. (ii) if 𝑓(𝑥) is a 𝑅𝐿 [(𝑖𝑖)−𝛽]-differentiable fuzzy-valued func-
tion, then
Theorem 5. Let 𝑓(𝑥) ∈ 𝐶F [𝑎, 𝑏] ∩ 𝐿F [𝑎, 𝑏]; the fuzzy Riem-
ann-Liouville integral of fuzzy-valued function 𝑓 is defined by
𝛽 𝑅𝐿 𝛽 𝛽
𝛽 𝛽 𝛽 (𝑅𝐿 𝐷𝑎+ 𝑓) (𝑥0 ; 𝑟) = [ 𝐷𝑎+ 𝑓 (𝑥0 ; 𝑟) ,𝑅𝐿 𝐷𝑎+ 𝑓 (𝑥0 , 𝑟)] ,
(𝐼𝑎+ 𝑓) (𝑥; 𝑟) = [(𝐼𝑎+ 𝑓) (𝑥; 𝑟) , (𝐼𝑎+ 𝑓) (𝑥; 𝑟)] , 0 ≤ 𝑟 ≤ 1,
(11)
(6)
where
where
𝑥 𝑓 (𝑡; 𝑟) 𝑑𝑡
𝛽 1
(𝐼𝑎+ 𝑓) (𝑥; 𝑟) = ∫ , 0 ≤ 𝑟 ≤ 1,
Γ (𝛽) 𝑎 (𝑥 − 𝑡)1−𝛽
(7) 𝑅𝐿 𝛽 1 𝑑𝑛 𝑥 𝑓 (𝑡; 𝑟) 𝑑𝑡
( 𝐷𝑎+ 𝑓) (𝑥0 ; 𝑟) = [ ∫ ] ,
𝛽 1 𝑥 𝑓
(𝑡; 𝑟) 𝑑𝑡 Γ (𝑛 − 𝛽) 𝑑𝑥𝑛 𝑎 (𝑥 − 𝑡)𝛽−𝑛+1 𝑥=𝑥
(𝐼𝑎+ 𝑓) (𝑥; 𝑟) = ∫ , 0 ≤ 𝑟 ≤ 1. 0
Γ (𝛽) 𝑎 (𝑥 − 𝑡)1−𝛽
𝑅𝐿 𝛽 1 𝑑𝑛 𝑥 𝑓 (𝑡, 𝑟) 𝑑𝑡
Now, we define the fuzzy Riemann-Liouville fractional ( 𝐷𝑎+ 𝑓) (𝑥0 ; 𝑟) = [ ∫ ] .
Γ (𝑛 − 𝛽) 𝑑𝑥𝑛 𝑎 (𝑥 − 𝑡)𝛽−𝑛+1 𝑥=𝑥
derivatives of order 𝑛 − 1 < 𝛽 < 𝑛 for fuzzy-valued function 0
(12)
𝑓 (which is a direct extension of strongly generalized H-
differentiability [20] and lateral type of H-differentiability
[36] in the fractional literature) as follows.
Definition 8. Let 𝑓 ∈ 𝐶(𝑛)F [𝑎, 𝑏] ∩ 𝐿F1 [𝑎, 𝑏], and 𝑓(𝑛) is
𝑛
Definition 6. Let 𝑓 ∈ 𝐶 [𝑎, 𝑏] ∩ (𝑛)F
𝐿F1 [𝑎, 𝑏],
let 𝑥0 ∈ (𝑎, 𝑏), integrable; then the right fuzzy Caputo derivative of 𝑓 for
𝛽
𝑥
and let Φ(𝑥) = (1/Γ(𝑛 − 𝛽)) ∫𝑎 (𝑓(𝑡)𝑑𝑡/(𝑥 − 𝑡)𝛽−𝑛+1 ) (𝑛 = 𝑛 − 1 < 𝛽 < 𝑛 and 𝑥 ∈ [𝑎, 𝑏] is denoted by (𝐶𝐷𝑎+ 𝑓)(𝑥) ∈ E
[𝛽] + 1; 𝑥 > 𝑎). We say that 𝑓 is fuzzy Riemann-Liouville and defined by
fractional differentiable of order 𝛽, at 𝑥0 , if there exists an
𝛽
element (𝐷𝑎+ 𝑓)(𝑥0 ) ∈ E, such that for all ℎ > 0 sufficiently 𝛽 1 𝑥
small ( 𝐶𝐷𝑎+ 𝑓) (𝑥) = ⊙ ∫ (𝑥 − 𝑡)−𝛽+𝑛−1 ⊙ 𝑓(𝑛) (𝑡) 𝑑𝑡.
Γ (𝑛 − 𝛽) 𝑎
(13)
(i)
Theorem 12 (see [2]). Let 𝑓(𝑥), 𝑔(𝑥) be continuous-fuzzy- Since 𝐷(𝑘−1) 𝑓 is 𝐶[(i) − 𝛽], we get the following:
valued functions; suppose that 𝑐1 , 𝑐2 are constant; then
L [(𝑐1 ⊙ 𝑓 (𝑥)) ⊕ (𝑐2 ⊙ 𝑔 (𝑥))] 𝐶 𝐶 𝐶
𝛽 𝛽 𝛽
(25) ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) = [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)]
= (𝑐1 ⊙ L [𝑓 (𝑥)]) ⊕ (𝑐2 ⊙ L [𝑔 (𝑥)]) .
𝐶 𝛽 𝐶 𝛽
Lemma 13 (see [2]). Let 𝑓(𝑥) be continuous fuzzy-value = [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)] .
function on [0, ∞) and let 𝜆 ∈ R; then
(32)
L [𝜆 ⊙ 𝑓 (𝑥)] = 𝜆 ⊙ L [𝑓 (𝑥)] . (26)
Lemma 14 (see [2]). Let 𝑓(𝑥) be continuous fuzzy-value Hence, we get
function and 𝑔(𝑥) ≥ 0. Suppose that (𝑓(𝑥) ⊙ 𝑔(𝑥)) ⊙ 𝑒−𝑝𝑥 is
improper fuzzy Riemann-integrable on [0, ∞); then
∞ 𝐶 𝛽 𝐶 𝛽
∫ (𝑓 (𝑥) ⊙ 𝑔 (𝑥)) ⊙ 𝑒−𝑝𝑥 𝑑𝑥 ℓ [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)] = ℓ [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)]
0
∞ ∞
= 𝑝𝛽 ℓ [𝑓 (𝑥; 𝑟)]
= (∫ 𝑔 (𝑥) 𝑓 (𝑥, 𝑟) 𝑒−𝑝𝑥 𝑑𝑥, ∫ 𝑔 (𝑥) 𝑓 (𝑥, 𝑟) 𝑒−𝑝𝑥 𝑑𝑥) .
0 0
𝑛−1
(27) (𝑘)
− ( ∑ 𝑝𝛽−𝑘−1 𝑓 ) (0; 𝑟) ,
Theorem 15 (first translation theorem (see [2])). Let 𝑓 be 𝑘=0
continuous fuzzy value function and let L[𝑓(𝑥)] = 𝐹(𝑝). Then 𝐶 𝛽 𝐶 𝛽
𝑎𝑥 ℓ [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)] = ℓ [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)]
L [𝑒 ⊙ 𝑓 (𝑥)] = 𝐹 (𝑝 − 𝑎) , (28)
where 𝑒𝑎𝑥 is real value function. = 𝑝𝛽 ℓ [𝑓 (𝑥; 𝑟)]
Theorem 16 (derivative theorem). Suppose that 𝑓(𝑥) ∈ Then, we obtained the following:
𝐶(𝑛)F [0, ∞) ∩ 𝐿F [0, ∞); then
𝛽 𝑛−1
L [( 𝐶𝐷𝑎+ 𝑓) (𝑥)]
𝑝𝛽 L [𝑓 (𝑥)] ⊖ ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0)
(29) 𝑘=0
𝑛−1
𝛽 𝛽−𝑘−1 (𝑘)
= 𝑝 L [𝑓 (𝑡)] ⊖ ( ∑ 𝑝 𝑓 ) (0) 𝛽 𝛽
𝑘=0
= (ℓ [(𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟)] , ℓ [(𝐶𝐷𝑎+ 𝑓) (𝑥; 𝑟)]) .
(34)
if 𝐷(𝑘−1) 𝑓 is 𝐶[(𝑖) − 𝛽]-differentiable, and also
𝛽
L [( 𝐶𝐷𝑎+ 𝑓) (𝑥)] Then, we have
𝑛−1 (30)
= − ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0) ⊖ (−𝑝𝛽 L [𝑓 (𝑡)]) 𝑛−1
𝑘=0 𝑝𝛽 L [𝑓 (𝑥)] ⊖ ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0)
𝑘=0
if 𝐷(𝑘−1) 𝑓 is 𝐶[(𝑖𝑖) − 𝛽]-differentiable. (35)
𝐶 𝛽 𝐶 𝛽
Proof. For arbitrary fixed 𝑟 ∈ [0, 1] we have = L [( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) , ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟)] .
Now, we assume that 𝐷(𝑘−1) 𝑓 is the 𝐶[(ii) − 𝛽]; then for differential equations under Caputo’s H-differentiability. First
arbitrary fixed 𝑟 ∈ [0, 1] we have we consider the nonhomogeneous fuzzy fractional differen-
tial equation of order 0 < 𝛽 < 1 together with nonhomoge-
𝑛−1
neous initial condition
− ( ∑ 𝑝𝛽−𝑘−1 𝑓(𝑘) ) (0) ⊖ (−𝑝𝛽 L [𝑓 (𝑡)])
𝑘=0 2𝛽 2𝛽
(𝐶𝐷0+ Φ) (𝑥) ⊕ (𝐶𝐷0+ ) (𝑥)
= (−𝑓2 (𝑡; 𝑟) , −𝑓1 (𝑡; 𝑟)) ,
= 𝜑 [𝑥; Φ (𝑥)] , 0 < 𝛽 < 1, (42)
𝑛−1
𝛽−𝑘−1 (𝑘)
−𝑓2 (𝑡; 𝑟) = − ( ∑ 𝑝 𝑓 ) (0; 𝑟) + 𝑝𝛽 ℓ [𝑓 (𝑥; 𝑟)] , Φ (0) = 𝑓 (𝑥) ∈ E,
𝑘=0
where 𝐻1 (𝑝; 𝑟) and 𝐾1 (𝑝; 𝑟) are solutions of system (44). where 𝐻2 (𝑝; 𝑟) and 𝐾2 (𝑝; 𝑟) are solutions of system (49). Then
Then, by using the inverse Laplace transform, Φ(𝑥; 𝑟) and by making use of the inverse Laplace transform, Φ(𝑥; 𝑟) and
Φ(𝑥; 𝑟) are computed as follows: Φ(𝑥; 𝑟) are computed as follows:
Φ (𝑥, 𝑟) = ℓ−1 [𝐻1 (𝑝; 𝑟)] Φ (𝑥; 𝑟) = ℓ−1 [𝐻2 (𝑝; 𝑟)]
𝑡 = Φ (0; 𝑟) 𝐺2 (𝑡)
= Φ (0; 𝑟) 𝐺1 (𝑡) + ∫ 𝐺1 (𝑡; 𝜏) 𝜑 (𝑡 − 𝜏; 𝑟) 𝑑𝜏,
0 𝑡
+ ∫ 𝐺2 (𝑡; 𝜏) 𝜑 (𝑡 − 𝜏; 𝑟) 𝑑𝜏,
0 ≤ 𝑟 ≤ 1, 0
−1
Φ (𝑥; 𝑟) = ℓ [𝐾1 (𝑝; 𝑟)] 0 ≤ 𝑟 ≤ 1,
(52)
𝑡 Φ (𝑥; 𝑟) = ℓ−1 [𝐾2 (𝑝; 𝑟)]
= Φ (0; 𝑟) 𝐺1 (𝑡) + ∫ 𝐺1 (𝑡; 𝜏) 𝜑 (𝑡 − 𝜏; 𝑟) 𝑑𝜏,
0
= Φ (0; 𝑟) 𝐺2 (𝑡)
0 ≤ 𝑟 ≤ 1.
𝑡
(47) + ∫ 𝐺2 (𝑡; 𝜏) 𝜑 (𝑡 − 𝜏; 𝑟) 𝑑𝜏,
0
Case II. Let us consider that 𝐷Φ(𝑥) is 𝐶[(ii) − 𝛽]- 𝐺2 (𝑡 | 0) = ℓ−1 [Φ (𝑥; 𝑟)] . (53)
differentiable; then (43) can be written as follows:
5. Examples
ℓ [𝜑 (𝑥, Φ (𝑥) ; 𝑟)] = 𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝2𝛽−1 Φ (0, 𝑟)
In this section, we consider several examples, and we
𝛽 𝛽−1 find explicit expressions for fractional Green’s functions,
+ 𝑝 ℓ [Φ (𝑥, 𝑟)] − 𝑝 Φ (0, 𝑟) ,
including some versions of the fuzzy harmonic oscillator
ℓ [𝜑 (𝑥, Φ (𝑥) ; 𝑟)] = 𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝2𝛽−1 Φ (0, 𝑟) equations and fuzzy relaxation equation under Caputo’s H-
differentiability.
+ 𝑝𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝𝛽−1 Φ (0, 𝑟) ,
Example 1 (Green’s function for fuzzy damped harmonic
(49) oscillator). Let us examine the following FFDE:
where 𝐶 2𝛽 𝐶 𝛽
𝑎 ⊙ ( 𝐷0+ Φ) (𝑥) + 𝑏 ⊙ ( 𝐷0+ Φ) (𝑥)
𝜑 (𝑥, Φ (𝑥) ; 𝑟) = min {𝜑 (𝑥, 𝑢) | 𝑢 ∈ [Φ (𝑥; 𝑟) , Φ (𝑥; 𝑟)]} , (54)
= 𝜆𝑤2 ⊙ Φ (𝑥) + 𝜑 (𝑥) , 0 < 𝛽 < 1,
0 ≤ 𝑟 ≤ 1, Φ (0) ∈ E,
𝜑 (𝑥, Φ (𝑥) ; 𝑟) = max {𝜑 (𝑥, 𝑢) | 𝑢 ∈ [Φ (𝑥; 𝑟) , Φ (𝑥; 𝑟)]} ,
where 𝑎, 𝑏 are real positive constants and 𝑤 is the frequency of
0 ≤ 𝑟 ≤ 1. the harmonic oscillator. We will solve this example according
(50) to the two following cases for 𝜆 ∈ R.
Case I. Suppose that 𝜆 ∈ R+ = (0, +∞); then taking Laplace In the case 𝛽 = 1, we report
transform on the both sides of above equation, we have the
following: 𝐺1 (𝑥 | 0) = 𝑥𝐸2, 2 (𝜆𝑤2 𝑥2 ) (69)
𝐶 2𝛽
L [( 𝐷𝑎+ Φ) (𝑥)] = L (𝜆𝑤2𝛽 ⊙ Φ (𝑥) + 𝜑 (𝑥)) . (63) which is the classic Green’s function associated with fuzzy the
driven harmonic oscillator.
Using 𝐶[(i) − 𝛽]-differentiability, we obtain
Taking 𝜑(𝑥) = 0̃, we have a homogeneous fuzzy fractional
2𝛽
𝜆𝑤 ℓ [Φ (𝑥; 𝑟)] + ℓ [𝜑 (𝑥; 𝑟)] differential equation whose solution under 𝐶[(𝑖) − 𝛽]-H-
differentiability is given by
= 𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝2𝛽−1 Φ (0, 𝑟) ,
(64) Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸2𝛽 (𝜆𝑤2𝛽 𝑥2𝛽 ) , 0 ≤ 𝑟 ≤ 1,
2𝛽
𝜆𝑤 ℓ [Φ (𝑥; 𝑟)] + ℓ [𝜑 (𝑥; 𝑟)] (70)
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸2𝛽 (𝜆𝑤2𝛽 𝑥2𝛽 ) , 0 ≤ 𝑟 ≤ 1.
= 𝑝2𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑝2𝛽−1 Φ (0, 𝑟) .
For 𝛽 = 1, we get
Then, after some manipulations the following result is
reported: Φ (𝑥; 𝑟) = Φ (0; 𝑟) cos 𝑤𝑡, 0 ≤ 𝑟 ≤ 1,
(71)
(𝑝2𝛽 − 𝜆𝑤2𝛽 ) ℓ [Φ (𝑥; 𝑟)] = 𝑝2𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] , Φ (𝑥; 𝑟) = Φ (0; 𝑟) cos 𝑤𝑡, 0 ≤ 𝑟 ≤ 1.
(𝑝2𝛽 − 𝜆𝑤2𝛽 ) ℓ [Φ (𝑥; 𝑟)] = 𝑝2𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] . Case II. Let us suppose that 𝜆 ∈ R− = (−∞, 0); then using
𝐶
(65) [(ii)−𝛽]-differentiability and Theorem 16 and their inverses,
fractional Green’s function will be obtained similar to (68).
By taking inverse of Laplace on the both sides of (75) we have
Example 3 (Green’s functions for fuzzy fractional relaxation
𝑝2𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] equation). The next step is to consider the following FFDE:
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ℓ [−1 ],
𝑝2𝛽 − 𝜆𝑤2𝛽 𝛽
[ ] 𝑏 ⊙ ( 𝐶𝐷0+ Φ) (𝑥) = 𝜆𝑤2 ⊙ Φ (𝑥) + 𝜑 (𝑥) ,
0 ≤ 𝑟 ≤ 1,
0 < 𝛽 < 1, (72)
2𝛽−1
𝑝 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)]
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ℓ−1 [ ], Φ (0) ∈ E,
𝑝2𝛽 − 𝜆𝑤2𝛽
0 ≤ 𝑟 ≤ 1. where 𝑎 = 0 in (54) and 0 < 𝛽 < 1. We solve this example
(66) according to the two following cases corresponding to 𝜆 ∈ R.
Finally, we determine the solution of FFDE as follows: Case I. Suppose that 𝜆 ∈ R+ = (0, +∞); then taking Laplace
transform on the both sides of above equation, we have
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸2𝛽 (𝜆𝑤2𝛽 𝑥2𝛽 )
𝐶 𝛽
𝑥 L [𝑏 ⊙ ( 𝐷𝑎+ Φ) (𝑥)] = L [𝜆𝑤2 ⊙ Φ (𝑥) + 𝜑 (𝑥)] . (73)
2𝛽−1 2𝛽 2𝛽
+ ∫ 𝜑 (𝑡 − 𝑦; 𝑟) 𝑦 𝐸2𝛽, 2𝛽 (𝜆𝑤 𝑦 ) 𝑑𝑦,
0
Using 𝐶[(i) − 𝛽]-differentiability, we get
0 ≤ 𝑟 ≤ 1,
𝜆𝑤𝛽 ℓ [Φ (𝑥; 𝑟)] + ℓ [𝜑 (𝑥; 𝑟)]
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸2𝛽 (𝜆𝑤2𝛽 𝑥2𝛽 )
𝑥 = 𝑏𝑝𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑏𝑝𝛽−1 Φ (0, 𝑟) ,
+ ∫ 𝜑 (𝑡 − 𝑦; 𝑟) 𝑦2𝛽−1 𝐸2𝛽, 2𝛽 (𝜆𝑤2𝛽 𝑦2𝛽 ) 𝑑𝑦, (74)
0 𝜆𝑤𝛽 ℓ [Φ (𝑥; 𝑟)] + ℓ [𝜑 (𝑥; 𝑟)]
0 ≤ 𝑟 ≤ 1.
(67) = 𝑏𝑝𝛽 ℓ [Φ (𝑥, 𝑟)] − 𝑏𝑝𝛽−1 Φ (0, 𝑟) .
To show the corresponding Green’s function under 𝐶[(i) − Then, after some manipulations we get the following:
𝛽]-differentiability, which is solution of (62) satisfying the
condition Φ(0) = 0̃ and taking 𝜑(𝑥) = 𝛿(𝑥), we use the (𝑏𝑝𝛽 − 𝜆𝑤2 ) ℓ [Φ (𝑥; 𝑟)] = 𝑏𝑝𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] ,
linearity of the inverse Laplace transform to obtain
(𝑏𝑝𝛽 − 𝜆𝑤2 ) ℓ [Φ (𝑥; 𝑟)] = 𝑏𝑝𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] .
2𝛽−1 2𝛽 2𝛽
𝐺1 (𝑥 | 0) ≡ 𝑥 𝐸2𝛽, 2𝛽 (𝜆𝑤 𝑥 ) . (68) (75)
10 Abstract and Applied Analysis
By taking inverse of Laplace on the both sides of (75) we have Case II. Suppose that 𝜆 ∈ R− = (−∞, 0); then using
𝐶
[(ii)−𝛽]-differentiability and Theorem 16, and their inverses,
𝑏𝑝𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)] fractional Green’s function will be obtained similar to (78).
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ℓ−1 [ ],
𝑏𝑝𝛽 − 𝜆𝑤2
[ ] Now, we provide some examples such that the obtained
0 ≤ 𝑟 ≤ 1, solutions coincide with the previously reported solution in
integer case [37].
𝑏𝑝𝛽−1 Φ (0, 𝑟) + ℓ [𝜑 (𝑥; 𝑟)]
Φ (𝑥; 𝑟) = Φ (0; 𝑟) ℓ−1 [ ], Example 4. For a special case, let us consider the following
𝑏𝑝𝛽 − 𝜆𝑤2 FFDE:
𝐶
0 ≤ 𝑟 ≤ 1. 𝛽
( 𝐷0+ Φ) (𝑥) = 𝜆 ⊙ Φ (𝑥) , 0 < 𝛽, 𝑥 < 1,
(76) (82)
Thus, we determine the solution of FFDE as follows: Φ (0) ∈ E,
2
𝜆𝑤 𝛽 where 𝑦 represents the number of radionuclides present in a
Φ (𝑥; 𝑟) = Φ (0; 𝑟) 𝐸𝛽 ( 𝑥 ) given radioactive and 𝜆 is a decay constant.
𝑏
9 1020
8
1015
7
6
1010
Log(y(x))
y(x)
4 105
3
100
2
1 10−5
0 0.2 0.4 0.6 0.8 1 10−30 10−25 10−20 10−15 10−10 10−5 100
x
Log(x)
Figure 1: Solution of Example 4, Case I, 𝛽 = 1.
Figure 4: Solution of Example 4, Case II, 𝛽 = 0.5.
1016 3
14
10 2.8
2.6
1012
2.4
1010
2.2
Log(y(x))
108
y(x)
2
6
10 1.8
104 1.6
1.4
102
1.2
100 1
10−30 10−25 10−20 10−15 10−10 10−5 100 0 0.2 0.4 0.6 0.8 1
Log(x) x
3 1020
2.5
1015
2
1010
Log(y(x))
y(x)
1.5
105
1
100
0.5
0 10−5
0 0.2 0.4 0.6 0.8 1 10−30 10−25 10−20 10−15 10−10 10−5 100
x Log(x)
Caputo’s differentiability were applied. Some particular cases [15] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory
were also presented, involving some fuzzy versions of the and Applications of Fractional Differential Equations, vol. 204
harmonic oscillator. To this end, some needed concepts are of North-Holland Mathematics Studies, Elsevier Science B.V.,
given like the Caputo derivatives, fuzzy Laplace transforms, Amsterdam, The Netherland, 2006.
and their inverses, and so forth. For future research, we [16] V. Lakshmikantham, S. Leela, and J. Vasundhara Devi, Theory of
extend the boundary value problem for FDE [38] to the Fractional Dynamic Systems, Cambridge Scientific, Cambridge,
fractional case. UK, 2009.
[17] V. Lakshmikantham and R. N. Mohapatra, Theory of Fuzzy
Differential Equations and Inclusions, vol. 6 of Series in Math-
Acknowledgments ematical Analysis and Applications, Taylor & Francis, London,
UK, 2003.
The authors thank the referees for careful reading and helpful [18] V. Lakshmikantham and A. S. Vatsala, “Basic theory of frac-
suggestions on the improvement of the paper. tional differential equations,” Nonlinear Analysis: Theory, Meth-
ods & Applications, vol. 69, no. 8, pp. 2677–2682, 2008.
References [19] R. P. Agarwal, V. Lakshmikantham, and J. J. Nieto, “On the
concept of solution for fractional differential equations with
[1] A. Ahmadian, M. Suleiman, S. Salahshour, and D. Baleanu, “A uncertainty,” Nonlinear Analysis: Theory, Methods & Applica-
Jacobi operational matrix for solving a fuzzy linear fractional tions, vol. 72, no. 6, pp. 2859–2862, 2010.
differential equation,” Advances in Difference Equations, vol. [20] B. Bede and S. G. Gal, “Generalizations of the differentiability
2013, article 104, pp. 1–29, 2013. of fuzzy-number-valued functions with applications to fuzzy
[2] T. Allahviranloo and M. B. Ahmadi, “Fuzzy Laplace trans- differential equations,” Fuzzy Sets and Systems, vol. 151, no. 3,
forms,” Soft Computing, vol. 14, no. 3, pp. 235–243, 2010. pp. 581–599, 2005.
[3] T. Allahviranloo, S. Salahshour, and S. Abbasbandy, “Explicit [21] B. Bede, I. J. Rudas, and A. L. Bencsik, “First order linear
solutions of fractional differential equations with uncertainty,” fuzzy differential equations under generalized differentiability,”
Soft Computing, vol. 16, no. 2, pp. 297–302, 2012. Information Sciences, vol. 177, no. 7, pp. 1648–1662, 2007.
[4] T. Allahviranloo, S. Abbasbandy, S. Salahshour, and A. [22] I. Perfilieva, “Fuzzy transforms: theory and applications,” Fuzzy
Hakimzadeh, “A new method for solving fuzzy linear differen- Sets and Systems, vol. 157, no. 8, pp. 993–1023, 2006.
tial equations,” Computing, vol. 92, no. 2, pp. 181–197, 2011. [23] I. Perfilieva, H. De Meyer, B. De Baets, and D. Plšková, “Cauchy
[5] T. Allahviranloo and S. Salahshour, “A new approach for solving problem with fuzzy initial condition and its approximate
first order fuzzy differential equations,” Communications in solution with the help of fuzzy transform,” in Proceedings of
Computer and Information Science, vol. 81, part 5, part 6, pp. IEEE International Conference on Fuzzy Systems (FUZZ ’08), pp.
522–531, 2010. 2285–2290, June 2008.
[6] J. Li, A. Zhao, and J. Yan, “The Cauchy problem of fuzzy [24] S. Salahshour, T. Allahviranloo, S. Abbasbandy, and D. Baleanu,
differential equations under generalized differentiability,” Fuzzy “Existence and uniqueness results for fractional differential
Sets and Systems, vol. 200, pp. 1–24, 2012. equations with uncertainty,” Advances in Difference Equations,
[7] S. Salahshour and T. Allahviranloo, “Application of fuzzy dif- vol. 2012, article 112, pp. 1–12, 2012.
ferential transform method for solving fuzzy Volterra integral [25] M. Mazandarani and A. V. Kamyad, “Modified fractional Euler
equations,” Applied Mathematical Modelling, vol. 37, no. 3, pp. method for solving fuzzy fractional initial value problem,” Com-
1016–1027, 2013. munications in Nonlinear Science and Numerical Simulation, vol.
[8] S. Salahshour and A. Allahviranloo, “Applications of fuzzy 18, no. 1, pp. 12–21, 2013.
Laplace transforms,” Soft Computing, vol. 17, no. 1, pp. 145–178, [26] R. P. Agarwal, S. Arshad, D. O’Regan, and V. Lupulescu,
2013. “Fuzzy fractional integral equations under compactness type
[9] A. Arara, M. Benchohra, N. Hamidi, and J. J. Nieto, “Fractional condition,” Fractional Calculus and Applied Analysis, vol. 15, no.
order differential equations on an unbounded domain,” Nonlin- 4, pp. 572–590, 2012.
ear Analysis: Theory, Methods & Applications, vol. 72, no. 2, pp. [27] J. Xu, Z. Liao, and Z. Hu, “A class of linear differential dynamical
580–586, 2010. systems with fuzzy initial condition,” Fuzzy Sets and Systems,
[10] Y. I. Babenko, Heat and Mass Transfer, Chemia, Leningrad, vol. 158, no. 21, pp. 2339–2358, 2007.
Russia, 1986. [28] M. Friedman, M. Ma, and A. Kandel, “Numerical solutions
[11] R. L. Bagley, “On the fractional order initial value problem of fuzzy differential and integral equations,” Fuzzy Sets and
and its engineering applications,” in Fractional Calculus and Systems, vol. 106, no. 1, pp. 35–48, 1999.
Its Applications, K. Nishimoto, Ed., pp. 12–20, College of [29] M. Ma, M. Friedman, and A. Kandel, “Numerical solutions of
Engineering, Nihon University, Tokyo, Japan, 1990. fuzzy differential equations,” Fuzzy Sets and Systems, vol. 105,
[12] H. Beyer and S. Kempfle, “Definition of physically consistent no. 1, pp. 133–138, 1999.
damping laws with fractional derivatives,” Zeitschrift für Ange- [30] H.-J. Zimmermann, Fuzzy Set Theory and Its Applications,
wandte Mathematik und Mechanik, vol. 75, no. 8, pp. 623–635, Kluwer Academic Publishers, Boston, Mass, USA, 2nd edition,
1995. 1992.
[13] K. Diethelm and N. J. Ford, “Analysis of fractional differential [31] M. L. Puri and D. A. Ralescu, “Fuzzy random variables,” Journal
equations,” Journal of Mathematical Analysis and Applications, of Mathematical Analysis and Applications, vol. 114, no. 2, pp.
vol. 265, no. 2, pp. 229–248, 2002. 409–422, 1986.
[14] I. Podlubny, Fractional Differential Equations, vol. 198 of Math- [32] H.-C. Wu, “The improper fuzzy Riemann integral and its
ematics in Science and Engineering, Academic Press, San Diego, numerical integration,” Information Sciences, vol. 111, no. 1–4, pp.
Calif, USA, 1999. 109–137, 1998.
Abstract and Applied Analysis 13
Research Article
Fractional-Order Total Variation Image Restoration Based on
Primal-Dual Algorithm
Copyright © 2013 Dali Chen et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This paper proposes a fractional-order total variation image denoising algorithm based on the primal-dual method, which provides
a much more elegant and effective way of treating problems of the algorithm implementation, ill-posed inverse, convergence
rate, and blocky effect. The fractional-order total variation model is introduced by generalizing the first-order model, and the
corresponding saddle-point and dual formulation are constructed in theory. In order to guarantee 𝑂(1/𝑁2 ) convergence rate, the
primal-dual algorithm was used to solve the constructed saddle-point problem, and the final numerical procedure is given for
image denoising. Finally, the experimental results demonstrate that the proposed methodology avoids the blocky effect, achieves
state-of-the-art performance, and guarantees 𝑂(1/𝑁2 ) convergence rate.
1. Introduction It has been proved that this model is able to deal with
the blocky problem. However, it tends to cause the sign
Since the work of Rudin et al. [1], total variation (TV) of uplifting effect and formation of artifacts around edges.
minimization problems arise in many image processing For this problem, an improved fourth-order PDE model,
applications for regularizing inverse problems where one replacing the Laplacian operator of diffusivity function by the
expects the recovered image or signal to be piecewise constant gradient operator, was proposed in [8].
[2–5]. The typical total variation model, for example, ROF In this paper, our interest focuses on the second gen-
model, has been proved to be able to achieve a good tradeoff eralization which deals with fractional-order differentiation.
between edge preservation and noise removal [1]. However, Fractional calculus is a rapidly growing mathematical disci-
it tends to produce the so-called blocky (staircase) effects on pline, which provides an important tool for nonlocal field
the images because it favors a piecewise constant solution theories [9]. Recently, it has been greatly studied in computer
in bounded variation (BV) space [6]. In order to deal vision [10–16], and the main reason for this development
with blocky effects, the modification of TV model, which is the expectation that the use of this theory will lead to a
generalizes the differential order in regularization term, has much more elegant and effective way of treating problems of
aroused the more and more attentions of numerous scholars. blocky effect and detailed information protection. Specially,
The improved methods of TV model are divided into the fractional-order total variation (TV) models play an
two kinds: the high-order derivative and the fractional-order important role for image denoising, inpainting, and motion
derivative. The first one replaces the first-order derivative estimation [17–21]. So far, the methods adopted to deal
in regularization term by the high-order derivative. For with the fractional-order total variation problem are divided
example, a fourth-order partial differential equation-(PDE- into three kinds: (1) solving the associated Euler-Lagrange
based) denoising mode was proposed by [7], in which the equation, which is a nonlinear partial differential equation
regularized solution is obtained by solving the minimization (PDE) [17, 18]; (2) using the methods based on duality [6];
of potential function of second-order derivative of the image. (3) using the method based on majorization-minimization
2 Abstract and Applied Analysis
(MM) algorithm [22]. Although the convergence of these where √(𝐷1𝛼 𝑢)2 + (𝐷2𝛼 𝑢)2 is the discretization of |𝐷𝛼 𝑢|, 𝐷𝛼
methods has been studied in the corresponding literatures, is the fractional-order derivative operator, and 𝐷𝛼 𝑢 :=
the rate of convergence for them has not been discussed yet.
(𝐷1𝛼 𝑢, 𝐷2𝛼 𝑢)𝑇 . 𝐷1𝛼 and 𝐷2𝛼 are linear operators corresponding
For this problem, in this paper, a primal-dual algorithm is
to horizontal and vertical fractional-order derivative.
used to solve the fractional-order total variation problem,
From Grünwald-Letnikov fractional derivative definition
which is able to guarantee 𝑂(1/𝑁2 ) convergence rate.
[31], the finite fractional-order forward difference can be
The primal-dual algorithm was first presented by [23]
obtained by
and named as the primal-dual hybrid gradient (PDHG)
method in [24]. In this method, each iteration updates both
a primal and a dual variable. It is thus able to avoid some of 𝐾−1 𝐾−1
the difficulties that arise when working only on the primal 𝐷1𝛼 𝑢𝑖,𝑗 = ∑ 𝑤𝑘(𝛼) 𝑢𝑖+𝑘,𝑗 , 𝐷2𝛼 𝑢𝑖,𝑗 = ∑ 𝑤𝑘(𝛼) 𝑢𝑖,𝑗+𝑘 , (4)
or dual side [25, 26]. The convergence of the primal-dual 𝑘=0 𝑘=0
algorithm has been studied in [27, 28]. Recently a unified
form of primal-dual algorithm was presented by [29, 30],
which demonstrated that, in some case, these algorithms can where 𝑤𝑘(𝛼) = (−1)𝑘+1 C𝛼𝑘 , C𝛼𝑘 = Γ(𝛼 + 1)/(Γ(𝑘 + 1)Γ(𝛼 − 𝑘 + 1))
also achieve the 𝑂(1/𝑁2 ) rate of convergence. In our paper, denotes the generalized binomial coefficient, and Γ(𝑥) is the
a new image denoising method is proposed, in which the Gamma function. In addition, the coefficients 𝑤(𝛼) can also
primal-dual algorithm is used to solve the fractional-order be obtained recursively from
total variation denoising model. The proposed denoising
method is able to avoid the blocky effect, achieves state- 𝛼+1
of-the-art performance, and guarantees the 𝑂(1/𝑁2 ) rate of 𝑤0(𝛼) = −1, 𝑤𝑘(𝛼) = (1 − (𝛼)
) 𝑤𝑘−1 , 𝑘 = 1, 2, . . . .
𝑘
convergence. (5)
This paper is organized as follows. Section 1 introduces
prior work, focusing on the main problems with existing
methods that are addressed by our model. In Section 2, the When 𝛼 = 1, 𝑤𝑘1 = 0 for 𝑘 > 1 and (4) is the first-order
fractional-order total variation denoising model is described, forward derivative as usual.
and the corresponding saddle-point and dual formulation are To describe (2) in matrix algebra language, we reord
constructed in theory. Based on this, the primal-dual algo- the image matrix 𝑢 and 𝑓 row wisely into the vector 𝑥
rithm was used to solve the constructed saddle-point prob- and 𝑧, associating the (𝑖, 𝑗) element of the two-dimensional
lem, and the final numerical implementation is presented structure with the element (𝑗 − 1)𝑛 + 𝑖 of the vector structure,
for image denoising. Experimental evaluation is presented in 𝑥(𝑗−1)𝑛+𝑖 = 𝑢𝑖,𝑗 , and 𝑧(𝑗−1)𝑛+𝑖 = 𝑓𝑖,𝑗 . We have 𝑥 ∈ R𝑁
Section 3 and the paper is concluded in Section 4. and 𝑧 ∈ R𝑁, where 𝑁 = 𝑛2 . The (𝑖, 𝑗) component of the
fractional-order derivative 𝐷𝛼 𝑢 can thus be represented as a
2. Fractional-Order Total Variation Model and multiplication of the vector 𝑥 ∈ R𝑁 by a matrix 𝐴 𝑚 ∈ R2×𝑁,
Primal-Dual Algorithm for 𝑚 = 1, 2, . . . , 𝑁:
𝑇
2.1. Model Description. Let 𝑓𝑖,𝑗 = 𝑢𝑖,𝑗 + V𝑖,𝑗 denote the {
{
𝐾−1 𝐾−1
{
{ ( ∑ 𝑤𝑘(𝛼) 𝑥𝑚+𝑘 , ∑ 𝑤𝑘(𝛼) 𝑥𝑚+𝑛𝑘 , )
observed noisy image, where (𝑖, 𝑗)𝑇 denotes the location with {
{
{ 𝑘=0 𝑘=0
a rectangular image domain Ω ∈ R𝑛×𝑛 and V is white {
{ if (𝑚 mod 𝑛) ≠0 and 𝑚 ≤ 𝑁 − 𝑛
{
{
Gaussian noise. The typical total variation (TV) denoising {
{ 𝑇
{
{ 𝐾−1
model estimates the desired clean image 𝑢𝑖,𝑗 by solving the {
{(0, ∑ 𝑤 (𝛼)
𝑥 ) ,
{
{ 𝑘 𝑚+𝑛𝑘
following finite-dimensional optimization problem: {
{ 𝑘=0
𝐴 𝑚 𝑥 = {if (𝑚 mod 𝑛) = 0 and 𝑚 ≤ 𝑁 − 𝑛 (6)
𝜆 {
{ 𝑇
2 {
{
𝑢̂ = arg min {𝐸 (𝑢) := TV (𝑢) + 𝑓 − 𝑢2 } , (1) {
{ (
𝐾−1
∑ 𝑤(𝛼)
𝑥 , 0) ,
𝑢∈Ω 2 {
{ 𝑘 𝑚+𝑘
{
{ 𝑘=0
{
{
where TV(𝑢) is the regularization term, ‖𝑓 − 𝑢‖22 is the data {
{ if (𝑚 mod 𝑛) ≠0 and 𝑚 > 𝑁 − 𝑛
{
{
fidelity term, ‖𝑢‖] is ]-norm of 𝑢, and 𝜆 is regularization {
{(0, 0)𝑇 ,
{
{
{if (𝑚 mod 𝑛) = 0 and 𝑚 > 𝑁 − 𝑛.
parameter which controls the degree of smoothing. In this
paper, we consider a fractional-order total variation model,
defined as
From this definition, the discrete version of the prime
𝜆 2
𝑢̂ = arg min {𝐸 (𝑢) := TV𝛼 (𝑢) + 𝑓 − 𝑢2 } , (2) fractional-order total variation (FOTV) model (2) can be
𝑢∈Ω 2 written as
where TV𝛼 (𝑢) is obtained by the following formula:
𝑁
2
𝜆
2
TV𝛼 (𝑢) = √(𝐷1𝛼 𝑢) + (𝐷2𝛼 𝑢) , (3) min𝑁 {𝐸 (𝑥) := ∑ 𝐴 𝑚 𝑥2 + ‖𝑥 − 𝑧‖22 } . (7)
1 𝑥∈𝑅 𝑚=1 2
Abstract and Applied Analysis 3
Algorithm 1
in which case we can interpret 𝛿(⋅) as the indicator function which vanishes only if (𝑥, 𝑦) is itself the saddle point [29].
for the unit ball in the dual norm, and 𝑦 = (𝑦1 , 𝑦2 , . . . , 𝑦𝑁)𝑇 ∈ The primal-dual gap 𝐺(𝑥, 𝑦) is a measure of closeness of the
2 𝑇 primal-dual (𝑥, 𝑦) to the primal-dual solution, and we use it
𝑅2𝑁 with 𝑦𝑚 = (𝑦𝑚 1
, 𝑦𝑚 ) ∈ 𝑅2 . Since 𝐹 = 𝐹∗∗ if 𝐹 is closed
to design the stopping criterion for our numerical algorithm
and convex, we have
in this paper.
‖𝐴𝑥‖𝐸 = sup ⟨𝑦, 𝐴𝑥⟩ − 𝛿 (𝑦) . (10)
𝑦∈𝑅2𝑁
2.3. Primal-Dual Method for Solving Saddle-Point Problem.
Substituting (10) into (8), we can obtain the saddle-point The first-order primal-dual method summarized in [29]
formulation of FOTV model defined by for convex problem was applied in this paper to solve the
saddle-point problem described by (11), which is described
𝜆
min sup ⟨𝑦, 𝐴𝑥⟩ − 𝛿 (𝑦) + ‖𝑥 − 𝑧‖22 . (11) by Algorithm 1.
𝑥∈𝑅𝑁 𝑦∈𝑅2𝑁 2 Let 𝐿 = ‖𝐴‖ and 𝜏𝑡 𝜎𝑡 𝐿2 ≤ 1; then there exists a saddle-point
̂ 𝑦)
(𝑥, ̂ such that 𝑥𝑡 → 𝑥̂ and 𝑦𝑡 → 𝑦. ̂ Since (𝜆/2)‖𝑥 − 𝑧‖22
The minimization problem of (11) can be solved exactly as
is uniformly convex such that it has a Lipschitz continuous
1 𝑇 gradient, the output (𝑥𝑡 , 𝑦𝑡 ) possesses a rate of convergence of
𝑥=𝑧− 𝐴 𝑦. (12)
𝜆 𝑂(1/𝑁2 ) [29]. More specifically, we assume that the constant
stepsizes are used; that is, 𝜏𝑡 = 𝜏 > 0, 𝜎𝑡 = 𝜎 > 0, and 𝜃𝑡 = 𝜃
Substituting (12) into (11) yields the following dual problem:
for all 𝑡 ≥ 0. If 𝜏𝜎𝐿2 < 1; then the 𝑂(1/𝑁) convergence rate
2 can be guaranteed for 𝜃 = 1, and 𝑂(1/√𝑁) convergence rate
max − (𝐴𝑇 𝑦 − 𝜆𝑧2 + 𝛿 (𝑦)) , (13)
𝑦∈𝑅2𝑁 can be guaranteed for 𝜃 = 0 [30].
4 Abstract and Applied Analysis
To summarize, our entire noise removal algorithm in a iteration of our method performs the following updates on
form of a pseudocode is done in following: prime variable 𝑢 ∈ R𝑛×𝑛 and dual variable 𝑝 ∈ R𝑛×𝑛×2 :
(1) initialization: 𝑝𝑘 + 𝜎𝑘 𝐷𝛼 𝑢𝑘
𝑝𝑘+1 = , (19)
max {1, 𝑝𝑘 + 𝜎𝑘 𝐷𝛼 𝑢𝑘 }
(i) for a given noisy image 𝑓 of size 𝑛 × 𝑛, we reord
the image 𝑓 row wisely into the vector 𝑧 ∈ 𝑅𝑁 𝑢𝑘 − 𝜏𝑘 div𝛼 𝑝𝑘+1 + 𝜏𝑘 𝜆𝑓
with 𝑁 = 𝑛 × 𝑛, 𝑧(𝑗−1)𝑛−𝑖 = 𝑓𝑖,𝑗 ; 𝑢𝑘+1 = , (20)
1 + 𝜏𝑘 𝜆
(ii) compute the fractional-order discrete operator
𝐴 according to (6); where the dual variable 𝑝 = (𝑝1 , 𝑝2 )𝑇 corresponds with 𝑦
(iii) initialize the regularization parameter 𝜆, itera- and the discrete fractional divergence operator div𝛼 is adjoint
tion number 𝑇, and maximum permissible error to the discrete fractional gradient operator 𝐷𝛼 . Note that
𝐸mpe ; (19) and (20), respectively, correspond with the solutions of
(iv) set 𝑡 = 0, 𝑥0 = 𝑧, 𝐿 = ‖𝐴‖, 𝜏0 = 1/𝐿2 , 𝜎0 = 1/𝐿2 , problem (A) and (B). The updates of the other parameters
and 𝛾 = 0.7𝜆; are the same as the description in the foregoing pseudocode,
and the number of iteration is fixed to ensure the correct
timing sequence. After the iteration, the denoised image 𝑢 is
(2) iteration: compute 𝑥 by the following steps.
outputted, the next image frame is captured, and the same
Step 1. Solve the problem (A) by process is repeated until the assignment is over.
𝑡+1 𝑦𝑚𝑡
+ 𝜎𝑡 𝐴 𝑚 𝑥𝑡 3. Experiments and Analysis
𝑦𝑚 = 𝑡 , 𝑚 = 1, 2, . . . , 𝑁. (16)
max {1, 𝑦𝑚 + 𝜎𝑡 𝐴 𝑚 𝑥𝑡 2 }
3.1. Restraint of Block Effect. The blocky effect is the main
drawback produced by the typical first-order total variation
Step 2. Substitute 𝑦𝑡+1 into the (B) and solve the problem (B) denoising algorithm. In this section, some experiments are
by given to assess the capability of reducing block effect of our
proposed fractional-order total variation (FOTV) denoising
𝑥𝑡 − 𝜏𝑡 𝐴𝑇 𝑦𝑡+1 + 𝜏𝑡 𝜆𝑧 algorithm. Firstly, the one-dimensional signal is used as the
𝑥𝑡+1 = . (17)
1 + 𝜏𝑡 𝜆 test signal, 𝑦(𝑡) = 4 sin(2𝜋𝑡) + 8 sin(3𝜋𝑡). We contaminate
the given signal using the additive white Gaussian noise
Step 3. Compute 𝜃𝑡 = 1/√1 + 2𝛾𝜏𝑡 , 𝜏𝑡+1 = 𝜏𝑡 𝜃𝑡 , and 𝜎𝑡+1 = (AWGN) with standard deviation (SD) of 0.5. The corrupted
𝜎𝑡 /𝜃𝑡 . signal is shown in Figure 1(a). The proposed FOTV denoising
algorithm with 𝛼 = 1 and 𝛼 = 1.5 is used to process the
Step 4. Update 𝑥 by 𝑥𝑡+1 = (1 + 𝜃𝑡 )𝑥𝑡+1 − 𝜃𝑡 𝑥𝑡 . contaminated signal, respectively. When 𝛼 = 1, our FOTV
algorithm is the typical ROF denoising algorithm based on
Step 5. Compute the primal-dual gap 𝑔𝑡+1 according to (15) primal-dual method. In this experiment, we set 𝜆 = 0.01, and
as follows: the results at the 1000th iteration are, respectively, shown in
Figures 1(b) and 1(c). The difference is obvious: while the first-
𝑁
𝑇 order TV denoising algorithm approximates the observed
𝑔𝑡+1 = ∑ 𝐴 𝑚 𝑥𝑡+1 2 − (𝑦𝑡+1 ) 𝐴𝑥𝑡+1
signal with a step signal, the fractional-order algorithm with
𝑚=1
(18) a piecewise planar signal which looks more natural and does
𝜆 1 2 not produce false edges.
+ 𝑥𝑡+1 − 𝑧 + 𝐴𝑇 𝑦𝑡+1 . We now consider a two-dimensional image “Lena” of
2 𝜆 2
size 512 × 512. The image is degraded by using AWGN
Step 6. If (𝑔𝑡+1 < 𝐸mpe ) or (𝑡 > 𝑇), then we terminate the with standard deviation of 10 and the result is shown in
iteration and output 𝑥𝑡+1 ; otherwise, go back to Step 1. Figure 2(a). This degraded image is, respectively, fed into the
proposed FOTV denoising algorithm with 𝛼 = 1 and 𝛼 = 1.5
Due to the simplicity of our model, it is easy to be as initial condition, and the time evolution of the algorithms
implemented and can be effectively accelerated on parallel begins. The results at the 1000th iteration are, respectively,
hardware such as field programmable gata array (FPGA) and shown in Figures 2(b) and 2(c). The blocky effects are obvious
graphics processing unit (GPU). In order to employ the huge in Figure 2(b), while the Figure 2(c) looks more natural and
computational power and the parallel processing capabilities does not produce blocky effect.
of FPGA to obtain a fully accelerated implementation of our In conclusion, the fractional-order TV algorithm can
denoising method, the numerical method should be working reduce blocky effect effectively comparing with the tradi-
on the regular grids. For this purpose, the implementation of tional first-order TV algorithm.
our algorithm on FPGA is described as the following strategy.
First, an image frame is captured and read in the processor 3.2. Analysis of Denoising Performance. The aim of this
from a live camera. For the captured frame 𝑓 ∈ R𝑛×𝑛 , the 𝑘th section is to analyze the denoising performance of the FOTV
Abstract and Applied Analysis 5
10 10
5 5
y
0 y 0
−5 −5
−10 −10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t t
(a) Noisy signal (b) Processed by FOTV with 𝛼 = 1
10
y
0
−5
−10
0 0.2 0.4 0.6 0.8 1
t
(c) Processed by FOTV with 𝛼 = 1.5
algorithm. For this purpose, three famous 512 × 512 test Peppers image. From the figure, we can obtain the following
images, called Barbara, Lena, and Peppers, are used in the conclusions. Firstly, the PSNR reaches a maximum between
experiments, as shown in Figure 3. In order to quantify the 𝛼 = 1 and 𝛼 = 2. Secondly, the PSNR at 𝛼 = 1 is lower than
denoised image, we consider the peak-signal-to-noise ratio PSNR at 𝛼 > 1, which is owing to the blocky effect. Finally, the
(PSNR), which has been largely used in the literature and PSNR decreases rapidly as 𝛼 tends to zero, and when 𝛼 = 0,
commonly applied to determine the quality of a processed the PSNR reaches a minimum. According to these facts, we
image. It can be calculated by the following formula: can select the fractional-order 𝛼 between 𝛼 = 1.2 and 𝛼 = 2.
𝑉𝐿2 𝑀𝑁 In order to analyze the denoising performance, the addi-
PSNR = 10log10 2
, (21) tive white Gaussian noise (AWGN) with standard deviation
∑𝑁 𝑀
̂ (𝑖, 𝑗))
𝑖=1 ∑𝑗=1 (𝑢 (𝑖, 𝑗) − 𝑢 (SD) of 10, 20, and 30 is, respectively, added in the three
where 𝑢 is the original image, 𝑢̂ is the denoised image, and 𝑉𝐿 test images. Five denoising algorithms are used to process
is the maximal gray level of the image. these given noisy images, which are the improved Perona and
Firstly, in order to decide the value of the fractional- Malik (IP-M) model [33], fourth order (F-O) PDE model [7],
order 𝛼, we study the relation between the PSNR and 𝛼. improved fourth order (IF-O) PDE model [8], ROF model
Figure 4 shows the relation between the PSNR and 𝛼 on Lena [1], and the proposed FOTV denoising algorithm. In this
and Peppers image corrupted by the additive white Gaussian experiment, we set 𝜎 = 5 and Δ𝑡 = 0.25 for IP-M, F-O-PDE,
noise with standard deviation (SD) of 20. The left image is and IF-O-PDE models, and we set 𝜆 = 0.07, 𝜏 = 0.1, and
the result of Lena image and the right one is the result of 𝐾 = 20 for ROF model. These parameters are able to ensure
6 Abstract and Applied Analysis
the best denoising performance of the corresponding denois- the PSNR is, the better the denoising performance is. For easy
ing model. Based on the conclusion of the foregoing experi- observation, the biggest PSNR values are shown in boldface.
ment, we set 𝛼 = 1.8 for FOTV denoising algorithm, and the From the table, it is obvious that the PSNR of our proposed
denoising results are shown in Table 1. In the table, the first algorithm is bigger than that of the other four algorithms,
column lists the given image and the second column lists the so we can conclude that our FOTV denoising algorithm
standard deviation of noisy image. The PSNRs of the denoised outperforms the other algorithms.
images processed by the different denoising algorithms are In order to further verify the denoising performance of
listed under the corresponding denoising model. The bigger our proposed algorithm, Figure 5 shows the denoised images
Abstract and Applied Analysis 7
(c) Peppers
32 X: 1 32
Y: 30.75
X: 1.7 X: 2
X: 1.5 X: 2
30 Y: 31.32 Y: 31.27
30 X: 1
Y: 31.04 Y: 30.96
Y: 30.66
28
28
26
PSNR
PSNR
26
24
24
22
20 22
18 20
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4
𝛼 𝛼
(a) Lena (b) Peppers
Figure 4: Relation between the fractional-order 𝛼 and PSNR on Lena (a) and Peppers (b) image corrupted by the additive white Gaussian
noise with standard deviation (SD) of 20.
8 Abstract and Applied Analysis
(a) Barbara with 20 SD Gaussian noise (b) Partial view of (a) (c) IP-M algorithm
(d) IF-O PDE algorithm (e) ROF algorithm (f) FOTV algorithm
Figure 5: Comparison of four algorithms for the noisy image with additive Gaussian white noise of 𝜎 = 20.
processed by four denoising algorithms. The first figure is iteration. Figure 6(a) shows the relation between the iteration
the famous image “Barbara” corrupted by the additive white and PSNR. The horizontal axis of this figure is iteration
Gaussian noise with standard deviation (SD) of 20. In order number and the vertical axis is PSNR. The different colorful
to show the comparison clearly, the partial enlarged view curve denotes the result obtained from the different noisy
of the noisy image is shown in the second figure. The image. The red dotted line is the result of the noisy image
denoised result of IP-M algorithm is shown in the right-top with 𝜎 = 30, the green dashed line is the result of the noisy
figure. The second row shows the denoised images processed image with 𝜎 = 20, and the red real line is the result of the
by IF-O PDE algorithm, ROF algorithm, and our FOTV noisy image with 𝜎 = 10. From the figure, we can see that
algorithm. In Figures 5(c) and 5(d), although the detailed the PSNR values are stable after 50 iterations, which indicate
information is preserved, there are a lot of noises unremoved. that our proposed FOTV denoising algorithm is convergent.
The result in Figure 5(e) looks blocky. Only the result of our In addition, it can be seen that the bigger the variance of white
model, as shown in Figure 5(f), looks natural and does not noise is, the longer the stopping times are and the lower the
produce false edges. So we conclude that our proposed FOTV PSNRs are.
algorithm is able to achieve a better tradeoff between edge Figure 6 plots the convergence of the FOTV algorithm for
preservation and noise removal. the above experiment together with the theoretical 𝑂(1/𝑁2 )
and 𝑂(1/𝑁4 ) rate. For easy observation, the horizontal axis
3.3. The Convergence and the Rate of Convergence. The is the log curve of iteration number and the vertical axis is
convergence and the rate of convergence are two important the log curve of primal-dual gap calculated in Step 5. The
factors for evaluating the performance of the denoising dashed line shows the theoretical 𝑂(1/𝑁2 ) rate and the dotted
method. In this section, some experiments are given to line shows the theoretical 𝑂(1/𝑁4 ) rate. The red line shows
show the convergence and convergence rate of our proposed the convergence of FOTV algorithm in the noisy image with
fractional-order total variation (FOTV) denoising algorithm. 𝜎 = 30, the green line shows the convergence of FOTV
For this purpose, the proposed FOTV algorithm is, algorithm in the noisy image with 𝜎 = 20, and the blue line
respectively, used to process the image “Lena” corrupted by shows the convergence of FOTV algorithm in the noisy image
the additive white Gaussian noise with standard deviation with 𝜎 = 10. From this figure, it can be seen that the smaller
(SD) of 10, 20, and 30, and the PSNR is recorded at each the variance of white noise is, the faster convergence is. In
Abstract and Applied Analysis 9
36 106
34
32 104
30
28 102
Error
PSNR
26
24 100
22
20 10−2
18
0 50 100 150 200 100 101 102 103
Iteration number Iterations
Figure 6: Relation among the iteration, PSNR, and primal-dual gap on the noisy Lena image with 𝜎 = 10, 𝜎 = 20, and 𝜎 = 30.
addition, we can see that the FOTV algorithm converges at [3] C. Vogel and M. Oman, “Fast, robust total variation-based
𝑂(1/𝑁2 ) rate firstly and 𝑂(1/𝑁4 ) rate in the end. reconstruction of noisy, blurred images,” IEEE Transactions on
Image Processing, vol. 7, no. 6, pp. 813–824, 1998.
[4] F. Alter, S. Durand, and J. Froment, “Adapted total variation
4. Conclusion for artifact free decompression of JPEG images,” Journal of
Mathematical Imaging and Vision, vol. 23, no. 2, pp. 199–211,
A fractional-order total variation image denoising algorithm 2005.
based on primal-dual method was proposed in this paper. [5] F. Li, C. Shen, and C. Li, “Multiphase soft segmentation with
The main contributions are as follows: (1) the saddle point total variation and H 1 regularization,” Journal of Mathematical
and dual formulation of the proposed fractional-order total Imaging and Vision, vol. 37, no. 2, pp. 98–111, 2010.
variation model are constructed in theory; (2) the primal- [6] J. Zhang, Z. Wei, and L. Xiao, “Adaptive fractional-order multi-
dual algorithm was used to solve the constructed saddle- scale method for image denoising,” Journal of Mathematical
point problem, in which 𝑂(1/𝑁2 ) convergence rate can Imaging and Vision, vol. 43, no. 1, pp. 39–49, 2012.
be guaranteed; (3) the experiments demonstrate that the [7] Y. L. You and M. Kaveh, “Fourth-order partial differential
proposed methodology is able to avoid the blocky effect, equations for noise removal,” IEEE Transactions on Image
achieve state-of-the-art performance, and guarantee 𝑂(1/𝑁2 ) Processing, vol. 9, no. 10, pp. 1723–1730, 2000.
convergence rate. [8] M. Hajiaboli, “A self-governing fourth-order nonlinear dif-
fusion filter for image noise removal,” IPSJ Transactions on
Computer Vision and Applications, vol. 2, pp. 94–103, 2010.
Acknowledgments [9] R. Herrmann, Fractional Calculus: An Introduction for Physi-
cists, World Scientific, New Jersey, NJ, USA, 2011.
This work was supported by National Natural Science Foun-
dation of China (no. 61201378), Scientific Research Fund of [10] S.-C. Liu and S. Chang, “Dimension estimation of discrete-time
fractional Brownian motion with applications to image texture
Liaoning Provincial Education Department (L2012073), and
classification,” IEEE Transactions on Image Processing, vol. 6, no.
Fundamental Research Funds for the Central Universities 8, pp. 1176–1184, 1997.
(N110304001).
[11] D. Chen, Y. Chen, and D. Xue, “Digital fractional order
Savitzky-Golay differentiator,” IEEE Transactions on Circuits
References and Systems II, vol. 58, no. 11, pp. 758–762, 2011.
[12] S. Didas, B. Burgeth, A. Imiya, and J. Weickert, “Regularity and
[1] L. Rudin, S. Osher, and E. Fatemi, “Nonlinear total variation scale-space properties of fractional high order linear filtering,”
based noise removal algorithms,” Physica D, vol. 60, no. 1–4, pp. in 5th International Conference on Scale Space and PDE Methods
259–268, 1992. in Computer Vision, Scale-Space 2005, pp. 13–25, deu, April
[2] J.-F. Aujol, “Some first-order algorithms for total variation based 2005.
image restoration,” Journal of Mathematical Imaging and Vision, [13] B. Ninness, “Estimation of 1/𝑓 Noise,” IEEE Transactions on
vol. 34, no. 3, pp. 307–327, 2009. Information Theory, vol. 44, no. 1, pp. 32–46, 1998.
10 Abstract and Applied Analysis
[14] I. Petras, D. Sierociuk, and I. Podlubny, “Identification of [32] S. Boyd and L. Vandenberghe, Convex Optimization, Cambridge
parameters of a half-order system,” IEEE Transactions on Signal University Press, New York, NY, USA, 2004.
Processing, vol. 60, no. 10, pp. 5561–5566, 2012. [33] P. Perona and J. Malik, “Scale-space and edge detection using
[15] D. Chen, Y. Chen, and D. Xue, “1-D and 2-D digital fractional- anisotropic diffusion,” IEEE Transactions on Pattern Analysis
order Savitzky-Golay differentiator,” Signal, Image and Video and Machine Intelligence, vol. 12, no. 7, pp. 629–639, 1990.
Processing, vol. 6, no. 3, pp. 503–511, 2012.
[16] Y. F. Pu, J. L. Zhou, and X. Yuan, “Fractional differential mask:
a fractional differential-based approach for multiscale texture
enhancement,” IEEE Transactions on Image Processing, vol. 19,
no. 2, pp. 491–511, 2010.
[17] D. Chen, H. Sheng, Y. Chen, and D. Xue, “Fractional-order vari-
ational optical flow model for motion estimation,” Philosophical
Transactions of the Royal Society A, vol. 371, no. 1990, 2013.
[18] J. Bai and X. Feng, “Fractional-order anisotropic diffusion for
image denoising,” IEEE Transactions on Image Processing, vol.
16, no. 10, pp. 2492–2502, 2007.
[19] P. Guidotti and J. V. Lambers, “Two new nonlinear nonlocal
diffusions for noise reduction,” Journal of Mathematical Imaging
and Vision, vol. 33, no. 1, pp. 25–37, 2009.
[20] E. Cuesta, M. Kirane, and S. A. Malik, “Image structure preserv-
ing denoising using generalized fractional time integrals,” Signal
Processing, vol. 92, no. 2, pp. 553–563, 2012.
[21] M. Janev, S. Pilipović, T. Atanacković, R. Obradović, and
N. Ralević, “Fully fractional anisotropic diffusion for image
denoising,” Mathematical and Computer Modelling, vol. 54, no.
1-2, pp. 729–741, 2011.
[22] D. Chen, S. Sun, C. Zhang, Y. Chen, and D. Xue, “Fractional-
order TV-L 2 model for image denoising,” Central European
Journal of Physics, 2013.
[23] K. Arrow, L. Hurwicz, and H. Uzawa, “Studies in linear and non-
linear programming,” in Stanford Mathematical Studies in the
Social Sciences, Stanford University Press, Palo Alto, Calif, USA,
1958.
[24] M. Zhu and T. Chan, “An efficient primal-dual hybrid gradient
algorithm for total variation image restoration,” CAM Report
08-34, UCLA, Los Angeles, Calif, USA, 2008.
[25] A. Chambolle, “An Algorithm for Total Variation Minimization
and Applications,” Journal of Mathematical Imaging and Vision,
vol. 20, no. 1-2, pp. 89–97, 2004.
[26] E. Esser, X. Zhang, and T. F. Chan, “A general framework
for a class of first order primal-dual algorithms for convex
optimization in imaging science,” SIAM Journal on Imaging
Sciences, vol. 3, no. 4, pp. 1015–1046, 2010.
[27] S. Bonettini and V. Ruggiero, “On the convergence of primal-
dual hybrid gradient algorithms for total variation image
restoration,” Journal of Mathematical Imaging and Vision, vol.
44, no. 3, pp. 236–253, 2012.
[28] B. He and X. Yuan, “Convergence analysis of primal-dual
algorithms for a saddle-point problem: from contraction per-
spective,” SIAM Journal on Imaging Sciences, vol. 5, no. 1, pp.
119–149, 2012.
[29] A. Chambolle and T. Pock, “A first-order primal-dual algorithm
for convex problems with applications to imaging,” Journal of
Mathematical Imaging and Vision, vol. 40, no. 1, pp. 120–145,
2011.
[30] Y. Chen, G. Lan, and Y. Ouyang, “Optimal primal-dual methods
for a class of saddle point problems,” CAM Report 13-31, UCLA,
Los Angeles, Calif, USA, 2013.
[31] I. Podlubny, Fractional Differential Equations, Academic Press,
New York, NY, USA, 1999.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 810625, 9 pages
http://dx.doi.org/10.1155/2013/810625
Research Article
Local Observability of Systems on Time Scales
Zbigniew Bartosiewicz
Faculty of Computer Science, Bialystok University of Technology, Wiejska 45a, 15-351 Bialystok, Poland
Copyright © 2013 Zbigniew Bartosiewicz. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
Analytic systems on an arbitrary time-scale are studied. As particular cases they include continuous-time and discrete-time systems.
Several local observability properties are considered. They are characterized in a unified way using the language of real analytic
geometry, ideals of germs of analytic functions, and their real radicals. It is shown that some properties related to observability are
preserved under various discretizations of continuous-time systems.
If T = R, then (1) is the standard continuous-time system Example 3. Let 𝜑 = 𝑥𝑖 be the 𝑖th coordinate function on R𝑛 .
Then 𝜑 = 𝑒𝑖 —the (row) vector of the standard basis of R𝑛
𝑥̇(𝑡) = 𝑓 (𝑥 (𝑡) , 𝑢 (𝑡)) , with 1 at the 𝑖-th position. For any 𝑡0 ∈ T we have
(2)
𝑦 (𝑡) = ℎ (𝑥 (𝑡)) . (Γ𝑔𝑡0 𝑥𝑖 ) (𝑥) = 𝑒𝑖 𝑔 (𝑥) = 𝑔𝑖 (𝑥) . (8)
For T = Z (1) takes the form Observe that if 𝜇(𝑡0 ) > 0 then
𝑥 (𝑡 + 1) − 𝑥 (𝑡) = 𝑓 (𝑥 (𝑡) , 𝑢 (𝑡)) , 1
(3) (Γ𝑔𝑡0 𝜑) (𝑥) = (𝜑 (𝑥 + 𝜇 (𝑡0 ) 𝑔 (𝑥)) − 𝜑 (𝑥)) . (9)
𝑦 (𝑡) = ℎ (𝑥 (𝑡)) . 𝜇 (𝑡0 )
This can be rewritten in a more standard shift form as On the other hand for 𝜇(𝑡0 ) = 0 we obtain (Γ𝑔𝑡0 𝜑)(𝑥) =
𝜑 (𝑥)𝑔(𝑥), which is the standard Lie derivative of the function
𝑥 (𝑡 + 1) = 𝑓 (𝑥 (𝑡) , 𝑢 (𝑡)) + 𝑥 (𝑡) =: 𝑔 (𝑥 (𝑡) , 𝑢 (𝑡)) , 𝜑 along the vector field 𝑔. In general, when operator Γ𝑔𝑡0 does
(4)
𝑦 (𝑡) = ℎ (𝑥 (𝑡)) . not depend on 𝑡0 , we will denote it by Γ𝑔 . This happens, for
instance, if the time-scale is homogeneous.
As there is a simple passage from 𝑓 to 𝑔 and vice versa, all Let 𝜎𝑔𝑡0 : 𝐶𝜔 (R𝑛 ) → 𝐶𝜔 (R𝑛 ) be another map related to
statements for (3) may be translated to statements for (4). the function 𝑔 and the time 𝑡0 defined by
Remark 1. The equation 𝑥(𝑡 + 1) = 𝑔(𝑥(𝑡), 𝑢(𝑡)) may be (𝜎𝑔𝑡0 𝜑) (𝑥) := 𝜑 (𝑥 + 𝜇 (𝑡0 ) 𝑔 (𝑥)) . (10)
studied on an arbitrary set 𝑋 or on an analytic manifold
𝑀, if analyticity of the system is essential. But then we If 𝜇(𝑡0 ) = 0, then 𝜎𝑔𝑡0 is the identity map. In general we have
cannot pass to form (3), as to do this we need a linear space an obvious property
structure. Thus, one can argue that (4) is more general than
(3). However, we concentrate here on local analytic problems, Proposition 4. The map 𝜎𝑔𝑡0 is an endomorphism of the
for which R𝑛 is general enough. algebra 𝐶𝜔 (R𝑛 ).
By 𝛾(𝑡, 𝑡0 , 𝑥0 , 𝑢) we denote the solution of the equation
𝑥Δ = 𝑓(𝑥, 𝑢) corresponding to control 𝑢 and the initial For 𝜇(𝑡0 ) > 0 the operators Γ𝑔𝑡0 and 𝜎𝑔𝑡0 are related by the
condition 𝑥(𝑡0 ) = 𝑥0 and evaluated at time 𝑡. following equality:
Let 𝑥1 , 𝑥2 ∈ R𝑛 and 𝑡0 ∈ T. Then 𝑥1 and 𝑥2 are called
indistinguishable at time 𝑡0 if 𝜎𝑔𝑡0 𝜑 = 𝜑 + 𝜇 (𝑡0 ) Γ𝑔𝑡0 𝜑. (11)
ℎ (𝛾 (𝑡, 𝑡0 , 𝑥1 , 𝑢)) = ℎ (𝛾 (𝑡, 𝑡0 , 𝑥2 , 𝑢)) (5) We also have the following generalization of the Leibniz rule.
1
Proposition 6. (i) The states 𝑥1 and 𝑥2 are indistinguishable
(Γ𝑔𝑡0 𝜑) (𝑥) := ∫ 𝜑 (𝑥 + 𝑠𝜇 (𝑡0 ) 𝑔 (𝑥)) 𝑑𝑠 ⋅ 𝑔 (𝑥) , (7)
at time 𝑡0 if and only if for every 𝜑 ∈ 𝐻𝑡0 , 𝜑(𝑥1 ) = 𝜑(𝑥2 ).
0
(ii) The states 𝑥1 and 𝑥2 are indistinguishable if and only if
where 𝜑 ∈ 𝐶𝜔 (R𝑛 ) and 𝜑 is the gradient of 𝜑 (a row vector). for every 𝜑 ∈ 𝐻, 𝜑(𝑥1 ) = 𝜑(𝑥2 ).
Abstract and Applied Analysis 3
Proof. The statement (ii) was shown in [1] for continuous- We say that Σ is robustly locally observable at 𝑥0 (RLO(𝑥0 ))
time systems (T = R). In this case (i) and (ii) mean the if there is a neighborhood 𝑈 of 𝑥0 such that Σ is weakly locally
same since indistinguishability at some 𝑡0 is equivalent to observable at 𝑥 for every 𝑥 ∈ 𝑈.
indistinguishability. For an arbitrary time-scale (ii) follows Robust local observability was introduced in [3] for
from (i). The statement (i) for an arbitrary time-scale was continuous-time systems under the name “stable local
shown in [8]. Analyticity of the control system and the observability.” It means that the weak local observability at 𝑥0
functions from 𝐻 is essential in the proof. is stable or robust with respect to small perturbations of the
initial condition 𝑥0 .
Remark 7. Proposition 6 allows us to use the same language We call Σ weakly locally observable (WLO) (strongly locally
of analytic functions on R𝑛 to study different observability observable (SLO), and robustly locally observable (RLO),
properties of analytic systems on arbitrary time-scales as long resp.), when it is weakly locally observable (strongly locally
as these properties are defined via the indistinguishability observable and robustly locally observable, resp.) at every
relation. It also implies that indistinguishability is an equiv- 𝑥 ∈ R𝑛 .
alence relation. This is not true for smooth systems (see [9]) Let 𝑑H(𝑥0 ) denote the linear space of the differentials of
or for analytic partially defined systems (see [10], where a functions from H taken at 𝑥0 . The following theorem is a
different definition was developed to preserve this property simple extension of the result from [1].
for partially defined systems).
Let H denote the subalgebra of 𝐶𝜔 (R𝑛 ) generated by 𝐻. Theorem 11. (a) If dim 𝑑H(𝑥0 ) = 𝑛, then Σ is SLO(𝑥0 ).
It will be called the observation algebra of the system Σ. The (b) (𝑓𝑜𝑟 𝑎𝑙𝑙 𝑥0 ∈ R𝑛 : dim 𝑑H(𝑥0 ) = 𝑛) ⇒ (𝑓𝑜𝑟
elements of H are obtained by substituting functions from 𝐻 𝑎𝑙𝑙 𝑥0 ∈ R𝑛 : Σ is SLO (𝑥0 )) ⇒ (∃𝑋—a real analytic set in
into polynomials of several variables with real coefficients. In R𝑛 : 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑥0 ∉ 𝑋 : dim 𝑑H(𝑥0 ) = 𝑛).
particular, all constant functions belong to H.
From Proposition 6 we get the following. Let us denote the condition dim 𝑑H(𝑥0 ) = 𝑛 by HK(𝑥0 )
(Hermann-Krener condition at 𝑥0 ). The second part of
Proposition 8. The states 𝑥1 and 𝑥2 are indistinguishable if Theorem 11 says that if we are interested in strong local
and only if for every 𝜑 ∈ H, 𝜑(𝑥1 ) = 𝜑(𝑥2 ). observability at large, that is, at each point, then condition
HK(𝑥0 ) is satisfied almost everywhere, so the gap between
We say that Σ is observable if any two distinct states are sufficient condition and necessary condition for strong local
distinguishable. observability at large is quite narrow. However, when one is
From the definition and Proposition 8 we obtain the interested in local observability at a particular point of the
following characterization. state space, the Hermann-Krener condition may be far from
being necessary (see [2]).
Proposition 9. Σ is observable if and only if for any distinct We have a nice gradation of different local observability
𝑥1 , 𝑥2 ∈ R𝑛 there is 𝜑 ∈ H such that 𝜑(𝑥1 ) ≠𝜑(𝑥2 ). concepts.
The condition stated in Proposition 9 is difficult to check. Proposition 12. HK(𝑥0 ) ⇒ Σ is SLO(𝑥0 ) ⇒ Σ is RLO(𝑥0 ) ⇒
This is one of the reasons that a weaker concept of local Σ is WLO(𝑥0 ).
observability seems to be more interesting. There are many None of the implications in Proposition 12 may be, in
different concepts of local observability and one concept has general, reversed.
often a few different names. For the first two concepts we
follow the terminology used in [4]. Example 13. (a) Let 𝑥 ∈ R, 𝑥Δ = 0, and 𝑦 = 𝑥3 . Thus, 𝐻 =
We say that Σ is weakly locally observable at 𝑥0 (WLO(𝑥0 )) {𝑥3 }. The system is observable, so it is also strongly locally
if there is a neighborhood 𝑈 of 𝑥0 such that for every 𝑥 ∈ 𝑈, observable at any 𝑥. But the Hermann-Krener condition fails
𝑥 and 𝑥0 are distinguishable. at 𝑥 = 0.
(b) Let 𝑥 ∈ R, 𝑥Δ = 0, and 𝑦 = 𝑥2 . Then 𝐻 = {𝑥2 }. The
Remark 10. Weak local observability at 𝑥0 is in fact a weak system is robustly locally observable at 0, but it is not strongly
property. It holds, for example, for the system locally observable at this point.
(c) Let 𝑥 ∈ R2 , 𝑥Δ = 0, and 𝑦 = 𝑥12 + 𝑥22 . The system
𝑥Δ = 0, 𝑦 = 𝑥12 + ⋅ ⋅ ⋅ + 𝑥𝑛2 , (15) is weakly locally observable at 0, but it is not robustly locally
observable at this point.
But we have an important, though obvious, global equiv-
at 𝑥0 = 0 ∈ R𝑛 . Since all solutions of 𝑥Δ = 0 are constant,
alence.
time does not influence indistinguishability relation. To
distinguish points we have to use only the output function. Proposition 14. Σ is RLO(𝑥0 ) for every 𝑥0 ∈ R𝑛 if and only if
Clearly, it takes different values at 0 and any other point, so Σ is WLO(𝑥0 ) for every 𝑥0 ∈ R𝑛 . In other words, Σ is RLO if
we can distinguish 𝑥0 from any of its neighbors. Observe that and only if Σ is WLO.
local observability fails at any 𝑥0 ≠0, if 𝑛 ≥ 2.
We say that Σ is strongly locally observable at 𝑥0 (SLO(𝑥0 )) By O𝑥 we will denote the algebra of germs at 𝑥 of analytic
if there is a neighborhood 𝑈 of 𝑥0 such that for every distinct functions on R𝑛 (see Appendix B). Let 𝐽𝑥 be the ideal of O𝑥
𝑥1 , 𝑥2 ∈ 𝑈, 𝑥1 and 𝑥2 are distinguishable. generated by germs at 𝑥 of functions from H that vanish at 𝑥.
4 Abstract and Applied Analysis
For 𝑥 ∈ R𝑛 and for an ideal 𝐽 of O𝑥 , let 𝑍(𝐽) be the germ Now, for a point 𝑥 ∈ R𝑛 , we define a sequence of ideals
at 𝑥 of the zero-set of 𝐽. Since 𝐽 is finitely generated (O𝑥 is in O𝑥 related to the system Σ. Let 𝐼𝑥(0) := (0) and 𝐼𝑥(𝑘+1) :=
Noetherian), 𝑍(𝐽) is well defined. Let 𝐼(𝑍(𝐽)) be the ideal of
√ 𝐷(H𝑥 ∪ 𝐼𝑥(𝑘) ). It is clear that instead of H𝑥 in this definition
R
O𝑥 consisting of all germs of analytic functions that vanish on
𝑍(𝐽). one can take the previously defined ideal 𝐽𝑥 ; that is, 𝐼𝑥(𝑘+1) =
√ 𝐷(𝐽𝑥 ∪ 𝐼𝑥(𝑘) ). This leads to the following generalization of
R
R 𝐽
(a) Σ is WLO(𝑥0 ) if and only if √ 𝑥0 = 𝑚𝑥0 . √ 𝐷(H𝑥 ∪ 𝐼𝑥(𝑘+1) ). This means that 𝐼𝑥(𝑘+1) ⊆ 𝐼𝑥(𝑘+2) . Since the
R
(b) HK(𝑥0 ) if and only if 𝐽𝑥0 = 𝑚𝑥0 . ring O𝑥 is Noetherian the sequence of ideals 𝐼𝑥(𝑘) must stabilize
at some 𝑠.
Proof. (a) Σ is not weakly locally observable at 𝑥0 if and
only if arbitrarily close to 𝑥0 there is 𝑥 such that 𝑥0 and Now we can characterize robust local observability.
𝑥 are indistinguishable. By Lemma 15, the last statement
is equivalent to the condition 𝑍(𝐽𝑥0 ) ≠{𝑥0 }, which in turn Theorem 19. System Σ is RLO(𝑥0 ) if and only if 𝐼𝑥(𝑠)0 = O𝑥0 for
means that 𝐼(𝑍(𝐽𝑥0 )) ≠𝐼({𝑥0 }). But 𝐼({𝑥0 }) = 𝑚𝑥0 , so from some 𝑠 > 0.
Theorem B.1 the last inequality is equivalent to the condition
R 𝐽 The proof of Theorem 19 will rely on several lemmas. They
√ 𝑥0 ≠𝑚𝑥0 . This gives the equivalence of both sides in (a).
appeared in a similar form in [3]. However, there were a few
(b) It is enough to prove the proposition for 𝑥 = 0 in R𝑛 . flaws in the proofs, which are now corrected.
Let G be a family of analytic functions on some open set
Observe that 𝑑H(𝑥) = 𝑑𝐽𝑥 (𝑥) for every 𝑥. Thus, if 𝑈 ⊂ R𝑛 . Denote by 𝑆𝑥 (G) the germ at 𝑥 of the level set of G
𝐽0 = 𝑚0 , then 𝑑H(0) contains all the differentials 𝑑𝑥𝑖 for that passes through 𝑥. Thus
𝑖 = 1, . . . , 𝑛, which are linearly independent.
On the other hand, the condition dim 𝑑H(0) = 𝑛 implies 𝑆𝑥 (G) = {𝑦 ∈ 𝑈 : 𝜑 (𝑥) = 𝜑 (𝑦) for 𝜑 ∈ G}𝑥 . (17)
that in a neighborhood 𝑈 of 0, there are functions 𝜑1 , . . . , 𝜑𝑛
The set-germ 𝑆𝑥 (G) is a germ of analytic set. One of the
whose germs belong to 𝐽0 and the differentials at 0, 𝑑𝜑𝑖 (0)
representatives of 𝑆𝑥 (G) is the analytic set in 𝑈 : {𝑦 ∈ 𝑈 :
are linearly independent. We may write 𝜑𝑖 = ∑𝑗 Φ𝑖𝑗 𝑥𝑗 for
𝜑(𝑦) = 𝜑(𝑥) for 𝜑 ∈ G}.
some analytic functions Φ𝑖𝑗 on 𝑈 (sufficiently small). Then
𝑑𝜑𝑖 (0) = ∑𝑗 Φ𝑖𝑗 (0)𝑑𝑥𝑗 . This means that the matrix Φ = Lemma 20. Let 𝑈 be an open subset of R𝑛 and let 𝑋 be an
(Φ𝑖𝑗 ) is invertible at 0 and then in some neighborhood of 0. analytic set in 𝑈. Consider a family G of analytic functions on
Let Ψ = Φ−1 . Then the germs of elements of Ψ are in O𝑥 𝑈. If for every 𝑥 ∉ 𝑋 : 𝑆𝑥 (G) = {𝑥}, then for every 𝑥 ∈ 𝑋 :
and 𝑑𝑥𝑖 = ∑𝑗 Ψ𝑖𝑗 𝑑𝜑𝑗 which means that 𝜑1 , . . . , 𝜑𝑛 generate 𝑆𝑥 (G) ⊂ 𝑋𝑥 .
𝑚0 .
Proof. Suppose that there is 𝑥 ∈ 𝑋 such that 𝑆𝑥 (G) ⊄ 𝑋𝑥 .
Let 𝐺 be a subset of O𝑥 . By 𝐷(𝐺) we will denote the ideal of This means that for every representatives 𝑆̃𝑥 (G) and 𝑋 ̃𝑥 we
O𝑥 generated by Jacobians 𝜕(𝜑1 , . . . , 𝜑𝑛 )/𝜕(𝑥1 , . . . , 𝑥𝑛 ), where have 𝑆̃𝑥 (G) ⊄ 𝑋 ̃𝑥 . Take 𝑋̃𝑥 := 𝑋 and arbitrarily small
𝜑𝑖 ∈ 𝐺. Observe that these Jacobians are well defined on
neighborhood 𝑉 of 𝑥 in 𝑈. Let 𝑆̃𝑥 (G) be a representative of
germs of functions. If G is a family of real analytic functions
𝑆𝑥 (G) in 𝑉. Then there is 𝑦 ∈ 𝑆̃𝑥 (G) such that 𝑦 ∉ 𝑋.
on an open set 𝑈 in R𝑛 , then similarly we define the Jacobian
Take a sequence (𝑦𝑛 ) of such points converging to 𝑥. We may
ideal 𝐷(G) in O𝑈. Furthermore, there is a simple relation
assume that all these points belong to some (large enough)
between Jacobian ideals for functions and germs of functions.
representative 𝑌 of 𝑆𝑥 (G) and that 𝑌 is an analytic set. Only
If G is a family of analytic functions on 𝑈, and 𝑥0 ∈ 𝑈 then
finite number of points 𝑦𝑛 may be isolated points of 𝑌. This
we have
means that arbitrarily close to 𝑥 there is a point 𝑦𝑛 for which
𝐷(G)𝑥0 = 𝐷 (G𝑥0 ) . (16) 𝑆𝑦𝑛 (G) ≠{𝑦𝑛 }. Thus we get a contradiction.
Abstract and Applied Analysis 5
Lemma 21. Let 𝑈 be an open subset of R𝑛 and let G be a family Lemma 24. Let 𝑈 be an open subset of R𝑛 and let 𝜑1 , . . . , 𝜑𝑘
of analytic functions on 𝑈. For every 𝑥 ∈ 𝑈: if 𝑥 ∉ 𝑍(𝐷(G)), be analytic functions on 𝑈 whose gradients are linearly inde-
then 𝑆𝑥 (G) = {𝑥}. pendent at each point of 𝑈. Let 𝑌 = 𝑍(𝜑1 , . . . , 𝜑𝑘 ), 𝑥0 ∈ 𝑌, and
let G be a family of analytic functions on 𝑈.
Proof. If 𝑥 ∉ 𝑍(𝐷(G)), then there are functions 𝜑1 , . . . , 𝜑𝑛 ∈ If 𝑌 ⊂ 𝑍(𝐷(G ∪ {𝜑1 , . . . , 𝜑𝑘 })), then arbitrarily close to 𝑥0
𝐺 such that there is 𝑥 ∈ 𝑌 such that 𝑆𝑥 (G) ≠{𝑥}.
𝜕 (𝜑1 , . . . , 𝜑𝑛 )
(𝑥) ≠0. (18) Proof. Changing the coordinates we can obtain 𝜑𝑖 = 𝑥𝑖 , 𝑖 =
𝜕 (𝑥1 , . . . , 𝑥𝑛 ) 1, . . . , 𝑘. Let 𝜓1 , . . . , 𝜓𝑛−𝑘 ∈ G. Then for 𝑥 ∈ 𝑌 we have
Thus, the map 𝑥 → (𝜑1 (𝑥), . . . , 𝜑𝑛 (𝑥)) is injective in a
neighborhood of 𝑥. This implies that 𝑆𝑥 (G) = {𝑥}. ∇𝜑1 (𝑥)
[ .. ]
[ . ]
Lemma 22. Assume that 𝑋𝑥(𝑘)0 ≠𝑋𝑥(𝑘+1) for some 𝑘 ≥ 0. Then [ ]
0 [ ∇𝜑𝑘 (𝑥) ]
there is a neighborhood 𝑈 of 𝑥0 in R𝑛 and a representative 0 = det [
[ ∇𝜓1 (𝑥) ]
]
̃(𝑘+1) of 𝑋(𝑘+1) in 𝑈 such that for every 𝑥 ∈ 𝑈:
𝑋 [ ]
𝑥0 𝑥0 [ .. ]
[ . ]
̃(𝑘+1) ⇒ 𝑆𝑥 (H) = {𝑥} .
𝑥∉𝑋 (19) ∇𝜓 (𝑥)
𝑥0 [ 𝑛−𝑘 ]
(21)
Proof. First observe that 𝑍(𝐷(H|𝑈)) is a representative of 𝐼 0
𝑍(𝐷(H𝑥0 )). We proceed by induction. Let 𝑘 = 0 and assume [ ]
= det [( 𝜕𝜓𝑖 (𝑥)) ]
that 𝑋𝑥(1)0 ≠𝑋𝑥(0)0 = R𝑛𝑥0 . Take any neighborhood 𝑈 of 𝑥0 . 𝜕𝑥𝑗 𝑖=1,...,𝑛−𝑘
̃(1) is a [ 𝑗=1,...,𝑛 ]
Since 𝑋𝑥(1)0 = 𝑍(𝐷(H𝑥0 )), then 𝑍(𝐷(H|𝑈)) =: 𝑋 𝑥0
representative of 𝑋𝑥(1)0 in 𝑈. If 𝑥 ∉ 𝑋 ̃(1) , then by Lemma 21, 𝜕𝜓𝑖 (𝑥)
𝑥0 = det ( ) .
𝑆𝑥 (H) = 𝑆𝑥 (H|𝑈) = {𝑥}. 𝜕𝑥𝑗 𝑖=1,...,𝑛−𝑘
Now assume that the statement of the lemma holds for 𝑗=𝑘+1,...,𝑛
But 𝑋𝑥(𝑠)0 = 0 so in some neighborhood of 𝑥0 the level set- In the discretized system the ordinary derivative is replaced
germs of H must be trivial. This means that Σ is robustly by the delta derivative on the discrete time-scale.
locally observable at 𝑥0 . Thus, (23) is replaced with
̃(𝑠) = 𝑋
Necessity. Assume that 𝑋 ̃(𝑠+1) for some 𝑠 ≥ 0. From 𝑥 (𝑡 + 𝜇T (𝑡)) − 𝑥 (𝑡)
𝑥0 𝑥0 = 𝑓 (𝑥 (𝑡) , 𝑢 (𝑡)) , 𝑦 (𝑡) = ℎ (𝑥 (𝑡))
Lemma 25 it follows that Σ is not robustly locally observable. 𝜇T (𝑡)
(25)
The observation algebra H is generated by a single function on the scale of integers and on the quantum scales. Different
𝜓(𝑥1 , 𝑥2 ) = 𝑥13 + 𝑥2 , which means that the Hermann-Krener concepts of local observability for systems on arbitrary time-
condition does not hold at any point. The discretized system scales have been considered. We have established relations
ΣT is given by between these concepts and provided characterizations of
weak and robust local observability with the aid of certain
𝑥1 (𝑡 + 𝜇T (𝑡)) = 𝑥1 (𝑡) − 𝜇T (𝑡) 𝑢 (𝑡) , ideals of the ring of germs of analytic functions and real
radicals of those ideals. Equivalent geometric characteriza-
𝑥2 (𝑡 + 𝜇T (𝑡)) = 𝑥2 (𝑡) + 3𝜇T (𝑡) 𝑥12 (𝑡) 𝑢 (𝑡) , (27) tions have been given. Observation algebras from which the
ideals are obtained and the ideals themselves depend on the
𝑦 (𝑡) = 𝑥1 (𝑡)3 + 𝑥2 (𝑡) . time-scale on which the systems is defined, but once the
ideals are computed, the procedures and the criteria of local
The observation algebra HT contains now the functions observability are the same for all time-scales. This allows for
𝜓1 (𝑥1 , 𝑥2 ) = 𝑥13 + 𝑥2 and 𝜓2 (𝑥1 , 𝑥2 ) = 𝑥1 . The Hermann- unified treatment of observability of systems on arbitrary
Krener condition is then satisfied at all points and for all time-scales.
discrete time-scales T.
The language of time-scales allows for a natural descrip-
Remark 31. Hermann-Krener rank condition is equivalent tion of discretization of continuous-time systems: the ordi-
to the property that the ideal 𝐽𝑥0 is maximal; that is, it is nary derivative is replaced by delta derivative on a discrete
generated by the coordinate functions. One can show that this time-scale T. The paper contains preliminary results on
property is preserved when the ideal 𝐽𝑥0 is replaced with the preservation of properties related to observability under
ideals corresponding to systems ΣT if T contains 𝑡 with 𝜇T (𝑡) discretization. In particular Hermann-Krener rank condition
sufficiently small. In characterizations of weak and robust is preserved. Preservation of other properties, in particular
local observability there appear real radicals of ideals. It is weak and robust local observability, is stated as an open
not clear whether desired properties of the radicals like max- problem. To solve the problem one will have to study limit
imality (for WLO(𝑥0 )) or nonproperness (for RLO(𝑥0 ) are properties of real radicals for rings of germs of analytic
preserved under discretizations. Thus preservation of weak functions. This will be a subject of a future research.
and robust local observability under Euler discretization is
still an open problem. Appendices
We finish this discussion with a positive example. A. Calculus on Time-scales
Example 32. Let Σ be
A time-scale T is an arbitrary nonempty closed subset of the
𝑥1̇ = 𝑥2 𝑢, 𝑥2̇ = −𝑥1 𝑢, 𝑦= 𝑥12 + 𝑥22 . (28) set R of real numbers. In particular R, ℎZ for ℎ > 0 and 𝑞N :=
{𝑞𝑘 , 𝑘 ∈ N} for 𝑞 > 1 are time-scales. We assume that T is
The observation algebra of Σ is generated by 𝜑(𝑥1 , 𝑥2 ) = 𝑥12 + a topological space with the relative topology induced from
𝑥22 . R. If 𝑡0 , 𝑡1 ∈ T, then [𝑡0 , 𝑡1 ]T denotes the intersection of the
ordinary closed interval with T. Similar notation is used for
The discretization gives ΣT open, half-open or infinite intervals.
For 𝑡 ∈ T we define the forward jump operator 𝜎 : T → T
𝑥1Δ = 𝑥2 𝑢, 𝑥2Δ = −𝑥1 𝑢, 𝑦 = 𝑥12 + 𝑥22 , (29) by 𝜎(𝑡) := inf{𝑠 ∈ T : 𝑠 > 𝑡} if 𝑡 ≠ sup T and 𝜎(sup T) = sup T
when sup T is finite; the backward jump operator 𝜌 : T → T
where by 𝜌(𝑡) := sup{𝑠 ∈ T : 𝑠 < 𝑡} if 𝑡 ≠ inf T and 𝜌(inf T) = inf T
when inf T is finite; the forward graininess function 𝜇 : T →
𝑥 (𝑡 + 𝜇T (𝑡)) − 𝑥 (𝑡)
𝑥Δ (𝑡) = . (30) [0, ∞) by 𝜇(𝑡) := 𝜎(𝑡) − 𝑡; the backward graininess function
𝜇T (𝑡) ] : T → [0, ∞) by ](𝑡) := 𝑡 − 𝜌(𝑡).
The observation algebra of ΣT is also generated by 𝜑(𝑥1 , 𝑥2 ) = If 𝜎(𝑡) > 𝑡, then 𝑡 is called right-scattered, while if 𝜌(𝑡) < 𝑡,
it is called left-scattered. If 𝑡 < sup T and 𝜎(𝑡) = 𝑡, then 𝑡 is
𝑥12 + 𝑥22 .
called right-dense. If 𝑡 > inf T and 𝜌(𝑡) = 𝑡, then 𝑡 is left-
Thus, Σ and ΣT are both weakly locally observable at 𝑥 =
dense.
0. They are not weakly locally observable at any other point.
Example 27 describes a positive behavior of robust local The time-scale T is homogeneous, if 𝜇 and ] are constant
observability under discretization. In fact, the calculations are functions. When 𝜇 ≡ 0 and ] ≡ 0, then T = R or T is a
the same for all time-scales. closed interval (in particular a half-line). When 𝜇 is constant
and greater than 0, then T = 𝜇Z + 𝑐, for some 𝑐 ∈ R.
Let T 𝜅 := {𝑡 ∈ T : 𝑡 is nonmaximal or left-dense}. Thus
5. Conclusions 𝜅
T is obtained from T by removing its maximal point if this
We have shown that the methods of real analytic geometry point exists and is left-scattered.
and real algebra developed for continuous time systems may Let 𝑓 : T → R and 𝑡 ∈ T 𝜅 . The delta derivative of 𝑓 at
be used for systems on arbitrary time-scales, in particular 𝑡, denoted by 𝑓Δ (𝑡), is the real number with the property that
8 Abstract and Applied Analysis
given any 𝜀 there is a neighborhood 𝑈 = (𝑡 − 𝛿, 𝑡 + 𝛿)T such analytic functions that vanish on 𝐴. If 𝐼 is an ideal of O𝑥 , then
that 𝑍(𝐼) will denote the zero set-germ of 𝐼 (at 𝑥). Let us recall
that 𝑍(𝐼) is defined as the intersection of the set-germs 𝑍(𝜑𝑖 ),
(𝑓 (𝜎 (𝑡)) − 𝑓 (𝑠)) − 𝑓Δ (𝑡) (𝜎 (𝑡) − 𝑠) ≤ 𝜀 |𝜎 (𝑡) − 𝑠| (A.1) 𝑖 = 1, . . . , 𝑘, where 𝜑1 , . . . , 𝜑𝑘 are generators of the ideal 𝐼.
Since only finite intersections of set-germs are defined, we
for all 𝑠 ∈ 𝑈. If 𝑓Δ (𝑡) exists, then we say that 𝑓 is must use here the property that O𝑥 is Noetherian.
delta differentiable at 𝑡. Moreover, we say that 𝑓 is delta We have a natural duality between ideals and set-germs.
differentiable on T 𝑘 provided 𝑓Δ (𝑡) exists for all 𝑡 ∈ T 𝑘 . If 𝐼1 ⊂ 𝐼2 , then 𝑍(𝐼2 ) ⊂ 𝑍(𝐼1 ).
Let 𝑃 be any commutative ring with a unit and let 𝐼 be an
Example A.1. If T = R, then 𝑓Δ (𝑡) = 𝑓 (𝑡). If T = ℎZ, then ideal of 𝑃. Then the real radical of 𝐼, denoted by √ R
𝐼, is the
𝑓Δ (𝑡) = (𝑓(𝑡 + ℎ) − 𝑓(𝑡))/ℎ. If T = 𝑞N , then 𝑓Δ (𝑡) = (𝑓(𝑞𝑡) − set of all 𝑎 ∈ 𝑃 for which there is 𝑚 ∈ N, 𝑘 ∈ N ∪ {0} and
𝑓(𝑡))/((𝑞 − 1)𝑡). 𝑏1 , . . . , 𝑏𝑘 ∈ 𝑃 such that
A function 𝑓 : T → R is called rd-continuous provided
it is continuous at right-dense points in T and its left-sided 𝑎2𝑚 + 𝑏12 + ⋅ ⋅ ⋅ + 𝑏𝑘2 ∈ 𝐼. (B.1)
limits exist (finite) at left-dense points in T. If 𝑓 is continuous,
then it is rd-continuous. The real radical is an ideal in 𝑃 and it contains 𝐼. If 𝐼 is a proper
A function 𝑓 : T → R is called regressive, if 1 + ideal of 𝑃, then also √ R
𝐼 is proper. An ideal 𝐼 is called real if
𝜇(𝑡)𝑓(𝑡) ≠0 for all 𝑡 ∈ T. √𝐼 = 𝐼.
R
Research Article
Positive Solutions Using Bifurcation Techniques for
Boundary Value Problems of Fractional Differential Equations
Yansheng Liu
School of Mathematical Sciences, Shandong Normal University, Jinan 250014, China
Copyright © 2013 Yansheng Liu. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This paper is concerned with the existence of positive solutions for a class of boundary value problems of fractional differential
equations with parameter. The main tools used here are bifurcation techniques and topological degree theory. Finally, an example
is worked out to demonstrate the main result.
1. Introduction numbers 𝜆∗ and 𝜆∗∗ with 𝜆∗ < 𝜆∗∗ such that the above
system has at least two positive solutions for 𝜆 ∈ (0, 𝜆∗ ) and
During the last few decades, fractional calculus and fractional no solution for 𝜆 > 𝜆∗∗ under some suitable assumptions
differential equations have been studied extensively since such as the following.
fractional-order models are found to be more adequate
than integer-order models in some real-world problems. In (A1) There exists an interval [𝑎, 𝑏] ⊂ (0, 1) such that
fact, fractional derivatives provide an excellent tool for the lim𝑢 → +∞ 𝑓(𝑠, 𝑢)/𝑢 = +∞ uniformly with respect to
description of memory and hereditary properties of various 𝑠 ∈ [𝑎, 𝑏].
materials and processes. The mathematical modeling of sys-
tems and processes in the fields of physics, chemistry, aerody- In [9], Bai and Lü consider the following nonlinear frac-
namics, electrodynamics of complex medium, polymer rhe- tional differential equation Dirichlet-type boundary value
ology, and so forth involves derivatives of fractional order. problem:
For details and examples, see [1–7] and references therein.
Recently, there have been a few papers which deal with the 𝐷0𝛼+ 𝑢 (𝑡) + 𝑓 (𝑡, 𝑢 (𝑡)) = 0, 𝑡 ∈ (0, 1) ,
boundary value problem for fractional differential equation. (2)
For example, in [8], Tian and Liu investigated the following 𝑢 (0) = 𝑢 (1) = 0,
singular fractional boundary value problem (BVP, for short)
of the form where 1 < 𝛼 ≤ 2 is a real number and 𝐷0𝛼+ is the standard
Riemann-Liouville differentiation. The corresponding Green
𝐶
𝐷0𝛼+ 𝑢 (𝑡) + 𝜆𝑓 (𝑡, 𝑢 (𝑡)) = 0, 0 < 𝑡 < 1, function is derived. By means of some fixed point theorems
on cone, the existence and multiplicity of positive solutions
𝑢(𝑗) (0) = 0, 0 ≤ 𝑗 ≤ 𝑛 − 1, 𝑗 ≠2, (1) for BVP (2) were investigated.
In [10], Jiang and Yuan further investigated BVP (2).
𝑢 (1) = 0, Comparing with [9], they deduced some new properties of
the Green function, which extended the results of integer-
where 𝐶𝐷0𝛼+ is Caputo’s fractional derivatives, 𝑛 − 1 < 𝛼 ≤ 𝑛, order Dirichlet boundary value problems. Based on these new
𝑛 ≥ 4, and 𝑓 : (0, 1) × (0, +∞) → [0, +∞) is continuous; that properties and Krasnoselskii fixed point theorem, the exis-
is, 𝑓(𝑡, 𝑢) may be singular at 𝑡 = 0, 1 and 𝑢 = 0. By con- tence and multiplicity of positive solutions for BVP (2) were
structing a special cone, they obtained that there exist positive considered.
2 Abstract and Applied Analysis
In this paper, by using bifurcation techniques, we con- Then there exists a connected component C of T containing
sider the following boundary value problem of fractional dif- [𝑎, 𝑏] × 0 in R × 𝑉, and either
ferential equation:
(i) C is unbounded in R × 𝑉 or
𝐷0𝛼+ 𝑢 (𝑡) + 𝜂𝑓 (𝑡, 𝑢 (𝑡)) = 0, 𝑡 ∈ (0, 1) , (ii) C ∩ [(R \ [𝑎, 𝑏]) × 0] ≠0.
(3)
𝑢 (0) = 𝑢 (1) = 0,
Lemma 2 (Schmitt [16]). Let 𝑉 be a real reflexive Banach
where 1 < 𝛼 ≤ 2, 𝐷0𝛼+is the standard Riemann-Liouville space. Let 𝐺 : R × 𝑉 to 𝑉 be completely continuous, and let
differentiation, 𝑟 > 0 is a given constant, and 𝑓 : [0, 1] × 𝑎, 𝑏 ∈ R (𝑎 < 𝑏) be such that the solution of (4) is, a priori,
R+ → R+ is a given continuous function satisfying some bounded in 𝑉 for 𝜆 = 𝑎 and 𝜆 = 𝑏; that is, there exists an 𝑅 > 0
assumptions that will be specified later. such that
It is remarkable that the method used in references men-
tioned above was fixed point theorems and the same kind of 𝐺 (𝑎, 𝑢) ≠𝑢 ≠𝐺 (𝑏, 𝑢) (7)
conditions was used such that the nonlinearity 𝑓(𝑡, 𝑢) satisfies
for all 𝑢 with ‖𝑢‖ ≥ 𝑅. Furthermore, assume that
superlinear or sublinear condition at 0 and ∞, which is sim-
ilar to (A1). To the best of our knowledge, there is no paper deg (𝐼 − 𝐺 (𝑎, ⋅) , 𝐵𝑅 (0) , 0) ≠deg (𝐼 − 𝐺 (𝑏, ⋅) , 𝐵𝑅 (0) , 0) ,
studying such fractional differential equations using bifurca- (8)
tion ideas. As we know, the bifurcation technique is widely
used in solving BVP of integer-order differential equations for 𝑅 > 0 sufficiently large. Then there exists a closed connected
(see, e.g., [11–13] and references therein). In [14], by virtue set C of solutions of (4) that is unbounded in [𝑎, 𝑏] × 𝑉, and
of bifurcation ideas, the authors studied a kind of BVP of either
differential inclusions. The purpose of present paper is to
fill this gap. By using bifurcation techniques and topological (i) C is unbounded in 𝜆 direction or
degree theory, the existence of positive solutions of BVP (3) is (ii) there exists an interval [𝑐, 𝑑] such that (𝑎, 𝑏)∩(𝑐, 𝑑) = 0
investigated. The main features of present paper are as follows. and C bifurcates from infinity in [𝑐, 𝑑] × 𝑉.
First, the nonlinearity 𝑓(𝑡, 𝑢) is asymptotically linear at 0 and
∞, not super-linear or sub-linear (see the condition (H1) in Lemma 3 (Guo [17]). Let Ω be a bounded open set of real
Section 2 and example in Section 4). Next, the main method Banach space 𝐸, and let 𝐴 : Ω → 𝐸 be completely continuous.
used here is bifurcation techniques and topological degree, If there exists 𝑦0 ∈ 𝐸, 𝑦0 ≠𝜃 such that
not fixed point theorem on cone, which is different from the
references. 𝑥 ∈ 𝜕Ω, 𝜏 ≥ 0 ⇒ 𝑥 − 𝐴𝑥 ≠𝜏𝑦0 , (9)
The paper is organized as described below. At the end of
this section, for completeness, we list some results on bifur- then
cation theory from interval and topological degree of com-
deg (𝐼 − 𝐴, Ω, 𝜃) = 0. (10)
pletely continuous operators. Section 2 contains background
materials and preliminaries. In Section 3, by using bifurcation
techniques and topological degree theory, bifurcation results 2. Background Materials and Preliminaries
from infinity and trivial solution are established. Then the
main results of present paper are given and proved. Finally in For convenience, we present some necessary definitions and
Section 4, an example is worked out to demonstrate the main results on fractional calculus theory (see [6]).
results.
Definition 4. The fractional (arbitrary) order integral of the
Lemma 1 (Schmitt and Thompson [15]). Let 𝑉 be a real reflex- function ℎ ∈ 𝐿1 ([𝑎, 𝑏]) of order 𝛼 ∈ R+ is defined by
ive Banach space. Let 𝐺 : R × 𝑉 to 𝑉 be completely continuous
such that 𝐺(𝜆, 0) = 0, for all 𝜆 ∈ R. Let 𝑎, 𝑏 ∈ R (𝑎 < 𝑏) be 𝛼
𝑡
(𝑡 − 𝑠)𝛼−1
𝐼𝑎+ ℎ (𝑡) = ∫ ℎ (𝑠) 𝑑𝑠, (11)
such that 𝑢 = 0 is an isolated solution of the equation 𝑎 Γ (𝛼)
Lemma 6. Let 𝛼 > 0; then, the differential equation Lemma 9 (see [10]). The function 𝐺(𝑡, 𝑠) defined by (18) has
the following properties.
𝐷0𝛼+ 𝑢 (𝑡) = 0. (13)
(i) 𝐺(𝑡, 𝑠) > 0, ∀𝑡, 𝑠 ∈ (0, 1).
has solutions 𝑢(𝑡) = 𝑐1 𝑡𝛼−1 + 𝑐2 𝑡𝛼−2 + ⋅ ⋅ ⋅ + 𝑐𝑛 𝑡𝛼−𝑛 , for some (ii) The function 𝐺∗ (𝑡, 𝑠) =: 𝑡2−𝛼 𝐺(𝑡, 𝑠) has the following
𝑐𝑖 ∈ R, 𝑖 = 0, 1, 2, . . . , 𝑛, where 𝑛 is the smallest integer greater properties:
than or equal to 𝛼.
𝛼−1
𝑡 (1 − 𝑡) 𝑠(1 − 𝑠)𝛼−1 ≤ 𝐺∗ (𝑡, 𝑠)
Notice that 𝐷0𝛼+ 𝐼𝛼 ℎ(𝑡) = ℎ(𝑡) for all ℎ ∈ 𝐶(0, 1) ∩ 𝐿(0, 1). Γ (𝛼)
From Lemma 6, we deduce the following result.
1 (19)
≤ 𝑠(1 − 𝑠)𝛼−1 ,
1
Lemma 7. Assume that 𝑢 ∈ 𝐶(0, 1)∩𝐿 [0, 1] with a derivative Γ (𝛼)
of order 𝑛 that belongs to 𝐶(0, 1) ∩ 𝐿1 [0, 1]. Then ∀𝑡, 𝑠 ∈ [0, 1] .
𝐼0𝛼+ 𝐷0𝛼+ 𝑢 (𝑡) = 𝑢 (𝑡) + 𝑐1 𝑡 𝛼−1 𝛼−2
+ 𝑐2 𝑡 𝛼−𝑛
+ ⋅ ⋅ ⋅ + 𝑐𝑛 𝑡 , (14)
The basic space used in this paper is
for some 𝑐𝑖 ∈ R, 𝑖 = 0, 1, 2, . . . , 𝑛, where 𝑛 is the smallest integer 𝐸 =: {𝑢 ∈ 𝐶 [0, 1] : 𝑢 (0) = 𝑢 (1) = 0} . (20)
greater than or equal to 𝛼.
Obviously, 𝐸 is a Banach space with norm ‖𝑢‖ = max𝑡∈𝐽 |𝑢(𝑡)|
For more detailed results of fractional calculus, we refer (for all 𝑢 ∈ 𝐸).
the reader to [6]. Let
Now let us list the following assumption satisfied
throughout the paper. 𝑄 := {V ∈ 𝐸 : V (𝑡) ≥ (𝛼 − 1) 𝑡 (1 − 𝑡) V (𝑠) ≥ 0, ∀𝑠, 𝑡 ∈ (0, 1)} .
(21)
(H1) There exist two positive numbers 𝑟, 𝑅 with 𝑟 < 𝑅 and
functions 𝑎0 , 𝑎0 , 𝑏∞ , 𝑎0 ∈ 𝐶(𝐽, R+ ) with 𝑎0 (𝑡), 𝑎0 (𝑡), It is easy to see that 𝑄 is a cone of 𝐸. Moreover, from (21), we
𝑏∞ (𝑡), 𝑎0 (𝑡) ≢ 0 in any subinterval of [0, 1] such that have, for all V ∈ 𝑄,
𝑓 (𝑡, 𝑢) ⊂ [𝑎0 (𝑡) (𝑢 − 𝜉1 (𝑡, 𝑢)) , 𝑎0 (𝑡) (𝑢 + 𝜉2 (𝑡, 𝑢))] , V (𝑡) ≥ (𝛼 − 1) 𝑡 (1 − 𝑡) ‖V‖ , ∀𝑡 ∈ [0, 1] . (22)
For the sake of using bifurcation technique to investigate
∀ (𝑡, 𝑢) ∈ 𝐽 × [0, 𝑟] ,
BVP (3), we study the following fractional boundary value
𝑓 (𝑡, 𝑢) ⊂ [𝑏∞ (𝑡) (𝑢 − 𝜁1 (𝑡, 𝑢)) , 𝑏∞ (𝑡) (𝑢 + 𝜁2 (𝑡, 𝑢))] , problem with parameter 𝜆:
𝐷0𝛼+ 𝑦 (𝑡) + 𝜆𝑓 (𝑡, 𝑦 (𝑡)) = 0, 𝑡 ∈ (0, 1) ,
∀ (𝑡, 𝑢) ∈ 𝐽 × [𝑅, +∞) , (23)
(15) 𝑦 (0) = 𝑦 (1) = 0.
A function (𝜆, 𝑢) is said to be a solution of BVP (23) if
where 𝐽 = [0, 1], 𝜉𝑖 , 𝜁𝑖 ∈ 𝐶(𝐽 × R+ ) with 𝜉𝑖 (𝑡, 𝑡𝛼−2 𝑢) = 𝑜(𝑡𝛼−2 𝑢) (𝜆, 𝑢) satisfies (23). In addition, if 𝜆 > 0, 𝑢(𝑡) > 0 for 𝑡 ∈
as 𝑢 → 0 uniformly with respect to 𝑡 ∈ [0, 1] (𝑖 = 1, 2), and (0, 1), then (𝜆, 𝑢) is said to be a positive solution of BVP (23).
𝜁𝑖 (𝑡, 𝑡𝛼−2 𝑢) = 𝑜(𝑡𝛼−2 𝑢) as 𝑢 → +∞ uniformly with respect to Obviously, if 𝜆 > 0, 𝑢 ∈ 𝑄 \ {𝜃} is a solution of BVP (23), then
𝑡 ∈ [0, 1] (𝑖 = 1, 2). by (22), we know that (𝜆, 𝑢) is a positive solution of BVP (23),
To solve BVP (3), we first consider the following linear where 𝜃 denotes the zero element of Banach space 𝐸.
boundary problem of fractional differential equation: Define
𝐷0𝛼+ 𝑢 (𝑡) + 𝑔 (𝑡) = 0, 𝑡 ∈ (0, 1) , 𝑓 (𝑡, 𝑢) , (𝑡, 𝑢) ∈ 𝐽 × R+ ,
(16) 𝑓 (𝑡, 𝑢) = { (24)
𝑢 (0) = 𝑢 (1) = 0, 𝑓 (𝑡, 0) , (𝑡, 𝑢) ∈ 𝐽 × (−∞, 0) .
where 𝑔 ∈ 𝐶[0, 1]. We cite the following two lemmas from Then 𝑓(𝑡, 𝑢) ≥ 0 on 𝐽 × R. Let
references. 1
𝐴 𝜆 V (𝑡) =: 𝜆 ∫ 𝐺∗ (𝑡, 𝑠) 𝑓 (𝑠, 𝑠𝛼−2 V (𝑠)) 𝑑𝑠, ∀V ∈ 𝑄. (25)
Lemma 8 (see [9]). Given 𝑔 ∈ 𝐶[0, 1], then 0
Lemma 11. The operator 𝐿 𝑎 defined by (28) is completely con- ∀ (𝑡, 𝑢) ∈ 𝐽 × [0, 𝑟] .
tinuous and has a unique characteristic value 𝜆 1 (𝑎), which is (33)
positive, real, and simple and the corresponding eigenfunction Therefore, by virtue of (25), we know
𝜙(𝑡) is of one sign in (0, 1); that is, 𝜙(𝑡) = 𝜆 1 (𝑎)𝐿 𝑎 𝜙(𝑡) for all
𝑡 ∈ 𝐽. 1
𝑤𝑛 (𝑡) ≤ 𝜇𝑛 ∫ 𝐺∗ (𝑡, 𝑠) 𝑎0 (𝑠) (𝑠𝛼−2 𝑤𝑛 (𝑠)
Notice that the operator 𝐿 𝑎 can be regarded as 𝐿 𝑎 : 0
Consequently, 𝜆 1 (𝑎0 ) ≤ 𝜇 ≤ 𝜆 1 (𝑎0 ), which contradicts Let 𝜓∗ be the positive eigenfunction of 𝐿∗𝑎0 corresponding to
𝜇 ∈ [𝑐, 𝑑]. Therefore, there exists 𝛿1 ∈ (0, 𝑟) such that 𝜆 1 (𝑎0 ). Then
V ≠𝐴 𝜆 V, ∀𝜆 ∈ [𝑐, 𝑑] , ∀V ∈ 𝐸 with 0 < ‖V‖ ≤ 𝛿1 . (38)
⟨V𝑛 , 𝜓∗ ⟩ ≥ 𝜆 (1 − 𝜎) ⟨𝐿 𝑎0 V𝑛 , 𝜓∗ ⟩
= 𝜆 (1 − 𝜎) ⟨V𝑛 , 𝐿∗𝑎0 𝜓∗ ⟩
0 (47)
Lemma 15. For 𝜇 ∈ (0, 𝜆 1 (𝑎 )), there exists 𝛿1 ∈ (0, 𝑟) such
that 𝜓∗
= 𝜆 (1 − 𝜎) ⟨V𝑛 , ⟩.
deg (𝐼 − 𝐴 𝜇 , 𝐵𝛿 , 0) = 1, ∀𝛿 ∈ (0, 𝛿1 ] . (39) 𝜆 1 (𝑎0 )
This together with ⟨V𝑛 , 𝜓∗ ⟩ > 0 guarantees that
Proof. Notice that [0, 𝜇] ∩ [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] = 0. From
Lemma 14, there exists 𝛿1 ∈ (0, 𝑟) such that 𝜆 (1 − 𝜎) ≤ 𝜆 1 (𝑎0 ) , (48)
V ≠𝐴 𝜆 V, ∀𝜆 ∈ [0, 𝜇] , ∀V ∈ 𝐶 [0, 1] with 0 < ‖V‖ ≤ 𝛿1 ,
which is a contradiction. Therefore, (44) holds. By Lemma 3,
(40)
for each 𝜆 > 𝜆 1 (𝑎0 ), there exists 𝛿2 > 0 such that
which means
deg (𝐼 − 𝐴 𝜆 , 𝐵𝛿 , 0) = 0, ∀𝛿 ∈ (0, 𝛿2 ] . (49)
V ≠𝜏𝐴 𝜇 V, ∀𝜏 ∈ [0, 1] , ∀V ∈ 𝐶 [0, 1] with 0 < ‖V‖ ≤ 𝛿1 .
(41)
Therefore, by the homotopy invariance of topological Theorem 17. [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] is a bifurcation interval of
degree, we have positive solutions from the trivial solution for BVP (23); that is,
there exists an unbounded component C0 of positive solutions
deg (𝐼 − 𝐴 𝜇 , 𝐵𝛿 , 0) = deg (𝐼, 𝐵𝛿 , 0) = 1, ∀𝛿 ∈ (0, 𝛿1 ] .
of BVP (23), which meets [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] × {0}. Moreover,
(42) there exists no bifurcation interval of positive solutions from the
trivial solution which is disjointed with [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )].
Lemma 16. For 𝜆 > 𝜆 1 (𝑎0 ), there exists 𝛿2 ∈ (0, 𝑟) such that Proof. By virtue of (27) and Lemma 10, we need only to prove
that there exists an unbounded component C0 of Σ, which
deg (𝐼 − 𝐴 𝜆 , 𝐵𝛿 , 0) = 0, ∀𝛿 ∈ (0, 𝛿2 ] . (43) meets [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] × {0}, and there exists no bifurcation
Proof. First we prove that for 𝜆 > 𝜆 1 (𝑎0 ), there exists 𝛿2 ∈ interval of Σ from the trivial solution which is disjointed with
(0, 𝑟) such that [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )].
For fixed 𝑛 ∈ N with 𝜆 1 (𝑎0 ) − (1/𝑛) > 0, by Lemmas 15
V − 𝐴 𝜆 V ≠𝜏𝜑0 , ∀𝜏 ≥ 0, ∀V ∈ 𝐶 [0, 1] with 0 < ‖V‖ ≤ 𝛿2 , and 16 and their proof, there exists 𝑟 > 0 such that all of the
(44) conditions of Lemma 1 are satisfied with 𝐺(𝜆, 𝑢) = 𝐴 𝜆 𝑢,
where 𝜑0 is the positive eigenfunctions of 𝐿 𝑎0 corresponding 𝑎 = 𝜆 1 (𝑎0 ) − (1/𝑛), and 𝑏 = 𝜆 1 (𝑎0 ) + (1/𝑛). This together with
to 𝜆 1 (𝑎0 ). Lemma 10 guarantees that there exists a closed connected set
If this is false, then there exist V𝑛 ∈ 𝐶[0, 1] with ‖V𝑛 ‖ → 0 C𝑛 of Σ containing [𝜆 1 (𝑎0 ) − (1/𝑛), 𝜆 1 (𝑎0 ) + (1/𝑛)] × 0 in
(𝑛 → +∞) and 𝜏𝑛 ≥ 0 such that R+ ×𝐶[0, 1]. From Lemma 14, the case (ii) of Lemma 1 cannot
occur. Thus, C𝑛 bifurcates from [𝜆 1 (𝑎0 )−(1/𝑛), 𝜆 1 (𝑎0 )+1/𝑛]×
V𝑛 − 𝐴 𝜆 V𝑛 = 𝜏𝑛 𝜑0 . (45) 0 and is unbounded in R+ × 𝐶[0, 1]. Moreover, for any closed
By Lemma 10, we have V𝑛 (𝑡) > 0 in (0, 1). From 𝜆 > interval [𝑐, 𝑑] ⊂ [𝜆 1 (𝑎0 ) − 1/𝑛, 𝜆 1 (𝑎0 ) + 1/𝑛] \ [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )],
𝜆 1 (𝑎0 ), there exists 𝜎 > 0 satisfying 𝜆 > (1 − 𝜎)𝜆 1 (𝑎0 ). Then by Lemma 14, there exists 𝛿1 > 0 such that the set {V ∈
condition (H1) guarantees that there exists 𝛿 ∈ (0, 𝑟) such 𝐶[0, 1] : (𝜆, V) ∈ Σ, 0 < ‖V‖ ≤ 𝛿1 , 𝜆 ∈ [𝑐, 𝑑]} = 0.
Therefore, C𝑛 must be bifurcated from [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )] × {0},
that |𝜉1 (𝑡, 𝑡𝛼−2 𝑢)| < 𝜎𝑡𝛼−2 𝑢 for 𝑢 ∈ (0, 𝛿). Noticing ‖V𝑛 ‖ →
which implies that C𝑛 can be regarded as C0 . In addition,
0(𝑛 → +∞), there exists 𝑁 > 0 such that ‖V𝑛 ‖ < 𝛿 for 𝑛 > 𝑁.
using Lemma 14 again, there exists no bifurcation interval of
Consequently, by virtue of (25) and (45), for 𝑛 > 𝑁, we know
positive solutions from the trivial solution which is disjointed
V𝑛 (𝑡) = 𝐴 𝜆 V𝑛 (𝑡) + 𝜏𝑛 𝜑0 with [𝜆 1 (𝑎0 ), 𝜆 1 (𝑎0 )].
1 By a process similar to the above, one can obtain the fol-
≥ 𝜆 ∫ 𝐺∗ (𝑡, 𝑠) 𝑎0 (𝑠) lowing conclusions.
0
× (𝑠𝛼−2 V𝑛 (𝑠) − 𝜉1 (𝑠, 𝑠𝛼−2 V𝑛 (𝑠))) 𝑑𝑠 (46) Lemma 18. Let [𝑐, 𝑑] ⊂ R+ be a compact interval with
[𝜆 1 (𝑏∞ ), 𝜆 1 (𝑏∞ )] ∩ [𝑐, 𝑑] = 0. Then there exists 𝑅1 > 𝑅 such
1
that
≥ 𝜆 (1 − 𝜎) ∫ 𝐺∗ (𝑡, 𝑠) 𝑎0 (𝑠) 𝑠𝛼−2 V𝑛 (𝑠) 𝑑𝑠
0
𝑢 ≠𝐴 𝜆 𝑢, ∀𝜆 ∈ [𝑐, 𝑑] , ∀𝑢 ∈ 𝐶 [0, 1] with ‖𝑢‖ ≥ 𝑅1 .
= 𝜆 (1 − 𝜎) 𝐿 𝑎0 V𝑛 (𝑡) . (50)
6 Abstract and Applied Analysis
Lemma 19. For 𝜇 ∈ (0, 𝜆 1 (𝑏∞ )), there exists 𝑅1 > 𝑅 such that In fact, from assumption (H2), it follows that there exists
𝜀 > 0 such that
deg (𝐼 − 𝐴 𝜇 , 𝐵𝑅 , 0) = 1, ∀𝑅 ≥ 𝑅1 . (51)
𝜂+𝜀 1
∫ [𝑠 (1 − 𝑠)]𝛼−1 ℎ (𝑠) 𝑑𝑠 < 1. (55)
Lemma 20. For 𝜆 > 𝜆 1 (𝑏∞ ), there exists 𝑅2 > 𝑅 such that Γ (𝛼) 0
Research Article
Approximation of Eigenvalues of Sturm-Liouville Problems by
Using Hermite Interpolation
Copyright © 2013 M. M. Tharwat and S. M. Al-Harbi. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
Eigenvalue problems with eigenparameter appearing in the boundary conditions usually have complicated characteristic
determinant where zeros cannot be explicitly computed. In this paper, we use the derivative sampling theorem “Hermite
interpolations” to compute approximate values of the eigenvalues of Sturm-Liouville problems with eigenvalue parameter in one
or two boundary conditions. We use recently derived estimates for the truncation and amplitude errors to compute error bounds.
Also, using computable error bounds, we obtain eigenvalue enclosures. Also numerical examples, which are given at the end of the
paper, give comparisons with the classical sinc method and explain that the Hermite interpolations method gives remarkably better
results.
where 𝑓𝑁(𝑡) is the truncated series as follows: 0 < ] ≤ 1, then for 0 < 𝜀 ≤ min{𝜋/𝜎, 𝜎/𝜋, 1/√𝑒}, we have,
[22],
𝑛𝜋 2
𝑓𝑁 (𝑡) = ∑ [𝑓 ( ) 𝑆𝑛 (𝑡)
𝜎 A (𝜀, 𝑓)∞
|𝑛|≤𝑁
(4)
𝑛𝜋 sin (𝜎𝑡 − 𝑛𝜋) 4𝑒1/4 𝜋
+𝑓 ( ) 𝑆𝑛 (𝑡)] . ≤ {√3𝑒 (1 + 𝜎) + (( ) 𝐴 + 𝑀𝑓 ) 𝜌 (𝜀)
𝜎 𝜎 𝜎 (] + 1) 𝜎
1
It is proved in [22] that if 𝑓(𝑡) ∈ PW2𝜎 and 𝑓(𝑡) is sufficiently + (𝜎 + 2 + log (2)) 𝑀𝑓 } 𝜀 log ( ) ,
𝜀
smooth in the sense that there exists 𝑘 ∈ Z+ such that 𝑡𝑘 𝑓(𝑡) ∈ (10)
𝐿2 (R), then for 𝑡 ∈ R, |𝑡| < 𝑁𝜋/𝜎, we have
where
𝑅𝑁 (𝑓) (𝑡)
3𝜎 𝜎 ]
𝐴 := (𝑓 (0) + 𝑀𝑓 ( ) ) ,
≤ 𝑇𝑁,𝑘,𝜎 (𝑡) 𝜋 𝜋
(11)
1
𝜉𝑘,𝜎 𝐸𝑘 |sin 𝜎𝑡|2 1 1 𝜌 (𝜀) := 𝛾 + 10 log ( ) ,
:= ( + ) 𝜀
√3(𝑁 + 1)𝑘 (𝑁𝜋 − 𝜎𝑡)3/2 (𝑁𝜋 + 𝜎𝑡)3/2
and 𝛾 := lim𝑛 → ∞ [∑𝑛𝑘=1 1/𝑘 − log 𝑛] ≅ 0.577216 is the Euler-
𝜉𝑘,𝜎 (𝜎𝐸𝑘 + 𝑘𝐸𝑘−1 ) |sin 𝜎𝑡|2 Mascheroni constant.
+
𝜎(𝑁 + 1)𝑘 The classical [23] sampling theorem of Whittaker, Kotel-
nikov, and Shannon (WKS) for 𝑓 ∈ PW2𝜎 is the series
1 1
×( + ), representation as follows:
√𝑁𝜋 − 𝜎𝑡 √𝑁𝜋 + 𝜎𝑡
(5) ∞
𝑛𝜋
𝑓 (𝑡) = ∑ 𝑓 ( ) 𝑆𝑛 (𝑡) , 𝑡 ∈ R, (12)
𝑛=−∞ 𝜎
where the constants 𝐸𝑘 and 𝜉𝑘,𝜎 are given by
where the convergence is absolute and uniform on R and
∞
𝑘 2 𝜎𝑘+1/2 it is uniform on compact sets of C cf. [23–25]. Series (12),
𝐸𝑘 := √ ∫ 𝑡 𝑓 (𝑡) 𝑑𝑡, 𝜉𝑘,𝜎 := . (6) which is of Lagrange interpolation type, has been used to
−∞ 𝜋𝑘+1 √1 − 4−𝑘 compute eigenvalues of second-order eigenvalue problems,
The amplitude error occurs when approximate samples are see for example, [17, 26–29]. The use of (12) in numerical
used instead of the exact ones, which we cannot compute. It analysis is known as the sinc method established by Stenger
is defined to be et al., cf. [30–32]. The aim of this paper is to investigate
the possibilities of using Hermite interpolations rather than
A (𝜀, 𝑓) (𝑡) Lagrange interpolations, to compute the eigenvalues numer-
ically. Notice that, due to Paley-Wiener’s theorem [33] 𝑓 ∈
∞
𝑛𝜋 𝑛𝜋 PW2𝜎 if and only if there is 𝑔(⋅) ∈ 𝐿2 (−𝜎, 𝜎) such that
= ∑ [{𝑓 ( ) − 𝑓̃ ( )} 𝑆𝑛2 (𝑡)
𝑛=−∞ 𝜎 𝜎 𝜎
1
𝑓 (𝑡) = ∫ 𝑔 (𝑥) 𝑒𝑖𝑥𝑡 𝑑𝑥. (13)
𝑛𝜋 ̃ ( 𝑛𝜋 )} sin (𝜎𝑡 − 𝑛𝜋) 𝑆 (𝑡)] , √2𝜋 −𝜎
+ {𝑓 ( )−𝑓 𝑛
𝜎 𝜎 𝜎
Therefore, 𝑓 (𝑡) ∈ PW2𝜎 , that is, 𝑓 (𝑡) also has an expansion of
𝑡 ∈ R,
the form (12). However, 𝑓 (𝑡) can also be obtained by term-
(7) by-term differentiation formula of (12) as follows:
̃
where 𝑓(𝑛𝜋/𝜎) and 𝑓 ̃ (𝑛𝜋/𝜎) are approximate samples of ∞
𝑛𝜋
𝑓(𝑛𝜋/𝜎) and 𝑓 (𝑛𝜋/𝜎), respectively. Let us assume that the 𝑓 (𝑡) = ∑ 𝑓 ( ) 𝑆𝑛 (𝑡) , (14)
𝑛=−∞ 𝜎
differences 𝜀𝑛 := 𝑓(𝑛𝜋/𝜎) − 𝑓(𝑛𝜋/𝜎), ̃ 𝜀𝑛 := 𝑓 (𝑛𝜋/𝜎) −
̃ (𝑛𝜋/𝜎), and 𝑛 ∈ Z are bounded by a positive number 𝜀,
𝑓 see [23, page 52] for convergence. Thus, the use of Hermite
that is, |𝜀𝑛 |, |𝜀𝑛 | ≤ 𝜀. If 𝑓(𝑡) ∈ PW2𝜎 satisfies the natural decay interpolations will not cost any additional computational
conditions efforts since the samples 𝑓(𝑛𝜋/𝜎) will be used to compute
both 𝑓(𝑡) and 𝑓 (𝑡) according to (12) and (14), respectively.
𝑛𝜋 𝑛𝜋
𝜀𝑛 ≤ 𝑓 ( ) , Now, we consider the following differential equations:
𝜀𝑛 ≤ 𝑓 ( ) , (8)
𝜎 𝜎
𝑀𝑓 ℓ (𝑦) := −𝑦 (𝑥, 𝜇) + 𝑞 (𝑥) 𝑦 (𝑥, 𝜇) = 𝜇2 𝑦 (𝑥, 𝜇) ,
𝑓 (𝑡) ≤ ]+1 , 𝑡 ∈ R − {0} , (9) (15)
|𝑡| 𝑥 ∈ [0, 1] ,
Abstract and Applied Analysis 3
with the following boundary conditions: These zeros are real and simple. The function Δ(𝜇) is an
entire function of 𝜇. We aim to approximate Δ(𝜇) and hence
𝑎1 𝑦 (0, 𝜇) + 𝑎2 𝑦 (0, 𝜇) = 𝜇2 (𝑎1 𝑦 (0, 𝜇) + 𝑎2 𝑦 (0, 𝜇)) , (16) its zeros, that is, the eigenvalues by the use of the Hermite
Interpolation. The idea is to split Δ(𝜇) into two parts, one
𝑏1 𝑦 (1, 𝜇) + 𝑏2 𝑦 (1, 𝜇) = 𝜇2 (𝑏1 𝑦 (1, 𝜇) + 𝑏2 𝑦 (1, 𝜇)) , (17) is known and the other is unknown, but lies in a Paley-
Wiener space. Then we approximate the unknown part to
where 𝜇 is a complex spectral parameter, 𝑞(⋅) is assumed to get the approximate Δ(𝜇) and then compute the approximate
be real valued and continuous on [0, 1], and 𝑎𝑖 , 𝑏𝑖 , 𝑎𝑖 , 𝑏𝑖 ∈ R, zeros. Using the method of variation of constants, the solution
𝑖 = 0, 1 satisfying 𝑦(𝑥, 𝜇) satisfies Volterra integral equation as follows:
((𝑎1 , 𝑎2 ) = (0, 0) or 𝑎1 𝑎2 − 𝑎1 𝑎2 > 0) , 𝑦 (𝑥, 𝜇) = (𝑎2 − 𝑎2 𝜇2 ) cos 𝜇𝑥
(18) (21)
((𝑏1 , 𝑏2 ) = (0, 0) or 𝑏1 𝑏2 − 𝑏1 𝑏2 > 0) . sin 𝜇𝑥
− (𝑎1 − 𝑎1 𝜇2 ) + 𝑇 [𝑦] (𝑥, 𝜇) ,
𝜇
The eigenvalue problem (15)–(17) will be denoted by
where 𝑇 is the Volterra operator defined by
Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ) when (𝑎1 , 𝑎2 ) ≠ (𝑏1 , 𝑏2 ). It is a Sturm-
(0, 0) ≠
Liouville problem when the eigenparameter 𝜇 appears lin- 𝑥 sin 𝜇 (𝑥 − 𝑡)
early in both boundary conditions. The classical problem 𝑇 [𝑦] (𝑥, 𝜇) = ∫ 𝑞 (𝑡) 𝑦 (𝑡, 𝜇) 𝑑𝑡. (22)
0 𝜇
when 𝑎1 = 𝑎2 = 𝑏1 = 𝑏2 = 0, which we denote by Π(𝑞, 𝑎, 𝑏, 0,
0) has a countable set of real and simple eigenvalues with ∞ Differentiating (21), we get
as the only possible limit point, [34, 35]. In [14], the authors
used Hermite-type sampling series (1) to compute the eigen- 𝑦 (𝑥, 𝜇) = (𝑎2 𝜇2 − 𝑎2 ) 𝜇 sin 𝜇𝑥
values of problem Π(𝑞, 𝑎, 𝑏, 0, 0) numerically. In [36], see (23)
̃ [𝑦] (𝑥, 𝜇) ,
+ (𝑎1 𝜇2 − 𝑎1 ) cos 𝜇𝑥 + 𝑇
also [37], Annaby and Tharwat proved that Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 )
has a denumerable set of real and simple eigenvalues with
̃ is the Volterra operator
where 𝑇
∞ as the limit point using techniques similar of those
established in [38–40], where also sampling theorems have 𝑥
been established. Similar results are established in [38] for the ̃ [𝑦] (𝑥, 𝜇) = ∫ cos 𝜇 (𝑥 − 𝑡) 𝑞 (𝑡) 𝑦 (𝑡, 𝜇) 𝑑𝑡.
𝑇 (24)
0
problem when the eigenparameter appears in one condition,
that is, when 𝑎1 = 𝑎2 = 0, (𝑏1 , 𝑏2 ) ≠ (0, 0) or equivalently Define 𝑓(⋅, 𝜇) and 𝑔(⋅, 𝜇) to be
when (𝑎1 , 𝑎2 ) ≠ (0, 0) and 𝑏1 = 𝑏2 = 0. These problems will ̃ [𝑦] (𝑥, 𝜇) . (25)
𝑓 (𝑥, 𝜇) := 𝑇 [𝑦] (𝑥, 𝜇) , 𝑔 (𝑥, 𝜇) := 𝑇
be denoted by Π(𝑞, 𝑎, 𝑏, 0, 𝑏 ), Π(𝑞, 𝑎, 𝑏, 𝑎 , 0), respectively.
The aim of the present work is to compute the eigenval- In the following, we will make use of the estimates [44] as
ues of Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ), Π(𝑞, 𝑎, 𝑏, 0, 𝑏 ), and Π(𝑞, 𝑎, 𝑏, 𝑎 , 0) follows:
numerically by the Hermite interpolations with an error sin 𝑧
analysis. This method is based on sampling theorem, Hermite |cos 𝑧| ≤ 𝑒|I𝑧| , ≤ 𝑐0 𝑒|I𝑧| ,
(26)
interpolations, but applied to regularized functions. Hence, 𝑧 1 + |𝑧|
avoiding any (multiple) integration and keeping the number where 𝑐0 is some constant (we may take 𝑐0 ≃ 1.72). For
of terms in the Cardinal series manageable. It has been convenience, we define the constants by
demonstrated that the method is capable of delivering higher
1
order estimates of the eigenvalues at a very low cost, see [41–
𝜏 := ∫ 𝑞 (𝑡) 𝑑𝑡, 𝑐1 := 𝑎2 + 𝑐0 𝑎1 ,
43]. In Sections 2 and 3 we derive the Hermite interpolation 0
technique to compute the eigenvalues of Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ) and
Π(𝑞, 𝑎, 𝑏, 0, 𝑏 ) with error estimates, respectively. The last 𝑐2 := 𝑎2 + 𝑐0 𝑎1 , 𝑐3 := 𝑐0 𝜏,
section involves some illustrative examples.
𝑐4 := exp 𝑐3 , 𝑐5 := max {𝑐1 , 𝑐2 , 𝑏1 + 𝑏2 𝜏, 𝑏1 + 𝑏2 𝜏} .
2. Treatment of Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ) (27)
From (21) and (25), we get
In this section, we derive approximate values of the eigenval-
ues of Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ). Let 𝑦(⋅, 𝜇) denote the solution of (15) 𝑓 (𝑥, 𝜇)
satisfying the following initial conditions:
𝑥 sin 𝜇 (𝑥 − 𝑡)
=∫ 𝑞 (𝑡) [ (𝑎2 − 𝑎2 𝜇2 ) cos 𝜇𝑡
𝑦 (0, 𝜇) = 𝑎2 − 𝑎2 𝜇2 ,
𝑦 (0, 𝜇) = 𝑎1 𝜇2 − 𝑎1 . (19) 0 𝜇
Thus, 𝑦(⋅, 𝜇) satisfies the boundary condition (16). The eigen- sin 𝜇𝑡 (28)
− (𝑎1 − 𝑎1 𝜇2 ) ] 𝑑𝑡
values of the problem Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ) are the zeros of the 𝜇
function as follows:
𝑥 sin 𝜇 (𝑥 − 𝑡)
+∫ 𝑞 (𝑡) 𝑓 (𝑡, 𝜇) 𝑑𝑡.
Δ (𝜇) := (𝑏1 𝜇2 − 𝑏1 ) 𝑦 (1, 𝜇) + (𝑏2 𝜇2
− 𝑏2 ) 𝑦 (1, 𝜇) . (20) 0 𝜇
4 Abstract and Applied Analysis
Lemma 1. For 0 ≤ 𝑥 ≤ 1, 𝜇 ∈ C, the following estimates hold: Combining (31) and (32), we obtain 0 ≤ 𝑥 ≤ 1, 𝜇 ∈ C,
𝑓 (𝑥, 𝜇)
1
2 𝑐 2
𝑐3 𝑐4 (𝑐1 + 𝑐2 𝜇 ) |I𝜇|𝑥 ≤ 𝑒|I𝜇|𝑥 0 ∫ 𝑞 (𝑡) [𝑎2 + 𝑎2 𝜇
𝑓 (𝑥, 𝜇) ≤ 𝑒 , (29) 1 + 𝜇 0
1 + 𝜇 (33)
2
+ (𝑎1 + 𝑎1 𝜇 ) 𝑐0 𝑡] 𝑑𝑡
2
𝜏𝑐3 𝑐4 (𝑐1 + 𝑐2 𝜇 ) |I𝜇|𝑥 𝑥
𝑔 (𝑥, 𝜇) ≤ 𝑒 . (30)
1 + 𝜇 + 𝑐0 𝑒|I𝜇|𝑥 ∫ 𝑒−|I𝜇|𝑡 𝑞 (𝑡) 𝑓 (𝑡, 𝜇) 𝑑𝑡.
0
Applying Gronwall’s inequality, cf. for example, [34, page 51],
yields 𝜇 ∈ C,
Proof. We divide 𝑓(⋅, 𝜇) into two parts 𝑓1 (⋅, 𝜇) and 𝑓2 (⋅, 𝜇) and
estimate each of them. Indeed, for 𝑥 ∈ [0, 1] and 𝜇 ∈ C we 𝑒−|I𝜇|𝑥 𝑓 (𝑥, 𝜇)
have 1
𝑐 2
≤ [ 0 ∫ 𝑞 (𝑡) [𝑎2 + 𝑎2 𝜇
1 + 𝜇 0
𝑓1 (𝑥, 𝜇)
2
+ (𝑎1 + 𝑎1 𝜇 ) 𝑐0 𝑡] 𝑑𝑡]
𝑥 sin 𝜇 (𝑥 − 𝑡)
= ∫ 𝑞 (𝑡) [ (𝑎2 − 𝑎2 𝜇2 ) cos 𝜇𝑡
0 𝜇 𝑥
× exp (𝑐0 ∫ 𝑞 (𝑡) 𝑑𝑡)
sin 𝜇𝑡 (34)
] 𝑑𝑡
0
− (𝑎1 − 𝑎1 𝜇2 )
𝜇 1
𝑐0 2
𝑥 ≤[ ∫ 𝑞 (𝑡) [𝑎2 + 𝑎2 𝜇
𝑐0 (𝑥 − 𝑡) 1 + 𝜇 0
≤ 𝑒|I𝜇|𝑥 ∫ 𝑞 (𝑡)
0 1 + 𝜇 (𝑥 − 𝑡)
2
+ (𝑎1 + 𝑎1 𝜇 ) 𝑐0 𝑡] 𝑑𝑡]
2 2
× [ 𝑎2 + 𝑎2 𝜇 + (𝑎1 + 𝑎1 𝜇 )
1
× exp (𝑐0 ∫ 𝑞 (𝑡) 𝑑𝑡) ,
𝑐0 𝑡 0
× ] 𝑑𝑡
1 + 𝜇 𝑡 from which we get
𝑐0 𝑥
𝑥
2 𝑓 (𝑥, 𝜇)
≤ 𝑒|I𝜇|𝑥 ∫ 𝑞 (𝑡) [𝑎2 + 𝑎2 𝜇
1 + 𝜇 𝑥 0
2 2
2 |I𝜇|𝑥 [ 𝑐0 [𝑎2 + 𝑎2 𝜇 + (𝑎1 + 𝑎1 𝜇 ) 𝑐0 ]
+ (𝑎1 + 𝑎1 𝜇 ) 𝑐0 𝑡] 𝑑𝑡 ≤𝑒 [
1 + 𝜇
1
[
𝑐0 2
≤ 𝑒|I𝜇|𝑥 ∫ 𝑞 (𝑡) [𝑎2 + 𝑎2 𝜇 1 1
(35)
1 + 𝜇 0 ]
× ∫ 𝑞 (𝑡) 𝑑𝑡] exp (𝑐0 ∫ 𝑞 (𝑡) 𝑑𝑡)
2 0 0
+ (𝑎1 + 𝑎1 𝜇 ) 𝑐0 𝑡] 𝑑𝑡. ]
(31) 2
𝑐3 𝑐4 (𝑐1 + 𝑐2 𝜇 ) |I𝜇|𝑥
= 𝑒 .
1 + 𝜇
Moreover, 0 ≤ 𝑥 ≤ 1, 𝜇 ∈ C, Then from (25) and (29), we obtain the estimate (30).
Now we split Δ(𝜇) into two parts via
𝑥 Δ (𝜇) = G (𝜇) + S (𝜇) , (36)
sin 𝜇 (𝑥 − 𝑡)
𝑓2 (𝑥, 𝜇) = ∫ 𝑞 (𝑡) 𝑓 (𝑡, 𝜇) 𝑑𝑡
0 𝜇 where G(𝜇) is known part
𝑥
𝑐0 (𝑥 − 𝑡) |I𝜇|(𝑥−𝑡) G (𝜇)
≤∫ 𝑒
0 1 + 𝜇 (𝑥 − 𝑡) sin 𝜇
= (𝑏1 𝜇2 − 𝑏1 ) [(𝑎2 − 𝑎2 𝜇2 ) cos 𝜇 − (𝑎1 − 𝑎1 𝜇2 ) ]
𝜇
× 𝑞 (𝑡) 𝑓 (𝑡, 𝜇) 𝑑𝑡
𝑥
+(𝑏2 𝜇2 − 𝑏2 )[(𝑎2 𝜇2 − 𝑎2 )𝜇 sin 𝜇+(𝑎1 𝜇2 − 𝑎1 ) cos 𝜇] ,
𝑞 (𝑡) 𝑓 (𝑡, 𝜇) 𝑑𝑡.
|I𝜇|𝑥 −|I𝜇|𝑡
≤ 𝑐0 𝑒 ∫ 𝑒 (32) (37)
0
Abstract and Applied Analysis 5
and S(𝜇) is unknown part Proof. Since S(𝜇) is entire, then also F𝜃,𝑚 (𝜇) is entire in 𝜇.
Combining the estimates | sin 𝑧/𝑧| ≤ (𝑐0 /(1 + |𝑧|))𝑒|I𝑧| and
S (𝜇) = (𝑏1 𝜇2 − 𝑏1 ) 𝑓 (1, 𝜇) + (𝑏2 𝜇2 − 𝑏2 ) 𝑔 (1, 𝜇) . (38) (39), we obtain
𝑚
Then, from Lemma 1, we have the following lemma. 𝑐0
F𝜃,𝑚 (𝜇) ≤ (
|I𝜇|𝑚𝜃
) 𝑒
1 + 𝜃 𝜇
Lemma 2. The function S(𝜇) is entire in 𝜇 and the following (45)
2 2
estimate holds: 𝑐3 𝑐4 𝑐5 (1 + 𝜇 ) |I𝜇|
⋅ 𝑒 , 𝜇 ∈ C,
2 2 1 + 𝜇
𝑐3 𝑐4 𝑐5 (1 + 𝜇 ) |I𝜇| (39)
S (𝜇) ≤ 𝑒 .
1 + 𝜇 leading to (42). Therefore, we get
𝑁 and so
̃ 𝜃,𝑚,𝑁 (𝜇) := ∑ [F
F ̃ 𝜃,𝑚 ( 𝑛𝜋 ) 𝑆2 (𝜇)
𝜎 𝑛 −𝑚
𝑛=−𝑁 sin 𝜃𝜇∗ ̃ 𝜃,𝑚,𝑁 (𝜇∗ )
G (𝜇∗ ) + ( ) F
𝜃𝜇∗
𝑛𝜋 sin (𝜎𝜇 − 𝑛𝜋)
̃ (
+F ) 𝑆𝑛 (𝜇)] , (59)
𝜃,𝑚
𝜎 𝜎 sin 𝜃𝜇∗ −𝑚
≤ ∗
(𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A (𝜀)) .
𝜃𝜇∗
𝑁 > 𝑚.
(52) ̃ 𝜃,𝑚,𝑁(𝜇∗ ) is given and
−𝑚
Since G(𝜇∗ ) + (sin 𝜃𝜇∗ /𝜃𝜇∗ ) F
∗ ∗ −𝑚 ∗
Using standard methods for solving initial problems, we may | sin 𝜃𝜇 /𝜃𝜇 | (𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A(𝜀)) has computable
assume that for |𝑛| < 𝑁, upper bound, we can define an enclosure for 𝜇∗ by solving
𝑛𝜋
̃ 𝜃,𝑚 ( 𝑛𝜋 ) < 𝜀,
the following system of inequalities:
F𝜃,𝑚 ( ) − F
𝜎 𝜎 sin 𝜃𝜇∗ −𝑚
(53) − ∗
(𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A (𝜀))
𝑛𝜋
F𝜃,𝑚 ( ) − F ̃ ( 𝑛𝜋 ) < 𝜀, 𝜃𝜇∗
𝜎 𝜃,𝑚
𝜎
−𝑚
sin 𝜃𝜇∗ ̃ 𝜃,𝑚,𝑁 (𝜇∗ )
for a sufficiently small 𝜀. From (42) we can see that F𝜃,𝑚 (𝜇) ≤ G (𝜇∗ ) + ( ) F (60)
satisfies the condition (9) when 𝑚 > 4 and therefore 𝜃𝜇∗
whenever 0 < 𝜀 ≤ min{𝜋/𝜎, 𝜎/𝜋, 1/√𝑒} we have sin 𝜃𝜇∗ −𝑚
̃ 𝜃,𝑚,𝑁 (𝜇) ≤ A (𝜀) , 𝜇 ∈ R, ≤ ∗
(𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A (𝜀)) .
F𝜃,𝑚,𝑁 (𝜇) − F (54) 𝜃𝜇∗
where there is a positive constant 𝑀F𝜃,𝑚 for which, cf. (10), Its solution is an interval containing 𝜇∗ , and over which
−𝑚 ̃
the graph G(𝜇∗ ) + (sin 𝜃𝜇∗ /𝜃𝜇∗ ) F ∗
𝜃,𝑚,𝑁 (𝜇 ) is squeezed
2𝑒1/4 √ 𝜋 between the graphs as follows:
A (𝜀) := { 3𝑒 (1 + 𝜎) + ( 𝐴 + 𝑀F𝜃,𝑚 ) 𝜌 (𝜀)
𝜎 𝜎
(55) sin 𝜃𝜇∗ −𝑚
1 − ∗
(𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A (𝜀)) ,
+ (𝜎 + 2 + log (2)) 𝑀F𝜃,𝑚 } 𝜀 log ( ) . 𝜃𝜇∗
𝜀 (61)
sin 𝜃𝜇∗ −𝑚
Here ∗
(𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A (𝜀)) .
𝜃𝜇∗
3𝜎 𝜎
𝐴 := (F (0) + 𝑀 ) ,
𝜋 𝜃,𝑚 𝜋 F𝜃,𝑚 Using the fact that
(56)
1
𝜌 (𝜀) := 𝛾 + 10 log ( ) . ̃ 𝜃,𝑚,𝑁 (𝜇) → F𝜃,𝑚 (𝜇)
F (62)
𝜀
Abstract and Applied Analysis 7
uniformly over any compact set and since 𝜇∗ is a simple root, and 𝜃 is chosen sufficiently small for which |𝜃𝜇| < 𝜋.
we obtain the following for large 𝑁 and sufficiently small 𝜀: Therefore, 𝜃, 𝑚 must be chosen so that for |𝜇| < 𝑁𝜋/𝜎
𝜕 sin 𝜃𝜇 −𝑚 ̃
(G (𝜇) + ( ) F𝜃,𝑚,𝑁 (𝜇)) ≠
0 (63) 𝑚 > 4, 𝜃 ∈ (0, 1) , 𝜃𝜇 < 𝜋. (69)
𝜕𝜇 𝜃𝜇
Let 𝜇∗ be an eigenvalue and let 𝜇𝑁 be its approximation. Thus,
in a neighborhood of 𝜇∗ . Hence, the graph of Δ(𝜇∗ ) = 0 and Δ̃ 𝑁(𝜇𝑁) = 0. From (68) we have |Δ ̃ 𝑁(𝜇∗ )| ≤
G(𝜇) + (sin 𝜃𝜇/𝜃𝜇)−𝑚 F ̃ 𝜃,𝑚,𝑁(𝜇) intersects the graphs ∗ ∗ −𝑚 ∗
| sin 𝜃𝜇 /𝜃𝜇 | (𝑇𝑁,𝑚−4,𝜎 (𝜇 ) + A(𝜀)). Now we estimate the
−| sin 𝜃𝜇/𝜃𝜇| (𝑇𝑁,𝑚−4,𝜎 (𝜇) + A(𝜀)) and | sin 𝜃𝜇/𝜃𝜇|−𝑚
−𝑚
error |𝜇∗ − 𝜇𝑁| for an eigenvalue 𝜇∗ .
(𝑇𝑁,𝑚−4,𝜎 (𝜇) + A(𝜀)) at two points with abscissae
𝑎− (𝜇∗ , 𝑁, 𝜀) ≤ 𝑎+ (𝜇∗ , 𝑁, 𝜀) and the solution of the system of 2
Theorem 5. Let 𝜇∗ be an eigenvalue of Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ). For
inequalities (60) is the interval sufficient large 𝑁 we have the following estimate:
𝐼𝜀,𝑁 := [𝑎− (𝜇∗ , 𝑁, 𝜀) , 𝑎+ (𝜇∗ , 𝑁, 𝜀)] (64) −𝑚
∗ sin 𝜃𝜇𝑁 𝑇𝑁,𝑚−4,𝜎 (𝜇𝑁) + A (𝜀)
∗ 𝜇 − 𝜇𝑁 < . (70)
and in particular 𝜇 ∈ 𝐼𝜀,𝑁. Summarizing the above 𝜃𝜇𝑁 inf 𝜁∈𝐼𝜀,𝑁 Δ (𝜁)
discussion, we arrive at the following lemma which is similar
to that of [26]. Proof. Since Δ(𝜇𝑁) − Δ ̃ 𝑁(𝜇𝑁) = Δ(𝜇𝑁) − Δ(𝜇∗ ), then from
2 (68) and after replacing 𝜇 by 𝜇𝑁, we obtain
Lemma 4. For any eigenvalue 𝜇∗ , we can find 𝑁0 ∈ Z+ and
sufficiently small 𝜀 such that 𝜇∗ ∈ 𝐼𝜀,𝑁 for 𝑁 > 𝑁0 . Moreover, sin 𝜃𝜇 −𝑚
∗ 𝑁
we get Δ (𝜇𝑁) − Δ (𝜇 ) ≤ (𝑇𝑁,𝑚−4,𝜎 (𝜇𝑁) + A (𝜀)) .
𝜃𝜇𝑁
[𝑎− (𝜇∗ , 𝑁, 𝜀) , 𝑎+ (𝜇∗ , 𝑁, 𝜀)] → {𝜇∗ } (71)
(65)
as 𝑁 → ∞, 𝜀 → 0. Using the mean value theorem yields that for some 𝜁 ∈ 𝐽𝜀,𝑁 :=
[min(𝜇∗ , 𝜇𝑁), max(𝜇∗ , 𝜇𝑁)],
Proof. Since all eigenvalues of Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ) are sim-
ple, then for large 𝑁 and sufficiently small 𝜀 we have ∗
(𝜇 − 𝜇𝑁) Δ (𝜁)
(𝜕/𝜕𝜇)(G(𝜇) + (sin 𝜃𝜇/𝜃𝜇)−𝑚 F̃ 𝜃,𝑚,𝑁(𝜇)) > 0, in a neighbor-
hood of 𝜇∗ . Choose 𝑁0 such that sin 𝜃𝜇𝑁 −𝑚
≤ (𝑇𝑁,𝑚−4,𝜎 (𝜇𝑁) + A (𝜀)) , 𝜁 ∈ 𝐽𝜀,𝑁 ⊂ 𝐼𝜀,𝑁.
sin 𝜃𝜇 −𝑚 ̃ 𝜃𝜇𝑁
G (𝜇) + ( ) F𝜃,𝑚,𝑁0 (𝜇) (72)
𝜃𝜇
(66)
sin 𝜃𝜇 −𝑚 Since the eigenvalues are simple, then for sufficiently large
= ± (𝑇𝑁0 ,𝑚−4,𝜎 (𝜇) + A (𝜀)) 𝑁inf 𝜁∈𝐼𝜀,𝑁 |Δ (𝜁)| > 0 and we get (70).
𝜃𝜇
has two distinct solutions which we denote by 𝑎− (𝜇∗ , 𝑁0 , 𝜀) ≤
𝑎+ (𝜇∗ , 𝑁0 , 𝜀). The decay of 𝑇𝑁,𝑚−4,𝜎 (𝜇) → 0 as 𝑁 → ∞ and 3. The Case of Π(𝑞, 𝑎, 𝑏, 0, 𝑏 )
A(𝜀) → 0 as 𝜀 → 0 will ensure the existence of the solutions This section includes briefly a treatment similarly to
𝑎− (𝜇∗ , 𝑁, 𝜀) and 𝑎+ (𝜇∗ , 𝑁, 𝜀) as 𝑁 → ∞ and 𝜀 → 0. For the that of the previous section for the eigenvalue problem
second point we recall that F ̃ 𝜃,𝑚,𝑁(𝜇) → F𝜃,𝑚 (𝜇) as 𝑁 → Π(𝑞, 𝑎, 𝑏, 0, 𝑏 ) introduced in Section 1. Notice that condition
∞ and as 𝜀 → 0. Hence, by taking the limit we obtain (18) implies that the analysis of problem Π(𝑞, 𝑎, 𝑏, 0, 𝑏 ) is
−𝑚 not included in that of Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ). Let 𝜓(⋅, 𝜇) denote the
sin 𝜃𝜇∗
G (𝑎+ (𝜇∗ , ∞, 0)) + ( ) F𝜃,𝑚 (𝑎+ (𝜇∗ , ∞, 0)) = 0, solution of (15) satisfying the following initial conditions:
𝜃𝜇∗
−𝑚 𝜓 (0, 𝜇) = 𝑎2 , 𝜓 (0, 𝜇) = −𝑎1 . (73)
sin 𝜃𝜇∗
G (𝑎− (𝜇∗ , ∞, 0)) + ( ) ∗
F𝜃,𝑚 (𝑎− (𝜇 , ∞, 0)) = 0.
𝜃𝜇∗ Thus, 𝜓(⋅, 𝜇) satisfies the boundary condition (16). The
(67) eigenvalues of the problem Π(𝑞, 𝑎, 𝑏, 0, 𝑏 ) are the zeros of the
function as follows:
That is, Δ(𝑎+ ) = Δ(𝑎− ) = 0. This leads us to conclude that
𝑎+ = 𝑎− = 𝜇∗ , since 𝜇∗ is a simple root. Ω (𝜇) := (𝑏1 𝜇2 − 𝑏1 ) 𝜓 (1, 𝜇) + (𝑏2 𝜇2 − 𝑏2 ) 𝜓 (1, 𝜇) . (74)
̃ 𝑁(𝜇) := G(𝜇) + (sin 𝜃𝜇/𝜃𝜇)−𝑚 F
Let Δ ̃ 𝜃,𝑚,𝑁(𝜇). Then (50)
Recall that Π(𝑞, 𝑎, 𝑏, 𝑎 , 𝑏 ) has denumerable set of real and
and (54) imply
simple eigenvalues, cf. [38]. Using the method of variation
−𝑚 of constants, the solution 𝜓(𝑥, 𝜇) satisfies Volterra integral
̃ 𝑁 (𝜇) ≤ sin 𝜃𝜇 (𝑇𝑁,𝑚−4,𝜎 (𝜇) + A (𝜀)) ,
Δ (𝜇) − Δ
𝜃𝜇 equation as follows:
(68)
𝑁𝜋 sin 𝜇𝑥
𝜇 < 𝜓 (𝑥, 𝜇) = 𝑎2 cos 𝜇𝑥 − 𝑎1 + 𝑇 [𝜓] (𝑥, 𝜇) , (75)
𝜎 𝜇
8 Abstract and Applied Analysis
where 𝑇 is the Volterra operator defined in (22). Differentiat- reconstruct the functions R𝜃,𝑚 (𝜇) via the following sampling
ing (75), we get formula:
̃ [𝜓] (𝑥, 𝜇) , R𝜃,𝑚 (𝜇)
𝜓 (𝑥, 𝜇) = −𝑎2 𝜇 sin 𝜇𝑥 − 𝑎1 cos 𝜇𝑥 + 𝑇 (76)
∞
̃ is the Volterra operator defined in (24). Define 𝑛𝜋 2
where 𝑇 = ∑ [R𝜃,𝑚 ( ) 𝑆𝑛 (𝜇)
𝜎 (86)
ℎ1 (⋅, 𝜇) and ℎ2 (⋅, 𝜇) to be 𝑛=−∞
and U(𝜇) is the unknown one Since all eigenvalues are real, then from now on we restrict
ourselves to 𝜇 ∈ R. Since 𝜇𝑚−2 R𝜃,𝑚 (𝜇) ∈ 𝐿2 (R), the
U (𝜇) := (𝑏1 𝜇2 − 𝑏1 ) ℎ1 (1, 𝜇) + (𝑏2 𝜇2 − 𝑏2 ) ℎ2 (1, 𝜇) . (80) truncation error, cf. (5), is given for |𝜇| < 𝑁𝜋/𝜎 by
R𝜃,𝑚 (𝜇) − R𝜃,𝑚,𝑁 (𝜇) ≤ 𝑇𝑁,𝑚−2,𝜎 (𝜇) , (88)
Then, as in the previous section, U(𝜇) is entire in 𝜇 for each
𝑥 ∈ [0, 1] for which where
2 𝑇𝑁,𝑚−2,𝜎 (𝜇)
𝑐1 𝑐3 𝑐4 𝑐6 (1 + 𝜇 ) |I𝜇|
U (𝜇) ≤ 𝑒 , 𝜇 ∈ C, (81)
1 + 𝜇 2
𝜉𝑚−2,𝜎 𝐸𝑚−2 sin 𝜎𝜇 1 1
:= ( + )
where 𝑐6 := max{|𝑏1 | + |𝑏2 |𝜏, |𝑏1 | + |𝑏2 |𝜏}. √3(𝑁 + 1)𝑚−2 (𝑁𝜋 − 𝜎𝜇)
3/2
(𝑁𝜋 + 𝜎𝜇)
3/2
where 𝑁
̃ 𝜃,𝑚 ( 𝑛𝜋 2
:= ∑ [R ) 𝑆𝑛 (𝜇)
𝜎
12𝜃2 Γ [2𝑚 − 3] + Γ [2𝑚 − 1] 𝑛=−𝑁
(90)
𝜔0 := √ . (85)
𝜃2𝑚−1 Γ [2 (𝑚 + 1)]
̃ ( 𝑛𝜋 ) sin (𝜎𝜇 − 𝑛𝜋) 𝑆𝑛 (𝜇)] ,
+R 𝜃,𝑚
𝜎 𝜎
Thus, R𝑚,𝜃 (𝜇) belongs to the Paley-Wiener space PW2𝜎 with
𝜎 = 1 + 𝑚𝜃. Since R𝜃,𝑚 (𝜇) ∈ PW2𝜎 ⊂ PW22𝜎 , then we can 𝑁 > 𝑚.
Abstract and Applied Analysis 9
Using standard methods for solving initial problems, we and 𝜃 is chosen sufficiently small for which |𝜃𝜇| < 𝜋.
may assume that for |𝑛| < 𝑁 Therefore, 𝜃, 𝑚 must be chosen so that for |𝜇| < 𝑁𝜋/𝜎
R ( 𝑛𝜋 ) − R
𝑚 > 2, 𝜃 ∈ (0, 1) , 𝜃𝜇 < 𝜋.
𝜃,𝑚 ̃ 𝜃,𝑚 ( 𝑛𝜋 ) < 𝜀, (99)
𝜎 𝜎 Let 𝜇∗ be an eigenvalue and 𝜇𝑁 be its approximation. Thus
(91)
̃ 𝑁(𝜇𝑁) = 0. From (98) we have |Ω
̃ 𝑁(𝜇∗ )| ≤
R ( 𝑛𝜋 ) − R ̃ ( 𝑛𝜋 ) < 𝜀, Ω(𝜇∗ ) = 0 and Ω
𝜃,𝑚
𝜎 𝜃,𝑚
𝜎 ∗ ∗ −𝑚 ∗
| sin 𝜃𝜇 /𝜃𝜇 | (𝑇𝑁,𝑚−2,𝜎 (𝜇 ) + A(𝜀)). Now we estimate the
error |𝜇∗ − 𝜇𝑁| for an eigenvalue 𝜇∗ . Finally we have the
for a sufficiently small 𝜀. From (83) we can see that R𝜃,𝑚 (𝜇) following estimate.
satisfies the condition (9) when 𝑚 > 2 and therefore
whenever 0 < 𝜀 ≤ min{𝜋/𝜎, 𝜎/𝜋, 1/√𝑒} we have 2
Theorem 7. Let 𝜇∗ be an eigenvalue of the problem
̃ 𝜃,𝑚,𝑁 (𝜇) ≤ A (𝜀) ,
R𝜃,𝑚,𝑁 (𝜇) − R Π(𝑞, 𝑎, 𝑏, 0, 𝑏 ). For sufficient large 𝑁 we have the following
𝜇 ∈ R, (92)
estimate
−𝑚
where there is a positive constant 𝑀R𝜃,𝑚 for which, cf. (10), ∗ sin 𝜃𝜇𝑁 𝑇𝑁,𝑚−2,𝜎 (𝜇𝑁) + A (𝜀)
𝜇 − 𝜇𝑁 < . (100)
and 𝜃𝜇𝑁 inf 𝜁∈I𝜀,𝑁 Ω (𝜁)
2𝑒1/4 √ 𝜋
A (𝜀) := { 3𝑒 (1 + 𝜎) + ( 𝐴 + 𝑀R𝜃,𝑚 ) 𝜌 (𝜀) 4. Numerical Examples
𝜎 𝜎
(93)
1 This section includes two detailed worked examples illus-
+ (𝜎 + 2 + log (2)) 𝑀R𝜃,𝑚 } 𝜀 log ( ) . trating the above technique. Examples 1 and 2 computed in
𝜀
[27, 45] with the classical sinc method, where only truncation
Here error analysis is considered, respectively. It is clearly seen that
3𝜎 our new method (Hermite interpolations) gives remarkably
𝜎
𝐴 := (R (0) + 𝑀 ), better results than in [27, 45], see also [41–43]. We indicate in
𝜋 𝜃,𝑚 𝜋 R𝜃,𝑚 these examples the effect of the amplitude error in the method
(94)
1 by determining enclosure intervals for different values of
𝜌 (𝜀) := 𝛾 + 10 log ( ) .
𝜀 𝜀. We also indicate the effect of the parameters 𝑚 and 𝜃
by several choices. Each example is exhibited via figures
As in the above section, we have the following lemma. that accurately illustrate the procedure near to some of
2
the approximated eigenvalues. More explanations are given
Lemma 6. For any eigenvalue 𝜇∗ of the problem below. Recall that 𝑎± (𝜇) and 𝑏± (𝜇) are defined by
Π(𝑞, 𝑎, 𝑏, 0, 𝑏 ), we can find 𝑁0 ∈ Z+ and sufficiently
small 𝜀 such that 𝜇∗ ∈ I𝜀,𝑁 for 𝑁 > 𝑁0 , where −𝑚
̃ 𝑁 (𝜇) ± sin 𝜃𝜇 (𝑇𝑁,𝑚−3,𝜎 (𝜇) + A (𝜀)) ,
𝑎± (𝜇) = Δ
𝜃𝜇
I𝜀,𝑁 := [𝑏− (𝜇∗ , 𝑁, 𝜀) , 𝑎𝑏+ (𝜇∗ , 𝑁, 𝜀)] , (95)
𝑁𝜋
𝜇 < ,
𝑏− , 𝑏+ are the solutions of the inequalities 𝜎
(101)
sin 𝜃𝜇 −𝑚
−𝑚
− (𝑇𝑁,𝑚−2,𝜎 (𝜇) + A (𝜀))
𝜃𝜇 ̃ 𝑁 (𝜇) ± sin 𝜃𝜇 (𝑇𝑁,𝑚−2,𝜎 (𝜇) + A (𝜀)) ,
𝑏± (𝜇) = Ω 𝜃𝜇
(96)
sin 𝜃𝜇 −𝑚
≤Ω ̃ 𝑁 (𝜇) ≤ 𝑁𝜋
𝜃𝜇 (𝑇𝑁,𝑚−2,𝜎 (𝜇) + A (𝜀)) . 𝜇 < ,
𝜎
(102)
Moreover, we get
respectively. Recall also that the enclosure intervals 𝐼𝜀,𝑁 :=
[𝑏− (𝜇∗ , 𝑁, 𝜀) , 𝑏+ (𝜇∗ , 𝑁, 𝜀)] → {𝜇∗ } [𝑎− , 𝑎+ ] and I𝜀,𝑁 := [𝑏− , 𝑏+ ] are determined by solving
(97)
as 𝑁 → ∞, 𝜀 → 0. 𝑁𝜋
𝑎± (𝜇) = 0, 𝜇 < , (103)
−𝑚 ̃
𝜎
̃ 𝑁(𝜇) := K(𝜇) + (sin 𝜃𝜇/𝜃𝜇)
Let Ω R𝜃,𝑚,𝑁(𝜇). Then (88)
𝑁𝜋
and (92) imply 𝑏± (𝜇) = 0, 𝜇 < . (104)
𝜎
̃ 𝑁 (𝜇)
Ω (𝜇) − Ω respectively. We would like to mention that Mathematica
has been used to obtain the exact values for the three
sin 𝜃𝜇 −𝑚
𝑁𝜋 examples where eigenvalues cannot be computed concretely.
≤ (𝑇𝑁,𝑚−2,𝜎 (𝜇) + A (𝜀)) , 𝜇 <
𝜃𝜇 𝜎 Mathematica is also used in rounding the exact eigenvalues,
(98) which are square roots.
10 Abstract and Applied Analysis
0.08 0.06
0.06
0.04
0.04
0.02 0.02
0.00 0.00
−0.02
−0.02
−0.04
−0.04
−0.06
1.030 1.035 1.040 1.045 1.050 1.055 1.060 1.030 1.035 1.040 1.045 1.050 1.055 1.060
b− b−
Ω(𝜇) Ω(𝜇)
b+ b+
Figure 1: The enclosure interval dominating 𝜇1 for 𝑚 = 10, 𝑁 = 15, Figure 2: The enclosure interval dominating 𝜇1 for 𝑚 = 10, 𝑁 = 15,
𝜃 = 1/5, and 𝜀 = 10−5 . 𝜃 = 1/5, and 𝜀 = 10−10 .
Ω (𝜇) = (−1 + 2𝜇2 ) cos [√1 + 𝜇2 ] 3.510 3.515 3.520 3.525 3.530
(106)
b−
− (−1 + 𝜇2 ) √1 + 𝜇2 sin [√1 + 𝜇2 ] . Ω(𝜇)
b+
The function K(𝜇) will be
Figure 3: The enclosure interval dominating 𝜇2 for 𝑚 = 10, 𝑁 = 15,
K (𝜇) = (−1 + 2𝜇2 ) cos [𝜇] − 𝜇 (−1 + 𝜇2 ) sin [𝜇] . (107) 𝜃 = 1/5, and 𝜀 = 10−5 .
Example 2. The boundary value problem [45] G (𝜇) = (1 + 𝜇2 ) ((1 − 𝜇2 ) cos 𝜇 − 𝜇 sin 𝜇) . (110)
−𝑦 (𝑥, 𝜇) − 𝑦 (𝑥, 𝜇) = 𝜇2 𝑦 (𝑥, 𝜇) 0 ≤ 𝑥 ≤ 1, As is clearly seen, the eigenvalues cannot be computed
(108) explicitly. As in the previous example, Figures 5, 6, 7, and 8
𝑦 (0, 𝜇) = 𝜇2 𝑦 (0, 𝜇) , 𝑦 (1, 𝜇) = 𝜇2 𝑦 (1, 𝜇) , illustrate the results of Tables 4, 5, 6, and 7.
Abstract and Applied Analysis 11
Table 1: With 𝑁 = 15, the approximation 𝜇𝑘,𝑁 and the exact solution 𝜇𝑘 for different choices of 𝑚 and 𝜃.
𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
Exact 𝜇𝑘 1.0493258679653497 3.5207214555369464 6.505146961583527 9.578576417519093
𝜇𝑘,𝑁
𝑚=6
𝜃 = 1/9 1.0493258487568435 3.5207214557864277 6.5051469611825405 9.578576417536349
𝜃 = 1/12 1.0493291624256957 3.5207212860735546 6.5051469673141336 9.578576393056661
𝑚 = 10
𝜃 = 1/5 1.0493258679653554 3.5207214555369624 6.505146961583516 9.57857641751921
𝜃 = 1/8 1.049325865990155 3.520721455899672 6.505146961464727 9.578576417559768
Table 2: Absolute error |𝜇𝑘 − 𝜇𝑘,𝑁 |.
𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
𝑚=6
𝜃 = 1/9 1.92085 × 10−8 2.49481 × 10−10 4.00987 × 10−10 1.72555 × 10−11
𝜃 = 1/12 3.29446 × 10−6 1.69463 × 10−7 5.73061 × 10−9 2.44624 × 10−8
𝑚 = 10
𝜃 = 1/5 5.77316 × 10−15 1.59872 × 10−14 1.15463 × 10−14 1.1724 × 10−13
𝜃 = 1/8 1.97519 × 10−9 3.62725 × 10−10 1.188 × 10−10 4.0675 × 10−11
Table 3: For 𝑁 = 15, 𝑚 = 10, and 𝜃 = 1/5, the exact solution 𝜇𝑘 are all inside the interval [𝑏− , 𝑏+ ] for different values of 𝜀.
𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
Exact 𝜇𝑘 1.0493258679653497 3.5207214555369464 6.505146961583527 9.578576417519093
I𝜀,𝑁 , 𝜀 = 10−5 [1.04294069, 1.05557896] [3.51981844, 3.52162396] [6.50375768, 6.50653831] [9.55222305, 9.60712093]
I𝜀,𝑁 , 𝜀 = 10−10 [1.04932561, 1.049326118] [3.52072141, 3.52072149] [6.50514690, 6.50514702] [9.57857529, 9.57857754]
𝐸8 (R𝜃,𝑚 ) = 4.51845 × 108 , 𝐸7 (R𝜃,𝑚 ) = 2.29709 × 105 , ] = 1, 𝑀R𝜃,𝑚 = 4.55609 × 104 .
Table 4: With 𝑁 = 40, the approximation 𝜇𝑘,𝑁 and the exact solution 𝜇𝑘 for different choices of 𝜃.
𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
Exact 𝜇𝑘 0.4828692021748484 1.966318052350425 4.827089429919572 7.919684444168381
𝜇𝑘,𝑁
𝑚=8
𝜃 = 1/32 0.48286920221045176 1.96631805234574 4.827089429919605 7.919684444168366
𝜃 = 1/35 0.4828692337692527 1.966318047624416 4.8270894299720776 7.91968444416245
𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
𝑚=8
𝜃 = 1/32 3.56034 × 10−11 4.68492 × 10−12 3.28626 × 10−14 1.5099 × 10−14
𝜃 = 1/35 3.15944 × 10−8 4.72601 × 10−9 5.25047 × 10−11 5.93126 × 10−12
Table 6: The approximation 𝜇𝑘,𝑁 and the exact solution 𝜇𝑘 for 𝑁 = 40, 𝑚 = 14 and 𝜃 = 1/26.
Table 7: For 𝑁 = 40, 𝑚 = 14 and 𝜃 = 1/26, the exact solution 𝜇𝑘 are all inside the interval [𝑎− , 𝑎+ ] for different values of 𝜀.
𝜇𝑘 𝜇1 𝜇2 𝜇3 𝜇4
Exact 𝜇𝑘 0.4828692021748484 1.966318052350425 4.827089429919572 7.919684444168381
I𝜀,𝑁 ,, 𝜀 = 10−5 [0.47918888, 0.48651557] [1.96592879, 1.96670680] [4.82707252, 4.82710633] [7.91968171, 7.919687175]
I𝜀,𝑁 , 𝜀 = 10−10 [0.48284084, 0.48289756] [1.96631794, 1.96631815] [4.82708919, 4.82708966] [7.919684437, 7.919684450]
𝐸10 (F𝜃,𝑚 ) = 2.83057 × 1018 , 𝐸9 (F𝜃,𝑚 ) = 1.12829 × 1014 , ] = 1, 𝑀F𝜃,𝑚 = 1.57716 × 107 .
12 Abstract and Applied Analysis
0.03
0.4
0.02
0.2
0.01
0.0 0.00
−0.01
−0.2
−0.02
−0.4 −0.03
−0.04
3.510 3.515 3.520 3.525 3.530 7.91967 7.91968 7.91968 7.91969
b− a−
Ω(𝜇) Δ(𝜇)
b+ a+
Figure 4: The enclosure interval dominating 𝜇2 for 𝑚 = 10, 𝑁 = 15, Figure 7: The enclosure interval dominating 𝜇4 for 𝑚 = 14, 𝑁 = 40,
𝜃 = 1/5, and 𝜀 = 10−10 . 𝜃 = 1/26, and 𝜀 = 10−5 .
0.15 0.03
0.10 0.02
0.05 0.01
0.00 0.00
−0.05 −0.01
−0.10 −0.02
−0.15 −0.03
0.42 0.44 0.46 0.48 0.50 0.52 0.54 7.91967 7.91968 7.91968 7.91969
a− a−
Δ(𝜇) Δ(𝜇)
a+ a+
Figure 5: The enclosure interval dominating 𝜇1 for 𝑚 = 14, 𝑁 = 40, Figure 8: The enclosure interval dominating 𝜇4 for 𝑚 = 14, 𝑁 = 40,
𝜃 = 1/26, and 𝜀 = 10−5 . 𝜃 = 1/26, and 𝜀 = 10−10 .
Acknowledgment
0.10
This research has been supported by a grant from the Institute
0.05 of Scientific Research at Umm AL-Qura University, Saudi
Arabia.
0.00
−0.05 References
−0.10 [1] E. H. Doha, A. H. Bhrawy, and R. M. Hafez, “A Jacobi-
Jacobi dual-Petrov-Galerkin method for third- and fifth-order
−0.15 differential equations,” Mathematical and Computer Modelling,
vol. 53, no. 9-10, pp. 1820–1832, 2011.
0.42 0.44 0.46 0.48 0.50 0.52 0.54 [2] E. H. Doha, A. H. Bhrawy, and M. A. Saker, “Integrals of
Bernstein polynomials: an application for the solution of high
a−
even-order differential equations,” Applied Mathematics Letters,
Δ(𝜇)
a+
vol. 24, no. 4, pp. 559–565, 2011.
[3] E. Tohidi, A. H. Bhrawy, and K. Erfani, “A collocation method
Figure 6: The enclosure interval dominating 𝜇1 for 𝑚 = 14, 𝑁 = 40, based on Bernoulli operational matrix for numerical solution
𝜃 = 1/26, and 𝜀 = 10−10 . of generalized pantograph equation,” Applied Mathematical
Abstract and Applied Analysis 13
Modelling, vol. 37, no. 6, pp. 4283–4294, 2013. [19] J. R. Higgins, G. Schmeisser, and J. J. Voss, “The sampling
[4] E. H. Doha and A. H. Bhrawy, “Efficient spectral-Galerkin theorem and several equivalent results in analysis,” Journal of
algorithms for direct solution for second-order differential Computational Analysis and Applications, vol. 2, no. 4, pp. 333–
equations using Jacobi polynomials,” Numerical Algorithms, vol. 371, 2000.
42, no. 2, pp. 137–164, 2006. [20] G. Hinsen, “Irregular sampling of bandlimited 𝐿𝑝 -functions,”
[5] A. H. Bhrawy, “A Jacobi-Gauss-Lobatto collocation method Journal of Approximation Theory, vol. 72, no. 3, pp. 346–364,
for solving generalized Fitzhugh-Nagumo equation with time- 1993.
dependent coefficients,” Applied Mathematics and Computation, [21] D. Jagerman and L. Fogel, “Some general aspects of the sampling
vol. 222, pp. 255–264, 2013. theorem,” IRE Transactions on Information Theory, vol. 2, pp.
[6] A. H. Bhrawy, M. A. Abdelkawy, and A. Biswas, “Cnoidal and 139–146, 1956.
snoidal wave solutions to coupled nonlinear wave equations by [22] M. H. Annaby and R. M. Asharabi, “Error analysis associated
the extended Jacobi’s elliptic function method,” Communica- with uniform Hermite interpolations of bandlimited functions,”
tions in Nonlinear Science and Numerical Simulation, vol. 18, no. Journal of the Korean Mathematical Society, vol. 47, no. 6, pp.
4, pp. 915–925, 2013. 1299–1316, 2010.
[7] A. Imani, A. Aminataei, and A. Imani, “Collocation method via [23] J. R. Higgins, Sampling Theory in Fourier and Signal Analysis:
Jacobi polynomials for solving nonlinear ordinary differential Foundations, Oxford University Press, Oxford, UK, 1996.
equations,” International Journal of Mathematics and Mathe- [24] P. L. Butzer, J. R. Higgins, and R. L. Stens, “Sampling theory of
matical Sciences, vol. 2011, Article ID 673085, 11 pages, 2011. signal analysis,” in Development of Mathematics 1950–2000, pp.
[8] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional 193–234, Birkhäuser, Basel, Switzerland, 2000.
Calculus: Models and Numerical Methods, vol. 3 of Series on [25] P. L. Butzer, G. Schmeisser, and R. L. Stens, “An introduction
Complexity, Nonlinearity and Chaos, World Scientific, Hacken- to sampling analysis,” in Non Uniform Sampling: Theory and
sack, NJ, USA, 2012. Practices, F. Marvasti, Ed., pp. 17–121, Kluwer, New York, NY,
[9] D. Baleanu, A. H. Bhrawy, and T. M. Taha, “Two efficient USA, 2001.
generalized Laguerre spectral algorithms for fractional initial [26] A. Boumenir, “Higher approximation of eigenvalues by the
value problems,” Abstract and Applied Analysis, vol. 2013, Article sampling method,” BIT Numerical Mathematics, vol. 40, no. 2,
ID 546502, 10 pages, 2013. pp. 215–225, 2000.
[10] D. Baleanu, A. H. Bhrawy, and T. M. Taha, “A modified gen- [27] A. Boumenir, “The sampling method for Sturm-Liouville prob-
eralized Laguerre spectral methods for fractional differential lems with the eigenvalue parameter in the boundary condition,”
equations on the half line,” Abstract and Applied Analysis, vol. Numerical Functional Analysis and Optimization, vol. 21, no. 1-2,
2013, Article ID 413529, 12 pages, 2013. pp. 67–75, 2000.
[11] F. Ghaemi, R. Yunus, A. Ahmadian, S. Salahshour, M. B. [28] A. Boumenir and B. Chanane, “Eigenvalues of Sturm-Liouville
Suleiman, and S. F. Saleh, “Application of fuzzy fractional kinetic systems using sampling theory,” Applied Analysis, vol. 62, pp.
equations to modelling of the acid hydrolysis reaction,” Abstract 323–334, 1996.
and Applied Analysis, vol. 2013, Article ID 610314, 19 pages, 2013.
[29] M. M. Tharwat, A. H. Bhrawy, and A. Yildirim, “Numerical
[12] A. H. Bhrawy and A. S. Alofi, “The operational matrix of frac- computation of the eigenvalues of a discontinuous Dirac system
tional integration for shifted Chebyshev polynomials,” Applied using the Sinc method with error analysis,” International Journal
Mathematics Letters, vol. 26, no. 1, pp. 25–31, 2013. of Computer Mathematics, vol. 89, no. 15, pp. 2061–2080, 2012.
[13] A. H. Bhrawy and M. M. Al-Shomrani, “A shifted Legendre [30] J. Lund and K. Bowers, Sinc Methods for Quadrature and
spectral method for fractional-order multi-point boundary Differential Equations, SIAM, Philadelphia, Pa, USA, 1992.
value problems,” Advances in Difference Equations, vol. 2012, p.
[31] F. Stenger, “Numerical methods based on Whittaker cardinal, or
8, 2012.
Sinc functions,” SIAM Review, vol. 23, no. 2, pp. 165–224, 1981.
[14] M. H. Annaby and R. M. Asharabi, “Computing eigenvalues of
[32] F. Stenger, Numerical Methods Based on Sinc and Analytic
Sturm-Liouville problems by Hermite interpolations,” Numeri-
Functions, Springer, New York, NY, USA, 1993.
cal Algorithms, vol. 60, no. 3, pp. 355–367, 2012.
[33] R. P. Boas, Entire Functions, Academic Press, New York, NY,
[15] M. M. Tharwat, A. H. Bhrawy, and A. S. Alofi, “Computing
USA, 1954.
eigenvalues of discontinuous Sturm-Liouville problems with
eigenparameter in all boundary conditions using Hermite [34] M. S. P. Eastham, Theory of Ordinary Differential Operators, Van
approximation,” Abstract and Applied Analysis, vol. 2013, Article Nostrand Reinhold, London, UK, 1970.
ID 498457, 14 pages, 2013. [35] B. M. Levitan and I. S. Sargsjan, “Introduction to spectral the-
[16] M. M. Tharwat, A. H. Bhrawy, and A. S. Alofi, “Approximation ory: selfadjoint ordinary differential operators,” in Translation
of eigenvalues of discontinuous Sturm-Liouville problems with of Mthematical Monographs, vol. 39, American Mathematical
eigenparameter in all boundary conditions,” Boundary Value Society, Providence, RI, USA, 1975.
Problems, vol. 2013, p. 132, 2013. [36] M. H. Annaby and M. M. Tharwat, “On sampling theory and
[17] M. M. Tharwat, A. H. Bhrawy, and A. Yildirim, “Numerical eigenvalue problems with an eigenparameter in the boundary
computation of eigenvalues of discontinuous Sturm-Liouville conditions,” SUT Journal of Mathematics, vol. 42, no. 2, pp. 157–
problems with parameter dependent boundary conditions 176, 2006.
using Sinc method,” Numerical Algorithms, vol. 63, no. 1, pp. 27– [37] M. M. Tharwat, “Discontinuous Sturm-Liouville problems and
48, 2013. associated sampling theories,” Abstract and Applied Analysis,
[18] G. R. Grozev and Q. I. Rahman, “Reconstruction of entire vol. 2011, Article ID 610232, 30 pages, 2011.
functions from irregularly spaced sample points,” Canadian [38] C. T. Fulton, “Two-point boundary value problems with
Journal of Mathematics, vol. 48, no. 4, pp. 777–793, 1996. eigenvalue parameter contained in the boundary conditions,”
14 Abstract and Applied Analysis
Research Article
Approximate Solutions of Fisher’s Type Equations with
Variable Coefficients
Copyright © 2013 A. H. Bhrawy and M. A. Alghamdi. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
The spectral collocation approximations based on Legendre polynomials are used to compute the numerical solution of time-
dependent Fisher’s type problems. The spatial derivatives are collocated at a Legendre-Gauss-Lobatto interpolation nodes. The
proposed method has the advantage of reducing the problem to a system of ordinary differential equations in time. The four-stage
A-stable implicit Runge-Kutta scheme is applied to solve the resulted system of first order in time. Numerical results show that the
Legendre-Gauss-Lobatto collocation method is of high accuracy and is efficient for solving the Fisher’s type equations. Also the
results demonstrate that the proposed method is powerful algorithm for solving the nonlinear partial differential equations.
tions based on the Legendre polynomials, and in Section 4 the (𝑢, V)𝑤 = ∑ 𝑢 (𝑥𝑁,𝑗 ) V (𝑥𝑁,𝑗 ) 𝜛𝑁,𝑗 , (10)
𝑗=0
proposed method is implemented to obtain some numerical
results for three problems of Fisher-type equations with
where 𝑥𝑁,𝑗 and 𝜛𝑁,𝑗 are the nodes and the corresponding
known exact solutions. Finally, a brief conclusion is provided
weights of the Legendre-Gauss-Lobatto quadrature formula
in Section 5.
on the interval (−1, 1), respectively.
2. Legendre Polynomials
3. Legendre Spectral Collocation Method
The Legendre polynomials 𝐿 𝑘 (𝑥) (𝑘 = 0, 1, . . .,) satisfy the
following Rodrigues’ formula: Because of the pseudospectral method is an efficient and
accurate numerical scheme for solving various problems in
(−1)𝑘 𝑘 𝑘 physical space, including variable coefficient and singularity
𝐿 𝑘 (𝑥) = 𝑘
𝐷 ((1 − 𝑥2 ) ) ; (2)
(see, [54, 55]), we propose this method based on Legendre
2 𝑘!
polynomials for approximating the solution of the nonlinear
we recall also that 𝐿 𝑘 (𝑥) is a polynomial of degree 𝑘, and generalized Burger-Fisher model equation and Fisher model
therefore, the 𝑞th derivative of 𝐿 𝑘 (𝑥) is given by with variable coefficient.
𝑘−𝑞
(𝑞)
𝐿 𝑘 (𝑥) = ∑ 𝐶𝑞 (𝑘, 𝑖) 𝐿 𝑖 (𝑥) , (3) 3.1. (1+1)-Dimensional Generalized Burger-Fisher Equation.
𝑖=0(𝑘+𝑖=even) In this subsection, we derive a Legendre pseudospectral
Abstract and Applied Analysis 3
algorithm to solve numerically the generalized Burger-Fisher Subsequently, the second-order spatial partial derivatives
problem: of 𝑢(𝑥, 𝑡) may be written at the same collocation nodes as
𝑁 𝑁
𝑢𝑡 + ]𝑢𝛿 𝑢𝑥 − 𝑢𝑥𝑥 − 𝛾𝑢 (1 − 𝑢𝛿 ) = 0, (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] , 1
𝑢𝑥𝑥 (𝑥𝑛 , 𝑡) = ∑ ( ∑ 𝐿 𝑗 (𝑥𝑖 ) (𝐿 𝑗 (𝑥𝑛 )) 𝜛𝑁,𝑖 ) 𝑢 (𝑥𝑖 , 𝑡)
(11) 𝑖=0 𝑗=0 ℎ𝑗
𝑁
where 𝐷 = {𝑥 : −1 ≤ 𝑥 ≤ 1}. Subject to
= ∑𝐵𝑛𝑖 𝑢 (𝑥𝑖 , 𝑡)
𝑖=0
𝑢 (𝑥, 𝑡) = 𝑔 (𝑡) , 𝑥 = −1, 1, (12)
𝑁
𝑢 (𝑥, 0) = 𝑓 (𝑥) , 𝑥 ∈ 𝐷. (13) = ∑𝐵𝑛𝑖 𝑢𝑖 ,
𝑖=0
(19)
In the following, we shall derive an efficient algorithm for
the numerical solution of (11)–(13). Let the approximation
where
of 𝑢(𝑥, 𝑡) be given in terms of the Legendre polynomials
expansion: 𝑁
1
𝐵𝑛𝑖 = ∑ 𝐿 𝑗 (𝑥𝑖 ) (𝐿 𝑗 (𝑥𝑛 )) 𝜛𝑁,𝑖 . (20)
𝑁 ℎ
𝑗=0 𝑗
𝑇
𝑢 (𝑥, 𝑡) = ∑ 𝑎𝑗 (𝑡) 𝐿 𝑗 (𝑥) , a = (𝑎0 , 𝑎1 , . . . , 𝑎𝑁) . (14)
𝑗=0 In collocation methods, one specifically seeks the approx-
imate solution such that the problem (11) is satisfied exactly
Making use of relations (8) and (10) gives at the Legendre Gauss-Lobatto set of interpolation points 𝑥𝑛 ;
𝑛 = 1, . . . , 𝑁 − 1. The approximation is exact at the 𝑁 −
𝑁 1 collocation points. Therefore, (11) after using relations (17)–
1 𝑁
𝑢 (𝑥, 𝑡) = ∑ ( ∑𝐿 (𝑥 ) 𝐿 (𝑥) 𝜛𝑁,𝑖 𝑢 (𝑥𝑖 , 𝑡)) (15) (20), can be written as
𝑗=0 ℎ𝑗 𝑖=0 𝑗 𝑖 𝑗
𝑁
𝑢𝑛⋅ (𝑡) + ]𝑢𝑛𝛿 (𝑡) ∑𝐴 𝑛𝑖 𝑢𝑖 (𝑡)
or equivalently 𝑖=0
𝑁 (21)
𝑁 𝑁
1 − ∑𝐵𝑛𝑖 𝑢𝑖 (𝑡) − 𝛾𝑢𝑛 (𝑡) (1 − 𝑢𝑛𝛿 (𝑡)) = 0,
𝑢 (𝑥, 𝑡) = ∑ ( ∑ 𝐿 𝑗 (𝑥𝑖 ) 𝐿 𝑗 (𝑥) 𝜛𝑁,𝑖 ) 𝑢 (𝑥𝑖 , 𝑡) . (16)
𝑖=0 𝑗=0 ℎ𝑗
𝑖=0
𝑛 = 1, . . . , 𝑁 − 1,
The Gauss-Lobatto points were introduced by way
of (9). We then saw that the polynomial approximation where 𝑢𝑛 (𝑡) = 𝑢(𝑥𝑛 , 𝑡) and 𝑢𝑛⋅ (𝑡) = 𝜕𝑢𝑛⋅ (𝑡)/𝜕𝑡.
𝑢(𝑥, 𝑡) can be characterized by (𝑁 + 1) nodal values 𝑢(𝑥𝑖 , 𝑡). Now the two values 𝑢0 (𝑡) and 𝑢𝑁(𝑡) can be determined
The approximation of the spatial partial derivatives of first- from the boundary conditions (12), then (21) can be reformu-
order for 𝑢(𝑥, 𝑡) can be computed at the Legendre Gauss- lated as
Lobatto interpolation nodes as
𝑁−1
𝑁 𝑁
𝑢𝑛⋅ (𝑡) + ]𝑢𝑛𝛿 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡)
1 𝑖=1
𝑢𝑥 (𝑥𝑛 , 𝑡) = ∑ ( ∑ 𝐿 𝑗 (𝑥𝑖 ) (𝐿 𝑗 (𝑥𝑛 )) 𝜛𝑁,𝑖 ) 𝑢 (𝑥𝑖 , 𝑡)
𝑖=0 ℎ
𝑗=0 𝑗 𝑁−1
(22)
− ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡) + ]𝑢𝑛𝛿 (𝑡) 𝑑𝑛 (𝑡) − 𝑑̃𝑛 (𝑡)
𝑁
𝑖=1
= ∑𝐴 𝑛𝑖 𝑢 (𝑥𝑖 , 𝑡) (17)
𝑖=0 − 𝛾𝑢𝑛 (𝑡) (1 − 𝑢𝑛𝛿 (𝑡)) = 0, 𝑛 = 1, . . . , 𝑁 − 1,
𝑁
= ∑𝐴 𝑛𝑖 𝑢𝑖 (𝑡) , 𝑛 = 0, 1, . . . , 𝑁, where
𝑖=0
𝑑𝑛 (𝑡) = 𝐴 𝑛0 𝑢0 (𝑡) + 𝐴 𝑛𝑁𝑢𝑁(𝑡),
where (23)
𝑑̃𝑛 (𝑡) = 𝐵𝑛0 𝑢0 (𝑡) + 𝐵𝑛𝑁𝑢𝑁(𝑡).
𝑁
1
𝐴 𝑛𝑖 = ∑ 𝐿 𝑗 (𝑥𝑖 ) (𝐿 𝑗 (𝑥𝑛 )) 𝜛𝑁,𝑖 , Approximation (22) automatically satisfies the boundary
𝑗=0 ℎ𝑗 (18) conditions (12), but we need an initial condition for each
of the 𝑢𝑛 (𝑡) to integrate (22) in time. The initial condi-
𝑢𝑖 (𝑡) = 𝑢 (𝑥𝑖 , 𝑡) . tion is usually taken to be the interpolant of the initial
4 Abstract and Applied Analysis
function 𝑓(𝑥); that is 𝑢𝑛 (0) = 𝑓(𝑥𝑛 ). Therefore, the approx- Table 1: Absolute errors for Example 1.
imation of (11)–(13) is reduced to the solution of system of
𝑥 𝑡 𝐸(𝑥, 𝑡) 𝑥 𝑡 𝐸(𝑥, 𝑡)
ordinary differential equations in time. Consider
−1 5.56 × 10−11 −1 8.93 × 10−11
𝑁−1 𝑁−1
−0.5 1.46 × 10−8 −0.5 8.23 × 10−9
𝑢𝑛⋅ (𝑡) + ]𝑢𝑛𝛿 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡) − ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡)
0 0.1 1.95 × 10−8 0 0.2 1.41 × 10−8
𝑖=1 𝑖=1
0.5 1.55 × 10−8 0.5 1.17 × 10−8
+ ]𝑢𝑛𝛿 (𝑡) 𝑑𝑛 (𝑡) − 𝑑̃𝑛 (𝑡) − 𝛾𝑢𝑛 (𝑡) (1 − 𝑢𝑛𝛿 (𝑡)) = 0, (24) 1 5.56 × 10−11 1 8.93 × 10−11
𝑛 = 1, . . . , 𝑁 − 1,
which can be written in the matrix form
𝑢𝑛 (0) = 𝑓 (𝑥𝑛 ) .
𝑈⋅ (𝑡) = 𝐹 (𝑡, 𝑢 (𝑡)) ,
Let us denote (30)
𝑇
𝑈 (0) = 𝑓,
𝑈⋅ (𝑡) = [𝑢1⋅ (𝑡) , 𝑢2⋅ (𝑡) , . . . , 𝑢𝑁−1
⋅
(𝑡)] ,
where
𝑇
𝑈 (0) = [𝑢1 (0), 𝑢2 (0), . . . , 𝑢𝑁−1 (0)] ,
𝐹𝑛 (𝑡, 𝑢 (𝑡))
𝑇
𝑓 = [𝑓(𝑥1 ), 𝑓(𝑥2 ), . . . , 𝑓(𝑥𝑁−1 )] , 𝑁−1
= 𝑏 (𝑡) ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡) + 𝑏 (𝑡) 𝑑̃𝑛 (𝑡) + 𝑐𝑢𝑛 (𝑡) (1 − 𝑢𝑛 (𝑡)) ,
𝑇
𝐹 (𝑡, 𝑢 (𝑡)) = [𝐹1 (𝑡, 𝑢 (𝑡)) , 𝐹2 (𝑡, 𝑢 (𝑡)) , . . . , 𝐹𝑁−1 (𝑡, 𝑢 (𝑡))] , 𝑖=1
𝑁−1 𝑁−1 𝑛 = 1, . . . , 𝑁 − 1.
𝐹𝑛 (𝑡, 𝑢 (𝑡)) = − ]𝑢𝑛𝛿 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡) + ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡) (31)
𝑖=1 𝑖=1
− ]𝑢𝑛𝛿 (𝑡) 𝑑𝑛 (𝑡) + 𝑑̃𝑛 (𝑡) + 𝛾𝑢𝑛 (𝑡) (1 − 𝑢𝑛𝛿 (𝑡)) , 4. Numerical Examples
×10−8
4
0.7
1.0 3 1.0
0.6 E
̃
u 2
0.5
1
0.4
0
−1.0 0.5 −1.0 0.5 t
t
−0.5 −0.5
0.0 0.0
x x
0.5 0.5
0.0 0.0
1.0 1.0
(a) (b)
Figure 1: The result of the L-GL-C method at 𝛾 = 10−2 and 𝑁 = 20. (a) The approximate solution. (b) The absolute error.
0.6
(35)
0.5
where 𝐷 = {𝑥 : −1 < 𝑥 < 1}. Subject to
0.4
−1.0 −0.5 0.0 0.5 1.0
x 𝑢 (1, 𝑡)
u(x, 0) ̃ (x, 0.5)
u
1 1 ]𝛿
̃ (x, 0)
u u(x, 0.9) = ( − tanh [
u(x, 0.5) ̃ (x, 0.9)
u 2 2 2 (𝛿 + 1)
1/𝛿
Figure 2: The curves of approximate solutions and the exact ] 𝛾 (𝛿 + 1)
solutions of problem (32) at 𝑡 = 0.0, 𝑡 = 0.5, and 𝑡 = 0.9 with 𝛾 = × (1 − ( + ) 𝑡)]) ,
𝛿+1 ]
10−2 and 𝑁 = 20.
𝑢 (−1, 𝑡)
1 1 ]𝛿
= ( + tanh [
−2
In case of 𝛾 = 10 and 𝑁 = 20, the approximate solu- 2 2 2 (𝛿 + 1)
tion and absolute errors of problem (32) are displayed in ] 𝛾 (𝛿 + 1) 1/𝛿
Figures 1(a) and 1(b), respectively. In Figure 2, we plotted the × (1 + ( + ) 𝑡)]) ,
curves of approximate solutions and exact solutions of pro- 𝛿+1 ]
blem (32) for different values of 𝑡 (0.0,0.5 and 0.9) with 𝛾 = 1 1 ]𝛿 1/𝛿
10−2 and 𝑁 = 20. It is clear from this figure that approximate 𝑢 (𝑥, 0) = ( − tanh [ 𝑥]) , 𝑥 ∈ 𝐷.
2 2 2 (𝛿 + 1)
solutions and exact solutions completely coincide for the
chosen values of 𝑡. (36)
6 Abstract and Applied Analysis
×10−10
8
1.0 6 1.0
0.502
E 4
̃
u
0.500 2
0
−1.0 0.5 t −1.0 0.5 t
−0.5 −0.5
0.0 0.0
x x
0.5 0.5
1.0 0.0 1.0 0.0
(a) (b)
Figure 3: The result of the L-GL-C method at ] = 𝛾 = 10−2 , 𝛿 = 1, and 𝑁 = 20. (a) The approximate solution. (b)The absolute error.
0.5030 The absolute errors for problem (35) are listed in Table 2
using the L-GL-C method with ] = 𝛾 = 10−2 , 𝑁 = 20, and
0.5025 various choices of 𝛿.
To illustrate the effectiveness of the Legendre pseudospec-
0.5020
u and u
̃
×10−8
Figure 5: The result of the L-GL-C method at ] = 𝛾 = 10−2 , 𝛿 = 2, and 𝑁 = 20. (a) The approximate solution. (b)The absolute error.
1 𝑎 𝜇 5𝑎
𝑢 (−1, 𝑡) = coth ( 𝑡 + 𝑐) sech2 (− + 𝑡) The exact solution of (38) is
4 6 2 12
1 𝑎 𝜇𝑥 5𝑎
1 𝜇 5𝑎 1 𝑢 (𝑥, 𝑡) = coth ( 𝑡 + 𝑐) sech2 ( + 𝑡)
+ tanh (− + 𝑡) + , 4 6 2 12
2 2 12 2 (41)
(39) 1 𝜇𝑥 5𝑎 1
+ tanh ( + 𝑡) + .
2 2 12 2
1 𝜇𝑥
𝑢 (𝑥, 0) = coth (𝑐) sech2 ( ) Table 3 lists the absolute errors for problem (38) using
4 2
(40) the L-GL-C method. From numerical results of this table, it
1 𝜇𝑥 1 can be concluded that the numerical solutions are in excellent
+ tanh ( ) + , 𝑥 ∈ 𝐷.
2 2 2 agreement with the exact solutions.
8 Abstract and Applied Analysis
[19] M. R. Eslahchi, M. Dehghan, and M. Parvizi, “Application of Applied Mathematics and Computation, vol. 216, no. 8, pp. 2472–
the collocation method for solving nonlinear fractional integro- 2478, 2010.
differential equations,” Journal of Computational and Applied [39] X. Y. Chen, Numerical methods for the Burgers-Fisher equation
Mathematics, vol. 257, pp. 105–128, 2013. [M.S. thesis], University of Aeronautics and Astronautics, Nan-
[20] Y. Yangy and Y. Huang, “Spectral-collocation methods for jing, China, 2007.
fractional pantograph delay-integrodifferential equations,” [40] R. E. Mickens and A. B. Gumel, “Construction and analysis of
Advances in Mathematical Physics. In press. a non-standard finite difference scheme for the Burgers-Fisher
[21] A. H. Bhrawy and A. S. Alofi, “The operational matrix of frac- equation,” Journal of Sound and Vibration, vol. 257, no. 4, pp.
tional integration for shifted Chebyshev polynomials,” Applied 791–797, 2002.
Mathematics Letters, vol. 26, no. 1, pp. 25–31, 2013. [41] R. E. Mickens, “A best finite-difference scheme for the Fisher
[22] R. A. Fisher, “The wave of advance of advantageous genes,” equation,” Numerical Methods for Partial Differential Equations,
Annals of Eugenics, vol. 7, pp. 335–369, 1937. vol. 10, no. 5, pp. 581–585, 1994.
[23] A. J. Khattak, “A computational meshless method for the gen- [42] R. E. Mickens, “Relation between the time and space step-
eralized Burger’s-Huxley equation,” Applied Mathematical Mod- sizes in nonstandard finite-difference schemes for the Fisher
elling, vol. 33, no. 9, pp. 3718–3729, 2009. equation,” Numerical Methods for Partial Differential Equations,
[24] N. F. Britton, Reactiondiffusion Equations and Their Applications vol. 15, no. 1, pp. 51–55, 1997.
to Biology, Academic Press, London, UK, 1986. [43] N. Parekh and S. Puri, “A new numerical scheme for the Fisher
[25] D. A. Frank, Diffusion and Heat Exchange in Chemical Kinetics, equation,” Journal of Physics A, vol. 23, no. 21, pp. L1085–L1091,
Princeton University Press, Princeton, NJ, USA, 1955. 1990.
[26] A. Wazwaz, “The extended tanh method for abundant solitary [44] R. Rizwan-Uddin, “Comparison of the nodal integral method
wave solutions of nonlinear wave equations,” Applied Mathe- and nonstandard finite-difference schemes for the Fisher equa-
matics and Computation, vol. 187, no. 2, pp. 1131–1142, 2007. tion,” SIAM Journal on Scientific Computing, vol. 22, no. 6, pp.
[27] W. Malfliet, “Solitary wave solutions of nonlinear wave equa- 1926–1942, 2001.
tions,” American Journal of Physics, vol. 60, no. 7, pp. 650–654, [45] R. Chernma, “Exact and numerical solutions of tiie generalized
1992. fisher equation,” Reports on Mathematkxl Physics, vol. 47, pp.
[28] Y. Tan, H. Xu, and S.-J. Liao, “Explicit series solution of travel- 393–411, 2001.
ling waves with a front of Fisher equation,” Chaos, Solitons and [46] A. Şahin, I. Dag, and B. Saka, “A B-spline algorithm for the
Fractals, vol. 31, no. 2, pp. 462–472, 2007. numerical solution of Fisher’s equation,” Kybernetes, vol. 37, no.
[29] H. N. A. Ismail, K. Raslan, and A. A. A. Rabboh, “Ado- 2, pp. 326–342, 2008.
mian decomposition method for Burger’s-Huxley and Burger’s- [47] J. Roessler and H. Hüssner, “Numerical solution of the 1 +
Fisher equations,” Applied Mathematics and Computation, vol. 2 dimensional Fisher’s equation by finite elements and the
159, no. 1, pp. 291–301, 2004. Galerkin method,” Mathematical and Computer Modelling, vol.
[30] J. Canosa, “Diffusion in nonlinear multiplicative media,” Journal 25, no. 3, pp. 57–67, 1997.
of Mathematical Physics, vol. 10, no. 10, pp. 1862–1868, 1969. [48] J. Roessler and H. Hüssner, “Numerical solution of the 1 +
[31] A. Öǧün and C. Kart, “Exact solutions of Fisher and generalized 2 dimensional Fisher’s equation by finite elements and the
Fisher equations with variable coefficients,” Acta Mathematicae Galerkin method,” Mathematical and Computer Modelling, vol.
Applicatae Sinica, vol. 23, no. 4, pp. 563–568, 2007. 25, no. 3, pp. 57–67, 1997.
[32] M. D. Gunzburger, L. S. Hou, and W. Zhu, “Fully discrete finite [49] E. H. Doha, A. H. Bhrawy, R. M. Hafez, and M. A. Abdelkawy,
element approximations of the forced Fisher equation,” Journal “A Chebyshev-Gauss-Radau scheme for nonlinear hyperbolic
of Mathematical Analysis and Applications, vol. 313, no. 2, pp. system of first order,” Applied Mathematics and Information
419–440, 2006. Science, vol. 8, no. 2, pp. 1–10, 2014.
[33] I. Dag, A. Şahin, and A. Korkmaz, “Numerical investigation [50] A. H. Bhrawy, “A Jacobi-Gauss-Lobatto collocation method
of the solution of Fisher’s equation via the B-spline galerkin for solving generalized Fitzhugh-Nagumo equation with time-
method,” Numerical Methods for Partial Differential Equations, dependent coefficients,” Applied Mathematics and Computation,
vol. 26, no. 6, pp. 1483–1503, 2010. vol. 222, pp. 255–264, 2013.
[34] M. Bastani and D. K. Salkuyeh, “A highly accurate method to [51] E. H. Doha, D. Baleanu, A. H. Bhrawy, and M. A. Abdelkawy, “A
solve Fisher’s equation,” Pramana, vol. 78, no. 3, pp. 335–346, Jacobi collocation method for solving nonlinear Burgers’-type
2012. equations,” Abstract and Applied Analysis, vol. 2013, Article ID
[35] R. C. Mittal and R. K. Jain, “Numerical solutions of nonlinear 760542, 12 pages, 2013.
Fisher’s reaction-diffusion equation with modified cubic B- [52] A. H. Bhrawy, L. M. Assas, and M. A. Alghamdi, “Fast spec-
spline collocation method,” Mathematical Sciences, vol. 7, article tral collocation method for solving nonlinear time-delayed Bur-
12, 2013. gers-type equations with positive power terms,” Abstract and
[36] J. Gazdag and J. Canosa, “Numerical solution of Fisher’s equa- Applied Analysis, vol. 2013, Article ID 741278, 12 pages, 2013.
tion,” Journal of Applied Probability, vol. 11, pp. 445–457, 1974. [53] A. H. Bhrawy, L. M. Assas, and M. A. Alghamdi, “An efficient
[37] T. Zhao, C. Li, Z. Zang, and Y. Wu, “Chebyshev-Legendre spectral collocation algorithm for nonlinear Phi-four equa-
pseudo-spectral method for the generalised Burgers-Fisher tions,” Boundary Value Problems, vol. 2013, article 87, 16 pages,
equation,” Applied Mathematical Modelling, vol. 36, no. 3, pp. 2013.
1046–1056, 2012. [54] A. Saadatmandi, “Bernstein operational matrix of fractional
[38] G. Gürarslan, “Numerical modelling of linear and nonlin- derivatives and its applications,” Applied Mathematical Mod-
ear diffusion equations by compact finite difference method,” elling, 2013.
10 Abstract and Applied Analysis
Research Article
A Jacobi Collocation Method for Solving Nonlinear
Burgers-Type Equations
Copyright © 2013 E. H. Doha et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We solve three versions of nonlinear time-dependent Burgers-type equations. The Jacobi-Gauss-Lobatto points are used as
collocation nodes for spatial derivatives. This approach has the advantage of obtaining the solution in terms of the Jacobi parameters
𝛼 and 𝛽. In addition, the problem is reduced to the solution of the system of ordinary differential equations (SODEs) in time. This
system may be solved by any standard numerical techniques. Numerical solutions obtained by this method when compared with the
exact solutions reveal that the obtained solutions produce high-accurate results. Numerical results show that the proposed method
is of high accuracy and is efficient to solve the Burgers-type equation. Also the results demonstrate that the proposed method is a
powerful algorithm to solve the nonlinear partial differential equations.
with the Legendre spectral method to solve time-dependent the system of (𝑁 + 1) ordinary differential equations (ODEs)
partial differential equations and gave error estimates of the in time. This system can be solved by one of the possible
method. Tang and Ma [17] introduced the Legendre spectral methods of numerical analysis such as the Euler method,
method together with the Fourier approximation in spatial Midpoint method, and the Runge-Kutta method. Finally, the
for time-dependent first-order hyperbolic equations with accuracy of the proposed method is demonstrated by test
periodic boundary conditions. Recently, the author of [18] problems.
proposed an accurate numerical algorithm to solve the gen- The remainder of the paper is organized as follows.
eralized Fitzhugh-Nagumo equation with time-dependent In the next section, we introduce some properties of the
coefficients. Jacobi polynomials. In Section 3, the way of constructing
In [20], Bateman introduced the one-dimensional quasi- the Gauss-Lobatto collocation technique for nonlinear time-
linear parabolic partial differential equation, while Burgers dependent Burgers-type equations is described using the
[21] developed it as mathematical modeling of turbulence, Jacobi polynomials, and in Section 4 the proposed method
and it is referred as one-dimensional Burgers’ equation. is applied to three problems of nonlinear time-dependent
Many authors gave different solutions for Burgers’ equation Burgers-type equations. Finally, some concluding remarks
by using various methods. Kadalbajoo and Awasthi [22] are given in Section 5.
and Gülsu [23] used a finite-difference approach method to
find solutions of one-dimensional Burgers’ equation. Crank-
2. Some Properties of Jacobi Polynomials
Nicolson scheme for Burgers’ equation is developed by Kim,
[24]. Nguyen and Reynen [25, 26], Gardner et al. [27, 28] (𝛼,𝛽)
The standard Jacobi polynomials of degree 𝑘 (𝑃𝑘 (𝑥), 𝑘 =
and Kutluay et al. [29] used methods based on the Petrov- 0, 1, . . .) with the parameters 𝛼 > −1, 𝛽 > −1 are satisfying
Galerkin, Least-Squares finite-elements, and B-spline finite the following relations:
element methods to solve Burgers’ equation. A method
based on collocation of modified cubic B-splines over finite (𝛼,𝛽) (𝛼,𝛽)
𝑃𝑘 (−𝑥) = (−1)𝑘 𝑃𝑘 (𝑥) ,
elements has been investigated by Mittal and Jain in [30].
In this work, we propose a J-GL-C method to numerically
(𝛼,𝛽) (−1)𝑘 Γ (𝑘 + 𝛽 + 1)
solve the following three nonlinear time-dependent Burgers’- 𝑃𝑘 (−1) = , (4)
type equations: 𝑘!Γ (𝛽 + 1)
respectively. Now, we introduce the following discrete inner and accordingly, (14) takes the form
product and norm:
𝑢 (𝑥, 𝑡)
𝑁
(𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
(𝑢, V)𝑤(𝛼,𝛽) = ∑ 𝑢 (𝑥𝑁,𝑗 ) V (𝑥𝑁,𝑗 ) 𝜛𝑁,𝑗 , 𝑁
1 𝑁 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
𝑗=0 (8) = ∑( ∑𝑃 (𝑥𝑁,𝑖 ) 𝜛𝑁,𝑖 𝑢 (𝑥𝑁,𝑖 , 𝑡)) 𝑃𝑗 (𝑥) ,
𝑗=0 ℎ𝑗 𝑖=0 𝑗
‖𝑢‖𝑤(𝛼,𝛽) = (𝑢, 𝑢)1/2
𝑤(𝛼,𝛽)
. (15)
For 𝛼 = 𝛽, one recovers the ultraspherical polynomials or equivalently takes the form
(symmetric Jacobi polynomials) and for 𝛼 = 𝛽 = ∓1/2, 𝛼 =
𝛽 = 0, the Chebyshev of the first and second kinds 𝑢 (𝑥, 𝑡)
and the Legendre polynomials, respectively; and for the
nonsymmetric Jacobi polynomials, the two important special 𝑁 𝑁
1 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
cases 𝛼 = −𝛽 = ±1/2 (the Chebyshev polynomials of the third = ∑ (∑ 𝑃𝑗 (𝑥𝑁,𝑖 ) 𝑃𝑗 (𝑥) 𝜛𝑁,𝑖 ) 𝑢 (𝑥𝑁,𝑖 , 𝑡) .
𝑖=0 ℎ
𝑗=0 𝑗
and fourth kinds) are also recovered.
(16)
3. Jacobi Spectral Collocation Method The spatial partial derivatives with respect to 𝑥 in (9) can be
computed at the J-GL-C points to give
Since the collocation method approximates the differential
equations in physical space, it is very easy to implement and (𝛼,𝛽)
be adaptable to various problems, including variable coeffi- 𝑢𝑥 (𝑥𝑁,𝑛 , 𝑡)
cient and nonlinear differential equations (see, for instance 𝑁 𝑁
[4, 6]). In this section, we develop the J-GL-C method to 1 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
= ∑(∑ 𝑃𝑗 (𝑥𝑁,𝑖 ) (𝑃𝑗 (𝑥𝑁,𝑛 )) 𝜛𝑁,𝑖 )𝑢(𝑥𝑁,𝑖 , 𝑡)
numerically solve the Burgers-type equations. ℎ
𝑖=0 𝑗=0 𝑗
𝑁
3.1. (1 + 1)-Dimensional Burgers’ Equation. In 1939, Burg- (𝛼,𝛽)
ers has simplified the Navier-Stokes equation by dropping = ∑𝐴 𝑛𝑖 𝑢 (𝑥𝑁,𝑖 , 𝑡) , 𝑛 = 0, 1, . . . , 𝑁,
𝑖=0
the pressure term to obtain his one-dimensional Burgers’
equation. This equation has many applications in applied (𝛼,𝛽)
𝑢𝑥𝑥 (𝑥𝑁,𝑛 , 𝑡)
mathematics, such as modeling of gas dynamics [36, 37],
modeling of fluid dynamics, turbulence, boundary layer 𝑁 𝑁
1 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
behavior, shock wave formation, and traffic flow [38]. In this = ∑ (∑ 𝑃𝑗 (𝑥𝑁,𝑖 ) (𝑃𝑗 (𝑥𝑁,𝑛 )) 𝜛𝑁,𝑖 )𝑢(𝑥𝑁,𝑖 , 𝑡)
subsection, we derive a J-GL-C method to solve numerically 𝑖=0 ℎ
𝑗=0 𝑗
the (1 + 1)-dimensional Burgers’ model problem:
𝑁
(𝛼,𝛽)
𝑢𝑡 + ]𝑢𝑢𝑥 − 𝜇𝑢𝑥𝑥 = 0; (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] , (9) = ∑𝐵𝑛𝑖 𝑢(𝑥𝑁,𝑖 , 𝑡) ,
𝑖=0
(17)
where
where
𝐷 = {𝑥 : −1 ≤ 𝑥 ≤ 1} , (10)
𝑁
1 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
subject to the boundary conditions 𝐴 𝑛𝑖 = ∑ 𝑃𝑗 (𝑥𝑁,𝑖 ) (𝑃𝑗 (𝑥𝑁,𝑛 )) 𝜛𝑁,𝑖 ,
𝑗=0 ℎ𝑗
𝑢 (−1, 𝑡) = 𝑔1 (𝑡) , 𝑢 (1, 𝑡) = 𝑔2 (𝑡) , 𝑡 ∈ [0, 𝑇] , (11) (18)
𝑁
1 (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽)
and the initial condition 𝐵𝑛𝑖 = ∑ 𝑃𝑗 (𝑥𝑁,𝑖 ) (𝑃𝑗 (𝑥𝑁,𝑛 )) 𝜛𝑁,𝑖 .
𝑗=0 ℎ𝑗
𝑢 (𝑥, 0) = 𝑓 (𝑥) , 𝑥 ∈ 𝐷. (12)
Making use of (17) and (18) enables one to rewrite (9) in the
Now we assume that form:
𝑁 𝑁 𝑁
Using Equation (19) and using the two-point boundary and physics applications, gas dynamic, and traffic flow. The
conditions (11) generate a system of (𝑁 − 1) ODEs in time: Burger-Fisher equation can be written in the following form:
𝑁−1 𝑁−1
𝑢𝑡 = 𝑢𝑥𝑥 − ]𝑢𝑢𝑥 + 𝛾𝑢 (1 − 𝑢) ; (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] , (26)
𝑢̇𝑛 (𝑡) + ]𝑢𝑛 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡) − 𝜇 ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡)
𝑖=1 𝑖=1 (21) where
+]𝑑𝑛 (𝑡) − 𝜇𝑑̃𝑛 (𝑡) = 0, 𝑛 = 1, . . . , 𝑁 − 1,
𝐷 = {𝑥 : −1 < 𝑥 < 1} , (27)
where
subject to the boundary conditions
𝑑𝑛 (𝑡) = 𝐴 𝑛𝑜 𝑔1 (𝑡) + 𝐴 𝑛𝑁𝑔2 (𝑡) ,
(22) 𝑢 (−1, 𝑡) = 𝑔1 (𝑡) , 𝑢 (1, 𝑡) = 𝑔2 (𝑡) , (28)
𝑑̃𝑛 (𝑡) = 𝐵𝑛𝑜 𝑔1 (𝑡) + 𝐵𝑛𝑁𝑔2 (𝑡) .
and the initial condition
Then the problem (9)–(12) transforms to the SODEs:
𝑢 (𝑥, 0) = 𝑓 (𝑥) , 𝑥 ∈ 𝐷. (29)
𝑁−1 𝑁−1
𝑢̇𝑛 (𝑡) + ]𝑢𝑛 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡) − 𝜇 ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡)
𝑖=1 𝑖=1
The same procedure of Section 3.1 can be used to reduce
(23) (26)–(29) to the system of nonlinear differential equations
+ ]𝑑𝑛 (𝑡) − 𝜇𝑑̃𝑛 (𝑡) = 0, 𝑛 = 1, . . . , 𝑁 − 1, in the unknown expansion coefficients of the sought-for
semianalytical solution. This system is solved by using the
(𝛼,𝛽)
𝑢𝑛 (0) = 𝑓 (𝑥𝑁,𝑛 ) , implicit Runge-Kutta method.
which may be written in the following matrix form: 3.3. (1 + 1)-Dimensional Generalized Burgers-Huxley Equa-
tion. The Huxley equation is a nonlinear partial differential
u̇(𝑡) = F (𝑡, 𝑢 (𝑡)) , equation of second order of the form
(24)
u (0) = f,
𝑢𝑡 − 𝑢𝑥𝑥 − 𝑢 (𝑘 − 𝑢) (𝑢 − 1) = 0; 𝑘 ≠
0. (30)
where
It is an evolution equation that describes the nerve propaga-
𝑇
u̇(𝑡) = [𝑢̇1 (𝑡) , 𝑢̇2 (𝑡) , . . . , 𝑢̇𝑁−1 (𝑡)] , tion [47] in biology from which molecular CB properties can
be calculated. It also gives a phenomenological description of
𝑇
f = [𝑓 (𝑥𝑁,1 ) , 𝑓 (𝑥𝑁,2 ) , . . . , 𝑓 (𝑥𝑁,𝑁−1 )] , the behavior of the myosin heads II. In addition to this non-
linear evolution equation, combined forms of this equation
𝑇
F (𝑡, 𝑢 (𝑡)) = [𝐹1 (𝑡, 𝑢 (𝑡)) , 𝐹2 (𝑡, 𝑢 (𝑡)) , . . . , 𝐹𝑁−1 (𝑡, 𝑢 (𝑡))] , and Burgers’ equation will be investigated. It is interesting to
point out that this equation includes the convection term 𝑢𝑥
𝑁−1 𝑁−1
and the dissipation term 𝑢𝑥𝑥 in addition to other terms. In this
𝐹𝑛 (𝑡, 𝑢 (𝑡)) = −]𝑢𝑛 (𝑡) ∑ 𝐴 𝑛𝑖 𝑢𝑖 (𝑡) + 𝜇 ∑ 𝐵𝑛𝑖 𝑢𝑖 (𝑡) subsection, we derive J-GL-C method to solve numerically
𝑖=1 𝑖=1
the (1+1)-dimensional generalized Burgers-Huxley equation:
− ]𝑑𝑛 (𝑡) + 𝜇𝑑̃𝑛 (𝑡) , 𝑛 = 1, . . . , 𝑁 − 1.
(25) 𝑢𝑡 + ]𝑢𝛿 𝑢𝑥 − 𝑢𝑥𝑥 − 𝜂𝑢 (1 − 𝑢𝛿 ) (𝑢𝛿 − 𝛾) = 0,
(31)
The SODEs (24) in time may be solved using any standard (𝑥, 𝑡) ∈ 𝐷 × [0, 𝑇] ,
technique, like the implicit Runge-Kutta method.
where
3.2. (1 + 1)-Dimensional Burger-Fisher Equation. The Burger-
Fisher equation is a combined form of Fisher and Burgers’ 𝐷 = {𝑥 : −1 < 𝑥 < 1} , (32)
equations. The Fisher equation was firstly introduced by
Fisher in [39] to describe the propagation of a mutant subject to the boundary conditions:
gene. This equation has a wide range of applications in a
large number of the fields of chemical kinetics [40], logistic 𝑢 (−1, 𝑡) = 𝑔1 (𝑡) , 𝑢 (1, 𝑡) = 𝑔2 (𝑡) , (33)
population growth [41], flame propagation [42], population
in one-dimensional habitual [43], neutron population in a and the initial condition:
nuclear reaction [44], neurophysiology [45], autocatalytic
chemical reactions [19], branching the Brownian motion 𝑢 (𝑥, 0) = 𝑓 (𝑥) , 𝑥 ∈ 𝐷. (34)
processes [40], and nuclear reactor theory [46]. Moreover,
the Burger-Fisher equation has a wide range of applications in The same procedure of Sections 3.1 and 3.2 is used to solve
various fields of financial mathematics, applied mathematics numerically (30)–(34).
Abstract and Applied Analysis 5
[ −]𝛿 + 𝛿√] + 4𝜂 (1 + 𝛿) ]]
2
choices of the Jacobi parameters 𝛼 and 𝛽 reveals that the [𝛾 𝛾
present method is very effective and convenient for all choices = [ + tanh [𝑥𝛾 ]] ,
2 2 4 (𝛿 + 1)
of 𝛼 and 𝛽. We consider the following three examples. [ [ ]]
× (𝐴 − (]𝛾 (1+𝛿−𝛾) (√]2 +4𝜂 (1 + 𝛿)−])) where 𝑢(𝑥, 𝑡) and 𝑢̃(𝑥, 𝑡), is the exact solution and the
approximate solution at the point (𝑥, 𝑡), respectively.
1/𝛿 In the cases of 𝛾 = 10−3 , ] = 𝜂 = 𝛿 = 1, and 𝑁 = 4,
−1 ]] Table 1 lists the comparison of absolute errors of problem
× (2(𝛿 + 1)2 ) 𝑡) ]] ,
(35) subject to (36) and (37) using the J-GL-C method for
]] different choices of 𝛼 and 𝛽 with references [19], in the
interval [0, 1]. Moreover in Tables 2 and 3, the absolute errors
𝑢 (𝐵, 𝑡)
of this problem with 𝛼 = 𝛽 = 1/2 and various choices of
𝑥, 𝑡 for 𝛿 = 1 (3), in both intervals [0, 1] and [−1, 1], are
[𝛾 𝛾 given, respectively. In Table 4, maximum absolute errors with
=[ +
2 2 various choices of (𝛼, 𝛽) for both values of 𝛿 = 1, 3 are given
[ where ] = 𝛾 = 𝜂 = 0.001, in both intervals [0, 1] and [−1, 1].
Moreover, the absolute errors of problem (35) are shown in
[ −]𝛿 + 𝛿√] + 4𝜂 (1 + 𝛿)
2 Figures 1, 2, and 3 for 𝛿 = 1, 2, and 3 with values of parameters
× tanh [𝛾 listed in their captions, respectively, while in Figure 4, we
4 (𝛿 + 1)
[ plotted the approximate solution of this problem where 𝛼 = 0,
𝛽 = 1, ] = 𝜂 = 𝛾 = 10−3 , and 𝑁 = 12 for 𝛿 = 1. These
× (𝐵 − (]𝛾 (1+𝛿−𝛾) (√]2 +4𝜂 (1+𝛿)−])) figures demonstrate the good accuracy of this algorithm for
all choices of 𝛼, 𝛽, and 𝑁 and moreover in any interval.
1/𝛿
Example 2. Consider the nonlinear time-dependent one
2 −1 ]]
× (2(𝛿 + 1) ) 𝑡) ]] , dimensional Burgers-type equation:
]]
(36) 𝑢𝑡 + ]𝑢𝑢𝑥 − 𝜇𝑢𝑥𝑥 = 0; (𝑥, 𝑡) ∈ [𝐴, 𝐵] × [0, 𝑇] , (40)
6 Abstract and Applied Analysis
Table 1: Comparison of absolute errors of Example 1 with results from different articles, where 𝑁 = 4, 𝛾 = 10−3 , and ] = 𝜂 = 𝛿 = 1.
Table 2: Absolute errors with 𝛼 = 𝛽 = 1/2, 𝛿 = 1 and various choices of 𝑥, 𝑡 for Example 1.
𝑥 𝑡 𝐴 𝐵 ] 𝛾 𝜂 𝑁 𝐸 𝐴 𝐵 𝐸
0.0 0.1 0 1 0.001 0.001 0.001 12 7.04 × 10−12 −1 1 2.47 × 10−11
0.1 5.61 × 10−11 4.19 × 10−12
0.2 7.29 × 10−11 7.01 × 10−12
0.3 8.26 × 10−12 4.59 × 10−11
0.4 4.37 × 10−11 5.74 × 10−11
0.5 2.47 × 10−11 1.44 × 10−11
0.6 7.01 × 10−12 2.16 × 10−11
0.7 5.74 × 10−11 3.74 × 10−11
0.8 2.15 × 10−11 1.03 × 10−10
0.9 1.03 × 10−10 1.25 × 10−10
1 4.64 × 10−12 4.64 × 10−12
0.0 0.2 0 1 0.001 0.001 0.001 12 1.33 × 10−11 −1 1 4.92 × 10−11
0.1 1.11 × 10−10 8.33 × 10−12
0.2 1.46 × 10−10 1.42 × 10−11
0.3 1.65 × 10−11 9.20 × 10−11
0.4 8.74 × 10−11 1.15 × 10−10
0.5 4.92 × 10−11 2.87 × 10−11
0.6 1.42 × 10−11 4.30 × 10−11
0.7 1.14 × 10−10 7.49 × 10−11
0.8 4.30 × 10−11 2.05 × 10−10
0.9 2.05 × 10−10 2.51 × 10−10
1 1.09 × 10−11 1.09 × 10−11
subject to the boundary conditions If we apply the generalized tanh method [48], then we find
that the analytical solution of (40) is
𝑐 𝑐 𝑐 𝑐 𝑐 𝑐
𝑢 (𝐴, 𝑡) = − tanh [ (𝐴 − 𝑐𝑡)] , 𝑢 (𝑥, 𝑡) = − tanh [ (𝑥 − 𝑐𝑡)] . (43)
] ] 2𝜇 ] ] 2𝜇
(41)
𝑐 𝑐 𝑐 In Table 5, the maximum absolute errors of (40) subject to
𝑢 (𝐵, 𝑡) = − tanh [ (𝐵 − 𝑐𝑡)] ,
] ] 2𝜇 (41) and (42) are introduced using the J-GL-C method, with
various choices of (𝛼, 𝛽) in both intervals [0, 1] and [−1, 1].
and the initial condition Absolute errors between exact and numerical solutions of this
problem are introduced in Table 6 using the J-GL-C method
for 𝛼 = 𝛽 = 1/2 with 𝑁 = 20, and ] = 10, 𝜇 = 0.1 and
𝑐 𝑐 𝑐 𝑐 = 0.1 in both intervals [0, 1] and [−1, 1]. In Figures 5, 6,
𝑢 (𝑥, 0) = − tanh [ 𝑥] , 𝑥 ∈ [𝐴, 𝐵] . (42)
] ] 2𝜇 and 7, we displayed the absolute errors of problem (40) for
Abstract and Applied Analysis 7
Table 3: Absolute errors with 𝛼 = 𝛽 = 1/2, 𝛿 = 3 and various choices of 𝑥, 𝑡 for Example 1.
𝑥 𝑡 𝐴 𝐵 ] 𝛾 𝜂 𝑁 𝐸 𝐴 𝐵 𝐸
0.0 0.1 0 1 0.001 0.001 0.001 12 5.62 × 10−8 −1 1 8.32 × 10−9
0.1 6.50 × 10−6 2.99 × 10−6
0.2 4.67 × 10−6 1.65 × 10−6
0.3 3.08 × 10−6 2.44 × 10−6
0.4 1.65 × 10−6 3.09 × 10−6
0.5 8.33 × 10−9 1.97 × 10−6
0.6 1.65 × 10−6 4.67 × 10−6
0.7 3.09 × 10−6 3.79 × 10−6
0.8 4.67 × 10−6 6.53 × 10−6
0.9 6.53 × 10−6 35.32 × 10−6
1 5.66 × 10−8 5.66 × 10−8
0.0 0.2 0 1 0.001 0.001 0.001 12 2.27 × 10−7 −1 1 2.54 × 10−8
0.1 12.98 × 10−6 5.97 × 10−6
0.2 9.32 × 10−6 3.32 × 10−6
0.3 6.16 × 10−6 4.89 × 10−6
0.4 3.30 × 10−6 6.21 × 10−6
0.5 2.54 × 10−8 3.93 × 10−6
0.6 3.32 × 10−6 9.36 × 10−6
0.7 6.21 × 10−6 7.58 × 10−6
0.8 9.36 × 10−6 13.10 × 10−6
0.9 13.10 × 10−6 70.74 × 10−6
1 2.27 × 10−7 2.27 × 10−7
Table 4: Maximum absolute errors with various choices of (𝛼, 𝛽) for both values of 𝛿 = 1, 3 Example 3.
𝛼 𝛽 𝐴 𝐵 ] 𝛾 𝜂 𝛿 𝑁 𝑀𝐸 𝐴 𝐵 𝑀𝐸
different numbers of collation points and different choices Example 3. Consider the nonlinear time-dependent one-
of 𝛼 and 𝛽 in interval [0, 1] with values of parameters being dimensional generalized Burger-Fisher-type equation:
listed in their captions. Moreover, in Figure 8, we see that,
the approximate solution and the exact solution are almost
coincided for different values of 𝑡 (0, 0.5 and 0.9) of problem
(40) where ] = 10, 𝜇 = 0.1, 𝑐 = 0.1, 𝛼 = 𝛽 = −0.5, and 𝑁 = 20 𝑢𝑡 = 𝑢𝑥𝑥 − ]𝑢𝛿 𝑢𝑥 + 𝛾𝑢 (1 − 𝑢𝛿 ) ; (𝑥, 𝑡) ∈ [𝐴, 𝐵] × [0, 𝑇] ,
in interval [−1, 1]. (44)
8 Abstract and Applied Analysis
×10−7 0.00050005
8
10
̃ (x, t)
6 1.0 0.00050000
E 4
u
0.00049995
2
0 −10 5 t
0.0 0.5 t −5
0
x
0.5 5
x 10
0
0.0
1.0 Figure 4: The approximate solution of problem (35) where 𝛼 = 0,
𝛽 = 1, ] = 𝜂 = 𝛾 = 10−3 , and 𝑁 = 12 for 𝛿 = 1.
Figure 1: The absolute error of problem (35) where 𝛼 = 𝛽 = 0,
] = 𝜂 = 𝛾 = 10−3 , and 𝑁 = 4 for 𝛿 = 1.
×10−9
−7 1.5
×10
4 1.0
1
3 1.0 E
5
E 2
1 0
0.0 0.5
0 t
0.5 t
0.2
0.5
0.4 x
x 0.6
0.0
0.8 1.0
0.0
Figure 5: The absolute error of problem (40) where ] = 10, 𝜇 = 0.1,
Figure 2: The absolute error of problem (35) where 𝛼 = 𝛽 = 1/2,
𝑐 = 0.1, −𝛼 = 𝛽 = 1/2, and 𝑁 = 12.
] = 𝜂 = 𝛾 = 10−3 , and 𝑁 = 4 for 𝛿 = 2.
×10−10
×10−6
4
2 3 1.0
1.0
E 1 E 2
1
0 0
0.5 0.0 0.5 t
0.2 t
0.4
0.5
x 0.6 x
0.8
0.0 0.0
1.0 1.0
Figure 3: The absolute error of problem (35) where −𝛼 = 𝛽 = 1/2, Figure 6: The absolute error of problem (40) where ] = 10, 𝜇 = 0.1,
] = 𝜂 = 𝛾 = 10−3 , and 𝑁 = 4 for 𝛿 = 3. 𝑐 = 0.1, 𝛼 = 𝛽 = 1/2, and 𝑁 = 20.
Abstract and Applied Analysis 9
Table 5: Maximum absolute errors with various choices of (𝛼, 𝛽) for Table 6: Absolute errors with 𝛼 = 𝛽 = 1/2 and various choices of
Example 2. 𝑥, 𝑡 for Example 2.
𝑥 𝑡 𝐴 𝐵 𝜇 ] 𝑁 𝐸 𝐴 𝐵 𝐸
𝛼 𝛽 𝐴 𝐵 𝜇 ] 𝑁 𝑀𝐸 𝐴 𝐵 𝑀𝐸
0.0 0.1 0 1 0.1 10 20 1.34 × 10−12 −1 1 9.21 × 10−11
0 0 0 1 0.1 10 4 1.49 × 10−6 −1 1 5.62 × 10−7 0.1 8.71 × 10−11 8.70 × 10−11
−10
0.2 1.07 × 10 8.16 × 10−11
1/2 1/2 2.45 × 10−6 8.25 × 10−7 0.3 1.07 × 10−10
7.61 × 10−11
−1/2 −1/2 6.51 × 10−7 6.51 × 10−7 0.4 1.01 × 10−10 7.01 × 10−11
−11
0.5 9.21 × 10 6.38 × 10−11
−1/2 1/2 3.84 × 10−6 4.67 × 10−7 0.6 8.16 × 10−11
5.70 × 10−11
−11
1/2 −1/2 1.20 × 10−6 1.20 × 10−6 0.7 7.01 × 10 4.88 × 10−11
−11
0.8 5.70 × 10 3.85 × 10−11
0 0 0 1 0.1 10 16 1.43 × 10−9 −1 1 1.06 × 10−9 0.9 3.85 × 10−11
2.30 × 10−11
−12
1 1.34 × 10 1.34 × 10−12
1/2 1/2 1.62 × 10−9 1.18 × 10−9 −12
0.0 0.2 0 1 0.1 10 20 1.77 × 10 −1 1 3.35 × 10−10
−1/2 −1/2 6.70 × 10−10 4.45 × 10−10 0.1 2.71 × 10−10 3.17 × 10−10
−10
−1/2 1/2 6.68 × 10−10 4.49 × 10−10 0.2 3.61 × 10 2.98 × 10−10
−10
0.3 3.78 × 10 2.77 × 10−10
1/2 −1/2 6.67 × 10−10 4.23 × 10−10 0.4 3.64 × 10−10
2.55 × 10−10
−10
0.5 3.34 × 10 2.31 × 10−10
−10
0.6 2.98 × 10 2.03 × 10−10
−10
0.7 2.55 × 10 1.71 × 10−10
−10
0.8 2.02 × 10 1.30 × 10−10
0.9 1.30 × 10−10 7.70 × 10−11
−12
×10−9 1 1.77 × 10 1.77 × 10−12
1.5 1.0
1
E 0.5 and the initial condition
0 1/𝛿
0.5 1 1 ]𝛿
0.0 t 𝑢 (𝑥, 0) = [ − tanh [ (𝑥)]] , 𝑥 ∈ [𝐴, 𝐵] .
2 2 2 (𝛿 + 1)
0.5 (46)
x
0.0 The exact solution of (44) is
1.0
Figure 7: The absolute error of problem (40) where ] = 10, 𝜇 = 0.1, 𝑢 (𝑥, 𝑡)
𝑐 = 0.1, 𝛼 = 𝛽 = −1/2, and 𝑁 = 16. 1/𝛿
1 1 ]𝛿 ] 𝛾 (𝛿+1)
= [ − tanh [ (𝑥−( + ) 𝑡)]] .
2 2 2 (𝛿+1) 𝛿+1 ]
subject to the boundary conditions (47)
Table 7: Comparison of absolute errors of Example 3 with results from [19], where 𝑁 = 4, 𝛼 = 𝛽 = 0, and various choices of 𝑥, 𝑡.
𝑥 𝑡 𝛾 ] 𝛿 [19] 𝐸 𝑥 𝑡 𝛾 ] 𝛿 [19] 𝐸
0.1 0.005 0.001 0.001 1 9.69 × 10−6 1.85 × 10−6 0.1 0.0005 1 1 2 1.40 × 10−3 3.83 × 10−5
0.001 1.94 × 10−6 3.72 × 10−7 0.0001 2.80 × 10−4 3.88 × 10−5
0.01 1.94 × 10−5 3.72 × 10−6 0.001 2.80 × 10−3 3.76 × 10−5
0.5 0.005 9.69 × 10−6 7.04 × 10−6 0.5 0.0005 1.35 × 10−3 2.32 × 10−5
0.001 1.94 × 10−6 1.41 × 10−6 0.0001 2.69 × 10−4 2.38 × 10−5
0.01 1.94 × 10−5 1.41 × 10−5 0.001 2.69 × 10−3 2.25 × 10−5
0.9 0.005 9.69 × 10−6 3.21 × 10−6 0.9 0.0005 1.28 × 10−3 1.58 × 10−5
0.001 1.94 × 10−6 6.42 × 10−7 0.0001 2.55 × 10−4 1.55 × 10−5
0.01 1.94 × 10−5 6.42 × 10−6 0.001 2.55 × 10−3 1.61 × 10−5
0.1 0.1
0.2 4.28 × 10−9 0.2 1.96 × 10−10
0.0010
0.3 4.66 × 10−9 0.3 3.52 × 10−10
0.4 4.93 × 10−9 0.4 5.62 × 10−10
0.009
0.5 5.05 × 10−9 0.5 7.31 × 10−10
0.6 4.93 × 10−9 0.6 7.87 × 10−10
0.008
−0.4 −0.2 0.0 0.2 0.4 0.7 4.49 × 10−9 0.7 8.23 × 10−10
x 0.8 3.61 × 10−9 0.8 8.17 × 10−10
0.9 2.26 × 10−9 0.9 8.05 × 10−10
u(x, 0.0) ̃ (x, 0.5)
u
̃ (x, 0.0)
u u(x, 0.9) 1.0 1.13 × 10−11 1.0 1.09 × 10−11
u(x, 0.5) ̃ (x, 0.9)
u
×10−10
8
6 1.0
4
E
2
0
−1.0 0.5 t
−0.5
0.0
x
×10−10 0.5
0.0
6 1.0
1.0
4
Figure 10: The absolute error of problem (44) where 𝛼 = 𝛽 = 1/2
E 2
and ] = 𝛾 = 10−2 at 𝑁 = 20.
0
−1.0 0.5
t
−0.5
0.0
5. Conclusion
x
0.5 An efficient and accurate numerical scheme based on the J-
0.0 GL-C spectral method is proposed to solve nonlinear time-
1.0
dependent Burgers-type equations. The problem is reduced
Figure 9: The absolute error of problem (44) where 𝛼 = 𝛽 = 0 and to the solution of a SODEs in the expansion coefficient of
] = 𝛾 = 10−2 at 𝑁 = 20. the solution. Numerical examples were given to demonstrate
Abstract and Applied Analysis 11
[19] H. N. A. Ismail, K. Raslan, and A. A. A. Rabboh, “Ado- [36] A. Veksler and Y. Zarmi, “Wave interactions and the analysis of
mian decomposition method for Burger’s-Huxley and Burger’s- the perturbed Burgers equation,” Physica D, vol. 211, no. 1-2, pp.
Fisher equations,” Applied Mathematics and Computation, vol. 57–73, 2005.
159, no. 1, pp. 291–301, 2004. [37] A. Veksler and Y. Zarmi, “Freedom in the expansion and
[20] H. Bateman, “Some recent researchers on the motion of fluids,” obstacles to integrability in multiple-soliton solutions of the
Monthly Weather Review, vol. 43, pp. 163–170, 1915. perturbed KdV equation,” Physica D, vol. 217, no. 1, pp. 77–87,
[21] J. M. Burgers, “A mathematical model illustrating the theory 2006.
of turbulence,” in Advances in Applied Mechanics, pp. 171–199, [38] Z.-Y. Ma, X.-F. Wu, and J.-M. Zhu, “Multisoliton excitations for
Academic Press, New York, NY, USA, 1948. the Kadomtsev-Petviashvili equation and the coupled Burgers
[22] M. K. Kadalbajoo and A. Awasthi, “A numerical method based equation,” Chaos, Solitons and Fractals, vol. 31, no. 3, pp. 648–
on Crank-Nicolson scheme for Burgers’ equation,” Applied 657, 2007.
Mathematics and Computation, vol. 182, no. 2, pp. 1430–1442, [39] R. A. Fisher, “The wave of advance of advantageous genes,”
2006. Annals of Eugenics, vol. 7, pp. 335–369, 1937.
[23] M. Gülsu, “A finite difference approach for solution of Burgers’ [40] A. J. Khattak, “A computational meshless method for the
equation,” Applied Mathematics and Computation, vol. 175, no. generalized Burger’s-Huxley equation,” Applied Mathematical
2, pp. 1245–1255, 2006. Modelling, vol. 33, no. 9, pp. 3718–3729, 2009.
[24] P. Kim, “Invariantization of the Crank-Nicolson method for [41] N. F. Britton, Reaction-Diffusion Equations and their Applica-
Burgers’ equation,” Physica D, vol. 237, no. 2, pp. 243–254, 2008. tions to Biology, Academic Press, London, UK, 1986.
[25] H. Nguyen and J. Reynen, “A space-time least-square finite [42] D. A. Frank, Diffusion and Heat Exchange in Chemical Kinetics,
element scheme for advection-diffusion equations,” Computer Princeton University Press, Princeton, NJ, USA, 1955.
Methods in Applied Mechanics and Engineering, vol. 42, no. 3, [43] A.-M. Wazwaz, “The extended tanh method for abundant
pp. 331–342, 1984. solitary wave solutions of nonlinear wave equations,” Applied
[26] H. Nguyen and J. Reynen, “A space-time finite element approach Mathematics and Computation, vol. 187, no. 2, pp. 1131–1142,
to Burgers’ equation,” in Numerical Methods for Non-Linear 2007.
Problems, C. Taylor, E. Hinton, D. R. J. Owen, and E. Onate, Eds., [44] W. Malfliet, “Solitary wave solutions of nonlinear wave equa-
vol. 2, pp. 718–728, 1984. tions,” American Journal of Physics, vol. 60, no. 7, pp. 650–654,
[27] L. R. T. Gardner, G. A. Gardner, and A. Dogan, “A Petrov- 1992.
Galerkin finite element scheme for Burgers’ equation,” The [45] Y. Tan, H. Xu, and S.-J. Liao, “Explicit series solution of
Arabian Journal for Science and Engineering C, vol. 22, no. 2, pp. travelling waves with a front of Fisher equation,” Chaos, Solitons
99–109, 1997. and Fractals, vol. 31, no. 2, pp. 462–472, 2007.
[28] L. R. T. Gardner, G. A. Gardner, and A. Dogan, in A Least- [46] J. Canosa, “Diffusion in nonlinear multiplicate media,” Journal
Squares Finite Element Scheme For Burgers Equation, Mathe- of Mathematical Physics, vol. 31, pp. 1862–1869, 1969.
matics, University of Wales, Bangor, UK, 1996. [47] A.-M. Wazwaz, “Travelling wave solutions of generalized
[29] S. Kutluay, A. Esen, and I. Dag, “Numerical solutions of forms of Burgers, Burgers-KdV and Burgers-Huxley equations,”
the Burgers’ equation by the least-squares quadratic B-spline Applied Mathematics and Computation, vol. 169, no. 1, pp. 639–
finite element method,” Journal of Computational and Applied 656, 2005.
Mathematics, vol. 167, no. 1, pp. 21–33, 2004. [48] E. Fan and Y. C. Hon, “Generalized tanh method extended
[30] R. C. Mittal and R. K. Jain, “Numerical solutions of nonlinear to special types of nonlinear equations,” Zeitschrift fur Natur-
Burgers’ equation with modified cubic B-splines collocation forschung A, vol. 57, no. 8, pp. 692–700, 2002.
method,” Applied Mathematics and Computation, vol. 218, no.
15, pp. 7839–7855, 2012.
[31] E. H. Doha and A. H. Bhrawy, “Efficient spectral-Galerkin algo-
rithms for direct solution of fourth-order differential equations
using Jacobi polynomials,” Applied Numerical Mathematics, vol.
58, no. 8, pp. 1224–1244, 2008.
[32] E. H. Doha and A. H. Bhrawy, “A Jacobi spectral Galerkin
method for the integrated forms of fourth-order elliptic dif-
ferential equations,” Numerical Methods for Partial Differential
Equations, vol. 25, no. 3, pp. 712–739, 2009.
[33] S. Kazem, “An integral operational matrix based on Jacobi
polynomials for solving fractional-order differential equations,”
Applied Mathematical Modelling, vol. 37, no. 3, pp. 1126–1136,
2013.
[34] A. Ahmadian, M. Suleiman, S. Salahshour, and D. Baleanu, “A
Jacobi operational matrix for solving a fuzzy linear fractional
differential equation,” Advances in Difference Equations, vol.
2013, p. 104, 2013.
[35] E. H. Doha, A. H. Bhrawy, and R. M. Hafez, “A Jacobi-
Jacobi dual-Petrov-Galerkin method for third- and fifth-order
differential equations,” Mathematical and Computer Modelling,
vol. 53, no. 9-10, pp. 1820–1832, 2011.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 316978, 6 pages
http://dx.doi.org/10.1155/2013/316978
Research Article
Mappings for Special Functions on Cantor Sets and
Special Integral Transforms via Local Fractional Operators
Copyright © 2013 Yang Zhao et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The mappings for some special functions on Cantor sets are investigated. Meanwhile, we apply the local fractional Fourier
series, Fourier transforms, and Laplace transforms to solve three local fractional differential equations, and the corresponding
nondifferentiable solutions were presented.
1 4.5
0.9
4
0.8
0.7 3.5
0.6
3
0.5
0.4 2.5
0.3 2
0.2
1.5
0.1
0 1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.7
problems are presented. Finally, in Section 4, the conclusions
are presented. 0.6
0.5
2. Mappings for Special Functions on
Cantor Sets 0.4
∞
𝑓 (𝑥) = 𝑥2𝛼 , 𝑥2𝛼𝑘
(3) cos𝛼 𝑥𝛼 = ∑ (−1)𝑘 , (6)
𝑘=0
Γ (1 + 2𝛼𝑘)
and its graph is shown in Figure 1.
The Mittag-Leffler functions on Cantor sets are given by with graph in Figure 4.
[24, 25] Hyperbolic sine on Cantor sets is defined by [24, 25]
∞
𝑥𝛼𝑘 ∞
𝑥𝛼(2𝑘+1)
𝐸𝛼 (𝑥𝛼 ) = ∑ , (4) sinh𝛼 𝑥𝛼 = ∑ , (7)
𝑘=0
Γ (1 + 𝑘𝛼) 𝑘=0
Γ [1 + 𝛼 (2𝑘 + 1)]
and we draw the corresponding graph in Figure 2. and we draw its graphs as shown in Figure 5.
Abstract and Applied Analysis 3
1 2.5
0.9
0.8
2
0.7
0.6
1.5
0.5
0.4
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
∞
The local fractional Fourier coefficients read as
𝛼 𝑥2𝛼𝑘
cosh𝛼 𝑥 = ∑ , (8) 1 𝑇
𝑘=0
Γ (1 + 2𝛼𝑘) 𝑎0 = ∫ 𝑓 (𝑥) (𝑑𝑥)𝛼 ,
𝑇𝛼 0
and its graph is shown in Figure 6. 2 𝛼 𝑇
𝑎𝑘 = ( ) ∫ 𝑓 (𝑥) sin𝛼 (𝑘𝛼 𝜔0𝛼 𝑥𝛼 ) (𝑑𝑥)𝛼 , (13)
Following (4)–(8), we have 𝑇 0
where 𝑖𝛼 is a fractal unit of an imaginary number [24, 26–32]. We notice that the above results are obtained from
If for 𝜀, 𝛿 > 0 and 𝜀, 𝛿 ∈ 𝑅, 𝑓(𝑥) satisfies the condition Pythagorean theorem in the generalized Hilbert space
[24–26] [24, 26–28].
Let 𝑓(𝑥) ∈ 𝐶𝛼 (−∞, ∞). The local fractional Fourier
𝑓 (𝑥) − 𝑓 (𝑥0 ) < 𝜀 ;
𝛼
(10) transform of 𝑓(𝑥) is suggested by [24, 29–32]
𝐹𝛼 {𝑓 (𝑥)} = 𝑓𝜔𝐹,𝛼 (𝜔)
for 𝑥 ∈ [𝑎, 𝑏] we write it as follows:
1 ∞ (14)
= ∫ 𝐸𝛼 (−𝑖𝛼 𝜔𝛼 𝑥𝛼 ) 𝑓 (𝑥) (𝑑𝑥)𝛼 .
𝑓 (𝑥) ∈ 𝐶𝛼 (𝑎, 𝑏) . (11) Γ (1 + 𝛼) −∞
4 Abstract and Applied Analysis
The inverse formula is expressed as follows [24, 29–32]: Submitting (20)-(21) into (18), we obtain
𝑛=1
Example 2. We now consider the following differential equa-
Here, we give a particular solution in the following form:
tion on Cantor sets:
∞
𝑦𝑝 (𝑥) = 𝑎0 + ∑ 𝐴 𝑛 sin𝛼 (𝑛𝛼 𝑥𝛼 ) 𝑑2𝛼 𝑥
+ 𝑝𝑥 = 𝑓 (𝑡) , +∞ > 𝑡 > −∞, (26)
𝑛=1 𝑑2𝛼 𝑡
(20)
∞
𝛼 𝛼 subject to the initial value condition
+ ∑ 𝐵𝑛 cos𝛼 (𝑛 𝑥 ) .
𝑛=1
𝑑𝛼 𝑥
= 0, 𝑥 (0) = 0, (27)
Following (20), we have 𝑑𝛼 𝑡 𝑡=0
∞
𝑦𝑝(𝛼) (𝑥) = ∑ 𝐴 𝑛 𝑛𝛼 cos𝛼 (𝑛𝛼 𝑥𝛼 ) where 𝑝 is constant and 𝑓(𝑡) is the local fractional continuous
𝑛=1 function so that its local fractional Fourier transform exists.
(21) Application of local fractional Fourier transform gives
∞
+ ∑ 𝐵𝑛 𝑛𝛼 sin𝛼 (𝑛𝛼 𝑥𝛼 ) .
𝑛=1 −𝜔2𝛼 𝑥𝜔𝐹,𝛼 (𝜔) + 𝑝𝑥𝜔𝐹,𝛼 (𝜔) = 𝑓𝜔𝐹,𝛼 (𝜔) , (28)
Abstract and Applied Analysis 5
so that References
2𝛼
(−𝜔 + 𝑝) 𝑥𝜔𝐹,𝛼 (𝜔) = 𝑓𝜔𝐹,𝛼 (𝜔) . (29) [1] G. E. Andrews, R. Askey, and R. Roy, Special Functions,
Cambridge University Press, Cambridge, UK, 1999.
From (29), we have
[2] Á. Elbert and A. Laforgia, “On some properties of the gamma
𝑓𝜔𝐹,𝛼 (𝜔) function,” Proceedings of the American Mathematical Society,
𝑥𝜔𝐹,𝛼 (𝜔) = . (30) vol. 128, no. 9, pp. 2667–2673, 2000.
(−𝜔2𝛼 + 𝑝)
[3] N. Virchenko, S. L. Kalla, and A. Al-Zamel, “Some results on a
Therefore, taking the inverse formula of local fractional generalized hypergeometric function,” Integral Transforms and
Fourier transform, we have Special Functions, vol. 12, no. 1, pp. 89–100, 2001.
𝑝−(1/2) 𝑡 [4] J. Choi and P. Agarwal, “Certain unified integrals associated
𝑥 (𝑡) = − ∫ 𝑓 (𝑡 − 𝜏) sin𝛼 (𝑝1/2 𝜏𝛼 ) (𝑑𝜏)𝛼 . (31) with Bessel functions,” Boundary Value Problems, vol. 2013, no.
Γ (1 + 𝛼) −∞ 1, pp. 1–9, 2013.
Example 3. Let us find the solution to the differential equa- [5] P. J. McNamara, “Metaplectic Whittaker functions and crystal
tion on Cantor sets bases,” Duke Mathematical Journal, vol. 156, no. 1, pp. 1–31, 2011.
𝑑2𝛼 𝑥 𝑑𝛼 𝑥 [6] E. W. Barnes, “The theory of the G-function,” Quarterly Journal
+ − 2𝑥 = 𝑓 (𝑡) , 𝑡 > 0, (32) of Mathematics, vol. 31, pp. 264–314, 1899.
𝑑2𝛼 𝑡 𝑑𝛼 𝑡
[7] R. Floreanini and L. Vinet, “𝑈𝑞 (𝑠𝑙(2)) and 𝑞-special functions,”
subject to the initial value condition Contemporary Mathematics, vol. 160, p. 85, 1994.
𝑑𝛼 𝑥 [8] H. M. Srivastava, K. C. Gupta, and S. P. Goyal, The H-Functions
= 0, 𝑥 (0) = 0, (33)
𝑑𝛼 𝑡 𝑡=0 of One and Two Variables with Applications, South Asian
Publishers, New Delhi, India, 1982.
where 𝑓(𝑡) is the local fractional continuous function so that [9] R. Gorenflo, A. A. Kilbas, and S. V. Rogosin, “On the generalized
its local fractional Laplace transform exists. Mittag-Leffler type functions,” Integral Transforms and Special
Taking the local fractional Laplace transform, from (32), Functions, vol. 7, no. 3-4, pp. 215–224, 1998.
we have [10] R. K. Saxena and M. Saigo, “Certain properties of fractional
(𝑠2𝛼 𝑥𝑠𝐿,𝛼 (𝑠) − 𝑠𝛼 𝑥 (0) − 𝑥(𝛼) (0)) + (𝑠𝛼 𝑥𝑠𝐿,𝛼 (𝑠) − 𝑥 (0)) calculus operators associated with generalized Mittag-Leffler
(34) function,” Fractional Calculus & Applied Analysis, vol. 8, no. 2,
pp. 141–154, 2005.
+ 2𝑥𝑠𝐿,𝛼 (𝑠) = 𝑓𝑠𝐿,𝛼 (𝑠) ,
[11] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
so that Calculus Models and Numerical Methods, Series on Complexity,
Nonlinearity and Chaos, World Scientific, Singapore, 2012.
𝑓𝑠𝐿,𝛼 (𝑠)
𝑥𝑠𝐿,𝛼 (𝑠) = . (35) [12] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory
𝑠2𝛼 + 𝑠𝛼 − 2 and Applications of Fractional Differential Equations, vol. 204,
When the local fractional convolution of two functions is Elsevier, Amsterdam, The Netherlands, 2006.
given by [24] [13] J. Sabatier, O. P. Agrawal, and J. A. T. Machado, Advances in
𝑡 Fractional Calculus: Theoretical Developments and Applications
1
𝑓1 (𝑡) ∗ 𝑓2 (𝑡) = ∫ 𝑓1 (𝑡 − 𝜏) 𝑓2 (𝜏) (𝑑𝜏)𝛼 (36) in Physics and Engineering, Springer, New York, NY, USA, 2007.
Γ (1 + 𝛼) 0 [14] S. Hu, Y. Q. Chen, and T. S. Qiu, Fractional Processes and
and the local fractional Laplace transform of 𝑓1 (𝑡) ∗ 𝑓2 (𝑡) is Fractional-Order Signal Processing: Techniques and Applications,
[24] Springer, New York, NY, USA, 2012.
𝐿,𝛼 𝐿,𝛼 [15] F. Mainardi and R. Gorenflo, “On Mittag-Leffler-type functions
𝐿 𝛼 {𝑓1 (𝑡) ∗ 𝑓2 (𝑡)} = 𝑓𝑠,1 (𝑠) 𝑓𝑠,2 (𝑠) , (37) in fractional evolution processes,” Journal of Computational and
the inverse formula of the local fractional Laplace transform Applied Mathematics, vol. 118, no. 1-2, pp. 283–299, 2000.
together with the local fractional convolution theorem gives [16] R. K. Saxena, A. M. Mathai, and H. J. Haubold, “Fractional
the solution reaction-diffusion equations,” Astrophysics and Space Science,
𝑡 vol. 305, no. 3, pp. 289–296, 2006.
1
𝑥 (𝑡) = ∫ 𝑓 (𝑡 − 𝜏) (𝐸𝛼 (−2𝜏𝛼 ) + 𝐸𝛼 (𝜏𝛼 )) (𝑑𝜏)𝛼 . [17] Y. Li, Y. Chen, and I. Podlubny, “Mittag-Leffler stability of
Γ (1 + 𝛼) 0 fractional order nonlinear dynamic systems,” Automatica, vol.
(38) 45, no. 8, pp. 1965–1969, 2009.
[18] E. C. de Oliveira, F. Mainardi, and J. Vaz Jr., “Models
4. Conclusions based on Mittag-Leffler functions for anomalous relaxation in
dielectrics,” The European Physical Journal Special Topics, vol.
In this work, we investigated the mappings for special 193, no. 1, pp. 161–171, 2011.
functions on Cantor sets and special integral transforms [19] D. S. F. Crothers, D. Holland, Y. P. Kalmykov, and W. T. Coffey,
via local fractional calculus, namely, the local fractional “The role of Mittag-Leffler functions in anomalous relaxation,”
Fourier series, Fourier transforms, and Laplace transforms, Journal of Molecular Liquids, vol. 114, no. 1-3, pp. 27–34, 2004.
respectively. These transformations were applied successfully [20] E. Scalas, R. Gorenflo, and F. Mainardi, “Fractional calculus and
to solve three local fractional differential equations, and the continuous-time finance,” Physica A, vol. 284, no. 1–4, pp. 376–
nondifferentiable solutions were reported. 384, 2000.
6 Abstract and Applied Analysis
Research Article
New Wavelets Collocation Method for
Solving Second-Order Multipoint Boundary Value Problems
Using Chebyshev Polynomials of Third and Fourth Kinds
Copyright © 2013 W. M. Abd-Elhameed et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
This paper is concerned with introducing two wavelets collocation algorithms for solving linear and nonlinear multipoint boundary
value problems. The principal idea for obtaining spectral numerical solutions for such equations is employing third- and fourth-
kind Chebyshev wavelets along with the spectral collocation method to transform the differential equation with its boundary
conditions to a system of linear or nonlinear algebraic equations in the unknown expansion coefficients which can be efficiently
solved. Convergence analysis and some specific numerical examples are discussed to demonstrate the validity and applicability of
the proposed algorithms. The obtained numerical results are comparing favorably with the analytical known solutions.
of beams and plate deflection theory [23], obstacle problems in some form in any scheme of the numerical solution,
[24], Troesch’s problem relating to the confinement of a and it is well known that other numerical methods do not
plasma column by radiation pressure [25, 26], temperature perform well near singularities. Finally, due to their rapid
distribution of the radiation fin of trapezoidal profile [21, convergence, Chebyshev wavelets collocation method does
27], and a number of other engineering applications. Many not suffer from the common instability problems associated
authors have used numerical and approximate methods with other numerical methods.
to solve second-order BVPs. The details about the related The main aim of this paper is to develop two new spectral
numerical methods can be found in a large number of papers algorithms for solving second-order multipoint BVPs based
(see, for instance, [21, 23, 24, 28]). The Walsh wavelets and on shifted third- and fourth-kind Chebyshev wavelets. The
the semiorthogonal B-spline wavelets are used in [23, 29] method reduces the differential equation with its boundary
to construct some numerical algorithms for the solution of conditions to a system of algebraic equations in the unknown
second-order BVPs with Dirichlet and Neumann boundary expansion coefficients. Large systems of algebraic equations
conditions. Na [21] has found the numerical solution of may lead to greater computational complexity and large
second-, third-, and fourth-order BVPs by converting them storage requirements. However the third- and fourth-kind
into initial value problems and then applying a class of Chebyshev wavelets collocation method reduces drastically
methods like nonlinear shooting, method of reduced physical the computational complexity of solving the resulting alge-
parameters, method of invariant imbedding, and so forth. braic system.
The presented approach in this paper can be applied to both The structure of the paper is as follows. In Section 2,
BVPs and IVPs with a slight modification, but without the we give some relevant properties of Chebyshev polynomi-
transformation of BVPs into IVPs or vice versa. als of third and fourth kinds and their shifted ones. In
Wavelets theory is a relatively new and an emerging area Section 3, the third- and fourth-kind Chebyshev wavelets are
in mathematical research. It has been applied to a wide range constructed. Also, in this section, we ascertain the conver-
of engineering disciplines; particularly, wavelets are very gence of the Chebyshev wavelets series expansion. Two new
successfully used in signal analysis for wave form repre- shifted Chebyshev wavelets collocation methods for solving
sentation and segmentations, time frequency analysis, and second-order linear and nonlinear multipoint boundary
fast algorithms for easy implementation. Wavelets permit value problems are implemented and presented in Section 4.
the accurate representation of a variety of functions and In Section 5, some numerical examples are presented to
operators. Moreover, wavelets establish a connection with fast show the efficiency and the applicability of the presented
numerical algorithms, (see [30, 31]). algorithms. Some concluding remarks are given in Section 6.
The application of Legendre wavelets for solving differen-
tial and integral equations is thoroughly considered by many
2. Some Properties of 𝑉𝑘 (𝑥) and 𝑊𝑘 (𝑥)
authors (see, for instance, [32, 33]). Also, Chebyshev wavelets
are used for solving some fractional and integral equations The Chebyshev polynomials 𝑉𝑘 (𝑥) and 𝑊𝑘 (𝑥) of third and
(see, [34, 35]). fourth kinds are polynomials of degree 𝑘 in 𝑥 defined,
Chebyshev polynomials have become increasingly crucial respectively, by (see [38])
in numerical analysis, from both theoretical and practical
points of view. It is well known that there are four kinds cos (𝑘 + (1/2)) 𝜃 sin (𝑘 + (1/2)) 𝜃
of Chebyshev polynomials, and all of them are special cases 𝑉𝑘 (𝑥) = , 𝑊𝑘 (𝑥) = ,
cos (𝜃/2) sin (𝜃/2)
of the more widest class of Jacobi polynomials. The first (1)
and second kinds are special cases of the symmetric Jacobi
polynomials (i.e., ultraspherical polynomials), while the third where 𝑥 = cos 𝜃; also they can be obtained explicitly as
and fourth kinds are special cases of the nonsymmetric Jacobi (𝛼,𝛽)
two particular cases of Jacobi polynomials 𝑃𝑘 (𝑥) for the
polynomials. In the literature, there is a great concentration two nonsymmetric cases correspond to 𝛽 = −𝛼 = ±1/2.
on the first and second kinds of Chebyshev polynomials 𝑇𝑛 (𝑥) Explicitly, we have
and 𝑈𝑛 (𝑥) and their various uses in numerous applications,
(see, for instance, [36]). However, there are few articles that 2
(2𝑘 𝑘!)
concentrate on the other two types of Chebyshev polynomi- 𝑉𝑘 (𝑥) = 𝑃𝑘(−1/2,1/2) (𝑥) ,
als, namely, third and fourth kinds 𝑉𝑛 (𝑥) and 𝑊𝑛 (𝑥), either (2𝑘)!
from theoretical or practical point of view and their uses (2)
𝑘 2
in various applications (see, e.g., [37]). This motivates our (2 𝑘!)
interest in such polynomials. We therefore intend in this work 𝑊𝑘 (𝑥) = 𝑃𝑘(1/2,−1/2) (𝑥) .
(2𝑘)!
to use them in a marvelous application of multipoint BVPs
arising in physics. It is readily seen that
There are several advantages of using Chebyshev wavelets
approximations based on collocation spectral method. First, 𝑊𝑘 (𝑥) = (−1)𝑘 𝑉𝑘 (−𝑥) . (3)
unlike most numerical techniques, it is now well established
that they are characterized by exponentially decaying errors. Hence, it is sufficient to establish properties and relations for
Second, approximation by wavelets handles singularities in 𝑉𝑛 (𝑥) and then deduce their corresponding properties and
the problem. The effect of any such singularities will appear relations for 𝑊𝑛 (𝑥) (by replacing 𝑥 by −𝑥).
Abstract and Applied Analysis 3
The polynomials 𝑉𝑛 (𝑥) and 𝑊𝑛 (𝑥) are orthogonal on wavelet. When the dilation parameter 𝑎 and the translation
(−1, 1); that is, parameter 𝑏 vary continuously, then we have the following
1
family of continuous wavelets:
∫ 𝜔1 (𝑥) 𝑉𝑘 (𝑥) 𝑉𝑗 (𝑥) 𝑑𝑥
−1 𝑡−𝑏
𝜓𝑎,𝑏 (𝑡) = |𝑎|−1/2 𝜓 ( ), 𝑎, 𝑏 ∈ R, 𝑎 ≠
0. (13)
1 𝑎
= ∫ 𝜔2 (𝑥) 𝑊𝑘 (𝑥) 𝑊𝑗 (𝑥) 𝑑𝑥 (4)
−1 Each of the third- and fourth-kind Chebyshev wavelets
𝜓𝑛𝑚 (𝑡) = 𝜓(𝑘, 𝑛, 𝑚, 𝑡) has four arguments: 𝑘, 𝑛 ∈ N, 𝑚 is
𝜋, 𝑘 = 𝑗, the order of the polynomial 𝑉𝑚∗ (𝑡) or 𝑊𝑚∗ (𝑡), and 𝑡 is the
={
0, 𝑘 ≠
𝑗, normalized time. They are defined explicitly on the interval
[0, 1] as
where
1+𝑥 1−𝑥 𝜓𝑛𝑚 (𝑡)
𝜔1 (𝑥) = √ , 𝜔2 (𝑥) = √ , (5)
1−𝑥 1+𝑥 2 (𝑘+1)/2
{
{ 𝑉∗ (2𝑘 𝑡 − 𝑛) ,
{
{ √𝜋 𝑚
and they may be generated by using the two recurrence {
{
{
{
relations {
{
resp.,
{
{
{ 2(𝑘+1)/2 𝑛 𝑛+1
𝑉𝑘 (𝑥) = 2𝑥𝑉𝑘−1 (𝑥) − 𝑉𝑘−2 (𝑥) , 𝑘 = 2, 3, . . . , (6) ={ 𝑊𝑚∗ (2𝑘 𝑡 − 𝑛) , 𝑡 ∈ [ 𝑘 , 𝑘 ] , 0 ⩽ 𝑚 ⩽ 𝑀,
{
{ √𝜋 2 2
{
{
with the initial values {
{ 0 ⩽ 𝑛 ⩽ 2𝑘 − 1,
{
{
{
{
𝑉0 (𝑥) = 1, 𝑉1 (𝑥) = 2𝑥 − 1, (7) {
{0, otherwise.
{
𝑊𝑘 (𝑥) = 2𝑥 𝑊𝑘−1 (𝑥) − 𝑊𝑘−2 (𝑥) , 𝑘 = 2, 3, . . . , (8) (14)
with the initial values
3.1. Function Approximation. A function 𝑓(𝑡) defined over
𝑊0 (𝑥) = 1, 𝑊1 (𝑥) = 2𝑥 + 1. (9) [0, 1] may be expanded in terms of Chebyshev wavelets as
kinds are defined on [0, 1], respectively, as 𝑓 (𝑡) = ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝑡) , (15)
𝑛=0 𝑚=0
𝑉𝑛∗ (𝑥) = 𝑉𝑛 (2𝑥 − 1) , 𝑊𝑛∗ (𝑥) = 𝑊𝑛 (2𝑥 − 1) . (10)
where
All results of Chebyshev polynomials of third and fourth 1
kinds can be easily transformed to give the corresponding 𝑐𝑛𝑚 = (𝑓 (𝑡) , 𝜓𝑛𝑚 (𝑡))𝜔∗ = ∫ 𝜔𝑖∗ 𝑓 (𝑡) 𝜓𝑛𝑚 (𝑡) 𝑑𝑡, (16)
𝑖
results for their shifted ones. 0
The orthogonality relations of 𝑉𝑛∗ (𝑡) and 𝑊𝑛∗ (𝑡) on [0, 1]
are given by and the weights 𝑤𝑖∗ , 𝑖 = 1, 2, are given in (12).
Assume that 𝑓(𝑡) can be approximated in terms of
1 Chebyshev wavelets as
∫ 𝑤1∗ 𝑉𝑚∗ (𝑡) 𝑉𝑛∗ (𝑡) 𝑑𝑡
0
2𝑘 −1 𝑀
1 𝑓 (𝑡) ≃ ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝑡) . (17)
= ∫ 𝑤2∗ 𝑊𝑚∗ (𝑡) 𝑊𝑛∗ (𝑡) 𝑑𝑡 (11) 𝑛=0 𝑚=0
0
𝜋
{ , 𝑚 = 𝑛, 3.2. Convergence Analysis. In this section, we state and prove
= {2 a theorem to ascertain that the third- and fourth-kind Cheby-
0, 𝑚 ≠
𝑛, shev wavelets expansion of a function 𝑓(𝑡), with bounded
{
second derivative, converges uniformly to 𝑓(𝑡).
where
𝑡 1−𝑡 Theorem 1. Assume that a function 𝑓(𝑡) ∈ 𝐿2𝜔∗ [0, 1], 𝜔1∗ =
𝜔1∗ =√ 𝜔2∗ =√
1
, . (12)
√𝑡/(1 − 𝑡) with |𝑓 (𝑡)| ⩽ 𝐿, can be expanded as an infinite
1−𝑡 𝑡
series of third-kind Chebyshev wavelets; then this series con-
3. Shifted Third- and Fourth-Kind verges uniformly to 𝑓(𝑡). Explicitly, the expansion coefficients
in (16) satisfy the following inequality:
Chebyshev Wavelets
Wavelets constitute of a family of functions constructed from 2√2𝜋 𝐿𝑚2
𝑐𝑛𝑚 < , ∀𝑛 ⩾ 0, 𝑚 > 1. (18)
dilation and translation of single function called the mother (𝑛 + 1)5/2 (𝑚4 − 1)
4 Abstract and Applied Analysis
2(𝑘+1)/2 (𝑛+1)/2
𝑘
2√2𝜋 𝐿𝑚2
𝑐𝑛𝑚 = ∫ 𝑓 (𝑡) 𝑉𝑚∗ (2𝑘 𝑡 − 𝑛) 𝜔1∗ (2𝑘 𝑡 − 𝑛) 𝑑𝑡. 𝑐𝑛𝑚 < . (24)
√𝜋 𝑛/2𝑘 (𝑛 + 1)5/2 (𝑚4 − 1)
(19)
Remark 2. The estimation in (18) is also valid for the coef-
If we make use of the substitution 2𝑘 𝑡 − 𝑛 = cos 𝜃 in (19), then ficients of fourth-kind Chebyshev wavelets expansion. The
we get proof is similar to the proof of Theorem 1.
2(−𝑘+1)/2 𝜋 cos 𝜃 + 𝑛
𝑐𝑛𝑚 = ∫ 𝑓( ) 4. Solution of Multipoint BVPs
√𝜋 0 2𝑘
In this section, we present two Chebyshev wavelets col-
cos (𝑚 + (1/2)) 𝜃 √ 1 + cos 𝜃 location methods, namely, third-kind Chebyshev wavelets
× sin 𝜃𝑑𝜃
cos (𝜃/2) 1 − cos 𝜃 collocation method (3CWCM) and fourth-kind Chebyshev
wavelets collocation method (4CWCM), to numerically solve
2(−𝑘+3)/2 𝜋 cos 𝜃 + 𝑛 1 𝜃 the following multipoint boundary value problem (BVP):
= ∫ 𝑓( ) cos (𝑚 + ) 𝜃 cos ( ) 𝑑𝜃
√𝜋 0 2 𝑘 2 2
𝑎 (𝑥) 𝑦 (𝑥) + 𝑏 (𝑥) 𝑦 (𝑥) + 𝑐 (𝑥) 𝑦 (𝑥)
(−𝑘+1)/2 𝜋
2 cos 𝜃 + 𝑛 (25)
= ∫ 𝑓( )
√𝜋 0 2𝑘 + 𝑓 (𝑥, 𝑦 ) + 𝑔 (𝑥, 𝑦) = 0, 0 ⩽ 𝑥 ⩽ 1,
𝜋 2𝑘 −1 𝑀
𝐿
⩽ ∫ 𝛾𝑚 (𝜃) 𝑑𝜃
+ ∑ ∑ 𝑐𝑛𝑚 𝑏 (𝑥) 𝜓𝑛𝑚 (𝑥)
25𝑘/2 √2𝜋 0 𝑛=0 𝑚=1
𝐿√𝜋 1 1 1 2𝑘 −1 𝑀
⩽ [ ( + ) (23)
2(5𝑘+1)/2 𝑚+1 𝑚 𝑚+2 + ∑ ∑ 𝑐𝑛𝑚 𝑐 (𝑥) 𝜓𝑛𝑚 (𝑥) (28)
1 1 1 𝑛=0 𝑚=0
+ ( + )]
𝑚 𝑚−1 𝑚+1 2𝑘 −1 𝑀
+ 𝑓 (𝑥, ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝑥))
𝐿√2𝜋 1 1
= 5𝑘/2 [ 2 + ] 𝑛=0 𝑚=1
2 𝑚 + 2𝑚 𝑚2 − 1
2𝑘 −1 𝑀
2𝐿√2𝜋 𝑚2 + 𝑔 (𝑥, ∑ ∑ 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝑥)) .
< 5𝑘/2 ( 4 ).
2 𝑚 −1 𝑛=0 𝑚=0
Abstract and Applied Analysis 5
Now, the application of the typical collocation method (see, Example 2. Consider the second-order linear BVP (see, [39,
e.g., [5]) gives 40]):
𝑚1 2𝑘 −1 𝑀
= 6 cosh 𝑥 + (2 + 𝑥 − 𝑥2 + sinh 𝑥) sinh 𝑥,
= ∑ ∑ ∑ 𝜇𝑖 𝑐𝑛𝑚 𝜓𝑛𝑚 (𝜂𝑖 ) + 𝛿1 . 0 < 𝑥 < 1,
𝑖=1 𝑛=0 𝑚=0 (34)
2 2
𝑦 (0) + 𝑦 ( ) = sinh ( ) ,
Equations (29) and (30) generate 2𝑘 (𝑀 + 1) equations in 3 3
the unknown expansion coefficients, 𝑐𝑛𝑚 , which can be 1 4 1 4
solved with the aid of the well-known Newton’s iterative 𝑦 (1) + 𝑦 ( ) = sinh ( ) + sinh 1,
2 5 2 5
method. Consequently, we get the desired approximate solu-
tion 𝑦𝑘,𝑀(𝑥) given by (27). with the exact solution 𝑦(𝑥) = sinh 𝑥. In Table 4, the maxi-
mum absolute error 𝐸 is listed for 𝑘 = 0 and various values
of 𝑀, while in Table 5 we give a comparison between the
5. Numerical Examples best errors resulted from the application of various methods
for Example 3. This table shows that our two algorithms are
In this section, the presented algorithms in Section 4 are
more accurate if compared with the two methods developed
applied to solve both of linear and nonlinear multipoint
in [40, 41].
BVPs. Some examples are considered to illustrate the effi-
ciency and applicability of the two proposed algorithms. Example 4. Consider the second-order nonlinear BVP (see
[42]):
Example 1. Consider the second-order nonlinear BVP (see
[6, 28]): 𝑦 + (1 + 𝑥 + 𝑥3 ) 𝑦2 = 𝑓 (𝑥) , 0 < 𝑥 < 1,
3 2 2 1 2 1 7
𝑦 + 𝑦 + (𝑦 ) + 1 = 0, 0 < 𝑥 < 1, 𝑦 (0) = 𝑦 ( ) + 𝑦 ( ) − 0.0286634, (35)
8 1089 6 9 3 9
(31)
1 1 2 1 7
𝑦 (0) = 0, 𝑦 ( ) = 𝑦 (1) . 𝑦 (1) = 𝑦 ( ) + 𝑦 ( ) − 0.0401287,
3 5 9 2 9
where
The two proposed methods are applied to the problem for
the case corresponding to 𝑘 = 0 and 𝑀 = 8. The numerical 1
𝑓 (𝑥) = [ − 6 cos (𝑥 − 𝑥2 ) + sin (𝑥 − 𝑥2 )
solutions are shown in Table 1. Due to nonavailability of the 9
exact solution, we compare our results with Haar wavelets
method [6], ADM solution [28] and ODEs Solver from × (−3(1 − 2𝑥)2 + (1 + 𝑥 + 𝑥3 ) sin (𝑥 − 𝑥2 ))] .
Mathematica which is carried out by using Runge-Kutta
method. This comparison is also shown in Table 1. (36)
6 Abstract and Applied Analysis
0.070 0.5
0.050
0.4
0.030
0.3
ln|y(x)|
0.020
y(x)
0.015
0.2
0.010
0.1
0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
x x
Exact k = 0, M = 3 Exact k = 1, M = 2
k = 0, M = 4 k = 0, M = 2 k = 1, M = 3 k = 1, M = 1
The exact solution of (35) is given by 𝑦(𝑥) = (1/3) sin(𝑥 − to obtain an approximate solution of 𝑦(𝑥). If we make use of
𝑥2 ). In Table 6, the maximum absolute error 𝐸 is listed for (27), then the approximate solution 𝑦0,3 (𝑥) can be expanded
𝑘 = 2 and various values of 𝑀, and in Table 7 we give a in terms of third-kind Chebyshev wavelets as
comparison between best errors resulted from the application
of various methods for Example 4. This table shows that 2 2
𝑦0,3 (𝑥) = 𝑐0,0 √ + 𝑐 √ (4𝑥 − 3)
our two algorithms are more accurate if compared with the 𝜋 0,1 𝜋
method developed in [42]. (42)
2
+ 𝑐0,2 √ (16𝑥2 − 20𝑥 + 5) .
Example 5. Consider the second-order singular linear BVP: 𝜋
𝑦 + 𝑓 (𝑥) 𝑦 = 𝑔 (𝑥) , 0 < 𝑥 < 1, If we set
1 2
𝑦 (0) + 16𝑦 ( ) = 3√4 𝑒, V𝑖 = √ 𝑐 , 𝑖 = 0, 1, 2, (43)
4 (37) 𝜋 0,𝑖
3 then (42) reduces to the form
𝑦 (1) + 16𝑦 ( ) = 3√𝑒3 ,
4
4
𝑦0,3 (𝑥) = V0 + V1 (4𝑥 − 3) + V2 (16𝑥2 − 20𝑥 + 5) . (44)
where
1 If we substitute (44) into (39), then the residual of (39) is
{
{3𝑥, 0 ⩽ 𝑥 ⩽ ,
𝑓 (𝑥) = { 2 (38) given by
{2𝑥, 1 < 𝑥 ⩽ 1,
2
{ 2 𝑅 (𝑥) = 2V2 + [V1 + V2 (8𝑥 − 5)]
and 𝑔(𝑥) is chosen such that the exact solution of (37) is
𝑦(𝑥) = 𝑥(1 − 𝑥)𝑒𝑥 . In Table 8, the maximum absolute error 𝐸 − 4 [V0 + V1 (4𝑥 − 3) (45)
is listed for 𝑘 = 1 and various values of 𝑀, while in Figure 2,
we give a comparison between the exact solution of (37) with + V2 (16𝑥2 − 20𝑥 + 5)] − 2.
three approximate solutions.
We enforce the residual to vanish at the first root of 𝑉3∗ (𝑥) =
Example 6. Consider the following nonlinear second-order 64𝑥3 −112𝑥2 +56𝑥−7, namely, at 𝑥1 = 0.18825509907063323,
BVP: to get
2
𝑦 + (𝑦 ) − 64𝑦 = 32, 0 < 𝑥 < 1, (39) 6.10388V22 + 0.5V12 − 3.49396V2 V1 − 2.60388V2
(46)
1 + 4.49396V1 − 2V0 = 1.
𝑦 (0) + 𝑦 ( ) = 1, (40)
4
Furthermore, the use of the boundary conditions (40) and
1 (41) yields
4𝑦 ( ) − 𝑦 (1) = 0, (41)
2
2V0 − 5V1 + 6V2 = 1,
with the exact solution 𝑦(𝑥) = 16𝑥2 . We solve (39) using (47)
3CWCM for the case corresponding to 𝑘 = 0 and 𝑀 = 3, 3V0 − 5V1 − 5V2 = 0.
Abstract and Applied Analysis 7
𝑥 3CWCM 4CWCM Haar method [6] ADM method [28] ODEs solver from Mathematica
0.1 0.06560 0.06560 0.06561 0.0656 0.06560
0.3 0.16587 0.16587 0.16588 0.1658 0.16587
0.5 0.22369 0.22369 0.22369 0.2236 0.22369
0.7 0.23820 0.23820 0.23821 0.2382 0.23820
0.9 0.20920 0.20920 0.20910 0.2092 0.20920
𝑘 𝑀 10 11 12 13 14 15
3CWCM 5.173 ⋅ 10−9 7.184 ⋅ 10−11 4.418 ⋅ 10−12 5.018 ⋅ 10−14 2.442 ⋅ 10−15 2.220 ⋅ 10−16
0
4CWCM 3.324 ⋅ 10−10 2.533 ⋅ 10−11 2.811 ⋅ 10−12 1.665 ⋅ 10−14 1.110 ⋅ 10−15 2.220 ⋅ 10−16
𝑀 4 5 6 7 8 9
3CWCM 4.644 ⋅ 10−6 1.001 ⋅ 10−8 1.840 ⋅ 10−9 2.547 ⋅ 10−10 6.247 ⋅ 10−11 5.681 ⋅ 10−12
1
4CWCM 2.15 ⋅ 10−6 5.247 ⋅ 10−9 7.548 ⋅ 10−10 1.004 ⋅ 10−10 2.154 ⋅ 10−11 4.257 ⋅ 10−12
𝑀 8 9 10 11 12 13
3CWCM 4.444 ⋅ 10−10 2.759 ⋅ 10−11 4.389 ⋅ 10−13 1.643 ⋅ 10−14 4.441 ⋅ 10−16 2.220 ⋅ 10−16
4CWCM 4.478 ⋅ 10−9 2.179 ⋅ 10−10 1.527 ⋅ 10−12 2.975 ⋅ 10−14 4.441 ⋅ 10−16 2.220 ⋅ 10−16
𝑀 4 5 6 7 8 9
3CWCM 2.881 ⋅ 10−3 3.441 ⋅ 10−3 1.212 ⋅ 10−5 1.933 ⋅ 10−5 1.990 ⋅ 10−6 3.010 ⋅ 10−8
4CWCM 1.241 ⋅ 10−3 8.542 ⋅ 10−4 7.526 ⋅ 10−6 1.002 ⋅ 10−6 6.321 ⋅ 10−7 2.354 ⋅ 10−9
𝑀 7 8 9 10 11 12 13 14
3CWCM 7.1 ⋅ 10−6 3.2 ⋅ 10−7 1.3 ⋅ 10−8 7.4 ⋅ 10−10 1.3 ⋅ 10−11 3.5 ⋅ 10−13 9.3 ⋅ 10−15 8.1 ⋅ 10−16
4CWCM 2.8 ⋅ 10−5 1.5 ⋅ 10−6 6.3 ⋅ 10−8 2.4 ⋅ 10−9 7.8 ⋅ 10−11 2.3 ⋅ 10−12 6.3 ⋅ 10−14 1.7 ⋅ 10−15
8 Abstract and Applied Analysis
The solution of the nonlinear system of (46) and (47) gives [6] Siraj-ul-Islam, I. Aziz, and B. Šarler, “The numerical solution of
second-order boundary-value problems by collocation method
V0 = 10, V1 = 5, V2 = 1, (48) with the Haar wavelets,” Mathematical and Computer Modelling,
vol. 52, no. 9-10, pp. 1577–1590, 2010.
and consequently [7] A. H. Bhrawy, A. S. Alofi, and S. I. El-Soubhy, “An exten-
sion of the Legendre-Galerkin method for solving sixth-order
𝑦0,3 (𝑥) = 10 + 5 (4𝑥 − 3) + (16𝑥2 − 20𝑥 + 5) = 16𝑥2 , (49) differential equations with variable polynomial coefficients,”
Mathematical Problems in Engineering, vol. 2012, Article ID
896575, 13 pages, 2012.
which is the exact solution. [8] E. H. Doha and W. M. Abd-Elhameed, “Efficient solutions of
multidimensional sixth-order boundary value problems using
Remark 3. It is worth noting here that the obtained numerical symmetric generalized Jacobi-Galerkin method,” Abstract and
results in the previous solved six examples are very accurate, Applied Analysis, vol. 2012, Article ID 749370, 19 pages, 2012.
although the number of retained modes in the spectral [9] W. M. Abd-Elhameed, “Efficient spectral Legendre dual-Petrov-
expansion is very few, and again the numerical results are Galerkin algorithms for the direct solution of (2𝑛 + 1)th-
comparing favorably with the known analytical solutions. order linear differential equations,” Journal of the Egyptian
Mathematical Society, vol. 17, no. 2, pp. 189–211, 2009.
6. Concluding Remarks [10] A. H. Bhrawy and W. M. Abd-Elhameed, “New algorithm for
the numerical solutions of nonlinear third-order differential
In this paper, two algorithms for obtaining numerical spectral equations using Jacobi-Gauss collocation method,” Mathemat-
wavelets solutions for second-order multipoint linear and ical Problems in Engineering, vol. 2011, Article ID 837218, 14
nonlinear boundary value problems are analyzed and dis- pages, 2011.
cussed. Chebyshev polynomials of third and fourth kinds [11] A. H. Bhrawy, A. S. Alofi, and S. I. El-Soubhy, “Spectral shifted
are used. One of the advantages of the developed algorithms Jacobi tau and collocation methods for solving fifth-order
is their availability for application on singular boundary boundary value problems,” Abstract and Applied Analysis, vol.
2011, Article ID 823273, 14 pages, 2011.
value problems. Another advantage is that high accurate
approximate solutions are achieved using a few number of [12] M. Moshiinsky, “Sobre los problemas de condiciones a la
terms of the approximate expansion. The obtained numerical frontiera en una dimension de caracteristicas discontinuas,”
Boletin de La Sociedad Matematica Mexicana, vol. 7, article 125,
results are comparing favorably with the analytical ones.
1950.
[13] S. P. Timoshenko, Theory of Elastic Stability, McGraw-Hill Book,
Acknowledgment New York, NY, USA, 2nd edition, 1961.
[14] R. P. Agarwal and I. Kiguradze, “On multi-point boundary value
The authors would like to thank the referee for his valuable problems for linear ordinary differential equations with singu-
comments and suggestions which improved the paper in its larities,” Journal of Mathematical Analysis and Applications, vol.
present form. 297, no. 1, pp. 131–151, 2004.
[15] Z. Du, “Solvability of functional differential equations with
References multi-point boundary value problems at resonance,” Computers
& Mathematics with Applications, vol. 55, no. 11, pp. 2653–2661,
[1] E. H. Doha, W. M. Abd-Elhameed, and Y. H. Youssri, “Efficient 2008.
spectral-Petrov-Galerkin methods for the integrated forms [16] W. Feng and J. R. L. Webb, “Solvability of 𝑚-point boundary
of third- and fifth-order elliptic differential equations using value problems with nonlinear growth,” Journal of Mathematical
general parameters generalized Jacobi polynomials,” Applied Analysis and Applications, vol. 212, no. 2, pp. 467–480, 1997.
Mathematics and Computation, vol. 218, no. 15, pp. 7727–7740, [17] H. B. Thompson and C. Tisdell, “Three-point boundary value
2012. problems for second-order, ordinary, differential equations,”
[2] W. M. Abd-Elhameed, E. H. Doha, and Y. H. Youssri, “Efficient Mathematical and Computer Modelling, vol. 34, no. 3-4, pp. 311–
spectral-Petrov-Galerkin methods for third- and fifth-order 318, 2001.
differential equations using general parameters generalized [18] M. R. Scott and H. A. Watts, “SUPORT—a computer code for
Jacobi polynomials,” Quaestiones Mathematicae, vol. 36, no. 1, two-point boundary-value problems via orthonormalization,”
pp. 15–38, 2013. Sandia Labs Report 75-0198, Sandia Laboratories, Albuquerque,
[3] E. H. Doha, W. M. Abd-Elhameed, and A. H. Bhrawy, “New NM, USA, 1975.
spectral-Galerkin algorithms for direct solution of high even- [19] M. R. Scott and H. A. Watts, “Computational solution of linear
order differential equations using symmetric generalized Jacobi two-point boundary value problems via orthonormalization,”
polynomials,” Collectanea Mathematica, vol. 64, no. 3, pp. 373– SIAM Journal on Numerical Analysis, vol. 14, no. 1, pp. 40–70,
394, 2013. 1977.
[4] E. H. Doha, A. H. Bhrawy, and R. M. Hafez, “On shifted Jacobi [20] M. R. Scott and W. H. Vandevender, “A comparison of several
spectral method for high-order multi-point boundary value invariant imbedding algorithms for the solution of two-point
problems,” Communications in Nonlinear Science and Numerical boundary-value problems,” Applied Mathematics and Compu-
Simulation, vol. 17, no. 10, pp. 3802–3810, 2012. tation, vol. 1, no. 3, pp. 187–218, 1975.
[5] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, [21] T. Y. Na, Computational Methods in Engineering Boundary Value
Spectral Methods, Springer, Berlin, Germany, 2006. Problems, vol. 145, Academic Press, New York, NY, USA, 1979.
Abstract and Applied Analysis 9
[22] K. E. Bisshopp and D. C. Drucker, “Large deflection of cantilever [40] F. Geng, “Solving singular second order three-point bound-
beams,” Quarterly of Applied Mathematics, vol. 3, pp. 272–275, ary value problems using reproducing kernel Hilbert space
1945. method,” Applied Mathematics and Computation, vol. 215, no.
[23] W. Glabisz, “The use of Walsh-wavelet packets in linear bound- 6, pp. 2095–2102, 2009.
ary value problems,” Computers & Structures, vol. 82, no. 2-3, [41] F. Z. Geng, “A numerical algorithm for nonlinear multi-
pp. 131–141, 2004. point boundary value problems,” Journal of Computational and
[24] Siraj-ul-Islam, M. A. Noor, I. A. Tirmizi, and M. A. Khan, Applied Mathematics, vol. 236, no. 7, pp. 1789–1794, 2012.
“Quadratic non-polynomial spline approach to the solution of [42] A. Saadatmandi and M. Dehghan, “The use of sinc-collocation
a system of second-order boundary-value problems,” Applied method for solving multi-point boundary value problems,”
Mathematics and Computation, vol. 179, no. 1, pp. 153–160, 2006. Communications in Nonlinear Science and Numerical Simula-
[25] S. Robert and J. Shipman, “Solution of Troesch’s two-point tion, vol. 17, no. 2, pp. 593–601, 2012.
boundary value problems by shooting techniques,” Journal of
Computational Physics, vol. 10, pp. 232–241, 1972.
[26] E. Wiebel, Confinement of a Plasma Column by Radiation
Pressure in the Plasma in a Magnetic Field, Stanford University
Press, Stanfor, Calif, USA, 1958.
[27] H. H. Keller and E. S. Holdrege, “Radiation heat transfer for
annular fins of trapezoidal profile,” International Journal of High
Performance Computing Applications, vol. 92, pp. 113–116, 1970.
[28] M. Tatari and M. Dehgan, “The use of the Adomian decomposi-
tion method for solving multipoint boundary value problems,”
Physica Scripta, vol. 73, pp. 672–676, 2006.
[29] M. Lakestani and M. Dehghan, “The solution of a second-order
nonlinear differential equation with Neumann boundary con-
ditions using semi-orthogonal B-spline wavelets,” International
Journal of Computer Mathematics, vol. 83, no. 8-9, pp. 685–694,
2006.
[30] A. Constantmldes, Applied Numerical Methods with Personal
Computers, McGraw-Hill, New York, NY, USA , 1987.
[31] D. E. Newland, An Introduction to Random Vibrations, Spectral
and Wavelet Analysis, Longman Scientific and Technical, New
York, NY, USA, 1993.
[32] M. Razzaghi and S. Yousefi, “Legendre wavelets method for the
solution of nonlinear problems in the calculus of variations,”
Mathematical and Computer Modelling, vol. 34, no. 1-2, pp. 45–
54, 2001.
[33] M. Razzaghi and S. Yousefi, “Legendre wavelets method for
constrained optimal control problems,” Mathematical Methods
in the Applied Sciences, vol. 25, no. 7, pp. 529–539, 2002.
[34] E. Babolian and F. Fattahzadeh, “Numerical solution of differ-
ential equations by using Chebyshev wavelet operational matrix
of integration,” Applied Mathematics and Computation, vol. 188,
no. 1, pp. 417–426, 2007.
[35] L. Zhu and Q. Fan, “Solving fractional nonlinear Fredholm
integro-differential equations by the second kind Chebyshev
wavelet,” Communications in Nonlinear Science and Numerical
Simulation, vol. 17, no. 6, pp. 2333–2341, 2012.
[36] E. H. Doha, W. M. Abd- Elhameed, and Y. H. Youssri, “Second
kind Chebyshev operational matrix algorithm for solving dif-
ferential equations of Lane-Emden type,” New Astronomy, vol.
23-24, pp. 113–117, 2013.
[37] E. H. Doha, W. M. Abd-Elhameed, and M. A. Bassuony, “New
algorithms for solving high even-order differential equations
using third and fourth Chebyshev-Galerkin methods,” Journal
of Computational Physics, vol. 236, pp. 563–579, 2013.
[38] J. C. Mason and D. C. Handscomb, Chebyshev Polynomials,
Chapman & Hall, New York, NY, USA, 2003.
[39] Y. Lin, J. Niu, and M. Cui, “A numerical solution to nonlinear
second order three-point boundary value problems in the
reproducing kernel space,” Applied Mathematics and Computa-
tion, vol. 218, no. 14, pp. 7362–7368, 2012.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 513808, 9 pages
http://dx.doi.org/10.1155/2013/513808
Research Article
Numerical Solution of a Class of Functional-Differential
Equations Using Jacobi Pseudospectral Method
Copyright © 2013 A. H. Bhrawy et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The shifted Jacobi-Gauss-Lobatto pseudospectral (SJGLP) method is applied to neutral functional-differential equations (NFDEs)
with proportional delays. The proposed approximation is based on shifted Jacobi collocation approximation with the nodes of
Gauss-Lobatto quadrature. The shifted Legendre-Gauss-Lobatto Pseudo-spectral and Chebyshev-Gauss-Lobatto Pseudo-spectral
methods can be obtained as special cases of the underlying method. Moreover, the SJGLP method is extended to numerically
approximate the nonlinear high-order NFDE with proportional delay. Some examples are displayed for implicit and explicit forms
of NFDEs to demonstrate the computation accuracy of the proposed method. We also compare the performance of the method with
variational iteration method, one-leg 𝜃-method, continuous Runge-Kutta method, and reproducing kernel Hilbert space method.
proportional delays using the shifted Jacobi polynomials on 𝐿2𝑤(𝜃,𝜗) with respect to the weight function are defined as
the interval [0, 𝐿]; we dedicate the shifted Jacobi-Gauss- follows:
Lobatto pseudospectral (SJGLP) method to find the approx-
1
imate solution 𝑢𝑁(𝑥). Approximate semianalytical solution (𝑢, V)𝑤(𝜃,𝜗) = ∫ 𝑢 (𝑥) V (𝑥) 𝑤(𝜃,𝜗) (𝑥) 𝑑𝑥,
with high accuracy can be obtained by selecting a limited −1 (4)
number of Gauss-Lobatto collocation points for the linear
and nonlinear high-order NFDEs with proportional delays. ‖𝑢‖𝑤(𝜃,𝜗) = (𝑢, 𝑢)1/2
𝑤(𝜃,𝜗)
.
For suitable collocation points we use the (𝑁 − 𝑚 + 1) nodes
of the shifted Jacobi-Gauss-Lobatto interpolation on (0, 𝐿) in The set of Jacobi polynomials forms a complete 𝐿2𝑤(𝜃,𝜗) -
which the nodes are distinct and lie between 0 and 𝐿, and orthogonal system, and
the two endpoints of the domain (0 and 𝐿) are used as the
(𝜃,𝜗)
first and last collocation points, respectively. The algorithms 𝐽𝑘 (𝜃,𝜗) := ℎ𝑘
given in [16] can be obtained as special cases from the 𝑤
proposed algorithms. Finally, the accuracy of the proposed 2𝜃+𝜗+1 Γ (𝑘 + 𝜃 + 1) Γ (𝑘 + 𝜗 + 1)
method is showed by test problems. From the results, these = .
(2𝑘 + 𝜃 + 𝜗 + 1) Γ (𝑘 + 1) Γ (𝑘 + 𝜃 + 𝜗 + 1)
algorithms are extremely efficient and accurate for solving (5)
NFDEs.
The paper is arranged in the following way. In the (𝛼,𝛽) (𝛼,𝛽)
next section, some basic properties of Jacobi polynomials Let us denote 𝑃𝐿,𝑘 (𝑥) = 𝐽𝑘 ((2𝑥/𝐿) − 1), 𝐿 > 0. By the
which are required in the present paper are given, and shifted Jacobi polynomial of degree 𝑘 and by using (1) and (2);
in Section 3, the way of constructing the pseudospectral then we deduce that
technique for NFDEs with proportional delays is described
(𝛼,𝛽) Γ (𝑘 + 𝛽 + 1)
using the shifted Jacobi polynomials. In Section 4, we 𝑃𝐿,𝑘 (0) = (−1)𝑘 , (6)
investigate the shifted Jacobi-Gauss-Lobatto pseudo-spectral Γ (𝛽 + 1) 𝑘!
(SJGLP) method for solving nonlinear high-order NFDEs
with proportional delays. Some numerical results exhibit- (𝛼,𝛽)
(−1)𝑘−𝑞 Γ (𝑘 + 𝛽 + 1) (𝑘 + 𝛼 + 𝛽 + 1)𝑞
ing the accuracy of the proposed algorithm are given in 𝐷𝑞 𝑃𝐿,𝑘 (0) = , (7)
𝐿𝑞 Γ (𝑘 − 𝑞 + 1) Γ (𝑞 + 𝛽 + 1)
Section 5. Finally, a conclusion is given in Section 6.
(𝛼,𝛽) Γ (𝑚 + 𝑘 + 𝛼 + 𝛽 + 1) (𝛼+𝑚,𝛽+𝑚)
𝐷𝑚 𝑃𝐿,𝑘 (𝑥) = 𝑃 (𝑥) . (8)
2. Preliminaries 𝐿𝑚 Γ (𝑘 + 𝛼 + 𝛽 + 1) 𝐿,𝑘−𝑚
In this section, we briefly recall some properties of the Jacobi The following inner product and norm
polynomials (𝐽𝑘(𝜃,𝜗) (𝑥), 𝑘 = 0, 1, . . ., 𝜃 > −1, 𝜗 > −1), which
𝐿
satisfy the following relations: (𝛼,𝛽)
(𝑢, V)𝑤(𝛼,𝛽) = ∫ 𝑢 (𝑥) V (𝑥) 𝑤𝐿 (𝑥) 𝑑𝑥,
𝐿 0
(9)
𝐽𝑘(𝜃,𝜗) (−𝑥) = (−1)𝑘 𝐽𝑘(𝜃,𝜗) (𝑥) ,
‖V‖𝑤(𝛼,𝛽) = (V, V)1/2(𝛼,𝛽) ,
𝐿 𝑤𝐿
(−1)𝑘 Γ (𝑘 + 𝜗 + 1)
𝐽𝑘(𝜃,𝜗) (−1) = , (1)
𝑘!Γ (𝜗 + 1) are defined on the weighted space 𝐿2𝑤(𝛼,𝛽) (0, 𝐿) subject to the
𝐿
Γ (𝑘 + 𝜃 + 1) (𝛼,𝛽) (𝛼,𝛽)
𝐽𝑘(𝜃,𝜗) (1) = . weight function 𝑤𝐿 (𝑥) = (𝐿 − 𝑥)𝛼 𝑥𝛽 . Moreover, 𝑃𝐿,𝑘 (𝑥)
𝑘!Γ (𝜃 + 1) forms a complete 𝐿2𝑤(𝛼,𝛽) (0, 𝐿)-orthogonal system.
𝐿
The 𝑞th derivative of Jacobi polynomials of degree 𝑘 can be According to (5), we get
given by
(𝛼,𝛽) 2 𝐿 𝛼+𝛽+1 (𝛼,𝛽) (𝛼,𝛽)
𝑃
Γ (𝑗 + 𝜃 + 𝜗 + 𝑞 + 1) (𝜃+𝑞,𝜗+𝑞) 𝐿,𝑘 𝑤(𝛼,𝛽) = ( 2 ) ℎ𝑘 := ℎ𝐿,𝑘 . (10)
𝐷(𝑞) 𝐽𝑘(𝜃,𝜗)
𝐿
(𝑥) = 𝑞 𝐽 (𝑥) . (2)
2 Γ (𝑗 + 𝜃 + 𝜗 + 1) 𝑘−𝑞
3. Linear High-Order NFDE with
These polynomials are the only polynomials arising as eigen- Proportional Delay
functions of the following singular Sturm-Liouville equation:
In this section, we shall investigate solutions to NFDEs with
2 proportional delays of the form
(1 − 𝑥 ) 𝜙 (𝑥) + [𝜗 − 𝜃 + (𝜃 + 𝜗 + 2) 𝑥] 𝜙 (𝑥)
(3)
(𝑚)
+ 𝑛 (𝑛 + 𝜃 + 𝜗 + 1) 𝜙 (𝑥) = 0. (𝑢(𝑥) + 𝑎 (𝑥) 𝑢 (𝛾𝑚 𝑥))
𝑚−1 (11)
Let 𝑤(𝜃,𝜗) (𝑥) = (1 − 𝑥)𝜃 (1 + 𝑥)𝜗 ; then we define the weighted = 𝛽𝑢 (𝑥) + ∑ 𝑏𝑛 (𝑥) 𝑢(𝑛) (𝛾𝑛 𝑥) + 𝑓 (𝑥) , 𝑥 ≥ 0,
space 𝐿2𝑤(𝜃,𝜗) as usual. The inner product and the norm of 𝑛=0
Abstract and Applied Analysis 3
(𝛼,𝛽) (𝛼,𝛽)
with the initial conditions where 𝑥𝐿,𝑁,𝑗 and 𝜛𝐿,𝑁,𝑗 are the nodes and the corresponding
weights of the shifted Jacobi-Gauss-Lobatto quadrature for-
𝑚−1
mula on the interval (0, 𝐿), respectively. Obviously,
∑ 𝑐𝑖𝑛 𝑢(𝑛) (0) = 𝜆 𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1, (12)
𝑛=0
where 𝑎 and 𝑏𝑛 (𝑛 = 0, 1, . . . , 𝑚 − 1) are given functions, (𝑢, V)𝑤(𝛼,𝛽) ,𝑁 = (𝑢, V)𝑤(𝛼,𝛽) , ∀𝑢, V ∈ 𝑆2𝑁−1 . (16)
𝐿 𝐿
meanwhile, 𝛽, 𝛾𝑛 , 𝑐𝑖𝑛 , and 𝜆 𝑖 are constants with 0 < 𝛾𝑛 < 1
(𝑛 = 0, 1, . . . , 𝑚). This class of equations plays an important
role in modeling phenomena of the real world.
In the pseudo-spectral methods [25–28], one needs to The shifted Jacobi-Gauss-Lobatto pseudo-spectral meth-
exactly satisfy the differential equation at specified collo- od for solving (28) and (29) is to seek 𝑢𝑁(𝑥) ∈ 𝑆𝑁(0, 𝐿), such
cation points in the domain of solution. Generally, the that
distribution of the collocation nodes can be freely chosen,
but an accurate approximations are obtained by selecting the
collocation nodes as the zeros of the orthogonal polynomials. (𝛼,𝛽) (𝛼,𝛽) (𝛼,𝛽) (𝑚)
(𝑢 (𝑥𝐿,𝑁−𝑚,𝑘 ) + 𝑎 (𝑥𝐿,𝑁−𝑚,𝑘 ) 𝑢 (𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 ))
For shifted Jacobi polynomials, two commonly used quadra-
ture and collocation nodes, namely, (i) shifted Jacobi-Gauss (𝛼,𝛽)
= 𝛽𝑢 (𝑥𝐿,𝑁−𝑚,𝑘 )
nodes (in the interior of the domain) and (ii) shifted Jacobi-
Gauss-Lobatto nodes (in the interior and at the two endpoints 𝑚−1
(𝛼,𝛽) (𝛼,𝛽)
of the domain). + ∑ 𝑏𝑛 (𝑥𝐿,𝑁−𝑚,𝑘 ) 𝑢(𝑛) (𝛾𝑛 𝑥𝐿,𝑁−𝑚,𝑘 )
Now, we will present the shifted Jacobi-Gauss-Lobatto 𝑛=0
(𝛼,𝛽) (17)
type quadratures. Let 𝑥𝑁,𝑗 , 0 ⩽ 𝑗 ⩽ 𝑁, be the nodes of the (𝛼,𝛽)
(𝛼,𝛽) + 𝑓 (𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
Jacobi-Gauss-Lobatto interpolation on (−1, 1), and let 𝜛𝑁,𝑗 ,
0 ⩽ 𝑗 ⩽ 𝑁, be the corresponding weights. Throughout 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
this paper, we assume that 𝑥𝐿,𝑁,𝑗 , 0 ⩽ 𝑗 ⩽ 𝑁, stands for
𝑚−1
the nodes of the shifted Jacobi-Gauss-Lobatto interpolation
(𝛼,𝛽) ∑ 𝑐𝑖𝑛 𝑢(𝑛) (0) = 𝜆 𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1,
on the interval (0, 𝐿). Thus 𝑥𝐿,𝑁,𝑗 , 0 ⩽ 𝑗 ⩽ 𝑁, and their 𝑛=0
(𝛼,𝛽) (𝛼,𝛽)
corresponding weights are 𝜛𝐿,𝑁,𝑗 = (𝐿/2)𝛼+𝛽+1 𝜛𝑁,𝑗 , 0 ⩽ 𝑗 ⩽
𝑁. Let 𝑆𝑁(0, 𝐿) be the set of all polynomials of degree ≤ 𝑁.
One gets for any 𝜙 ∈ 𝑆2𝑁−1 (0, 𝐿), (𝛼,𝛽)
where the 𝑥𝐿,𝑁−𝑚,𝑘 , 𝑘 = 1, 2, . . . , 𝑁 − 𝑚 − 1, are distinct and
(𝛼,𝛽) (𝛼,𝛽)
𝐿 lie between 0 and 𝐿, 𝑥𝐿,𝑁−𝑚,0 = 0, and 𝑥𝐿,𝑁−𝑚,𝑁−𝑚 = 𝐿.
𝛼 𝛽
∫ (𝐿 − 𝑥) 𝑥 𝜙 (𝑥) 𝑑𝑥 For simplicity in presentation and without loss of generality,
0 assume that 𝑎(𝑥) ≡ 1. We now derive the collocation
𝐿 𝛼+𝛽+1 1 𝐿 algorithm for solving (28) and (29). To do this, consider the
=( ) ∫ (1 − 𝑥)𝛼 (1 + 𝑥)𝛽 𝜙 ( (𝑥 + 1)) 𝑑𝑥 solution is approximated by a truncated Jacobi expansion
2 −1 2
(13)
𝐿 𝛼+𝛽+1 𝑁 (𝛼,𝛽) 𝐿 (𝛼,𝛽)
=( ) ∑𝜛𝑁,𝑗 𝜙 ( (𝑥𝑁,𝑗 + 1)) 𝑁
2 𝑗=0 2 (𝛼,𝛽) 𝑇
𝑢𝑁 (𝑥) = ∑ 𝑎ℎ 𝑃𝐿,ℎ (𝑥) , a = (𝑎0 , 𝑎1 , . . . , 𝑎𝑁) . (18)
𝑁 ℎ=0
(𝛼,𝛽) (𝛼,𝛽)
= ∑𝜛𝐿,𝑁,𝑗 𝜙 (𝑥𝐿,𝑁,𝑗 ) .
𝑗=0
We first approximate 𝑢(𝑥) and 𝑢𝑛 (𝑥) as (32). By substitut-
Let us first introduce some basic notations that will be ing these approximation in (28), we get
used in the sequel. We set
Then, by virtue of (8), we deduce that Thus (22) with relation (23) can be written as a linear
algebraic system. To do this, let us consider
𝑁
Γ (ℎ + 𝛼 + 𝛽 + 𝑚 + 1) (𝛼+𝑚,𝛽+𝑚) 𝑇
∑ 𝑎ℎ ( 𝑃 (𝑥) a = (𝑎0 , 𝑎1 , . . . , 𝑎𝑁) ,
ℎ=0
𝐿𝑛 Γ (ℎ + 𝛼 + 𝛽 + 1) 𝐿,ℎ−𝑚
𝑓𝑘 = 𝑓 (𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, (24)
Γ (ℎ + 𝛼 + 𝛽 + 𝑚 + 1) (𝛼+𝑚,𝛽+𝑚)
+ 𝛾𝑚𝑚 𝑚 𝑃 (𝛾𝑚 𝑥)) 𝑇
𝐿 Γ (ℎ + 𝛼 + 𝛽 + 1) 𝐿,ℎ−𝑚 f = (𝑓0 , 𝑓1 , . . . , 𝑓𝑁−𝑚 , 𝜆 0 , . . . , 𝜆 𝑚−1 ) .
𝑁
(𝛼,𝛽) The matrix system associated with (22) and (23) becomes
= 𝛽 ∑ 𝑎ℎ 𝑃𝐿,ℎ (𝑥)
ℎ=0
𝑚−1
𝑚−1 𝑁
Γ (ℎ + 𝛼 + 𝛽 + 𝑛 + 1) (𝐴 + 𝛾𝑚𝑚 𝐵 + 𝛽𝐶 + ∑ 𝛾𝑛𝑛 𝐷𝑛 + 𝐸) a = f, (25)
+ ∑ ∑ 𝛾𝑛(𝑛) 𝑏𝑛 (𝑥) 𝑎ℎ 𝑛=0
𝑛=0 ℎ=0 𝐿𝑛 Γ (ℎ + 𝛼 + 𝛽 + 1)
(𝛼+𝑛,𝛽+𝑛) where the matrices 𝐴, 𝐵, 𝐶, 𝐷𝑖 , 𝑖 = 1, 2, . . . , 𝑚 − 1, and
× 𝑃𝐿,ℎ−𝑛 (𝛾𝑛 𝑥) + 𝑓 (𝑥) . 𝐸 are given explicitly in the following. If we denote that
(20) 𝐴 = (𝑎𝑘𝑗 )0<𝑘,𝑗<𝑁, 𝐵 = (𝑏𝑘𝑗 )0<𝑘,𝑗<𝑁, 𝐶 = (𝑐𝑘𝑗 )0<𝑘,𝑗<𝑁, 𝐷𝑛 =
𝑛
(𝑑𝑘𝑗 )0<𝑘,𝑗<𝑁, 𝑛 = 1, 2, . . . , 𝑚 − 1, and 𝐸 = (𝑒𝑘𝑗 )0<𝑘,𝑗<𝑁, then the
𝑛
Also, by substituting (32) in (29), we obtain elements 𝑎𝑘𝑗 , 𝑏𝑘𝑗 , 𝑐𝑘𝑗 , and 𝑑𝑘𝑗 are given by
𝑚−1 𝑁
(𝛼,𝛽)
𝑎𝑘𝑗
∑ ∑ 𝑐𝑖𝑛 𝑎ℎ 𝐷(𝑛) 𝑃𝐿,ℎ (0) = 𝜆 𝑖 . (21)
𝑛=0 ℎ=0
{ Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1)
{
{
{
{ 𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1)
{
{
To find the solution 𝑢𝑁(𝑥), we first collocate (20) at the { (𝛼+𝑚,𝛽+𝑚)
{ ×𝑃𝐿,𝑗−𝑚 (𝛼,𝛽)
(𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
(𝑁 − 𝑚 + 1) shifted Jacobi roots that yields ={
{
{ 𝑗 = 0, 1, . . . , 𝑁,
{
{
{
{ 0, 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁,
𝑁 {
{
Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1) (𝛼+𝑚,𝛽+𝑚) (𝛼,𝛽) 𝑗 = 0, 1, . . . , 𝑁,
∑𝑎𝑗 ( 𝑃 (𝑥𝐿,𝑁−𝑚,𝑘 ) {
𝑗=0 𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1) 𝐿,𝑗−𝑚
𝑏𝑘𝑗
Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1)
+ 𝛾𝑚𝑚 𝑚 Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1)
𝐿 Γ (𝑗 + 𝛼 + 𝛽 + 1) {
{
{
{ 𝑚 (𝑗 + 𝛼 + 𝛽 + 1)
{ 𝐿 Γ(𝛼+𝑚,𝛽+𝑚)
{
{
(𝛼+𝑚,𝛽+𝑚) (𝛼,𝛽) {
{ ×𝑃𝐿,𝑗−𝑚 (𝛼,𝛽)
(𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
× 𝑃𝐿,𝑗−𝑚 (𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 ) ) ={
{
{ 𝑗 = 0, 1, . . . , 𝑁,
{
{
{
{ 0, 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁,
𝑁
(𝛼,𝛽) (𝛼,𝛽)
{
{
= 𝛽∑ 𝑎𝑗 𝑃𝐿,𝑗 (𝑥𝐿,𝑁−𝑚,𝑘 ) { 𝑗 = 0, 1, . . . , 𝑁,
𝑗=0
𝑐𝑘𝑗
𝑚−1 𝑁
(𝛼,𝛽)
+ ∑ ∑𝛾𝑛(𝑛) 𝑏𝑛 (𝑥𝐿,𝑁−𝑚,𝑘 ) 𝑎𝑗 (𝛼,𝛽)
{ −𝑃
𝑛=0 𝑗=0 { 𝐿,𝑗 (𝛼,𝛽)
= { × (𝑥𝐿,𝑁−𝑚,𝑘 ), 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = 0, 1, . . . , 𝑁,
Γ (𝑗 + 𝛼 + 𝛽 + 𝑛 + 1) (𝛼+𝑛,𝛽+𝑛) {
× 𝑃
(𝛼,𝛽)
(𝛾𝑛 𝑥𝐿,𝑁−𝑚,𝑘 ) {0, 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁, 𝑗 = 0, 1, . . . , 𝑁,
𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1) 𝐿,𝑗−𝑛
𝑛
𝑑𝑘𝑗
(𝛼,𝛽)
+ 𝑓 (𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚.
(𝛼,𝛽)
(22) { −𝑏𝑛 (𝑥𝐿,𝑁−𝑚,𝑘 )
{
{
{
{ Γ (𝑗 + 𝛼 + 𝛽 + 𝑛 + 1)
{ ×
{
{
{ 𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1)
Next, (21), after using (7), can be written as {
{ (𝛼+𝑛,𝛽+𝑛) (𝛼,𝛽)
= { ×𝑃𝐿,𝑗−𝑛 (𝛾𝑛 𝑥𝐿,𝑁−𝑚,𝑘 ) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
{
{
𝑚−1 𝑁
Γ (𝑗 + 𝛽 + 1) (𝑗 + 𝛼 + 𝛽 + 1)𝑛 {
{ 𝑗 = 0, 1, . . . , 𝑁,
{
{
∑ ∑ (−1)𝑗−𝑛 𝑐𝑖𝑛 𝑎𝑗 = 𝜆𝑖, {
{ 0, 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁,
𝐿𝑛 Γ (𝑗 − 𝑛 + 1) Γ (𝑛 + 𝛽 + 1) {
{
𝑛=0 𝑗=0 (23)
{ 𝑗 = 0, 1, . . . , 𝑁.
𝑖 = 0, 1, . . . , 𝑚 − 1. (26)
Abstract and Applied Analysis 5
0, 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑑𝑚 (𝛼,𝛽)
{
{ 𝑢 (𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 )) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚.
{
{ 𝑑𝑥𝑚
{
{ 𝑗 = 0, 1, . . . , 𝑁,
{
{
{
{ 𝑚−1 (30)
{
{ 𝑗−𝑛
{ ∑ (−1)
{
= { 𝑛=0
{
{ ×𝑐𝑘−𝑁−𝑚+1,𝑛 Now, we approximate the numerical solution as a trun-
{
{
{
{ Γ(𝑗+𝛽+1)(𝑗+𝛼+𝛽+1)𝑛 cated series expansion of shifted Jacobi polynomial in the
{
{ × 𝑛 , 𝑘 = 𝑁−𝑚+1, . . . , 𝑁,
{
{ form
{
{ 𝐿 Γ(𝑗−𝑛+1) Γ(𝑛+𝛽+1)
{
{ 𝑗 = 0, 1, . . . , 𝑁.
(27) 𝑁
(𝛼,𝛽)
𝑢𝑁 (𝑥) = ∑𝑎𝑗 𝑃𝐿,𝑗 (𝑥) . (31)
𝑗=0
In the case of 𝑎(𝑥) ≠0, 𝑏𝑛 (𝑥) ≠
0, 𝑛 = 0, 1, . . . , 𝑚 − 1, and
𝛽 ≠ 0, the linear system (25) can be solved by forming explic-
itly the LU factorization; that is, 𝐴+𝛾𝑚𝑚 𝐵+𝛽𝐶+∑𝑚−1 𝑛
𝑛=0 𝛾𝑛 𝐷𝑛 +
Accordingly, (30) can be written as
𝐸 = LU. The expense of calculating LU factorization is 𝑂(𝑁3 )
operations, and the expense of solving the linear system (25), 𝑁
𝑑𝑚 (𝛼,𝛽) (𝛼,𝛽)
provided that the factorization is known, is 𝑂(𝑁2 ). ∑ 𝑎𝑗 𝑃 (𝑥𝐿,𝑁−𝑚,𝑘 )
𝑗=0 𝑑𝑥𝑚 𝐿,𝑗
0 ≤ 𝑥 ≤ 𝐿, 𝑘 = 0, 1, . . . , 𝑁 − 𝑚.
𝑁
Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1)
𝑑𝑚 (𝛼,𝛽) ∑ 𝑎𝑗 𝛾1 (
𝑢 (𝑥𝐿,𝑁−𝑚,𝑘 ) 𝐿𝑛 Γ (𝑗 + 𝛼 + 𝛽 + 1)
𝑑𝑥𝑚 𝑗=0
6 Abstract and Applied Analysis
(𝛼+𝑚,𝛽+𝑚) (𝛼,𝛽)
× 𝑃𝐿,𝑗−𝑚 (𝛾1 𝑥𝐿,𝑁−𝑚,𝑘 ) ) , . . . , 0.35
0.30
𝑁
Γ (𝑗 + 𝛼 + 𝛽 + 𝑚 + 1) 0.25
∑𝑎𝑗 𝛾𝑚𝑚 ( 𝑛
𝑗=0 𝐿 Γ (𝑗 + 𝛼 + 𝛽 + 1) 0.20
u(x)
0.15
(𝛼+𝑚,𝛽+𝑚) (𝛼,𝛽)
× 𝑃𝐿,𝑗−𝑚 (𝛾𝑚 𝑥𝐿,𝑁−𝑚,𝑘 ) ) ) , 0.10
0.05
𝑘 = 0, 1, . . . , 𝑁 − 𝑚.
0.00
(33) 0 2 4 6 8 10
x
Exact solution
Finally, to find the unknown expansion coefficients 𝑎𝑗 , 𝑗 = Approximate method at N = 20
0, 1, . . . , 𝑁, we implement any iteration technique to solve a
Figure 1: Comparison of the approximate solution with the exact
system of 𝑁 + 1 nonlinear algebraic equations resulting from solution for 𝛼 = −1/2, 𝛽 = 1/2 at 𝑁 = 28 for Example 1.
the combination of 𝑁 − 𝑚 + 1 nonlinear algebraic equations
resulting from (33) and 𝑚 linear algebraic equations resulting
from (29). ×10−12
4.0
3.0
5. Numerical Results
2.0
In this section, we will carry out three test examples to study
1.0
Error
+ (0.32𝑥 − 0.5) 𝑒−0.8𝑥 + 𝑒−𝑥 , 𝑥 ≥ 0, (34) Example 2. Let us consider the second-order NFDE with
proportional delay
𝑢 (0) = 0,
𝑥 1 𝑥 𝜋
𝑢 (𝑥) = 𝑢 ( ) − 𝑥𝑢 ( ) −
2 2 2 2
which has the exact solution 𝑥𝑒−𝑥 . 𝜋𝑥 𝜋𝑥
× (2 cos ( ) + 𝜋𝑥 sin ( )
2 2 (35)
Table 1 lists the absolute error using Jacobi pseudospectral
method for three choices of 𝛼, 𝛽 at 𝑁 = 20 in the interval + 2𝜋 sin (𝜋𝑥) ) , 𝑥 ∈ [0, 6] ,
[0, 1]. We compare the errors obtained by the proposed
method with variational iteration (VI) method [32], the one- 𝑢 (0) = 0, 𝑢 (0) = 𝜋,
leg 𝜃 method [14, 33] with 𝜃 = 0.8, and RKHSM method [31].
The graph of analytical solution and approximate solution in which enjoys exact solution 𝑢(𝑥) = sin(𝜋𝑥).
long interval [0, 10] for 𝑁 = 28 and 𝛼 = −1/2, 𝛽 = 1/2
is displayed in Figure 1 to make it easier to compare with In Table 2, we introduce the absolute error using the
analytical solution. Moreover, in this case the graph of the proposed method at 𝑁 = 24 with various choices of 𝛼 and
error is given in Figure 2. Consequently, we conclude that the 𝛽. The resulting graph of (35) for the presented method in
approximate solution by Jacobi pseudospectral method agree the case of 𝛼 = 𝛽 = 1 at 𝑁 = 24 and the analytic solution are
very well with the exact solution. shown in Figure 3.
Abstract and Applied Analysis 7
Table 1: Absolute errors using SJGLP method with various choices of 𝑁, 𝛼, and 𝛽 for Example 1.
Table 2: Absolute errors using SJGLP method with various choices where
of 𝛼 and 𝛽 for Example 2.
1 𝑥/12
𝑥 𝛼 = 𝛽 = −1/2 𝛼 = 𝛽 = 1/2 𝛼=𝛽=0 𝑓 (𝑥) = 𝑒 ( − 13752 sin (2𝑥) − 54𝑒7𝑥/12 cos2 (8𝑥)
54
0.0 4.13 ⋅ 10−16 3.92 ⋅ 10−16 1.11 ⋅ 10−16
0.1 6.76 ⋅ 10−10 1.29 ⋅ 10−9 3.57 ⋅ 10−10 + 1727 cos (2𝑥) + 6𝑒𝑥/36
0.2 5.01 ⋅ 10−9 3.70 ⋅ 10−9 5.86 ⋅ 10−10
8𝑥 8𝑥
0.3 3.24 ⋅ 10−9 9.52 ⋅ 10−9 1.71 ⋅ 10−9 × (48 sin ( ) + 575 cos ( ))
3 3
0.4 4.14 ⋅ 10−8 5.17 ⋅ 10−7 2.16 ⋅ 10−9
0.5 2.61 ⋅ 10−8 2.88 ⋅ 10−8 2.27 ⋅ 10−9 − 18𝑒𝑥/12 (cos (4𝑥) − 24 sin (4𝑥))
1.0 3.72 ⋅ 10−8 1.01 ⋅ 10−8 2.13 ⋅ 10−8
2.0 7.32 ⋅ 10−8 3.89 ⋅ 10−9 3.59 ⋅ 10−8 + 2𝑒𝑥/4 (13752 sin (8𝑥) −1727 cos (8𝑥)) ).
3.0 2.19 ⋅ 10−9 2.84 ⋅ 10−8 1.85 ⋅ 10−7
(37)
4.0 3.52 ⋅ 10−7 7.78 ⋅ 10−7 3.01 ⋅ 10−7
5.0 3.73 ⋅ 10−6 1.24 ⋅ 10−6 1.80 ⋅ 10−6
6.0 3.42 ⋅ 10−6 2.97 ⋅ 10−6 8.74 ⋅ 10−6
The exact solution of the problem is 𝑢(𝑥) = 𝑒𝑥/3 cos(8𝑥).
0.0 References
u(x)
[16] A. H. Bhrawy, L. M. Assas, E. Tohidi, and M. A. Alghamdi, [33] W. Wang, T. Qin, and S. Li, “Stability of one-leg 𝜃-methods
“Legendre-Gauss collocation method for neutral functional- for nonlinear neutral differential equations with proportional
differential equations with proportional delays,” Advances in delay,” Applied Mathematics and Computation, vol. 213, no. 1, pp.
Differential Equations, vol. 2013, article 63, 16 pages, 2013. 177–183, 2009.
[17] Z.-Z. Sun and Z.-B. Zhang, “A linearized compact difference [34] Z. Jackiewicz and B. Zubik-Kowal, “Spectral collocation and
scheme for a class of nonlinear delay partial differential equa- waveform relaxation methods for nonlinear delay partial differ-
tions,” Applied Mathematical Modelling, vol. 37, no. 3, pp. 742– ential equations,” Applied Numerical Mathematics, vol. 56, no.
752, 2013. 3-4, pp. 433–443, 2006.
[18] L. F. Cordero and R. Escalante, “Segmented tau approximation
for test neutral functional differential equations,” Applied Math-
ematics and Computation, vol. 187, no. 2, pp. 725–740, 2007.
[19] M. Muslim, “Approximation of solutions to history-valued
neutral functional differential equations,” Computers & Math-
ematics with Applications, vol. 51, no. 3-4, pp. 537–550, 2006.
[20] O. R. Işik, Z. Güney, and M. Sezer, “Bernstein series solutions of
pantograph equations using polynomial interpolation,” Journal
of Difference Equations and Applications, vol. 18, no. 3, pp. 357–
374, 2012.
[21] E. K. Ifantis, “An existence theory for functional-differential
equations and functional-differential systems,” Journal of Dif-
ferential Equations, vol. 29, no. 1, pp. 86–104, 1978.
[22] A. Iserles and Y. Liu, “On neutral functional-differential equa-
tions with proportional delays,” Journal of Mathematical Analy-
sis and Applications, vol. 207, no. 1, pp. 73–95, 1997.
[23] C. He, X. Lv, and J. Niu, “A new method based on the RKHSM
for solving systems of nonlinear IDDEs with proportional
delays,” Abstract and Applied Analysis, vol. 2013, Article ID
541935, 13 pages, 2013.
[24] E. Ishiwata, “On the attainable order of collocation methods for
the neutral functional-differential equations with proportional
delays,” Computing, vol. 64, no. 3, pp. 207–222, 2000.
[25] A. H. Bhrawy and A. S. Alofi, “A Jacobi-Gauss collocation
method for solving nonlinear Lane-Emden type equations,”
Communications in Nonlinear Science and Numerical Simula-
tion, vol. 17, no. 1, pp. 62–70, 2012.
[26] A. H. Bhrawy and M. A. Alghamdi, “A shifted Jacobi-Gauss-
Lobatto collocation method for solving nonlinear fractional
Langevin equation involving two fractional orders in different
intervals,” Boundary Value Problems, vol. 2012, article 62, 13
pages, 2012.
[27] M. Maleki, I. Hashim, M. Tavassoli Kajani, and S. Abbas-
bandy, “An adaptive pseudospectral method for fractional order
boundary value problems,” Abstract and Applied Analysis, vol.
2012, Article ID 381708, 19 pages, 2012.
[28] A. H. Bhrawy, “A Jacobi-Gauss-Lobatto collocation method
for solving generalized Fitzhugh-Nagumo equation with time-
dependent coefficients,” Applied Mathematics and Computation,
vol. 222, pp. 255–264, 2013.
[29] W. Wang, Y. Zhang, and S. Li, “Stability of continuous Runge-
Kutta-type methods for nonlinear neutral delay-differential
equations,” Applied Mathematical Modelling, vol. 33, no. 8, pp.
3319–3329, 2009.
[30] J. X. Kuang and Y. H. Cong, Stability of Numerical Methods
for Delay Differential Equations, Science Press, Beijing, China,
2005.
[31] X. Lv and Y. Gao, “The RKHSM for solving neutral functional-
differential equations with proportional delays,” Mathematical
Methods in the Applied Sciences, vol. 36, no. 6, pp. 642–649, 2013.
[32] X. Chen and L. Wang, “The variational iteration method for
solving a neutral functional-differential equation with propor-
tional delays,” Computers & Mathematics with Applications, vol.
59, no. 8, pp. 2696–2702, 2010.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 828764, 8 pages
http://dx.doi.org/10.1155/2013/828764
Research Article
Numerical Solution of a Kind of Fractional Parabolic Equations
via Two Difference Schemes
Copyright © 2013 A. Atangana and D. Baleanu. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
A kind of parabolic equation was extended to the concept of fractional calculus. The resulting equation is, however, difficult to
handle analytically. Therefore, we presented the numerical solution via the explicit and the implicit schemes. We presented together
the stability and convergence of this time-fractional parabolic equation with two difference schemes. The explicit and the implicit
schemes in this case are stable under some conditions.
Definition 2 (see [7–16]). The Riemann-Liouville fractional derivative of order 𝛼 for 𝑓 in respect to the 𝑥𝑖 the function as
integral operator of order 𝛼 ≥ 0, of a function 𝑓 ∈ 𝐶𝜇 , 𝜇 ≥ −1, follows:
is defined as
𝑥𝑖
1
𝜕𝑥𝑖 𝑓 (𝑥𝑗 )𝑥 =𝑡 𝑑𝑡,
𝑚−𝛼−1 𝑚
𝑥 𝑎𝜕𝛼x 𝑓 = ∫ (𝑥𝑖 − 𝑡) (8)
𝛼 1 Γ (𝑚 − 𝛼) 𝑎
∫ (𝑥 − 𝑡)𝛼−1 𝑓 (𝑡) 𝑑𝑡,
𝑗
𝐽 𝑓 (𝑥) = 𝛼 > 0, 𝑥 > 0,
Γ (𝛼) 0 (2)
where 𝜕𝑥𝑚𝑖 is the usual partial derivative of integer order 𝑚.
0
𝐽 𝑓 (𝑥) = 𝑓 (𝑥) .
3. Examination of the Numerical Solution via
Properties of the operator can be found in [7–15]; we only
mention the following: Difference Schemes
This section is devoted to the discussion underpinning the
for 𝑓 ∈ 𝐶𝜇 , 𝜇 ≥ −1, 𝛼, 𝛽 ≥ 0, 𝛾 > −1, numerical simulation of the solution above (1) via the explicit
scheme [19–23] and the implicit scheme [23–30]. However,
𝐽𝛼 𝐽𝛽 𝑓 (𝑥) = 𝐽𝛼+𝛽 𝑓 (𝑥) , 𝐽𝛼 𝐽𝛽 𝑓 (𝑥) = 𝐽𝛽 𝐽𝛼 𝑓 (𝑥) , before we present the numerical schemes, we must assume
(3) that (1) has a unique and sufficiently smooth solution [23]. In
𝛾 Γ (𝛾 + 1) 𝛼+𝛾 addition, to present the numerical schemes, we let 𝑥𝑙 = 𝑙ℎ,
𝐽𝛼 𝑥 = 𝑥 . 0 ≤ 𝑙 ≤ 𝑀, 𝑀ℎ = 𝐿, 𝑡𝑘 = 𝑘𝜏, 0 ≤ 𝑘 ≤ 𝑁, and 𝑁𝜏 = 𝑇; ℎ
Γ (𝛼 + 𝛾 + 1)
is the space step size, and 𝑀 and 𝑁 are grid points. We will
Definition 3. The Caputo fractional-order derivative is given start with the implicit scheme.
as follows [7–10]:
3.1. Implicit Scheme for the Main Problem. It is important to
𝑥 𝑑 𝑓 (𝑡)𝑛 recall that the finite difference approximation for the second-
𝐶 𝛼 1
0 𝐷𝑥 (𝑓 (𝑥)) = ∫ (𝑥 − 𝑡)𝑛−𝛼−1 𝑑𝑡, order spatial derivative is known as follows [23]:
Γ (𝑛 − 𝛼) 0 𝑑𝑡𝑛 (4)
𝑛 − 1 ≤ 𝛼 ≤ 𝑛. 𝜕2 V (𝑥𝑙 , 𝑡𝑘+1 ) V (𝑥𝑙+1 , 𝑡𝑘+1 ) − 2V (𝑥𝑙 , 𝑡𝑘+1 ) + V (𝑥𝑙−1 , 𝑡𝑘+1 )
=
𝜕𝑥2 ℎ2
Definition 4. The Riemann-Liouville fractional-order deriva- + 𝑂 (ℎ2 ) .
tive is given as follows [8–16]: (9)
1 𝑑𝑛 𝑥
𝐷𝑥𝛼 (𝑓 (𝑥)) = ∫ (𝑥 − 𝑡)𝑛−𝛼−1 𝑓 (𝑡) 𝑑𝑡, The discretization of the Caputo-type time-fractional-order
Γ (𝑛 − 𝛼) 𝑑𝑥𝑛 0 (5) derivative can be presented as follows:
𝑛 − 1 ≤ 𝛼 ≤ 𝑛. 𝜕𝛼 V (𝑥𝑙 , 𝑡𝑘+1 )
𝜕𝑡𝛼
Definition 5. The Jumarie fractional-order derivative is given
as follows [16]: 𝜏−𝛼
= (V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
Γ (2 − 𝛼)
1 𝑑𝑛 𝑥
𝐷𝑥𝛼 (𝑓 (𝑥)) = ∫ (𝑥 − 𝑡)𝑛−𝛼−1 {𝑓 (𝑡) − 𝑓 (0)} 𝑑𝑡, 𝑘
Γ (𝑛 − 𝛼) 𝑑𝑥𝑛 0 + ∑ [V (𝑥𝑙 , 𝑡𝑘+1−𝑗 ) − V (𝑥𝑙 , 𝑡𝑘−𝑗 )] (10)
𝑗=1
𝑛 − 1 ≤ 𝛼 ≤ 𝑛.
(6) 1−𝛼 1−𝛼
× [(𝑗 + 1) − (𝑗) ]) ,
Lemma 6. If 𝑚 − 1 < 𝛼 ≤ 𝑚, 𝑚 ∈ N, and 𝑓 ∈ 𝐶𝜇𝑚 , 𝜇 ≥ −1, 𝜕V (𝑥𝑙 , 𝑡𝑘+1 ) V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
= + 𝑂 (Δ𝑡) .
then 𝜕𝑡 Δ𝑡
𝐷𝛼 𝐽𝛼 𝑓 (𝑥) = 𝑓 (𝑥) , Now, substituting (9) and (10) into (1), we obtained the follo-
wing expression:
𝑚−1
𝑥𝑘
𝐽𝛼 𝐷0𝛼 𝑓 (𝑥) = 𝑓 (𝑥) − ∑ 𝑓(𝑘) (0+ ) , (7) V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 ) 𝜏−𝛼
𝑘=0
𝑘! +
Δ𝑡 Γ (2 − 𝛼)
𝑥 > 0.
× (V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
Definition 7 (partial derivatives of fractional order [7, 8, 11, 𝑘
18]). Assume now that 𝑓(x) is a function of 𝑛 variables 𝑥𝑖 , + ∑ [V (𝑥𝑙 , 𝑡𝑘+1−𝑗 ) − V (𝑥𝑙 , 𝑡𝑘−𝑗 )]
𝑖 = 1, . . . , 𝑛, also of class 𝐶 on 𝐷 ∈ R𝑛 . We define partial 𝑗=1
Abstract and Applied Analysis 3
1−𝛼 1−𝛼
× [(𝑗 + 1) − (𝑗) ]) It is important to inform that if 𝑘 = 0, then the term of the
sum of the right-hand side automatically vanished. Then, (13)
V (𝑥𝑙+1 , 𝑡𝑘+1 ) − 2V (𝑥𝑙 , 𝑡𝑘+1 ) + V (𝑥𝑙−1 , 𝑡𝑘+1 ) can be divided as follows:
− 𝑎𝑙𝑘 ( )
ℎ2
𝜏−𝛼 𝑎1
+ 𝛿V (𝑥𝑙 , 𝑡𝑘 ) = 𝑓 (𝑥𝑙 , 𝑡𝑘 ) . (Δ𝑡 + + 2 𝑙2 ) V𝑙1 − 𝑎𝑙1 V𝑙+1
1
− 𝑎𝑙1 V𝑙−1
1
Γ (2 − 𝛼) ℎ
(11) (14)
𝜏−𝛼
For ease, let = (Δ𝑡 + + 𝛿) V𝑙0 + 𝑓𝑙0 , 𝑘 = 0,
Γ (2 − 𝛼)
V𝑙𝑘 = V (𝑥𝑙 , 𝑡𝑘 ) , 𝑓𝑙𝑘 = 𝑓 (𝑥𝑙 , 𝑡𝑘 ) ,
(12) 𝜏−𝛼 𝑎𝑘
1−𝛼 1−𝛼 (Δ𝑡 + + 2 𝑙2 ) V𝑙𝑘+1 − 𝑎𝑙𝑘 V𝑙+1
𝑘+1
− 𝑎𝑙𝑘 V𝑙−1
𝑘+1
𝑎 (𝑥𝑙 , 𝑡𝑘 ) = 𝑎𝑙𝑘 , (𝑗 + 1) − (𝑗) = 𝑏𝑗 . Γ (2 − 𝛼) ℎ
𝜏−𝛼 𝑎1 V1𝑘+1
Δ𝑡 + + 2 𝑙2 −𝑎1𝑘 0 0 0 ⋅⋅⋅ 0
Γ (2 − 𝛼) ℎ
𝑎1 V2𝑘+1 )
( 𝜏−𝛼 d 0 0 )(
( −𝑎2𝑘 Δ𝑡 + + 2 22 )( )
( Γ (2 − 𝛼) ℎ d 0 0 )( )
( 0 )( .. )
( d 𝜏−𝛼 𝑎 1 )( . )
( 𝑘
+ 2 𝑚−2 𝑘 )( )
( .. −𝑎𝑚−2 Δ𝑡 +
Γ (2 − 𝛼) ℎ2
−𝑎𝑚−2 )(
(
)
)
. 𝑘+1
V𝑚−1
𝜏−𝛼 𝑎1
0 0 𝑘
0 0 −𝑎𝑚−1 Δ𝑡 + + 2 𝑙2 𝑘+1
( Γ (2 − 𝛼) ℎ ) ( V𝑚−1 )
(16)
𝑎1𝑘
𝑓1𝑘 − V (0, 𝑡𝑘 )
ℎ2
( 𝑓2𝑘 )
( .. )
( )
=( . ).
( 𝑘
𝑓𝑚−2 )
( )
𝑘
𝑘 𝑎𝑚−1
𝑓𝑚−1 − V (0, 𝑡𝑘 )
( ℎ2 )
3.1.1. Stability of the Implicit Difference Scheme. In this To analyze the stability, we exploit the Fourier method
subsection, we present the stability analysis of the implicit [22], and the expression of 𝜁𝑘 can be defined as follows:
difference scheme for solving the time-fractional parabolic
equation (1). To accomplish this, we let 𝜁𝑙𝑘 = V𝑙𝑘 − 𝑉𝑙𝑘 , with ℎ ℎ
{
{ 𝜁𝑙𝑘 , if 𝑥𝑙 − < 𝑥 ≤ 𝑥𝑙 + ,
{
{ 2 2
𝑉𝑙𝑘 being the approximate solution of the main problem at
𝜁𝑘 (𝑥) = { 𝑙 = 1, 2, . . . , 𝑀 − 1, (17)
the point (𝑥𝑙 , 𝑡𝑘 ), 𝑘 = 1, . . . , 𝑁; 𝑙 = 1, . . . , 𝑀; in addition, 𝜁𝑘 is {
{
{ ℎ
the transpose of the matrix [𝜁1𝑘 , 𝜁2𝑘 , . . . , 𝜁𝑀 𝑘
]. 0, if 𝐿 − < 𝑥 ≤ 𝐿.
{ 2
4 Abstract and Applied Analysis
Then, the function 𝜁𝑘 (𝑥) can be expressed in Fourier series as And for 𝑘 ≥ 1, we have
follows:
𝜏−𝛼 𝜌ℎ 𝑎𝑘
𝑚=∞
2𝑖𝜋𝑚𝑘 (Δ𝑡 + + 2sin2 ( ) 𝑙2 ) 𝛿𝑘+1
𝜁𝑘 (𝑥) = ∑ 𝛿𝑚 (𝑚) exp [ ], Γ (2 − 𝛼) 2 ℎ
𝑚=−∞ 𝐿
(18) 𝜏−𝛼 𝜌ℎ
𝐿 = (Δ𝑡 + + 2sin2 ( ) 𝛿 − 𝑑1 ) 𝛿𝑘 (25)
1 2𝑖𝜋𝑚𝑥 Γ (2 − 𝛼) 2
𝛿𝑘 (𝑥) = ∫ 𝜌𝑘 (𝑥) exp [ ] 𝑑𝑥.
𝐿 0 𝐿
𝜏−𝛼 𝑘−1
It was proven in [22, 29, 30] that − ∑ [𝛿 ] 𝑑 + 𝑏 𝛿 .
Γ (2 − 𝛼) 𝑗=1 𝑘−𝑗 𝑗 𝑘+1 0
2 2 𝑚=∞
𝜌 = ∑ 𝛿𝑘 (𝑚)2 . (19) Observe that 𝛿 ≤ 𝑎𝑙0 /ℎ2 + 𝑑1 then, the following is obtained:
2
𝑚=−∞
Let us now examine the stability of the implicit scheme of the (Δ𝑡 + 𝜏−𝛼 /Γ (2 − 𝛼) + 2sin2 (𝜌ℎ/2) 𝛿 − 𝑑1 )
𝛿1 = 𝛿0 . (26)
main problem (Δ𝑡 + 𝜏−𝛼 /Γ (2 − 𝛼) + 2sin2 (𝜌ℎ/V) (𝑎𝑙0 /ℎ2 ))
But we have that Here, 𝑇𝑙𝑘+1 is the truncate or the remainder term of the
approximation and has the following expression:
𝑘−1
∑ 𝑑𝑗 = 1 − 𝑏𝑘+1 , 0 ≤ 𝑑𝑗 ≤ 1. (31) 𝜏−𝛼 𝑘−1 𝑘−𝑗
𝑗=1 𝑇𝑙𝑘+1 = V (𝑥𝑙 , 𝑡𝑘+1 ) + ∑𝛽 𝑏
Γ (2 − 𝛼) 𝑗=1 𝑙 𝑗
Therefore,
+ 𝑎 (𝑥𝑙 , 𝑡𝑘 ) [V (𝑥𝑙+1 , 𝑡𝑘+1 )
−𝛼 2
Δ𝑡 + 𝜏 /Γ (2 − 𝛼) + 2sin (𝜌ℎ/2) 𝛿 − 𝑑1 −2V (𝑥𝑙 , 𝑡𝑘+1 ) + V (𝑥𝑙−1 , 𝑡𝑘+1 )]
𝛿𝑘+1 ≤ [ ] 𝛿0 .
Δ𝑡 + 𝜏−𝛼 /Γ (2 − 𝛼) + 2sin2 (𝜌ℎ/2) (𝑎𝑙𝑘 /ℎ2 )
(32) + (V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )) − 𝑓 (𝑥𝑙 , 𝑡𝑘 ) + 𝛿V (𝑥𝑙 , 𝑡𝑘 ) .
(36)
Thus, It follows from (3), (4), and (5) that,
𝜕𝛼 V (𝑥𝑙 , 𝑡𝑘+1 )
𝛿𝑘+1 ≤ 𝛿0 . (33)
+ 𝐷1 𝜏
𝜕𝑡𝛼
𝜏−𝛼
=
Theorem 9. The implicit difference scheme for the time- Γ (2 − 𝛼)
fractional parabolic equation (1) is stable providing that, for
all (𝑙, 𝑘), 𝛿 ≤ 𝑎𝑙𝑘 /ℎ2 + 𝑑1 (𝑙 = 1, . . . , 𝑁; 𝑘 = 1, . . . , 𝑀). × (V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
Proof. From (19) and Lemma 8, we obtain
𝑘−1
2 (37)
𝜁 ≤ 𝜁0 , (34) + ∑ V (𝑥𝑙 , 𝑡𝑘−𝑗 ) 𝑑𝑗 + 𝑑𝑘+1 V (𝑥𝑙 , 𝑡0 )) ,
2 2
𝑗=1
and this proves that the implicit difference scheme for the
𝜕2 V (𝑥𝑙 , 𝑡𝑘+1 )
time-fractional parabolic equation (1) is stable. + ℎ2 𝐷2
𝜕𝑥2
Remark 10. It is observed that, from Theorem 9, the statement V (𝑥𝑙+1 , 𝑡𝑘+1 ) − 2V (𝑥𝑙 , 𝑡𝑘+1 ) + V (𝑥𝑙−1 , 𝑡𝑘+1 )
of stability of implicit difference scheme for time-fractional = ,
parabolic equation (1) depends on the evolution of the ℎ2
function 𝑎(𝑥, 𝑡). It follows that the stability condition can 𝜕V (𝑥𝑙 , 𝑡𝑘+1 ) V (𝑥𝑙 , 𝑡𝑘+1 ) − V (𝑥𝑙 , 𝑡𝑘 )
change in time advancement and space position. + Δ𝑡𝐷3 = .
𝜕𝑡 Δ𝑡
Thus, from (36) and (37), we have the following:
3.1.2. Convergence Analysis of the Implicit Difference Scheme.
Assuming that V(𝑥𝑙 , 𝑡𝑘 ) (𝑙 = 0, . . . , 𝑀 − 1; 𝑘 = 0, . . . , 𝑁 − 1) is 𝑇𝑙𝑘+1 ≤ 𝐷 (𝜏1+𝛼 + ℎ2 𝜏𝛼 + Δ𝑡𝜏𝛼 ) , (38)
the exact solution of (1) at the point (𝑥𝑙 , 𝑡𝑘 ), then, by defining,
𝛽𝑙𝑘 = V(𝑥𝑙 , 𝑡𝑘 ) − V𝑙𝑘 , and 𝛽𝑘 is the transpose of the matrix where 𝐷1 , 𝐷2 , 𝐷3 , and 𝐷 are constant. The interested reader
(𝛽1𝑘 , 𝛽2𝑘 , . . . , 𝛽𝑁
𝑘
); here, 𝛽0 is neglected because of being equal can find the error analysis of the chosen fractional derivative
to zero. Therefore, we have the following relation for the (the Caputo fractional derivative) in [29, 30].
implicit difference scheme for the time-fractional parabolic −1
equation (1) and from (13): Lemma 11. One has that ‖𝛽𝑘 ‖∞ ≤ 𝐷(𝑏𝑘+1 ) [𝜏1+𝛼 + 𝜏𝛼 ℎ2 +
Δ𝑡𝜏𝛼 ] is true for all 𝑘 = 0, 1, . . . , 𝑀 − 1.
𝜏−𝛼 𝑎0
(Δ𝑡 + + 2 𝑙2 ) 𝛽𝑙1 − 𝑎𝑙0 𝛽𝑙+1
1
Proof. Again we employ the induction method to achieve
Γ (2 − 𝛼) ℎ this, so that, for 𝑘 = 0,
− 𝑎𝑙0 𝛽𝑙−1
1
− 𝑓𝑙0 = 𝑇𝑙1 , for 𝑘 = 0, 𝜏−𝛼 𝑎0 1 1 1
(Δ𝑡 + + 2 𝑙2 ) 𝛽𝑙1 − 𝑎𝑙0 𝛽𝑙+1 − 𝑎𝑙0 𝛽𝑙−1
≥ 𝛽𝑙 ,
Γ (2 − 𝛼) ℎ
𝜏−𝛼 𝑎𝑘 𝜏−𝛼
(Δ𝑡 + + 2 𝑙2 ) 𝛽𝑙𝑘+1 − (Δ𝑡 + + 𝛿) 𝛽𝑙𝑘 (39)
Γ (2 − 𝛼) ℎ Γ (2 − 𝛼)
1 𝜏−𝛼 𝑎𝑙0 1 0 1 0 1
− 𝑎𝑙𝑘 𝛽𝑙+1
𝑘+1
− 𝑎𝑙𝑘 𝛽𝑙−1
𝑘+1
− 𝑓𝑙𝑘 𝑇𝑙 = (Δ𝑡 + + 2 ) 𝛽 − 𝑎 𝛽 − 𝑎 𝛽
𝑙 𝑙−1
Γ (2 − 𝛼) ℎ2 𝑙 𝑙 𝑙+1
𝜏−𝛼 𝑘−1 𝑘−𝑗
≥ 𝛽𝑙1 ,
(40)
=− ∑ 𝛽 𝑏 + 𝑇𝑙𝑘 , for 𝑘 ≥ 1.
Γ (2 − 𝛼) 𝑗=1 𝑙 𝑗
−1
(35) 𝐷(𝑏1 ) (𝜏1+𝛼 + ℎ2 𝜏𝛼 + Δ𝑡𝜏𝛼 ) ≥ 𝑇𝑙1 .
6 Abstract and Applied Analysis
Now, assuming that, for all 𝑗 = 0, . . . , 𝑀 − 2, ‖𝛽𝑗 ‖∞ ≤ The discretization of the Caputo-type time-fractional-order
−1
𝐷(𝑏𝑗+1 ) [𝜏1+𝛼 + 𝜏𝛼 ℎ2 + Δ𝑡𝜏𝛼 ], then, derivative can be presented as follows:
+ 𝛿V (𝑥𝑙 , 𝑡𝑘 ) = 𝑓 (𝑥𝑙 , 𝑡𝑘 ) .
≥ 𝛽𝑙𝑘+1 ,
−1 For simplicity, let
(𝑏𝑘+1 ) × 𝐷 [𝜏1+𝛼 + 𝜏𝛼 ℎ2 + Δ𝑡𝜏𝛼 ] ≥ 𝛽𝑙𝑘+1 .
(42)
V𝑙𝑘 = V (𝑥𝑙 , 𝑡𝑘 ) , 𝑓𝑙𝑘 = 𝑓 (𝑥𝑙 , 𝑡𝑘 ) ,
This completes the proof. (47)
1−𝛼 1−𝛼
𝑎 (𝑥𝑙 , 𝑡𝑘 ) = 𝑎𝑙𝑘 , (𝑗 + 1) − (𝑗) = 𝑏𝑗 .
Theorem 12. The implicit difference scheme of the time-
fractional parabolic equation (1) is convergent, and there exists
Then, (37) can be rewritten as follows
a constant 𝐷 such that,
V (𝑥𝑙 , 𝑡𝑘 ) − V𝑙𝑘 ≤ 𝐷 (𝜏1+𝛼 + 𝜏𝛼 ℎ2 + Δ𝑡𝜏𝛼 ) , V𝑙𝑘+1 = V𝑙𝑘 [𝐵1 + 1 − 𝑟𝑙𝑘 − 𝛿] + V𝑙+1
𝑘
[𝑟𝑙𝑘 −𝐵1 ]
(43)
𝑓𝑜𝑟 (𝑙 = 0, . . . , 𝑁 − 1; 𝑘 = 0, . . . , 𝑀 − 1) . 𝑘 (48)
𝑘+1−𝑗 𝑘−𝑗
+ 𝑟𝑙𝑘 V𝑙−1
𝑘
+ 𝐵1 𝑓𝑙𝑘 − ∑ [V𝑙 − V𝑙 ] 𝑏𝑗 ,
𝑗=1
3.2. Explicit Difference Scheme for Time-Fractional Parabolic
Equation. It is important to recall that the finite-difference
approximation for the second-order spatial derivative is where 𝐵1 = Γ(2 − 𝛼)𝜏𝛼 and 𝑟𝑙𝑘 = (Γ(2 − 𝛼)𝜏𝛼 /ℎ2 )𝑎𝑙𝑘 .
known as follows:
𝜕2 V (𝑥𝑙 , 𝑡𝑘 ) V (𝑥𝑙+1 , 𝑡𝑘 ) − 2V (𝑥𝑙 , 𝑡𝑘 ) + V (𝑥𝑙−1 , 𝑡𝑘 ) 3.2.1. Stability of the Explicit Difference Scheme of the Time-
= + 𝑂 (ℎ2 ) .
𝜕𝑥2 ℎ2 Fractional Parabolic Equation. Following the discussion pre-
(44) sented earlier for the analysis of the implicit scheme, we
Abstract and Applied Analysis 7
obtain the following roundoff error equation from explicit Now, making use of the induction hypothesis, we arrive at the
scheme equation (1): following:
where 2 𝜎ℎ
𝛿𝑘+1 ≤ [𝐵1 + 1 − sin ( ) 𝑟𝑙 − 𝛿] 𝛿0 ≤ 𝛿0 .
𝑘
(59)
2
𝑑𝑗 = 𝑏𝑗+1 − 𝑏𝑗 . (50)
This completes the proof.
Again, we suppose that 𝛽𝑙𝑘 in (39) can be given in exponential
form as follows: Theorem 14. The explicit difference scheme of time-fractional
parabolic equation (1) is stable under the condition that, for all
𝛽𝑙𝑘 = 𝛿𝑘 𝑒(−𝑖𝜎𝑙𝑘) . (51)
(𝑙 = 0, 1, . . . , 𝑁 − 1; 𝑘 = 0, . . . , 𝑀 − 1), 𝑟𝑙𝑘 ≤ 1 − 𝛿 + 𝐵1 and
Then, replacing this form into (49), we arrive at the following 𝑟𝑙𝑘 ≥ 𝐵1 − 𝛿.
expression:
Proof. It is straightforward from (11) and Lemma 13, that,
𝜎ℎ
𝛿𝑘+1 = 𝛿𝑘 [𝐵1 + 1 − sin ( ) 𝑟𝑙𝑘 − 𝛿]
2
2
2 𝜁 ≤ 𝜁0 (60)
2 2
𝑘−1
(52)
− ∑ 𝛿𝑘−𝑗 𝑑𝑗 + 𝛿0 𝑑𝑘+1 , for 𝑘 = 0, . . . , 𝑀 − 1. and this proves that the explicit difference scheme for the
𝑗=1 time-fractional parabolic equation (3) is stable.
[4] A. Yakar and M. E. Koksal, “Existence results for solutions diffusion equation,” Journal of Computational Physics, vol. 213,
of nonlinear fractional differential equations,” Abstract and no. 1, pp. 205–213, 2006.
Applied Analysis, vol. 2012, Article ID 267108, 12 pages, 2012. [22] C. M. Chen, F. Liu, I. Turner, and V. Anh, “A Fourier method
[5] A. Atangana and J. F. Botha, “Generalized groundwater flow for the fractional diffusion equation describing sub-diffusion,”
equation using the concept of variable order derivative,” Bound- Journal of Computational Physics, vol. 227, no. 2, pp. 886–897,
ary Value Problems, vol. 2013, article 53, 2013. 2007.
[6] A. Atangana and S. C. Oukouomi Noutchie, “Stability and con- [23] S. B. Yuste and L. Acedo, “An explicit finite difference method
vergence of a tme-fractional variable order Hantush equation and a new von Neumann-type stability analysis for fractional
for a deformable aquifer,” Abstract and Applied Analysis, vol. diffusion equations,” SIAM Journal on Numerical Analysis, vol.
2013, Article ID 691060, 8 pages, 2013. 42, no. 5, pp. 1862–1874, 2005.
[7] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional [24] I. Podlubny, A. Chechkin, T. Skovranek, Y. Chen, and B. M.
Calculus Models and Numerical Methods Series on Complexity, Vinagre Jara, “Matrix approach to discrete fractional calculus
Nonlinearity and Chaos, World Scientific, 2012. II: partial fractional differential equations,” Journal of Compu-
[8] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory tational Physics, vol. 228, no. 8, pp. 3137–3153, 2009.
and Applications of Fractional Differential Equations, vol. 204 [25] E. Hanert, “On the numerical solution of space-time fractional
of North-Holland Mathematics Studies, Elsevier Science B.V., diffusion models,” Computers and Fluids, vol. 46, no. 1, pp. 33–
Amsterdam, The Netherlands, 2006. 39, 2011.
[9] K. B. Oldham and J. Spanier, The Fractional Calculus, vol. 111 of [26] P. Zhuang, F. Liu, V. Anh, and I. Turner, “Numerical methods for
Mathematics in Science and Engineering, Academic Press, New the variable-order fractional advection-diffusion equation with
York, NY, USA, 1974. a nonlinear source term,” SIAM Journal on Numerical Analysis,
[10] B. Ross, Ed., Fractional Calculus and Its Applications, vol. 457 of vol. 47, no. 3, pp. 1760–1781, 2009.
Lecture Notes in Mathematics, Springer, Berlin, Germany, 1975. [27] R. Lin, F. Liu, V. Anh, and I. Turner, “Stability and conver-
[11] Y. Luchko and R. Gorenflo, “The initial value problem for some gence of a new explicit finite-difference approximation for the
fractional differential equations with the Caputo derivative,” variable-order nonlinear fractional diffusion equation,” Applied
Preprint Series A08–98, Freic Universitat Berlin, 1998, Fachbre- Mathematics and Computation, vol. 212, no. 2, pp. 435–445,
ich Mathematik and Informatik. 2009.
[12] A. Atangana and E. Alabaraoye, “Solving system of fractional [28] J. Crank and P. Nicolson, “A practical method for numerical
partial differential equations arisen in the model of HIV evaluation of solutions of partial differential equations of the
infection of CD4+ cells and attractor one-dimensional Keller- heat conduction type,” Proceedings of the Cambridge Philosoph-
Segel equation,” Advances in Difference Equations, vol. 2013, ical Society, vol. 43, no. 1, pp. 50–67, 1947.
article 94, 2013. [29] K. Diethelm, N. J. Ford, and A. D. Freed, “Detailed error analysis
[13] I. Podlubny, Fractional Differential Equations, vol. 198 of Math- for a fractional Adams method,” Numerical Algorithms, vol. 36,
ematics in Science and Engineering, Academic Press, San Diego, no. 1, pp. 31–52, 2004.
Calif, USA, 1999. [30] C. P. Li and C. X. Tao, “On the fractional Adams method,”
[14] A. Atangana and A. Secer, “A Note on fractional order deriva- Computers and Mathematics with Applications, vol. 58, no. 8, pp.
tives and table of fractional derivatives of some special func- 1573–1588, 2009.
tions,” Abstract and Applied Analysis, vol. 2013, Article ID
279681, 8 pages, 2013.
[15] K. S. Miller and B. Ross, An Introduction to the Fractional
Calculus and Fractional Differential Equations, John Wiley &
Sons, New York, NY, USA, 1993.
[16] A. Atangana and A. Kilicman, “Analytical solutions of the space-
time fractional derivative of advection dispersion equation,”
Mathematical Problems in Engineering, vol. 2013, Article ID
853127, 9 pages, 2013.
[17] X.-J. Yang, D. Baleanu, and J. A. Tenreiro Machado, “Systems of
navier-stokes equations on cantor sets,” Mathematical Problems
in Engineering, vol. 2013, Article ID 769724, 8 pages, 2013.
[18] F. Mainardi, “Fractional calculus: some basic problems in
continuum and statistical mechanics,” in Fractals and Fractional
Calculus in Continuum Mechanics, vol. 378 of CISM Courses and
Lectures, pp. 291–348, Springer, Vienna, Austria, 1997.
[19] M. M. Meerschaert and C. Tadjeran, “Finite difference approx-
imations for fractional advection-dispersion flow equations,”
Journal of Computational and Applied Mathematics, vol. 172, no.
1, pp. 65–77, 2004.
[20] Y. Zhang, “A finite difference method for fractional partial
differential equation,” Applied Mathematics and Computation,
vol. 215, no. 2, pp. 524–529, 2009.
[21] C. Tadjeran, M. M. Meerschaert, and H. P. Scheffler, “A second-
order accurate numerical approximation for the fractional
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 176180, 8 pages
http://dx.doi.org/10.1155/2013/176180
Research Article
Existence Results for a Class of Fractional Differential Equations
with Periodic Boundary Value Conditions and with Delay
Copyright © 2013 Hadi Karami et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We discuss the existence and uniqueness of solution for two types of fractional order ordinary and delay differential equations.
Fixed point theorems are the main tool used here to establish the existence and uniqueness results. First we use Banach contraction
principle to prove the uniqueness of solution and then Krasnoselskii’s fixed point theorem to show the existence of the solution
under certain conditions in a Banach space.
associated with boundary conditions We denote 𝐷𝑎𝛼+ 𝑢(𝑡) as 𝐷𝑎𝛼 𝑦(𝑡) and 𝐷𝑎−𝛼+ 𝑢(𝑡) as 𝐷𝑎−𝛼 𝑢(𝑡).
Further 𝐷0𝛼+ 𝑢(𝑡) and 𝐷0−𝛼+ 𝑢(𝑡) are referred as 𝐷𝛼 𝑢(𝑡) and
𝑢 (0) = lim+ 𝑡1−𝛼 𝑢 (𝑡) = 𝑢 (1) , (3) 𝐷−𝛼 𝑢(𝑡), respectively.
𝑡→0
where 𝐷𝛼 is the standard Riemann-Liouville fractional Definition 2. A two-parameter function of the Mittag-Leffler
derivative and 𝑓 is a continuous function. Here 𝑢𝑡 (⋅) repre- type is defined by
sents the properitoneal state from time −𝜏 up to time 𝑡 which
is defined by 𝑢𝑡 (𝜃) = 𝑢(𝑡 + 𝜃), −𝜏 ≤ 𝜃 ≤ 0. We proved ∞
𝑧𝑘
the uniqueness of existence solutions for (2) with periodic 𝐸𝛼,𝛽 (𝑧) = ∑ , (𝛼 > 0, 𝛽 > 0) . (10)
𝑘=0
Γ (𝛼𝑘 + 𝛽)
boundary condition (3) under some further conditions.
For investigating to establish an existence theorem, we
also consider a class of nonlinear delayed fractional differen- Definition 3. The beta function is usually defined by
tial equations of the form
1
𝐷𝛼 𝑢 (𝑡) = 𝑢 (𝑡) + 𝑓 (𝑡, 𝑢 (𝑡) , 𝑢 (𝑡 − 𝜏)) , 𝑡 ∈ (0, 1) , 𝐵 (𝑧, 𝑤) = ∫ 𝜏𝑧−1 (1 − 𝜏)𝑤−1 𝑑𝜏,
(4) 0 (11)
𝑢 (𝑡) = 𝜙 (𝑡) , 𝑡 ∈ [−𝜏, 0] , 0 < 𝛼 < 1,
(Re (𝑧) > 0, Re (𝑤) > 0) .
with periodic boundary condition
and we have also the following expression for the beta
𝑢 (0) = lim+ 𝑡1−𝛼 𝑢 (𝑡) = 𝑐 = 𝑢 (1) , function:
𝑡→0
(5)
𝐷−(1−𝛼) 𝑢(𝑡)𝑡=0 = 𝑐Γ (𝛼) . Γ (𝑧) Γ (𝑤)
𝐵 (𝑧, 𝑤) = . (12)
Γ (𝑧 + 𝑤)
The paper has been organized as follows. In Section 2 we
give basic definitions and preliminary. Unique solution of Theorem 4 (Arzela-Ascoli’s theorem). A subset of 𝐶[𝑎, 𝑏] is
(2)-(3) under some conditions is proved in Section 3. The compact if and only if it is closed, bounded, and equicontinuous.
existence solution of (4)-(5) under some assumptions has
been presented in Section 4. Theorem 5 (Banach’s fixed point theorem). Consider a metric
space 𝑋 = (𝑋, 𝑑), where 𝑋 ≠0. Suppose that 𝑋 is complete and
2. Preliminaries 𝑇 : 𝑋 → 𝑋 is a contraction on 𝑋. Then 𝑇 has precisely one
fixed point.
For the convenience of the readers, we firstly present the
necessary definitions from the fractional calculus theory and Theorem 6 (Krasnoselskii’s fixed point theorem). Let 𝐵 be a
functional analysis. These definitions and results can be found nonempty closed convex subset of a Banach space (𝑋, ‖ ⋅ ‖).
in the literature [3, 7, 38]. Suppose that 𝑇1 , and 𝑇2 map 𝐵 into 𝑋 such that
Let 𝐶[0, 1] be the Banach space of all continuous real
functions defined on [0, 1] with the norm (1) for any 𝑥, 𝑦 ∈ 𝐵 we have 𝑇1 𝑥 + 𝑇2 𝑦 ∈ 𝐵,
‖𝑢‖ =: max {|𝑢 (𝑡)| : 𝑡 ∈ [0, 1]} . (6) (2) 𝑇1 is a contraction,
Let 𝐶𝑟 [0, 1], 𝑟 ≥ 0, be the space of all functions 𝑓 such that (3) 𝑇2 is continuous and 𝑇2 (𝐵) is contained in a compact
𝑡𝑟 𝑢(𝑡) ∈ 𝐶[0, 1] which is a Banach space when endowed with set.
the norm
Then there exists 𝑧 ∈ 𝐵 such that z = 𝑇1 𝑧 + 𝑇2 𝑧.
‖𝑢‖𝑟 =: max {𝑡𝑟 |𝑢 (𝑡)| : 𝑡 ∈ [0, 1]} . (7)
Definition 1. For a function 𝑢 defined on an interval [𝑎, 𝑏], the 3. Uniqueness of Solution
Riemann-Liouville fractional integral of 𝑢 of order 𝛼 > 0 is
defined by In this section we prove (2) with boundary condition (3) and
𝑡 another condition on 𝑓 has a unique solution. Before proving,
1
𝐷𝑎−𝛼+ 𝑢 (𝑡) = ∫ (𝑡 − 𝑠)𝛼−1 𝑢 (𝑠) 𝑑𝑠, 𝑡 > 𝑎, (8) we need to introduce some notations that will be provided in
Γ (𝛼) 𝑎 the following.
and Riemann-Liouville fractional derivative of 𝑢(𝑡) of order Let Ω = {𝑢 : [−𝜏, 1] → R, 𝑢 ∈ 𝐶1−𝛼 [0, 1]}. Consider the
𝛼 > 0 defined by operator 𝑁 : Ω → Ω defined by
𝑑𝑛 𝜙 (𝑡) , 𝑡 ∈ [−𝜏, 0] ,
𝐷𝑎𝛼+ 𝑢 (𝑡) = {𝐷𝑎−(𝛼−𝑛) 𝑢 (𝑡)} , 𝑛 − 1 < 𝛼 ≤ 𝑛, (9) {
𝑑𝑡𝑛
+
{
{
𝑁𝑢 (𝑡) = { 1 (13)
provided that the right-hand side of the pervious equation is {
{
pointwise defined on (𝑎, +∞). ∫ 𝐺 (𝑡, 𝑠) 𝑓 (𝑠, 𝑢𝑠 ) 𝑑𝑠, 𝑡 ∈ (0, 1) ,
{ 0 1,𝛼
Abstract and Applied Analysis 3
0, 𝑡 ∈ (0, 1) , 𝑇𝑧1 (𝑡) − 𝑇𝑧2 (𝑡)1−𝛼
𝑥 (𝑡) = { (15)
𝜙 (𝑡) , 𝑡 ∈ [−𝜏, 0] . 1
≤ 𝐾 max {𝑡1−𝛼 ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑠𝛼−1 𝑑𝑠} (22)
𝑡∈[0,1] 0
1−𝛼
For each 𝑧 ∈ 𝐶1−𝛼 ([0, 1], R) with lim𝑡 → 0+ 𝑡 𝑧(𝑡) = 𝑐 we
denote 𝑧 the function defined by × 𝑧1 − 𝑧2 1−𝛼 .
Moreover,
𝑧 (𝑡) , 𝑡 ∈ (0, 1) ,
𝑧 (𝑡) = { (16)
0, 𝑡 ∈ [−𝜏, 0] . 1
𝑡1−𝛼 ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑠𝛼−1 𝑑𝑠
0
If 𝑢(⋅) satisfies the integral equation, (23)
Γ (𝛼) 𝐸𝛼,𝛼 (1) Γ(𝛼)2
≤ ( + 1) 𝐸𝛼,𝛼 (1) Γ .
1 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) (2𝛼)
𝑢 (𝑡) = ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑓 (𝑠, 𝑢𝑠 ) 𝑑𝑠, (17)
0
Indeed we have
we can decompose 𝑢(⋅) as 𝑢(𝑡) = 𝑧(𝑡) + 𝑥(𝑡), 0 < 𝑡 < 1, which
implies 𝑢𝑡 = 𝑧𝑡 + 𝑥𝑡 for every 0 < 𝑡 < 1, and function 𝑧(⋅) 1
satisfies 𝑡1−𝛼 ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑠𝛼−1 𝑑𝑠
0
1 Γ (𝛼) 𝐸𝛼,𝛼 (𝑡𝛼 )
𝑧 (𝑡) = ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑓 (𝑠, 𝑧𝑠 + 𝑥𝑠 ) 𝑑𝑠. (18) ≤
0 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)
∞ 1
1
Set 𝐶0 = {𝑧 ∈ (𝐶1−𝛼 [0, 1], R) : lim𝑡 → 0+ 𝑡1−𝛼 𝑧(𝑡) = 𝑐}. 𝐶0 ×∑ ∫ (1 − 𝑠)𝛼−1 𝑠𝛼−1 𝑑𝑠
is Banach space with the norm ‖ ⋅ ‖𝑟 . Let 𝑇 : 𝐶0 → 𝐶0 be 𝑖=0 Γ (𝛼𝑖 + 𝛼) 0
defined by ∞
1
+ 𝑡𝛼−1 ∑ (24)
1 𝑖=0 Γ (𝛼𝑖 + 𝛼)
𝑇𝑧 (𝑡) = ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑓 (𝑠, 𝑧𝑠 + 𝑥𝑠 ) 𝑑𝑠, 0 < 𝑡 < 1. (19)
0 𝑡 Γ (𝛼) 𝐸𝛼,𝛼 (𝑡𝛼 )
× ∫ (𝑡 − 𝑠)𝛼−1 𝑠𝛼−1 𝑑𝑠
0 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)
Note that as operator 𝑁 has a fixed point, equivalently 𝑇 has
a fixed point and so instead we try to prove that 𝑇 has a fixed 1
𝛼
𝐷 𝑦 (𝑡) − 𝑦 (𝑡) = ℎ (𝑡) , (0 < 𝑡 < 1, 0 < 𝛼 < 1) , (31) This completes the proof.
Abstract and Applied Analysis 5
Now we prove our main result using Lemma 8 and two ≤ (𝑀1 ‖𝑢‖1−𝛼 + 𝑀2 ‖𝑢‖21−𝛼 )
more assumptions which follow next.
Γ (𝛼) 𝐸𝛼,𝛼 (𝑡𝛼 )
(H1) We assume that 𝑓(𝑡, 𝑢, V) can be written as 𝑓1 (𝑡, 𝑢) + × ( 𝐸𝛼,𝛼+1 (1)
1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)
𝑓2 (𝑡, 𝑢, V), where 𝑓1 , 𝑓2 are Lipschitz continuous.
Moreover assume that the function 𝑓1 , 𝑓2 satisfies the
following relations: + 𝑡𝐸𝛼,𝛼+1 (𝑡𝛼 ) ) ≤ 𝑟.
(45)
𝑓1 (𝑡, 𝑢 (𝑡)) ≤ 𝑀1 ‖𝑢 (𝑡)‖1−𝛼 ,
(41) (ii) We will prove that 𝑇1 is a contraction:
𝑓2 (𝑡, 𝑢 (𝑡) , V (𝑡)) ≤ 𝑀2 ‖𝑢 (𝑡)‖1−𝛼 ‖V(𝑡)‖1−𝛼 .
𝑇1 𝑢 − 𝑇1 V1−𝛼
(H2) Let Ω = 𝐶1−𝛼 ([−𝜏, 1], R) denote collection of the = 𝑡1−𝛼 𝑇1 𝑢 (𝑡) − 𝑇1 V (𝑡) ,
space of all function 𝑢 such that 𝑡1−𝛼 𝑢(𝑡) ∈ 𝐶[0, 1].
1
Define the set 𝐵 = {𝑢 ∈ Ω : 𝑡1−𝛼 |𝑢| ≤ 𝑟}, where 𝑟
≤ 𝑡1−𝛼 ∫ 𝐺1,𝛼 (𝑡, 𝑠)
satisfies 0
(𝑀1 ‖𝑢‖1−𝛼 + 𝑀2 ‖𝑢‖21−𝛼 ) × 𝑓1 (𝑠, 𝑢 (𝑠)) − 𝑓1 (𝑠, V (𝑠)) 𝑑𝑠,
1
Γ (𝛼) 𝐸𝛼,𝛼 (𝑡𝛼 ) ≤ 𝐿 𝑓1 𝑡1−𝛼 ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑢 − V 𝑑𝑠,
× ( 𝛼
𝐸𝛼,𝛼+1 (1) + 𝑡𝐸𝛼,𝛼+1 (𝑡 )) ≤ 𝑟.
(46)
1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)
0
1
(42)
≤ 𝐿 𝑓1 𝑡1−𝛼 ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑠𝛼−1 𝑠1−𝛼 𝑢 − V 𝑑𝑠
0
1
≤ 𝑡1−𝛼 ∫ 𝐺1,𝛼 (𝑡, 𝑠) × (𝑢𝑛 (𝑠) − 𝑢 (𝑠)
0
+ 𝑢𝑛 (𝑠 − 𝜏) − 𝑢 (𝑠 − 𝜏)) 𝑑𝑠
× 𝑓1 (𝑠, 𝑢 (𝑠))
1
+𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠 ≤ 2𝐿 𝑓2 𝑡1−𝛼 ∫ 𝐺1,𝛼 (𝑡, 𝑠) 𝑠𝛼−1
0
1
On the other hand for 0 ≤ 𝑡1 < 𝑡2 ≤ 1 and 𝑢 ∈ 𝐵 we have × ∫ 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) (1 − 𝑠)𝛼−1
𝑡2
× 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠
1−𝛼 Γ (𝛼) 𝐸𝛼,𝛼 (𝑡2𝛼 )
𝑡1 𝑇2 𝑢 (𝑡1 ) − 𝑡21−𝛼 𝑇2 𝑢 (𝑡2 )
−
1 − Γ (𝛼) 𝐸𝛼,𝛼 (1)
1
= 𝑡11−𝛼 ∫ 𝐺1,𝛼 (𝑡1 , 𝑠) 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠 1
0 × ∫ 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼−1 ) (1 − 𝑠)𝛼−1
𝑡2
1
− 𝑡21−𝛼 ∫ 𝐺1,𝛼 (𝑡2 , 𝑠) 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠 ×𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠 .
0
Γ (𝛼) 𝐸 (𝑡𝛼 ) (48)
≤
𝛼,𝛼 1
1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) Then we have
1−𝛼
𝑡1 𝑡1 𝑇2 𝑢 (𝑡1 ) − 𝑡21−𝛼 𝑇2 𝑢 (𝑡2 )
× ∫ 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) (1 − 𝑠)𝛼−1
0 ≤ 𝑀2 ‖𝑢‖21−𝛼
× 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠
Γ (𝛼) 𝛼
𝐸 (𝑡 ) − 𝐸𝛼,𝛼 (𝑡2 )
𝛼
×(
𝑡1
𝛼 𝛼−1 1 − Γ (𝛼) 𝐸𝛼,𝛼 (1) 𝛼,𝛼 1
+ 𝑡11−𝛼 ∫ 𝐸𝛼,𝛼 ((𝑡1 − 𝑠) ) (𝑡1 − 𝑠)
0 𝑡
1
× 𝑓2 (𝑠, 𝑢 (𝑠) , 𝑢 (𝑠 − 𝜏)) 𝑑𝑠 × ∫ 𝐸𝛼,𝛼 ((1 − 𝑠)𝛼 ) (1 − 𝑠)𝛼−1 𝑑𝑠
0
Then by Arzela-Ascoli’s theorem we conclude that 𝑇2 is [13] A. El-Sayed, “Fractional order differential equation,” Kyungpook
compact. By using Krasnoselskii’s theorem there exists 𝑧 ∈ 𝐵 Mathematical Journal, vol. 28, no. 2, pp. 18–22, 1988.
such that 𝑧 is a fixed point of 𝑇. This completes the proof. [14] A. El-Sayed, “On the fractional differential equations,” Applied
Mathematics and Computation, vol. 49, no. 2-3, pp. 205–213,
1992.
5. Conclusions
[15] A. El-Sayed, “Nonlinear functional-differential equations of
We considered two types of nonlinear delay fractional dif- arbitrary orders,” Nonlinear Analysis. Theory, Methods & Appli-
ferential equations (FDE) with periodic boundary conditions cations. An International Multidisciplinary Journal. Series A:
involving Remann-Liouville fractional derivative possessing Theory and Methods, vol. 33, no. 2, pp. 181–186, 1998.
with a lower terminal at 0. In order to obtain the results in [16] R. P. Agarwal, Y. Zhou, and Y. He, “Existence of fractional
this paper, we have shown the existence and the uniqueness of neutral functional differential equations,” Computers & Math-
solution for a class of nonlinear delayed FDE by Banach con- ematics with Applications, vol. 59, no. 3, pp. 1095–1100, 2010.
traction principle. Then using Krasnoselskii’s fixed point the- [17] V. Daftardar-Gejji and A. Babakhani, “Analysis of a system
orem we established an existence theorem for a different type of fractional differential equations,” Journal of Mathematical
Analysis and Applications, vol. 293, no. 2, pp. 511–522, 2004.
of the equation that we have proven its uniqueness theorem.
[18] Y. Liu, “Existence of solutions for impulsive differential models
on half lines involving Caputo fractional derivatives,” Commu-
Acknowledgment nications in Nonlinear Science and Numerical Simulation, vol. 18,
no. 10, pp. 2604–2625, 2013.
The authors would like thank to the anonymous referees for
[19] D. Delbosco and L. Rodino, “Existence and uniqueness for a
their valuable comments and suggestions. nonlinear fractional differential equation,” Journal of Mathe-
matical Analysis and Applications, vol. 204, no. 2, pp. 609–625,
References 1996.
[20] J. Deng and L. Ma, “Existence and uniqueness of solutions
[1] W. Boyce and R. Diprima, Elementary Differential Equations of initial value problems for nonlinear fractional differential
and Boundary Value Problems, John Wiley & Sons, New York, equations,” Applied Mathematics Letters, vol. 23, no. 6, pp. 676–
NY, USA, 1997. 680, 2010.
[2] F. Moon and M. Johnson, Dynamics and Chaos in Manufactur- [21] K. Diethelm and N. J. Ford, “Analysis of fractional differential
ing Processes, chapter 1, John Wiley & Sons, New York, NY, USA, equations,” Journal of Mathematical Analysis and Applications,
1998. vol. 265, no. 2, pp. 229–248, 2002.
[3] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory and [22] A. El-Sayed, A. El-Mesiry, and H. El-Saka, “On the fractional-
Application of Fractional Differential Equations, Elsevier Science order logistic equation,” Applied Mathematics Letters, vol. 20,
B.V., Amsterdam, The Netherlands, 2006. no. 7, pp. 817–823, 2007.
[4] S. Das, Functional Fractional Calculus for System Identification [23] K. M. Furati and N.-E. Tatar, “Long time behavior for a
and Controls, Springer, Berlin, Germany, 2008. nonlinear fractional model,” Journal of Mathematical Analysis
[5] R. Hilfer, Ed., Applications of Fractional Calculus in Physics, and Applications, vol. 332, no. 1, pp. 441–454, 2007.
World Scientific, River Edge, NJ, USA, 2000. [24] S. B. Hadid, “Local and global existence theorems on differential
[6] N. Heymans and I. Podlubny, “Physical interpretation of initial equations of non-integer order,” Journal of Fractional Calculus,
conditions for fractional differential equations with Riemann- vol. 7, pp. 101–105, 1995.
Liouville fractional derivatives,” Rheologica Acta, vol. 45, no. 5, [25] S. Hadid, B. Masaedeh, and S. Momani, “On the existence of
pp. 765–771, 2006. maximal and minimal solutions of differential equations of
[7] I. Podlubny, Fractional Differential Equations, Academic Press, non-integer order,” Journal of Fractional Calculus, vol. 9, pp. 41–
London, UK, 1999. 44, 1996.
[8] I. Podlubny, “Geometric and physical interpretation of frac- [26] R. W. Ibrahim and S. Momani, “On the existence and unique-
tional integration and fractional differentiation,” Fractional ness of solutions of a class of fractional differential equations,”
Calculus & Applied Analysis, vol. 5, no. 4, pp. 367–386, 2002, Journal of Mathematical Analysis and Applications, vol. 334, no.
Dedicated to the 60th anniversary of Prof. Francesco Mainardi. 1, pp. 1–10, 2007.
[9] I. Podlubny, I. Petráš, B. M. Vinagre, P. O’Leary, and L. Dorčák, [27] V. Lakshmikantham and A. S. Vatsala, “Basic theory of frac-
“Analogue realizations of fractional-order controllers. Frac- tional differential equations,” Nonlinear Analysis. Theory, Meth-
tional order calculus and its applications,” Nonlinear Dynamics, ods & Applications. An International Multidisciplinary Journal.
vol. 29, no. 1–4, pp. 281–296, 2002. Series A: Theory and Methods, vol. 69, no. 8, pp. 2677–2682,
[10] J. Sabatier, O. P. Agrawal, and J. A. T. Machado, Advances in 2008.
Fractional Calculus: Theoretical Developments and Applications [28] M. Benchohra, J. Henderson, S. K. Ntouyas, and A. Ouahab,
in Physics and Engineering, Springer, New York, NY, USA, 2007. “Existence results for fractional order functional differential
[11] J. T. Machado, V. Kiryakova, and F. Mainardi, “Recent history of equations with infinite delay,” Journal of Mathematical Analysis
fractional calculus,” Communications in Nonlinear Science and and Applications, vol. 338, no. 2, pp. 1340–1350, 2008.
Numerical Simulation, vol. 16, no. 3, pp. 1140–1153, 2011. [29] M. Belmekki, J. J. Nieto, and R. Rodrı́guez-López, “Existence
[12] J. T. Machado, A. M. Galhano, and J. J. Trujillo, “On devel- of periodic solution for a nonlinear fractional differential equa-
opment of fractional calculus during the last fifty years,” in tion,” Boundary Value Problems, vol. 2009, Article ID 324561, 18
Scientometrics, Springer, New York, NY, USA, 2013. pages, 2009.
8 Abstract and Applied Analysis
Research Article
Application of Fuzzy Fractional Kinetic Equations to
Modelling of the Acid Hydrolysis Reaction
Copyright © 2013 Ferial Ghaemi et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
In view of the usefulness and a great importance of the kinetic equation in specific chemical engineering problems, we discuss
the numerical solution of a simple fuzzy fractional kinetic equation applied for the hemicelluloses hydrolysis reaction. The fuzzy
approximate solution is derived based on the Legendre polynomials to the fuzzy fractional equation calculus. Moreover, the
complete error analysis is explained based on the application of fuzzy Caputo fractional derivative. The main advantage of the
present method is its superior accuracy which is obtained by using a limited number of Legendre polynomials. The method is
computationally interesting, and the numerical results demonstrate the effectiveness and validity of the method for solving fuzzy
fractional differential equations.
lignocellulosic substances. It is revealed that the diffusivity of matrix for fractional derivatives was introduced by [53] and
sulfuric acid in agricultural residues is extraordinary higher applied with spectral methods for numerical solution of
than in hard wood [7]. It is with this motivation to investigate multiterm linear and nonlinear fractional differential equa-
the effect of sulfuric acid concentration on both of hydrolysis tions subject to initial conditions which were developed by
of oil palm empty fruit bunch (OPEFB) fiber hydrolyzing and Kazem et al. [52] who applied fractional Legendre orthogonal
analyzing the appropriate kinetic model. The OPEFB biomass functions for solving these types of equations. Subsequently,
includes cellulose, hemicellulose, and lignin. In this study, to Doha et al. [55] introduced a new efficient Chebyshev spec-
obtain an in depth understanding to the hydrolysis reaction tral algorithms for solving linear and nonlinear multiterm
and an optimum reaction condition for the process, the fractional orders differential equations. Thereafter, Bhrawy et
reaction kinetics at a complete set of reaction conditions were al. [60] propose a method to approximate multiterm frac-
investigated. The respective rate constants were determined tional differential equations with variable coefficients using
as functions of temperature and sulfuric acid concentration a quadrature shifted Legendre tau approach. Consequently,
by a nonlinear regression analysis. this way has been followed by several authors [54, 57, 59, 61].
On the other hand, the study of fuzzy differential equa-
1.2. Fractional Kinetic Equation Model and Numerical Meth- tions is rapidly developing as a new field of fuzzy mathe-
ods. It is now well established that fractional kinetic equa- matics. The fuzzy differential equations have been studied by
tions represent an appropriate model to describe physical several authors [62–70]. Lately, Agarwal et al. [71] proposed
phenomena such as diffusion in porous media with fractal the concept of solutions for fractional differential equations
geometry, kinematics in viscoelastic media, relaxation pro- with uncertainty. They have considered Riemann-Liouville’s
cesses in complex systems (including viscoelastic materials, differentiability with a fuzzy initial condition to solve FFDEs.
glassy materials, synthetic polymers, and biopolymers), prop- Afterward, several authors have studied the existence and
agation of seismic waves, and the rate of change of chemical uniqueness of the solution of the fuzzy fractional differential
composition of a star [8–18]. equations (FFDEs) under different types of fuzzy fractional
These equations are obtained from the classical kinetic differentiability [72–75]; nevertheless, it is a little bit surpris-
equation by replacing the first- or second-order derivative by ing that few papers reported numerical methods for solving
fractional derivative [19–30]. In the nonstochastic situation, FFDEs [75–78].
fractional kinetic equations have been studied by Kochubeı̆ The scope of this paper is to derive an explicit formula
[31], Saichev and Zaslavsky [17], Zaslavsky [18], Haubold and for fuzzy fractional-order derivative of shifted fractional Leg-
Mathai [32], and Saxena et al. [33, 34].
endre polynomials of any degree in terms of shifted fractional
In the few years, the analytical and numerical meth-
Legendre polynomials themselves, in the fuzzy Caputo sense
ods for solving fractional differential equations (FDEs)
with order (0 < V ≤ 1). Also, we are concerned with the direct
have attracted much more consideration of mathematicians.
solution technique for solving the fuzzy fractional kinetic
Although some studies have been exploited to solve FDEs
equation (FFKE) which is extracted by using the kinetics data
analytically [35–38], most of them do not have an exact
of the acid hydrolysis reaction subject to nonhomogeneous
analytical solution. Therefore, numerical and approximation
initial conditions.
techniques have been used for solving these equations. Some
In this paper, the fuzzy fractional derivative of the
of the most common methods are homotopy perturbation
proposed FFKE is approximated based on the shifted frac-
method [39], Haar wavelet method [40], Spline collocation
tional Legendre polynomials presented in [52], and then the
method [41], fractional difference method (FDM) [42], power
operational matrix of the fractional Caputo derivative of
series method [43], Adomian decomposition method (ADM)
[44], Kronecker convolution product [45], spectral methods order (0 < V ≤ 1) is specified, and we apply the shifted
[46], He’s variational iteration method [47], and homotopy fractional Legendre spectral tau (SCT) method to construct
analysis method [48]. the spectral solution for such problem. To the best of the
It is somewhat extraordinary that recently the orthogonal authors knowledge, such approach has not been employed for
functions received remarkable utilizing for the fractional- solving linear fractional kinetic equations under uncertainty.
order differential equations [49, 50]. Much efforts have been The paper is organized as follows. In Section 2, we
made to develop accurate algorithms using tau and collo- introduce some necessary definitions of fuzzy sets, fractional
cation based on operational matrices of some orthogonal calculus theory, kinetic equations, relevant properties of Leg-
polynomials such as block pulse functions [51], Legendre endre polynomials, and some of the main properties of fuzzy
polynomials [52, 53], Chebyshev polynomials [54–56], Jacobi fractional derivatives. Section 3 is devoted to presentation
polynomials [57, 58], and Laguerre polynomials [59]. The of the governing fractional kinetic equation. In Section 4,
main characteristic behind the approach using this technique the proposed method is explained for numerical solution
is that it reduces FDEs to those of solving a system of algebraic of the FFDEs. In Section 5, the derived FFKE is solved
equations thus notably simplifying the problem. based on the different values of constant coefficients in the
Furthermore, spectral methods have been found very equation by applying the presented technique, and the error
robust tools for solving many types of fractional differential of the approximate solution is depicted to demonstrate the
equations which have inspired many authors to apply them effectiveness of the method. Finally, some conclusions are
for these kinds of equations. The shifted Legendre operational drawn.
Abstract and Applied Analysis 3
Definition 1. Let 𝑢 be a fuzzy set in R. 𝑢 is called a fuzzy Definition 4 (see [84]). Let 𝑓 and 𝑔 be the two fuzzy-number-
number if valued functions on the interval [𝑎, 𝑏], that is, 𝑓, 𝑔 : [𝑎, 𝑏] →
RF . The uniform distance between fuzzy-number-valued
(i) 𝑢 is normal: there exists 𝑥0 ∈ R such that 𝑢(𝑥0 ) = 1, functions is defined by
(ii) 𝑢 is convex: for all 𝑥, 𝑦 ∈ R and 0 ≤ 𝜆 ≤ 1, it holds
that 𝐷∗ (𝑓, 𝑔) := sup 𝐷 (𝑓 (𝑥) , 𝑔 (𝑥)) . (9)
𝑥∈[𝑎,𝑏]
𝑢 (𝜆𝑥 + (1 − 𝜆) 𝑦) ≥ min {𝑢 (𝑥) , 𝑢 (𝑦)} , (1)
Remark 5 (see [83]). Let 𝑓 : [𝑎, 𝑏] → RF be fuzzy contin-
(iii) 𝑢 is upper semicontinuous: for any 𝑥0 ∈ R, it holds uous. Then, from property (iv) of Hausdorff distance, we can
that define
𝑢 (𝑥0 ) ≥ lim± 𝑢 (𝑥) , (2) 𝐷 (𝑓 (𝑥) , 0̃) = sup max {𝑓1𝑟 (𝑥) , 𝑓2𝑟 (𝑥)} , ∀𝑥 ∈ [𝑎, 𝑏] .
𝑥 → 𝑥0 𝑟∈[0,1]
(10)
(iv) [𝑢]0 = supp(𝑢) is a compact subset of R.
Definition 6 (see [85]). Let 𝑥, 𝑦 ∈ RF . If there exists 𝑧 ∈ RF
In this paper, the set of all fuzzy numbers is denoted by such that 𝑥 = 𝑦 ⊕ 𝑧, then 𝑧 is called the H-difference of 𝑥 and
RF . 𝑦, and it is denoted by 𝑥 ⊖ 𝑦.
Definition 2. Let 𝑢 ∈ RF and 𝑟 ∈ [0, 1]. The 𝑟-cut of 𝑢 is In this paper, the sign “⊖” always stands for H-difference,
the crisp set [𝑢]𝑟 that contains all elements with membership and note that 𝑥 ⊕ 𝑦 ≠𝑥 + (−𝑦). Also, throughout the
degree in 𝑢 greater than or equal to 𝑟, that is, paper, the Hukuhara-diference and generalized Hukuhara-
differentiability are assumed to be existed.
[𝑢]𝑟 = {𝑥 ∈ R | 𝑢 (𝑥) ≥ 𝑟} . (3)
For a fuzzy number 𝑢, its 𝑟-cuts are closed intervals in R, and Theorem 7 (see [69]). Let 𝐹 : (𝑎, 𝑏) → RF be a function,
we denote them by and denote [𝐹(𝑡)]𝑟 = [𝑓𝑟 (𝑡), 𝑔𝑟 (𝑡)], for each 𝑟 ∈ [0, 1]. Then,
[𝑢]𝑟 = [𝑢1𝑟 , 𝑢2𝑟 ] . (4) (1) if 𝐹 is (1)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are
differentiable functions and
According to Zadeh’s extension principle, the operation 𝑟
of addition on RF is defined as follows: [𝐹 (𝑡)] = [𝑓𝑟 (𝑡) , 𝑔𝑟 (𝑡)] , (11)
(𝑢 + V) (𝑥) = sup min {𝑢 (𝑦) , V (𝑥 − 𝑦)} , 𝑥 ∈ R,
𝑦∈R
(5) (2) if 𝐹 is (2)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are
differentiable functions and
and scalar multiplication of a fuzzy number is given by
𝑟
𝑥 [𝐹 (𝑡)] = [𝑔𝑟 (𝑡) , 𝑓𝑟 (𝑡)] . (12)
{𝑢 ( ) , 𝑘 > 0,
(𝑘 ⊙ 𝑢) (𝑥) = { 𝑘 (6)
̃ Definition 8 (see [86]). Consider the 𝑛 × 𝑛 linear system of
{0, 𝑘 = 0,
equations
where 0̃ ∈ RF .
𝑎11 𝑥1 + 𝑎12 𝑥2 + ⋅ ⋅ ⋅ + 𝑎1𝑛 𝑥𝑛 = 𝑦1 ,
Definition 3 (see [80]). The distance 𝐷(𝑢, V) between two 𝑎21 𝑥1 + 𝑎22 𝑥2 + ⋅ ⋅ ⋅ + 𝑎2𝑛 𝑥𝑛 = 𝑦2 ,
fuzzy numbers 𝑢 and V is defined as
(13)
..
𝐷 (𝑢, V) = sup 𝑑H ([𝑢]𝑟 , [V]𝑟 ) , (7) .
𝑟∈[0,1]
𝑎𝑛1 𝑥1 + 𝑎𝑛2 𝑥2 + ⋅ ⋅ ⋅ + 𝑎𝑛𝑛 𝑥𝑛 = 𝑦𝑛 .
where
𝑑H ([𝑢]𝑟 , [V]𝑟 ) = max {𝑢1𝑟 − V1𝑟 , 𝑢2𝑟 − V2𝑟 } (8) The matrix form of the earlier equations is
where the coefficient matrix 𝐴 = (𝑎𝑖𝑗 ), 1 ≤ 𝑖, 𝑗 ≤ 𝑛 is a crisp 2.2. Fractional Legendre Polynomials. The shifted Legendre
𝑛 × 𝑛 matrix and 𝑦𝑖 ∈ RF , 1 ≤ 𝑖 ≤ 𝑛. This system is called a polynomials are generated from the three-term recurrence
fuzzy linear system (FLS). relation:
A function 𝑓(𝑥), square integrable in (0, 1], may be expressed The production and destruction of species is described by
in terms of FLFs as kinetic equations governing the change of the number density
𝑁𝑖 of species 𝑖 over time [32–34], that is,
∞
𝑓 (𝑥) = ∑𝑎𝑖 FL𝛼𝑖 (𝑥) , (29) 𝑑
𝑁 (𝑡) = 𝑎𝑁𝑖 (𝑡) , (34)
𝑖=0
𝑑𝑡 𝑖
where the coefficients 𝑎𝑗 are given by where 𝑎 is a constant. The destruction rate of the particles of
type 𝑖 is given by
1
𝑎𝑗 = 𝛼 (2𝑖 + 1) ∫ 𝑓 (𝑥) FL𝛼𝑖 (𝑥) 𝑤 (𝑥) 𝑑𝑥, 𝑑
𝑁 (𝑡) = −𝑏𝑁𝑖 (𝑡) , (35)
0 (30) 𝑑𝑡 𝑖
𝑗 = 0, 1, 2, . . . . where 𝑏 is a constant. Then, the residual effect can be taken as
𝑚−1
where 𝑐 = 𝑏 − 𝑎. The solution of (36) can be seen as a simple
𝑓 (𝑥) ≃ 𝑓𝑚 (𝑥) = ∑ 𝑎𝑖 FL𝛼𝑖 𝑇
(𝑥) = 𝐴 Φ (𝑥) , (31) function
𝑖=0
𝑁𝑖 (𝑡) = 𝑁0 𝑒−𝑐𝑡 . (37)
with A fractional production-destruction equation can be
𝑇
obtained from a standard production-destruction equation
𝐴 = [𝑎0 , 𝑎1 , . . . , 𝑎𝑚−1 ] , by considering a fractional integral in place of a classical
(32) integral. A fractional production-destruction equation can
𝑇
Φ (𝑥) = [FL𝛼0 (𝑥) , FL𝛼1 (𝑥) , . . . , FL𝛼𝑚−1 (𝑥)] . be obtained, after dropping 𝑖, as
Theorem 12 (see [52]). Let the function 𝐷𝑘𝛼 𝑓(𝑥) ∈ 𝐶(0, 1] for 𝑁 (𝑡) − 𝑁0 = −𝑐V 0 𝐷𝑡−V 𝑁 (𝑡) , V > 0, (38)
𝑘 = 0, 1, . . . , 𝑚, (2𝑚 + 1) ≤ 𝛼 and P𝑚 = Span{FL𝛼𝑖 (𝑥)}𝑚−1
𝑖=0 . If where 0 𝐷𝑡−V represents the Riemann-Liouville fractional inte-
𝑓𝑚 = 𝐴𝑇 Φ(𝑥) is the best approximation to 𝑓 from P𝑚 , then the gral.
error bound is presented as follows: It is worth to note here the alternative approach of
Mainardi [93], for solving the fractional relaxation-oscillation
𝑀𝛼 1 equation, that is, the fractional production-destruction equa-
𝑓 (𝑥) − 𝑓𝑚 (𝑥)𝑤 ≤ √ , (33) tion with the Caputo fractional-order derivative.
Γ (𝑚𝛼 + 1) (2𝑚 + 1) 𝛼
As it can be seen, the fractional kinetic equation is
obtained by replacing the standard integral with the frac-
where 𝑀𝛼 ≥ |𝐷𝑚𝛼 𝑓(𝑥)|, 𝑥 ∈ (0, 1].
tional Riemann-Liouville integral. In the present research,
The later theorem proved that the approximate function firstly, we replace the Caputo fractional-order derivative with
based on the fractional Legendre polynomials converges to the classical derivative in the kinetic equation, and then the
the function 𝑓. derived fractional kinetic equation in the sense of the fuzzy
setting is solved by the proposed technique.
2.3. Fractional Kinetic Equations. Chemical kinetics as a
2.4. Fuzzy Fractional Differentiability. In the fractional lit-
science began in the middle of the 19th century, when
erature, Mittag-Leffler function plays an important role in
Wilhelmy [88] was apparently the first to recognize that the
the theory of fractional calculus/fuzzy fractional calculus and
rate at which a chemical reaction proceeds follows definite
fractional differential equations/fuzzy fractional differential
laws, and although his work paved the way for the law of
equations which is defined by [19, 30, 94]:
mass action of Waage and Guldberg [89], it attracted little
attention until it was taken up by Ostwald towards the end of ∞
𝑧𝑘
the century, as discussed by Laidler [90]. Wilhelmy realized 𝐸𝛼,𝛽 (𝑧) = ∑ , 𝛼 > 0, 𝛽 > 0. (39)
that chemical rates depended on the concentrations of the 𝑘=0
Γ (𝛼𝑘 + 𝛽)
reactants [91]. Also, it is useful to review some basic definitions and
Here, we provide a brief definition to illustrate the theorems related to the FFDEs. For more details see [72, 74,
chemical kinetic process. One can find more details in [91, 75]. Firstly, we present some notations which are used later in
92]. the paper.
Definition 13. Chemical kinetics is the study of the rate at (i) 𝐿R𝑝F (𝑎, 𝑏), 1 ≤ 𝑝 ≤ ∞ is the set of all fuzzy-valued
which a chemical process occurs. Besides information about measurable functions 𝑓 on [𝑎, 𝑏], where ‖𝑓‖𝑝 =
the speed at which reactions occur, kinetics also sheds light on 1 1/𝑝
the reaction mechanism (exactly how the reaction occurs). (∫0 (𝑑(𝑓(𝑡), 0))𝑝 𝑑𝑡) .
6 Abstract and Applied Analysis
(ii) 𝐶RF [𝑎, 𝑏] is a space of fuzzy-valued functions which For the sake of simplicity, we say that the fuzzy-valued
are continuous on [𝑎, 𝑏]. function 𝑓 𝑐 [(1) − V] is differentiable if it is differentiable as
(iii) 𝐶𝑛RF [𝑎, 𝑏] indicates the set of all fuzzy-valued func- in Definition 16 case (i), and 𝑓 is 𝑐 [(2) − V] differentiable if it
tions which are continuous up to order 𝑛. is differentiable as in Definition 16 case (ii), and so on for the
other cases.
(iv) 𝐴𝐶RF [𝑎, 𝑏] denotes the set of all fuzzy-valued func-
tions which are absolutely continuous. Theorem 17 (see [74]). Let 0 < V ≤ 1 and 𝑓 ∈ 𝐴𝐶RF [𝑎, 𝑏];
then the fuzzy Caputo fractional derivative exists almost
Note that one can find easily these notations in the crisp everywhere on (𝑎, 𝑏), and for all 0 ≤ 𝑟 ≤ 1, one has
context in [30] and references therein.
V
Definition 14 (see [74]). Let 𝑓 ∈ 𝐶RF [𝑎, 𝑏] ∩ 𝐿RF [𝑎, 𝑏]. (𝑐 𝐷𝑎+ 𝑓) (𝑥; 𝑟)
The Riemann-Liouville integral of fuzzy-valued function 𝑓 is 𝑥 𝑓 (𝑡) 𝑑𝑡𝑟 𝑥 𝑓 (𝑡) 𝑑𝑡 𝑟
defined as 1 − 1 +
=[ ∫ V , ∫ ]
𝑥 𝑓
Γ (1 − V) 𝑎 (𝑥 − 𝑡) Γ (1 − V) 𝑎 (𝑥 − 𝑡)V
V 1 (𝑡) 𝑑𝑡
(RL 𝐼𝑎+ 𝑓) (𝑥) = ∫ , 𝑥 > 𝑎, 0 < V ≤ 1. 1−V
Γ (V) 𝑎 (𝑥 − 𝑡)1−V = [(𝐼𝑎+ 𝐷𝑓−𝑟 ) (𝑥) , (𝐼𝑎+
1−V
𝐷𝑓+𝑟 ) (𝑥)] ,
(40) (46)
Definition 15 (see [74]). Let 𝑓 ∈ 𝐶RF [𝑎, 𝑏] ∩ 𝐿RF [𝑎, 𝑏]. Then when 𝑓 is (1) differentiable, and
𝑓 is said to be Caputo’s H-differentiable at 𝑥 when
V
𝑐 V RL 𝛽 (𝑐 𝐷𝑎+ 𝑓) (𝑥; 𝑟)
(i) ( 𝐷𝑎+ 𝑓) (𝑥) =( 𝐷𝑎+ [𝑓 (𝑡) ⊖ 𝑓 (𝑎)]) (𝑥) ,
𝑟
𝑥 𝑓 (𝑡) 𝑑𝑡 𝑥 𝑓 (𝑡) 𝑑𝑡 𝑟
(41) 1 + 1 −
𝑐 V RL 𝛽 =[ ∫ , ∫ ]
(ii) ( 𝐷𝑎+ 𝑓) (𝑥) =( 𝐷𝑎+ [−𝑓 (𝑎) ⊖ (−𝑓 (𝑡))]) (𝑥) . Γ (1 − V) 𝑎 (𝑥 − 𝑡)V Γ (1 − V) 𝑎 (𝑥 − 𝑡)V
1−V
Definition 16 (see [74]). Let 𝑓 : 𝐿RF [𝑎,𝑏] ∩ 𝐶RF [𝑎, 𝑏] and 𝑥0 ∈ = [(𝐼𝑎+ 𝐷𝑓+𝑟 ) (𝑥) , (𝐼𝑎+
1−V
𝐷𝑓−𝑟 ) (𝑥)] ,
𝑥
(𝑎, 𝑏) and Φ(𝑥) = (1/Γ(1 − V)) ∫𝑎 (𝑓(𝑡)/(𝑥 − 𝑡)V )𝑑𝑡. We say that (47)
𝑓(𝑥) is fuzzy Caputo fractional differentiable of order 0 < V ≤
1 at 𝑥0 , if there exists an element (𝑐 𝐷𝑎+
V
𝑓)(𝑥0 ) ∈ 𝐶𝐸 [𝑎, 𝑏] such when 𝑓 is (2) differentiable.
that for all 0 ≤ 𝑟 ≤ 1, ℎ > 0,
Xylose (g/L)
5
4
model that will be utilized to analyze this experimental data 3
is revealed by the equation as follows:
2
xylan
→ xylose
→ Decomposed products 1
(49) 0
𝑘1 𝑘2
𝐴 → 𝐵 → 𝐶, 0 50 100 150 200
Time (min)
where 𝐴 = xylan; 𝐵 = xylose; 𝐶 = Decomposition products; Xcal
𝑘1 = sugar release rate; 𝑘2 = sugar decomposition rate. Xexp
Material balance for components “𝐴” and “𝐵” for the first-
order kinetics gives Figure 1: Comparison between calculated and experimental value
of xylose in hydrolysis of OPEFB with 2% H2 SO4 at 120∘ C [95].
𝑑𝐶𝐴 (𝑡)
− = 𝑘1 𝐶𝐴 (𝑡) (50)
𝑑𝑡 9
Figure 3: Comparison between calculated and experimental value The problem is addressed to the best shifted fractional
of xylose in hydrolysis of OPEFB with 6% H2 SO4 at 120∘ C [95]. Legendre approximation, as we use fractional Legendre’s
nodes.
in which FL𝛼𝑖 (𝑥) is as the same in (26), and ∑∗ means addition where 𝑀V ≥ sup𝑟∈[0,1] max𝑥∈[𝑥0 ,1] {|𝐷V 𝑦1𝑟 (𝑥)|, |𝐷V 𝑦2𝑟 (𝑥)|} and
with respect to ⊕ in RF . 𝑦𝑟 (𝑥) = [𝑦1𝑟 (𝑥), 𝑦2𝑟 (𝑥)].
Remark 19. Practically, only the first (𝑚)-terms shifted frac- Proof. Considering Definition 4 for which we know that
tional Legendre polynomials are considered. So, we have
𝐷∗ (𝑦 (𝑥) , 𝑦𝑚 (𝑥))
𝑚−1
𝑦 (𝑥) ≃ 𝑦𝑚 (𝑥) = ∑ ∗ 𝑎𝑗 ⊙ FL𝛼𝑖 (𝑥) = 𝐴𝑇𝑚 ⊙ Φ𝑚 (𝑥) ; (58) = sup 𝐷 (𝑦 (𝑥) , 𝑦𝑚 (𝑥))
𝑥∈[𝑥0 ,1]
𝑗=0
= sup sup max {𝑦1𝑟 (𝑥) − 𝑦𝑚,1
𝑟
(𝑥) , 𝑦2𝑟 (𝑥) − 𝑦𝑚,2
𝑟
(𝑥)}
hence 𝑥∈[𝑥 ,1] 𝑟∈[0,1]
0
𝑚−1 = sup max {𝑦1𝑟 (𝑥) − 𝑦𝑚,1
𝑟
(𝑥)𝑤 , 𝑦2𝑟 (𝑥) − 𝑦𝑚,2
𝑟
(𝑥)𝑤 } .
𝑟 𝑟 ∗ 𝑟
𝑦 (𝑥) ≃ 𝑦𝑚 (𝑥) = ∑ 𝑎𝑗 ⊙ FL𝛼𝑖 (𝑥) , (59) 𝑟∈[0,1]
𝑗=0 (63)
Abstract and Applied Analysis 9
𝐷∗ (𝑦 (𝑥) , 𝑦𝑚 (𝑥))
V
𝐷 𝑦1 (𝑥) 1
≤ sup max { √ ,
Absolute error
𝑟∈[0,1] Γ (𝑚V + 1) (2𝑚 + 1) 𝛼
V 10−11
𝐷 𝑦2 (𝑥) 1 (64)
√ }
Γ (𝑚V + 1) (2𝑚 + 1) 𝛼
𝑀V 1
≤ √ ,
Γ (𝑚V + 1) (2𝑚 + 1) V
10−12
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
in which 𝑀 ≥ sup𝑟∈[0,1] max𝑥∈[𝑥0 ,1] {|𝐷V 𝑦1𝑟 (𝑥)|, |𝐷V 𝑦2𝑟 (𝑥)|}
V
r-cuts
and 𝑦𝑟 (𝑥) = [𝑦1𝑟 (𝑥), 𝑦2𝑟 (𝑥)].
= 𝛼 = 0.5 = 𝛼 = 0.85
= 𝛼 = 0.75 = 𝛼 = 0.95
4.2. Operational Matrix of FLFs
Figure 4: The absolute error for different values V = 𝛼 with 𝑚 = 12,
Lemma 23. The fuzzy Caputo fractional derivative of order 𝑘1 = 0.012, and 𝑘2 = 0.001.
0 < V ≤ 1 over the shifted fractional legendre functions can
be described in the form of
10−8
𝑖
𝑐 Γ (𝑘𝛼 + 1) 𝑘𝛼−V
𝐷V FL𝛼𝑖 (𝑥) = ∑ 𝑏𝑘,𝑖
𝑥 , (65)
𝑘=0
Γ (𝑘𝛼 − V + 1)
where 𝑏𝑘,𝑖 = 0 if 𝑏𝑘,𝑖 = (Γ(𝑘𝛼 + 1)/Γ(𝑘𝛼 − V + 1))𝑏𝑘𝑗 . 10−9
Absolute error
where 𝑐 𝐷(V) is the 𝑚×𝑚 operational matrix of fuzzy fractional in which 𝜃𝑖,𝑗,𝑘 are acquired by
V
Caputo’s derivative of FLFs and 𝑐 𝐷 Φ(𝑥) ∈ 𝐶RF [𝑎, 𝑏]. So,
using (67) and Lemma 23, we extend the operational matrix 𝜃𝑖,𝑗,𝑘
of Caputo fractional-order derivative of FLFs in the sense of
the fuzzy setting as follows. = 𝛼 (2𝑗 + 1)
𝑗
Theorem 24. Let one assume that Φ(𝑥) is the FLF vector. 𝐷𝑖,𝑗 (V) (−1)𝑖+𝑗+𝑘+𝑙 (𝑖 + 𝑘)! (𝑙 + 𝑗)!Γ (𝑘𝛼 + 1)
×∑ ,
is the 𝑚-square operational matrix of fuzzy fractional Caputo’s 𝑙=0 (𝑖 − 𝑘)!(𝑘!)2 Γ(𝑘𝛼−V+1)(𝑗 − 𝑙)!(𝑙!)2 (𝑘𝛼 + 𝑙𝛼 + 𝛼 − V)
(V)
derivative of order 0 < V ≤ 1. Then, the elements of 𝐷𝑖,𝑗 are
𝑗 = 0, 1, . . . , 𝑚 − 1.
achieved as
(69)
𝑖
(V)
𝐷𝑖,𝑗 = ∑ 𝜃𝑖,𝑗,𝑘 , 𝑖 = 0, 1, . . . , 𝑚 − 1, (68)
𝑘=⌈𝛼⌉
Note that in 𝐷(V) , the first ⌈V⌉ rows are all zero.
10 Abstract and Applied Analysis
𝛼 = 0.95 V
𝛼 = 0.5 (𝑐 𝐷0+ 𝐶𝐵 ) (𝑥) = 𝐶𝐵 (𝑥) ⊕ 𝑔 (𝑥) , 0 < V ≤ 1,
𝛼 = 0.75 𝛼=1 (71)
Figure 6: The absolute error for different values 𝛼 with V = 0.75,
𝐶𝐵 (0) = 𝐶𝐵0 ∈ RF ,
𝑘1 = 0.012, 𝑚 = 7, and 𝑘2 = 0.001.
where 𝐶𝐵 (𝑥) : 𝐿RF ∩ 𝐶RF is a continuous fuzzy-valued func-
V
tion, 𝑐 𝐷0+ (𝑥) denotes the fuzzy Caputo fractional derivative
Proof. If we consider the fractional derivative in the crisp of order V, and 𝑔(𝑥) : [0, 1] → RF .
context, then the proof of this theorem is obtained from Firstly, we state the unknown fuzzy functions 𝐶𝐵 (𝑥),
Theorem 2 in [52]. In the same way, if we consider the (𝑐 𝐷0V+ 𝐶𝐵 )(𝑥) and known fuzzy function 𝑔(𝑥) in terms of the
fuzzy Caputo fractional derivative, then analogously to the FLFs as follows:
demonstration of Theorem 1 in [52] and by using Lemma 23,
𝑚−1
we can prove the relation (69). 𝑟
𝐶𝐵𝑟 (𝑥) ≃ 𝐶𝐵 𝑟𝑚 (𝑥) = ∑ ∗ 𝑎𝑗 ⊙ FL𝛼𝑗 (𝑥) = 𝐴𝑇𝑚 ⊙ Φ𝑟𝑚 (𝑥) ,
Also, we can exhibit the operational matrix 𝐷(V) in the 𝑗=0
Remark 25. If V = 1, then Theorem 24 gives the operational (V) (𝑟) (𝑟)
matrix for the shifted Legendre polynomials function as [(𝑐 𝐷 𝐶𝐵𝑚 ) (𝑥)] = [𝑅𝑚 (𝑥) ⊕ 𝑔𝑚 (𝑥) ⊕ 𝐶𝐵𝑚 (𝑥)] ,
described in Section 2.2.
𝑟 ∈ [0, 1] .
(75)
4.3. The Application of the FLFs for Solving Linear FFKE. In
this section, we are concerned with providing a numerical
solution to linear FFKE which was described in Section 3. We Let ⟨⋅, ⋅⟩RF denote the fuzzy inner product over 𝑋RF =
approximate the fuzzy function 𝑦(𝑥) by means of the FLFs 𝐿2RF ([0, 1]).As in a typical tau method (see [100, 101]), we
and then substitute in the FFKE to derive the approximate make (𝑚 − 1) linear fuzzy equations from the following inner
solution. The method reduces the fuzzy fractional initial product as
Abstract and Applied Analysis 11
= 𝛼 = 0.5 = 𝛼 = 0.75
Approximate solution
Approximate solution
2 2
1 1
0 0
1 1
1 1
0.5 0.5
r-cu 0.5 r-cu 0.5
ts ts
0 0 x 0 0 x
= 𝛼 = 0.85 = 𝛼 = 0.95
Approximate solution
2 2
Approximate solution
1 1
0 0
1 1
1 1
0.5 0.5 0.5
r-cu r-cu 0.5
ts 0 0 x ts 0 0 x
Figure 7: The fuzzy approximate solution for different values of fractional orders V = 𝛼 with 𝑚 = 8, 𝑘1 = 0.012, and 𝑘2 = 0.001.
10−9 10−8
10−10
10−11 10−9
Absolute error
Absolute error
10−12
10−13 10−10
10−14
10−15 10−11
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts r-cuts
Figure 8: The absolute error for different values V = 𝛼 with 𝑚 = 8, Figure 9: The absolute error for different values 𝑚 with V = 𝛼 =
𝑘1 = 0.014, and 𝑘2 = 0.001. 0.85, 𝑘1 = 0.014, and 𝑘2 = 0.001.
⟨𝑅𝑚 (𝑥; 𝑟) , FL𝛼𝑖 (𝑥; 𝑟)⟩RF = 0̃, 𝑖 = 0, 1, . . . , 𝑚 − 2, 𝑟 ∈ [0, 1] , From (76), we make (𝑚 − 1) fuzzy linear algebraic
(76) equations which are as follows in the expanded form:
1 𝑚−2
where ⟨𝑅𝑚 (𝑥; 𝑟), FL𝛼𝑖 (𝑥; 𝑟)⟩RF = [(FR) ∫0 𝑅𝑚 (𝑥) ⊙ (𝑥), (𝑟)
(𝑟)
∑ ∗ 𝑎𝑗 ⊙ {⟨𝐷(V) FL𝛼𝑗 (𝑥) , FL𝛼𝑖 (𝑥)⟩ − ⟨FL𝛼𝑗 (𝑥) , FL𝛼𝑖 (𝑥)⟩}
FL𝛼𝑖 (𝑥) ⊙ 𝑤(𝑥)𝑑𝑥] . 𝑗=0
12 Abstract and Applied Analysis
10−5 the exact solution reveals that our method is efficient and
convenient.
Now, we consider (54) with the fuzzy initial condition.
10−6
It means that the initial concentration of xylose is a fuzzy
number. So, we have
Absolute error
10−7
Example 27. Consider the following FFDE:
10−8 𝑐 V
𝐷0+ 𝐶𝐵 (𝑡) ⊕ 𝑘2 𝐶𝐵 (𝑡) = 𝑘1 𝐶𝐴 0 exp (−𝑘1 𝑡) , 0 < V ≤ 1,
0 < 𝑟 ≤ 1.
5. Numerical Results (83)
In order to illustrate the effectiveness and accuracy of the
proposed method, we carry it out for solving the fractional Employing the proposed method described in Section 4,
kinetic equation which was derived in Section 3 in the sense we can derive the operational matrix of the FLFs based on
of the fuzzy concept. Comparison of the results obtained with the fuzzy fractional derivative of the Caputo type and the
Abstract and Applied Analysis 13
= 𝛼 = 0.5 = 𝛼 = 0.75
Approximate solution
Approximate solution 2 2
1 1
0 0
1 1
1 1
0.5 0.5
r-cu 0.5 r-cu 0.5
ts x ts
0 0 0 0 x
= 𝛼 = 0.85 = 𝛼 = 0.95
Approximate solution
Approximate solution
2 2
1 1
0 0
1 1
1 1
0.5 0.5
r-cu 0.5 r-cu 0.5
ts
ts 0 0 x 0 0 x
Figure 11: The fuzzy approximate solution for different values of fractional orders V = 𝛼 with 𝑚 = 12, 𝑘1 = 0.014, and 𝑘2 = 0.001.
10−6 10−8
10−8
10−9
Absolute error
Absolute error
−10
10
10−12
10−10
10−14
10−16 10−11
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cuts r-cuts
Figure 12: The absolute error for different values V = 𝛼 with 𝑚 = 6, Figure 13: The absolute error for different values 𝑚 with V = 𝛼 =
𝑘1 = 0.05, and 𝑘2 = 0.05. 0.95, 𝑘1 = 0.05, and 𝑘2 = 0.05.
2 2 2
−8 ∑ 𝜃1,0,𝑘 ∑ 𝜃1,1,𝑘 ∑ 𝜃1,2,𝑘
10
𝑘=1 𝑘=1 𝑘=1 (88)
10−9 0 0 0
= ( 1.8381 0 0) .
10−10 −1.1751 4.3392 0
0.019
Using Theorem 26, we can derive the fuzzy residual 𝐺12 = 𝐺22 = (0.0017) . (90)
function for (81). Afterward, employing the inner product 1.000
explained in Section 4.3 and (81), we can generate fuzzy linear
algebraic system with (𝑚−1) equations which is in the matrix Putting (88) and (90) in (85), we reach a fuzzy linear algebraic
form as follows: equations system. Unknown coefficients vectors 𝐴 12 and 𝐴 22
are obtained by solving this system the values for 𝑟-cut = 1 are
𝐴 1 𝑇𝑚 [𝐷(V) + 𝐼] = 𝐺1 𝑇𝑚 , as
(85)
𝐴 2 𝑇𝑚 [𝐷(V) + 𝐼] = 𝐺2 𝑇𝑚 . 1.0060
𝐴 12 = 𝐴 22 = (0.0059) , (91)
0.0000
Additionally, we approximate fuzzy initial conditions (81) by
using (80) to produce the last equation which is needed to and from the earlier vector, we can attain the approximate
complete our fuzzy algebraic linear equations system. So, we fuzzy function as
have
𝐶𝐵2 1 = 𝐶𝐵2 2 = 1 + 0.0120𝑥(3/4) − 1.1301 × 10−4 𝑥(3/2) . (92)
𝐶𝐵𝑟 01 ≃ 𝐴 1 𝑇𝑚 Φ𝑚 (𝑥) = (0.5 + 0.5𝑟) ,
(86) Now, in the following results, we will reach the approxi-
𝐶𝐵𝑟 02 ≃ 𝐴 2 𝑇𝑚 Φ𝑚 (𝑥) = (1.5 − 0.5𝑟) . mate fuzzy solutions for (81) regarding different values of 𝑘1
and 𝑘2 from Table 1. Actually, we achieve the concentration
From (85) and (86), (𝑚)-fuzzy linear equations are produced of xylose after a specific time from the stated fuzzy fractional
which on can solve it easily using Definitions 8 and 9 and kinetic equation (81).
the method presented in [86] to find the unknown fuzzy In Table 2, we present the approximate solution and the
coefficients, {𝑎𝑗 }𝑚−1
𝑗=0 , of the fuzzy approximate solution. absolute error with 𝑘1 = 0.012, 𝑚 = 7, and 𝑘2 = 0.001
We solved the problem by applying the technique at 𝑇 = 1. As expected, the absolute error demonstrates the
described in Section 4. With 𝑚 = 3, V = 𝛼 = 0.75 and spectral accuracy of the proposed method. The absolute error
from Table 1, we assume that 𝑘1 = 0.012, 𝑘2 = 0.001. We for V = 𝛼 = 0.5, 075, 0.85, 0.95 is plotted in Figure 4. As we
approximate the solution in terms of the lower and upper can see, the absolute error increases gradually but remains
approximate fuzzy functions as in the interval [10−10 , 10−12 ]. This error can be explained
by the computer round-off errors that prevent any further
2 accuracy improvements. This error can be explained by the
𝑇
𝐶𝐵2 1 (𝑥; 𝑟) = ∑𝑎𝑗 𝑟1 FL0.75
𝑗 (𝑥) = [𝑎0 𝑟1 𝑎1 𝑟1 𝑎2 𝑟1 ] Φ (𝑥) , computer round-off errors that prevent any further accuracy
𝑗=0
improvements. Also, Figure 5 shows the absolute error for
(87)
2 different number of the FLFs which is clear that decreasing
𝑇
𝐶𝐵2 2 (𝑥; 𝑟) = ∑𝑎𝑗 𝑟2 FL0.75
𝑗 (𝑥) = [𝑎0 𝑟2 𝑎1 𝑟2 𝑎2 𝑟2 ] Φ (𝑥) , the the absolute error decreases with the increasing number
𝑗=0 of functions occurring. Moreover, in Figure 6, the absolute
Abstract and Applied Analysis 15
= 𝛼 = 0.5 = 𝛼 = 0.75
Approximate solution
Approximate solution
2 1.5
1 1
0 0.5
1 1
1 0.5 1
0.5
r-cu 0. 5 r-cu 0.5
ts x ts
0 0 0 0 x
= 𝛼 = 0.95 =𝛼=1
Approximate solution
2 1.5
Approximate solution
1 1
0 0.5
1 1
1 1
0.5 r-cu0.5
r-cu 0.5
ts
0.5
ts 0 x 0 x
0 0
Figure 15: The fuzzy approximate solution for different values of fractional orders V = 𝛼 with 𝑚 = 10, 𝑘1 = 0.05, and 𝑘2 = 0.05.
Table 2: The result of the proposed method for Example 27 with V = 𝛼 = 0.75, 𝑚 = 7, 𝑘1 = 0.012, and 𝑘2 = 0.001.
error obtained by our method for V = 0.75 and 𝑚 = 7. As considerably, but the method could achieve a suitable error
one can see, the increase in the amount of 𝛼 cannot reduce in the interval [10−9 , 10−14 ]. The absolute error for V = 0.85
the absolute error noticeably. The approximated solutions are and the various choices of 𝑚 and 𝛼 by the presented method
evaluated for V = 𝛼 = 0.5, 0.75, 0.85, 0.95 and 𝑚 = 8. The are shown in Figures 9 and 10, respectively, to make it easier to
results of the numerical simulations are plotted in Figure 7. compare with the analytic solution. We show the approximate
The approximate solutions for V = 𝛼 = 0.85, 𝑚 = 9, 𝑘1 = solution in Figure 11 for V = 𝛼 = 0.5, 0.75, 0.85, 0.95 with
0.014, and 𝑘2 = 0.001 are obtained in Table 3. The obtained 𝑚 = 12.
numerical results are in very good agreement with the exact Table 4 exhibits the error of analytical and numerical
solutions. In Figure 8, the absolute errors for the problem solutions. Numerical results demonstrate the good accuracy
of V = 𝛼 = 0.5, 0.75, 0.85, 0.95 by the proposed method of the proposed method. The absolute error is evaluated
with 𝑚 = 8 are given. It is obvious that the improvement of for various choices of 𝛼 in Figure 12. We see that as V, 𝛼
the value of the V and 𝛼 could not affect the absolute error approaches 1, the solution of the fuzzy fractional kinetic
16 Abstract and Applied Analysis
Table 3: The result of the proposed method for Example 27 with V = 𝛼 = 0.85, 𝑚 = 9, 𝑘1 = 0.014, and 𝑘2 = 0.001.
Table 4: The result of the proposed method for Example 27 with V = 𝛼 = 0.95, 𝑚 = 11, 𝑘1 = 0.05, and 𝑘2 = 0.05.
equation approaches that of the integer order fuzzy kinetic differential equations like fuzzy fractional oscillation differ-
equations. In Figure 13, again we see that the method could ential equations of the distributed order.
reach a high accuracy for the approximate solution, but the
increase in the number of the FLFS could not affect the
References
absolute error considerably. Also, in Figure 14, the method
could hit the better accuracy when the values of the fractional [1] A. Rodrı́guez-Chong, J. A. Ramı́rez, G. Garrote, and M.
derivative and fractional order of the Legendre functions Vázquez, “Hydrolysis of sugar cane bagasse using nitric acid:
were the same. In Figure 15, we show the fuzzy approximate a kinetic assessment,” Journal of Food Engineering, vol. 61, no. 2,
solutions which were evaluated for various values of V = 𝛼. pp. 143–152, 2004.
[2] Y. Lu and N. S. Mosier, “Kinetic modeling analysis of maleic
acid-catalyzed hemicellulose hydrolysis in corn stover,” Biotech-
nology and Bioengineering, vol. 101, no. 6, pp. 1170–1181, 2008.
6. Conclusion [3] Y. Sun, X. Lu, S. Zhang, R. Zhang, and X. Wang, “Kinetic study
for Fe(NO3 )3 catalyzed hemicellulose hydrolysis of different
In this paper, a spectral method has been adapted for numer- corn stover silages,” Bioresource Technology, vol. 102, no. 3, pp.
ical solution of fuzzy fractional kinetic equation of order 0 < 2936–2942, 2011.
V ≤ 1. A specific family of the orthogonal polynomials which [4] J. F. Saeman, “Kinetics of wood Saccharification-Hydrolysis of
is so-called FLFs was used as the approximation basis. The cellulose and decomposition of sugars in dilute acid at high
proposed method is characterized by its simplicity, efficiency, temperature,” Industrial and Engineering Chemistry, vol. 37, pp.
and high accuracy. The accuracy and validity of the presented 43–52, 1945.
method were demonstrated through numerical simulation [5] W. Faith, “Development of the Scholler Process in The United
for the derived FFKE. Using the proposed method, we States,” Industrial & Engineering Chemistry, vol. 37, pp. 9–11,
could reach the suitable approximation of the amount of the 1945.
concentration value of xylose after a determined time that is [6] M. Neureiter, H. Danner, C. Thomasser, B. Saidi, and R.
important to analyze the kinetic data in the chemical process. Braun, “Dilute-acid hydrohlysis of sugarcane bagasse at varying
A direction for further research would be to extend the conditions,” Applied Biochemistry and Biotechnology A, vol. 98–
presented method to the other types of the fuzzy fractional 100, pp. 49–58, 2002.
Abstract and Applied Analysis 17
[7] J. S. Kim, Y. Y. Lee, and R. W. Torget, “Cellulose hydrolysis under [25] K. Diethelm, The Analysis of Fractional Differential Equations.
extremely low sulfuric acid and high-temperature conditions,” An Application-Oriented Exposition Using Differential Operators
Applied Biochemistry and Biotechnology A, pp. 91–93, 331–340, of Caputo Type, vol. 2004 of Lecture Notes in Mathematics,
2001. Springer, Berlin, Germany, 2010.
[8] M. Caputo, “Linear model of dissipation whose Q is almost [26] R. Gorenflo and F. Mainardi, “Fractional calculus: integral
frequency independent, II, Geophys,” Journal of the Royal and differential equations of fractional order,” in Fractals and
Astronomical Society, vol. 13, pp. 529–539, 1967. Fractional Calculus in Continuum Mechanics (Udine, 1996), A.
[9] D. D. Clayton, Principles of Stellar Evolution and Nucleosynthe- Carpinteri and F. Mainardi, Eds., vol. 378 of CISM Courses and
sis, The University of Chicago Press, Chicago, Ill, USA, 2nd Lectures, pp. 223–276, Springer, Vienna, Austria, 1997.
edition, 1983. [27] A. Kadem and D. Baleanu, “Fractional radiative transfer
[10] W. G. Glöckle and T. F. Nonnenmacher, “Fox function represen- equation within Chebyshev spectral approach,” Computers &
tation of non-Debye relaxation processes,” Journal of Statistical Mathematics with Applications, vol. 59, no. 5, pp. 1865–1873,
Physics, vol. 71, no. 3-4, pp. 741–757, 1993. 2010.
[11] R. Hilfer, “Fractional time evolution,” in Applications of Frac- [28] K. S. Miller and B. Ross, An Introduction to the Fractional Calcu-
tional Calculus in Physics, pp. 87–130, World Scientific Publish- lus and Fractional Differential Equations, A Wiley-Interscience
ing, River Edge, NJ, USA, 2000. Publication, John Wiley & Sons, New York, NY, USA, 1993.
[12] V. L. Kobelev, E. P. Romanov, L. Y. Kobelev, and Y. L. Kobelev, [29] K. B. Oldham and J. Spanier, The Fractional Calculus, Academic
“Relaxational and diffusive processes in fractal space,” Izvestiya Press, New York, NY, USA, 1974.
Akademii Nauk. Seriya Fizicheskaya, vol. 62, no. 12, pp. 2401– [30] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
2408, 1998. ematics in Science and Engineering, Academic Press, San Diego,
[13] F. Mainardi, “Fractional diffusive waves in viscoelastic solids,” Calif, USA, 1999.
in Nonlinear Waves in Solids, J. L. Wegner and F. R. Norwood, [31] A. N. Kochubeı̆, “The Cauchy problem for evolution equations
Eds., pp. 93–97, ASME, New York, NY, USA, 1995. of fractional order,” Journal of Differential Equations, vol. 25, pp.
[14] F. Mainardi, “Fractal calculus: some basic problems in contin- 967–974, 1989.
uum and statistical mechanics,” in Fractals and Fractional Cal-
[32] H. J. Haubold and A. M. Mathai, “The fractional kinetic
culus in Continuum Mechanics, A. Carpinteri and F. Mainardi,
equation and thermonuclear functions,” Astrophysics and Space
Eds., pp. 291–348, Springer, Wien, Austria, 1997.
Science, vol. 273, no. 1–4, pp. 53–63, 2000.
[15] F. Mainardi and M. Tomirotti, “Seismic pulse propagation
[33] R. K. Saxena, A. M. Mathai, and H. J. Haubold, “On fractional
with constant Q and stable probability distributions,” Annali di
kinetic equations,” Astrophysics and Space Science, vol. 282, no.
Geofisica, vol. 40, no. 5, pp. 1311–1325, 1997.
1, pp. 281–287, 2002.
[16] R. Metzler, W. G. Glöckle, and T. F. Nonnenmacher, “Fractional
model equation for anomalous diffusion,” Physica A, vol. 211, no. [34] R. K. Saxena, A. M. Mathai, and H. J. Haubold, “Unified
1, pp. 13–24, 1994. fractional kinetic equation and a fractional diffusion equation,”
Astrophysics and Space Science, vol. 290, no. 3-4, pp. 299–310,
[17] A. I. Saichev and G. M. Zaslavsky, “Fractional kinetic equations:
2004.
solutions and applications,” Chaos, vol. 7, no. 4, pp. 753–764,
1997. [35] A. Atangana and A. Kilicman, “Analytical solutions of the space-
[18] G. M. Zaslavsky, “Fractional kinetics of Hamiltonian chaotic time fractional derivative of advection dispersion equation,”
systems,” in Applications of Fractional Calculus in Physics, pp. Mathematical Problems in Engineering, vol. 2013, Article ID
203–239, World Scientific Publishing, River Edge, NJ, USA, 853127, 9 pages, 2013.
2000. [36] A. M. A. El-Sayed, I. L. El-Kalla, and E. A. A. Ziada, “Analytical
[19] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional and numerical solutions of multi-term nonlinear fractional
Calculus. Models and Numerical Methods, vol. 3 of Series on orders differential equations,” Applied Numerical Mathematics,
Complexity, Nonlinearity and Chaos, World Scientific Publish- vol. 60, no. 8, pp. 788–797, 2010.
ing, Hackensack, NJ, USA, 2012. [37] H. Jiang, F. Liu, I. Turner, and K. Burrage, “Analytical solu-
[20] D. Baleanu, Z. B. Güvenc, and J. A. Tenreiro Machado, New tions for the multi-term time-space Caputo-Riesz fractional
Trends in Nanotechnology and Fractional Calculus Applications, advection-diffusion equations on a finite domain,” Journal of
Springer, New York, NY, USA, 2010. Mathematical Analysis and Applications, vol. 389, no. 2, pp. 1117–
[21] D. Baleanu and J. I. Trujillo, “A new method of finding 1127, 2012.
the fractional Euler-Lagrange and Hamilton equations within [38] Z. Odibat, S. Momani, and H. Xu, “A reliable algorithm of
Caputo fractional derivatives,” Communications in Nonlinear homotopy analysis method for solving nonlinear fractional
Science and Numerical Simulation, vol. 15, no. 5, pp. 1111–1115, differential equations,” Applied Mathematical Modelling, vol. 34,
2010. no. 3, pp. 593–600, 2010.
[22] P. L. Butzer and U. Westphal, “An introduction to fractional [39] N. H. Sweilam and M. M. Khader, “Exact solutions of some
calculus,” in Applications of Fractional Calculus in Physics, pp. coupled nonlinear partial differential equations using the
1–85, World Scientific Publishing, River Edge, NJ, USA, 2000. homotopy perturbation method,” Computers & Mathematics
[23] M. M. Djrbashian, Harmonic Analysis and Boundary Value with Applications, vol. 58, no. 11-12, pp. 2134–2141, 2009.
Problems in the Complex Domain, vol. 65 of Operator Theory: [40] M. ur Rehman and R. A. Khan, “A numerical method for solving
Advances and Applications, Birkhäuser, Basel, Switzerland, 1993. boundary value problems for fractional differential equations,”
[24] K. Diethelm, N. J. Ford, A. D. Freed, and Yu. Luchko, “Algo- Applied Mathematical Modelling, vol. 36, no. 3, pp. 894–907,
rithms for the fractional calculus: a selection of numerical 2012.
methods,” Computer Methods in Applied Mechanics and Engi- [41] A. Pedas and E. Tamme, “Spline collocation methods for
neering, vol. 194, no. 6–8, pp. 743–773, 2005. linear multi-term fractional differential equations,” Journal of
18 Abstract and Applied Analysis
Computational and Applied Mathematics, vol. 236, no. 2, pp. Applied Mathematical Modelling, vol. 37, no. 3, pp. 1126–1136,
167–176, 2011. 2013.
[42] S. Momani and Z. Odibat, “Numerical comparison of methods [59] A. H. Bhrawy, M. M. Alghamdi, and T. M. Taha, “A new
for solving linear differential equations of fractional order,” modified generalized Laguerre operational matrix of fractional
Chaos, Solitons & Fractals, vol. 31, no. 5, pp. 1248–1255, 2007. integration for solving fractional differential equations on the
[43] Z. M. Odibat and N. T. Shawagfeh, “Generalized Taylor’s half line,” Advances in Difference Equations, vol. 2012, article 179,
formula,” Applied Mathematics and Computation, vol. 186, no. 2012.
1, pp. 286–293, 2007. [60] A. H. Bhrawy, A. S. Alofi, and S. S. Ezz-Eldien, “A quadrature
[44] N. T. Shawagfeh, “Analytical approximate solutions for nonlin- tau method for fractional differential equations with variable
ear fractional differential equations,” Applied Mathematics and coefficients,” Applied Mathematics Letters, vol. 24, no. 12, pp.
Computation, vol. 131, no. 2-3, pp. 517–529, 2002. 2146–2152, 2011.
[45] A. Kilicman and Z. A. A. Al Zhour, “Kronecker operational [61] S. Esmaeili and M. Shamsi, “A pseudo-spectral scheme for the
matrices for fractional calculus and some applications,” Applied approximate solution of a family of fractional differential equa-
Mathematics and Computation, vol. 187, no. 1, pp. 250–265, 2007. tions,” Communications in Nonlinear Science and Numerical
[46] A. Kadem, Y. Luchko, and D. Baleanu, “Spectral method Simulation, vol. 16, no. 9, pp. 3646–3654, 2011.
for solution of the fractional transport equation,” Reports on [62] S. Abbasbandy, T. A. Viranloo, Ó. López-Pouso, and J. J.
Mathematical Physics, vol. 66, no. 1, pp. 103–115, 2010. Nieto, “Numerical methods for fuzzy differential inclusions,”
[47] S. Abbasbandy, “An approximation solution of a nonlinear Computers & Mathematics with Applications, vol. 48, no. 10-11,
equation with Riemann-Liouville’s fractional derivatives by He’s pp. 1633–1641, 2004.
variational iteration method,” Journal of Computational and [63] S. Abbasbandy, E. Babolian, and M. Allame, “Numerical solu-
Applied Mathematics, vol. 207, no. 1, pp. 53–58, 2007. tion of fuzzy max-min systems,” Applied Mathematics and
[48] S. Abbasbandy and A. Shirzadi, “Homotopy analysis method for Computation, vol. 174, no. 2, pp. 1321–1328, 2006.
multiple solutions of the fractional Sturm-Liouville problems,” [64] T. Allahviranloo, N. Ahmady, and E. Ahmady, “Numerical
Numerical Algorithms, vol. 54, no. 4, pp. 521–532, 2010. solution of fuzzy differential equations by predictor-corrector
[49] C. Hwang and Y. P. Shih, “Laguerre operational matrices for method,” Information Sciences, vol. 177, no. 7, pp. 1633–1647,
fractional calculus and applications,” International Journal of 2007.
Control, vol. 34, no. 3, pp. 577–584, 1981. [65] T. Allahviranloo, S. Abbasbandy, S. Salahshour, and A.
[50] G. Maione, “A digital, noninteger order, differentiator using Hakimzadeh, “A new method for solving fuzzy linear differen-
laguerre orthogonal sequences,” International Journal of Intel- tial equations,” Computing, vol. 92, no. 2, pp. 181–197, 2011.
ligent Systems, vol. 11, pp. 77–81, 2006. [66] E. Babolian, H. Sadeghi Goghary, and S. Abbasbandy, “Numer-
[51] Y. Li and N. Sun, “Numerical solution of fractional differential ical solution of linear Fredholm fuzzy integral equations of the
equations using the generalized block pulse operational matrix,” second kind by Adomian method,” Applied Mathematics and
Computers & Mathematics with Applications, vol. 62, no. 3, pp. Computation, vol. 161, no. 3, pp. 733–744, 2005.
1046–1054, 2011. [67] B. Bede and S. G. Gal, “Generalizations of the differentiability
[52] S. Kazem, S. Abbasbandy, and S. Kumar, “Fractional-order of fuzzy-number-valued functions with applications to fuzzy
Legendre functions for solving fractional order differential differential equations,” Fuzzy Sets and Systems, vol. 151, no. 3,
equations,” Applied Mathematical Modelling, vol. 37, pp. 5498– pp. 581–599, 2005.
5510, 2013. [68] A. Bencsik, B. Bede, J. Tar, and J. Fodor, “Fuzzy differential
[53] A. Saadatmandi and M. Dehghan, “A new operational matrix equations in modeling hydraulic differential servo cylinders,”
for solving fractional-order differential equations,” Computers in Proceedings of the 3rd Romanian-Hungarian Joint Symposium
& Mathematics with Applications, vol. 59, no. 3, pp. 1326–1336, on Applied Computational Intellidence (SACI ’06), Timisoara,
2010. Romania, 2006.
[54] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A Chebyshev [69] Y. Chalco-Cano and H. Román-Flores, “On new solutions of
spectral method based on operational matrix for initial and fuzzy differential equations,” Chaos, Solitons and Fractals, vol.
boundary value problems of fractional order,” Computers & 38, no. 1, pp. 112–119, 2008.
Mathematics with Applications, vol. 62, no. 5, pp. 2364–2373, [70] S. Salahshour and T. Allahviranloo, “Applications of fuzzy
2011. Laplace transforms,” Soft Computing, pp. 1–14, 2012.
[55] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “Efficient [71] R. P. Agarwal, V. Lakshmikantham, and J. J. Nieto, “On the
Chebyshev spectral methods for solving multi-term fractional concept of solution for fractional differential equations with
orders differential equations,” Applied Mathematical Modelling, uncertainty,” Nonlinear Analysis. Theory, Methods & Applica-
vol. 35, no. 12, pp. 5662–5672, 2011. tions, vol. 72, no. 6, pp. 2859–2862, 2010.
[56] M. M. Khader, T. S. El Danaf, and A. S. Hendy, “A computational [72] T. Allahviranloo, S. Salahshour, and S. Abbasbandy, “Explicit
matrix method for solving systems of high order fractional solutions of fractional differential equations with uncertainty,”
differential equations,” Applied Mathematical Modelling, vol. 37, Soft Computing, vol. 16, no. 2, pp. 297–302, 2012.
pp. 4035–4050, 2013. [73] T. Allahviranloo, Z. Gouyandeh, and A. Armand, “Fuzzy frac-
[57] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new tional differential equations under generalized fuzzy Caputo
Jacobi operational matrix: an application for solving fractional derivative,” Journal of Intelligent and Fuzzy Systems, 2013.
differential equations,” Applied Mathematical Modelling, vol. 36, [74] S. Salahshour, T. Allahviranloo, S. Abbasbandy, and D. Baleanu,
no. 10, pp. 4931–4943, 2012. “Existence and uniqueness results for fractional differential
[58] S. Kazem, “An integral operational matrix based on Jacobi equations with uncertainty,” Advances in Difference Equations,
polynomials for solving fractional-order differential equations,” vol. 2012, article 112, 2012.
Abstract and Applied Analysis 19
[75] S. Salahshour, T. Allahviranloo, and S. Abbasbandy, “Solving reactor,” in Proceedings of International Chemical, Biological and
fuzzy fractional differential equations by fuzzy Laplace trans- Environmental Engineering (IPCBEE ’12), vol. 38, IACSIT Press,
forms,” Communications in Nonlinear Science and Numerical 2012.
Simulation, vol. 17, no. 3, pp. 1372–1381, 2012. [96] S. Abbasbandy and M. Amirfakhrian, “A new approach to
[76] A. Ahmadian, M. Suleiman, S. Salahshour, and D. Baleanu, “A universal approximation of fuzzy functions on a discrete set
Jacobi operational matrix for solving a fuzzy linear fractional of points,” Applied Mathematical Modelling, vol. 30, no. 12, pp.
differential equation,” Advances in Difference Equations, vol. 1525–1534, 2006.
2013, article 104, 2013. [97] S. Abbasbandy and M. Amirfakhrian, “Best approximation of
[77] A. Ahmadian, M. Suleiman, and S. Salahshour, “An opera- fuzzy functions,” Journal of Nonlinear Studies, vol. 14, pp. 88–
tional matrix based on legendre polynomials for solving fuzzy 103, 2007.
fractional-order differential equations,” Abstract and Applied [98] G. A. Anastassiou, “Fuzzy approximation by fuzzy convolution
Analysis, vol. 2013, Article ID 505903, 29 pages, 2013. type operators,” Computers & Mathematics with Applications,
[78] M. Mazandarani and A. V. Kamyad, “Modified fractional Euler vol. 48, no. 9, pp. 1369–1386, 2004.
method for solving fuzzy fractional initial value problem,” Com- [99] H. Huang and C. Wu, “Approximation of fuzzy functions by
munications in Nonlinear Science and Numerical Simulation, vol. regular fuzzy neural networks,” Fuzzy Sets and Systems, vol. 177,
18, no. 1, pp. 12–21, 2013. pp. 60–79, 2011.
[79] P. Diamond and P. Kloeden, Metric Spaces of Fuzzy Sets. Theory [100] M. Dehghan and A. Saadatmandi, “A tau method for the one-
and Applications, World Scientific Publishing, River Edge, NJ, dimensional parabolic inverse problem subject to temperature
USA, 1994. overspecification,” Computers & Mathematics with Applications,
[80] D. Dubois and H. Prade, “Towards fuzzy differential calculus vol. 52, no. 6-7, pp. 933–940, 2006.
part 3: differentiation,” Fuzzy Sets and Systems, vol. 8, no. 3, pp. [101] J. H. Freilich and E. L. Ortiz, “Numerical solution of systems of
225–233, 1982. ordinary differential equations with the Tau method: an error
[81] R. Goetschel, Jr. and W. Voxman, “Elementary fuzzy calculus,” analysis,” Mathematics of Computation, vol. 39, no. 160, pp. 467–
Fuzzy Sets and Systems, vol. 18, no. 1, pp. 31–43, 1986. 479, 1982.
[82] H.-J. Zimmermann, Fuzzy Set Theory—And Its Applications,
Kluwer Academic Publishers, Boston, Mass, USA, 2nd edition,
1992.
[83] G. A. Anastassiou and S. G. Gal, “On a fuzzy trigonometric
approximation theorem of Weierstrass-type,” Journal of Fuzzy
Mathematics, vol. 9, no. 3, pp. 701–708, 2001.
[84] G. A. Anastassiou, Fuzzy Mathematics: Approximation Theory,
vol. 251 of Studies in Fuzziness and Soft Computing, Springer,
Berlin, Germany, 2010.
[85] O. Kaleva, “Fuzzy differential equations,” Fuzzy Sets and Sys-
tems, vol. 24, no. 3, pp. 301–317, 1987.
[86] T. Allahviranloo and M. Afshar Kermani, “Solution of a fuzzy
system of linear equation,” Applied Mathematics and Computa-
tion, vol. 175, no. 1, pp. 519–531, 2006.
[87] M. Friedman, M. Ming, and A. Kandel, “Fuzzy linear systems,”
Fuzzy Sets and Systems, vol. 96, no. 2, pp. 201–209, 1998.
[88] L. F. Wilhelmy, “ber das Gesetz, nach welchem die Einwirkung
der Suren auf Rohrzucker stattfindet,” Poggendorff ’s Annalen
der Physik und Chemie, vol. 81, pp. 413–433, 499–526, 1850.
[89] P. Waage and C. M. Guldberg, “Studier over affiniteten forhan-
dlinger,” VIdenskabs-Selskabet I ChrIstIana, vol. 35–40, pp. 111–
120, 1864.
[90] K. J. Laidler, The World of Physical Chemistry, Oxford University
Press, Oxford, UK, 1993.
[91] A. Cornish-Bowden, Fundamentals of Enzyme Kinetics, John
Wiley & Sons, New York, NY, USA, 4th edition, 2012.
[92] M. Robson Wright, An Introduction to Chemical Kinetics, John
Wiley & Sons, New York, NY, USA, 2004.
[93] F. Mainardi, “Fractional relaxation-oscillation and fractional
diffusion-wave phenomena,” Chaos, Solitons and Fractals, vol.
7, no. 9, pp. 1461–1477, 1996.
[94] P. Humbert and R. P. Agarwal, “Sur la fonction de Mittag-Leffler
et quelques-unes de ses généralisations,” Bulletin des Sciences
Mathématiques. 2e Série, vol. 77, pp. 180–185, 1953.
[95] S. Faridah Salleh, R. Yunus, M. Farid Atan, and D. R. Awg
Biak, “Kinetic studies on acid hydrolysis of OPEFB in a batch
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 413529, 12 pages
http://dx.doi.org/10.1155/2013/413529
Research Article
A Modified Generalized Laguerre Spectral Method for
Fractional Differential Equations on the Half Line
Copyright © 2013 D. Baleanu et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This paper deals with modified generalized Laguerre spectral tau and collocation methods for solving linear and nonlinear
multiterm fractional differential equations (FDEs) on the half line. A new formula expressing the Caputo fractional derivatives
of modified generalized Laguerre polynomials of any degree and for any fractional order in terms of the modified generalized
Laguerre polynomials themselves is derived. An efficient direct solver technique is proposed for solving the linear multiterm FDEs
with constant coefficients on the half line using a modified generalized Laguerre tau method. The spatial approximation with its
(𝛼,𝛽)
Caputo fractional derivatives is based on modified generalized Laguerre polynomials 𝐿 𝑖 (𝑥) with 𝑥 ∈ Λ = (0, ∞), 𝛼 > −1, and
𝛽 > 0, and 𝑖 is the polynomial degree. We implement and develop the modified generalized Laguerre collocation method based on
the modified generalized Laguerre-Gauss points which is used as collocation nodes for solving nonlinear multiterm FDEs on the
half line.
they also presented pseudospectral approximation for a class equipped with the following inner product and norm:
of FDEs. Yüzbaşı [28] proposed Bessel collocation method
for the approximate solution of the Bagley-Torvik equation. (𝑢, V)𝑤(𝛼,𝛽) = ∫ 𝑢 (𝑥) V (𝑥) 𝑤(𝛼,𝛽) (𝑥) 𝑑𝑥,
Atabakzadeh et al. [29] extended the application of the shifted Λ
(2)
Chebyshev operational matrix for obtaining an analytical
approximation solution for linear and nonlinear multiorder ‖V‖𝑤(𝛼,𝛽) = (V, V)1/2
𝑤(𝛼,𝛽)
.
FDE. Moreover, the authors of [30] extended the application
(𝛼,𝛽)
of the Legendre operational matrix for treating the fractional Next, let 𝐿 𝑖 (𝑥) be the modified generalized Laguerre
order fuzzy differential equations. Recently, Bhrawy et al. [31, polynomial of degree 𝑖 for 𝛼 > −1 and 𝛽 > 0, that is defined
32] proposed the operational matrices of Riemann-Liouville by
fractional integration of Chebyshev and modified generalized
Laguerre polynomials which are employed together with (𝛼,𝛽) 1 −𝛼 𝛽𝑥 𝑖 𝑖+𝛼 −𝛽𝑥
𝐿𝑖 (𝑥) = 𝑥 𝑒 𝜕𝑥 (𝑥 𝑒 ) , 𝑖 = 1, 2, . . . . (3)
spectral tau method for solving linear FDEs on finite and 𝑖!
semi-infinite intervals, respectively.
The fundamental aim of this paper is to extend the According to (2.3)-(2.4) of [11] for 𝛼 > −1 and 𝛽 > 0, we have
application of spectral tau and collocation methods based
(𝛼,𝛽) (𝛼+1,𝛽)
on modified generalized Laguerre polynomials to solve the 𝜕𝑥 𝐿 𝑖 (𝑥) = −𝛽𝐿 𝑖−1 (𝑥) , (4)
linear and nonlinear multiterm fractional initial value prob-
(𝛼,𝛽) 1 (𝛼,𝛽)
lems on the half line. We propose the modified generalized 𝐿 𝑖+1 (𝑥) = [(2𝑖 + 𝛼 + 1 − 𝛽𝑥) 𝐿 𝑖 (𝑥)
Laguerre spectral tau (MGLT) approximation for obtaining 𝑖+1
a direct solution technique to solve linear multiterm FDE (𝛼,𝛽)
− (𝑖 + 𝛼) 𝐿 𝑖−1 (𝑥)] , 𝑖 = 1, 2, . . . ,
on the half line. This method requires the Caputo fractional
derivatives of any fractional order of the modified generalized (5)
Laguerre polynomials of any degree which is already stated
(𝛼,𝛽) (𝛼,𝛽)
and proved. where 𝐿 0 (𝑥) = 1 and 𝐿 1 (𝑥) = −𝛽𝑥 + Γ(𝛼 + 2)/Γ(𝛼 + 1).
The modified generalized Laguerre spectral collocation The set of modified generalized Laguerre polynomials is
(MGLC) approximation, which is more reliable, is employed the 𝐿2𝑤(𝛼,𝛽) (Λ)-orthogonal system, namely,
to obtain approximate solution of nonlinear multiterm FDE
∞
on the half line with leading fractional differential operator (𝛼,𝛽) (𝛼,𝛽)
of order ] (𝑚 − 1 < ] ≤ 𝑚) and 𝑚 initial conditions.
∫ 𝐿𝑗 (𝑥) 𝐿 𝑘 (𝑥) 𝑤(𝛼,𝛽) (𝑥) 𝑑𝑥 = ℎ𝑘 𝛿𝑗𝑘 , (6)
0
To be more precise, in such approximation, the nonlinear
FDE is collocated only at (𝑁 − 𝑚 + 1) zeros of modified where 𝛿𝑗𝑘 is the Kronecker function and ℎ𝑘 = Γ(𝑘 + 𝛼 +
generalized Laguerre polynomials. Therefore, the problem 1)/𝛽𝛼+1 𝑘!.
reduces to a system of (𝑁 + 1) nonlinear algebraic equations The modified generalized Laguerre polynomials of degree
which may be solved by any standard technique to find the 𝑖 on the interval Λ, are given by
spectral solution 𝑢𝑁(𝑥). To the best of the our knowledge,
𝑖
the proposed algorithms have not been introduced for the
(𝛼,𝛽) Γ (𝑖 + 𝛼 + 1) 𝛽𝑘
numerical solution of multiterm FDEs on the half line. 𝐿𝑖 (𝑥) = ∑ (−1)𝑘 𝑥𝑘 ,
𝑘=0
Γ (𝑘 + 𝛼 + 1) (𝑖 − 𝑘)!𝑘! (7)
The plan of the paper is as follows. In the next section, we
introduce basic properties of modified generalized Laguerre
𝑖 = 0, 1, . . . ,
polynomials. In Section 3, the Caputo fractional derivative of
the modified generalized Laguerre polynomials is proved. In (𝛼,𝛽)
Section 4, we develop modified generalized Laguerre tau and where 𝐿 𝑖 (0) = Γ(𝑖 + 𝛼 + 1)/Γ(𝛼 + 1)Γ(𝑖 + 1).
collocation spectral methods for solving multiterm FDEs. In The special value
Section 5, several numerical examples are implemented. Also,
(𝛼,𝛽) (−1)𝑞 𝛽𝑞 Γ (𝑖 + 𝛼 + 1)
a conclusion is given in Section 6. 𝐷𝑞 𝐿 𝑖 (0) = , 𝑖 ⩾ 𝑞, (8)
(𝑖 − 𝑞)!Γ (𝑞 + 𝛼 + 1)
In particular applications, the modified generalized Theorem 2. The Caputo fractional derivative of order ] for
Laguerre polynomials up to degree 𝑁 + 1 are considered. modified generalized Laguerre polynomials is given by
Then, we have
𝑁
(𝛼,𝛽) (𝛼,𝛽)
𝑁 𝐷] 𝐿 𝑖 (𝑥) = ∑ Ψ] (𝑖, 𝑗) 𝐿 𝑗 (𝑥) , 𝑖 = ⌈]⌉ , . . . , 𝑁,
(𝛼,𝛽)
𝑢𝑁 (𝑥) = ∑𝑎𝑗 𝐿 𝑗 (𝑥) . (10) 𝑗=0
𝑗=0 (16)
then the standard modified generalized Laguerre tau approx- Let us also denote
imation to (23) is to find 𝑢𝑁 ∈ 𝑆𝑁(Λ) such that
𝑟−1 𝐴 = (𝑎𝑘𝑗 )0<𝑘,𝑗<𝑁,
(𝛼,𝛽) (𝛼,𝛽)
(𝐷] 𝑢𝑁, 𝐿 𝑘 (𝑥))𝑤(𝛼,𝛽) + ∑𝛾𝑖 (𝐷𝜁𝑖 𝑢𝑁, 𝐿 𝑘 (𝑥))𝑤(𝛼,𝛽)
𝑖=1 𝐵𝑠 = (𝑏𝑘𝑗
𝑠
) , (29)
0<𝑘,𝑗<𝑁;𝑠=1,2,...,𝑟−1
(𝛼,𝛽)
+ 𝛾𝑟 (𝑢𝑁, 𝐿 𝑘 (𝑥))𝑤(𝛼,𝛽) (26)
(𝛼,𝛽) 𝐶 = (𝑐𝑘𝑗 )0<𝑘,𝑗<𝑁,
= ((𝑔, 𝐿 𝑘 (𝑥))𝑤(𝛼,𝛽) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
(𝑖)
𝑢𝑁 (0) = 𝑑𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1. where
(𝛼,𝛽) (𝛼,𝛽)
(𝐷] 𝐿 𝑗 (𝑥) , 𝐿 𝑘 (𝑥)) (𝛼,𝛽) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = 0, 1, . . . , 𝑁,
𝑎𝑘𝑗 = { 𝑘−𝑁+𝑚−1 (𝛼,𝛽)
𝑤
𝐷 𝐿 𝑗 (0) , 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁, 𝑗 = 0, 1, . . . , 𝑁,
(𝛼,𝛽) (𝛼,𝛽)
𝑠 (𝐷𝜁𝑠 𝐿 𝑗 (𝑥) , 𝐿 𝑘 (𝑥)) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = 0, 1, . . . , 𝑁, 𝑠 = 1, 2, . . . , 𝑟 − 1,
𝑏𝑘𝑗 ={ 𝑤(𝛼,𝛽) (30)
0, otherwise,
(𝛼,𝛽) (𝛼,𝛽)
(𝐿 (𝑥), 𝐿 𝑘 (𝑥)) , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = 0, 1, . . . , 𝑁,
𝑐𝑘𝑗 = { 𝑗 𝑤(𝛼,𝛽)
0, otherwise,
then by virtue of the orthogonality relation of modified and (16) and after some manipulation, we may deduce that
𝑠
generalized Laguerre polynomials (6) and making use of (8) the elements of 𝑎𝑘𝑗 , 𝑏𝑘𝑗 , 𝑠 = 1, 2, . . . , 𝑟 − 1, and 𝑐𝑘𝑗 are given by
Abstract and Applied Analysis 5
{ℎ𝑘 Ψ] (𝑗, 𝑘) ,
{ 𝑘 = 0, 1, . . . , 𝑁 − 𝑚, 𝑗 = 𝑚, 𝑚 + 1, . . . , 𝑁,
𝑎𝑘𝑗 = { (−1)𝑘−𝑁+𝑚−1 𝛽(𝑘−𝑁+𝑚−1) Γ (𝑖 + 𝛼 + 1)!
{ , 𝑘 = 𝑁 − 𝑚 + 1, . . . , 𝑁, 𝑗 = 0, 1, . . . , 𝑁,
{ (𝑖 − 𝑘 + 𝑁 − 𝑚 + 1)! Γ (𝑘 − 𝑁 + 𝑚 + 𝛼 + 2)
ℎ , 𝑘 = 𝑗 = 0, 1, . . . , 𝑁 − 𝑚,
𝑐𝑘𝑗 = { 𝑘
0, otherwise.
Therefore, (28) can be written in the following matrix form: generalized Laguerre roots 𝑥𝑁−𝑚+1,𝑘 , which immediately
yields
𝑟−1
(𝐴 + ∑𝛾𝑖 𝐵𝑖 + 𝛾𝑟 𝐶) a = g. (32) 𝑁
(𝛼,𝛽)
𝑖=1 ∑𝑎𝑗 𝐷] 𝐿 𝑗 (𝑥𝑁−𝑚+1,𝑘 )
𝑗=0
4.2. Collocation Method for Nonlinear Multiterm FDEs. It is 𝑁
(𝛼,𝛽)
known that the collocation method approximates differential = 𝐹 (𝑥𝑁−𝑚+1,𝑘, ∑𝑎𝑗 𝐷𝛽1 𝐿 𝑗 (𝑥𝑁−𝑚+1,𝑘 ) , . . . ,
equations in physical space, so it is easy to apply to various 𝑗=0
problems, including variable coefficient and nonlinear dif- 𝑁
(𝛼,𝛽)
ferential equations (see, for instance, [36]). In this section, ∑ 𝑎𝑗 𝐷𝛽𝑘 𝐿 𝑗 (𝑥𝑁−𝑚+1,𝑘 )) ,
an algorithm for solving fractional differential equation is 𝑗=0
proposed based on modified generalized Laguerre spectral
𝑘 = 0, 1, . . . , 𝑁 − 𝑚,
collocation (MGLC) method in a half line. Here, we study the
(37)
following multiterm FDE:
with (33) written in the form
𝐷] 𝑢 (𝑥) = 𝐹 (𝑥, 𝑢 (𝑥) , 𝐷𝛽1 𝑢 (𝑥) , . . . , 𝐷𝛽𝑘 𝑢 (𝑥)) , in Λ,
𝑁
(33) (𝛼,𝛽)
∑𝑎𝑗 𝐿 𝑗 (0) = 𝑑𝑖 , 𝑖 = 0, 1, . . . , 𝑚 − 1. (38)
with initial conditions 𝑗=0
By virtue of (16), the fractional derivatives 𝐷] 𝑢(𝑥), Example 3. This example concerns a class of Bagley-Torvik
equation given by the initial value problem
𝐷𝛽1 𝑢(𝑥), . . . , 𝐷𝛽𝑘 𝑢(𝑥) can be expressed explicitly in terms of
the expansion coefficients 𝑎𝑗 . 8
The modified generalized Laguerre spectral collocation 𝐷2 𝑢 (𝑥) + 𝐷3/2 𝑢 (𝑥) + 𝑢 (𝑥) = 𝑥3 + 6𝑥 + 𝑥1.5 ,
Γ (0.5)
method for solving approximately (33), (34) is to find 𝑢𝑁(𝑥) ∈
𝑆𝑁(Λ) such that 𝑢 (0) = 0, 𝑢 (0) = 0, 𝑥 ∈ [0, 10] .
(39)
𝐷] 𝑢𝑁 (𝑥) = 𝐹 (𝑥, 𝑢𝑁 (𝑥) , 𝐷𝛽1 𝑢𝑁 (𝑥) , . . . , 𝐷𝛽𝑘 𝑢𝑁 (𝑥)) ,
(36) The solution is 𝑢(𝑥) = 𝑥3 .
is collocated at 𝑥𝑁−𝑚+1,𝑘 , 𝑘 = 0, 1, . . . , 𝑁 − 𝑚. In other This equation arises in the treatment of the motion
words, we have to collocate (36) at the (𝑁 − 𝑚 + 1) modified of a large thin plate immersed in a Newtonian fluid.
6 Abstract and Applied Analysis
5 5
4 4
−Log 10 error
−Log 10 error
3 3
2 2
1 1
0 0
20 30 40 50 60 70 80 90 20 30 40 50 60 70 80 90
N N
Figure 1: Maximum absolute error by using MGLT (Gauss) with the Figure 2: Maximum absolute error by using MGLT (Radau) with
various choices of 𝑁. the various choices of 𝑁.
2000
quadratures with general parameters 𝛼 and 𝛽. The maximum
absolute errors obtained by MGLT method for various
choices of 𝑁, 𝛼, and 𝛽 are given in Figures 1 and 2. Moreover, 1000
the approximate solutions obtained by the proposed method
for 𝛼 = 0, 𝛽 = 2, and two choices of 𝑁 are shown in Figure 3 0
0 5 10 15
to make it easier to compare with the analytic solution.
x
Example 4. Consider the equation Exact solution
uN (x) at N = 20
𝐷2 𝑢 (𝑥) + 𝐷] 𝑢 (𝑥) + 𝑢 (𝑥) = 𝑔 (𝑥) , uN (x) at N = 40
(40)
𝑢 (0) = 0, 𝑢 (0) = 0, 0 < ] < 1, 𝑥 ∈ [0, 10] , Figure 3: Comparing the exact and approximate solutions at 𝑁 =
20, 40, with 𝛼 = 0 and 𝛽 = 2.
where
1
𝑔 (𝑥) = 𝑥4 − 𝑥3 + 12𝑥2 − 3𝑥
2
(41) The solution of this problem is obtained by applying the
3 24 technique described in Section 4. The maximum absolute
− 𝑥3−] + 𝑥4−] ,
Γ (4 − ]) Γ (5 − ]) error for 𝛾 = {0.1 and 0.01} and various choices of 𝑁, 𝛼, 𝛽,
and ] are shown in Tables 2 and 3; moreover, the right-hand
and the exact solution is given by 𝑢(𝑥) = 𝑥4 − (1/2)𝑥3 (see side is treated by modified generalized Laguerre Gauss and
Table 1). modified generalized Laguerre-Gauss-Radau quadratures.
Table 1: Maximum absolute error, with various choices of 𝑁, for Example 4 in which 𝑥 ∈ [0, 10].
Table 2: Maximum absolute error for 𝛾 = 1/10, with various choices of 𝑁, for Example 5 in which 𝑥 ∈ [0, 10].
Table 3: Maximum absolute error for 𝛾 = 1/100, with various choices of 𝑁, for Example 5 in which 𝑥 ∈ [0, 10].
5.5
4
5
4.5 3
−Log 10 error
−Log 10 error
4
2
3.5
3 1
2.5
2 0
10 20 30 40 50
10 20 30 40 50
N
N
𝛼 = 1 and 𝛽 = 4 𝛼 = 1 and 𝛽 = 4
𝛼 = 3 and 𝛽 = 3 𝛼 = 3 and 𝛽 = 3
𝛼 = 3 and 𝛽 = 6
𝛼 = 3 and 𝛽 = 6
Figure 4: Maximum absolute error (case I) by using MGLT (Gauss) Figure 6: Maximum absolute error (case II) by using MGLT
with the various choices of 𝑁. (Gauss) with various choices of 𝑁.
3.5 3.5
3 3
2.5 2.5
−Log 10 error
−Log 10 error
2 2
1.5 1.5
1 1
0.5 0.5
0 0
10 20 30 40 50
10 20 30 40 50
N
N
𝛼 = 1 and 𝛽 = 4
𝛼 = 1 and 𝛽 = 4
𝛼 = 3 and 𝛽 = 3 𝛼 = 3 and 𝛽 = 3
𝛼 = 3 and 𝛽 = 6 𝛼 = 3 and 𝛽 = 6
Figure 5: Maximum absolute error (case I) by using MGLT (Radau) Figure 7: Maximum absolute error (case II) by using MGLT
with the various choices of 𝑁. (Radau) with various choices of 𝑁.
Table 4: Maximum absolute error for ] = 3/2, with various choices of 𝑁, for Example 5 in which 𝑥 ∈ (0, 1).
0.01 0.01
0.008 0.008
uN (x)
uN (x)
0.006 0.006
0.004 0.004
0.002 0.002
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x
Figure 8: Comparing the exact and approximate solutions by using Figure 9: Comparing the exact and approximate solutions by using
MGLT (Gauss) at 𝑁 = 8, 64, with 𝛼 = 2 and 𝛽 = 4. MGLT (Radau) at 𝑁 = 8, 64, with 𝛼 = 2 and 𝛽 = 4.
Table 5: Maximum absolute error for ] = 3/2, with various choices of 𝑁, for Example 9 in which 𝑥 ∈ (0, 1).
𝑁 𝛼 𝛽 MGLC 𝛼 𝛽 MGLC
3 2.66 ⋅ 10−2 2.61 ⋅ 10−2
6 2.27 ⋅ 10−3 3.04 ⋅ 10−3
0 1 2 3
9 1.30 ⋅ 10−3 1.36 ⋅ 10−3
12 5.68 ⋅ 10−4 7.52 ⋅ 10−4
3 1.39 ⋅ 10−2 1.07 ⋅ 10−2
6 1.87 ⋅ 10−3 1.86 ⋅ 10−3
2 4 3 6
9 8.47 ⋅ 10−4 6.94 ⋅ 10−4
12 4.02 ⋅ 10−4 3.63 ⋅ 10−4
×10−6 10
2.5
6
2
1.5 4
1
2
0.5
0
0
0 2 4 6 8 10 12 14 0 1 2 3 4
x x
Figure 11: The absolute error for 𝛾 = 1/100, ] = 3/4, 𝛼 = 2, and uN (x), = 1.2 uN (x), = 1.6
𝛽 = 2 at 𝑁 = 20. u(x), = 1.2 u(x), = 1.6
uN (x), = 1.4 uN (x), = 1.8
u(x), = 1.4 u(x), = 1.8
80
The maximum absolute errors at ] = 1.5 for various choices
60 of 𝛼, 𝛽, and 𝑁 in the interval [0, 1] are shown in Table 5.
Moreover, Figures 14 and 15 display a comparison between
40 the curves of exact solutions and the approximate solutions
20 at 𝛼 = 0 and 𝛽 = 1 of proposed problem subject to 𝑢(0) = 0
for the four different fractional orders ] = 0.2, 0.4, 0.6, and
0 0.8 in case of 𝑁 = 12 and 𝑁 = 16, respectively. From all
0 2 4 6 8 10
the Figures 12–15, it can be seen that the numerical solutions
x are in complete agreement with the exact solutions for all
uN (x), = 1.2 uN (x), = 1.6 values of ]. Also, from the numerical results implemented
u(x), = 1.2 u(x), = 1.6 in this example, the classical Laguerre polynomial (𝛼 = 0,
uN (x), = 1.4 uN (x), = 1.8 𝛽 = 1), which is used most frequently in practice, is not the
u(x), = 1.4 u(x), = 1.8 best one, especially when we are approximating the solution
of fractional differential equations.
Figure 12: Comparing the exact and approximate solutions at 𝑁 =
12, 𝛼 = 0, 𝛽 = 1, and ] = 1.2, 1.4, 1.6, and 1.8.
6. Conclusions
In this paper, we have proposed two efficient spectral meth-
Comparison between the curves of exact solutions and ods based on modified generalized Laguerre polynomials for
the approximate solutions at 𝛼 = 0 and 𝛽 = 1 of proposed tackling linear and nonlinear FDEs on the half line. In these
problem subject to 𝑢(0) = 𝑢 (0) = 0 for the four different methods, the problem is reduced to the solution of a system
fractional orders ] = 1.2, 1.4, 1.6, and 1.8 in case of 𝑁 = 12 of algebraic equations in the expansion coefficient of the
and 𝑁 = 16 are shown in Figures 12 and 13, respectively. solution. Numerical examples were given to demonstrate the
Abstract and Applied Analysis 11
[20] J. Song, F. Yin, X. Cao, and F. Lu, “Fractional variational Communications in Nonlinear Science and Numerical Simula-
iteration method versus Adomian’s decomposition method in tion, vol. 17, no. 1, pp. 62–70, 2012.
some fractional partial differential equations,” Journal of Applied [37] K. Diethelm and N. J. Ford, “Numerical solution of the Bagley-
Mathematics, vol. 2013, Article ID 392567, 10 pages, 2013. Torvik equation,” BIT. Numerical Mathematics, vol. 42, no. 3, pp.
[21] X. Zhang, L. Liu, Y. Wu, and Y. Lu, “The iterative solutions of 490–507, 2002.
nonlinear fractional differential equations,” Applied Mathemat- [38] S. S. Ray and R. K. Bera, “Analytical solution of the Bagley
ics and Computation, vol. 219, no. 9, pp. 4680–4691, 2013. Torvik equation by Adomian decomposition method,” Applied
[22] A. H. Bhrawy, A. S. Alofi, and S. S. Ezz-Eldien, “A quadrature Mathematics and Computation, vol. 168, no. 1, pp. 398–410, 2005.
tau method for fractional differential equations with variable [39] Y. Çenesiz, Y. Keskin, and A. Kurnaz, “The solution of the
coefficients,” Applied Mathematics Letters, vol. 24, no. 12, pp. Bagley-Torvik equation with the generalized Taylor collocation
2146–2152, 2011. method,” Journal of the Franklin Institute, vol. 347, no. 2, pp. 452–
[23] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “Efficient 466, 2010.
Chebyshev spectral methods for solving multi-term fractional [40] M. A. Z. Raja, J. A. Khan, and I. M. Qureshi, “Solution of
orders differential equations,” Applied Mathematical Modelling, fractional order system of Bagley-Torvik equation using evolu-
vol. 35, no. 12, pp. 5662–5672, 2011. tionary computational intelligence,” Mathematical Problems in
[24] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A Chebyshev Engineering, vol. 2011, Article ID 675075, 18 pages, 2011.
spectral method based on operational matrix for initial and [41] S. S. Ray, “On Haar wavelet operational matrix of general order
boundary value problems of fractional order,” Computers & and its application for the numerical solution of fractional
Mathematics with Applications, vol. 62, no. 5, pp. 2364–2373, Bagley Torvik equation,” Applied Mathematics and Computa-
2011. tion, vol. 218, no. 9, pp. 5239–5248, 2012.
[25] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new
Jacobi operational matrix: an application for solving fractional
differential equations,” Applied Mathematical Modelling, vol. 36,
no. 10, pp. 4931–4943, 2012.
[26] A. Ahmadian, M. Suleiman, S. Salahshour, and D. Baleanu, “A
Jacobi operational matrix for solving a fuzzy linear fractional
differential equation,” Advances in Difference Equations, 2013,
article 104.
[27] C. Li, F. Zeng, and F. Liu, “Spectral approximations to the
fractional integral and derivative,” Fractional Calculus and
Applied Analysis, vol. 15, no. 3, pp. 383–406, 2012.
[28] S. Yüzbaşı, “Numerical solution of the Bagley-Torvik equation
by the Bessel collocation method,” Mathematical Methods in the
Applied Sciences, vol. 36, no. 3, pp. 300–312, 2013.
[29] M. H. Atabakzadeh, M. H. Akrami, and G. H. Erjaee, “Cheby-
shev operational matrix method for solving multi-order frac-
tional ordinary differential equations,” Applied Mathematical
Modelling, 2013.
[30] A. Ahmadian, M. Suleiman, and S. Salahshour, “An opera-
tional matrix based on Legendre polynomials for solving fuzzy
fractional-order differential equations,” Abstract and Applied
Analysis, vol. 2013, Article ID 505903, 29 pages, 2013.
[31] A. H. Bhrawy and A. S. Alofi, “The operational matrix of frac-
tional integration for shifted Chebyshev polynomials,” Applied
Mathematics Letters, vol. 26, no. 1, pp. 25–31, 2013.
[32] A. H. Bhrawy, M. Alghamdi, and T. M. Taha, “A new modified
generalized Laguerre operational matrix of fractional integra-
tion for solving fractional differential equations on the half line,”
Advances in Difference Equations, vol. 2012, p. 179, 2012.
[33] B.-Y. Guo and X.-Y. Zhang, “A new generalized Laguerre
spectral approximation and its applications,” Journal of Compu-
tational and Applied Mathematics, vol. 181, no. 2, pp. 342–363,
2005.
[34] G. Szegő, Orthogonal Polynomials, vol. 23, American Mathe-
matical Society. Colloquium Publications, 1985.
[35] B.-Y. Guo and C. Zhang, “The spectral method for high order
problems with proper simulations of asymptotic behaviors at
infinity,” Journal of Computational and Applied Mathematics,
vol. 237, no. 1, pp. 269–294, 2013.
[36] A. H. Bhrawy and A. S. Alofi, “A Jacobi-Gauss collocation
method for solving nonlinear Lane-Emden type equations,”
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 816803, 11 pages
http://dx.doi.org/10.1155/2013/816803
Research Article
Numerical Modeling of Fractional-Order Biological Systems
Fathalla A. Rihan1,2
1
Department of Mathematical Sciences, College of Science, UAE University, Al Ain 15551, UAE
2
Department of Mathematics, Faculty of Science, Helwan University, Cairo 11795, Egypt
Copyright © 2013 Fathalla A. Rihan. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We provide a class of fractional-order differential models of biological systems with memory, such as dynamics of tumor-immune
system and dynamics of HIV infection of CD4+ T cells. Stability and nonstability conditions for disease-free equilibrium and
positive equilibria are obtained in terms of a threshold parameter R0 (minimum infection parameter) for each model. We provide
unconditionally stable method, using the Caputo fractional derivative of order 𝛼 and implicit Euler’s approximation, to find a
numerical solution of the resulting systems. The numerical simulations confirm the advantages of the numerical technique and
using fractional-order differential models in biological systems over the differential equations with integer order. The results may
give insight to infectious disease specialists.
The fractional derivative of order 𝛼 ∈ (𝑛 − 1, 𝑛) of 𝑓(𝑡) is We next provide a class of fractional-order differential
defined by two (nonequivalent) ways: models to describe the dynamics of tumour-immune system
interactions.
(i) Riemann-Liouville fractional derivative: take frac-
tional integral of order (𝑛 − 𝛼), and then take 𝑛th 2. Fractional Model of Tumor-Immune System
derivative as follows:
Immune system is one of the most fascinating schemes from
𝑑𝑛
𝐷∗𝛼 𝑓 (𝑡) = 𝐷∗𝑛 𝐼𝑎𝑛−𝛼 𝑓 (𝑡) , 𝐷∗𝑛 = 𝑛, 𝑛 = 1, 2, . . . . the point of view of biology and mathematics. The immune
𝑑𝑡 system is complex, intricate, and interesting. It is known to be
(2) multifunctional and multipathway, so most immune effectors
do more than one job. Also each function of the immune
(ii) Caputo-fractional derivative: take 𝑛th derivative, and system is typically done by more than one effector, which
then take a fractional integral of order (𝑛 − 𝛼) makes it more robust. The reason of using FODEs is that they
are naturally related to systems with memory which exists in
𝐷𝛼 𝑓 (𝑡) = 𝐼𝑎𝑛−𝛼 𝐷∗𝑛 𝑓 (𝑡) , 𝑛 = 1, 2, . . . . (3) tumor-immune interactions.
Ordinary and delay differential equations have long been
We notice that the definition of time-fractional derivative used in modeling cancer phenomena [33–37], but fractional-
of a function 𝑓(𝑡) at 𝑡 = 𝑡𝑛 involves an integration and order differential equations have short history in modeling
calculating time-fractional derivative that requires all the past such systems with memory. The authors in [1] used a system
history, that is, all the values of 𝑓(𝑡) from 𝑡 = 0 to 𝑡 = of fractional-order differential equations in modeling cancer-
𝑡𝑛 . For the concept of fractional derivative, we will adopt immune system interaction. The model includes two immune
Caputo’s definition which is a modification of the Riemann- effectors: 𝐸1 (𝑡), 𝐸2 (𝑡) (such as cytotoxic T cells and natural
Liouville definition and has the advantage of dealing properly killer cells), interacting with the cancer cells, 𝑇(𝑡), with a
with initial value problems. The following remark addresses Holling function of type III. (Holling type III describes a
some of the main properties of the fractional derivatives and situation in which the number of prey consumed per predator
integrals (see [8, 29–31]). initially rises slowly as the density of prey increases but then
levels off with further increase in prey density. In other words
Remark 2. Let 𝛽, 𝛾 ∈ R+ and 𝛼 ∈ (0, 1). Then the response of predators to prey is depressed at low prey
density, then levels off with further increase in prey density.)
(i) if 𝐼𝑎𝛽 : 𝐿1 → 𝐿1 and 𝑓(𝑡) ∈ 𝐿1 , then 𝐼𝑎𝛽 𝐼𝑎𝛾 𝑓(𝑡) = The model takes the form
𝐼𝑎𝛽+𝛾 𝑓(𝑡);
𝐷𝛼 𝑇 = 𝑎𝑇 − 𝑟1 𝑇𝐸1 − 𝑟2 𝑇𝐸2 ,
(ii) lim𝛽 → 𝑛 𝐼𝑎𝛽 𝑓(𝑥) = 𝐼𝑎𝑛 𝑓(𝑡) uniformly on [𝑎, 𝑏], 𝑛 = 1, 2,
𝑡 𝑇2 𝐸1
3, . . ., where 𝐼𝑎1 𝑓(𝑡) = ∫0 𝑓(𝑠)𝑑𝑠; 𝐷𝛼 𝐸1 = −𝑑1 𝐸1 + , 0 < 𝛼 ≤ 1,
𝑇2 + 𝑘1 (5)
(iii) lim𝛽 → 0 𝐼𝑎𝛽 𝑓(𝑡) = 𝑓(𝑡) weakly;
𝑇2 𝐸2
(iv) if 𝑓(𝑡) is absolutely continuous on [𝑎, 𝑏], then 𝐷𝛼 𝐸2 = −𝑑2 𝐸2 + ,
lim𝛼 → 1 𝐷∗𝛼 𝑓(𝑡) = 𝑑𝑓(𝑡)/𝑑𝑡; 𝑇2 + 𝑘1
(v) thus 𝐷∗𝛼 𝑓(𝑡) = (𝑑/𝑑𝑡)𝐼∗1−𝛼 𝑓(𝑡) (Riemann-Liouville where 𝑇 ≡ 𝑇(𝑡), 𝐸1 ≡ 𝐸1 (𝑡), 𝐸2 ≡ 𝐸2 (𝑡), and 𝑎, 𝑟1 , 𝑟2 , 𝑑1 ,
sense) and 𝐷𝛼 𝑓(𝑡) = 𝐼∗1−𝛼 (𝑑/𝑑𝑡)𝑓(𝑡) (Caputo sense). 𝑑2 , 𝑘1 , and 𝑘2 are positive constants. The interaction terms
in the second and third equations of model (5) satisfy the
The generalized mean value theorem and another property crossreactivity property of the immune system. It has been
are defined in the following remark [32]. assumed that (𝑑1 𝑘1 /(1 − 𝑑1 )) ≪ (𝑑2 𝑘2 /(1 − 𝑑2 )) to avoid
Abstract and Applied Analysis 3
the nonbiological interior solution where both immune To ease the analysis and stability of the steady states with
effectors coexist. The equilibrium points of the system (5) are meaningful parameters and minimize sensitivity (or robust-
ness) of the model, we nondimensionalize the bilinear system
(8) by taking the rescaling
𝑑1 𝑘1 𝑎
E0 = (0, 0, 0) , E1 = (√ , , 0) ,
(1 − 𝑑1 ) 𝑟1
(6) 𝐸 𝑇 𝑝1 𝑇0
𝑥= , 𝑦= , 𝜔= ,
𝑑𝑘 𝑎 𝐸0 𝑇0 𝑡𝑠 𝐸0
E2 = (√ 2 2 , 0, ) .
(1 − 𝑑2 ) 𝑟2 𝑝2 𝑠1 𝑝4
𝜃= , 𝜎= , 𝑎= , (9)
𝑡𝑠 𝑡𝑠 𝐸0 𝑡𝑠
The first equilibrium E0 is the nave, the second E1 is the
memory, and the third E2 is endemic according to the value 𝑝6 𝐸0
𝑏 = 𝑝5 𝑇0 , 1= , 𝜏 = 𝑡𝑠 𝑡.
of the tumor size. Stability analysis shows that the nave state is 𝑡𝑠
unstable. However, the memory state is locally asymptotically
stable if 𝑑1 < 𝑑2 and 𝑑1 < 1. While the endemic state is
locally asymptotically stable if 𝑑2 < 𝑑1 and 𝑑2 < 1, there Therefore, after the previous substitution into (8) and replac-
is bifurcation at 𝑑1 = 1. The stability of the memory state ing 𝜏 by 𝑡, the model becomes
depends on the value of one parameter, namely, the immune
effector death rate. 𝐷𝛼1 𝑥 = 𝜎 + 𝜔𝑥𝑦 − 𝜃𝑥,
Now we modify model (5) to include three populations (10)
of the activated immune-system cells, 𝐸(𝑡); the tumor cells, 𝐷𝛼2 𝑦 = 𝑎𝑦 (1 − 𝑏𝑦) − 𝑥𝑦.
𝑇(𝑡); and the concentration of IL-2 in the single tumor-site
compartment, 𝐼𝐿 (𝑡). We consider the classic bilinear model
that includes Holling function of type I and external effector 2.1. Equilibria and Local Stability of Model (10). The steady
cells 𝑠1 and input of IL-2, 𝑠2 . (holling Type I is a linear states of the reduced model (10) are again the intersection of
relationship, where the predator is able to keep up with the null clines 𝐷𝛼1 𝑥 = 0, 𝐷𝛼2 𝑦 = 0. If 𝑦 = 0, the tumor-
increasing density of prey by eating them in direct proportion free equilibrium is at E0 = (𝑥, 𝑦) = (𝜎/𝜃, 0). This steady state
to their abundance in the environment. If they eat 10% of the is always exist, since 𝜎/𝜃 > 0. From the analysis, it is easy
prey at low density, they continue to eat 10% of them at high to prove that the tumor-free equilibrium E0 = (𝜎/𝜃, 0) of the
densities.) The interactions of the three populations are then model (10) is asymptotically stable if threshold parameter (the
governed by the fractional-order differential model: minimum tumor-clearance parameter) R0 = 𝑎𝜃/𝜎 < 1 and
unstable if R0 > 1.
𝐷𝛼1 𝐸 = 𝑠1 + 𝑝1 𝐸𝑇 − 𝑝2 𝐸 + 𝑝3 𝐸𝐼𝐿 , However, if 𝑦 ≠0, the steady states are obtained by solving
0 < 𝛼𝑖 ≤ 1, 𝑖 = 1, 2, (8)
𝜔𝑦 − 𝜃 𝜔𝑥1
𝐽 (E1 ) = ( 1 ). (13)
𝐷𝛼2 𝑇 = 𝑝4 𝑇 (1 − 𝑝5 𝑇) − 𝑝6 𝐸𝑇. −𝑦1 𝑎 − 2𝑎𝑏𝑦1 − 𝑥1
4 Abstract and Applied Analysis
Proposition 4. Assume that the endemic equilibrium E1 the healthy CD4+ T-cells is (1 − ((𝐻 + 𝐼)/𝐻max )), and the
exists and has nonnegative coordinates. If R0 = 𝑎𝜃/𝜎 < 1, proliferation of infected cells is neglected. The parameter 𝑠 is
then tr(𝐽(E1 )) > 0 and E1 is unstable. the source of CD4+ T cells from precursors, 𝜇𝐻 is the natural
death rate of CD4+ T cells (𝜇𝐻𝐻max > 𝑠, cf. [39, page 85]),
Proof. Since 𝑟 is their growth rate (thus, 𝑟 > 𝜇𝐻 in general), and 𝐻max is
their carrying capacity. The parameter 𝑘1 represents the rate
𝜔2 − 𝜔 (𝑎𝑏 + 𝑏𝜃) − 𝑎𝑏2 𝜃 of infection of T cells with free virus. 𝑘1 is the rate at which
tr (𝐽 (E1 )) = infected cells become actively infected. 𝜇𝐼 is a blanket death
2𝑏𝜔
term for infected cells to reflect the assumption that we do not
𝜔 − 𝑎𝑏 √ 2 initially know whether the cells die naturally or by bursting.
+ 𝑎 (𝑏𝜃 + 𝜔)2 − 4𝑎𝑏𝜔 (𝑎𝜃 − 𝜎),
2𝑎𝑏𝜔 In addition, 𝜇𝑏 is the lytic death rate for infected cells. Since
(14) 𝑀 viral particles are released by each lysing cell, this term is
multiplied by the parameter 𝑀 to represent the source for free
then inequality tr(𝐽(E1 )) > 0 is true if virus (assuming a one-time initial infection). Finally, 𝜇𝑉 is the
loss rate of virus. The initial conditions for infection by free
𝑎 [𝜔2 − 𝜔 (𝑎𝑏 + 𝑏𝜃) − 𝑎𝑏2 𝜃] virus are 𝐻(0) = 𝐻0 , 𝐼(0) = 𝐼0 , and 𝑉(0) = 𝑉0 .
(15)
Theorem 6. According to Remark 3 and the fact that 𝑠 ≥ 0,
> (𝑎𝑏 − 𝜔) √𝑎2 (𝑏𝜃 + 𝜔)2 − 4𝑎𝑏𝜔 (𝑎𝜃 − 𝜎). then the system (16) has a unique solution (𝐻, 𝐼, 𝑉)𝑇 which
remains in R3+ and bounded by 𝐻max ; [11].
Therefore, when 𝑎𝜃 < 𝜎, we have 𝜔2 −𝜔𝑏(𝑎+𝜃)−𝑎𝜃𝑏2 > 0, and
hence both sides of the inequality are positive. Therefore if the 3.1. Equilibria and Local Stability of Model (16). To evaluate
equilibrium point E1 exists and has nonnegative coordinates, the equilibrium points of system (16), we put 𝐷𝛼1 𝐻(𝑡) =
then tr(𝐽(E1 )) > 0 and the point (E1 ) is unstable, whenever 𝐷𝛼2 𝐼(𝑡) = 𝐷𝛼3 𝑉(𝑡) = 0. Then the infection-free equilibrium
R0 = 𝑎𝜃/𝜎 < 1. point (uninfected steady state) is E∗0 = (𝐻0 , 0, 0), and
endemic equilibrium point (infected steady state) is E∗+ =
Similarly, we arrive at the following proposition. (𝐻∗ , 𝐼∗ , 𝑉∗ ), where
We also assume that Then the characteristic equation of the linearized system is
2−1
{𝑟 − 𝜇𝐻 + [(𝑟 − 𝜇𝐻) + 4𝑟𝑠𝐻max ]
1/2
} 𝑃 (𝜆) = 𝜆3 + 𝑎1 𝜆2 + 𝑎2 𝜆 + 𝑎3 = 0, (25)
𝐻0 = . (20)
−1
2𝑟𝐻max 𝑎1 = 𝜇𝐼 + 𝜇𝑉 + 𝑘1 𝐻∗ + 𝐿∗ ,
𝑎2 = 𝐿∗ (𝜇𝐼 + 𝜇𝑉 + 𝑘1 𝐻∗ ) + 𝜇𝐼 (𝜇𝑉 + 𝑘1 𝐻∗ )
The uninfected steady state is asymptotically stable if all of the
eigenvalues 𝜆 of the Jacobian matrix 𝐽(E∗0 ), given by (18), have 𝑟𝑉∗
negative real parts. The characteristic equation det(𝐽(E0 ) − − 𝑘12 𝐻∗ 𝑉∗ − 𝑘1 𝐻∗ (𝑀𝜇𝑏 − ),
𝐻max (26)
𝐼) = 0 becomes
∗
𝑟𝜇𝑉 𝑉
2𝑟𝐻0 𝑎3 = 𝑘1 𝐻∗ [𝑘1 𝑀𝜇𝑏 𝑉∗ − − 𝐿∗ 𝑀𝜇𝑏 ]
(𝜆 + 𝜇𝐻 − 𝑟 + ) (𝜆2 + 𝐵𝜆 + 𝐶) = 0, (21) 𝐻max
𝐻max
+ 𝐿∗ 𝜇𝐼 (𝜇𝑉 + 𝑘1 𝐻∗ ) − 𝜇𝐼 𝑘12 𝐻∗ 𝑉∗ .
where 𝐵 = 𝜇𝐼 + 𝑘𝐼 𝐻0 + 𝜇𝑉 and 𝐶 = 𝜇𝐼 (𝑘1 𝐻0 + 𝜇𝑉 ) − 𝑘1 𝜇𝑏 𝑀𝐻0 .
Hence, the three roots of the characteristic equation (21) are The infected steady state E∗+ is asymptotically stable if all of
the eigenvalues have negative real parts. This occurs if and
𝑟𝐻0 2 only if Routh-Hurwitz conditions are satisfied; that is, 𝑎1 > 0,
𝜆 1 = −𝜇𝐻 + 𝑟 − ≡ −√(𝑟 − 𝜇𝐻) + 4𝑟𝑠𝐻max
−1 < 0,
𝑎3 > 0, and 𝑎1 𝑎2 > 𝑎3 .
𝐻max
(22)
1
𝜆 2,3 = [−𝐵 ± √𝐵2 − 4𝐶] . 4. Implicit Euler’s Scheme for FODEs
2
Since most of the fractional-order differential equations do
Proposition 8. If R∗0 ≡ (𝑘1 𝜇𝑏 𝑀𝐻0 )/𝜇𝐼 (𝜇𝑉 + 𝑘1 𝐻0 ) < 1, then not have exact analytic solutions, approximation and numer-
𝐶 > 0 and the three roots of the characteristic equation (21) will ical techniques must be used. Several numerical methods
have negative real parts. have been proposed to solve the fractional-order differential
equations [18, 42, 43]. In addition, most of resulting biological
Corollary 9. In case of uninfected steady state E∗0 , one has systems are stiff. (One definition of the stiffness is that the
three cases. global accuracy of the numerical solution is determined by
stability rather than local error, and implicit methods are
(i) If R∗0 < 0, the uninfected state is asymptotically more appropriate for it.) The stiffness often appears due to
stable and the infected steady state E+ does not exist the differences in speed between the fastest and slowest
(unphysical). components of the solutions and stability constraints. In
(ii) If R∗0 = 1, then 𝐶 = 0, and from (21) implies that one addition, the state variables of these types of models are
eigenvalue 𝜆 = 0 and the remaining two eigenvalues very sensitive to small perturbations (or changes) in the
have negative real parts. The uninfected and infected parameters occur in the model. Therefore, efficient use of a
steady states collide, and there is a transcritical bifur- reliable numerical method that based in general on implicit
cation. formulae for dealing with stiff problems is necessary.
Consider biological models in the form of a system of
(iii) If R∗0 > 1, then 𝐶 < 0, and thus at least one eigenvalue FODEs of the form
will be positive real root. Thus, the uninfected state
E0 is unstable, and the endemically infected state E∗+ 𝐷𝛼 𝑋 (𝑡) = 𝐹 (𝑡, 𝑋 (𝑡) , P) , 𝑡 ∈ [0, 𝑇] , 0 < 𝛼 ≤ 1,
emerges. (27)
𝑋 (0) = 𝑋0 .
To study the local stability of the positive infected steady
states E∗+ for R∗0 > 1, we consider the linearized system of Here 𝑋(𝑡) = [𝑥1 (𝑡), 𝑥2 (𝑡), . . . , 𝑥𝑛 (𝑡)]𝑇 , P is the set of param-
(16) at E∗+ . The Jacobian matrix at E∗+ becomes eters appear in the model, and 𝐹(𝑡, 𝑋(𝑡)) satisfies the Lipschitz
condition
−𝑟𝐻∗ ‖𝐹 (𝑡, 𝑋 (𝑡) , P) − 𝐹 (𝑡, 𝑌 (𝑡) , P)‖ ≤ 𝐾 ‖𝑋 (𝑡) − 𝑌 (𝑡)‖ ,
−𝐿∗ −𝑘1 𝐻∗
𝐻max
∗
𝐾 > 0,
𝐽 (E∗+ ) = (𝑘1 𝑉 −𝜇𝐼 𝑘1 𝐻∗ ). (23) (28)
𝑘1 𝑉∗ 𝑀𝜇𝑏 − (𝑘1 𝐻∗ + 𝜇𝑉 ) where 𝑌(𝑡) is the solution of the perturbed system.
( )
Theorem 10. Problem (27) has a unique solution provided that
Here the Lipschitz condition (28) is satisfied and 𝐾𝑇𝛼 /Γ(𝛼 + 1) < 1.
𝑟 (2𝐻∗ + 𝐼∗ ) Proof. Using the definitions of Section 1, we can apply a
𝐿∗ = − [𝜇𝐻 − 𝑟 + 𝑘1 𝑉∗ + ]. (24)
𝐻max fractional integral operator to the differential equation (27)
6 Abstract and Applied Analysis
‖L𝑋 (𝑡) − L𝑌 (𝑡)‖ ≤ ‖𝑋 − 𝑌‖ . (32) then the first-order approximation method for the compu-
tation of Caputo’s fractional derivative is then given by the
This implies that our problem has a unique solution. expression
1.5 2.5
𝛼 = 0.75 𝛼 = 0.75
2
E2 (t)
1.5
E1 (t)
1
0.5 T(t) T(t)
0.5
E1 (t) E2 (t)
0 0
0 20 40 60 80 100 0 20 40 60 80 100
Time
Time
1.8 1.6
𝛼 = 0.95 𝛼 = 0.75
1.6 1.4
1.4
1.2
Effector cells E1 (t) E2 (t)
1.2 E2 (t)
T(t), E1 (t), E2 (t)
1
1
0.8
0.8
0.6
0.6
T(t) 0.4
0.4
0.2 0.2
E1 (t)
0 0
0 20 40 60 80 100 0 0.5 1 1.5 2 2.5
Time Tumor
Figure 1: Numerical simulations of the FODEs model (5) when 𝛼 = 0.75 and 𝛼 = 0.95 with (𝑎 = 𝑟1 = 𝑟2 = 1; 𝑑1 = 0.3; 𝑑2 = 0.7; 𝑘1 = 0.3;
𝑘2 = 0.7) and when 𝑑1 = 0.7; 𝑑2 = 0.3. The system converges to the stable steady states E1 , E2 . The fractional derivative damps the oscillation
behavior.
Theorem 12. The fully implicit numerical approximation (35), Since (1 − (𝜌0 /G𝛼,ℎ )) ≥ 1 for all G𝛼,ℎ , then
to test problem (36) for all 𝑡 ≥ 0, is consistent and uncon-
ditionally stable. 𝜁1 ≤ 𝜁0 , (39)
𝑛
Proof. We assume that the approximate solution of (36) is of
𝜁𝑛 ≤ 𝜁𝑛−1 + ∑𝜔𝑗(𝛼) (𝜁𝑛−𝑗 − 𝜁𝑛−𝑗+1 ) , 𝑛 ≥ 2.
the form 𝑢(𝑡𝑛 ) ≈ 𝑈𝑛 ≡ 𝜁𝑛 , and then (36) can be reduced to (40)
𝑗=2
𝜁2 ≤ 𝜁1 . (42)
or
Repeating the process, we have from (40)
𝜁𝑛−1 + ∑𝑛𝑗=2 𝜔𝑗(𝛼) (𝜁𝑛−𝑗 − 𝜁𝑛−𝑗+1 ) + 𝜌1 /G𝛼,ℎ
𝜁𝑛 = , 𝑛 ≥ 2. 𝑛
(1 − (𝜌0 /G𝛼,ℎ )) 𝜁𝑛 ≤ 𝜁𝑛−1 + ∑𝜔𝑗(𝛼) (𝜁𝑛−𝑗 − 𝜁𝑛−𝑗+1 ) ≤ 𝜁𝑛−1 , (43)
(38) 𝑗=2
8 Abstract and Applied Analysis
4 12
3.5
10
3
2.5
2 6
1.5
4
1
2
0.5
0
0 50 100 150 0 50 100 150
Time Time
𝛼=1 𝛼=1
𝛼 = 0.9 𝛼 = 0.9
𝛼 = 0.7 𝛼 = 0.7
12
10
8
Tumor cells, T(t)
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Effector cells, E(t)
𝛼=1
𝛼 = 0.9
𝛼 = 0.7
Figure 2: Numerical simulations of the FODEs model (10) when 𝑠 = 0.1181, 𝜔 = 0.1184, 𝜃 = 0.1747, 𝑟 = 0.636, and 𝑏 = 0.002. 𝛼1 = 𝛼2 =
1, 0.9, 0.7. The endemic state E2 is locally asymptotically stable when R0 = 𝑎𝜃/𝜎 > 1. The fractional derivative damps the oscillation behavior.
since each term in the summation is negative. Thus 𝜁𝑛 ≤ equilibrium points (for infection-free and endemic cases) are
𝜁𝑛−1 ≤ 𝜁𝑛−2 ≤ ⋅ ⋅ ⋅ ≤ 𝜁0 . With the assumption that 𝜁𝑛 = |𝑈𝑛 | ≤ the same in both integer-order (when 𝛼 = 1) and fractional-
𝜁0 = |𝑈0 |, which entails ‖𝑈𝑛 ‖ ≤ ‖𝑈0 ‖, we have stability. order (𝛼 < 1) models. We notice that, in the endemic steady
states, the fractional-order derivative damps the oscillation
Of course this numerical technique can be used both for behavior. From the graphs, we can see that FODEs have rich
linear and for nonlinear problems, and it may be extended dynamics and are better descriptors of biological systems
to multiterm FODEs. For more details about stability and than traditional integer-order models.
convergence of the fractional Euler method, we refer to [19, The numerical simulations for the HIV FODE model (16)
45]. (with parameter values given in the captions) are displayed in
Figure 4. We note that the solution of the model, with various
4.2. Numerical Simulations. We employed the implicit Euler’s values of 𝛼, continuously depends on the time-fractional
scheme (35) to solve the resulting biological systems of derivative but arrives to the equilibrium points. The displayed
FODEs (5), (10), and (16). Interesting numerical simulations solutions, in the figure, confirm that the fractional order of
of the fractional tumor-immune models (5) and (10), with the derivative plays the role of time delay in the system. We
step size ℎ = 0.05 and 0.5 < 𝛼 ≤ 1 and parameters values should also note that although the equilibrium points are
given in the captions, are displayed in Figures 1, 2, and 3. The the same for both integer-order and fractional-order models,
Abstract and Applied Analysis 9
0.75 0.35
0.7
0.3
0.65
0.6 0.25
Effector cells, E(t)
𝛼=1 𝛼=1
𝛼 = 0.9 𝛼 = 0.9
𝛼 = 0.8 𝛼 = 0.8
0.35
0.3
0.25
Tumor cells, T(t)
0.2
0.15
0.1
0.05
0
0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75
Effector cells, E(t)
𝛼=1
𝛼 = 0.9
𝛼 = 0.8
Figure 3: Numerical simulations of the FODEs model (10) when 𝑠 = 0.1181, 𝜔 = 0.1184, 𝜃 = 0.3747, 𝑟 = 1.636, and 𝑏 = 0.002. 𝛼1 = 𝛼2 =
1, 0.9, 0.7. The infection-free steady state E0 is locally asymptotically stable when R0 = 𝑎𝜃/𝜎 < 1.
the solution of the fractional order model tends to the fixed We provided unconditionally stable numerical technique,
point over a longer period of time. using the Caputo fractional derivative of order 𝛼 and implicit
Euler’s approximation, for the resulting system. The numer-
5. Conclusions ical technique is suitable for stiff problems. The solution
of the system at any time 𝑡∗ is continuously depends on
In fact, fractional-order differential equations are generaliza- all the previous states at 𝑡 ≤ 𝑡∗ . We have seen that the
tions of integer-order differential equations. Using fractional- presence of the fractional differential order leads to a notable
order differential equations can help us to reduce the errors increase in the complexity of the observed behaviour and play
arising from the neglected parameters in modeling biological the role of time lag (or delay term) in ordinary differential
systems with memory and systems distributed parameters. In model. We have obtained stability conditions for disease-
this paper, we presented a class of fractional-order differential free equilibrium and nonstability conditions for positive
models of biological systems with memory to model the equilibria. We should also mention that one of the basic
interaction of immune system with tumor cells and with reasons of using fractional-order differential equations is that
HIV infection of CD4+ T cells. The models possess non- fractional-order differential equations are, at least, as stable
negative solutions, as desired in any population dynamics. as their integer-order counterpart. In addition, the presence
We obtained the threshold parameter R0 that represents the of a fractional differential order in a differential equation can
minimum tumor-clearance parameter or minimum infection lead to a notable increase in the complexity of the observed
free for each model. behaviour, and the solution continuously depends on all
10 Abstract and Applied Analysis
1200
𝛼=1 600
800 500 𝛼=1
600 400
𝛼 = 0.9 300
400
𝛼 = 0.7 200 𝛼 = 0.9
𝛼 = 0.7
200 100
0 0
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
Time Time
×104
Initial density of HIV RNA
10
×104
8 10
𝛼=1 𝛼=1
V(t)
6
5
4 𝛼 = 0.9
𝛼 = 0.9 0
2 𝛼 = 0.7 800 𝛼 = 0.7
600 1000 1200
400 600 800
0 200 200 400
0 20 40 60 80 100 120 140 160 180 200 I(t) 0 0
H(t)
Time
Figure 4: Numerical simulations of the FODEs model (16) when 𝑠 = 20 day−1 mm−3 , 𝜇𝐻 = 0.02 day−1 , 𝜇𝐼 = 0.26 day−1 , 𝜇𝑏 = 0.24 day−1 ,
𝜇𝑉 = 2.4 day−1 , 𝑘1 = 2.4 × 10−5 mm3 day−1 , 𝑘1 = 2 × 10−5 mm3 day−1 , 𝑟 = 0.3 day−1 , 𝐻max = 1500 mm−3 , and 𝑀 = 1400. The trajectories of
the system approach to the steady state E∗+ .
the previous states. The analysis can be extended to other [8] R. Hilfer, Ed., Applications of Fractional Calculus in Physics,
models related to the immune response systems. World Scientific, River Edge, NJ, USA, 2000.
[9] G. M. Zaslavsky, “Chaos, fractional kinetics, and anomalous
transport,” Physics Reports, vol. 371, no. 6, pp. 461–580, 2002.
Acknowledgments [10] S. B. Yuste, L. Acedo, and K. Lindenberg, “Reaction front in an
A + B → C reaction-subdiffusion process,” Physical Review E,
The work is funded by the UAE University, NRF Project no. vol. 69, no. 3, Article ID 036126, pp. 1–36126, 2004.
NRF-7-20886. The author is grateful to Professor Changpin [11] W. Lin, “Global existence theory and chaos control of fractional
Li’s valuable suggestions and referees’ constructive com- differential equations,” Journal of Mathematical Analysis and
ments. Applications, vol. 332, no. 1, pp. 709–726, 2007.
[12] H. Sheng, Y. Q. Chen, and T. S. Qiu, Fractional Processes and
Fractional-Order Signal Processing, Springer, New York, NY,
References USA, 2012.
[13] K. Assaleh and W. M. Ahmad, “Modeling of speech signals
[1] E. Ahmed, A. Hashish, and F. A. Rihan, “On fractional order using fractional calculus,” in Proceedings of the 9th International
cancer model,” Journal of Fractional Calculus and Applied Symposium on Signal Processing and its Applications (ISSPA ’07),
Analysis, vol. 3, no. 2, pp. 1–6, 2012. Sharjah, United Arab Emirates, February 2007.
[2] A. A. M. Arafa, S. Z. Rida, and M. Khalil, “Fractional modeling [14] Y. Ferdi, “Some applications of fractional order calculus to
dynamics of HIV and CD4+ T-cells during primary infection,” design digital filters for biomedical signal processing,” Journal
Nonlinear Biomedical Physics, vol. 6, no. 1, article 1, 2012. of Mechanics in Medicine and Biology, vol. 12, no. 2, Article ID
[3] K. S. Cole, “Electric conductance of biological systems,” Cold 12400088, 13 pages, 2012.
Spring Harbor Symposia on Quantitative Biology, pp. 107–116, [15] W.-C. Chen, “Nonlinear dynamics and chaos in a fractional-
1993. order financial system,” Chaos, Solitons and Fractals, vol. 36, no.
[4] A. M. A. El-Sayed, A. E. M. El-Mesiry, and H. A. A. El-Saka, “On 5, pp. 1305–1314, 2008.
[16] A. Rocco and B. J. West, “Fractional calculus and the evolution
the fractional-order logistic equation,” Applied Mathematics
of fractal phenomena,” Physica A, vol. 265, no. 3, pp. 535–546,
Letters, vol. 20, no. 7, pp. 817–823, 2007.
1999.
[5] H. Xu, “Analytical approximations for a population growth [17] V. D. Djordjević, J. Jarić, B. Fabry, J. J. Fredberg, and D.
model with fractional order,” Communications in Nonlinear Stamenović, “Fractional derivatives embody essential features
Science and Numerical Simulation, vol. 14, no. 5, pp. 1978–1983, of cell rheological behavior,” Annals of Biomedical Engineering,
2009. vol. 31, no. 6, pp. 692–699, 2003.
[6] L. Debnath, “Recent applications of fractional calculus to [18] K. Diethelm, N. J. Ford, and A. D. Freed, “A predictor-corrector
science and engineering,” International Journal of Mathematics approach for the numerical solution of fractional differential
and Mathematical Sciences, no. 54, pp. 3413–3442, 2003. equations,” Nonlinear Dynamics, vol. 29, no. 1–4, pp. 3–22, 2002.
[7] A. M. A. El-Sayed, “Nonlinear functional-differential equations [19] C. Li and F. Zeng, “The finite difference methods for fractional
of arbitrary orders,” Nonlinear Analysis. Theory, Methods & ordinary differential equations,” Numerical Functional Analysis
Applications, vol. 33, no. 2, pp. 181–186, 1998. and Optimization, vol. 34, no. 2, pp. 149–179, 2013.
Abstract and Applied Analysis 11
[20] J. C. Trigeassou, N. Maamri, J. Sabatier, and A. Oustaloup, “State [39] A. S. Perelson, D. E. Kirschner, and R. De Boer, “Dynamics of
variables and transients of fractional order differential systems,” HIV infection of CD4+ T cells,” Mathematical Biosciences, vol.
Computers & Mathematics with Applications, vol. 64, no. 10, pp. 114, no. 1, pp. 81–125, 1993.
3117–3140, 2012. [40] G. A. Bocharov and F. A. Rihan, “Numerical modelling in
[21] X. Zhang and Y. Han, “Existence and uniqueness of positive biosciences using delay differential equations,” Journal of Com-
solutions for higher order nonlocal fractional differential equa- putational and Applied Mathematics, vol. 125, no. 1-2, pp. 183–
tions,” Applied Mathematics Letters, vol. 25, no. 3, pp. 555–560, 199, 2000.
2012. [41] R. V. Culshaw and S. Ruan, “A delay-differential equation model
[22] A. M. A. El-Sayed and M. Gaber, “The Adomian decomposition of HIV infection of CD4+ T-cells,” Mathematical Biosciences,
method for solving partial differential equations of fractal order vol. 165, no. 1, pp. 27–39, 2000.
in finite domains,” Physics Letters A, vol. 359, no. 3, pp. 175–182, [42] R. Anguelov and J. M.-S. Lubuma, “Nonstandard finite dif-
2006. ference method by nonlocal approximation,” Mathematics and
[23] H. Jafari and V. Daftardar-Gejji, “Solving a system of nonlinear Computers in Simulation, vol. 61, no. 3-6, pp. 465–475, 2003.
fractional differential equations using Adomian decomposi- [43] C. Li and F. Zeng, “Finite difference methods for fractional
tion,” Journal of Computational and Applied Mathematics, vol. differential equations,” International Journal of Bifurcation and
196, no. 2, pp. 644–651, 2006. Chaos in Applied Sciences and Engineering, vol. 22, no. 4, 28
[24] K. Diethelm and G. Walz, “Numerical solution of fractional pages, 2012.
order differential equations by extrapolation,” Numerical Algo- [44] F. A. Rihan, “Computational methods for delay parabolic
rithms, vol. 16, no. 3-4, pp. 231–253, 1997. and time-fractional partial differential equations,” Numerical
[25] L. Galeone and R. Garrappa, “On multistep methods for differ- Methods for Partial Differential Equations, vol. 26, no. 6, pp.
ential equations of fractional order,” Mediterranean Journal of 1556–1571, 2010.
Mathematics, vol. 3, no. 3-4, pp. 565–580, 2006. [45] Z. Odibat and S. Momani, “Numerical methods for nonlinear
[26] G. Wang, R. P. Agarwal, and A. Cabada, “Existence results partial differential equations of fractional order,” Applied Math-
and the monotone iterative technique for systems of nonlinear ematical Modelling, vol. 32, no. 1, pp. 28–39, 2008.
fractional differential equations,” Applied Mathematics Letters,
vol. 25, no. 6, pp. 1019–1024, 2012.
[27] N. J. Ford and A. C. Simpson, “The numerical solution of frac-
tional differential equations: speed versus accuracy,” Numerical
Algorithms, vol. 26, no. 4, pp. 333–346, 2001.
[28] V. Lakshmikantham, S. Leela, and J. Vasundhara, Theory of
Fractional Dynamic Systems, Cambridge Academic Publishers,
Cambridge. UK.
[29] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
ematics in Science and Engineering, Academic Press, San Diego,
Calif, USA, 1999.
[30] S. G. Samko, A. A. Kilbas, and O. I. Marichev, Fractional
integrals and derivatives, Gordon and Breach Science, Yverdon,
Switzerland, 1993.
[31] C. Li and Y. Ma, “Fractional dynamical system and its lineariza-
tion theorem,” Nonlinear Dynamics, vol. 71, no. 4, pp. 621–633,
2013.
[32] Z. M. Odibat and N. T. Shawagfeh, “Generalized Taylor’s
formula,” Applied Mathematics and Computation, vol. 186, no.
1, pp. 286–293, 2007.
[33] N. Bellomo, A. Bellouquid, J. Nieto, and J. Soler, “Multiscale
biological tissue models and flux-limited chemotaxis for mul-
ticellular growing systems,” Mathematical Models & Methods in
Applied Sciences, vol. 20, no. 7, pp. 1179–1207, 2010.
[34] A. Gökdoğan, A. Yildirim, and M. Merdan, “Solving a fractional
order model of HIV infection of CD+ T cells,” Mathematical and
Computer Modelling, vol. 54, no. 9-10, pp. 2132–2138, 2011.
[35] D. Kirschner and J. C. Panetta, “Modeling immunotherapy
of the tumor—immune interaction,” Journal of Mathematical
Biology, vol. 37, no. 3, pp. 235–252, 1998.
[36] F. A. Rihan, M. Safan, M. A. Abdeen, and D. Abdel Rahman,
“Qualitative and computational analysis of a mathematical
model for tumor-immune interactions,” Journal of Applied
Mathematics, vol. 2012, Article ID 475720, 19 pages, 2012.
[37] R. Yafia, “Hopf bifurcation in differential equations with delay
for tumor-immune system competition model,” SIAM Journal
on Applied Mathematics, vol. 67, no. 6, pp. 1693–1703, 2007.
[38] K. A. Pawelek, S. Liu, F. Pahlevani, and L. Rong, “A model of
HIV-1 infection with two time delays: mathematical analysis
and comparison with patient data,” Mathematical Biosciences,
vol. 235, no. 1, pp. 98–109, 2012.
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 740568, 11 pages
http://dx.doi.org/10.1155/2013/740568
Research Article
Oscillation Criteria for Fourth-Order Nonlinear Dynamic
Equations on Time Scales
Copyright © 2013 Xin Wu et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We establish some new oscillation criteria for nonlinear dynamic equation of the form (𝑎(𝑡)(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ )Δ + 𝑞(𝑡)𝑓(𝑥(𝜎(𝑡))) =
0 on an arbitrary time scale T with sup T = ∞, where 𝑎(𝑡), 𝑏(𝑡), 𝑐(𝑡) are positive rd-continuous functions. An example illustrating
the importance of our result is included.
on a time scale T, where 𝛾 ≥ 1 is a ratio of odd positive on an arbitrary time scale T with sup T = ∞, where 𝑝, 𝑞 ∈
∞
integers and the functions 𝑟𝑖 (𝑡) (𝑖 = 1, 2), 𝑞(𝑡) are positive 𝐶rd (T, (0, ∞)) with ∫𝑡 (1/𝑝(𝑠))Δ𝑠 < ∞ and there exists a
real-valued rd-continuous functions defined on T. 0
positive constant 𝐿 such that 𝑓(𝑦)/𝑦 ≥ 𝐿 for all 𝑦 ≠0; they
Erbe et al. in [15] were concerned with the oscillation of give a new oscillation result of (8).
the third-order nonlinear functional dynamic equation Motivated by the previous studies, in this paper, we will
Δ 𝛾 Δ study the oscillation criteria of the following fourth-order
(𝑎 (𝑡) [(𝑟 (𝑡) 𝑥Δ (𝑡)) ] ) + 𝑓 (𝑡, 𝑥 (𝑔 (𝑡))) = 0, (3) nonlinear dynamic equation:
on a time scale T, where 𝛾 is the quotient of odd positive Δ
Δ Δ
integers, 𝑎 and 𝑟 are positive rd-continuous functions on T, (𝑎 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ) ) + 𝑞 (𝑡) 𝑓 (𝑥 (𝜎 (𝑡))) = 0,
and 𝑔 : T → T satisfies lim𝑡 → ∞ 𝑔(𝑡) = ∞ and 𝑓 ∈
𝐶(T × R, R). The authors obtain some new oscillation criteria
𝑡 ∈ [𝑡0 , ∞)T ,
and extend many known results for oscillation of third-order
dynamic equations. (9)
Qi and Yu in [16] obtained some oscillation criteria for the
fourth-order nonlinear delay dynamic equation where T is a time scale with sup T = ∞ and 𝑡0 ∈ T is a
4
constant and [𝑡0 , ∞)T = [𝑡0 , ∞) ⋂ T. Throughout this paper,
𝑥Δ (𝑡) + 𝑝 (𝑡) 𝑥𝛾 (𝜏 (𝑡)) = 0, (4) we assume that the following conditions are satisfied:
on a time scale T, where 𝛾 is the ratio of odd positive
integers, 𝑝 is a positive real-valued rd-continuous function (H 1 ) 𝑎, 𝑏, 𝑐, 𝑞 ∈ 𝐶rd ([𝑡0 , ∞)T , (0, ∞)), 𝑏Δ (𝑡) ≥ 0 and 𝑐Δ (𝑡)
defined on T, 𝜏 ∈ 𝐶rd (T, T), 𝜏(𝑡) ≤ 𝑡, and lim𝑡 → ∞ 𝜏(𝑡) = ∞. ≥ 0.
Grace et al. in [17] were concerned with the oscillation of ∞ ∞ ∞
(H 2 ) ∫𝑡 (1/𝑎(𝑠))Δ𝑠 < ∫𝑡 (1/𝑏(𝑠))Δ𝑠 = ∫𝑡 (1/𝑐(𝑠))Δ𝑠 =
the fourth-order nonlinear dynamic equation ∞.
0 0 0
4
𝑥Δ (𝑡) + 𝑞 (𝑡) 𝑥𝜆 (𝑡) = 0, (5) (H 3 ) 𝑓 ∈ 𝐶(T, R) and there exists a positive constant 𝑀
on a time scale T, where 𝜆 is the ratio of odd positive such that for any 𝑢 ≠0, 𝑓(𝑢)/𝑢 ≥ 𝑀.
integers, 𝑞 is a positive real-valued rd-continuous function
defined on T. They reduce the problem of the oscillation of all By a solution of (9), we mean a nontrivial real-valued
1
solutions of (5) to the problem of oscillation of two second- function 𝑥 ∈ 𝐶rd ([𝑇𝑥 , ∞)T ) with 𝑇𝑥 ≥ 𝑡0 , which has the
order dynamic equations and give some conditions ensuring property that 𝑎(𝑡)(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ ∈ 𝐶rd 1
([𝑇𝑥 , ∞)T ) and
1
that all bounded solutions of (5) are oscillatory. satisfies (9) on [𝑇𝑥 , ∞)T , where 𝐶rd is the space of differ-
Grace et al. in [18] establish some new criteria for the entiable functions whose derivative is rd-continuous. The
oscillation of fourth-order nonlinear dynamic equations solutions vanishing in some neighborhood of infinity will be
2 Δ2 excluded from our consideration. A solution 𝑥(𝑡) of (9) is
(𝑎𝑥Δ ) (𝑡) + 𝑓 (𝑡, 𝑥𝜎 (𝑡)) = 0, 𝑡 ≥ 𝑡0 , (6) said to be oscillatory if it is neither eventually positive nor
eventually negative; otherwise it is called nonoscillatory.
where 𝑎 is a positive real-valued rd-continuous function
∞
satisfying that ∫𝑡 (𝜎(𝑠)/𝑎(𝑠))Δ𝑠 < ∞, 𝑓 : [𝑡0 , ∞) × R →
0 2. Some Auxiliary Lemmas
R is continuous satisfying sgn 𝑓(𝑡, 𝑥) = sgn 𝑥 and 𝑓(𝑡, 𝑥) ≤
𝑓(𝑡, 𝑦) for 𝑥 ≤ 𝑦 and 𝑡 ≥ 𝑡0 . They also investigate the case We shall employ the following lemmas.
of strongly superlinear and the case of strongly sublinear
equations subject to various conditions. Lemma 1. Assume that 𝑥(𝑡) is an eventually positive solution
Agarwal et al. in [19] were concerned with oscillatory of (9). Then there exists 𝑡1 ∈ [𝑡0 , ∞)T sufficiently large, such
behavior of a fourth-order half-linear delay dynamic equa- that, for 𝑡 ∈ [𝑡1 , ∞)T , one of the following cases holds:
tion with damping
Δ
Δ3
𝛾 Δ
Δ3
𝛾
𝛾 (1) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) < 0, (𝑐(𝑡)𝑥Δ (𝑡)) > 0,
(𝑟(𝑥 ) ) (𝑡) + 𝑝 (𝑡) (𝑥 ) (𝑡) + 𝑞 (𝑡) 𝑥 (𝜏 (𝑡)) = 0, (7) Δ Δ Δ
(𝑏(𝑡)(𝑐(𝑡)𝑥 (𝑡)) ) < 0,
on a time scale T with sup T = ∞, where 𝜆 is the ratio
of odd positive integers, 𝑟, 𝑝, 𝑞 are positive real-valued rd- (2) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ > 0,
Δ Δ Δ
continuous functions defined on T, 𝑟(𝑡) − 𝜇(𝑡)𝑝(𝑡) ≠0, 𝜏 ∈ (𝑏(𝑡)(𝑐(𝑡)𝑥 (𝑡)) ) < 0,
𝐶rd (T, T), 𝜏(𝑡) ≤ 𝑡, and 𝜏(𝑡) → ∞ as 𝑡 → ∞. They
establish some new oscillation criteria of (7). (3) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ > 0,
Δ Δ Δ
Zhang et al. in [20] were concerned with the oscillation of (𝑏(𝑡)(𝑐(𝑡)𝑥 (𝑡)) ) > 0,
a fourth-order nonlinear dynamic equation
Δ (4) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ < 0,
Δ3 Δ Δ
(𝑝𝑥 ) (𝑡) + 𝑞 (𝑡) 𝑓 (𝑥 (𝜎 (𝑡))) = 0, (8) Δ
(𝑏(𝑡)(𝑐(𝑡)𝑥 (𝑡)) ) > 0.
Abstract and Applied Analysis 3
Proof. Let 𝑥(𝑡) be an eventually positive solution of (9). Then Lemma 2 (see [12]). Assume that there exists 𝑇 ∈ T such
there is a 𝑡1 ≥ 𝑡0 , sufficiently large, such that, 𝑥(𝑡) > 0 for 𝑡 ≥ that 𝑈 satisfies
𝑡1 . By (9) we have
Δ
Δ Δ
(𝑎 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ) ) = −𝑞 (𝑡) 𝑓 (𝑥 (𝜎 (𝑡))) 𝑈 (𝑡) > 0, 𝑈Δ (𝑡) > 0, 𝑈ΔΔ (𝑡) > 0,
(15)
≤ −𝑀𝑞 (𝑡) 𝑥 (𝜎 (𝑡)) < 0, 𝑈ΔΔΔ (𝑡) ≤ 0, for 𝑡 ∈ [𝑇, ∞)T .
(10)
𝑡 𝜎(𝑠)
1 1
𝑥 (𝑡) ≤ 𝑥 (𝑡6 ) − 𝑀 ∫ Δ𝑠 → −∞, (14) ×∫ Δ𝑢 × (𝜎 (𝑠) 𝑐 (𝑠))−1 ) (20)
𝑡6 𝑐 (𝑠) 𝑡1 𝑏 (𝑢)
which contradicts 𝑥(𝑡) > 0 eventually. The proof is com- 1
− ] Δ𝑠 = ∞,
pleted. 4𝑄 (𝜎 (𝑠)) 𝑎 (𝑠)
4 Abstract and Applied Analysis
(1) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) < 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ > 0, Integrating the previous inequality from 𝑡0 to 𝑡 gives
(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ < 0, 𝑡 ∞
1 𝑄 (𝑠)
(2) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > Δ
0, (𝑐(𝑡)𝑥 (𝑡)) Δ
> 0, 𝑥 (𝑡0 ) − 𝑥 (𝑡) ≥ 𝑚 ∫ ∫ Δ𝑠Δ𝑢, (32)
𝑡0 𝑐 (𝑢) 𝑢 𝑏 (𝑠)
(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ < 0,
(3) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ > 0, which implies
(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ > 0, 𝑡 ∞
1 𝑄 (𝑠) 𝑥 (𝑡0 )
(4) 𝑥(𝑡) > 0, 𝑥Δ (𝑡) > 0, (𝑐(𝑡)𝑥Δ (𝑡))Δ < 0, ∫ ∫ Δ𝑠Δ𝑢 ≤ , (33)
𝑡0 𝑐 (𝑢) 𝑢 𝑏 (𝑠) 𝑚
(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ > 0.
If case (1) holds, then which contradicts assumption (18).
Let 𝐴(𝑡) = 𝑎(𝑡)(𝑏(𝑡)(𝑐(𝑡)𝑥Δ (𝑡))Δ )Δ . Integrating (27) from
Δ
Δ Δ 𝑡 to ∞ gives
(𝑎 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ) ) = −𝑞 (𝑡) 𝑓 (𝑥 (𝜎 (𝑡)))
∞ ∞
𝐴 (𝑠) 𝑄 (𝑠) 𝑄 (𝑠)
−𝑐 (𝑡) 𝑥Δ (𝑡) ≥ ∫ − Δ𝑠 ≥ −𝐴 (𝑡) ∫ Δ𝑠.
≤ −𝑀𝑞 (𝑡) 𝑥 (𝜎 (𝑡)) < 0, 𝑡 𝑏 (𝑠) 𝑡 𝑏 (𝑠)
(24) (34)
Abstract and Applied Analysis 5
𝑡
On the other hand, let 𝑈(𝑡) = ∫𝑡 𝑢(𝑠)Δ𝑠 for 𝑡 ∈ [𝑡1 , ∞)T , Combining (45) with (54) gives
1
Δ
where 𝑢(𝑡) = 𝑐(𝑡)𝑥 (𝑡); it is easy to check that 𝑈(𝑡) > 0, 𝜎(𝑡)
𝑀𝑑𝑞 (𝑡) ℎ2 (𝜎 (𝑡) , 𝑡0 ) ∫𝑡 (1/𝑏 (𝑠)) Δ𝑠 𝑅2 (𝑡)
𝑈Δ (𝑡) > 0, 𝑈ΔΔ (𝑡) > 0. In view of Δ
𝑅 (𝑡) ≤ − 2
− .
𝜎 (𝑡) 𝑐 (𝜎 (𝑡)) 𝑎 (𝑡)
Δ
(𝑏 (𝑡) 𝑢Δ (𝑡)) = 𝑏Δ (𝑡) 𝑢Δ (𝑡) + 𝑏𝜎 (𝑡) 𝑢ΔΔ (𝑡) < 0, (46) (55)
we get × 𝑄𝜎 (𝑠) Δ𝑠
𝑡 𝑡
𝑅 (𝑠) 𝑅2 (𝑠) 𝜎
𝑐 (𝑡) 𝑥 (𝑡) ≥ ∫ 𝑐 (𝑠) 𝑥Δ (𝑠) Δ𝑠. (51) −∫ [ + 𝑄 (𝑠)] Δ𝑠
𝑡1 𝑡2 𝑎 (𝑠) 𝑎 (𝑠)
In view of (49), we obtain that for all 𝑡 ∈ [𝑡𝑑 , ∞)T , ≤ 𝑅 (𝑡2 ) 𝑄 (𝑡2 )
𝜎(𝑠)
[ 𝑀𝑑𝑞 (𝑠) ℎ2 (𝜎 (𝑠) , 𝑡0 ) ∫𝑡2 (1/𝑏 (𝑢)) Δ𝑢
𝑡 𝑡
𝑥 (𝑡) 𝑐 (𝑡) 𝑥 (𝑡) ∫𝑡 𝑐 (𝑠) 𝑥Δ (𝑠) Δ𝑠 𝑑ℎ2 (𝑡, 𝑡0 ) (52) −∫ [
= ≥ 1
≥ . 𝑡2 𝜎 (𝑠) 𝑐 (𝜎 (𝑠))
𝑥Δ (𝑡) 𝑐 (𝑡) 𝑥Δ (𝑡) 𝑐 (𝑡) 𝑥Δ (𝑡) 𝑡
[
On the other hand, there exists 𝑡2 ≥ 𝑡𝑑 such that for any 𝑡 ∈
[𝑡2 , ∞)T , 1 ]
× 𝑄𝜎 (𝑠) − ] Δ𝑠,
4𝑄 (𝜎 (𝑠)) 𝑎 (𝑠)
𝑐 (𝑡) 𝑥Δ (𝑡) = 𝑐 (𝑡2 ) 𝑥Δ (𝑡2 ) ]
(58)
Δ Δ
𝑡 𝑏(𝑠) (𝑐 (𝑠) 𝑥 (𝑠))
+∫ Δ𝑠 which implies that
𝑏 (𝑠) (53)
𝑡2
𝜎(𝑠)
[ 𝑀𝑑𝑞 (𝑠) ℎ2 (𝜎 (𝑠) , 𝑡0 ) ∫𝑡2 (1/𝑏 (𝑢)) Δ𝑢
𝑡 𝑡
Δ 1 Δ
∫ [
≥ 𝑏(𝑡) (𝑐 (𝑡) 𝑥 (𝑡)) ∫ Δ𝑠. 𝜎 (𝑠) 𝑐 (𝜎 (𝑠))
𝑡2 𝑏 (𝑠) 𝑡2
[
It follows from (52) and (53) that (59)
𝜎 1 ]
𝑑ℎ2 (𝑡, 𝑡0 ) Δ × 𝑄 (𝑠) − ] Δ𝑠
𝑥 (𝑡) ≥ 𝑥 (𝑡) 4𝑄 (𝜎 (𝑠)) 𝑎 (𝑠)
𝑡 ]
𝑡 (54)
𝑑ℎ2 (𝑡, 𝑡0 ) ∫𝑡 (1/𝑏 (𝑠)) Δ𝑠 Δ
≤ 𝑅 (𝑡2 ) 𝑄 (𝑡2 ) − 𝑅 (𝑡) 𝑄 (𝑡) ≤ 𝑅 (𝑡2 ) 𝑄 (𝑡2 ) + 1,
Δ
≥ 2
𝑏(𝑡) (𝑐 (𝑡) 𝑥 (𝑡)) .
𝑡𝑐 (𝑡) which contradicts assumption (20).
Abstract and Applied Analysis 7
𝑠 (62)
∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 Δ Δ
× 1
Δ𝑠 −𝐴 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) ) )
𝑏 (𝑠)
Δ 𝑠 Δ Δ −1
𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) 𝑡 ∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 × (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) (𝑏(𝑐𝑥Δ ) ) (𝜎 (𝑡))) (67)
≥ 𝑡 ∫ 1
Δ𝑠,
∫𝑡 (1/𝑎 (𝑠)) Δ𝑠 𝑡2 𝑏 (𝑠)
1 (𝛼Δ (𝑡))+ 𝐴Δ (𝑡)
≤ 𝑅 (𝑡) + 𝛼𝜎 (𝑡) Δ
which implies that 𝛼 (𝑡) (𝑏 (𝑐𝑥Δ ) ) (𝜎 (𝑡))
Δ
𝑐 (𝑡) 𝑥Δ (𝑡) Δ Δ
( ) ≤ 0. (63) 𝐴 (𝑡) (𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) )
𝑡 𝑠
∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (𝑠)) Δ𝑠 − 𝛼𝜎 (𝑡) Δ Δ
2 1 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) (𝑏 (𝑐𝑥Δ ) ) (𝜎 (𝑡))
Hence, there exists 𝑡3 ∈ [𝑡2 , ∞)T such that
(𝛼Δ (𝑡))+ 𝐴Δ (𝑡)
𝑥 (𝑡) = 𝑥 (𝑡3 ) = 𝑅 (𝑡) + 𝛼𝜎 (𝑡) Δ
𝛼 (𝑡) (𝑏 (𝑐𝑥Δ ) ) (𝜎 (𝑡))
𝑡
𝑐 (𝑠) 𝑥Δ (𝑠) Δ
+∫ (𝑏(𝑐𝑥Δ ) (𝑡))
𝑡3
𝑠 V
∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV 𝜎 𝑅2 (𝑡)
− 𝛼 (𝑡) .
2 1
𝑎 (𝑡) 𝛼2 (𝑡) (𝑏(𝑐𝑥Δ )Δ ) (𝜎 (𝑡))
𝑠 V
∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV
× 2 1
Δ𝑠
𝑐 (𝑠) (64) By (61) and (65), we get
𝑐 (𝑡) 𝑥Δ (𝑡)
≥ 𝑡 𝑠 𝑥𝜎 (𝑡)
∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (𝑠)) Δ𝑠 𝑅Δ (𝑡) ≤ −𝑀𝑞 (𝑡) 𝛼𝜎 (𝑡) Δ
2 1
(𝑏(𝑐𝑥Δ ) ) (𝜎 (𝑡))
𝑠 V
𝑡 ∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV
×∫ 2 1
Δ𝑠. (𝛼Δ (𝑡))+
𝑡3 𝑐 (𝑠) + 𝑅 (𝑡)
𝛼 (𝑡)
Combining (62) with (64) gives Δ
𝑅2 (𝑡) (𝑏(𝑐𝑥Δ ) ) (𝑡)
𝜎
𝑡 𝑠 V
∫𝑡 (∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV/𝑐 (𝑠)) Δ𝑠 − 𝛼 (𝑡)
𝑎 (𝑡) 𝛼2 (𝑡) (𝑏(𝑐𝑥Δ )Δ ) (𝜎 (𝑡))
𝑥 (𝑡) ≥ 3 2 1
𝑡
∫𝑡 (1/𝑎 (𝑠)) Δ𝑠 (65)
1
Δ ≤ − 𝑀𝑞 (𝑡) 𝛼𝜎 (𝑡)
× 𝑏 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡)) .
8 Abstract and Applied Analysis
V
𝜎(𝑡) 𝑠 ∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 If case (4) holds, then
×∫ (∫ ( 1
ΔV)
𝑡3 𝑡2 𝑏 (V) (𝐴 (𝑡))Δ ≤ 𝑀𝑞 (𝑡) 𝑥 (𝜎 (𝑡)) < 0. (70)
V ∞ (73)
𝜎(𝑡) 𝑠 ∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 𝑧 ∫𝑠 𝑞 (𝑢) Δ𝑢
×∫ (∫ ( 1
ΔV) + 𝑀𝑥 (𝜎 (𝑡)) ∫ Δ𝑠 ≤ 0.
𝑡3 𝑡2 𝑏 (V) 𝑡 𝑎 (𝑠)
Letting 𝑧 → ∞ in this inequality, we obtain
−1
× (𝑐 (𝑠)) ) Δ𝑠 ∞
∞
∫𝑠 𝑞 (𝑢) Δ𝑢
Δ Δ
𝑏 (𝑡) (𝑐 (𝑡) 𝑥 (𝑡)) + 𝑀𝑥 (𝜎 (𝑡)) ∫ Δ𝑠 ≤ 0.
𝑡 𝑎 (𝑠)
𝜎(𝑡) −1
1 (74)
× (∫ Δ𝑠)
𝑡1 𝑎 (𝑠) Now we set
2 𝜎(𝑡)
[(𝛼Δ (𝑡))+ ] 𝑎 (𝑡) ∫𝑡 (1/𝑎 (𝑠)) Δ𝑠 𝑐 (𝑡) 𝑥Δ (𝑡)
+ 1
. 𝑅 (𝑡) = 𝛽 (𝑡) for 𝑡 ∈ [𝑡1 , ∞)T . (75)
𝑡 𝑥 (𝑡)
4𝛼𝜎 (𝑡) ∫𝑡 (1/𝑎 (𝑠)) Δ𝑠
1
Integrating the last inequality from 𝑡4 (𝑡4 ∈ [𝑡3 , ∞)T ) to 𝑡, 𝑐 (𝑡) 𝑥Δ (𝑡)
𝑅Δ (𝑡) = 𝛽Δ (𝑡)
we get 𝑥 (𝑡)
Δ
𝜎
(𝑐 (𝑡) 𝑥Δ (𝑡)) 𝑥 (𝑡) − 𝑐 (𝑡) 𝑥Δ (𝑡) 𝑥Δ (𝑡)
𝑡
[ +𝛽 (𝑡)
∫ [𝑀𝑞 (𝑠) 𝛼𝜎 (𝑠) 𝑥 (𝑡) 𝑥 (𝜎 (𝑡))
𝑡4 (76)
[ Δ
Δ
𝛽 (𝑡) (𝑐 (𝑡) 𝑥Δ (𝑡))
V ≤ 𝑅 (𝑡) + 𝛽𝜎 (𝑡)
𝜎(𝑠) 𝑧 ∫𝑡 (1/𝑎 (𝑢)) Δ𝑢 𝛽 (𝑡) 𝑥 (𝜎 (𝑡))
×∫ (∫ ( 1
ΔV)
𝑡3 𝑡2 𝑏 (V) 𝛽𝜎 (𝑡) 𝑥 (𝑡) 2
− 𝑅 (𝑡) .
𝛽2 (𝑡) 𝑐 (𝑡) 𝑥𝜎 (𝑡)
× (𝑐 (𝑧))−1 ) Δ𝑧 Since
(69)
𝑡
𝑐 (𝑠) 𝑥Δ (𝑠)
𝜎(𝑠)
1
−1 𝑥 (𝑡) ≥ 𝑥 (𝑡) − 𝑥 (𝑡1 ) ≥ ∫ Δ𝑠
× (∫ Δ𝑧) 𝑡1 𝑐 (𝑠)
𝑡1 𝑎 (𝑧) 𝑡
(77)
Δ 1
2 𝜎(𝑠) ≥ 𝑐 (𝑡) 𝑥 (𝑡) ∫ Δ𝑠,
[(𝛼Δ (𝑠))+ ] 𝑎 (𝑠) ∫𝑡 (1/𝑎 (𝑧)) Δ𝑧 ] 𝑡1 𝑐 (𝑠)
− 𝑠 ] Δ𝑠 1
Integrating the previous inequality from 𝑡1 to 𝑡, we get Then, condition (18) holds. By Lemma 2, we get
𝑡 (𝑡 − 𝑡0 ) (𝑡 − 2𝑡0 )
ℎ2 (𝑡, 𝑡0 ) = ∫ (𝜏 − 𝑡0 ) Δ𝜏 = ,
∞ 𝑡0 3
𝑡 𝜎 ∞ ∫ 𝑞 (V) ΔV (86)
[ 𝑀𝛽 (𝑠)
∫ [ ∫ 𝑢 Δ𝑢 (2𝑡 − 𝑡0 ) (2𝑡 − 2𝑡0 )
𝑡1 𝑏 (𝑠) 𝑠 𝑎 (𝑢) ℎ2 (𝜎 (𝑡) , 𝑡0 ) = ℎ2 (2𝑡, 𝑡0 ) = ≥ 𝑡2 ,
[ 3
2 𝜎(𝑠)
while 𝑡 sufficiently large. Let 𝛼(𝑡) = 1, 𝛽(𝑡) = 𝑡. We have that
𝑐 (𝑠) [(𝛽Δ (𝑠))+ ] ∫𝑡 (1/𝑐 (𝑢)) Δ𝑢 ] (80) if ≥ 1/2𝑑, then
− 𝑠
1
] Δ𝑠 𝑡
4𝛽𝜎 (𝑠) ∫𝑡 (1/𝑐 (𝑢)) Δ𝑢
1
] lim sup ∫ [ (𝑑𝑀𝑞 (𝑠) 𝑄 (𝜎 (𝑠)) ℎ2 (𝜎 (𝑠) , 𝑡0 )
𝑡→∞ 𝑡0
≤ 𝑅 (𝑡1 ) , 𝜎(𝑠)
×∫ (1/𝑏 (𝑢)) Δ𝑢)
𝑡1
which contradicts assumption (22). The proof is completed.
× (𝜎 (𝑠) 𝑐 (𝑠))−1
1 (87)
− ] Δ𝑠
4. Example 4𝑄 (𝜎 (𝑠)) 𝑎 (𝑠)
Finally, we give an example to illustrate our main result. 𝑡 𝑑 (/𝑠6/5 ) (1/𝑠) 𝑠2 𝑠4/5 1
≥ lim sup ∫ [ − ] Δ𝑠
𝑡→∞ 𝑡0 2𝑠𝑠3/5 4𝑠
Example 1. Consider the fourth-order nonlinear dynamic
equation 𝑑 1 𝑡
1
≥( − ) lim sup ∫ Δ𝑠 = ∞.
2 4 𝑡 → ∞ 𝑡0 𝑠
Δ Since
Δ Δ
(𝑡2 (𝑡1/5 (𝑡3/5 𝑥Δ (𝑡)) ) ) + 𝑓 (2𝑡) = 0, 𝑡 ∈ 2Z , (81) 𝑧 V
𝑡6/5 𝜎(𝑠) ∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV
∫ ( 2 1
) Δ𝑧
𝑡3 𝑐 (𝑧)
where > 0 is a constant, and
1/5 𝑧
1 𝜎(𝑠) ∫𝑡2 (1/V ) ΔV (88)
≥ ∫ Δ𝑧
𝑡1 𝑡3 𝑐 (𝑧)
𝑎 (𝑡) = 𝑡2 , 𝑏 (𝑡) = 𝑡1/5 , 𝑐 (𝑡) = 𝑡3/5 ,
(82) 1 𝜎(𝑠) 𝑧4/5 1
𝑞 (𝑡) = , 𝑓 (𝑢) = 𝑢 ln (3 + 𝑢 ) . 2 ≥ ∫ Δ𝑧 ≥ 𝑠6/5 ,
𝑡6/5 𝑡1 𝑡3 𝑐 (𝑧) 𝑡1
10 Abstract and Applied Analysis
we get References
[1] S. Hilger, “Analysis on measure chains—a unified approach to
𝑡 continuous and discrete calculus,” Results in Mathematics, vol.
[
lim sup ∫ [𝑀𝑞 (𝑠) 𝛼𝜎 (𝑠) 18, no. 1-2, pp. 18–56, 1990.
𝑡→∞ 𝑡4
[2] V. Kac and P. Cheung, Quantum Calculus, Universitext, 2002.
[
[3] M. Bohner and A. Peterson, Dynamic Equations on Time Scales:
𝑧 V
𝜎(𝑠) ∫𝑡 (∫𝑡 (1/𝑎 (𝑢)) Δ𝑢/𝑏 (V)) ΔV An Introduction with Applications, Birkhäuser, Boston, Mass,
×∫ ( 2 1
) Δ𝑧 USA, 2001.
𝑡3 𝑐 (𝑧) [4] M. Bohner and A. Peterson, Advances in Dynamic Equations on
Time Scales, Birkhäuser, Boston, Mass, USA, 2003.
𝜎(𝑠) −1
1 [5] T. Sun, H. Xi, and X. Peng, “Asymptotic behavior of solutions
× (∫ ( ) Δ𝑧) of higher-order dynamic equations on time scales,” Advances in
𝑡1 𝑎 (𝑧)
Difference Equations, vol. 2011, Article ID 237219, 14 pages, 2011.
2 𝜎(𝑠) [6] T. Sun, H. Xi, X. Peng, and W. Yu, “Nonoscillatory solutions
[(𝛼Δ (𝑠))+ ] 𝑎 (𝑠) ∫𝑡 (1/𝑎 (𝑧)) Δ𝑧 ]
for higher-order neutral dynamic equations on time scales,”
− 𝑠
1
] Δ𝑠
4𝛼𝜎 (𝑠) ∫𝑡 (1/𝑎 (𝑧)) Δ𝑧 Abstract and Applied Analysis, vol. 2010, Article ID 428963, 16
1
] pages, 2010.
6/5 6/5 [7] T. Sun, H. Xi, and W. Yu, “Asymptotic behaviors of higher order
𝑡 (/𝑠6/5 ) (1/𝑡1 ) 2 𝑠 nonlinear dynamic equations on time scales,” Journal of Applied
≥ lim sup ∫ [ ] Δ𝑠
Mathematics and Computing, vol. 37, no. 1-2, pp. 177–192, 2011.
𝑡→∞ 𝑡4 1/𝑡1
[ ] [8] T. Sun, W. Yu, and H. Xi, “Oscillatory behavior and comparison
𝑡 for higher order nonlinear dynamic equations on time scales,”
= 26/5 lim sup ∫ Δ𝑠 = ∞. Journal of Applied Mathematics & Informatics, vol. 30, no. 1-2,
𝑡→∞ 𝑡4 pp. 289–304, 2012.
(89) [9] T. S. Hassan, “Oscillation of third order nonlinear delay
dynamic equations on time scales,” Mathematical and Computer
Since Modelling, vol. 49, no. 7-8, pp. 1573–1586, 2009.
[10] L. Erbe, A. Peterson, and S. H. Saker, “Asymptotic behavior of
∞
∞ ∫𝑢 𝑞 (V) ΔV 1 solutions of a third-order nonlinear dynamic equation on time
∫ Δ𝑢 ≥ , (90) scales,” Journal of Computational and Applied Mathematics, vol.
𝑠 𝑎 (𝑢) 𝑠6/5 181, no. 1, pp. 92–102, 2005.
𝜎(𝑠) [11] L. Erbe, A. Peterson, and S. H. Saker, “Oscillation and asymp-
∫𝑡 (1/𝑐 (𝑢)) Δ𝑢 totic behavior of a third-order nonlinear dynamic equation,”
1
𝑠 ≤ 2, (91) The Canadian Applied Mathematics Quarterly, vol. 14, no. 2, pp.
∫𝑡 (1/𝑐 (𝑢)) Δ𝑢
1 129–147, 2006.
[12] L. Erbe, A. Peterson, and S. H. Saker, “Hille and Nehari
we obtain type criteria for third-order dynamic equations,” Journal of
Mathematical Analysis and Applications, vol. 329, no. 1, pp. 112–
∞ 131, 2007.
𝑡 𝜎 ∞ ∫ 𝑞 (V) ΔV
[ 𝑀𝛽 (𝑠)
lim sup ∫ [ ∫ 𝑢 Δ𝑢 [13] T. Li, Z. Han, S. Sun, and Y. Zhao, “Oscillation results for
𝑡→∞ 𝑡0 𝑏 (𝑠) 𝑠 𝑎 (𝑢) third order nonlinear delay dynamic equations on time scales,”
[ Bulletin of the Malaysian Mathematical Sciences Society, vol. 34,
2 𝜎(𝑠)
no. 3, pp. 639–648, 2011.
[(𝛽Δ (𝑠))+ ] 𝑐 (𝑠) ∫𝑡 (1/𝑐 (𝑢)) Δ𝑢 ] (92) [14] Y. Wang and Z. Xu, “Asymptotic properties of solutions of cer-
− 𝑠
1
] Δ𝑠 tain third-order dynamic equations,” Journal of Computational
𝜎
4𝛽 (𝑠) ∫𝑡 (1/𝑐 (𝑢)) Δ𝑢 and Applied Mathematics, vol. 236, no. 9, pp. 2354–2366, 2012.
1
]
[15] L. Erbe, T. S. Hassan, and A. Peterson, “Oscillation of third
𝑡
1 1 order nonlinear functional dynamic equations on time scales,”
≥ (2 − ) lim sup ∫ 2/5 Δ𝑠 = ∞.
4 𝑡 → ∞ 𝑡0 𝑠 Differential Equations and Dynamical Systems, vol. 18, no. 1-2,
pp. 199–227, 2010.
Hence, conditions (18), (20), (21), and (22) of Theorem 3 are [16] Y. Qi and J. Yu, “Oscillation criteria for fourth-order nonlinear
satisfied. By Theorem 3, we see that every solution 𝑥(𝑡) of delay dynamic equations,” Electronic Journal of Differential
Equations, vol. 2013, no. 79, pp. 1–17, 2013.
(81) is oscillatory if ≥ 1/2𝑑.
[17] S. R. Grace, M. Bohner, and S. Sun, “Oscillation of fourth-
order dynamic equations,” Hacettepe Journal of Mathematics
Acknowledgments and Statistics, vol. 39, no. 4, pp. 545–553, 2010.
[18] S. R. Grace, S. Sun, and Y. Wang, “On the oscillation of fourth
This project is supported by NNSF of China (11261005) order strongly superlinear and strongly sublinear dynamic
and NSF of Guangxi (2011GXNSFA018135, 2012GXNSFD equations,” Journal of Applied Mathematics and Computing,
A276040) and SF of ED of Guangxi (2013ZD061). 2013.
Abstract and Applied Analysis 11
Research Article
The Bernstein Operational Matrices for Solving
the Fractional Quadratic Riccati Differential Equations with
the Riemann-Liouville Derivative
Copyright © 2013 Dumitru Baleanu et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
We obtain the approximate analytical solution for the fractional quadratic Riccati differential equation with the Riemann-Liouville
derivative by using the Bernstein polynomials (BPs) operational matrices. In this method, we use the operational matrix for
fractional integration in the Riemann-Liouville sense. Then by using this matrix and operational matrix of product, we reduce
the problem to a system of algebraic equations that can be solved easily. The efficiency and accuracy of the proposed method are
illustrated by several examples.
𝛼
𝐵𝑖,𝑚 (𝑥) = ( 𝑚𝑖 ) 𝑥𝛼𝑖 (1 − 𝑥𝛼 )𝑚−𝑖 (𝑖 = 0, 1, . . . , 𝑚) that is different Definition 2 (see [4, 22–24]). Let 𝛼 ≥ 0 and 𝑛 = ⌈𝛼⌉; the
from the standard Bernstein polynomials. So, the operational operator 𝑎 𝐷𝑡𝛼 , defined by
matrices in this work are different from those in [18].
The organization of this paper is as follows. In Section 2, 𝛼 1 𝑑 𝑛 𝑡
𝑎 𝐷𝑡 𝑓 (𝑡) = ( ) ∫ (𝑡 − 𝑥)𝑛−𝛼−1 𝑓 (𝑥) 𝑑𝑥,
the Bernstein polynomials are introduced. Some basic defi- Γ (𝑛 − 𝛼) 𝑑𝑡 𝑎
nitions and properties of the fractional calculus and also the (7)
BPs operational matrix for the Riemann-Liouville fractional 𝑎 ≤ 𝑡 ≤ 𝑏,
integration are presented in Section 3. In Section 4, by BPs 0
operational matrices, we solve the fractional quadratic Riccati 𝑎 𝐷𝑡 𝑓 (𝑡) = 𝑓 (𝑡) ,
differential equation. In Section 5, we discuss the convergence
of the proposed method. In Section 6, several examples are is called the Riemann-Liouville fractional derivative operator
considered to evaluate the power and effectiveness of the of order 𝛼.
presented method. Some conclusions are summarized in the
Definition 3 (see [4, 22–24]). Let 𝛼 ≥ 0, 𝑛 = ⌈𝛼⌉, and
last section. 𝑐
𝑑𝑛 𝑓(𝑥)/𝑑𝑥𝑛 ∈ 𝐿 1 [𝑎, 𝑏]. The operator 𝑎 𝐷𝑡𝛼 , defined by
𝑐 𝛼 1 𝑡 𝑑𝑛 𝑓 (𝑥)
2. The Bernstein Polynomials and 𝑎 𝐷𝑡 𝑓 (𝑡) = ∫ (𝑡 − 𝑥)𝑛−𝛼−1 𝑑𝑥,
Their Properties Γ (𝑛 − 𝛼) 𝑎 𝑑𝑥𝑛
𝑎 ≤ 𝑡 ≤ 𝑏, (8)
On the interval [0, 1] we define the Bernstein polynomials
(BPs) of mth degree as follows [19]: 𝑐 0
𝑎 𝐷𝑡 𝑓 (𝑡) = 𝑓 (𝑡) ,
𝑚
𝐵𝑖,𝑚 (𝑥) = ( ) 𝑥𝑖 (1 − 𝑥)𝑚−𝑖 , 𝑖 = 0, 1, . . . , 𝑚. (3) is called the Caputo fractional derivative operator of order 𝛼.
𝑖
Set {𝐵0,𝑚 (𝑥), 𝐵1,𝑚 (𝑥), . . . , 𝐵𝑚,𝑚 (𝑥)} in the Hilbert space Lemma 4. If 𝛼 ≥ 0, 𝑛 = ⌈𝛼⌉, and 𝑎 ≤ 𝑡 ≤ 𝑏, then
𝐿2 [0, 1] is a complete basis. We can write Φ𝑚 (𝑥) = 𝐴𝑇𝑚 (𝑥), 𝑐
𝑇 (1) 𝑎 𝐷𝑡𝛼 𝑎 𝐼𝑡𝛼 𝑓 (𝑡) = 𝑓 (𝑡) , (9)
where 𝐴 is a matrix upper triangular, 𝑇𝑚 (𝑥) = [1, 𝑥, . . . , 𝑥𝑚 ] ,
𝑇
and Φ𝑚 (𝑥) = [𝐵0,𝑚 (𝑥), 𝐵1,𝑚 (𝑥), . . . , 𝐵𝑚,𝑚 (𝑥)] [20]. 𝑛−1
𝑓(𝑘) (𝑎)
𝑐
As a result, any polynomial of degree 𝑚 can be expanded (2) 𝑎 𝐼𝑡𝛼 𝑎 𝐷𝑡𝑎 𝑓 (𝑡) = 𝑓 (𝑡) − ∑ (𝑥 − 𝑎)𝑘 , (10)
in terms of linear combination of 𝐵𝑖,𝑚 (𝑥) (𝑖 = 0, 1, . . . , 𝑚) as 𝑘=0
𝑘!
given below: 𝑛−1
𝑐 𝑓(𝑘) (𝑎)
𝑚
(3) 𝑎 𝐷𝑡𝛼 𝑓 (𝑡) = 𝑎 𝐷𝑡𝛼 𝑓 (𝑡) − ∑ (𝑥 − 𝑎)𝑘−𝛼 .
𝑃 (𝑥) = ∑𝑐𝑖 𝐵𝑖,𝑚 (𝑥) = 𝑐𝑇 Φ𝑚 (𝑥) , (4)
𝑘=0
Γ (𝑘 − 𝛼 + 1)
𝑖=0 (11)
where
1 −1 1 Proof. See [22–24].
𝑐 = (∫ Φ (𝑥) Φ(𝑥)𝑇 𝑑𝑥) (∫ 𝑃 (𝑥) Φ (𝑥) 𝑑𝑥) . (5)
0 0 Theorem 5. One can get BPs operational matrix 𝐹𝛼 from
The approximation of functions within the Bernstein polyno- order (𝑚 + 1) × (𝑚 + 1) for the Riemann-Liouville fractional
mials and convergence analysis can be found in [20, 21]. integral as
𝑡
𝛼 1
𝑎 𝐼𝑡 Φ𝑚 (𝑡) = ∫ (𝑡 − 𝑥)𝛼−1 Φ𝑚 (𝑥) 𝑑𝑥 ≈ 𝐹𝛼 Φ𝑚 (𝑡) .
3. BPs Operational Matrix for the Γ (𝛼) 𝑎
(12)
Riemann-Liouville Fractional Integration
In this section, firstly, we give some basic definitions and Proof. See [21].
properties of the fractional calculus which are used further
in this paper.
4. BPs for Solving the Fractional Quadratic
𝛼 Riccati Differential Equation
Definition 1 (see [4, 22–24]). Let 𝛼 ≥ 0; the operator 𝑎 𝐼𝑡 ,
defined on 𝐿 1 [𝑎, 𝑏] by
Firstly, we use the initial conditions to reduce a given initial-
𝑡
𝛼 1 value problem to a problem with zero initial conditions.
𝑎 𝐼𝑡 𝑓 (𝑡) = ∫ (𝑡 − 𝑥)𝛼−1 𝑓 (𝑥) 𝑑𝑥, 𝑎 ≤ 𝑡 ≤ 𝑏, So, we define
Γ (𝛼) 𝑎 (6)
0 𝑦 (𝑡) = 𝑦̂ (𝑡) + 𝑧 (𝑡) ,
𝑎 𝐼𝑡 𝑓 (𝑡) = 𝑓 (𝑡) , (13)
is called the Riemann-Liouville fractional integral operator of ̂ is some known function that satisfied the initial
where 𝑦(𝑡)
order 𝛼. conditions (2) and 𝑧(𝑡) is a new unknown function.
Abstract and Applied Analysis 3
Substituting (13) in (1) and (2), we have an initial-value Therefore we can reduce (22) by (23)–(25) as
problem as follows:
𝐶𝑇 Φ𝑚 (𝑡) = 𝐹𝑇 Φ𝑚 (𝑡) + Φ𝑚 (𝑡)𝑇 𝐵̂ 𝐶𝛼 + Φ𝑚 (𝑡)𝑇 𝐴
̂𝐶̃𝛼 𝐶𝛼 .
𝛼
𝑎 𝐷𝑡 𝑧 (𝑡) = 𝑓̃ (𝑡) + ̃𝑏 (𝑡) 𝑧 (𝑡) + 𝑎̃ (𝑡) 𝑧2 (𝑡) , 𝑎 < 𝑡 ≤ 𝑏, (26)
(14)
Finally, we obtain the following nonlinear system of algebraic
subject to the initial conditions equation:
𝑧(𝑘) (𝑎) = 0, 𝑘 = 0, 1, . . . , ⌈𝛼⌉ − 1. (15) 𝐶 = 𝐹 + 𝐵̂ 𝐶𝛼 + 𝐴
̂𝐶̃𝛼 𝐶𝛼 , (27)
On the other hand, by (11) in Lemma 4 we can write such that by solving this system we can obtain the vector 𝐶.
𝑐 𝛼 𝛼 Then, we can get
𝑎 𝐷𝑡 𝑧 (𝑡) = 𝑎 𝐷𝑡 𝑧 (𝑡) . (16)
= 𝐼𝑡𝛼 𝑎 𝐷𝑡𝛼 𝑧 (𝑡) ≈ 𝐼𝑡𝛼 (𝐶𝑇 Φ𝑚 (𝑡)) By taking 𝑢(𝑡) = 𝑎 𝐷𝑡𝛼 𝑧(𝑡) we obtain the following fractional
(16) 𝑎 (18) 𝑎
integral equation:
= 𝐶𝑇 𝑎 𝐼𝑡𝛼 Φ𝑚 (𝑡) ≈ 𝐶𝑇 𝐹𝛼 Φ𝑚 (𝑡) 2
(12) 𝑢 (𝑡) = 𝑓̃ (𝑡) + ̃𝑏 (𝑡) 𝑎 𝐼𝑡𝛼 𝑢 (𝑡) + 𝑎̃ (𝑡) ( 𝑎 𝐼𝑡𝛼 𝑢 (𝑡)) ,
(31)
= 𝐶𝛼𝑇 Φ𝑚 (𝑡) , 𝑎 < 𝑡 ≤ 𝑏.
(21)
If we use the approximation 𝑢(𝑡) ≈ 𝐶𝑇 Φ𝑚 (𝑡), then the
where 𝐶𝛼𝑇 = 𝐶𝑇 𝐹𝛼 . problem (31) from space 𝐶1 [0, 1] reduces to the following
Now, by substituting (17)–(21) into (14), we obtain problem in space 𝑆𝑚 = Span{𝐵0,𝑚 (𝑡), 𝐵1,𝑚 (𝑡), . . . , 𝐵𝑚,𝑚 (𝑡)}:
𝐶𝑇 Φ𝑚 (𝑡) = 𝐹𝑇 Φ𝑚 (𝑡) + 𝐵𝑇 Φ𝑚 (𝑡) Φ𝑚 (𝑡)𝑇 𝐹𝛼𝑇 𝐶 + 𝐴𝑇 Φ𝑚 (𝑡) 𝐶𝑇 Φ𝑚 (𝑡) = 𝑓̃ (𝑡) + ̃𝑏 (𝑡) 𝑎 𝐼𝑡𝛼 (𝐶𝑇 Φ𝑚 (𝑡))
(32)
× (𝐶𝑇 𝐹𝛼 Φ𝑚 (𝑡) Φ𝑚 (𝑡)𝑇 𝐹𝛼𝑇 𝐶) 2
+ 𝑎̃ (𝑡) ( 𝑎 𝐼𝑡𝛼 (𝐶𝑇 Φ𝑚 (𝑡))) , 𝑎 < 𝑡 ≤ 𝑏.
= 𝐹𝑇 Φ𝑚 (𝑡) + 𝐵𝑇 Φ𝑚 (𝑡) Φ𝑚 (𝑡)𝑇 𝐶𝛼 + 𝐴𝑇 Φ𝑚 (𝑡)
Now, similar to Theorem 6.1 in [21], we propose the next
× (𝐶𝛼𝑇 Φ𝑚 𝑇
(𝑡) Φ𝑚 (𝑡) 𝐶𝛼 ) . theorem.
(22) Theorem 6. Suppose that 𝑢∗ (𝑡) ∈ 𝐶1 [0, 1] is the exact solution
of (31) and 𝜇𝑚 = 𝐽[𝑢𝑚 ] = Min𝑢∈𝑆𝑚 𝐽[𝑢], where
Then, from Lemma 3.5 in [20] we have
2
̃𝛼 ,
𝐶𝛼𝑇 Φ𝑚 (𝑥) Φ𝑚 (𝑥)𝑇 ≈ Φ𝑚 (𝑥)𝑇 𝐶 (23) 𝐽 [𝑢] = 𝑢 (𝑡) − 𝑓̃ (𝑡) − ̃𝑏 (𝑡) 𝑎 𝐼𝑡𝛼 𝑢 (𝑡) − 𝑎̃ (𝑡) ( 𝑎 𝐼𝑡𝛼 𝑢 (𝑡)) .
(33)
̂
𝐴𝑇 Φ𝑚 (𝑥) Φ𝑚 (𝑥)𝑇 ≈ Φ𝑚 (𝑥)𝑇 𝐴, (24)
𝑇 𝑇 𝑇̂
Then one has 𝜇𝑚 → 0 as 𝑚 → ∞ (i.e., 𝑢𝑚 (𝑡) → 𝑢∗ (𝑡) as
𝐵 Φ𝑚 (𝑥) Φ𝑚 (𝑥) ≈ Φ𝑚 (𝑥) 𝐵. (25) 𝑚 → ∞).
4 Abstract and Applied Analysis
Then by (23) and (25), (35) reduces to the following equation: 0.3
0.2
𝐶𝑇 Φ𝑚 (𝑡) = 𝐹𝑇 Φ𝑚 (𝑡) + (Φ𝑀(𝑡)𝑇 𝐵̂ + 𝐸̂𝐵𝑀) 𝐶𝛼 Exact for
(36) 0.1 𝛼=1
̃𝛼 + 𝐸̃𝑀 ) 𝐶𝛼 ) .
+ 𝐴𝑇 Φ𝑚 (𝑡) ((Φ𝑀(𝑡)𝑇 𝐶 0
𝐶𝛼
0 0.2 0.4 0.6 0.8 1
Equation (36) gets to (35) as 𝑀 → ∞, because from t
Lemma 3.1 in [21] 𝐸̂𝐵𝑀, 𝐸̃𝐶𝑀𝛼 → 0 as 𝑀 → ∞. Then by
Figure 2: Behavior of 𝑦10 (𝑡) in Example 1 for different 𝛼 and exact
deleting 𝐸̂𝐵𝑀, 𝐸̃𝐶𝑀𝛼 , taking 𝑀 = 𝑚, and using (24) in (36), we solution for 𝛼 = 1.
have
̂ 𝛼
𝐶𝑇 Φ𝑚 (𝑡) = 𝐹𝑇 Φ𝑚 (𝑡) + Φ𝑚 (𝑡)𝑇 𝐵𝐶
(37) subject to the initial condition as 𝑦(0) = 0. The exact solution
̂ + 𝐸̂𝑀) 𝐶
+ (Φ𝑀(𝑡)𝑇 𝐴 ̃𝛼 𝐶𝛼 ,
𝐴 of the equation for 𝛼 = 1 is given as
where, from Lemma 3.1 in [21], 𝐸̂𝐴 𝑀
→ 0 as 𝑀 → ∞. 𝑒2𝑡 − 1
Now, by taking 𝑀 = 𝑚 and deleting 𝐸̂𝐴 𝑀 𝑦 (𝑡) = . (39)
in (37), we get 𝑒2𝑡 + 1
(26). Obviously, if 𝑢̃𝑚 (𝑡) is solution of (32), then we have
𝑢̃𝑚 − 𝑢𝑚 → 0 as 𝑚 → ∞. Numerical results compared to [14] are given in Table 1 and
On the other hand, from Theorem 6 we obtained 𝑢̃𝑚 → also Figure 1 shows the absolute error for our method for 𝛼 =
𝑢∗ as 𝑚 → ∞. Therefore we can write 𝑢𝑚 → 𝑢∗ as 𝑚 → ∞ 1 and Figure 2 shows behavior 𝑦10 (𝑡) for different values of 𝛼.
and the proof is complete.
Example 2. Consider the following quadratic Riccati differ-
ential equation of fractional order [14]
𝛼
6. Illustrative Numerical Examples 0 𝐷𝑡 𝑦 (𝑡) = 2𝑦 (𝑡) − 𝑦(𝑡)2 + 1, 0 < 𝑡 ≤ 1, (40)
In this section, we apply our method with 𝑚 = 10 (BPs of subject to the initial condition as 𝑦(0) = 0. The exact solution
degree 𝑚 = 10) to solve the following examples. We define of the equation for 𝛼 = 1 is given as
𝑦𝑚 (𝑡) and 𝑦(𝑡) for the approximate solution and the exact
1 √2 − 1
solution, respectively. 𝑦 (𝑡) = 1 + √2 tanh (√2𝑡 + log ( )) . (41)
2 √2 + 1
Example 1. Consider the nonlinear Riccati differential equa-
tion [14]: Numerical results compared to [14] are given in Table 2 and
also, Figure 3 shows the absolute error for our method for 𝛼 =
𝛼
0 𝐷𝑡 𝑦 (𝑡) = −𝑦(𝑡)2 + 1, 0 < 𝑡 ≤ 1, (38) 1 and Figure 4 shows behavior 𝑦10 (𝑡) for different values of 𝛼.
Abstract and Applied Analysis 5
Table 1: Numerical results for 𝛼 = 1 and 𝑚 = 10 in Example 1 with Table 4: Numerical results for 𝛼 = 2.5 in Example 3 with
comparison to exact solution and [14]. comparison to [25, 26].
𝑡 Exact Present method Reference [14] 𝑡 Present method ADM [25] FDTM [26]
𝑚 = 10 𝑚 = 10
0.1 0.099668 0.099668 0.099668 0.1 0.000952 0.000952 0.000952
0.2 0.197375 0.197375 0.197375 0.2 0.005383 0.005383 0.005383
0.3 0.291313 0.291313 0.291313 0.3 0.014833 0.014833 0.014833
0.4 0.379949 0.379949 0.379944 0.4 0.030450 0.030450 0.030450
0.5 0.462117 0.462117 0.462078 0.5 0.053197 0.053197 0.053197
0.6 0.537050 0.537050 0.536857 0.6 0.083925 0.083925 0.083925
0.7 0.604368 0.604368 0.603631 0.7 0.123412 0.123412 0.123412
0.8 0.664037 0.664037 0.661706 0.8 0.172391 0.172391 0.172391
0.9 0.716298 0.716298 0.709919 0.9 0.231574 0.231574 0.231574
1 0.761594 0.761594 0.746032 1 0.301676 0.301676 0.301676
×10−7
Table 2: Numerical results for 𝛼 = 1 and 𝑚 = 10 in Example 2 with
comparison to exact solution and [14]. 1
𝑡 Present method
Exact Reference [14] 0.8
𝑚 = 10 Absolute error y(t)
0.1 0.110295 0.110295 0.110294 0.6
0.2 0.241977 0.241977 0.241965
0.3 0.395105 0.395105 0.395106 0.4
0.4 0.567812 0.567812 0.568115
0.2
0.5 0.756014 0.756014 0.757564
0.6 0.953566 0.953566 0.958259 0
0.7 1.152949 1.152949 1.163459 0 0.2 0.4 0.6 0.8 1
0.8 1.346364 1.365240 t
1.346364
0.9 1.526911 1.526911 1.554960 Figure 3: Behavior of the absolute error function in Example 2 for
1 1.689499 1.689499 1.723810 𝛼 = 1 and 𝑚 = 10.
Table 3: Numerical results for 𝛼 = 1.5 in Example 3 with with the initial conditions
comparison to [25–27].
𝑦(𝑘) (0) = 0, 𝑘 = 0, 1, . . . , 𝑛 − 1. (43)
𝑡 Present method ADM [25] FDTM [26] BPFs [27]
𝑚 = 10 This problem has been studied by using ADM [25], FDTM
0.1 0.023779 0.023790 0.023790 0.023800 [26], and BPFs [27]. Our results with 𝛼 = 1.5, 𝛼 = 2.5 are
0.2 0.067336 0.067330 0.067330 0.067335 compared to [25–27] in Tables 3 and 4. Therefore, we see that
0.3 0.123896 0.123896 0.123900
our method is very effective and obtained solutions that are
0.123886
in good agreement with the results in [25–27]. Also, Figure 5
0.4 0.191373 0.191362 0.191362 0.191368
shows behavior 𝑦𝑚 (𝑡) for different values of 𝛼.
0.5 0.268851 0.268856 0.268856 0.268862
0.6 0.356235 0.356238 0.356238 0.356244
0.7 0.453958 0.453950 0.453950 0.453956 7. Conclusion
0.8 0.562999 0.563007 0.563007 0.563014 In this paper, we proposed a numerical method for solving
0.9 0.685066 0.685056 0.685056 0.685067 the fractional quadratic Riccati differential equations by
1 0.822540 0.822511 0.822509 0.822525 the operational matrices of the Bernstein polynomials. We
applied operational matrix for fractional integration in the
Riemann-Liouville sense. Then by using this matrix and
operational matrix of product, we reduced the fractional
Example 3. Consider the nonlinear fractional differential
quadratic Riccati differential equation to a system of algebraic
equation:
equations that can be solved easily. Finally, examples have
been simulated to demonstrate the high performance of the
𝛼
0 𝐷𝑡 𝑦 (𝑡) = 𝑦2 (𝑡) + 1, 0 < 𝑡 ≤ 1, 𝑛 − 1 < 𝛼 ≤ 𝑛, (42) proposed method. We saw that the results were in good
6 Abstract and Applied Analysis
Research Article
Fractional Dynamics of Genetic Algorithms Using
Hexagonal Space Tessellation
J. A. Tenreiro Machado
Department of Electrical Engineering, Institute of Engineering, Polytechnic of Porto, Rua Dr. António Bernardino de Almeida 431,
4200-072 Porto, Portugal
Copyright © 2013 J. A. Tenreiro Machado. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
The paper formulates a genetic algorithm that evolves two types of objects in a plane. The fitness function promotes a relationship
between the objects that is optimal when some kind of interface between them occurs. Furthermore, the algorithm adopts an
hexagonal tessellation of the two-dimensional space for promoting an efficient method of the neighbour modelling. The genetic
algorithm produces special patterns with resemblances to those revealed in percolation phenomena or in the symbiosis found in
lichens. Besides the analysis of the spacial layout, a modelling of the time evolution is performed by adopting a distance measure and
the modelling in the Fourier domain in the perspective of fractional calculus. The results reveal a consistent, and easy to interpret,
set of model parameters for distinct operating conditions.
the search into the region of better performance. During the Bearing these ideas in mind this paper is organized
last decades a growing amount of successful application to as follows. Section 2 formulates the main algorithms and
real-world problems demonstrated that GAs are a powerful methods. Section 3 presents the experiments and analyzes the
and robust optimisation technique. results. Finally, Section 4 draws the main conclusions.
The hexagonal tessellation is a regular tiling of the
Euclidean plane, in which each vertex meets three hexagons 2. Main Algorithms and Methods
[17, 18]. There are two other regular tessellations of the plane,
namely, the triangular and the square tilings. Nevertheless, In this section are introduced briefly some aspects of FC and
the hexagonal tessellation constitutes the best way to divide Laplace transform and the computational implementation of
a given surface into regions of equal areas, while having GA.
the least total perimeter. This forms the so-called “hon-
eycomb conjecture” that dates back to the ancient Greek 2.1. Fractional Calculus. The most used definitions of a
mathematician Pappus of Alexandria (c. 290–c. 350) and fractional derivative of order 𝛼 are the Riemann-Liouville,
was proven in 1999 by Hales [19]. We find this structure Grünwald-Letnikov, and Caputo formulations [29]. Frac-
in nature, such as crystals or honeycombs, built by honey tional derivatives capture the history of past events, contrary
bees, and in man-made structures [20, 21], or even as art in to integer derivatives that are merely “local” operators. This
the famous Maurits Escher woodcuts and lithographs [22]. property has been recognized both in natural and man-made
Many other examples can be mentioned such as graphene and phenomena, where modelling becomes simpler using FC
superbenzene, substances with atoms arranged in a regular rather than building complicated integer order expressions.
hexagonal structure [23], or pineapples [24], a fruit with a Using the Laplace transform, for zero initial conditions,
rough skin having a hexagonal pattern of nodules. we have the expression
The three scientific concepts are put together for sim-
ulating and modelling an evolutionary process in a two- L {𝐷𝑡𝛼 𝑓 (𝑡)} = 𝑠𝛼 L {𝑓 (𝑡)} , (1)
dimensional space. First, it is considered a plane where some
kind of process evolves. The plane is discretized by means where 𝑠 and L denote the Laplace variable and operator,
of a regular hexagonal pattern, and the evolution consists of respectively.
the optimization using a standard GA. Second, the evolution In the scope of FC it is also important to mention the
of the GA population is described using a fractional order Mittag-Leffler function 𝐸𝛼 (𝑡) defined as [30–33]
model that approximates the numerical results. For that
∞
purpose, the best individual in each generation of the GA 𝑡𝑘
𝐸𝛼 (𝑡) = ∑ , 𝛼 ∈ C, Re (𝛼) > 0. (2)
population is analysed in the viewpoint of fitness function,
𝑘=0
Γ (𝛼𝑘 + 1)
compared with the previous case, and the result is converted
into the Fourier domain. An important aspect is also the The Mittag-Leffler function is a generalization of the
fitness function that measures the “performance” of each GA exponential and the power laws. The first occurs in phenom-
individual. In the two-dimensional space are considered two ena governed by integer dynamics, and the second emerges in
distinct types of objects, and it is assumed that a “good per- fractional dynamics. In particular, when 𝛼 = 1 yields 𝐸1 (𝑡) =
formance” corresponds to a spacial arrangement exhibiting 𝑒𝑡 , while, for large values of 𝑡, the asymptotic behaviour of the
some type of interface between them. By other words, it is ML leads to 𝐸𝛼 (−𝑡) ≈ (1/Γ(1 − 𝛼))(1/𝑡), 𝛼 ≠1, 0 < 𝛼 < 2.
assumed that some kind of cooperation, or synergy, exists Applying the Laplace transform
between the two objects, such that they should coexist close
to each other in space. The resulting time-space population 𝑠𝛼−1
reveals fractal characteristics and patterns resembling those L {𝐸𝛼 (±𝑎𝑡𝛼 )} = , (3)
𝑠𝛼 ∓ 𝑎
of percolation [25, 26], in the inanimate world, or of lichens,
when thinking in living organisms [27, 28]. The possible we verify the generalization from the exponential up to the
examples correspond only to possible interpretations of the Mittag-Leffler function, that is, from integer up to fractional
abstract algorithm implemented in the paper. Percolation powers of 𝑠.
is the phenomenon involved in the movement and filtering These results mean that standard methods in mod-
of fluids through porous materials. Nevertheless, in the last elling and control, such as transfer functions and frequency
years percolation brought a new light into many topics response, can be directly applied as long as we allow the
such as material science, epidemiology, or geology. On the substitution of integer orders by their fractional counterparts.
other hand, lichens are organisms consisting of two partners,
namely, a fungus and a green alga growing in a symbiotic 2.2. Genetic Algorithms. GAs are a computer method to
relationship. The body of a lichen consists of fungal filaments find approximate solutions in optimization problems. GAs
surrounding the cells of the algae. The basis of the symbiosis are implemented such that a population of 𝑁 possible
in lichens is that the fungus provides the algal protection and solutions evolves with successive iterations towards better
gains nutrients in return. Therefore, such examples are merely approximations. In the GA formulation it is necessary to
possible interpretations of the simulation results, but, in fact, define the genetic representation of the problem and the
an abstract formulation is the basis of the proposed study that fitness function that measures how successfully a given
primarily intends to model the GA evolution with FC tools. individual approximates the solution. In the GA execution
Abstract and Applied Analysis 3
0.7
0.6
0.5
0.4
𝑑(𝑡)
0.3
0.1
0
0 200 400 600 800 1000
𝑡
Re
0 10 20 30 40
0
−5
object type 2,” respectively. Moreover, different initializations
of the GA population, such that the three types of cells have
distinct probabilities, are also tested. Let us represent the −10
Im
90
0.012
80
70 0.01
60 0.008
𝐾 50 𝑝 0.006
40
30 0.004
20
0.002
10
0 0
0.1 0.1
0.8 0.8
𝑁
𝑁 (× 0.6
(× 0.6 10
10 00
00 )
) 0.4 𝑃0 0.4 𝑃0
10.2 1 0.2
Figure 6: Variation of the transfer function parameter 𝐾 versus 𝑁 Figure 7: Variation of the transfer function parameter 𝑝 versus 𝑁
and 𝑃0 . and 𝑃0 .
1.05
where 𝐾 denotes the gain, 𝑝 represents a pole of fractional
order 𝛼, and 𝜏 stands for a time delay. 1
For example, Figure 5 depicts the polar diagram of the 0.1
experimental result and approximation (6), that is, Re =
R{𝐷(𝚤𝜔)} versus Im = I{𝐷(𝚤𝜔)}, for 𝑁 = 100 and 𝑃0 = 0.8
{0.33, 0.33, 0.33}. Several experiments demonstrated that the 𝑁
(× 0.6
low frequency content of 𝐷(𝚤𝜔) is invariant with different 10
00
GA seeds, in opposition with the high frequency behaviour ) 0.4 𝑃0
that reflects the stochastic nature of the algorithm and reveals
noisy characteristics in the Fourier domain. Therefore, in the 10.2
sequel a bandwidth limitation is considered such that 𝜔 ≤ Figure 8: Variation of the transfer function parameter 𝛼 versus 𝑁
0.03. and 𝑃0 .
Given the large number of combinations of values for
𝑁 and 𝑃0 , the identification was performed automatically by
means of a second GA having a population of 500 individuals Figures 6, 7, 8, and 9 show the variation of the transfer
and terminating after calculating 500 iterations. Several tests function parameters {𝐾, 𝑝, 𝛼, 𝜏} versus 𝑁 and 𝑃0 .
demonstrated that good identification results were produced We observe that the parameters of the transfer function
by the fitness function: (6) have the following behaviour:
References
0.8
[1] S. G. Samko, A. A. Kilbas, and O. I. Marichev, Fractional Inte-
𝑁
(× 0.6 grals and Derivatives, Gordon and Breach Science, Amsterdam,
10
00 The Netherlands, 1993.
) 0.4 𝑃0
[2] K. S. Miller and B. Ross, An Introduction to the Fractional
1 0.2 Calculus and Fractional Differential Equations, John Wiley &
Sons, New York, NY, USA, 1993.
Figure 9: Variation of the transfer function parameter 𝜏 versus 𝑁 [3] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory and Ap-
and 𝑃0 . plications of Fractional Differential Equations, vol. 204 of North-
Holland Mathematics Studies, Elsevier Science B.V., Amsterdam,
The Netherlands, 2006.
(iii) 𝛼 has a fractional value and seems to be almost [4] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
independent of 𝑁 and grows slightly with 𝑃0 ; Calculus: Models and Numerical Methods, vol. 3 of Series on
(iv) 𝜏 has not a clear relationship with 𝑁 or 𝑃0 . It has a Complexity, Nonlinearity and Chaos, World Scientific, Boston,
Mass, USA, 2012.
small value and its average is 𝜏av = 0.3.
[5] A. B. Malinowska and D. F. M. Torres, Introduction to the Frac-
We verify that we can model the GA dynamical behaviour tional Calculus of Variations, Imperial College Press, London,
in terms of a simple fractional order model. In fact, stochastic UK, 2012.
results of the GA seem to be of minor influence in the [6] O. Heaviside, “On operators in physical mathematics,” Proceed-
proposed modelling scheme, not only due to the com- ings of the Royal Society, vol. 52, pp. 504–529, 1893.
prehensive variation of the parameters, but also because [7] R. L. Magin, Fractional Calculus in Bioengineering, Begell
several numerical experiments with distinct seeds lead to House, Redding, Calif, USA, 2006.
similar results. Nevertheless, we have a phenomenal mod- [8] C. M. Ionescu and R. De Keyser, “Relations between fractional-
elling perspective based on the analytical approximation and order model parameters and lung pathology in chronic obstruc-
supported by the outcome results. Therefore, a new challenge tive pulmonary disease,” IEEE Transactions on Biomedical
Engineering, vol. 56, no. 4, pp. 978–987, 2009.
is the inverse problem. By other words, the problem of
defining the fractional order model remains open and, as [9] C. Ionescu and J. T. MacHado, “Mechanical properties and
impedance model for the branching network of the sapping
a consequence, designing the GA rules that produce such
system in the leaf of Hydrangea Macrophylla,” Nonlinear
dynamical behaviour. In the scope of this problematic it can Dynamics, vol. 60, no. 1-2, pp. 207–216, 2010.
not be forgotten that the GA optimizes a symbiotic behaviour
[10] F. Mainardi, Fractional Calculus and Waves in Linear Viscoelas-
using a regular hexagonal space tessellation. Therefore, at a ticity: An Introduction to Mathematical Models, Imperial College
higher level several problems remain open such as the design Press, London, UK, 2010.
of other fitness functions, the effect of other tessellation [11] J. T. Machado, V. Kiryakova, and F. Mainardi, “Recent history of
methods, or the phenomena produced by a larger number of fractional calculus,” Communications in Nonlinear Science and
object types in the GA population. Numerical Simulation, vol. 16, no. 3, pp. 1140–1153, 2011.
[12] J. H. Holland, Adaptation in Natural and Artificial Systems, Uni-
4. Conclusions versity of Michigan Press, Ann Arbor, Mich, USA, 1975.
[13] D. E. Goldberg, Genetic Algorithms in Search Optimization, and
This paper presented a GA dynamical evolution and its Machine Learning, Addison-Wesley, Reading, Mass, USA, 1989.
description by means of a fractional model. The GA adopts [14] J. R. Koza, Genetic Programming: On the Programming of Com-
a tessellation of the space using regular hexagons in order to puters by Means of Natural Selection, MIT Press, Cambridge,
provide an efficient scheme to handle the neighbour cells in Mass, USA, 1992.
the numerical discretization. Furthermore, the GA includes [15] Z. Michalewicz, Genetic Algorithms + Data Structures = Evolu-
a fitness function that evaluates positively the interface tion Programs, Springer, Berlin, Germany, 1996.
Abstract and Applied Analysis 7
Research Article
An Operational Matrix Based on Legendre Polynomials for
Solving Fuzzy Fractional-Order Differential Equations
Copyright © 2013 Ali Ahmadian et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This paper deals with the numerical solutions of fuzzy fractional differential equations under Caputo-type fuzzy fractional
derivatives of order 𝛼 ∈ (0, 1). We derived the shifted Legendre operational matrix (LOM) of fuzzy fractional derivatives for the
numerical solutions of fuzzy fractional differential equations (FFDEs). Our main purpose is to generalize the Legendre operational
matrix to the fuzzy fractional calculus. The main characteristic behind this approach is that it reduces such problems to the
degree of solving a system of algebraic equations which greatly simplifies the problem. Several illustrative examples are included to
demonstrate the validity and applicability of the presented technique.
approximate or exact solution for FFDEs is a significant task generalized differentiability. Also, we introduce a suitable way
which has been aroused simultaneously with the emerging to estimate the nonlinear fuzzy fractional initial problems on
of FFDES, except for a few number of these equations, the interval [0, 1], by spectral shifted Legendre collocation
and we have hardship in finding their analytical solutions. method based on Legendre operational matrix, to find the
Consequently, there have been limited efforts to develop new approximate fuzzy solution. Finally, the accuracy of the pro-
methods for gaining approximate solutions which reasonably posed algorithms is demonstrated by several test problems.
estimate the exact solutions. Salahshour et al. [31] considered We note that the two shifted Legendre and shifted Jacobi
fuzzy laplace transforms for solving FFDEs under Riemann- operational matrices have been introduced by Saadatmandi
Liouville H-differentiability. Also Mazandarani and Kamyad and Dehghan [36] and Doha et al. [39], respectively, in the
[32] generalized the fractional Euler method for solving crisp concept. We, therefore, motivated our interest in the
FFDEs under Caputo-type derivative. shifted Legendre operational matrix in the fuzzy settings.
From another point of view, several methods have been This paper is organized as follows: In Section 2, we begin
exploited to solve fractional differential equations, and frac- by introducing some necessary definitions and mathematical
tional partial differential equations, fractional integrodiffer- preliminaries of the fuzzy calculus and fractional calculus.
ential equations such as Adomian’s decomposition method Some basic concepts, properties and theorems of fuzzy
[7], He’s variational iteration method [33], homotopy per- fractional calculus are presented in Section 3. Section 4 is
turbation method [34], and spectral methods [35, 36]. In devoted to the fuzzy Legendre functions and their properties.
this way, orthogonal functions have received considerable The shifted Legendre operational matrix of fuzzy fractional
attention in dealing with the various kinds of fractional derivative for solving fuzzy fractional differential equation is
differential equations. The main characteristic behind the obtained in Section 5. Section 6 illustrates the effectiveness
approach using this technique is that it reduces these prob- of the proposed method through solving several examples
lems to those of solving a system of algebraic equations thus which some of them are modelled based on the real phenom-
greatly simplifying the problem. Saadatmandi and Dehghan ena. Finally, a conclusion is given in the last section.
[36] presented the shifted Legendre operational matrix for
fractional derivatives and applied it with tau method for the 2. Preliminaries
numerical solution of fractional differential equations subject
to initial conditions. Also in [37–39], the authors derived new We give some definitions and introduce the necessary nota-
formulas using shifted Chebyshev polynomials and shifted tion which will be used throughout the paper, see, for exam-
Jacobi polynomials of any degree, respectively and applied ple, [40, 41]. Also for some definitions related to generalized
them together with tau and collocation spectral methods for fuzzy difference, one can find more in [42, 43].
solving multiterm linear and nonlinear fractional differential We denote the set of all real numbers by R. A fuzzy
equations. number is a mapping 𝑢̃ : R → [0, 1] with the following
The essential target of this paper is to recommend a properties:
suitable way to approximate FFDEs using a shifted Leg-
endre tau approach. This strategy demands a formula for (a) 𝑢̃ is upper semicontinuous,
fuzzy fractional-order Caputo derivatives of shifted Legendre (b) 𝑢̃ is fuzzy convex, that is, 𝑢̃(𝜆𝑥 + (1 − 𝜆)𝑦 ≥
polynomials of any degree which is provided and applied min{̃ 𝑢(𝑥), 𝑢̃(𝑦)} for all 𝑥, 𝑦 ∈ R, 𝜆 ∈ [0, 1],
together with the tau method for solving FFDEs with initial
conditions. Up till now, and to the best of our knowledge, (c) 𝑢̃ is normal, that is, ∃𝑥0 ∈ R for which 𝑢̃(𝑥0 ) = 1,
few methods corresponding to those mentioned previously (d) supp 𝑢̃ = {𝑥 ∈ R | 𝑢̃(𝑥) > 0} is the support of the 𝑢,
have been devoted to solve FFDEs and are traceless in and its closure cl (supp 𝑢̃) is compact.
the literature for FFDEs under Caputo differentiability. This
partially motivates our interest in the operational matrix of Let E be the set of all fuzzy number on R. The 𝛼-level set
fuzzy fractional derivative of shifted Legendre polynomials. of a fuzzy number 𝑢̃ ∈ E, 0 ≤ 𝑟 ≤ 1, denoted by [̃ 𝑢]𝑟 , is
Also another motivation is based on the reality that only a defined as
few terms of expansion of the shifted Legendre function is {𝑥 ∈ R | 𝑢̃ (𝑥) ≥ 𝑟} if 0 < 𝑟 ≤ 1
needed to reach to a high accuracy, therefore, it does not 𝑢]𝑟 = {
[̃ (1)
need to implement the method frequently for finding the cl (supp 𝑢̃) if 𝑟 = 0.
approximate results in each particular point.
It is clear that the 𝑟-level set of a fuzzy number is a closed
For finding the fuzzy solution, the shifted Legendre
operational matrix is generalized for the fuzzy fractional and bounded interval [̃ 𝑢(𝑟), 𝑢̃(𝑟)], where 𝑢̃(𝑟) denotes the
derivative (0 < 𝛼 < 1) which is based on the Legendre tau left-hand endpoint of [̃ 𝑢]𝑟 and 𝑢̃(𝑟) denotes the right-hand
method for solving numerically FFDEs with the fuzzy initial endpoint of [̃𝑢]𝑟 . Since each 𝑦 ∈ R can be regarded as a fuzzy
conditions. It is worthy to note here that the method based number 𝑦̃ defined by
on using the operational matrix of the Legendre orthogonal
function for solving FFDEs is computer oriented. 1 if 𝑡 = 𝑦,
𝑦̃ (𝑡) = { (2)
The aim of this paper is to introduce the shifted Legendre 0 if 𝑡 ≠𝑦,
operational matrix of fuzzy fractional derivative which is
based on Legendre tau method for solving FFDEs under R can be embedded in E.
Abstract and Applied Analysis 3
The addition and scalar multiplication of fuzzy number Definition 5 (see [47]). Let 𝑥, 𝑦 ∈ E. If there exists 𝑧 ∈ E such
in E are defined as follows: that 𝑥 = 𝑦 ⊕ 𝑧, and then 𝑧 is called the H-difference of 𝑥 and
𝑦, and it is denoted by 𝑥 ⊖ 𝑦.
𝑢 + ̃V, 𝑢̃ + ̃V) ,
(1) 𝑢̃ ⊕ ̃V = (̃
In this paper, the sign “⊖” always stands for H-difference
{(𝜆̃ 𝑢 (𝑟) , 𝜆̃
𝑢 (𝑟)) 𝜆 ≥ 0, (3)
and note that 𝑥 ⊕ 𝑦 ≠𝑥 + (−𝑦). Also throughout the paper
(2) (𝜆 ⊙ 𝑢̃) = {
(𝜆̃𝑢 (𝑟) , 𝜆̃
𝑢 (𝑟)) 𝜆 < 0. is assumed that the Hukuhara difference and Hukuhara
{ generalized differentiability existed.
The metric structure is given by the Hausdorff distance
𝐷 : E × E → R+ ⋃ 0, Definition 6 (see [42]). The generalized difference (g-
difference for short) of two fuzzy numbers 𝑢, V ∈ E is given
𝑢, ̃V) = sup max {𝑢̃ (𝑟) − ̃V (𝑟) , 𝑢̃ (𝑟) − ̃V (𝑟)} . (4)
𝐷 (̃ by its level sets as
𝑟∈[0,1]
It is easy to see that 𝐷 is a metric in E and has the following [𝑢 ⊖ g V]𝛼 = cl ⋃ ([𝑢]𝛽 ⊖gH [V]𝛽 ) ∀𝛼 ∈ [0, 1] , (8)
properties 𝛽≥𝛼
(i) 𝐷(̃
𝑢⊕𝑤 ̃, ̃V ⊕ 𝑤
̃) = 𝐷(̃𝑢, ̃V), for all 𝑢̃, ̃V, 𝑤
̃ ∈ E,
where the gH-difference ⊖gH is with interval operands [𝑢]𝛽
(ii) 𝐷(𝑘 ⊙ 𝑢̃, 𝑘 ⊙ ̃V) = |𝑘|𝐷(̃
𝑢, ̃V), for all 𝑘 ∈ R, 𝑢̃, ̃V ∈ E, and [V]𝛽 .
𝑢 ⊕ ̃V, 𝑤
(iii) 𝐷(̃ ̃ ⊕ 𝑒̃) ≤ 𝐷(̃ ̃) + 𝐷(̃V, 𝑒̃), for all 𝑢̃, ̃V, 𝑤
𝑢, 𝑤 ̃ ∈ E,
(iv) 𝐷(̃ ̃ ̃ ̃
𝑢 + ̃V, 0) ≤ 𝐷(𝑢, 0) + 𝐷(V, 0), for all 𝑢, V ∈ E, Proposition 7. The g-difference in Definition 6 is given by the
expression
(v) (E, 𝐷) is a complete metric space.
Definition 1. The property (iv) in the properties of the above [𝑢 ⊖ g V]𝛼 = [inf min {𝑢𝛽 − V𝛽 , 𝑢𝛽 − V𝛽 } ,
metric space suggests the definition of a function ‖ ⋅ ‖E : 𝑅 → 𝛽≥𝛼
̃ for all 𝑢 ∈ E that has the properties
E that ‖𝑢‖E := 𝐷(𝑢, 0), (9)
of usual norms. In [44], the properties of this function are
sup max {𝑢𝛽 − V𝛽 , 𝑢𝛽 − V𝛽 }] .
presented as follows: 𝛽≥𝛼
(i) ‖𝑢‖ ≥ 0, for all 𝑢 ∈ E and ‖𝑢‖ = 0 if and only if
𝑢 = 0̃, Proof. See [42].
(ii) ‖𝜆⋅𝑢‖ = |𝜆|⋅‖𝑢‖ and ‖𝑢+V‖ ≥ ‖𝑢‖+‖V‖, for all 𝑢, V ∈
E, for all 𝜆 ∈ R. The next proposition gives simplified notation for 𝑢 ⊖ g V
(iii) |‖𝑢‖ − ‖V‖| ≤ 𝐷(𝑢, V) and 𝐷(𝑢, V) ≤ ‖𝑢‖ + and V ⊖ g 𝑢.
‖V‖ for all 𝑢, V ∈ E.
Proposition 8. For any two fuzzy numbers 𝑢, V ∈ E the two g-
Definition 2 (see [45]). Let 𝑓 and 𝑔 be the two fuzzy- difference 𝑢 ⊖ g V and V ⊖ g 𝑢 exist and, for any 𝛼 ∈ [0, 1], one
number-valued functions on the interval [𝑎, 𝑏], that is, 𝑓, 𝑔 : 𝑢 ⊖ g V = −(V ⊖ g 𝑢) with
[𝑎, 𝑏] → E. The uniform distance between fuzzy-number-
valued functions is defined by [𝑢 ⊖ g V]𝛼 = [𝑑𝛼 , 𝑑𝛼 ] , [𝑢 ⊖ g V]𝛼 = [−𝑑𝛼 , −𝑑𝛼 ] , (10)
∗
𝐷 (𝑓, 𝑔) := sup 𝐷 (𝑓 (𝑥) , 𝑔 (𝑥)) . (5)
𝑥∈[𝑎,𝑏] where
Remark 3 (see [46]). Let 𝑓 : [𝑎, 𝑏] → E be fuzzy continuous.
Then from property (iv) of Hausdorff distance, we can define 𝑑𝛼 = inf (𝐷𝛼 ) , 𝑑𝛼 = sup (𝐷𝛼 ) , (11)
𝑟
𝐷 (𝑓 (𝑥) , 0̃) = sup max {𝑓𝑟 (𝑥) , 𝑓 (𝑥)} , ∀𝑥 ∈ [𝑎, 𝑏] . and the sets 𝐷𝛼 are
𝑟∈[0,1]
(6) 𝐷𝛼 = {𝑢𝛽 − V𝛽 | 𝛽 ≥ 𝛼} ∪ {𝑢𝛽 − V𝛽 | 𝛽 ≥ 𝛼} . (12)
Definition 4 (see [42]). Let 𝐾𝑐𝑛
be the space of nonempty
𝑛
compact and convex sets of R . The generalized Hukuhara Proof. See [42].
difference of two sets 𝐴, 𝐵 ∈ 𝐾𝑐𝑛 (gH-difference for short) is
defined as follows:
The following proposition prove that the g-difference is
(a) 𝐴 = 𝐵 + 𝐶 or well-defined.
𝐴 ⊖gH 𝐵 = 𝐶 ⇐⇒ { (7)
(b) 𝐵 = 𝐴 + (−1) 𝐶.
Proposition 9 (see [14]). For any fuzzy numbers 𝑢, V ∈ E the
In case (a) of the above equation, the gH-difference is g-difference 𝑢 ⊖ g V exists and it is a fuzzy number.
coincident with the H-difference. Thus the gH-difference is
a generalization of the H-difference. Proof. See [42].
4 Abstract and Applied Analysis
The following property holds for g-derivative. Theorem 13 (see [17]). Let 𝑓 : (𝑎, 𝑏) → E be a function and
denote [𝐹(𝑡)]𝑟 = [𝑓𝑟 (𝑡), 𝑔𝑟 (𝑡)], for each 𝑟 ∈ [0, 1]. Then
Proposition 10. Let 𝑢, V ∈ E be two fuzzy numbers, and then
(1) if 𝑓 is (1)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are
(i) 𝑢 ⊖ g V = 𝑢 ⊖gH V whenever the expressions on the right differentiable functions and
exist, in particular, 𝑢 ⊖ g 𝑢 = 0,
𝑟
(ii) (𝑢 + V)⊖ g V = 𝑢, [𝐹 (𝑡)] = [𝑓𝑟 (𝑡) , 𝑔𝑟 (𝑡)] , (17)
(iii) 0 ⊖ g (𝑢 ⊖ g V) = V ⊖ g 𝑢,
(iv) 𝑢 ⊖ g V = V ⊖ g 𝑢 = 𝑤 if and only if 𝑤 = −𝑤. (2) if 𝑓 is (2)-differentiable, then 𝑓𝑟 (𝑡) and 𝑔𝑟 (𝑡) are
Furthermore, 𝑤 = 0 if and only if 𝑢 = V. differentiable functions and
𝑟
In this paper, we consider the following definition which [𝐹 (𝑡)] = [𝑔𝑟 (𝑡) , 𝑓𝑟 (𝑡)] . (18)
was introduced by Bede and Gal in [14].
Definition 14 (see [42]). Let 𝑓 : (𝑎, 𝑏) → E and 𝑥0 ∈
Definition 11 (see [14]). Let 𝑓 : (𝑎, 𝑏) → E and 𝑥0 ∈ (𝑎, 𝑏). (𝑎, 𝑏). We say that 𝑓 is g-differentiable at 𝑥0 , if there exists
One says that 𝑓 is strongly generalized differentiable at 𝑥0 , if an element 𝑓 (𝑥0 ) ∈ E such that
there exists an element 𝑓 (𝑥) ∈ E, such that
𝑓 (𝑥 + ℎ) ⊖ g 𝑓 (𝑥)
(i) for all ℎ > 0 sufficiently small, ∃𝑓(𝑥0 + ℎ) ⊖ 𝑓 (𝑥0 ) = lim . (19)
𝑓(𝑥0 ), ∃𝑓(𝑥0 )⊖𝑓(𝑥0 −ℎ), and the limits (in the metric ℎ→0 ℎ
𝐷) Next we review one of the main results from Bede [15] for
𝑓 (𝑥0 + ℎ) ⊖ 𝑓 (𝑥0 ) fuzzy initial value problem (FIVP) under (1)-differentiability
lim which Nieto et al. [48] generalized this results for FIVP under
ℎ → 0+ ℎ
(13) (2)-differentiability (let ‖⋅‖ denote the usual Euclidean norm).
𝑓 (𝑥0 ) ⊖ 𝑓 (𝑥0 − ℎ)
= lim+ = 𝑓 (𝑥0 ) , Theorem 15 (see [15], characterization theorem). Let one
ℎ→0 ℎ
consider the fuzzy initial value problem
(ii) for all ℎ > 0 sufficiently small, ∃𝑓(𝑥0 ) ⊖ 𝑓(𝑥0 +
ℎ), ∃𝑓(𝑥0 − ℎ) ⊖ 𝑓(𝑥0 ), and the limits (in the metric 𝑦 = 𝑓 (𝑥, 𝑦 (𝑥)) ,
𝐷) (20)
𝑦 (𝑡0 ) = 𝑦0 ,
𝑓 (𝑥0 ) ⊖ 𝑓 (𝑥0 + ℎ)
lim+ where 𝑓 : [𝑥0 , 𝑥0 + 𝑎] × E → E is such that
ℎ→0 −ℎ
(14) 𝑟
𝑓 (𝑥0 − ℎ) ⊖ 𝑓 (𝑥0 ) (i) [𝑓(𝑥, 𝑦)]𝑟 = [𝑓𝑟 (𝑥, 𝑦, 𝑦), 𝑓 (𝑥, 𝑦, 𝑦)],
= lim+ = 𝑓 (𝑥0 ) ,
ℎ→0 −ℎ 𝑟
(ii) 𝑓𝑟 and 𝑓 are equicontinuous (i.e., for any 𝜖 > 0 there
(iii) for all ℎ > 0 sufficiently small, ∃𝑓(𝑥0 + ℎ) ⊖ is a 𝛿 > 0 such that |𝑓𝑟 (𝑥, 𝑦, 𝑧) − 𝑓𝑟 (𝑥1 , 𝑦1 , 𝑧1 )| < 𝜖
𝑟 𝑟
𝑓(𝑥0 ), ∃𝑓(𝑥0 −ℎ)⊖𝑓(𝑥0 ), and the limits (in the metric and |𝑓 (𝑥, 𝑦, 𝑧) − 𝑓 (𝑥1 , 𝑦1 , 𝑧1 )| < 𝜖 for all 𝑟 ∈ [0, 1],
𝐷) whenever (𝑥, 𝑦, 𝑧), (𝑥1 , 𝑦1 , 𝑧1 ) ∈ [𝑥0 , 𝑥0 + 𝑎] × R2 and
𝑓 (𝑥0 + ℎ) ⊖ 𝑓 (𝑥0 ) ‖(𝑥, 𝑦, 𝑧) − (𝑥1 , 𝑦1 , 𝑧1 )‖ < 𝛿 and uniformly bounded on
lim any bounded set,
ℎ → 0+ ℎ
(15) (iii) there exists an 𝐿 > 0 such that |𝑓𝑟 (𝑥2 , 𝑦2 , 𝑧2 ) −
𝑓 (𝑥0 − ℎ) ⊖ 𝑓 (𝑥0 )
= lim+ = 𝑓 (𝑥0 ) , 𝑓𝑟 (𝑥1 , 𝑦1 , 𝑧1 )| ≤ 𝐿 max{|𝑦2 − 𝑦1 |, |𝑧2 − 𝑧1 |} for all 𝑟 ∈
ℎ→0 −ℎ 𝑟 𝑟
[0, 1], |𝑓 (𝑥2 , 𝑦2 , 𝑧2 ) − 𝑓 (𝑥1 , 𝑦1 , 𝑧1 )| ≤ 𝐿 max{|𝑦2 −
(iv) for all ℎ > 0 sufficiently small, ∃𝑓(𝑥0 ) ⊖ 𝑓(𝑥0 + 𝑦1 |, |𝑧2 − 𝑧1 |} for all 𝑟 ∈ [0, 1].
ℎ), ∃𝑓(𝑥0 ) ⊖ 𝑓(𝑥0 − ℎ), and the limits (in the metric
𝐷) Then the FIVP (20) and system of ODEs
𝑓 (𝑥0 ) ⊖ 𝑓 (𝑥0 + ℎ) (𝑦𝑟 (𝑥)) = 𝑓𝑟 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
lim+
ℎ→0 −ℎ
(16) 𝑟
(𝑦𝑟 (𝑥)) = 𝑓 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
𝑓 (𝑥0 ) ⊖ 𝑓 (𝑥0 − ℎ)
= lim+ = 𝑓 (𝑥0 ) . (21)
ℎ→0 ℎ
𝑦𝑟 (𝑥0 ) = (𝑦0𝑟 ) ,
Remark 12. Throughout this paper, we say that 𝑓 is (1)-
differentiable on (𝑎, 𝑏), if 𝑓 is differentiable in the sense (i) 𝑦 (𝑥0 ) = (𝑦0𝑟 ) ,
of Definition 11 and also 𝑓 is (2)-differentiable on (𝑎, 𝑏), if 𝑓
is differentiable in the sense (ii) of Definition 11. are equivalent.
Abstract and Applied Analysis 5
Corollary 16 (see [48]). If we consider FIVP (20) under (2)- The matrix form of the above equations is
differentiability then the FIVP (20) and the following system of
ODEs are equivalent: 𝐴𝑋 = 𝑌, (28)
on [𝑎, ∞) and it is represented by (𝑓(𝑥; 𝑟), 𝑓(𝑥; 𝑟)). For any (29)
𝑛 𝑛
fixed 𝑟 ∈ [0, 1], assume that (𝑓(𝑥; 𝑟) and 𝑓(𝑥; 𝑟)) are Riemann- ∑𝑎𝑖𝑗 𝑥𝑗 = ∑𝑎𝑖𝑗 𝑥𝑗 = 𝑦𝑖 .
integrable on [𝑎, 𝑏] for every 𝑏 ≥ 𝑎, and assume that there are 𝑗=1 𝑗=1
two positive 𝑀(𝑟) and 𝑀(𝑟) such that:
If for a particular 𝑘, 𝑎𝑘𝑗 > 0, 1 ≤ 𝑗 ≤ 𝑛, we simply get
𝑏 𝑏
𝑛 𝑛
∫ 𝑓 (𝑥; 𝑟) 𝑑𝑥 ≤ 𝑀 (𝑟) , ∫ 𝑓 (𝑥; 𝑟) 𝑑𝑥 ≤ 𝑀 (𝑟) ∑ 𝑎𝑘𝑗 𝑥𝑗 = 𝑦𝑘 , ∑𝑎𝑘𝑗 𝑥𝑗 = 𝑦𝑘 .
𝑎 𝑎 (30)
𝑗=1 𝑗=1
for every 𝑏 ≥ 𝑎.
To solve fuzzy linear systems, one can refer to [51–53].
Definition 22 (see [54]). The Riemann-Liouville fractional Lemma 26. Let 0 < 𝛼 < 1 and 𝑓 ∈ 𝐶[𝑎, 𝑏] ∩ 𝐿 𝑝 [𝑎, 𝑏]. Then
derivatives of order 0 < 𝛼 < 1 of a function 𝑓 ∈ 𝐶[𝑎, 𝑏] the Caputo fractional derivatives are bounded for any 𝑥 ∈ [𝑎, 𝑏]
are expressed by and (1 ≤ 𝑝 < ∞) as
1 𝑑 𝑥 𝑓 (𝑡) 𝑑𝑡 𝑓 (𝑥)
𝛼
(𝐷𝑎+ 𝑓) (𝑥) = ∫ (𝑥 > 𝑎) , 𝑐 𝛼 𝑝
Γ (1 − 𝛼) 𝑑𝑥 𝑎 (𝑥 − 𝑡)𝛼 ( 𝐷𝑎+ 𝑓) (𝑥) ≤ (𝑥 − 𝑎)1−𝛼 = 𝑀𝑎𝛼 ,
|Γ (1 − 𝛼)| [1 − 𝛼]
𝑥𝑓
(𝑡) 𝑑𝑡 (40)
𝛼 1 𝑑 1
(𝐷𝑏− 𝑓)(𝑥) = − ∫ (𝑥 < 𝑏) . 𝑓 (𝑥)
Γ (1 − 𝛼) 𝑑𝑥 Γ (1 − 𝛼) 𝑎 (𝑡 − 𝑥)𝛼 𝑐 𝛼 𝑝 1−𝛼
( 𝐷𝑏+ 𝑓) (𝑥) ≤ |Γ (1 − 𝛼)| [1 − 𝛼] (𝑏 − 𝑥) = 𝑀𝑏𝛼 .
(33)
Definition 23 (see [55]). The fractional Caputo derivatives Proof. See [54, 57].
𝑐 𝛼
𝐷𝑎+ 𝑓(𝑥) and 𝑐 𝐷𝑏−
𝛼
𝑓(𝑥) on [𝑎, 𝑏] for 0 < 𝛼 < 1 are defined
via the above Riemann-Liouville fractional derivatives by 3. Fuzzy Caputo Fractional Derivatives
(𝑐 𝐷𝑎+
𝛼 𝛼
𝑓) (𝑥) = (𝐷𝑎+ [𝑓 (𝑡) − 𝑓 (𝑎)]) (𝑥) , In this section, some definitions and theorems related to the
𝛼
(34) fuzzy Caputo fractional derivatives are presented which are
( 𝑐 𝐷𝑏− 𝑓) (𝑥) = (𝐷𝑏−
𝛼
[𝑓 (𝑡) − 𝑓 (𝑏)]) (𝑥) , an extension of the fractional derivative in the crisp sense. The
generalized differentiability should be considered to expand
which can be simplified as the concept of Caputo fractional derivatives for the fuzzy
𝑓 (𝑎) space. For more details, see [14, 30].
𝛼
( 𝑐 𝐷𝑎+ 𝑓) (𝑥) = (𝐷𝑎+
𝛼
𝑓) (𝑥) − (𝑥 − 𝑎)−𝛼 ,
Γ (1 − 𝑎) Definition 27. Let 𝑓 : 𝐿E ∩ 𝐶E be a fuzzy set-value function
(35) 𝑥
𝛼 𝑓 (𝑏) and Φ(𝑥) = (1/Γ(1 − 𝛼)) ∫𝑎 (𝑓(𝑡)𝑑𝑡/(𝑥 − 𝑡)𝛼 ), and then 𝑓 is
( 𝑐 𝐷𝑏− 𝑓) (𝑥) = (𝐷𝑏−
𝛼
𝑓) (𝑥) − (𝑏 − 𝑥)−𝛼 . said to be g-Caputo fuzzy fractional differentiable at 𝑥, when
Γ (1 − 𝑎)
Also, the fractional Caputo derivative can be defined in a 𝛼
Φ (𝑥 + ℎ) ⊖ g Φ (𝑥)
( g 𝐷𝑎+ 𝑓) (𝑥) = lim , (41)
sense of integral form described in Definition 24. ℎ→0 ℎ
where
Definition 24 (see [56]). The Caputo definition of the
fractional-order derivative is defined as 𝑥 𝑓 (𝑡; 𝑟) 𝑑𝑡
g 𝛼 1
( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) =[ ∫ ],
𝑐 𝛼 1 𝑥 𝑓𝑛 (𝑡) Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼
𝐷 𝑓 (𝑥) = ∫ 𝑑𝑡, (42)
Γ (𝑛 − 𝛼) 0 (𝑥 − 𝑡)𝛼+1−𝑛 (36)
𝑥 𝑓
g 𝛼 1 (𝑡; 𝑟) 𝑑𝑡
𝑛 − 1 < 𝛼 ≤ 𝑛, 𝑛 ∈ N, ( 𝐷𝑎+ 𝑓) (𝑥; 𝑟) =[ ∫ ].
Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼
where 𝛼 > 0 is the order of the derivative and 𝑛 is the smallest
Remark 28. A fuzzy-valued function 𝑓 is 𝐶[1 − 𝛼]-
integer greater than 𝛼. For the Caputo derivative, we have
differentiable, if it is differentiable as in Definition 27, Case
𝑐
𝐷𝛼 𝐶 = 0, (𝐶 is a constant) , (37) (i), and it is 𝐶[2 − 𝛼]-differentiable, if it is differentiable as in
Definition 27, case (ii).
{ 0, for 𝛽 ∈ N0 , 𝛽 < ⌈𝛼⌉ ,
{
{
{ Theorem 29. Let 0 < 𝛼 < 1 and 𝑓 ∈ 𝐴𝐶E [𝑎, 𝑏], then Caputo
{
{ fuzzy fractional derivative exists almost everywhere on (𝑎, 𝑏)
𝑐 𝛼 𝛽
𝐷 𝑥 = { Γ (𝛽 + 1) 𝛽−𝛼 and for all 0 ≤ 𝑟 ≤ 1 we have
{
{ 𝑥 , for 𝛽 ∈ N0 , 𝛽 ≥ ⌈𝛼⌉
{
{ Γ (𝛽 + 1 − 𝛼)
{
{ or 𝛽 ∉ N, 𝛽 > ⌊𝛼⌋ . (𝐶𝐷𝑎𝛼+ 𝑓) (𝑥; 𝑟)
(38)
𝑥 𝑓 (𝑡) 𝑥 𝑓
1 1 (𝑡)
The ceiling function ⌈𝛼⌉ is used to denote the smallest integer =[ ∫ 𝛼 𝑑𝑡, ∫ 𝑑𝑡]
Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡) Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼
greater than or equal to 𝛼, and the floor function ⌊𝛼⌋ to denote
the largest integer less than or equal to 𝛼. Also N = {1, 2, . . .} (43)
and N0 = {0, 1, 2, . . .}.
when 𝑓 is (1)-differentiable, and
Definition 25 (see [1]). Similar to the differential equation of
integer order, the Caputo’s fractional differentiation is a linear (𝐶𝐷𝑎𝛼+ 𝑓) (𝑥; 𝑟)
operation, that is,
𝑥 𝑓 𝑥 𝑓 (𝑡)
1 (𝑡) 1
𝑐 𝛼 𝑐 𝛼
𝐷 (𝜆𝑓 (𝑥) + 𝜇𝑔 (𝑥)) = 𝜆 𝐷 𝑓 (𝑥) + 𝜇 𝐷 𝑔 (𝑥) , 𝑐 𝛼 =[ ∫ 𝑑𝑡, ∫ 𝑑𝑡]
(39) Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼 Γ (1 − 𝛼) 𝑎 (𝑥 − 𝑡)𝛼
where 𝜆 and 𝜇 are constants. (44)
Abstract and Applied Analysis 7
= [(𝐼𝑎1−𝛼 1−𝛼
+ 𝐷𝑓) (𝑥; 𝑟) , (𝐼𝑎+ 𝐷𝑓) (𝑥; 𝑟)] , (𝑐 𝐷𝑥𝛼+ 𝑦) (𝑥; 𝑟) = 𝑓𝑟 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
0
when 𝑓 is (2)-differentiable. 𝑟
(𝑐 𝐷𝑥𝛼+ 𝑦) (𝑥; 𝑟) = 𝑓 (𝑥, 𝑦𝑟 , 𝑦𝑟 ) ,
0
𝑛 𝑖𝛼 𝑐 (𝑛+1)𝛼 𝛼 (52)
𝑟 𝑥 𝑐 𝑖𝛼 𝑟 + 𝐷 𝑓 (𝑥0 ) (𝑛+1)𝛼 𝑦𝑟 (𝑥0 ) = (𝑐 𝑦0𝑟 ) ,
𝑓 (𝑥) = ∑ 𝐷 𝑓 (0 ) + 𝑥 ,
𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑛𝛼 + 𝛼 + 1)
𝑐 (𝑛+1)𝛼 𝛼
𝑦 (𝑥0 ) = (𝑐 𝑦0𝑟 ) .
𝑛 𝑖𝛼
𝑟 𝑥 𝑐 𝑖𝛼 𝑟 + 𝐷 𝑓 (𝑥0 ) (𝑛+1)𝛼
𝑓 (𝑥) = ∑ 𝐷 𝑓 (0 ) + 𝑥 ,
𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑛𝛼 + 𝛼 + 1) Proof. In the papers [15, 59], the authors proved for fuzzy
(49) ordinary differential equations and hybrid fuzzy differential
equations. The result for FFDEs is obtained analogously by
𝑟 𝑟
where 𝑐 𝐷𝛼 𝑓𝑟 (0)=𝑐 𝐷𝛼 𝑓𝑟 (𝑥)| 𝑥=0 , 𝑐 𝐷𝛼 𝑓 (0)=𝑐 𝐷𝛼 𝑓 (𝑥)| 𝑥=0 . using Theorem 2 in [15] and Theorem 3.1 in [59].
8 Abstract and Applied Analysis
L m (x)
differential equations by tau method. In this section, we 0
try to approximate fuzzy solution using shifted Legendre −0.2
polynomials under H-differentiability as follows. −0.4
−0.6
4.1. Properties of Shifted Legendre Polynomials. The Legendre
polynomials, denoted by 𝑃𝑛 (𝑧), are orthogonal with Legendre −0.8
weight function: 𝑤(𝑧) = 1 over [−1, 1], namely [60], −1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1 x
2
∫ 𝑃𝑛 (𝑧) 𝑃𝑚 (𝑧) 𝑑𝑧 = 𝛿 , (53)
−1 2𝑛 + 1 𝑛𝑚 m=3 m=6
m=4 m=7
where 𝛿𝑛𝑚 is the Kronecker function and can be specified with m=5 m=8
the help of following recurrence formula:
Figure 1: The shifted Legendre functions for different 𝑚.
𝑃0 (𝑧) = 1, 𝑃1 (𝑧) = 𝑧,
2𝑖 + 1 𝑖 (54)
𝑃𝑖+1 (𝑧) = 𝑧𝑃𝑖 (𝑧) − 𝑃 (𝑧) , 𝑖 = 1, 2, . . . . A function 𝑢(𝑥) of independent variable defined for 0 ≤ 𝑥 ≤ 1
𝑖+1 𝑖 + 1 𝑖−1 may be expanded in terms of shifted Legendre polynomials as
In order to use these polynomials on the interval [0, 1], 𝑚
we define the so-called shifted Legendre polynomials by 𝑢 (𝑥) = ∑𝑏𝑖 𝐿 𝑖 (𝑥) = Λ𝑇 Φ (𝑥) , (60)
introducing the change of variable 𝑧 = 2𝑥 − 1. Let the shifted 𝑖=0
Legendre polynomials 𝑃𝑛 (2𝑥 − 1) be denoted by 𝐿 𝑛 (𝑥). The where the shifted Legendre coefficients matrix 𝐵 and the
shifted Legendre polynomials are orthogonal with respect to shifted Legendre vector Φ(𝑥) are given by
the weight function 𝑤𝑠 (𝑥) = 1 in the interval (0, 1) with the
following orthogonality property: Λ𝑇 = [𝜆 1 , 𝜆 2 , . . . , 𝜆 𝑚 ] ,
1
(61)
1 Φ (𝑥) = [𝐿 0 (𝑥) , 𝐿 1 (𝑥) , . . . , 𝐿 𝑚 (𝑥)] .
𝑇
∫ 𝐿 𝑛 (𝑥) 𝐿 𝑚 (𝑥) 𝑑𝑥 = 𝛿 . (55)
0 2𝑛 + 1 𝑛𝑚
Also, the derivative of Φ(𝑥) can be expressed by
The shifted Legendre polynomials are generated from the
following three-term recurrence relation: 𝑑Φ (𝑥)
= 𝐷Φ (𝑥) , (62)
𝑑𝑥
(2𝑖 + 1) (2𝑥 − 1) 𝑖
𝐿 𝑖+1 (𝑡) = 𝐿 𝑖 (𝑥) − 𝐿 (𝑥) , where 𝐷 is the (𝑚+1)×(𝑚+1) operational matrix of derivative
𝑖+1 𝑖 + 1 𝑖−1
given by
𝑖 = 1, 2, . . . (56)
{ 2 (2𝑗 + 1) , for 𝑗 = 𝑖 − 𝑘,
𝐿 0 (𝑥) = 1, 𝐿 1 (𝑥) = 2𝑥 − 1. {
{
{ 𝑘 = 1, 3, . . . , 𝑚, if 𝑚 odd,
𝐷 = (𝑑𝑖𝑗 ) = { {
{
{ 𝑘 = 1, 3, . . . , 𝑚 − 1, if 𝑚 even,
The analytic form of the shifted Legendre polynomial 𝐿 𝑛 (𝑥) {
of degree 𝑛 is given by {0, otherwise.
(63)
𝑛 𝑛
(𝑛 + 𝑖)! 𝑥𝑖
𝐿 𝑛 (𝑥) = ∑(−1)𝑛+𝑖 = ∑ 𝑒 𝑥𝑖 , (57) The graph of some shifted Legendre polynomials (for 3 ≤ 𝑚 ≤
(𝑛 − 𝑖)! (𝑖!)2 𝑖=0 𝑖,𝑛
𝑖=0 8) shown in Figure 1 to depict their behaviors.
Now, we use the shifted Legender functions due to
in which
approximate a fuzzy function.
(𝑛 + 𝑖)!
𝑒𝑖,𝑛 = (−1)𝑛+𝑖 , (58)
(𝑛 − 𝑖)!(𝑖!)2 4.2. The Approximation of Fuzzy Function. In this section,
we propose a shifted Legendre approximation for the fuzzy-
where valued functions. To this end, we use Legendre’s nodes and
fuzzy shifted Legendre polynomials to calculate the fuzzy best
𝐿 𝑛 (0) = (−1)𝑛 , 𝐿 𝑛 (1) = 1. (59) approximation. For more details, see [61–67].
Abstract and Applied Analysis 9
Definition 34. For 𝑦 ∈ 𝐿E𝑝 (0, 1) ∩ 𝐶E (0, 1) and Legendre and also assume that 𝑦(𝑥) is 𝑐 [1 − 𝛼]-differentiable. Therefore
polynomial 𝐿 𝑛 (𝑥) a real valued function over (0, 1), the fuzzy using Theorem 32, we have
function is approximated by 𝑚 𝑖𝛼
𝑟
𝑦 (𝑥) − ∑ 𝑥 𝑖𝛼 𝑟 + 𝑥(𝑚+1)𝛼
𝐷 𝑦 (0 )
≤ 𝑀𝛼 ,
𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
∞∗
𝑦 (𝑥) = ∑ 𝑐𝑗 ⊙ 𝐿 𝑗 (𝑥) , 𝑥 ∈ (0, 1) , (64) 𝑚 𝑖𝛼
𝑗=0 𝑟
𝑦 (𝑥) − ∑ 𝑥 𝑖𝛼 𝑟 + 𝑥(𝑚+1)𝛼
𝐷 𝑦 (0 )
≤ 𝑀𝛼 ,
where the fuzzy coefficients 𝑐𝑗 are obtained by 𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
(71)
1
𝑐𝑗 = (2𝑗 + 1) ⊙ ∫ 𝑦 (𝑥) ⊙ 𝐿 𝑗 (𝑥) 𝑑𝑥, (65) in which |𝐷(𝑚+1)𝛼 𝑦𝑟 (𝑥0 )| < 𝑀𝛼 and |𝐷(𝑚+1)𝛼 𝑦𝑟 (𝑥0 )| < 𝑀𝛼 .
0
Proof. From Theorem 32, we have
in which 𝐿 𝑗 (𝑥) is the same as in (57), and ∑∗ means addition
𝑐 (𝑚+1)𝛼
with respect to ⊕ in E. 𝑚
𝑥𝑖𝛼 𝑐 𝑖𝛼 𝑟
𝐷 𝑦𝑟 (𝑥0 ) (𝑚+1)𝛼
𝑦𝑟 (𝑥) = ∑ +
𝐷 𝑦 (0 ) + 𝑥 ,
Remark 35. In actuality, only the first (𝑚 + 1)-terms shifted 𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
Legendre polynomials are considered. So we have 𝑐 (𝑚+1)𝛼
𝑚
𝑥𝑖𝛼 𝑐 𝑖𝛼 𝑟 𝐷 𝑦𝑟 (𝑥0 ) (𝑚+1)𝛼
𝑦𝑟 (𝑥) = ∑ +
𝐷 𝑦 (0 ) + 𝑥 ,
𝑚∗
𝑇 𝑖=0 Γ (𝑖𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
𝑦 (𝑥) ≃ 𝑦̃𝑚 (𝑥) = ∑ 𝑐𝑗 ⊙ 𝐿 𝑗 (𝑥) = 𝐶𝑚 ⊙ Φ𝑚 (𝑥) , (66) (72)
𝑗=0
and the following relation can be obtained:
hence 𝑚 𝑖𝛼
𝑟 +
𝑦 (𝑥) − ∑ 𝑥 𝑖𝛼 𝑟
𝐷 𝑦 (0 )
𝑚∗
𝑦𝑟 (𝑥) ≃ 𝑦̃𝑚
𝑟
(𝑥) = ∑ 𝑐𝑗𝑟 ⊙ 𝐿 𝑗 (𝑥) ,
𝑖=0 Γ (𝑖𝛼 + 1)
(67)
𝑗=0 𝑐 (𝑚+1)𝛼 𝑟
𝐷 𝑦 (𝑥0 ) (𝑚+1)𝛼 𝑥(𝑚+1)𝛼
𝑇
≤ 𝑥 ≤ 𝑀𝑟𝛼 ,
that the fuzzy shifted Legendre coefficient vector 𝐶𝑚+1 and Γ (𝑚𝛼 + 𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
shifted Legendre function vector Φ𝑚+1 (𝑥) are defined as 𝑚
𝑟 𝑥𝑖𝛼
𝑦 (𝑥) − ∑ 𝐷𝑖𝛼 𝑦𝑟 (0+ )
𝑇
𝐶𝑚 = [𝑐0 , 𝑐2 , . . . , 𝑐𝑚 ] , (68) Γ (𝑖𝛼 + 1)
𝑖=0
𝑐 (𝑚+1)𝛼 𝑟
Φ𝑚 (𝑥) = [𝐿 0 (𝑥) , 𝐿 1 (𝑥) , . . . , 𝐿 𝑚 (𝑥)] . (69) 𝐷 𝑦 (𝑥0 ) (𝑚+1)𝛼 𝑥(𝑚+1)𝛼
≤
𝑟
𝑥 ≤ 𝑀𝛼 ,
Γ (𝑚𝛼 + 𝛼 + 1) Γ (𝑚𝛼 + 𝛼 + 1)
Definition 36 (see [68]). A fuzzy-valued polynomial 𝑝̃∗ ∈ (73)
̃ is the best approximation to fuzzy function 𝑓 on 𝜒 =
∏ 𝑁 𝑟
{𝑥0 , 𝑥1 , 𝑥2 , . . . , 𝑥𝑁}, if that 𝑀𝑟𝛼 = |𝑐 𝐷(𝑚+1)𝛼 𝑦𝑟 (𝑥0 )| and 𝑀𝛼 = |𝑐 𝐷(𝑚+1)𝛼 𝑦𝑟 (𝑥0 )|.
max 𝐷 (𝑝̃∗ (𝑥𝑖 ) , 𝑓𝑖 ) = min { max 𝐷 (𝑝̃ (𝑥𝑖 ) , 𝑓𝑖 )} , Remark 39. If we consider Lemma 38 and define 𝑀𝛼 =
𝑟
𝑖=0,1,2,...,𝑁 ̃ ∏
𝑝∈ ̃
𝑁
𝑖=0,1,2,...,𝑁 max{𝑀𝑟𝛼 , 𝑀𝛼 }, then (71) can be stated in the following form,
(70) regarding to Section 2:
𝑚
̃ is the set of all fuzzy valued polynomials. 𝑥𝑖𝛼
in which ∏ 𝑁 𝐷 (𝑦 (𝑥) , ∑ 𝐷𝑖𝛼 𝑦 (0+ ))
𝑖=0 Γ (𝑖𝛼 + 1)
The problem is referred to as the best shifted Legendre
𝑚
approximation, as we use Legendre’s nodes. 𝑥𝑖𝛼
= sup max {𝑦𝑟 (𝑥) − ∑ 𝐷𝑖𝛼 𝑦𝑟 (0+ ) ,
𝑖=0 Γ (𝑖𝛼 + 1)
Theorem 37. The best approximation of a fuzzy function
𝑟∈[0,1]
𝑚 𝑖𝛼
based on the Legendre nodes exists and is unique. 𝑟 +
𝑦 (𝑥) − ∑ 𝑥 𝐷 𝑖𝛼 𝑟
𝑦 (0 ) }
Proof. The proof is an instantaneous outcome of Theo- 𝑖=0 Γ (𝑖𝛼 + 1)
rem 4.2.1 in [68].
𝑥(𝑛+1)𝛼 𝑟 𝑥(𝑛+1)𝛼
≤ sup max {𝑀𝑟𝛼 , 𝑀𝛼 }
Now, we want to show that the fuzzy approximation 𝑟∈[0,1] Γ (𝑛𝛼 + 𝛼 + 1) Γ (𝑛𝛼 + 𝛼 + 1)
converges of Legendre functions to function 𝑦(𝑥).
𝑥(𝑛+1)𝛼
E ≤ 𝑀𝛼 .
Lemma 38. Suppose that 𝑦(𝑥) ∈ 𝐴𝐶 (0, 1] ∩ and 𝐿E𝑝 (0, 1] Γ (𝑛𝛼 + 𝛼 + 1)
𝑐 𝛼 E (74)
𝐷 𝑦(𝑥) ∈ 𝐶 (0, 1], 0 < 𝛼 < 1, 0 ≤ 𝑥0 ≤ 𝑥, and 𝑥 ∈ (0, 1],
10 Abstract and Applied Analysis
Theorem 40. Let 𝑐 𝐷𝛼 𝑦(𝑥) ∈ 𝐶E (0, 1] ∩ 𝐿E𝑝 (0, 1], and 0 < From Remark 39 and Lemma 38, we have
𝛼 < 1. Also consider a sequence of finite dimensional fuzzy
𝑟 𝑀𝛼 1
space 𝑋E ⊂ 𝑋, 𝑚 ≥ 1, in which 𝑋E have dimension 𝑑𝑚+1 . 𝑦 (𝑥) − 𝑦̃𝑟 (𝑥) ≤
Additionally, 𝑋E have a basis {𝐿 𝑖 (𝑥)}𝑚 𝑚 Γ (𝑚𝛼 + 𝛼 + 1) √ (2 (𝑚 + 1) 𝛼 + 1) ,
𝑖=0 . If one assumes that
𝑦̃𝑚 (𝑥) = 𝐶𝑇 Φ is the best fuzzy approximation for fuzzy
function 𝑦(𝑥) from {𝐿 𝑖 (𝑥)}𝑚 𝑟 𝑀𝛼 1
𝑖=0 , then the error estimation is as 𝑦 (𝑥) − 𝑦̃𝑟 (𝑥) ≤
follows: 𝑚 Γ (𝑚𝛼 + 𝛼 + 1) √ (2 (𝑚 + 1) 𝛼 + 1) ,
(79)
lim 𝐷 (𝑦 (𝑥) , 𝑦̃𝑚 (𝑥)) = 0. (75)
𝑚→∞
and if 𝑚 → ∞, we get ‖𝑦𝑟 (𝑥) − 𝑦̃ 𝑟 (𝑥)‖ → 0, ‖𝑦𝑟 (𝑥) −
𝑚
𝑟
Proof. Let 𝑓 : 𝑅 → E be a fuzzy valued function such 𝑦̃𝑚 (𝑥)‖ → 0. Therefore, from Remark 39 and the definition
𝑟
that [𝑓(𝑥)]𝑟 = [𝑓𝑟 (𝑥), 𝑓 (𝑥)]. Also consider the fuzzy of Hausdorff distance in Section 2, it can be implied that
Taylor’s formula in Theorem 32, 𝑓𝑟 (𝑥) = ∑𝑚 𝑖𝛼
𝑖=0 (𝑥 /Γ(𝑖𝛼 + lim 𝐷 (𝑦 (𝑥) , 𝑦̃𝑚 (𝑥)) = 0,
𝑟 𝑚→∞ (80)
𝑖𝛼 𝑖𝛼 𝑟
1)) 𝑐 𝐷 𝑓𝑟 (0+ ) and 𝑓 (𝑥) = ∑𝑚 𝑖𝛼 𝑐 +
𝑖=0 (𝑥 /Γ(𝑖𝛼 + 1)) 𝐷 𝑓 (0 ).
From Lemma 38 we have which completes the proof.
𝑟 𝑥(𝑚+1)𝛼 Remark 41. The same result can be obtained for 𝑦(𝑥) under
𝑦 (𝑥) − 𝑓𝑟 (𝑥) ≤ 𝑀𝑟𝛼 , 𝑐
[2 − 𝛼]-differentiability.
Γ (𝑚𝛼 + 𝛼 + 1)
(76)
𝑟
𝑦 (𝑥) − 𝑓𝑟 (𝑥) ≤ 𝑀𝑟 𝑥(𝑚+1)𝛼 4.3. Operational Matrix of Caputo Fractional Derivative
𝛼 .
Γ (𝑚𝛼 + 𝛼 + 1)
Lemma 42. The fuzzy Caputo fractional derivative of order
𝑇 0 < 𝛼 < 1 over the shifted Legendre functions can be gained
From the assumption, 𝐶 Φ is the best fuzzy approximation
in the form of
to 𝑦 from {𝐿 𝑖 (𝑥)}𝑚
𝑖=0 , and 𝑓 ∈ 𝑋E , 𝑥 ∈ (0, 1]. So one has
𝑖
𝑐 Γ (𝑘 + 1) 𝑘−𝛼
𝑟 𝑥(𝑚+1)𝛼 𝐷𝛼 𝐿 𝑖 (𝑥) = ∑ 𝑒𝑘,𝑖
𝑥 ,
𝑦 (𝑥) − 𝑦̃𝑟 (𝑥) ≤ 𝑦𝑟 (𝑥) − 𝑓𝑟 (𝑥) ≤ 𝑀𝑟
(81)
, Γ (𝑘 − 𝛼 + 1)
𝑚 𝛼
Γ (𝑚𝛼 + 𝛼 + 1) 𝑘=0
where 𝑒𝑘,𝑖 = 0 when 𝑖 < ⌈𝛼⌉ and for 𝑖 ≥ ⌈𝛼⌉, and so one has
𝑟
𝑦 (𝑥) − 𝑦̃𝑟 (𝑥) ≤ 𝑦𝑟 (𝑥) − 𝑓𝑟 (𝑥) ≤ 𝑀𝑟 𝑥(𝑚+1)𝛼
𝑚 𝛼 , 𝑒𝑘,𝑖 = 𝑒𝑘,𝑖 .
Γ (𝑚𝛼 + 𝛼 + 1)
(77) Proof. Employing the analytic form of shifted Legendre
polynomials explained in Section 4.1 and Definition 25, we
and thus, taking into account Theorem 1 in [69] and above have:
relations, we have
𝑖
𝑟 2 𝐷𝛼 𝐿 𝑖 (𝑥) = ∑ 𝑒𝑘,𝑖 𝐷𝛼 ⊙ 𝑥𝑘−𝛼 . (82)
𝑦 (𝑥) − 𝑦̃𝑟 (𝑥)
𝑚 𝑘=0
𝑀𝑟,2 1
Lemma 43. Let 0 < 𝛼 < 1, and the integral of the product
≤ 𝛼
2
∫ 𝑥2(𝑚+1)𝛼 𝑑𝑥
Γ(𝑚𝛼 + 𝛼 + 1) 0 of the fuzzy Caputo fractional derivative with order 𝛼 over the
shifted legendre functions can be obtained by
𝑀𝑟,2
𝛼
≤ , 𝑗
Γ(𝑚𝛼 + 𝛼 + 1)2 (2 (𝑚 + 1) 𝛼 + 1)2 1 𝑖 𝑒𝑘,𝑖 𝑒𝑙,𝑗
Γ (𝑘 + 1)
∫ 𝐷𝛼 𝐿 𝑖 (𝑥) 𝐿 𝑗 (𝑥) 𝑑𝑥 = ∑ ∑ .
𝑟 2 (78) 0 (𝑘 + 𝑙 + 1) − 𝛼 Γ (𝑘 − 𝛼 + 1)
𝑦 (𝑥) − 𝑦̃𝑟 (𝑥) 𝑘=0 𝑙=0
𝑚 (83)
𝑟 2 Proof. Using Lemma 42 and the analytic form of shifted Leg-
≤ 𝑦𝑟 (𝑥) − 𝑓 (𝑥)
endre polynomials explained in Section 4.1, we can acquire:
𝑟,2
𝑀𝛼 1 1
≤ ∫ 𝑥2(𝑚+1)𝛼 𝑑𝑥 ∫ 𝐷𝛼 𝐿 𝑖 (𝑥) 𝐿 𝑗 (𝑥) 𝑑𝑥
Γ(𝑚𝛼 + 𝛼 + 1)2 0 0
𝑟,2 𝑗
(84)
𝑀𝛼
𝑖 𝑒𝑘,𝑖 𝑒𝑙,𝑗 Γ (𝑘 + 1) 1
(𝑘+𝑙+1)−𝛼−1
≤ . = ∑∑ ∫ 𝑥 𝑑𝑥.
Γ(𝑚𝛼 + 𝛼 + 1)2 (2 (𝑚 + 1) 𝛼 + 1)2 𝑘=0 𝑙=0
Γ (𝑘 − 𝛼 + 1) 0
Abstract and Applied Analysis 11
The operational matrix of different orthogonal functions where ∼ Λ is an (𝑚 + 1)-square product operational matrix
for solving differential equations was introduced in the crisp for the fuzzy vector Λ ̃ = [Λ, ̃ Using the above equation
̃ Λ].
concept [36, 37, 39]. Here, the Legendre operational matrix and by the orthogonal property equation (56), the elements
(LOM) in [36] is applied to the FFDEs using Caputo-type { ∼ Λ𝑖𝑗 }𝑚
𝑖,𝑗=0
can be computed from
derivative.
The Caputo fractional derivatives operator of order 0 <
𝑚 ∗
𝛼 < 1 of the vector Φ defined in (69) can be stated by ∼
Λ 𝑖𝑗 = (2𝑗 + 1) ∑ 𝜆 𝑘 ⊙ 𝑔𝑖𝑗𝑘 , (91)
𝐷𝛼 Φ (𝑥) ≃ 𝐷(𝛼) Φ (𝑥) , (85) 𝑘=0
where 𝐷(𝛼) is the (𝑚 + 1)-square operational matrix of where ∑∗ denotes the fuzzy summation and ⊙ indicates fuzzy
fractional Caputo’s derivative of Legendre functions. Regard- multiplication. Moreover, 𝑔𝑖𝑗𝑘 are obtained by
(𝛼)
ing the following theorem, the LOM elements 𝐷𝑖,𝑗 are
determined under Caputo fractional derivative. This theorem 1
is generalizing the operational matrix of derivatives of shifted 𝑔𝑖𝑗𝑘 = ∫ 𝐿 𝑖 (𝑥) 𝐿 𝑗 (𝑥) 𝐿 𝑘 (𝑥) 𝑑𝑥, (92)
0
Legendre given in Section 4.1 to the fractional calculus.
(𝛼)
that in the simpler form, it is given by
Theorem 44. Let Φ be Legendre functions vector. 𝐷𝑖,𝑗 is
the (𝑚 + 1)-square operational matrix of fractional Caputo’s 𝑑𝑗−𝑙 𝑑𝑙 𝑑𝑖−𝑙
(𝛼)
derivative of order 0 < 𝛼 < 1. Then the elements of 𝐷𝑖,𝑗 are {
{ , 𝑘 = 𝑖 + 𝑗 − 2𝑙;
{
{ (2𝑖 + 2𝑗 − 2𝑙 + 1) 𝑑𝑖+𝑗−𝑙
achieved as {
{
𝑔𝑖𝑗𝑘 ={ 𝑙 = 0, 1, . . . , 𝑗, (93)
𝑖 {
{
{
(𝛼)
𝐷𝑖,𝑗 = ∑ 𝜃𝑖,𝑗,𝑘 , (86) {0,
{ 𝑘 ≠𝑖 + 𝑗 − 2𝑙;
𝑘=⌈𝛼⌉ { 𝑙 = 0, 1, . . . , 𝑗,
in which ⟨𝐷𝛼 Φ(𝑥), Φ(𝑥)𝑇 ⟩ and 𝐸−1 are (𝑚 + 1)-square 5.1. General Linear FFDEs. Let us consider the general linear
matrixes defined as fuzzy fractional differential equation
1
⟨𝐷𝛼 Φ (𝑥) , Φ(𝑥)𝑇 ⟩ = {∫ 𝐷𝛼 𝐿 𝑖 (𝑥)𝐿 𝑗 (𝑥)𝑑𝑥} , (𝑐 𝐷0𝛼+ 𝑦) (𝑥) = 𝑎1 ⊙ 𝑦 (𝑥) ⊕ 𝑎2 ⊙ 𝑓 (𝑥) , 0 < 𝛼 ≤ 1,
0 𝑖,𝑗=0 (89) (94)
𝑦 (0) = 𝑦0 ∈ E,
𝐸−1 = diag {(2𝑖 + 1)}𝑚−1
𝑖=0 .
in which 𝑎𝑗 for 𝑗 = 1, 2 are fuzzy constant coefficients, 𝑦(𝑥) :
Hence, applying Lemma 43 and inserting the above matrixes
in the product 𝐷(𝛼) , the theorem be proved. 𝐿E ∩ 𝐶E is a continuous fuzzy-valued function, and 𝑐 𝐷0𝛼+
represents the fuzzy Caputo fractional derivative of order 𝛼.
Remark 45. If 𝛼 = 𝑛 ∈ N, then Theorem 44 gives the
mentioned result as in Section 4.1. Remark 47. In this section, considering more simplicity, it is
The following property of the product of two shifted assumed that for the fuzzy set-valued function 𝑦(𝑥) : 𝐼 ⊂
Legendre polynomials fuzzy vectors will also be applied R → 𝐿E ∩ 𝐶E , we present the functions 𝑦𝑟 , 𝑦𝑟 : 𝑅 → 𝐼 ⊂
which is the extension of the crisp case introduced in [36], 𝑅, 𝑟 ∈ [0, 1] by 𝑦−𝑟 (𝑥) = (𝑦(𝑥))𝑟− , 𝑦+𝑟 (𝑥) = (𝑦(𝑥))𝑟+ , for all 𝑥 ∈
𝐼, for all 𝑟 ∈ [0, 1], which are in the same previous manner,
̃ ≃ ∼ ΛΦ,
ΦΦ𝑇Λ (90) called the left and right 𝑟-cut functions of 𝑦(𝑥).
12 Abstract and Applied Analysis
Let 𝑟 ∈ [0, 1] and 0 < 𝛼 < 1. Then from (94), we have that Also, using relations (85) and (100), we obtain
𝑟 𝑟 𝑟
[𝑎1 ⊙ 𝑦 (𝑥)] = [(𝑎1 ⊙ 𝑦 (𝑥))− , (𝑎1 ⊙ 𝑦 (𝑥))+ ] 𝑐
𝐷𝛼 𝑦 (𝑥) ≃ 𝐶𝑇 ⊙ 𝐷𝛼 Φ𝑚 (𝑥) ≃ 𝐶𝑇 ⊙ 𝐷(𝛼) Φ𝑚 (𝑥) . (103)
𝑟 𝑟
= [(𝑎1 )− 𝑦(𝑥)𝑟− , (𝑎1 )+ 𝑦(𝑥)𝑟+ ] Substituting (100)–(103) in problem (94), the coefficients
(95) 𝑚
{𝑐𝑗(𝑟) } are specified by imposing the equation to be almost
= 𝑎1(𝑟) 𝑦(𝑥)(𝑟) , 𝑗=0
fuzzy exact in the Legendre operational matrix form. Now, we
𝑟 𝑟
[𝑎1 𝑦(𝑥)]± = (𝑎1 )± 𝑦(𝑥)𝑟± . establish fuzzy residual for the approximation of (94), when
[𝑦(𝑥)](𝑟) ≈ [𝑦̃𝑚 (𝑥)](𝑟) .
Therefore,
𝛼 𝑟 𝑟
( 𝑐 𝐷0+ 𝑦)± (𝑥) = (𝑎1 )± 𝑦(𝑥)𝑟± + (𝑎2 )± 𝑓(𝑥)𝑟± ,
𝑟
(96) Theorem 48. Let 𝑦𝑟 ∈ 𝐶E [0, 1] and 0 < 𝛼 ≤ 1, and then
(𝑟) (𝑟)
and thus [(𝑐 𝐷𝛼 𝑦̃𝑚 )(𝑥)] = [𝑅𝑚 (𝑥) ⊕ 𝑎2 ⊙ 𝑓̃𝑚 (𝑥) ⊕ 𝑎1 ⊙ 𝑦̃𝑚 (𝑥)] .
𝛼 𝑟 𝛼 𝑟
[( 𝑐 𝐷0+ 𝑦)− (𝑥) , ( 𝑐 𝐷0+ 𝑦)+ (𝑥)] (104)
(𝑟)
𝑟 𝑟 𝑟
= [(𝑎1 )− 𝑦(𝑥)𝑟− + (𝑎2 )− 𝑓(𝑥)𝑟− , (𝑎1 )+ 𝑦(𝑥)𝑟+ + (𝑎2 )+ 𝑓(𝑥)𝑟+ ] .
𝑟 Proof. Let 𝑟 ∈ [0, 1]. We have [(𝑐 𝐷0𝛼+ 𝑦̃𝑚 )(𝑥)] =
(97) [(𝑐 𝐷0𝛼+ 𝑦̃𝑚 )𝑟− (𝑥), (𝑐 𝐷0𝛼+ 𝑦̃𝑚 )𝑟+ (𝑥)], [𝑦̃𝑚 (𝑥)](𝑟) = (𝑟)
[𝑦𝑚− (𝑥),
(𝑟)
𝑦𝑚+ (𝑥)] and regarding to what was conversed previously,
Hence, we obtain
𝑟 𝑟
(𝑥) − (𝑎1 )± 𝑦̃𝑚 (𝑥)𝑟± − (𝑎2 )± 𝑓̃𝑚 (𝑥)𝑟± ,
𝛼 (𝑟)
𝛼 (𝑟) ( 𝑐 𝐷 𝑦̃𝑚± (𝑟)
) (𝑥) = 𝑅𝑚±
[( 𝑐 𝐷0+ 𝑦) (𝑥)]
(105)
𝛼 𝑟 𝛼 𝑟
= [( 𝑐 𝐷0+ 𝑦)− (𝑥) , ( 𝑐 𝐷0+ 𝑦)+ (𝑥)] so
𝑟 𝑟 𝑟 𝑟
= [(𝑎1 )− 𝑦(𝑥)𝑟− +(𝑎2 )− 𝑓(𝑥)𝑟− , (𝑎1 )+ 𝑦(𝑥)𝑟+ +(𝑎2 )+ 𝑓(𝑥)𝑟+ ] 𝛼
[( 𝑐 𝐷 𝑦̃𝑚 ) (𝑥)]
(𝑟)
(𝑟) (𝑟)
= [𝑎1 ⊙ 𝑦(𝑥)] + [𝑎2 ⊙ 𝑓(𝑥)] , 𝑟 ∈ [0, 1] , 𝛼
= [( 𝑐 𝐷 𝑦̃𝑚−
(𝑟)
) (𝑥) , ( 𝑐 𝐷 𝑦̃𝑚+
(𝑟) 𝛼
) (𝑥)]
(𝑟)
= [𝑎1 ⊙ 𝑦(𝑥) ⊕ 𝑎2 ⊙ 𝑓(𝑥)] , 𝑟 ∈ [0, 1] . (𝑟)
= [𝑅𝑚− (𝑟)
(𝑥) , 𝑅𝑚+ (𝑥)]
(98)
𝑟 𝑟
− [(𝑎1 )− 𝑦̃𝑚 (𝑥)𝑟− , (𝑎1 )+ 𝑦̃𝑚 (𝑥)𝑟+ ]
We can rewrite (94) in the operator form
𝑟 𝑟
(𝑎1(𝑟) 𝐼⊖ g 𝑐 𝐷𝛼 ) 𝑦(𝑟) = 𝑎2(𝑟) 𝑓(𝑟) , (99) − [(𝑎2 )− 𝑓̃𝑚 (𝑥)𝑟− , (𝑎2 )+ 𝑓̃𝑚 (𝑥)𝑟+ ]
(𝑟) (𝑟) (𝑟)
in which the fuzzy operator 𝑐 𝐷𝛼 = 𝐼1−𝛼 𝐷 is supposed to be = [𝑅𝑚 (𝑥)] ⊕ [(𝑎1 ) ⊙ 𝑦̃𝑚 (𝑥)] ⊕ [(𝑎2 ) ⊙ 𝑓̃𝑚 (𝑥)] .
compact on a fuzzy Banach space 𝑋E to 𝑋E [45, 70] and ⊖ g (106)
is the notation for g-difference.
For solving fuzzy fractional differential equation (94), Using g-difference, we have
(𝑟)
we attempt to find a fuzzy function 𝑦𝑚 ∈ 𝑋E ; there-
𝑐 𝛼 (𝑟) (𝑟)
fore, let ( 𝐷0+ 𝑦)(𝑥), 𝑦(𝑥) and 𝑓(𝑥) be approximated using [𝑅𝑚 (𝑥)] = [(𝑐 𝐷𝛼 𝑦̃𝑚 )(𝑥)]
Definition 34 as: (107)
(𝑟) (𝑟)
𝑚∗ ⊖ g [(𝑎1 ) ⊙ 𝑦̃𝑚 (𝑥)] ⊖ g [(𝑎2 ) ⊙ 𝑓̃𝑚 (𝑥)] ,
𝑇
𝑦 (𝑥) ≃ 𝑦̃𝑚 (𝑥) = ∑ 𝑐𝑗 ⊙ 𝐿 𝑗 (𝑥) = 𝐶𝑚 ⊙ Φ𝑚 , (100)
𝑗=0 or in the sense of fuzzy operator,
that (𝑟)
[𝑅𝑚 ] = (𝑎1(𝑟) 𝐼 ⊖ g 𝑐 𝐷𝛼 ) 𝑦(𝑟) ⊖ g 𝑎2(𝑟) 𝑓(𝑟) . (108)
𝑚∗
(𝑟)
[𝑦̃𝑚 (𝑥)] =∑ 𝑐𝑗(𝑟) ⊙ [𝐿 𝑗 (𝑥)] 𝑥 ∈ 𝐼 ⊂ 𝑅,
𝑗=0
(101)
𝑚∗ It is anticipated that the deriving fuzzy function [𝑦̃𝑚 (𝑥)](𝑟)
𝑓 (𝑥) ≃ 𝑓̃𝑚 (𝑥) = ∑ 𝑓𝑗 ⊙ 𝐿 𝑗 (𝑥) = 𝐹𝑚𝑇 ⊙ Φ𝑚 , will be a suitable approximation of the exact solution
(𝑟)
𝑗=0 ̃
[𝑦(𝑥)] . To this end, let 𝑋E = 𝐿2E ([0, 1]), and let ⟨⋅, ⋅⟩E
indicate the fuzzy inner product for 𝑋E . It is demanded that
where 𝐹𝑚+1 = [𝑓0 , 𝑓1 , . . . , 𝑓𝑚 ]𝑇 is obtained as (𝑟)
𝑅𝑚 satisfy
1
𝑓𝑗 = (2𝑗 + 1) ⊙ ∫ 𝑓 (𝑥) ⊙ 𝐿 𝑗 (𝑥) 𝑑𝑥. (102) (𝑟)
⟨𝑅𝑚 ̃
, 𝐿 𝑖 ⟩𝐸 = 0, 𝑖 = 0, 1, . . . , 𝑚 − 1, 𝑟 ∈ [0, 1] , (109)
0
Abstract and Applied Analysis 13
1
in which ⟨𝑅𝑚 (𝑟)
, 𝐿 𝑖 ⟩𝐸 = [(𝐹𝑅) ∫0 𝑅𝑚 (𝑥) ⊙ 𝐿 𝑖 (𝑥)𝑑𝑥](𝑟) . The left of this concept in the crisp context (see more in [69, 74–76]).
side is the shifted legendre coefficients associated with 𝐿 𝑗 . If Initially, we state the following lemma which can offer an
{𝐿 𝑖 }𝑚
𝑖=0 are the main members of shifted Legendre family Φ =
upper bound for approximating the error of Caputo fractional
{𝐿 𝑖 }∞ derivative. So we define the error vector 𝐸𝛼 as
𝑖=0 which is complete in 𝑋E , then (109) needs the main
(𝑟)
terms to be zero in the Fourier extension of 𝑅𝑚 with respect 𝑇
to Φ which is called tau method in the crisp context and it 𝐸𝛼 = 𝐷𝛼 Φ − 𝐷(𝛼) Φ = [𝐸0,𝛼 , 𝐸1,𝛼 , . . . , 𝐸𝑚,𝛼 ] , (114)
is in a similar manner with the meaning of fuzzy (for more
details, see [71–73]). where
To discover 𝑦̃𝑛(𝑟) , implementing (109) to (108), regarding 𝑚
the (100)–(103), we generate 𝑚 fuzzy linear equations as 𝐸𝑖,𝛼 = 𝐷𝛼 𝐿 𝑖 (𝑥) − ∑𝐷𝑖𝑗(𝛼) 𝐿 𝑗 (𝑥) , 𝑖 = 0, 1, . . . , 𝑚. (115)
𝑗=0
(𝑟)
⟨[𝑅𝑚 (𝑥)] , 𝐿 𝑖 (𝑥)⟩
Now, we can propose the next lemma by using Theorem 1
𝑐 𝛼 (𝑟) (𝑟) in [69] to depict the error bound of Caputo fractional
= ⟨[( 𝐷 𝑦̃𝑚 )(𝑥)] ⊖ g [(𝑎1 ) ⊙ 𝑦̃𝑚 (𝑥)] (110) derivative operator for the shifted Legendre polynomials.
(𝑟)
⊖ g [(𝑎2 ) ⊙ 𝑓̃𝑚 (𝑥)] , 𝐿 𝑖 (𝑥)⟩ = 0,
̃ Lemma 49. Let the error function of Caputo fractional deriva-
tive operator for Legendre polynomials 𝐸𝑖,𝛼 : [𝑥0 , 1] → R
for 𝑖 = 0, 1, . . . , 𝑚 − 1. Equation (110) generate (𝑚) set of fuzzy be 𝑚 + 1 times continuously differentiable for 0 < 𝑥0 ≤ 𝑥,
linear equations. These fuzzy linear equations can be acquired 𝑥 ∈ (0, 1]. Also 𝐸𝑖,𝛼 ∈ 𝐶𝑚+1 [𝑥0 , 1] and 𝛼 < 𝑚 + 1 and then the
as the following system of equations: error bound is presented as follows:
𝑚−1∗ |Γ (𝑖 + 1)| 𝑀𝛼 1
∑ 𝑐𝑗(𝑟) ⊙ {⟨𝐷(𝛼) 𝐿 𝑗 , 𝐿 𝑖 ⟩ − ⟨𝑎1(𝑟) 𝐿 𝑗 , 𝐿 𝑖 ⟩} 𝐸𝑖,𝛼 ≤ √ 𝑥−𝛼 .
|Γ (1 − 𝛼)| Γ (𝑚𝛼 + 𝛼 + 1) (2 (𝑚 + 1) 𝛼 + 1) 0
𝑗=0
(111) (116)
𝑚−1∗
= ∑ 𝑓𝑗(𝑟) ⊙ ⟨𝑎2(𝑟) 𝐿 𝑗 , 𝐿 𝑖 ⟩ , 𝑖 = 0, 1, . . . , 𝑚 − 1 Proof. Firstly, we acquire a bound for 𝑐
𝐷𝛼 𝑥𝑖 , regarding
𝑗=0 Definition 24, as
or
𝑐 Γ (𝑖 + 1) 𝑖−𝛼
𝐷𝛼 𝑥 𝑖 = 𝑥
𝑚−1∗ 1 Γ (𝑖 + 1 − 𝛼)
∑ 𝑐𝑗(𝑟) ⊙ {(𝐹𝑅) ∫ 𝐷(𝛼) 𝐿 𝑗 (𝑥) 𝐿 𝑖 (𝑥) 𝑑𝑡 (117)
𝑗=0 0 |Γ (𝑚 + 2)| −𝛼
≤ 𝑥 , 𝑖 = 0, 1, . . . , 𝑚.
|Γ (1 − 𝛼)| 0
1
− (𝐹𝑅) ∫ 𝑎1(𝑟) 𝐿 𝑗 (𝑥) 𝐿 𝑖 (𝑥) 𝑑𝑡} (112)
0 Then using Lemma 1 in [69], we have:
𝑚−1∗ 1 𝑖
= ∑ 𝑓𝑗(𝑟) ⊙ (𝐹𝑅) ∫ 𝑎2(𝑟) 𝐿 𝑗 (𝑥) 𝐿 𝑖 (𝑥) 𝑑𝑡. |Γ (𝑖 + 1)| −𝛼 𝑖
𝐷𝛼 𝐿 𝑖 (𝑥) = ∑ 𝑒𝑘,𝑖 𝐷𝛼 𝑥𝑘 ≤ 𝑥 ∑𝑒
𝑗=0 0
𝑘=0
|Γ (1 − 𝛼)| 0 𝑘=0 𝑘,𝑖
(118)
Afterwards, substitution of (100) in the initial condition of |Γ (𝑖 + 1)| −𝛼 |Γ (𝑖 + 1)| −𝛼
= 𝑥 𝐿 (1) = 𝑥 ,
(94) yields |Γ (1 − 𝛼)| 0 𝑖 |Γ (1 − 𝛼)| 0
𝑚∗
in which
𝑦 (0) = ∑ 𝑐𝑗(𝑟) ⊙ 𝐿 𝑗 (0) = 𝑦0 , (113)
𝑗=0 (𝑘 + 𝑖)!
𝑒𝑘,𝑖 = (−1)𝑘+𝑖 . (119)
that this equation be coupled with the previous fuzzy linear (𝑖 − 𝑘)!(𝑘!)2
equations and constructed (𝑚 + 1) fuzzy linear equations.
Now, utilizing Theorem 1 in [69], the lemma can be proved.
Clearly, after solving this fuzzy system, the coefficients {𝑐𝑗 }𝑚
𝑗=0
will be gained.
Therefore, the maximum norm of error vector 𝐸𝛼 is
5.1.1. Error Analysis. The aim of this section is to acquire the attained as
error bound for the Legendre approximation using shifted
Legendre polynomials. We consider the best shifted Legendre |Γ (𝑚 + 1)| 𝑀𝛼 1
approximation of a smooth fuzzy function under Caputo 𝐸𝛼 ∞ ≤ √ 𝑥−𝛼 .
derivative for 0 < 𝛼 ≤ 1 to reach the result. It should be |Γ (1 − 𝛼)| Γ (𝑚𝛼 + 𝛼 + 1) (2 (𝑚 + 1) 𝛼 + 1) 0
considered that the results of this section are the extension (120)
14 Abstract and Applied Analysis
Now we extend these results for the fuzzy case. Hence, the approximate and exact solutions of the fuzzy fractional
error of fuzzy Caputo fractional derivative is defined as differential equations, respectively. Then we rewrite (94) as
𝛼 (𝑟) (𝑟)
𝐷∗ (𝐷𝛼 Φ (𝑥) , 𝐷(𝛼) Φ (𝑥)) ( 𝑐 𝐷 𝑦𝑚±
(𝑟)
) (𝑥) − (𝑎1 )± 𝑦𝑚 (𝑥)(𝑟) (𝑟) (𝑟)
± = 𝐻𝑚± (𝑥) + (𝑎2 )± 𝑓𝑚 (𝑥)± ,
(124)
= sup 𝐷 (𝐷𝛼 Φ (𝑥) , 𝐷(𝛼) Φ (𝑥)) (𝑟)
𝑥∈[0,1] that 𝐻𝑚± (𝑥) is the fuzzy perturbation function that depends
only on 𝑦𝑚 (𝑥)(𝑟) ± . By subtracting (94) from above equation
= sup sup max {𝐷𝛼 Φ(𝑥)𝑟− (𝑥) − 𝐷(𝛼) Φ(𝑥)𝑟− (𝑥) , and using Lemmas 38 and 50, Theorem 40, and Definition 2,
𝑥∈[0,1] 𝑟∈[0,1]
one can obtain
𝛼 (𝑟)
𝐷 Φ(𝑥)𝑟+ (𝑥) − 𝐷(𝛼) Φ(𝑥)𝑟+ (𝑥)}
𝐻𝑚± (𝑥)E
𝛼
= sup max {𝐷 Φ(𝑥)𝑟− (𝑥) − 𝐷(𝛼) Φ(𝑥)𝑟− (𝑥)∞ , ≤ sup max {𝐷𝛼 Φ(𝑥)𝑟± − 𝐷𝛼 Φ(𝑥)𝑟± ∞ }
𝑟∈[0,1] 𝑟∈[0,1] (125)
𝛼
𝐷 Φ(𝑥)𝑟+ (𝑥) − 𝐷(𝛼) Φ(𝑥)𝑟+ (𝑥) } + (𝑎1 )± sup max {𝑦±𝑟 (𝑥) − 𝑦𝑚+1,±
(𝑟) 𝑟
(𝑥)∞ } .
∞
(121) 𝑟∈[0,1]
that 𝑀𝛼 = max{𝑀𝛼,− , 𝑀𝛼,+ }. We intend to find the fuzzy coefficients 𝐶𝑚 . Also by substitut-
ing (100) in the initial condition of nonlinear FFDE (128), we
Proof. It is straightforward from the definition of (𝐸𝛼 ), have
Definition 1, and Remark 3. 𝑦 (0) = 𝐶𝑇 Φ (0) = 𝑦 ,
0
(130)
For an error assessment of the approximation solution 𝑇
of (94), we assume that 𝑦𝑚 (𝑥) and 𝑦(𝑥) reveal the fuzzy 𝑦 (0) = 𝐶 Φ (0) = 𝑦0 .
Abstract and Applied Analysis 15
To find the approximate fuzzy solution 𝑦̃𝑚 (𝑥), we first Remark 53. If 𝛼 = 1, then the fuzzy fractional differential
collocate (129) at (𝑚) points. For appropriate collocation equations (132) are converted to fuzzy integer-order differen-
points we use the first (𝑚) shifted Legendre polynomials tial equations. So the exact solution of problem under (1)-
roots. These equations together with (130) generate (𝑚 + differentiability using Theorem 15 is as follows:
1) nonlinear fuzzy equations which can be solved using
Newton’s iterative method presented in [77]. 𝑌 (𝑥; 𝑟) = (0.5 + 0.5𝑟) 𝑒𝑥 ,
(135)
Remark 51. The solvability of system (129) and (130) is a 𝑌 (𝑥; 𝑟) = (1.5 − 0.5𝑟) 𝑒𝑥 , 0 < 𝑟 ≤ 1.
complicated problem and we cannot prove the existence and
uniqueness of such a fuzzy solution. But in our accomplish- By utilizing the technique explained in Section 5, (131) with
ment, we have solved this system, regarding the method in 𝜆 = 1 becomes:
[77], using MATLAB functions. In the assumed example, 𝑚
these functions have prospered to gain an accurate fuzzy ∑ 𝑐𝑗𝑟 [𝑑𝑖,𝑗
(𝛼)
− 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = 0 𝑖 = 0, . . . , 𝑚,
approximate solution of the system, even starting with a zero 𝑗=0
initial guess. (136)
𝑚
∑ 𝑐𝑟 𝑗 [𝑑𝑖,𝑗
(𝛼)
− 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = 0 𝑖 = 0, . . . , 𝑚,
6. Numerical Examples 𝑗=0
In this section, to demonstrate the effectiveness of the or we can rewrite it in the matrix form as
proposed method in the present paper, two different test
𝑇
examples are carried out. Also, the obtained numerical 𝐶𝑚,− [𝐷(𝛼) − 𝐼] Φ (𝑥) = 0,
solutions be compared with exact solutions. (137)
𝑇
𝐶𝑚,+ [𝐷(𝛼) − 𝐼] Φ (𝑥) = 0,
Example 52. Let us consider the following FFDE:
𝛼 in which (𝑐𝑗𝑟 ) = [𝑐𝑗,− 𝑟 𝑟
, 𝑐𝑗,+ ] and for 𝑗 = 0, 1, . . . , 𝑚. As it
( 𝑐 𝐷0+ 𝑦) (𝑥) = 𝜆 ⊙ 𝑦 (𝑥) , 0 < 𝛼 ≤ 1,
was described in Section 5, we produce 𝑚 fuzzy algebraic
(131)
equations multiplied above fuzzy residual system by 𝐿 𝑖 (𝑥) for
𝑦 (0; 𝑟) = [𝑦 (𝑟) , 𝑦0 (𝑟)] , 0 < 𝑟 ≤ 1,
0 𝑖 = 0, 1, . . . , 𝑚−1 using orthogonal property, and so we obtain
in which 𝑦(𝑥) : 𝐿E ∩ 𝐶E is a continuous fuzzy-valued function 𝑇
𝐶𝑚,− [𝐷(𝛼) − 𝐼] = 0,
and 𝑐 𝐷0𝛼+ denotes the fuzzy Caputo fractional derivative of (138)
order 𝛼. We solve this example according to two following 𝑇
𝐶𝑚,+ [𝐷(𝛼) − 𝐼] = 0.
cases for 𝜆 = 1, −1.
Case 1. Suppose that 𝜆 = 1, and then using 𝑐 [1 − 𝛼]- Now, using the initial condition (131), we have
differentiability and Theorem 33, we have the following: 𝑇
𝑦 (0; 𝑟) ≃ 𝐶𝑚,− Φ𝑚 = (0.5 + 0.5𝑟) ,
𝛼
( 𝑐 𝐷0+ 𝑦) (𝑥; 𝑟) = 𝑦 (𝑥) , 0 < 𝛼 ≤ 1, 𝑇
(139)
𝑦 (0; 𝑟) ≃ 𝐶𝑚,+ Φ𝑚 = (1.5 − 0.5𝑟) .
𝑦 (0; 𝑟) = 𝑦0 (𝑟) , 0 < 𝑟 ≤ 1,
(132) The above three equations generate a set of 𝑚 + 1 fuzzy linear
𝛼
( 𝑐 𝐷0+ 𝑦) (𝑥; 𝑟) = 𝑦 (𝑥) , 0 < 𝛼 ≤ 1, algebraic equations. As a result, for 𝛼 = 0.85 and 𝑚 = 3, one
can gain
𝑦 (0; 𝑟) = 𝑦0 (𝑟) , 0 < 𝑟 ≤ 1,
0 0 0 0
̃ 𝑟) = [0.5 + 0.5𝑟, 1.5 − 0.5𝑟].
in which 𝑦(0; [ 1.8639 0.3901 −0.1755 0.1095 ]
𝐷 (0.85)
=[
[−0.3901 4.5267 0.8696 −0.4078]
] (140)
where 𝐸𝛼,𝛼 is the classical Mittag-Leffler function Case 2. Suppose that 𝜆 = −1, then using 𝑐 [2 − 𝛼]-
differentiability and Theorem 33, the FDEs are obtained as
∞
𝑧𝑘 same as (132). Also the exact solution is given by
𝐸𝛼,𝛼 (𝑧) = ∑ . (134)
𝑘=0
Γ (𝛼 (𝑘 + 1))
𝑌 (𝑥; 𝑟) = (0.5 + 0.5𝑟) 𝐸𝛼,1 [−𝑥𝛼 ] , 0 < 𝛼 ≤ 1,
Note that for 𝛼 = 1 this representation is still valid since (141)
Γ(1) = 1 and 𝐸1,1 (𝑧) = 𝑒𝑧 . 𝑌 (𝑥; 𝑟) = (1.5 − 0.5𝑟) 𝐸𝛼,1 [−𝑥𝛼 ] , 0 < 𝑟 ≤ 1.
16 Abstract and Applied Analysis
10−1 10−2
10−2
Absolute error
Absolute error
10−3 10−3
10−4
10−5 10−4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cut r-cut
m=2 m=5
m=4 m=9
m=8 m = 11
Figure 2: The absolute error for different 𝑚 of Example 52, Case 1. Figure 4: The Absolute Error for different 𝑚 of Example 52, Case 2.
𝛼 = 0.85. 𝛼 = 0.75.
4.5
4 𝑌 (𝑥; 𝑟) = (0.5 + 0.5𝑟) 𝑒−𝑥 ,
3.5 (143)
3 𝑌 (𝑥; 𝑟) = (1.5 − 0.5𝑟) 𝑒−𝑥 , 0 < 𝑟 ≤ 1.
2.5
2 The approximate solution, exact solution and absolute errors
are depicted for Case 1 of Example 52 in Table 1 for 𝑥 = 1 with
1.5
𝛼 = .85. It can be seen that a few terms of the shifted Legendre
1
functions are required to achieve a suitable approximation
0.5 which demonstrate the applicability of the proposed method
0 0.2 0
0.2 0.4
0.4 0.6 0.8 0.8 0.6 for this problem. Also the absolute errors for 𝑚 = 2, 4, 8
1 1 r-cut
x with 𝛼 = 0.85 are plotted in Figure 2 which show the
decreasing of the error with the increasing of the number of
Figure 3: The Fuzzy approximate solution of Example 52, Case 1.
Legendre functions. The fuzzy approximate solution is shown
𝛼 = 0.85, 𝑚 = 4.
in Figure 3 for 𝛼 = 0.85, 𝑚 = 4.
̃ 𝑟) of
Remark 55. Figure 2 depicts the absolute errors for 𝑦(𝑥;
By applying the technique described in Sections 4 and 5, ̃ 𝑟),
Example 52, Case 1. In the same way, if we consider 𝑦(𝑥;
namely 𝑚 = 2, we may write the approximate solution for then analogously to the demonstration of Figure 2, we can
Cases 1 and 2 in the forms obtain the absolute errors.
Remark 54. If 𝛼 = 1, then the fuzzy fractional differential Remark 56. Figure 4 shows the absolute errors for 𝑦(𝑥; 𝑟) of
equation (131) with 𝜆 = −1 is converted to fuzzy integer-order Example 52, Case 2. In the same way, if we consider 𝑦(𝑥; 𝑟),
Abstract and Applied Analysis 17
Table 1: The result of the proposed method for Case 1 of Example 52 with 𝛼 = 0.85 and 𝑚 = 4.
Table 2: The result of the proposed method for Case 2 of Example 52 with 𝛼 = 0.75 and 𝑚 = 9.
0 < 𝛼 ≤ 1,
then analogously to the demonstration of Figure 4, we can
obtain the absolute errors. 𝑌 (𝑥; 𝑟) = (1.5 − 0.5𝑟) 𝐸𝛼,1 [𝑥𝛼 ]
𝑥
Example 57. Let us consider the following FFDE: + ∫ (𝑥 − 𝑡)𝛼−1 𝐸𝛼,𝛼 [−(𝑥 − 𝑡)𝛼 ] (𝑥 + 1) 𝑑𝑡,
0
(𝑐 𝐷0𝛼+ 𝑦) (𝑥) = 𝜆 ⊙ 𝑦 (𝑥) ⊕ (𝑥 + 1) , 0 < 𝛼 ≤ 1,
(144) 0 < 𝑟 ≤ 1.
𝑦 (0; 𝑟) = [𝑦̃ (𝑟) , 𝑦0 (𝑟)] 0 < 𝑟 ≤ 1. (146)
0
18 Abstract and Applied Analysis
Table 3: The result of the proposed method for Example 57 with 𝛼 = 0.95 and 𝑚 = 6.
Applying the shifted Legendre method with LOM technique Finally, the corresponding fuzzy approximate solution from
explained in Sections 4 and 5, we have (151) can be acquired.
For case 𝛼 = 0.95 and 𝑚 = 6, 𝑥 = 1 with 𝑟 = 0.1, we
𝑚 𝑚
obtain the fuzzy approximate solution in a series expansion
∑ 𝑐𝑗𝑟 [𝑑𝑖,𝑗
(𝛼)
+ 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = ∑ 𝑓𝑗𝑟 𝐿 𝑗 (𝑥; 𝑟) 𝑖 = 0, . . . , 𝑚,
as
𝑗=0 𝑗=0
10−5 10−3
10−4
−6
10
Absolute error
Absolute error
10−5
10−7
10−6
10−8 10−7
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r-cut r-cut
m=6 𝛼 = 0.75
m=9 𝛼 = 0.85
m = 12 𝛼 = 0.95
Figure 6: The absolute error for different 𝑚 of Example 57, 𝛼 = 0.95. Figure 8: The absolute error for different 𝛼 of Example 57, 𝑚 = 8.
1
1.8 0.9
Fuzzy approximate solution (y(x))
1.6 0.8
1.4 0.7
1.2 0.6
r-cut
1 0.5
0.8 0.4
0.3
0.6
0.2
0.4
0 0.1
0.5
x 0
1 0.8 1 1.15 1.2 1.25 1.3 1.35 1.4 1.45 1.5 1.55 1.6 1.65
0 0.2 0.4 0.6
r-cut y(1; r)
Figure 7: The fuzzy approximate solution of Example 57, 𝛼 = Figure 9: The approximate solution for 𝛼 = 0.75(∗), 0.85(∘),
0.95, 𝑚 = 6. and 0.95(+) of Example 57, 𝑚 = 8.
2𝑥2−𝛼 𝑥1−𝛼
Example 60. For our third example, consider the inhomoge- (𝑐 𝐷0+
𝛼
𝑦) (𝑥; 𝑟) + 𝑦 (𝑥; 𝑟) = − + 𝑥2 − 𝑥,
neous linear equation in [78] with fuzzy initial values, and so Γ (3 − 𝛼) Γ (2 − 𝛼)
we have:
𝑦 (0; 𝑟) = −1 + 𝑟, 0 < 𝛼 ≤ 1, 0 ≤ 𝑥 ≤ 1,
2𝑥2−𝛼 𝑥1−𝛼
Absolute error
×( − + 𝑥2 − 𝑥) 𝑑𝑡,
Γ (3 − 𝛼) Γ (2 − 𝛼) 10−10
0 ≤ 𝑟 ≤ 1,
𝑥 10−15
𝑦 (𝑥; 𝑟) = (1 − 𝑟) 𝐸𝛼,1 [−𝑥𝛼 ] + ∫ (𝑥 − 𝑡)𝛼−1 𝐸𝛼,𝛼 [−(𝑥 − 𝑡)𝛼 ]
0
2−𝛼 1−𝛼
2𝑥 𝑥 10−20
×( − + 𝑥2 − 𝑥) 𝑑𝑡, 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Γ (3 − 𝛼) Γ (2 − 𝛼)
r-cut
0 ≤ 𝑟 ≤ 1.
𝛼 = 0.65 𝛼 = 0.85
(156) 𝛼 = 0.75 𝛼 = 0.95
By exploiting the method proposed in Section 5, the Figure 10: The absolute error for different 𝛼 of Example 60, 𝑚 = 5.
equations are acquired by
𝑚 𝑚
∑ 𝑐𝑗𝑟 [𝑑𝑖,𝑗
(𝛼)
+ 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = ∑ 𝑓𝑗𝑟 𝐿 𝑗 (𝑥; 𝑟) 𝑖 = 0, . . . , 𝑚,
𝑗=0 𝑗=0 The approximate and exact solution of FFDEs (154) is
illustrated in Table 4 at 𝑥 = 1 with the absolute errors. In
𝑚 𝑚
𝑟 this table, three shifted Legendre polynomials are considered
∑𝑐𝑟 𝑗 [𝑑𝑖,𝑗
(𝛼)
+ 𝐼] 𝐿 𝑗 (𝑥; 𝑟) = ∑ 𝑓 𝑗 𝐿 𝑗 (𝑥; 𝑟) 𝑖 = 0, . . . , 𝑚, to derive the approximate solution. It is clear that with the
𝑗=0 𝑗=0
lower number of shifted Legendre functions, using proposed
(157) method, one can reach a resealable approximate solution.
or we can rewrite it in the matrix form as Also Figure 10 depicts the absolute error with 𝑚 = 5 and
𝑇 various values of 𝛼. This figure exhibits the applicability
𝐶𝑚,− [𝐷(𝛼) + 𝐼] Φ (𝑥) = 𝐹𝑚,−
𝑇
Φ (𝑥) ,
and validity of the proposed technique for this example.
(158) Furthermore, the fuzzy approximate solution for 𝛼 = 0.75
𝑇
𝐶𝑚,+ [𝐷(𝛼) + 𝐼] Φ (𝑥) = 𝐹𝑚,+
𝑇
Φ (𝑥) , with 𝑚 = 5 is shown in Figure 11 at 𝑥 = 1. Additionally,
in which (𝑐𝑗𝑟 ) = [𝑐𝑗,− 𝑟 𝑟
, 𝑐𝑗,+ ] and (𝑓𝑗𝑟 ) = [𝑓𝑗,−
𝑟 𝑟
, 𝑓𝑗,+ ] for 𝑗 = in Figure 12, the fuzzy approximate solution is compared
0, 1, . . . , 𝑚. As it was described in Section 5, we create 𝑚 fuzzy with different values of 𝛼. It can be seen that as 𝛼 tends to
algebraic equations multiplied in above system by 𝐿 𝑖 (𝑥) for 1, the solution of the fuzzy fractional differential equations
𝑖 = 0, 1, . . . , 𝑚−1 and implemented in the inner product using tends to that of the fuzzy integer-order differential equations.
orthogonal property, and so we gain: Finally, Figure 13 displays the absolute error for 𝛼 = 0.75 with
different values of 𝑚 which is obvious that with increasing the
𝑇
𝐶𝑚,− [𝐷(𝛼) + 𝐼] = 𝐹𝑚,−
𝑇
, number of shifted Legendre functions, the absolute error of
(159) the problem has been decreased gently.
𝑇
𝐶𝑚,+ [𝐷(𝛼) + 𝐼] = 𝐹𝑚,+
𝑇
.
Remark 61. Figures 10 and 13 illustrate the absolute errors for
Also for the initial condition (154), we have ̃ 𝑟) of Example 60. In the same way, if we consider 𝑦(𝑥;
𝑦(𝑥; ̃ 𝑟),
𝑇
𝑦 (0; 𝑟) ≃ 𝐶𝑚,− Φ𝑚 = −1 + 𝑟, then analogously to the demonstration of Figures 10 and 13,
(160) we can obtain the absolute errors.
𝑇
𝑦 (0; 𝑟) ≃ 𝐶𝑚,+ Φ𝑚 = 1 − 𝑟.
Now, we consider two applicable examples which are new
Equations (159) and (160) produce a set of 𝑚 + 1 fuzzy under uncertainty represented by fuzzy-valued functions.
linear algebraic equations. As a result, for 𝛼 = 0.75𝑟 = Firstly, a brief history is given in the deterministic case of
0.1 and 𝑚 = 3 the unknown coefficients 𝑐𝑗 can be Example 62. For more details, see [1, 79].
𝑇
achieved as 𝐶3,− = [−0.6914, 0.3147, 0.1144, 0.0083] and There are two kinds of electrical circuits which are related
𝑇
𝐶3,+ = [0.3725, −0.1418, 0.2954, −0.0903]. Also with these to the fractional calculus. Circuits of the first types are
assumptions, one has supposed to consist of capacitors and resistors, which are
described by conventional (integer-order) models. A circuit
0 0 0 0 expressing fractional order behavior is called fractance. Cir-
[ 1.7652 0.5884 −0.2263 0.1305 ]
𝐷 (0.75)
=[ ]
[−0.5884 3.6662 1.2114 −0.4979] . (161)
cuits of the second type may consist of resistors, capacitors,
and fractances. The term fractance was suggested initially
[ 1.5389 −0.6230 5.1081 1.8265 ] by Mehaute and Crepy [80]. Now, in order to scrutinize the
Abstract and Applied Analysis 21
Table 4: The result of the proposed method for Example 60 with 𝛼 = 0.95 and 𝑚 = 3.
1
Fuzzy approximate solution (y(x))
1
0.9
0.5 0.8
0.7
0
0.6
r-cut
0.5
−0.5
0.4
−1 0.3
1
0.2
0.5
x 0.1
0.8 1
0 0.4 0.6
0 0.2 0
r-cut −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4
Figure 11: The fuzzy approximate solution of Example 60, 𝛼 = y(1; r)
0.75, 𝑚 = 5. Figure 12: The fuzzy approximate solution for different 𝛼:(1: star,
0.95: square, 0.85: ∘, 0.75: ×, 0.65: pentagram) of Example 60, 𝑚 = 5.
10−10
𝑐 𝛼 𝑅
( 𝐷0+ 𝑦) (𝑥) = − 𝑦 (𝑥) + V (𝑥) , 0 ≤ 𝑥 ≤ 1,
𝐿 (162) 10−12
10−16
in which 𝑅 is the circuit resistance, and 𝐿 is a coefficient,
corresponding to the solenoid and 0 ≤ 𝑟 ≤ 1. Suppose that 10−18
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
V(𝑥) = sin(𝑥), 𝑅 = 1 ohm and 𝐿 = 1H, so (162) can be
r-cut
rewritten as
m=5 m = 14
𝑐 m = 10 m=3
( 𝐷0𝛼+ 𝑦) (𝑥) = −𝑦 (𝑥) + sin (𝑥) , 0 ≤ 𝑥 ≤ 1,
(163) Figure 13: The absolute error for different 𝑚 of Example 60, 𝛼 =
𝑦 (0; 𝑟) = [0.96 + 0.04𝑟, 1.01 + 0.01𝑟] . 0.75.
22 Abstract and Applied Analysis
Table 5: The result of the proposed method for Example 62 with 𝛼 = 0.85 and 𝑚 = 8.
Now, using [2 − 𝛼]-differentiability and Theorem 33, we conditions of (164), we can obtain the unknown fuzzy
have the following: coefficients {𝑐𝑗 }3𝑗=0 as
𝛼 𝑐0 = 0.7222, 𝑐1 = −0.0497,
( 𝑐 𝐷0+ 𝑦) (𝑥; 𝑟) + 𝑦 (𝑥; 𝑟) = sin (𝑥) ,
𝑐2 = 0.1272, 𝑐3 = −0.0650,
𝑦 (0; 𝑟) = (0.96 + 0.04𝑟) , 0 < 𝛼 ≤ 1, 0 ≤ 𝑥 ≤ 1, (168)
(164) 𝑐0 = 0.7488, 𝑐1 = −0.0611,
𝑐 𝛼
( 𝐷0+ 𝑦) (𝑥; 𝑟) + 𝑦 (𝑥; 𝑟) = sin (𝑥) ,
𝑐2 = 0.1317, 𝑐3 = −0.0674,
𝑦 (0; 𝑟) = (1.01 + 0.01𝑟) , 0 < 𝛼 ≤ 1, 0 ≤ 𝑥 ≤ 1.
and finally, the fuzzy approximate solution is given by
The exact solution of (163) under (2)-differentiability for
integer order is given by 𝑦 (𝑥, 0.1) = 0.9640 − 1.6420𝑥 + 2.7118𝑥2 − 1.2990𝑥3 ,
4
10−2 compartment does not appears, it is not often clear if the link
model is actually linked with a site of action, or if it exhibits
some pharmacodynamic related delay, or a combination of
10−3 the two. It is begun by representing drug concentration in
the effect compartment by the (Caputo) fractional differential
Absolute error
equation as follows:
10−4
(𝛼 𝐷0+
𝛼
𝐶𝑒) (𝑡) = 𝑘eo (−𝐶𝑒 (𝑡) + 𝑓 (𝑡)) , (170)
(171)
1.1
1 in which 𝑓(𝑡) = (Dose 𝑘𝑎 /𝑉(𝑘el −𝑘𝑎 ))(𝑒−𝑘el 𝑡 −𝑒−𝑘𝑎 𝑡 ) represents
0.9 drug concentration in the central compartment following, for
0.8 example, an oral administration of a dose, in this example
0.7 dose = 1 and parameters values are 𝑉 = 1, 𝑘𝑎 = 5.0, 𝑘el =
0.5, and 𝑘eo = 1.
1
0.8 The exact solution of (171) with 𝛼 = 1 under generalized
0.6 1
differentiability is given by
0.4 0.8
x 0.6
0.2 0.4 20 𝑡/2 5 −4𝑡
0 0
0.2 r-cut 𝑌 (𝑡; 𝑟) = 𝑒−𝑡 (− 𝑒 − 𝑒 + 3 + 0.5𝑟) ,
9 18
(172)
Figure 15: The fuzzy approximate solution of Example 62, 𝛼 = 20 5
0.75, 𝑚 = 10. 𝑌 (𝑡; 𝑟) = 𝑒 (− 𝑒𝑡/2 − 𝑒−4𝑡 + 4 − 0.5𝑟) .
−𝑡
9 18
We apply the method presented in Section 5 and solve this
10−2
problem for 𝑚 = 3, 𝛼 = 0.95, and 𝑟 = 0.2. Hence, we have
0 0 0 0
[ 1.9566 0.1432 −0.0743 0.0501 ]
𝐷 (0.95)
=[ ]
[−0.1432 5.4849 1.3447 −0.1823] ,
10−3
Absolute error
(173)
[ 1.8823 −0.2015 8.8258 0.5752 ]
𝐹4𝑇 = [−0.6537 −0.1886 0.2729 −0.1322] ,
10−4
and thereafter the fuzzy unknown coefficients for the fuzzy
approximation of (171) can be acquired easily by replacing the
above results in (110) and solving this algebraic fuzzy linear
equations system. So we have
10−5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
𝑐0 = 0.1744, 𝑐1 = −0.4368,
r-cut
m=3 m=9 𝑐2 = 0.0573, 𝑐3 = 0.0185,
m=6 m = 12 (174)
𝑐0 = 0.6417, 𝑐1 = −0.6635,
Figure 16: The absolute error for different 𝑚 of Example 62, 𝛼 =
0.85. 𝑐2 = 0.1043, 𝑐3 = 0.0095,
24 Abstract and Applied Analysis
Table 6: The result of the proposed method for Example 63 with 𝛼 = 0.95 and 𝑚 = 12.
𝑟
𝑟 𝑦1 2 𝑌(𝑟) Error 𝑦1 2𝑟 𝑌(𝑟) Error
0 −0.243121 −0.243123 1.232190𝑒 − 6 0.128454 0.128450 3.891957𝑒 − 6
0.1 −0.224543 −0.224544 1.365176𝑒 − 6 0.109875 0.109871 3.758969𝑒 − 6
0.2 −0.205964 −0.205965 1.498166𝑒 − 6 0.091296 0.091293 3.625981𝑒 − 6
0.3 −0.187385 −0.187387 1.631153𝑒 − 6 0.072717 0.072714 3.492989𝑒 − 6
0.4 −0.168806 −0.168808 1.764141𝑒 − 6 0.054139 0.054135 3.360001𝑒 − 6
0.5 −0.150227 −0.150229 1.897131𝑒 − 6 0.035560 0.035557 3.227013𝑒 − 6
0.6 −0.131648 −0.131651 2.030121𝑒 − 6 0.016981 0.016978 3.094025𝑒 − 6
0.7 −0.113070 −0.113072 2.163107𝑒 − 6 −0.001597 −0.001600 2.961033𝑒 − 6
0.8 −0.094491 −0.094493 2.296095𝑒 − 6 −0.020176 −0.020178 2.828050𝑒 − 6
0.9 −0.075912 −0.075914 2.429084𝑒 − 6 −0.038754 −0.038757 2.695061𝑒 − 6
1 −0.057333 −0.057336 2.562075𝑒 − 6 −0.057333 −0.057336 2.562075𝑒 − 6
𝛼
( 𝑐 𝐷0+ 𝑦) (𝑥) = 3𝐴 ⊙ 𝑦2 , 0 < 𝛼 < 1, 0 ≤ 𝑥 ≤ 1, 𝑖 = 0, . . . , 𝑚,
(176) (177)
𝑚 𝑚 2
𝑦 (0; 𝑟) = [0.5√𝑟, 0.2√1 − 𝑟 + 0.5] , ∑ 𝑐𝑗𝑟 𝑑𝑖,𝑗
(𝛼)
𝐿𝑗 (𝑥; 𝑟) − 3 (1 + 𝑟) ∑ (𝑐𝑗𝑟 𝐿 𝑗 (𝑥; 𝑟)) = 0
𝑗=0 𝑗=0
𝑖 = 0, . . . , 𝑚,
in which 𝑦(𝑥) : 𝐿E [0, 1] ∩ 𝐶E [0, 1] is a continuous fuzzy-
valued function and 𝑐 𝐷0𝛼+ denotes the fuzzy Caputo frac- or we can rewrite it in the matrix form as
tional derivative of order 𝛼. Also 𝐴 = [1+𝑟, 3−𝑟] is a constant
fuzzy number. 𝑇 2
𝐶𝑚,− 𝐷(𝛼) Φ (𝑥) − 3 (3 − 𝑟) [𝐶𝑚,−
𝑇
Φ (𝑥)] = 0,
For approximating the fuzzy solution of (162), using the (178)
𝑇 2
method described in Sections 4 and 5, if we approximate the 𝐶𝑚,+ 𝐷(𝛼) Φ (𝑥) − 3 (1 + 𝑟) [𝐶𝑚,+
𝑇
Φ (𝑥)] = 0,
solution by 𝑚 shifted Legendre functions, then it needs to
consider the first 𝑚 roots of the shifted Legendre polynomial in which (𝑐𝑗𝑟 ) = [𝑐𝑗,−
𝑟 𝑟
, 𝑐𝑗,+ ] and for 𝑗 = 0, 1, . . . , 𝑚. As it
𝐿 𝑚+1 (𝑥). Initially using (162), we have was described in Section 5, we produce 𝑚 fuzzy algebraic
Abstract and Applied Analysis 25
1
Fuzzy approximate solution (y(x))
1.5 0.9
1 0.8
0.7
0.5
0.6
0
r-cut
0.5
−0.5 0.4
1
0.8 0.3
0.6
0.2
0.4
x 1
0.2 0.8 0.1
0.6
0.4
0 0.2
0 r-cut 0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Figure 18: The fuzzy approximate solution of Example 63, 𝛼 = y(1; r)
0.95, 𝑚 = 8.
m=3
m=5
m=7
10−3
Figure 20: The approximate solution for different 𝑚 of Example 64,
𝛼 = 0.75.
10−4
Absolute error
10−5 1
0.8
0.6
10−6
0.4
0.2
10−7 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1
1 0.8
r-cut 0.8 0.6
0.6 0.4
0.4 t
m=3 m=9 x 0.2
0 0
0.2 r-cu
m=6 m = 12
Figure 21: The fuzzy approximate solution of Example 64, 𝛼 =
Figure 19: The absolute error for different 𝑚 of Example 63, 𝛼 = 0.75, 𝑚 = 3.
0.95.
Table 7: The approximate solution of the proposed method for Example 64 at 𝑥 = 1 with 𝑚 = 3.
𝛼=0.5 𝑟 𝛼=0.75 𝑟 𝛼=0.85 𝑟 𝛼=0.5 𝑟 𝛼=0.75 𝑟 𝛼=0.85 𝑟
𝑟 𝑦̃3 𝑦̃3 𝑦̃3 𝑦̃3 𝑦̃3 𝑦̃3
0 0.1402 0.1019 0.1040 0.8224 0.8856 0.9381
0.1 0.1434 0.1053 0.1080 0.3061 0.3250 0.3456
0.2 0.1468 0.1091 0.1123 0.2769 0.2917 0.3111
0.3 0.1503 0.1131 0.1170 0.2539 0.2646 0.2826
0.4 0.1539 0.1175 0.1222 0.2356 0.2418 0.2586
0.5 0.1578 0.1224 0.1278 0.2209 0.2223 0.2380
0.6 0.1618 0.1278 0.1340 0.2087 0.2056 0.2202
0.7 0.1662 0.1328 0.1409 0.1986 0.1912 0.2046
0.8 0.1706 0.1388 0.1485 0.1908 0.1785 0.1909
0.9 0.1752 0.1455 0.1571 0.1838 0.1674 0.1788
1 0.1778 0.1576 0.1680 0.1778 0.1576 0.1680
few number of shifted Legendre polynomials, desired results For future research, we will consider this method for
are available. Ultimately, the fuzzy approximate solution is solving FFDEs with order 1 < 𝛼 < 2. Also we will apply
described in Figure 21 with 𝛼 = 0.75 and 𝑚 = 3. it under Riemann-Liouville’s H-differentiability. Apart from
this, the other orthogonal functions like Jacobi polynomials
Remark 65. It is consequential to note that, for a crisp differ- will be extended for solving FFDEs.
ential equation of integer or fractional order, the problem
and F. Mainardi, Eds., vol. 378 of CISM Courses and Lectures, [25] M. Barkhordari Ahmadi and M. Khezerloo, “Fuzzy bivariate
pp. 291–348, Springer, Vienna, Austria, 1997. Chebyshev method for solving fuzzy Volterra-Fredholm inte-
[9] Y. A. Rossikhin and M. V. Shitikova, “Applications of fractional gral equations,” International Journal of Industrial Mathematics,
calculus to dynamic problems of linear and nonlinear heredi- vol. 3, no. 2, pp. 67–77, 2011.
tary mechanics of solids,” Applied Mechanics Reviews, vol. 50, [26] H. S. Goghary and M. S. Goghary, “Two computational meth-
no. 1, pp. 15–67, 1997. ods for solving linear Fredholm fuzzy integral equation of the
[10] S.-D. Lin, W.-C. Ling, K. Nishimoto, and H. M. Srivastava, second kind,” Applied Mathematics and Computation, vol. 182,
“A simple fractional-calculus approach to the solutions of the no. 1, pp. 791–796, 2006.
Bessel differential equation of general order and some of its [27] T. Lotfi and K. Mahdiani, “Fuzzy Galerkin method for solving
applications,” Computers & Mathematics with Applications, vol. Fredholm integral equations with error analysis,” International
49, no. 9-10, pp. 1487–1498, 2005. Journal of Industrial Mathematics, vol. 3, no. 4, pp. 237–249, 2011.
[11] P.-Y. Wang, S.-D. Lin, and H. M. Srivastava, “Remarks on [28] R. P. Agarwal, V. Lakshmikantham, and J. J. Nieto, “On the
a simple fractional-calculus approach to the solutions of the concept of solution for fractional differential equations with
Bessel differential equation of general order and some of its uncertainty,” Nonlinear Analysis. Theory, Methods & Applica-
applications,” Computers & Mathematics with Applications, vol. tions, vol. 72, no. 6, pp. 2859–2862, 2010.
51, no. 1, pp. 105–114, 2006. [29] T. Allahviranloo, S. Salahshour, and S. Abbasbandy, “Explicit
solutions of fractional differential equations with uncertainty,”
[12] Z. Odibat and S. Momani, “The variational iteration method:
Soft Computing, vol. 16, no. 2, pp. 297–302, 2012.
an efficient scheme for handling fractional partial differential
equations in fluid mechanics,” Computers & Mathematics with [30] S. Salahshour, T. Allahviranloo, S. Abbasbandy, and D. Baleanu,
Applications, vol. 58, no. 11-12, pp. 2199–2208, 2009. “Existence and uniqueness results for fractional differential
equations with uncertainty,” Advances in Difference Equations,
[13] J. He, D. Huang, and Z. He, “Study on sedimentary facies vol. 2012, article 112, 2012.
classified by main amplitude parameters of seismic wave and
[31] S. Salahshour, T. Allahviranloo, and S. Abbasbandy, “Solving
their variation characteristics,” Kuangwu Yanshi, vol. 19, no. 4,
fuzzy fractional differential equations by fuzzy Laplace trans-
pp. 86–89, 1999.
forms,” Communications in Nonlinear Science and Numerical
[14] B. Bede and S. G. Gal, “Generalizations of the differentiability Simulation, vol. 17, no. 3, pp. 1372–1381, 2012.
of fuzzy-number-valued functions with applications to fuzzy [32] M. Mazandarani and A. V. Kamyad, “Modified fractional Euler
differential equations,” Fuzzy Sets and Systems, vol. 151, no. 3, method for solving fuzzy fractional initial value problem,” Com-
pp. 581–599, 2005. munications in Nonlinear Science and Numerical Simulation, vol.
[15] B. Bede, “Note on “Numerical solution of fuzzy differential 18, no. 1, pp. 12–21, 2013.
equations by predictor-corrector method”,” Information Sci- [33] S. Momani and M. Aslam Noor, “Numerical methods for
ences. An International Journal, vol. 178, no. 7, pp. 1917–1922, fourth-order fractional integro-differential equations,” Applied
2008. Mathematics and Computation, vol. 182, no. 1, pp. 754–760,
[16] J. J. Buckley and T. Feuring, “Fuzzy differential equations,” Fuzzy 2006.
Sets and Systems, vol. 110, no. 1, pp. 43–54, 2000. [34] S. Momani and Z. Odibat, “Homotopy perturbation method
[17] Y. Chalco-Cano and H. Román-Flores, “On new solutions of for nonlinear partial differential equations of fractional order,”
fuzzy differential equations,” Chaos, Solitons and Fractals, vol. Physics Letters A, vol. 365, no. 5-6, pp. 345–350, 2007.
38, no. 1, pp. 112–119, 2008. [35] E. A. Rawashdeh, “Numerical solution of fractional integro-
[18] O. Kaleva, “Fuzzy differential equations,” Fuzzy Sets and Sys- differential equations by collocation method,” Applied Mathe-
tems, vol. 24, no. 3, pp. 301–317, 1987. matics and Computation, vol. 176, no. 1, pp. 1–6, 2006.
[19] H. Kim and R. Sakthivel, “Numerical solution of hybrid [36] A. Saadatmandi and M. Dehghan, “A new operational matrix
fuzzy differential equations using improved predictor-corrector for solving fractional-order differential equations,” Computers
method,” Communications in Nonlinear Science and Numerical & Mathematics with Applications, vol. 59, no. 3, pp. 1326–1336,
Simulation, vol. 17, no. 10, pp. 3788–3794, 2012. 2010.
[37] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “Efficient
[20] S. Salahshour and T. Allahviranloo, “Applications of fuzzy
Chebyshev spectral methods for solving multi-term fractional
Laplace transforms,” Soft Computing Journal, pp. 1–14, 2012.
orders differential equations,” Applied Mathematical Modelling,
[21] R. Ezzati and S. Ziari, “Numerical solution and error estimation vol. 35, no. 12, pp. 5662–5672, 2011.
of fuzzy Fredholm integral equation using fuzzy Bernstein [38] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A Chebyshev
polynomials,” Australian Journal of Basic and Applied Sciences, spectral method based on operational matrix for initial and
vol. 5, no. 9, pp. 2072–2082, 2011. boundary value problems of fractional order,” Computers &
[22] M. M. M. Otadi, “Numerical solution of fuzzy integral equations Mathematics with Applications, vol. 62, no. 5, pp. 2364–2373,
using Bernstein polynomials,” Australian Journal of Basic and 2011.
Applied Sciences, vol. 5, no. 7, pp. 724–728, 2011. [39] E. H. Doha, A. H. Bhrawy, and S. S. Ezz-Eldien, “A new
[23] M. A. F. Araghi and N. Parandin, “Numerical solution of fuzzy Jacobi operational matrix: an application for solving fractional
Fredholm integral equations by the Lagrange interpolation differential equations,” Applied Mathematical Modelling, vol. 36,
based on the extension principle,” Soft Computing, vol. 15, no. no. 10, pp. 4931–4943, 2012.
12, pp. 2449–2456, 2011. [40] R. Goetschel Jr. and W. Voxman, “Elementary fuzzy calculus,”
[24] N. Parandin and M. A. F. Araghi, “The approximate solution Fuzzy Sets and Systems, vol. 18, no. 1, pp. 31–43, 1986.
of linear fuzzy fredholm integral equations of the second kind [41] H.-J. Zimmermann, Fuzzy Set Theory—and Its Applications,
by using iterative interpolation,” World Academy of Science, Kluwer Academic Publishers, Dordrecht, The Netherlands,
Engineering and Technology, vol. 37, pp. 1036–1042, 2009. 1991.
28 Abstract and Applied Analysis
[42] B. Bede and L. Stefanini, “Generalized differentiability of fuzzy- [62] S. Abbasbandy and M. Amirfakhrian, “A new approach to
valued functions,” Fuzzy Sets and Systems. universal approximation of fuzzy functions on a discrete set
[43] L. Stefanini, “A generalization of Hukuhara difference and of points,” Applied Mathematical Modelling, vol. 30, no. 12, pp.
division for interval and fuzzy arithmetic,” Fuzzy Sets and 1525–1534, 2006.
Systems, vol. 161, no. 11, pp. 1564–1584, 2010. [63] S. Abbasbandy and M. Amirfakhrian, “Numerical approx-
[44] B. Bede and S. G. Gal, “Quadrature rules for integrals of fuzzy- imation of fuzzy functions by fuzzy polynomials,” Applied
number-valued functions,” Fuzzy Sets and Systems, vol. 145, no. Mathematics and Computation, vol. 174, no. 2, pp. 1001–1006,
3, pp. 359–380, 2004. 2006.
[45] G. A. Anastassiou, Fuzzy Mathematics: Approximation Theory, [64] S. Abbasbandy and B. Asady, “The nearest trapezoidal fuzzy
vol. 251 of Studies in Fuzziness and Soft Computing, Springer, number to a fuzzy quantity,” Applied Mathematics and Compu-
Berlin, Germany, 2010. tation, vol. 156, no. 2, pp. 381–386, 2004.
[46] G. A. Anastassiou and S. G. Gal, “On a fuzzy trigonometric [65] O. Kaleva, “Interpolation of fuzzy data,” Fuzzy Sets and Systems,
approximation theorem of Weierstrass-type,” Journal of Fuzzy vol. 61, no. 1, pp. 63–70, 1994.
Mathematics, vol. 9, no. 3, pp. 701–708, 2001. [66] R. Lowen, “A fuzzy Lagrange interpolation theorem,” Fuzzy Sets
[47] O. Kaleva, “A note on fuzzy differential equations,” Nonlinear and Systems, vol. 34, no. 1, pp. 33–38, 1990.
Analysis. Theory, Methods & Applications, vol. 64, no. 5, pp. 895– [67] C.-T. Yeh, “Trapezoidal and triangular approximations preserv-
900, 2006. ing the expected interval,” Fuzzy Sets and Systems, vol. 159, no.
[48] J. J. Nieto, A. Khastan, and K. Ivaz, “Numerical solution of 11, pp. 1345–1353, 2008.
fuzzy differential equations under generalized differentiability,” [68] S. Abbasbandy and M. Amirfakhrian, “Best Approximation of
Nonlinear Analysis. Hybrid Systems, vol. 3, no. 4, pp. 700–707, Fuzzy Functions,” Journal of Nonlinear Studies, vol. 14, pp. 88–
2009. 103, 2007.
[49] H.-C. Wu, “The improper fuzzy Riemann integral and its [69] S. Kazem, S. Abbasbandy, and Sunil Kumar, “Fractional-order
numerical integration,” Information Sciences, vol. 111, no. 1–4, pp. Legendre functions for solving fractional order differential
109–137, 1998. equations,” Applied Mathematical Modelling, vol. 37, no. 7, pp.
[50] B. Asady, S. Abbasbandy, and M. Alavi, “Fuzzy general linear 5498–5510, 2013.
systems,” Applied Mathematics and Computation, vol. 169, no. 1, [70] G. A. Anastassiou, “Fuzzy approximation by fuzzy convolution
pp. 34–40, 2005. type operators,” Computers & Mathematics with Applications,
[51] T. Allahviranloo and M. Afshar Kermani, “Solution of a fuzzy vol. 48, no. 9, pp. 1369–1386, 2004.
system of linear equation,” Applied Mathematics and Computa- [71] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang,
tion, vol. 175, no. 1, pp. 519–531, 2006. Spectral Methods in Fluid Dynamics, Springer Series in Com-
[52] M. Dehghan and B. Hashemi, “Iterative solution of fuzzy linear putational Physics, Springer, New York, NY, USA, 1988.
systems,” Applied Mathematics and Computation, vol. 175, no. 1, [72] J. H. Freilich and E. L. Ortiz, “Numerical solution of systems of
pp. 645–674, 2006. ordinary differential equations with the Tau method: an error
[53] M. Friedman, M. Ming, and A. Kandel, “Fuzzy linear systems,” analysis,” Mathematics of Computation, vol. 39, no. 160, pp. 467–
Fuzzy Sets and Systems, vol. 96, no. 2, pp. 201–209, 1998. 479, 1982.
[54] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory [73] E. L. Ortiz, “The tau method,” SIAM Journal on Numerical
and Applications of Fractional Differential Equations, vol. 204 Analysis, vol. 6, no. 3, pp. 480–492, 1969.
of North-Holland Mathematics Studies, Elsevier Science B.V., [74] S. Kazem, “An integral operational matrix based on Jacobi
Amsterdam, The Netherlands, 2006. polynomials for solving fractional-order differential equations,”
[55] K. S. Miller and B. Ross, An Introduction to the Fractional Calcu- Applied Mathematical Modelling, vol. 37, no. 3, pp. 1126–1136,
lus and Fractional Differential Equations, A Wiley-Interscience 2013.
Publication, John Wiley & Sons, New York, NY, USA, 1993. [75] J. C. Mason and D. C. Handscomb, Chebyshev Polynomials,
[56] K. Diethelm, N. J. Ford, A. D. Freed, and Yu. Luchko, “Algo- Chapman & Hall/CRC, Boca Raton, Fla, USA, 2003.
rithms for the fractional calculus: a selection of numerical [76] S. Nemati, P. M. Lima, and Y. Ordokhani, “Numerical solution
methods,” Computer Methods in Applied Mechanics and Engi- of a class of two-dimensional nonlinear Volterra integral equa-
neering, vol. 194, no. 6–8, pp. 743–773, 2005. tions using Legendre polynomials,” Journal of Computational
[57] S. G. Samko, A. A. Kilbas, and O. I. Marichev, Fractional and Applied Mathematics, vol. 242, pp. 53–69, 2013.
Integrals and Derivatives. Theory and Applications, Gordon and [77] S. Abbasbandy and R. Ezzati, “Newton’s method for solving a
Breach Science Publishers, Yverdon, Switzerland, 1993. system of fuzzy nonlinear equations,” Applied Mathematics and
[58] Z. M. Odibat and N. T. Shawagfeh, “Generalized Taylor’s Computation, vol. 175, no. 2, pp. 1189–1199, 2006.
formula,” Journal of Applied Mathematics and Computing, vol. [78] Z. Odibat and S. Momani, “An algorithm for the numerical
186, no. 1, pp. 286–293, 2007. solution of differential equations of fractional order,” Journal of
[59] S. Pederson and M. Sambandham, “Numerical solution of Applied Mathematics & Informatics, vol. 26, pp. 15–27, 2008.
hybrid fuzzy differential equation IVPs by a characterization [79] S. Das, Functional Fractional Calculus for System Identification
theorem,” Information Sciences, vol. 179, no. 3, pp. 319–328, 2009. and Controls, Springer, Berlin, Germany, 2008.
[60] J. Shen and T. Tang, High Order Numerical Methods and [80] A. Le Mehaute and G. Crepy, “Introduction to transfer and
Algorithms, Chinese Science Press, Beijing, China, 2005. motion in fractal media: the geometry of kinetics,” Solid State
[61] S. Abbasbandy and M. Amirfakhrian, “The nearest approx- Ionics, vol. 9-10, no. 1, pp. 17–30, 1983.
imation of a fuzzy quantity in parametric form,” Applied [81] C. Csajka and D. Verotta, “Pharmacokinetic-pharmacodynamic
Mathematics and Computation, vol. 172, no. 1, pp. 624–632, modelling: history and perspectives,” Journal of Pharmacokinet-
2006. ics and Pharmacodynamics, vol. 33, no. 3, pp. 227–279, 2006.
Abstract and Applied Analysis 29
Research Article
Two Efficient Generalized Laguerre Spectral Algorithms for
Fractional Initial Value Problems
Copyright © 2013 D. Baleanu et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We present a direct solution technique for approximating linear multiterm fractional differential equations (FDEs) on semi-
infinite interval, using generalized Laguerre polynomials. We derive the operational matrix of Caputo fractional derivative of
the generalized Laguerre polynomials which is applied together with generalized Laguerre tau approximation for implementing a
spectral solution of linear multiterm FDEs on semi-infinite interval subject to initial conditions. The generalized Laguerre pseudo-
spectral approximation based on the generalized Laguerre operational matrix is investigated to reduce the nonlinear multiterm
FDEs and its initial conditions to nonlinear algebraic system, thus greatly simplifying the problem. Through several numerical
examples, we confirm the accuracy and performance of the proposed spectral algorithms. Indeed, the methods yield accurate
results, and the exact solutions are achieved for some tested problems.
In this paper, the Caputo fractional derivative of gen- where ⌈]⌉ and ⌊]⌋ are the ceiling and floor functions,
eralized Laguerre operational matrix (GLOM) is stated respectively, while 𝑁 = {1, 2, . . .} and 𝑁0 = {0, 1, 2, . . .}.
and proved. The main aim of this paper is to extend the The Caputo’s derivative operator is a linear operation:
application of generalized Laguerre spectral tau method
based on GLOM to develop a direct solution technique 𝐷] (𝜆𝑓 (𝑥) + 𝜇𝑔 (𝑥)) = 𝜆𝐷] 𝑓 (𝑥) + 𝜇𝐷] 𝑔 (𝑥) , (5)
for the numerical solution of linear multi-term FDEs on
where 𝜆 and 𝜇 are constants.
a semi-infinite interval. Moreover, we develop the gener-
We recall below some relevant properties of the general-
alized Laguerre pseudo-spectral approximation based on
ized Laguerre polynomials (Szego [27] and Funaro [28]). let
the GLOM for reducing the nonlinear multi-term FDEs
subject to nonhomogeneous initial conditions to a system Λ = (0, ∞) and let 𝑤(𝛼) (𝑥) = 𝑥𝛼 𝑒−𝑥 be a weight function on
of nonlinear algebraic equations. Finally, the accuracy of the Λ in the usual sense. Define
proposed algorithms is demonstrated by test problems. The 𝐿2𝑤(𝛼) (Λ)
numerical results are given to show that the proposed spectral (6)
algorithms based on generalized Laguerre operational matrix = {V | V is measurable on Λ and ‖V‖𝑤(𝛼) < ∞} ,
of Caputo fractional derivatives are very effective for linear
and nonlinear FDEs. with the inner product and norm
The outline of the paper is as follows. In Section 2, we
present some preliminaries. Section is devoted to drive the (𝑢, V)𝑤(𝛼) = ∫ 𝑢 (𝑥) V (𝑥) 𝑤(𝛼) (𝑥) 𝑑𝑥,
Λ
GLOM of Caputo fractional derivative. In Section 4, we (7)
extend the generalized Laguerre spectral tau and pseudo- ‖V‖𝑤(𝛼) = (V, V)1/2
𝑤(𝛼)
.
spectral approximations based on the GLOM of fractional
derivative for solving multiorder linear and nonlinear FDEs. For 𝛼 > −1, the generalized Laguerre polynomials are given
Some numerical experiments are presented in Section 5. by
Finally, we conclude the paper with some remarks.
1 −𝛼 𝑥 𝑖 𝑖+𝛼 −𝑥
𝐿(𝛼)
𝑖 (𝑥) = 𝑥 𝑒 𝜕𝑥 (𝑥 𝑒 ) , 𝑖 = 1, 2, . . . . (8)
𝑖!
2. Some Basic Preliminaries
According to [29] for 𝛼 > −1, we get
The two most commonly used definitions are the Riemann-
Liouville operator and the Caputo operator. We give some 𝜕𝑥 𝐿(𝛼) (𝛼+1)
𝑖 (𝑥) = −𝐿 𝑖−1 (𝑥) ,
definitions and properties of fractional derivatives and gen-
eralized Laguerre polynomials. 𝐿(𝛼)
𝑖+1 (𝑥)
where 𝐿(𝛼)
𝑗 (0) = Γ(𝑗 + 𝛼 + 1)/Γ(𝛼 + 1)𝑗!, will be of important Then the derivative of the vector 𝜙(𝑥) can be expressed by
use later, for treating the initial conditions of the given FDEs.
Let 𝑢(𝑥) ∈ 𝐿2𝑤(𝛼) (Λ), then 𝑢(𝑥) may be expressed in terms 𝑑𝜙 (𝑥)
= D(1) 𝜙 (𝑥) , (21)
of generalized Laguerre polynomials as 𝑑𝑥
∞
𝑢 (𝑥) = ∑𝑎𝑗 𝐿(𝛼)
𝑗 (𝑥) , (13) where D(1) is the (𝑁 + 1) × (𝑁 + 1) operational matrix of the
𝑗=0 derivative given by
1 ∞
𝑎𝑗 = ∫ 𝑢 (𝑥) 𝐿(𝛼)
𝑗 (𝑥) 𝑤
(𝛼)
(𝑥) 𝑑𝑥, 𝑗 = 0, 1, 2, . . . . 0 0 0 0
0 0 ⋅⋅⋅ 0
ℎ𝑘 0 1 0 0 0
0 0 ⋅⋅⋅ 0
(14)
(1 1 0 0
0 0 ⋅⋅⋅ 0)
( 0)
In particular applications, the generalized Laguerre polyno- D(1) = − (1 1 1 0
0 0 ⋅⋅⋅ ). (22)
(1 1 1 1
0 0 ⋅⋅⋅ 0)
mials up to degree 𝑁 + 1 are considered. Then we have
.. .. .. ..
. .. ..
𝑁 . . ⋅ ⋅ ⋅ ..
. . . .
𝑢𝑁 (𝑥) = ∑𝑎𝑗 𝐿(𝛼)
𝑗 (𝑥) . (15) (1 1 1 1 1 ⋅ ⋅ ⋅ 1 0)
𝑗=0
By using (21), it is clear that
We will present the Laguerre-Gauss quadrature. Let
{𝑥𝑗(𝛼) , 𝜛𝑗(𝛼) } be the set of generalized Laguerre-Gauss quadra- 𝑑𝑛 𝜙 (𝑥) 𝑛
ture nodes and weights: 𝑛
= (D(1) ) 𝜙 (𝑥) , (23)
𝑑𝑥
𝑁
∫ 𝜙 (𝑥) 𝑤(𝛼) (𝑥) 𝑑𝑥 = ∑𝜙 (𝑥𝑗(𝛼) ) 𝜛𝑗(𝛼) . (16) where 𝑛 ∈ 𝑁 and the superscript in D(1) denotes matrix
Λ 𝑗=0 powers. Thus
For the generalized Laguerre-Gauss quadrature, {𝑥𝑗(𝛼) } are the 𝑛
D(𝑛) = (D(1) ) , 𝑛 = 1, 2, . . . . (24)
zeros of 𝐿(𝛼)
𝑖+1 (𝑥), and
where 𝑁
𝑔 (𝑥) ≃ ∑𝑔𝑖 𝐿(𝛼) 𝑇
𝑖 (𝑥) = 𝐺 𝜙 (𝑥) , (39)
𝑖 𝑗
𝑘+ℓ 𝑖=0
𝑆] (𝑖, 𝑗) = ∑ ∑ ((−1) 𝑗!Γ (𝑖 + 𝛼 + 1)
𝑘=⌈]⌉ℓ=0
where vector 𝐺 = [𝑔0 , . . . , 𝑔𝑁]𝑇 is known.
×Γ (𝑘 − ] + 𝛼 + ℓ + 1) ) By using Theorem 4 (relation equations (26), and (38)) we
have
(33)
((𝑖 − 𝑘)! (𝑗 − ℓ)!ℓ! 𝐷] 𝑢 (𝑥) ≃ 𝐶𝑇 𝐷] 𝜙 (𝑥) = 𝐶𝑇 𝐷(]) 𝜙 (𝑥) ,
× Γ (𝑘 − ] + 1) Γ (𝑘 + 𝛼 + 1)
𝐷𝛽𝑗 𝑢 (𝑥) ≃ 𝐶𝑇 𝐷𝛽𝑗 𝜙 (𝑥) = 𝐶𝑇 𝐷(𝛽𝑗 ) 𝜙 (𝑥) , 𝑗 = 1, . . . , 𝑘.
−1 (40)
× Γ (𝛼 + ℓ + 1) ) .
Accordingly, (32) can be written in a vector form as follows: Making use of (38)–(40), the residual 𝑅𝑁(𝑥) for (36) can be
given from
𝐷] 𝐿(𝛼)
𝑖 (𝑥) 𝑘
≃ [𝑆] (𝑖, 0) , 𝑆] (𝑖, 1) , 𝑆] (𝑖, 2) , . . . , 𝑆] (𝑖, 𝑁)] 𝜙 (𝑥) , (34) 𝑅𝑁 (𝑥) = (𝐶𝑇 𝐷(]) − 𝐶𝑇 ∑𝛾𝑗 D(𝛽𝑗 ) − 𝛾𝑘+1 𝐶𝑇 − 𝐺𝑇 ) 𝜙 (𝑥) .
𝑗=1
𝑖 = ⌈]⌉ , . . . , 𝑁. (41)
Abstract and Applied Analysis 5
4.2. Nonlinear Multiorder FDEs. In this section, we present If we apply the operational matrix formulation, the
the generalized Laguerre pseudo-spectral approximation in generalized Laguerre spectral tau method with 𝑁 = 2, we
combination with GLOM of fractional derivative to find the get
approximate solution 𝑢𝑁(𝑥).
Let us consider the nonlinear multi-term FDE 𝑢 (𝑥) = 𝑐0 𝐿(𝛼) (𝛼) (𝛼) 𝑇
0 (𝑥) + 𝑐1 𝐿 1 (𝑥) + 𝑐2 𝐿 2 (𝑥) = 𝐶 𝜙 (𝑥) .
(49)
(𝛼 + 1) (𝛼 + 2)
≃ 𝐹 (𝑥, 𝐶𝑇 𝜙 (𝑥𝑗(𝛼) ) , 𝐶𝑇 D(𝛽1 ) 𝜙 (𝑥𝑗(𝛼) ) , (47) 𝑐0 + (𝛼 + 1) 𝑐1 + 𝑐2 = 1,
2 (52)
𝑇 (𝛽𝑘 )
...,𝐶 D 𝜙 (𝑥𝑗(𝛼) )) . −𝑐1 − (𝛼 + 2) 𝑐2 − 1 = 0.
Combining (𝑁 − 𝑚 + 1) algebraic equations (47) with 𝑚 Solving the resulted system of algebraic equations (51)-
initial conditions (46) generates a system of (𝑁 + 1) non- (52) provides the unknown coefficients in terms of the
linear algebraic equations. This system may be evaluated by parameter 𝛼:
implementing Newton’s iterative method to find the spectral
solution 𝑢(𝑥). 𝑐0 = 𝛼 + 2, 𝑐1 = −1, 𝑐2 = 0. (53)
6 Abstract and Applied Analysis
Table 1: 𝑐0 , 𝑐1 , and 𝑐2 for different values of 𝛼 for Example 1. Table 2: 𝑐0 , 𝑐1 , and 𝑐2 for different values of 𝛼 for Example 2.
𝛼 𝑐0 𝑐1 𝑐2 𝛼 𝑐0 𝑐1 𝑐2
−0.5 1.5 −1 0 −0.5 0.75 −3 2
0 2 −1 0 0 2 −4 2
0.5 2.5 −1 0 0.5 3.75 −5 2
1 3 −1 0 1 6 −6 2
2 4 −1 0 2 12 −8 2
3 5 −1 0 3 20 −10 2
𝐿(𝛼) (𝛼 + 1) (𝛼 + 2)
1 (𝑥) 𝑐0 + (𝛼 + 1) 𝑐1 + 𝑐2 = 0.
𝑢 (𝑥) = (𝑐0 , 𝑐1 , 𝑐2 ) ( ) = 1 + 𝑥, (54) 2 (59)
𝐿(𝛼)
2 (𝑥) −𝑐1 − (𝛼 + 2) 𝑐2 = 0.
𝑢 (𝑥) = 𝑐0 𝐿(𝛼) (𝛼) (𝛼) 𝑇 Example 3. Consider linear initial value problem of fractional
0 (𝑥) + 𝑐1 𝐿 0 (𝑥) + 𝑐2 𝐿 2 (𝑥) = 𝐶 𝜙 (𝑥) .
(56)
order (see [31])
Here, we have
0 0 0 0 0 0 0 0 𝛼 𝑐0 𝑐1 𝑐2 𝑐3
(2) 0 0 0 0 (1) 1 0 0 0 −0.5 15/8 −45/4 15 −6
D =( ), D = −( ),
1 0 0 0 1 1 0 0 0 6 −18 18 −6
2 1 0 0 1 1 1 0 0.5 105/8 −105/4 21 −6
0 0 0 0 1 24 −36 24 −6
2 60 −60 30 −6
𝑆1/2 (1, 0) 𝑆1/2 (1, 1) 𝑆1/2 (1, 2) 𝑆1/2 (1, 3)
3 120 −90 36 −6
(1/2) ( )
D =( ,
𝑆1/2 (2, 0) 𝑆1/2 (2, 1) 𝑆1/2 (2, 2) 𝑆1/2 (2, 3))
𝑆1/2 (3, 0) 𝑆1/2 (3, 1) 𝑆1/2 (3, 2) 𝑆1/2 (3, 3) Example 4. Consider the following FDE:
( )
𝐷5/2 𝑢 (𝑥) + 𝐷2 𝑢 (𝑥) − 2𝐷1/2 𝑢 (𝑥) + 4𝑢 (𝑥) = 𝑔 (𝑥) ,
𝑔0 (68)
𝑔1 𝑢 (0) = 0, 𝑢 (0) = 0, 𝑥 ∈ Λ,
𝐺 = ( ).
𝑔2
𝑔3 where
(64)
2Γ (10) 17/2 Γ (10) 13/2
Therefore using (42), we obtain 𝑔 (𝑥) = 4𝑥9 + 72𝑥7 − 𝑥 + 𝑥 (69)
Γ (19/2) Γ (15/2)
𝑐0 + [2𝑐1 + 𝑆1/2 (1, 0)] 𝑐1
and the exact solution is 𝑢(𝑥) = 𝑥9 .
+ [3 + 𝑆1/2 (2, 0)] 𝑐2
Now, if we use the spectral tau approximation based on
+ [4 + 𝑆1/2 (3, 0)] 𝑐3 − 𝑔0 = 0, with 𝑁 = 9 and 𝑥 ∈ Λ, then we obtain
1 6
0 1 2 3 4 5 6 7
x
4
uN (x, 1.5) u(x, 1.7)
u(x, 1.5) uN (x, 1.9)
uN (x, 1.7) u(x, 1.9) 2
The solution of this problem is obtained by applying and the exact solution is 𝑢(𝑥) = 𝑥3 .
the generalized Laguerre-Gauss collocation method based on
generalized Laguerre operational matrix. The absolute error The solution of this problem is obtained by applying
between the exact and the approximate solution obtained by the generalized Laguerre-Gauss collocation method based on
the proposed method 𝛾 = 0.01, 𝛼 = 1 and 𝑁 = 30 is given in generalized Laguerre operational matrix for 𝜉 = 2.001, 𝜂 =
Figure 3. 1.001, and 𝜃 = 0.001. The exact solution and approximate
Abstract and Applied Analysis 9
15
Jacobi operational matrix: an application for solving fractional
10
differential equations,” Applied Mathematical Modelling, vol. 36,
no. 10, pp. 4931–4943, 2012.
5 [8] D. Funaro, “Estimates of Laguerre spectral projectors in Sobolev
spaces,” in Orthogonal Polynomials and Their Applications, C.
0 Brezinski, L. Gori, and A. Ronveaux, Eds., vol. 9, pp. 263–266,
0 0.5 1 1.5 2 2.5 3 Scientific Publishing, Singapore, 1991.
x [9] A. H. Bhrawy, A. S. Alofi, and S. S. Ezz-Eldien, “A quadrature
tau method for fractional differential equations with variable
Exact solution coefficients,” Applied Mathematics Letters, vol. 24, no. 12, pp.
uN (x) at N = 5
2146–2152, 2011.
uN (x) at N = 10
[10] J. Song, F. Yin, X. Cao, and F. Lu, “Fractional variational
Figure 4: Comparing the exact solution and the approximate iteration method versus Adomian’s decomposition method in
solutions at 𝑁 = 5, 10 and 𝛼 = 1. some fractional partial differential equations,” Journal of Applied
Mathematics, vol. 2013, Article ID 392567, 10 pages, 2013.
[11] H. Jafari, H. Tajadodi, and D. Baleanu, “A modified variational
iteration method for solving fractional Riccati differential equa-
solutions obtained by the proposed method for 𝛼 = 1 and tion by Adomian polynomials,” Fractional Calculus and Applied
two choices of 𝑁 are shown in Figure 4. Analysis, vol. 16, no. 1, pp. 109–122, 2013.
[12] A. A. Hemeda, “New iterative method: an application for
solving fractional physical differential equations,” Abstract and
6. Conclusion Applied Analysis, vol. 2013, Article ID 617010, 9 pages, 2013.
In this paper, the generalized Laguerre operational matrix of [13] G.-C. Wu and D. Baleanu, “Variational iteration method for
Caputo fractional derivative was derived. Furthermore, we fractional calculus: a universal approach by Laplace transform,”
have implemented the generalized Laguerre tau approxima- Advances in Difference Equations, vol. 2013, article 18, 2013.
tion in combination with the GLOM with the generalized [14] C. Yang and J. Hou, “An approximate solution of nonlinear
Laguerre family to solve the linear FDEs. In addition, com- fractional differential equation by Laplace transform and ado-
mian polynomials,” Journal of Information and Computational
bining the pseudo-spectral approximation and the GLOM
Science, vol. 10, no. 1, pp. 213–222, 2013.
of fractional derivative was applied to develop an accurate
[15] M. Gulsua, Y. Ozturka, and A. Anapalia, “Numerical approach
approximate solution of nonlinear FDEs. The generalized for solving fractional Fredholm integro-differential equation,”
Laguerre-Gauss quadrature points were used as a collocation International Journal of Computer Mathematics, 2013.
nodes. This method reduces the nonlinear FDEs to a system [16] H. S. Nik, S. Effati, S. S. Motsa, and S. Shateyi, “A new piecewise-
of algebraic equation in the expansion coefficients which spectral homotopy analysis method for solving chaotic systems
can be solved by any standard technique. The numerical of initial value problems,” Mathematical Problems in Engineer-
results demonstrate that the proposed spectral algorithms are ing, vol. 2013, Article ID 583193, 13 pages, 2013.
accurate and efficient. [17] E. H. Doha, A. H. Bhrawy, D. Baleanu, and S. S. Ezz-Eldien, “On
shifted Jacobi spectral approximations for solving fractional
differential equations,” Applied Mathematics and Computation,
References vol. 219, no. 15, pp. 8042–8056, 2013.
[1] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional [18] M. Maleki, I. Hashim, M. Tavassoli Kajani, and S. Abbas-
Calculus: Models and Numerical Methods, vol. 3 of Series on bandy, “An adaptive pseudospectral method for fractional order
Complexity, Nonlinearity and Chaos, World Scientific Publish- boundary value problems,” Abstract and Applied Analysis, vol.
ing, Hackensack, NJ, USA, 2012. 2012, Article ID 381708, 19 pages, 2012.
[2] K. S. Miller and B. Ross, An Introduction to the Fractional [19] A. Al-Rabtah, S. Momani, and M. A. Ramadan, “Solving linear
Calculus and Fractional Differential Equations, John Wiley & and nonlinear fractional differential equations using spline
Sons, New York, NY, USA, 1993. functions,” Abstract and Applied Analysis, vol. 2012, Article ID
426514, 9 pages, 2012.
[3] M. D. Ortigueira, “Introduction to fractional linear systems.
[20] A. H. Bhrawy, M. M. Alghamdi, and T. M. Taha, “A new
Part 1: continuous-time case,” IEE Proceedings: Vision, Image
modified generalized Laguerre operational matrix of fractional
and Signal Processing, vol. 147, no. 1, pp. 62–70, 2000.
integration for solving fractional differential equations on the
[4] S. Das, Functional Fractional Calculus for System Identification half line,” Advances in Difference Equations, vol. 2012, article 179,
and Controls, Springer, New York, NY, USA, 2008. 2012.
[5] C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, [21] S. Yuzbasi, “Numerical solution of the Bagley-Torvik equation
Spectral Methods in Fluid Dynamics, Springer, New York, NY, by the Bessel collocation method,” Mathematical Methods in the
USA, 1988. Applied Sciences, vol. 36, no. 3, pp. 300–312, 2013.
10 Abstract and Applied Analysis
Research Article
Local Fractional Series Expansion Method for Solving
Wave and Diffusion Equations on Cantor Sets
Copyright © 2013 Ai-Min Yang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We proposed a local fractional series expansion method to solve the wave and diffusion equations on Cantor sets. Some examples
are given to illustrate the efficiency and accuracy of the proposed method to obtain analytical solutions to differential equations
within the local fractional derivatives.
Let us consider the local fractional differential equation The convergent condition is
𝑡𝑖𝛼
𝑢𝑡𝑛𝛼 = 𝐿 𝛼 𝑢, (4) lim [ 𝑋 (𝑥)] = 0. (15)
𝑛→∞ Γ (1 + 𝑖𝛼) 𝑖
where 𝐿 is a linear local operator with respect to 𝑥, 𝑛 ∈ {1, 2}. This approach is termed the local fractional series expansion
In accordance with the results in [28, 47], there are method (LFSEM)
multiterm separated functions of independent variables 𝑡 and
𝑥, namely,
3. Applications to Wave and Diffusion
∞ Equations on Cantor Sets
𝑢 (𝑥, 𝑡) = ∑𝑇𝑖 (𝑡) 𝑋𝑖 (𝑥) , (5)
𝑖=0 In this section, four examples for wave and diffusion equa-
tions on Cantor sets will demonstrate the efficiency of
where 𝑇𝑖 (𝑡) and 𝑋𝑖 (𝑥) are local fractional continuous func- LFSEM.
tions.
Moreover, there is a nondifferential series term Example 1. Let us consider the diffusion equation on Cantor
𝑖𝛼 set
𝑡
𝑇𝑖 (𝑡) = 𝑝𝑖 , (6)
𝑢𝑡𝛼 (𝑥, 𝑡) − 𝑢𝑥2𝛼 (𝑥, 𝑡) = 0, 0<𝛼≤1 (16)
Γ (1 + 𝑖𝛼)
with the initial condition
where 𝑝𝑖 is a coefficient.
In view of (6), we may present the solution in the form 𝑥𝛼
𝑢 (𝑥, 0) = . (17)
Γ (1 + 𝛼)
∞ 𝑖𝛼
𝑡
𝑢 (𝑥, 𝑡) = ∑𝑝𝑖 𝑋 (𝑥) . (7) Following (12), we have recursive formula
𝑖=0 Γ (1 + 𝑖𝛼) 𝑖
𝜕2𝛼 𝑋𝑖 (𝑥)
𝑋𝑖+1 (x) = ,
Then, following (7), we have 𝜕𝑥2𝛼
(18)
∞
𝑡 𝑖𝛼 𝑥𝛼
𝑢 (𝑥, 𝑡) = ∑ 𝑋 (𝑥) . 𝑋0 (𝑥) = .
Γ (1 + 𝑖𝛼) 𝑖
(8) Γ (1 + 𝛼)
𝑖=0
Hence, we get
Hence,
𝑥𝛼
∞ ∞
𝑋0 (𝑥) = ,
1 1 Γ (1 + 𝛼)
𝑢𝑡𝑛𝛼 = ∑ 𝑡𝑖𝛼 𝑋𝑖+1 (𝑥) = ∑ 𝑡𝑖𝛼 𝑋𝑖+𝑛 (𝑥) ,
𝑖=0 Γ (1 + 𝑖𝛼) 𝑖=0 Γ (1 + 𝑖𝛼) 𝑋1 (𝑥) = 0,
(19)
∞ ∞
𝑡𝑖𝛼 𝑡𝑖𝛼 𝑋2 (𝑥) = 0,
𝐿 𝛼 𝑢 = 𝐿 𝛼 [∑ 𝑋𝑖 (𝑥)] = ∑ (𝐿 𝛼 𝑋𝑖 ) (𝑥) .
𝑖=0 Γ (1 + 𝑖𝛼) 𝑖=0 Γ (1 + 𝑖𝛼) ..
(9) .
Example 2. Let us consider the diffusion equation on Cantor Hence, using the relations (29), the recursive calculations
set yield
𝑥2𝛼
𝑢𝑡𝛼 (𝑥, 𝑡) − ⋅ 𝑢2𝛼 (𝑥, 𝑡) = 0, 0<𝛼≤1 (21) 𝑥𝛼
Γ (1 + 2𝛼) 𝑥 𝑋0 (𝑥) = ,
Γ (1 + 𝛼) (29)
with the initial condition
𝑋1 (𝑥) = 1,
𝑥2𝛼
𝑢 (𝑥, 0) = . (22) 𝑋2 (𝑥) = 0,
Γ (1 + 2𝛼)
𝑋3 (𝑥) = 0,
Following (12), we get
(30)
𝑋4 (𝑥) = 0,
2𝛼 2𝛼
𝑥 𝜕 𝑋i (𝑥)
𝑋𝑖+1 (𝑥) = , ..
Γ (1 + 2𝛼) 𝜕𝑥2𝛼 .
(23)
𝑥2𝛼
𝑋0 (𝑥) = . and so on.
Γ (1 + 2𝛼)
Finally, we obtain
By using the recursive formula (23), we get consequently
𝑥𝛼 𝑡2𝛼
𝑥2𝛼 𝑢 (𝑥, 𝑡) = + . (31)
𝑋0 (𝑥) = , Γ (1 + 𝛼) Γ (1 + 2𝛼)
Γ (1 + 2𝛼)
𝑥2𝛼 Example 4. Let us consider the wave equation on Cantor sets
𝑋1 (𝑥) = , [26, 30]
Γ (1 + 2𝛼)
(24)
2𝛼
𝑥 𝑢𝑡2𝛼 (𝑥, 𝑡) − cu2𝛼
𝑋2 (𝑥) = , 𝑥 (𝑥, 𝑡) = 0, 0 < 𝛼 ≤ 1, (32)
Γ (1 + 2𝛼)
.. where c is a constant.
. The initial condition is
[24] H. Jafari, H. Tajadodi, N. Kadkhoda, and D. Baleanu, “Frac- information of non-random functions,” Chaos, Solitons and
tional subequation method for Cahn-Hilliard and Klein- Fractals, vol. 40, no. 3, pp. 1428–1448, 2009.
Gordon equations,” Abstract and Applied Analysis, vol. 2013, [44] G. Jumarie, “Laplace’s transform of fractional order via the
Article ID 587179, 5 pages, 2013. Mittag-Leffler function and modified Riemann-Liouville deriv-
[25] S. Zhang, Q. A. Zong, D. Liu, and Q. Gao, “A generalized exp- ative,” Applied Mathematics Letters, vol. 22, no. 11, pp. 1659–1664,
function method for fractional Riccati differential equations,” 2009.
Communications in Fractional Calculus, vol. 1, pp. 48–52, 2010. [45] X. J. Yang and D. Baleanu, “Fractal heat conduction problem
[26] X. J. Yang, Advanced Local Fractional Calculus and Its Applica- solved by local fractional variation iteration method,” Thermal
tions, World Science, New York, NY, USA, 2012. Science, vol. 17, no. 2, pp. 625–628, 2013.
[27] X. J. Yang, “Local fractional integral transforms,” Progress in [46] W. H. Su, D. Baleanu, X.-J. Yang, and H. Jafari, “Damped wave
Nonlinear Science, vol. 4, pp. 1–225, 2011. equation and dissipative wave equation in fractal strings within
[28] X. J. Yang, Local Fractional Functional Analysis and Its Applica- the local fractional variational iteration method,” Fixed Point
tions, Asian Academic, Hong Kong, 2011. Theory and Applications, vol. 2013, no. 1, pp. 89–102, 2013.
[29] W. P. Zhong, X. J. Yang, and F. Gao, “A Cauchy problem for [47] M. S. Hu, R. P. Agarwal, and X.-J. Yang, “Local fractional Fourier
some local fractional abstract differential equation with fractal series with application to wave equation in fractal vibrating
conditions,” Journal of Applied Functional Analysis, vol. 8, no. 1, string,” Abstract and Applied Analysis, vol. 2012, Article ID
pp. 92–99, 2013. 567401, 15 pages, 2012.
[30] W. H. Su, X. J. Yang, H. Jafari, and D. Baleanu, “Fractional [48] G. A. Anastassiou and O. Duman, Advances in Applied Math-
complex transform method for wave equations on Cantor ematics and Approximation Theory, Springer, New York, NY,
sets within local fractional differential operator,” Advances in USA, 2013.
Difference Equations, vol. 2013, no. 1, pp. 97–107, 2013. [49] X. J. Yang, H. M. Srivastava, J. H. He, and D. Baleanu, “Cantor-
[31] M. S. Hu, D. Baleanu, and X. J. Yang, “One-phase problems type cylindrical-coordinate method for differential equations
for discontinuous heat transfer in fractal media,” Mathematical with local fractional derivatives,” Physics Letters A, vol. 377, no.
Problems in Engineering, vol. 2013, Article ID 358473, 3 pages, 28–30, pp. 1696–1700, 2013.
2013. [50] X. Yang, M. Liao, and J. Chen, “A novel approach to pro-
[32] K. M. Kolwankar and A. D. Gangal, “Local fractional Fokker- cessing fractal signals using the Yang-Fourier transforms,” in
Planck equation,” Physical Review Letters, vol. 80, no. 2, pp. 214– Proceedings of the International Workshop on Information and
217, 1998. Electronics Engineering (IWIEE ’12), vol. 29, pp. 2950–2954,
March 2012.
[33] A. Carpinteri and A. Sapora, “Diffusion problems in fractal
media defined on Cantor sets,” ZAMM Journal of Applied [51] W. Zhong, F. Gao, and X. Shen, “Applications of Yang-Fourier
Mathematics and Mechanics, vol. 90, no. 3, pp. 203–210, 2010. transform to local fractional equations with local fractional
derivative and local fractional integral,” Advanced Materials
[34] A. Carpinteri and P. Cornetti, “A fractional calculus approach to Research, vol. 461, pp. 306–310, 2012.
the description of stress and strain localization in fractal media,”
Chaos, Solitons and Fractals, vol. 13, no. 1, pp. 85–94, 2002. [52] W. P. Zhong and F. Gao, “Application of the Yang-Laplace trans-
forms to solution to nonlinear fractional wave equation with
[35] A. Carpinteri, B. Chiaia, and P. Cornetti, “The elastic problem fractional derivative,” in Proceedings of the 3rd International
for fractal media: basic theory and finite element formulation,” Conference on Computer Technology and Development, pp. 209–
Computers and Structures, vol. 82, no. 6, pp. 499–508, 2004. 213, ASME, 2011.
[36] A. Babakhani and V. Daftardar-Gejji, “On calculus of local
fractional derivatives,” Journal of Mathematical Analysis and
Applications, vol. 270, no. 1, pp. 66–79, 2002.
[37] F. Ben Adda and J. Cresson, “About non-differentiable func-
tions,” Journal of Mathematical Analysis and Applications, vol.
263, no. 2, pp. 721–737, 2001.
[38] Y. Chen, Y. Yan, and K. Zhang, “On the local fractional deriv-
ative,” Journal of Mathematical Analysis and Applications, vol.
362, no. 1, pp. 17–33, 2010.
[39] W. Chen, “Time-space fabric underlying anomalous diffusion,”
Chaos, Solitons and Fractals, vol. 28, no. 4, pp. 923–925, 2006.
[40] W. Chen, H. Sun, X. Zhang, and D. Korošak, “Anomalous diffu-
sion modeling by fractal and fractional derivatives,” Computers
& Mathematics with Applications, vol. 59, no. 5, pp. 1754–1758,
2010.
[41] A. K. Golmankhaneh, V. Fazlollahi, and D. Baleanu, “Newto-
nian mechanics on fractals subset of real-line,” Romania Reports
in Physics, vol. 65, pp. 84–93, 2013.
[42] A. Parvate and A. D. Gangal, “Fractal differential equations and
fractal-time dynamical systems,” Pramana, vol. 64, no. 3, pp.
389–409, 2005.
[43] G. Jumarie, “Probability calculus of fractional order and frac-
tional Taylor’s series application to Fokker-Planck equation and
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2013, Article ID 160681, 8 pages
http://dx.doi.org/10.1155/2013/160681
Research Article
Nonlinear Fractional Jaulent-Miodek and Whitham-Broer-Kaup
Equations within Sumudu Transform
Copyright © 2013 A. Atangana and D. Baleanu. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
We solve the system of nonlinear fractional Jaulent-Miodek and Whitham-Broer-Kaup equations via the Sumudu transform
homotopy method (STHPM). The method is easy to apply, accurate, and reliable.
FWBK equation (1) describes the dispersive long wave in Definition 5. The Jumarie Fractional order derivative is given
shallow water, where 𝑢(𝑥, 𝑡) is the field of horizontal velocity, as follows [24]:
V(𝑥, 𝑡) is the height which deviates from the equilibrium
position of liquid, and 𝛼 and 𝛽 are constants that represent 1 𝑑𝑛
𝐷𝑥𝛼 (𝑓 (𝑥)) =
different powers. If 𝛼 = 0 and 𝛽 = 1, (1) reduces to Γ (𝑛 − 𝛼) 𝑑𝑥𝑛
the classical long-wave equations which describe the shallow 𝑥
water wave with diffusion [6]. If 𝛼 = 1 and 𝛽 = 0, (8)
× ∫ (𝑥 − 𝑡)𝑛−𝛼−1 {𝑓 (𝑡) − 𝑓 (0)} 𝑑𝑡,
(1) becomes the modified Boussinesq equations [7, 8]. FJM 0
equation (2) appears in several areas of science such as 𝑛 − 1 ≤ 𝛼 ≤ 𝑛.
condense matter physics [9], fluid mechanics [10], plasma
physics [11], and optics [12] and associates with energy-
Lemma 6. If 𝑚 − 1 < 𝛼 ≤ 𝑚, 𝑚 ∈ N and 𝑓 ∈ 𝐶𝜇𝑚 , 𝜇 ≥ −1,
dependent Schrödinger potential [13, 14].
then
The paper is organized as follows. In Section 2, we intro-
duce briefly some of the basic tools of fractional order and 𝐷𝛼 𝐽𝛼 𝑓 (𝑥) = 𝑓 (𝑥) ,
of the Sumudu transform method. We show the numerical
𝑚−1
results in Section 4. The conclusions can be seen in Section 5. 𝑥𝑘 (9)
𝐽𝛼 𝐷0𝛼 𝑓 (𝑥) = 𝑓 (𝑥) − ∑ 𝑓(𝑘) (0+ ) , 𝑥 > 0.
𝑘=0
𝑘!
2. Basic Tools
Definition 7 (partial derivatives of fractional order [15, 16, 19]).
2.1. Properties and Definitions Assume now that 𝑓(x) is a function of 𝑛 variables 𝑥𝑖 , 𝑖 =
Definition 1 (see [15–24]). A real function 𝑓(𝑥), 𝑥 > 0, is said 1, . . . , 𝑛 also of class 𝐶 on 𝐷 ∈ R𝑛 . As an extension of
to be in the space 𝐶𝜇 , 𝜇 ∈ R if there exists a real number Definition 3, we define partial derivative of order 𝛼 for 𝑓 with
respect to 𝑥𝑖 the function
𝑝 > 𝜇, such that 𝑓(𝑥) = 𝑥𝑝 ℎ(𝑥), where ℎ(𝑥) ∈ 𝐶[0, ∞), and
it is said to be in space 𝐶𝜇𝑚 if 𝑓(𝑚) ∈ 𝐶𝜇 , 𝑚 ∈ N. 1 𝑥𝑖
𝑚−𝛼−1 𝑚
𝑎𝜕𝛼x 𝑓 = ∫ (𝑥𝑖 − 𝑡) 𝜕𝑥𝑖 𝑓 (𝑥𝑗 ) 𝑑𝑡, (10)
Γ (𝑚 − 𝛼) 𝑎 𝑥𝑗=𝑡
Definition 2 (see [15–24]). The Riemann-Liouville fractional
integral operator of order 𝛼 ≥ 0, of a function𝑓 ∈ 𝐶𝜇 , 𝜇 ≥ −1,
where 𝜕𝑥𝑚𝑖 is the usual partial derivative of integer order 𝑚.
is defined as
𝑥
1
𝐽𝛼 𝑓 (𝑥) = ∫ (𝑥 − 𝑡)𝛼−1 𝑓 (𝑡) 𝑑𝑡, 𝛼 > 0, 𝑥 > 0, 3. Background of Sumudu Transform
Γ (𝛼) 0 (4)
Definition 8 (see [25]). The Sumudu transform of a function
𝐽0 𝑓 (𝑥) = 𝑓 (𝑥) . 𝑓(𝑡), defined for all real numbers 𝑡 ≥ 0, is the function 𝐹𝑠 (𝑢),
Properties of the operator can be found in [15–23]; we defined by
mention only the following. ∞
For 𝑓 ∈ 𝐶𝜇 , 𝜇 ≥ −1, 𝛼, 𝛽 ≥ 0 and 𝛾 > −1 1 𝑡
𝑆 (𝑓 (𝑡)) = 𝐹𝑠 (𝑢) = ∫ exp [− ] 𝑓 (𝑡) 𝑑𝑡. (11)
0 𝑢 𝑢
𝐽𝛼 𝐽𝛽 𝑓 (𝑥) = 𝐽𝛼+𝛽 𝑓 (𝑥) , 𝐽𝛼 𝐽𝛽 𝑓 (𝑥) = 𝐽𝛽 𝐽𝛼 𝑓 (𝑥) ,
Theorem 9 (see [26]). Let 𝐺(𝑢) be the Sumudu transform of
𝛼 𝛾 Γ (𝛾 + 1) 𝛼+𝛾 (5) 𝑓(𝑡) such that
𝐽 𝑥 = 𝑥 .
Γ (𝛼 + 𝛾 + 1) (i) (𝐺(1/𝑠)/𝑠) is a meromorphic function, with singulari-
Definition 3. The Caputo fractional order derivative is given ties having Re[𝑠] ≤ 𝛾;
as follows [15–18]: (ii) there exist a circular region Γ with radius 𝑅 and positive
𝐶 𝛼 constants 𝑀 and 𝐾 with |𝐺(1/𝑠)/𝑠| < 𝑀𝑅−𝐾 , then the
0 𝐷𝑥 (𝑓 (𝑥))
function 𝑓(𝑡) is given by
1 𝑥 𝑑𝑛 𝑓 (𝑡)
= ∫ (𝑥 − 𝑡)𝑛−𝛼−1 𝑑𝑡, 𝑛 − 1 ≤ 𝛼 ≤ 𝑛. 1 𝛾+𝑖∞ 1 𝑑𝑠
Γ (𝑛 − 𝛼) 0 𝑑𝑡𝑛 𝑆−1 (𝐺 (𝑠)) = ∫ exp [𝑠𝑡] 𝐺 ( )
(6) 2𝜋𝑖 𝛾−𝑖∞ 𝑠 𝑠
(12)
Definition 4. The Riemann-Liouville fractional order deriva- 𝐺 (1/𝑠)
= ∑ residual [exp [𝑠𝑡] ].
tive is given as follows [16–24]: 𝑠
𝐷𝑥𝛼 (𝑓 (𝑥))
For the proof see [26].
𝑛 𝑥
1 𝑑
= ∫ (𝑥 − 𝑡)𝑛−𝛼−1 𝑓 (𝑡) 𝑑𝑡, 𝑛 − 1 ≤ 𝛼 ≤ 𝑛.
Γ (𝑛 − 𝛼) 𝑑𝑥𝑛 0 3.1. Basics of the Sumudu Transform Homotopy Perturbation
(7) Method. We illustrate the basic idea of this method [27–32]
Abstract and Applied Analysis 3
by considering a general fractional nonlinear nonhomoge- which is the coupling of the Sumudu transform and the
neous partial differential equation with the initial condition HPM using He’s polynomials. Comparing the coefficients of
of the following form: like powers of 𝑝, the following approximations are obtained
[29, 30]:
𝐷𝑡𝛼 𝑈 (𝑥, 𝑡) = 𝐿 (𝑈 (𝑥, 𝑡)) + 𝑁 (𝑈 (𝑥, 𝑡)) + 𝑓 (𝑥, 𝑡) , 𝛼 > 0,
(13)
𝑝0 :𝑈0 (𝑥, 𝑡) = 𝐺 (𝑥, 𝑡) ,
subject to the initial condition
𝑝1 :𝑈1 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑈0 (𝑥, 𝑡)) + 𝐻0 (𝑈)]] ,
𝐷0𝑘 𝑈 (𝑥, 0) = 𝑔𝑘 , (𝑘 = 0, . . . , 𝑛 − 1) ,
(14)
𝐷0𝑛 𝑈 (𝑥, 0) = 0, 𝑛 = [𝛼] , 𝑝2 :𝑈2 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑈1 (𝑥, 𝑡)) + 𝐻1 (𝑈)]] , (22)
where 𝐷𝑡𝛼 denotes without loss of generality the Caputo 𝑝3 :𝑈3 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑈2 (𝑥, 𝑡)) + 𝐻2 (𝑈)]] ,
fraction derivative operator, 𝑓 is a known function, 𝑁 is
the general nonlinear fractional differential operator, and 𝐿 𝑝𝑛 :𝑈𝑛 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑈𝑛−1 (𝑥, 𝑡)) + 𝐻𝑛−1 (𝑈)]] .
represents a linear fractional differential operator.
Applying the Sumudu transform on both sides of (10), we Finally, we approximate the analytical solution 𝑈(𝑥, 𝑡) by
obtain truncated series:
𝑆 [𝐷𝑡𝛼 𝑈 (𝑥, 𝑡]) = 𝑆 [𝐿 (𝑈 (𝑥, 𝑡))]
𝑁
(15)
+ 𝑆 [𝑁 (𝑈 (𝑥, 𝑡))] + 𝑆 [𝑓 (𝑥, 𝑡)] . 𝑈 (𝑥, 𝑡) = lim ∑ 𝑈𝑛 (𝑥, 𝑡) . (23)
𝑁→∞
𝑛=0
Using the property of the Sumudu transform, we have
𝑆 [𝑈 (𝑥, 𝑡)] = 𝑢𝛼 𝑆 [𝐿 (𝑈 (𝑥, 𝑡))] + 𝑢𝛼 𝑆 [𝑁 (𝑈 (𝑥, 𝑡))] The above series solutions generally converge very rapidly
(16) [29, 30].
+ 𝑢𝛼 𝑆 [𝑓 (𝑥, 𝑡)] + 𝑔 (𝑥, 𝑡) .
Now applying the Sumudu inverse on both sides of (12) we 4. Applications
obtain
In this section, we apply this method for solving the system
𝑈 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑈 (𝑥, 𝑡))] + 𝑢𝛼 𝑆 [𝑁 (𝑈 (𝑥, 𝑡))]] of the fractional differential equation. We will start with (1).
(17)
+ 𝐺 (𝑥, 𝑡) ,
4.1. Approximate Solution of (1). Following carefully the
where 𝐺(𝑥, 𝑡) represents the term arising from the known
steps involved in the STHPM, after comparing the terms of
function 𝑓(𝑥, 𝑡) and the initial conditions.
the same power of 𝑝 and choosing the appropriate initials
Now we apply the following HPM:
conditions, we arrive at the following series solutions:
∞
𝑈 (𝑥, 𝑡) = ∑ 𝑝𝑛 𝑈𝑛 (𝑥, 𝑡) . (18) 𝑐1
𝑛=0 𝑢0 (𝑥, 𝑡) = 𝐺 (𝑥, 𝑡) = − + 2𝑐1 √−𝛼 − 𝛽2 sech (𝑐1 𝑥) ,
𝑐2
The nonlinear term can be decomposed to
∞ V0 (𝑥, 𝑡) = 𝐺1 (𝑥, 𝑡)
𝑁𝑈 (𝑥, 𝑡) = ∑ 𝑝𝑛 H𝑛 (𝑈) , (19) 2
𝑛=0 = − 𝑐12 (𝛼 + 𝛽2 ) + 2𝑐12 (𝛼 + 𝛽2 ) sech(𝑐1 𝑥)
using the He’s polynomial H𝑛 (𝑈) given as
+ 2𝑐12 𝛽√−𝛼 − 𝛽2 sech (𝑐1 𝑥) tanh (𝑐1 𝑥) ,
𝑛 ∞
1 𝜕 [
H𝑛 (𝑈0 , . . . , 𝑈𝑛 ) = 𝑁 ( ∑𝑝𝑗 𝑈𝑗 (𝑥, 𝑡))] , 𝑢1 (𝑥, 𝑡) = 𝑆−1 [𝑢𝛼 𝑆 [𝐿 (𝑢0 (𝑥, 𝑡)) + 𝐻0 (𝑢)]]
𝑛! 𝜕𝑝𝑛 𝑗=0 (20)
[ ]
3
𝑛 = 0, 1, 2 . . . . 𝑐12 𝑡𝜂 sech(𝑐1 𝑥)
=
𝑐2 Γ (𝜂 + 1)
Substituting (15) and (16) gives
∞
× (𝑐1 𝑐2 𝛽√−𝛼 − 𝛽2 cos (2𝑐1 𝑥)
∑ 𝑝𝑛 𝑈𝑛 (𝑥, 𝑡)
𝑛=0
∞
+ 4𝑐1 𝑐2 (𝛼 + 𝛽2 ) sinh (𝑐1 𝑥) + √−𝛼 − 𝛽2
= 𝐺 (𝑥, 𝑡) + 𝑝 [𝑆−1 [𝑢𝛼 𝑆 [𝐿 ( ∑ 𝑝𝑛 𝑈𝑛 (𝑥, 𝑡))] (21)
𝑛=0 × ( − 3𝑐1 𝑐2 𝛽
∞
+ 𝑢𝛼 𝑆 [𝑁 ( ∑ 𝑝𝑛 𝑈𝑛 (𝑥, 𝑡))]]] , + (𝑐1 − 𝑐2 ) sinh (2𝑐1 𝑥) )) ,
𝑛=0
4 Abstract and Applied Analysis
1 5
− √−𝛼 − 𝛽2 sech (𝑐1 𝑥) 0.95
2
+ 41+𝜂 √−𝛼 − 𝛽2 ( − 2𝑐1 𝑐2 + 𝑐22 + 𝑐12 V (𝑥, 𝑡) = V0 (𝑥, 𝑡) + V1 (𝑥, 𝑡) + V2 (𝑥, 𝑡) + V3 (𝑥, 𝑡) + ⋅ ⋅ ⋅ .
(25)
× (1 + 𝑐22 (12𝛼 + 31𝛽2 )))
Figures 1, 2, 3, and 4 show the graphical representation of the
× cosh (2𝑐1 𝑥) + 42+𝜂 𝑐1 𝑐2 (−𝑐1 + 𝑐2 ) (𝛼 + 𝛽2 )
approximated solution of the system of nonlinear fractional
Whitham-Broer-Kaup equation for 𝜂 = 0.9, 𝜇 = 0.98, 𝑐1 =
× cosh (3𝑐1 𝑥) − 4𝜂 𝑐12 √−𝛼 − 𝛽2 cosh (4𝑐1 𝑥) 𝑐2 = 0.1, and 𝛽 = 𝛼 = 0.1.
Abstract and Applied Analysis 5
𝑐2 𝑐𝑥 2
𝑢0 (𝑥, 𝑡) = (1 − sech ( ) ) ,
8 2
1
𝑐𝑥 2
0.99 2200 V0 (𝑥, 𝑡) = 𝑐 sech ( ),
𝑢(𝑥, 𝑡)
2
150
0.98
𝑐5 𝑡𝜂 sech(𝑐𝑥/2)5
𝑢1 (𝑥, 𝑡) = −
100 128Γ (𝜂 + 1)
−100 𝑡
𝑐𝑥 3𝑐𝑥
0 50 × (192 cosh [ ] − 32 cosh [ ]
𝑥 2 2
100 𝑐𝑥 3𝑐𝑥
0 +3𝑐 (3 sinh [ ] + sinh [ ]))
2 2
Figure 2: Approximate solution of FWBK equation.
𝑐𝑥
× tanh [ ],
2
𝑐𝑥
𝑥 × (71 − cosh [𝑐𝑥] + 6𝑐 tanh [ ]) ,
−150 −100 −50 50 100 150 2
−0.0005 𝑢2 (𝑥, 𝑡)
4−10−𝜂 𝑐5 𝑡𝜂 (𝑐𝑥/2)15
−0.001
=(
Γ (1 + 𝜇) Γ (1 + 𝜂) Γ (0.5 + 𝜂) Γ (1 + 𝜇 + 𝜂)
𝑐𝑥 4
−0.0015 × (−32𝑐3 √𝜋𝑡𝜂 𝜇 cosh ( ) Γ (𝜇)
2
× Γ (1 + 𝜂 + 𝜇) Γ (1 + 2𝜂 + 𝜇)
Figure 3: Approximate solution of FWBK equation.
𝑐𝑥
× (221184 − 20532𝑐2 ) cosh ( )
2
3𝑐𝑥
+ 6 (−11008 + 4813𝑐2 ) cosh ( )
2
5𝑐𝑥 5𝑐𝑥
− 69120 cosh ( ) − 8622𝑐2 cosh ( )
2 2
7𝑐𝑥 7𝑐𝑥
0.002 + 10368 cosh ( ) + 267𝑐2 cosh ( )
2 2
0.001 2
200
(𝑥, 𝑡)
9𝑐𝑥 9𝑐𝑥
0 150 − 128 cosh ( ) + 9𝑐2 cosh ( )
−0.001 2 2
100 𝑐𝑥 𝑐𝑥
𝑡
−100 + 61032𝑐 sinh ( ) − 2772𝑐3 sinh ( )
2 2
0 50
𝑥 3𝑐𝑥 3𝑐𝑥
100 + 29040𝑐 sinh ( ) + 828𝑐3 sinh ( )
2 2
5𝑐𝑥 5𝑐𝑥
Figure 4: Approximate solution of FWBK equation. − 27312𝑐 sinh ( ) + 108𝑐3 sinh ( )
2 2
6 Abstract and Applied Analysis
𝑐𝑥 2
× (1 + 2𝜂 + 𝜇) sinh ( ) V2 (𝑥, 𝑡)
2
𝑐𝑥 13
× (−2𝑐 + sinh (𝑐𝑥)) = (2−17−2𝜂 𝑐4 𝑡𝜇 sech ( )
2
+ 1024𝑐3 𝑡𝜂 𝜂Γ (𝜂) Γ (1 + 𝜇) Γ 2
× (Γ(1 + 𝜇) Γ (1 + 𝜂) Γ (0.5 + 𝜇) Γ
𝑐𝑥 2 −1
× (1 + 2𝜂 + 𝜇) sinh ( ) × (1 + 𝜇 + 𝜂) Γ (1 + 3𝜇))
2
𝑐𝑥 3𝑐𝑥 𝑐𝑥 4
× (−15745 cosh ( ) + 12951 cosh ( ) × (3 × 44+𝜇 𝑐4 𝑡𝜂 cosh ( )
2 2 2
3𝑐𝑥 7𝑐𝑥 2
− 1175 cosh ( ) + cosh ( ) × Γ(1 + 𝜇) Γ (1 + 𝜂) Γ (0.5 + 𝜇)
2 2
𝑐𝑥
𝑐𝑥 × Γ (1 + 3𝜇) sinh ( )
− 6240𝑐 sinh ( ) 2
2
𝑐𝑥 3𝑐𝑥
3𝑐𝑥 × (896 cosh ( ) − 608 cosh ( )
+ 1728𝑐 sinh ( ) 2 2
2
5𝑐𝑥 𝑐𝑥
5𝑐𝑥 + 32 cosh ( ) + 78𝑐 sinh ( )
−96𝑐 sinh ( )) 2 2
2
3𝑐𝑥 5𝑐𝑥
2 𝑐𝑥 +3𝑐 sinh ( ) − 5𝑐 sinh ( ))
6 𝜂+𝜇
+ 2𝑐 𝑡 Γ(1 + 𝜂 + 𝜇) sinh ( ) 2 2
2
+ Γ (𝜂 + 1) Γ (1 + 𝜂 + 𝜇)
× (−235648 − 1154128𝑐2 + 15804𝑐4
× (15 × 4𝜇 Γ (0.5 + 𝜇)
2 4
− 16 (5584 − 7358𝑐 + 1125𝑐 )
𝑐𝑥 9
× (− 65536𝜇 cosh ( ) Γ (𝜇)
× cosh (𝑐𝑥) 2
0.0012
0.1 0.0011
𝑢(𝑥, 𝑡)
20
0.075 20 0.00088
(𝑥, 𝑡)
0.05 0.00066 15
0.0004
04
0.025 15 10 𝑡
0 −100
10 𝑡 0 5
−100
𝑥
5 100
0 0
𝑥
100
0 Figure 6: Approximate solution of FJM equation.
Figure 5: Approximate solution of FJM equation.
propagation,” Computer Physics Communications, vol. 182, no. [25] G. K. Watugala, “Sumudu transform: a new integral trans-
12, pp. 2464–2470, 2011. form to solve differential equations and control engineering
[7] F. Xie, Z. Yan, and H. Zhang, “Explicit and exact traveling wave problems,” International Journal of Mathematical Education in
solutions of Whitham-Broer-Kaup shallow water equations,” Science and Technology, vol. 24, no. 1, pp. 35–43, 1993.
Physics Letters A, vol. 285, no. 1-2, pp. 76–80, 2001. [26] S. Weerakoon, “Application of Sumudu transform to partial
[8] B. A. Kupershmidt, “Mathematics of dispersive water waves,” differential equations,” International Journal of Mathematical
Communications in Mathematical Physics, vol. 99, no. 1, pp. 51– Education in Science and Technology, vol. 25, no. 2, pp. 277–283,
73, 1985. 1994.
[9] W.-X. Ma, C.-X. Li, and J. He, “A second Wronskian formulation [27] J. Singh, D. Kumar, and A. Kılıçman, “Homotopy perturbation
of the Boussinesq equation,” Nonlinear Analysis. Theory, Meth- method for fractional gas dynamics equation using Sumudu
ods & Applications, vol. 70, no. 12, pp. 4245–4258, 2009. transform,” Abstract and Applied Analysis, vol. 2013, Article ID
934060, 8 pages, 2013.
[10] T. Hong, Y. Z. Wang, and Y. S. Huo, “Bogoliubov quasiparticles
carried by dark solitonic excitations in non-uniform Bose- [28] H. Eltayeb and A. Kılıçman, “A note on the Sumudu transforms
Einstein condensates,” Chinese Physics Letters, vol. 15, pp. 550– and differential equations,” Applied Mathematical Sciences, vol.
552, 1998. 4, no. 21-24, pp. 1089–1098, 2010.
[11] J.-f. Zhang, “Multiple soliton solutions of the dispersive long- [29] A. Kılıçman, H. Eltayeb, and P. R. Agarwal, “On Sumudu
wave equations,” Chinese Physics Letters, vol. 16, no. 1, pp. 4–5, transform and system of differ ential equations,” Abstract and
1999. Applied Analysis, vol. 2010, Article ID 598702, 11 pages, 2010.
[12] G. C. Das, J. Sarma, and C. Uberoi, “Explosion of a soliton in a [30] V. G. Gupta and B. Sharma, “Application of Sumudu transform
multicomponent plasma,” Physics of Plasmas, vol. 4, no. 6, pp. in reaction-diffusion systems and nonlinear waves,” Applied
2095–2100, 1997. Mathematical Sciences, vol. 4, no. 9–12, pp. 435–446, 2010.
[31] H. Eltayeb and A. Kılıçman, “A note on the Sumudu transforms
[13] S. Y. Lou, “A direct perturbation method: nonlinear Schrodinger
and differential equations,” Applied Mathematical Sciences, vol.
equation with loss,” Chinese Physics Letters, vol. 16, pp. 659–661,
4, no. 21–24, pp. 1089–1098, 2010.
1999.
[32] F. B. M. Belgacem, A. A. Karaballi, and S. L. Kalla, “Analyt-
[14] A. Atangana and A. H. Cloot, “Stability and convergence of the
ical investigations of the Sumudu transform and applications
space fractional variable-order Schrödinger equation,” Advan-
to integral production equations,” Mathematical Problems in
ces in Difference Equations, vol. 2013, article 80, 2013.
Engineering, no. 3-4, pp. 103–118, 2003.
[15] D. Baleanu, K. Diethelm, E. Scalas, and J. J. Trujillo, Fractional
[33] A. Abdon, “New cass of boundary value problems,” Informa-
Calculus Models and Numerical Methods, vol. 3 of Series on
tions Sciences Letters, vol. 1, no. 2, pp. 67–76, 2012.
Complexity, Nonlinearity and Chaos, World Scientific, Hacken-
sack, NJ, USA, 2012. [34] G.-C. Wu, “Adomian decomposition method for non-smooth
initial value problems,” Mathematical and Computer Modelling,
[16] A. A. Kilbas, H. M. Srivastava, and J. J. Trujillo, Theory and
vol. 54, no. 9-10, pp. 2104–2108, 2011.
Applications of Fractional Differential Equations, vol. 204 of
North-Holland Mathematics Studies, Elsevier Science B. V., [35] G. C. Wu and D. Baleanu, “Variational iteration method for
Amsterdam, The Netherlands, 2006. the Burgers’ flow with fractional derivatives-New Lagrange
multipliers,” Applied Mathematical Modelling, vol. 5, pp. 1012–
[17] K. B. Oldham and J. Spanier, The Fractional Calculus, Mathe-
1018, 2012.
matics in Science and Engineering, vol. 111, Academic Press, New
York, NY, USA, 1974. [36] G.-C. Wu and D. Baleanu, “Variational iteration method for
fractional calculus—a universal approach by Laplace trans-
[18] B. Ross, Ed., Fractional Calculus and Its Applications, vol. 457 of form,” Advances in Difference Equations, vol. 2013, article 18,
Lecture Notes in Mathematics, Springer, Berlin, Germany, 1975. 2013.
[19] Y. Luchko and R. Gorenflo, “The initial value problem for some [37] H. Xu, S.-J. Liao, and X.-C. You, “Analysis of nonlinear frac-
fractional differential equations with the Caputo derivative,” tional partial differential equations with the homotopy analysis
Preprint Series A08-98, Fachbreich Mathematik and Infor- method,” Communications in Nonlinear Science and Numerical
matik, Freic Universitat Berlin, 1998. Simulation, vol. 14, no. 4, pp. 1152–1156, 2009.
[20] A. Atangana and A. Secer, “A note on fractional order deriva- [38] J.-H. He, “Homotopy perturbation method with an auxiliary
tives and table of fractional derivatives of some special func- term,” Abstract and Applied Analysis, vol. 2012, Article ID
tions,” Abstract and Applied Analysis, vol. 2013, Article ID 857612, 7 pages, 2012.
279681, 8 pages, 2013.
[21] I. Podlubny, Fractional Differential Equations, vol. 198 of Math-
ematics in Science and Engineering, Academic Press, San Diego,
Calif, USA, 1999.
[22] S. G. Samko, A. A. Kilbas, and O. I. Marichev, Fractional
Integrals and Derivatives: Theory and Applications, Gordon and
Breach Science, Yverdon, Switzerland, 1993.
[23] K. S. Miller and B. Ross, An Introduction to the Fractional Calcu-
lus and Fractional Differential Equations, A Wiley-Interscience,
John Wiley & Sons, New York, NY, USA, 1993.
[24] G. Jumarie, “On the representation of fractional Brownian
motion as an integral with respect to (d𝑡)𝑎 ,” Applied Mathemat-
ics Letters, vol. 18, no. 7, pp. 739–748, 2005.