Volume 22, Number 5 May 2017 ISSN:1521-1398 PRINT,1572-9206 ONLINE Journal of Computational Analysis and Applications EUDOXUS PRESS,LLC
Volume 22, Number 5 May 2017 ISSN:1521-1398 PRINT,1572-9206 ONLINE
Journal of Computational Analysis and Applications EUDOXUS PRESS,LLC
Journal of Computational Analysis and Applications ISSNno.βs:1521-1398 PRINT,1572-9206 ONLINE SCOPE OF THE JOURNAL An international publication of Eudoxus Press, LLC (fourteen times annually) Editor in Chief: George Anastassiou Department of Mathematical Sciences, University of Memphis, Memphis, TN 38152-3240, U.S.A [email protected] http://www.msci.memphis.edu/~ganastss/jocaaa The main purpose of "J.Computational Analysis and Applications" is to publish high quality research articles from all subareas of Computational Mathematical Analysis and its many potential applications and connections to other areas of Mathematical Sciences. Any paper whose approach and proofs are computational,using methods from Mathematical Analysis in the broadest sense is suitable and welcome for consideration in our journal, except from Applied Numerical Analysis articles. Also plain word articles without formulas and proofs are excluded. The list of possibly connected mathematical areas with this publication includes, but is not restricted to: Applied Analysis, Applied Functional Analysis, Approximation Theory, Asymptotic Analysis, Difference Equations, Differential Equations, Partial Differential Equations, Fourier Analysis, Fractals, Fuzzy Sets, Harmonic Analysis, Inequalities, Integral Equations, Measure Theory, Moment Theory, Neural Networks, Numerical Functional Analysis, Potential Theory, Probability Theory, Real and Complex Analysis, Signal Analysis, Special Functions, Splines, Stochastic Analysis, Stochastic Processes, Summability, Tomography, Wavelets, any combination of the above, e.t.c. "J.Computational Analysis and Applications" is a peer-reviewed Journal. See the instructions for preparation and submission of articles to JoCAAA. Assistant to the Editor: Dr.Razvan Mezei,Lenoir-Rhyne University,Hickory,NC 28601, USA. Journal of Computational Analysis and Applications(JoCAAA) is published by EUDOXUS PRESS,LLC,1424 Beaver Trail Drive,Cordova,TN38016,USA,[email protected] http://www.eudoxuspress.com. Annual Subscription Prices:For USA and Canada,Institutional:Print $700, Electronic OPEN ACCESS. Individual:Print $350. For any other part of the world add $130 more(handling and postages) to the above prices for Print. No credit card payments. CopyrightΒ©2017 by Eudoxus Press,LLC,all rights reserved.JoCAAA is printed in USA. JoCAAA is reviewed and abstracted by AMS Mathematical Reviews,MATHSCI,and Zentralblaat MATH. It is strictly prohibited the reproduction and transmission of any part of JoCAAA and in any form and by any means without the written permission of the publisher.It is only allowed to educators to Xerox articles for educational purposes.The publisher assumes no responsibility for the content of published papers.
780
Editorial Board
Associate Editors of Journal of Computational Analysis and Applications
Francesco Altomare Dipartimento di Matematica Universita' di Bari Via E.Orabona, 4 70125 Bari, ITALY Tel+39-080-5442690 office +39-080-3944046 home +39-080-5963612 Fax [email protected] Approximation Theory, Functional Analysis, Semigroups and Partial Differential Equations, Positive Operators. Ravi P. Agarwal Department of Mathematics Texas A&M University - Kingsville 700 University Blvd. Kingsville, TX 78363-8202 tel: 361-593-2600 [email protected] Differential Equations, Difference Equations, Inequalities George A. Anastassiou Department of Mathematical Sciences The University of Memphis Memphis, TN 38152,U.S.A Tel.901-678-3144 e-mail: [email protected] Approximation Theory, Real Analysis, Wavelets, Neural Networks, Probability, Inequalities. J. Marshall Ash Department of Mathematics De Paul University 2219 North Kenmore Ave. Chicago, IL 60614-3504 773-325-4216 e-mail: [email protected] Real and Harmonic Analysis Dumitru Baleanu Department of Mathematics and Computer Sciences, Cankaya University, Faculty of Art and Sciences, 06530 Balgat, Ankara, Turkey, [email protected]
Fractional Differential Equations Nonlinear Analysis, Fractional Dynamics Carlo Bardaro Dipartimento di Matematica e Informatica Universita di Perugia Via Vanvitelli 1 06123 Perugia, ITALY TEL+390755853822 +390755855034 FAX+390755855024 E-mail [email protected] Web site: http://www.unipg.it/~bardaro/ Functional Analysis and Approximation Theory, Signal Analysis, Measure Theory, Real Analysis.
Martin Bohner Department of Mathematics and Statistics, Missouri S&T Rolla, MO 65409-0020, USA [email protected] web.mst.edu/~bohner Difference equations, differential equations, dynamic equations on time scale, applications in economics, finance, biology. Jerry L. Bona Department of Mathematics The University of Illinois at Chicago 851 S. Morgan St. CS 249 Chicago, IL 60601 e-mail:[email protected] Partial Differential Equations, Fluid Dynamics Luis A. Caffarelli Department of Mathematics The University of Texas at Austin Austin, Texas 78712-1082 512-471-3160 e-mail: [email protected] Partial Differential Equations George Cybenko Thayer School of Engineering
781
Dartmouth College 8000 Cummings Hall, Hanover, NH 03755-8000 603-646-3843 (X 3546 Secr.) e-mail:[email protected] Approximation Theory and Neural Networks Sever S. Dragomir School of Computer Science and Mathematics, Victoria University, PO Box 14428, Melbourne City, MC 8001, AUSTRALIA Tel. +61 3 9688 4437 Fax +61 3 9688 4050 [email protected] Inequalities, Functional Analysis, Numerical Analysis, Approximations, Information Theory, Stochastics. Oktay Duman TOBB University of Economics and Technology, Department of Mathematics, TR-06530, Ankara, Turkey, [email protected] Classical Approximation Theory, Summability Theory, Statistical Convergence and its Applications Saber N. Elaydi Department Of Mathematics Trinity University 715 Stadium Dr. San Antonio, TX 78212-7200 210-736-8246 e-mail: [email protected] Ordinary Differential Equations, Difference Equations Christodoulos A. Floudas Department of Chemical Engineering Princeton University Princeton,NJ 08544-5263 609-258-4595(x4619 assistant) e-mail: [email protected] Optimization Theory&Applications, Global Optimization J .A. Goldstein Department of Mathematical Sciences The University of Memphis Memphis, TN 38152 901-678-3130 [email protected]
Partial Differential Equations, Semigroups of Operators H. H. Gonska Department of Mathematics University of Duisburg Duisburg, D-47048 Germany 011-49-203-379-3542 e-mail: [email protected] Approximation Theory, Computer Aided Geometric Design John R. Graef Department of Mathematics University of Tennessee at Chattanooga Chattanooga, TN 37304 USA [email protected] Ordinary and functional differential equations, difference equations, impulsive systems, differential inclusions, dynamic equations on time scales, control theory and their applications Weimin Han Department of Mathematics University of Iowa Iowa City, IA 52242-1419 319-335-0770 e-mail: [email protected] Numerical analysis, Finite element method, Numerical PDE, Variational inequalities, Computational mechanics Tian-Xiao He Department of Mathematics and Computer Science P.O. Box 2900, Illinois Wesleyan University Bloomington, IL 61702-2900, USA Tel (309)556-3089 Fax (309)556-3864 [email protected] Approximations, Wavelet, Integration Theory, Numerical Analysis, Analytic Combinatorics Margareta Heilmann Faculty of Mathematics and Natural Sciences, University of Wuppertal GauΓstraΓe 20 D-42119 Wuppertal, Germany, [email protected]
782
Approximation Theory (Positive Linear Operators) Xing-Biao Hu Institute of Computational Mathematics AMSS, Chinese Academy of Sciences Beijing, 100190, CHINA [email protected] Computational Mathematics Jong Kyu Kim Department of Mathematics Kyungnam University Masan Kyungnam,631-701,Korea Tel 82-(55)-249-2211 Fax 82-(55)-243-8609 [email protected] Nonlinear Functional Analysis, Variational Inequalities, Nonlinear Ergodic Theory, ODE, PDE, Functional Equations. Robert Kozma Department of Mathematical Sciences The University of Memphis Memphis, TN 38152, USA [email protected] Neural Networks, Reproducing Kernel Hilbert Spaces, Neural Percolation Theory Mustafa Kulenovic Department of Mathematics University of Rhode Island Kingston, RI 02881,USA [email protected] Differential and Difference Equations Irena Lasiecka Department of Mathematical Sciences University of Memphis Memphis, TN 38152 PDE, Control Theory, Functional Analysis, [email protected]
Burkhard Lenze Fachbereich Informatik Fachhochschule Dortmund University of Applied Sciences Postfach 105018 D-44047 Dortmund, Germany e-mail: [email protected] Real Networks, Fourier Analysis, Approximation Theory
Hrushikesh N. Mhaskar Department Of Mathematics California State University Los Angeles, CA 90032 626-914-7002 e-mail: [email protected] Orthogonal Polynomials, Approximation Theory, Splines, Wavelets, Neural Networks Ram N. Mohapatra Department of Mathematics University of Central Florida Orlando, FL 32816-1364 tel.407-823-5080 [email protected] Real and Complex Analysis, Approximation Th., Fourier Analysis, Fuzzy Sets and Systems Gaston M. N'Guerekata Department of Mathematics Morgan State University Baltimore, MD 21251, USA tel: 1-443-885-4373 Fax 1-443-885-8216 Gaston.N'[email protected] [email protected] Nonlinear Evolution Equations, Abstract Harmonic Analysis, Fractional Differential Equations, Almost Periodicity & Almost Automorphy M.Zuhair Nashed Department Of Mathematics University of Central Florida PO Box 161364 Orlando, FL 32816-1364 e-mail: [email protected] Inverse and Ill-Posed problems, Numerical Functional Analysis, Integral Equations, Optimization, Signal Analysis Mubenga N. Nkashama Department OF Mathematics University of Alabama at Birmingham Birmingham, AL 35294-1170 205-934-2154 e-mail: [email protected] Ordinary Differential Equations, Partial Differential Equations Vassilis Papanicolaou Department of Mathematics
783
National Technical University of Athens Zografou campus, 157 80 Athens, Greece tel:: +30(210) 772 1722 Fax +30(210) 772 1775 [email protected] Partial Differential Equations, Probability Choonkil Park Department of Mathematics Hanyang University Seoul 133-791 S. Korea, [email protected] Functional Equations Svetlozar (Zari) Rachev, Professor of Finance, College of Business, and Director of Quantitative Finance Program, Department of Applied Mathematics & Statistics Stonybrook University 312 Harriman Hall, Stony Brook, NY 11794-3775 tel: +1-631-632-1998, [email protected] Alexander G. Ramm Mathematics Department Kansas State University Manhattan, KS 66506-2602 e-mail: [email protected] Inverse and Ill-posed Problems, Scattering Theory, Operator Theory, Theoretical Numerical Analysis, Wave Propagation, Signal Processing and Tomography Tomasz Rychlik Polish Academy of Sciences Instytut Matematyczny PAN 00-956 Warszawa, skr. poczt. 21 ul. Εniadeckich 8 Poland [email protected] Mathematical Statistics, Probabilistic Inequalities Boris Shekhtman Department of Mathematics University of South Florida Tampa, FL 33620, USA Tel 813-974-9710 [email protected]
Approximation Theory, Banach spaces, Classical Analysis T. E. Simos Department of Computer Science and Technology Faculty of Sciences and Technology University of Peloponnese GR-221 00 Tripolis, Greece Postal Address: 26 Menelaou St. Anfithea - Paleon Faliron GR-175 64 Athens, Greece [email protected] Numerical Analysis H. M. Srivastava Department of Mathematics and Statistics University of Victoria Victoria, British Columbia V8W 3R4 Canada tel.250-472-5313; office,250-477-6960 home, fax 250-721-8962 [email protected] Real and Complex Analysis, Fractional Calculus and Appl., Integral Equations and Transforms, Higher Transcendental Functions and Appl.,q-Series and q-Polynomials, Analytic Number Th. I. P. Stavroulakis Department of Mathematics University of Ioannina 451-10 Ioannina, Greece [email protected] Differential Equations Phone +3-065-109-8283 Manfred Tasche Department of Mathematics University of Rostock D-18051 Rostock, Germany [email protected] rostock.de Numerical Fourier Analysis, Fourier Analysis, Harmonic Analysis, Signal Analysis, Spectral Methods, Wavelets, Splines, Approximation Theory Roberto Triggiani Department of Mathematical Sciences University of Memphis Memphis, TN 38152 PDE, Control Theory, Functional
784
Analysis, [email protected]
Juan J. Trujillo University of La Laguna Departamento de Analisis Matematico C/Astr.Fco.Sanchez s/n 38271. LaLaguna. Tenerife. SPAIN Tel/Fax 34-922-318209 [email protected] Fractional: Differential Equations-Operators-Fourier Transforms, Special functions, Approximations, and Applications Ram Verma International Publications 1200 Dallas Drive #824 Denton, TX 76205, USA [email protected] Applied Nonlinear Analysis, Numerical Analysis, Variational Inequalities, Optimization Theory, Computational Mathematics, Operator Theory
Xiang Ming Yu Department of Mathematical Sciences Southwest Missouri State University Springfield, MO 65804-0094 417-836-5931 [email protected] Classical Approximation Theory, Wavelets Lotfi A. Zadeh Professor in the Graduate School and Director, Computer Initiative, Soft Computing (BISC) Computer Science Division University of California at Berkeley Berkeley, CA 94720 Office: 510-642-4959 Sec: 510-642-8271 Home: 510-526-2569 FAX: 510-642-1712 [email protected] Fuzzyness, Artificial Intelligence, Natural language processing, Fuzzy logic Richard A. Zalik Department of Mathematics Auburn University Auburn University, AL 36849-5310
USA. Tel 334-844-6557 office 678-642-8703 home Fax 334-844-6555 [email protected] Approximation Theory, Chebychev Systems, Wavelet Theory Ahmed I. Zayed Department of Mathematical Sciences DePaul University 2320 N. Kenmore Ave. Chicago, IL 60614-3250 773-325-7808 e-mail: [email protected] Shannon sampling theory, Harmonic analysis and wavelets, Special functions and orthogonal polynomials, Integral transforms Ding-Xuan Zhou Department Of Mathematics City University of Hong Kong 83 Tat Chee Avenue Kowloon, Hong Kong 852-2788 9708,Fax:852-2788 8561 e-mail: [email protected] Approximation Theory, Spline functions, Wavelets Xin-long Zhou Fachbereich Mathematik, Fachgebiet Informatik Gerhard-Mercator-Universitat Duisburg Lotharstr.65, D-47048 Duisburg, Germany e-mail:[email protected] duisburg.de Fourier Analysis, Computer-Aided Geometric Design, Computational Complexity, Multivariate Approximation Theory, Approximation and Interpolation Theory
785
Instructions to Contributors
Journal of Computational Analysis and Applications An international publication of Eudoxus Press, LLC, of TN.
Editor in Chief: George Anastassiou
Department of Mathematical Sciences University of Memphis
Memphis, TN 38152-3240, U.S.A.
1. Manuscripts files in Latex and PDF and in English, should be submitted via email to the Editor-in-Chief: Prof.George A. Anastassiou Department of Mathematical Sciences The University of Memphis Memphis,TN 38152, USA. Tel. 901.678.3144 e-mail: [email protected] Authors may want to recommend an associate editor the most related to the submission to possibly handle it. Also authors may want to submit a list of six possible referees, to be used in case we cannot find related referees by ourselves. 2. Manuscripts should be typed using any of TEX,LaTEX,AMS-TEX,or AMS-LaTEX and according to EUDOXUS PRESS, LLC. LATEX STYLE FILE. (Click HERE to save a copy of the style file.)They should be carefully prepared in all respects. Submitted articles should be brightly typed (not dot-matrix), double spaced, in ten point type size and in 8(1/2)x11 inch area per page. Manuscripts should have generous margins on all sides and should not exceed 24 pages. 3. Submission is a representation that the manuscript has not been published previously in this or any other similar form and is not currently under consideration for publication elsewhere. A statement transferring from the authors(or their employers,if they hold the copyright) to Eudoxus Press, LLC, will be required before the manuscript can be accepted for publication.The Editor-in-Chief will supply the necessary forms for this transfer.Such a written transfer of copyright,which previously was assumed to be implicit in the act of submitting a manuscript,is necessary under the U.S.Copyright Law in order for the publisher to carry through the dissemination of research results and reviews as widely and effective as possible.
786
4. The paper starts with the title of the article, author's name(s) (no titles or degrees), author's affiliation(s) and e-mail addresses. The affiliation should comprise the department, institution (usually university or company), city, state (and/or nation) and mail code. The following items, 5 and 6, should be on page no. 1 of the paper. 5. An abstract is to be provided, preferably no longer than 150 words. 6. A list of 5 key words is to be provided directly below the abstract. Key words should express the precise content of the manuscript, as they are used for indexing purposes. The main body of the paper should begin on page no. 1, if possible. 7. All sections should be numbered with Arabic numerals (such as: 1. INTRODUCTION) . Subsections should be identified with section and subsection numbers (such as 6.1. Second-Value Subheading). If applicable, an independent single-number system (one for each category) should be used to label all theorems, lemmas, propositions, corollaries, definitions, remarks, examples, etc. The label (such as Lemma 7) should be typed with paragraph indentation, followed by a period and the lemma itself. 8. Mathematical notation must be typeset. Equations should be numbered consecutively with Arabic numerals in parentheses placed flush right, and should be thusly referred to in the text [such as Eqs.(2) and (5)]. The running title must be placed at the top of even numbered pages and the first author's name, et al., must be placed at the top of the odd numbed pages. 9. Illustrations (photographs, drawings, diagrams, and charts) are to be numbered in one consecutive series of Arabic numerals. The captions for illustrations should be typed double space. All illustrations, charts, tables, etc., must be embedded in the body of the manuscript in proper, final, print position. In particular, manuscript, source, and PDF file version must be at camera ready stage for publication or they cannot be considered. Tables are to be numbered (with Roman numerals) and referred to by number in the text. Center the title above the table, and type explanatory footnotes (indicated by superscript lowercase letters) below the table. 10. List references alphabetically at the end of the paper and number them consecutively. Each must be cited in the text by the appropriate Arabic numeral in square brackets on the baseline. References should include (in the following order): initials of first and middle name, last name of author(s) title of article,
787
name of publication, volume number, inclusive pages, and year of publication. Authors should follow these examples: Journal Article 1. H.H.Gonska,Degree of simultaneous approximation of bivariate functions by Gordon operators, (journal name in italics) J. Approx. Theory, 62,170-191(1990). Book 2. G.G.Lorentz, (title of book in italics) Bernstein Polynomials (2nd ed.), Chelsea,New York,1986. Contribution to a Book 3. M.K.Khan, Approximation properties of beta operators,in(title of book in italics) Progress in Approximation Theory (P.Nevai and A.Pinkus,eds.), Academic Press, New York,1991,pp.483-495. 11. All acknowledgements (including those for a grant and financial support) should occur in one paragraph that directly precedes the References section. 12. Footnotes should be avoided. When their use is absolutely necessary, footnotes should be numbered consecutively using Arabic numerals and should be typed at the bottom of the page to which they refer. Place a line above the footnote, so that it is set off from the text. Use the appropriate superscript numeral for citation in the text. 13. After each revision is made please again submit via email Latex and PDF files of the revised manuscript, including the final one. 14. Effective 1 Nov. 2009 for current journal page charges, contact the Editor in Chief. Upon acceptance of the paper an invoice will be sent to the contact author. The fee payment will be due one month from the invoice date. The article will proceed to publication only after the fee is paid. The charges are to be sent, by money order or certified check, in US dollars, payable to Eudoxus Press, LLC, to the address shown on the Eudoxus homepage. No galleys will be sent and the contact author will receive one (1) electronic copy of the journal issue in which the article appears. 15. This journal will consider for publication only papers that contain proofs for their listed results.
788
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS
DAE SAN KIM AND TAEKYUN KIM
Abstract. Carlitz introduced the degenerate Bernoulli polynomials and de-
rived, among other things, the so-called degenerate Staudt-Clausen theoremfor the degenerate Bernoulli numbers as an analogue of the classical Staudt-Clausen theorem. In this paper, we consider the higher-order Carlitzβs degen-erate Bernoulli polynomials with umbral calculus viewpoint and derive new
identities and properties of those polynomials associated with special polyno-mials which are derived from umbral calculus.
1. Introduction
The degenerate Bernoulli polynomials Ξ²n (Ξ», x) (Ξ» = 0) are defined by Carlitz tobe
(1.1)t
(1 + Ξ»t)1Ξ» + 1
(1 + Ξ»t)xΞ» =
ββn=0
Ξ²n (Ξ», x)tn
n!, (Ξ» = 0) , (see [3, 4]) .
Ustinov rediscovered these polynomials in [18], which are called Korobov poly-
nomials of the second kind and denoted by k(Ξ»)n (x).
When x = 0, Ξ²n (Ξ») = Ξ²n (Ξ», 0) are called the degenerate Bernoulli numbers.Now, we observe that
(1.2) limΞ»β0
Ξ²n (Ξ», x) = Ξ²n (0, x) = Bn (x) , limΞ»ββ
Ξ»βnΞ²n (Ξ», Ξ»x) = bn (x) ,
where Bn (x) and bn (x) are the Bernoulli polynomials of the first kind and of thesecond kind.
The first few degenerate Bernoulli polynomials are given by Ξ²0 (Ξ», x) = 1, Ξ²1 (Ξ», x)= x β 1
2 + 12Ξ», Ξ²2 (Ξ», x) = x2 β x + 1
6 β16Ξ»
2, Ξ²3 (Ξ», x) = x3 β 32x
2 + 12x β
32Ξ»x
2 +32Ξ»x + 1
4Ξ»3 β 1
4Ξ», . . . . As an analogue of the classical Staudt-Clausen theorem forBernoulli numbers, Carlitz proved the so called degenerate Staudt-Clausen theoremfor Ξ²n (Ξ»), (Ξ» a rational number) (see [3, 19, 20]). The generalized falling factorials(x|Ξ»)n for any Ξ» β C are defined as
(x|Ξ»)0 = 1, (x|Ξ»)n = x (xβ Ξ») Β· Β· Β· (xβ Ξ» (nβ 1)) , (for n > 0) .
Carlitz also found in [4] the following relation expressing sums of generalizedfalling factorals in terms of degenerate Bernoulli polynomials: for integers l,m withl β₯ 1,m β₯ 0,
(1.3)lβ1βi=0
(i|Ξ»)m =1
m+ 1(Ξ²m+1 (Ξ», l)β Ξ²m+1 (Ξ»)) ,
2000 Mathematics Subject Classification. 05A19, 05A40, 11B83.Key words and phrases. Higher-order degenerate Bernoulli polynomial, Umbral calculus.
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
789 DAE SAN KIM et al 789-811
2 DAE SAN KIM AND TAEKYUN KIM
which, by letting Ξ»β 0, becomes the familiar relation
(1.4)lβ1βi=0
im =1
m+ 1(Bm+1 (l)βBm+1) .
For r β N, the Bernoulli polynomials of the second kind of order r are definedby the generating function to be
(1.5)
(t
log (1 + t)
)r
(1 + t)x
=
ββn=0
b(r)n (x)tn
n!, (see [16]) ,
and the Bernoulli polynomials of order r are given by
(1.6)
(t
et β 1
)r
ext =
ββn=0
B(r)n (x)
tn
n!, (see [2, 5β7, 9]) .
When x = 0, B(r)n = B
(r)n (0), b
(r)n = b
(r)n (0) are called the Bernouli numbers of
the first kind of order r and of the second kind of order r. For Β΅ β C with Β΅ = 1,the Frobenius-Euler polynomials with order s β N are defined by the generatingfunction to be
(1.7)
(1β Β΅et β Β΅
)s
ext =ββ
n=0
H(s)n (x|Β΅)
tn
n!, (see [1, 10β12]) .
When x = 0, H(s)n (Β΅) = H
(s)n (0|Β΅) are called the Frobenius-Euler numbers of order
s. As is well known, the Stirling number of the second kind is defined by thegenerating function to be
(1.8)(et β 1
)n= n!
ββl=n
S2 (l,m)tl
l!, (n β Zβ₯0) , (see [16, 17]) .
For n β₯ 0, the Stirling number of the first kind is given by
(x)n = x (xβ 1) Β· Β· Β· (xβ (nβ 1)) =nβ
l=0
S1 (n, l)xl, (see [13, 15, 16, 21]) .
Let F be the set of all formal power series in the variable t:
(1.9) F =
f (t) =
ββk=0
aktk
k!
β£β£β£β£β£ ak β C
.
Let P = C [x] and let Pβbe the vector space of all linear functionals on P.β¨L| p (x)β©denotes the action of the linear functional L on p (x) which satisfies β¨L+M | p (x)β© =β¨L| p (x)β©+β¨M | p (x)β©, and β¨cL| p (x)β© = c β¨L| p (x)β©, where c is a complex constant.The linear functional β¨f (t)| Β·β© on P is defined as
(1.10) β¨f (t)|xnβ© = an, (n β₯ 0) , where f (t) β F .Thus, by (1.9) and (1.10), we get
(1.11)β¨tkβ£β£xnβ© = n!Ξ΄n,k, (n, k β₯ 0) , (see [14, 16]) ,
where Ξ΄n,k is the Kronecker symbol.
Let fL (t) =ββ
k=0β¨L|xkβ©
k! tk. Then, by (1.11), we get β¨fL (t)|xnβ© = β¨L|xnβ©. So,the map L 7β fL (t) is a vector space isomorphism from Pβ onto F . Henceforth,F denotes both the algebra of formal power series in t and the vector space of all
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
790 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 3
linear functionals on P, and so an element f (t) of F will be thought of as both aformal power series and a linear functional. We call F the umbral algebra and theumbral calculus is the study of umbral algebra. The order o (f (t)) of a power seriesf (t) = 0 is the smallest integer k for which the coefficient of tk does not vanish(see [14, 16]). If o (f (t)) = 0, then f (t) is called an invertible series; if o (f (t)) = 1,then f (t) is called a delta series. Let f (t) , g (t) be a delta series and an invertibleseries, respectively. Then there exists a unique sequence sn (x)(deg sn (x) = n)
such thatβ¨g (t) f (t)
kβ£β£β£ sn (x)
β©= n!Ξ΄n,k, for k β₯ 0. Such a sequence sn (x) is called
the Sheffer sequence for (g (t) , f (t))which is denoted by sn (x) βΌ (g (t) , f (t)) (see[14, 16]). The sequence sn (x) is Sheffer for (g (t) , f (t)) if and only if
(1.12)1
g(f (t)
)eyf(t) =ββk=0
sk (y)
k!tk, (y β C) , (see [11, 17]) ,
where f (t) is the compositional inverse of f (t) with f (f (t)) = f(f (t)
)= t.
Let f (t) , g (t) β F and p (x) β P. Then we see that
(1.13) f (t) =ββk=0
β¨f (t)|xk
β© tkk!, p (x) =
ββk=0
β¨tkβ£β£ p (x)
β© xkk!.
From (1.13), we have
(1.14) tkp (x) = p(k) (x) =dkp (x)
dxk, eytp (x) = p (x+ y) .
By (1.14), we get β¨eyt| p (x)β© = p (y) .For sn (x) βΌ (g (t) , f (t)), we have the following equations ([16]):
(1.15) f (t) sn (x) = nsnβ1 (x) , (n β₯ 1) , sn (x+ y) =nβ
j=0
(n
j
)sj (x) pnβj (y) ,
where pn (x) = g (t) sn (x),(1.16)
sn+1 (x) =
(xβ gβ² (t)
g (t)
)1
f β² (t)sn (x) , sn (x) =
nβj=0
1
j!
β¨g(f (t)
)β1f (t)
jβ£β£β£xnβ©xj ,
and
(1.17)
β¨f (t)|xp (x)β© = β¨βtf (t)| p (x)β© ,
d
dxsn (x) =
nβ1βl=0
(n
l
)β¨f (t)
β£β£xnβlβ©sl (x) , (n β₯ 1) .
In particular, for pn (x) βΌ (1, f (t)), qn (x) βΌ (1, g (t)), we note that
(1.18) qn (x) = x
(f (t)
g (t)
)n
xβ1pn (x) , (n β₯ 1) .
Let us assume that sn (x) βΌ (g (t) , f (t)), rn (x) βΌ (h (t) , l (t)). Then we have
(1.19) sn (x) =nβ
m=0
Cn,mrm (x) , (n β₯ 0) ,
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
791 DAE SAN KIM et al 789-811
4 DAE SAN KIM AND TAEKYUN KIM
where
(1.20) Cn,m =1
m!
β¨h(f (t)
)g(f (t)
) (l (f (t)))mβ£β£β£β£β£xn
β©, (see [16]) .
In this paper, we consider, for any positive integer r, the degenerate Bernoulli
polynomials Ξ²(r)n (Ξ», x) of order r which are defined by the generating function to
be
(1.21)
(t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)xΞ» =
ββn=0
Ξ²(r)n (Ξ», x)
tn
n!, (r β Zβ₯0) .
From (1.20) and (1.21), we note that
(1.22) Ξ²(r)n (Ξ», x) βΌ
((Ξ» (et β 1)
eΞ»t β 1
)r
,1
Ξ»
(eΞ»t β 1
)).
That is, Ξ²(r)n (Ξ», x) is the Sheffer polynomial for the pair(
g (t) =
(Ξ» (et β 1)
eΞ»t β 1
)r
, f (t) =1
Ξ»
(eΞ»t β 1
)).
The purpose of this paper is to give new identities and properties of the higher-order degenerate Bernoulli polynomials associated with special polynomials whichare derived from umbral calculus.
2. Higher-order degenerate Bernoulli polynomials
For n β₯ 0, we note that
xn βΌ (1, t) ,
(Ξ» (et β 1)
eΞ»t β 1
)r
Ξ²(r)n (Ξ», x) βΌ
(1,
1
Ξ»
(eΞ»t β 1
)),
From (1.18), we can derive the following equation:(Ξ» (et β 1)
eΞ»t β 1
)r
Ξ²(r)n (Ξ», x) = x
(Ξ»t
eΞ»t β 1
)n
xβ1xn = x
(Ξ»t
eΞ»t β 1
)n
xnβ1(2.1)
=nβ1βl=0
(nβ 1
l
)Ξ»lB
(n)l xnβl, (n β₯ 1) .
Thus, by (2.1), we get
Ξ²(r)n (Ξ», x)(2.2)
=nβ1βl=0
(nβ 1
l
)Ξ»lB
(n)l
(eΞ»t β 1
Ξ» (et β 1)
)r
xnβl
=nβ1βl=0
(nβ 1
l
)Ξ»lB
(n)l
(t
et β 1
)r (eΞ»t β 1
Ξ»t
)r
xnβl
=
nβ1βl=0
(nβ 1
l
)Ξ»lB
(n)l
(t
et β 1
)r(r!
ββk=0
S2 (k + r, r)Ξ»k
(k + r)!tk
)xnβl
=nβ1βl=0
(nβ 1
l
)Ξ»lB
(n)l
(t
et β 1
)r nβlβk=0
(nβlk
)(k+rr
)S2 (k + r, r)Ξ»kxnβlβk
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
792 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 5
=nβ1βl=0
nβlβk=0
(nβ1l
)(nβlk
)(k+rr
) S2 (k + r, r)Ξ»k+lB(n)l
(t
et β 1
)r
xnβlβk
=nβ1βl=0
nβlβk=0
(nβ1l
)(nβlk
)(k+rr
) S2 (k + r, r)Ξ»k+lB(n)l B
(r)nβlβk (x) .
Therefore, by (2.2), we obtain the following theorem.
Theorem 2.1. For n β₯ 1, we have
Ξ²(r)n (Ξ», x) =
nβ1βl=0
nβlβk=0
(nβ1l
)(nβlk
)(k+rr
) S2 (k + r, r)Ξ»k+lB(n)l B
(r)nβkβl (x) .
Remark. When x = 0 and r = 1, we get
(2.3) Ξ²n (Ξ») =nβ1βl=0
nβlβk=0
1
k + 1
(nβ 1
l
)(nβ lk
)Ξ»k+lB
(n)l Bnβkβl.
From (1.18), (1.22) and
(x|Ξ»)n = Ξ»n(xΞ»
)n
=nβ
m=0
S1 (n,m)Ξ»nβmxm βΌ(
1,1
Ξ»
(eΞ»t β 1
)).
We note that
Ξ²(r)n (Ξ», x) =
nβm=0
S1 (n,m)Ξ»nβm
(eΞ»t β 1
Ξ» (et β 1)
)r
xm(2.4)
=
nβm=0
S1 (n,m)Ξ»nβm
(t
et β 1
)r (eΞ»t β 1
Ξ»t
)r
xm
=
nβm=0
S1 (n,m)Ξ»nβm
(t
et β 1
)r mβk=0
(mk
)(k+rr
)S2 (k + r, r)Ξ»kxmβk
= Ξ»nnβ
m=0
mβk=0
(mk
)(k+rr
)S1 (n,m)S2 (k + r, r)Ξ»kβmB(r)mβk (x) .
Therefore, by (2.4), we obtain the following theorem.
Theorem 2.2. For n β₯ 0, we have
Ξ²(r)n (Ξ», x) = Ξ»n
nβm=0
mβk=0
(mk
)(k+rr
)S1 (n,m)S2 (k + r, r)Ξ»kβmB(r)mβk (x) .
Remark. For r = 1 and x = 0, we get an expression for the degenerate Bernoullinumbers:
(2.5) Ξ²n (Ξ») = Ξ»nnβ
m=0
mβk=0
1
k + 1
(m
k
)S1 (n,m)Ξ»kβmBmβk.
Here we use the conjugation representation.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
793 DAE SAN KIM et al 789-811
6 DAE SAN KIM AND TAEKYUN KIM
For Ξ²(r)n (Ξ», x) βΌ
(g (t) =
(Ξ»(etβ1)eΞ»tβ1
)r
, f (t) = 1Ξ»
(eΞ»t β 1
)), we observe thatβ¨
g(f (t)
)β1f (t)
jβ£β£β£xnβ©(2.6)
=
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r (1
Ξ»log (1 + Ξ»t)
)jβ£β£β£β£β£xn
β©
=Ξ»βj
β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£ j!ββl=j
S1 (l, j)Ξ»l
l!tlxn
β©
=j!Ξ»βjnβ
l=j
(n
l
)S1 (l, j)Ξ»l
β¨ ββm=0
Ξ²(r)m (Ξ»)
tm
m!
β£β£β£β£β£xnβl
β©
=j!Ξ»βjnβ
l=j
(n
l
)S1 (l, j)Ξ»lΞ²
(r)nβl (Ξ») .
Therefore, by (1.16) and (2.6), we obtain the following theorem.
Theorem 2.3. For n β₯ 0, r β₯ 1, we have
Ξ²(r)n (Ξ», x) =
nβj=0
Ξ»βj
nβl=j
(n
l
)S1 (l, j)Ξ»lΞ²
(r)nβl (Ξ»)
xj .
Remark. Recall that
(2.7)
(Ξ» (et β 1)
eΞ»t β 1
)r
Ξ²(r)n (Ξ», x) βΌ
(1,
1
Ξ»
(eΞ»t β 1
)), (x|Ξ»)n βΌ
(1,
1
Ξ»
(eΞ»t β 1
)).
Thus, by (2.7), we get
(2.8)
(Ξ» (et β 1)
eΞ»t β 1
)r
Ξ²(r)n (Ξ», x) = (x|Ξ»)n , and
eΞ»t β 1
Ξ»(x|Ξ»)n = n (x|Ξ»)nβ1 .
From (2.8), we have(et β 1
)rΞ²(r)n (Ξ», x) =
(eΞ»t β 1
Ξ»
)r
(x|Ξ»)n(2.9)
=
(n)r (x|Ξ»)nβr , if r β€ n0 , if r > n.
By (2.9), we get
trΞ²(r)n (Ξ», x) =
(n)r Ξ»
nβr(
tetβ1
)r (xΞ»
)nβr
, if r β€ n0 , if r > n
(2.10)
=
(n)r Ξ»
nβrβnβr
m=0 S1 (nβ r,m)Ξ»βmB(r)m (x) , if r β€ n
0 , if r > n.
Therefore, from (1.14) and (2.10), we have(2.11)(
d
dx
)r
Ξ²(r)n (Ξ», x) =
(n)r Ξ»
nβrβnβr
m=0 S1 (nβ r,m)Ξ»βmB(r)m (x) , if r β€ n
0 , if r > n.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
794 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 7
In particular,
(2.12)d
dxΞ²n (Ξ», x) =
nΞ»nβ1
βnβ1m=0 S1 (nβ 1,m)Ξ»βmBm (x) , if r β€ n
0 , if r > n.
To proceed further, we recall that the Ξ»-Daehee polynomials D(r)n,Ξ» (x) of order r
are given by
(2.13)
(Ξ» log (1 + t)
(1 + t)Ξ» β 1
)r
(1 + t)x
=ββ
n=0
D(r)n,Ξ» (x)
tn
n!, (see [12, 16]) .
From (1.5), (1.11) and (2.13), we have
Ξ²(r)n (Ξ», y)(2.14)
=
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)yΞ»
β£β£β£β£β£xnβ©
=nβ
l=0
(n
l
)Ξ»lb
(r)l
( yΞ»
)β¨ log (1 + Ξ»t)
Ξ»(
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£β£xnβl
β©
=
nβl=0
(n
l
)Ξ»lb
(r)l
( yΞ»
)β¨ ββm=0
D(r)
m, 1λλm
tm
m!
β£β£β£β£β£xnβl
β©
=nβ
l=0
(n
l
)Ξ»lb
(r)l
( yΞ»
)D
(r)
nβl, 1λλnβl
=Ξ»nnβ
l=0
(n
l
)D
(r)
nβl, 1Ξ»b(r)l
( yΞ»
),
and
Ξ²(r)n (Ξ», y)(2.15)
=
β¨ ββl=0
Ξ²(r)l (Ξ», y)
tl
l!
β£β£β£β£β£xnβ©
=
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)yΞ»
β£β£β£β£β£xnβ©
=
β¨(Ξ»t
log (1 + Ξ»t)
)rβ£β£β£β£ log (1 + Ξ»t)
Ξ»(
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)yΞ» xn
β©
=
β¨(Ξ»t
log (1 + Ξ»t)
)rβ£β£β£β£ ββl=0
D(r)l,Ξ»β1
( yΞ»
)Ξ»ltl
l!xn
β©
=
nβl=0
(n
l
)Ξ»lD
(r)l,Ξ»β1
( yΞ»
)β¨ ββm=0
b(r)m Ξ»mtm
m!
β£β£β£β£β£xnβl
β©
=nβ
l=0
(n
l
)Ξ»lD
(r)l,Ξ»β1
( yΞ»
)b(r)nβlΞ»
nβl
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
795 DAE SAN KIM et al 789-811
8 DAE SAN KIM AND TAEKYUN KIM
=Ξ»nnβ
l=0
(n
l
)b(r)nβlD
,(r)l,Ξ»β1
( yΞ»
).
Therefore, by (2.14) and (2.15), we obtain the following theorem.
Theorem 2.4. For n β₯ 0, we havenβ
l=0
(n
l
)D
(r)
nβl, 1Ξ»b(r)l
(xΞ»
)=
nβl=0
(n
l
)b(r)nβlD
(r)l,Ξ»β1
(xΞ»
)= Ξ»βnΞ²(r)
n (Ξ», x) .
Recalling that
Ξ²(r)n (Ξ», x) βΌ
(g (t) =
(Ξ» (et β 1)
eΞ»t β 1
)r
, f (t) =1
Ξ»
(eΞ»t β 1
)),
we observe that
β¨g(f (t)
)β1f (t)
jβ£β£β£xnβ©
(2.16)
=j!Ξ»βjnβ
l=j
S1 (l, j)
(n
l
)Ξ»l
β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£xnβl
β©
=j!Ξ»βjnβ
l=j
S1 (l, j)
(n
l
)Ξ»l
β¨ log (1 + Ξ»t)
Ξ»(
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£β£
(Ξ»t
log (1 + Ξ»t)
)r
xnβl
β©
=j!Ξ»βjnβ
l=j
S1 (l, j)
(n
l
)Ξ»l
β¨ log (1 + Ξ»t)
Ξ»(
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£β£
ββm=0
b(r)m
Ξ»m
m!tmxnβl
β©
=j!Ξ»βjnβ
l=j
S1 (l, j)
(n
l
)Ξ»l
nβlβm=0
(nβ lm
)Ξ»mb(r)m
β¨ log (1 + Ξ»t)
Ξ»(
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£β£xnβlβm
β©
=j!Ξ»βjnβ
l=j
S1 (l, j)
(n
l
)Ξ»l
nβlβm=0
(nβ lm
)Ξ»mb(r)m D
(r)nβlβm,Ξ»β1Ξ»
nβlβm
=j!Ξ»nnβ
l=j
nβlβm=0
(n
l
)(nβ lm
)S1 (l, j)Ξ»βjb(r)m D
(r)nβlβm,Ξ»β1 .
From (1.16) and (2.16), we have(2.17)
Ξ²(r)n (Ξ», x) = Ξ»n
nβj=0
nβ
l=j
nβlβm=0
(n
l
)(nβ lm
)S1 (l, j)Ξ»βjb(r)m D
(r)nβlβm,Ξ»β1
xj .
Remark. We have
limΞ»β0
Ξ²(r)n (Ξ», x) = Ξ²(r)
n (0, x) = B(r)n (x) ,
limΞ»β0
D(r)n,Ξ» (x) = (x)n ,
limΞ»ββ
Ξ»βnΞ²(r)n (Ξ», Ξ»x) = b(r)n (x) ,
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
796 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 9
limΞ»ββ
Ξ»βnD(r)n,Ξ» (Ξ»x) = B(r)
n (x) .
where r > 0.
From (1.22), we note that
(2.18) pn (x) =
(Ξ» (et β 1)
eΞ»t β 1
)r
Ξ²(r)n (Ξ», x) = (x|Ξ»)n βΌ
(1,
1
Ξ»
(eΞ»t β 1
)).
By (2.18) and (1.15), we get
(2.19) Ξ²(r)n (Ξ», x+ y) =
nβj=0
(n
j
)Ξ²(r)j (Ξ», x) (y|Ξ»)nβj ,
and, by (1.14) and (1.15), we get
(2.20)1
Ξ»
(eΞ»t β 1
)Ξ²(r)n (Ξ», x) = nΞ²
(r)nβ1 (Ξ», x) .
From (2.20), we have
(2.21) Ξ²(r)n (Ξ», x+ Ξ»)β Ξ²(r)
n (Ξ», x) = nλβ(r)nβ1 (Ξ», x) .
Therefore, by (2.17), (2.19) and (2.20), we obtain the following theorem.
Theorem 2.5. For n β₯ 0, we have
Ξ²(r)n (Ξ», x) = Ξ»n
nβj=0
nβ
l=j
nβlβm=0
(n
l
)(nβ lm
)S1 (l, j)Ξ»βjb(r)m D
(r)nβlβm,Ξ»β1
xj ,
and
Ξ²(r)n (Ξ», x+ Ξ») =
nβj=0
(n
j
)Ξ²(r)j (Ξ», x) (Ξ»|Ξ»)nβj
= nλβ(r)nβ1 (Ξ», x) + Ξ²(r)
n (Ξ», x) .
For Ξ²(r)n (Ξ», x) βΌ
(g (t) =
(Ξ»(etβ1)eΞ»tβ1
)r
, f (t) = 1Ξ»
(eΞ»t β 1
)), we note that
gβ² (t)
g (t)(2.22)
= (log g (t))β²
=r(log Ξ»+ log
(et β 1
)β log
(eΞ»t β 1
))β²=r
t
( ββl=0
Bl (1)tl
l!β
ββl=0
Bl (1)Ξ»ltl
l!
)
=r
t
ββl=1
Bl (1)(1β Ξ»l
) tll!
=r
ββl=0
Bl+1 (1)(1β Ξ»l+1
) tl
(l + 1)!.
By (2.22), we get
gβ² (t)
g (t)Ξ²(r)n (Ξ», x)(2.23)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
797 DAE SAN KIM et al 789-811
10 DAE SAN KIM AND TAEKYUN KIM
=rββl=0
Bl+1 (1)(1β Ξ»l+1
) tl
(l + 1)!Ξ»n
Γnβ
m=0
mβk=0
(mk
)(k+rr
)S1 (n,m)S2 (k + r, r)Ξ»kβmB(r)mβk (x)
=Ξ»nr
nβm=0
mβk=0
mβkβl=0
S1 (n,m)S2 (k + r, r)Ξ»kβm
Γmβkβl=0
Bl+1 (1)(1β Ξ»l+1
) 1
(l + 1)!(mβ k)lB
(r)mβkβl (x)
=Ξ»nrnβ
m=0
mβk=0
mβkβl=0
1
mβ k β l + 1
(mk
)(mβk
l
)(k+rr
) (Ξ»kβm β Ξ»1βl
)Γ S1 (n,m)S2 (k + r, r)Bmβkβl+1 (1)B
(r)l (x)
=rnβ
l=0
nβm=l
mβk=l
1
k β l + 1
(mk
)(kl
)(mβk+r
r
) (Ξ»nβk β Ξ»nβl+1)S1 (n,m)
Γ S2 (mβ k + r, r)Bkβl+1 (1)B(r)l (x) .
From (1.16) and (2.23), we have
Ξ²(r)n+1 (Ξ», x)(2.24)
=xΞ²(r)n (Ξ», xβ Ξ»)β eβΞ»t g
β² (t)
g (t)Ξ²(r)n (Ξ», x)
=xΞ²(r)n (Ξ», xβ Ξ»)β r
nβl=0
(nβ
m=l
mβk=l
1
k β l + 1
(mk
)(kl
)(mβk+r
r
) (Ξ»nβk β Ξ»nβl+1)
Γ S1 (n,m)S2 (mβ k + r, r)Bkβl+1 (1))B(r)l (xβ Ξ») .
Therefore, by (2.24), we obtain the following theorem.
Theorem 2.6. For n β₯ 0, we have
Ξ²(r)n+1 (Ξ», x)
=xΞ²(r)n (Ξ», xβ Ξ»)β r
nβl=0
(nβ
m=l
mβk=l
1
k β l + 1
(mk
)(kl
)(mβk+r
r
) (Ξ»nβk β Ξ»nβl+1)
ΓS1 (n,m)S2 (mβ k + r, r)Bkβl+1 (1))B(r)l (xβ Ξ») .
By Ξ²(r)n (Ξ», x) βΌ
(g (t) =
(Ξ»(etβ1)eΞ»tβ1
)r
, f (t) = 1Ξ»
(eΞ»t β 1
)), we getβ¨
f (t)β£β£xnβl
β©(2.25)
=
β¨1
Ξ»log (1 + Ξ»t)
β£β£β£β£xnβl
β©=Ξ»β1
β¨ ββm=1
(β1)mβ1
Ξ»m (mβ 1)!tm
m!
β£β£β£β£β£xnβl
β©
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
798 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 11
=Ξ»β1 (β1)nβlβ1
Ξ»nβl (nβ l β 1)!
= (βΞ»)nβlβ1
(nβ l β 1)!.
From (1.17) and (2.25), we have
(2.26)d
dxΞ²(r)n (Ξ», x) = n!
nβ1βl=0
(βΞ»)nβlβ1
l! (nβ l)Ξ²(r)l (Ξ», x) .
Let n β₯ 1. Then, by (1.11) and (1.17), we get
Ξ²(r)n (Ξ», y)(2.27)
=
β¨ ββl=0
Ξ²(r)l (Ξ», y)
tl
l!
β£β£β£β£β£xnβ©
=
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)yΞ»
β£β£β£β£β£xnβ©
=
β¨βt
((t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)yΞ»
)β£β£β£β£β£xnβ1
β©
=
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
βt (1 + Ξ»t)yΞ»
β£β£β£β£β£xnβ1
β©
+
β¨(βt
(t
(1 + Ξ»t)1Ξ» β 1
)r)(1 + Ξ»t)
yΞ»
β£β£β£β£β£xnβ1
β©.
The first term of (2.27) is
(2.28) y
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)yβλλ
β£β£β£β£β£xnβ1
β©= yΞ²
(r)nβ1 (Ξ», y β Ξ») .
For the second term of (2.27), we observe that
(2.29) βt
(t
(1 + Ξ»t)1Ξ» β 1
)r
= r
(t
(1 + Ξ»t)1Ξ» β 1
)rβ1
βt
(t
(1 + Ξ»t)1Ξ» β 1
),
where
βt
(t
(1 + Ξ»t)1Ξ» β 1
)(2.30)
=(1 + Ξ»t)
1Ξ» β 1β t (1 + Ξ»t)
1Ξ»β1(
(1 + Ξ»t)1Ξ» β 1
)2=
(1 + Ξ»t)1Ξ» β 1β t
(1 + Ξ»t)
1Ξ»β1 β (1 + Ξ»t)
β1β t (1 + Ξ»t)
β1((1 + Ξ»t)
1Ξ» β 1
)2=β 1
1 + Ξ»t
t
(1 + Ξ»t)1Ξ» β 1
+1
t(2.31)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
799 DAE SAN KIM et al 789-811
12 DAE SAN KIM AND TAEKYUN KIM
Γ
t
(1 + Ξ»t)1Ξ» β 1
β 1
(1 + Ξ»t)
(t
(1 + Ξ»t)1Ξ» β 1
)2 .
Thus, by (2.29) and (2.30), we get
βt
(t
(1 + Ξ»t)1Ξ» β 1
)r
(2.32)
=β r
(1 + Ξ»t)
(t
(1 + Ξ»t)1Ξ» β 1
)r
+r
t
(
t
(1 + Ξ»t)1Ξ» β 1
)r
β 1
1 + Ξ»t
(t
(1 + Ξ»t)1Ξ» β 1
)r+1 .
From (2.32), we note that the second term of (2.27) is
β r
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)yβλλ
β£β£β£β£β£xnβ1
β©(2.33)
+ r
β¨(1 + Ξ»t)
yΞ»
β£β£β£β£β£β£1t(
t
(1 + Ξ»t)1Ξ» β 1
)r
β 1
1 + Ξ»t
(t
(1 + Ξ»t)1Ξ» β 1
)r+1xnβ1
β©
=β rΞ²(r)nβ1 (Ξ», y β Ξ»)
+r
n
β¨(1 + Ξ»t)
yΞ»
β£β£β£β£β£β£(
t
(1 + Ξ»t)1Ξ» β 1
)r
β 1
1 + Ξ»t
(t
(1 + Ξ»t)1Ξ» β 1
)r+1xn
β©
=β rΞ²(r)nβ1 (Ξ», y β Ξ»)
+r
n
β¨(t
(1 + t)1Ξ» β 1
)r
(1 + Ξ»t)yΞ»
β£β£β£β£β£xnβ©
β r
n
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r+1
(1 + Ξ»t)yβλλ
β£β£β£β£β£β£xnβ©
=β rΞ²(r)nβ1 (Ξ», y β Ξ») +
r
nΞ²(r)n (Ξ», y)β r
nΞ²(r+1)n (Ξ», y β Ξ») .
By (2.27), (2.28) and (2.33), we get
(2.34)(
1β r
n
)Ξ²(r)n (Ξ», x) = (xβ r)Ξ²(r)
nβ1 (Ξ», xβ Ξ»)β r
nΞ²(r+1)n (Ξ», xβ Ξ») .
Therefore, by (2.34), we obtain the following theorem.
Theorem 2.7. For n β₯ 1, we have
Ξ²(r+1)n (Ξ», xβ Ξ») =
(1β n
r
)Ξ²(r)n (Ξ», x) +
(nrxβ n
)Ξ²(r)nβ1 (Ξ», xβ Ξ») .
Here we compute
(2.35)
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r (1
Ξ»log (1 + Ξ»t)
)mβ£β£β£β£β£xn
β©
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
800 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 13
in two different ways.On one hand, it is equal to
Ξ»βm
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
(log (1 + Ξ»t))m
β£β£β£β£β£xnβ©
(2.36)
=Ξ»βm
β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£m!ββ
l=m
S1 (l,m)Ξ»l
l!tlxn
β©
=m!Ξ»βmnβ
l=m
(n
l
)S1 (l,m)Ξ»l
β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£xnβl
β©
=m!Ξ»βmnβ
l=m
(n
l
)S1 (l,m)Ξ»l
β¨ ββk=0
Ξ²(r)k (Ξ»)
tk
k!
β£β£β£β£β£xnβl
β©
=m!Ξ»βmnβ
l=m
(n
l
)S1 (l,m)Ξ»lΞ²
(r)nβl (Ξ») .
On the other hand, it is equal toβ¨βt
((t
(1 + Ξ»t)1Ξ» β 1
)r (1
Ξ»log (1 + Ξ»t)
)m)β£β£β£β£β£xnβ1
β©(2.37)
=
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
βt
(1
Ξ»log (1 + Ξ»t)
)mβ£β£β£β£β£xnβ1
β©
+
β¨(βt
(t
(1 + Ξ»t)1Ξ» β 1
)r)(1
Ξ»log (1 + Ξ»t)
)mβ£β£β£β£β£xnβ1
β©.
The first term of (2.37) is
m
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r (1
Ξ»log (1 + Ξ»t)
)mβ1
(1 + Ξ»t)β1
β£β£β£β£β£xnβ1
β©(2.38)
=mΞ»β(mβ1)
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)β1
β£β£β£β£β£ (log (1 + Ξ»t))mβ1
xnβ1
β©
=mΞ»β(mβ1)
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)β1
β£β£β£β£β£ (mβ 1)!ββ
l=mβ1
S1 (l,mβ 1)Ξ»ltl
l!xnβ1
β©
=m!Ξ»β(mβ1)nβ1β
l=mβ1
(nβ 1
l
)S1 (l,mβ 1)Ξ»l
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)βΞ»
Ξ»
β£β£β£β£β£xnβ1βl
β©
=m!Ξ»β(mβ1)nβ1β
l=mβ1
(nβ 1
l
)S1 (l,mβ 1)Ξ»lΞ²
(r)nβ1βl (Ξ»,βΞ») .
For the second term of (2.37), we recall that
βt
(t
(1 + Ξ»t)1Ξ» β 1
)r
(2.39)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
801 DAE SAN KIM et al 789-811
14 DAE SAN KIM AND TAEKYUN KIM
=β r
(1 + Ξ»t)
(t
(1 + Ξ»t)1Ξ» β 1
)r
+r
t
(
t
(1 + Ξ»t)1Ξ» β 1
)r
β 1
1 + Ξ»t
(t
(1 + Ξ»t)1Ξ» β 1
)r+1 .
Now, the second term of (2.37) is
Ξ»βm
β¨βt
(t
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£ (log (1 + Ξ»t))mxnβ1
β©(2.40)
=m!Ξ»βmnβ1βl=m
(nβ 1
l
)S1 (l,m)Ξ»l
β¨βt
(t
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£xnβ1βl
β©
=m!Ξ»βmnβ1βl=m
(nβ 1
l
)S1 (l,m)Ξ»l
βr
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
(1 + Ξ»t)βλλ
β£β£β£β£β£xnβlβ1
β©
+r
nβ l
β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£xnβl
β©
β r
nβ l
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r+1
(1 + Ξ»t)βΞ»
Ξ»
β£β£β£β£β£β£xnβl
β©=m!Ξ»βm
nβ1βl=m
(nβ 1
l
)S1 (l,m)Ξ»l
ΓβrΞ²(r)
nβ1βl (Ξ»,βΞ») +r
nβ lΞ²(r)nβl (Ξ»)β r
nβ lΞ²(r+1)nβl (Ξ»,βΞ»)
.
From (2.35), (2.36), (2.37), and (2.40), we have
m!Ξ»βmnβ
l=m
(n
l
)S1 (l,m)Ξ»lΞ²
(r)nβl (Ξ»)
=m!Ξ»β(mβ1)nβ1β
l=mβ1
(nβ 1
l
)S1 (l,mβ 1)Ξ»lΞ²
(r)nβlβ1 (Ξ»,βΞ»)
+m!Ξ»βmnβ1βl=m
(nβ 1
l
)S1 (l,m)Ξ»l
(βrΞ²(r)
nβ1βl (Ξ»,βΞ») +r
nβ lΞ²(r)nβl (Ξ»)
β r
nβ lΞ²(r+1)nβl (Ξ»,βΞ»)
),
where nβ 1 β₯ m β₯ 1.After simplification and modification, we get: for nβ 1 β₯ m β₯ 1,
nβmβl=0
(n
l
)S1 (nβ l,m)Ξ»nβlΞ²
(r)l (Ξ»)(2.41)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
802 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 15
=Ξ»
nβmβl=0
(nβ 1
l
)S1 (nβ l β 1,mβ 1)Ξ»nβlβ1Ξ²
(r)l (Ξ»,βΞ»)
+nβmβ1β
l=0
(nβ 1
l
)S1 (nβ l β 1,m)Ξ»nβlβ1
Γ(βrΞ²(r)
l (Ξ»,βΞ») +r
l + 1Ξ²(r)l+1 (Ξ»)β r
l + 1Ξ²(r+1)l+1 (Ξ»,βΞ»)
)=Ξ»
nβmβl=0
(nβ 1
l
)S1 (nβ l β 1,mβ 1)Ξ»nβlβ1Ξ²
(r)l (Ξ»,βΞ»)
β rnβmβ1β
l=0
(nβ 1
l
)S1 (nβ l β 1,m)Ξ»nβlβ1Ξ²
(r)l (Ξ»,βΞ»)
+r
n
nβmβ1βl=0
(n
l + 1
)S1 (nβ l β 1,m)Ξ»nβlβ1Ξ²
(r)l+1 (Ξ»)
β r
n
nβmβ1βl=0
(n
l + 1
)S1 (nβ l β 1,m)Ξ»nβlβ1Ξ²
(r+1)l+1 (Ξ»,βΞ»)
=Ξ»
nβmβl=0
(nβ 1
l
)S1 (nβ l β 1,mβ 1)Ξ»nβlβ1Ξ²
(r)l (Ξ»,βΞ»)
β rnβmβ1β
l=0
(nβ 1
l
)S1 (nβ l β 1,m)Ξ»nβlβ1Ξ²
(r)l (Ξ»,βΞ»)
+r
n
nβmβl=0
(n
l
)S1 (nβ l,m)Ξ»nβlΞ²
(r)l (Ξ»)
β r
n
nβmβl=0
(n
l
)S1 (nβ l,m)Ξ»nβlΞ²
(r+1)l (Ξ»,βΞ») .
Therefore, by (2.41), we obtain the following theorem.
Theorem 2.8. For nβ 1 β₯ m β₯ 1, we have
(1β r
n
) nβmβl=0
(n
l
)S1 (nβ l,m)Ξ»nβlΞ²
(r)l (Ξ»)
=Ξ»nβmβl=0
(nβ 1
l
)S1 (nβ l β 1,mβ 1)Ξ»nβlβ1Ξ²
(r)l (Ξ»,βΞ»)
β rnβmβ1β
l=0
(nβ 1
l
)S1 (nβ l β 1,m)Ξ»nβlβ1Ξ²
(r)l (Ξ»,βΞ»)
β r
n
nβmβl=0
(n
l
)S1 (nβ l,m)Ξ»nβlΞ²
(r+1)l (Ξ»,βΞ») .
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
803 DAE SAN KIM et al 789-811
16 DAE SAN KIM AND TAEKYUN KIM
For r > s β₯ 1, by (1.19), (1.20) and (1.22), we get
(2.42) Ξ²(r)n (Ξ», x) =
nβm=0
Cn,mΞ²(s)m (Ξ», x) ,
where
Cn,m =1
m!
β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβs
tm
β£β£β£β£β£β£xnβ©
(2.43)
=1
m!
β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβsβ£β£β£β£β£β£ tmxn
β©
=
(n
m
)β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβsβ£β£β£β£β£β£xnβm
β©
=
(n
m
)β¨ ββl=0
Ξ²(rβs)l (Ξ»)
tl
l!
β£β£β£β£β£xnβm
β©
=
(n
m
)Ξ²(rβs)nβm (Ξ») .
Therefore, by (2.42)and (2.43), we obtain the following theorem.
Theorem 2.9. For r > s β₯ 1, we have
Ξ²(r)n (Ξ», x) =
nβm=0
(n
m
)Ξ²(rβs)nβm (Ξ»)Ξ²(s)
m (Ξ», x) .
Remark. Replacing x by x+ Ξ» in Theorem 2.7, we have
(2.44) Ξ²(r+1)n (Ξ», x) =
(1β n
r
)Ξ²(r)n (Ξ», x+ Ξ») +
(nrx+
n
rΞ»β n
)Ξ²(r)nβ1 (Ξ», x) .
From Theorem 2.5, we note that
Ξ²(r)n (Ξ», x+ Ξ») =
nβj=0
(n
j
)Ξ²(r)j (Ξ», x) (Ξ»|Ξ»)nβj(2.45)
= nλβ(r)nβ1 (Ξ», x) + Ξ²(r)
n (Ξ», x) .
Substituting (2.45) into (2.44), we get(2.46)
Ξ²(r+1)n (Ξ», x) =
(1β n
r
)Ξ²(r)n (Ξ», x) +
n
r(x+ (Ξ»β 1) r β (nβ 1)Ξ»)Ξ²
(r)nβ1 (Ξ», x) .
By using this and induction on r, it is shown in [[19]] that
(2.47) Ξ²(r)n (Ξ», x) = r
(n
r
) rβ1βk=0
(β1)rβ1βk
Οrβ1,k (Ξ», x, n)Ξ²nβk (Ξ», x)
nβ k,
where
Οr,k (Ξ», x, n) =β
1β€ik<ikβ1<Β·Β·Β·<i1β€r
kβj=1
(x+ (Ξ»β 1) ij β (nβ j)Ξ») .
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
804 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 17
For r > s β₯ 1, by (1.19), (1.20) and (1.22), we have
(2.48) Ξ²(s)n (Ξ», x) =
nβm=0
Cn,mΞ²(r)m (Ξ», x) ,
where
Cn,m =1
m!
β¨((1 + Ξ»t)
1Ξ» β 1
t
)rβs
tm
β£β£β£β£β£β£xnβ©
(2.49)
=
(n
m
)β¨((1 + Ξ»t)
1Ξ» β 1
t
)rβsβ£β£β£β£β£β£xnβm
β©.
Observe here that((1 + Ξ»t)
1Ξ» β 1
t
)rβs
(2.50)
=
(e
1Ξ» log(1+Ξ»t) β 1
t
)rβs
=1
trβs
(e
1Ξ» log(1+Ξ»t) β 1
)rβs
=1
trβs(r β s)!
ββl=rβs
S2 (l, r β s) (log (1 + Ξ»t))l
Ξ»ll!
= (r β s)!ββ
l=rβs
S2 (l, r β s)l!
(log (1 + Ξ»t)
Ξ»t
)l
tlβ(rβs)
= (r β s)!ββl=0
S2 (l + r β s, r β s)(l + r β s)!
tl(
log (1 + Ξ»t)
Ξ»t
)l+rβs
= (r β s)!ββl=0
S2 (l + r β s, r β s)(l + r β s)!
tl (l + r β s)!
Γββk=0
S1 (k + l + r β s, l + r β s) (Ξ»t)k
(k + l + r β s)!
= (r β s)!ββk=0
ββl=0
S1 (k + l + r β s, l + r β s)
Γ S2 (l + r β s, r β s) Ξ»k
(k + l + r β s)!tk+l
= (r β s)!ββj=0
βk+l=j
S1 (k + l + r β s, l + r β s)
Γ S2 (l + r β s, r β s) Ξ»k
(k + l + r β s)!tk+l
= (r β s)!ββj=0
jβk=0
S1 (j + r β s, j β k + r β s)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
805 DAE SAN KIM et al 789-811
18 DAE SAN KIM AND TAEKYUN KIM
Γ S2 (j β k + r β s, r β s) Ξ»k
(j + r β s)!tj .
From (2.49) and (2.50), we have
Cn,m
(2.51)
=
(n
m
)(r β s)!
Γ
β¨ ββj=0
(jβ
k=0
S1 (j + r β s, j β k + r β s)S2 (j β k + r β s, r β s) Ξ»kj!
(j + r β s)!
)tj
j!
β£β£β£β£β£β£xnβm
β©
=
(n
m
)(r β s)!
nβmβk=0
S1 (nβm+ r β s, nβmβ k + r β s)
Γ S2 (nβmβ k + r β s, r β s) Ξ»k (nβm)!
(nβm+ r β s)!
=
(nm
)(nβm+rβs
rβs
) nβmβk=0
S1 (nβm+ r β s, nβmβ k + r β s)S2 (nβmβ k + r β s, r β s)Ξ»k.
Therefore, by (2.48) and (2.51), we obtain the following theorem.
Theorem 2.10. For r > s β₯ 1, we have
Ξ²(s)n (Ξ», x)
=nβ
m=0
(nm
)(nβm+rβs
rβs
) nβmβk=0
S1 (nβm+ r β s, nβmβ k + r β s)
ΓS2 (nβmβ k + r β s, r β s)Ξ»kΞ²(r)m (Ξ», x)
=nβ
m=0
(nm
)(nβm+rβs
rβs
) nβmβk=0
S1 (nβm+ r β s, k + r β s)
ΓS2 (k + r β s, r β s)Ξ»nβmβkΞ²(r)m (Ξ», x) .
For (x|Ξ»)n βΌ(1, 1
Ξ»
(eΞ»t β 1
)), by (1.19), (1.20) and (1.22), we get
(2.52) Ξ²(r)n (Ξ», x) =
nβm=0
Cn,m (x|Ξ»)m ,
where
Cn,m =1
m!
β¨(t
(1 + Ξ»t)1Ξ» β 1
)r
tm
β£β£β£β£β£xnβ©
(2.53)
=1
m!
β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£ tmxnβ©
=
(n
m
)β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£xnβm
β©
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
806 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 19
=
(n
m
)β¨ ββk=0
Ξ²(r)k (Ξ»)
tk
k!
β£β£β£β£β£xnβm
β©
=
(n
m
)Ξ²(r)nβm (Ξ») .
Therefore, by (2.52) and (2.53), we obtain the following theorem.
Theorem 2.11. For n β₯ 0, we have
Ξ²(r)n (Ξ», x) =
nβm=0
(n
m
)Ξ²(r)nβm (Ξ») (x|Ξ»)m .
Remark. For n β₯ 0, we get(2.54)
(x|Ξ»)n =
nβm=0
(nm
)(nβm+r
r
) nβmβk=0
S1 (nβm+ r, k + r)S2 (k + r, r)Ξ»nβmβk
Ξ²(r)m (Ξ», x) .
By (1.6) and (1.12), we easily get
(2.55) B(s)n (x) βΌ
((et β 1
t
)s
, t
), (s β N) .
From (1.19), (1.20), (1.22) and (2.55), we have
(2.56) B(s)n (x) =
nβm=0
Cn,mΞ²(r)m (Ξ», x) ,
where
Cn,m =1
m!
β¨ (Ξ»(etβ1)eΞ»tβ1
)r
(etβ1
t
)s (1
Ξ»
(eΞ»t β 1
))m
β£β£β£β£β£β£β£β£xn
β©(2.57)
=1
m!Ξ»m
β¨(Ξ»t
eΞ»t β 1
)r (et β 1
t
)r (t
et β 1
)s (eΞ»t β 1
)mβ£β£β£β£xnβ©=
1
m!Ξ»m
β¨(Ξ»t
eΞ»t β 1
)r (et β 1
t
)r (t
et β 1
)sβ£β£β£β£m!
ββl=m
S2 (l,m)Ξ»l
l!tlxn
β©
= Ξ»βmnβ
l=m
(n
l
)S2 (l,m)Ξ»l
β¨(Ξ»t
eΞ»t β 1
)r (et β 1
t
)r (t
et β 1
)sβ£β£β£β£xnβl
β©
= Ξ»βmnβ
l=m
(n
l
)S2 (l,m)Ξ»l
β¨(et β 1
t
)r (t
et β 1
)sβ£β£β£β£ ββk=0
B(r)k
Ξ»k
k!tkxnβl
β©
= Ξ»βmnβ
l=m
(n
l
)S2 (l,m)Ξ»l
nβlβk=0
(nβ lk
)B
(r)k Ξ»k
β¨(et β 1
t
)r (t
et β 1
)sβ£β£β£β£xnβlβk
β©.
Case 1. For r > s β₯ 1, we haveβ¨(et β 1
t
)r (t
et β 1
)sβ£β£β£β£xnβlβk
β©(2.58)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
807 DAE SAN KIM et al 789-811
20 DAE SAN KIM AND TAEKYUN KIM
=
β¨(et β 1
t
)rβsβ£β£β£β£β£xnβlβk
β©
=
β¨(r β s)!
ββj=0
S2 (j + r β s, r β s) j!(j + r β s)!
tj
j!
β£β£β£β£β£β£xnβlβk
β©
= (r β s)!S2 (nβ l β k + r β s, r β s) (nβ l β k)!
(nβ l β k + r β s)!
=S2 (nβ l β k + r β s, r β s)/(nβ l β k + r β s
r β s
)Case 2. For r = s β₯ 1, we get
(2.59)
β¨(et β 1
t
)r (t
et β 1
)sβ£β£β£β£xnβlβk
β©=β¨
1|xnβlβkβ©
= Ξ΄0,nβlβk = Ξ΄k,nβl.
Case 3. For s > r β₯ 1, we have
(2.60)
β¨(et β 1
t
)r (t
et β 1
)sβ£β£β£β£xnβlβk
β©=
β¨(t
et β 1
)sβrβ£β£β£β£β£xnβlβk
β©= B
(sβr)nβlβk.
Therefore, by (2.56), (2.57), (2.58), (2.59) and (2.60), we obtain the following the-orem.
Theorem 2.12. Let n β₯ 0. Then we have
B(s)n (x) =
nβm=0
Ξ»βm
nβl=m
nβlβk=0
(nl
)(nβlk
)(nβlβk+rβs
rβs
)S2 (l,m)
Γ S2 (nβ l β k + r β s, r β s)Ξ»k+lB(r)k
Ξ²(r)m (Ξ», x) , if r > s β₯ 1,
Ξ»nnβ
m=0
Ξ»βm
nβl=m
(n
l
)S2 (l,m)B
(r)nβl
Ξ²(r)m (Ξ», x) , if r = s β₯ 1,
nβm=0
Ξ»βm
nβ
l=m
nβlβk=0
(n
l
)(nβ lk
)S2 (l,m)
ΓΞ»k+lB(r)k B
(sβr)nβlβk
Ξ²(r)m (Ξ», x) , if s > r β₯ 1.
Remark. Let r > s β₯ 1. Then we get(2.61)
Ξ²(r)n (Ξ», x) =
nβm=0
Ξ»βm
nβl=m
nβlβk=0
(n
l
)(nβ lk
)S1 (l,m)Ξ»k+lb
(s)k Ξ²
(rβs)nβlβk (Ξ»)
B(s)
m (x) .
For r = s β₯ 1, we have
(2.62) Ξ²(r)n (Ξ», x) = Ξ»n
nβm=0
Ξ»βm
nβ
l=m
(n
l
)S1 (l,m) b
(s)nβl
B(s)
m (x) .
If s > r β₯ 1, then we note that
Ξ²(r)n (Ξ», x)(2.63)
=Ξ»nnβ
m=0
Ξ»βm
nβl=m
nβlβk=0
nβlβkβi=0
(nl
)(nβlk
)(nβlβk+sβr
sβr
)S1 (l,m)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
808 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 21
Γ S1 (nβ l β k + sβ r, i+ sβ r)S2 (i+ sβ r, sβ r)Ξ»βib(s)k
B(s)
m (x) .
From (1.7) and (1.12), we get
(2.64) H(s)n (x|Β΅) βΌ
((et β Β΅1β Β΅
)s
, t
).
By (1.19), (1.20), (1.22) and (2.64), we have
(2.65) Ξ²(r)n (Ξ», x) =
nβm=0
Cn,mH(s)m (x|Β΅) ,
where
Cn,m
(2.66)
=1
m!
β¨ (e
1Ξ» log(1+Ξ»t) β Β΅
/1β Β΅
)s(Ξ»(e
1Ξ» log(1+Ξ»t) β 1
)/elog(1+Ξ»t) β 1
)r ( 1
Ξ»log (1 + Ξ»t)
)mβ£β£β£β£β£β£xn
β©
=1
m!Ξ»m (1β Β΅)s
β¨((1 + Ξ»t)
1Ξ» β Β΅
)s( t
(1 + Ξ»t)1Ξ» β 1
)r
(log (1 + Ξ»t))m
β£β£β£β£β£xnβ©
=1
Ξ»m (1β Β΅)s
nβl=m
(n
l
)S1 (l,m)Ξ»l
β¨((1 + Ξ»t)
1Ξ» β Β΅
)sβ£β£β£( t
(1 + Ξ»t)1Ξ» β 1
)r
xnβl
β©
=1
Ξ»m (1β Β΅)s
nβl=m
(n
l
)S1 (l,m)Ξ»l
nβlβk=0
(nβ lk
)Ξ²(r)k (Ξ»)
β¨((1 + Ξ»t)
1Ξ» β Β΅
)sβ£β£β£xnβlβkβ©.
It is easy to show thatβ¨((1 + Ξ»t)
1Ξ» β Β΅
)sβ£β£β£xnβlβkβ©
(2.67)
=sβ
i=0
(s
i
)(βΒ΅)
sβi
β¨ ββj=0
(i|Ξ»)jtj
j!
β£β£β£β£β£β£xnβlβk
β©
=sβ
i=0
(s
i
)(βΒ΅)
sβi(i|Ξ»)nβlβk .
From (2.66) and (2.67), we have
Cn,m
(2.68)
=1
Ξ»m (1β Β΅)s
nβl=m
(n
l
)S1 (l,m)Ξ»l
nβlβk=0
(nβ lk
)Ξ²(r)k (Ξ»)
sβi=0
(s
i
)(βΒ΅)
sβi(i|Ξ»)nβlβk
=1
Ξ»m (1β Β΅)s
nβl=m
nβlβk=0
sβi=0
(n
l
)(nβ lk
)(s
i
)S1 (l,m)Ξ»l (βΒ΅)
sβiΞ²(r)k (Ξ») (i|Ξ»)nβlβk .
Therefore, by (2.65) and (2.68), we obtain the following theorem.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
809 DAE SAN KIM et al 789-811
22 DAE SAN KIM AND TAEKYUN KIM
Theorem 2.13. For Β΅ β C with Β΅ = 1, n β₯ 0, we have
Ξ²(r)n (Ξ», x)
=1
(1β Β΅)s
nβm=0
Ξ»βm
nβl=m
nβlβk=0
sβi=0
(n
l
)(nβ lk
)(s
i
)S1 (l,m)Ξ»l (βΒ΅)
sβi
Γ Ξ²(r)k (Ξ») (i|Ξ»)nβlβk
H(s)
m (x|Β΅) .
Remark. For n β₯ 0, we have
H(s)n (x|Β΅)
=
nβm=0
1
Ξ»m
nβmβl=0
nβlβk=m
(n
l
)(nβ lk
)S2 (k,m)Ξ»k+lB
(r)l H
(r)nβlβkβj (Β΅)
Ξ²(r)m (Ξ», x) .
ACKNOWLEDGEMENTS.The work reported in this paper was conducted dur-ing the sabbatical year of Kwangwoon University in 2014.
References
1. S. Araci and M. Acikgoz, A note on the Frobenius-Euler numbers and poly-nomials associated with Bernstein polynomials, Adv. Stud. Contemp. Math.(Kyungshang) 22 (2012), no. 3, 399β406. MR 2976598
2. A. Bayad and T. Kim, Identities for the Bernoulli, the Euler and the Genoc-chi numbers and polynomials, Adv. Stud. Contemp. Math. (Kyungshang) 20(2010), no. 2, 247β253. MR 2656975 (2011d:11274)
3. L. Carlitz, A degenerate Staudt-Clausen theorem, Arch. Math. (Basel) 7 (1956),28β33. MR 0074436 (17,586a)
4. , Degenerate Stirling, Bernoulli and Eulerian numbers, Utilitas Math.15 (1979), 51β88. MR 531621 (80i:05014)
5. D. Ding and J. Yang, Some identities related to the Apostol-Euler and Apostol-Bernoulli polynomials, Adv. Stud. Contemp. Math. (Kyungshang) 20 (2010),no. 1, 7β21. MR 2597988 (2011k:05030)
6. S. Gaboury, R. Tremblay, and B.-J. Fugere, Some explicit formulas for cer-tain new classes of Bernoulli, Euler and Genocchi polynomials, Proc. JangjeonMath. Soc. 17 (2014), no. 1, 115β123. MR 3184467
7. Y. He and W. Zhang, A convolution formula for Bernoulli polynomials, ArsCombin. 108 (2013), 97β104. MR 3060257
8. F. T. Howard, Explicit formulas for degenerate Bernoulli numbers, DiscreteMath. 162 (1996), no. 1-3, 175β185. MR 1425786 (97m:11024)
9. K.-W. Hwang, D. V. Dolgy, D. S. Kim, T. Kim, and S. H. Lee, Some the-orems on Bernoulli and Euler numbers, Ars Combin. 109 (2013), 285β297.MR 3087218
10. Hassan Jolany, R. Eizadi Alikelaye, and S. Sharif Mohamad, Some results onthe generalization of Bernoulli, Euler and Genocchi polynomials, Acta Univ.Apulensis Math. Inform. (2011), no. 27, 299β306. MR 2896405 (2012k:11026)
11. D. S. Kim and T. Kim, Some identities of Bernoulli and Euler polynomialsarising form umbral calculus, Adv. Stud. Contemp. Math. (Kyungshang) 23(2013), no. 1, 159β171. MR 3059325
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
810 DAE SAN KIM et al 789-811
HIGHER-ORDER DEGENERATE BERNOULLI POLYNOMIALS 23
12. D. S. Kim, T. Kim, S.-H. Lee, and J.-J. Seo, A note on the lambda-Daeheepolynomials, Int. J. Math. Anal. (Ruse) 7 (2013), no. 61-64, 3069β3080.MR 3162167
13. T. Kim, Some identities for the Bernoulli, the Euler and the Genocchi numbersand polynomials, Adv. Stud. Contemp. Math. (Kyungshang) 20 (2010), no. 1,23β28. MR 2597989 (2011a:11211)
14. , Identities involving Laguerre polynomials derived from umbral calculus,Russ. J. Math. Phys. 21 (2014), no. 1, 36β45. MR 3182545
15. S.-H. Rim and J. Jeong, Identities on the modified q-Euler and q-Bernsteinpolynomials and numbers with weight, J. Comput. Anal. Appl. 15 (2013), no. 1,39β44. MR 3076716
16. S. Roman, The umbral calculus, Pure and Applied Mathematics, vol. 111, Aca-demic Press, Inc. [Harcourt Brace Jovanovich, Publishers], New York, 1984.MR 741185 (87c:05015)
17. E. Sen, Theorems on Apostol-Euler polynomials of higher order arising fromEuler basis, Adv. Stud. Contemp. Math. (Kyungshang) 23 (2013), no. 2, 337β345. MR 3088764
18. A. V. Ustinov, Korobov polynomials and umbral analysis, ChebyshevskiΔ± Sb. 4(2003), no. 4(8), 137β152. MR 2097912 (2005m:33018)
19. M. Wu and H. Pan, Sums of products of the degenerate Euler numbers, Ad-vances in Difference Equations (2014), 2014:40.
20. P. T. Young, Degenerate Bernoulli polynomials, generalized factorial sums, andtheir applications, J. Number Theory 128 (2008), no. 4, 738β758. MR 2400037(2009b:11045)
21. Z. Zhang and H. Yang, Some closed formulas for generalized Bernoulli-Eulernumbers and polynomials, Proc. Jangjeon Math. Soc. 11 (2008), no. 2, 191β198.MR 2482602 (2010a:11036)
Department of Mathematics, Sogang University, Seoul 121-742, Republic of Korea
E-mail address: [email protected]
Department of Mathematics, Kwangwoon University, Seoul 139-701, Republic of Ko-
reaE-mail address: [email protected]
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
811 DAE SAN KIM et al 789-811
KOROBOV POLYNOMIALS OF THE SEVENTH KIND AND OFTHE EIGHTH KIND
DAE SAN KIM, TAEKYUN KIM, TOUFIK MANSOUR, AND JONG-JIN SEO
Abstract. In this paper, we consider the Korobov polynomials of the seventh kindand of the eighth kind. We present several explicit formulas and recurrence relationsfor these polynomials. In addition, we establish connections between our polynomialsand several known families of polynomials.
1. Introduction
The degenerate Bernoulli polynomials are the degenerate version of Bernoulli polyno-mials introduced by Calitz [3, 4]. On the other hand, the Korobov polynomials of thefirst kind are the first degenerate version of the Bernoulli polynomials of the secondkind, see [13,14].
In recent years, many researchers studied various kinds of degenerate versions of fam-ilies polynomials like Bernoulli polynomials, Euler polynomials, falling factorial poly-nomials, Bell polynomials and their variants, see [6β10] and references therein. Alongthis line of research, we introduced in [8, 9] four kinds of new degenerate versions ofBernoulli polynomials of the second kind, called the Korobov polynomials of the third,fourth, fifth, and sixth kind.
Here, we will discuss two other degenerate versions of Bernoulli polynomials of thesecond kind, namely, the Korobov polynomials of the seventh and eighth kind. We willinvestigate some properties, explicit expressions, recurrence relations, and connectionswith other families polynomials with the help of umbral calculus (see [10, 15, 16]). Todo that, we recall some families polynomials. The Bernoulli polynomials of the secondkind bn(x) are given by the generating function
t
log(1 + t)(1 + t)x =
βnβ₯0
bn(x)tn
n!.(1.1)
For x = 0, bn = bn(0) are called the Bernoulli numbers of the second kind. The Daeheepolynomials Dn(x) are defined by the generating function
log(1 + t)
t(1 + t)x =
βnβ₯0
Dn(x)tn
n!.(1.2)
2010 Mathematics Subject Classification. 05A19, 05A40, 11B83.Key words and phrases. Korobov polynomials of the seventh kind and of the eighth kind, Umbral
calculus.1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
812 DAE SAN KIM et al 812-824
2 DAE SAN KIM, TAEKYUN KIM, TOUFIK MANSOUR, AND JONG-JIN SEO
When x = 0, Dn = Dn(0) are called the Daehee numbers. The Krobov polynomialsKn(Ξ», x) of the first kind are given by
Ξ»t
(1 + t)Ξ» β 1(1 + t)x =
βnβ₯0
Kn(Ξ», x)tn
n!.(1.3)
When x = 0, Kn(Ξ») = Kn(Ξ», 0) are called the Korobov numbers of the first kind. Thedegenerate falling factorial polynomials (x)n,Ξ» were defined in [7] by the generatingfunction
(1 + Ξ»)xΞ»
(1+t)Ξ»β1Ξ» =
βnβ₯0
(x)n,Ξ»tn
n!.(1.4)
Clearly, limΞ»β0(x)n,Ξ» = (x)n, the nth falling factorial polynomial. These polynomialscan be defined as (x)n,Ξ» βΌ (1, f(t)), where
f(t) =
(1 +
Ξ»2t
log(1 + Ξ»)
) 1Ξ»
β 1 and f(t) =log(1 + Ξ»)
Ξ»
(1 + t)Ξ» β 1
Ξ».(1.5)
Note that we write sn(x) βΌ (g(t), f(t)) ifβ
nβ₯0 sn(x) tn
n!= 1
g(f(t))exf(t), where f(t) is the
compositional inverse of f(t), see [15, 16]. The degenerate Stirling numbers of the firstkind S1(n, k | Ξ»), n β₯ k β₯ 0, were given in [7] by the generating function
1
k!
((1 + t)Ξ» β 1
Ξ»
)k
=βnβ₯k
S1(n, k | Ξ»)tn
n!,(1.6)
so that, in the notation of umbral calculus, S1(n, k | Ξ») = 1k!
β¨((1+t)Ξ»β1
Ξ»
)k|xnβ©
. Then,
it was shown in [7] that
(x)n,Ξ» =nβ
k=0
(log(1 + Ξ»)
Ξ»
)k
S1(n, k | Ξ»)xk
with
S1(n, k | Ξ») =nβ
m=k
S1(n,m)S2(m, k)Ξ»mβk,
where limΞ»β0 S1(n, k | Ξ») = S1(n, k) is the Stirling number of the first kind.
Here, we introduce Korobov polynomials of the seventh kindKn,7(Ξ», x) and of the eighthkind Kn,9(Ξ», x), respectively given by
log(1 + Ξ»t)
Ξ» log(1 + t)(1 + Ξ»)
xΞ»
(1+t)Ξ»β1Ξ» =
βnβ₯0
Kn,7(Ξ», x)tn
n!,(1.7)
log(1 + Ξ»t)
(1 + t)Ξ» β 1(1 + Ξ»)
xΞ»
(1+t)Ξ»β1Ξ» =
βnβ₯0
Kn,8(Ξ», x)tn
n!.(1.8)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
813 DAE SAN KIM et al 812-824
KOROBOV POLYNOMIALS OF THE SEVENTH KIND AND OF THE EIGHTH KIND 3
When x = 0, Kn,7(Ξ») = Kn,7(Ξ», 0) and Kn,8(Ξ») = Kn,8(Ξ», 0) are called the Korobovnumbers of the seventh kind and of the eighth kind, respectively. We observe that
limΞ»β0
Ξ»t
(1 + t)Ξ» β 1(1 + t)x = lim
Ξ»β0
log(1 + Ξ»t)
Ξ» log(1 + t)(1 + Ξ»)
xΞ»
(1+t)Ξ»β1Ξ»
= limΞ»β0
log(1 + Ξ»t)
(1 + t)Ξ» β 1(1 + Ξ»)
xΞ»
(1+t)Ξ»β1Ξ» =
t
log(1 + t)(1 + t)x,
which implies that limΞ»β0Kn(Ξ», x) = limΞ»β0Kn,7(Ξ», x) = limΞ»β0Kn,8(Ξ», x) = bn(x).It is immediate to see that Kn,7(Ξ», x) and Kn,8(Ξ», x) are Sheffer sequences (see [15,16])
for the respective pairs(
Ξ» log(1+f(t))log(1+Ξ»f(t))
, f(t))
and(
(1+f(t))Ξ»β1log(1+Ξ»f(t))
, f(t))
, where f(t) is given
in (1.5). Thus, (1.7) and (1.8) can be presented as
Kn,7(Ξ», x) βΌ(Ξ» log(1 + f(t))
log(1 + Ξ»f(t)), f(t)
)=
log(
1 + Ξ»2tlog(1+Ξ»)
)log(1 + Ξ»f(t))
, f(t)
,(1.9)
Kn,8(Ξ», x) βΌ(
(1 + f(t))Ξ» β 1
log(1 + Ξ»f(t)), f(t)
)=
(Ξ»2t
log(1+Ξ»)
log(1 + Ξ»f(t)), f(t)
).(1.10)
In the next two sections, we will use umbral calculus in order to study some properties,explicit formulas, recurrence relations and identities about the Korobov polynomialsof the seventh kind and of the eighth kind. In last section, we present connectionsbetween our polynomials and several known families of polynomials.
2. Explicit expressions
In this section, we present several explicit formulas for the Korobov polynomials of theseventh kind and of the eighth kind, namely Kn,7(Ξ», x) and Kn,8(Ξ», x).
Theorem 2.1. For all n β₯ 0,
Kn,7(Ξ», x) =nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)Knββ,7(Ξ»)xk
=nβ
k=0
nββ=k
nβββm=0
(n
β
)(nβ βm
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)bmDnβββmΞ»
nβββmxk,
Kn,8(Ξ», x) =nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)Knββ,8(Ξ»)xk
=nβ
k=0
nββ=k
nβββm=0
(n
β
)(nβ βm
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)Km(Ξ»)DnβββmΞ»
nβββmxk.
Proof. We proceed the proof by using the conjugation representation for Sheffer se-quences (see [15,16]): sn(x) =
βnk=0
1k!β¨g(f(t))β1f(t)k|xnβ©xk, for any sn(x) βΌ (g(t), f(t)).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
814 DAE SAN KIM et al 812-824
4 DAE SAN KIM, TAEKYUN KIM, TOUFIK MANSOUR, AND JONG-JIN SEO
Thus, by (1.9), we have
Kn,7(Ξ», x) =nβ
k=0
1
k!
β¨log(1 + Ξ»t)
Ξ» log(1 + t)
logk(1 + Ξ»)((1 + t)Ξ» β 1)k
Ξ»2k|xnβ©xk
=nβ
k=0
logk(1 + Ξ»)
Ξ»k
β¨log(1 + Ξ»t)
Ξ» log(1 + t)|((1 + t)Ξ» β 1)k
k!Ξ»kxnβ©xk,
which, by (1.6) and (1.7), implies
Kn,7(Ξ», x) =nβ
k=0
logk(1 + Ξ»)
Ξ»k
β¨log(1 + Ξ»t)
Ξ» log(1 + t)|βββ₯k
S1(β, k|Ξ»)tβ
β!xn
β©xk
=nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)
β¨log(1 + Ξ»t)
Ξ» log(1 + t)|xnββ
β©xk(2.1)
=nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)Knββ,7(Ξ»)xk.
On the other hand, by (2.1), we have
Kn,7(Ξ», x) =nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)
β¨log(1 + Ξ»t)
Ξ»t| t
log(1 + t)xnββ
β©xk,
which, by (1.1) and (1.2), we obtain
Kn,7(Ξ», x)
=nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)
β¨log(1 + Ξ»t)
Ξ»t|βmβ₯0
bmtm
m!xnββ
β©xk
=nβ
k=0
nββ=k
nβββm=0
(n
β
)(nβ βm
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)bm
β¨log(1 + Ξ»t)
Ξ»t|xnβββm
β©xk
=nβ
k=0
nββ=k
nβββm=0
(n
β
)(nβ βm
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)bm
β¨βjβ₯0
DjΞ»j t
j
j!|xnβββm
β©xk
=nβ
k=0
nββ=k
nβββm=0
(n
β
)(nβ βm
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)bmDnβββmΞ»
nβββmxk,
which completes the proof of formulas for kn,7(Ξ», x).
Now let us deal with the case Kn,8(Ξ», x). Similarly, by using the conjugation represen-tation for Sheffer sequences, (1.10) and (1.6), we obtain
Kn,8(Ξ», x) =nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)
β¨log(1 + Ξ»t)
(1 + t)Ξ» β 1|xnββ
β©xk(2.2)
=nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)Knββ,8(Ξ»)xk.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
815 DAE SAN KIM et al 812-824
KOROBOV POLYNOMIALS OF THE SEVENTH KIND AND OF THE EIGHTH KIND 5
On the other hand, by (2.2), we have
Kn,8(Ξ», x) =nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)
β¨log(1 + Ξ»t)
Ξ»t| Ξ»t
(1 + t)Ξ» β 1xnββ
β©xk,
which, by (1.3) and (1.2), we obtain
Kn,8(Ξ», x)
=nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)
β¨log(1 + Ξ»t)
Ξ»t|βmβ₯0
Km(Ξ»)tm
m!xnββ
β©xk
=nβ
k=0
nββ=k
nβββm=0
(n
β
)(nβ βm
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)Km(Ξ»)
β¨log(1 + Ξ»t)
Ξ»t|xnβββm
β©xk
=nβ
k=0
nββ=k
nβββm=0
(n
β
)(nβ βm
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)Km(Ξ»)DnβββmΞ»
nβββmxk,
(2.3)
which completes the proof.
Now, we express our polynomials in terms of the degenerate falling factorial polyno-mials.
Theorem 2.2. For all n β₯ 0,
Kn,7(Ξ», x) =nβ
β=0
(n
β
)( nβββm=0
(nβ βm
)Ξ»nβββmbmDnβββm
)(x)β,Ξ»,
Kn,8(Ξ», x) =nβ
β=0
(n
β
)( nβββm=0
(nβ βm
)Ξ»nβββmKm(Ξ»)Dnβββm
)(x)β,Ξ».
Proof. By (1.9), we have
Kn,7(Ξ», y) =
β¨log(1 + Ξ»t)
Ξ» log(1 + t)(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ» |xn
β©=
β¨log(1 + Ξ»t)
Ξ» log(1 + t)|(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ» xn
β©,
which, by (1.4), implies
Kn,7(Ξ», y) =
β¨log(1 + Ξ»t)
Ξ» log(1 + t)|βββ₯0
(y)β,Ξ»tβ
β!xn
β©=
nββ=0
(n
β
)(y)β,Ξ»
β¨log(1 + Ξ»t)
Ξ» log(1 + t)|xnββ
β©.
Therefore, by (2.1), we obtain
Kn,7(Ξ», y) =nβ
β=0
(n
β
)( nβββm=0
(nβ βm
)Ξ»nβββmbmDnβββm
)(y)β,Ξ»,
which completes the proof for Kn,7(Ξ», y).
By using similar arguments as above together with (1.10) and (1.4), we obtain
Kn,8(Ξ», y) =nβ
β=0
(n
β
)(y)β,Ξ»
β¨log(1 + Ξ»t)
(1 + t)Ξ» β 1|xnββ
β©.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
816 DAE SAN KIM et al 812-824
6 DAE SAN KIM, TAEKYUN KIM, TOUFIK MANSOUR, AND JONG-JIN SEO
Therefore, by (2.2) and (2.3), we have
Kn,8(Ξ», y) =nβ
β=0
(n
β
)( nβββm=0
(nβ βm
)Ξ»nβββmKm(Ξ»)Dnβββm
)(y)β,Ξ»,
which completes the proof.
In the next theorem, we find explicit formulas for the coefficient of xj in Kn,7(Ξ», x) andKn,8(Ξ», x).
Theorem 2.3. For all n β₯ 0 and s = 7, 8,
Kn,s(Ξ», x)
=nβ
j=0
(nβ
k=j
kββ=0
ββm=0
(β1)ββm
β!
(β
m
)(k
j
)(m|Ξ»)kβj
logj(1 + Ξ»)
Ξ»jS1(n, k|Ξ»)Kβ,s(Ξ»)
)xj.
Proof. By (1.4) and (1.9), we have
Ξ» log(1 + f(t))
log(1 + Ξ»f(t))Kn,7(Ξ», x) = (x)n,Ξ» =
nβk=0
logk(1 + Ξ»)
Ξ»kS1(n, k|Ξ»)xk βΌ (1, f(t)).
Thus,
Kn,7(Ξ», x) =nβ
k=0
logk(1 + Ξ»)
Ξ»kS1(n, k|Ξ»)
log(1 + Ξ»f(t))
Ξ» log(1 + f(t))xk
=nβ
k=0
kββ=0
logk(1 + Ξ»)
Ξ»kS1(n, k|Ξ»)
Kβ,7(Ξ»)
β!(f(t))βxk.(2.4)
Note that
(f(t))βxk =ββ
m=0
(β
m
)(β1)ββm
(1 + Ξ»2t/ log(1 + Ξ»)
)m/Ξ»xk
=ββ
m=0
kβj=0
(β
m
)(β1)ββm(m|Ξ»)j (Ξ»/ log(1 + Ξ»))j
(k
j
)xkβj.
Therefore,
Kn,7(Ξ», x)
=nβ
k=0
kββ=0
ββm=0
kβj=0
(β1)ββm(m|Ξ»)kβjlogj(1 + Ξ»)
Ξ»jS1(n, k|Ξ»)
Kβ,7(Ξ»)
β!
(β
m
)(k
j
)xj
=nβ
j=0
(nβ
k=j
kββ=0
ββm=0
(β1)ββm
β!
(β
m
)(k
j
)(m|Ξ»)kβj
logj(1 + Ξ»)
Ξ»jS1(n, k|Ξ»)Kβ,7(Ξ»)
)xj.
By (1.4) and (1.10), we have
(1 + f(t))Ξ» β 1
log(1 + Ξ»f(t))Kn,8(Ξ», x) = (x)n,Ξ» =
nβk=0
logk(1 + Ξ»)
Ξ»kS1(n, k|Ξ»)xk βΌ (1, f(t)).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
817 DAE SAN KIM et al 812-824
KOROBOV POLYNOMIALS OF THE SEVENTH KIND AND OF THE EIGHTH KIND 7
Thus, by using the above arguments, we obtain
Kn,8(Ξ», x)
=nβ
j=0
(nβ
k=j
kββ=0
ββm=0
(β1)ββm
β!
(β
m
)(k
j
)(m|Ξ»)kβj
logj(1 + Ξ»)
Ξ»jS1(n, k|Ξ»)Kβ,8(Ξ»)
)xj,
which completes the proof.
In the next theorem, we express Korobov polynomials of seventh and eighth kinds interms of Korobov polynomials of fifth and sixth kinds.
Theorem 2.4. For all n β₯ 0 and s = 7, 8,
Kn,s(Ξ», x) =nβ
β=0
(n
β
)DnββΞ»
nββKβ,sβ2(Ξ», x).
Proof. Recall that Korobov polynomials of the fifth kind (see [9]) is given by
t
log(1 + t)(1 + Ξ»)
xΞ»
(1+t)Ξ»β1Ξ» =
βnβ₯0
Kn,5(Ξ», x)tn
n!.
So, by (1.7), we have
Kn,7(Ξ», y) =
β¨log(1 + Ξ»t)
Ξ»t| t
log(1 + t)(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ» xn
β©=
nββ=0
(n
β
)Kβ,5(Ξ», y)
β¨log(1 + Ξ»t)
Ξ»t|xnββ
β©,
which, by (1.2), implies Kn,7(Ξ», y) =βn
β=0
(nβ
)Kβ,5(Ξ», y)DnββΞ»
nββ.
Recall that Korobov polynomials of the sixth kind (see [9]) is defined by
Ξ»t
(1 + t)Ξ» β 1(1 + Ξ»)
xΞ»,y
(1+t)Ξ»β1Ξ» =
βnβ₯0
Kn,6(Ξ», x)tn
n!.
Similarly, by (1.2) and (1.8), we obtain Kn,8(Ξ», y) =βn
β=0
(nβ
)Kβ,6(Ξ», y)DnββΞ»
nββ, asclaimed.
In the next theorem, we express our polynomials Kn,7(Ξ», x) and Kn,8(Ξ», x) in terms
of degenerate Bernoulli numbers Ξ²(n)β (Ξ») of order n, which are given by the generating
function
tn
((1 + Ξ»t)1/Ξ» β 1)n=βββ₯0
Ξ²(n)β (Ξ»)
tβ
β!.(2.5)
Theorem 2.5. For all n β₯ 1 and s = 7, 8,
Kn,s(Ξ», x)
nβj=0
(nβ
k=j
kββ=0
ββm=0
(β1)ββm(nβ1kβ1
)(βm
)(kj
)β!
(m|Ξ»)kβj
(log(1 + Ξ»)
Ξ»
)j
Ξ²(n)nβk(Ξ»)Kβ,s(Ξ»)
)xj.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
818 DAE SAN KIM et al 812-824
8 DAE SAN KIM, TAEKYUN KIM, TOUFIK MANSOUR, AND JONG-JIN SEO
Proof. It is not hard to see that logn(1+Ξ»)xn
Ξ»n βΌ (1, Ξ»t/ log(1 + Ξ»)). Thus, by (1.9), wehave
Ξ» log(1 + f(t))
log(1 + Ξ»f(t))Kn,7(Ξ», x) = x
(Ξ»t
log(1+Ξ»)
(1 + Ξ»2t/ log(1 + Ξ»))1/Ξ» β 1
)n
xβ1 logn(1 + Ξ»)xn
Ξ»n
=logn(1 + Ξ»)
Ξ»nx
(r
(1 + Ξ»r)1/Ξ» β 1
)n
|r=Ξ»t/ log(1+Ξ») xnβ1,
which, by (2.5), implies
Ξ» log(1 + f(t))
log(1 + Ξ»f(t))Kn,7(Ξ», x) =
logn(1 + Ξ»)
Ξ»nxβkβ₯0
Ξ²(n)k (Ξ»)
1
k!
(Ξ»t
log(1 + Ξ»)
)k
xnβ1
=nβ1βk=0
(nβ 1
k
)Ξ²(n)k (Ξ»)
(log(1 + Ξ»)
Ξ»
)nβk
xnβk
=nβ
k=0
(nβ 1
k β 1
)Ξ²(n)nβk(Ξ»)
(log(1 + Ξ»)
Ξ»
)k
xk.
On the other hand, by (2.4), we have
log(1 + Ξ»f(t))
Ξ» log(1 + f(t))xk
=kβ
β=0
ββm=0
kβj=0
Kβ,7(Ξ»)
β!
(β
m
)(k
j
)(β1)ββm(m|Ξ»)kββj
Ξ»kβj
logkβj(1 + Ξ»)xj.
Therefore, the polynomials Kn,7(Ξ», x) is given by
nβj=0
(nβ
k=j
kββ=0
ββm=0
(β1)ββm(nβ1kβ1
)(βm
)(kj
)β!
(m|Ξ»)kβj
(log(1 + Ξ»)
Ξ»
)j
Ξ²(n)nβk(Ξ»)Kβ,7(Ξ»)
)xj.
By using similar argument as above with using (1.10), we obtain the formula for the nthKorobov polynomial kn,8(Ξ», x) of the eighth kind (we leave the details for the interestedreader).
3. Recurrences
In this section, we present several recurrences for the Korobov polynomials of theseventh kind and of the eighth kind. Note that, by (1.9), (1.10) and the fact that(x)n,Ξ» βΌ (1, f(t)), we obtain Kn,d(Ξ», x+ y) =
βnj=0
(nj
)Kj,d(Ξ», x)(y)nβj,Ξ», for d = 7, 8.
Proposition 3.1. For all n β₯ 1 and s = 7, 8,
Kn,s(Ξ», x) + nKnβ1,s(Ξ», x)
=nβ
m=0
(nβ
k=m
nββ=k
(n
β
)(k
m
)(1|Ξ»)kβm
logm(1 + Ξ»)
Ξ»mS1(β, k|Ξ»)Knββ,s(Ξ»)
)xm.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
819 DAE SAN KIM et al 812-824
KOROBOV POLYNOMIALS OF THE SEVENTH KIND AND OF THE EIGHTH KIND 9
Proof. It is well-known that if sn(x) βΌ (g(t), f(t)), then we have f(t)sn(x) = nsnβ1(x)
(see [15,16]). Thus, by (1.9) and (1.10), we obtain
((1 + Ξ»2t
log(1+Ξ»)
) 1Ξ» β 1
)Kn,s(Ξ», x) =
nKnβ1,s(Ξ», x), which implies Kn,s(Ξ», x) + nKnβ1,s(Ξ», x) =(
1 + Ξ»2tlog(1+Ξ»)
) 1Ξ»Kn,s(Ξ», x).
By Theorem 2.1 we have
Kn,s(Ξ», x) + nKnβ1,s(Ξ», x)
=nβ
k=0
nββ=k
(n
β
)logk(1 + Ξ»)
Ξ»kS1(β, k|Ξ»)Knββ,s(Ξ»)
(1 +
Ξ»2t
log(1 + Ξ»)
) 1Ξ»
xk
=nβ
k=0
nββ=k
kβm=0
(n
β
)logkβm(1 + Ξ»)
Ξ»kβmS1(β, k|Ξ»)Knββ,s(Ξ»)(1|Ξ»)m
tm
m!xk
=nβ
k=0
nββ=k
kβm=0
(n
β
)(k
m
)logkβm(1 + Ξ»)
Ξ»kβmS1(β, k|Ξ»)Knββ,s(Ξ»)(1|Ξ»)mx
kβm
=nβ
m=0
(nβ
k=m
nββ=k
(n
β
)(k
m
)(1|Ξ»)kβm
logm(1 + Ξ»)
Ξ»mS1(β, k|Ξ»)Knββ,s(Ξ»)
)xm.
which completes the proof.
In the next result, we express ddxKn,7(Ξ», x) and d
dxKn,8(Ξ», x) in terms of Kn,7(Ξ», x) and
Kn,8(Ξ», x), respectively.
Proposition 3.2. For all n β₯ 0 and s = 7, 8,
d
dxKn,s(Ξ», x) =
log(1 + Ξ»)
Ξ»2
nβ1ββ=0
(n
β
)(Ξ»)nββKβ,s(Ξ», x).
Proof. Note that ddxsn(x) =
βnβ1β=0
(nβ
)β¨f(t)|xnβββ©sβ(x), for all sn(x) βΌ (g(t), f(t)), see
[15, 16]. So, for sn(x) = Kn,s(Ξ», x), it remains to compute A = β¨f(t)|xnβββ©. By (1.9)
and (1.10), we have A = log(1+Ξ»)Ξ»2 β¨
βjβ₯1(Ξ»)j
tj
j!|xnβββ© = log(1+Ξ»)
Ξ»2 (Ξ»)nββ, which completes
the proof.
Theorem 3.3. For all n β₯ 1 and s = 7, 8,
Kn,s(Ξ», x) =x log(1 + Ξ»)
Ξ»
nβ1ββ=0
(nβ 1
β
)(Ξ»β 1)nβ1ββKβ,s(Ξ», x)
+1
n
nββ=0
nβββm=0
(n
β
)(nβ β)nβββmus(β,m),
where
u7(β) = bβ
(βΞ»)nβββm(x)m,Ξ» β (β1)nβββmKm,7(Ξ», x),
u8(β) = Kβ(Ξ»)
(βΞ»)nβββm(x)m,Ξ» β
(Ξ»β 1
nβ ββm
)Km,8(Ξ», x)
.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
820 DAE SAN KIM et al 812-824
10 DAE SAN KIM, TAEKYUN KIM, TOUFIK MANSOUR, AND JONG-JIN SEO
Proof. Since the similarity between Kn,7(Ξ», x) and Kn,8(Ξ», x) (see (1.9) and (1.10)), weomit the proof of the case Kn,8(Ξ», x) and give only the details of the case Kn,7(Ξ», x).By (1.9), we have
Kn,7(Ξ», y) =
β¨d
dt
(log(1 + Ξ»t)
Ξ» log(1 + t)(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ»
)|xnβ1
β©= A+B,(3.1)
whereB =
β¨ddt
log(1+Ξ»t)Ξ» log(1+t)
(1 + Ξ»)yΞ»
(1+t)Ξ»β1Ξ» |xnβ1
β©andA =
β¨log(1+Ξ»t)Ξ» log(1+t)
ddt
(1 + Ξ»)yΞ»
(1+t)Ξ»β1Ξ» |xnβ1
β©.
First, we compute the term B.
B =
β¨log(1 + Ξ»t)
Ξ» log(1 + t)(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ»
log(1 + Ξ»)
Ξ»y(1 + t)Ξ»β1|xnβ1
β©=y log(1 + Ξ»)
Ξ»
β¨log(1 + Ξ»t)
Ξ» log(1 + t)(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ» |(1 + t)Ξ»β1xnβ1
β©=y log(1 + Ξ»)
Ξ»
nβ1ββ=0
(nβ 1
β
)(Ξ»β 1)β
β¨log(1 + Ξ»t)
Ξ» log(1 + t)(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ» |xnβ1ββ
β©
=y log(1 + Ξ»)
Ξ»
nβ1ββ=0
(nβ 1
β
)(Ξ»β 1)βKnβ1ββ,7(Ξ», y)
=y log(1 + Ξ»)
Ξ»
nβ1ββ=0
(nβ 1
β
)(Ξ»β 1)nβ1ββKβ,7(Ξ», y).
Now, we compute the first term A,
A =
β¨t
log(1 + t)
1
t
1
1 + Ξ»tβ log(1 + Ξ»t)
Ξ» log(1 + t)(1 + t)β1
(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ» |xnβ1
β©=
1
n
β¨1
1 + Ξ»tβ log(1 + Ξ»t)
Ξ» log(1 + t)(1 + t)β1
(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ» | t
log(1 + t)xnβ©
=1
n
β¨1
1 + Ξ»tβ log(1 + Ξ»t)
Ξ» log(1 + t)(1 + t)β1
(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ» |
βββ₯0
bβtβ
β!xn
β©.
Note that 11+Ξ»tβ log(1+Ξ»t)
Ξ» log(1+t)(1 + t)β1 has order at least one. Thus,
A =1
n
nββ=0
(n
β
)bβ
β¨(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ» | 1
1 + Ξ»txnββ
β©ββ¨
log(1 + Ξ»t)
Ξ» log(1 + t)(1 + Ξ»)
yΞ»
(1+t)Ξ»β1Ξ» | 1
1 + txnββ
β©=
1
n
nββ=0
(n
β
)bβ
nβββm=0
(βΞ»)m(nβ β)m
β¨βkβ₯0
(y)k,Ξ»tk
k!|xnβββm
β©
βnβββm=0
(β1)m(nβ β)m
β¨βkβ₯0
Kk,7(Ξ», y)tk
k!|xnβββm
β©,
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
821 DAE SAN KIM et al 812-824
KOROBOV POLYNOMIALS OF THE SEVENTH KIND AND OF THE EIGHTH KIND 11
which implies
A =1
n
nββ=0
nβββm=0
(n
β
)bβ
(βΞ»)m(nβ β)m(y)nβββm,Ξ» β (β1)m(nβ β)mKnβββm,7(Ξ», y)
,
Hence, by substituting the expressions of A and B in (3.1), we complete the proof.
4. Connections with families of polynomials
In this section, we present some examples on the connections with families of poly-nomials. To do that, we recall for any two Sheffer sequences sn(x) βΌ (g(t), f(t)) andrn(x) βΌ (h(t), β(t)), we have that sn(x) =
βnm=0Cn,mrm(x), where (see [15,16])
Cn,m =1
m!
β¨h(f(t))
g(f(t))(β(f(t)))m|xn
β©.(4.1)
We start with the connection to Bernoulli polynomials B(s)n (x) of order s. Recall that
the Bernoulli polynomials B(s)n (x) of order s are defined by the generating function(
tetβ1
)sext =
βnβ₯0B
(s)n (x) t
n
n!, equivalently,
B(s)n (x) βΌ
((et β 1
t
)s
, t
)(4.2)
(see [2, 5, 15]). In the next result, we express our polynomials in terms of Bernoullipolynomials of order s.
Theorem 4.1. Let d = 7, 8. For all n β₯ 0, Kn,d(Ξ», x) =βn
k=0Cn,mB(s)m (x), where
Cn,m =nβmββ=0
nβββmβk=0
(nβ
)(k+mm
)(k+ss
) Kβ,d(Ξ»)S2(k + s, s)logk+m(1 + Ξ»)
Ξ»k+mS1(nβ β, k +m|Ξ»).
Proof. Since the similarity between Kn,7(Ξ», x) and Kn,8(Ξ», x) (see (1.9) and (1.10)), weomit the proof of the case Kn,8(Ξ», x) and give only the details of the case Kn,7(Ξ», x).
Let Kn,7(Ξ», x) =βn
m=0Cn,mB(s)m (x). So, by (1.9) and (4.2), we have
Cn,m =1
m!
β¨(ef(t) β 1)s
f s(t)
log(1 + Ξ»t)
Ξ» log(1 + t)fm(t)|xn
β©
=1
m!
β¨(ef(t) β 1)s
f s(t)fm(t)| log(1 + Ξ»t)
Ξ» log(1 + t)xn
β©
=1
m!
β¨(ef(t) β 1)s
f s(t)fm(t)|
βββ₯0
Kβ,7(Ξ»)tβ
β!xn
β©
=1
m!
nββ=0
(n
β
)Kβ,7(Ξ»)
β¨s!βkβ₯0
S2(k + s, s)fk+m(t)
(k + s)!|xnββ
β©.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
822 DAE SAN KIM et al 812-824
12 DAE SAN KIM, TAEKYUN KIM, TOUFIK MANSOUR, AND JONG-JIN SEO
Thus,
Cn,m =s!
m!
nβmββ=0
nβββmβk=0
(n
β
)Kβ,7(Ξ»)S2(k + s, s)
β¨fk+m(t)
(k + s)!|xnββ
β©
=s!
m!
nβmββ=0
nβββmβk=0
(n
β
)Kβ,7(Ξ»)S2(k + s, s)
logk+m(1 + Ξ»)
(k + s)!Ξ»k+m
β¨((1 + t)Ξ» β 1)k+m
Ξ»k+m|xnββ
β©
=s!
m!
nβmββ=0
nβββmβk=0
(n
β
)Kβ,7(Ξ»)S2(k + s, s)
logk+m(1 + Ξ»)
(k + s)!Ξ»k+m(k +m)!S1(nβ β, k +m|Ξ»).
Therefore,
Cn,m =nβmββ=0
nβββmβk=0
(nβ
)(k+mm
)(k+ss
) Kβ,7(Ξ»)S2(k + s, s)logk+m(1 + Ξ»)
Ξ»k+mS1(nβ β, k +m|Ξ»),
as required.
Similar techniques as in the proof of the previous theorem, we can express our poly-nomials Kn,7(Ξ», x), Kn,8(Ξ», x) in terms of other families. Below we present two ex-amples, where we leave the proofs to the interested reader. In the first example,we express our polynomials in terms of Frobenius-Euler polynomials. Note that the
Frobenius-Euler polynomials H(s)n (x|Β΅) of order s are defined by the generating function(
1βΒ΅etβΒ΅
)sext =
βnβ₯0H
(s)n (x|Β΅) t
n
n!, (Β΅ = 1), or equivalently, H
(s)n (x|Β΅) βΌ
((etβΒ΅1βΒ΅
)s, t)
(see [1, 11,12]).
Theorem 4.2. For all n β₯ 0 and d = 7, 8, Kn,d(Ξ», x) =βn
m=0Cn,mH(s)m (x|Β΅), where
Cn,m =nβmββ=0
sβk=0
nβββmβj=k
k!(j+mm
)(nβ
)(sk
)(1β Β΅)k
logj+m(1 + Ξ»)
Ξ»j+mS1(nβ β, j +m|Ξ»)S2(j, k)Kβ,d(Ξ»).
For what follows, we define the associated sequence for 1 β (1 + Ξ»2t/ log(1 + Ξ»))β1/Ξ»,namely (x)(n,Ξ»). Thus,
(x)(n,Ξ») βΌ (1, 1β (1 + Ξ»2t/ log(1 + Ξ»))β1/Ξ»).
Recall here that (x)n βΌ (1, et β 1), (x)(n) βΌ (1, 1 β eβt), (x)n,Ξ» βΌ (1, (1 + Ξ»2t/ log(1 +Ξ»))1/Ξ» β 1) and (1 + Ξ»2t/ log(1 + Ξ»))1/Ξ» β 1 β et β 1, as Ξ» β 0. Now, we ready topresent our second example.
Theorem 4.3. For all n β₯ 0 and d = 7, 8, Kn,d(Ξ», x) =βn
m=0Cn,m(x)(m,Ξ»), where
Cn,m =nβ
β=0
(β1)nβββm
(n
β
)(nβ βm
)(nβ 1β β)nβββmKβ,d(Ξ»).
ACKNOWLEDGEMENTS.The work reported in this paper was conducted during thesabbatical year of Kwangwoon University in 2014.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
823 DAE SAN KIM et al 812-824
KOROBOV POLYNOMIALS OF THE SEVENTH KIND AND OF THE EIGHTH KIND 13
References
[1] S. Araci and M. Acikgoz, A note on the Frobenius-Euler numbers and polynomials associatedwith Bernstein polynomials, Adv. Stud. Contemp. Math. 22(3) (2012) 399β406.
[2] A. Bayad and T. Kim, Identities involving values of Bernstein, q-Bernoulli, and q-Euler polyno-mials, Russ. J. Math. Phys. 18(2) (2011) 133β143.
[3] L. Carlitz, A degenerate Staudt-Clausen theorem, Arch. Math. 7 (1956) 28β33.[4] L. Carlitz, Degenerate Stirling, Bernoulii and Eulerian numbers, Utilitas Math. 15 (1979) 51-88.[5] D. Ding and J. Yang, Some identities related to the Apostol-Euler and Apostol-Bernoulli poly-
nomials, Adv. Stud. Contemp. Math. 20(1) (2010) 7β21.[6] D.S. Kim and T. Kim, On degenerate Bell numbers and polynomials, preprint.[7] D.S. Kim and T. Kim, Degenerate falling factorial polynomials, preprint.[8] D.S. Kim and T. Kim, Korobov polynomials of the third kind and of the fourth kind, preprint.[9] D.S. Kim, T. Kim, H.I. Kwon and T. Mansour, Korobov polynomials of the fifth kind and of the
sixth kind, preprint.[10] D. S. Kim and T. Kim, Some identities of Korobov-type polynomials associated with p-adic
integrals on Zp, Adv. Difference Equ. 2015 (2015), DOI: 10.1186/s13662-015-0602-8.[11] T. Kim, Identities involving Laguerre polynomials derived from umbral calculus, Russ. J. Math.
Phys. 21 (2014) 36β45.[12] T. Kim and T. Mansour, Umbral calculus associated with Frobenius-type Eulerian polynomials,
Russ. J. Math. Phys. 21 (2014) 484β493 .[13] N.M. Korobov, On some properties of special polynomials, Proceedings of the IV International
Conference βModern Problems of Number Theory and its Applicationsβ (Russian)(Tula, 2001),Vol. 1, 2001, 40β49.
[14] N.M. Korobov, Speical polynomials and their applications, in βDiophantine Approcimationsβ,Mathematical Notes (Russian), Vol. 2, Izd. Moskov. Univ., Moscow, 1996, 77β89.
[15] S. Roman, The umbral calculus, Pure and Applied Mathematics 111, Academic Press, New York,1984.
[16] S. Roman, More on the umbral calculus, with emphasis on the q-umbral calculus, J. Math. Anal.Appl. 107 (1985) 222β254.
[17] A.V. Ustinov, Korobov polynomials and umbral calculus, ChebyshevskiΔ± Sb. 4:8 (2003) 137β152.
Department of Mathematics, Sogang University, Seoul 121-742, South Korea
E-mail address: [email protected]
Department of Mathematics, Kwangwoon University, Seoul 139-701, South Korea
E-mail address: [email protected]
University of Haifa, Department of Mathematics, 3498838 Haifa, Israel
E-mail address: [email protected]
Department of Applied Mathematics, Pukyong National University, Busan 48513, Re-public of Korea.
E-mail address: [email protected]
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
824 DAE SAN KIM et al 812-824
Some identities on the higher-order twisted q-Euler numbersand polynomials
C. S. Ryoo
Department of Mathematics, Hannam University, Daejeon 306-791, Korea
Abstract : In this paper we investigate some interesting symmetric identities for twisted q-Euler
polynomials of higher order in complex field.
Key words : Symmetric properties, power sums, Euler numbers and polynomials, twisted q-Euler
numbers and polynomials.
2000 Mathematics Subject Classification : 11B68, 11S40, 11S80.
1. Introduction
Euler numbers and polynomials possess many interesting properties and arising in many areas
of mathematics, mathematical physics and statistical physics. Many mathematicians have studied
in the area of the q- extension of Euler numbers and polynomials(see [1, 2, 3, 5, 6, 7, 8, 9, 11,
13]). Recently, Y. Hu studied several identities of symmetry for Carlitzβs q-Bernoulli numbers and
polynomials in complex field(see [3]). D. Kim et al.[4] derived some identities of symmetry for
(h, q)-extension of higher-order Euler numbers and polynomials. D. V. Dolgy et al.[2] derived some
identities of symmetry for higher-order generalized q-Euler polynomials. In this paper, we establish
some interesting symmetric identities for twisted q-Euler polynomials of higher order in complex
field. The purpose of this paper is to present a systemic study of the twisted q-Euler numbers and
polynomials of higher-order by using the multiple q-Euler zeta function. Throughout this paper, the
notations N,Z,R, and C denote the sets of positive integers, integers, real numbers, and complex
numbers, respectively, and Z+ := N βͺ 0. We assume that q β C with |q| < 1. Throughout this
paper we use the notation:
[x]q =1β qx
1β q(cf. [1, 2, 3, 5]) .
Note that limqβ1[x] = x. Let Ξ΅ be the pN -th root of unity(see [10, 12, 13]).
In [5], T. Kim introduced the multiple q-Euler zeta function which interpolates higher-order
q-Euler polynomials at negative integers as follows:
ΞΆq,r(s, x) = [2]rq
ββm1,Β·Β·Β· ,mr=0
(β1)βr
j=1 mjqβr
j=1 mj
[m1 + Β· Β· Β·+mr + x]sq, (1)
where s β C and x β R, with x = 0,β1,β2, . . ..
Recently, D. V. Dolgy et al.[2] considered some symmetric identities for higher-order generalized
q-Euler polynomials. The Euler polynomials of order r β N attached to Ο are also defined by the
generating function: (2dβ1βl=0
Ο(l)(β1)le(x+l)t
edt + 1
)r
=ββ
m=0
E(r)m,Ο(x)
tm
m!. (2)
When x = 0, E(r)n,Ο = E
(r)n,Ο(0) are called the Euler numbers E
(r)n,Ο attached to Ο(see [2, 4]).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
825 C. S. Ryoo 825-830
For h β Z, Ξ±, k β N, and n β Z+, we introduced the higher order twisted q-Euler polynomials
with weight Ξ± as follows(see [7]):
E(Ξ±)n,q,Ξ΅(k|x) =
[2]kq(1β qΞ±)n
nβl=0
(n
l
)(β1)l
qΞ±lx
(1 + Ξ΅qΞ±l+h) Β· Β· Β· (1 + Ξ΅qΞ±l+hβk+1).
In the special case, x = 0, E(Ξ±)n,q,w(k|0) = E
(Ξ±)n,q,w(k) are called the higher-order twisted q-Euler
numbers with weight Ξ±.
We consider the higher order q-Euler polynomials of order r attached to Ο twisted by ramified
roots of unity as follows(see [10]):
ββn=0
E(r)n,Ο,Ξ΅,q(x)
tn
n!= 2r
ββm1,...,mr=0
(βΞ΅)βr
j=0 mj
(rβ
i=1
Ο(mi)
)e[x+
βrj=1 mj ]qt.
In the special case x = 0, the sequence E(r)n,Ο,Ξ΅,q(0) = E
(r)n,Ο,Ξ΅,q are called the n-th q-Euler numbers of
order r attached to Ο twisted by ramified roots of unity.
As is well known, the higher-order twisted q-Euler polynomials E(k)n,q,Ξ΅(x) are defined by the
following generating function to be
F (k)q,Ξ΅ (t, x) = [2]kq
ββm1,Β·Β·Β· ,mk=0
(β1)m1+Β·Β·Β·+mkΞ΅m1+Β·Β·Β·+mke[m1+Β·Β·Β·+mk+x]qt
=
ββn=0
E(k)n,q,Ξ΅(x)
tn
n!,
(3)
where k β N. When x = 0, E(k)n,q,Ξ΅ = E
(k)n,q,Ξ΅(0) are called the higher-order twisted q-Euler numbers
E(k)n,q,Ξ΅. Observe that if q β 1, Ξ΅β 1, then E
(k)n,q,Ξ΅ β E
(k)n and E
(k)n,q,Ξ΅(x)β E
(k)n,Ο(x).
By using (3) and Cauchy product, we have
E(k)n,q,Ξ΅(x) =
nβl=0
(n
l
)qlxE
(k)l,q,Ξ΅[x]nβl
q
= (qxE(k)q,Ξ΅ + [x]q)n,
(4)
with the usual convention about replacing (E(k)q,Ξ΅ )n by E
(k)n,q,Ξ΅.
By using complex integral and (3), we can also obtain the multiple twisted q-l-function as
follows:
l(k)q,Ξ΅ (s, x) =1
Ξ(s)
β« β
0
F (k)q,Ξ΅ (βt, x)tsβ1dt
= [2]kq
ββm1,Β·Β·Β· ,mk=0
(β1)βk
j=1 mjΞ΅βk
j=1 mj
[m1 + Β· Β· Β·+mk + x]sq,
(5)
where s β C and x β R, with x = 0,β1,β2, . . ..
By using Cauchy residue theorem, the value of multiple twisted q-l-function at negative integers
is given explicitly by the following theorem:
Theorem 1. Let k β N and n β Z+. We obtain
l(k)q,Ξ΅ (βn, x) = E(k)n,q,Ξ΅(x).
The purpose of this paper is to obtain some interesting identities of the power sums and
the higher-order twisted q-Euler polynomials E(k)n,q,Ξ΅(x) using the symmetric properties for multiple
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
826 C. S. Ryoo 825-830
twisted q-l-function. In this paper, if we take Ξ΅ = 1 in all equations of this article, then [2] are the
special case of our results.
2. Symmetry identities for multiple twisted q-l-function
In this section, by using the similar method of [2, 3, 4], expect for obvious modifications, we
investigate some symmetric identities for higher-order twisted q-Euler polynomials E(k)n,q,Ξ΅(x). We
assume that Ξ΅ be the pN -th root of unity. Let w1, w2 β N with w1 β‘ 1 (mod 2), w2 β‘ 1 (mod 2). For
h β Z, k β N and n β Z+, we obtain certain symmetry identities for multiple twisted q-l-function.
Observe that [xy]q = [x]qy [y]q for any x, y β C. In (5), we derive next result by substitute
w2x+w2
w1(j1 + Β· Β· Β·+ jk) for x in and replace q and Ξ΅ by qw1 and Ξ΅w1 , respectively.
1
[2]kqw1
l(k)qw1 ,Ξ΅w1 (s, w2x+
w2
w1(j1 + Β· Β· Β·+ jk))
=ββ
m1,Β·Β·Β· ,mk=0
(β1)βk
j=1 mjΞ΅w1βk
j=1 mj
[m1 + Β· Β· Β·+mk + w2x+w2
w1(j1 + Β· Β· Β·+ jk)]sqw1
=
ββm1,Β·Β·Β· ,mk=0
(β1)βk
j=1 mjΞ΅w1
βkj=1 mj[
w1(m1 + Β· Β· Β·+mk) + w1w2x+ w2(j1 + Β· Β· Β·+ jk)
w1
]sqw1
=
ββm1,Β·Β·Β· ,mk=0
(β1)βk
j=1 mjΞ΅w1
βkj=1 mj
[w1(m1 + Β· Β· Β·+mk) + w1w2x+ w2(j1 + Β· Β· Β·+ jk)]sq[w1]sq
= [w1]sq
ββm1,Β·Β·Β· ,mk=0
(β1)βk
j=1 mjΞ΅w1
βkj=1 mj
[w1(m1 + Β· Β· Β·+mk) + w1w2x+ w2(j1 + Β· Β· Β·+ jk)]sq
= [w1]sq
ββm1,Β·Β·Β· ,mk=0
w2β1βi1,Β·Β·Β· ,ik=0
(β1)βk
j=1 mjΞ΅w1βk
j=1 mj
[w1(m1 + Β· Β· Β·+mk) + w1w2x+ w2(j1 + Β· Β· Β·+ jk)]sq
= [w1]sq
ββm1,Β·Β·Β· ,mk=0
w2β1βi1,Β·Β·Β· ,ik=0
(β1)βk
j=1(dw2mj+ij)
Γ Ξ΅w1
βkj=1(dw2mj+ij)
Γ([w1(dw2m1 + i1) + Β· Β· Β·+ w1(dw2mk + ik) + w1w2x+ w2(j1 + Β· Β· Β·+ jk)]sq
)β1
= [w1]sq
ββm1,Β·Β·Β· ,mk=0
w2β1βi1,Β·Β·Β· ,ik=0
(β1)βk
j=1 mj (β1)βk
j=1 ij
Γ Ξ΅dw1w2
βkj=1 mjΞ΅w1
βkj=1 ij
Γ([w1w2(x+ dm1 + Β· Β· Β·+ dmk) + w1(i1 + Β· Β· Β·+ ik) + w2(j1 + Β· Β· Β·+ jk)]sq
)β1
(6)
Thus, from (6), we can derive the following equation.
[w2]sq[2]kqw1
w1β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w2βk
l=1 jl
Γ l(k)qw1 ,Ξ΅w1 (s, w2x+w2
w1(j1 + Β· Β· Β·+ jk))
= [w1]sq[w2]sq
ββm1,Β·Β·Β· ,mk=0
w2β1βi1,Β·Β·Β· ,ik=0
w1β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1(jl+il+ml)
Γ Ξ΅dw1w2βk
l=1 mlΞ΅w1βk
l=1 ilΞ΅w2βk
l=1 jl
Γ([w1w2(x+ dm1 + Β· Β· Β·+ dmk) + w1(i1 + Β· Β· Β·+ ik) + w2(j1 + Β· Β· Β·+ jk)]sq
)β1
(7)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
827 C. S. Ryoo 825-830
By using the same method as (7), we have
[w1]sq[2]kqw2
w2β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w1
βkl=1 jl
Γ l(k)qw2 ,Ξ΅w2 (s, w1x+w1
w2(j1 + Β· Β· Β·+ jk))
= [w1]sq[w2]sq
ββm1,Β·Β·Β· ,mk=0
w2β1βj1,Β·Β·Β· ,jk=0
w1β1βi1,Β·Β·Β· ,ik=0
(β1)βk
l=1(jl+il+ml)
Γ Ξ΅dw1w2
βkl=1 mlΞ΅w2
βkl=1 ilΞ΅w1
βkl=1 jl
Γ([w1w2(x+ dm1 + Β· Β· Β·+ dmk) + w1(j1 + Β· Β· Β·+ jk) + w2(i1 + Β· Β· Β·+ ik)]sq
)β1
(8)
Therefore, by (7) and (8), we have the following theorem.
Theorem 2. Let w1, w2 β N with w1 β‘ 1 (mod 2), w2 β‘ 1 (mod 2). For h β Z , we obtain
[w2]sq[2]kqw2
w1β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w2
βkl=1 jl l
(k)qw1 ,Ξ΅w1
(s, w2x+
w2
w1(j1 + Β· Β· Β·+ jk)
)
= [w1]sq[2]kqw1
w2β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w1βk
l=1 jl l(k)qw2 ,Ξ΅w2
(s, w1x+
w1
w2(j1 + Β· Β· Β·+ jk)
) (9)
By (9) and Theorem 1, we obtain the following theorem.
Theorem 3. Let w1, w2 β N with w1 β‘ 1 (mod 2), w2 β‘ 1 (mod 2). For h β Z, k β N and
n β Z+, we obtain
[w2]sq[2]kqw2
w1β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w2
βkl=1 jlE
(k)n,qw1 ,Ξ΅w1
(w2x+
w2
w1(j1 + Β· Β· Β·+ jk)
)
= [w1]sq[2]kqw1
w2β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w1βk
l=1 jlE(k)n,qw2 ,Ξ΅w2
(w1x+
w1
w2(j1 + Β· Β· Β·+ jk)
).
(10)
From (4), we note that
E(k)n,q,Ξ΅(x+ y) = (qx+yE(k)
n,q,Ξ΅ + [x+ y]q)n
=
nβi=0
(n
i
)qxiE
(k)i,q,Ξ΅(y)[x]nβi
q .(11)
with the usual convention about replacing (E(k)q,Ξ΅ )n by E
(k)n,q,Ξ΅.
By (11), we have
w1β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w2
βkl=1 jlE
(k)n,qw1 ,Ξ΅w1
(w2x+
w2
w1(j1 + Β· Β· Β·+ jk)
)
=
w1β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w2βk
l=1 jl
Γnβ
i=0
(n
i
)qw2i(j1+Β·Β·Β·+jk)E
(k)i,qw1 ,Ξ΅w1 (w2x)
[w2
w1(j1 + Β· Β· Β·+ jk)
]nβi
qw1
=
w1β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w2βk
l=1 jl
Γnβ
i=0
(n
i
)qw2(nβi)
βkl=1 jlE
(k)nβi,qw1 ,Ξ΅w1 (w2x)
[w2
w1(j1 + Β· Β· Β·+ jk)
]iqw1
(12)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
828 C. S. Ryoo 825-830
Hence we have the following theorem.
Theorem 4. Let w1, w2 β N with w1 β‘ 1 (mod 2), w2 β‘ 1 (mod 2). For k β N and n β Z+,
we obtain
w1β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w2
βkl=1 jlE
(k)n,qw1 ,Ξ΅w1
(w2x+
w2
w1(j1 + Β· Β· Β·+ jk)
)
=nβ
i=0
(n
i
)[w2]iq[w1]βi
q E(k)nβi,qw1 ,Ξ΅w1 (w2x)
w1β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w2βk
l=1 jl [j1 Β· Β· Β·+ jk]iqw2 .
For each integer n β₯ 0, let
S(k)n,i,q,Ξ΅(w) =
wβ1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅βk
l=1 jl [j1 Β· Β· Β·+ jk]iq.
The above sum S(k)n,i,q,Ξ΅(w) is called the alternating q-power sums.
By Theorem 4, we have
[2]kqw2 [w1]nq
w1β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w2βk
l=1 jlE(k)n,qw1 ,Ξ΅w1
(w2x+
w2
w1(j1 + Β· Β· Β·+ jk)
)
= [2]kqw2
nβi=0
(n
i
)[w2]iq[w1]nβi
q E(k)nβi,qw1 ,Ξ΅w1 (w2x)S(k)n,i,qw2 ,Ξ΅w2 (w1)
(13)
By using the same method as in (13), we have
[2]kqw1 [w2]nq
w2β1βj1,Β·Β·Β· ,jk=0
(β1)βk
l=1 jlΞ΅w1
βkl=1 jlE
(k)n,qw2 ,Ξ΅w2
(w1x+
w1
w2(j1 + Β· Β· Β·+ jk)
)
= [2]kqw1
nβi=0
(n
i
)[w1]iq[w2]nβi
q E(k)nβi,qw2 ,Ξ΅w2 (w1x)S(k)n,i,qw1 ,Ξ΅w1 (w2)
(14)
Therefore, by (13) and (14) and Theorem 3, we have the following theorem.
Theorem 5. Let w1, w2 β N with w1 β‘ 1 (mod 2), w2 β‘ 1 (mod 2). For k β N and n β Z+,
we obtain
[2]kqw2
nβi=0
(n
i
)[w2]iq[w1]nβi
q E(k)nβi,qw1 ,Ξ΅w1 (w2x)S(k)n,i,qw2 ,Ξ΅w2 (w1)
= [2]kqw1
nβi=0
(n
i
)[w1]iq[w2]nβi
q E(k)nβi,qw2 ,Ξ΅w2 (w1x)S(k)n,i,qw1 ,Ξ΅w1 (w2).
By Theorem 5, we obtain the interesting symmetric identity for the higher-order twisted q-Euler
numbers E(k)n,q,Ξ΅ in complex field.
Corollary 6. Let w1, w2 β N with w1 β‘ 1 (mod 2), w2 β‘ 1 (mod 2). For k β N and n β Z+,
we obtain
[2]kqw2
nβi=0
(n
i
)[w2]iq[w1]nβi
q S(k)n,i,qw2 ,Ξ΅w2 (w1)E(k)nβi,qw1 ,Ξ΅w1
= [2]kqw1
nβi=0
(n
i
)[w1]iq[w2]nβi
q S(k)n,i,qw1 ,Ξ΅w1 (w2)E(k)nβi,qw2 ,Ξ΅w2 .
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
829 C. S. Ryoo 825-830
REFERENCES
1. M. Cenkci, The p-adic generalized twisted (h, q)-Euler-l-function and its applications, Adv.
Stud. Contemp. Math., 15(2007), 34-47.
2. D. V. Dolgy, D.S. Kim, T.G. Kim, J.J. Seo, Identities of Symmetry for Higher-Order Gener-
alized q-Euler Polynomials, Abstract and Applied Analysis, 2014(2014), Article ID 286239, 6
pages.
3. Yuan He, Symmetric identities for Carlitzβs q-Bernoulli numbers and polynomials, Adv.
Difference Equ., 246(2013), 10 pages.
4. D. Kim, T. Kim, J.-J. Seo, Identities of symmetric for (h, q)-extension of higher-order Euler
polynomials, Applied Mathemtical Sciences 8 (2014), 3799-3808.
5. T. Kim, New approach to q-Euler polynomials of higher order, Russ. J. Math. Phys.
17(2010), 218-225.
6. T. Kim, Barnes type multiple q-zeta function and q-Euler polynomials, J. phys. A : Math.
Theor. 43(2010) 255201(11pp).
7. H. Y. Lee, N. S. Jung, J. Y. Kang, C. S. Ryoo, Some identities on the higher-order-twisted
q-Euler numbers and polynomials with weight Ξ±, Adv. Difference Equ., 2012:21(2012), 10pp.
8. E.-J. Moon, S.-H. Rim, J.-H. Jin, S.-J. Lee, On the symmetric properties of higher-order
twisted q-Euler numbers and polynomials, Adv. Difference Equ., 2010, Art ID 765259, 8pp.
9. H. Ozden, Y. Simsek, I. N. Cangul, Euler polynomials associated with p-adic q-Euler measure,
Gen. Math., 15(2007), 24-37.
10. C. S. Ryoo, On the generalized Barnes type multiple q-Euler polynomials twisted by ramified
roots of unity, Proc. Jangjeon Math. Soc. 13(2010), 255-263.
11. C. S. Ryoo, A note on the weighted q-Euler numbers and polynomials, Adv. Stud. Contemp.
Math., 21(2011), 47-54.
12. Y. Simsek, q-analogue of twisted l-series and q-twisted Euler numbers, Journal of Number
Theory, 110(2005), 267-278.
13. Y. Simsek, Twisted (h, q)-Bernoulli numbers and polynomials related to twisted (h, q)-zeta
function and L-function, J. Math. Anal. Appl., 324(2006), 790-804.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
830 C. S. Ryoo 825-830
UMBRAL CALCULUS ASSOCIATED WITH NEW DEGENERATE
BERNOULLI POLYNOMIALS
DAE SAN KIM, TAEKYUN KIM, AND JONG-JIN SEO
Abstract. In this paper, we introduce new degenerate Bernoulli polynomi-als which are derived from umbral calculus and investigate some interesting
properties of those polynomials.
1. Introduction
The Bernoulli polynomials are defined by the generating function
(1.1)t
et β 1ext =
ββn=0
Bn (x)tn
n!, (see [1β14]) .
When x = 0, Bn = Bn (0) are called the ordinary Bernoulli numbers. From (1.1),we note that
(1.2) Bn (x) =nβ
l=0
(n
l
)Blx
nβl, (n β₯ 0) , (see [13]) .
Thus, by (1.2), we get
(1.3)d
dxBn (x) = nBnβ1 (x) , (n β N) .
In [2], L. Carlitz introduced the degenerate Bernoulli polynomials which aregiven by the generating function
(1.4)t
(1 + Ξ»t)1Ξ» β 1
(1 + Ξ»t)xΞ» =
ββn=0
Ξ²n (x | Ξ»)tn
n!.
When x = 0, Ξ²n (0 | Ξ») = Ξ²n (Ξ») are called Carlitzβs degenerate Bernoulli num-bers (see [2]).
Thus, by (1.4), we get
(1.5) Ξ²n (x | Ξ») =nβ
l=0
(n
l
)Ξ²l (Ξ») (x | Ξ»)nβl , (n β₯ 0) ,
where (x | Ξ»)n = x (xβ Ξ») Β· Β· Β· (xβ Ξ» (nβ 1)).Let C be the field of complex numbers and let F be the set of all formal power
series in the variable t over C with
F =
f (t) =
ββk=0
aktk
k!
β£β£β£β£β£ ak β C
.
2010 Mathematics Subject Classification. 11B83, 11B75, 05A19, 05A40.Key words and phrases. Degenerate Bernoulli polynomial, Higher-order degenerate Bernoulli
polynomial, Umbral calculus.
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
831 DAE SAN KIM et al 831-840
2 DAE SAN KIM, TAEKYUN KIM, AND JONG-JIN SEO
Let P = C [x] and Pβ denotes the vector space of all linear functionals on P. Theaction of the linear functional L β Pβ on a polynomial p (x) is denoted by β¨L| p (x)β©,and linearly extended as β¨cL+ cβ²Lβ²| p (x)β© = c β¨L| p (x)β©+cβ² β¨Lβ²| p (x)β©, where c andcβ² β C.
For f (t) =ββ
k=0 aktk
k! β F , we define a linear functional on P by setting
(1.6) β¨f (t)|xnβ© = an
for all n β₯ 0, (see [1, 5, 13]).Thus, by (1.6), we get
(1.7)β¨tkβ£β£xnβ© = n!Ξ΄n,k, (n, k β₯ 0) , (see [7, 13]) ,
where Ξ΄n,k is the Kroneckerβs symbol.
Let fL (t) =ββ
k=0
β¨L|xk
β©tk
k! . Then we have β¨fL (t)|xnβ© = β¨L|xnβ© (n β₯ 0). Themapping L 7β fL (t) is a vector space isomorphism from Pβ onto F . Henceforth,F will denote both the algebra of formal power series in t and the vector space ofall linear functionals on P, and so an element f (t) of F will be thought of as botha formal power seires and a linear functional. We shall call F the umbral algebra.The umbral calculus is the study of umbral algebra and can be also desribed as asystematic study of the class of Sheffer sequences. The order o (f) of the non-zeropower series f (t) is the smallest integer k for which the coefficient of tk does notvanish (see [12, 13]).
For f (t) , g (t) β F with o (f) = 1 and o (g) = 0, there exists a unique sequencesn (x) of polynomials such thatβ¨
g (t) f (t)kβ£β£β£ sn (x)
β©= n!Ξ΄n,k, (n, k β₯ 0) .
The sequence sn (x) is called the Sheffer sequence for (g (t) , f (t)) which is de-noted by sn (x) βΌ (g (t) , f (t))(see [10, 13]).
Let f (t) β F and p (x) β P. Then by (1.7), we get
(1.8)β¨eytβ£β£ p (x)
β©= p (y) , β¨f (t) g (t)| p (x)β© = β¨g (t)| f (t) p (x)β© ,
and
(1.9) f (t) =ββk=0
β¨f (t)|xk
β© tkk!, p (x) =
ββk=0
β¨tkβ£β£ p (x)
β© xkk!, (see [13]) .
By (1.9), we easily get
(1.10) p(k) (0) =β¨tkβ£β£ p (x)
β©=β¨
1| p(k) (x)β©, (k β₯ 0) ,
where p(k) (0) denotes the k-th derivative of p (x) with respect to x at x = 0.From (1.10), we have
(1.11) tkp (x) = p(k) (x) .
In [13], it is known that
(1.12) sn (x) βΌ (g (t) , f (t)) ββ 1
g(f (t)
)exf(t) =
ββn=0
sn (x)tn
n!,
where f (t) is the compositional inverse of f (t) such that f(f (t)
)= f (f (t)) = t.
From (1.7), we can easily derive
(1.13) eytp (x) = p (x+ y) , where p (x) β P = C [x] .
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
832 DAE SAN KIM et al 831-840
UMBRAL CALCULUS ASSOCIATED WITH NEW DEGENERATE BERNOULLI POLYNOMIALS3
For p (x) β P, we haveβ¨eyt β 1
t
β£β£β£β£ p (x)
β©=
Λ y
0
p (u) du, β¨f (t)|xp (x)β© = β¨βtf (t)| p (x)β© .
Let f1 (t) , f2 (t) , . . . , fm (t) β F . Then we have
(1.14) β¨f1 (t) f2 (t) Β· Β· Β· fm (t)|xnβ© =β(
n
i1, . . . , im
)β¨f1 (t)|xi1
β©Β· Β· Β·β¨fm (t)|xim
β©where the sum is over all nonnegative integers i1, . . . , im such that i1+ Β· Β· Β·+ im = n.
In this paper, we introduce new degenerate Bernoulli polynomials which aredifferent Carlitzβs degenerate Bernoulli polynomials and investigate some interestingproperties of those polynomials.
2. Umbral calculus and degenerate Bernoulli polynomials
From (1.1) and (1.13), we have
(2.1) Bn (x) βΌ(et β 1
t, t
), (n β₯ 0) .
Now, we introduce the new degenerate Bernoulli polynomials which are derivedfrom Sheffer sequence as follows:
(2.2) Ξ²n,Ξ» (x) βΌ
((1 + Ξ»t)
1Ξ» β 1
t, t
), (n β₯ 0) .
From (1.12) and (2.2), we have
(2.3)ββ
n=0
Ξ²n,Ξ» (x)tn
n!=
t
(1 + Ξ»t)1Ξ» β 1
ext.
When x = 0, Ξ²n,Ξ» = Ξ²n,Ξ» (0) are called the degenerate Bernoulli numbers.Note that
ββn=0
limΞ»β0
Ξ²n,Ξ» (x)tn
n!= lim
Ξ»β0
t
(1 + Ξ»t)1Ξ» β 1
ext(2.4)
=t
et β 1ext
=ββ
n=0
Bn (x)tn
n!.
Thus, by (2.4), we get
(2.5) limΞ»β0
Ξ²n,Ξ» (x) = Bn (x) , (n β₯ 0) .
From (2.3), we have
ββn=0
Ξ²n,Ξ» (x)tn
n!=
( ββl=0
Ξ²l,Ξ»tl
l!
)( ββm=0
xm
m!tm
)(2.6)
=ββ
n=0
(nβ
l=0
(n
l
)Ξ²l,Ξ»x
nβl
)tn
n!.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
833 DAE SAN KIM et al 831-840
4 DAE SAN KIM, TAEKYUN KIM, AND JONG-JIN SEO
Thus, by (2.6), we get
Ξ²n,Ξ» (x) =nβ
l=0
(n
l
)Ξ²l,Ξ»x
nβl, (n β₯ 0) ,
and
d
dxΞ²n,Ξ» (x) =
nβl=1
(n
l
)Ξ²l,Ξ» (nβ l)xnβlβ1(2.7)
= nnβ1βl=0
(nβ 1
l
)Ξ²l,Ξ»x
nβ1βl
= nΞ²nβ1,Ξ» (x) .
From (1.11) and (2.3), we have
(2.8)t
(1 + Ξ»t)1Ξ» β 1
xn = Ξ²n,Ξ» (x) , (n β₯ 0) ,
and
(2.9) tΞ²n,Ξ» (x) =d
dxΞ²n,Ξ» (x) = nΞ²nβ1,Ξ» (x) , (n β₯ 1) .
Thus, by (2.8) and (2.9), we get
Λ x+y
x
Ξ²n,Ξ» (u) du(2.10)
=1
n+ 1Ξ²n+1,Ξ» (x+ y)β Ξ²n+1,Ξ» (x)
=eyt β 1
tΞ²n,Ξ» (x)
=ββk=1
yk
k!tkβ1Ξ²n,Ξ» (x) .
From (2.9), we have
(2.11) Ξ²n (x) = t
1
n+ 1Ξ²n+1,Ξ» (x)
.
Thus, by (2.11), we getβ¨eyt β 1
t
β£β£β£β£Ξ²n,Ξ» (x)
β©=
β¨eyt β 1
β£β£ 1
n+ 1Ξ²n+1,Ξ» (x)
β©(2.12)
=
Λ y
0
Ξ²n,Ξ» (u) du.
Therefore, by (2.12), we obtain the following theorem.
Theorem 1. For n β₯ 0, we haveβ¨eyt β 1
t
β£β£β£β£Ξ²n,Ξ» (x)
β©=
Λ y
0
Ξ²n,Ξ» (u) du.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
834 DAE SAN KIM et al 831-840
UMBRAL CALCULUS ASSOCIATED WITH NEW DEGENERATE BERNOULLI POLYNOMIALS5
For r β N, the degenerate Bernoulli polynomials of order r are defined by thegenerating function
(2.13)
(t
(1 + Ξ»t)1Ξ» β 1
)r
ext =ββ
n=0
Ξ²(r)n,Ξ» (x)
tn
n!.
When x = 0, Ξ²(r)n,Ξ» = Ξ²
(r)n,Ξ» (0) are called the higher order degenerate Bernoulli
numbers.Indeed, limΞ»β0 Ξ²
(r)n,Ξ» (x) = B
(r)n (x), where B
(r)n (x) are the higher-order Bernoulli
polynomials which are defined by the generating function(t
et β 1
)r
ext =ββ
n=0
B(r)n (x)
tn
n!.
From (2.13), we have
(2.14) Ξ²(r)n,Ξ» (x) =
nβl=0
(n
l
)Ξ²(r)l,Ξ»x
nβl, (n β₯ 0) ,
and
d
dxΞ²(r)n,Ξ» (x) =
nβl=0
(n
l
)Ξ²(r)l,Ξ» (nβ l)xnβlβ1(2.15)
= nnβ1βl=0
(nβ 1
l
)Ξ²(r)l,Ξ»x
nβlβ1
= nΞ²(r)nβ1,Ξ» (x) , (n β₯ 1) .
By (2.8) and (2.13), we easily get
(2.16) Ξ²(r)n,Ξ» =
βl1+Β·Β·Β·+lr=n
(n
l1, . . . , lr
)Ξ²l1,Ξ» Β· Β· Β·Ξ²lr,Ξ».
Thus, by (2.14) and (2.16), we see that Ξ²(r)n,Ξ» (x) is a monic polynomial of degree
n with coefficients in Q (Ξ»).From (2.14) and (2.15), we can deriveΛ x+y
x
Ξ²(r)n,Ξ» (u) du =
1
n+ 1
Ξ²(r)n+1,Ξ» (x+ y)β Ξ²(r)
n+1,Ξ» (x)
(2.17)
=eyt β 1
tΞ²(r)n,Ξ» (x) .
If sn (x) βΌ (g (t) , t), then sn (x) is called an Appell sequence.From (1.12) and (2.13), we have
(2.18) Ξ²(r)n,Ξ» (x) βΌ
(((1 + Ξ»t)
1Ξ» β 1
t
)r
, t
), (n β₯ 0) .
Thus, by (2.18), we note that Ξ²(r)n,Ξ» (x) is the Appell sequence for
((1+Ξ»t)
1Ξ» β1
t
)r
.
From (2.18), we have
(2.19)
((1 + Ξ»t)
1Ξ» β 1
t
)r
Ξ²(r)n,Ξ» (x) βΌ (1, t) , xn βΌ (1, t) , (n β₯ 0) .
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
835 DAE SAN KIM et al 831-840
6 DAE SAN KIM, TAEKYUN KIM, AND JONG-JIN SEO
Thus, by (2.19), we get
(2.20) xn =
((1 + Ξ»t)
1Ξ» β 1
t
)r
Ξ²(r)n,Ξ» (x) , (n β₯ 0) .
We observe that((1 + Ξ»t)
1Ξ» β 1
t
)r
(2.21)
=1
tr
(e
1Ξ» log(1+Ξ»t) β 1
)r=
1
trr!
ββl=r
S2 (l, r)Ξ»βl (log (1 + Ξ»t))l
l!
=1
trr!
ββl=0
S2 (l + r, r)Ξ»β(l+r) 1
(l + r)!(l + r)!
ββn=l+r
S1 (n, l + r)Ξ»ntn
n!
=1
trr!
ββl=0
S2 (l + r, r)Ξ»β(l+r)ββn=l
S1 (n+ r, l + r)Ξ»n+r
(n+ r)!tn+r
=
ββn=0
(nβ
l=0
S2 (l + r, r)S1 (n+ r, l + r)Ξ»nβl 1(n+rr
)) tn
n!
By (2.20) and (2.21), we get(2.22)
xm =mβ
n=0
nβl=0
S2 (l + r, r)S1 (n+ r, l + r)Ξ»nβl
(mn
)(n+rr
)Ξ²(r)mβn,Ξ» (x) , (m β₯ 0) .
Therefore, by (2.22), we obtain the following theorem.
Theorem 2. For m β₯ 0, we have
xm =
mβn=0
nβl=0
S2 (l + r, r)S1 (n+ r, l + r)Ξ»nβl
(mn
)(n+rr
)Ξ²(r)mβn,Ξ» (x) ,
where S1 (m,n) and S2 (m,n) are the Stirling numbers of the first kind and of thesecond kind defined by
(x)n =nβ
l=0
S1 (n, l)xl,
xn =nβ
l=0
S2 (n, l) (x)l .
From (1.11) and (2.18), we have
(2.23) tΞ²(r)n,Ξ» (x) = t
1
n+ 1Ξ²(r)n+1,Ξ» (x)
, (n β₯ 0) ,
and β¨eyt β 1
t
β£β£β£β£Ξ²(r)n,Ξ» (x)
β©=
β¨eyt β 1
β£β£ 1
n+ 1Ξ²(r)n+1,Ξ» (x)
β©(2.24)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
836 DAE SAN KIM et al 831-840
UMBRAL CALCULUS ASSOCIATED WITH NEW DEGENERATE BERNOULLI POLYNOMIALS7
=
Λ y
0
Ξ²(r)n,Ξ» (u) du.
Moreover,
Ξ²(r)n,Ξ»(2.25)
=
β¨(t
(1 + Ξ»t)1Ξ» β 1
)rβ£β£β£β£β£xnβ©
=β
n=i1+Β·Β·Β·+ir
(n
i1, . . . , ir
)β¨t
(1 + Ξ»t)1Ξ» β 1
β£β£β£β£β£xi1β©Β· Β· Β·
β¨t
(1 + Ξ»t)1Ξ» β 1
β£β£β£β£β£xirβ©
and
(2.26) Ξ²n,Ξ» =
β¨t
(1 + Ξ»t)1Ξ» β 1
β£β£β£β£β£xnβ©, (n β₯ 0) .
Therefore, by (2.24), (2.25) and (2.26), we obtain the following theorem.
Theorem 3. For n β₯ 0, we haveβ¨eyt β 1
t
β£β£β£β£Ξ²(r)n,Ξ» (x)
β©=
Λ y
0
Ξ²(r)n,Ξ» (u) du,
and
Ξ²(r)n,Ξ» =
βn=i1+Β·Β·Β·+ir
(n
i1, . . . , ir
)Ξ²i1,Ξ» Β· Β· Β·Ξ²ir,Ξ».
Let Pn = p (x) β C [x]| deg p (x) β€ n , (n β₯ 0). For p (x) β Pn, we assumethat
(2.27) p (x) =nβ
k=0
bkΞ²k,Ξ» (x) .
From (2.2), we have
(2.28)
β¨(1 + Ξ»t)
1Ξ» β 1
ttk
β£β£β£β£β£Ξ²n,Ξ» (x)
β©= n!Ξ΄n,k, (n, k β₯ 0) .
Thus, by (2.27) and (2.28), we getβ¨(1 + Ξ»t)
1Ξ» β 1
ttk
β£β£β£β£β£ p (x)
β©=
nβl=0
bl
β¨(1 + Ξ»t)
1Ξ» β 1
ttk
β£β£β£β£β£Ξ²l,Ξ» (x)
β©(2.29)
=nβ
l=0
bll!Ξ΄l,k = k!bk.
Hence,
bk =1
k!
β¨(1 + Ξ»t)
1Ξ» β 1
ttk
β£β£β£β£β£ p (x)
β©(2.30)
=1
k!
β¨(1 + Ξ»t)
1Ξ» β 1
t
β£β£β£β£β£ p(k) (x)
β©,
where p(k) (x) = dk
dxk p (x).Therefore, by (2.27) and (2.30), we obtain the following theorem.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
837 DAE SAN KIM et al 831-840
8 DAE SAN KIM, TAEKYUN KIM, AND JONG-JIN SEO
Theorem 4. Let p (x) β Pn. Then we have
p (x) =nβ
k=0
bkΞ²k,Ξ» (x) ,
where
bk =1
k!
β¨(1 + Ξ»t)
1Ξ» β 1
t
β£β£β£β£β£ p(k) (x)
β©.
Let p (x) β Pn with p (x) = Ξ²(r)n,Ξ» (x). Then, we have
(2.31) p(k) (x) =
(d
dx
)k
Ξ²(r)n,Ξ» (x) = k!
(n
k
)Ξ²(r)nβk,Ξ» (x) .
Let us assume that
(2.32) p (x) = Ξ²(r)n,Ξ» (x) =
nβk=0
bkΞ²k,Ξ» (x) .
Then, by Theorem 5, we get
bk =1
k!
β¨(1 + Ξ»t)
1Ξ» β 1
t
β£β£β£β£β£ p(k) (x)
β©(2.33)
=
(n
k
)β¨(1 + Ξ»t)
1Ξ» β 1
t
β£β£β£β£β£Ξ²(r)nβk,Ξ» (x)
β©
=
(n
k
)β¨(1 + Ξ»t)
1Ξ» β 1
t
β£β£β£β£β£(
t
(1 + Ξ»t)1Ξ» β 1
)r
xnβk
β©
=
(n
k
)β¨1
β£β£β£β£β£β£(
t
(1 + Ξ»t)1Ξ» β 1
)rβ1
xnβk
β©
=
(n
k
)Ξ²(rβ1)nβk,Ξ».
Therefore, by (2.32) and (2.33), we obtain the following theorem.
Theorem 5. For r β N and n β₯ 0, we have
Ξ²(r)n,Ξ» (x) =
nβk=0
(n
k
)Ξ²(rβ1)nβk,λβk,Ξ» (x) .
Let p (x) β Pn with p (x) =βn
k=0 b(r)k Ξ²
(r)k,Ξ» (x). By (2.18), we get
β¨((1 + Ξ»t)
1Ξ» β 1
t
)r
tk
β£β£β£β£β£ p (x)
β©=
nβl=0
b(r)l
β¨((1 + Ξ»t)
1Ξ» β 1
t
)r
tk
β£β£β£β£β£Ξ²(r)l,Ξ» (x)
β©(2.34)
=nβ
l=0
b(r)l l!Ξ΄l,k = k!b
(r)k .
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
838 DAE SAN KIM et al 831-840
UMBRAL CALCULUS ASSOCIATED WITH NEW DEGENERATE BERNOULLI POLYNOMIALS9
Thus, by (2.34), we get
(2.35) b(r)k =
1
k!
β¨((1 + Ξ»t)
1Ξ» β 1
t
)r
tk
β£β£β£β£β£ p (x)
β©.
Theorem 6. For p (x) β Pn, we have
p (x) =nβ
k=0
b(r)k Ξ²
(r)k,Ξ» (x) ,
where
b(r)k =
1
k!
β¨((1 + Ξ»t)
1Ξ» β 1
t
)r
tk
β£β£β£β£β£ p (x)
β©
=1
k!
β¨((1 + Ξ»t)
1Ξ» β 1
t
)rβ£β£β£β£β£ p(k) (x)
β©,
where p(k) (x) =(
ddx
)kp (x).
References
1. S. Araci, M. Acikgoz, and A. Kilicman, Extended p-adic q-invariant integralson Zp associated with applications of umbral calculus, Adv. Difference Equ.(2013), 2013:96, 14. MR 3055847
2. L. Carlitz, Degenerate Stirling, Bernoulli and Eulerian numbers, Utilitas Math.15 (1979), 51β88. MR 531621 (80i:05014)
3. R. Dere and Y. Simsek, Applications of umbral algebra to some special polyno-mials, Adv. Stud. Contemp. Math. (Kyungshang) 22 (2012), no. 3, 433β438.MR 2976601
4. Q. Fang and T. Wang, Umbral calculus and invariant sequences, Ars Combin.101 (2011), 257β264. MR 2828068 (2012e:05045)
5. S. Gaboury, R. Tremblay, and B.-J. Fugere, Some explicit formulas for cer-tain new classes of Bernoulli, Euler and Genocchi polynomials, Proc. JangjeonMath. Soc. 17 (2014), no. 1, 115β123. MR 3184467
6. Y. He and W. Zhang, A convolution formula for Bernoulli polynomials, ArsCombin. 108 (2013), 97β104. MR 3060257
7. D. S. Kim and T. Kim, Some identities of Bell polynomials, Science ChinaMathematics (2015).
8. , Some identities of degenerate special polynomials, Open Math. 13(2015), Art. 37. MR 3353617
9. D. S. Kim, T. Kim, S.-H. Lee, and S.-H. Rim, Umbral calculus and Eulerpolynomials, Ars Combin. 112 (2013), 293β306. MR 3112583
10. T. Kim, Identities involving Laguerre polynomials derived from umbral calculus,Russ. J. Math. Phys. 21 (2014), no. 1, 36β45. MR 3182545
11. , Barnesβ type multiple degenerate Bernoulli and Euler polynomials,Appl. Math. Comput. 258 (2015), 556β564. MR 3323091
12. T. Kim and T. Mansour, Umbral calculus associated with Frobenius-type Euler-ian polynomials, Russ. J. Math. Phys. 21 (2014), no. 4, 484β493. MR 3284958
13. S. Roman, The umbral calculus, Pure and Applied Mathematics, vol. 111, Aca-demic Press, Inc. [Harcourt Brace Jovanovich, Publishers], New York, 1984.MR 741185 (87c:05015)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
839 DAE SAN KIM et al 831-840
10 DAE SAN KIM, TAEKYUN KIM, AND JONG-JIN SEO
14. Y. Simsek, Generating functions of the twisted Bernoulli numbers and polyno-mials associated with their interpolation functions, Adv. Stud. Contemp. Math.(Kyungshang) 16 (2008), no. 2, 251β278. MR 2404639 (2009f:11021)
ACKNOWLEDGEMENTS.The work reported in this paper was conducted dur-ing the sabbatical year of Kwangwoon University in 2014.
Department of Mathematics, Sogang University, Seoul 121-742, Republic of KoreaE-mail address: [email protected]
Department of Mathematics, Kwangwoon University, Seoul 139-701, Republic of Ko-rea
E-mail address: [email protected]
Department of Applied Mathematics,Department of Applied Mathematics, PukyongNational University, Busan 48513, Re- public of Korea.
E-mail address: [email protected]
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
840 DAE SAN KIM et al 831-840
Regularization Smoothing Approximation of Fuzzy Parametric
Variational Inequality Constrained Stochastic Optimization
Heng-you Lan
Department of Mathematics, Sichuan University of Science & Engineering,
Zigong, Sichuan 643000, PR China
E-mail: [email protected]
Abstract. This work is motivated by the fact that very little is known about thefuzzy parametric variational inequalities constrained stochastic optimization prob-lems in finite dimension real numeral spaces, which are studied more difficult be-cause of the existence of random variable and fuzzified version. Based on the notionof quasi-Monte Carlo estimate and method of centres with entropic regularization,we develop a class of new regularization smoothing approximation approaches todiscretize the stochastic optimization problem with continuous random variable,and construct a centre iterative algorithm for approximating the optimal solution-s of the stochastic optimization problems. Further, we give some comprehensiveconvergence theorems of optimal solutions for the resulting optimization problem.Finally, a numerical illustration is analyzed.
Key Words and Phrases. Regularization smoothing approximation, fuzzy para-metric variational inequality, Stochastic optimization problem, centre iterative al-gorithm with quasi-Monte Carlo estimate, comprehensive convergence.
AMS Subject Classification. 49J40, 65K05, 90C30, 90C33
1 Introduction
As all we know, mathematical program with equilibrium constraints is a constrained optimizationproblem in which the essential constraints are defined by a parametric variational inequality. Thisclass of problems can be regarded as a generalization of a bilevel programming problem and ittherefore plays an important role in many fields such as transportation, communication networks,structural mechanics, economic equilibrium, multilevel game, and mathematical programming itself.See, for example, [1β7] and the reference therein. Moreover, in order to describe the uncertainties,Monica [5] considered the Bochner integrability setting, a measure space of indices and use randomfuzzy mappings, and presented random fixed point theorems with random fuzzy mappings, extensionsof the ones with random data.
In this paper, we study approximation of optimal solutions for the following fuzzy parametricvariational inequality constrained stochastic optimization problem in n-dimension real numeral setRn:
minx,y(Β·)
EΟ[f(x, y(Ο), Ο)]
s.t. x β U β Rn,y(Ο) β C(x, Ο)β¨F (x, y(Ο), Ο), z(Ο)β y(Ο)β©β₯βΌ 0, for all z(Ο) β C(x, Ο) and a.e. Ο β Ξ©,
(1.1)
where EΟ denotes the mathematical expectation with respect to the random variable Ο β Ξ© onprobability space (Ξ©,A,Ξ), f : Rn+m ΓΞ©β R and F : Rn+m ΓΞ©β Rm are two nonlinear randomfunctions, C : RnΓΞ©β 2R
m
is a multi-valued random function, β¨F (x, y(Ο), Ο), z(Ο)β y(Ο)β©β₯βΌ 0 are
fuzzy inequalities (also called fuzzy stochastic variational inequality problems, in short, V IΟ), β β₯βΌ β
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
841 Heng-you Lan 841-857
denotes the fuzzified version of ββ₯β with the linguistic interpretation βapproximately greater thanor equal toβ, and βa.e.β is the abbreviation for almost everywhereβ.
Remark 1.1. Problem (1.1) is brand new in the literature including and can be thought asa generalized version of some problems, includes a number of stochastic mathematical programwith equilibrium constraints (SMPEC), mathematical programs with fuzzy equilibrium constraints(MPFEC) mathematical program with equilibrium constraints (MPEC) and mathematical programwith complementarity constraints (MPCC) have been studied by many authors as special cases. See,for example, [1β4, 6, 8β14] and the references therein, and the following examples.
Example 1.1 If Ξ© is a singleton, then problem (1.1) reduces to the following MPFEC:
min g(x, u)s.t. x β U,
u solves V I (G(x, Β·), D(x)),
(1.2)
where g : Rn+m β R and G : Rn+m β Rm is a continuously differentiable function, D : Rn β 2Rm
isa set valued function, and u solves V I(G(x, Β·), D(x)) if and only if u β D(x) and β¨G(x, u), zβ uβ©β₯βΌ 0for all z β D(x). Problem (1.2) was introduced and studied by Hu and Liu [12] and Lan et al.[13]. Moreover, Hu and Liu [12] pointed out βalthough a powerful theory has been developed forvariational inequalities, the parameterized setting in MPEC makes these problems very difficult tosolve, and due to the vagueness involved in real world problems, the MPEC problem in a fuzzyenvironment becomes an important problem both in theory and in practiceβ, and βproblem (1.2)is a constrained optimization problem whose constraints include some fuzzy parametric variationalinequalities.
In 2013, inspired by the works of Hu and Liu [12] and other researchers, Lan et al. [13] constructedan iterative algorithm for finding a solution of a class of mathematical program problems with fuzzyparametric variational inequality constraints by using a new smoothing approach based on a versionof the method of centres with entropic regularization techniques. In fact, the tolerance approach andentropic regularization technique have been successfully proposed in solving various problems, whichare important numerical methods for solving fuzzy variational inequalities in a fuzzy environmentand nonlinear semi-infinite programming problems. See, for example, [3, 8, 14β21] and the referencestherein.
Example 1.2. Since a solution satisfying a fuzzy inequality system to a membership degreeclose to 1 is a near optimal solution to the corresponding regular inequality problem [22], if y(Ο) β‘ vfor all Ο β Ξ© and the degree for the fuzzy inequalities in (1.1) is close to 1, then problem (1.1) isequivalent to the following SMPEC:
min EΟ[f(x, v, Ο)]s.t. x β U, Ο β Ξ©
v solves V I(F (x, Β·, Ο), C(x, Ο)),(1.3)
where V I(F (x, Β·, Ο), C(x, Ο)) denotes the variational inequality problem defined by the pair (F (x, Β·,Ο), C(x, Ο)) for all x β Rn and Ο β Ξ©. In 2003, Lin et al. [9] considered problem (1.3) and showedthat SMPEC can be thought as a generalization of MPEC, and proposed a smoothing implicitprogramming method to establish a comprehensive convergence theory for the lower-level wait-and-see model. Further, there are many stochastic formulations of MPEC proposed in the recentdiscussions. For related works, we refer readers to [1, 3, 6, 8, 10, 11]. However, there has been verylittle study on applications of these theories and approaches to (1.1).
Over years of development, optimization approaches have become one of the most promisingtechniques for engineering applications and an MPEC is a hard problem because its constraintsfail to satisfy a standard constraint qualification at any feasible point [23]. However, since theexistence of the random variable Ο and the fuzzified version β β₯
βΌ β mean that (1.1) involves multiplecomplementarity-type constraints, it is more difficult to solve problem (1.1) than to solve an ordinaryMPCC, MPEC, MPFEC or SMPEC generally. Therefore, our focus in this paper is to develop aclass of new regularization smoothing approximation approaches to define some parameters of theobjective function fuzzy yielded by fuzzy constraints, and consider the approximation-solvability foran equivalent stochastic parametric optimization problem of problem (1.1).
2
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
842 Heng-you Lan 841-857
Motivated and inspired by the above works, we shall give some preliminaries needed throughoutthe whole paper in Section 2. Specially, by using the notion of tolerance approach and the fuzzy settheory, we show that the fuzzy parametric variational inequality constrained stochastic optimizationproblem (1.1) and a fuzzy complementarity constrained optimization problem can be converted to aregular nonlinear parametric optimization problem. In Sections 3, we will construct a centre iterativealgorithm and develop a class of new regularization smoothing approximation approach for solvingthe stochastic fuzzy optimization based on quasi-Monte Carlo estimate, and establish comprehensiveconvergence theorems of the solution. We also report some numerical simulation analysis results inSection 4.
2 Preliminaries
Throughout in this paper, we assumption that (Ξ©,A,Ξ) is a complete Ο-finite measure space andthe probability measure Ξ of our considered space (Ξ©,A,Ξ) is non-atomic. Let B(Rm) be the classof Borel Ο-fields in Rm and P (U) denote the power set of a vector space U .
Definition 2.1. (i) A function y : Ξ© β Rm is said to be measurable, if for any B β B(Rm),Ο β Ξ© : y(Ο) β B β A).
(ii) The multi-valued function Ξ¨ : Ξ©β P (U) is called said to be measurable, if for any B β B(U),Ξ¨β1(B) = Ο β Ξ© : Ξ¨(Ο) β©B = β β A.
(iii) A multi-valued random function Ξ¦ : Rn Γ Ξ© β 2Rm
is said to be measurable, if for anyx β Rn, Ξ¦(x, Β·) is measurable.
(iv) F : Rn+m Γ Ξ© β Rm is called a random and continuously differentiable function, whenF (x, z, Ο) = ΞΆ(Ο) is measurable for any x β Rn and z β Rm, and F (Β·, Β·, Ο) is continuously differen-tiable for all Ο β Ξ©.
Definition 2.2. Let C,Cβ : Rn Γ Ξ©β 2Rm
be two multi-valued random function. Then(i) C(x, Ο) is said to be convex cone, if C(x, Β·) is convex cone for every x β Rn, that is,
Ξ±y(Β·) + Ξ²w(Β·) β C(x, Β·) for any positive scalarsΞ±, Ξ² and all measurable function y(Β·), w(Β·) β C(x, Β·);
(ii) Cβ(x, Ο) is called polar (dual) cone of C(x, Ο) β Rm for x β Rn and Ο β Ξ©, if Cβ(x, Β·) ispolar (dual) cone for every x β Rn, i.e.
β¨ΞΎ, Ξ½(Β·)β© β₯ 0 βΞΎ β Rm and for each measurable function Ξ½(Β·) β C(x, Β·).
In other words, the polar (dual) cone Cβ(x, Ο) can be expressed as follows:
Cβ(x, Ο) = ΞΎ β Rmβ₯ β¨ΞΎ, Ξ½(Ο)β© β₯ 0 βΞ½(Ο) β C(x, Ο) .
Definition 2.3. Let Ο > 0 and Ο > 0 be constants. a function M : Rn β Rm is said to beHolder continuous on K β Rm with order Ο and Holder constant Ο if
β₯M(u)βM(v)β₯ β€ Οβ₯uβ vβ₯Ο, βu, v β K.
holds for all u and v in K.Remark 2.1. If Ο = 1, then the definition of Holder continuity reduces to definition of Lipschitz
continuity. We note that for two different positive numbers Ο and Οβ², Holder continuous functionswith order Ο and those with order Οβ² constitute different subclasses. For example, the functionM(u) :=
ββ₯uβ₯ for all u β K β Rm is Holder continuous with order Ο = 1
2 , but not Lipschitzcontinuous.
In the sequel, we give some preparations needed later to approximating the optimal solutionsof problem (1.1). First, we propose discretization of the stochastic objective function in (1.1) withcontinuous random variable.
3
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
843 Heng-you Lan 841-857
Lemma 2.1. Let ΞΆ : Ξ© β [0,+β) be the continuous probability density function of Ο. Thenthe objective function in (1.1) can be represented as
EΟ(f(x, y(Ο), Ο) =1
L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο), (2.1)
where Ξ©L := Ο1, Ο2, Β· Β· Β· , ΟL is a uniformly distributed sample set from Ξ©.Proof. Let Ξ© be a sample space, which is usually denoted using set notation, and the possible
outcomes are listed as elements in the set. If Ξ© is unbounded, under some mild conditions, we canapproximate the problem by a sequence of programs with bounded sampling spaces (see [11]) formore details. In the sequel, let Ξ© be a bounded rectangle. In particular, without loss of generality,we assume that Ξ© = [0, 1]ΞΊ. Let ΞΆ : Ξ© β [0,+β) be the continuous probability density function ofΟ. Then the objective function in (1.1) can be represented as
EΟ(f(x, y(Ο), Ο) =
β«Ξ©
f(x, y(Ο), Ο)ΞΆ(Ο)dΟ.
Based on quasi-Monte Carlo method in [24], now we estimate numerical integration to the ob-jective function in problem (1.1). Roughly speaking, given a function Ο : Ξ© β R, the quasi-Monte Carlo estimate for EΟ[Ο(Ο)] is obtained by taking a uniformly distributed sample set Ξ©L :=Ο1, Ο2, Β· Β· Β· , ΟL from Ξ© and letting EΟ[Ο(Ο)] β 1
L
βΟβΞ©L
Ο(Ο). This implies that (2.1) holds. 2
Next, we consider the random membership functions of each fuzzy stochastic inequality andstochastic fuzzy objective yielded by the fuzzy constraints in (1.1).
Let the membership function for each fuzzy stochastic inequality β¨F (x, y(Ο), Ο), z β y(Ο)β©β₯βΌ 0 asfollows: for all x β Rn and any z β C(x, Ο),
¡Ωz(x, y(Ο), Ο) =
1, if β¨F (x, y(Ο), Ο), z β y(Ο)β© β₯ 0,Β΅z(β¨F (x, y(Ο), Ο), z β y(Ο)β©), if β¨F (x, y(Ο), Ο), z β y(Ο)β© β [βtz, 0),0, if β¨F (x, y(Ο), Ο), z β y(Ο)β© < βtz,
(2.2)
specify the degree to which the regular inequality β¨F (x, y(Ο), Ο), zβy(Ο)β© β₯ 0 is satisfied, where Ξ©z isa fuzzy set actually determined by the fuzzy stochastic inequality in Rn+mΓΞ©, tz β₯ 0 is the tolerancelevel which can be tolerated by decision makers in the accomplishment of the fuzzy stochasticinequality β¨F (x, y(Ο), Ο), zβ y(Ο)β©β₯βΌ 0. We usually assume that Β΅z(β¨F (x, y(Ο), Ο), zβ y(Ο)β©) β [0, 1]and it is continuous and strictly increasing over [βtz, 0). Fig. 1 shows different shapes of suchmembership functions.
Β΅z(γF(x, y(Ο), Ο), z β y(Ο)γ)
1
βtz0
Ξ±
Β΅β1
Ξ©z
(Ξ±) γF(x, y(Ο), Ο), z β y(Ο)γ
¡Ωz
(x, y(Ο), Ο)
Figure 1: The membership function ¡Ωz(x, y(Ο), Ο).
4
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
844 Heng-you Lan 841-857
Similarly, the random membership function of the objective, Β΅S0(x, y(Ο), Ο), is defined as follows:
Β΅S0(x, y(Ο), Ο) =
1, if EΟ[f(x, y(Ο), Ο)] < f,
Β΅0(EΟ[f(x, y(Ο), Ο)]), if EΟ[f(x, y(Ο), Ο)] β [f, f),
0, if EΟ[f(x, y(Ο), Ο)] β₯ f ,(2.3)
where f and f are two parameters defined as follows:
f = min EΟ[f(x, y(Ο), Ο)]s.t. x β U, Ο β Ξ©,
β¨F (x, y(Ο), Ο), z β y(Ο)β© β₯ 0, βz β C(x, Ο)(2.4)
andf = min EΟ[f(x, y(Ο), Ο)]
s.t. x β U, Ο β Ξ©,β¨F (x, y(Ο), Ο), z β y(Ο)β© β₯ βtz, βz β C(x, Ο).
(2.5)
By [22, 25], one can know that studying such a problem (1.1) is related to finding βalmostoptimalβ solutions for a general convex minimization problem (see also [13, 14, 17]). Thus, weextend the idea and have the following result.
Lemma 2.2. Let C(x, Ο) be a convex cone for all x β Rn and Ο β Ξ©. Then the problem V IΟ,i.e., finding y(Ο) β C(x, Ο) such that
β¨F (x, y(Ο), Ο), z(Ο)β y(Ο)β©β₯βΌ 0, βz(Ο) β C(x, Ο), (2.6)
is equivalent to the fuzzy complementarity problem of finding y(Ο) β Rm such that
y(Ο) β C(x, Ο), β¨F (x, y(Ο), Ο), y(Ο)β©=βΌ 0, F (x, y(Ο), Ο) ββΌ C
β(x, Ο), (2.7)
where β=βΌβ denotes the fuzzified version of β=β with the linguistic interpretation βapproximatelyequal toβ, βββΌβ denotes the fuzzified version of βββ with the linguistic interpretation βapproximatelyinβ and Cβ(x, Ο) is a polar (dual) cone of C(x, Ο) β Rm for all x β Rn and Ο β Ξ©.
Proof. We start by showing that problem (2.7)β problem (2.6). For any x β Rn and Ο β Ξ©,suppose that yβ(Ο) is a solution of problem (2.7), then we have
β¨F (x, yβ(Ο), Ο), yβ(Ο)β©=βΌ 0 (2.8)
andβ¨F (x, yβ(Ο), Ο), Ο (Ο)β©β₯βΌ 0, βΟ (Ο) β C(x, Ο). (2.9)
Combining (2.8) and (2.9), we have β¨F (x, yβ(Ο), Ο), Ο (Ο)β yβ(Ο)β©β₯βΌ 0 for all Ο (Ο) β C(x, Ο). Thus,yβ(Ο) is also a solution of problem (2.6) for all Ο β Ξ©.
Now we show that problem (2.6) β problem (2.7). Let yβ(Ο) be the solution of problem (2.6)with the membership degree Ξ± β [0, 1] for every Ο β Ξ©. According to the tolerance approach [15, 21],by (2.2), we have
β¨F (x, yβ(Ο), Ο), Ο (Ο)β yβ(Ο)β© β₯ Β΅β1
Ξ©z(Ξ±) β₯ βtz, βΟ (Ο) β C(x, Ο), (2.10)
where for all Ο (Ο) β C(x, Ο) and any Ο β Ξ©, Β΅β1
Ωzis the inverse functions of ¡Ωz
(x, Β·, Ο) and tz > 0 is
the tolerance level which a decision maker can tolerate in the accomplishment of the fuzzy inequalityβ¨F (x, y(Ο), Ο), Ο (Ο)β y(Ο)β©β₯βΌ 0. Suppose that for tz β₯ 0 and tz < 0, either
β¨F (x, yβ(Ο), Ο), yβ(Ο)β© > tz or β¨F (x, yβ(Ο), Ο), yβ(Ο)β© < tz
is true. For any x β Rn and each Ο β Ξ©, since C(x, Ο) is a convex cone, we have β¨F (x, yβ(Ο), Ο), yβ(Ο)β© β₯βtzΞ»β1 , tz β₯ 0 when Ο (Ο) = Ξ»yβ(Ο) with Ξ» > 1, and β¨F (x, yβ(Ο), Ο), yβ(Ο)β© β€ tz, when Ο (Ο) = 0. If
5
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
845 Heng-you Lan 841-857
β¨F (x, yβ(Ο), Ο), yβ(Ο)β© > tz for tz β₯ 0, This leads to a contradiction. If β¨F (x, yβ(Ο), Ο), yβ(Ο)β© < tzfor tz < 0, then tz β€ tz. There lies a contradiction. Therefore,
tz β€ β¨F (x, yβ(Ο), Ο), yβ(Ο)β© β€ tz,
for tz β₯ 0 and tz < 0, that is, β¨F (x, yβ(Ο), Ο), yβ(Ο)β©=βΌ 0. Furthermore, from (2.10), we have for anyΟ (Ο) β C(x, Ο),
β¨F (x, yβ(Ο), Ο), Ο (Ο)β© β₯ β¨F (x, yβ(Ο), Ο), yβ(Ο)β© β tz β₯ tz β tz.
This implies that β¨F (x, yβ(Ο), Ο), Ο (Ο)β©β₯βΌ 0 for all Ο (Ο) β C(x, Ο). Hence, we have F (x, yβ(Ο), Ο) ββΌ
Cβ(x, Ο). Therefore, yβ(Ο) for any Ο β Ξ© is also a solution of problem (2.7). This completes theproof. 2
Based on Lemma 2.2 and the work of [21], we have the following results.Lemma 2.3. Let C(x, Ο) is a convex cone with polar (dual) cone Cβ(x, Ο) for all (x, Ο) β RnΓΞ©
and Ξ± be a new variable. Then the stochastic optimization problem (1.1) can eventually be expressedas the following regular semi-infinite optimization problem with finitely many variables x, y(Ο), Οand Ξ±:
max Ξ±s.t. Β΅S0
(x, y(Ο), Ο) β₯ Ξ±,¡Ωz
(x, y(Ο), Ο) β₯ Ξ±,(x, y(Ο), Ο) β S,0 β€ Ξ± β€ 1,
(2.11)
where S = (x, y(Ο), Ο) β Rn+m Γ Ξ©|x β U, Ο β Ξ©, F (x, y(Ο), Ο) ββΌ C
β(x, Ο), and the randommembership functions Β΅S0
and ¡Ωzare the same as in (2.3) and (2.2), respectively.
Proof. It follows from Lemma 2.1 that, in order to find a solution to the stochastic optimizationproblem (1.1) with C(x, Ο) being a convex cone for (x, Ο) β RnΓΞ©, we should consider the followingstochastic fuzzy complementarity constrained optimization problem:
min EΟ[f(x, y(Ο), Ο)]s.t. x β U, Ο β Ξ©, y(Ο) β C(x, Ο),
β¨F (x, y(Ο), Ο), y(Ο)β©=βΌ 0,F (x, y(Ο), Ο) β
βΌ Cβ(x, Ο).
(2.12)
Since a global minimum is often required for practical problems, by the work of [21] and the descrip-tion of the fuzzy stochastic inequalities (2.6), and a solution of problem (2.12) can be taken as thesolution with the highest membership in the fuzzy decision set and eventually obtained by solvingthe following regular nonlinear parametric optimization problem:
max(x,y(Ο),Ο)βS
minΒ΅S0
(x, y(Ο), Ο), ¡Ωz(x, y(Ο), Ο)
,
which implies that the result holds for the new variable Ξ±. 2
Remark 2.2. Moreover, if the membership functions ¡S0and ¡Ωz
in Lemma 2.3 are invertible,then from (2.11), we get
max Ξ±
s.t. (x, y(Ο), Ο)0 β₯ Β΅β1
S0(Ξ±),
(x, y(Ο), Ο)Cβ β₯ Β΅β1
Ξ©z(Ξ±),
(x, y(Ο), Ο) β S,0 β€ Ξ± β€ 1,
(2.13)
where (x, y(Ο), Ο)0 and (x, y(Ο), Ο)Cβ can be followed by (2.3) and (2.2), respectively.
6
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
846 Heng-you Lan 841-857
3 Regularization smoothing approximation algorithms
In this section, based on the βmethod of centresβ with entropic regularization, we develop a classof new smoothing approach and construct a centre iterative algorithm for solving the stochasticfuzzy optimization (1.1), and give the solution theorems.
In the sequel, we first give the following assumption (HC) for convenience: Define
C(x, Ο) := Ο (Ο) β Rm|D(x)Ο (Ο) β₯ 0, D(x) = [di(x)] is an l Γmmatrix, di(x) is the ith row ofD(x), βi = 1, 2, Β· Β· Β· , l. (3.1)
STEP I. The random membership function of the fuzzy stochastic inequalities in (1.1) can bespecified under condition (HC).
From (3.1), it is easy to see that the multi-valued operator C(x, Ο) is a convex cone for any(x, Ο) β Rn Γ Ξ©, and can be shown that F (x, y(Ο), Ο) β Cβ(x, Ο) if and only if there exists anonnegative random vector r(Ο) = (r1(Ο), r2(Ο), Β· Β· Β· , rl(Ο))T β Rl such that
F (x, y(Ο), Ο) = r1(Ο)dT1 (x) + r2(Ο)dT2 (x) + Β· Β· Β·+ rl(Ο)dTl (x) = DT (x)r(Ο), (3.2)
that is, for every i = 1, 2, Β· Β· Β· , l, dβ²i(x)F (x, y(Ο), Ο) β₯ 0, where dβ²i(x) is normal to di(x) (see [17, 26]).It follows that the fuzzy stochastic optimization problem (2.12) can be rewritten as the following
generalized stochastic optimization problem with fuzzy stochastic inequality constraints:
min EΟ[f(x, y(Ο), Ο)]s.t. x β U, Ο β Ξ©,
di(x)y(Ο) β₯ 0, i = 1, 2, Β· Β· Β· , l,β¨F (x, y(Ο), Ο), y(Ο)β©β₯βΌ 0,β¨βF (x, y(Ο), Ο), y(Ο)β©β₯βΌ 0,dβ²i(x)F (x, y(Ο), Ο) β₯
βΌ 0, i = 1, 2, Β· Β· Β· , l,
(3.3)
and each fuzzy stochastic inequality in (3.3) can be represented by a fuzzy set Sj (i.e., represent
of Ξ©z in (2.11) or (2.13)) with corresponding random membership function Β΅Sj(x, y(Ο), Ο) for j =
1, 2, Β· Β· Β· , l + 2. To specify the membership functions Β΅Sj, j = 1, 2, Β· Β· Β· , l + 2, similar treatment to
(2.2), we define the membership functions as follows:
Β΅S1(x, y(Ο), Ο) =
1, if β¨F (x, y(Ο), Ο), y(Ο)β© β₯ 0,Β΅1(β¨F (x, y(Ο), Ο), y(Ο)β©), if β¨F (x, y(Ο), Ο), y(Ο)β© β [βt1, 0),0, if β¨F (x, y(Ο), Ο), y(Ο)β© < βt1,
Β΅S2(x, y(Ο), Ο) =
1, if β¨βF (x, y(Ο), Ο), y(Ο)β© β₯ 0,Β΅2(β¨βF (x, y(Ο), Ο), y(Ο)β©), if β¨βF (x, y(Ο), Ο), y(Ο)β© β [βt2, 0),0, if β¨βF (x, y(Ο), Ο), y(Ο)β© < βt2,
Β΅Si+2(x, y(Ο), Ο) =
1, if dβ²iF (x, y(Ο), Ο) β₯ 0,Β΅i+2(dβ²iF (x, y(Ο), Ο)), if dβ²iF (x, y(Ο), Ο) β [βti+2, 0),0, if dβ²iF (x, y(Ο), Ο) < βti+2,
(3.4)
where ti+2 β₯ 0 for i = 1, 2, Β· Β· Β· , l, is the tolerance level which one decision maker can tolerate in theaccomplishment of the fuzzy stochastic inequalities in (3.3).
STEP II. Discrete approximation of problem (1.1) with condition (HC) need to be given.By Lemma 2.1 and STEP I, we have the following problem as an appropriate discrete approxi-
mation of problem (3.3):
min 1L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο)
s.t. x β U β Rn, Ο β Ξ©L,di(x)y(Ο) β₯ 0, i = 1, 2, Β· Β· Β· , l,β¨F (x, y(Ο), Ο), y(Ο)β©β₯βΌ 0,β¨βF (x, y(Ο), Ο), y(Ο)β©β₯βΌ 0,dβ²i(x)F (x, y(Ο), Ο) β₯
βΌ 0, i = 1, 2, Β· Β· Β· , l.
(3.5)
7
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
847 Heng-you Lan 841-857
We note that the sample set Ξ©L is chosen to be asymptotically dense in Ξ©. Especially, it followsfrom (2.4), (2.5) and (3.5) that the appropriate discrete approximation of two parameter f and fcan be shown as follows, respectively:
f = min 1L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο)
s.t. x β U β Rn, Ο β Ξ©L,di(x)y(Ο) β₯ 0, i = 1, 2, Β· Β· Β· , l,β¨F (x, y(Ο), Ο), y(Ο)β© β₯ 0,β¨βF (x, y(Ο), Ο), y(Ο)β© β₯ 0,dβ²i(x)F (x, y(Ο), Ο) β₯ 0, i = 1, 2, Β· Β· Β· , l
(3.6)
andf = min 1
L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο)
s.t. x β U β Rn, Ο β Ξ©L,di(x)y(Ο) β₯ 0, i = 1, 2, Β· Β· Β· , l,β¨F (x, y(Ο), Ο), y(Ο)β© β₯ βt1,β¨βF (x, y(Ο), Ο), y(Ο)β© β₯ βt2,dβ²i(x)F (x, y(Ο), Ο) β₯ βti+2, i = 1, 2, Β· Β· Β· , l.
(3.7)
STEP III. A new centre iterative method for solving problem (3.8) should be adopt.It follows from (2.13), (2.3) and (3.4)-(3.7) that an optimal solution of the stochastic optimization
problem (1.1) can be obtained by approximating for the following stochastic parametric optimizationproblem:
max Ξ±
s.t. Β΅β1
S0(Ξ±)β 1
L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο) β₯ 0,
Β΅β1
S1(Ξ±)β β¨F (x, y(Ο), Ο), y(Ο)β© β₯ 0,
Β΅β1
S2(Ξ±) + β¨F (x, y(Ο), Ο), y(Ο)β© β₯ 0,
βΒ΅β1
Sj(Ξ±) + dβ²jβ2F (x, y(Ο), Ο) β₯ 0, j = 3, 4, Β· Β· Β· , l + 2,
di(x)y(Ο) β₯ 0, i = 1, 2, Β· Β· Β· , l,0 β€ Ξ± β€ 1, x β U, Ο β Ξ©L.
(3.8)
It is interested in developing an efficient algorithm to solve (3.8) based on a framework of centreiterations. This iterative approach can be traced back to Huardβs work [28]. The basic concepts areeasy to understand and very adaptive to new developments. To describe the approach, we denotethe feasible domain of (3.8) by a set V and define some terminologies. A general assumption forthis approach is that V is bounded and convex, and the interior of V is nonempty.
Definition 3.1 For any given point (x, y(Ο), Ο, Ξ±) in the convex domain V , we define the βdis-tance L of (x, y(Ο), Ο, Ξ±) to the boundary of V β by a continuous function
L((x, y(Ο), Ο, Ξ±), V ) = mini=1,2,Β·Β·Β· ,l
j=3,4,Β·Β·Β· ,l+2
Ξ±, 1β Ξ±, Β΅β1
S0(Ξ±)β 1
L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο),
Β΅β1
S1(Ξ±)β β¨F (x, y(Ο), Ο), y(Ο)β©, Β΅β1
S2(Ξ±) + β¨F (x, y(Ο), Ο), y(Ο)β©,
βΒ΅β1
Sj(Ξ±) + dβ²jβ2F (x, y(Ο), Ο), di(x)y(Ο)
.
Definition 3.2 Let a distance function L((x, y(Ο), Ο, Ξ±), V ) be defined on a convex domain V .Then a point (x, y(Ο), Ο, Ξ±) β V is called the βcentre of V β, if it maximizes the distance functionL((x, y(Ο), Ο, Ξ±), V ), i.e.,
(x, y(Ο), Ο, Ξ±) : L((x, y(Ο), Ο, Ξ±), V ) = max L((x, y(Ο), Ο, Ξ±), V )| (x, y(Ο), Ο, Ξ±) β V .
Thus, a new centre iterative method for problem (3.8) could be described as follows.
8
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
848 Heng-you Lan 841-857
Algorithm 3.1. Step 1. Taking a point (xk, yk(Ο), Ο, Ξ±k) in V , then we consider the distanceL in convex domain Wk = V β© (x, y(Ο), Ο, Ξ±)|Ξ± β₯ Ξ±k.
Step 2. Solving the maximal problem maxL((x, y(Ο), Ο, Ξ±),Wk)| (x, y(Ο), Ο, Ξ±) β Wk anddenoting the new iterative point (xk+1, yk+1(Ο), Ο, Ξ±k+1) as a centre of Wk, then we have
(xk+1, yk+1(Ο), Ο, Ξ±k+1) :
L((xk+1, yk+1(Ο), Ο, Ξ±k+1),Wk) = maxL((x, y(Ο), Ο, Ξ±),Wk)| (x, y(Ο), Ο, Ξ±) βWk,
where
L((x, y(Ο), Ο, Ξ±),Wk) = mini=1,2,Β·Β·Β· ,l
j=3,4,Β·Β·Β· ,l+2
Ξ±β Ξ±k, Ξ±, 1β Ξ±, Β΅β1
S0(Ξ±)β 1
L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο),
Β΅β1
S1(Ξ±)β β¨F (x, y(Ο), Ο), y(Ο)β©, Β΅β1
S2(Ξ±) + β¨F (x, y(Ο), Ο), y(Ο)β©,
βΒ΅β1
Sj(Ξ±) + dβ²jβ2F (x, y(Ο), Ο), di(x)y(Ο)
is the distance function defined on the convex domain Wk.
Step 3. Start working again with (xk+1, yk+1(Ο), Ο, Ξ±k+1) instead of (xk, yk(Ο), Ο, Ξ±k) and go toStep 1.
It follows from the properties introduced in [28, Lemma 2.2] and Algorithm 3.1 that the majorcomputational work lies in the determination of the centres required, i.e., at the kth iteration, thefollowing βmin-max problemβ should be solved:
β minx,y(Ο),Ο,Ξ±
L((x, y(Ο), Ο, Ξ±),Wk) = minx,y(Ο),Ο,Ξ±
max i=1,2,Β·Β·Β· ,lj=3,4,Β·Β·Β· ,l+2
Ξ±k β Ξ±, βΞ±, Ξ±β 1,
1L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο)β Β΅β1
S0(Ξ±),
β¨F (x, y(Ο), Ο), y(Ο)β© β Β΅β1
S1(Ξ±),
ββ¨F (x, y(Ο), Ο), y(Ο)β© β Β΅β1
S2(Ξ±),
βdβ²jβ2F (x, y(Ο), Ο) + Β΅β1
Sj(Ξ±), βdi(x)y(Ο)
.
(3.9)
STEP IV A class of new regularization smoothing approximation algorithms is developed undercondition (HC).
Since the maximal membership function (see [12]) in the βmin-maxβ problem (3.9) is non-differentiability, it is easy to see that one major difficulty encountered is to develop a class ofnew smoothing approximation methods, which are based on the notion of newly proposed βentropicregularization procedureβ (see [18]).
Algorithm 3.2.Step 1. Set k = 0, give the initial iterate (x0, y0(Ο), Ο, Ξ±0) which is an interior point of V defined
by (3.8), a sufficiently small constant Ο΅ > 0, and an upper bound Q which is the maximum numberof unconstrained minimizations to be performed.
Step 2. Starting from (xk, yk(Ο), Ο, Ξ±k), apply a standard quasi-Newton line search of MAT-LAB software to solve the unconstrained smooth convex program (3.6), (3.7) and the followingunconstrained smooth convex program:
β minx,y(Ο),Ο,Ξ±
LΞ³((x, y(Ο), Ο, Ξ±),Wk)
= 1Ξ³ ln
exp[Ξ³(Ξ±k β Ξ±)] + exp[Ξ³(βΞ±)] + exp[Ξ³(Ξ±β 1)]
+ exp[Ξ³( 1L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο)β Β΅β1
S0(Ξ±))]
+ exp[Ξ³(β¨F (x, y(Ο), Ο), y(Ο)β© β Β΅β1
S1(Ξ±))]
+ exp[Ξ³(ββ¨F (x, y(Ο), Ο), y(Ο)β© β Β΅β1
S2(Ξ±))]
+l+2βj=3
exp[Ξ³(βdβ²jβ2F (x, y(Ο), Ο) + Β΅β1
Sj(Ξ±))]
+lβ
i=1
exp[Ξ³(βdi(x)y)]
(3.10)
9
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
849 Heng-you Lan 841-857
with a sufficiently large Ξ³. Denote its solution by (xk+1, yk+1(Ο), Ο, Ξ±k+1) in the light of Algorithm3.1.
Step 3. If k > 1 and β₯(xk+1, yk+1(Ο), Ο, Ξ±k+1) β (xk, yk(Ο), Ο, Ξ±k)β₯2 β€ Ο΅, then the computationterminates with (xk+1, yk+1(Ο), Ο, Ξ±k+1) as the solution. If k > Q, then the computation terminateswith a failure.
Step 4. k β k + 1 and go to Step 2.From Algorithm 3.2, it follows that minx,y(Ο),Ο,Ξ± LΞ³((x, y(Ο), Ο, Ξ±),Wk) provides a centre of Wk,
as Ξ³ ββ. By using a moderately large Ξ³, we can obtain an accurate approximation. Also because ofthe special βlog-exponentialβ form of LΞ³((x, y(Ο), Ο, Ξ±),Wk), we can avoid most overflow problems incomputation. Moreover, since problem (3.10) is an unconstrained, smooth, and convex optimizationprogram, the commonly used solution methods, such as the quasi-Newton line search of MATLABsoftware, can be readily applied.
Remark 3.1. We note that Algorithm 3.1 appears in Step 2 of Algorithm 3.2. It is the fuzzyconstraints in (1.1) that yields a fuzzy objective. Hence, a class of new and interesting regularizationsmoothing approximation approaches must be chosen to define two parameters in (3.6) and (3.7), andto employ for solving problem (3.10) which is equivalent to the stochastic parametric optimizationproblem (3.8).
STEP V Comprehensive convergence theorems based on Algorithm 3.2 should be proved.In the sequel, we first give the following lemmas and results.Lemma 3.1. Let the function Ο : Ξ©β R be continuous. Then we have
limLββ
1
L
βΟβΞ©L
Ο(Ο)ΞΆ(Ο) =
β«Ξ©
Ο(Ο)ΞΆ(Ο)dΟ.
Proof. Taking N = L, Is = Ξ©, J = Ξ©L, xi = Οi (i = 1, 2, Β· Β· Β· ) and f = ΟΞΆ, then from the results(2.2) and (2.3) given in Chapter 2 of [24, pp. 13-14], the result holds. This completes the proof. 2
Remark 3.2. By Lemma 3.1, we know immediately that
limLββ
1
L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο) =
β«Ξ©
f(x, y(Ο), Ο)ΞΆ(Ο)dΟ (3.11)
and particularly,
limLββ
1
L
βΟβΞ©L
ΞΆ(Ο) =
β«Ξ©
ΞΆ(Ο)dΟ = 1. (3.12)
Lemma 3.2. If Ο(x) is continuous, strictly increasing and linear over a convex set U in Rn,then its inverse Οβ1 is linear.
Theorem 3.1. Suppose that condition (HC) holds, the set U β Rn is nonempty and bounded,F : Rn+m Γ Ξ©β Rm is continuously differentiable, and f : Rn+m Γ Ξ©β R is Holder continuous in(x, y(Β·)) on U Γ Rm with order Ο > 0 and Holder constant Ο(Ο) > 0 for all Ο β Ξ©L satisfyingβ«
Ξ©
Ο(Ο)dΞΆ(Ο) < +β.
Then(i) problem (3.6) has at least one optimal solution when L is large enough;(ii) (xβ, yβ(Β·)) is an optimal solution of problem (2.4) when xβ is an accumulation point of the
sequence xL and yβ(Β·) is defined by
yβ(Ο) := maxi=1,2,Β·Β·Β· ,l
βF (xβ, 0, Ο),βdβ²i(xβ)F (xβ, 0, Ο), 0, Ο β Ξ©. (3.13)
Proof. (i) Let FL be the feasible region of problem (3.6). It is not difficult to see that FL is anonempty and closed set and the objective function of problem (3.6) is bounded below on FL. Thus,
10
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
850 Heng-you Lan 841-857
there exists a sequence (xk, yk(Ο))ΟβΞ©L β FL such that
limkββ
1
L
βΟβΞ©L
f(xk, yk(Ο), Ο)ΞΆ(Ο)
= inf(x,y(Ο),Ο)ΟβΞ©L
1
L
βΟβΞ©L
f(x, y(Ο), Ο)ΞΆ(Ο). (3.14)
It follows from the boundedness of U and the Holder continuity of f that the sequence xk and thefunction f are bounded.
On the other hand, noting that (xk, yk(Ο))ΟβΞ©L β FL for every k, we have
0 β€ yk(Ο) β₯ 1
L
βΟβΞ©L
F (xk, yk(Ο), Ο)ΞΆ(Ο) β₯ 0, (3.15)
where the symbol β₯ means the two vectors are perpendicular to each other. Assume that thesequence yk(Ο)ΟβΞ©L is unbounded. Taking a subsequence if necessary, let
limkββΟβΞ©L
β₯yk(Ο)β₯ = +β, limkββΟβΞ©L
yk(Ο)
β₯yk(Ο)β₯= y(Ο), β₯y(Ο)β₯ = 1. (3.16)
Then, for all Ο β Ξ©L, dividing (3.15) by β₯yk(Ο)β₯ and letting k β +β, we have for any x β U ,
0 β€ y(Ο) β₯ 1
L
βΟβΞ©L
F (x, y(Ο), Ο)ΞΆ(Ο) β₯ 0.
This contradicts (3.16) by the continuous differentiability of F , and so yk(Ο) is bounded for eachΟ β Ξ©L with ΞΆ(Ο) > 0. For any Ο β Ξ©L with ΞΆ(Ο) = 0, we redefine yk(Ο) by
yk(Ο) := maxi=1,2,Β·Β·Β· ,l
βF (xk, 0, Ο),βdβ²i(xk)F (xk, 0, Ο), 0.
Hence, the sequence (xk, yk(Ο))ΟβΞ©L is bounded and (3.14) remains valid. Therefore, the closenessof FL implies that any accumulation point of (xk, yk(Ο))ΟβΞ©L
must be an optimal solution ofproblem (3.6).
(ii) By the assumptions, the sequence xL contains a subsequence converging to xβ. Withoutloss of generality, we suppose limLββ xL = xβ.
Firstly, we prove that (xβ, yβ(Β·)) is feasible to problem (3.6). To this end, we define
yL(Ο) := maxi=1,2,Β·Β·Β· ,l
βF (xL, 0, Ο),βdβ²i(xL)F (xL, 0, Ο), 0
, Ο β Ξ©. (3.17)
It is obvious that (xβ, yL(Ο))ΟβΞ©Lis feasible to problem (3.6) for every L. Since F (xβ, yβ(Ο), Ο) β₯ 0
by the definition (3.13), it is sufficient to show that
(yβ(Ο))TF (xβ, yβ(Ο), Ο) = 0, Ο β Ξ©. (3.18)
Let Ο β Ξ© be fixed. Since the sample set Ξ©L is chosen to be asymptotically dense in Ξ©, there existsa sequence ΟL of samples such that ΟL β Ξ©L for each L and limLββ ΟL = Ο. Thus, we obtain
(yL(ΟL))TF (xL, yL(ΟL), ΟL) = 0, L = 1, 2, Β· Β· Β· .
Letting Lβ +β and taking the continuity of the functions F (x, y(Β·), Β·) on the compact set Ξ© intoaccount, we have
(yβ(Ο))TF (xβ, yβ(Ο), Ο) = 0.
By the arbitrariness of Ο β Ξ©, now we know that (3.18) immediately holds. This completes theproof of the feasibility of (xβ, yβ(Β·)) in (3.6).
11
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
851 Heng-you Lan 841-857
Next, let (x, y(Β·)) be an arbitrary feasible solution of (3.6). It follows from the results of (i) andobvious that (x, y(Ο), Ο)ΟβΞ©L
is feasible to problem (3.6) for any L. Moreover, from the Hocontinuityof f , we have
1
L
βΟβΞ©L
[f(xL, yL(Ο), Ο)β f(xL, yL(Ο), Ο)
]ΞΆ(Ο)
=1
L
βΟβΞ©L
[f(xL, yL(Ο), Ο)β f(xL, yL(Ο), Ο)
]ΞΆ(Ο)
β€ 1
L
βΟβΞ©L
[β₯yL(Ο)β yL(Ο)β₯Ο(Ο)
]Β· ΞΆ(Ο)β 0 as k ββ, w.p.1.
which along with Lemma 3.1 yields
limLββ
1
L
βΟβΞ©L
f(xk, yk(Ο), Ο) = limLββ
1
L
βΟβΞ©L
f(xβ, yβ(Ο), Ο) = EΟ[f(xβ, yβ(Ο), Ο)] w.p.1,
which indicates that (xβ, yβ(Β·)) is an optimal solution of problem (1.1) with probability one andthe feasibility of (xL, yL(Ο), Ο)ΟβΞ©L in (3.6) that (xL, yL(Ο), Ο)ΟβΞ©L is also an optimal solution ofproblem (3.6). Thus, since f is Holder continuous in (x, y(Β·)) on U Γ Rm, we obtain
1
L
βΟβΞ©L
[f(xβ, yβ(Ο), Ο)β f(x, y(Ο), Ο)
]ΞΆ(Ο)
β€ 1
L
βΟβΞ©L
[f(xβ, yβ(Ο), Ο)β f(xL, yL(Ο), Ο)
]ΞΆ(Ο)
β€ 1
L
βΟβΞ©L
β£β£f(xβ, yβ(Ο), Ο)β f(xL, yL(Ο), Ο)β£β£ ΞΆ(Ο)
β€ 1
L
βΟβΞ©L
ΞΆ(Ο) Β·[β₯xL β xββ₯+ β₯yL(Ο)β yβ(Ο)β₯
]Ο(Ο). (3.19)
It follows from (3.12) that the sequence
1L
βΟβΞ©L
ΞΆ(Ο)
is bounded. This yields
limLββ
1
L
βΟβΞ©L
β£β£β£f(xβ, yβ(Ο), Ο)β f(xL, yL(Ο), Ο)β£β£β£ΞΆ(Ο) = 0. (3.20)
Thus, by letting Lβ +β in (3.19) and taking (3.11) and (3.20) into account, we haveβ«Ξ©
f(xβ, yβ(Ο), Ο)ΞΆ(Ο)dΟ β€β«Ξ©
f(x, y(Ο), Ο)ΞΆ(Ο)dΟ,
which implies that xβ together with yβ(Β·) constitutes an optimal solution of problem (3.6). Thiscompletes the proof. 2
Similarly, by Lemma 3.1, (3.11), (3.12) and proof of Theorem 3.1, we have the following result.Theorem 3.2. Assume that condition (HC) holds, and f , F and U are the same as in Theorem
3.1. Then(i) problem (3.7) has at least one optimal solution when L is large enough;(ii) (xβ, yβ(Β·)) is an optimal solution of problem (2.5) when xβ is an accumulation point of the
sequence xL and yβ(Β·) is defined by
yβ(Ο) := maxi=1,2,Β·Β·Β· ,l
βt1 β F (xβ, 0, Ο),βt2 + F (xβ, 0, Ο),βti+2 β dβ²i(xβ)F (xβ, 0, Ο), 0, Ο β Ξ©.
Now, consider the case that the membership function of each fuzzy stochastic inequality andthe objective function EΟ[f(x, y(Ο), Ο)] in (3.3) is continuous, strictly increasing, and linear over the
12
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
852 Heng-you Lan 841-857
corresponding tolerance interval. A commonly used example in fuzzy set theory is that Ο(x) = 1βbxΞ²with b > 0 and Ξ² > 1. In this case, from the theory of convex analysis [27], Lemma 3.2, and Theorems3.1 and 3.2, we have the following simple result.
Theorem 3.3. Suppose that condition (HC) holds. If F : Rn+mΓΞ©β Rm is monotone in thesecond variable, and ¡Ωz
(x, y(Ο), Ο) is continuous, strictly increasing and linear for all z β C(x, Ο)and any (x, y(Ο), Ο) β Rn+mΓΞ©, then we can find an optimal solution (xβ, yβ(Β·)) of the stochastic op-timization problem (1.1) by solving the following stochastic parametric optimization problem: (3.8),which can be readily approximated by the iterative sequence (xk+1, yk+1(Ο), Ο, Ξ±k+1) generatedby Algorithm 3.2.
4 Simulation analysis
In this section, we shall give an example to illustrate the validity of our approaches.Taking n = 2,m = 3, l = 2, U = [1, 14]Γ [1, 14], d1(x) = (β1,β1, 3), d2(x) = (βn2, 1,β1),
f(x, y(Ο), Ο) = (x1 β y1)2 + x2y2 + 2Ο, F (x, y(Ο), Ο) =
β2x1 + y1 β 3y2 + Οx1 + x2 + 3y1 β y3 β 2Οβ2x2 + y2 + 2y3 β Ο
,
C(x, Ο) =
y(Ο) = (y1, y2, y3)T β R3
β£β£β£β£ ( d1(x)d2(x)
)y(Ο) β₯ 0
,
and Letting dβ²1(x) = (0, 3, 1) and dβ²2(x) = (1, 2, 0) in (3.3), and ΞΆ(Οβ) = pβ (β = 1, 2, Β· Β· Β· , L) in (3.5),then we have
Ο(x, y(Ο), Ο) := EΟ[(x1 β y1)2 + x2y2 + 2Ο] = 1L
βLβ=1[(x1 β y1)2 + x2y2 + 2Οβ]pβ,
f1(x, y(Ο), Ο) = β¨F (x, y(Ο), Ο), y(Ο)β© = β2x1y1 + x1y2 + x2y2 β 2x2y3 + y21+Οy1 β 2Οy2 + 2y23 β Οy3,
f2(x, y(Ο), Ο) = β¨βF (x, y(Ο), Ο), y(Ο)β© = 2x1y1 β x1y2 β x2y2 + 2x2y3 β y21βΟy1 + 2Οy2 β 2y23 + Οy3,
f3(x, y(Ο), Ο) = dβ²1(x)F (x, y(Ο), Ο) = 3x1 + x2 + 9y1 + y2 β y3 β 7Ο,f4(x, y(Ο), Ο) = dβ²2(x)F (x, y(Ο), Ο) = 2x2 + 7y1 β 3y2 β 2y3 β 3Ο.
Thus, problem (3.5) is equivalent to the following generalized fuzzy stochastic inequality constrainedoptimization program:
min Ο(x, y(Ο), Ο)s.t. 1 β€ x1, x2 β€ 14, y1, y2, y3 β₯ 0,
βy1 β y2 + y3 β₯ 0, β2y1 + y2 β y3 β₯ 0,fΞΉ(x, y(Ο), Ο) β₯
βΌ 0, ΞΉ = 1, 2, 3, 4,
(4.1)
with the membership function Β΅SΟ(x, y(Ο), Ο) (Ο = 1, 2, 3, 4), being specified as t1 = 9, t2 = 2, t3 =
6, t4 = 10,
Β΅S1(x, y(Ο), Ο) =
1, if f1(x) β₯ 0,
1β f1(x,y(Ο),Ο)9 , if f1(x, y(Ο), Ο) β [β9, 0),
0, if f1(x, y(Ο), Ο) < β9,
Β΅S2(x, y(Ο), Ο) =
1, if f2(x, y(Ο), Ο) β₯ 0,
1β f2(x,y(Ο),Ο)2 , if f2(x, y(Ο), Ο) β [β2, 0),
0, if f2(x) < β2,
Β΅S3(x, y(Ο), Ο) =
1, if f3(x, y(Ο), Ο) β₯ 0,
1β f3(x,y(Ο),Ο)6 , if f3(x, y(Ο), Ο) β [β6, 0),
0, if f3(x, y(Ο), Ο) < β6,
Β΅S4(x, y(Ο), Ο) =
1, if f4(x) β₯ 0,
1β f4(x,y(Ο),Ο)10 , if f4(x, y(Ο), Ο) β [β10, 0),
0, if f4(x) < β10,
13
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
853 Heng-you Lan 841-857
and
Β΅S0(x, y(Ο), Ο) =
1, if Ο(x, y(Ο), Ο) < f,
fβΟ(x,y(Ο),Ο)
fβf, if Ο(x, y(Ο), Ο) β [f, f),
0, if β¨Ο(x, y(Ο), Ο) β₯ f ,
where
f = min Ο(x, y(Ο), Ο)s.t. 1 β€ x1, x2 β€ 14, y1, y2, y3 β₯ 0,
βy1 β y2 + 3y3 β₯ 0, β2y1 + y2 β y3 β₯ 0,fΞΉ(x, y(Ο), Ο) β₯ 0, ΞΉ = 1, 2, 3, 4,
(4.2)
and
f = min Ο(x, y(Ο), Ο)s.t. 1 β€ x1, x2 β€ 14, y1, y2, y3 β₯ 0, ,
βy1 β y2 + 3y3 β₯ 0, β2y1 + y2 β y3 β₯ 0,f1(x, y(Ο), Ο) β₯ β9, f2(x, y(Ο), Ο) β₯ β2,f3(x, y(Ο), Ο) β₯ β6, f4(x, y(Ο), Ο) β₯ β10.
(4.3)
By Bellman and Zadehβs method of fuzzy decision making [15] and Algorithm 3.2, now we knowthat the conditions of Theorem 3.3 hold, and so an optimal solution of the problem (4.1) canbe obtained by solving the following unconstrained and smooth nonlinear parametric optimizationproblem:
minx,y(Ο),Ο,Ξ±1Ξ³ ln
exp[Ξ³(Ξ±k β Ξ±)] + exp[Ξ³(βΞ±)] + exp[Ξ³(Ξ±β 1)]
+ exp[Ξ³(Ο(x, y(Ο), Ο)β (f β Ξ±(f β f)))]+ exp[Ξ³(f1(x, y(Ο), Ο)β 9(1β Ξ±))]+ exp[Ξ³(f2(x, y(Ο), Ο)β 2(1β Ξ±))]+ exp[Ξ³(βf3(x, y(Ο), Ο) + 6(1β Ξ±))]+ exp[Ξ³(βf4(x, y(Ο), Ο) + 10(1β Ξ±))]+ exp[Ξ³(y1 + y2 β 3y3)] + exp[Ξ³(2y1 β y2 + y3)]+ exp[Ξ³(1β x1)] + exp[Ξ³(1β x2)] + exp[Ξ³(x1 β 14)]
+ exp[Ξ³(x2 β 14)] + exp[Ξ³(βy1)] + exp[Ξ³(βy2)] + exp[Ξ³(βy3)]
(4.4)
with Ξ³ being sufficiently large, where the optimal values of f and f are obtained by computing (4.2)and (4.3), respectively.
Choosing x0 = (4.0000, 2.0000), y0(Ο) = (1.0547, 1.0564, 0.1574) and Ξ±0 = 0.2 and setting L = 3with the probability p1 = 0.1590, p2 = 0.6821, p3 = 0.1589, and Ο΅ = 10β5, Q = 106 and fixed Ξ³ = 12,then for each iteration of Algorithm 3.2, we first generate the random variable Ο by using normrndfunction (that is, normal distribution function) of MATLAB 7.0 software. Secondly, we solve fromproblems (4.2) and (4.3) to problem (4.4) in turn by the commonly used quasi-Newton line searchof MATLAB software 7.0.
Here, the first layer iteration searching optimization is to solve problems (4.2) and (4.3), respec-tively. And the second optimizing process is to find the optimal solution of problem (4.1) via solvingthe unconstrained and smooth nonlinear parametric optimization problem (4.4). We only presentfour optimal solution (xβ, yβ(Β·)) with respect to the random variable Ο and the corresponding mem-bership degree Ξ±β for whole stochastic optimization problem, which is listed in Table 1. Further,Table 2 show that each iteration calculation results including iterative solutions with the randomvariable Ο = 0.128808 for the second optimizing process to this problem. The results for every stepin Table 2 (i.e., k = 0, 1, 2, Β· Β· Β· , 11) come from the first layer iteration process, which are too muchand so they are omitted.
14
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
854 Heng-you Lan 841-857
Table 1: The optimal solution with the random variable and membership degree
k (xβ, yβ(Ο), Ο) Ξ±β
1 (0.984674, 1.158197, 0.036645, 0.228989, 0.128690, 0.128808) 0.9543632 (0.987781, 1.219403, 0.024577, 0.187961, 0.097964, 0.345629) 0.9496523 (0.984337, 1.156955, 0.036586, 0.232004, 0.127346, 0.191080) 0.9531854 (0.988056, 1.226168, 0.023580, 0.184104, 0.095786, 0.353731) 0.949391
Table 2: Data for Computational results with the random variable Ο = 0.128808
k (xk, yk) Ξ±k Iterations No.0 (4.0000, 2.0000, 1.0547, 1.0564, 0.1574) 0.2 461 (0.989779,1.252773,0.027679,0.173089,0.106607) 0.871896 162 (0.985831,1.166544,0.034618,0.217474,0.125437) 0.924577 133 (0.984928,1.159731,0.036190,0.226463,0.128032) 0.945141 154 (0.984726,1.158504,0.036552,0.228473,0.128555) 0.951749 145 (0.984687,1.158268,0.036623,0.228867,0.128658) 0.953643 96 (0.984677,1.158241,0.036639,0.228955,0.128681) 0.954166 47 (0.984676,1.158237,0.036638,0.228959,0.128678) 0.954309 38 (0.984676,1.158233,0.036638,0.228962,0.128678) 0.954348 59 (0.984674,1.158198,0.036645,0.228989,0.128690) 0.954359 110 (0.984674,1.158197,0.036645,0.228989,0.128690) 0.954362 111 (0.984674,1.158197,0.036645,0.228989,0.128690) 0.954363
5 Concluding remarks
In this paper, by developing a class of new regularization smoothing approximation approaches,we investigated approximation solvability of the following fuzzy parametric variational inequalityconstrained stochastic optimization problems in n-dimension real numeral set Rn:
minx,y(Β·)
EΟ[f(x, y(Ο), Ο)]
s.t. x β U,y(Ο) β C(x, Ο),β¨F (x, y(Ο), Ο), z(Ο)β y(Ο)β©β₯βΌ 0, βz(Ο) β C(x, Ο),
(5.1)
which has been very little studied by right of the known theories and approaches in the literature.It is because the existence of the random variable and the fuzzified version mean that (5.1) involvesmultiple complementarity-type constraints, and solving problem (5.1) is more difficult than solvingan ordinary mathematical program with (fuzzy) equilibrium constraints or stochastic mathematicalprogram with equilibrium constraints.
Based on the notion of tolerance approach with entropic regularization and fuzzy set theory,we first showed that solving the stochastic optimization problem with fuzzy parametric variationalinequality constraints is equivalent to solving a fuzzy complementarity constrained stochastic opti-mization problem, which can be converted to a regular nonlinear parametric optimization problemwith continuous random variables. Then, we constructed a centre iterative algorithm and developeda class of new regularization smoothing approximation approaches for solving a problem with contin-uous random variables based on quasi-Monte Carlo estimate and entropic regularization technique,and discussed a comprehensive convergence theory for approximating the resulting optimizationproblem. Finally, numerical example was provided to illustrate our main results applying quasi-Newton line search of MATLAB software.
We remark that in the paper, based on the concept that fuzzy constraints should yield a fuzzy
15
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
855 Heng-you Lan 841-857
objective, we must choose a class of new regularization smoothing approximation approaches todefine the objective function value of two optimization problems as the parameters in an equivalentstochastic parametric optimization problem. Hence, the problem presented in this paper is brandnew and the method is also new and interesting.
Whether the corresponding results of Theorem 3.3 hold when the objective function is a fuzzystochastic function, the constraints are fuzzy implicit variational inequalities (such as in [14, 17]) orelliptic inequalities subject to physical phenomenon, and the numerical testing is some large-scaleapplications, which are still open questions to be solved in further research.
Acknowledgments
This work has been partially supported by Sichuan Province Cultivation Fund Project of Academ-ic and Technical Leaders, and the Scientific Research Project of Sichuan University of Science &Engineering (2015RC07).
References
[1] S.I. Birbil, G. Gurkan and O. Listes, Solving stochastic mathematical programs with complementarityconstraints using simulation, Math. Oper. Res. 31(4) (2006), 739β760.
[2] Y. Chen and Q.J. Hu, A SQP algorithm based on a smoothing lower order penalty function for inequalityconstrained optimization, J. Comput. Anal. Appl. 11(3) (2009), 481β491.
[3] X.J. Chen, H.L. Sun and Roger J.B. Wets, Regularized mathematical programs with stochastic equilib-rium constraints: estimating structural demand models, SIAM J. Optim. 25(1) (2015), 53β75.
[4] Q.J. Hu, W.Y. Chen and Y.H. Xiao, An improved active set feasible SQP algorithm for the solution ofinequality constrained optimization problems, J. Comput. Anal. Appl. 11(1) (2009), 54β63.
[5] M. Patriche, Bayesian abstract fuzzy economies, random quasi-variational inequalities with random fuzzymappings and random fixed point theorems, Fuzzy Sets and Systems 245 (2014), 125β136.
[6] M. Patriksson, On the applicability and solution of bilevel optimization models in transportation science:a study on the existence, stability and computation of optimal solutions to stochastic mathematicalprograms with equilibrium constraints, Transportation. Res. B 42 (2008), 843β860.
[7] F. Toyasaki, P. Daniele and T. Wakolbinger, A variational inequality formulation of equilibrium modelsfor end-of-life products with nonlinear constraints, European J. Oper. Res. 236(1) (2014), 340β350.
[8] J. Zhang, Y.Q. Zhang and L.W. Zhang, A sample average approximation regularization method for astochastic mathematical program with general vertical complementarity constraints, J. Comput. Appl.Math. 280 (2015), 202β216.
[9] G.H. Lin, X.J. Chen, M. Fukushima, Smoothing Implicit Programming Approaches for Stochastic Math-ematical Programs with Linear Complementarity Constraints, Technical Report 2003-006, Departmentof Applied Mathematics and Physics, Graduate School of Informatics, Kyoto University, Kyoto, Japan,2003.
[10] G.H. Lin, X.J. Chen and M. Fukushima, Solving stochastic mathematical programs with equilibriumconstraints via approximation and smoothing implicit programming with penalization, Math. Program.Ser. B 116 (2009), 343β368.
[11] H. Xu, An implicit programming approach for a class of stochastic mathematical programs with linearcomplementarity constraints, SIAM J. Optim. 16 (2006), 670β696.
[12] C.F. Hu and F.B. Liu, Solving mathematical programs with fuzzy equilibrium constraints, Comput.Math. Appl. 58 (2009), 1844β1851.
[13] H.Y. Lan, C.J. Liu and T.X. Lu, Method of centres for solving mathematical programs with fuzzyparametric variational inequality constraints, in: X.S. Zhang et al. (Eds.), 11th International Symposiumon Operations Research and its Applications in Engineering, Technology and Management, Vol. 2013,Issue 644CP, 2013, pp. 194β199.
[14] H.Y. Lan and J.J. Nieto, Solving implicit mathematical programs with fuzzy variational inequalityconstraints based on the method of centres with entropic regularization, Fuzzy Optim. Decis. Mak. (inpress, March 2015) http://dx.doi.org/10.1007/s10700-015-9207-7
16
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
856 Heng-you Lan 841-857
[15] R. Bellman and L.A. Zadeh, Decision making in a fuzzy environment, Management Sci. 17B (1970),141β164.
[16] Z.T. Gong, X.X. Liu and X. Feng, Almost ideal statistical convergence and strongly almost ideallacunary convergence of sequences of fuzzy numbers with respect to the Orlicz functions, J. Comput.Anal. Appl. 19(3) (2015), 418β425.
[17] H.Y. Lan, An approach for solving fuzzy implicit variational inequalities with linear membership func-tions, Comput. Math. Appl. 55(3) (2008), 563β572.
[18] E.Y. Pee and J.O. Royset, On solving large-scale finite minimax problems using exponential smoothing,J. Optim. Theory Appl. 148(2) (2011), 390β421.
[19] D.M. Yuan and X.L. Cheng, Method of fundamental solutions with an optimal regularization techniquefor the Cauchy problem of the modified Helmholtz equation, J. Comput. Anal. Appl. 14(1) (2012), 54β66.
[20] Z.H. Zhang, Asymptotic representations in stochastic process approximations, J. Comput. Anal. Appl.18(4) (2015), 672β684.
[21] H.J. Zimmermann, Fuzzy Set Theory and Its Applications (2nd edition), Kluwer Academic, Dordrecht,1991.
[22] C.F. Hu, Solving Systems of Fuzzy Inequalities, Ph.D thesis, North Carolina State University, ProQuestLLC, Ann Arbor, MI, 1997.
[23] Y. Chen and M. Florian, The nonlinear bilevel programming problem, formulations, regularity andoptimality conditions, Optimization 32 (1995), 193β209.
[24] H. Niederreiter, Random number generation and quasi-Monte Carlo methods, in: CBMS-NSF RegionalConference Series in Applied Mathematics, vol. 63, Society for Industrial and Applied Mathematics,Philadelphia, PA, vi+241 pp, 1992.
[25] O.G. Mancino and G. Stampacchia, Convex programming and variational inequalities, J. Optim. TheoryAppl. 9 (1972), 3β23.
[26] H.F. Wang and H.L. Liao, Variational inequality with fuzzy convex cone, J. Global Optim. 14(4) (1999),395β414.
[27] R.T. Rockafellar, Convex Analysis, Princeton, NJ: Princeton Univ. Press, 1970.
[28] P. Huard, Resolution of mathematical programming with nonlinear constraints by the method of centresin nonlinear programming, in: Nonlinear Programming (eds. J. Abadie), North-Holland, Amsterdam,1967, pp. 207β219.
17
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
857 Heng-you Lan 841-857
The Split Common Fixed Point Problem
for Demicontractive Mappings
in Banach Spaces
Li Yang1, Fuhai Zhao2 and Jong Kyu Kim3
1School of Science, South West University of Science and Technology,Mianyang, Sichuan 621010, P. R. China
e-mail: [email protected]
2School of Science, South West University of Science and Technology,Mianyang, Sichuan 621010, P. R. China
e-mail: [email protected]
3Department of Mathematics Education, Kyungnam University,Changwon, Gyeongnam 51767, Korea
e-mail: [email protected]
Abstract. In this paper, based on the work by Moudafi and inspired by Takahashi and Xu, we tryto investigate the split common fixed point problems for the class of demicontractive mappings inthe setting of two Banach spaces, and obtain the strong and weak convergence theorems. The resultspresented in the paper improve and extend some recent well-known corresponding results.
Keywords: split common fixed point problem; demicontractive mapping; Demiclosed principle,weak and strong convergence theorems.
2010 AMS Subject Classification: 47H09, 49J25.
1 Introduction and Preliminaries
The split common fixed point problem was introduced by Moudafi [1] in 2010. Moudafiproposed an iteration scheme and obtained a weak convergence theorem of the splitcommon fixed point problem for demicontractive mappings in the setting of twoHilbert spaces. Since then, many authors investigated the split common fixed pointproblems of other nonlinear mappings in the setting of two Hilbert spaces (see [2-7]).At the beginning of 2015, Takahashi [8] first attempted to introduce and consider thesplit feasibility problem and split common null point problem in the setting of oneHilbert space and one Banach space. By using hybrid methods and Halperns typemethods under suitable conditions, some strong and weak convergence theorems forsuch problems are obtained. The results presented in [8] seem to be the first outsideHilbert spaces. This naturally brings us to solve the split common fixed point problemfor demicontractive mappings in the setting of two Banach space.
Let E1 and E2 be two real Banach spaces, and A : E1 β E2 be a boundedlinear operator such that A 6= 0. The split common fixed point problem (SCFP) fornonlinear mappings S and T is to find a point x β E1 such thst
x β F (S) and Ax β F (T ), (1.1)
0Corresponding author: Jong Kyu Kim([email protected])
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
858 Li Yang et al 858-863
where F (S) and F (T ) denote the sets of fixed points of S and T , respectively. Weuse Ξ to denote the set of solutions of SCFP for mappings S and T , that is,
Ξ = x β F (S)| Ax β F (T ).
In this paper, we use the following algorithm to approximate a split common fixedpoint of demicontractive mappings in the setting of two Banach spaces.
Algorithm: Let E1 and E2 be two real Banach spaces, A : E1 β E2 be a boundedlinear operator, Aβ be the adjoint operator of A and Ji be the normalized dualitymapping from Ei to 2E
βi , i = 1, 2. Now, we define the iterative scheme xn:
Let x1 β E1 be arbitrary, for all n β₯ 1, set
yn = xn + Ξ³Jβ11 AβJ2(T β I)Axn, (1.2)
xn+1 = (1β Ξ±n)yn + Ξ±nS(yn), (1.3)
where S : E1 β E1 and T : E2 β E2 are two demicontractive mappings.
Under some suitable conditions, the iterative scheme xn is shown to convergeweakly and strongly to a split common fixed point of demicontractive mappings Tand S. Our result extends the split common fixed point problem from Hilbert spacesto Banach spaces.
In order to solve this problem mentioned above, we recall the following conceptsand results.
Let E be a real Banach space with norm β Β· β and let Eβ be the dual space of E,We denote the value of yβ β E at x β E by γx, yβγ. When xn is a sequence in E, wedenote the strong convergence of xn to x β E by xn β x and the weak convergenceby xn x.
We recall that T : E β E is demicontractive (see for example [9]) if there existsa constant Ξ· β [0, 1) such that
βTxβ qβ2 β€ βxβ qβ2 + Ξ·βxβ Txβ2, β(x, q) β E Γ F (T ). (1.4)
An operator satisfying (1.4) will be referred to as a Ξ·-demicontractive mapping.It is worth noting that the class of demicontractive maps contains important op-
erators such as the quasi-nonexpansive maps and the strictly pseudocontractive mapswith fixed points.
A mapping T : E β E is called quasi-nonexpansive, if
βTxβ qβ β€ βxβ qβ
for all (x, q) β E Γ F (T ). A mapping T : E β E is strictly pseudocontractive, if
βTxβ Tyβ2 β€ βxβ yβ2 + Ξ²βxβ y β (Txβ Ty)β2
for all (x, y) β E Γ E and for some Ξ² β [0, 1).A mapping T : E β E is called demiclosed at zero, if for any sequence xn β E
and x β E, we have
xn x, (I β T )(xn)β 0β x β F (T ).
A mapping S : E β E is said to be semi-compact, if for any sequence xn in Esuch that βxn β Sxnβ β 0, (n β β), there exists subsequence xnj
of xn suchthat xnj
converges strongly to xβ β E.
2
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
859 Li Yang et al 858-863
The normalized duality mapping J from E to 2Eβ
is defined by
Jx = xβ β Eβ : γx, xβγ = βxβ2 = βxββ2, βx β E.
Let U = x β E : βxβ = 1. The norm of E is said to be Gateaux differentiable iffor each x, y β U , the limit
limtβ0
βx+ tyβ β βxβt
exists.
In the case, E is called smooth. E is smooth if and only if J is single-valued. Wedenote the single-valued normalized duality mapping by J .
The modulus of convexity of E is defined by
Ξ΄E(Ξ΅) = inf
1β βx+ yβ2
: βxβ β€ 1, βyβ β€ 1, βxβ yβ β₯ Ξ΅,
for every Ξ΅ with 0 β€ Ξ΅ β€ 2. A Banach space E is said to be uniformly convex ifΞ΄(Ξ΅) > 0 for every Ξ΅ > 0. E is said to be p-uniformly convex, if there exists a constanta > 0 such that Ξ΄E(Ξ΅) β₯ aΞ΅p for all 0 < Ξ΅ β€ 2.
Let ΟE : [0,β)β [0,β) be the modulus of smoothness of E defined by
ΟE(t) = sup1
2(βx+ yβ+ βxβ yβ)β 1 : x β U, βyβ β€ t.
A Banach space E is said to be uniformly smooth if ΟE(t)t β 0 as t β 0. Let q be
a fixed real number with q > 1. Then a Banach space E is said to be q-uniformlysmooth if there exists a constant b > 0 such that ΟE(t) β€ btq for all t > 0. It is wellknown that every q-uniformly smooth Banach space is uniformly smooth.
A Banach space E is said to satisfy the Opialβs condition [10] if for any sequencexn β E, xn x implies
lim supnββ
βxn β xβ < lim supnββ
βxn β yβ,
for all y β E with y 6= x.
Lemma 1.1. [11] Let E be a 2-uniformly convex Banach space. Then the followinginequality holds:
βΞ»x+ (1β Ξ»)yβ2 β€ Ξ»βxβ2 + (1β Ξ»)βyβ2 β Ξ»(1β Ξ»)cβxβ yβ2, βx, y β E, (1.5)
where 0 β€ Ξ» β€ 1, c = Β΅(1) > 0,
Β΅(t) : = infΞ»βxβ2 + (1β Ξ»)βyβ2 β βΞ»x+ (1β Ξ»)yβ2
Ξ»(1β Ξ»):
0 < Ξ» < 1, x, y β E and βxβ yβ = t
> 0.
Lemma 1.2. [11] Let E be a 2-uniformly smooth Banach space with the best smooth-ness constants ΞΊ > 0. Then the following inequality holds:
βx+ yβ2 β€ βxβ2 + 2γy, Jyγ+ 2βΞΊyβ2,
for all x, y β E.
3
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
860 Li Yang et al 858-863
2 Main Results
Lemma 2.1. Let E1 be a real 2-uniformly convex and 2-uniformly smooth Banachspaces with the best smoothness constant ΞΊ satisfying 0 < ΞΊ < 1β
2, E2 be a real
Banach space, and A : E1 β E2 be a bounded linear operator. Let S : E1 β E1 be Ξ²-demicontractive and T : E2 β E2 be Ξ·-demicontractive with F (S) 6= β and F (T ) 6= β .Then the sequence xn generated by algorithm (1.2)-(1.3) is Fejer-monotone withrespect to Ξ = x β F (S)|Ax β F (T ), that is, for every z β Ξ,
βxn+1 β zβ β€ βxn β zβ, βn β N,
where 0 < Ξ³ < min
1βΞ·βAβ2 ,
1β2ΞΊβAβ2
and Ξ±n β (0, 1β Ξ²
c ], Ξ² < c = Β΅(1).
Proof. Let z β Ξ. Then z β F (S) and Az β F (T ). It follows from Lemma 1.1 and(1.3) that
βxn+1 β zβ2 = β(1β Ξ±n)yn + Ξ±nS(yn)β zβ2
= β(1β Ξ±n)(yn β z) + Ξ±n(S(yn)β z)β2
β€ (1β Ξ±n)βyn β zβ2 + Ξ±nβS(yn)β zβ2
βΞ±n(1β Ξ±n)cβS(yn)β ynβ2 (2.1)
β€ (1β Ξ±n)βyn β zβ2 + Ξ±nβyn β zβ2 + Ξ±nΞ²βS(yn)β ynβ2
βΞ±n(1β Ξ±n)cβS(yn)β ynβ2
β€ βyn β zβ2 β Ξ±n(cβ Ξ² β cΞ±n)βS(yn)β ynβ2,
where c = Β΅(1).On the other hand, It follows from (1.2) and Lemma 1.2 that
βyn β zβ2 = βxn + Ξ³Jβ11 AβJ2(T β I)Axn β zβ2
= βxn β z + Ξ³Jβ11 AβJ2(T β I)Axnβ2
β€ βΞ³Jβ11 AβJ2(T β I)Axnβ2 + 2Ξ³γxn β z,AβJ2(T β I)Axnγ+2ΞΊ2βxn β zβ2
β€ Ξ³2βAβ2β(T β I)Axnβ2 + 2Ξ³γAxn βAz, J2(T β I)Axnγ+2ΞΊ2βxn β zβ2
= Ξ³2βAβ2β(T β I)Axnβ2 + 2ΞΊ2βxn β zβ2
2Ξ³γAxn β TAxn + TAxn βAz, J2(T β I)Axnγβ€ Ξ³2βAβ2β(T β I)Axnβ2 + 2ΞΊ2βxn β zβ2
β2Ξ³β(T β I)Axnβ2 + 2Ξ³γTAxn βAz, J2(T β I)Axnγβ€ (Ξ³2βAβ2 β 2Ξ³)β(T β I)Axnβ2 + 2ΞΊ2βxn β zβ2
+Ξ³(βTAxn βAzβ2 + β(T β I)Axnβ2)
β€ 7(Ξ³2βAβ2 β 2Ξ³)β(T β I)Axnβ2 + 2ΞΊ2βxn β zβ2
+Ξ³(βAxn βAzβ2 + Ξ·|(Axn β TAxnβ2 + β(T β I)Axnβ2)
β€ (2ΞΊ2 + Ξ³βAβ2)βxn β zβ2 β Ξ³(1β Ξ· β Ξ³βAβ2)β(T β I)Axnβ2,
where Aβ is the adjoint operator of A and Ji is the normalized duality mapping fromEi to 2E
βi , i = 1, 2.
In addition, since 0 < ΞΊ < 1β2
and 0 < Ξ³ < 1β2ΞΊ2
βAβ2 , 0 < Ξ³βAβ2 + 2ΞΊ2 < 1, so we
haveβyn β zβ2 β€ βxn β zβ2 β Ξ³(1β Ξ· β Ξ³βAβ2)β(T β I)Axnβ2. (2.2)
4
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
861 Li Yang et al 858-863
It follows from (2.1) and (2.2) that
βxn+1 β zβ2 β€ βxn β zβ2 β Ξ³(1β Ξ· β Ξ³βAβ2)β(T β I)Axnβ2 (2.3)
βΞ±n(cβ Ξ² β cΞ±n)βS(yn)β ynβ2.
Finally, by the assumptions on Ξ³ and Ξ±n, we obtain the desired result.
Theorem 2.2. Let E1 be a real 2-uniformly convex and 2-uniformly smooth Banachspace satisfying Opialβs condition with the best smoothness constant ΞΊ satisfying 0 <ΞΊ < 1β
2, and E2 be a real Banach space. Let A : E1 β E2 be a bounded linear
operator, S : E1 β E1 and T : E2 β E2 be two demicontractive mappings withconstants Ξ² and Ξ· with F (S) 6= β and F (T ) 6= β , respectively. Assume that I β Sand I β T are demiclosed at zero. If Ξ 6= β , then the sequence xn generated byalgorithm (1.2)-(1.3) converges weakly to a split common fixed point x β Ξ, for 0 <Ξ³ < min
1βΞ·βAβ2 ,
1β2ΞΊβAβ2
, Ξ±n β (Ξ΄, 1β Ξ²
c βΞ΄), Ξ² < c = Β΅(1), and for small enough Ξ΄ > 0.
Proof. From (2.3) and the fact that 0 < Ξ³ < min
1βΞ·βAβ2 ,
1β2ΞΊβAβ2
and Ξ±n β (Ξ΄, 1β Ξ²
cβΞ΄),we obtain that the sequence βxnβzβ is monotonically decreasing and thus convergesto some positive real limit l(z). From (2.3), we have
Ξ³(1β Ξ· β Ξ³βAβ2)β(I β T )Axnβ2 β€ βxn β zβ2 β βxn+1 β zβ2.
Therefore,limnββ
β(I β T )Axnβ = 0. (2.4)
From the Fejer-monotonicity of xn, it follows that the sequence is bounded. Denot-ing by x a weak-cluster point of xn. Let k = 0, 1, 2, ... be the sequence of indices,such that xnk
x, as k ββ. Then from (2.4) and demiclosedness of I β T at zero,we obtain T (Ax) = Ax, that is, Ax β F (T ).
Now, by setting yn = xn + Ξ³Jβ11 AβJ2(I β T )Axn, it follows that ynk x. Again
from (2.3), we obtain
Ξ±n(cβ Ξ² β cΞ±n)βyn β S(yn)β2 β€ βxn β zβ2 β βxn+1 β zβ2.
Using the convergence of the sequence βxn β zβ, we get
limnββ
βyn β S(yn)β = 0, (2.5)
which combined with the demiclosedness of I β S at zero and the weak convergenceof ynk
to y yields S(x) = x. Hence, x β F (S) and therefore x β Ξ. Since E1 satisesOpialβs condition, we know that xn converges weakly to x β Ξ.
Theorem 2.3. Let E1 be a real 2-uniformly convex and 2-uniformly smooth Banachspace satisfying Opialβs condition with the best smoothness constant ΞΊ satisfying 0 <ΞΊ < 1β
2, and E2 be a real Banach space. Let A : E1 β E2 be a bounded linear
operator, S : E1 β E1 and T : E2 β E2 be two demicontractive mappings withconstants Ξ² and Ξ· with F (S) 6= β and F (T ) 6= β , respectively. Assume that I β Sand I β T are demiclosed at zero. If Ξ 6= β and S is semi-compact, then the sequencexn generated by algorithm (1.2)-(1.3) converges strongly to a split common fixedpoint x β Ξ, for 0 < Ξ³ < min 1βΞ·
βAβ2 ,1β2ΞΊβAβ2 , Ξ±n β (Ξ΄, 1 β Ξ²
c β Ξ΄), Ξ² < c = Β΅(1), and
for a small enough Ξ΄ > 0.
Proof. It follows from (1.2) that
βxn β ynβ = βJ1(xn β yn)β = βΞ³AβJ2(I β T )Axnβ,
5
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
862 Li Yang et al 858-863
and so, from (2.4) we havelimnββ
βxn β ynβ = 0. (2.6)
Since S is semi-compact, from (2.5), there exist subsequence ynj of yn such that
ynj converges strongly to xβ β E1. Using (2.6), we know that xnj
convergesstrongly to xβ. By Theorem 2.2, we know that xn converges weakly to x, so wehave xβ = x. Since lim
nβββxn β xβ exists and lim
jβββxnj β xβ = 0, we know that xn
converges strongly to x β Ξ.
Acknowledgements: The first author was Supported by Scientific Reserch Fundof Sichuan Provincial Education Department (No.15ZA0112) and third author wassupported by the Basic Science Research Program through the National ResearchFoundation(NRF) Grant funded by Ministry of Education of the republic of Ko-rea(2014046293).
References
[1] A. Moudafi, The split common fixed point problem for demi-contractive map-pings, Inverse Problem, Vol. 26(2010), Article ID 055007, 2010.
[2] S.S. Chang, L. Wang, Y.K. Tang and L. Yang, The split common fixed fointproblem for total asymptotically strictly pseudocontractive mappings, Journalof Applied Mathematics, Vol. 2012, Article ID 385638, 2012.
[3] S.S. Chang, J.K. Kim, Y.J. Cho and J.Y. Sim,, Weak and strong convergencetheorems of solution to split feasibility problem for nonspreading type mappingin Hilbert spaces, Fixed Point Theory and Appl., Vol. 2014:11, 2014.
[4] X.F. Zhang, L. Wang, Z.L. Ma and W. Duan, The strong convergence theoremsfor split common fixed point problems of asymptotically nonexpansive mappingsin Hilbert spaces, Journal of Inequalities and Appl., Vol. 2015:1, 2015.
[5] J. Quan, S.S. Chang and X. Zhang, Multiple-set split feasibility problems fork-strictly pseudononspreading mapping in Hilbert spaces, Abstract and AppliedAnalysis, Vol. 2013, Article ID 342545, doi:10.1155/ 2013/342545.
[6] S.S. Chang, Y.J. Cho, J.K. Kim, W.B. Zhang and L. Yang, Multiple-set splitfeasibility problems for asymptotically strict pseudocontractions, Abstract andApplied Analysis, Vol. 2012, Article ID 491760, doi:10.1155/2012/491760.
[7] A. Moudafi, A note on the split common fixed point problem for quasi-nonexpansive operators, Nonlinear Anal.TMA, Vol. 74(2011), 4083-4087.
[8] W. Takahashi, The split feasibility problem in Banach spaces, Nonlinear ConvexAnal., Vol. 15(6)(2015), 1349-1355.
[9] S. Maruster and C. Popirlan On the Mann-type iteration and convex feasibilityproblem J. Comput. Appl. Math., Vol. 212(2008), 390396.
[10] Z. Opial, Weak convergence of the sequence of successive approximations fornonexpansive mappings, Bull. Amer. Math. Soc., Vol. 73(1967), 591-597.
[11] H.K. Xu, Inequalities in Banach spaces with applications, Nonlinear Anal. TMA,Vol. 16(1991), 1127-1138.
6
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
863 Li Yang et al 858-863
Iterated binomial transform of the k-Lucas sequence
Nazmiye Yilmaz and Necati Taskara
Department of Mathematics, Faculty of Science,Selcuk University, Campus, 42075, Konya - Turkey
[email protected] and [email protected]
Abstract
In this study, we apply βrβ times the binomial transform to k-Lucassequence. Also, the Binet formula, summation, generating function ofthis transform are found using recurrence relation. Finally, we give theproperties of iterated binomial transform with classical Lucas sequence.
Keywords: k-Lucas sequence, iterated binomial transform, Pell se-quence.
Ams Classification: 11B65, 11B83.
1 Introduction and Preliminaries
There are so many studies in the literature that concern about the specialnumber sequences such as Fibonacci, Lucas and generalized Fibonacci anadLucas numbers (see, for example [1]-[3], and the references cited therein). InFibonacci and Lucas numbers, there clearly exists the term Golden ratio whichis defined as the ratio of two consecutive of these numbers that converges to
Ξ± = 1+β5
2 . It is also clear that the ratio has so many applications in, specially,Physics, Engineering, Architecture, etc.[4]. Also, many generalizations of theFibonacci sequence have been introduced and studied matrix applications ofthis sequence in [13]-[16].
For n β₯ 1, k-Lucas sequence is defined by the recursive equation:
Lk,n+1 = kLk,n + Lk,nβ1, Lk,0 = 2 and Lk,1 = k. (1.1)
In addition, some matrix-based transforms can be introduced for a givensequence. Binomial transform is one of these transforms and there are alsoother ones such as rising and falling binomial transforms(see [5]-[12]). Given
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
864 Nazmiye Yilmaz et al 864-873
an integer sequence X = x0, x1, x2, . . ., the binomial transform B of thesequence X, B (X) = bn , is given by
bn =nβ
i=0
(n
i
)xi.
In [10], authors gave the application of the several class of transforms tothe k-Lucas sequence. For example, for n β₯ 1, authors obtained recurrencerelation of the binomial transform for k-Lucas sequence
bk,n+1 = (2 + k) bk,n β kbk,nβ1, bk,0 = 2 and bk,1 = k + 2.
Falcon [11] studied the iterated application of some Binomial transforms tothe k-Fibonacci sequence. For example, author obtained recurrence relationof the iterated binomial transform for k-Fibonacci sequence
c(r)k,n+1 = (2r + k) c
(r)k,n β
(r2 + kr β 1
)c(r)k,nβ1, c
(r)k,0 = 0 and c
(r)k,1 = 1.
Motivated by [11, 12], the goal of this paper is to apply iteratively the bino-mial transform to the k-Lucas sequence. Also, the properties of this transformare found by recurrence relation. Finally, the relation of between the trans-form and the iterated binomial transform of k-Fibonacci sequence by derivingnew formulas are illustrated.
2 Iterated Binomial Transform of k-Lucas Sequences
In this section, we will mainly focus on iterated binomial transforms of k -Lucas sequences to get some important results. In fact, we will also presentthe recurrence relation, Binet formula, summation, generating function of thetransform and relationships betweeen of the transform and iterated binomialtransform of k-Fibonacci sequence.
The iterated binomial transform of the k-Lucas sequences is demonstrated
by B(r)k =
b(r)k,n
, where b
(r)k,n is obtained by applying βrβ times the binomial
transform to k-Lucas sequence. It is obvious that b(r)k,0 = 2 and b
(r)k,1 = 2r + k.
The following lemma will be the key proof of the next theorems.
Lemma 2.1 For n β₯ 0 and r β₯ 1, the following equality hold:
b(r)k,n+1 = b
(r)k,n +
nβj=0
(n
j
)b(rβ1)k,j+1.
2
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
865 Nazmiye Yilmaz et al 864-873
Proof. By using definition of binomial transform and the well known binomialequality (
n+ 1
i
)=
(n
i
)+
(n
iβ 1
),
we obtain
b(r)k,n+1 =
n+1βj=0
(n+ 1
j
)b(rβ1)k,j
=
n+1βj=1
(n+ 1
j
)b(rβ1)k,j + b
(rβ1)k,0
b(r)k,n+1 =
n+1βj=1
(n
j
)b(rβ1)k,j +
n+1βj=1
(n
j β 1
)b(rβ1)k,j + b
(rβ1)k,0
=
n+1βj=0
(n
j
)b(rβ1)k,j +
n+1βj=0
(n
j β 1
)b(rβ1)k,j
=nβ
j=0
(n
j
)b(rβ1)k,j +
nβj=β1
(n
j
)b(rβ1)k,j+1
= b(r)k,n +
nβj=0
(n
j
)b(rβ1)k,j+1
which is desired result.In [10], the authors obtained the following equality for binomial transform
of k-Lucas sequences. However, in here, we obtain the equality in terms of iter-ated binomial transform of the k-Lucas sequences as a consequence of Lemma2.1. To do that we take r = 1 in Lemma 2.1:
bk,n+1 = bk,n +
nβj=0
(n
j
)Lk,j+1.
Theorem 2.1 For n β₯ 0 and r β₯ 1, the recurrence relation of sequenceb(r)k,n
is
b(r)k,n+1 = (2r + k) b
(r)k,n β
(r2 + kr β 1
)b(r)k,nβ1, (2.1)
with initial conditions b(r)k,0 = 2 and b
(r)k,1 = 2r + k.
3
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
866 Nazmiye Yilmaz et al 864-873
Proof. The proof will be done by induction steps on r and n.First of all, for r = 1, from the equality 2.2 in [10], it is true bk,n+1 =
(2 + k) bk,n β kbk,nβ1.Let us consider definition of iterated binomial transform, then we have
b(r)k,2 = k2 + 2rk + 2r2 + 2.
The initial conditions are
b(r)k,0 = 2 and b
(r)k,1 = 2r + k.
Hence, for n = 1, the Eq. (2.1) is true, that is b(r)k,2 = (2r + k) b
(r)k,1β
(r2 + kr β 1
)b(r)k,0.
Actually, by assuming the Eq. (2.1) holds for all (r β 1, n) and (r, nβ 1),that is,
b(rβ1)k,n+1 = (2r β 2 + k) b
(rβ1)k,n β
((r β 1)2 + k (r β 1)β 1
)b(rβ1)k,nβ1,
andb(r)k,n = (2r + k) b
(r)k,nβ1 β
(r2 + kr β 1
)b(r)k,nβ2.
Now, by taking into account Lemma 2.1, we obtain
b(r)k,n+1 = b
(r)k,n +
nβj=0
(n
j
)b(rβ1)k,j+1
=
nβj=0
(n
j
)b(rβ1)k,j +
nβj=0
(n
j
)b(rβ1)k,j+1
=nβ
j=1
(n
j
)(b(rβ1)k,j + b
(rβ1)k,j+1
)+ b
(rβ1)k,0 + b
(rβ1)k,1 .
By reconsidering our assumption, we write
b(r)k,n+1 =
nβj=1
(n
j
)(b(rβ1)k,j + (2r β 2 + k) b
(rβ1)k,j β
(r2 β 2r + kr β k
)b(rβ1)k,jβ1
)+ b
(rβ1)k,0 + b
(rβ1)k,1
= (2r + k β 1)
nβj=1
(n
j
)b(rβ1)k,j β
(r2 β 2r + kr β k
) nβj=1
(n
j
)b(rβ1)k,jβ1 + b
(rβ1)k,0 + b
(rβ1)k,1
= (2r + k β 1)
nβj=0
(n
j
)b(rβ1)k,j β
(r2 β 2r + kr β k
) nβj=1
(n
j
)b(rβ1)k,jβ1 + b
(rβ1)k,0 + b
(rβ1)k,1
β (2r + k β 1) b(rβ1)k,0
= (2r + k β 1) b(r)k,n β
(r2 β 2r + kr β k
) nβj=1
(n
j
)b(rβ1)k,jβ1 + (2β 2r β k) b
(rβ1)k,0 + b
(rβ1)k,1 .
4
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
867 Nazmiye Yilmaz et al 864-873
Then we have
b(r)k,n+1 β (2r + k β 1) b
(r)k,n = β
(r2 β 2r + kr β k
) nβj=1
(n
j
)b(rβ1)k,jβ1 + 4β 2r β k.
(2.2)By taking nβ nβ 1, it is
b(r)k,n = (2r + k β 1) b
(r)k,nβ1 β
(r2 β 2r + kr β k
) nβ1βj=1
(nβ 1
j
)b(rβ1)k,jβ1 + 4β 2r β k
= (2r + k β 1) b(r)k,nβ1 β
(r2 β 2r + kr β k
) nβj=1
[(n
j
)β(nβ 1
j β 1
)]b(rβ1)k,jβ1 + 4β 2r β k
= (2r + k β 1) b(r)k,nβ1 β
(r2 β 2r + kr β k
) nβj=1
(n
j
)b(rβ1)k,jβ1
+(r2 β 2r + kr β k
) nβj=1
(nβ 1
j β 1
)b(rβ1)k,jβ1 + 4β 2r β k
b(r)k,n = (2r + k β 1) b
(r)k,nβ1 β
(r2 β 2r + kr β k
) nβj=1
(n
j
)b(rβ1)k,jβ1
+(r2 β 2r + kr β k
) nβ1βj=0
(nβ 1
j
)b(rβ1)k,j + 4β 2r β k
= (2r + k β 1) b(r)k,nβ1 β
(r2 β 2r + kr β k
) nβj=1
(n
j
)b(rβ1)k,jβ1
+(r2 β 2r + kr β k
)b(r)k,nβ1 + 4β 2r β k
=(r2 + kr β 1
)b(r)k,nβ1 β
(r2 β 2r + kr β k
) nβj=1
(n
j
)b(rβ1)k,jβ1 + 4β 2r β k.
Hence, we have
b(r)k,n β
(r2 + kr β 1
)b(r)k,nβ1 = β
(r2 β 2r + kr β k
) nβj=1
(n
j
)b(rβ1)k,jβ1 + 4β 2rβ k.
If last expression put in place in the equation (2.2), then we get
b(r)k,n+1 = (2r + k β 1) b
(r)k,n + b
(r)k,n β
(r2 + kr β 1
)b(r)k,nβ1
= (2r + k) b(r)k,n β
(r2 + kr β 1
)b(r)k,nβ1
5
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
868 Nazmiye Yilmaz et al 864-873
which completed the proof of this theorem.
The characteristic equation of sequenceb(r)k,n
in (2.1) is
Ξ»2β (2r + k)Ξ»+ r2 + krβ 1 = 0. Let Ξ»1 and Ξ»2 be the roots of this equation.
Then, Binetβs formulas of sequenceb(r)k,n
can be expressed as
b(r)k,n =
(k +βk2 + 4
2+ r
)n
+
(k ββk2 + 4
2+ r
)n
. (2.3)
In here, we obtain the equalities given in [10] in terms of iterated binomialtransform of the k-Lucas sequences as a consequence of Theorem 2.1. To dothat we take r = 1 in Theorem 2.1 and the Eq. (2.3):
bk,n+1 = (2 + k) bk,n β kbk,nβ1,
and
bk,n =
(k + 2 +
βk2 + 4
2
)n
+
(k + 2β
βk2 + 4
2
)n
.
Now, we give the sum of iterated binomial transform for k-Lucas sequences.
Theorem 2.2 Sum of sequenceb(r)k,n
is
nβ1βi=0
b(r)k,i =
(r2 + kr β 1
)b(r)k,nβ1 β b
(r)k,n β k β 2r + 2
r2 + kr β k β 2r.
Proof. By considering Eq. (2.3), we have
nβ1βi=0
b(r)k,i =
nβ1βi=0
(Ξ»i1 + Ξ»i2
).
Then we obtainnβ1βi=0
b(r)k,i =
(Ξ»n1 β 1
Ξ»1 β 1
)+
(Ξ»n2 β 1
Ξ»2 β 1
).
Afterward, by taking into account equations Ξ»1.Ξ»2 = r2 +krβ1 and Ξ»1 +Ξ»2 =k + 2r, we conclude
nβ1βi=0
b(r)k,i =
(r2 + kr β 1
)b(r)k,nβ1 β b
(r)k,n β k β 2r + 2
r2 + kr β k β 2r.
6
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
869 Nazmiye Yilmaz et al 864-873
Note that, if we take r = 1 in Theorem 2.2, we obtain the summation ofbinomial transform for k-Lucas sequence:
nβ1βi=0
bk,i = bk,n β kbk,nβ1 + k
Theorem 2.3 The generating function of the iterated binomial transform forLk,n is
ββi=0
b(r)k,ix
i =2β (2r + k)x
1β (2r + k)x+ (r2 + kr β 1)x2.
Proof. Assume that b (k, x, r) =ββi=0
b(r)k,ix
i is the generating function of the
iterated binomial transform for Lk,n. From Theorem 2.1, we obtain
b (k, x, r) = b(r)k,0 + b
(r)k,1x+
ββi=2
((2r + k) b
(r)k,iβ1 β
(r2 + kr β 1
)b(r)k,iβ2
)xi
= b(r)k,0 + b
(r)k,1xβ (2r + k) b
(r)k,0x+ (2r + k)x
ββi=0
b(r)k,ix
i
β(r2 + kr β 1
)x2ββi=0
b(r)k,ix
i
= b(r)k,0 +
(b(r)k,1 β (2r + k) b
(r)k,0
)x+ (2r + k)xb (k, x, r)
β(r2 + kr β 1
)x2 b (k, x, r) .
Now rearrangement of the equation implies that
b (k, x, r) =b(r)k,0 +
(b(r)k,1 β (2r + k) b
(r)k,0
)x
1β (2r + k)x+ (r2 + kr β 1)x2,
which equals to theββi=0
b(r)k,ix
i in theorem. Hence, the result.
In here, we obtain the generating function given in [10] in terms of iteratedbinomial transform of the k-Lucas sequences as a consequence of Theorem 2.3.To do that we take r = 1 in Theorem 2.3:
ββi=0
bk,ixi =
2β (2 + k)x
1β (2 + k)x+ kx2.
7
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
870 Nazmiye Yilmaz et al 864-873
In the following theorem, we present the relationship between the iteratedbinomial transform of k-Lucas sequence and iterated binomial transform ofk-Fibonacci sequence.
Theorem 2.4 For n > 0, the relationship of between the transformsb(r)k,n
and
c(r)k,n
is illustrated by following way:
b(r)k,n = c
(r)k,n+1 β
(r2 + kr β 1
)c(r)k,nβ1, (2.4)
where b(r)k,n is the iterated binomial transform of k-Lucas sequence and c
(r)k,n is
the iterated binomial transform of k-Fibonacci sequence.
Proof. By using the Eq.(2.4), let be
b(r)k,n = Xc
(r)k,n+1 + Y c
(r)k,nβ1.
If we take n = 1 and 2, we have the systemb(r)k,1 = Xc
(r)k,2 + Y c
(r)k,0,
b(r)k,2 = Xc
(r)k,3 + Y c
(r)k,1.
By considering definition of the iterated binomial transforms for k-Lucas, k-Fibonacci sequence and Cramer rule for the system, we obtain
2r + k = (2r + k)X,k2 + 2rk + 2r2 + 2 =
(3r2 + 3rk + k2 + 1
)X + Y
andX = 1 and Y = β
(r2 + kr β 1
)which is completed the proof of this theorem.
Note that, if we take r = 1 in Theorem 2.4, we obtain the relationshipof between the binomial transform for k-Lucas sequence and the binomialtransform for k-Fibonacci sequence:
bk,n = ck,n+1 β kck,nβ1.
Corollary 2.1 We should note that choosing k = 1 in the all results of section2, it is actually obtained some properties of the iterated binomial transform forclassical Lucas sequence such that the recurrence relation, Binet formula, sum-mation, generating function and relationship of between binomial transformsfor Fibonacci and Lucas sequences.
8
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
871 Nazmiye Yilmaz et al 864-873
Corollary 2.2 We should note that choosing k = 2 in the all results of section2, it is actually obtained some properties of the iterated binomial transform forclassical Pell-Lucas sequence such that the recurrence relation, Binet formula,summation, generating function and relationship of between binomial trans-forms for Pell and Pell-Lucas sequences.
Conclusion 2.1 In this paper, we define the iterated binomial transform fork-Lucas sequence and present some properties of this transform. By the resultsin Sections 2 of this paper, we have a great opportunity to compare and obtainsome new properties over this transform. This is the main aim of this paper.Thus, we extend some recent result in the literature.
In the future studies on the iterated binomial transform for number se-quences, we expect that the following topics will bring a new insight.
(1) It would be interesting to study the iterated binomial transform for Fi-bonacci and Lucas matrix sequences,
(2) Also, it would be interesting to study the iterated binomial transform forPell and Pell-Lucas matrix sequences.
Acknowledgement 2.1 A part of this study presented at Sharjah-BAE βTheSecond International Conference on Mathematics and Statistics (AUS-ICMSβ15)β.This research is supported by TUBITAK and Selcuk University Scientific Re-search Project Coordinatorship (BAP).
References
[1] T. KOSHY, Fibonacci and Lucas Numbers with Applications, John Wileyand Sons Inc, NY, 2001.
[2] S. FALCON and A. PLAZA, On k-Fibonacci numbers of arithmetic in-dexes, Applied Mathematics and Computation, 208, 2009, 180-185.
[3] S. FALCON, On the k-Lucas number, International Journal of Contem-porary Mathematical Sciences, Vol. 6, no. 21, 2011, 1039 - 1050.
[4] L. MAREK-CRNJAC, The mass spectrum of high energy elementaryparticles via El Naschieβs golden mean nested oscillators, the Dunkerly-Southwell eigenvalue theorems and KAM, Chaos, Solutions & Fractals,18(1), 2003, 125-133.
9
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
872 Nazmiye Yilmaz et al 864-873
[5] H. PRODINGER, Some information about the binomial transform, TheFibonacci Quarterly, 32 (5), 1994, 412-415.
[6] K.W. CHEN, Identities from the binomial transform, Journal of NumberTheory, 124, 2007, 142-150.
[7] S. FALCON and A. PLAZA, Binomial Transforms of k-Fibonacci Se-quence, International Journal of Nonlinear Sciences and Numerical Sim-ulation, 10(11-12), 2009, 1527-1538.
[8] N. YILMAZ and N. TASKARA, Binomial Transforms of the Padovanand Perrin Matrix Sequences, Abstract and Applied Analysis, ArticleNumber: 497418 Published: 2013.
[9] Y. YAZLIK, N. YILMAZ, N. TASKARA, The Generalized (s, t)-matrixSequenceβs Binomial Transforms, Gen. Math. Notes, Vol. 24, No. 1, 2014,pp.127-136.
[10] P. BHADOURIA, D. JHALA, B. SINGH, Binomial Transforms of the k-Lucas Sequences and its Properties, Journal of mathematics and computerscience, 8, 2014, 81-92.
[11] S. FALCON, Iterated Binomial Transforms of the k-Fibonacci Sequence,British Journal of Mathematics & Computer Science, 4(22), 2014, 3135-3145.
[12] N. YILMAZ, N. TASKARA, On the Properties of Iterated BinomialTransforms for the Padovan and Perrin Matrix Sequences, MediterraneanJournal of Mathematics, DOI 10.1007/s00009-015-0612-5, 2015.
[13] Y. YAZLIK and N. TASKARA, A note on generalized k-Horadam se-quence, Computers & Mathematics with Applications, 63, 2012, 36-41.
[14] Y. YAZLIK and N. TASKARA, On the Inverse of Circulant Matrix viaGeneralized k-Horadam Numbers, Applied Mathematics and Computa-tion, 223, 2013, 191-196.
[15] Y. YAZLIK and N. TASKARA, Spectral norm, Eigenvalues and Determi-nant of Circulant Matrix involving the Generalized k-Horadam numbers,Ars Combinatoria, 104, 2012, 505-512.
[16] Y. YAZLIK and N. TASKARA, On the norms of an r-circulant matrixwith the generalized k-Horadam numbers, Journal of Inequalities and Ap-plications, 2013.1, 2013, 1-8.
10
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
873 Nazmiye Yilmaz et al 864-873
Nielsen xed point theory for digital images
Ozgur EGE1, Ismet KARACA2
1 Celal Bayar University, Department of Mathematics, 45140 Manisa, Turkey
E-mail: [email protected] Ege University, Department of Mathematics, 35100 Izmir, Turkey
E-mail: [email protected]
September 1, 2015
Abstract
In this paper, we introduce the Nielsen xed point theory in digital images. We also deal with someimportant properties of the Nielsen number and calculate the Nielsen number of some digital images. We getsome new results using digital covering maps and Nielsen number.
Keywords: Fixed point, Nielsen number, digital homotopy.2010 Mathematics Subject Classication. 47H10, 54H25, 68U10.
1 Introduction
Digital topology is often used in computer graphics, pattern recognition and image processing. This topic has been studiedby important researchers such as Rosenfeld, Kong, Kopperman, Boxer, Karaca, Han, etc. Their goal is to determine notonly similarities but also dierences between digital images and topology.
Fixed point theory with applications is an important area in topology. This theory continues to develop with newcomputations and come out of new invariants. Nielsen xed point theorem is a notable theorem in this theory because itgives a way to count xed points. One of the main goals in digital topology is to classify digital images. For this reason,we use the Nielsen number which is a powerful invariant for digital images.
In 1920s, Jakop Nielsen introduced the Nielsen theory and the Nielsen number. He focused on both the existenceproblem of xed points and the problem of determining the minimal number of xed points in the homotopy classes. Hedid this by introducing the Nielsen number of a self map. This number is a homotopy invariant lower bound for the numberof xed points of the map. In this area, there are signicant works [10, 11, 12, 16, 17, 18, 19].
Boxer [6] introduces the digital covering space and showed that the existence of digital universal covering spaces. Boxerand Karaca [7] classify digital covering spaces using the conjugacy class corresponding to a digital covering space. Boxerand Karaca [8] study digital versions of some properties of covering spaces from algebraic topology. Karaca and Ege [20]get some results related to the simplicial homology groups of 2D digital images. Ege and Karaca [13] give characteristicproperties of the simplicial homology groups.
This paper is organized as follows. The second section provides the general notions of digital images, digital homotopy,digital covering spaces and digital homology groups. In Section 3 we present the Nielsen xed point theorem for digitalimages, give some examples and properties. In Section 4 we discuss about the relation between Nielsen theory and digitaluniversal covering space. We nally make some conclusions about this topic.
2 Preliminaries
A digital image consists of a pair (X,ΞΊ), where Z is the set of integers, X β Zn for some positive integer n, and ΞΊ indicatesan adjacency relation for the members of X.
Denition 2.1. [3]. For a positive integer l with 1 β€ l β€ n and two distinct points p = (p1, p2, . . . , pn), q = (q1, q2, . . . , qn) βZn, p and q are cl-adjacent, if(1) there are at most l indices i such that |pi β qi| = 1, and
(2) for all other indices j such that |pj β qj | 6= 1, pj = qj .
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
874 Ozgur EGE et al 874-880
2
The notation cl represents the number of points q β Zn that are adjacent to a given point p β Zn. Thus, in Z,we have c1 = 2-adjacency; in Z2, we have c1 = 4-adjacency and c2 = 8-adjacency; in Z3, we have c1 = 6-adjacency,c2 = 18-adjacency, and c3 = 26-adjacency [5]. A ΞΊ-neighbor of p β Zn [3] is a point of Zn that is ΞΊ-adjacent to p.
A digital interval [2] is dened by [a, b]Z = z β Z | a β€ z β€ b where a, b β Z and a < b. A digital image X β Zn isΞΊ-connected [15] if and only if for every pair of dierent points x, y β X, there is a set x0, x1, . . . , xr of points of a digitalimage X such that x = x0, y = xr and xi and xi+1 are ΞΊ-neighbors where i = 0, 1, . . . , r β 1.
Denition 2.2. [3]. Let X β Zn0 and Y β Zn1 be digital images with ΞΊ0-adjacency and ΞΊ1-adjacency, respectively. Afunction f : X ββ Y is said to be (ΞΊ0, ΞΊ1)-continuous if for every ΞΊ0-connected subset U of X, f(U) is a ΞΊ1-connectedsubset of Y . We say that such a function is digitally continuous.
Proposition 2.3. [3]. Let X β Zn0 and Y β Zn1 be digital images with ΞΊ0-adjacency and ΞΊ1-adjacency, respectively. Thefunction f : X β Y is (ΞΊ0, ΞΊ1)-continuous if and only if for every ΞΊ0-adjacent points x0, x1 of X, either f(x0) = f(x1)or f(x0) and f(x1) are ΞΊ1-adjacent in Y .
In a digital image X, if there is a (2, ΞΊ)-continuous function f : [0,m]Z β X such that f(0) = x and f(m) = y, thenthere exists a digital ΞΊ-path [6] from x to y. If f(0) = f(m), then we say that f is digital ΞΊ-loop and the point f(0) is thebase point of the loop f . When a digital loop f is a constant function, it is said to be a trivial loop.
Denition 2.4. Let (X,ΞΊ0) β Zn0 and (Y, ΞΊ1) β Zn1 be digital images. A function f : X β Y is called a (ΞΊ0, ΞΊ1)-isomorphism [2] if f is (ΞΊ0, ΞΊ1)-continuous and bijective and fβ1 : Y β X is (ΞΊ1, ΞΊ0)-continuous.
Denition 2.5. [3]. Let (X,ΞΊ0) β Zn0 and (Y, ΞΊ1) β Zn1 be digital images. We say that two (ΞΊ0, ΞΊ1)-continuous functionsf, g : X β Y are digitally (ΞΊ0, ΞΊ1)-homotopic in Y if there is a positive integer m and a function H : X Γ [0,m]Z β Y suchthatβ’ for all x β X, H(x, 0) = f(x) and H(x,m) = g(x);β’ for all x β X, the induced function Hx : [0,m]Z β Y dened by
Hx(t) = H(x, t) for all t β [0,m]Z,
is (2, ΞΊ1)-continuous; andβ’ for all t β [0,m]Z, the induced function Ht : X β Y dened by
Ht(x) = H(x, t) for all x β X,
is (ΞΊ0, ΞΊ1)-continuous.The function H is called a digital (ΞΊ0, ΞΊ1)-homotopy between f and g. If these functions are digitally (ΞΊ0, ΞΊ1)-homotopic,
it is denoted f '(ΞΊ0,ΞΊ1) g. The digital (ΞΊ0, ΞΊ1)-homotopy relation [3] is equivalence among digitally continuous functionsf : (X,ΞΊ0)β (Y, ΞΊ1).
If f : [0,m1]Z β X and g : [0,m2]Z β X are digital ΞΊ-paths with f(m1) = g(0), then dene the product (f β g) :[0,m1 +m2]Z β X [3] by
(f β g)(t) =f(t), t β [0,m1]Zg(tβm1), t β [m1,m1 +m2]Z.
Let f and fβ²be ΞΊ-loops in a digital image (X,x0). We say f
β²is a trivial extension of f [3] if there are sets of ΞΊ-paths
f1, . . . , fr and F1, . . . , Fp in X such that:(1) r β€ p,(2) f = f1 β . . . β fr,(3) f
β²= F1 β . . . β Fp,
(4) There are indices 1 β€ i1 < i2 < . . . < ir β€ p such that Fij = fj , 1 β€ j β€ r and i 6= i1, . . . , ir implies Fi is a trivialloop.
If f, g : [0,m]Z β X are ΞΊ-paths such that f(0) = g(0) and f(m) = g(m), then a homotopy
H : [0,m]Z Γ [0,M ]Z β X
between f and g such that for all t β [0,M ]Z, H(0, t) = f(0) and H(m, t) = f(m), holds the endpoints xed.Two loops f, f0 with the same base point x0 β X belong to the same loop class [f ]X if they have trivial extensions that
can be joined by a homotopy that holds the endpoints xed (see [4]).Let (E, ΞΊ) be a digital image and let Ξ΅ be a positive integer. The ΞΊ-neighborhood of e0 β E with radius Ξ΅ is the set
NΞΊ(e0, Ξ΅) = e β E | lΞΊ(e0, e) β€ Ξ΅ βͺ e0,
where lΞΊ(e0, e) is the length of a shortest ΞΊ-path from e0 to e in E (see [14]).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
875 Ozgur EGE et al 874-880
3
Denition 2.6. [6]. Let (E, ΞΊ0) and (B, ΞΊ1) be digital images. A map p : E β B is called a (ΞΊ0, ΞΊ1)-covering map if thefollowings are true:1. p is a (ΞΊ0, ΞΊ1)-continuous surjection.2. for each b β B, there exists an indexing set M such that pβ1(b) can be indexed as pβ1(b) = ei|i βM and the followingconditions hold:β’ pβ1(NΞΊ1(b, 1)) =
βiβM NΞΊ0(ei, 1),
β’ if i, j βM , i 6= j, then NΞΊ0(ei, 1) β©NΞΊ0(ej , 1) = β ,β’ the restriction map p|NΞΊ0 (ei,1) : NΞΊ0(ei, 1)β NΞΊ1(b, 1) is a (ΞΊ0, ΞΊ1)-isomorphism for all i βM .
Let (E, ΞΊ0), (B, ΞΊ1) and (X,ΞΊ2) be digital images, let p : E β B be a (ΞΊ0, ΞΊ1)-covering map, and f : X β B be(ΞΊ2, ΞΊ1)-continuous. A lifting of f with respect to p is a (ΞΊ2, ΞΊ0)-continuous function f : X β E such that p f = f (see[14]).
Denition 2.7. [6]. Let (E0, p0, B) be a (ΞΊ0, ΞΊB)-covering. Suppose C is a set of (ΞΊE , ΞΊB)-coverings of B such that forevery (E, p,B) β C, there is a (ΞΊ0, ΞΊE)-covering (E0, pE , E). Then the pair (E0, p0) is a universal covering space of B forthe set C.
Denition 2.8. [21]. Let S be a set of nonempty subset of a digital image (X,ΞΊ). Then the members of S are calledsimplexes of (X,ΞΊ), if the followings hold:β’ If p and q are distinct points of s β S, then p and q are ΞΊ-adjacent,β’ If s β S and β 6= t β s, then t β S.
An m-simplex is a simplex S such that |S| = m + 1. For a digital m-simplex P , if Pβ²is a nonempty proper subset of
P , then Pβ²is called a face of P .
Denition 2.9. [1]. Let (X,ΞΊ) be a nite collection of digital m-simplices, 0 β€ m β€ d for some non-negative integer d. Ifthe followings hold, then (X,ΞΊ) is called a nite digital simplicial complex :β’ If P belongs to X, then every face of P also belongs to X.β’ If P,Q β X, then P β©Q is either empty or a common face of P and Q.
Denition 2.10. [1]. Let (X,ΞΊ) β Zn be a digital oriented simplicial complex with m-dimension. CΞΊq (X) is a free abeliangroup with basis all digital (ΞΊ, q)-simplices in X. A homomorphism βq : C
ΞΊq (X) ββ CΞΊqβ1(X) called the boundary operator.
If Ο = [v0, . . . , vq] is an oriented simplex with 0 < q β€ m, βq is dened by
βqΟ = βq[v0, . . . , vq] =
qβi=0
(β1)i[v0, . . . , vi, . . . , vq]
where vi means the vertex vi is to be deleted from the array.
We remark that for q < 0,m < q, since CΞΊq (X) is the trivial group, the operator βq is the trivial homomorphism forq β€ 0, m < q. We notice that βqβ1 βq = 0 [1] for q β₯ 0.
Denition 2.11. [1]. Let (X,ΞΊ) β Zn be a digital oriented simplicial complex with m-dimension.β’ ZΞΊq (X) = Ker βq is called the group of digital simplicial q-cycles.β’ BΞΊq (X) = Im βq+1 is called the group of digital simplicial q-boundaries.β’ HΞΊ
q (X) = ZΞΊq (X)/BΞΊq (X) is called the qth digital simplicial homology group.
3 Nielsen Theory for Digital Images
Let (X,ΞΊ) be a digital image and let f : X β X be a digital map. The xed point set of f is Fix(f) = x β X : f(x) = x.The main object of study in topological xed point theory is the minimum number of xed points which is denoted byM [f ] among all digital maps (ΞΊ, ΞΊ)-homotopic to f . For example, M [f ] = 0 means that there is a digital map g which is(ΞΊ, ΞΊ)-homotopic to f such that g(x) 6= x for all x β X.
To calculate M [f ] we have to examine the xed point sets of every map homotopic to f . In the xed point theory, it ismade use of a homotopy invariant, called the Nielsen number of f . Its computation requires only a knowledge of the mapf itself.
Denition 3.1. Let f : (X,ΞΊ1) ββ (Y, ΞΊ2) be a (ΞΊ1, ΞΊ2)-continuous map where (X,ΞΊ1) and (Y, ΞΊ2) are digital images.Then f induces homomorphisms fβ : H
ΞΊ1β (X) ββ HΞΊ2
β (Y ) and fβ can be thought of as a homomorphisms of the integers.The integer deg(f) to which the number 1 gets sent is called the degree of the map f .
Denition 3.2. Let (X,ΞΊ) be a digital image, A β X and f : Aβ X a digital map. We dene the xed point index of fas ind(f) = deg(F ) where F (x) = xβ f(x) and x β X.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
876 Ozgur EGE et al 874-880
4
Some properties of xed point index can be given as follows. We don't prove these because they are proved similarly inAlgebraic Topology.
1. (Homotopy invariance) Let A β X Γ [0,m]Z be digital image with ΞΊ-adjacency and F : Aβ X be a digital map suchthat
Fix(F ) = (x, t) β A : F (x, t) = x.Then ind(f0) = ind(fm), where ft = F (β, t) for 0 β€ t < m and a positive integer m.
2. (Commutativity) Let (X,ΞΊ1) and (Y, ΞΊ2) be digital images and let f : A β Y and g : B β X be digital (ΞΊ1, ΞΊ2)-continuous maps, respectively, where A β X, B β Y . Then Fix(gf) = Fix(fg) and ind(fg) = ind(gf).
Now we dene Nielsen number for digital images.
Denition 3.3. Let (X,ΞΊ) be a digital image and f : (X,ΞΊ) β (X,ΞΊ) a self-map. Two xed points x, y β Fix(f) areNielsen related if and only if there is a ΞΊ-path c : [0,m]Z β X satisfying c(0) = x, c(m) = y and the ΞΊ-paths c, f c arexed end point homotopic, i.e. there is a digital map H : [0, n]Z Γ [0,m]Z β X satisfying
H(t, 0) = c(t), H(t,m) = f c(t), H(0, s) = x, H(n, s) = y.
This is an equivalence relation, hence Fix(f) splits into disjoint Nielsen classes. A xed point class F is essential if itsindex is nonzero. The number of essential xed point classes is called the Nielsen number of f , denoted N(f).
We give some characteristic examples about the Nielsen number.
Example 3.4. Let (X,ΞΊ) be a digital image. If f : X β X is a constant digital map, then N(f) = 1.
Since the boundary Bd(In+1) of an (n+1)-cube In+1 is homeomorphic to n-sphere Sn, we can represent a digital sphereby using the boundary of a digital cube. Boxer [5] denes sphere-like digital image as Sn = [β1, 1]n+1
Z \ 0n+1 β Zn+1,where 0n denotes the origin of Zn.
Example 3.5. S1 = c0 = (1, 0), c1 = (1, 1), c2 = (0, 1), c3 = (β1, 1), c4 = (β1, 0), c5 = (β1,β1), c6 = (0,β1), c7 =(1,β1) is digital 1-sphere with 4-adjacency in Z2. Let f : (S1, 4) β (S1, 4) be a digital map of degree 1. Then f can beconsidered as identity map and is (4, 4)-homotopic to a xed point free map. Thus we have N(f) = 0.
Let's give some important properties of Nielsen number for digital images.
Theorem 3.6. Let (X,ΞΊ) be any digital image. If f '(ΞΊ,ΞΊ) g : X β X, then N(f) = N(g).
Proof. We must show that there is a bijection between sets of essential classes of f and g. Let H(t, s) be a digital (ΞΊ, ΞΊ)-
homotopy from f and g. For every Nielsen class A β Fix(f), there is one Aβ²β Fix(H) containing A. Let
B = x β X | (x,m) β Aβ².
So B is a Nielsen class of g or is empty. From homotopy invariance index property, we have ind(f,A) = ind(g,B). If A isessential, then B is essential. As a result, we nd a map from the set of essential classes of f to the set of essential classesof g.
On the other hand, H(x,mβ t) gives the inverse map. Consequently, we get N(f) = N(g).
Theorem 3.7. Let (X,ΞΊ) be a digital image and f : X β X be a digital map. Any digital map g digital (ΞΊ, ΞΊ)-homotopicto f has at least N(f) xed points.
Proof. Using Theorem 3.6, we have N(f) = N(g). Since each essential Nielsen class of g is nonempty, we get M [g] β₯ N(g)where
M [g] = min#Fixg | g '(ΞΊ,ΞΊ) f : X β X.
Theorem 3.8. Let (X,ΞΊ) and (Y, ΞΊβ²) be any digital images, f : X β Y and g : Y β X be digital maps. Then N(g f) =
N(f g).
Proof. For digital maps f and g, if we use commutativity property of xed point index, i.e. Fix(f g) = Fix(g f) andind(f g) = ind(g f), we have a bijection which preserves index between the sets of essential Nielsen classes. As a result,we have N(g f) = N(f g).
Lemma 3.9. Let A β X be digital image with ΞΊ-adjacency and f : X β X be digital map such that f(X) β A, where Xis any digital image with ΞΊ-adjacency. If fA : Aβ A is the restriction of f , then N(fA) = N(f).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
877 Ozgur EGE et al 874-880
5
Proof. Let i : Aβ X be inclusion map. Assume that g : X β A is given by g(x) = f(x). Using Theorem 3.8, we conclude
N(f) = N(i g) = N(g i) = N(fA)
because i g = f and g i = fA.
Theorem 3.10. Let (X,ΞΊ) and (Y, ΞΊβ²) be any two digital images. Suppose that h : X β Y is a digital (ΞΊ, ΞΊ
β²)-homotopy
equivalence and let the diagram
Xf //
h
X
h
Y
g// Y
be digital (ΞΊ, ΞΊβ²)-homotopy commutative, i.e. h f '(ΞΊ,ΞΊ
β²) g h. Then N(f) = N(g).
Proof. Assume that the digital (ΞΊβ², ΞΊ)-homotopy inverse of h is m : Y β X. Then m h '(ΞΊ,ΞΊ) 1X and h m '(ΞΊ
β²,ΞΊ
β²) 1Y .
By Theorem 3.6 and Theorem 3.8, we have
N(f) = N(f(mh)) = N((fm)h) = N(h(fm))
= N((hf)m)
= N((gh)m)
= N(g(hm))
= N(g).
Theorem 3.11. Let (X,ΞΊ) be a digital image and f : (X,ΞΊ) β (X,ΞΊ) be a digital map. N(f) is a lower bound for thenumber of xed points in the homotopy class of f , i.e.
0 β€ N(f) β€M [f ] := min#Fixg | g '(ΞΊ,ΞΊ) f : X β X.
Proof. By the denition of Nielsen number, we have N(f) β₯ 0. On the other hand, since we know that each Nielsen classcontains at least one xed point of f , we conclude that 0 β€ N(f) β€M [f ].
4 Nielsen Theory and Digital Universal Covering Spaces
Since there is a connection between the digital fundamental group and the digital universal covering of a space, same resultscan be also obtained by the lifts of the considered maps to the digital universal coverings. Let (X,ΞΊ) be a digital image andp : X β X be the digital universal covering of X, with group Ο of covering transformations. Let f : X β X be a lifting off , i.e., have a commutative diagram
Xf //
p
X
p
X
f// X
If fβ²is another lifting of f , then f
β²= Ξ± f for some Ξ± β Ο. The set of all liftings of f is Ξ± f | Ξ± β Ο.
For any Ξ± β Ο, f Ξ± is a lifting of f and so we have Ξ±β² f = f Ξ± for some Ξ±
β²β Ο. This denes a homomorphism
Ο : Ο β Ο given by Ο(Ξ±) = Ξ±β². Dene the Reidemeister action of Ο on Ο as follows:
Ο Γ Ο β Ο
(Ξ³, Ξ±) 7β Ξ³Ξ±Ο(Ξ³)β1,
where Ξ³, Ξ± β Ο. This denes an equivalence relation. We say that the Reidemeister classes of its equivalence classes. Theset of the Reidemeister classes determined by Ο is denoted by R[Ο] = [Ξ±] | Ξ± β Ο.
Theorem 4.1. Let (X,ΞΊ) be a digital image, f : X β X be a digital map and f : X β X be a lifting of f . Then [Ξ±] = [Ξ±β²]
if and only if p(Fix(Ξ± f)) = p(Fix(Ξ±β² f)), where p : X β X is a digital covering map of f and Ξ±, Ξ±
β²β Ο.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
878 Ozgur EGE et al 874-880
6
Proof. For necessary condition, since the xed point sets of any two digital homotopic maps are same, we conclude that
[Ξ±] = [Ξ±β²]β Ξ± '(ΞΊ,ΞΊ) Ξ±
β²
β Ξ± f '(ΞΊ,ΞΊ) Ξ±β² f
β Fix(Ξ± f) = Fix(Ξ±β² f)
β p(Fix(Ξ± f)) = p(Fix(Ξ±β² f)).
For sucient condition, let a β p(Fix(Ξ± f)) = p(Fix(Ξ±β² f)). Then we have p(a) = a and p(aβ²) = a. Moreover, we get
Ξ± f(a) = a = 1X(a) and Ξ±β² f(a) = aβ² = 1X(aβ²)
Finally, we can say
Ξ±(f(a)) = Ξ±β²(f(a)) β Ξ± = Ξ±
β²β Ξ± '(ΞΊ,ΞΊ) Ξ±
β²,
where a is any point of X. As a result, we have [Ξ±] = [Ξ±β²].
Corollary 4.2. If p(Fix(Ξ±f)) is any xed point class, then
Fix(f) =β
[Ξ±]βR[Ο]
p(Fix(Ξ±f)),
where [Ξ±] is a Reidemeister class.
Lemma 4.3. Let (X,ΞΊ), (X, ΞΊ) be digital images, f : X β X be a digital map and f : X β X be any lifting of f . Thenany two points in p(Fix(f)) β Fix(f) are Nielsen related. We have also
Fix(f) =βΛfβ²
p(Fix(f β²))
where f β² is a lifting of f .
Proof. Let a and b be any two points in p(Fix(f)). We say that
p(a) = a, p(b) = b, f(a) = a, f(b) = b,
for some a, b β Fix(f). We denote a ΞΊ-path from a to b in X by ΞΈ. There is a digital homotopy H such that
H : X Γ [0,m]Z β X
H(x, 0) = ΞΈ and H(x,m) = f ΞΈbecause X is ΞΊ-connected digital image. Then we have p H = H : X Γ [0,mβ²]Z β X is a digital homotopy between twoΞΊ-paths
ΞΈ = p ΞΈ and f ΞΈ = p (f ΞΈ)which join two points a, b β Fix(f). As a result, a and b are Nielsen related points.
Now we prove the latter statement. Let u β Fix(f) and u β pβ1(u). Then f(u) = u and pβ1(u) = u. Moreover,
f β²(u) = u because f β² is a lifting of f . We can say the following result.
f β²(u) = u β p f β²(u) = p(u) = u β u β p(Fix(f β²)).
Consequently, we have Fix(f) =βΛfβ²
p(Fix(f β²)).
Let p : X β X be a digital universal covering map. Let
OX = Ξ± β X β X : p Ξ± = p
denote the group of deck transformations of this digital covering map.
Lemma 4.4. Let C be the set of liftings of f and f , f β² β C. If p(Fix(f)) = p(Fix(f β²)) 6= β , then there is an Ξ± β OX such
that Ξ± f = f β² Ξ±.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
879 Ozgur EGE et al 874-880
7
Proof. If p(Fix(f)) = p(Fix(f β²)) 6= β , then there are two points x, xβ² such that p(x) = p(xβ²) where
x β Fix(f) β f(x) = x and xβ² β Fix(f β²) β f β²(xβ²) = xβ² .
Since Ξ± β OX , i.e. p Ξ± = p, we have Ξ±(x) = xβ² . We conclude that
f β² Ξ±(x) = f β²(xβ²) = xβ² = Ξ±(x) = Ξ±(f(x)) = Ξ± f(x).
As a result, we have Ξ±f = f β²Ξ±.
Lemma 4.5. Let f β² = Ξ±fΞ±β1 for an Ξ± β OX . Then p(Fix(f)) = p(Fix(f β²)).
Proof. By assumption, we have p(Ξ±(x)) = p(x). If u β p(Fix(f)), then p(x) = u and f(x) = x. Since
p f(x) = u β p Ξ±β1 f β² Ξ±(x) = p f β²(xβ²) = u,
we have u β p(Fix(f β²)). As a result, p(Fix(f)) = p(Fix(f β²)).
5 Conclusion
The essential aim of this paper is to determine xed point properties for a digital image. This work can play an importantrole in digital images because Nielsen theory gives an information about the number of xed points of a map. Since theNielsen number is a powerful invariant in digital images, we think that this work will be useful for xed point theory,especially Nielsen theory.
References
[1] H. Arslan, I. Karaca and A. Oztel, Homology groups of n-dimensional digital images, XXI. Turkish National Mathe-matics Symposium, B, 113 (2008).
[2] L. Boxer, Digitally continuous functions, Pattern Recognition Letters, 15, 833839 (1994).
[3] L. Boxer, A classical construction for the digital fundamental group, Journal of Mathematical Imaging and Vision, 10,5162 (1999).
[4] L. Boxer, Properties of digital homotopy, Journal of Mathematical Imaging and Vision, 22, 1926 (2005).
[5] L. Boxer, Homotopy properties of sphere-like digital images, Journal of Mathematical Imaging and Vision, 24, 167175(2006).
[6] L. Boxer, Digital products, wedges and covering spaces, Journal of Mathematical Imaging and Vision, 25, 159171(2006).
[7] L. Boxer and I. Karaca, The classication of digital covering spaces, Journal of Mathematical Imaging and Vision, 32,2329 (2008).
[8] L. Boxer and I. Karaca, Some properties of digital covering spaces, Journal of Mathematical Imaging and Vision, 37,1726 (2010).
[9] L. Boxer, I. Karaca, A. Oztel, Topological invariants in digital images, Journal of Mathematical Sciences: Advancesand Applications, 11(2), 109140 (2011).
[10] R.F. Brown, Retraction methods in Nielsen xed point theory, Pacic Journal of Mathematics, 115(2), 277297(1984).
[11] R.F. Brown, Nielsen xed point theory on manifolds, Banach Center Publications, 49(1), 1927 (1999).
[12] R.F. Brown, M. Furi, L. Gorniewicz and B. Jiang, Handbook of Topological Fixed Point Theory, Springer (2005).
[13] O. Ege and I. Karaca, Fundamental properties of simplicial homology groups for digital images, American Journal ofComputer Technology and Application, 1(2), 2542 (2013).
[14] S.E. Han, Non-product property of the digital fundamental group, Information Sciences, 171, 7391 (2005).
[15] G.T. Herman, Oriented surfaces in digital spaces, CVGIP:Graphical Models and Image Processing, 55, 381396 (1993).
[16] J. Jezierski, The Nielsen number product formula for coincidence, Fund. Math., 134, 183212 (1989).
[17] J. Jezierski and W. Marzantowicz, Homotopy Methods in Topological Fixed and Periodic Points Theory, Springer(2006).
[18] B. Jiang, Estimation of the Nielsen numbers, Chinese Math., 5, 330339 (1964).
[19] B. Jiang, Lectures on Nielsen xed point theory, Contemp. Math., 14 (1983).
[20] I. Karaca and O. Ege, Some results on simplicial homology groups of 2D digital images, International Journal ofInformation and Computer Science, 1(8), 198203 (2012).
[21] E. Spanier, Algebraic Topology, McGraw-Hill, New York (1966).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
880 Ozgur EGE et al 874-880
A FIXED POINT THEOREM AND STABILITY OF
ADDITIVE-CUBIC FUNCTIONAL EQUATIONS IN MODULAR
SPACES
CHANG IL KIM, GILJUN HANβ, AND SEONG-A SHIM
Abstract. In this paper, we investigate a fixed point theorem for a mapping
without the condition of bounded orbit in a modular space, whose inducedmodular is lower semi-continunous. Using this fixed point theorem, we prove
the generalized Hyers-Ulam stability for an additive-cubic functional equation
in modular spaces without 42-conditions and the convexity.
1. Introduction and preliminaries
The question of stability for a generic functional equation was originated in 1940by Ulam [14]. Concerning a group homomorphism, Ulam posted the question ask-ing how likely to an automorphism a function should behave in order to guaranteethe existence of an automorphism near such functions. Hyers [3] gave the first affir-mative partial answer to the question of Ulam for Banach spaces. Hyersβ theoremwas generalized by Aoki [1] for additive mappings and by Rassias [12] for linearmappings by considering an unbounded Cauchy difference, the latter of which hasinfluenced many developments in the stability theory. This area is then referred toas the generalized Hyers-Ulam stability. In 1994, P. Gavruta [2] generalized thesetheorems for approximate additive mappings controlled by the unbounded Cauchydifference with regular conditions.
A problem that mathematicians has dealt with is βhow to generalize the classicalfunction space Lpβ. A first attempt was made by Birnhaum and Orlicz in 1931. Thisgeneralization found many applications in differential and intergral equations withkernls of nonpower types. The more abstract generalization was given by Nakano[10] in 1950 based on replacing the particular integral form of the functional byan abstract one that satisfies some good properties. This functional was calledmodular. This idea was refined, generalized by Musielak and Orlicz [8] in 1959 andstudied by many authors ([4], [7], [11], [19]).
Recently, Sadeghi [13] presented a fixed point method to prove the general-ized Hyers-Ulam stability of functional equations in modular spaces with the 42-condition and Wongkum, Chaipunya, and Kumam [15] proved the fixed point theo-rem and the generalized Hyers-Ulam stability for quadratic mappings in a modularspace whose modular is convex, lower semi-continuous but do not satisfy the 42-condition.
2010 Mathematics Subject Classification. 39B52, 39B72, 47H09.Key words and phrases. fixed point theorem, stability, additive-cubic functional equation,
modular space.* Corresponding author.
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
881 CHANG IL KIM et al 881-893
2 CHANG IL KIM, GILJUN HAN, AND SEONG-A SHIM
In this paper, we investigate a fixed point theorem in modular spaces, whoseinduced modular is lower semi-continuous, for a mapping with some conditionsin place of the condition of bounded orbit and using this fixed point theorem,we will prove the generalized Hyers-Ulam stability for the following additive-cubicfunctional equation
(1.1) f(2x+ y) + f(2xβ y)β 2f(x+ y)β 2f(xβ y)β 2f(2x) + 4f(x) = 0
in modular spaces without 42-conditions and the convexity.In fact, the equation (1.1) has been studied in various spaces. For example, in
quasi-Banach spaces ([9]), in F-spaces ([16]), in non-Archimedean fuzzy normedspaces ([17]), and in intuitionistic fuzzy normed spaces ([18]), etc. Unlike Banachspaces and F-spaces, due to the absence of the triangle inequality in modular spaces,we need subtle caculations in the proofs of Lemma 1.4 and Theorem 2.2.
Definition 1.1. Let X be a vector space over a field K(R or C).(1) A generalized functional Ο : X ββ [0,β] is called a modular if
(M1) Ο(x) = 0 if and only if x = 0,(M2) Ο(Ξ±x) = Ο(x) for every scalar Ξ± with |Ξ±| = 1, and(M3) Ο(Ξ±x + Ξ²y) β€ Ο(x) + Ο(y) for all x, y β X and for all nonnegative real
numbers Ξ±, Ξ² with Ξ±+ Ξ² = 1.(2) If (M3) is replaced by
(M4) Ο(Ξ±x+ Ξ²y) β€ Ξ±Ο(x) + Ξ²Ο(y)for all x, y β X and for all nonnegative real numbers Ξ±, Ξ² with Ξ±+ Ξ² = 1, then wesay that Ο is a convex modular.
Remark 1.2. Let Ο be a modular on a vector space X. Then by (M1) and (M3),we can easily show that for any positive real number Ξ΄ with Ξ΄ < 1,
Ο(Ξ΄x) β€ Ο(x)
for all x β X.
For any modular Ο on X, the modular space XΟ is defined by
XΟ := x β X | Ο(Ξ»x)β 0 as Ξ»β 0.
Let XΟ be a modular space and let xn be a sequence in XΟ. Then (i) xnis called Ο-convergent to a point x β XΟ if Ο(xn β x) β 0 as n β β, (ii) xn iscalled Ο-Cauchy if for any Ξ΅ > 0, there is a k β N such that Ο(xn β xm) < Ξ΅ forall m,n β N with n,m β₯ k, and (iii) a subset K of XΟ is called Ο-complete if eachΟ-Cauchy sequence is Ο-convergent to an element of K.
Another unnatural behavior one usually encounter is that the convergence of asequence xn to x does not imply that cxn converges to cx for some c β K.Thus, many mathematicians imposed some additional conditions for a modular tomeet in order to make the multiples of xn converge naturally. Such preferencesare referred to mostly under the term related to the 42-conditions.
A modular space XΟ is said to satisfy the 42-condition if there exists k β₯ 2 suchthat Ο(2x) β€ kΟ(x) for all x β XΟ. Some authors varied the notion so that onlyk > 0 is required and called it the 42-type condition. In fact, one may see thatthese two notions coincide. There are still a number of equivalent notions relatedto the 42-conditions.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
882 CHANG IL KIM et al 881-893
A FIXED POINT THEOREM AND STABILITY OF ADDITIVE-CUBIC... 3
In [5], Khamsi proved a series of fixed point theorems in modular spaces where themodulars do not satisfy 42-conditions. His results exploit one unifying hypothesisin which the boundedness of an orbit is assumed.
Lemma 1.3. (see [5]) Let XΟ be a modular space whose induced modular is lowersemi-continuous and let C β XΟ be a Ο-complete subset. If T : C ββ C is aΟ-contraction, that is, there is a constant L β [0, 1) such that
Ο(Txβ Ty) β€ LΟ(xβ y), βx, y β C
and T has a bounded orbit at a point x0 β C, that is,
supΟ(Tnx0 β Tmx0) | n,m β N βͺ 0 <β
then the sequence Tnx0 is Ο-convergent to a point w β C.
Now, we will prove a fixed point theorem in modular spaces where the map Tdo not assume to be the boundedness of an orbit. Our results exploit one unifyinghypothesis in which some conditions are assumed.
Lemma 1.4. Let XΟ be a modular space whose induced modular is lower semi-continuous and let C β XΟ be a Ο-complete subset. Let T : C ββ C be a mappingsuch that
(1.2) 2Tx = T2x, βx β C.
Suppose that there is a constant L β [0, 1) with
(1.3) Ο(2Txβ 2Ty) β€ LΟ(xβ y), βx, y β C
and Ο(Txo β xo) < β at xo β C. Then the sequence Tn x0
4 is Ο-convergent tosome point w β C and
(1.4) Ο(x04β w) β€ 2
1β LΟ(Tx0 β x0).
Proof. By (M1) and (M3), we have Ο(Txβ Ty) β€ Ο(2Tx β 2Ty) and so, by (1.3),T is a Ο-contration. Hence we have
Ο(1
2T 2x0 β
1
2x0) β€ Ο(T 2x0 β Tx0) + Ο(Tx0 β x0)
β€ (L+ 1)Ο(Tx0 β x0).
Let Gx = 2Tx for all x β C. By (1.3), we have
Ο(1
2Tnx0 β
1
2x0) β€ Ο(Tnx0 β Tx0) + Ο(Tx0 β x0)
= Ο(1
2G(Tnβ1x0)β 1
2Gx0
)+ Ο(Tx0 β x0)
β€ LΟ(1
2Tnβ1x0 β
1
2x0
)+ Ο(Tx0 β x0)
for all n β N with n β₯ 2 and by induction, we have
Ο(1
2Tnx0 β
1
2x0) β€ Ξ£nβ1k=0L
kΟ(Tx0 β x0) β€ 1
1β LΟ(Tx0 β x0)
for all n β N. For any non-negative integers m,n with m > n,
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
883 CHANG IL KIM et al 881-893
4 CHANG IL KIM, GILJUN HAN, AND SEONG-A SHIM
Ο(1
4Tnx0 β
1
4Tmx0) β€ Ο(
1
2Tnx0 β
1
2x0) + Ο(
1
2Tmx0 β
1
2x0)
β€ 2
1β LΟ(Tx0 β x0).
(1.5)
By (1.2), T has a bounded orbit at a point x0
4 β C and thus by Lemma 1.3, Tn x0
4 is Ο-convergent to a point Ο β C. Since Ο is lower semi-continuous, by taking n = 0and mββ in (1.5), we have (1.4).
If Ο is convex, then Lemma 1.4 can be replaced by the following lemma.
Lemma 1.5. All conditions in Lemma 1.4 are assumed. Suppose that Ο is convexand 0 β€ L < 2. Then the sequence Tn x0
4 is Ο-convergent to some point w β Cand
(1.6) Ο(x04β w) β€ 1
2β LΟ(Tx0 β x0).
Proof. By (M1) and (M4), we have Ο(Tx β Ty) β€ 12Ο(2Tx β 2Ty) and since 0 β€
L < 2, by (1.3), T is a Ο-contration. Hence by (M4), we have
Ο(1
2T 2x0 β
1
2x0) β€ 1
2Ο(T 2x0 β Tx0) +
1
2Ο(Tx0 β x0)
β€ (1
4L+
1
2)Ο(Tx0 β x0).
Let Gx = 2Tx for all x β C. By (1.3), we have
Ο(1
2Tnx0 β
1
2x0) β€ 1
2Ο(Tnx0 β Tx0) +
1
2Ο(Tx0 β x0)
= Ο(1
2G(Tnβ1x0)β 1
2Gx0
)+
1
2Ο(Tx0 β x0)
β€ 1
2LΟ(1
2Tnβ1x0 β
1
2x0
)+
1
2Ο(Tx0 β x0)
for all n β N with n β₯ 2 and by induction, we have
Ο(1
2Tnx0 β
1
2x0) β€ Ξ£nβ1k=0
Lk
2k+1Ο(Tx0 β x0) β€ 1
2β LΟ(Tx0 β x0)
for all n β N. For any non-negative integers m,n with m > n,
Ο(1
4Tnx0 β
1
4Tmx0) β€ 1
2Ο(
1
2Tnx0 β
1
2x0) +
1
2Ο(
1
2Tmx0 β
1
2x0)
β€ 1
2β LΟ(Tx0 β x0).
The rest of the proof is similar to Lemma 1.4.
Let Ο be a modular on X, V a linear space. Define a set M by
M := g : V ββ XΟ | g(0) = 0and a generalized function Ο on M by
Ο(g) := infc > 0 | Ο(g(x)) β€ cΟ(x, x), βx β V ,for each g βM, where Ο : V 2 ββ [0,β) a mapping. Then M is a linear space, Ο isa modular on M. Furthermore, if Ο is convex, then Ο is also convex([15]).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
884 CHANG IL KIM et al 881-893
A FIXED POINT THEOREM AND STABILITY OF ADDITIVE-CUBIC... 5
Lemma 1.6. Let V be a linear space, XΟ a Ο-complete modular space, where Ο islower semi-continuous and f : V ββ XΟ a mapping with f(0) = 0. Let Ο : V 2 ββ[0,β) be a mapping. Then we have the following :
(1) MΟ = M and MΟ is Ο-complete.(2) Ο is lower semi-continuous.
Proof. (1) By the definition of MΟ, MΟ = M. Let Ξ΅ > 0 be given. Take any Ο-Cauchy sequence gn in MΟ. Then there is an l β N such that for n,m β N withn,m β₯ l,
(1.7) Ο(gn(x)β gm(x)) β€ Ξ΅Ο(x, x)
for all x β V . Hence gn(x) is a Ο-Cauchy sequence in XΟ for all x β X. Since XΟ
is Ο-complete, there is a mapping g : V ββ XΟ such that Ο(gn(x)β g(x)) ββ 0 asn βββ for all x β X. Then there is an m β N such that
Ο(gm(0)β g(0)) = Ο(g(0)) β€ Ξ΅
and hence g βMΟ. Since Ο is a lower semi-continuous, by (1.7), we have
Ο(gn(x)β g(x)) β€ lim infmβββ
Ο(gn(x)β gm(x)) β€ Ξ΅Ο(x, x)
for all x β X. Hence MΟ is Ο-complete.(2) Suppose that gn is a sequence in MΟ which is Ο-convergent to g βMΟ. Let
Ξ΅ > 0. Then for any n β N, there is a positive real number cn such that
Ο(gn) β€ cn β€ Ο(gn) + Ξ΅
and so
Ο(g(x)) β€ lim infnββ
Ο(gn(x)) β€ lim infnββ
cnΟ(x, x) β€(
lim infnββ
Ο(gn(x)) + Ξ΅)Ο(x, x)
for all x β X. Hence Ο is lower semi-continuous.
2. The generalized Hyers-Ulam stability for (1.1) in modular spaces
Throughout this section, we assume that every modular is lower semi-continuous.In this section, we will prove the generalized Hyers-Ulam stability for (1.1) by usingour fixed point theorem. We can easily show the following lemma.
Lemma 2.1. Let X and Y be vector spaces. Let f : X ββ Y satisfies (1.1) andf(0) = 0. Then we have :
(1) f is additive if and only if f(2x) = 2f(x) for all x β X.(2) f is cubic if and only if f(2x) = 8f(x) for all x β X.
For any mapping g : X ββ Y , let
ga(x) =1
4(g(2x)β 8g(x)), gc(x) = g(2x)β 2g(x)
and
Dg(x, y) = g(2x+ y) + g(2xβ y)β 2g(x+ y)β 2g(xβ y)β 2g(2x) + 4g(x).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
885 CHANG IL KIM et al 881-893
6 CHANG IL KIM, GILJUN HAN, AND SEONG-A SHIM
Theorem 2.2. Let V be a linear space, XΟ a Ο-complete modular space. Supposethat f : V ββ XΟ satisfies f(0) = 0 and
(2.1) Ο(Df(x, y)) β€ Ο(x, y)
for all x, y β V , where Ο : V 2 ββ [0,β) is a mapping such that
(2.2) Ο(2x, 2y) β€ LΟ(x, y)
for some L with 0 β€ L < 1 and for all x, y β V . Then there exists a uniqueadditive-cubic mapping F : V ββ XΟ such that
(2.3) Ο(F (x)β 3
16f(x)) β€ 4
1β LΟ(x, x)
for all x β V , where Ο(x, y) = Ο(x, 2y) + Ο(x, y) + Ο(0, y).
Proof. Let Ο(x, y) = Ο(x, 2y) + Ο(x, y) + Ο(0, y). Then by Lemma 1.6, MΟ = M isΟ-complete and Ο is lower semi-continuous.
Define Ta : MΟ ββ MΟ by Tag(x) = 12g(2x) for all g β MΟ and all x β V . Let
g, h β MΟ. Suppose that Ο(g β h) β€ c for some positive real number c. Then by(2.3) and Remark 1.2, we have
Ο(Tag(x)β Tah(x)) β€ Ο(g(2x)β h(2x)) β€ LcΟ(x, x)
for all x β V and so Ο(Tag β Tah) β€ LΟ(g β h). Hence Ta is a Ο-contraction. By(2.1), we get
(2.4) Ο(f(x) + f(βx)) β€ Ο(0, x),
(2.5) Ο(f(3x)β 4f(2x) + 5f(x)) β€ Ο(x, x),
and
(2.6) Ο(f(4x)β 2f(3x)β 2f(2x)β 2f(βx) + 4f(x)) β€ Ο(x, 2x),
for all x β V . By (2.4), (2.5), and (2.6), we obtain
Ο(Tafa(x)β fa(x)) = Ο(1
2fa(2x)β fa(x)) β€ Ο(x, 2x) + Ο(x, x) + Ο(0, x) = Ο(x, x)
for all x β V and hence we have
(2.7) Ο(Tafa β fa) β€ 1.
Let Gg = 2Tag for all g βMΟ. Then
Gg(x) = g(2x)
for all g β MΟ and for all x β V . Suppose that Ο(g β h) β€ c for some positive realnumber c, where g, h β MΟ. Then Ο(g(x) β h(x)) β€ cΟ(x, x) for all x β V and by(2.2), we have
Ο(Gg(x)βGh(x)) = Ο(g(2x)β h(2x)) β€ cΟ(2x, 2x) β€ cLΟ(x, x)
for all x β V . Hence Ο(Gg βGh) β€ cL and so
Ο(Gg βGh) β€ LΟ(g β h).
Since Ta is linear, by Lemma 1.4, there is an A β MΟ such that Tnafa4 is Ο-
convergent to A. In fact, we get
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
886 CHANG IL KIM et al 881-893
A FIXED POINT THEOREM AND STABILITY OF ADDITIVE-CUBIC... 7
(2.8) limnββ
Ο(1
2n+2fa(2nx)βA(x)) = 0
for all x β V . Since Ο is lower semi-continuous, we get
Ο(TaAβA) β€ lim infnββ
Ο(TaAβ Tn+1a
fa4
) β€ lim infnββ
LΟ(Aβ Tnafa4
) = 0
and hence A is a fixed point of Ta in MΟ. Replacing x and y by 2nx and 2ny in(2.1), respectively, by (2.2), we have
Ο(1
2n+2Dfa(2nx, 2ny))
β€ Ο(1
2n+3Df(2n+1x, 2n+1y)) + Ο(
1
2nDf(2nx, 2ny))
β€ Ln+1Ο(x, y) + LnΟ(x, y)
for all x, y β V and for all n β N. Hence we get
(2.9) limnββ
Ο( 1
2n+2Dfa(2nx, 2ny)
)= 0
for all x, y β V . Note that
Ο( 1
2n+10Dfa(2nx, 2ny)β 1
28DA(x, y)
)β€ Ο( 1
2n+9fa(2n(2x+ y))β 1
27A(2x+ y)
)+ Ο( 1
2n+8fa(2n(2xβ y))β 1
26A(2xβ y)
)+ Ο( 1
2n+72fa(2n(x+ y))β 1
252A(x+ y)
)+ Ο( 1
2n+62fa(2n(xβ y))β 1
242A(xβ y)
)+ Ο( 1
2n+52fa(2n2x)β 1
232A(2x)
)+ Ο( 1
2n+52fa(2n(xβ y))β 1
234A(xβ y)
)for all x, y β V and for all n β N. Hence we have
(2.10) limnββ
Ο( 1
2n+10Dfa(2nx, 2ny)β 1
28DA(x, y)
)= 0
for all x, y β V . Since
Ο( 1
29DA(x, y)
)β€ Ο( 1
2n+10Dfa(2nx, 2ny)β 1
28DA(x, y)
)+Ο( 1
2n+10Dfa(2nx, 2ny)
)for all x, y β V and for all n β N, by (2.9) and (2.10), we get
(2.11) DA(x, y) = 0
for all x, y β V . By (1.4) in Lemma 1.4, we get
(2.12) Ο(Aβ 1
4fa) β€ 2
1β L.
Define Tc : MΟ ββ MΟ by Tcg(x) = 18g(2x) for all g β MΟ and all x β V . By
(2.4), (2.5), and (2.6), we obtain
Ο(1
23fc(2x)β fc(x)) β€ Ο(x, x)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
887 CHANG IL KIM et al 881-893
8 CHANG IL KIM, GILJUN HAN, AND SEONG-A SHIM
for all x β V and hence
(2.13) Ο(Tcfc β fc) β€ 1.
Let Hg = 2Tcg for all g βMΟ. Then
Hg(x) =1
4g(2x).
for all g β MΟ and for all x β V . Let g, h β MΟ. Suppose that Ο(g β h) β€ c forsome positive real number c. Then Ο(g(x)β h(x)) β€ cΟ(x, x) for all x β V and by(2.3), we get
Ο(Hg(x)βHh(x)) = Ο(1
4g(2x)β 1
4h(2x)) β€ cΟ(2x, 2x) β€ cLΟ(x, x)
for all x β V . Hence Ο(Hg βHh) β€ cL and so
Ο(Hg βHh) β€ LΟ(g β h).
Since Tc is linear, by Lemma 1.4, there is a C β MΟ such that Tnc 14fc is Ο-
convergent to C. Since Ο is lower semi-continuous, we get
Ο(TcC β C) β€ lim infnββ
Ο(TcC β Tn+1c
1
4fc) β€ lim inf
nββLΟ(C β Tnc
1
4fc) = 0
and hence C is a fixed point of Tc in MΟ. Replacing x and y by 2nx and 2ny in(2.1), respectively, by (2.2), we have
Ο(1
23n+2Dfc(2
nx, 2ny))
β€ Ο(1
23n+1Df(2n+1x, 2n+1y)) + Ο(
1
23nDf(2nx, 2ny))
β€ Ln+1Ο(x, y) + LnΟ(x, y)
for all x, y β V . Hence we get
limnββ
Ο( 1
23n+2Dfc(2
nx, 2ny))
= 0
for all x, y β V . Similar to A, we have
(2.14) DC(x, y) = 0
for all x, y β V and by (1.4) in Lemma 1.4, we get
Ο(C(x)β 1
4fc(x)) β€ 2
1β LΟ(x, x)
for all x β X. Hence we have
(2.15) Ο(C β 1
4fc) β€
2
1β L.
Let F = 18C β
12A. Since A is a fixed point of Ta, A(2x) = 2A(x) for all x β X
and similarly, C(2x) = 8C(x) for all x β X. By Lemma 2.1, A is additive and Cis cubic. Hence F is an additive-cubic mapping. Since f(x) = 1
6fc(x)β 23fa(x), we
have
Ο(F β 3
16f) β€ Ο(Aβ 1
4fa) + Ο(
1
4C β 1
16fc) β€ Ο(Aβ 1
4fa) + Ο(C β 1
4fc),
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
888 CHANG IL KIM et al 881-893
A FIXED POINT THEOREM AND STABILITY OF ADDITIVE-CUBIC... 9
and hence by (2.12) and (2.15), we have (2.3).To prove the uniquness of F , let K : V ββ XΟ be another additive-cubic map-
ping with (2.3). By (2.3), we get
Ο(1
4K(x)β 1
4F (x)) β€ Ο(K(x)β 3
16f(x)) + Ο(F (x)β 3
16f(x)))
β€ 8
1β LΟ(x, x)
for all x β V and so
Ο(1
16Ka(x)β 1
16Fa(x)) β€ Ο(
1
32K(2x)β 1
32F (2x)) + Ο(
1
4K(x)β 1
4F (x)))
β€ 8(1 + L)
1β LΟ(x, x)
for all x β V . Since Fa and Ka are fixed points of Ta, we have
Ο(1
16Ka(x)β 1
16Fa(x)) = Ο(
1
16TnaKa(x)β 1
16Tna Fa(x))
β€ 8(1 + L)
1β LLnΟ(x, x)
for all x β V and for all n β N. Letting n β β in the last inequality, we haveFa = Ka and similarly, we have Fc = Kc. Thus F = K.
Comparing the results in a modular and a convex modular, we may see that thecoefficient in the case of convex modular is smaller.
Theorem 2.3. Suppose that every assumption of Theorem 2.2 holds, Ο is convexand 0 β€ L < 2. Then there exists a unique additive-cubic mapping F : V ββ XΟ
such that
(2.16) Ο(F (x)β 3
16f(x)) β€ 5
32(2β L)Ο(x, x)
for all x β V , where Ο(x, y) = Ο(x, 2y) + Ο(x, y) + Ο(0, y).
Proof. Define Ta : MΟ ββ MΟ by Tag(x) = 12g(2x) for all g β MΟ and all x β V .
Let g, h β MΟ. Suppose that Ο(g β h) β€ c for some positive real number c. Thenby (2.3) and (M4), we have
Ο(Tag(x)β Tah(x)) β€ 1
2Ο(g(2x)β h(2x)) β€ 1
2LcΟ(x, x)
for all x β V and so Ο(Tag β Tah) β€ 12LΟ(g β h). Hence Ta is a Ο-contraction. By
(2.1), we get
(2.17) Ο(f(x) + f(βx)) β€ Ο(0, x),
(2.18) Ο(f(3x)β 4f(2x) + 5f(x)) β€ Ο(x, x),
and
(2.19) Ο(f(4x)β 2f(3x)β 2f(2x)β 2f(βx) + 4f(x)) β€ Ο(x, 2x),
for all x β V . By (2.17), (2.18), and (2.19), we obtain
Ο(1
2fa(2x)β fa(x)) β€ 1
8Ο(x, 2x) +
1
4Ο(x, x) +
1
4Ο(0, x) β€ 1
4Ο(x, x)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
889 CHANG IL KIM et al 881-893
10 CHANG IL KIM, GILJUN HAN, AND SEONG-A SHIM
for all x β V and hence
Ο(Tafa(x)β fa(x)) β€ 1
4Ο(x, x)
for all g βMΟ and all x β V . Hence we have
(2.20) Ο(Tafa β fa) β€ 1
4.
Let Gg = 2Tag for all g βMΟ. Then
Gg(x) = g(2x)
for all x β V . Suppose that Ο(g β h) β€ c for some positive real number c. ThenΟ(g(x)β h(x)) β€ cΟ(x, x) for all x β V and by (2.2), we have
Ο(Gg(x)βGh(x)) = Ο(g(2x)β h(2x)) β€ cΟ(2x, 2x) β€ cLΟ(x, x)
for all x β V . Hence Ο(Gg βGh) β€ cL and so
Ο(Gg βGh) β€ LΟ(g β h).
Since Ta is linear, by Lemma 1.5, there is an A β MΟ such that Tnafa4 is Ο-
convergent to A. In fact, we get
limnββ
Ο(1
2n+2fa(2nx)βA(x)) = 0
for all x β V . Since Ο is lower semi-continuous, we get
Ο(TaAβA) β€ lim infnββ
Ο(TaAβ Tn+1a
fa4
) β€ lim infnββ
LΟ(Aβ Tnafa4
) = 0
and hence A is a fixed point of Ta in MΟ. Similar to Theorem 2.2, we have
(2.21) DA(x, y) = 0
for all x, y β V and by (1.6) in Lemma 1.5 and (2.20), we get
(2.22) Ο(Aβ 1
4fa) β€ 1
4(2β L).
Define Tc : MΟ ββ MΟ by Tcg(x) = 18g(2x) for all g β MΟ and all x β V . By
(2.17), (2.18), and (2.19), we obtain
Ο(1
23fc(2x)β fc(x)) β€ 1
4Ο(x, x)
for all x β V and hence
(2.23) Ο(Tcfc β fc) β€1
4.
Let Hg = 2Tcg for all g βMΟ. Then
Hg(x) =1
4g(2x).
for all g β MΟ and for all x β V . Let g, h β MΟ. Suppose that Ο(g β h) β€ c forsome positive real number c. Then Ο(g(x)β h(x)) β€ cΟ(x, x) for all x β V and by(2.3), we get
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
890 CHANG IL KIM et al 881-893
A FIXED POINT THEOREM AND STABILITY OF ADDITIVE-CUBIC... 11
Ο(Hg(x)βHh(x)) = Ο(1
4g(2x)β 1
4h(2x)) β€ cΟ(2x, 2x) β€ cLΟ(x, x)
for all x β V . Hence Ο(Gg βGh) β€ cL and so
Ο(Hg βHh) β€ LΟ(g β h).
Since Tc is linear, by Lemma 1.5, there is a C β MΟ such that Tnc 14fc is Ο-
convergent to C. Since Ο is lower semi-continuous, we get
Ο(TcC β C) β€ lim infnββ
Ο(TcC β Tn+1c
1
4fc) β€ lim inf
nββLΟ(C β Tnc
1
4fc) = 0
and hence C is a fixed point of Tc in MΟ. Similar to Theorem 2.2, we get
(2.24) DC(x, y) = 0
for all x, y β V and by (1.6) in Lemma 1.5, we get
Ο(C(x)β 1
4fc(x)) β€ 1
4(2β L)Ο(x, x)
for all x β X. Hence we have
(2.25) Ο(C β 1
4fc) β€
1
4(2β L).
Let F = 18C β
12A. Since A is a fixed point of Ta, A(2x) = 2A(x) for all x β X
and similarly, C(2x) = 8C(x) for all x β X. By Lemma 2.1, A is additive and Cis cubic. Hence F is an additive-cubic mapping. Since f(x) = 1
6fc(x)β 23fa(x), by
(2.22) and (2.25), we have
Ο(F β 3
16f) β€ 1
2Ο(Aβ 1
4fa) +
1
2Ο(
1
4C β 1
16fc) β€
1
2Ο(Aβ 1
4fa) +
1
8Ο(C β 1
4fc).
and hence we have (2.16). The rest of the proof is similar to Theorem 2.2.
Remark 2.4. Sadeghi [13] proved the generalized Hyers-Ulam stability of func-tional equations in modular spaces with the 42-condition and in [15], authorsproved the stability of mappings f : V ββ XΟ and Ο : V 2 ββ [0,β) satisfyingf(0) = 0,
(2.26) limnββ
Ο(2nx, 2ny)
4n= 0, Ο(2x, 2x) β€ 4LΟ(x, x), βx, y β V,
andΟ(4f(x+ y) + 4f(xβ y)β 8f(x)β 8f(y)) β€ Ο(x, y), βx, y β V
for some real number L with 0 β€ L < 12 whose codomain is equipped with a convex
and lower semi-continuous modular without 42-conditions. Our results guaranteethe stability of an additive-cubic mapping, whose induced modular is lower semi-continuous without the convexity and 42-conditions if 0 β€ L < 1
4 . Further, in [15],authors left whether the multiple of 4 on the left side of the inequality (6) can bedropped as a problem. We can solve the problem by using Lemma 1.4 and its proofis similar to the proof in Theorem 2.2.
In fact, suppose that Ο : V 2 ββ [0,β) is a mapping with (2.26) and that0 β€ L < 1
4 . Let f : V ββ XΟ be a mapping such that f(0) = 0 and
(2.27) Ο(f(x+ y) + f(xβ y)β 2f(x)β 2f(y)) β€ Ο(x, y)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
891 CHANG IL KIM et al 881-893
12 CHANG IL KIM, GILJUN HAN, AND SEONG-A SHIM
for all x, y β X. Let Ο(x, y) = Ο(x, y) for all x, y β V , f0(x) = 4f(x), andTg(x) = 1
4g(2x). Then by (2.27), we have
Ο(Tf0(x)β f0(x)) = Ο(1
4f0(2x)β f0(x)) β€ Ο(x, x)
for all x β V and so
(2.28) Ο(Tf0 β f0) β€ 1.
Moreover, by (2.26), we have
(2.29) Ο(2Tg β 2Th) β€ 4LΟ(g β h).
Since 0 β€ L < 14 , by Lemma 1.4, there is a fixed point Q βMΟ such that Tn f04 =
Tnf converges to Q in XΟ and
(2.30) Ο(Q(x)β f(x)) β€ 2
1β 4LΟ(x, x)
for all x β V . We can show that Q is a quadratic mapping ([15]).Further, suppose that Ο is convex and 0 β€ L < 1. Then (2.28) and (2.29) can be
replaced by
Ο(Tf0 β f0) β€ 1
4and
Ο(2Tg β 2Th) β€ 2LΟ(g β h),
respectively. By Lemma 1.5 and (1.6),
Ο(Q(x)β f(x)) β€ 1
4(2β 2L)Ο(x, x) =
1
8(1β L)Ο(x, x)
for all x β V .
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publicationof this paper.
AcknowledgmentThe corresponding author was supported by the research fund of Dankook Uni-
versity in 2014.
References
[1] T. Aoki, On the stability of the linear transformation in Banach spaces, J. Math. Soc. Japan
2(1950), 64-66.[2] P. Gavruta, A generalization of the Hyer-Ulam-Rassias stability of approximately additive
mappings, J. Math. Anal. Appl. 184(1994), 431-436.
[3] D. H. Hyers, On the stability of the linear functional equation, Proc. Nat. Acad. Sci. 27(1941),222-224.
[4] M. A. Kamasi and W. M. Kozlowski, Fixed point thoery in modular function spaces,
Birkhauser, 2015.[5] M. A. Khamsi, Quasicontraction mappings in modular spaces without 42-condition., Fixed
Point Theory and Applications, 2008(2008), 1-6.
[6] M. A. Khamsi, W. M. Kozlowski, and S. Reich, Fixed point theory in Modular spaces,Nonlinear analysis, Theory, Method and Applications, 14(1990), 935-953.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
892 CHANG IL KIM et al 881-893
A FIXED POINT THEOREM AND STABILITY OF ADDITIVE-CUBIC... 13
[7] W. A. Luxemburg, Banach function spaces, Ph. D. thesis, Delft Univrsity of technology, Delft,The Netherlands, 1959.
[8] J. Musielak and W. Orlicz, On modular spaces, Studia Mathematica, 18(1959), 591-597.
[9] A. Najati and G. Z. Eskandani, Stability of a mixed additive and cubic functional equationin quasi-Banach spaves, J. Math. Anal. Appl., 342(2008), 1318-1331.
[10] H. Nakano, Modular semi-ordered spaces, Tokyo, Japan, 1959.
[11] W. Orlicz, Collected Papers, Vols. I, II, PWN, Warszawa, 1988.[12] Th. M. Rassias, On the stability of the linear mapping in Banach sapces, Proc. Amer. Math.
Sco. 72(1978), 297-300.[13] G. Sadeghi, A fixed point approach to stability of functional equations in modular spaces,
Bulletin of the Malaysian Mathematical Sciences Society. Second Series, 37(2014), 333-344.
[14] S. M. Ulam, Problems in Modern Mathematics, Wiley, New York; 1964.[15] K. Wongkum, P. Chaipunya, and P. Kumam, On the generalized Ulam-Hyers-Rassias stability
of quadratic mappings in modular spaces without 42-conditions, 2015(2015), 1-6.
[16] T. Z. Xu, J. M. Rassias, and W. X. Xu, Stability of a general mixed additive-cubic equationin F-spaces, J. Comp. Anal. Appl., 14(2012), 1026-1037.
[17] T. Z. Xu, J. M. Rassias, and W. X. Xu, Stability of a general mixed additive-cubic functional
equation in non-Archimedean fuzzy normed spaces, J. Math. Phys. 51(2010).[18] T. Z. Xu, J. M. Rassias, and W. X. Xu, Intuitionistic fuzzy stability of a general mixed
additive-cubic equation, J. Math. Phys. 51(2010).
[19] S. Yamamuro, On conjugate spaces of Nakano spaces, Trans. Amer. Math. Soc. 90(1959)291-311.
Department of Mathematics Education, Dankook University, 152, Jukjeon-ro, Suji-
gu, Yongin-si, Gyeonggi-do, 448-701, KoreaE-mail address: [email protected]
Department of Mathematics Education, Dankook University, 152, Jukjeon-ro, Suji-
gu, Yongin-si, Gyeonggi-do, 448-701, KoreaE-mail address: [email protected]
Department of Mathematics, Sungshin Womenβs University, 249-1, Dongseon-Dong3-Ga, SeongBuk-Gu, Seoul 136-742, Korea
E-mail address: [email protected]
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
893 CHANG IL KIM et al 881-893
Results on value-shared of admissible function and
non-admissible function in the unit disc β
Hong-Yan Xua
a Department of Informatics and Engineering, Jingdezhen Ceramic Institute,
Jingdezhen, Jiangxi, 333403, China
<e-mail: [email protected]>
Abstract
In this paper, we consider the uniqueness problem of admissible functions andnon-admissible functions sharing some values in the unit disc. We obtain: If f1 isadmissible and f2 is inadmissible satisfying lim
rβ1βT (r, f2) = β, aj(j = 1, 2, . . . ; q)
be q distinct complex numbers. Then(i) f1(z), f2(z) can share at most three values a1, a2, a3 IM ;(ii) f1(z), f2(z) can share at most five values aj(j = 1, 2, . . . ; 5) with reduced
weight 1. Our results of this paper are improvement of the uniqueness theoremsof meromorphic functions sharing some values on the whole complex plane whichgiven by Yi and Cao.Key words: uniqueness; meromorphic function; admissible; non-admissible.Mathematical Subject Classification (2010): 30D 35.
1 Introduction and Main Results
In what follows, we shall assume that reader is familiar with the fundamental results andthe standard notations of the Nevanlinna value distribution theory of meromorphic func-tions such as the proximity function m(r, f), counting function N(r, f), characteristicfunction T (r, f), the first and second main theorems, lemma on the logarithmic deriva-tives etc. of Nevalinna theory, (see Hayman [7] , Yang [16] and Yi and Yang [19]). For ameromorphic function f , S(r, f) denotes any quantity satisfying S(r, f) = o(T (r, f)) forall r outside a possible exceptional set of finite logarithmic measure.
We use C to denote the open complex plane, C := Cββ to denote the extended
complex plane, and D = z : |z| < 1 to denote the unit disc.
Let f, g be two non-constant meromorphic functions in D and a β C. If E(a,D, f) =E(a,D, g), we say f and g share a CM(counting multiplicities) in D. If E(a,D, f) =
βThis work was supported by NSFC(11561033, 11301233, 61202313), the Natural Science Foun- dationof Jiangxi Province in China 20132BAB211001, 20151BAB201008), and the Foundation of EducationDepartment of Jiangxi (GJJ14644) of China.
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
894 Hong-Yan Xu 894-906
E(a,D, g), we say f and g share a IM(ignoring multiplicities) in D. If D is replaced byC, we give the simple notation as before, E(a, f), E(a, f) and so on(see [16]).
R.Nevanlinna [12] proved the following well-known theorems.
Theorem 1.1 (see [12]) If f and g are two non-constant meromorphic functions thatshare five distinct values a1, a2, a3, a4, a5 IM in C, then f(z) β‘ g(z).
Theorem 1.2 (see [12]) If f and g are two distinct non-constant meromorphic functionsthat share four distinct values a1, a2, a3, a4 CM in C, then f is a Mobius transformationof g , two of the shared values, say a1 and a2 are Picard values, and the cross ratio(a1, a2, a3, a4) = β1.
After their very work, the uniqueness of meromorphic functions with shared valuesin the whole complex plane attracted many investigations (see [16]). In 1987 and 1988,Yi [17, 18] dealt with the problems of multiple values and uniquness of meromorphicfunctions sharing some values in the whole complex plane by adopting L.Yangβs methodand obtained some results which improved the concerning theorems due to Gopalakrishnaand Bhoosnurmathβs [6], Ueda [14]. To state the theorems, we will explain some notationsas follows.
Let f(z) be a non-constant meromorphic function, an arbitrary complex number
a β C, and k be a positive integer. We use Ek)(a, f) to denote the set of zeros of f β a,with multiplicities no greater than k, in which each zero counted only once. We say thatf(z) and g(z) share the value a with reduced weight k, if Ek)(a, f) = Ek)(a, g).
In 1987, Yi [17] obtained the uniqueness theorems concerning multiple values of mero-morphic functions as follows.
Theorem 1.3 (see [17, 19, Theorem 3.15]). Let f(z) and g(z) be two non-constantmeromorphic functions, aj(j = 1, 2, . . . , q) be q distinct complex numbers, and let kj(j =1, 2, . . . , q) be positive integers or β satisfying
(1) k1 β₯ k2 β₯ Β· Β· Β· β₯ kq β₯ 1.
IfEkj)(aj , f) = Ekj)(aj , g) (j = 1, 2, . . . , q)
and
(2)
qβj=3
kjkj + 1
> 2,
then f(z) β‘ g(z).
In recent, it is an interesting topic to investigate the uniqueness with shared valuesin the subregion of the complex plane such as the unit disc, an angular domain, see[1, 2, 9, 10, 11, 15, 20, 21, 22]. In 1999, Fang [5] studied the uniqueness problem ofadmissible meromorphic functions in the unit disc D sharing two sets and three sets.Later, there were some results of uniqueness of meromorphic function in the unit discconcerning admissible functions. To state some uniqueness theorems of meromorphicfunctions in the unit disc D, we need the following basic notations and definitions ofmeromorphic functions in D(see [3], [4], [8]).
2
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
895 Hong-Yan Xu 894-906
Definition 1.1 Let f be a meromorphic function in D and limrβ1β T (r, f) =β. Then
Ξ±(f) := lim suprβ1β
T (r, f)
β log(1β r)
is called the index of inadmissibility of f . If Ξ±(f) =β, f is called admissible.
Definition 1.2 Let f be a meromorphic function in D and limrβ1β T (r, f) =β. Then
Ο(f) := lim suprβ1β
log+ T (r, f)
β log(1β r)
is called the order (of growth) of f .
For admissible functions, the following theorem plays a very important role in studiesthe uniqueness problems of meromorphic functions in the unit disc.
Theorem 1.4 (see [13, Theorem 3]). Let f be an admissible meromorphic function inD, q be a positive integer and a1, a2, . . . , aq be pairwise distinct complex numbers. Then,for r β 1β, r 6β E,
(q β 2)T (r, f) β€qβj=1
N
(r,
1
f β aj
)+ S(r, f),
where E β (0, 1) is a possibly occurring exceptional set withβ«E
dr1βr <β. If the order of
f is finite, the remainder S(r, f) is a O(
log 11βr
)without any exceptional set.
In 2005, Titzhoff [13] investigated the uniqueness of two admissible functions in theunit disc D by using the Second Main Theorem for admissible functions (Theorem 1.4)and obtained the five values theorem in the unit disc D as follows.
Theorem 1.5 (see [13]). If two admissible function f, g share five distinct values, thenf β‘ g.
In 2009, Mao and Liu [11] gave a different method to investigate the uniquenessproblem of meromorphic functions in unit disc and obtained the following results.
Theorem 1.6 (see [11]). Let f, g be two meromorphic functions in D, aj β C(j =1, 2, . . . , 5) be five distinct values, and β(ΞΈ0, Ξ΄)(0 < Ξ΄ < Ο) be an angular domain such
that for some a β C,
(3) lim suprβ1β
log n(r,β(ΞΈ0, Ξ΄/2), f(z) = a)
log 11βr
= Ο > 1.
If f and g share aj(j = 1, 2, . . . , 5) IM in β(ΞΈ0, Ξ΄), then f(z) β‘ g(z).
Remark 1.1 In fact, the condition (3) implies that f is admissible in the unit disc.Therefore, Theorem 1.6 is one result of uniqueness of admissible functions in the unitdisc.
3
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
896 Hong-Yan Xu 894-906
For admissible functions in the unit disc D, from Theorem 1.4, using the same argu-ment as in the proofs of Theorem 1.3, we can easily get the following results.
Theorem 1.7 Let f1(z) and f2(z) be two admissible meromorphic functions in D, aj(j =1, 2, . . . , q) be q distinct complex numbers, and let kj(j = 1, 2, . . . , q) be positive integersor β satisfying (1). If f1(z) and f2(z) satisfy
(4) Ekj)(aj ,D, f1) = Ekj)(aj ,D, f2) (j = 1, 2, . . . , q)
and (2), then f1(z) β‘ f2(z), where Ek)(a,D, f) to denote the set of zeros of f β a in D,with multiplicities no greater than k, in which each zero counted only once.
Remark 1.2 For a β C and a positive integer k, we can say that f1(z), f2(z) share thevalue a in D with reduced weight k, if Ek)(a,D, f1) = Ek)(a,D, f2).
Similar to the corollary of Theorem 1.3 (see [19, Corollary,pp.181.]), we can get thefollowing corollary.
Corollary 1.1 Let f1(z) and f2(z) be two admissible meromorphic functions in D, aj(j =1, 2, . . . , q) be q distinct complex numbers, and let kj(j = 1, 2, . . . , q) be positive integersor β satisfying (1) and (4).
(i) if q = 7, then f1(z) β‘ f2(z);(ii) if q = 6 and k3 β₯ 2, then f1(z) β‘ f2(z);(iii) if q = 5, k3 β₯ 3 and k5 β₯ 2, then f1(z) β‘ f2(z);(iv) if q = 5 and k4 β₯ 4, then f1(z) β‘ f2(z);(v) if q = 5, k3 β₯ 5 and k4 β₯ 3, then f1(z) β‘ f2(z);(vi) if q = 5, k3 β₯ 6 and k4 β₯ 2, then f1(z) β‘ f2(z).
Remark 1.3 In Theorem 1.5, the conclusion f(z) β‘ g(z) holds when q = 5 and kj =β (j = 1, 2, . . . , 5). From Corollary 1.1, we can get that f1(z) β‘ f2(z) when q = 5 andkj (j = 1, 2, . . . , 5) satisfy any of the four conditions (i)-(iv). Hence, Corollary 1.1 is animprovement of Theorem 1.5.
For non-admissible functions, the following theorem also plays a very important rolein studies theirs uniqueness problems.
Theorem 1.8 (see [13, Theorem 2]). Let f be a meromorphic function in D and limrβ1β
T (r, f) =β, q be a positive integer and a1, a2, . . . , aq be pairwise distinct complex num-bers. Then, for r β 1β, r 6β E,
(q β 2)T (r, f) β€qβj=1
N
(r,
1
f β aj
)+ log
1
1β r+ S(r, f).
Remark 1.4 (i) In contrast to admissible functions, the term log 11βr in Theorem 1.8
does not necessarily enter the remainder S(r, f) because the non-admissible function f
may have T (r, f) = O(
log 11βr
).
(ii) If 0 < Ξ±(f) < β, we can see that S(r, f) = o(
log 11βr
)holds in Theorem 1.8
without a possible exception set.
4
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
897 Hong-Yan Xu 894-906
From Theorem 1.8 and Remark 1.4, we can see that the uniqueness of non-admissiblefunctions is more intricate than the case of admissible functions.
In this paper, we will deal with the uniqueness problem of non-admissible functionsin D. We use Ξ₯Ξ± to denote the class of non-admissible functions satisfying the condition:Ξ±(f) = Ξ±(0 < Ξ± <β) for f β Ξ₯Ξ±. For the class Ξ₯Ξ±, we get the following results
Theorem 1.9 Let f(z) β Ξ₯Ξ±, aj(j = 1, 2, . . . , q) are q distinct complex numbers. Ifq = 5 + [ 2
k + k+1kΞ± ], then there does not exist g(z)(6β‘ f(z)) β Ξ₯Ξ± satisfying
(5) Ek)(aj ,D, f) = Ek)(aj ,D, g), (j = 1, 2, . . . , q),
where [x] denotes the largest integer less than or equal to x.
Corollary 1.2 Let f(z) β Ξ₯Ξ±. Then f(z) is uniquely determined in Ξ₯Ξ± by one of thefollowing cases:
(i) if f have seven point-sets E1)(aj ,D, f)(j = 1, 2, . . . , 7) and Ξ± > 1;
(ii) if f have six point-sets E2)(aj ,D, f)(j = 1, 2, . . . , 6) and Ξ± > 32 ;
(iii) if f have five point-sets E3)(aj ,D, f)(j = 1, 2, . . . , 5) and Ξ± > 4.
Remark 1.5 For Corollary 1.1, we can see that the conclusion (iii) in Corollary 1.1 is animprovement of Theorem 1.6. In fact, the conclusion of Theorem 1.6 is that non-constantmeromorphic function f is uniquely determined in D by five point-sets Eβ)(aj ,D, f)(j =1, 2, . . . , 5) and Ξ±(f) =β.
Theorem 1.10 Let Ξ± > 12 and f(z) β Ξ₯Ξ±, aj(j = 1, 2, . . . , 5) be five distinct complexnumbers. Then f(z) is uniquely determined in Ξ₯ by three point-sets E3)(aj ,D, f)(j =
1, 2, 3) and two point-sets E2)(aj ,D, f)(j = 4, 5).
Furthermore, for the uniqueness of regular inadmissibility functions we obtain thefollowing theorems
Theorem 1.11 Let aj(j = 1, 2, . . . , q) be q distinct complex numbers, and let kj(j =1, 2, . . . , q) be positive integers orβ satisfying (1). If f1(z), f2(z) be non-constant regularinadmissibility functions satisfying 0 < Ξ±(f1), Ξ±(f2) <β, (4) and
(6)
qβj=3
kjkj + 1
β 2 >2
Ξ±(f1) + Ξ±(f2),
then f1(z) β‘ f2(z).
From Theorem 1.11, similar to Corollary 1.1, we can get the following results easily.
Corollary 1.3 Let aj(j = 1, 2, . . . , q) be q distinct complex numbers, and let kj(j =1, 2, . . . , q) be positive integers or β satisfying (1), Ξ± := minΞ±(f1), Ξ±(f2). And letf1(z), f2(z) be non-constant regular inadmissibility functions satisfying 0 < Ξ±(f1), Ξ±(f2) <β and (4),
(i) if Ξ± > 1, q = 7 and k7 β₯ 2, then f1(z) β‘ f2(z);(ii) if Ξ± > 1, q = 6 and k6 β₯ 4, then f1(z) β‘ f2(z);
5
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
898 Hong-Yan Xu 894-906
(iii) if Ξ± > 2 and q = 7, then f1(z) β‘ f2(z);(iv) if Ξ± > 3, q = 6 and k3 β₯ 2, then f1(z) β‘ f2(z);(v) if Ξ± > 6, q = 5, k3 β₯ 3 and k5 β₯ 2, then f1(z) β‘ f2(z);(vi) if Ξ± > 10, q = 5 and k4 β₯ 4, then f1(z) β‘ f2(z);(vii) if Ξ± > 12, q = 5, k3 β₯ 5 and k4 β₯ 3, then f1(z) β‘ f2(z);(viii) if Ξ± > 42, q = 5, k3 β₯ 6 and k4 β₯ 2, then f1(z) β‘ f2(z).
Remark 1.6 In Corollary 1.1, f1(z), f2(z) are all admissible functions, that is, Ξ±(f1) =β and Ξ±(f2) =β. From the conclusions of Corollary 1.3, we see that f1(z) β‘ f2(z) holdswhen f1(z), f2(z) are non-constant regular inadmissibility functions with minΞ±(f1), Ξ±(f2)> ΞΆ and ΞΆ a positive constant. Hence, Corollary 1.3 is an improvement of Corollary 1.1.
The following theorem will show that an admissible function can share sufficientlymany values concerning multiple values with another inadmissible function.
Theorem 1.12 If f1 is admissible and f2 is inadmissible satisfying limrβ1β T (r, f2) =β, aj(j = 1, 2, . . . , q) be q distinct complex numbers, and let kj(j = 1, 2, . . . , q) be positiveintegers or β satisfying (1). Then
(7)
qβj=2
kjkj + 1
β 2 > 0
and (4) do not hold at same time.
Corollary 1.4 If f1 is admissible and f2 is inadmissible satisfying limrβ1β T (r, f2) =β, aj(j = 1, 2, . . . , q) be q distinct complex numbers. Then
(i) f1(z), f2(z) can share at most three values a1, a2, a3 IM ;(ii) f1(z), f2(z) can share at most five values aj(j = 1, 2, . . . , 5) with reduced weight
1;And any one of the following cases can not hold
(iii) q = 4 and Ek1)(a1,D, f1) = Ek1)(a1,D, f2) (k1 β₯ 6), E6)(a2,D, f1) = E6)(a2,D,f2), E2)(a3,D, f1) = E2)(a3,D, f2) and E1)(a4,D, f1) = E1)(a4,D, f2);
(iv) q = 4 and Ek1)(a1,D, f1) = Ek1)(a1,D, f2) (k1 β₯ 3), E3)(a2,D, f1) = E3)(a2,D,f2), E2)(a3,D, f1) = E2)(a3,D, f2) and E2)(a4,D, f1) = E2)(a4,D, f2);
(v) q = 5 and Ek)(ai,D, f1) = Ek)(a,D, f2) (k β₯ 2, i = 1, 2), E1)(aj ,D, f1) =
E1)(aj ,D, f2) (j = 3, 4, 5).
2 Some Lemmas
To prove our results, we will require the following lemmas.
Lemma 2.1 (see [13, Lemma 1]). Let f(z), g(z) satisfy limrβ1β T (r, f) = β andlimrβ1β T (r, g) =β. If there is a K β (0,β) with
T (r, f) β€ KT (r, g) + S(r, f) + S(r, g),
then each S(r, f) is also an S(r, g).
6
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
899 Hong-Yan Xu 894-906
Lemma 2.2 (see [19, Lemma 3.4]). Let f(z) be a non-constant meromorphic function,a be an arbitrary complex number, and k be a positive integer. Then
N
(r,
1
f β a
)β€ k
k + 1Nk)
(r,
1
f β a
)+
1
k + 1N
(r,
1
f β a
),
and
N
(r,
1
f β a
)β€ k
k + 1Nk)
(r,
1
f β a
)+
1
k + 1T (r, f) +O(1),
where Nk)
(r, 1fβa
)are denoted by the zeros of f β a in |z| β€ r, whose multiplicities are
not greater than k and are counted only once.
From Lemma 2.2 and Theorems 1.4 and 1.8, we can get the following Lemma
Lemma 2.3 Let f(z) be a meromorphic function in D and limrβ1β T (r, f) =β, aj(j =1, 2, . . . , q) be q distinct complex numbers, and kj(j = 1, 2, . . . , q) be positive integers orβ. If f is an admissible function, thenq β 2β
qβj=1
1
kj + 1
T (r, f) β€qβj=1
kjkj + 1
Nkj)
(r,
1
f β aj
)+ S(r, f);
If f is a non-admissible function, thenq β 2βqβj=1
1
kj + 1
T (r, f) β€qβj=1
kjkj + 1
Nkj)
(r,
1
f β aj
)+ log
1
1β r+ S(r, f),
where S(r, f) is stated as in Theorem 1.4.
3 Proofs of Theorems 1.9 and 1.10
3.1 The Proof of Theorem 1.9
Suppose that there exists g(z) β Ξ₯Ξ± satisfying (5) and f(z) 6β‘ g(z). Without loss ofgenerality, we can assume that aj(j = 1, 2, . . . , q) are all finite numbers, otherwise, asuitable linear transformation will be done. Since f(z), g(z) β Ξ₯Ξ±, from Lemma 2.3, wehave
(8)
(q β 2β q
k + 1
)T (r, f) β€ k
k + 1
qβj=1
Nk)
(r,
1
f β aj
)+ log
1
1β r+ S(r, f).
If follows form (5) that
(9)
qβj=1
Nk)
(r,
1
f β aj
)β€ N
(r,
1
f β g
)β€ T (r, f) + T (r, g) +O(1).
7
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
900 Hong-Yan Xu 894-906
From (8) and (9), we have(qk
k + 1β 3k + 2
k + 1
)T (r, f) β€ k
k + 1T (r, g) + log
1
1β r+ S(r, f).
Similarly, we have(qk
k + 1β 3k + 2
k + 1
)T (r, g) β€ k
k + 1T (r, f) + log
1
1β r+ S(r, g).
Combining the above two inequalities, we get
(10)
(qk
k + 1β 4k + 2
k + 1
)T (r, f) + T (r, g) β€ 2 log
1
1β r+ S(r, f) + S(r, g).
Since f(z), g(z) β Ξ₯Ξ± and 0 < Ξ± <β, from the definition of index and q = 5+[ 2k + k+1
kΞ± ],
we have for 0 < Ξ΅ < Ξ±β k+1qkβ4kβ2 , there exists a sequence rm β 1β such that
(11) T (rm, f) > (Ξ±β Ξ΅) log1
1β rm, T (rm, g) > (Ξ±β Ξ΅) log
1
1β rm,
for all m β β. From f(z), g(z) β Ξ₯Ξ± and the assumptions of Theorem 1.9, we can see
that S(r, f) = o(
log 11βr
)and S(r, g) = o
(log 1
1βr
). From this fact and (10)-(11), we
have
(12)
2
(qk
k + 1β 4k + 2
k + 1
)(Ξ±β Ξ΅)β 2
log
1
1β rm< o
(log
1
1β rm
).
From (12) and 2(qkk+1 β
4k+2k+1
)(Ξ± β Ξ΅) β 2 > 0, we can get a contradiction. Hence, we
have f(z) β‘ g(z).Thus, this completes the proof of Theorem 1.9.
3.2 The Proof of Theorem 1.10
Suppose that there exists g(z) β Ξ₯Ξ± satisfying f(z) 6β‘ g(z) and
E3)(aj ,D, f) = E3)(aj ,D, g), (j = 1, 2, 3)(13)
E2)(aj ,D, f) = E2)(aj ,D, g), (j = 4, 5).
Without loss of generality, we can assume that aj(j = 1, 2, . . . , 5) are all finite num-bers, otherwise, a suitable linear transformation will be done. Since f(z), g(z) β Ξ₯Ξ±,from Lemma 2.3, we have(
5β 2β 3
4β 2
3
)T (r, f)(14)
β€3
4
3βj=1
N3)
(r,
1
f β aj
)+
2
3
5βj=4
N2)
(r,
1
f β aj
)+ log
1
1β r+ S(r, f)
β€3
4
3βj=1
N3)
(r,
1
f β aj
)+
5βj=4
N2)
(r,
1
f β aj
)+ log1
1β r+ S(r, f).
8
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
901 Hong-Yan Xu 894-906
From (13), we have
3βj=1
N3)
(r,
1
f β aj
)+
5βj=4
N2)
(r,
1
f β aj
)β€ N
(r,
1
f β g
)β€ T (r, f) + T (r, g) +O(1).
From this inequality and (14), we have
(15)5
6T (r, f) β€ 3
4T (r, g) + log
1
1β r+ S(r, f).
Similarly, we have
(16)5
6T (r, g) β€ 3
4T (r, f) + log
1
1β r+ S(r, g).
Since f(z), g(z) β Ξ₯Ξ± and Ξ± > 12, from the definition of index, we have for any Ξ΅(0 <Ξ΅ < Ξ± β 12), there exists a sequence rm β 1β satisfying (11) for all m β β. Fromthis fact and (15)-(16), we have
(17)
(1
6(Ξ±β Ξ΅)β 2
)log
1
1β rm< o
(log
1
1β rm
).
Since Ξ± > 12 and 0 < Ξ΅ < Ξ±β 12, we have 16 (Ξ±β Ξ΅)β 2 1
6 (Ξ±β Ξ΅)β 2 > 0, a contradiction.Hence, we have f(z) β‘ g(z).
Thus, this completes the proof of Theorem 1.10.
4 Proof of Theorem 1.11
Without loss of generality, we may assume that all aj(j = 1, 2, . . . , q) are finite, otherwise,a suitable Mobius transformation will be done. From Lemma 2.3, we haveq β 2β
qβj=1
1
kj + 1
T (r, f1) β€qβj=1
kjkj + 1
Nkj)
(r,
1
f1 β aj
)+ log
1
1β r(18)
+ S(r, f1).
From (1), we have
(19)1
2β€ kqkq + 1
β€ Β· Β· Β· β€ k2
k2 + 1β€ k1
k1 + 1β€ 1.
9
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
902 Hong-Yan Xu 894-906
From (18) and (19), we have qβj=1
kjkj + 1
β 2
T (r, f1)(20)
β€ k3
k3 + 1
qβj=1
Nkj)
(r,
1
f1 β aj
)+
2βj=1
(kj
kj + 1β k3
k3 + 1
)Nkj)
(r,
1
f1 β aj
)+ log
1
1β r+ S(r, f1)
β€ k3
k3 + 1
qβj=1
Nkj)
(r,
1
f1 β aj
)+
2βj=1
(kj
kj + 1β k3
k3 + 1
)T (r, f1)
+ log1
1β r+ S(r, f1).
If f1(z) 6β‘ f2(z), from the assumptions of Theorem 1.11, we have
(21)
qβj=1
Nkj)
(r,
1
f1 β aj
)β€ N
(r,
1
f1 β f2
)β€ T (r, f1) + T (r, f2) +O(1).
From this inequality, we have
(22)
qβj=3
kjkj + 1
+k3
k3 + 1β 2
T (r, f1) β€ k3
k3 + 1T (r, f2) + log
1
1β r+ S(r, f1).
Similarly, we have
(23)
qβj=3
kjkj + 1
+k3
k3 + 1β 2
T (r, f2) β€ k3
k3 + 1T (r, f1) + log
1
1β r+ S(r, f2).
Since 0 < Ξ±(f1), Ξ±(f2) <β, we have S(r, f1) = o(
log 11βr
), S(r, f2) = o
(log 1
1βr
). And
from the definition of index, for any Ξ΅ satisfying
0 < 2Ξ΅ < min
Ξ±(f1), Ξ±(f2), Ξ±(f1) + Ξ±(f2)β 2βqj=3
kjkj+1
,
there exists a sequence rm β 1β such that
(24) T (rm, f1) > (Ξ±(f1)β Ξ΅) log1
1β rm, T (rm, f2) > (Ξ±(f2)β Ξ΅) log
1
1β rm,
for all mββ. From (22)-(24), we have
(25)
(Ξ±(f1) + Ξ±(f2)β 2Ξ΅)
qβj=3
kjkj + 1
β 2
log1
1β rm< o
(log
1
1β rm
).
10
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
903 Hong-Yan Xu 894-906
Since 0 < 2Ξ΅ < Ξ±(f1)+Ξ±(f2)β 2
Ξ£qj=3
kjkj+1
, we have (Ξ±(f1)+Ξ±(f2)β2Ξ΅)βqj=3
kjkj+1β2 > 0,
a contradiction. Hence, we have f1(z) β‘ f2(z).Thus, this completes the proof of Theorem 1.11.
5 Proofs of Theorem 1.12 and Corollary 1.4
5.1 The Proof of Theorem 1.12
We will employ the proof by contradiction, that is, suppose that (4) and (7) hold at thesame time. Since f1(z) is admissible, from Lemma 2.3, and by using the same argumentas in Theorem 1.11, we can easily get qβ
j=3
kjkj + 1
+2k2
k2 + 1β 2
T (r, f1) β€ k2
k2 + 1(T (r, f1) + T (r, f2)) + S(r, f1),
that is,
(26)
qβj=2
kjkj + 1
β 2
T (r, f1) β€ k2
k2 + 1T (r, f2) + S(r, f1).
Set K =βqj=2
kjkj+1 β 2. If K > 0, from (26), we have
(27) T (r, f1) β€ K β²T (r, f2) + S(r, f1),
where K β² = 1K
k2k2+1 . Since kj > 0(j = 1, 2, . . . , q), we have K β² > 0 as K > 0. From this
and Lemma 2.1, we can get that each S(r, f1) is also an S(r, f2). Since f1(z) is admissibleand f2(z) is non-admissible, we can get T (r, f2) = S(r, f1). Thus, we have
(28) T (r, f2) = S(r, f1) = S(r, f2) = o(T (r, f2)).
Since limrβ1β T (r, f2) =β and (28), we can get a contradiction.Hence, we prove that (4) and (7) do not hold at the same time.
6 The Proof of Corollary 1.4
(i) Suppose that f1(z), f2(z) share four values aj(j = 1, 2, 3, 4) IM , that is, kj =β(j =1, 2, 3, 4). Since f1(z) is admissible, from Theorem 1.4, we have
(29) 2T (r, f1) β€4βj=1
N
(r,
1
f1 β aj
)+ S(r, f1).
Since f1(z), f2(z) share four values aj(j = 1, 2, 3, 4) IM , we have
(30)4βj=1
N
(r,
1
f1 β aj
)β€ N
(r,
1
f1 β f2
)β€ T (r, f1) + T (r, f2) +O(1).
11
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
904 Hong-Yan Xu 894-906
From (29) and (30), we have
(31) T (r, f1) β€ T (r, f2) + S(r, f1).
By Lemma 2.1, similar to the proof of Theorem 1.12, we have
T (r, f2) = S(r, f1) = S(r, f2) = o(T (r, f2)).
From this and limrβ1β T (r, f2) =β, we can get a contradiction.Thus, this completes (i) of Corollary 1.4.Similar to the proof of Corollary 1.4 (i), we can prove (iii),(iv) and (v) of Corollary
1.4 easily. Here we omit the detail.(ii) Suppose that f1, f2 share six values aj(j = 1, 2, . . . , 6) with reduced weight 1, that
is,
(32) E1)(aj ,D, f1) = E1)(aj ,D, f2), (j = 1, 2, . . . , 6),
and k1 = k2 = Β· Β· Β· = k6 = 1. Then, we can deduce that
6βj=2
kjkj + 1
β 2 = 5Γ 1
2β 2 =
1
2> 0.
From this and the conclusion of Theorem 1.12, we get a contradiction.Thus, this completes the proof of Corollary 1.4.
Competing interests
The authors declare that they have no competing interests.
Authorβs contributions
HYX, LZY and CFY completed the main part of this article, HYX corrected the maintheorems. All authors read and approved the final manuscript.
References
[1] T. B. Cao, H. X. Yi, Analytic functions sharing three values DM in one angular domain,J. Korean Math. Soc. 45(6)(2008), 1523-1534.
[2] T. B. Cao, H. X. Yi, On the uniqueness of meromorphic functions that share four valuesin one angular domain, J. Math. Anal. Appl. 358 (2009), 81-97.
[3] T. B. Cao, H. X. Yi, The growth of solutions of linear differential equations with coefficientsof iterated order in the unit disc, J. Math. Anal. Appl. 319 (2006), 278-294.
12
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
905 Hong-Yan Xu 894-906
[4] Z. X. Chen, K. H. Shon, The growth of solutions of differential equations with coefficientsof small growth in the disc, J. Math. Anal. Appl. 297 (2004), 285-304.
[5] M. L. Fang, On the uniqueness of admissible meromorphic functions in the unit disc, Sci.China A 42 (1999), 367-381.
[6] H. S. Gopalakrishna, S. S. Bhoosnurmath, Uniqueness theorem for meromorphic functions,Math. Scand. 39 (1976), 125-130.
[7] W. K. Hayman, Meromorphic Functions, Oxford Univ. Press, London, 1964.
[8] J. Heittokangas, On complex differential equations in the unit disc, Ann. Acad. Sci. Fenn.Math. Diss. 122 (2000), 1-54.
[9] W. C. Lin, S. Mori, K. Tohge, Uniqueness theorems in an angular domain, Tohoku Math.J. 58 (2006), 509-527.
[10] W. C. Lin, S. Mori, H. X. Yi, Uniqueness theorems of entire functions with shared-set inan angular domain, Acta Mathematica Sinica 24 (2008), 1925-1934.
[11] Z. Q. Mao, H. F. Liu, Meromorphic functions in the unit disc that share values in anangular domain, J. Math. Anal. Appl. 359 (2009), 444-450.
[12] R. Nevanlinna, Le theoreme de Picard-Borel et la theorie des fonctions meromorphes,Reprinting of the 1929 original, Chelsea Publishing Co. New York, 1974(in Frech).
[13] F. Titzhoff, Slowly growing functions sharing values, Fiz. Mat. Fak. Moksl. Semin. Darb.8 (2005), 143-164.
[14] H. Ueda, Unicity theorems for meromorphic or entire functions, Kodai Math. J. 6 (1983),26-36.
[15] Z. J. Wu, A remark on uniqueness theorems in an angular domain, Proc. Japan Acad. Ser.A 64 (6) (2008), 73-77.
[16] L. Yang, Value Distribution Theory, Springer/Science Press, Berlin/Beijing, 1993/1982.
[17] H. X. Yi, A note on the questions of multiple values and uniquness of meromorphic func-tions, Chin. Quart. J. Math. 2 (1) (1987), 99-104.
[18] H. X. Yi, Notes on a theorem of H. Ueda, J. Shandong Univ. 23 (4) (1988), 71-77.
[19] H. X. Yi, C. C. Yang, Uniqueness Theory of Meromorphic Functions, Science Press/ K-luwer., Beijing, 2003.
[20] Q. C. Zhang, Meromorphic functions sharing values in an angular domain, J. Math. Anal.Appl. 349 (2009), 100-112.
[21] J. H. Zheng, On uniqueness of meromorphic functions with shared values in some angulardomains, Canad J. Math. 47 (2004), 152-160.
[22] J. H. Zheng, On uniqueness of meromorphic functions with shared values in one angulardomains, Complex Var. Elliptic Equ. 48 (2003), 777-785.
13
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
906 Hong-Yan Xu 894-906
COMPOSITIONS INVOLVING SCHUR HARMONICALLY
CONVEX FUNCTIONS
HUAN-NAN SHI AND JING ZHANGβ
Abstract. The decision theorem of the Schur harmonic convexity for the
compositions involving Schur harmonically convex functions is established and
used to determine the Schur harmonic convexity of some symmetric functions.
2010 Mathematics Subject Classification: Primary 26D15; 05E05; 26B25
Keywords: Schur harmonically convex function; harmonically convex func-
tion; composite function; symmetric function
1. Introduction
Throughout the article, R denotes the set of real numbers, x = (x1, x2, . . . , xn)
denotes n-tuple (n-dimensional real vectors), the set of vectors can be written as
Rn = x = (x1, x2, . . . , xn) : xi,β R, i = 1, 2, . . . , n ,
Rn++ = x = (x1, x2, . . . , xn) : xi > 0, i = 1, 2, . . . , n,
Rn+ = x = (x1, x2, . . . , xn) : xi β₯ 0, i = 1, 2, . . . , n.
In particular, the notations R, R++ and R+ denote R1, R1++ and R1
+, respec-
tively.
The following conclusion is proved in reference [1, p. 91], [2, p. 64-65].
Theorem A. Let the interval [a, b] β R, Ο : Rn β R, f : [a, b] β R and
Ο(x1, x2, . . . , xn) = Ο(f(x1), f(x2), . . . , f(xn)) : [a, b]n β R.
β J. Zhang: Correspondence author.
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
907 HUAN-NAN SHI et al 907-922
2 H.-N. SHI AND J. ZHANG
(i) If Ο is increasing and Schur-convex and f is convex, then Ο is Schur-convex.
(ii) If Ο is increasing and Schur-concave and f is concave, then Ο is Schur-
concave.
(iii) If Ο is decreasing and Schur-convex and f is concave, then Ο is Schur-
convex.
(iv) If Ο is increasing and Schur-convex and f is increasing and convex, then Ο
is increasing and Schur-convex.
(v) If Ο is decreasing and Schur-convex and f is decreasing and concave, then
Ο is increasing and Schur-convex.
(vi) If Ο is increasing and Schur-convex and f is decreasing and convex, then
Ο is decreasing and Schur-convex.
(vii) If Ο is decreasing and Schur-convex and f is increasing and concave, then
Ο is decreasing and Schur-convex.
(viii) If Ο is decreasing and Schur-concave and f is decreasing and convex, then
Ο is increasing and Schur-concave.
Theorem A is very effective for determine of the Schur-convexity of the composite
functions.
The Schur harmonically convex functions were proposed by Chu et al. [3, 4, 5]
in 2009. The theory of majorization was enriched and expanded by using this
concepts. Regarding the Schur harmonically convex functions, the aim of this
paper is to establish the following theorem which is similar to Theorem A.
Theorem 1. Let the interval [a, b] β R++, Ο : Rn++ β R++, f : [a, b]β R++ and
Ο(x1, x2, . . . , xn) = Ο(f(x1), f(x2), . . . , f(xn)) : [a, b]n β R++.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
908 HUAN-NAN SHI et al 907-922
COMPOSITIONS INVOLVING SCHUR HARMONICALLY CONVEX FUNCTIONS 3
(i) If Ο is increasing and Schur harmonically convex and f is harmonically
convex, then Ο is Schur harmonically convex.
(ii) If Ο is increasing and Schur harmonically concave and f is harmonically
concave, then Ο is Schur harmonically concave.
(iii) If Ο is decreasing and Schur harmonically convex and f is harmonically
concave, then Ο is Schur harmonically convex.
(iv) If Ο is increasing and Schur harmonically convex and f is increasing and
harmonically convex, then Ο is increasing and Schur harmonically convex.
(v) If Ο is decreasing and Schur harmonically convex and f is decreasing and
harmonically concave, then Ο is increasing and Schur harmonically convex.
(vi) If Ο is increasing and Schur harmonically convex and f is decreasing and
harmonically convex, then Ο is decreasing and Schur harmonically convex.
(vii) If Ο is decreasing and Schur harmonically convex and f is increasing and
harmonically concave, then Ο is decreasing and Schur harmonically convex.
(viii) If Ο is decreasing and Schur harmonically concave and f is decreasing and
harmonically convex, then Ο is increasing and Schur harmonically concave.
2. Definitions and lemmas
In order to prove our results, in this section we will recall useful definitions and
lemmas.
Definition 1. [1, 2] Let x = (x1, x2, . . . , xn) and y = (y1, y2, . . . , yn) β Rn.
(i) x β₯ y means xi β₯ yi for all i = 1, 2, . . . , n.
(ii) Let Ξ© β Rn, Ο: Ξ© β R is said to be increasing if x β₯ y implies Ο(x) β₯
Ο(y). Ο is said to be decreasing if and only if βΟ is increasing.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
909 HUAN-NAN SHI et al 907-922
4 H.-N. SHI AND J. ZHANG
Definition 2. [1, 2] Let x = (x1, x2, . . . , xn) and y = (y1, y2, . . . , yn) β Rn.
We say y majorizes x (x is said to be majorized by y), denoted by x βΊ y,
ifβki=1 x[i] β€
βki=1 y[i] for k = 1, 2, . . . , n β 1 and
βni=1 xi =
βni=1 yi, where
x[1] β₯ x[2] β₯ Β· Β· Β· β₯ x[n] and y[1] β₯ y[2] β₯ Β· Β· Β· β₯ y[n] are rearrangements of x and y
in a descending order.
Definition 3. [1, 2] Let x = (x1, x2, . . . , xn) and y = (y1, y2, . . . , yn) β Rn.
(i) A set Ξ© β Rn is said to be a convex set if
Ξ±x + (1β Ξ±)y = (Ξ±x1 + (1β Ξ±)y1, Ξ±x2 + (1β Ξ±)y2, . . . , Ξ±xn + (1β Ξ±)yn) β Ξ©
for all x,y β Ξ©, and Ξ± β [0, 1].
(ii) Let Ξ© β Rn be convex set. A function Ο: Ξ© β R is said to be a convex
function on Ξ© if
Ο (Ξ±x + (1β Ξ±)y) β€ Ξ±Ο(x) + (1β Ξ±)Ο(y)
holds for all x,y β Ξ©, and Ξ± β [0, 1]. Ο is said to be a concave function on
Ξ© if and only if βΟ is convex function on Ξ©.
(iii) Let Ξ© β Rn. A function Ο: Ξ© β R is said to be a Schur-convex function
on Ξ© if x βΊ y on Ξ© implies Ο (x) β€ Ο (y) . A function Ο is said to be a
Schur-concave function on Ξ© if and only if βΟ is Schur-convex function on
Ξ©.
Lemma 1. (Schur-convex function decision theorem)[1, 2] : Let Ξ© β Rn be sym-
metric and have a nonempty interior convex set. Ξ©0 is the interior of Ξ©. Ο :
Ξ© β R is continuous on Ξ© and differentiable in Ξ©0. Then Ο is the Schur β
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
910 HUAN-NAN SHI et al 907-922
COMPOSITIONS INVOLVING SCHUR HARMONICALLY CONVEX FUNCTIONS 5
convex (or Schurβ concave, respectively) function if and only if Ο is symmetric
on Ξ© and
(x1 β x2)
(βΟ
βx1β βΟ
βx2
)β₯ 0(or β€ 0, respectively) (1)
holds for any x β Ξ©0.
Definition 4. [6] Let Ξ© β Rn++.
(i) A set Ξ© is said to be a harmonically convex set ifxy
Ξ»x + (1β Ξ»)yβ Ξ©
for every x,y β Ξ© and Ξ» β [0, 1], where xy =βni=1 xiyi and
1
x=( 1
x1,
1
x2, . . . ,
1
xn
).
(ii) Let Ξ© β Rn++ be a harmonically convex set. A function Ο : Ξ© β R++ be
a continuous function, then Ο is called a harmonically convex (or concave,
respectively) function, if
Ο
(1
Ξ±x + 1βΞ±
y
)β€ (or β₯, respectively)
1Ξ±
Ο(x) + 1βΞ±Ο(y)
holds for any x,y β Ξ©, and Ξ± β [0, 1].
(iii) A function Ο : Ξ© β R++ is said to be a Schur harmonically convex (or
concave, respectively) function on Ξ© if1
xβΊ 1
yimplies Ο(x) β€ (or β₯,
respectively) Ο(y).
By Definition 4, it is not difficult to prove the following propositions.
Proposition 1. Let Ξ© β Rn++ be a set, and let1
Ξ©= ( 1
x1,
1
x2, . . . ,
1
xn
):
(x1, x2, . . . , xn) β Ξ©. Then Ο : Ξ© β R++ is a Schur harmonically convex (or
concave, respectively) function on Ξ© if and only if Ο(1
x) is a Schur-convex (or
concave, respectively) function on1
Ξ©.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
911 HUAN-NAN SHI et al 907-922
6 H.-N. SHI AND J. ZHANG
In fact, for any u,v β 1
Ξ©, there exist x,y β Ξ© such that u =
1
x,v =
1
y.
Let u βΊ v, that is1
xβΊ 1
y, if Ο : Ξ© β R++ is a Schur harmonically convex
(or concave, respectively) function on Ξ©, then Ο(x) β€ (or β₯, respectively)Ο(y),
namely, Ο(1
u) β€ (or β₯, respectively)Ο(
1
v), this means that Ο(
1
x) is a Schur-convex
(or concave, respectively) function on1
Ξ©. The necessity is proved. The sufficiency
can be similar to proof.
Proposition 2. f : [a, b](β R++) β R++ is harmonically convex (or concave,
respectively) if and only if g(x) =1
f( 1x )
is concave (or convex, respectively) on[1
b,
1
a
].
In fact, for any x, y β[
1
b,
1
a
], then
1
x,
1
yβ [a, b]. If f : [a, b](β R++) β R++ is
harmonically convex (or concave, respectively), then
f
(1
Ξ±x+ (1β Ξ±)y
)β€ (or β₯, respectively)
1Ξ±
f( 1x )
+ 1βΞ±f( 1
y )
,
this is
1
f( 1Ξ±x+(1βΞ±)y )
β₯ (or β€, respectively)Ξ±
f( 1x )
+1β Ξ±f( 1
y ),
this means that g(x) =1
f( 1x )
is concave (or convex, respectively) on
[1
b,
1
a
]. The
necessity is proved. The sufficiency can be similar to proof.
Lemma 2. (Schur harmonically convex function decision theorem)[5] Let Ξ© β Rn++
be a symmetric and harmonically convex set with inner points, and let Ο : Ξ©β R++
be a continuously symmetric function which is differentiable on interior Ξ©0. Then
Ο is Schur harmonically convex (or Schur harmonically concave, respectively) on Ξ©
if and only if
(x1 β x2)
(x21βΟ(x)
βx1β x22
βΟ(x)
βx2
)β₯ 0 (or β€ 0, respectively), x β Ξ©0. (2)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
912 HUAN-NAN SHI et al 907-922
COMPOSITIONS INVOLVING SCHUR HARMONICALLY CONVEX FUNCTIONS 7
3. Proof of main results
Proof of Theorem 1. We only give the proof of Theorem 1 (vi) in detail.
Similar argument leads to the proof of the rest part.
If Ο is increasing and Schur harmonically convex and f is decreasing and harmon-
ically convex, then by Proposition 1, it follows that Ο(1
x1,
1
x2, . . . ,
1
xn) is decreasing
and Schur convex, and by Proposition 2, it follows that g(x) =1
f( 1x )
is decreasing
and concave on
[1
b,
1
a
]. And then from Theorem A (iii), it follows that
Ο
(1
g(x1),
1
g(x2), . . . ,
1
g(xn)
)= Ο
(f(
1
x1), f(
1
x2), . . . , f(
1
xn)
)
is increasing and Schur-convex. Again by Proposition 1, it follows that
Ο(x1, x2, . . . , xn) = Ο(f(x1), f(x2), . . . , f(xn))
is decreasing and Schur harmonically convex.
4. Applications
Let x = (x1, x2, . . . , xn) β Rn. Its elementary symmetric functions are
Er(x) = Er(x1, x2, . . . , xn) =β
1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
xij , r = 1, 2, . . . , n,
and defined E0(x) = 1, and Er(x) = 0 for r < 0 or r > n. The dual forms of the
elementary symmetric functions are
Eβr (x) = Eβ
r (x1, x2, . . . , xn) =β
1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
xij , r = 1, 2, . . . , n,
and defined Eβ0 (x) = 1, and Eβ
r (x) = 0 for r < 0 or r > n.
It is well-known that Er(x) is a increasing and Schur-concave function on Rn+[1].
In [7, 6], Shi proved that Eβr (x) is a increasing and Schur-concave function on
Rn+.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
913 HUAN-NAN SHI et al 907-922
8 H.-N. SHI AND J. ZHANG
Theorem 2. For r = 1, 2, . . . , n, n β₯ 2, Er(x) and Eβr (x) are Schur harmonically
convex function on Rn++.
Proof. Noting that
Er(x) = x1x2Erβ2(x3, x4, . . . , xn) + (x1 + x2)Erβ1(x3, x4, . . . , xn)
+ Er(x3, x4, . . . , xn), r = 1, 2, . . . , n,
then
(x1 β x2)
(x21βEr(x)
βx1β x22
βEr(x)
βx2
)=(x1 β x2)[x21(x2Erβ2(x3, x4, . . . , xn) + Erβ1(x3, x4, . . . , xn))β
x22(x1Erβ2(x3, x4, . . . , xn) + Erβ1(x3, x4, . . . , xn))]
=(x1 β x2)2[x1x2Erβ2(x3, x4, . . . , xn) + (x1 + x2)Erβ1(x3, x4, . . . , xn)] β₯ 0,
by Lemma 2, it follows that Er(x) is Schur harmonically convex on Rn++.
By a direct, though tedious, calculation, and according to Lemma 2, Eβ1 (x),
Eβ2 (x) is Schur harmonically convex on Rn++. When r > 2, it is easy to see that
Eβr (x) = Eβ
r (x1, x2, . . . , xn) = Eβr (x2, x3, . . . , xn)Γ
β2β€i1<i2<Β·Β·Β·<irβ1β€n
(x1 +rβ1βj=1
xij ),
then
logEβr (x) = logEβ
r (x2, x3, . . . , xn) +β
2β€i1<i2<Β·Β·Β·<irβ1β€n
log(x1 +rβ1βj=1
xij ).
Now, it leads to
1
Eβr (x)
βEβr (x)
βx1=
β2β€i1<i2<Β·Β·Β·<irβ1β€n
1
x1 +βrβ1j=1 xij
,
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
914 HUAN-NAN SHI et al 907-922
COMPOSITIONS INVOLVING SCHUR HARMONICALLY CONVEX FUNCTIONS 9
and then
βEβr (x)
βx1= Eβ
r (x)Γ β3β€i1<i2<Β·Β·Β·<irβ1β€n
1
x1 +βrβ1j=1 xij
+β
3β€i1<i2<Β·Β·Β·<irβ2β€n
1
x1 + x2 +βrβ2j=1 xij
.By the same arguments,
βEβr (x)
βx2= Eβ
r (x)Γ β3β€i1<i2<Β·Β·Β·<irβ1β€n
1
x2 +βrβ1j=1 xij
+β
3β€i1<i2<Β·Β·Β·<irβ2β€n
1
x1 + x2 +βrβ2j=1 xij
.Thus,
(x1 β x2)
(x21βEβ
r (x)
βx1β x22
βEβr (x)
βx2
)
=(x1 β x2)Eβr (x)Γ
[ β3β€i1<i2<Β·Β·Β·<irβ1β€n
(x21
x1 +βrβ1j=1 xij
β x22
x2 +βrβ1j=1 xij
)+
(x21 β x22) Β·β
3β€i1<i2<Β·Β·Β·<irβ2β€n
1
x1 + x2 +βrβ2j=1 xij
]
=(x1 β x2)2Eβr (x)Γ
[ β3β€i1<i2<Β·Β·Β·<irβ1β€n
x1x2 + (x1 + x2)βrβ1j=1 xij
(x1 +βrβ1j=1 xij )(x2 +
βrβ1j=1 xij )
+
(x1 + x2) Β·β
3β€i1<i2<Β·Β·Β·<irβ2β€n
1
x1 + x2 +βrβ2j=1 xij
]β₯ 0,
by Lemma 2, it follows that Eβr (x) is Schur harmonically convex on Rn++.
For x = (x1, x2, . . . , xn) β Rn, the complete symmetric functions cn(x, r) are
defined as
cn(x, r) =β
i1+i2+Β·Β·Β·+in=r
nβj=1
xijj , r = 1, 2, . . . , n,
where c0(x, r) = 1, r β 1, 2, . . . , n, i1, i2, . . . , in are non-negative integers.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
915 HUAN-NAN SHI et al 907-922
10 H.-N. SHI AND J. ZHANG
The dual forms of the complete symmetric functions cβn(x, r) are
cβn(x, r) =β
i1+i2+Β·Β·Β·+in=r
nβj=1
ijxj , r = 1, 2, . . . , n,
where ij(j = 1, 2, . . . , n) are non-negative integers.
Guan [8] discussed the Schur-convexity of cn(x, r) and proved that cn(x, r) is
increasing and Schur-convex on Rn++. Subsequently, Chu et al. [5] proved that
cn(x, r) is Schur harmonically convex on Rn++.
Zhang and Shi [9] proved that cβn(x, r) is increasing, Schur-concave and Schur
harmonically convex on Rn++.
In the following, we prove that the Schur harmonic convexity of the composite
functions involving the above symmetric functions and their dual form by using
Theorem 1.
Theorem 3. The following symmetric functions are increasing and Schur harmon-
ically convex on (0, 1)n, r = 1, 2, . . . , n,
Er
(1 + x
1β x
)=
β1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
1 + xij1β xij
, (3)
Eβr
(1 + x
1β x
)=
β1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
1 + xij1β xij
, (4)
cn
(1 + x
1β x, r
)=
βi1+i2+Β·Β·Β·+in=r
nβj=1
(1 + xj1β xj
)ij(5)
and
cβn
(1 + x
1β x, r
)=
βi1+i2+Β·Β·Β·+in=r
nβj=1
ij
(1 + xj1β xj
). (6)
Proof. Let f(x) =1 + x
1β x, x β (0, 1). Then f(x) > 0, f
β²(x) =
2
(1β x)2> 0, so f is
increasing on (0, 1).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
916 HUAN-NAN SHI et al 907-922
COMPOSITIONS INVOLVING SCHUR HARMONICALLY CONVEX FUNCTIONS 11
And let g(x) =1
f( 1x )
=xβ 1
x+ 1. Then g
β²β²(x) = β 4
(x+ 1)3< 0, this means that
1
f( 1x )
is concave on (1,β), by Proposition 2, it follows that f is harmonically convex
on (0, 1). Since Er(x), Eβr (x), cn(x, r) and cβn(x, r) are all increasing and Schur
harmonically convex function on Rn++, by Theorem 1 (iv), it follows that Theorem
3 holds.
Remark 1. By Lemma 2, Xia and Chu [10] proved that Er
(1 + x
1β x
)is Schur har-
monically convex on (0, 1)n. By the properties of Schur harmonically convex func-
tion, Shi and Zhang [11] proved that Eβr
(1 + x
1β x
)is Schur harmonically convex on
(0, 1)n. By Theorem 1, we give a new proof.
Theorem 4. The following symmetric functions are increasing and Schur harmon-
ically convex on Rn++, r = 1, 2, . . . , n,
Er
(x
1r
)=
β1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
x1rij, (7)
Eβr
(x
1r
)=
β1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
x1rij, (8)
cn
(x
1r , r)
=β
i1+i2+Β·Β·Β·+in=r
nβj=1
xijrj (9)
and
cβn
(x
1r , r)
=β
i1+i2+Β·Β·Β·+in=r
nβj=1
ijx1rj . (10)
Proof. For r β₯ 1, let p(x) = x1r , x β R++. Then p
β²(x) = 1
rx1rβ1 > 0, so p is
increasing on R++.
And let q(x) =1
p( 1x )
= x1r = p(x). Then q
β²β²(x) = 1
r ( 1r β 1)x
1rβ2 β€ 0, this means
that1
p( 1x )
is concave on R++, by Proposition 2, it follows that p is harmonically
convex on R++. Since Er(x), Eβr (x), cn(x, r) and cβn(x, r) are all increasing and
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
917 HUAN-NAN SHI et al 907-922
12 H.-N. SHI AND J. ZHANG
Schur harmonically convex function on Rn++, by Theorem 1 (iv), it follows that
Theorem 4 holds.
Remark 2. By Lemma 2, Chu and Lv [3] proved that the Hamyβs symmetric function
Er
(x
1r
)is Schur harmonically convex on Rn++. Later, K. Z. Guan and R. K. Guan
[12] further studied the harmonic convexity of the generalized Hamy symmetric
function.
By Lemma 2, Meng et al. [13] proved that the dual form of the Hamyβs symmetric
function Eβr
(x
1r
)is Schur harmonically convex on Rn++.
By Lemma 2, Chu and Sun [4] proved that cn
(x
1r , r)
is Schur harmonically
convex on Rn++ .
By Theorem 1, we give a new proof.
Since f(x) =1 + x
1β xis increasing and harmonically convex on (0, 1), from Theo-
rem 1 (iv) and Theorem 4, it follows
Theorem 5. The following symmetric functions are increasing and Schur harmon-
ically convex on (0, 1)n, r = 1, 2, . . . , n,
Er
((1 + x
1β x
) 1r
)=
β1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
(1 + xij1β xij
) 1r
, (11)
Eβr
((1 + x
1β x
) 1r
)=
β1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
(1 + xij1β xij
) 1r
, (12)
cn
((1 + x
1β x
) 1r
, r
)=
βi1+i2+Β·Β·Β·+in=r
nβj=1
(1 + xj1β xj
) ijr
(13)
and
cβn
((1 + x
1β x
) 1r
, r
)=
βi1+i2+Β·Β·Β·+in=r
nβj=1
ij
(1 + xj1β xj
) 1r
. (14)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
918 HUAN-NAN SHI et al 907-922
COMPOSITIONS INVOLVING SCHUR HARMONICALLY CONVEX FUNCTIONS 13
Remark 3. By Lemma 2, Long and Chu [14] proved that Eβr
((1+x1βx
) 1r
)is Schur
harmonically convex on (0, 1)n. By Theorem 1, we give a new proof.
Theorem 6. The following symmetric functions are increasing and Schur harmon-
ically convex on (0, 1)n, r = 1, 2, . . . , n,
Er
(x
1β x
)=
β1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
xij1β xij
, (15)
Eβr
(x
1β x
)=
β1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
xij1β xij
, (16)
cn
(x
1β x, r
)=
βi1+i2+Β·Β·Β·+in=r
nβj=1
(xj
1β xj
)ij(17)
and
cβn
(x
1β x, r
)=
βi1+i2+Β·Β·Β·+in=r
nβj=1
ij
(xj
1β xj
). (18)
Proof. Let h(x) =x
1β x, x β (0, 1). Then h
β²(x) =
1
(1β x)2> 0, so h is increasing
on (0, 1).
And let k(x) =1
h( 1x )
= xβ 1. Then kβ²β²(x) = 0, this means that
1
h( 1x )
is concave
on (1,β), by Proposition 2, it follows that h is harmonically convex on (0, 1).
Since Er(x), Eβr (x), cn(x, r) and cβn(x, r) are all increasing and Schur harmonically
convex function on Rn++, by Theorem 1 (iv), it follows that Theorem 5 holds.
Remark 4. By the judgement theorem of Schur harmonic convexity for a class of
symmetric functions, Shi and Zhang [15] proved that Er
(x
1β x
)is Schur harmon-
ically convex on (0, 1)n. Here by Theorem 1, we give a new proof.
By the properties of Schur harmonically convex function, Shi and Zhang [11]
proved that Eβr
(x
1β x
)is Schur harmonically convex on
[1
2, 1
)n. By Theorem
1, this conclusion is extended to the collection (0, 1)n.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
919 HUAN-NAN SHI et al 907-922
14 H.-N. SHI AND J. ZHANG
By Lemma 2, Sun et al. [16] proved that cn
(x
1β x, r
)is Schur harmonically
convex on [0, 1)n, here by Theorem 1, we give a new proof.
Since f(x) =x
1β xis increasing and harmonically convex on (0, 1), from Theo-
rem 1 (iv) and Theorem 4, it follows
Theorem 7. The following symmetric functions are increasing and Schur harmon-
ically convex on (0, 1)n, r = 1, 2, . . . , n,
Er
((x
1β x
) 1r
)=
β1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
(xij
1β xij
) 1r
, (19)
Eβr
((x
1β x
) 1r
)=
β1β€i1<i2<Β·Β·Β·<irβ€n
rβj=1
(xij
1β xij
) 1r
, (20)
cn
((x
1β x
) 1r
, r
)=
βi1+i2+Β·Β·Β·+in=r
nβj=1
(xj
1β xj
) ijr
(21)
and
cβn
((x
1β x
) 1r
, r
)=
βi1+i2+Β·Β·Β·+in=r
nβj=1
ij
(xj
1β xj
) 1r
. (22)
Remark 5. By Lemma 2, Sun [17] proved that Er
((x
1β x
) 1r
)is Schur harmon-
ically convex on [0, 1)n. Here by Theorem 1, we give a new proof.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication
of this article.
Acknowledgments
The work was supported by the Importation and Development of High-Caliber
Talents Project of Beijing Municipal Institutions (Grant No. IDHT201304089).
Thanks for the help.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
920 HUAN-NAN SHI et al 907-922
COMPOSITIONS INVOLVING SCHUR HARMONICALLY CONVEX FUNCTIONS 15
References
[1] A. W. Marshall, I. Olkin, and B. C. Arnold, Inequalities: Theory of Majorization and Its
Application (Second Edition), Springer, New York, 2011.
[2] B. Y. Wang, Foundations of Majorization Inequalities, Beijing Normal University Press,
Beijing, 1990. (in Chinese)
[3] Y. M. Chu, and Y. P. Lv, The Schur harmonic convexity of the Hamy symmetric function
and its applications, Journal of Inequalities and Applications, 2009, Article ID 838529, 10
pages.
[4] Y. M. Chu, and T. C. Sun, The Schur harmonic convexity for a class of symmetric functions,
Acta Mathematica Scientia, 2010, 30B(5), 1501-1506.
[5] Y. M. Chu, G. D. Wang, and X. H. Zhang, The Schur multiplicative and harmonic convexities
of the complete symmetric function, Mathematische Nachrichten, 2011, 284(5-6), 653-663.
[6] H. N. Shi, Theory of Majorization and Analytic Inequalities, Harbin Institute of Technology
Press, Harbin, 2012. (in Chinese)
[7] H. N. Shi, Schur-concavity and Schur-geometrically convexity of dual form for elementary
symmetric function with applications, RGMIA Research Report Collection, 2007, 10(2),
http://rgmia.vu.edu.au/v10n2.html
[8] K. Z. Guan, Schur-convexity of the complete symmetric function, Mathematical Inequalities
& Applications, 2006, 9(4), 567-576.
[9] K. S. Zhang, and H. N. Shi, Schur convexity of dual form of the complete symmetric function,
Mathematical Inequalities & Applications, 2013, 16(4), 963-970.
[10] W. F. Xia, and Y. M. Chu, On Schur convexity of some symmetric functions, Journal of
Inequalities and Applications, 2010, Article ID 543250, 12 pages.
[11] H. N. Shi, and J. Zhang, Schur-convexity, Schur geometric and Schur harmonic convexities of
dual form of a class symmetric functions, Journal of Mathematical Inequalities, 2014, 8(2),
349-358.
[12] K. Z. Guan, and R. K. Guan, Some properties of a generalized Hamy symmetric function
and its applications, Journal of Mathematical Analysis and Applications, 2011, 376, 494-505.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
921 HUAN-NAN SHI et al 907-922
16 H.-N. SHI AND J. ZHANG
[13] Junxia Meng, Yuming Chu, and Xiaomin Tang, The Schur-harmonic-convexity of dual form
of the Hamy symmetric function, Matematicki Vesnik, 2010, 62(1), 37-46.
[14] B. Y. Long, and Y. M. Chu, The Schur convexity and inequalities for a class of symmetric
functions, Acta Mathematica Scientia, 2012, 32A(1), 80-89. (in Chinese)
[15] H. N. Shi, and J. Zhang, Some new judgement theorems of Schur geometric and Schur har-
monic convexities for a class of symmetric functions, Journal of Inequalities and Applications
2013, 2013: 527, doi: 10.1186/1029-242X-2013-527.
[16] M. B. Sun, N. B. Chen, and S. H. Li, Some properties of a class of symmetric functions and
its applications, Mathematische Nachrichten, 2014, 287(13), 1530-1544, doi: 10.1002/man-
a.201300073.
[17] M. B. Sun, The Schur convexity for two classes of symmetric functions, Sci. Sin. Math., 2014,
44(6), 633-656, doi: 10.1360/N012013-00157. (in Chinese)
(H.-N. Shi) Department of Electronic Information, Teacherβs College, Beijing Union
University, Beijing 100011, P.R.China
E-mail address: [email protected], [email protected]
(J. Zhang) Basic Courses Department of Beijing Union University, Beijing 100101,
P.R.China
E-mail address: [email protected], [email protected]
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
922 HUAN-NAN SHI et al 907-922
A NOTE ON DEGENERATE GENERALIZED q-GENOCCHI
POLYNOMIALS
JONGKYUM KWON1, JIN-WOO PARK2, AND SANG JO YUN3,β
Abstract. In this paper,we consider degenerate generalized q-Genocchi poly-
nomials arising from p-adic fermionic q-integral on Zp. We found some inter-esting identities of these polynomials.
1. Introduction
Let p be a fixed odd prime number. Throughout this paper, Zp, Qp and Cp willdenote the ring of p-adic rational integers, the field of p-adic rational numbers andthe completion of the algebraic closure of Qp, respectively. Let Ξ½p be the normalized
exponential valuation of Cp with |p|p = pβΞ½p(p) = 1p . Let q be an indeterminate in
Cp such that |q β 1|p < pβ1pβ1 .
Let f(x) be a continuous function on Zp. Then the p-adic fermionic q-integralon Zp is defined by Kim to beβ«
Zpf(x)dΒ΅βq(x) = lim
Nββ
pNβ1βx=0
f(x)Β΅βq(x+ pNZp
)=
[2]q2
limNββ
pNβ1βx=0
f(x)(β1)xqx, (see [9]).
(1.1)
Thus, by (1.1), we get
q
β«Zpf(x+ 1)dΒ΅βq(x) +
β«Zpf(x)dΒ΅βq(x) = [2]qf(0), (1.2)
and
qnβ«Zpf(x+n)dΒ΅βq(x)+(β1)nβ1
β«Zpf(x)dΒ΅βq(x) = [2]q
nβ1βl=0
f(l)ql(β1)nβ1βl, (1.3)
where n β N (see [5-10, 12]).It is known that the q-Euler polynomials are given by the generating function as
follows: β«Zpe(x+y)tdΒ΅βq(y) =
[2]qqet + 1
ext =ββn=0
En,q(x)tn
n!. (1.4)
When x = 0, En,q = En,q(0) are called q-Euler numbers (see [5, 9, 12]).
1991 Mathematics Subject Classification. 05A10, 05A19.β corresponding author.
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
923 JONGKYUM KWON et al 923-928
2 JONGKYUM KWON1, JIN-WOO PARK2, AND SANG JO YUN3,β
Recently, degenerate q-Euler polynomials are introduced by he generating func-tion as follows:
[2]q
q(1 + Ξ»t)1Ξ» + 1
(1 + Ξ»t)xΞ» =
ββn=0
En,q,Ξ»(x)tn
n!, (see [9]). (1.5)
It is known that the q-Genocchi polynomials are given by the generating functionas follows: β«
Zpte(x+y)tdΒ΅βq(y) =
t[2]qqet + 1
ext =ββn=0
Gn,q(x)tn
n!. (1.6)
When x = 0, Gn,q = En,q(0) are called q-Genocchi numbers (see [1, 2, 4, 6-8]).Now, the degenerate q-Genocchi polynomials are introduced by the generating
function as follows:
t[2]q
q(1 + Ξ»t)1Ξ» + 1
(1 + Ξ»t)xΞ» =
ββn=0
Gn,q,Ξ»(x)tn
n!. (1.7)
Note that limΞ»β0 Gn,q,Ξ»(x) = Gn,q(x), (n β₯ 0), (see [3]).For d β N with d β‘ (mod 2) and (d, p) = 1, we set
X = limββN
ZdpNZ, Xβ =β
0<a<dp, a-p
(a+ dpZp) ,
and
a+ dpNZp =x β X | x β‘ a ( mod dpN )
,
where a β Z with 0 β€ a < dpN β 1.For d β N with d β‘ 1 (mod 2), let us assume that Ο is a Dirichlet character with
conductor d. Now, we consider the generalized q-Genocchi polynomials attached toΟ which are given by the generating function to beβ«
X
Ο(y)te(x+y)tdΒ΅βq(y) =
(t[2]q
qdedt + 1
dβ1βa=0
qq(β1)aΟ(a)eat
)ext
=ββn=0
Gn,q,Ο(x)tn
n!, (see [4-6, 8]).
(1.8)
When x = 0, Gn,q,Ο = Gn,q,Ο(0) are called generalized q-Genocchi numbers attachedto Ο.
One of the most recent papers on the theory of Genocchi polynomials and num-bers is the paper T. Kim(see [6-8]), which deals mainly with the theory of Genocchipolynomials and numbers. Facts on Bernoulli polynomials and Euler polynomials,to which Genocchi polynomials may be related, has been derived in Volkenbornintegral (see [3]). While a lot of the properties of Genocchi polynomials bear astriking resemblance to the properties of Beroulli and Euler polynomials, someproperties are rather different. Note that Genocchi polynomials occur naturally inthe areas of elementary number theory, complex analytic number theory, homo-topy theory, differential topology, theory of modular forms, p-adic analytic numbertheory, quantum physics (see [1-13]).
In the viewpoint of (1.8), we consider degenerate generalized q-Genocchi poly-nomials which are derived from the fermionic q-integral on Zp. The purpose ofthis paper is to investigate some properties and identities of degenerate generalizedq-Genocchi polynomials.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
924 JONGKYUM KWON et al 923-928
DEGENERATE GENERALIZED q-GENOCCHI POLYNOMIALS 3
2. Degenerate generalized q-Genocchi polynomials
In this section, we assume that Ξ», t β Cp with |Ξ»t|p < pβ1pβ1 . For d β N with
d β‘ 1 (mod 2) , let Ο be a Dirichletβs character with conductor d.In the viewpoint of (1.8), we consider degenerate generalized q-Genocchi poly-
nomials which are given by the generating function to beβ«X
Ο(y)t(1 + Ξ»t)x+yΞ» dΒ΅βq(y)
=
(t[2]q
qd(1 + Ξ»t)dΞ» + 1
dβ1βa=0
qa(β1)aΟ(a)(1 + Ξ»t)aΞ»
)(1 + Ξ»t)
xΞ»
=ββn=0
Gn,Ο,q,Ξ»(x)tn
n!,
(2.1)
where d β N with d β‘ 1 (mod 2).From (1.5) and (2.1), we have
ββn=0
Gn,Ο,q,Ξ»(x)tn
n!=
(t[2]q
qd(1 + Ξ»t)dΞ» + 1
dβ1βa=0
qa(β1)aΟ(a)(1 + Ξ»t)a+xΞ»
)
=[2]q[2]qd
dβ1βa=0
qa(β1)aΟ(a)
(t[2]qd
qd(1 + Ξ»t)dΞ» + 1
(1 + Ξ»t)dΞ»a+xd
)
=ββn=0
([2]q[2]qd
dβ1βa=0
qa(β1)aΟ(a)Gn,qd,Ξ»d
(a+ x
d
)dntn+1
n!
).
(2.2)
Thus, by (2.2), we get
Gn,Ο,q,Ξ»(x) =ndnβ1[2]q
[2]qd
dβ1βa=0
qa(β1)aΟ(a)Gnβ1,qd,Ξ»d
(a+ x
d
), (n β₯ 0). (2.3)
Therefore, by (2.3), we obtain the following theorem.
Theorem 2.1. For d β N with d β‘ 1 (mod 2), n β₯ 0, we have
Gn,Ο,q,Ξ»(x) =
β«X
Ο(y)t(x+ y|Ξ»)ndΒ΅βq(y)
=n[2]q[2]qd
dnβ1dβ1βa=0
Ο(a)qa(β1)aGnβ1,qd,Ξ»d
(a+ x
d
),
where
(x|Ξ»)n =x(xβ Ξ») Β· Β· Β· (xβ Ξ»(nβ 1))
=Ξ»n(xΞ»
)n.
For n β₯ 0, we observe that
(x+ y|Ξ»)n =Ξ»n(x+ y
Ξ»
)n
= Ξ»nnβl=0
S1(n, l)
(x+ y
Ξ»
)l=
nβl=0
S1(n, l)Ξ»nβl(x+ y)l.
(2.4)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
925 JONGKYUM KWON et al 923-928
4 JONGKYUM KWON1, JIN-WOO PARK2, AND SANG JO YUN3,β
By (2.4), we getβ«X
Ο(y)t(x+ y|Ξ»)ndΒ΅βq(y) =nβl=0
S1(n, l)Ξ»nβlβ«X
Ο(y)t(x+ y)ldΒ΅βq(y)
=nβl=0
S1(n, l)Ξ»nβlGl,q,Ο(x), (n β₯ 0),
(2.5)
where S1(n, l) is the Stirling number of the first kind. Therefore, by (2.5), we obtainthe following theorem.
Theorem 2.2. For n β₯ 0, we have
Gn,Ο,q,Ξ»(x) =nβl=0
S1(n, l)Ξ»nβlGl,q,Ο(x).
By replacing t by 1Ξ» (eΞ»t β 1) in (2.1), we getβ«
X
Ο(y)1
Ξ»(eΞ»t β 1)e(x+y)tdΒ΅βq(y) =
ββm=0
Gm,Ο,q,Ξ»(x)1
m!
(1
Ξ»(eΞ»t β 1)
)m=ββm=0
Gm,Ο,q,Ξ»(x)Ξ»βmββn=m
S2(n,m)Ξ»n
n!tn
=ββn=0
(nβ
m=0
Ξ»nβmS2(n,m)Gm,Ο,q,Ξ»(x)
)tn
n!,
(2.6)
where S2(n,m) is the Stirling number of the second kind.From (1.8), we note thatβ«
X
Ο(y)te(x+y)tdΒ΅βq(y) =
(t[2]q
qdedt + 1
) dβ1βa=0
qa(β1)aΟ(a)e(a+x)t
=ββn=0
Gn,q,Ο(x)tn
n!.
(2.7)
By multiplying t on the both side (2.6), we getβ«X
Ο(y)te(x+y)tdΒ΅βq(y) =ββn=0
(nβ
m=0
1
eΞ»t β 1Ξ»nβm+1S2(n,m)Gm,Ο,q,Ξ»(x)
)tn+1
n!.
(2.8)
Therefore, by (2.6), (2.7) and (2.8), we obtain the following theorem.
Theorem 2.3. For n β₯ 0, we have
Gn,q,Ο(x) =nβ1βm=0
n
eΞ»t β 1Ξ»nβmS2(nβ 1,m)Gm,Ο,q,Ξ»(x).
Let d β N with d β‘ 1 (mod 2). From (1.3), we have
qdβ«X
(x+ d|Ξ»)nΟ(x)dΒ΅βq(x) +
β«X
(x|Ξ»)nΟ(x)dΒ΅βq(x)
=[2]q
dβ1βa=0
Ο(a)qq(β1)a(a|Ξ»)n, (n β₯ 0).
(2.9)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
926 JONGKYUM KWON et al 923-928
DEGENERATE GENERALIZED q-GENOCCHI POLYNOMIALS 5
Therefore, by Theorem 2.1 and (2.8), we obtain the following theorem.
Theorem 2.4. For d β N with d β‘ 1 (mod 2), n β₯ 0, we have
qdGn,Ο,q,Ξ»(d) + Gn,Ο,q,Ξ» = t[2]q
dβ1βa=0
Ο(a)qq(β1)a(a|Ξ»)n,
where Gn,Ο,q,Ξ» = Gn,Ο,q,Ξ»(0) are called degenerate generalized q-Genocchi numbersattached to Ο.
Now, we observe that
ββn=0
Gn,Ο,q,Ξ»(x)tn
n!=
(t[2]q
βdβ1a=0 q
a(β1)aΟ(a)(1 + Ξ»t)aΞ»
qd(1 + Ξ»t)dΞ» + 1
)(1 + Ξ»t)
xΞ»
=
( ββm=0
Gn,Ο,q,Ξ»tm
m!
)( ββl=0
(x|Ξ»)ltl
l!
)
=ββn=0
(nβ
m=0
(n
m
)Gm,Ο,q,Ξ»(x|Ξ»)nβm
)tn
n!.
(2.10)
Thus, by comparing the coefficients on the both sides, we obtain the followingtheorem.
Theorem 2.5. For n β₯ 0, we have
Gn,Ο,q,Ξ»(x) =nβ
m=0
(n
m
)Gm,Ο,q,Ξ»(x|Ξ»)nβm.
Now, we observe that
(β1)n
n!Gn,Ο,q,Ξ» =
(β1)n
n!
β«X
Ο(x)t(x|Ξ»)ndΒ΅βq(x)
=Ξ»nβ«X
(βxΞ» + nβ 1
n
)Ο(x)tdΒ΅βq(x)
=Ξ»nnβl=0
(nβ 1
l β 1
)(β1)l
Ξ»ll!
β«X
Ο(x)t(x| β Ξ»)ldΒ΅βq(x)
=nβl=0
(nβ 1
l β 1
)Ξ»nβl(β1)l
1
l!Gl,Ο,q,βΞ»
=nβl=1
(nβ 1
l β 1
)Ξ»nβl(β1)l
Gl,Ο,q,βΞ»l!
.
(2.11)
Therefore, by (2.11), we obtain the following theorem.
Theorem 2.6. For n β₯ 0, we have
(β1)n
n!Gn,Ο,q,Ξ» =
nβl=1
(nβ 1
l β 1
)Ξ»nβl(β1)l
Gl,Ο,q,βΞ»l!
.
Note that
limΞ»β0Gn,Ο,q,Ξ»(x) = Gn,q,Ο(x), (n β₯ 0).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
927 JONGKYUM KWON et al 923-928
6 JONGKYUM KWON1, JIN-WOO PARK2, AND SANG JO YUN3,β
References
[1] S. Araci, M. Acikgoz, H. Jolany, and Y. He, Identities involving q-Genocchi Numbers and
polynomials, Notes on Num. Theo. and Disc. Math., 20 (2014), no.5, 64-74.[2] S. Araci, M. Acikgoz, and F. Qi, On the q-Genocchi Numbers and polynomials with weight
zero and their applications, Nonlinear Func. Anal. and Appl., 18 (2013), 193-203.[3] L. Carlitz, Degenerate Stirling, Bernoulli and Eulerian numbers, Util. Math., 15 (1979),
51-88.
[4] M. Cenkci, M. Can, and V. Kurt, q-Extensions of Genocchi Numbers, J. Korean Math. Soc.,43 (2006), 183-198.
[5] T. Kim, Some identities on the q-Euler polynomials of higher-order and q-Stirling numbers
by the fermionic p-adic integral on Zp, Russ. J. Math. Phys., 16 (2009), 484-491.[6] T. Kim, On the Multiple q-Genocchi and Euler Numbers, Russ. J. Math. Phys., 15 (2008),
481-486.
[7] T. Kim, On the q-Extension of Euler and Genocchi Numbers, J. Math. Anal. Appl. 326(2007), 1458-1465.
[8] T. Kim, A note on the q-Genocchi Numbers and polynomials, J. Ineq. and Appl., 2007
Article ID 71452, 8pages,(2007).[9] T. Kim, D. S. Kim and D. V. Dolgy, Degenerate q-Euler polynomials, Adv. Difference Equ.,
2015 (2015), 13662.
[10] D. S. Kim, T. Kim, D. V. Dolgy and T. Komatsu, Barnes-type degenerate Bernoulli polyno-mials, Adv. Stud. Contemp. Math., 25 (2015), 121-146.
[11] Q. M. Luo and F. Qi, Relationship between generalized Bernoulli numbers and polynomialsand generalized Euler numbers and polynomials, Adv. Stud. Contemp. Math., 7 (2003), 11-18.
[12] S. H. Rim and J. Jeong, On the modified q-Euler numbers of higher-order with weight, Adv.
Stud. Contemp. Math., 22 (2012), 93-98.[13] E. Sen, Theorems on Apostol-Euler polynomials of higher-order arising from Euler basis,
Adv. Stud. Contemp. Math., 23 (2013), 337-345..
1 Department of Mathematics educations and RINS, Gyeongsang National Univer-
sity, JinJu, 52828, Republic of KoreaE-mail address: [email protected]
2 Department of Mathematics education, Daegu University, Gyeongsan, 712-714, Ko-rea.
E-mail address: [email protected]
3 Department of Mathematics education, Daegu University, Gyeongsan, 712-714, Ko-
rea
E-mail address: [email protected]
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
928 JONGKYUM KWON et al 923-928
Cubic soft ideals in BCK/BCI-algebras
Young Bae Jun1, G. Muhiuddin2, Mehmet Ali Ozturk3 and
Eun Hwan Roh4,β1Department of Mathematics Education, Gyeongsang National University, Jinju 660-701, Korea2Department of Mathematics, University of Tabuk, P. O. Box 741, Tabuk 71491, Saudi Arabia
3Department of Mathematics, Faculty of Arts and Sciences, AdΔ±yaman University 02040
AdΔ±yaman, Turkey4Department of Mathematics Education, Chinju National University of Education, Jinju
660-756, Korea
Abstract. The notions of cubic soft -subalgebras and (closed) cubic soft ideals in BCK/BCI-algebras are intro-
duced, and related properties are investigated. Relations between cubic soft subalgebras, cubic soft -subalgebras
and (closed) cubic soft ideals are discussed. Conditions for a cubic soft subalgebras to be a (closed) cubic soft
ideals are provided. Characterizations of cubic soft ideals are considered. R-union and R-intersection of cubic soft
ideals are discussed.
1. Introduction
To solve complicated problems in economics, engineering, and environment, we canβt success-
fully use classical methods because of various uncertainties typical for those problems. Uncer-
tainties canβt be handled using traditional mathematical tools but may be dealt with using a wide
range of existing theories such as probability theory, theory of (intuitionistic) fuzzy sets, theory
of vague sets, theory of interval mathematics, and theory of rough sets. However, all of these
theories have their own difficulties which are pointed out in [8]. Maji et al. [5] and Molodtsov [8]
suggested that one reason for these difficulties may be due to the inadequacy of the parametriza-
tion tool of the theory. To overcome these difficulties, Molodtsov [8] introduced the concept of
soft set as a new mathematical tool for dealing with uncertainties that is free from the difficulties
that have troubled the usual theoretical approaches. Molodtsov pointed out several directions
for the applications of soft sets. At present, works on the soft set theory are progressing rapidly.
Maji et al. [5] described the application of soft set theory to a decision making problem. Maji
et al. [6] also studied several operations on the theory of soft sets. Jun et al. [2, 4] applied the
notion of soft sets to BCK/BCI-algebras and d-algebras.
02010 Mathematics Subject Classification: 06F35, 03G25, 06D72..0Keywords: cubic soft subalgebra, cubic soft -subalgebra, (closed) cubic soft ideal, R-union, R-intersection.
β The corresponding author. Tel: +82 55 740 12320E-mail: [email protected] (Y. B. Jun); [email protected] (G. Muhiuddin);
[email protected] (M. A. Ozturk); [email protected] (E. H. Roh)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
929 Young Bae Jun et al 929-940
Young Bae Jun, Seok Zun Song and Sun Shin Ahn
Combining cubic sets and soft sets, Muhiuddin and Al-roqi [10] introduced the notions of
(internal, external) cubic soft sets, P-cubic (resp. R-cubic) soft subsets, R-union (resp. R-
intersection, P-union, P-intersection) of cubic soft sets, and the complement of a cubic soft
set. They investigated several related properties, and applied the notion of cubic soft sets to
BCK/BCI-algebras. In [9], Muhiuddin et al. considered several basic operations of cubic soft
sets, namely, βANDβ operation and βORβ operation based on the P-order and the R-order.
They provided an example to illustrate that the R-union of two internal cubic soft sets might
not be internal. They also discussed conditions for the R-union of two internal cubic soft sets
to be an internal cubic soft set, and investigated several properties of cubic soft subalgebras in
BCK/BCI-algebras based on a parameter.
In this paper, we introduce the notions of cubic soft -subalgebras and (closed) cubic soft ideals
in BCK/BCI-algebras, and investigate related properties. We consider relations between cubic
soft subalgebras, cubic soft -subalgebras and (closed) cubic soft ideals, and provide conditions
for a cubic soft subalgebras to be a (closed) cubic soft ideals. We discuss characterizations of
cubic soft ideals. We show that the R-intersection of cubic soft ideals is a cubic soft ideal. We
also show that if parameter sets are mutually disjoint then the R-union of cubic soft ideals is a
cubic soft ideal. We provide an example to show that the R-union of cubic soft ideals is not a
cubic soft ideal when parameter sets are not disjoint.
2. Preliminaries
An algebra (X; β, 0) of type (2, 0) is called a BCI-algebra if it satisfies the following axioms:
(I) (βx, y, z β X) (((x β y) β (x β z)) β (z β y) = 0),
(II) (βx, y β X) ((x β (x β y)) β y = 0),
(III) (βx β X) (x β x = 0),
(IV) (βx, y β X) (x β y = 0, y β x = 0 β x = y).
If a BCI-algebra X satisfies the following identity:
(V) (βx β X) (0 β x = 0),
then X is called a BCK-algebra. Any BCK/BCI-algebra X satisfies the following conditions:
(a1) (βx β X) (x β 0 = x),
(a2) (βx, y, z β X) (x β y = 0 β (x β z) β (y β z) = 0, (z β y) β (z β x) = 0),
(a3) (βx, y, z β X) ((x β y) β z = (x β z) β y),
(a4) (βx, y, z β X) (((x β z) β (y β z)) β (x β y) = 0).
We can define a partial ordering β€ by x β€ y if and only if x β y = 0. A BCK-algebra X is said
to be with condition (S) if, for all x, y β X, the set z β X | z β x β€ y has a greatest element,
written x y. A BCI-algebra X is said to be p-semisimple if its BCK-part is equal to 0. In a
p-semisimple BCI-algebra, the following conditions are valid:
(a5) (βx, y β X) (0 β (x β y) = y β x).
(a6) (βx, y β X) (x β (x β y) = y).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
930 Young Bae Jun et al 929-940
Cubic soft ideals in BCK/BCI-algebras
A nonempty subset S of a BCK/BCI-algebra X is called a subalgebra of X if x β y β S for
all x, y β S. A subset I of a BCK/BCI-algebra X is called an ideal of X if it satisfies, for all
x, y β X, the following conditions:
(b1) 0 β I,
(b2) x β y β I, y β I β x β I.
An ideal I of a BCI-algebra X is said to be closed if 0 β x β I for all x β I. We refer the reader
to the books [1, 7] for further information regarding BCK/BCI-algebras.
By an interval number we mean a closed subinterval a = [aβ, a+] of I, where 0 β€ aβ β€ a+ β€ 1.
Denote by [I] the set of all interval numbers. Let us define what is known as refined minimum
and refined maximun (briefly, rmin and rmax) of two elements in [I ]. We also define the symbols
βΒΊβ, βΒΉβ, β=β in case of two elements in [I]. Consider two interval numbers a1 :=[aβ1 , a+
1
]and
a2 :=[aβ2 , a+
2
]. Then
rmin a1, a2 =[min
aβ1 , aβ2
, min
a+
1 , a+2
],
rmax a1, a2 =[max
aβ1 , aβ2
, max
a+
1 , a+2
],
a1 ΒΊ a2 if and only if aβ1 β₯ aβ2 and a+1 β₯ a+
2 ,
and similarly we may have a1 ΒΉ a2 and a1 = a2. To say a1 Γ a2 (resp. a1 βΊ a2) we mean a1 ΒΊ a2
and a1 6= a2 (resp. a1 ΒΉ a2 and a1 6= a2).
Let X be a nonempty set. A function A : X β [I] is called an interval-valued fuzzy set
(briefly, an IVF set) in X. Let [I]X stand for the set of all IVF sets in X. For every A β [I]X and
x β X, A(x) = [Aβ(x), A+(x)] is called the degree of membership of an element x to A, where
Aβ : X β I and A+ : X β I are fuzzy sets in X which are called a lower fuzzy set and an upper
fuzzy set in X, respectively. For simplicity, we denote A = [Aβ, A+].
Molodtsov [8] defined the soft set in the following way: Let U be an initial universe set and E
be a set of parameters. Let P(U) denotes the power set of U and A β E.
Definition 2.1 ([8]). A pair (F,A) is called a soft set over U, where F is a mapping given by
F : A β P(U).
In other words, a soft set over U is a parameterized family of subsets of the universe U. For
Ξ΅ β A, F (Ξ΅) may be considered as the set of Ξ΅-approximate elements of the soft set (F,A). Clearly,
a soft set is not a set. For illustration, Molodtsov considered several examples in [8].
Definition 2.2 ([3]). Let U be a universe. By a cubic set in U we mean a structure
A = γx,A(x), Ξ»(x)γ | x β Uin which A is an IVF set in U and Ξ» is a fuzzy set in U.
In what follows, a cubic set A = γx, Β΅A(x), Ξ»A(x)γ | x β U is simply denoted by A =
γΒ΅A, Ξ»Aγ, and denote by CU the collection of all cubic sets in U.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
931 Young Bae Jun et al 929-940
Young Bae Jun, Seok Zun Song and Sun Shin Ahn
Definition 2.3 ([10]). Let U be an initial universe set and let E be a set of parameters. A cubic
soft set over U is defined to be a pair (F , A) where F is a mapping from A to CU and A β E.
Note that the pair (F , A) can be represented as the following set:
(F , A) := F (Ξ΅) | Ξ΅ β A where F (Ξ΅) = γΒ΅F (Ξ΅), Ξ»F (Ξ΅)γ.
3. Cubic soft ideals
In what follows, let U be an initial universe set which is a BCK/BCI-algebra unless otherwise
specified.
Definition 3.1 ([10]). A cubic soft set (F , A) over U is said to be a cubic soft BCK/BCI-
algebra over U based on a parameter Ξ΅ (briefly, Ξ΅-cubic soft subalgebra over U) if there exists a
parameter Ξ΅ β A such that
Β΅F (Ξ΅)(x β y) ΒΊ rminΒ΅F (Ξ΅)(x), Β΅F (Ξ΅)(y) (3.1)
Ξ»F (Ξ΅)(x β y) β€ maxΞ»F (Ξ΅)(x), Ξ»F (Ξ΅)(y) (3.2)
for all x, y β U. If (F , A) is an Ξ΅-cubic soft subalgebra over U for all Ξ΅ β A, we say that (F , A)
is a cubic soft subalgebra over U .
Definition 3.2. Let U be a BCK-algebra with the condition (S). Given a parameter Ξ΅ β A, a
cubic soft set (F , A) over U is said to be a cubic soft -subalgebra over U based on Ξ΅ (briefly,
Ξ΅-cubic soft -subalgebra over U) if it satisfies the following conditions:
Β΅F (Ξ΅)(x y) ΒΊ rminΒ΅F (Ξ΅)(x), Β΅F (Ξ΅)(y) (3.3)
Ξ»F (Ξ΅)(x y) β€ maxΞ»F (Ξ΅)(x), Ξ»F (Ξ΅)(y) (3.4)
for all x, y β U. If (F , A) is an Ξ΅-cubic soft -subalgebra over U for all Ξ΅ β A, we say that (F , A)
is a cubic soft -subalgebra over U .
Definition 3.3. Given a parameter Ξ΅ β A, a cubic soft set (F , A) over U is said to be a cubic soft
ideal over U based on Ξ΅ (briefly, Ξ΅-cubic soft ideal over U) if it satisfies the following conditions:
Β΅F (Ξ΅)(0) ΒΊ Β΅F (Ξ΅)(x), Ξ»F (Ξ΅)(0) β€ Ξ»F (Ξ΅)(x), (3.5)
Β΅F (Ξ΅)(x) ΒΊ rminΒ΅F (Ξ΅)(x β y), Β΅F (Ξ΅)(y) (3.6)
Ξ»F (Ξ΅)(x) β€ maxΞ»F (Ξ΅)(x β y), Ξ»F (Ξ΅)(y) (3.7)
for all x, y β U. If (F , A) is an Ξ΅-cubic soft ideal over U for all Ξ΅ β A, we say that (F , A) is a
cubic soft ideal over U .
Example 3.4. Let (Y, β, 0) be a BCI-algebra and consider the adjoint BCI-algebra (Z,β, 0) of
the additive group (Z, +, 0) of integers. Then the direct product U := Y Γ Z of Y and Z is a
BCI-algebra (see [1]). For any Ξ΅ β A, let (F , A) be a soft set over U defined by
Β΅F (Ξ΅)(x) =
a = [aβ, a+]( 6= [0, 0]) if x β Y Γ N0,
[0,0] otherwise,
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
932 Young Bae Jun et al 929-940
Cubic soft ideals in BCK/BCI-algebras
Ξ»F (Ξ΅)(x) =
s if x β Y Γ N0,
t otherwise,
where N is the set of natural numbers, N0 = N βͺ 0 and s, t β [0, 1] with s < t. Then (F , A) is
an Ξ΅-cubic soft ideal over U .
Example 3.5. Let U = 0, a, b, c be a BCI-algebra with the following Cayley table 1:
Table 1. Cayley table of the operation β
β 0 a b c
0 0 0 0 c
a a 0 0 c
b b b 0 c
c c c c 0
Let (F , A) be a cubic soft set over U , where A = Ξ΅1, Ξ΅2, Ξ΅3, with the tabular representation
in Table 2.
Table 2. Tabular representation of the cubic soft set (F , A)
Ξ΅1 Ξ΅2 Ξ΅3
0 γ[0.5, 0.8], 0.6γ γ[0.8, 1.0], 0.1γ γ[0.4, 0.6], 0.7γa γ[0.8, 0.9], 0.7γ γ[0.3, 0.7], 0.8γ γ[0.1, 0.2], 0.7γb γ[0.1, 0.7], 0.5γ γ[0.3, 0.7], 0.8γ γ[0.1, 0.7], 0.3γc γ[0.2, 0.6], 0.9γ γ[0.3, 0.7], 0.8γ γ[0.3, 0.6], 0.2γ
Then (F , A) is not an Ξ΅1-cubic soft ideal over U since
Β΅F (Ξ΅1)(0) = [0.5, 0.8] [0.8, 0.9] = Β΅F (Ξ΅1)(a).
We know that (F , A) is an Ξ΅2-cubic soft ideal over U . (F , A) is not an Ξ΅3-cubic soft ideal over
U since
Β΅F (Ξ΅3)(a) = [0.1, 0.2] [0.3, 0.6] = rminΒ΅F (Ξ΅3)(a β c), Β΅F (Ξ΅3)(c).Proposition 3.6. If (F , A) is an Ξ΅-cubic soft ideal over U , then
(βx, y β U)(x β€ y β Β΅F (Ξ΅)(y) ΒΉ Β΅F (Ξ΅)(x), Ξ»F (Ξ΅)(y) β₯ Ξ»F (Ξ΅)(x)
).
Proof. Let x, y β U be such that x β€ y. Then x β y = 0, and so
Β΅F (Ξ΅)(x) ΒΊ rminΒ΅F (Ξ΅)(x β y), Β΅F (Ξ΅)(y)= rminΒ΅F (Ξ΅)(0), Β΅F (Ξ΅)(y) = Β΅F (Ξ΅)(y)
and
Ξ»F (Ξ΅)(x) β€ maxΞ»F (Ξ΅)(x β y), Ξ»F (Ξ΅)(y)= maxΞ»F (Ξ΅)(0), Ξ»F (Ξ΅)(y) = Ξ»F (Ξ΅)(y).
This completes the proof. Β€
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
933 Young Bae Jun et al 929-940
Young Bae Jun, Seok Zun Song and Sun Shin Ahn
Proposition 3.7. Let (F , A) be an Ξ΅-cubic soft ideal over U for a parameter Ξ΅ β A. If the
inequality x β y β€ z holds in U , then
Β΅F (Ξ΅)(x) ΒΊ rminΒ΅F (Ξ΅)(y), Β΅F (Ξ΅)(z)and Ξ»F (Ξ΅)(x) β€ maxΞ»F (Ξ΅)(y), Ξ»F (Ξ΅)(z).Proof. Assume that x β y β€ z for all x, y, z β U . Then
Β΅F (Ξ΅)(x β y) ΒΊ rminΒ΅F (Ξ΅)((x β y) β z), Β΅F (Ξ΅)(z)= rminΒ΅F (Ξ΅)(0), Β΅F (Ξ΅)(z) = Β΅F (Ξ΅)(z)
(3.8)
and
Ξ»F (Ξ΅)(x β y) β€ maxΞ»F (Ξ΅)((x β y) β z), Ξ»F (Ξ΅)(z)= maxΞ»F (Ξ΅)(0), Ξ»F (Ξ΅)(z) = Ξ»F (Ξ΅)(z),
(3.9)
which implies from (3.6) and (3.7) that
Β΅F (Ξ΅)(x) ΒΊ rminΒ΅F (Ξ΅)(x β y), Β΅F (Ξ΅)(y)ΒΊ rminΒ΅F (Ξ΅)(y), Β΅F (Ξ΅)(z)
and
Ξ»F (Ξ΅)(x) β€ maxΞ»F (Ξ΅)(x β y), Ξ»F (Ξ΅)(y)β€ maxΞ»F (Ξ΅)(y), Ξ»F (Ξ΅)(z).
This completes the proof. Β€
Theorem 3.8. In a BCK-algebra U with the condition (S), every Ξ΅-cubic soft ideal (F , A) over
U is an Ξ΅-cubic soft -subalgebra over U for all Ξ΅ β A.
Proof. Let Ξ΅ β A. Since U has the condition (S), we have (x y) β x β€ y for all x, y β U. Hence
(3.6), (3.7) and Proposition 3.6 imply that
Β΅F (Ξ΅)(x y) ΒΊ rminΒ΅F (Ξ΅)((x y) β x), Β΅F (Ξ΅)(x) ΒΊ rminΒ΅F (Ξ΅)(x), Β΅F (Ξ΅)(y)and
Ξ»F (Ξ΅)(x y) β€ maxΞ»F (Ξ΅)((x y) β x), Ξ»F (Ξ΅)(x) β€ maxΞ»F (Ξ΅)(x), Ξ»F (Ξ΅)(y)for all x, y β U. Therefore (F , A) is an Ξ΅-cubic soft -subalgebra over U for all Ξ΅ β A. Β€
Theorem 3.9. In a BCK-algebra U, if (F , A) is an Ξ΅-cubic soft ideal over U , then it is an
Ξ΅-cubic soft subalgebra over U for all Ξ΅ β A.
Proof. Let (F , A) be an Ξ΅-cubic soft ideal over U where Ξ΅ β A. For any x, y β U, we have
Β΅F (Ξ΅)(x β y) ΒΊ rminΒ΅F (Ξ΅)((x β y) β x), Β΅F (Ξ΅)(x)
= rminΒ΅F (Ξ΅)(0), Β΅F (Ξ΅)(x)
= Β΅F (Ξ΅)(x)
and Ξ»F (Ξ΅)(x β y) β€ maxΞ»F (Ξ΅)((x β y) β x), Ξ»F (Ξ΅)(x) = maxΞ»F (Ξ΅)(0), Ξ»F (Ξ΅)(x) = Ξ»F (Ξ΅)(x).
Therefore (F , A) is an Ξ΅-cubic soft subalgebra over U . Β€
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
934 Young Bae Jun et al 929-940
Cubic soft ideals in BCK/BCI-algebras
Theorem 3.9 is not true in a BCI-algebra. In fact, the Ξ΅-cubic soft ideal (F , A) in Example
3.4 is not an Ξ΅-cubic soft subalgebra over U since
Β΅F (Ξ΅) ((0, 0) β (0, 1)) = Β΅F (Ξ΅)(0,β1) = [0, 0] a = [aβ, a+]
= rminΒ΅F (Ξ΅)(0, 0), Β΅F (Ξ΅)(0, 1)
and/or
Ξ»F (Ξ΅) ((0, 0) β (0, 1)) = Ξ»F (Ξ΅)(0,β1) = t s = maxΞ»F (Ξ΅)(0, 0), Ξ»F (Ξ΅)(0, 1)
.
Definition 3.10. Let U be a BCI-algebra and Ξ΅ β A. An Ξ΅-cubic soft ideal (F , A) over U is
said to be closed if Β΅F (Ξ΅)(0 β x) ΒΊ Β΅F (Ξ΅)(x) and Ξ»F (Ξ΅)(0 β x) β€ Ξ»F (Ξ΅)(x) for all x β U.
Example 3.11. The Ξ΅2-cubic soft ideal (F , A) in Example 3.5 is closed.
Theorem 3.12. In a BCI-algebra, every closed cubic soft ideal is a cubic soft subalgebra.
Proof. Let (F , A) be a closed cubic soft ideal over U . Then Β΅F (Ξ΅)(0 βx) ΒΊ Β΅F (Ξ΅)(x) and Ξ»F (Ξ΅)(0 βx) β€ Ξ»F (Ξ΅)(x) for all x β U. It follows from (a3), (3.6) and (3.7) that
Β΅F (Ξ΅)(x β y) ΒΊ rminΒ΅F (Ξ΅)((x β y) β x), Β΅F (Ξ΅)(x)= rminΒ΅F (Ξ΅)(0 β y), Β΅F (Ξ΅)(x)ΒΊ rminΒ΅F (Ξ΅)(y), Β΅F (Ξ΅)(x)
and
Ξ»F (Ξ΅)(x β y) ΒΊ rminΞ»F (Ξ΅)((x β y) β x), Ξ»F (Ξ΅)(x)= maxΞ»F (Ξ΅)(0 β y), Ξ»F (Ξ΅)(x)β€ maxΞ»F (Ξ΅)(y), Ξ»F (Ξ΅)(x)
for all x, y β U . Therefore (F , A) is a cubic soft ideal over U . Β€
We provide a condition for a cubic soft subalgebra over U to be a (closed) cubic soft ideal over
U .
Theorem 3.13. In a p-semisimple BCI-algebra U , every cubic soft subalgebra over U is a closed
cubic soft ideal over U .
Proof. Let (F , A) be a cubic soft subalgebra over a p-semisimple BCI-algebra U and let Ξ΅ β A
be a parameter. For every x β U , we have
Β΅F (Ξ΅)(0) = Β΅F (Ξ΅)(x β x) ΒΊ rminΒ΅F (Ξ΅)(x), Β΅F (Ξ΅)(x) = Β΅F (Ξ΅)(x),
Ξ»F (Ξ΅)(0) = Ξ»F (Ξ΅)(x β x) β€ maxΞ»F (Ξ΅)(x), Ξ»F (Ξ΅)(x) = Ξ»F (Ξ΅)(x).(3.10)
Using (3.1), (3.2) and (3.10), we get
Β΅F (Ξ΅)(0 β x) ΒΊ rminΒ΅F (Ξ΅)(0), Β΅F (Ξ΅)(x) = Β΅F (Ξ΅)(x),
Ξ»F (Ξ΅)(0 β x) β€ maxΞ»F (Ξ΅)(0), Ξ»F (Ξ΅)(x) = Ξ»F (Ξ΅)(x).(3.11)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
935 Young Bae Jun et al 929-940
Young Bae Jun, Seok Zun Song and Sun Shin Ahn
For any x, y β U , we have
Β΅F (Ξ΅)(x) = Β΅F (Ξ΅)(y β (y β x)) ΒΊ rminΒ΅F (Ξ΅)(y), Β΅F (Ξ΅)(y β x)= rminΒ΅F (Ξ΅)(y), Β΅F (Ξ΅)(0 β (x β y))ΒΊ rminΒ΅F (Ξ΅)(x β y), Β΅F (Ξ΅)(y)
and
Ξ»F (Ξ΅)(x) = Ξ»F (Ξ΅)(y β (y β x)) β€ maxΞ»F (Ξ΅)(y), Ξ»F (Ξ΅)(y β x)= maxΞ»F (Ξ΅)(y), Ξ»F (Ξ΅)(0 β (x β y))β€ maxΞ»F (Ξ΅)(x β y), Ξ»F (Ξ΅)(y)
by using (a6), (3.1), (3.2), (a5) and (3.11). Therefore (F , A) is a closed cubic soft ideal over
U . Β€Corollary 3.14. If a BCI-algebra U satisfies any one of the following conditions:
β’ U = 0 β x | x β U,β’ every element of U is minimal,
β’ (βx, y β U) (x β (0 β y) = y β (0 β x)),
β’ (βx β U) (0 β x = 0 β x = 0),
β’ (βx, y β U) ((x β y) β z = x β (y β z)),
β’ (βx, y β U) (x β y = y β x),
β’ (βx β U) (0 β x = x),
β’ (βx, y, z β U) ((x β y) β (x β z) = z β y),
then every cubic soft subalgebra over U is a closed cubic soft ideal over U .
Theorem 3.15. For a cubic soft set (F , A) over a BCK-algebra U with condition (S) and a
parameter Ξ΅ β A, the following are equivalent.
(i) (F , A) is an Ξ΅-cubic soft ideal over U .
(ii) For every x, y, z β U , if x β€ yz, then Β΅F (Ξ΅)(x) ΒΊ rminΒ΅F (Ξ΅)(y), Β΅F (Ξ΅)(z) and Ξ»F (Ξ΅)(x) β€maxΞ»F (Ξ΅)(y), Ξ»F (Ξ΅)(z).
Proof. Assume that (F , A) is an Ξ΅-cubic soft ideal over U and x β€ y z for all x, y, z β U . Then
Β΅F (Ξ΅)(x) ΒΊ rminΒ΅F (Ξ΅)(x β (y z)), Β΅F (Ξ΅)(y z)= rminΒ΅F (Ξ΅)(0), Β΅F (Ξ΅)(y z)= Β΅F (Ξ΅)(y z)
ΒΊ rminΒ΅F (Ξ΅)(y), Β΅F (Ξ΅)(z)and
Ξ»F (Ξ΅)(x) β€ maxΞ»F (Ξ΅)(x β (y z)), Ξ»F (Ξ΅)(y z)= maxΞ»F (Ξ΅)(0), Ξ»F (Ξ΅)(y z)= Ξ»F (Ξ΅)(y z)
β€ maxΞ»F (Ξ΅)(y), Ξ»F (Ξ΅)(z).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
936 Young Bae Jun et al 929-940
Cubic soft ideals in BCK/BCI-algebras
Conversely suppose that (ii) is valid. Since 0 β€ x x for all x β U , it follows from (ii) that
Β΅F (Ξ΅)(0) ΒΊ rminΒ΅F (Ξ΅)(x), Β΅F (Ξ΅)(x) = Β΅F (Ξ΅)(x)
and
Ξ»F (Ξ΅)(0) β€ maxΞ»F (Ξ΅)(x), Ξ»F (Ξ΅)(x) = Ξ»F (Ξ΅)(x)
for all x β U . Since x β€ (x β y) y for all x, y β U , we have
Β΅F (Ξ΅)(x) ΒΊ rminΒ΅F (Ξ΅)(x β y), Β΅F (Ξ΅)(y)and
Ξ»F (Ξ΅)(x) β€ maxΞ»F (Ξ΅)(x β y), Ξ»F (Ξ΅)(y)for all x, y β U . Therefore (F , A) is an Ξ΅-cubic soft ideal over U . Β€
Theorem 3.16. Given a parameter Ξ΅ β A, a cubic soft set (F , A) over U is an Ξ΅-cubic soft ideal
over U if and only if the nonempty sets
Β΅βF (Ξ΅)[Ξ΄1, Ξ΄2] :=x β U | Β΅F (Ξ΅)(x) ΒΊ [Ξ΄1, Ξ΄2]
and
Ξ»βF (Ξ΅)(t) :=x β U | Ξ»F (Ξ΅)(x) β€ t
are ideals of U for all [Ξ΄1, Ξ΄2] β [I] and t β [0, 1].
Proof. Assume that a cubic soft set (F , A) over U is an Ξ΅-cubic soft ideal over U . Suppose that
Β΅βF (Ξ΅)[Ξ΄1, Ξ΄2]β©Ξ»βF (Ξ΅)(t) 6= β for all [Ξ΄1, Ξ΄2] β [I] and t β [0, 1]. Obviously, 0 β Β΅βF (Ξ΅)[Ξ΄1, Ξ΄2]β©Ξ»βF (Ξ΅)(t).
Let x and y be elements of U such that xβy β Β΅βF (Ξ΅)[Ξ΄1, Ξ΄2] and y β Β΅βF (Ξ΅)[Ξ΄1, Ξ΄2]. Then Β΅F (Ξ΅)(xβy) ΒΊ[Ξ΄1, Ξ΄2] and Β΅F (Ξ΅)(y) ΒΊ [Ξ΄1, Ξ΄2]. It follows from (3.6) that
Β΅F (Ξ΅)(x) ΒΊ rminΒ΅F (Ξ΅)(x β y), Β΅F (Ξ΅)(y) ΒΊ rmin[Ξ΄1, Ξ΄2], [Ξ΄1, Ξ΄2] = [Ξ΄1, Ξ΄2].
Hence x β Β΅βF (Ξ΅)[Ξ΄1, Ξ΄2]. Now if x β y, y β Ξ»βF (Ξ΅)(t), then Ξ»F (Ξ΅)(x β y) β€ t and Ξ»F (Ξ΅)(y) β€ t.
Using (3.7), we have Ξ»F (Ξ΅)(x) β€ maxΞ»F (Ξ΅)(x β y), Ξ»F (Ξ΅)(y) β€ t, and so x β Ξ»βF (Ξ΅)(t). Therefore
Β΅βF (Ξ΅)[Ξ΄1, Ξ΄2] and Ξ»βF (Ξ΅)(t) are ideals of U.
Conversely, suppose that Β΅βF (Ξ΅)[Ξ΄1, Ξ΄2] and Ξ»βF (Ξ΅)(t) are ideals of U for all [Ξ΄1, Ξ΄2] β [I] and
t β [0, 1]. Assume that there exists a β U such that Β΅F (Ξ΅)(0) Β΅F (Ξ΅)(a) or Ξ»F (Ξ΅)(0) > Ξ»F (Ξ΅)(a).
Let Β΅F (Ξ΅)(0) = [0β, 0+] and Β΅F (Ξ΅)(a) = [aβ, a+]. Then 0β < aβ and 0+ < a+ which imply that
0β < Ξ΄1 < aβ and 0+ < Ξ΄2 < a+, that is,
Β΅F (Ξ΅)(0) = [0β, 0+] < [Ξ΄1, Ξ΄2] < [aβ, a+]
by taking [Ξ΄1, Ξ΄2] :=[
12(0β + aβ), 1
2(0+ + a+)
]. Hence 0 /β Β΅βF (Ξ΅)[Ξ΄1, Ξ΄2]. Also 0 /β Ξ»βF (Ξ΅)(at) where
at = Ξ»F (Ξ΅)(a). This is a contradiction, and so (3.5) is valid. Assume that there exist a, b β U
such that
Β΅F (Ξ΅)(a) rminΒ΅F (Ξ΅)(a β b), Β΅F (Ξ΅)(b) (3.12)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
937 Young Bae Jun et al 929-940
Young Bae Jun, Seok Zun Song and Sun Shin Ahn
or
Ξ»F (Ξ΅)(a) > maxΞ»F (Ξ΅)(a β b), Ξ»F (Ξ΅)(b). (3.13)
For the case (3.12), let Β΅F (Ξ΅)(a) = [Ξ΄1, Ξ΄2], Β΅F (Ξ΅)(a β b) = [Ξ³1, Ξ³2] and Β΅F (Ξ΅)(b) = [Ξ³3, Ξ³4]. Then
[Ξ΄1, Ξ΄2] βΊ rmin[Ξ³1, Ξ³2], [Ξ³3, Ξ³4] = [minΞ³1, Ξ³3, minΞ³2, Ξ³4].
Hence Ξ΄1 < minΞ³1, Ξ³3 and Ξ΄2 < minΞ³2, Ξ³4. Taking
[Ο1, Ο2] = 12
(Β΅F (Ξ΅)(a) + rminΒ΅F (Ξ΅)(a β b), Β΅F (Ξ΅)(b)
)
implies that
[Ο1, Ο2] = 12([Ξ΄1, Ξ΄2] + [minΞ³1, Ξ³3, minΞ³2, Ξ³4])
=[
12(Ξ΄1 + minΞ³1, Ξ³3) , 1
2(Ξ΄2 + minΞ³2, Ξ³4)
].
It follows that
minΞ³1, Ξ³3 > Ο1 = 12(Ξ΄1 + minΞ³1, Ξ³3) > Ξ΄1,
minΞ³2, Ξ³4 > Ο2 = 12(Ξ΄2 + minΞ³2, Ξ³4) > Ξ΄2,
and so that
[minΞ³1, Ξ³3, minΞ³2, Ξ³4] Γ [Ο1, Ο2] Γ [Ξ΄1, Ξ΄2] = Β΅F (Ξ΅)(a).
Therefore a /β Β΅βF (Ξ΅)[Ο1, Ο2]. On the other hand, we know that
Β΅F (Ξ΅)(a β b) = [Ξ³1, Ξ³2] ΒΊ [minΞ³1, Ξ³3, minΞ³2, Ξ³4] Γ [Ο1, Ο2],
Β΅F (Ξ΅)(b) = [Ξ³3, Ξ³4] ΒΊ [minΞ³1, Ξ³3, minΞ³2, Ξ³4] Γ [Ο1, Ο2],
which imply that a β b, b β Β΅βF (Ξ΅)[Ο1, Ο2]. This is a contradiction, and so
Β΅F (Ξ΅)(x) ΒΊ rminΒ΅F (Ξ΅)(x β y), Β΅F (Ξ΅)(y)for all x, y β U. Now, (3.13) implies that there exists t0 β (0, 1) such that
Ξ»F (Ξ΅)(a) β₯ t0 > maxΞ»F (Ξ΅)(a β b), Ξ»F (Ξ΅)(b).Hence a β b, b β Ξ»βF (Ξ΅)(t0) but a /β Ξ»βF (Ξ΅)(t0). This is a contradiction, and therefore
Ξ»F (Ξ΅)(x) β€ maxΞ»F (Ξ΅)(x β y), Ξ»F (Ξ΅)(y)for all x, y β U. Consequently, (F , A) is an Ξ΅-cubic soft ideal over U . Β€
Definition 3.17 ([10]). The R-union of cubic soft sets (F , A) and (G , B) over U is a cubic soft
set (H , C) where C = A βͺB and
H (Ξ΅) =
F (Ξ΅) if Ξ΅ β A \B,
G (Ξ΅) if Ξ΅ β B \ A,
F (Ξ΅) βͺR G (Ξ΅) if Ξ΅ β A β©B
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
938 Young Bae Jun et al 929-940
Cubic soft ideals in BCK/BCI-algebras
for all Ξ΅ β C. This is denoted by (H , C) = (F , A) dR (G , B). Also the R-intersection of cubic
soft sets (F , A) and (G , B) over U is a cubic soft set (H , C) where C = A βͺB and
H (Ξ΅) =
F (Ξ΅) if Ξ΅ β A \B,
G (Ξ΅) if Ξ΅ β B \ A,
F (Ξ΅) β©R G (Ξ΅) if Ξ΅ β A β©B
for all Ξ΅ β C. This is denoted by (H , C) = (F , A) eR (G , B).
Theorem 3.18. If (F , A) and (G , B) are cubic soft ideals over U , then so is the R-intersection
(H , C) = (F , A) eR (G , B) of (F , A) and (G , B).
Proof. Straightforward. Β€
Theorem 3.19. Let (F , A) and (G , B) be cubic soft ideals over U . If A and B are disjoint,
then the R-union of (F , A) and (G , B) is a cubic soft ideal over U .
Proof. By means of Definition 3.17, we can write (F , A) dR (G , B) = (H , C), where C = AβͺB
and for all Ξ΅ β C,
H (Ξ΅) =
F (Ξ΅) if Ξ΅ β A \B,
G (Ξ΅) if Ξ΅ β B \ A,
F (Ξ΅) βͺR G (Ξ΅) if Ξ΅ β A β©B
Since A β© B = β , either Ξ΅ β A \ B or Ξ΅ β B \ A for all Ξ΅ β C. If Ξ΅ β A \ B, then H (Ξ΅) = F (Ξ΅)
is a cubic soft ideal over U . If Ξ΅ β B \ A, then H (Ξ΅) = G (Ξ΅) is a cubic soft ideal over U . Hence
(H , C) = (F , A) dR (G , B) is a cubic soft ideal over U . Β€
The following example shows that Theorem 3.19 is not valid if A and B are not disjoint.
Example 3.20. Let U = 0, a, b, c be a BCI-algebra with the Cayley table in Table 3.
Table 3. Cayley table of the operation β
β 0 a b c
0 0 a b c
a a 0 c b
b b c 0 a
c c b a 0
Consider sets of parameters A := Ξ΅1, Ξ΅2, Ξ΅3 and B := Ξ΅3, Ξ΅4. Then A and B are not disjoint.
Let (F , A) and (G , B) be cubic soft sets over U with the tabular representations in Table 4 and
Table 5, respectively.
Then (F , A) and (G , B) are cubic soft ideals over U , and the R-union (H , C) = (F , A)dR(G , B)
of (F , A) and (G , B) is represented by Table 6.
Note that
Β΅H (Ξ΅3)(c) = [0.4, 0.7] [0.6, 0.8] = rminΒ΅H (Ξ΅3)(c β a), Β΅H (Ξ΅3)(a)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
939 Young Bae Jun et al 929-940
Young Bae Jun, Seok Zun Song and Sun Shin Ahn
Table 4. Tabular representation of the cubic soft set (F , A)
Ξ΅1 Ξ΅2 Ξ΅3
0 γ[0.6, 0.8], 0.3γ γ[0.5, 0.9], 0.4γ γ[0.7, 0.9], 0.3γa γ[0.3, 0.7], 0.5γ γ[0.2, 0.5], 0.7γ γ[0.6, 0.8], 0.8γb γ[0.3, 0.7], 0.5γ γ[0.3, 0.6], 0.7γ γ[0.4, 0.7], 0.8γc γ[0.3, 0.7], 0.5γ γ[0.2, 0.5], 0.6γ γ[0.4, 0.7], 0.5γ
Table 5. Tabular representation of the cubic soft set (G , B)
Ξ΅3 Ξ΅4
0 γ[0.7, 1.0], 0.2γ γ[0.4, 0.8], 0.1γa γ[0.3, 0.7], 0.7γ γ[0.2, 0.6], 0.3γb γ[0.6, 0.8], 0.4γ γ[0.2, 0.6], 0.6γc γ[0.3, 0.7], 0.7γ γ[0.4, 0.8], 0.6γ
Table 6. Tabular representation of the cubic soft set (H , C)
Ξ΅1 Ξ΅2 Ξ΅3 Ξ΅4
0 γ[0.6, 0.8], 0.3γ γ[0.5, 0.9], 0.4γ γ[0.7, 1.0], 0.2γ γ[0.4, 0.8], 0.1γa γ[0.3, 0.7], 0.5γ γ[0.2, 0.5], 0.7γ γ[0.6, 0.8], 0.7γ γ[0.2, 0.6], 0.3γb γ[0.3, 0.7], 0.5γ γ[0.3, 0.6], 0.7γ γ[0.6, 0.8], 0.4γ γ[0.2, 0.6], 0.6γc γ[0.3, 0.7], 0.5γ γ[0.2, 0.5], 0.6γ γ[0.4, 0.7], 0.5γ γ[0.4, 0.8], 0.6γ
and/or Ξ»H (Ξ΅3)(a) = 0.7 0.5 = maxΞ»H (Ξ΅3)(a β b), Ξ»H (Ξ΅3)(b). Hence the R-union (H , C) =
(F , A) dR (G , B) of (F , A) and (G , B) is not a cubic soft ideal over U .
References
[1] Y. Huang, BCI-algebra, Science Press, Beijing 2006.[2] Y. B. Jun, Soft BCK/BCI-algerbas, Comput. Math. Appl. 56 (2008) 1408β1413.[3] Y. B. Jun, C. S. Kim and K. O. Yang, Cubic sets, Ann. Fuzzy Math. Inform. 4(1) (2012) 83β98.[4] Y. B. Jun, K. J. Lee, C. H. Park, Soft set theory applied to ideals in d-algebras, Comput. Math. Appl. 57
(2009) 367β378.[5] P. K. Maji, A. R. Roy, R. Biswas, An application of soft sets in a decision making problem, Comput. Math.
Appl. 44 (2002) 1077β1083.[6] P. K. Maji, R. Biswas, A. R. Roy, Soft set theory, Comput. Math. Appl. 45 (2003) 555β562.[7] J. Meng and Y. B. Jun, BCK-algebras, Kyungmoon Sa Co., Seoul, 1994.[8] D. Molodtsov, Soft set theory - First results, Comput. Math. Appl. 37 (1999) 19β31.[9] G. Muhiuddin, F. Feng and Y. B. Jun, Subalgerbas of BCK/BCI-algebras based on cubic soft sets, The
Scientific World Journal, Volume 2014, Article ID 458638, 9 pages.[10] G. Muhiuddin and Abdullah M. Al-roqi, Cubic soft sets with applications in BCK/BCI-algebras, Ann.
Fuzzy Math. Inform. 8(2) (2014) 291β304.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
940 Young Bae Jun et al 929-940
Hyers-Ulam stability of the delayed homogeneous
matrix difference equation with constructive
method
Soon-Mo Jung1 and Young Woo Nam2
1,2Mathematics Section, College of Science and Technology, Hongik University,30016 Sejong, Republic of Korea
1E-mail: [email protected]: [email protected]
Abstract
We prove Hyers-Ulam stability of the first order delayed homogeneous matrix differ-ence equation xi+p = A(i)xi for all integers i.
1 Introduction
Throughout this paper, we denote by C, N, N0, and Z the set of all complex numbers, of allpositive integers, of all nonnegative integers, and the set of all integers, respectively. Given afixed positive integer n, let (Cn, β₯ Β· β₯n) be a complex normed space, each of whose elements isa column vector, and let CnΓn be a vector space consisting of all (nΓ n) complex matrices.We choose a norm β₯ Β· β₯nΓn on CnΓn which is compatible with β₯ Β· β₯n, i.e., both norms obey
β₯ABβ₯nΓn β€ β₯Aβ₯nΓnβ₯Bβ₯nΓn and β₯Axβ₯n β€ β₯Aβ₯nΓnβ₯xβ₯n (1.1)
for all A,B β CnΓn and x β Cn.A matrix difference equation is a difference equation with matrix coefficients in which the
value of vector at one point depends on the values of preceding points.In this paper, we prove Hyers-Ulam stability of the first order delayed homogeneous matrix
difference equation
xi+p = A(i)xi (1.2)
for all integers i β Z, where each transition matrix A(i) is nonsingular and p is a fixed integerlarger than 1. More precisely, we prove that if a vector sequence yiiβZ of Cn satisfies theinequality
β₯yi+p βA(i)yiβ₯n β€ Ξ΅
for all i β Z, then there exists a solution xiiβZ to the delayed matrix difference equation(1.2) such that the bound for β₯yi β xiβ₯n depends on Ξ΅ and the transition matrices A(i) only.We refer the reader to [1, 2, 3, 4, 6] for the exact definition of Hyers-Ulam stability.
0Key words and phrases: difference equation; matrix difference equation; delayed matrix difference equa-tion; Hyers-Ulam stability; approximation.
02010 Mathematics Subject Classification: Primary 39A45, 39B82; Secondary 39A06, 39B42.
1
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
941 Soon-Mo Jung et al 941-948
2 Hyers-Ulam stability of matrix difference equation
2 Preliminaries
Throughout this paper, the transition matrix A(i) of CnΓn is defined by
A(i) =
a11(i) a12(i) Β· Β· Β· a1n(i)a21(i) a22(i) Β· Β· Β· a2n(i)
......
. . ....
an1(i) an2(i) Β· Β· Β· ann(i)
for any integer i. We moreover assume that every A(i) is nonsingular. We will use thefollowing abbreviation.
Ξ¦(j, k) :=
jβ1βi=k
A(i) = A(j β 1)A(j β 2) Β· Β· Β·A(k) (for j > k),
InΓn (for j = k),
(2.1)
where we set Ξ¦(j, k) :=(Ξ¦(k, j)
)β1= A(j)β1A(j + 1)β1 Β· Β· Β·A(k β 1)β1 for j < k and InΓn
denotes the (nΓn) identity matrix. Sometimes, we use Ξ¦(j) and Ξ¦β1(k, j) instead of Ξ¦(j, 0)
and(Ξ¦(k, j)
)β1, respectively.
In the following lemma, we introduce some properties of Ξ¦(j, k) without proof.
Lemma 2.1 Assume that n is a fixed positive integer. If the transition matrix A(i) of CnΓn
is nonsingular for any integer i, then it holds that
(i) Ξ¦(j + 1, k) = A(j)Ξ¦(j, k);
(ii) Ξ¦β1(j, k + 1) = A(k)Ξ¦β1(j, k);
(iii) A(k β 1)β1Ξ¦β1(j, k) = Ξ¦β1(j, k β 1)
for all integers j and k.
3 Hyers-Ulam stability of xi+p = A(i)xi
We now prove our main theorem concerning Hyers-Ulam stability of the delayed homoge-neous matrix difference equation (1.2). Obviously, our theorem is a generalization and animprovement of [5, Theorem 2.1].
Theorem 3.1 Assume that n > 0 and p > 1 are fixed integers and Ξ΅ is a nonnegative realnumber. For all integers i, assume that A(i) is a nonsingular (nΓ n) complex-valued matrixfor which there exists a constant K > 0 such that
ββj=0
β₯β₯β₯β₯β₯β₯(
jβk=0
A(i+ kp)
)β1β₯β₯β₯β₯β₯β₯nΓn
β€ K (3.1)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
942 Soon-Mo Jung et al 941-948
Soon-Mo Jung and Young Woo Nam 3
for all integers i. If a sequence yiiβZ of Cn satisfies the inequalityβ₯β₯yi+p βA(i)yiβ₯β₯nβ€ Ξ΅ (3.2)
for all integers i, then there exists a unique solution TiiβZ to the first order delayed homo-geneous matrix difference equation (1.2) such thatβ₯β₯Ti β yiβ₯β₯n β€ KΞ΅ (3.3)
for each integer i.
Proof. In view of (3.2), there exists a sequence Ξ΅iiβZ of Cn such that
yi+p βA(i)yi = Ξ΅i (3.4)
for all integers i and
supiβZ
β₯β₯Ξ΅iβ₯β₯n β€ Ξ΅. (3.5)
First, we use the induction on m to prove
yi+mp =
(mβ1βk=0
A(i+ kp)
)yi +
mβ1βj=0
mβ1βk=j+1
A(i+ kp)
Ξ΅i+jp (3.6)
for all i β Z and m β N0. Obviously, the equality (3.6) is true for m β 0, 1. Assume thatthe equality (3.6) is true for some positive integer m. It then follows from (3.4) and (3.6)that
yi+(m+1)p = A(i+mp)yi+mp + Ξ΅i+mp
=
(mβk=0
A(i+ kp)
)yi +
mβ1βj=0
mβk=j+1
A(i+ kp)
Ξ΅i+jp + Ξ΅i+mp
=
(mβk=0
A(i+ kp)
)yi +
mβj=0
mβk=j+1
A(i+ kp)
Ξ΅i+jp,
which follows from (3.6) by replacing m with m+ 1.
If we set
Ti(m) :=
(mβk=0
A(i+ kp)
)β1
yi+(m+1)p
for all i β Z and m β N0, then it follows from (3.6) that
Ti(m) = yi +
mβj=0
(jβ
k=0
A(i+ kp)
)β1
Ξ΅i+jp. (3.7)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
943 Soon-Mo Jung et al 941-948
4 Hyers-Ulam stability of matrix difference equation
Let m and n be nonnegative integers with n > m. Then, by (3.7), we have
Ti(n)β Ti(m) =
nβj=m+1
(jβ
k=0
A(i+ kp)
)β1
Ξ΅i+jp
for any fixed integer i. In view of (3.5) and (3.1), we further get
β₯β₯Ti(n)β Ti(m)β₯β₯n
=
β₯β₯β₯β₯β₯β₯nβ
j=m+1
(jβ
k=0
A(i+ kp)
)β1
Ξ΅i+jp
β₯β₯β₯β₯β₯β₯n
β€nβ
j=m+1
β₯β₯β₯β₯β₯β₯(
jβk=0
A(i+ kp)
)β1
Ξ΅i+jp
β₯β₯β₯β₯β₯β₯n
β€nβ
j=m+1
β₯β₯β₯β₯β₯β₯(
jβk=0
A(i+ kp)
)β1β₯β₯β₯β₯β₯β₯nΓn
β₯β₯Ξ΅i+jp
β₯β₯n
β€ Ξ΅nβ
j=m+1
β₯β₯β₯β₯β₯β₯(
jβk=0
A(i+ kp)
)β1β₯β₯β₯β₯β₯β₯nΓn
β 0, as mββ,
for every i β Z. Hence, Ti(m)mβN0 is a Cauchy sequence for each fixed i β Z, and we candefine
Ti := limmββ
Ti(m) (3.8)
for each i β Z.By (3.4), (3.7), and (3.8), we obtain
Ti+p βA(i)Ti = yi+p +
ββj=0
(jβ
k=0
A(i+ (k + 1)p
))β1
Ξ΅i+(j+1)p
βA(i)yi βββj=0
(jβ
k=1
A(i+ kp)
)β1
Ξ΅i+jp
= yi+p +
ββj=0
(j+1βk=1
A(i+ kp)
)β1
Ξ΅i+(j+1)p
βA(i)yi βββj=0
(jβ
k=1
A(i+ kp)
)β1
Ξ΅i+jp
= yi+p +ββj=1
(jβ
k=1
A(i+ kp)
)β1
Ξ΅i+jp
βA(i)yi βββj=0
(jβ
k=1
A(i+ kp)
)β1
Ξ΅i+jp
= 0
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
944 Soon-Mo Jung et al 941-948
Soon-Mo Jung and Young Woo Nam 5
for all i β Z. Moreover, it follows from (3.5), (3.1), (3.7), and (3.8) that
β₯β₯Ti β yiβ₯β₯n =
β₯β₯β₯β₯β₯β₯ββj=0
(jβ
k=0
A(i+ kp)
)β1
Ξ΅i+jp
β₯β₯β₯β₯β₯β₯n
β€ββj=0
β₯β₯β₯β₯β₯β₯(
jβk=0
A(i+ kp)
)β1
Ξ΅i+jp
β₯β₯β₯β₯β₯β₯n
β€ββj=0
β₯β₯β₯β₯β₯β₯(
jβk=0
A(i+ kp)
)β1β₯β₯β₯β₯β₯β₯nΓn
β₯β₯Ξ΅i+jp
β₯β₯n
β€ KΞ΅
for all i β Z.Finally, we prove the uniqueness of the sequence TiiβZ. Assume that UiiβZ is another
solution to the difference equation (1.2). By applying the induction on m, we prove that
Ui =
(mβk=0
A(i+ kp)
)β1
Ui+(m+1)p (3.9)
for any m β N0. Obviously, (3.9) is true for m = 0. Assume now that (3.9) is true for someinteger m β₯ 0. It then follows from (1.2) and (3.9) that
Ui =
(mβk=0
A(i+ kp)
)β1
Ui+(m+1)p
=
(m+1βk=0
A(i+ kp)
)β1
A(i+ (m+ 1)p
)Ui+(m+1)p
=
(m+1βk=0
A(i+ kp)
)β1
Ui+(m+2)p,
which can be obtained from (3.9) by replacing m with m+1. Thus, by (3.1), (3.3), and (3.9),we have β₯β₯Ti β Ui
β₯β₯n
=
β₯β₯β₯β₯β₯β₯(
mβk=0
A(i+ kp)
)β1 (Ti+(m+1)p β Ui+(m+1)p
)β₯β₯β₯β₯β₯β₯n
β€
β₯β₯β₯β₯β₯β₯(
mβk=0
A(i+ kp)
)β1β₯β₯β₯β₯β₯β₯nΓn
β₯β₯Ti+(m+1)p β yi+(m+1)p
β₯β₯n
+
β₯β₯β₯β₯β₯β₯(
mβk=0
A(i+ kp)
)β1β₯β₯β₯β₯β₯β₯nΓn
β₯β₯yi+(m+1)p β Ui+(m+1)p
β₯β₯n
β€ 2KΞ΅
β₯β₯β₯β₯β₯β₯(
mβk=0
A(i+ kp)
)β1β₯β₯β₯β₯β₯β₯nΓn
β 0, as mββ,
for all i β Z, which implies the uniqueness of TiiβZ.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
945 Soon-Mo Jung et al 941-948
6 Hyers-Ulam stability of matrix difference equation
4 Examples
At a glance, the condition (3.1) would seem too strong so that we could seldom find practicalexamples. But we get rid of such a misunderstanding through introducing a few examplesfor the sequence A(i)iβZ of transition matrices which satisfy the condition (3.1).
Example 4.1 Let us set n = 1 and p = 3. If A(i) = 23 is a (1Γ 1) matrix for every integeri, then we have
ββj=0
β£β£β£β£β£β£(
jβk=0
A(i+ 3k)
)β1β£β£β£β£β£β£ =
ββj=0
(23 Β· 23 Β· Β· Β· 23οΈΈ οΈ·οΈ· οΈΈ
j+1
)β1=
ββj=0
2β3(j+1) =1
7,
i.e., the condition (3.1) is satisfied with K = 17 .
Assume that a sequence yiiβZ of complex numbers satisfies the inequalityβ£β£yi+3 β 23yiβ£β£ β€ Ξ΅
for all integers i, where Ξ΅ is an arbitrarily given nonnegative real number. Then, accordingto Theorem 3.1, there exists a unique sequence xiiβZ of complex numbers such that
xi+3 = 23xi (4.1)
and
|xi β yi| β€1
7Ξ΅
for all integers i.Indeed, the delayed difference equation (4.1) is strongly related to the nonlinear difference
equation
xi+1 = 2i+1 β 22i+1
xi.
Example 4.2 We consider the difference equation with two variables given as
(ui+1
vi+1
)=
β22i+1
(i2 + 2i+ 2)(i2 + 1)vi β (i2 + 2i+ 2)2i+1
2i+2
i2 + 2i+ 2β 22i+1
(i2 + 2i+ 2)(i2 + 1)ui
for all integers i, where uiiβZ and viiβZ are sequences of complex numbers. By a straight-forward calculation, we show that
(ui+2
vi+2
)=
4(i2 + 1)
i2 + 4i+ 50
04(i2 + 1)
i2 + 4i+ 5
(ui
vi
)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
946 Soon-Mo Jung et al 941-948
Soon-Mo Jung and Young Woo Nam 7
for all integers i. We now define the (2Γ 2) matrix A(i) by
A(i) :=
4(i2 + 1)
i2 + 4i+ 50
04(i2 + 1)
i2 + 4i+ 5
=4(i2 + 1)
(i+ 2)2 + 1I2Γ2
for every integer i, where I2Γ2 denotes the (2Γ 2) identity matrix. Then we have
(jβ
k=0
A(i+ 2k)
)β1
=(i+ 2j + 2)2 + 1
(i2 + 1)4j+1I2Γ2
for all nonnegative integers j. Hence, we see that
ββj=0
β₯β₯β₯β₯β₯β₯(
jβk=0
A(i+ 2k)
)β1β₯β₯β₯β₯β₯β₯β
=
ββj=0
(i+ 2j + 2)2 + 1
(i2 + 1)4j+1
β€ββj=0
(|i|+ 2j + 2)2 + 1
(i2 + 1)4j+1
β€ββj=0
1
4j+1+
ββj=0
1
2
j + 1
4j+
ββj=0
(j + 1)2
4j
β€ 1
3+
ββj=0
1
2
1
2j+
ββj=0
9
4
1
2j
=35
6,
i.e., the condition (3.1) is satisfied with K = 356 .
Let Ξ΅ is an arbitrarily given nonnegative real number. Assume that a sequence yiiβZ ofC2 satisfies the inequality β₯β₯yi+2 βA(i)yi
β₯β₯β β€ Ξ΅
for all integers i. Then, due to Theorem 3.1 with n = 2, p = 2, and K = 356 , there exists
a unique solution xiiβZ to the delayed homogeneous matrix difference equation (1.2) suchthat β₯β₯xi β yiβ₯β₯β β€ 35
6Ξ΅
for any integer i.
Acknowledgment. The first author was supported by Basic Science Research Programthrough the National Research Foundation of Korea (NRF) funded by the Ministry of Edu-cation (No. 2013R1A1A2005557).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
947 Soon-Mo Jung et al 941-948
8 Hyers-Ulam stability of matrix difference equation
References
[1] S. Czerwik, Functional Equations and Inequalities in Several Variables, World Sci. Publ., Singapore, 2002.
[2] D. H. Hyers, On the stability of the linear functional equation, Proc. Natl. Acad. Sci. USA 27 (1941),222β224.
[3] D. H. Hyers, G. Isac and Th. M. Rassias, Stability of Functional Equations in Several Variables,Birkhauser, Boston, 1998.
[4] S.-M. Jung, Hyers-Ulam-Rassias Stability of Functional Equations in Nonlinear Analysis, Springer Opti-mization and Its Applications Vol. 48, Springer, New York, 2011.
[5] S.-M. Jung, Hyers-Ulam stability of the first-order matrix difference equations, Adv. Difference Equ. 2015(2015), no. 170, 13 pages.
[6] S. M. Ulam, A Collection of Mathematical Problems, Interscience Publishers, New York, 1960. Reprintedas: Problems in Modern Mathematics, John Wiley & Sons, Inc., New York, 1964.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
948 Soon-Mo Jung et al 941-948
Mathematical analysis of (n + 3)-dimensional virus dynamics model
A. M. Elaiw and N. H. AlShamrani
Department of Mathematics, Faculty of Science, King Abdulaziz University,
P.O. Box 80203, Jeddah 21589, Saudi Arabia.
Emails: a m [email protected] (A. Elaiw), [email protected]. (N. AlShamrani).
Abstract
An (n + 3)-dimensional nonlinear mathematical model for the virus dynamics with humoral immunity
and n-stages of infected cells is proposed and analyzed. Two threshold parameters, the basic reproduction
number, RM0 and the humoral immunity number, RM
1 are derived. Utilizing Lyapunov functions and LaSalleβs
invariance principle, the global asymptotic stability of all steady states of the model is obtained. An example
is presented and some numerical simulations are conducted in order to illustrate the dynamical behavior.
Keywords: Virus dynamics; global stability; humoral immunity; Lyapunov function.
1 Introduction
During the past decades many human viruses have been found such as HIV, HBV, HCV and HTLV-I. To un-
derstand the virus dynamics, several mathematical models for virus dynamics have been proposed and analyzed
(see e.g. [1]-[16]). One of the most important features of mathematical models is the global stability of steady
states which gives us a detailed information and enhances our understanding about the virus dynamics. There-
fore several researchers studied the global stability of virus dynamics models (see e.g. [5], [6], [7], [8], [9], [11],
[12], [13], [14], [19], [20]). Some of these papers consider a single-infected stage for infected cells (see e.g. [5],
[6], [7], [11], [12] and [14]). Other works consider double-infected stages for infected cells, the first stage is the
latently infected cells which contain viruses but do not produce it and the second stage is the actively infected
cells which produce new viruses (see e.g. [8] [9], [19] and [20]). As reported in [21], [22] and [23], due to ongoing
viral replication in the virus dynamics process such as HIV, the time from the contact of viruses and uninfected
target cells to the death of the cells modeled by dividing the process into n short stages y1 β y2 β .... β yn.
Georgescu and Hsieh [20] have proposed a virus dynamics model with multi-staged infected cells. However, the
model does not consider the immune response.
It should be pointed out that the immune response plays an important role in controlling the disease
progression. There are two main responses for immune system, Cytotoxic T Lymphocyte (CTL) immune
response and humoral immune response. The function of the CTL cells is to kill the infected cells. The humoral
immunity is based on the B cells which produce antibodies to attack the viruses [1]. It is mentioned in [24]
that, in malaria, the antibodies are more effective than CTL cells [24]. Several works incorporate the humoral
immune response into the virus dynamics models (see e.g. [25]-[31]). Elaiw and AlShamrani [29], [30] studied
the global stability of virus dynamics models with double-infected stages for infected cells.
The aim of this paper is to study a general virus dynamics model with multi-staged infected cells and
humoral immunity. Our model is an improvement of the model presented in [20] by taking into account the
humoral immune response, and by assuming a more general incidence rate which includes the form given in
[20]. We use Lyapunov functions and LaSalleβs invariance principle to prove the global stability of all the steady
states of the model. We show that there exist two bifurcation parameters, the basic reproduction number RM0
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
949 A. M. Elaiw et al 949-967
and the humoral immunity number RM1 . We establish a set of sufficient conditions which guarantee the global
stability of all steady states of the model.
2 The model
In this section we propose the following model:
x = Ξ»β dxβ g(x, v), (1)
y1 = g(x, v)β a1Ο1(y1), (2)
yi = aiβ1Οiβ1(yiβ1)β aiΟi(yi), i = 2, 3, ..., n, (3)
v = anΟn(yn)β pzv β uv, (4)
z = rzv β bz. (5)
All parameters and variables have the same identifications given in Section 1. The model is a generalization
of several existing model by considering general functions for: (i) the incidence rate of infection g(x, v); (ii)
the production rates of infected cells g(x, v) and aiβ1Οiβ1(yiβ1), i = 2, ..., n; (iii) the removal rate of infected
cells aiΟi(yi), i = 1, ..., n; (iii) the production rate of viruses anΟn(yn). Functions g and Οi are continuously
differentiable and satisfy the following conditions:
Condition C1. (i) g(x, v) > 0, g(0, v) = g(x, 0) = 0 for all x, v > 0 and
(ii) βg(x,v)βx > 0, βg(x,v)βv > 0, βg(x,0)
βv > 0 for all x, v > 0.
Condition C2. (i) g(x, v) β€ v βg(x,0)βv for all x, v > 0 and
(ii)
(βg(x, 0)
βv
)β²> 0 for all x, v > 0.
Condition C3. (i) Οi(yi) > 0 for all yi > 0, Οi(0) = 0, i = 1, 2, ..., n,
(ii) Οβ²i(yi) > 0 for all yi > 0, i = 1, 2, ..., n, and
(iii) there is Ξ±i > 0, i = 1, ..., n such that Οi(yi) β₯ Ξ±iyi for all yi > 0.
3 Properties of solutions
In this section, we study some properties of the solutions of the model such as the non-negativity and bound-
edness.
Proposition 1. Suppose that Conditions C1 and C3 are hold. Then there exist positive numbers Mj ,
j = 1, 2, ..., n+ 2, such that the compact set
Ξ =
(x, y1, ..., yn, v, z) β Rn+3β₯0 : 0 β€ x β€M1, 0 β€ yi β€Mi, 0 β€ v β€Mn+1, 0 β€ z β€Mn+2, i = 1, ..., n
is positively invariant.
Proof. Since
x |x=0= Ξ» > 0,
y1 |y1=0= g(x, v) β₯ 0 for all x, v β [0,β),
yi |yi=0= aiβ1Οiβ1(yiβ1) β₯ 0 for all yiβ1 β [0,β), i = 2, 3, ..., n,
v |v=0= anΟn(yn) β₯ 0 for all yn β [0,β),
z |z=0= 0,
Then, the orthant Rn+3β₯0 is positively invariant for system (1)-(5).
To show the boundedness of the solutions we let G1(t) = x(t) + y1(t), then
G1 = Ξ»β dxβ a1Ο1(y1) β€ Ξ»β dxβ a1Ξ±1y1 β€ Ξ»β Ξ΄1 (x+ y1) β€ Ξ»β Ξ΄1G1,
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
950 A. M. Elaiw et al 949-967
where Ξ΄1 = mind, a1Ξ±1. It follows that,
G1(t) β€ eβΞ΄1t(G1(0)β Ξ»
Ξ΄1
)+Ξ»
Ξ΄1.
Hence, 0 β€ G1(t) β€ M1 if G1(0) β€ M1 for t β₯ 0 where M1 = λδ1. The non-negativity of x and y1 implies that,
0 β€ x(t), y1(t) β€M1 if x(0) + y1(0) β€M1. From Eq. (3) and Condition C3, we have
y2 = a1Ο1(y1)β a2Ο2(y2) β€ a1Ο1(M1)β a2Ξ±2y2.
It follows that, 0 β€ y2(t) β€M2 if y2(0) β€M2, where M2 =a1Ο1(M1)
a2Ξ±2. Similarly, we can show 0 β€ yi(t) β€Mi if
yi(0) β€Mi, where Mi =aiβ1Οiβ1(Miβ1)
aiΞ±ii = 3, ..., n. Finally, we let G2(t) = v(t) + p
r z(t), then
G2 = anΟn(yn)β uv β pb
rz
β€ anΟn(Mn)β Ξ΄2(v +
p
rz)
= anΟn(Mn)β Ξ΄2G2,
where Ξ΄2 = minu, b. It follows that, 0 β€ G2(t) β€ Mn+1 if G2(0) β€ Mn+1, where Mn+1 =anΟn(Mn)
Ξ΄2. Since
v(t) and z(t) are non-negative, then 0 β€ v(t) β€ Mn+1 and 0 β€ z(t) β€ Mn+2 if v(0) + pr z(0) β€ Mn+1, where
Mn+2 = rpMn+1. Therefore, all the variables of the model are bounded and the region Ξ is positively invariant
with respect to model (1)-(5).
4 The steady states and biological bifurcations
In this section, we prove the existence of the steady states of system (1)-(5) and derive two bifurcation param-
eters.
Lemma 1. Assume that Conditions C1-C3 are satisfied, then there exist two bifurcation parameters RM0 >
RM1 > 0 such that
(i) if RM0 β€ 1, then the system has only one positive steady state Q0 β Ξ.
(ii) if RM1 β€ 1 < RM0 , then then the system has only two positive steady states Q0 β Ξ and Q1 β Ξ, and
(iii) if RM1 > 1, then then the system has three positive steady states Q0 β Ξ, Q1 β Ξ and Q2 βΞ.
Proof. At any steady state E(x, y1, ..., yn, v, z), the following equations hold:
Ξ»β dxβ g(x, v) = 0, (6)
g(x, v)β a1Ο1(y1) = 0, (7)
aiβ1Οiβ1(yiβ1)β aiΟi(yi) = 0, i = 2, ..., n, (8)
anΟn(yn)β uv β pzv = 0, (9)
(rv β b) z = 0. (10)
Eq. (10) has two possibilities, z = 0 and v =b
r. When z = 0, then from Eqs. (6)-(9) we get
Ξ»β dx = g(x, v) =
iβj=1
ajaj
aiΟi(yi) =
nβj=1
ajaj
uv, i = 1, ..., n, (11)
The continuity and strictly increasing properties of Οi imply that Οβ1i exists and it is also continuous and strictly
increasing [32]. Define fi(v) = Οβ1i
((iβ
j=1
ajaj
)(nβj=1
ajaj
)uvai
), i = 1, 2, ..., n, then fi(0) = 0 and fi(v) > 0 for all
v > 0. From Eq. (11), we get
yi = fi(v), x = x0 β1
d
nβj=1
ajaj
uv, (12)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
951 A. M. Elaiw et al 949-967
and
g
x0 β 1
d
nβj=1
ajaj
uv, v
β nβj=1
ajaj
uv = 0, (13)
where x0 = Ξ»/d. Condition C1 implies that Eq. (13) has two possible solutions v = 0 and v 6= 0. If v = 0, then
from Eq. (12), we get the disease-free steady state Q0 = (x0,
n+2οΈ· οΈΈοΈΈ οΈ·0, ..., 0, 0). Let us consider the case v 6= 0. Define
Ξ¨1 (v) = g
x0 β 1
d
nβj=1
ajaj
uv, v
β nβj=1
ajaj
uv = 0.
We have, Ξ¨1(0) = 0, and Ξ¨1(v) = βΞ» < 0.where v = Ξ»u
(nβj=1
ajaj
). Moreover,
Ξ¨β²1 (0) = βud
nβj=1
ajaj
βg(x0, 0)
βx+βg(x0, 0)
βvβ
nβj=1
ajaj
u.
From Condition C1 we have βg(x0,0)βx = 0, then
Ξ¨β²1 (0) = u
nβj=1
ajaj
1
u
nβj=1
ajaj
βg(x0, 0)
βvβ 1
.
Therefore, if 1u
(nβj=1
ajaj
)βg(x0,0)βv > 1, then Ξ¨β²1 (0) > 0 and there exists a v1 β (0, v) such that Ξ¨1(v1) = 0.
Substituting v = v1 in Eq. (6) and letting
Ξ¨2(x) = Ξ»β dxβ g(x, v1) = 0.
According to Condition C1, Ξ¨2 is a strictly decreasing, Ξ¨2(0) = Ξ» > 0 and Ξ¨2(x0) = βg(x0, v1) < 0. Thus,
there exists a unique x1 β (0, x0) such that Ξ¨2(x1) = 0. On the other hand, from Eq. (12) we have yi,1 =
fi(v1) > 0, i = 1, ..., n. It follows that, a endemic steady state without humoral immune response Q1 =
(x1, y1,1, ..., yn,1, v1, 0) exists when 1u
(nβj=1
ajaj
)βg(x0,0)βv > 1. Let us define the basic reproduction number as:
RM0 =1
u
nβj=1
ajaj
βg(x0, 0)
βv.
The other possibility of Eq. (10) is v = v2 =b
r. Let
Ξ¨3(x) = Ξ»β dxβ g(x, v2) = 0.
Clearly, Ξ¨3 is a strictly decreasing, Ξ¨3(0) = Ξ» > 0 and Ξ¨3(x0) = βg(x0, v2) < 0. Thus, there exists a unique
x2 β (0, x0) such that Ξ¨3(x2) = 0. It follows that,
yi,2 = Οβ1i
iβj=1
ajaj
g(x2, v2)
ai
> 0.
Further, z2 =u
p(RM1 β 1), where
RM1 =1
u
nβj=1
ajaj
g(x2, v2)
v2
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
952 A. M. Elaiw et al 949-967
represents the humoral immunity number. It follows that, if RM1 > 1, then there exists a endemic steady state
with humoral immune response Q2 = (x2, y1,2, ..., yn,2, v2, z2).
Now we show that Q0, Q1 β Ξ and Q2 βΞ. Clearly, Q0 β Ξ. We have x1 β (0, x0), then
0 < x1 <Ξ»
dβ€ Ξ»
Ξ΄1= M1.
From Eq. (11), we get
a1Ξ±1y1,1 β€ a1Ο1(y1,1) = Ξ»β dx1 < Ξ»β 0 < y1,1 <Ξ»
a1Ξ±1β€M1.
Also, from Eq. (8), we have
a2Ξ±2y2,1 β€ a2Ο2(y2,1) = a1Ο1(y1,1) < a1Ο1(M1)β 0 < y2,1 <a1Ο1(M1)
a2Ξ±2= M2.
Consequently, for i = 3, ..., n, we have
aiΞ±iyi,1 β€ aiΟi(yi,1) = aiβ1Οiβ1(yiβ1,1) < aiβ1Οiβ1(Miβ1)β 0 < yi,1 <aiβ1Οiβ1(Miβ1)
aiΞ±i= Mi.
Eq. (9) implies that,
uv1 = anΟn(yn,1) < anΟn(Mn)β 0 < v1 <anΟn(Mn)
uβ€ anΟn(Mn)
Ξ΄2= Mn+1.
We have also z1 = 0, then Q1 β Ξ. Similarly, one can show that 0 < x2 < M1 and 0 < yi,2 < Mi, i = 1, ..., n.
Now we show that if RM1 > 1, then 0 < v2 < Mn+1 and 0 < z2 < Mn+2. From Eq. (9) we have
uv2 + pv2z2 = anΟn(yn,2).
Then
uv2 < anΟn(yn,2) < anΟn(Mn)β 0 < v2 <anΟn(Mn)
uβ€Mn+1,
pv2z2 < anΟn(yn,2) < anΟn(Mn)β 0 < z2 <ranΟn(Mn)
pbβ€Mn+2.
Then, Q2 βΞ. Clearly from Condition C2, we have
RM1 =1
u
nβj=1
ajaj
g(x2, v2)
v2β€ 1
u
nβj=1
ajaj
βg(x2, 0)
βv<
1
u
nβj=1
ajaj
βg(x0, 0)
βv= RM0 .
5 Global stability analysis
In this section, we study the global stability of system (1)-(5) by constructing suitable Lyapunov functionals.
The stability of the disease-free steady state Q0 will be given in the following result.
Theorem 1. Let Conditions C1-C3 hold true and RM0 β€ 1, then Q0 is globally asymptotically stable (GAS)
in Ξ.
Proof. Define
V0(x, y1, ..., yn, v, z) = xβ x0 ββ« x
x0
limvβ0+
g(x0, v)
g(Ξ·, v)dΞ· +
nβi=1
iβ1βj=1
ajaj
yi +
nβj=1
ajaj
v +p
r
nβj=1
ajaj
z, (14)
where0βj=1
ajaj
= 1. It is seen that, V0(x, y1, ..., yn, v, z) > 0 for all x, y1, ..., yn, v, z > 0, while V0(x0,
n+2οΈ· οΈΈοΈΈ οΈ·0, ..., 0, 0) = 0.
We calculate dV0
dt along the solutions of model (1)-(5) as:
dV0dt
=
(1β lim
vβ0+
g(x0, v)
g(x, v)
)x+
nβi=1
iβ1βj=1
ajaj
yi +
nβj=1
ajaj
v +p
r
nβj=1
ajaj
z. (15)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
953 A. M. Elaiw et al 949-967
We have
nβi=1
iβ1βj=1
ajaj
yi = g(x, v)β a1Ο1(y1) +nβi=2
iβ1βj=1
ajaj
(aiβ1Οiβ1(yiβ1)β aiΟi(yi))
= g(x, v)β
nβj=1
ajaj
anΟn(yn).
Then
dV0dt
= dx0
(1β lim
vβ0+
g(x0, v)
g(x, v)
)(1β x
x0
)+ g(x, v) lim
vβ0+
g(x0, v)
g(x, v)β u
nβj=1
ajaj
v β pb
r
nβj=1
ajaj
z
= dx0
(1β βg(x0, 0)/βv
βg(x, 0)/βv
)(1β x
x0
)+ u
nβj=1
ajaj
1
u
nβj=1
ajaj
g(x, v)
v
βg(x0, 0)/βv
βg(x, 0)/βvβ 1
v
β pb
r
nβj=1
ajaj
z. (16)
From (i) of Condition C2, we have
dV0dtβ€ dx0
(1β βg(x0, 0)/βv
βg(x, 0)/βv
)(1β x
x0
)+ u
nβj=1
ajaj
(RM0 β 1)v β pb
r
nβj=1
ajaj
z. (17)
From (ii) of Condition C1, we get (1β βg(x0, 0)/βv
βg(x, 0)/βv
)(1β x
x0
)β€ 0,
where the equality occurs at x = x0. Therefore, if RM0 β€ 1, then dV0
dt β€ 0 for all x, v, z > 0. One can easily
show that dV0
dt = 0 occurs at Q0. Using LaSalleβs invariance principle, we derive that Q0 is GAS.
To prove the global stability of the two steady states Q1 and Q2, we need the following condition on the
incidence rate function.
Condition C4. (1β g(x, vi)
g(x, v)
)(g(x, v)
g(x, vi)β v
vi
)β€ 0, x, v > 0, i = 1, 2
Lemma 2. Suppose that Conditions C1-C4 are satisfied and RM0 > 1. Then x1, x2, v1, v2 exist satisfying
sgn(x2 β x1) = sgn(v1 β v2) = sgn(RM1 β 1).
Proof. From Condition C1, for x1, x2, v1, v2 > 0, we have
(g(x2, v2)β g(x1, v2))(x2 β x1) > 0, (18)
(g(x1, v2)β g(x1, v1)) (v2 β v1) > 0. (19)
Using Condition C4 with i = 1, x = x1 and v = v2 we get
(g(x1, v2)v1 β g(x1, v1)v2) (g(x1, v2)β g(x1, v1)) < 0. (20)
It follows from inequality (19) that
(g(x1, v2)v1 β g(x1, v1)v2)(v1 β v2) > 0. (21)
First, we claim sgn(x2 β x1) = sgn(v1 β v2). Suppose this is not true, i.e., sgn(x2 β x1) = sgn(v2 β v1). Using
the conditions of the steady states Q1 and Q2 we have
(Ξ»β dx2)β (Ξ»β dx1) = g(x2, v2)β g(x1, v1)
= (g(x2, v2)β g(x1, v2)) + (g(x1, v2)β g(x1, v1)).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
954 A. M. Elaiw et al 949-967
Therefore, from inequalities (18) and (19) we get:
sgn (x1 β x2) = sgn (x2 β x1) ,
which leads to a contradiction. Thus, sgn (x2 β x1) = sgn (v1 β v2) . Using the steady state conditions for Q1
we have 1u
(nβj=1
ajaj
)g(x1, v1)
v1= 1, then
RM1 β 1 =1
u
nβj=1
ajaj
g(x2, v2)
v2β 1
u
nβj=1
ajaj
g(x1, v1)
v1
=1
u
nβj=1
ajaj
[ 1
v2(g(x2, v2)β g(x1, v2)) +
1
v1v2(g(x1, v2)v1 β g(x1, v1)v2)
].
Thus, from inequalities (18) and (21) we get sgn(RM1 β 1) = sgn(v1 β v2).
Theorem 2. Assume that Conditions C1-C4 are satisfied. If RM1 β€ 1 < RM0 , then Q1 is GAS in Ξ.
Proof. Define:
V1(x, y1, ..., yn, v, z) = xβ x1 ββ« x
x1
g(x1, v1)
g(Ξ·, v1)dΞ· +
nβi=1
iβ1βj=1
ajaj
yi β yi,1 β yiβ«
yi,1
Οi(yi,1)
Οi(Ξ·)dΞ·
+
nβj=1
ajaj
v1H
(v
v1
)+p
r
nβj=1
ajaj
z. (22)
We note that, V1 is positive and reaches its global minimum at Q1. Calculating the time derivative of V1 along
the trajectories of system (1)-(5), we obtain
dV1dt
=
(1β g(x1, v1)
g(x, v1)
)(Ξ»β dxβ g(x, v)) +
(1β Ο1(y1,1)
Ο1(y1)
)(g(x, v)β a1Ο1(y1))
+
nβi=2
iβ1βj=1
ajaj
(1β Οi(yi,1)
Οi(yi)
)(aiβ1Οiβ1(yiβ1)β aiΟi(yi))
+
nβj=1
ajaj
(1β v1v
)(anΟn(yn)β uv β pzv) +
p
r
nβj=1
ajaj
(rzv β bz) . (23)
We havenβi=2
iβ1βj=1
ajaj
(aiβ1Οiβ1(yiβ1)β aiΟi(yi)) = a1Ο1(y1)β
nβj=1
ajaj
anΟn(yn). (24)
Then,
dV1dt
=
(1β g(x1, v1)
g(x, v1)
)(Ξ»β dx) + g(x, v)
g(x1, v1)
g(x, v1)β Ο1(y1,1)g(x, v)
Ο1(y1)
+ a1Ο1(y1,1)βnβi=2
iβ1βj=1
ajaj
aiβ1Οi(yi,1)Οiβ1(yiβ1)
Οi(yi)
+nβi=2
iβ1βj=1
ajaj
aiΟi(yi,1)β
nβj=1
ajaj
uv β
nβj=1
ajaj
anv1Οn(yn)
v
+
nβj=1
ajaj
uv1 +
nβj=1
ajaj
pv1z βpb
r
nβj=1
ajaj
z. (25)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
955 A. M. Elaiw et al 949-967
Using the steady state conditions for Q1:
Ξ» = dx1 + g(x1, v1),
g(x1, v1) =
iβj=1
ajaj
aiΟi(yi,1) =
nβj=1
ajaj
uv1, i = 1, ..., n.
We obtain
dV1dt
=
(1β g(x1, v1)
g(x, v1)
)(dx1 β dx) + g(x1, v1)
(1β g(x1, v1)
g(x, v1)
)+ g(x1, v1)
g(x, v)
g(x, v1)
β g(x1, v1)Ο1(y1,1)g(x, v)
Ο1(y1)g(x1, v1)+ (n+ 1) g(x1, v1)β g(x1, v1)
nβi=2
Οi(yi,1)Οiβ1(yiβ1)
Οi(yi)Οiβ1(yiβ1,1)
β g(x1, v1)v
v1β g(x1, v1)
v1Οn(yn)
vΟn(yn,1)+ p
nβj=1
ajaj
(v1 β b
r
)z. (26)
We can rewrite Eq. (26) as follows
dV1dt
= dx1
(1β g(x1, v1)
g(x, v1)
)(1β x
x1
)+ g(x1, v1)
[g(x, v)
g(x, v1)β v
v1
]+ g(x1, v1)
[(n+ 2)β g(x1, v1)
g(x, v1)β Ο1(y1,1)g(x, v)
Ο1(y1)g(x1, v1)
βnβi=2
Οi(yi,1)Οiβ1(yiβ1)
Οi(yi)Οiβ1(yiβ1,1)β v1Οn(yn)
vΟn(yn,1)
]+ p
nβj=1
ajaj
(v1 β v2) z
= dx1
(1β g(x1, v1)
g(x, v1)
)(1β x
x1
)+ g(x1, v1)
(1β g(x, v1)
g(x, v)
)(g(x, v)
g(x, v1)β v
v1
)+ g(x1, v1)
[(n+ 3)β g(x1, v1)
g(x, v1)β Ο1(y1,1)g(x, v)
Ο1(y1)g(x1, v1)β
nβi=2
Οi(yi,1)Οiβ1(yiβ1)
Οi(yi)Οiβ1(yiβ1,1)
β v1Οn(yn)
vΟn(yn,1)β vg(x, v1)
v1g(x, v)
]+ p
nβj=1
ajaj
(v1 β v2) z. (27)
From Conditions C1 and C4, we get that, the first and second terms of Eq. (27) are less than or equal to
zero. Since the geometrical mean is less than or equal to the arithmetical mean, then (n + 3) β€ g(x1,v1)g(x,v1)
+
Ο1(y1,1)g(x,v)Ο1(y1)g(x1,v1)
+nβi=2
Οi(yi,1)Οiβ1(yiβ1)Οi(yi)Οiβ1(yiβ1,1)
+ v1Οn(yn)vΟn(yn,1)
+ vg(x,v1)v1g(x,v)
. Lemma 2 implies that, if RM1 β€ 1, then v1 β€ v2.
It follows that, dV1
dt β€ 0 for all x, yi, v, z > 0, i = 1, ..., n. The solutions of system (1)-(5) are limited to Ξ©,
the largest invariant subset of
(x, y1, ..., yn, v, z) : dV1
dt = 0
. We have dV1
dt = 0 at the singleton Q1. Thus,
the global asymptotic stability of the endemic steady state without humoral immune response Q1 follows from
LaSalleβs invariance principle.
Theorem 3. Let Conditions C1-C4 are satisfied and RM1 > 1, then Q2 is GAS inΞ.
Proof. We construct a Lyapunov functional as follows:
V2(x, y1, ..., yn, v, z) = xβ x2 ββ« x
x2
g(x2, v2)
g(Ξ·, v2)dΞ· +
nβi=1
iβ1βj=1
ajaj
yi β yi,2 β yiβ«
yi,2
Οi(yi,2)
Οi(Ξ·)dΞ·
+
nβj=1
ajaj
v2H
(v
v2
)+p
r
nβj=1
ajaj
z2H
(z
z2
). (28)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
956 A. M. Elaiw et al 949-967
Note that V2 > 0 for all x, y1, ..., yn, v, z > 0 and V2(x2, y1,2, ..., yn,2, v2, z2) = 0. Function V2 satisfies:
dV2dt
=
(1β g(x2, v2)
g(x, v2)
)(Ξ»β dxβ g(x, v)) +
(1β Ο1(y1,2)
Ο1(y1)
)(g(x, v)β a1Ο1(y1))
+nβi=2
iβ1βj=1
ajaj
(1β Οi(yi,2)
Οi(yi)
)(aiβ1Οiβ1(yiβ1)β aiΟi(yi))
+
nβj=1
ajaj
(1β v2v
)(anΟn(yn)β uv β pzv) +
p
r
nβj=1
ajaj
(1β z2z
)(rzv β bz) . (29)
Using Eq. (24), we get
dV2dt
=
(1β g(x2, v2)
g(x, v2)
)(Ξ»β dx) + g(x, v)
g(x2, v2)
g(x, v2)β Ο1(y1,2)g(x, v)
Ο1(y1)+ a1Ο1(y1,2)
βnβi=2
iβ1βj=1
ajaj
aiβ1Οi(yi,2)Οiβ1(yiβ1)
Οi(yi)+
nβi=2
iβ1βj=1
ajaj
aiΟi(yi,2)
β
nβj=1
ajaj
uv β
nβj=1
ajaj
anv2Οn(yn)
v+
nβj=1
ajaj
uv2
+
nβj=1
ajaj
pv2z βpb
r
nβj=1
ajaj
z β p
nβj=1
ajaj
vz2 +pb
r
nβj=1
ajaj
z2. (30)
Using the steady state conditions for Q2:
Ξ» = dx2 + g(x2, v2), v2 =b
r,
g(x2, v2) =
iβj=1
ajaj
aiΟi(yi,2) =
nβj=1
ajaj
(uv2 + pv2z2) , i = 1, ..., n,
we get
dV2dt
=
(1β g(x2, v2)
g(x, v2)
)(dx2 β dx) + g(x2, v2)
(1β g(x2, v2)
g(x, v2)
)+ g(x2, v2)
g(x, v)
g(x, v2)
β g(x2, v2)Ο1(y1,2)g(x, v)
Ο1(y1)g(x2, v2)+ (n+ 1) g(x2, v2)β g(x2, v2)
nβi=2
Οi(yi,2)Οiβ1(yiβ1)
Οi(yi)Οiβ1(yiβ1,2)
β g(x2, v2)v
v2β g(x2, v2)
v2Οn(yn)
vΟn(yn,2). (31)
We can rewrite Eq. (31) as follows:
dV2dt
= dx2
(1β g(x2, v2)
g(x, v2)
)(1β x
x2
)+ g(x2, v2)
(1β g(x, v2)
g(x, v)
)(g(x, v)
g(x, v2)β v
v2
)+ g(x2, v2)
[(n+ 3)β g(x2, v2)
g(x, v2)β Ο1(y1,2)g(x, v)
Ο1(y1)g(x2, v2)β
nβi=2
Οi(yi,2)Οiβ1(yiβ1)
Οi(yi)Οiβ1(yiβ1,2)
β v2Οn(yn)
vΟn(yn,2)β vg(x, v2)
v2g(x, v)
]. (32)
We note from Conditions C1 and C4 and the relationship between the arithmetical and geometrical means that,
we obtain dV2
dt β€ 0 for all x, y1, ..., yn, v, z > 0. The solutions of model (1)-(5) are limited to Ξ, the largest
invariant subset of
(x, y1, ..., yn, v, z) : dV2
dt = 0
. It is easy to see that dV2
dt = 0 occurs at Q2. The global
asymptotic stability of Q2 follows from LaSalleβs invariance principle.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
957 A. M. Elaiw et al 949-967
6 Example and numerical simulations
In this section, we introduce an example and perform some numerical simulations to confirm our theoretical
results. By using the Lyapunov direct method, we have established a set of conditions on the functions g(x, v)
and Οi(yi) and on the parameters RM0 and RM1 ensuring the global asymptotic stability of the steady states of
model (1)-(5). We consider the following model with two stages (i.e. n = 2):
x = Ξ»β dxβ Οxv
(1 + Ξ³x) (1 + Ξ΄v), (33)
y1 =Οxv
(1 + Ξ³x) (1 + Ξ΄v)β a1y1, (34)
y2 = a1y1 β a2y2, (35)
v = a2y2 β pzv β uv, (36)
z = rzv β bz, (37)
where Ο β (0,β) and Ξ³, Ξ΄ β [0,β). In this example we have
Οi(yi) = yi, i = 1, ..., n, g(x, v) =Οxv
(1 + Ξ³x) (1 + Ξ΄v),
which guarantee that Condition C3 holds true. Now, we verify Conditions C1, C2 and C4. Clearly, g(x, v) > 0,
g(0, v) = g(x, 0) = 0 for all x, v β (0,β), and
βg(x, v)
βx=
Οv
(1 + Ξ³x)2 (1 + Ξ΄v),
βg(x, v)
βv=
Οx
(1 + Ξ³x) (1 + Ξ΄v)2 ,
βg(x, 0)
βv=
Οx
1 + Ξ³x.
Then, for all x, v β (0,β), we have βg(x,v)βx > 0, βg(x,v)βv > 0 and βg(x,0)
βv > 0. Therefore Condition C1 is satisfied.
We have also
g(x, v) =Οxv
(1 + Ξ³x) (1 + Ξ΄v)β€ Οxv
1 + Ξ³x= v
βg(x, 0)
βv,(
βg(x, 0)
βv
)β²=
Ο
(1 + Ξ³x)2> 0 for all x > 0.
It follows that, C2 is satisfied. Moreover,(1β g(x, vi)
g(x, v)
)(g(x, v)
g(x, vi)β v
vi
)= β Ξ΄ (v β vi)2
vi (1 + Ξ΄v) (1 + Ξ΄vi)< 0 for all v, vi β (0,β), i = 1, 2.
Thus, C4 is satisfied and the global stability results demonstrated in Theorems 1-3 are guaranteed. The
parameters RM0 and RM1 are given by:
RM0 =a1a2Ο
a1a2u
x01 + Ξ³x0
, RM1 =a1a2Ο
a1a2u
x2(1 + Ξ³x2) (1 + Ξ΄v2)
. (38)
Now, we will perform some numerical simulations for the model (33)-(37). The values of some parameters of
the example are listed in Table 1. The other parameters Ο, r and Ξ³ will be varied. All computations are carried
out by MATLAB.
We are interested to study the following cases:
Case (A): Effect of Ο and r on the stability of steady states:
In this case, we have chosen three different initial conditions:
IC(1): x(0) = 400, y1(0) = y2(0) = 1, v(0) = 0.2 and z(0) = 0.5,
IC(2): x(0) = 600, y1(0) = y2(0) = 2, v(0) = 0.5 and z(0) = 1,
IC(3): x(0) = 800, y1(0) = 5, y2(0) = 3, v(0) = 0.9 and z(0) = 1.5.
The evolution of the dynamics of model (33)-(37) was observed over a time interval [0, 500]. We fix the value
of Ξ³ = 0.5 and change the values of parameters Ο and r to get three sets as follows:
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
958 A. M. Elaiw et al 949-967
Table 1: The values of the parameters of model (33)-(37).
Parameter Value Parameter Value Parameter Value
Ξ» 10 a1 1 p 0.5
d 0.01 a2 1.5 r Varied
Ξ² Varied a1 0.5 b 0.3
Ξ³ Varied a2 1
Ξ΄ 0.1 u 3
Set (I): We choose, Ο = 4 and r = 0.3. Using the values of the parameters given in Table 1, we compute
RM0 = 0.89 < 1 and RM1 = 0.80 < 1, which means that the system has a disease-free steady state Q0 and it
is GAS based on Theorem 1. Evidently, Figures 1-5 show that, the states of the system eventually approach
Q0 = (1000, 0, 0, 0, 0) for the three initial conditions IC(1)-IC(3). This case corresponds to the healthy state
where the viruses are cleared.
Set (II): We take Ο = 5 and r = 0.3. With such choice we have, RM1 = 0.99 < 1 < RM0 = 1.11. Consequently,
Lemma 1 and Theorem 2 state that, Q1 exists and it is GAS. Figures 1-5 show that the numerical simulations
illustrate our theoretical results given in Theorem 2. We observe that, the trajectory of the system will converge
toQ1 = (140.43, 8.60, 2.87, 0.96, 0) for the three initial conditions IC(1)-IC(3). This case corresponds to a chronic
infection but with inactive immune response.
Set (III): We choose, Ο = 5 and r = 1. Then, we calculate RM0 = 1.11 > 1 and RM1 = 1.08 > 1, this means
that, the system has three steady states Q0, Q1 and Q2. Thus, from Theorem 3, Q2 is GAS. From Figures 1-5, we
observe a consistency between the numerical results and theoretical results of Theorem 3. We observe that, the
trajectory of the system show oscillating behavior for a period before reaching Q2 = (709.56, 2.90, 0.97, 0.3, 0.45),
in the same time frame for the three initial conditions IC(1)-IC(3).
Case (B): Effect of Ξ³ on the stability of the steady states
Let us consider Ο and r be fixed. In this case, we take the values of Ο = 5 and r = 1, and consider different
values of Ξ³. Here we take the initial condition as given in IC(1), while the evolution of the dynamics of model
(33)-(37) was observed over a time interval [0, 600]. Table 2 contains the values of the bifurcation parameters
RM0 and RM1 with different values of Ξ³ of model (33)-(37).
Table 2: The values of the threshold parameters RM0 and RM1 with different values of Ξ³ of model (33)-(37).
Different values of Ξ³ RM0 RM1 The equilibria
0.30 1.85 1.79 Q2 = (517.67, 4.82, 1.61, 0.3, 4.72)
0.40 1.39 1.34 Q2 = (637.35, 3.63, 1.21, 0.3, 2.06)
0.54 1.03 0.996 Q1 = (763.16, 2.37, 0.79, 0.26, 0)
0.55 1.01 0.98 Q1 = (926.89, 0.73, 0.24, 0.08, 0)
0.60 0.92 0.90 Q0 = (1000, 0, 0, 0, 0)
0.70 0.79 0.77 Q0 = (1000, 0, 0, 0, 0)
Table 2 and Figures 6-10 show that, when Ξ³ is increased, the infection rate is decreased which leads to an
increase in the concentration of the uninfected cells and a decrease on the concentrations of the (first/second)
stage of infected cells, free viruses and B cells.
Case (C): Effect of the multiple stages of infected cells on the dynamics of virus dynamics:
To show the effect of multiple stages of infected cells on the dynamical behavior of the virus, we consider
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
959 A. M. Elaiw et al 949-967
the following model with single stage of infected cells and compare it with model (33)-(37):
x = Ξ»β dxβ Οxv
(1 + Ξ³x) (1 + Ξ΄v), (39)
y1 =Οxv
(1 + Ξ³x) (1 + Ξ΄v)β a1y1, (40)
v = a1y1 β pzv β uv, (41)
z = rzv β bz. (42)
Consequently, the bifurcation parameters for this system are given by:
Rsingle0 =a1Ο
a1u
x01 + Ξ³x0
, Rsingle1 =a1Ο
a1u
x2(1 + Ξ³x2) (1 + Ξ΄v2)
. (43)
Since ai < ai, then from Eqs. (38) and (43) we have
RM0 =a1a2Ο
a1a2u
x01 + Ξ³x0
<a1Ο
a1u
x01 + Ξ³x0
= Rsingle0 ,
RM1 =a1a2Ο
a1a2u
x2(1 + Ξ³x2) (1 + Ξ΄v2)
<a1Ο
a1u
x2(1 + Ξ³x2) (1 + Ξ΄v2)
= Rsingle1 .
Here we consider the following initial condition: x(0) = 400, y1(0) = 0.5, y2(0) = 1, v(0) = 0.2 and z(0) = 0.5.
The evolution of the dynamics of models (33)-(37) and (39)-(42) was observed over a time interval [0, 600]. Let
us consider the values of parameters listed in Table 1 and choose the values Ο = 3.5, r = 1.5 and Ξ³ = 0.5. By
calculating the bifurcation parameters for systems (33)-(37) and (39)-(42), we obtain
RM0 = 0.78 < 1.16 = Rsingle0 , RM1 = 0.76 < 1.14 = Rsingle1 .
Therefore, with the same values of the parameters, the steady state Q0 is stable for system (33)-(37) but unstable
for system (39)-(42). The presence of multiple stages of infected cells reduces the infection progress. Figures
11-14 show a comparison between the evolution of the uninfected cells, infected cells, free virus particles and B
cells of the two systems (33)-(37) and (39)-(42). We observe that, the concentration of uninfected cells of the
model with three stages of infected cells is larger than that of system with only one single stage of infected cells
(see Figures 11), while the concentrations of first stage of infected cells, viruses and B cells with three stages
are less than that of system with a single stage of infected cells (see Figures 12-14). From a biological point of
view, the multiple stages of infected cells plays a similar role as antiviral treatment in eliminating the virus. We
observe that, if the number of stages of infected cells is increased, then the viral replication is suppressed and
the viruses can be cleared from the body. This give us some suggestions on new drugs to increase the number
of stages of infected cells.
7 Conclusion
We have studied a general virus dynamics model with humoral immunity. We have assumed that the infected
cells passes through n-stages to produce mature viruses. We have obtained two bifurcation parameters, the basic
reproduction number and the humoral immunity number. We have established a set of sufficient conditions
which guarantee the global stability of the model. The global asymptotic stability of the three steady states, Q0,
Q1 and Q2 has been investigated by constructing Lyapunov functionals and using LaSalleβs invariance principle.
To support our theoretical results, we have presented an example and conducted some numerical simulations.
8 Acknowledgment
This article was funded by the Deanship of Scientific Research (DSR), King Abdulaziz University, Jeddah. The
authors, therefore, acknowledge with thanks DSR technical and financial support.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
960 A. M. Elaiw et al 949-967
0 50 100 150 200 250 300 350 400 450 500100
200
300
400
500
600
700
800
900
1000
Un
infe
cted
tar
get
cell
s
Time
Set (I)
Set (II)
Set (III)
Figure 1: The uninfected cells for model (33)-(37).
0 50 100 150 200 250 300 350 400 450 5000
1
2
3
4
5
6
7
8
9
10
Fir
st s
tage
of
infe
cted
cel
ls
Time
Set (II)
Set (III)
Set (I)
Figure 2: The first stage infected cells for model (33)-(37).
0 50 100 150 200 250 300 350 400 450 5000
0.5
1
1.5
2
2.5
3
Sec
ond
sta
ge o
f in
fect
ed c
ells
Time
Set (III)
Set (II)
Set (I)
Figure 3: The second stage infected cells for model (33)-(37).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
961 A. M. Elaiw et al 949-967
0 50 100 150 200 250 300 350 400 450 5000
0.2
0.4
0.6
0.8
1
Fre
e vi
rus
par
ticl
es
Time
Set (II)
Set (III)
Set (I)
Figure 4: The free virus particles for model (33)-(37).
0 50 100 150 200 250 300 350 400 450 5000
0.5
1
1.5
2
2.5
3
3.5
B c
ells
Time
Sets (I) & (II) Set (III)
Figure 5: The B cells for model (33)-(37).
0 100 200 300 400 500 600
400
500
600
700
800
900
1000
Time
Un
infe
cted
tar
get
cell
s
Ξ³=0.30
Ξ³=0.54
Ξ³=0.60
Figure 6: The uninfected target cells for model (33)-(37) under different values of Ξ³.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
962 A. M. Elaiw et al 949-967
0 100 200 300 400 500 6000
2
4
6
8
10
12
14
Time
Fir
st s
tage
of
infe
cted
cel
ls
Ξ³=0.30
Ξ³=0.54
Ξ³=0.60
Figure 7: The first stage infected cells for model (33)-(37) under different values of Ξ³.
0 100 200 300 400 500 6000
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Time
Sec
ond
sta
ge o
f in
fect
ed c
ells
Ξ³=0.30
Ξ³=0.54
Ξ³=0.60
Figure 8: The second stage infected cells for model (33)-(37) under different values of Ξ³.
0 100 200 300 400 500 6000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time
Fre
e vi
rus
par
ticl
es
Ξ³=0.30
Ξ³=0.54
Ξ³=0.60
Figure 9: The free virus particles for model (33)-(37) under different values of Ξ³.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
963 A. M. Elaiw et al 949-967
0 100 200 300 400 500 6000
2
4
6
8
10
12
Time
B c
ell
s
Ξ³=0.30
Ξ³=0.54
Ξ³=0.60
Figure 10: The B cells for model (33)-(37) under different values of Ξ³.
0 100 200 300 400 500 600400
500
600
700
800
900
1000
Un
infe
cted
tar
get
cell
s
Time
For two stages of infected
For single stage of infected
Figure 11: Comparison on the concentration of the uninfected cells for systems (33)-(37) and (39)-(42).
0 100 200 300 400 500 6000
0.5
1
1.5
2
2.5
Time
Fir
st s
tage
of
infe
cted
cel
ls
For two stages of infected
For single stage of infected
Figure 12: Comparisons on the concentration of the first stage of infected cells for systems (33)-(37) and
(39)-(42).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
964 A. M. Elaiw et al 949-967
0 100 200 300 400 500 6000
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Fre
e vi
rus
par
ticl
es
Time
For two stages of infected
For single stage of infected
Figure 13: Comparisons on the concentration of the free virus particles for systems (33)-(37) and (39)-(42).
0 100 200 300 400 500 6000
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
B c
ells
Time
For two stages of infected
For single stage of infected
Figure 14: Comparisons on the concentration of the B cells for systems (33)-(37) and (39)-(42).
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
965 A. M. Elaiw et al 949-967
References
[1] M. A. Nowak and R. M. May, Virus dynamics: Mathematical Principles of Immunology and Virology,
Oxford Uni., Oxford, (2000).
[2] C. Connell McCluskey, Yu Yang, Global stability of a diffusive virus dynamics model with general incidence
function and time delay, Nonlinear Anal. Real World Appl., 25 (2015), 64-78.
[3] P. K. Roy, A. N. Chatterjee, D. Greenhalgh, and Q. J.A. Khan, Long term dynamics in a mathematical
model of HIV-1 infection with delay in different variants of the basic drug therapy model, Nonlinear Anal.
Real World Appl., 14 (2013), 1621-1633.
[4] A. M. Elaiw, and N. H. AlShamrani, Stability analysis of general viral infection models with humoral
immunity, Journal of Nonlinear Science and Applications, 9 (2016), 684-704.
[5] M. Y. Li and L. Wang, Backward bifurcation in a mathematical model for HIV infection in vivo with
anti-retroviral treatment, Nonlinear Anal. Real World Appl., 17 (2014), 147-160.
[6] G. Huang, W. Ma, Y. Takeuchi, Global properties for virus dynamics model with BeddingtonβDeAngelis
functional response, Appl. Math. Lett., 22 (2009), 1690-1693.
[7] A.M. Elaiw and S.A. Azoz, Global properties of a class of HIV infection models with Beddington-DeAngelis
functional response, Math. Methods Appl. Sci., 36 (2013), 383-394.
[8] A.M. Elaiw, Global properties of a class of HIV models, Nonlinear Anal. Real World Appl., 11 (2010),
2253-2263.
[9] A. M. Elaiw, Global properties of a class of virus infection models with multitarget cells, Nonlinear Dynam.,
69 (2012), 423-435.
[10] A. M. Elaiw, Global stability analysis of humoral immunity virus dynamics model including latently infected
cells, Journal of Biological Dynamics, DOI:10.1080/17513758.2015.1056846.
[11] L. Wang and M. Y. Li, Mathematical analysis of the global dynamics of a model for HIV infection of CD4+
T cells, Math. Biosci., 200(1) (2006), 44-57.
[12] K. Hattaf, N. Yousfi and A. Tridane, Mathematical analysis of a virus dynamics model with general incidence
rate and cure rate, Nonlinear Anal. Real World Appl., 13(4) (2012), 1866-1872.
[13] K. Wang, A. Fan and A. Torres, Global properties of an improved hepatitis B virus model, Nonlinear Anal.
Real World Appl., 11 (2010), 3131-3138.
[14] X. Song and A. Neumann, Global stability and periodic solution of the viral dynamics, J. Math. Anal. Appl.,
329 (2007), 281-297.
[15] A. U. Neumann, N. P. Lam, H. Dahari, D. R. Gretch, T. E. Wiley, T. J, Layden and A. S. Perelson, Hepatitis
C viral dynamics in vivo and the antiviral efficacy of interferon-alpha therapy, Science, 282 (1998), 103-107.
[16] L. Wang, M. Y. Li, and D. Kirschner, Mathematical analysis of the global dynamics of a model for HTLV-I
infection and ATL progression, Math. Biosci., 179 (2002) 207-217.
[17] A. M. Elaiw, and N. A. Almuallem, Global properties of delayed-HIV dynamics models with differential
drug efficacy in co-circulating target cells, Applied Mathematics and Computation, 265 (2015), 1067-1089.
[18] A. M. Elaiw, and N. A. Almuallem, Global dynamics of delay-distributed HIV infection models with differ-
ential drug efficacy in cocirculating target cells, Mathematical Methods in the Applied Sciences, 39 (2016),
4-31.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
966 A. M. Elaiw et al 949-967
[19] A. Korobeinikov, Global properties of basic virus dynamics models, Bull. Math. Biol., 66 (2004), 879-883.
[20] P. Georgescu and Y.H. Hsieh, Global stability for a virus dynamics model with nonlinear incidence of
infection and removal, SIAM J. Appl. Math., 67 (2) (2006), 337-353.
[21] G. Huang, Y. Takeuchi and W. Ma, Lyapunov functionals for delay differential equations model of viral
infections, SIAM J. Appl. Math., 70(7) (2010), 2693-2708.
[22] Z. Grossman, M. Polis, M. B. Feinberg, Z. Grossman, I. Levi, S. Jankelevich, R. Yarchoan, J. Boon, F. de
Wolf, J. M.A. Lange, J. Goudsmit, D. S. Dimitrov and W. E. Paul, Ongoing HIV dissemination during
HAART, Nat. Med., 5 (1999), 1099-1104.
[23] J. Wang and S. Liu, The stability analysis of a general viral infection model with distibuted delays and
multi-staged infected progression, Commun. Nonlinear Sci. Numer. Simul., 20(1) (2015), 263β272.
[24] J. A. Deans and S. Cohen, Immunology of malaria, Ann. Rev. Microbiol. 37 (1983), 25-49.
[25] A. Murase, T. Sasaki and T. Kajiwara, Stability analysis of pathogen-immune interaction dynamics, J.
Math. Biol., 51 (2005), 247-267.
[26] M. A . Obaid and A. M. Elaiw, Stability of virus infection models with antibodies and chronically infected
cells, Abstr. Appl. Anal, (2014) Article ID 650371.
[27] S. Wang and D. Zou, Global stability of in host viral models with humoral immunity and intracellular delays,
Appl. Math. Model., 36 (2012), 1313-1322.
[28] T. Wang, Z. Hu and F. Liao, Stability and Hopf bifurcation for a virus infection model with delayed humoral
immunity response, J. Math. Anal. Appl., 411 (2014) 63-74.
[29] A. M. Elaiw and N. H. AlShamrani, Global properties of nonlinear humoral immunity viral infection models,
Int. J. Biomath., 8(5) (2015), 1550058, 53 pages.
[30] A. M. Elaiw and N. H. AlShamrani, Global stability of humoral immunity virus dynamics models with
nonlinear infection rate and removal, Nonlinear Anal. Real World Appl., 26 (2015), 161-190.
[31] T. Wang, Z. Hu, F. Liao and W. Ma, Global stability analysis for delayed virus infection model with general
incidence rate and humoral immunity, Math. Comput. Simulation, 89 (2013), 13-22.
[32] R. Larson and B. H. Edwards, Calculus of a single variable, Cengage Learning, Inc., USA, (2010).
[33] J.K. Hale and S. Verduyn Lunel, Introduction to functional differential equations, Springer-Verlag, New
York, (1993).
[34] A.S. Perelson and P.W. Nelson, Mathematical analysis of HIV-1 dynamics in vivo, SIAM Rev., 41 (1999),
3β44.
[35] P. De. Leenheer and H. L. Smith, Virus dynamics: A global analysis, SIAM J. Appl. Math., 63 (2003),
1313-1327.
[36] G. Huang, W. Ma, Y. Takeuchi, Global properties for virus dynamics model with BeddingtonβDeAngelis
functional response, Appl. Math. Lett. 22 (2009), 1690β1693.
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
967 A. M. Elaiw et al 949-967
On the dynamics of a certain four-order fractional difference equations
Chang-you Wang1,2, Xiao-jing Fang1,2, Rui Li1*
1. College of Automation, Chongqing University of Posts and Telecommunications, Chongqing 400065 P. R. China
2. Key Laboratory of Industrial Internet of Things & Networked Control of Ministry of Education, Chongqing University of Posts and Telecommunications,
Chongqing 400065 P.R. China Abstract: This paper is concerned with the following rational recursive sequences
1 2 1 21 1
3 3
, 0,1, ,n n n nn n
n n
x x y yx y nA By C Dx
β β β β+ +
β β
= = =+ +
οΌ
where the parameters , , ,A B C D are positive constants. The initial condition 3 2, ,x xβ β
1 0,x xβ and 3 2 1 0, , ,y y y yβ β β are arbitrary nonnegative real numbers. We give sufficient
conditions under which the equilibrium (0,0) of the system is globally asymptotically stable, which extends and includes corresponding results obtained in the cited references [12-17]. Moreover, the asymptotic behavior of others equilibrium points is also studied. Our approach to the problem is based on new variational iteration method for the more general nonlinear difference equations and inequality skills as well as the linearization techniques.
Keywords: recursive sequences; equilibrium point; asymptotical stability; positive solutions. 1. Introduction
Nonlinear Difference equations have been studied because they model numerous real life problems in biology, ecology, physics, economics and so forth [1-5]. Today, with the dramatically development of computer-based computational techniques, difference equations are found to be much appropriate mathematical representations for computer simulation, experiment and computation, which play an important role in realistic applications [6]. Therefore, recently there has been an increasing interest in the study of qualitative analysis of rational difference equations. And the present cardinal problem of
* Corresponding author at: College of Automation, Chongqing University of Posts and Telecommunications, Chongqing 400065 P. R. China, Chongqing 400065, PR China. Email address: [email protected]
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
968 Chang-you Wang et al 968-976
2
asymptotic behavior of solutions for a rational difference equation has received extensive attention from researchers (see, e.g., [7-11] and the references therein).
Elabbasy [12] obtained the form of the solutions of the following rational difference system
1 11 1
-1 -11 1n n
n nn n n n
x yx yx y y xβ β
+ += =Β± + +β
οΌ (1.1)
with nonzero real number initial conditions. In particular, Clark and Kulenovic [13, 14] discussed the global stability properties and
asymptotic behavior of solutions for the recursive sequence
1 1, , 0,1, ,n nn n
n n
x yx y na cy b dx+ += = =+ +
(1.2)
where , , , (0, )a b c d β β and the initial conditions 0x and 0y are arbitrary nonnegative
numbers. In 2012, Zhang et al. [15] investigated the stability character and asymptotic behavior of the solution for the system of difference equations
-2 -21 1
2 1 2 1
, , 0,1, ,n nn n
n n n n n n
x yx y nB y y y A x x x+ +
β β β β
= = =+ +
(1.3)
where , (0, )A Bβ β , and the initial conditions 3 2 1 0 3 2 1 0, , , , , , , (0, ).x x x x y y y yβ β β β β β β β
Recently, the following nonlinear two-dimensional difference systems
1 1 2 21 1( , ), ( , )n n t n s n n s n tx x y y y xΟ Ο+ β β + β β= = , (1.4)
where 1 1 2 2, , ,t s s t are all positive integers, was studied by Liu et al. [16], in which they
gave some sufficient conditions such that every positive solution of this equation converges to the unique equilibrium point. More recently, in [17] the authors studied analogous results for the system of difference equations
1 1 1 1,n nx yn n n n n nx ax by e y cy dx eβ β+ β + β= + = + , (1.5)
where , , ,a b c d are positive constants and the initial values 1 0 1 0, , ,x x y y are positive
numbers. For more related work, one can refer to [18-22] and references therein. Inspired by the above works, the essential problem we consider in this paper is the
asymptotic behavior of the solution for the difference equation
1 2 1 21 1
3 3
, 0,1, ,n n n nn n
n n
x x y yx y nA By C Dx
β β β β+ +
β β
= = =+ +
οΌ (1.6)
where the initial conditions 3 2 1 0, , , (0, )x x x xβ β β β β , 3 2 1 0, , , (0, )y y y yβ β β β β and , , ,A B C D
are positive constants. This paper proceeds as follows. In Section 2, we introduce some definitions and
preliminary results. The main results and their proofs are given in Section 3.
2. Preliminaries
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
969 Chang-you Wang et al 968-976
3
Let ,x yI I be some intervals of real numbers and 4 4: x y xf I I IΓ β , 4 4: x y yg I I IΓ β be
continuously differentiable functions. Then for every initial conditions ( , )i i x yx y I Iβ Γ ,
( 3, 2, 1,0)i = β β β , the system of difference equations
1 -1 -2 -3 -1 -2 -3
1 -1 -2 -3 -1 -2 -3
( , , , , , , , ),0,1, 2, ,
( , , , , , , , ),n n n n n n n n n
n n n n n n n n n
x f x x x x y y y yn
y g x x x x y y y y+
+
=β§=β¨ =β©
(2.1)
has a unique solution n 3( , )n nx y β=β . A point ( , ) x yx y I Iβ Γ is called an equilibrium point
of (2.1) if ( )( , , , , , , , ), . , , , , , ,x f x x x x y y y y y g x x x x y y y y= = , i. e., ( , ) ( , )n nx y x y=
for all 0n β₯ . Interval x yI IΓ is called invariant for system (2.1) if, for all 0n > , ,n x n yx I y Iβ β
when the initial conditions 3 2 1 0 3 2 1 0, , , , , ,x yx x x x I y y y y Iβ β β β β ββ β .
Definition 2.1 Assume that ( , )x y is a fixed point of (2.1). Then
(i) ( , )x y is said to be stable relative to x yI IΓ if for every 0Ξ΅ > , there exits 0Ξ΄ >
such that for any initial conditions ,( ) ( 3, 2, 1,0)i i x yx y I I iβ Γ =β β β , with 0
3 iix x Ξ΄
=ββ <β ,
0
3 iiy y Ξ΄
=ββ <β , implies ,n nx x y yΞ΅ Ξ΅β < β < .
(ii) ( , )x y is called an attractor relative to x yI IΓ if for all ,( )i i x yx y I Iβ Γ
( 3, 2, 1,0)i = β β β , ,n n n nlim x x lim y yββ ββ= = .
(iii) ( , )x y is called asymptotically stable relative to x yI IΓ if it is stable and an
attractor. (iv) Unstable if it is not stable.
Theorem 2.1 Assume that ( 1) ( ( )) , 0,1,X n F X n n+ = = , is a system of difference
equations and X is the equilibrium point of this system i.e., ( )F X X= .
(i) If all eigenvalues of the Jacobian matrix FJ , evaluated at X lie inside the open
unit disk 1Ξ» < , then X is locally asymptotically stable.
(ii) If all eigenvalues of the Jacobian matrix FJ , evaluated at X has modulus greater
than one then X is unstable. Definition 2.2 Let , , ,p q s t be four nonnegative integers such that p q s t n+ = + = .
Splitting 1 2 1 2( , ) ( , , , , , , , )n nx y x x x y y y= into ( , ) ( [ ] , [ ] , [ ] , [ ] )p q s tx y x x y y= ,
where [ ]x Ο denotes a vector with Ο -components of x , we say that the function
1 2 1 2( , , , , , , , )n nf x x x y y y possesses a mixed monotone property in subsets 2nI of 2nR if ([ ] ,[ ] ,[ ] ,[ ] )p q s tf x x y y is monotone nondecreasing in each component of
[ ] , [ ]p sx y and is monotone nonincreasing in each component of [ ] , [ ]q tx y for 2( , ) nx y Iβ . In particular, if 0q t= = , then it is said to be monotone nondecreasing
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
970 Chang-you Wang et al 968-976
4
in 2nI .
3. The Main Results In this section, we investigate the asymptotic behavior of the equilibrium points of the
systems (1.6). It is easy to know that the systems (1.6) have four equilibrium points (0, 0) , (0, )C , ( , 0)A , and (( ) / (1 ), ( ) / (1 ))A BC BD C AD BD+ β + β . Theorem 3.1 The equilibrium point (0, 0) of (1.6) is locally asymptotically stable. Proof. We can easily obtain that the linearized system of (1.6) about the equilibrium point (0, 0) is
1n nDΟ Ο+ = (3.1)
where
1
2
3
1
2
3
0 0 0 0 0 0 0 01 0 0 0 0 0 0 00 1 0 0 0 0 0 00 0 1 0 0 0 0 0
,0 0 0 0 0 0 0 00 0 0 0 1 0 0 00 0 0 0 0 1 0 00 0 0 0 0 0 1 0
n
n
n
nn
n
n
n
n
xxxx
Dyyyy
Ο
β
β
β
β
β
β
β‘ β€ β‘ β€β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯= =β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β£ β¦β£ β¦
. (3.2)
Thus, the characteristic equation of (3.2) is
8( ) 0f Ξ» Ξ»= = . This shows that all the roots of characteristic equation lie inside unit disk. So the equilibrium 0,0οΌ οΌ is locally asymptotically stable. Theorem 3.2 Let
1 1 1
1 1 1
( , , , , , , , ),0,1,
( , , , , , , , ),n n n n k n n n k
n n n n k n n n k
x f x x x y y yn
y g x x x y y y+ β β β β
+ β β β β
=β§=β¨ =β©
, (3.3)
[ , ]a b be an interval of real numbers and assume that 1 1:[ , ] [ , ] [ , ]k kf a b c d a b+ +Γ β and 1 1:[ , ] [ , ] [ , ]k kg a b c d c d+ +Γ β are two continuous functions satisfying the mixed monotone
property. If there exit
0 1 0 1 0 0min , , , max , , , k k k km x x x x x x Mβ β + β β +β€ β€ β€ ,
and
0 1 0 1 0 0min , , , max , , , k k k kn y y y y y y Nβ β + β β +β€ β€ β€
such that
0 0 0 0 0 0 0 0 0 0([ ] ,[ ] ,[ ] ,[ ] ) ([ ] ,[ ] ,[ ] ,[ ] ) ,p q s t p q s tm f m M n N f M m N n Mβ€ β€ β€ (3.4)
and
0 0 0 0 0 0 0 0 0 0([ ] ,[ ] ,[ ] ,[ ] ) ([ ] ,[ ] ,[ ] ,[ ] )p q s t p q s tn g m M n N g M m N n Nβ€ β€ β€ , (3.5)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
971 Chang-you Wang et al 968-976
5
then there exit 20 0( , ) [ , ]m M m Mβ and 2
0 0( , ) [ , ]n N n Nβ satisfying
([ ] ,[ ] ,[ ] ,[ ] ), ([ ] ,[ ] ,[ ] ,[ ] )p q s t p q s tM f M m N n m f m M n N= = , (3.6)
and ([ ] ,[ ] ,[ ] ,[ ] ), ([ ] ,[ ] ,[ ] ,[ ] )p q s t p q s tN g M m N n n g m M n N= = . (3.7)
Moreover, if m M= and n N= , then the system (3.3) has a unique equilibrium point
0 0 0 0( , ) [ , ] [ , ]x y m M n Nβ Γ and every solution of (3.3) converges to ( , )x y .
Proof. Using 0 0,m M and 0 0,n N as two couples of initial iteration, we construct four
sequences , , i i im M n and ( 1, 2, )iN i = from the following equations
1 1 1 1 1 1 1 1([ ] ,[ ] ,[ ] ,[ ] ), ([ ] ,[ ] ,[ ] ,[ ] )i i p i q i s i t i i p i q i s i tm f m M n N M f M m N nβ β β β β β β β= = ,
and
1 1 1 1 1 1 1 1([ ] ,[ ] ,[ ] ,[ ] ), ([ ] ,[ ] ,[ ] ,[ ] )i i p i q i s i t i i p i q i s i tn g m M n N N g M m N nβ β β β β β β β= = .
It is obvious from the mixed monotone property of functions f and g that the
sequences , , i i im M n and ( 1, 2, )iN i = possess the following monotone property
0 1 1 0 ,i im m m M M Mβ€ β€ β€ β€ β€ β€ β€ β€ (3.8)
and
0 1 1 0 ,i in n n N N Nβ€ β€ β€ β€ β€ β€ β€ β€ (3.9)
where =0,1,2,i . Moreover, one has
i l im x Mβ€ β€ for ( 1) 1, 0,1,2, .l k i iβ₯ + + = (3.10)
and
i l in y Nβ€ β€ for ( 1) 1, 0,1,2, .l k i iβ₯ + + = (3.11)
Set lim , lim , lim , lim ,i i i ii i i i
m m M M n n N Nββ ββ ββ ββ
= = = = (3.12)
then liminf limsup , liminf limsup .i i i ii i i i
m x x M n y y Nββ ββ ββ ββ
β€ β€ β€ β€ β€ β€ (3.13)
By the continuity of f and g , we have
([ ] ,[ ] ,[ ] ,[ ] ), ([ ] ,[ ] ,[ ] ,[ ] )p q s t p q s tM f M m N n m f m M n N= = , (3.14)
and ([ ] ,[ ] ,[ ] ,[ ] ), ([ ] ,[ ] ,[ ] ,[ ] )p q s t p q s tN g M m N n n g m M n N= = . (3.15)
Moreover, if ,m M n N= = , then lim , limi ii im M x x n N y y
ββ ββ= = = = = = , and then the
proof is complete. Theorem 3.3 If ,A C B D= = , the equilibrium point 0,0οΌ οΌ of the systems (1.6) is a global attractor for any initial conditions
83 2 1 0 3 2 1 0( , , , , , , , ) (0, )x x x x y y y y Aβ β β β β β β .
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
972 Chang-you Wang et al 968-976
6
Proof. Let 4 4( , ) : (0, ) (0, ) (0, ) (0, )f g β Γ β β β Γ β be a function defined by
1 23 2 1 0 3 2 1 0
3
( , , , , , , , ) n n
n
x xf x x x x y y y yA By
β ββ β β β β β
β
=+
,
and
1 23 2 1 0 3 2 1 0
3
( , , , , , , , ) n n
n
y yg x x x x y y y yA Bx
β ββ β β β β β
β
=+
.
We can easily see that the functions f and g possess a mixed monotone property in
subsets 8(0, )A of 8R . Let
00 0 3 2 1 0 3 2 1 0 0 0max , , , , , , , , 0M AM N x x x x y y y y n m
Bβ β β β β β
β= = < = < .
We have 2 20 0
0 00 0
,m Mm MA BN A Bn
β€ β€ β€+ +
(3.16)
2 20 0
0 00 0
,n Nn NA BM A Bm
β€ β€ β€+ +
(3.17)
Then from (1.6) and Theorem 3.2, there exit 0 0, [ , ]m M m Mβ , 0 0, [ , ]n N n Nβ
satisfying 2 2
, ,m Mm MA BN A Bn
= =+ +
(3.18)
2 2
, .n Nn NA BM A Bm
= =+ +
(3.19)
In view of
0 0m M M A Bn A Bn A BN< < < + < + < + ,
and
0 0n N N A Bm A Bm A BM< < < + < + < + ,
thus, one has 0M m N n= = = = . (3.20)
It follows by Theorem 3.2 that the equilibrium point 0,0οΌ οΌ of (1.6) is a global attractor. The proof is complete. Theorem 3.4 The equilibrium point 0,0οΌ οΌ of the system (1.6) is global asymptotically stability for any initial conditions
83 2 1 0 3 2 1 0( , , , , , , , ) (0, )x x x x y y y y Aβ β β β β β β .
Proof. The result follows from Theorems 3.1 and 3.3. Theorem 3.5 The equilibrium point (0, )C , ( , 0)A of the system (1.6) is unstable. Proof. We can easily obtain that the linearized system of the system (1.6) about the equilibrium (0, )C is
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
973 Chang-you Wang et al 968-976
7
*1n nDΟ Ο+ = , (3.21)
where
1
2
3 *
1
2
3
0 0 0 0 0 0 0 01 0 0 0 0 0 0 00 1 0 0 0 0 0 00 0 1 0 0 0 0 0
,0 0 0 - 0 1 1 00 0 0 0 1 0 0 00 0 0 0 0 1 0 00 0 0 0 0 0 1 0
n
n
n
nn
n
n
n
n
xxxx
Dy Dyyy
Ο
β
β
β
β
β
β
β‘ β€ β‘ β€β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯= =β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β£ β¦β£ β¦
The characteristic equation of the systems (3.20) is 5 3( )= ( .P Ξ» Ξ» Ξ» Ξ»β β1) (3.22)
It is obvious that (1) 1, (2) 160P P= β = . It follows by the intermediate value theorem for continuous function that there exists 1Ξ» > so that ( ) 0P Ξ» = . Therefore, one of the roots of characteristic equation (3.22) lies outside unit disk. According to Theorem 2.1, the equilibrium (0, )C is unstable.
Similarly, we can obtain that the unique equilibrium ( , 0)A is unstable.
Theorem 3.6 If 1BD < , the equilibrium point ( , ) ( , )1 1A BC C ADx y
BD BD+ +
=β β
is locally
asymptotically stable. If 1BD > , the equilibrium point ( , )x y is unstable. Proof. We can easily obtain that the linearized system of (1.6) about the equilibrium ( , )x y is
*1n nDΟ Ο+ = , (3.23)
where
1
2
3 *
1
2
3
0 0 0 0 0 0 0 -1 0 0 0 0 0 0 00 1 0 0 0 0 0 00 0 1 0 0 0 0 0
,0 0 0 - 0 1 1 00 0 0 0 1 0 0 00 0 0 0 0 1 0 00 0 0 0 0 0 1 0
n
n
n
nn
n
n
n
n
x Bxxx
Dy Dyyy
Ο
β
β
β
β
β
β
β‘ β€ β‘ β€β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯= =β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β’ β₯ β’ β₯β£ β¦β£ β¦
The characteristic equation of (3.23) is 8( ) 0.P BDΞ» Ξ»= β = (3.24)
In view of 1BD < , this shows that all the roots of characteristic equation lie inside unit disk, so the unique equilibrium ( , )x y is locally asymptotically stable. If 1BD > , one of the roots of characteristic equation lie outside unit disk, so the unique equilibrium
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
974 Chang-you Wang et al 968-976
8
( , )x y is unstable.
4. Conclusions This paper presents the use of a variational iteration method for systems of nonlinear difference equations. This technique is a powerful tool for solving various difference equations and can also be applied to other nonlinear differential equations in mathematical physics. The variational iteration method provides an efficient method to handle the nonlinear structure. We have dealt with the problem of global asymptotic stability analysis for a class of nonlinear difference equations. The general sufficient conditions have been obtained to ensure the existence, uniqueness and global asymptotic stability of the equilibrium point (0,0) for the nonlinear difference equation. These criteria generalize and improve some known results in [12-17]. Moreover, the asymptotic behavior of others equilibrium points is also studied. In addition, the sufficient conditions that we obtained are very simple, which provide flexibility for the application and analysis of nonlinear difference equation.
Remark: Our model and results are different from the existence ones such as those of References [12-17]. In particular, the new variational iteration method can be applied to the models of References [12-17] and the more general nonlinear difference equations. In some sense, we enrich the theoretical results of the difference equations.
Acknowledgements This work is supported by Science Fund for Distinguished Young Scholars (cstc2014jc yjjq40004) of China, the National Nature Science Fund (Project nos.11372366 and 61503053) of China, the Science and Technology Project of Chongqing Municipal Education Committee (Grants no. kj1400423) of China, and the excellent talents project of colleges and universities in Chongqing of China.
References 1. Elsayed EM, Iricanin BD: On a max-type and a min-type difference equation. Appl.
Math. Comput. 215, 608-614 (2009) 2. Li WT, Zhang YH, Su YH: Global attractivity in a higher-order nonlinear difference
equation. Acta Mathematica Scientia 25, 59-66 (2005)
3. Yang ZG, Xu, DY: Mean square exponential stability of impulsive stochastic difference equations. Appl. Math. Lett. 20, 938-945 (2007)
4. Jia XM, Hu LX, Li WT: Dynamics of a rational difference equation. Advances in Difference Equations. 2010, Article ID 970720 (2010)
5. Li WT, Sun HR: Global attractivity in a rational recursive sequence. Dynamic Systems and Applications 11, 339-345 (2002)
6. Elaydi S: An Introduction to Difference Equations, third ed., Springer, New York
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
975 Chang-you Wang et al 968-976
9
(2005) 7. Muroya Y, Ishiwata E, Guglielmi N: Global stability for nonlinear difference equations
with variable coefficients. J. Math. Anal. Appl. 334, 232-247 (2007) 8. SteviΔ S: Solutions of a max-type system of difference equations. Appl. Math. Comput.
218, 9825-9830 (2012) 9. Papaschinopoulos G, Radin M, Schinas CJ: Study of the asymptotic behavior of the
solutions of three systems of difference equations of exponential form. Appl. Math. Comput. 218, 5310-5318 (2012)
10. Migda J: Asymptotically polynomial solutions of difference equations, Advances in Difference Equations 2013, 92 (2013)
11. Hu LX, He WS, Xia HM: Global asymptotic behavior of a rational difference equation. Appl. Math. Comput. 218, 7818-7828 (2012)
12. Elsayed EM: Solutions of rational difference systems of order two. Mathematical and Computer Modelling 55, 378-384 (2012)
13. Clark D, Kulenovic MRS: A coupled system of rational difference equations. Computers and Mathematics with Applications 43, 849-867(2002)
14. Clark D, Kulenovic MRS, Selgrade JF: Global asymptotic behavior of a two- dimensional difference equation modelling competition. Nonlinear Anal. 52, 1765- 1776 (2003)
15. Zhang QH, Yang LH, Liu JZ: Dynamics of a system of rational third-order difference equation. Advances in Difference Equations 2012, 136 (2012)
16. Liu WP, Yang XF, Liu XZ: Dynamics of a family of two-dimensional difference systems. Appl. Math. Comput. 219, 5949-5955 (2013)
17. Papaschinopoulos G, Ellina G, Papadopoulos KB: Asymptotic behavior of the positive solutions of an exponential type system of difference equations. Appl. Math. Comput. 245, 181-190 (2014)
18. Elsayed EM: Behavior and Expression of the Solutions of Some Rational Difference Equations. Journal of Computational Analysis and Applications 15, 73-81 (2013)
19. Zhou LQ: Global asymptotic stability of cellular neural networks with proportional delays. Nonlinear Dyn. 77, 41-47(2014)
20. Elsayed EM, Cinar C: On the solutions of some systems of difference equations. Utilitas Mathematica 93, 279-289 (2014)
21. Hien LV: A novel approach to exponential stability of nonlinear non-autonomous difference equations with variable delays. Appl. Math. Lett. 38, 7-13 (2014)
22. Kulenovic MRS, Pilav E, Silic E: Local dynamics and global attractivity of a certain second-order quadratic fractional difference equation. Advances in Difference Equations 2014, 68 (2014)
J. COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO.5, 2017, COPYRIGHT 2017 EUDOXUS PRESS, LLC
976 Chang-you Wang et al 968-976
TABLE OF CONTENTS, JOURNAL OF COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO. 5, 2017
Higher-Order Degenerate Bernoulli Polynomials, Dae San Kim, and Taekyun Kim,β¦β¦β¦789
Korobov Polynomials of the Seventh Kind and of the Eighth Kind, Dae San Kim, Taekyun Kim, Toufik Mansour, and Jong-Jin Seo,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦.812
Some Identities on the Higher-Order Twisted q-Euler Numbers and Polynomials, C. S. Ryoo,825
Umbral Calculus Associated With New Degenerate Bernoulli Polynomials, Dae San Kim, Taekyun Kim, and Jong-Jin Seo,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦831
Regularization Smoothing Approximation of Fuzzy Parametric Variational Inequality Constrained Stochastic Optimization, Heng-you Lan,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦841
The Split Common Fixed Point Problem for Demicontractive Mappings in Banach Spaces, Li Yang, Fuhai Zhao, and Jong Kyu Kim,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦.858
Iterated Binomial Transform of the k-Lucas Sequence, Nazmiye Yilmaz and Necati Taskara,864
Nielsen Fixed Point Theory for Digital Images, Ozgur Ege, and Ismet Karaca,β¦β¦β¦β¦β¦..874
A Fixed Point Theorem and Stability of Additive-Cubic Functional Equations in Modular Spaces, Chang Il Kim, Giljun Han, and Seong-A Shim,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦881
Results on Value-Shared of Admissible Function and Non-Admissible Function in the Unit Disc, Hong-Yan Xu,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦894
Compositions Involving Schur Harmonically Convex Functions, Huan-Nan Shi, and Jing Zhang,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦907
A Note on Degenerate Generalized q-Genocchi Polynomials, Jongkyum Kwon, Jin-Woo Park, and Sang Jo Yun,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦..923
Cubic Soft Ideals in BCK/BCI-Algebras, Young Bae Jun, G. Muhiuddin, Mehmet Ali Ozturk, and Eun Hwan Roh,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦.929
Hyers-Ulam Stability of the Delayed Homogeneous Matrix Difference Equation with Constructive Method, Soon-Mo Jung, and Young Woo Nam,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦941
Mathematical Analysis of (n + 3)-Dimensional Virus Dynamics Model, A. M. Elaiw, and N. H. AlShamrani,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦..949
TABLE OF CONTENTS, JOURNAL OF COMPUTATIONAL ANALYSIS AND APPLICATIONS, VOL. 22, NO. 5, 2017
(continued)
On the Dynamics of a Certain Four-Order Fractional Difference Equations, Chang-you Wang, Xiao-jing Fang, and Rui Li,β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦β¦968