Mathematica Vol. 10, No. 2, 2018

Page 1


Acta Universitatis Sapientiae The scientific journal of Sapientia Hungarian University of Transylvania (Cluj-Napoca, Romania) publishes original papers and surveys in several areas of sciences written in English. Information about the appropriate series can be found at the Internet address http://www.acta.sapientia.ro. Editor-in-Chief László DÁVID Main Editorial Board Zoltán KÁSA Laura NISTOR

András KELEMEN

Ágnes PETHŐ Emőd VERESS

Acta Universitatis Sapientiae, Mathematica Executive Editor Róbert SZÁSZ (Sapientia Hungarian University of Transylvania, Romania) Editorial Board Sébastien FERENCZI (Institut de Mathématiques de Luminy, France) Kálmán GYŐRY (University of Debrecen, Hungary) Zoltán MAKÓ (Sapientia Hungarian University of Transylvania, Romania) Ladislav MIŠÍK (University of Ostrava, Czech Republic) János TÓTH (Selye University, Slovakia) Adrian PETRUŞEL (Babeş-Bolyai University, Romania) Alexandru HORVÁTH (Sapientia Hungarian University of Transylvania, Romania) Árpád BARICZ (Babeş-Bolyai University, Romania) Csaba SZÁNTÓ (Babeş-Bolyai University, Romania) Szilárd ANDRÁS (Babeş-Bolyai University, Romania) Assistant Editor Pál KUPÁN (Sapientia Hungarian University of Transylvania, Romania) Contact address and subscription: Acta Universitatis Sapientiae, Mathematica Sapientia Hungarian University of Transylvania RO 400112 Cluj-Napoca Str. Matei Corvin nr. 4. Email: acta-math@acta.sapientia.ro Each volume contains two issues.

Sapientia University

Scientia Publishing House

ISSN 1844-6094 http://www.acta.sapientia.ro


Acta Universitatis Sapientiae

Mathematica Volume 10, Number 2, 2018

Sapientia Hungarian University of Transylvania Scientia Publishing House



Contents

L. Bognár, A. Joós, B. Nagy An improvement for a mathematical model for distributed vulnerability assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 A. A. Bouchentouf, A. Guendouzi, A. Kandouci Performance and economic analysis of Markovian Bernoulli feedback queueing system with vacations, waiting server and impatient customers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218 C. Carpintero, N. Rajesh, E. Rosas On real valued ω-continuous functions . . . . . . . . . . . . . . . . . . . . . . . . 242 K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan Rejection sampling of bipartite graphs with given degree sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249 S. K. Mohapatra, T. Panigrahi Some sufficient conditions for certain class of meromorphic multivalent functions nvolving Cho-Kwon-Srivastava operator . 276 F. Qi, A.-Q. Liu Alternative proofs of some formulas for two tridiagonal determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287 S. Pirzada, M. Imran Computing metric dimension of compressed zero divisor graphs associated to rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298 E. Rosas, C. Carpintero, J. Sanabria Θ-modifications on weak spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319

201


P. Sahoo, H. Karmakar Uniqueness theorems related to weighted sharing of two sets . . 329 Abdullah, F. A. Shah Scaling functions on the spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340 A. Sofo Integrals of polylogarithmic functions with negative argument 347 M. Şan, H. Irmak A note on some relations between certain inequalities and normalized analytic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368 M. Tărnăuceanu Finite groups with a certain number of cyclic subgroups II . . . . 375 M. Zákány New classes of local almost contractions . . . . . . . . . . . . . . . . . . . . . . 378 M. Y. Yilmaz, M. Bektaş Slant helices of (k, m)-type in E4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395 E. Hashemi, M. Yazdanfar, A. Alhevaz On extensions of Baer and quasi-Baer modules . . . . . . . . . . . . . . . 402 Contents of volume 10, 2018 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418

202


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 203–217 DOI: 10.2478/ausm-2018-0017

An improvement for a mathematical model for distributed vulnerability assessment László Bognár

Antal Joós

University of Dunaújváros, Hungary email: bognarl@uniduna.hu

University of Dunaújváros, Hungary email: joosanti@gmail.com

Bálint Nagy University of Dunaújváros, Hungary email: nagyb@uniduna.hu

Abstract. Hadarics et. al. gave a Mathematical Model for Distributed Vulnerability Assessment. In this model the extent of vulnerability of a specific company IT infrastructure is measured by the probability of at least one successful malware attack when the users behaviour is also incorporated into the model. The different attacks are taken as independent random experiments and the probability is calculated accordingly. The model uses some input probabilities related to the characteristics of the different threats, protections and user behaviours which are estimated by the corresponding relative frequencies. In this paper this model is further detailed, improved and a numerical example is also presented.

1

Introduction

In recent decades information and infocommunication devices have become widely used. Besides their advantages previously unknown threats and malicious codes [8], [9] appeared. Traditionally measuring cyber risk usually consist of testing malicious activity [3] and penetration testing [10], [1]. Information can be obtained from the traffic of the network hence interactive metrics can 2010 Mathematics Subject Classification: 60A99, 94C99 Key words and phrases: vulnerability, probability, relative frequency

203


204

L. BognaĚ r et al.

be evolved [5],[2], [7]. The behaviour of the users is usually regarded as a factor of secondary importance which can result in a model not adequately representing real life situations. In an adequate model for assessing vulnerability of a specific business all three factors should be considered: 1. Malicious activity from the outher world threatening the IT network of the business. 2. Not properly protected elements of the IT network at the business. 3. Dangerous behaviours of users inside the business.

2

The model

Most of the notation of [4] will be used. For completeness these notations are to be reviewed. Let L{l1 , . . . , lτ } be the set of all available threat landscapes. In what follows a specific landscape will be used denoted by l. Let Tall be the set of all possible malware. Let T = {t1 , . . . , tk } be the set of all possible malware inside l. Let U = {u1 , . . . , ur } be the set of all users. Let D = {d1 , . . . , dm } be the set of all possible devices inside l. Let P = {p1 , . . . , pn } be the set of all available protections inside l. Let UT = {ut1 , . . . , uti } be the set of all possible user tricks used by any malware inside l. An integrated measure of vulnerability accounting for all three sources (attacker ingenuity, infrastructure weakness and adverse user behaviour) can be constructed. For any given threat or class of threats for which the requisite IT infrastructure vulnerability and user facilitation is known, we can obtain a best estimate of: 1. The probability that an attacker will use a particular threat or class of threats against the enterprise (pprev ). The probability pprev is estimated by pprev (t, l) =

number of computers infected by t inside l number of all computers inside l

for t ∈ T . Note, that pprev can be based on a measurement or estimation and must be related to a time interval. Let Pprev =

t1 t2 ... tk p pprev (t1 ) pprev (t2 ) . . . pprev (tk )


Distributed vulnerability assessment

205

be a vector. This means if we examine a particular attack, then the probability that this attack is in the form of the threat t1 is pprev (t1 ), etc. 2. The probability that the enterprise’s IT infrastructure will allow the attack to be carried out successfully (pdevice ). To elaborate the estimation of pdevice first some auxiliary probabilities are defined and estimated. The probability pprot (t, p) is introduced pprot (t, p) =

number of successful attempts of t through the protection p number of all attempts of t through the protection p

for any t ∈ T and p ∈ P. Let

Pprot

p1 p2 t1 pprot (t1 , p1 ) pprot (t1 , p2 ) = t2 pprot (t2 , p1 ) pprot (t2 , p2 ) .. .. .. . . .

... pn . . . pprot (t1 , pn ) . . . pprot (t2 , pn ) .. ... .

tk pprot (tk , p1 ) pprot (tk , p2 ) . . . pprot (tk , pn ) be a k × n matrix. This means that the probability of a successful attempt of t1 through the protection p1 is pprot (t1 , p1 ), etc. The value zdevice−elements (d, t) is introduced

1 if t can work on d zdevice−elements (d, t) = 0 if t can not work on d (or shortly zdev−elem (d, t)) for any t ∈ T and d ∈ D. Let Zdevice−elements d1 = d2 .. .

t1

t2

zdev−elem (d1 , t1 ) zdev−elem (d2 , t1 ) .. .

zdev−elem (d1 , t2 ) zdev−elem (d2 , t2 ) .. .

... ... ...

tk zdev−elem (d1 , tk ) zdev−elem (d2 , tk ) .. .

... dm zdev−elem (dm , t1 ) zdev−elem (dm , t2 ) . . . zdev−elem (dm , tk )

be an m × k matrix. The value zdevice−prot−install (d, p) is introduced

1 if d does not have the protection p zdevice−prot−install (d, p) = 0 if d has the protection p


206

L. Bognár et al.

(or shortly zd−p−i ) for any d ∈ D and p ∈ P. Let Zdevice−prot−install d1 = d2 .. .

p1 zd−p−i (d1 , p1 ) zd−p−i (d2 , p1 ) .. .

p2 zd−p−i (d1 , p2 ) zd−p−i (d2 , p2 ) .. .

... ... ...

pn zd−p−i (d1 , pn ) zd−p−i (d2 , pn ) .. .

... dm zd−p−i (dm , p1 ) zd−p−i (dm , p2 ) . . . zd−p−i (dm , pn )

be an m × n matrix. Let Pdevice−prot−install−dj p1

t1 t = 2 .. .

p2

... pn pd−p−i−dj (t1 , p1 ) pd−p−i−dj (t1 , p2 ) . . . pd−p−i−dj (t1 , pn ) pd−p−i−dj (t2 , p1 ) pd−p−i−dj (t2 , p2 ) . . . pd−p−i−dj (t2 , pn ) .. .. .. . . ... .

tk pd−p−i−dj (tk , p1 ) pd−p−i−dj (tk , p2 ) . . . pd−p−i−dj (tk , pn ) be a k × n matrix where pd−p−i−dj (tx , py ) = max{pprot (tx , py ), zd−p−i (dj , py )} for any j ∈ {1, . . . , m}, x ∈ {1, . . . , k} and y ∈ {1, . . . , n}. This means that if the threat t1 can work on dj , then the probability of a successful attempts of the threat t1 through the protection p1 on the device dj is pd−p−i−dj (t1 , p1 ), etc. The probability pdevice−prot−dj (t) is introduced pdevice−prot−dj (t) =

for all

p

min protecting

pprot (t, p) dj

for any t ∈ T . Let p Pdevice−prot−dj

t1 = t2 .. .

pdevice−prot−dj (t1 ) pdevice−prot−dj (t2 ) .. .

tk pdevice−prot−dj (tk ) be the column vector where pdevice−prot−dj (tx )


Distributed vulnerability assessment

207

= min{pd−p−i−dj (tx , p1 ), pd−p−i−dj (tx , p2 ), . . . , pd−p−i−dj (tx , pn )} for any j ∈ {1, . . . , m} and x ∈ {1, . . . , k}. This means that if the threat t1 can work on dj , then the probability of a successful attempts of the threat t1 through any protection protecting the device dj is pdevice−prot−dj (t1 ), etc. The probability pdevice−prot (d, t) is introduced pdevice−prot (d, t) =

for all

p

min pprot (t, p) protecting d

for any t ∈ T and d ∈ D. Let Pdevice−prot d1 = d2 .. .

t1

t2

pdevice−prot (d1 , t1 ) pdevice−prot (d2 , t1 ) .. .

pdevice−prot (d1 , t2 ) pdevice−prot (d2 , t2 ) .. .

... ... ...

tk pdevice−prot (d1 , tk ) pdevice−prot (d2 , tk ) .. .

... dm pdevice−prot (dm , t1 ) pdevice−prot (dm , t2 ) . . . pdevice−prot (dm , tk )

be an m × k matrix where pdevice−prot (dx , ty ) = pdevice−prot−dx (ty ) for any x ∈ {1, . . . , m} and y ∈ {1, . . . , k}. The probability pdevice (d, t) is introduced pdevice (d, t) = zdecive−elements (d, t) · pdevice−prot (d, t) for any t ∈ T and d ∈ D. Let

Pdevice

d1 = d2 .. .

t1 pdevice (d1 , t1 ) pdevice (d2 , t1 ) .. .

t2 pdevice (d1 , t2 ) pdevice (d2 , t2 ) .. .

... ... ...

tk pdevice (d1 , tk ) pdevice (d2 , tk ) .. .

... dm pdevice (dm , t1 ) pdevice (dm , t2 ) . . . pdevice (dm , tk )

be an m × k matrix where pdevice (dx , ty ) = zdev−elem (dx , ty ) · pdevice−prot (dx , ty ) for any x ∈ {1, . . . , m} and y ∈ {1, . . . , k}. This means that the probability of a successful attempts of the threat t1 through any protection protecting the device d1 is pdevice (d1 , t1 ), etc.


208

L. Bognár et al.

3. The probability that users of the enterprise’s IT infrastructure will provide sufficient facilitation for the attack to succeed (puser ). The pusertrick (t, ut) probability is introduced pusertrick (t, ut) =

number of attempts of t where t used ut number of all attempts of t

for any t ∈ T and ut ∈ UT . Let Pusertrick

t1 = t2 .. .

ut1 ut2 pusertrick (t1 , ut1 ) pusertrick (t1 , ut2 ) pusertrick (t2 , ut1 ) pusertrick (t2 , ut2 ) .. .. . .

... uti . . . pusertrick (t1 , uti ) . . . pusertrick (t2 , uti ) .. . ...

tk pusertrick (tk , ut1 ) pusertrick (tk , ut2 ) . . . pusertrick (tk , uti ) be a k × i matrix. This means that the probability that the threat t1 uses usertrick ut1 is pusertrick (t1 , ut1 ), etc. The puser−usertrick (u, ut) probability is introduced puser−usertrick (u, ut) =

number of successful attempts of ut on u number of all attempts of ut on u

(or shortly pu−utrick (u, ut)) for any u ∈ U and ut ∈ UT . Let Puser−usertrick ut1 ut2 u1 pu−utrick (u1 , ut1 ) pu−utrick (u1 , ut2 ) = u2 pu−utrick (u2 , ut1 ) pu−utrick (u2 , ut2 ) .. .. .. . . . ur

... uti . . . pu−utrick (u1 , uti ) . . . pu−utrick (u2 , uti ) .. ... .

pu−utrick (ur , ut1 ) pu−utrick (ur , ut2 ) . . . pu−utrick (ur , uti )

be an r × i matrix. This means that the probability that the user u1 uses usertrick ut1 is pu−utrick (u1 , ut1 ), etc. From the probabilities pusertrick and puser−usertrick we can calculate the probability puser (u, t) which is the probability that the threat t infects using at least one usertrick through the user u. This is puser (u, t)


Distributed vulnerability assessment Y

=1− for all

ut

209

(1 − pusertrick (t, ut) · puser−usertrick (u, ut))

used by

t

for any u ∈ U, t ∈ T and ut ∈ UT . Let

Puser

t1 t2 u1 puser (u1 , t1 ) puser (u1 , t2 ) = u2 puser (u2 , t1 ) puser (u2 , t2 ) .. .. .. . . . ur

... tk . . . puser (u1 , tk ) . . . puser (u2 , tk ) .. ... .

puser (ur , t1 ) puser (ur , t2 ) . . . puser (ur , tk )

be an r × k matrix where puser (u1 , t1 ) = 1 − (1 − pusertrick (t1 , ut1 ) · pu−utrick (u1 , ut1 )) ·(1 − pusertrick (t1 , ut2 ) · pu−utrick (u1 , ut2 )) · . . . ·(1 − pusertrick (t1 , uti ) · pu−utrick (u1 , uti )), etc. This means that the probability that the threat t1 infects using at least one usertrick through the user u1 is puser (u1 , t1 ), etc.

2.1

The probability of infection

These three probabilities (pprev , pdevice , puser ) can be combined to obtain an overall probability of malicious success, (provided each relevant combination of attack, user, and component of IT infrastructure is accounted for) [6]. The (pprev , pdevice , puser ) values are related to a given threat, a given user and a given device. The aggregated vulnerability would be an index of the whole organization related to all of the users, all of the devices and all of the possible threats. The probability of the infection is ps which is the probability that the investigated landscape will be infected by at least one malware. This can be calculated in the following form Y (1 − puser (t, u) · pdevice (t, d) · pprev (t, l)) ps = 1 − for all t,u and d for any u ∈ U, t ∈ T and d ∈ D. The followings were assumed: 1. the attacker usage of the given threat, the IT infrastructure allowance and the user acceptance are different from each other,


210

L. BognaĚ r et al.

2. all of the attack attempts are independent from each other, 3. the computer usage behaviours of all users are the same and equal to the average usage in the organization. Observe the calculated ps value is related to the same time interval as the original pprev was related to.

3

A numerical example

Let T = {t1 , . . . , t4 } be the set of malware. Let U = {u1 , . . . , u7 } be the set of all users. Let D = {d1 , d2 , d3 } be the set of all devices. Let P = {p1 , . . . , p5 } be the set of all protections. Let UT = {ut1 , . . . , ut6 } be the set of all user tricks used by any malware in T . Let Pprev =

t1 t2 t3 t4 0.25 0.25 0.25 0.25

and

Pprot

t1 = t2 t3 t4

p1 0.01 0.11 0.21 0.31

p2 0.02 0.12 0.22 0.32

p3 0.03 0.13 0.23 0.33

p4 0.04 0.14 0.24 0.34

This means that the probability of a successful protection p1 is 0.01, etc. Let t1 t2 d1 1 0 Zdevice elements = d2 0 1 d3 0 1

p5 0.02 0.15 . 0.25 0.35

attempt of t1 through the t3 t4 0 0 . 1 0 0 1

This means that t1 can work on d1 , t2 can not work Let p1 p2 p3 d 1 0 1 Zdevice prot install = 1 d2 0 1 1 d3 1 0 0

on d1 , etc. p4 p5 0 1 . 0 1 1 1

This means that d1 does not have the protection p1 , d1 has the protection p2 , etc.


Distributed vulnerability assessment

211

Thus t1 Pprot install d1 = t2 t3 t4

p1 1 1 1 1

p2 p3 0.02 1 0.12 1 0.22 1 0.32 1

p4 p5 0.04 1 0.14 1 . 0.24 1 0.34 1

Observe pd−p−i−d1 (t1 , p1 ) = max{pprot (t1 , p1 ), zd−p−i (d1 , p1 )} = max{0.01, 1} = 1, pd−p−i−d1 (t1 , p2 ) = max{pprot (t1 , p2 ), zd−p−i (d1 , p2 )} = max{0.02, 0} = 0.02, etc. This means that the probability of a successful attempts of the threat t1 through the protection p1 on the device d1 is pd−p−i−d1 (t1 , p1 ), etc. Similarly

Pprot intall d2

Pprot intall d3

t1 = t2 t3 t4

p1 p2 p3 0.01 1 1 0.11 1 1 0.21 1 1 0.31 1 1

t1 = t2 t3 t4

p1 1 1 1 1

p2 0.02 0.12 0.22 0.32

p4 p5 0.04 1 0.14 1 , 0.24 1 0.34 1

p3 p4 p5 0.03 1 1 0.13 1 1 . 0.23 1 1 0.33 1 1

Furthermore

Pdevice prot D1

t1 = t2 t3 t4

P 0.02 0.12 . 0.22 0.32

Observe pdevice−prot−d1 (t1 ) = min{pd−p−i−d1 (t1 , p1 ), pd−p−i−d1 (t1 , p2 ), . . . , pd−p−i−d1 (t1 , pn )} min{1, 0.02, 0.03, 1, 1} = 0.02. This means that if the threat t1 can work on d1 , then the probability of a successful attempts of the threat t1 through any protection protecting the


212

L. Bognár et al.

device d1 is 0.02, etc. Similarly

Pdevice−prot−d2

Pdevice−prot−d3

t1 = t2 t3 t4

P 0.01 0.11 , 0.21 0.31

t1 = t2 t3 t4

P 0.02 0.12 . 0.22 0.32

Thus Pdevice−prot =

d1 d2 d3

t1 t2 t3 t4 0.02 0.12 0.22 0.32 . 0.01 0.11 0.21 0.31 0.02 0.12 0.22 0.32

Observe pdevice−prot (d1 , t1 ) = pdevice−prot−d1 (t1 ), pdevice−prot (d1 , t2 ) = pdevice−prot−d1 (t2 ), etc. This means that if the threat t1 can work on d1 , then the probability of a successful attempts of the threat t1 through any protection protecting the device d1 is 0.02, etc. Furthermore

Pdevice

t1 t2 t3 t4 d1 0.02 0 0 0 = . d2 0 0.11 0.21 0 d3 0 0.12 0 0.32

Observe pdevice (d1 , t1 ) = zdev−elem (d1 , t1 ) · pdevice−prot (d1 , t1 ) = 0.02 · 1 = 0.02, pdevice (d1 , t2 ) = zdev−elem (d1 , t2 ) · pdevice−prot (d1 , t2 ) = 0.12 · 0 = 0, etc. This means that the probability of a successful attempts of the threat t1 through any protection protecting the device d1 is 0.02. Since t2 can not work


Distributed vulnerability assessment

213

on d1 , the probability of a successful attempts of the threat t2 through any protection protecting the device d1 is 0, etc. Let

Pusertrick

ut1 0.141 0.151 0.161 0.171

t1 = t2 t3 t4

ut2 0.142 0.152 0.162 0.172

ut3 0.143 0.153 0.163 0.173

ut4 0.144 0.154 0.164 0.174

ut5 0.145 0.155 0.165 0.175

ut6 0.146 0.156 . 0.166 0.176

This means that the probability that the threat t1 uses usertrick ut1 is 0.141, etc. Observe the sum of the probabilities in any row is not greater than 1. Let

Puser usertrick

u1 u2 u = 3 u4 u5 u6 u7

ut1 0.031 0.041 0.051 0.061 0.071 0.081 0.091

ut2 0.032 0.042 0.052 0.062 0.072 0.082 0.092

ut3 0.033 0.043 0.053 0.063 0.073 0.083 0.093

ut4 0.034 0.044 0.054 0.064 0.074 0.084 0.094

ut5 0.035 0.045 0.055 0.065 0.075 0.085 0.095

ut6 0.036 0.046 0.056 . 0.066 0.076 0.086 0.096

This means that the probability that the user u1 uses usertrick ut1 is 0.031, etc. Thus

Puser

u1 u2 u = 3 u4 u5 u6 u7

t1 0.028516 0.036891 0.045206 0.053460 0.061655 0.069791 0.077868

t2 0.030477 0.039418 0.048290 0.057094 0.065830 0.074498 0.083099

t3 0.032434 0.041939 0.051366 0.060716 0.069989 0.079185 0.088305

t4 0.034388 0.044455 0.054434 . 0.064326 0.074132 0.083852 0.093487

Observe puser (u1 , t1 ) = 1 − (1 − pusertrick (t1 , ut1 ) · pu−utrick (u1 , ut1 )) · (1 − pusertrick (t1 , ut2 ) · pu−utrick (u1 , ut2 )) · . . . · (1 − pusertrick (t1 , uti ) · pu−utrick (u1 , uti )) = 1 − (1 − 0.141 · 0.031) · (1 − 0.142 · 0.032) · . . . · (1 − 0.146 · 0.036) = 0.028516,


214

L. Bognár et al.

etc. Therefore, ps = 1 − (1 − puser (t1 , u1 ) · pdevice (t1 , d1 ) · pprev (t1 )) · (1 − puser (t1 , u2 ) · pdevice (t1 , d1 ) · pprev (t1 )) · . . . · (1 − puser (t4 , u7 ) · pdevice (t4 , d3 ) · pprev (t4 )) = 1 − (1 − 0.028516 · 0.02 · 0.25) · (1 − 0.036891 · 0.02 · 0.25) · . . . · (1 − 0.093487 · 0.32 · 0.25) = 0.079774. This means that the probability of the infection of the investigated company with users u1 , . . . , u7 , devices d1 , d2 , d3 , protections p1 , . . . , p5 and matrices as above is 0.079774. Thus we get that the probability of an infection by at least one malware is 0.079774.

4

Simulations

In this section results of simulation studies are presented. Businesses with different sizes (different number of devices and users) are modelled and the ps probabilities are calculated when certain number of threats are present. The results are summarized in Table 1 and Table 2. The Micro (Small, Medium, Big, resp.) business is a company (or department) with 10 (50, 100, 1000, resp.) devices and 10 (50, 100, 1000, resp.) users. In real life the probabilities pprev , pprot , pusertrick and puser−usertrick can be estimated by relative frequencies but in the simulations these were estimated by random uniform probabilities. In the Table 1 the probabilities pprev (pprot , pusertrick , puser−usertrick , resp.) are in the interval [0.9, 1] ([0, 0.1], [0, 0.1], [0, 0.1], resp.). The results in the Table 1 correspond to the case when the number of protections is 5 and the number of usertrick is 5. The probability 0.25 in the cell of the third row of the second column in Table 1 means that the approximate probability of ps is 0.25 if there are 10 devices, 10 users in the company, the number of threats is 10, the number of protections is 5, the number of usertricks is 5, the random elements of the vector Pprev lie on the interval [0.9, 1], the random elements of the matrix Pprot lie on the interval [0, 0.1], the random elements of the matrix Pusertrick lie on the interval [0, 0.1] and the random elements of the matrix Puser−usertrick lie on the interval [0, 0.1]. Of course the matrices Zdevice−elements and Zdevice−prot−install are random matrices with elements 0 or 1. Observe that if the number of the devices (or users) or the number of the threats is large, then the probability is close to 1.


Distributed vulnerability assessment

215

Table 1: The values of ps probabilities in case of different business sizes

10

Micro devices=10 users=10 0.25

50

0.75

threats

100 1000

0.85 0.999999 99715744

Small devices=50 users=50 0.999935 91999547 0.999999 99996973 1 1

Medium devices=100 users=100 1

Big devices=1000 users=1000 1

1

1

1 1

1 1

The probabilities in Table 1 can be regarded as overestimates of the real ps probabilities since the sum of the elements in the random vector Pprev is greater than 1. In the Table 2 the probabilities pprev (pprot , pusertrick , puser−usertrick , resp.) are in the interval [0, 0.1] ([0, 0.1], [0, 0.1], [0, 0.1], resp.). The results in the Table 2 correspond to the case when the number of protections is 5 and the number of usertrick is 5. Table 2: The values of ps probabilities in case of different business sizes

10 50

Micro devices=10 users=10 0.02 0.07

100

0.15

1000

0.7

threats

Small devices=50 users=50 0.25 0.85 0.996973 10258718 0.999999 99998364

Medium devices=100 users=100 75 0.9986016 7849174 0.999999 99963790 1

Big devices=1000 users=1000 1 1 1 1

The difference between the Table 1 and Table 2 is the input random data pprev .


216

5

L. Bognár et al.

Conclusions

From the simulation studies it can be seen that the model presented can be used for defining an index number reflecting the state of vulnerability of a certain company against cyber attacks. However these simulations also show that this model has constraints of applicability because if the size of the company is big enough, then the probability ps is very close to 1 and no distinction can be made between the vulnerability of different companies. To overcome these constrains of the applicability it can be used either only to a smaller part of a large network or to a randomly selected smaller sample of users and devices. This index can be a good measuring tool of comparing the vulnerability of different parts of a company or comparing the state of vulnerability of a company at different time instances. Comparing different user behaviours can give valuable pieces of information for the company managements about the needs of improving employees awareness against cyber attacks.

References [1] M. T. Chapman, Establishing metrics to manage the human layer, ISSA Security Education Awareness Special Interest Group, 2013. [2] M. T. Chapman, Advanced Persistent Testing: How to fight bad phishing with good, PhishLine, 2015. http://www.phishline.com/advancedpersistent-testing-ebook [3] S. E. Edwards, R. Ford, G. Szappanos, Effectively testing APT defenses, Virus Bulletin Conference, Prague, Czech Republic, 2015. [4] K. Hadarics, K. Györffy, B. Nagy, L. Bognár, A. Arrott, F. Leitold, Mathematical Model of Distributed Vulnerability Assessment, In: Jaroslav Dočkal, Milan Jirsa, Josef Kaderka, Proceedings of Conference SPI 2017: Security and Protection of Information. Brno, 2017.07.01-2017.07.02. Brno: University of Defence, (2017), 45–57. (ISBN:978-80-7231-414-0) [5] F. Lalonde Levesque, J. M. Fernandez, A. Somayaji, Risk prediction of malware victimization based on user behavior, Malicious and Unwanted Software: The Americas (MALWARE), 2014 9th International Conference on. IEEE, 2014.


Distributed vulnerability assessment

217

[6] F. Leitold, A. Arrott, K. Hadarics, Quantifying cyber-threat vulnerability by combining threat intelligence, IT infrastructure weakness, and user susceptibility, 24th Annual EICAR Conference, Nuremberg, Germany, 2016. [7] F. Leitold, K. Hadarics, Measuring security risk in the cloud-enabled enterprise, In: Dr Fernando C Colon Osorio, 7th International Conference on Malicious and Unwanted Software (MALWARE), Fajardo, Puerto Rico, 2012.10.16-2012.10.18. Piscataway (NJ): IEEE, (2012), 62–66. (ISBN:9781-4673-4880-5) [8] NIST SP 800-53r4 Security and Privacy Controls for Federal Information Systems and Organizations, 2013. [9] NIST SP 800-83r1 Guide to Malware Incident Prevention and Handling for Desktops and Laptops, 2013. [10] Pwnie Express, Vulnerability assessment and penetration testing across the enterprise, Whitepaper, 2014. http://www.pwnieexpress.com

Received: January 7, 2018


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 218–241 DOI: 10.2478/ausm-2018-0018

Performance and economic analysis of Markovian Bernoulli feedback queueing system with vacations, waiting server and impatient customers Amina Angelika Bouchentouf Laboratory Mathematics, Djillali Liabes University of Sidi Bel Abbes, Algeria email: bouchentouf− amina@yahoo.fr

Abdelhak Guendouzi

Abdeldjebbar Kandouci

Laboratory of Stochastic Models, Statistic and Applications, Dr. Tahar Moulay University of Saida, Algeria email: a.guendouzi@yahoo.com

Laboratory of Stochastic Models, Statistic and Applications, Dr. Tahar Moulay University of Saida, Algeria email: kandouci1974@yahoo.fr

Abstract. This paper concerns the analysis of a Markovian queueing system with Bernoulli feedback, single vacation, waiting server and impatient customers. We suppose that whenever the system is empty the sever waits for a random amount of time before he leaves for a vacation. Moreover, the customer’s impatience timer depends on the states of the server. If the customer’s service has not been completed before the impatience timer expires, the customer leaves the system, and via certain mechanism, impatient customer may be retained in the system. We obtain explicit expressions for the steady-state probabilities of the queueing model, using the probability generating function (PGF). Further, we obtain some important performance measures of the system and formulate a cost model. Finally, an extensive numerical study is illustrated. 2010 Mathematics Subject Classification: 60K25, 68M20, 90B22 Key words and phrases: Markovian queueing models, vacations, impatience, Bernoulli feedback, waiting server, probability generating function, cost model

218


Vacation queueing model with waiting server and customers’ impatience 219

1

Introduction

Queueing models with vacations have a great impact in many real life situations, such models occur naturally in different fields such as computer and communication systems, flexible manufacturing systems, telephone services, production line systems, machine operating systems, post offices, etc. Over the past few decades, vacation queueing systems have paid attention of many researchers, excellent surveys on queueing systems with vacations can be found in Doshi [9] and Takagi [16] and in the monographs of Tian [17] and Ke [11]. In recent years, there has been gowning interest in the study of queueing systems with impatient customers (balking and reneging). For related literature, interested readers may refer to Shin and Choo [15], El-Paoumy and Nabwey [10], Kumar et al. [12], Kumar and Sharma [13], Bouchentouf et al. [7], Baek et al. [6], Bouchentouf and Messabihi [8] and references therein. The studies of queueing models with impatient customers were ranked depending on the causes of the impatience behavior. In queueing literature, models where customers may be impatient because of server vacations have been extensively analyzed. Yue [20] presented the optimal performance analysis of an M/M/1/N queueing system with balking, reneging and server vacation. Altman and Yechiali [2] gave the analysis of some queueing models such as M/M/1, M/G/1 and M/M/c queues with server vacations and customer impatience, both single and multiple vacation cases were studied. Further, Altman and Yechiali [3] investigated the infinite server queue with vacations and impatient customers. They obtained the probability generating function of the number of customers in the model and derived the performance measures of the system. Queueing systems with vacations and synchronized reneging have been done by Adan et al. [1]. Wu and Ke [19] presented computational algorithm and parameter optimization for a multi-server system with unreliable servers and impatient customers. Later, the model given in Altman and Yechiali [2] were extended by Yue et al. [21] by considering a variant of the multiple vacation policy which includes both single vacation and multiple vacations. In Padmavathy et al. [14], authors studied the steady state behavior of the vacation queues with impatient customers and a waiting server. Further, the transient solution of a M/M/1 multiple vacation queueing model with impatient customers has been investigated by Ammar [4]. Then, a study of single server Markovian queueing system with vacations and impatience timers which depend of the state of the server was presented in Yue et al. [22]. Recently, in Ammar [5], author established the transient solution of an M/M/1 vacation queue with a waiting server and impatient customers.


220

A. A. Bouchentouf, A. Guendouzi and A. Kandouci

The main objective of this article is to study an M/M/1 vacation queueing system with Bernoulli feedback, waiting server, reneging, and retention of reneged customers. It is supposed that whenever the busy period ended the server waits a random duration of time before beginning on a vacation. Moreover, we assume that the impatience timers of customers depend on the server’s states. We obtain the steady-state solution of the queueing model, using the probability generating function (PGF). Further, we give explicit expressions of useful measures of effectiveness and formulate a cost model. Then, we present a sensitive numerical experiments to illuminate the interests of our theoretical results and to show the impact of the diverse parameters on the behavior of the system. Finally, an appropriate economic analysis is carried out numerically. The model analyzed in this paper has a number of applications in practice. In most studies cited earlier, authors considered that the server leaves the system once the system is empty, but in many practical life situations the server waits a certain period of time before he leaves the system even if there is no customers, especially when we deal with a human behavior, examples can be found in post offices, banks, hospitals, etc. Further, our study has another great scope, in most studies mentioned in the above literature, the basis of the research is the supposition that customers may be impatient because of server vacations. However, there are many situations where the customer can become impatient due to the long wait in the queue even if the server is present in the system, another example when the customer may leave the system during busy period is when he cannot see the server state, these situations can be found in telecommunication systems, call centers and production inventory systems. The rest of the paper is organized in the following manner. In Section 2, we describe the model. In Section 3, we present the stationary analysis for the queueing model. In Section 4, we obtain different performance measures and formulate a cost model. Section 5 presents numerical results in the form of Tables and Figures. Finally, in Section 6 we conclude the paper.

2

System model

Consider a M/M/1 vacation queueing model with Bernoulli feedback, waiting server, reneging and retention of reneged customers. The model studied in this paper is based on following assumptions: ∗ Customers arrive into the system according to a Poisson process with


Vacation queueing model with waiting server and customers’ impatience 221 arrival rate λ, the service time is assumed to be exponentially distributed with parameter µ. The service discipline is FCFS and there is infinite space for customers to wait. ∗ When the busy period is finished the server waits a random duration of time before beginning on a vacation. This waiting duration is exponentially distributed with parameter η. ∗ If the server comes back from a vacation to an empty system he waits passively the first arrival, then he begins service. Otherwise, if there are customers waiting in the queue at the end of a vacation, the server starts immediately a busy period. That is single vacation policy. The period of vacation has an exponential distribution with parameter γ. ∗ Whenever a customer arrives at the system and finds the server on vacation (respectively. busy), he activates an impatience timer T0 (respectively. T1 ), which is exponentially distributed with parameter ξ0 (respectively. ξ1 ). If the customer’s service has not been completed before the impatience timer expires, the customer may abandon the queue. We suppose that the customers timers are independent and identically distributed random variables and independent of the number of waiting customers. ∗ Each reneged customer may leave the system without getting service with probability α and may be retained in the system with probability α 0 = (1−α). ∗ After completion of each service, the customer can either leave the system definitively with probability β or return to the system and join the end of the queue with probability β 0 , where β + β 0 = 1.

3

Stationary analysis

In this section, we use the probability generating function (PGF) to obtain the steady-state solution of the queueing system. Let L(t) be the number of customers in the system at time t, and J(t) denotes the state of the server at time t such that

1, when the server is in a busy period; J(t) = 0, otherwise. Clearly, the process {(L(t); J(t)); t ≥ 0} is a continuous-time Markov process with state space Ω = {(j, n) : j = 0, 1, n = 0, 1, ...}. Let Pj,n = lim P{J(t) = j, L(t) = n}, j = 0, 1, n = 0, 1, ..., (j, n) ∈ Ω, n→∞ denote the system state probabilities.


222

A. A. Bouchentouf, A. Guendouzi and A. Kandouci

Then, the steady-state balance equations of our model are given as follows: (λ + γ)P0,0 = αξ0 P0,1 + ηP1,0 ,

(1)

(λ + γ + nαξ0 )P0,n = λP0,n−1 + (n + 1)αξ0 P0,n+1 , n ≥ 1,

(2)

(λ + η)P1,0 = γP0,0 + (βµ + αξ1 )P1,1 ,

(3)

(λ + βµ + nαξ1 ) P1,n = λP1,n−1 + γP0,n + (βµ + (n + 1)αξ1 )P1,n+1 , (4) n ≥ 1, Theorem 1 If we have a single server Bernoulli feedback queueing system with single vacation, waiting server, server’s states-dependent reneging and retention of reneged customers, then 1. The steady-state probability P0,. is given by γαξ0 + δ1 K0 (1)(1 − γ) P0,. = P0,0 . (5) γK0 (1) 2. The steady-state probability P1,. is given by λ γ βµ η γ αξ1 K1 (1) + K2 (1) − K3 (1) P1,. = e λ + η αξ1 αξ1 αξ1 (6) βµ + αξ1 + λ+η

βµ η K1 (1) + K2 (1) αξ1 αξ1

αξ0 − δ1 K0 (1) δ2 K0 (1)

P0,0 ,

where

P0,0 =

λ δ1 δ2 K0 (1) + δ2 (αξ0 − δ1 K0 (1)) + e αξ1 γδ2 K0 (1)

γ + λ+η

βµ + αξ1 λ+η

αξ0 − δ1 K0 (1) δ2 K0 (1)

βµ η K1 (1) + K2 (1) αξ1 αξ1

γ αξ1 K3 (1)

−1 , (7)

Zz K0 (z) =

γ

(1 − s) αξ0 0

λ −1 − αξ s

e

0

ds,


Vacation queueing model with waiting server and customers’ impatience 223 Zz −1

K1 (z) =

s s

βµ αξ1

e

λs − αξ

1

Zz

0

and

βµ

(1 − s)−1 s αξ1 e

ds, K2 (z) =

λs − αξ

1

ds,

0

βµ Zz γ λ λ K0 (s) − αξ +1 − αξ s αξ1 αξ 0 1 1− K3 (z) = s (1 − s) e 0 ds. K (1) 0 0

Proof. Let

∞ X

Gj (z) =

Pj,n zn , j = 0, 1.

n=0

Then, multiplying Equation (2) by zn , using Equations (1) and (3) and summing all possible values of n, we get αξ0 (1 − z)G00 (z) − (λ(1 − z) + γ)G0 (z) = − {δ1 P00 + δ2 P11 } ,

(8)

with δ1 = where G00 (z) =

γη λ+η

and δ2 =

η(βµ + αξ1 ) λ+η

,

d dz G0 (z).

In the same manner, from Equations (3) and (4) we obtain αξ1 z(1−z)G10 (z)−(λz−βµ)(1−z)G1 (z) = −γzG0 (z)+(βµ(1−z)+ηz)P1,0 . (9) Next, let Γ = δ1 P00 + δ2 P11 . Then, for z 6= 1, Equation (8) can be rewritten as follows λ γ Γ 0 G0 (z) − + G0 (z) = − . (10) αξ0 αξ0 (1 − z) αξ0 (1 − z) γ

−λ

Multiplying both sides of Equation (10) by e αξ0 (1 − z) αξ0 , then integrating from 0 to z, we obtain

γ λ Γ z − αξ αξ0 G0 (z) = e (1 − z) 0 G0 (0) − K0 (z) , (11) αξ0 with

Zz K0 (z) =

γ

(1 − s) αξ0 0

λ −1 − αξ s

e

0

ds.

(12)


224

A. A. Bouchentouf, A. Guendouzi and A. Kandouci

Since G0 (1) =

∞ X

P0,n > 0 and z = 1 is the root of the denominator of the

n=0

right hand side of Equation (11), so z = 1 must be the root of the numerator of the right hand side of Equation (11). Thus, we get Γ K0 (1). (13) P0,0 = G0 (0) = αξ0 This implies P0,0 =

δ2 K0 (1) P1,1 . αξ0 − δ1 K0 (1)

(14)

P1,1 =

αξ0 − δ1 K0 (1) P0,0 . δ2 K0 (1)

(15)

Consequently

Next, substituting Equation (13) into (11), we obtain

γ λ K0 (z) z − αξ αξ0 G0 (z) = e (1 − z) 0 1 − P0,0 . K0 (1)

(16)

For z 6= 1 and z 6= 0, Equation (9) can be rewritten as follows λ βµ 0 G1 (z) − − G1 (z) αξ1 αξ1 z η γ βµ + P1,0 − G0 (z). = αξ1 z αξ1 (1 − z) αξ1 (1 − z) −

λ

z

(17)

βµ

Then, we multiply both sides of Equation (17) by e αξ1 z αξ1 , we get βµ λ d − αξ z αξ e 1 z 1 G1 (z) dz

(18) βµ η γ − λ z βµ = + P1,0 − G0 (z) e αξ1 z αξ1 . αξ1 z αξ1 (1 − z) αξ1 (1 − z) Integrating from 0 to z, we have

βµ λ βµ η z − αξ αξ1 G1 (z) = e z 1 K1 (z) + K2 (z) P1,0 αξ1 αξ1 Z βµ γ z − λs − (1 − s)−1 s αξ1 e αξ1 G0 (s)ds , αξ1 0

(19)


Vacation queueing model with waiting server and customers’ impatience 225 where

Zz −1

s s

K1 (z) =

βµ αξ1

e

λs − αξ

1

Zz

0

βµ

(1 − s)−1 s αξ1 e

ds, K2 (z) =

λs − αξ

1

ds.

(20)

0

Using Equation (14) and substituting Equation (16) into (19), we get

βµ λz βµ η γ − αξ αξ1 G1 (z) = e z 1 K1 (z) + K2 (z) P1,0 − K3 (z)P0,0 , (21) αξ1 αξ1 αξ1 with

βµ Zz γ λ λ K0 (s) − αξ +1 − αξ s αξ1 αξ 0 1 s (1 − s) e 0 ds. K3 (z) = 1− K0 (1) 0

(22)

Next, putting z = 1 in Equation (8), we get the probability that the server ∞ X P0,n , is on vacation, P0,. = G0 (1) = n=0

P0,. =

δ1 P0,0 + δ2 P1,1 γ

.

(23)

And, putting z = 1 in Equation (21), we find the probability that the server ∞ X P1,n , is in busy period, P1,. = G1 (1) =

P1,. = e

λ αξ1

n=0

γ η βµ K1 (1) + K2 (1) P1,0 − K3 (1)P0,0 . αξ1 αξ1 αξ1

From Equation (3), it yields γ βµ + αξ1 P1,0 = P0,0 + P1,1 . λ+η λ+η

(24)

(25)

Substituting Equation (25) into (24), we have

λ γ βµ η γ αξ1 P1,. = e K1 (1) + K2 (1) − K3 (1) P0,0 λ + η αξ1 αξ1 αξ1 +

βµ η βµ + αξ1 K1 (1) + K2 (1) P1,1 . αξ1 αξ1 λ+η

(26)

Next, substituting Equation (15) into (23), we get (5). Then, substituting Equation (15) into (26), we obtain (6).


226

A. A. Bouchentouf, A. Guendouzi and A. Kandouci

Finally, using the normalizing condition ∞ X

P0,n +

n=0

∞ X

P1,n = 1,

n=0

which is equivalent to (27)

P0,. + P1,. = 1. And substituting Equations (15), (23) and (26) into (27), we find (7)

4

Performance measures and cost model

4.1

Performance measures

In this subpart useful performance measures are presented. ∗ The probability that the server is in a busy period (PB ). P(Busy period) = PB = P1,. . ∗ The probability that the server is on vacation (PV ). P(Vacation period) = PV = 1 − P(Busy period). ∗ The probability that the server is idle during busy period (PI ). PI = P1,0 . ∗ The average number of customers in the system when the server is taking vacation (E(L0 )). From Equation (8), using L’Hopital rule, we have E(L0 ) = lim G00 (z) = z→1

−λP0,. + γE(L0 ) . −αξ0

This implies E(L0 ) =

λ γ + αξ0

P0,. .

∗ The average number of customers in the system when the server is in busy period (E(L1 )).


Vacation queueing model with waiting server and customers’ impatience 227 From Equation (9), using L’Hopital rule, we get λ − βµ γ 0 E(L1 ) = lim G1 (z) = P1,. + E(L0 ) z→1 αξ1 αξ1 (βµ + αξ1 )(αξ0 − δ1 K0 (1)) βµ γ+ P0,0 . + αξ1 (λ + η) δ2 K0 (1) ∗ The average number of customers in the system (E(L)). E(L) = E(L0 ) + E(L1 ). ∗ The average number of customers in the queue (E(Lq )). E(Lq ) =

+∞ X n=0

+∞ X nP0n + (n − 1)P1n n=1

= E(L) − (P1,. − P1,0 ). ∗ The mean waiting time of a customer in the system (Ws ). Ws =

E(L0 ) + E(L1 ) E(L) = . λ λ

∗ The expected number of customers served per unit of time (Ecs ). Ecs = βµ(P1,. − P1,0 ). ∗ The average rates of reneging and retention of impatient customers during vacation period. Rren0 = αξ0 E(L0 ), Rret0 = (1 − α)ξ0 E(L0 ). ∗ The average rates of reneging and retention of impatient customers during busy period. Rren1 = αξ1 E(L1 ), Rret1 = (1 − α)ξ1 E(L1 ). Thus, ∗ The average rate of abandonment of a customer due to impatience (Rren ). Rren = Rren0 + Rren1 . ∗ The average rate of retention of impatient customers (Rret ). Rret = Rret0 + Rret1 .


228

A. A. Bouchentouf, A. Guendouzi and A. Kandouci

4.2

Cost model

This subpart is devoted to develop a model for the costs incurred in the queueing system using the following symbols: • C1 : Cost per unit time when the server is working during busy period. • C2 : Cost per unit time when the server is idle during busy period. • C3 : Cost per unit time when the server is on vacation. • C4 : Cost per unit time when a customer joins the queue and waits for service. • C5 : Cost per service per unit time. • C6 : Cost per unit time when a customer reneges. • C7 : Cost per unit time when a customer is retained. • C8 : Cost per unit time when a customer returns to the system as a feedback customer. Let ∗ R be the revenue earned by providing service to a customer. ∗ Γ be the total expected cost per unit time of the system. Γ = C1 PB + C2 PI + C3 PV + C4 E(Lq ) + C6 Rren + C7 Rret + µ(C5 + β 0 C8 ). ∗ ∆ be the total expected revenue per unit time of the system. ∆ = Rµ(1 − PV − P1,0 ). ∗ Θ be the total expected profit per unit time of the system. Θ = ∆ − Γ.

5 5.1

Numerical analysis Impact of system parameters on performance measures

Different performance measures of interest computed under different scenarios are given. These measures are obtained by using a MATLAB program coded by the authors. To illustrate the system numerically, the values for default parameters are considered using the following cases


Vacation queueing model with waiting server and customers’ impatience 229 • Table 1: λ = 1.00 : 0.05 : 1.45, µ = 2.00, η = 0.10, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85, β = 0.50, and α = 0.50. • Table 2: λ = 1.50, µ = 2.00 : 0.40 : 5.60, η = 0.10, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85, β = 0.50, and α = 0.50. • Table 3: λ = 1.50, µ = 2.00, η = 0.10, γ = 0.10, ξ0 = 0.50 : 0.05 : 0.95, ξ1 = 0.85, β = 0.50, and α = 0.50. • Table 4: λ = 1.50, µ = 2.00, η = 0.10, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85 : 0.05 : 1.30, β = 0.50, and α = 0.50. • Table 5: λ = 1.50, µ = 2.00, η = 0.10, γ = 0.10 : 0.05 : 0.55, ξ0 = 0.50, ξ1 = 0.85, β = 0.50, and α = 0.50. • Table 6: λ = 1.50, µ = 2.00, η = 0.10 : 0.05 : 0.55, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85, β = 0.50, and α = 0.50. • Table 7: λ = 1.50, µ = 2.00, η = 0.10, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85, β = 0.10 : 0.10 : 1.00, and α = 0.50. • Table 8: λ = 1.50, µ = 2.00, η = 0.10, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85, β = 0.50, and α = 0.10 : 0.10 : 1.00.

5.2

General comments

∗ From Table 1 it is clearly seen that with the increases of the arrival rate λ, P0,0 and PV decrease, while PB increases. Thus, the mean number of customers in the system during the busy period E(L1 ) increases significatively, which leads to an increase in the number of customers served Ecs . Moreover, E(L0 ) is not monotone with λ, while Ws increases as the arrival rate increases, this implies an increases in the average reneging and retention rates Rren and Rret . ∗ According to Table 2 we see that along the increases of the service rate µ, P0,0 , PV , E(L0 ) and Ecs increase, whereas PB and E(L1 ) both decrease, as it should be expected. Moreover, with the increase in µ, the mean waiting time of a customer in the system Ws deceases, this leads to a decrease in Rren and Rret . Obviously, the higher the service rate, the smaller the average rate of abandonment and the larger the number of customers served. ∗ From Table 3 we remark that when the reneging rate during vacation period ξ0 increases, PB , Ws , E(L0 ) and E(L1 ) decrease, while P0,0 , PV , Rren and Rret increase. Consequently, Ecs decreases. As intuitively expected, the bigger the rate of reneging, the smaller the number of customers served.


230

A. A. Bouchentouf, A. Guendouzi and A. Kandouci Table 1: Performance measures vs. λ

λ 1.00 1.05 1.10 1.15 1.20 1.25 1.30 1.35 1.40 1.45

P0,0 0.0272 0.0248 0.0227 0.0208 0.0191 0.0176 0.0161 0.0148 0.0137 0.0126

PB 0.7720 0.7795 0.7869 0.7943 0.8017 0.8090 0.8163 0.8234 0.8305 0.8375

PV 0.2280 0.2205 0.2131 0.2057 0.1983 0.1910 0.1837 0.1766 0.1695 0.1625

E(L0 ) 0.6840 0.6931 0.7002 0.7052 0.7083 0.7094 0.7087 0.7063 0.7021 0.6964

E(L1 ) 0.7883 0.8654 0.9439 1.0237 1.1049 1.1874 1.2713 1.3566 1.4434 1.5315

Ws 1.4022 1.4169 1.4296 1.4407 1.4505 1.4591 1.4667 1.4735 1.4797 1.4853

Rren 0.5060 0.5411 0.5762 0.6114 0.6466 0.6820 0.7175 0.7531 0.7890 0.8250

Rret 0.5060 0.5411 0.5762 0.6114 0.6466 0.6820 0.7175 0.7531 0.7890 0.8250

Ecs 0.5440 0.5589 0.5738 0.5886 0.6034 0.6180 0.6325 0.6469 0.6610 0.6750

Rret 0.7457 0.6775 0.6179 0.5652 0.5182 0.4758 0.4374 0.4025 0.3707 0.3416

Ecs 0.7543 0.8225 0.8821 0.9348 0.9818 1.0242 1.0626 1.0975 1.1293 1.1584

Table 2: Performance measures vs. µ µ 2.00 2.40 2.80 3.20 3.60 4.00 4.40 4.80 5.20 5.60

P0,0 0.0144 0.0160 0.0174 0.0186 0.0197 0.0207 0.0216 0.0224 0.0231 0.0238

PB 0.8143 0.7938 0.7757 0.7597 0.7455 0.7328 0.7214 0.7111 0.7017 0.6931

PV 0.1857 0.2062 0.2243 0.2403 0.2545 0.2672 0.2786 0.2889 0.2983 0.3069

E(L0 ) 0.7959 0.8839 0.9614 1.0300 1.0909 1.1453 1.1941 1.2383 1.2786 1.3154

E(L1 ) 1.2864 1.0741 0.8883 0.7240 0.5775 0.4459 0.3268 0.2187 0.1201 0.0300

Ws 1.3882 1.3053 1.2331 1.1694 1.1123 1.0607 1.0140 0.9713 0.9325 0.8970

Rren 0.7457 0.6775 0.6179 0.5652 0.5182 0.4758 0.4374 0.4025 0.3707 0.3416

∗ According to Table 4, we observe that along the increases of the reneging rate during busy period ξ1 , PB , E(L1 ) and Ws decrease, this leads to a decrease in Ecs . Further, as expected, the increasing of ξ1 implies an increase in P0,0 , PV , E(L0 ), Rren and Rret . ∗ Table 5 illustrates that PB increases with increasing values of the vacation rate γ, while P0,0 is not monotonic with γ. Further, PV , Ws , E(L0 ) and E(L1 ) decrease with the increase of γ, this implies an increase in Ecs . On the other hand, Rren and Rret decrease significantly as the vacation rate increases, which agrees with the intuitive expectation; the higher the rate of vacation, the bigger the probability of busy period and the greater the number of customers served.


Vacation queueing model with waiting server and customers’ impatience 231 Table 3: Performance measures vs. ξ0 ξ0 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95

P0,0 0.0130 0.0134 0.0139 0.0143 0.0148 0.0153 0.0158 0.0163 0.0167 0.0172

PB 0.8374 0.8372 0.8370 0.8369 0.8367 0.8365 0.8364 0.8362 0.8361 0.8360

PV 0.1626 0.1628 0.1630 0.1631 0.1633 0.1635 0.1636 0.1638 0.1639 0.1640

E(L0 ) 0.6506 0.6106 0.5752 0.5438 0.5157 0.4904 0.4675 0.4466 0.4276 0.4101

E(L1 ) 1.5209 1.5117 1.5036 1.4964 1.4899 1.4842 1.4790 1.4742 1.4699 1.4659

Ws 1.4477 1.4148 1.3859 1.3601 1.3371 1.3164 1.2976 1.2806 1.2650 1.2507

Rren 0.8253 0.8256 0.8260 0.8263 0.8266 0.8269 0.8272 0.8275 0.8278 0.8281

Rret 0.8253 0.8256 0.8260 0.8263 0.8266 0.8269 0.8272 0.8275 0.8278 0.8281

Ecs 0.6747 0.6744 0.6740 0.6737 0.6734 0.6731 0.6728 0.6725 0.6722 0.6719

Rret 0.8380 0.8504 0.8623 0.8737 0.8846 0.8951 0.9052 0.9150 0.9244 0.9334

Ecs 0.6620 0.6496 0.6377 0.6263 0.6154 0.6049 0.5948 0.5850 0.5756 0.5666

Table 4: Performance measures vs. ξ1 ξ1 0.85 0.90 0.95 1.00 1.05 1.10 1.15 1.20 1.25 1.30

P0,0 0.0131 0.0136 0.0140 0.0145 0.0149 0.0153 0.0157 0.0161 0.0164 0.0168

PB 0.8310 0.8248 0.8189 0.8132 0.8077 0.8024 0.7974 0.7925 0.7878 0.7833

PV 0.1690 0.1752 0.1811 0.1868 0.1923 0.1976 0.2026 0.2075 0.2122 0.2167

E(L0 ) 0.7242 0.7508 0.7763 0.8007 0.8241 0.8467 0.8683 0.8892 0.9093 0.9288

E(L1 ) 1.4598 1.3951 1.3364 1.2828 1.2338 1.1886 1.1469 1.1083 1.0723 1.0389

Ws 1.4560 1.4306 1.4084 1.3890 1.3719 1.3568 1.3435 1.3317 1.3211 1.3117

Rren 0.8380 0.8504 0.8623 0.8737 0.8846 0.8951 0.9052 0.9150 0.9244 0.9334

∗ According to Table 6, it is clearly observed that with the increase in the waiting server rate η, the probability of busy period PB decreases which leads to a decrease in the mean number of customers served Ecs ; this is because Ws , PV and E(L0 ) increase with η, which implies an increase in Rren , Rret and P0,0 . On the other hand the number of customers in the system during busy period E(L1 ) increases; the reason is that the size of the system during vacation period becomes large with η.


232

A. A. Bouchentouf, A. Guendouzi and A. Kandouci Table 5: Performance measures vs. γ

γ 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55

P0,0 0.0284 0.0290 0.0290 0.0286 0.0281 0.0276 0.0270 0.0264 0.0258 0.0252

PB 0.7420 0.7933 0.8276 0.8522 0.8706 0.8850 0.8965 0.9059 0.9138 0.9204

PV 0.2580 0.2067 0.1724 0.1478 0.1294 0.1150 0.1035 0.0941 0.0862 0.0796

E(L0 ) 0.9674 0.6890 0.5172 0.4032 0.3234 0.2654 0.2218 0.1882 0.1617 0.1404

E(L1 ) 2.1932 1.9385 1.7849 1.6856 1.6179 1.5699 1.5348 1.5085 1.4884 1.4728

Ws 2.1071 1.7517 1.5348 1.3925 1.2942 1.2235 1.1711 1.1311 1.1001 1.0755

Rren 1.1740 0.9961 0.8879 0.8172 0.7685 0.7336 0.7077 0.6882 0.6730 0.6610

Rret 1.1740 0.9961 0.8879 0.8172 0.7685 0.7336 0.7077 0.6882 0.6730 0.6610

Ecs 0.6646 0.7106 0.7414 0.7635 0.7801 0.7930 0.8033 0.8118 0.8189 0.8249

Rret 0.8914 0.9669 1.0483 1.1338 1.2223 1.3130 1.4054 1.4992 1.5940 1.6897

Ecs 0.6532 0.6369 0.6243 0.6142 0.6060 0.5992 0.5935 0.5886 0.5843 0.5806

Table 6: Performance measures vs. η η 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55

P0,0 0.0161 0.0187 0.0208 0.0224 0.0237 0.0248 0.0258 0.0266 0.0272 0.0278

PB 0.7919 0.7579 0.7316 0.7107 0.6936 0.6794 0.6674 0.6571 0.6483 0.6405

PV 0.2081 0.2421 0.2684 0.2893 0.3064 0.3206 0.3326 0.3429 0.3517 0.3595

E(L0 ) 0.8919 1.0375 1.1502 1.2400 1.3132 1.3741 1.4254 1.4694 1.5074 1.5407

E(L1 ) 1.5729 1.6647 1.7899 1.9383 2.1034 2.2811 2.4684 2.6632 2.8639 3.0696

Ws 1.6432 1.8015 1.9601 2.1189 2.2778 2.4368 2.5959 2.7550 2.9142 3.0735

Rren 0.8914 0.9669 1.0483 1.1338 1.2223 1.3130 1.4054 1.4992 1.5940 1.6897

∗ The effect of non-feedback probability β is presented in Table 7, we see that PB and Ws both decrease with increasing values of β. Further, as expected, P0,0 , PV and E(L0 ) increase as β increases, whereas E(L1 ) decreases with increasing values of β; this is because the mean system size during vacation period increases with β. Further, it is well shown that Rren and Rret both decrease along the increasing of non-feedback probability β, which results in the increase of Ecs .


Vacation queueing model with waiting server and customers’ impatience 233 Table 7: Performance measures vs. β β 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00

P0,0 0.0020 0.0038 0.0060 0.0083 0.0104 0.0123 0.0140 0.0155 0.0169 0.0181

PB 0.9741 0.9503 0.9221 0.8932 0.8658 0.8410 0.8189 0.7995 0.7822 0.7669

PV 0.0259 0.0497 0.0779 0.1068 0.1342 0.1590 0.1811 0.2005 0.2178 0.2331

E(L0 ) 0.1109 0.2128 0.3336 0.4578 0.5752 0.6815 0.7760 0.8594 0.9333 0.9991

E(L1 ) 4.3719 3.6255 2.9646 2.3968 1.9133 1.4995 1.1416 0.8282 0.5510 0.3038

Ws 2.9885 2.5589 2.1988 1.9031 1.6590 1.4541 1.2784 1.1251 0.9895 0.8686

Rren 1.1207 0.9596 0.8246 0.7137 0.6221 0.5453 0.4794 0.4219 0.3711 0.3257

Rret 1.1207 0.9596 0.8246 0.7137 0.6221 0.5453 0.4794 0.4219 0.3711 0.3257

Ecs 0.3793 0.5404 0.6754 0.7863 0.8779 0.9547 1.0206 1.0781 1.1289 1.1743

Rret 5.0220 2.5817 1.6724 1.1631 0.8250 0.5784 0.3878 0.2344 0.1074 0.0000

Ecs 0.9420 0.8546 0.7833 0.7246 0.6750 0.6324 0.5952 0.5625 0.5334 0.5074

Table 8: Performance measures vs. α α 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00

P0,0 0.0019 0.0049 0.0076 0.0101 0.0126 0.0151 0.0178 0.0205 0.0231 0.0258

PB 0.9710 0.9273 0.8916 0.8623 0.8375 0.8162 0.7976 0.7813 0.7667 0.7537

PV 0.0290 0.0727 0.1084 0.1377 0.1625 0.1838 0.2024 0.2187 0.2333 0.2463

E(L0 ) 0.2900 0.5454 0.6502 0.6886 0.6964 0.6892 0.6746 0.6562 0.6362 0.6157

E(L1 ) 6.3941 3.4759 2.4282 1.8756 1.5315 1.2957 1.1238 0.9926 0.8892 0.8056

Ws 4.4560 2.6808 2.0523 1.7095 1.4853 1.3233 1.1989 1.0992 1.0170 0.9475

Rren 0.5580 0.6454 0.7167 0.7754 0.8250 0.8676 0.9048 0.9375 0.9666 0.9926

∗ The impact of non-retention probability α is shown in Table 8. As intuitively expected, along the increase of α, PB and E(L1 ) decrease, while PV increases as α increases. Further, E(L0 ) is not monotonic with the probability of non-retention. Moreover, Ws and Rret both decrease with increasing of α whereas Rren increases with the probability α, this leads to a decrease of Ecs . This is quite reasonable; the smaller the probability of retaining impatient customers, the larger the average rate of reneged customers and the smaller the number of customers served.


234

A. A. Bouchentouf, A. Guendouzi and A. Kandouci

5.3

Economic analysis

In this subpart, a sensitive economic analysis of the model is performed numerically and the results are discussed appropriately. We present the variation in total expected cost, total expected revenue and total expected profit with the change in different parameters of the system. For the whole numerical study we fix the costs at C1 = 5, C2 = 3, C3 = 5, C4 = 3, C5 = 4, C6 = 3, C7 = 2, C8 = 2, and R = 50.

Impact of arrival rate λ We examine the impact of λ by keeping all other variables fixed, to this end we take λ = 1.00 : 0.05 : 1.45, µ = 2.00, η = 0.10, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85, β = 0.50, and α = 0.50. Results of the analysis are summarized in Table 9 and Figure 1. Table 9: Γ, ∆ and Θ for different values of λ

Γ

1.05 23.04 55.89 32.84

1.15 23.86 58.86 35.00

1.20 24.26 60.33 36.06

1.25 24.67 61.80 37.12

Θ

Γ

1.35 25.48 64.68 39.20

1.40 25.88 66.10 40.21

1.45 26.29 67.50 41.20

Θ

80

Γ , ∆ and Θ

40

60

60 50 40 30

Γ , ∆ and Θ

1.30 25.07 63.25 38.17

100

1.10 23.45 57.38 33.92

120

1.00 22.63 54.39 31.76

70

λ Γ ∆ Θ

1.0

1.1

1.2

1.3

1.4

λ

Figure 1: Γ, ∆ and Θ vs. λ

2.0

2.5

3.0

3.5

4.0

4.5

5.0

5.5

µ

Figure 2: Γ, ∆ and Θ vs. µ

Following the obtained results we observe that Γ, ∆, and Θ all increase with the increasing of the arrival rate λ. This result agrees with our intuition; the number of the customers in the system increases with the increasing of


Vacation queueing model with waiting server and customers’ impatience 235 λ, therefore a large number of customers is served. Consequently, the total expected profit increases.

Impact of service rate µ To check the impact of service rate µ, the values of the parameters are chosen as follows: λ = 1.50, µ = 2.00 : 0.40 : 5.60, η = 0.10, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85, β = 0.50, and α = 0.50. Table 10: Γ, ∆ and Θ for different values of µ µ Γ ∆ Θ

2.00 27.53 75.43 47.89

2.40 28.88 82.25 53.37

2.80 30.31 88.21 57.90

3.20 31.80 93.48 61.67

3.60 33.35 98.18 64.82

4.00 34.95 102.4 67.46

4.40 36.58 106.2 69.67

4.80 38.25 109.7 71.49

5.20 39.94 112.9 72.98

5.60 41.66 115.8 74.17

According to Table 10 and Figure 2 we see that Γ and ∆ increase with increasing values of µ, this generates an increase in Θ. This result makes perfect sense, the higher the service rate, the greater the total expected profit of the system.

Impact of reneging rates ξ0 and ξ1 Let’s study the effect of reneging rates in vacation and busy periods ξ0 and ξ1 , to this end we consider the following cases • Table 11: λ = 1.50, µ = 2.00, η = 0.10, γ = 1.00, ξ0 = 2.00 : 0.50 : 6.50, ξ1 = 0.85, β = 0.50, and α = 0.50. • Table 12: λ = 1.50, µ = 2.00, η = 0.10, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85 : 0.05 : 1.30, β = 0.50, and α = 0.50. Table 11: Γ, ∆ and Θ for different values of ξ0 ξ0 Γ ∆ Θ

2.00 23.97 78.59 54.62

2.50 23.99 78.47 54.48

3.00 24.01 78.37 54.36

3.50 24.02 78.28 54.26

4.00 24.03 78.21 54.18

4.50 24.04 78.15 54.11

5.00 24.05 78.10 54.06

5.50 24.05 78.06 54.01

From Tables 11 and 12 and Figures 3 and 4 we observe that

6.00 24.05 78.03 53.97

6.50 24.06 78.01 53.95


236

A. A. Bouchentouf, A. Guendouzi and A. Kandouci Table 12: Γ, ∆ and Θ for different values of ξ1 0.85 26.24 66.20 39.95

0.95 26.19 63.77 37.58

1.00 26.17 62.63 36.45

1.05 26.17 61.54 35.36

1.10 26.17 60.48 34.31

1.15 26.18 59.47 33.29

1.20 26.19 58.50 32.31

1.25 26.20 57.56 31.36

Θ

Γ

1.30 26.22 56.65 30.43

Θ

40

50

Γ , ∆ and Θ

60 50

30

30

40

Γ , ∆ and Θ

70

60

80

Γ

0.90 26.21 64.96 38.74

70

ξ1 Γ ∆ Θ

2

3

4

5

6

ξ0

Figure 3: Γ, ∆ and Θ vs. ξ0

0.9

1.0

1.1

1.2

1.3

ξ1

Figure 4: Γ, ∆ and Θ vs. ξ1

∗ As expected, along the increasing of ξ0 , Γ increases while Θ and ∆ decrease with ξ0 , this is because the average rate of reneged customers increases with ξ0 . Therefore the number of customers served decreases, which results in the decrease of the total expected profit. ∗ With the increase of ξ1 , ∆ decreases, while Γ is not monotonic with the parameter ξ1 . Further, Θ decreases with the increasing values of the impatience rate, this is because the number of customers in the system decreases with ξ1 , this implies a decrease in PB which results in the decrease of Ecs .

Impact of vacation rate γ To examine the impact of the vacation rate γ on the total expected profit, we take λ = 1.50, µ = 2.00, η = 0.10, γ = 0.10 : 0.05 : 0.55, ξ0 = 0.50, ξ1 = 0.85, β = 0.50, and α = 0.50.


Vacation queueing model with waiting server and customers’ impatience 237 From Table 13 and Figure 5 it is easily seen that the increases of the vacation rate γ implies a decrease in Γ and a considerable increase in ∆ and Θ. This is quite explicable; as γ increases the vacation duration decreases and the server switches to busy period during which customers are served. This leads to a significant increase in the total expected profit. Table 13: Γ, ∆ and Θ for different values of γ 0.10 30.58 66.46 35.87

0.20 26.61 74.13 47.53

0.25 25.61 76.34 50.72

0.30 24.93 78.00 53.06

0.35 24.45 79.29 54.85

0.40 24.09 80.33 56.24

0.45 23.81 81.18 57.37

0.50 23.60 81.89 58.29

Θ

Γ

Θ

50

Γ , ∆ and Θ

40

70 60 50

20

30

30

40

Γ , ∆ and Θ

0.55 23.43 82.49 59.06

60

80

Γ

0.15 28.11 71.06 42.94

70

γ Γ ∆ Θ

0.1

0.2

0.3

0.4

0.5

γ

Figure 5: Γ, ∆ and Θ vs. γ

0.1

0.2

0.3

0.4

0.5

η

Figure 6: Γ, ∆ and Θ vs. η

Impact of waiting rate of a server η Here, we examine the sensitivity of the total expected profit versus the waiting server rate η. For this case, we put λ = 1.50, µ = 2.00, η = 0.10 : 0.05 : 0.55, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85, β = 0.50, and α = 0.50. The numerical results are presented in Table 14 and Figure 6. From the obtained results we remark that with the increase in η, total expected cost Γ increases, while ∆ and Θ monotonically decease with the parameter η. This is due to the fact that the probability of busy period during which service is provided decreases with the parameter η. Therefore, the total expected profit decreases considerably.


238

A. A. Bouchentouf, A. Guendouzi and A. Kandouci Table 14: Γ, ∆ and Θ for different values of η

η Γ ∆ Θ

0.10 27.26 65.31 38.04

0.15 28.30 63.68 35.38

0.20 29.38 62.42 33.04

0.25 30.49 61.42 30.92

0.30 31.62 60.60 28.98

0.35 32.77 59.92 27.15

0.40 33.93 59.34 25.42

0.45 35.09 58.85 23.75

0.50 36.27 58.43 22.15

0.55 37.45 58.06 20.60

Impact of non-retention probability α To study the impact of α on the total expect profit, we choose the parameters values as follows: λ = 1.50, µ = 2.00, η = 0.10, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85, β = 0.50, and α = 0.10 : 0.10 : 1.00. Table 15: Γ, ∆ and Θ for different values of α 0.10 46.85 94.20 47.34

0.20 34.38 85.45 51.07

0.3 30.05 78.32 48.27

0.40 27.75 72.45 44.69

Γ

0.60 25.26 63.24 37.97

0.80 23.88 56.25 32.36

0.90 23.39 53.34 29.95

Θ

Γ

Γ , ∆ and Θ

80

1.00 22.98 50.74 27.76

Θ

40

60 20

20

40

Γ , ∆ and Θ

0.70 24.49 59.52 35.03

80

0.50 26.29 67.50 41.20

60

α Γ ∆ Θ

0.2

0.4

0.6

0.8

α

Figure 7: Γ, ∆ and Θ vs. α

1.0

0.2

0.4

0.6

0.8

1.0

β

Figure 8: Γ, ∆ and Θ vs. β

According to Table 15 and Figure 7 we observe that the increases of nonretention probability α implies a decrease in Γ, ∆ and Θ. A slight increase is observed in Θ when the parameter α is below a certain value, (α = 0.2). Therefore, we can see that the probability of retaining reneged customers α 0 has a noticeable effect on the total expected profit of the system. This


Vacation queueing model with waiting server and customers’ impatience 239 is because the number of customers served increases with the parameter α 0 . Thus, it is quite clear that the probability of retention has a positive impact in the economy.

Impact of non-feedback probability β Here, we put λ = 1.50, µ = 2.00, η = 0.10, γ = 0.10, ξ0 = 0.50, ξ1 = 0.85, β = 0.10 : 0.10 : 1.00, and α = 0.50. The numerical results obtained for this situation are given in Table 16 and Figure 8.

Table 16: Γ, ∆ and Θ for different values of β β Γ ∆ Θ

0.10 35.32 94.81 59.49

0.20 32.26 90.07 57.80

0.30 29.65 84.42 54.77

0.40 27.45 78.63 51.18

0.50 25.57 73.15 47.57

0.60 23.94 68.19 44.24

0.70 22.49 63.78 41.29

0.80 21.17 59.89 38.72

0.90 19.96 56.44 36.48

1.00 18.83 53.37 34.54

From the obtained results, it is clearly shown that Γ, ∆ and Θ monotonically decrease as non-feedback probability β increases. The reason is that the number of the customers in the system decreases with the increasing of β, which leads to a decrease in the total expected profit.

6

Conclusion

In this paper we studied an M/M/1 Bernoulli feedback queueing system with single exponential vacation, waiting server, reneging and retention of reneged customers, wherein the impatience timers of customers depend on the states of the server. The explicit expressions of the steady-state probabilities are obtained, using probability generating functions (PGFs). Useful measures of effectiveness of the queueing system are presented and a cost model is developed. Finally, an extensive numerical study is presented. Our system can be considered as a generalized version of the existing queueing models given by Yue et al.[22] and Ammar [5] associated with several practical situations. The model considered in this paper can be extended to multiserver queueing system with delayed state-dependent service times, breakdowns and repairs.


240

A. A. Bouchentouf, A. Guendouzi and A. Kandouci

References [1] I. Adan, A. Economou, S. Kapodistria, Synchronized reneging in queueing systems with vacations, Queueing Syst, 62 (1-2) (2009), 1–33. [2] E. Altman, U. Yechiali, Analysis of customers’ impatience in queues with server vacation, Queueing Syst, 52 (4) (2006), 261–279. [3] E. Altman, U. Yechiali, Infinite server queues with systems’ additional task and impatient customers, Probab. Eng. Inf. Sci, 22 (4) (2008), 477– 493. [4] S. I. Ammar, Transient analysis of an M/M/1 queue with impatient behavior and multiple vacations, Appl. Math. Comput, 260 (C) (2015), 97–105. [5] S. I. Ammar, Transient solution of an M/M/1 vacation queue with a waiting server and impatient customers, Journal of the Egyptian Mathematical Society, 25 (3) (2017), 337–342. [6] J. Baek, O. Dudina, C. Kim, A queueing system with heterogeneous impatient customers and consumable additional items, Int. J. Appl. Math. Comput. Sci., 27 (2) (2017), 367–384. [7] A. A. Bouchentouf, M. Kadi, A. Rabhi, Analysis of two heterogeneous server queueing model with balking, reneging and feedback, Math. Sci. Appl. E Notes, 2 (2) (2014), 10–21. [8] A. A. Bouchentouf, A. Messabihi, Heterogeneous two-server queueing system with reverse balking and reneging, OPSEARCH, 55 (2) (2018), 251– 267. [9] B. T. Doshi, Queueing systems with vacations-A survey Queueing Syst, 1 (1) (1986), 29–66. [10] M. S. El-Paoumy, H. A. Nabwey, The Poissonian Queue with Balking Function, Reneging and two Heterogeneous Servers, International Journal of Basic and Applied Sciences, 11 (6) (2011), 149–152. [11] J. C. Ke, C. H. Wu, Z. G. Zhang, Recent development in vacation queueing models: a short survey, Int. J. Oper. Res., 7 (4) (2010), 3–8.


Vacation queueing model with waiting server and customers’ impatience 241 [12] R. Kumar, N. K. Jain, B. K. Som, Optimization of an M/M/1/N feedback queue with retention of reneged customers, Operations research and decisions, 24 (3) (2014), 45–58. [13] R. Kumar, S. K. Sharma, A Multi-Server Markovian Feedback Queue with Balking Reneging and Retention of Reneged Customers, AMO-Advanced Modeling and Optimization, 16 (2) (2014), 395–406. [14] R. Padmavathy, K. Kalidass, K. Ramanath, Vacation queues with impatient customers and a waiting server, Int. Jour. of Latest Trends in Soft. Eng., 1 (1) (2011), 10–19. [15] Y. W. Shin, T. S. Choo, M/M/s queue with impatient customers and retrials, Applied Mathematical Modelling, 33 (6) (2009), 2596–2606. [16] H. Takagi, Queueing Analysis: A Foundation of Performance Evaluation, Volume 1: Vacation and Priority System, Elsevier, Amsterdam, (1991). [17] N. Tian, Z. Zhang, Vacation Queueing Models-Theory and Applications, Springer-Verlag, New York, (2006). [18] K. H. Wang, Y. C. Chang, Cost analysis of a finite M/M/R queueing system with balking, reneging and server breakdowns, Mathematical Methods of OR, 56 (2) (2002), 169–180. [19] C. H. Wu, J. C. Ke, Computational algorithm and parameter optimization for a multi-server system with unreliable servers and impatient customers, J. Comput. Appl. Math., 235 (2010), 547–562. [20] D. Yue, Y. Zhang, W. Yue, Optimal performance analysis of an M/M/1/N queue system with balking, reneging and server vacation, International Journal of Pure and Applied Mathematics, 28 (1) (2006), 101–115. [21] D. Yue, W. Yue, Z. Saffer, X. Chen, Analysis of an M/M/1 queueing system with impatient customers and a variant of multiple vacation policy, Journal of Industrial & Management Optimization, 10 (1) (2014), 89–112. [22] D. Yue, W. Yue, G. Zhao, Analysis of an M/M/1 queue with vacations and impatience timers which depend on the server’s states, Journal of Industrial and Management Optimization, 12 (2) (2016), 653–666. Received: July 21, 2017


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 242–248 DOI: 10.2478/ausm-2018-0019

On real valued ω-continuous functions C. Carpintero

N. Rajesh

Department of Mathematics, Universidad De Oriente, Venezuela email: carpintero.carlos@gmail.com

Department of Mathematics, Rajah Serfoji Govt. College, India email: nrajesh topology@yahoo.co.in

E. Rosas Department of Mathematics, Universidad De Oriente, Venezuela Department of Natural Sciences and Exact, Universidad de la Costa, Colombia email: ennisrafael@gmail.com, erosas@cuc.edu.co

Abstract. The aim of this paper is to introduce and study upper and lower ω-continuous functions. Some characterizations and several properties concerning upper (resp. lower) ω-continuous functions are obtained.

1

Introduction

Generalized open sets play a very important role in General Topology and they are now the research topics of many topologist worldwide. Indeed a significant theme in General Topology and Real analysis concerns the various modified forms of continuity, separation axioms etc. by utilizing generalized open sets. Recently, as generalization of closed sets, the notion of ω-closed sets were introduced and studied by Hdeib [4]. Several characterizations and properties of ω-closed sets were provided in [1, 2, 3, 4, 5]. Various types of functions play a significant role in the theory of classical point set topology. A great number of papers dealing with such functions have appeared, and a good many of them 2010 Mathematics Subject Classification: 54C05, 54C601, 54C08, 54C10 Key words and phrases: ω-closed space, ω-open sets, ω-continuous functions

242


On real valued ω-continuous functions

243

have been extended to the setting of multifunction. The purpose of this paper is to define upper and lower ω-continuous functions. Also, some characterizations and several properties concerning upper (lower) ω-continuous functions are obtained.

2

Preliminaries

Throughout this paper, spaces (X, τ) and (Y, σ) (or simply X and Y) always mean topological spaces in which no separation axioms are assumed unless explicitly stated. Let A be a subset of a space X. For a subset A of (X, τ), Cl(A) and Int(A) denote the closure of A with respect to τ and the interior of A with respect to τ, respectively. A point x ∈ X is called a condensation point of A if for each U ∈ τ with x ∈ U, the set U ∩ A is uncountable. A is said to be ω-closed [4] if it contains all its condensation points. The complement of an ω-closed set is said to be an ω-open set. It is well known that a subset W of a space (X, τ) is ω-open if and only if for each x ∈ W, there exists U ∈ τ such that x ∈ U and U\W is countable. The intersection (resp. union) of all ω-closed (resp. ω-open) set containing (resp. contained in) A ⊂ X is called the ω-closure (resp. ω-interior) of A and is denoted by ω Cl(A) (resp. ω Int(A)). The family of all ω-open, ω-closed sets of (X, τ) is, respectively denoted by ωO(X), ωC(X). We set ωO(X, x) = {A : A ∈ ωO(X) and x ∈ A} and ωC(X, x) = {A : A ∈ ωC(X) and x ∈ A}. The ω-θ-closure [3] of A, denoted by ω Clθ (A), is defined to be the set of all x ∈ X such that A ∩ ω Cl(U) 6= ∅ for every U ∈ ωO(X, x). A subset A is called ω-θ-closed [3] if and only if A = ω Clθ (A). The complement of ω-θ-closed set is called ω-θ-open. A subset A is called ω-regular if and only if it is ω-θ-open and ω-θ-closed. The family of all ω-regular sets of (X, τ) is denoted by ωR(X). We set ωR(x) = {A : A ∈ ωR(X) and x ∈ A}. A topological space X is said to be ω-closed if every cover of X by ω-open sets has a finite subcover whose ω-closures cover X. Finally we recall that a function f : (X, τ) → (Y, σ) is ω-continuous at the point x ∈ X if for each open set V of Y containing f(x) there exists an ω-open set U in X containing x such that f(U) ⊂ V. If f has the property at each point x ∈ X, then it is said to be ω-continuous [5].

3

On upper and lower ω-continuous functions

Definition 1 A function f : X → R is said to be: (i) lower (resp. upper) ω-continuous at x1 if to each α > 0, there exists an


244

C. Carpintero, N. Rajesh and E. Rosas ω-open set Ux1 such that f(x) > f(x1 ) − α (resp. f(x) < f(x1 ) + α) for all x ∈ Ux1 ;

(ii) lower (resp. upper) ω-continuous if it is respectively so at each point of X. Example 1 Consider X = R with topology τ = {∅, R}, then τω = {∅, R, R \ Q} ∪ {(R \ Q) ∪ A : where A is a subset of Q}. Define f : X → R as follows: f = χR\Q . f is lower ω-continuous but is not upper ω-continuous. In the same form if we define g : X → R as follows: g = χQ , g is upper ω-continuous but is not lower ω-continuous. Theorem 1 A function f : X → R is lower ω-continuous if and only if for each α ∈ R, the set {x ∈ X : f(x) ≤ α} is ω-closed. Proof. Since the family of sets T = {R, ∅} ∪ {(α, ∞) : α ∈ R} forms a topology on R, f is lower ω-continuous if and only if f is ω-continuous from X into the topological space (R, T ). But (−∞, α] is a closed set in (R, T ) and hence f−1 ((−∞, α]) is ω-closed in X. But f−1 ((−∞, α]) = {x ∈ X : f(x) ≤ α}. Therefore, {x ∈ X : f(x) ≤ α} is ω-closed. Corollary 1 A subset A of X is ω-open if and only if the characteristic function χA is lower ω-continuous. Similarly for upper ω-continuity, we have the following characterization. Theorem 2 A function f : X → R is upper ω-continuous if and only if for each α ∈ R, the set {x ∈ X : f(x) ≥ α} is ω-closed. Corollary 2 A subset A of X is ω-closed if and only if the characteristic function χA is upper ω-continuous. Theorem 3 Let {fα : α ∈ Λ} be a family of lower ω-continuous functions from X into R, then the function M(x) = supα∈Λ fα (x) (if it exists) is lower ω-continuous. Proof. Let λ ∈ R and T M(x) < λ. Then fα (x) < λ, for all α ∈ Λ. Now {x ∈ X : M(x) ≤ λ} = {x ∈ X : fα (x) ≤ λ}. But each fα being lower ωα∈Λ

continuous, by Theorem 1, each set {x ∈ X : fα (x) ≤ λ} is ω-closed in X. Since any intersection of ω-closed sets is ω-closed, M is lower ω-continuous.


On real valued ω-continuous functions

245

Theorem 4 Let Λ be a finite index set and {fα : α ∈ Λ} be a family of lower ω-continuous functions from X into R, then the function m(x) = minα∈Λ {fα (x)} (if it exists) is lower ω-continuous. Proof. It is enough to prove the case, when m(x) = min{f1 (x), f2 (x)}. Let λ ∈ R and x0 ∈ X, since f1 , f2 are lower ω-continuous from X into R, there exists ω-open sets U1 (x0 ) (resp. U2 (x0 )) such that f1 (x) > f1 (x0 ) + λ for all x ∈ U1 (x0 ) (resp. f2 (x) > f2 (x0 ) + λ for all x ∈ U2 (x0 ) ). It follows that for all x ∈ U1 (x0 ) ∩ U2 (x0 ), we obtain that m(x) > m(x0 ) + λ for all x ∈ U1 (x0 ) ∩ U2 (x0 ). In consequence, the result follows. Remark 1 If Λ be an infinite index set and {fα : α ∈ Λ} be a family of lower ω-continuous functions from X into R. Then the function m(x) = infα∈Λ {fα (x)} (if it exists) may not be lower ω-continuous. Example 2 For each natural number n, define fn = χ(− 1 , 1 ) then m(x) = χ{0} , n n is not lower ω-continuous. Theorem 5 Let {fα : α ∈ Λ} be a family of upper ω-continuous function from X into R, then the function g(x) = infα∈Λ {fα (x)} (if it exists) is upper ω-continuous. Proof. Similar to the proof of Theorem 3.

Theorem 6 Let Λ be a finite index set and {fα : α ∈ Λ} be a family of upper ω-continuous functions from X into R, then the function M(x) = maxα∈Λ {fα (x)} (if it exists) is upper ω-continuous. Proof. Similar to the proof of Theorem 4.

Remark 2 If Λ be an infinite index set and {fα : α ∈ Λ} be a family of upper ω-continuous functions from X into R. Then the function m(x) = supα∈Λ {fα (x)} (if it exists) may not be upper ω-continuous. Example 3 Similar to Example 2. Theorem 7 Let X be an ω-closed space and let f : X → R be a lower ωcontinuous function. Then f assumes the value m = infx∈X {f(x)}.


246

C. Carpintero, N. Rajesh and E. Rosas

Proof. Let α ∈ R be such that α > m. Then f being the lower ω-continuous, the set Tα = {x ∈ X : f(x) ≤ α} is a nonempty (by the property of infimum) ωclosed set in X. The family {Tα : α ∈ R and α > m} is a collection of nonempty ω-closed sets with finite intersection property in the ω-closed space X; hence T it has nonempty intersection. Let x∗ ∈ Tα . Then f(x∗ ) = m. α>m

Theorem 8 If X is ω-closed, then any upper ω-continuous function f : X → R attains the value M = supx∈X {f(x)}. Proof. Similar to Theorem 7.

Remark 3 If a real valued function f : X → R from an ω-closed space is lower ω-continuous as well as upper ω-continuous, then it is bounded and attains its bounds. Definition 2 Let f : X → Y be a function, where X is a topological space and Y is a poset. Then (i) f is said to be lower (resp. upper) ω-continuous if f−1 ({y ∈ Y : y ≤ y0 }) (resp. f−1 ({y ∈ Y : y ≥ y0 })) is ω-closed in X for each y0 ∈ Y. (ii) a partial order relation ≤ on a topological space X is said to be lower (resp. upper) compatible if the set {x ∈ X : x ≤ x0 } (resp. {x ∈ X : x ≥ x0 }) is ω-closed for each x0 ∈ X. Theorem 9 A topological space X is ω-closed if and only if X has a maximal element with respect to each upper compatible partial order on X. Proof. Suppose that X is not ω-closed. Then there exists a net {xλ : λ ∈ Λ} which has no ω-accumulation point, where Λ is a well-ordered index set. We define the set Aα = X\ω Clθ ({xβ : β > α}). We claim that for each x ∈ X, x ∈ Aα for some α. In fact, x is contained in some ω-regular set R such that R ∩ {xβ : β ≥ λ} = ∅ for some β. Consider R = {R ∈ ωR(X) : R ∩ {xβ : β ≥ λ} = ∅ for some β}. Let λR be the smallest index such that R ∩ {xβ : β ≥ λR } = ∅; let λx be the smallest element of M = {λR : R ∈ R}. We define the relation ≤ on X as follows: x ≤ y if and only if Aλx ⊂ Aλy , that is, if and only if X\ω Cl({xβ : β ≥ λx }) ⊂ X\ω Cl({xβ : β ≥ λy }), that is, if and only if ω Cl({xβ : β ≥ λy }) ⊂ ω Cl({xβ : β ≥ λx }), that is, if and only if λx ≤ λy . Clearly ≤ is a partial order relation on X. We claim that λx is the first element of M for which x ∈ Aλx . In fact if α < λx and x ∈ Aα , then x ∈ / ω Cl({xβ : β ≥ α}). Then


On real valued ω-continuous functions

247

there exists R ∈ ωR(X) such that R ∩ {xβ : β ≥ α} = ∅, a contradiction. It is obvious that for the corresponding λx there exists an Rλx ∈ R such that Rλx ∩ {xβ : β ≥ λx } = ∅ and for any α < λx , Rλx ∩ {xβ : β ≥ λx } 6= ∅. Also, Rλx ∩ {xβ : β ≥ λx } = ∅. Then Rλx ∩ ω Cl({xβ : β ≥ λx }) = ∅, that is Rλx ⊂ X\ω Cl({xβ : β ≥ λx }) = Aλx and this happens for every x ∈ X. To show ≤ is upper compatible, it is sufficient to show that {x ∈ X : x ≥ x0 } is ω-closed for every x0 ∈ X. If possible, for some x0 ∈ X, {x ∈ X : x ≥ x0 } is not ω-closed, that is, there exists y ∈ ω Cl({x ∈ X : x ≥ x0 }) such that y < x0 , Rλy is an ω-regular set containing y such that x ∈ Rλy with x > y, that is, λx > λy , that is x ∈ X\ω Cl({xβ : β ≥ λy }) = Aλy . But λx is the first index such that x ∈ Aλx and thus we arrive at a contradiction. Hence, ≤ is upper compatible. Further, (X, ≤) has no maximal element; in fact, if there be any, say x0 , then for some fixed ω, ω Cl({xβ : β ≥ λ}) ⊂ ω Cl({xβ : β ≥ α}) for every α ∈ M, that is, xλ ∈ ω Cl({xβ : β ≥ α}), for all α ∈ M, a contradiction. Conversely, let S be a linearly ordered subset of the topological upper compatible poset X. We denote by Sx the set {y ∈ X : y ≥ x}. As the partial order on X is upper compatible, each Sx is ω-closed. Since S is a linearly ordered subset T T of X, {Sx : x∗∈ X} has finite intersection property. Then Sx 6= ∅. Let x∗ ∈ Sx . Then x ≥ x, for x∈S

x∈S

all x ∈ S. Therefore, by Zorn’s lemma X has a maximal element.

Theorem 10 A topological space X is ω-closed if and only if X has a maximal element with respect to each lower compatible partial order on X. Proof. Similar to Theorem 9.

Theorem 11 A topological space X is ω-closed if and only if each upper ωcontinuous function from X into a poset assumes a maximal value. Proof. Suppose that X is not ω-closed, then there exists a net {xλ : λ ∈ M} with no ω-accumulation point, where M is a well-ordered set. We assumes that the topology on M is the order topology. Now, for each β ∈ M, Aβ = ω Cl({xλ : λ ≥ β}). We define a function f : X → M as follows: f(x) = βx , where βx is the first element of the β’s for which x ∈ / Aβ . This is well defined because from the fact that M is well-ordered, obviously, f(x) has no maximal element. We define the relation ≤ on X as follows: x ≤ y if and only if f(x) ≤ f(y). Clearly, ≤ is a partial order relation on X. Now, for each x ∈ X, Sx = f−1 ({z ∈ Y : z ≥ f(x)}) = {y ∈ X : y ≥ x}. As f is ω-continuous, each Sx is ω-closed and hence ≤ is an upper compatible partial order relation on X. Then X being an


248

C. Carpintero, N. Rajesh and E. Rosas

ω-closed space, by Theorem 9 it has a maximal element x∗ . Therefore, f(x∗ ) is the maximal element of f(X). Theorem 12 A topological space X is ω-closed if and only if each lower ωcontinuous function from X into a poset assumes a minimum value. Proof. Similar to Theorem 11.

References [1] K. Al-Zoubi, B. Al-Nashef, The topology of ω-open subsets, Al-Manarah J., (9) (2003), 169–179. [2] A. Al-Omari, M. S. M. Noorani, Contra-ω-continuous and almost ωcontinuous functions, Int. J. Math. Math. Sci., (9) (2007), 169–179. [3] A. Al-Omari, T. Noiri, M. S. M. Noorani, Weak and strong forms of ω-continuous functions, Int. J. Math. Math. Sci., (9) (2009), 1–13. [4] H. Z. Hdeib, ω-closed mappings, Rev. Colombiana Mat., 16 (1982),65–78. [5] H. Z. Hdeib, ω-continuous functions, Dirasat J., 16 (2) (1989),136–142.

Received: July 6, 2017


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 249–275 DOI: 10.2478/ausm-2018-0020

Rejection sampling of bipartite graphs with given degree sequence Koko K. Kayibi

U. Samee

Department of Physics and Mathematics, University of Hull, UK email: kokokayibi@hull.ac.uk

Department of Mathematics, Islamia College for Science and Commerce, Srinagar, India email: pzsamee@yahoo.co.in

S. Pirzada

Mohammad Ali Khan

Department of Mathematics, University of Kashmir, Srinagar, India email: pirzadasd@kashmiruniversity.ac.in

Department of Mathematics, University of Lethbridge, Canada email: ma.khan@uleth.ca

Abstract. Let A = (a1 , a2 , ..., an ) be a degree sequence of a simple bipartite graph. We present an algorithm that takes A as input, and outputs a simple bipartite realization of A, without stalling. The running time of the algorithm is (n1 n2 ), where ni is the number of vertices in the part i of the bipartite graph. Then we couple the generation algorithm with a rejection sampling scheme to generate a simple realization of A uniformly at random. The best algorithm we know is the implicit one due to Bayati, Kim Saberi (2010) that has a running time of O(mamax ), Pand n where m = 21 i=1 ai and amax is the maximum of the degrees, but does not sample uniformly. Similarly, the algorithm presented by Chen et al. (2005) does not sample uniformly, but nearly uniformly. The realization of A output by our algorithm may be a start point for the edge-swapping Markov Chains pioneered by Brualdi (1980) and Kannan et al.(1999).

2010 Mathematics Subject Classification: 05C07, 65C05 Key words and phrases: degree sequence, contraction of a degree sequence, degree sequence bipartition, contraction of a graph, deletion of a graph, ecological occurrence matrix

249


250

1

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

Introduction

A graph G(V(G), E(G)) is said to be bipartite if its vertex set V(G) can be partitioned into two different sets V1 (G) and V2 (G) with V(G) = V1 (G) ∪ V2 (G) such that uv ∈ E if u ∈ V1 and v ∈ V2 . The graphs considered can have possible parallel edges and loops unless otherwise stated. The Degree Sequence Problem is to find some or all graphs with a given degree sequence [30, 34]. More detailed analysis of the Degree Sequence Problem and its relevance can be found in [29]. It is much researched upon for its relevance in network modelling in Ecology, Social Sciences, chemical compounds and biochemical networks in the cell. Especially, ecological occurrence matrices, such as the Darwin finches tables, are (0, 1) matrices whose rows are indexed by species of animals and columns are islands, and the (i, j) entry is 1 if animal i lives in island j, and is 0 otherwise. Moreover the row sums and columns sums are fixed by field observation of these islands. These occurrence matrices are thus bipartite graphs G with a fixed degree sequence in which V1 (G) is the set of animals and V2 (G) is the set of islands. Researchers in Ecology [8, 9, 15, 31] are highly interested in sampling easily and uniformly ecological occurrence tables, so that by using Monte Carlo methods, they can approximate test statistics to prove or disprove some null hypothesis about competitions amongst animals. Several algorithms are known to sample random realizations of degree sequences, and each one of them has its strengths and limitations. Most of these use Monte Carlo Markov chain methods based on edge-swapping [6, 9, 10, 11, 12, 13, 18, 22, 21, 24]. Since to start a Markov chain still requires to have a realisation of the degree sequence A, many algorithms are proposed that generate such a realisation [1, 3, 5, 2, 36]. Most of these algorithms are based on random matching methods. In particular, algorithms proposed in [1, 3, 8] are based on inserting edges sequentially according to some probability scheme. The basic ideas of the algorithm presented in the present paper can be seen as implementing a ”dual sequential method”, as it inserts sequentially vertices instead of edges. In the theory of the Tutte polynomial, there are two operations, deletion and contraction, that are dual to each other, see [7] for more details on this topic. Let G be a graph having n vertices and m edges. In G, the operation of deleting an edge e = (vi , vj ) means removing the edge e and the graph thus obtained, denoted by G\e, is a graph on n vertices and m−1 edges where both the degrees of vertices vi and vj decrease by 1. The operation of contracting the graph G by e = (vi , vj ) consists of deleting the edge e and identifying the vertices vi and vj . The graph thus obtained, denoted by G/e, is a graph on n − 1 vertices and m − 1 edges where the new vertex obtained by identifying


Rejection sampling of bipartite graphs with given degree sequence

251

vi and vj has degree ai + aj − 2. Deletion is said to be the dual of contraction as the incidence matrix of G\e is orthogonal to the incidence matrix of G∗ /e, where G∗ is the dual of G if G is planar. If A is a degree sequence having n entries, it can easily be shown that random matching methods used in [1, 2, 3, 5, 36] are equivalent to starting from a known realization G of A, delete all the edges one by one, and keeping track of the degrees of vertices after each deletion, until one reaches the empty graph having n vertices. Then, reconstructing a random realization of A consists of taking the reverse of the deletion. That is, starting from the empty graph on n vertices, re-insert edges one by one by choosing which edge to insert according to the degrees of the vertices and some probability scheme depending on the stage of the algorithm, and subject to not getting double edges if one would like to get simple graphs or not linking two vertices on the same part if one wants to get bipartite graphs. The algorithm presented in this paper is based on the dual operation of contraction that has been slightly modified to suit our purpose. It is equivalent to starting from a known realization G of A, contract all the edges one by one, and keeping track of the verticesPafter each contraction, until one reaches the graph with one vertex and 21 n1 ai loops. Then, reconstructing a random realization of A consists of reversing the process of contraction. That is, starting from a graph with one vertex P and 21 n1 ai loops, the algorithm re-inserts vertices one by one by choosing the vertices to be joined according to the degrees of the vertices and some probability that depends on the stage of the algorithm. But, to construct a bipartite realization, we force the algorithm to insert first all the vertices in V1 (G) and then all the vertices in V2 (G). While algorithms that are based on reversing the deletion operation [1, 3] are easy to implement, our algorithm seems more complex as one has to satisfy not only the degree conditions on the vertices, but also some added graphical structures imposed by the contraction. But this is more of a bonus than an inconvenience, as, apart from the fact that the running time is even better, the extra structure allows an easier analysis of the algorithm. Moreover, the internal structure imposed by the contraction operation allows the algorithm to avoid most of the shortcomings of the previous algorithms. In fact, not only the algorithm never restarts, but the algorithm also allows to sample all bipartite realizations with equal probabilities, making their approximate counting much easier than by the importance sampling used in [1, 3]. Better still, this technique can be extended to construct k-partite realizations of a k-partite degree sequence A, for k ≥ 3, where a k-partite degree sequence is defined in a natural way extending the definition of a bipartite degree sequence.


252

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

The present paper uses the following notations and terminology. Two edges e and f in E(G) are said to be multiple edges if they have the same end vertices (in Matroid Theory, multiple edges are said to be parallel ). A simple bipartite graph is without multiple edges and contains no loops. The degree ai of a vertex vi is the number of edges incident to vi with a loop contributing twice to the degree of vi . The degree sequence of a graph G is formed by listing the degrees of vertices of G. If A = (a1 , a2 , ..., an ) is a sequence of integers and G is a bipartite graph that has A as its degree sequence, we say that G is a realization of A, and such a sequence of integers is called a bipartite degree sequence. Thus entries of A can be partitioned as A1 and A2 , where Ai denotes the degree sequence of the part Vi (G). We write Vi and |Ai | to denote the set of vertices with degrees in Ai and the sum of entries in Ai respectively . In the sequel, we denote a bipartite degree sequence A as (A1 : A2 ) and the pair (A1 : A2 ) is called a bipartition of A. Remark 1 If A = (A1 : A2 ) is a bipartite degree sequence having n entries, and A1 and A2 have respectively n1 and n2 entries, then the following are true. 1. n1 + n2 = n 2. |A1 | = |A2 |. 3. The maximal entry of A1 is less or equal to n2 and vice versa. Conversely, any partition of entries of A into two sets B1 and B2 satisfying Observation 1 is a bipartition of A. In the sequel, we make use of Rejection Sampling to sample all realizations of the degree sequence with equal probability. Indeed, let S = S1 , ...,P Sr be a set of structures, where Si is obtained with probability π(Si ) such that i π(Si ) = 1. That is, the set of π(Si ) is a probability distribution function. Let min(π) be the minimal probability amongst all π(Si ). The Rejection Sampling scheme consists of generating Si , then accept it with probability min(π) π(S ) or reject it i

with probability 1 − min(π) π(Si ) . It is easy to see that every structure would then be sample with the same probability min(π). This paper is organized as follows. We first define what is called a recursion chain of a degree sequence, then we present routines for constructing all bipartite realizations. The next section presents criteria and routines to generate simple bipartite realizations only. Then these basic routines are coupled with a rejection sampling routine to get a uniform distribution on the set of all simple bipartite realizations.


Rejection sampling of bipartite graphs with given degree sequence

2 2.1

253

Construction all bipartite realizations of given degrees Recursion chain of degree sequences

Let G be a graph with n vertices and m edges. Throughout we assume that the vertices and edges of G are labelled v1 , v2 , · · · , vn . Let A = (a1 , · · · , an ) be the degree sequence of G, where ai is the degree of the vertex vi . Define an arithmetic operation on A, called contraction, as follows. For an ordered pair (ai , aj ) of entries ai and aj of A with i 6= j, the operation of contraction by (ai , aj ) means changing ai to ai +aj and deleting the entry aj from A. We write A/(i, j) to denote the new sequence thus obtained. We call the sequence A/(i, j) the (i, j)-minor or simply a minor of A. The following example illustrates this operation for a bipartite degree sequence. Example 1 Let A = (4, 3, 3 : 3, 3, 2, 2), where a1 = 4, a2 = 3, a3 = 3 and a4 = 3, a5 = 3, a6 = 2, a7 = 2. We have A/(1, 2) = (7, 3, 3, 3, 2, 2) and A/(4, 2) = (4, 3, 6, 3, 2, 2). Let A be the sequence of integers. A is said to be graphic if there is a graph G, not necessarily bipartite, such that G has A as its degree sequence. Moreover, it is trivial to observe that a sequence of integers is graphic if and only if the sum of its entries is even. Theorem 1 A sequence A is graphic if and only if all its minors are graphic. Proof. Obviously, if A is graphic, then A/(ai , aj ) is graphic, as the sum of its entries is even, by definition of contraction. Now suppose that A/(ai , aj ) is graphic and G 00 is a realization of A/(ai , aj ). To prove that A is also graphic, we present an algorithm, much used in the sequel, that constructs a realization of A, denoted by G, from G 00 . Algorithm AddVertex() Step 1. To G 00 add an isolated vertex labelled vj (as in Figure 1). Step 2 If the degree of vj is aj , stop, output G. Else Step 3. Amongst the ai0 edges incident to vi , counting loops twice, choose one edge e = (vi , vk ) with probability π(e) and connect e to vj so that e becomes (vj , vk ). Go to Step 2. Now, in G the degree of vj is aj by Step 2 of algorithm AddVertex(). Moreover, by the definition of contraction the degree of vi is equal to ai + aj in


254

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

G 00 . Since AddVertex() takes aj edges away from vi , the degree of vi is ai in G. Moreover all other vertices are left unchanged by AddVertex(). Thus G is a realization of A. vi’

G’

vi

vj

G

Figure 1: Construction of a graph G from its contract-minor G 00 To help the intuition, observe that if G 00 is a realization of A/(ai , aj ) and G is a realization of A constructed by AddVertex(), then G 00 is obtained from G by contraction of the edge (vi , vj ). Now, mimicking the process of recursive contraction of matroid as used in the theory of the Tutte polynomial, we define a process of recursive contraction for a degree sequence. A recursion chain of a degree sequence A is a unary tree rooted at A where nodes are integer sequences and every node, except for the root, is a minor of the preceding one. The recursive procedure of contraction is carried on from the root A until a node with a single entry is reached. See Figure 2 for an illustration. As for the Tutte polynomial, the amazing fact, which is then used to construct all the realizations of A is that the order of contraction is immaterial. Despite this basic fact, we still impose a particular order to ease many proofs in the sequel. Notes on notations. For the sake of convenience, we denote by A(i) the node of a recursion chain of a degree sequence A, where i is the number of entries in the node. Thus we denote the root A by A(n) , the next node by A(n−1) , and so on until the last node A(1) . Similarly, we denote by G(i) the realization of A(i) . The n entries of A are labelled from 1 to n. To keep tract of the vertices, we preserve the labelling of entries of A into its minors so that when a contraction by the pair (ai , aj ) is performed, the new vertex is labelled ai , the label aj is deleted, and all other entries keep the labelling they have before the contraction. In this paper, we consider the recursion chain, called the accumulating recursion chain, constructed as follows. Let A = (A1 : A2 ). We order A = (a1 , a2 , ..., an ) as (b1 , b2 , ..., bn : c1 , c2 , ..., cn ), where 1 2 A1 = (b1 , b2 , ..., bn ) and A2 = (c1 , c2 , ..., cn ), such that b1 ≥ b2 ≥ ... ≥ bn 1 2 1 and c1 ≥ c2 ≥ ... ≥ cn and n1 + n2 = n. Below is the pseudocode for the 2


Rejection sampling of bipartite graphs with given degree sequence

255

recursive construction of the accumulating recursion chain of a bipartite degree sequence. (4,3,3: 3,3,2,2) ( a1 , a7 )

(6,3,3, 3,3,2) ( a1 , a6 )

(8,3,3, 3,3) (a ,a ) 1

5

(11,3,3, 3) (a ,a ) 1

4

(14,3,3) ( a1 , a3 )

(17,3) ( a1 , a2 )

(20)

Figure 2: The recursion chain of the bipartition (4, 3, 3 : 3, 3, 2, 2). Nodes of the chain are labelled from A(7) = A to A(1) . Notice that we only perform contractions (v1 , vlast ). Algorithm ConstructBipartiteRecursionChain() Given a bipartite degree sequence A = (a1 , a2 , ..., an ) = (b1 , b2 , ..., bn : 1 c1 , c2 , ..., cn ) with b1 ≥ b2 ≥ ... ≥ bn and c1 ≥ c2 ≥ ... ≥ cn . Let i = n. 2

(1)

1 (2)

2

(n)

Step 1 If i = 1, stop, return {A , A , ..., A }. Else (i−1) (i) Step 2 Let A = A /(1, i). That is, get the (i − 1)th recursive minor of A by contracting the (i)th recursive minor by its first entry and the last entry. Step 3 Decrement i by 1 and go back to Step 1. The accumulation recursion chain of A is denoted by W = (A(1) , A(2) , ..., A(n) ). The following algorithm generates all the bipartite realizations of A. The graph constructed is not necessarily simple. Loosely speaking, this algorithm consists of reversing the recursive process of contraction as implemented by (1) ConstructBipartiteRecursionChain(). This algorithm starts from G the sole


256

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan (1)

(2)

realization of A , and by calling AddVertex() recursively it constructs G , (3) (n) (n) then G , and so on until G that is a realization of A = A. The only conditions imposed on the choice of edges is that up to the nth iteration, only 1 edges (v1 , vj ), with j ≤ n1 , are constructed. That is, we insert vertices of V1 . After the nth iteration, only edges (vk , vj ), with k > n1 and j ≤ n1 , are con1 (1)

(2)

structed. That is, we insert vertices of V2 . We call the graphs G , G ,..., (n) G the partial realizations of A. Algorithm ConstructBipartiteRealization() Given W = (A(1) , A(2) , ..., A(n) ), the bipartite accumulating recursion chain of A, do the following. Step 1. Let i = 1 and build the realization of the node A(1) , denoted P by G(1) , 1 which is the graph consisting of one vertex and m loops, where m = 2 ni=1 ai . Step 2. Let G = G(i) . If G has n vertices, stop, return G. Else, Step 3. Using G(i) and A(i+1) as input, Call Algorithm AddVertex() to construct G(i+1) as a realization of A(i+1) . If i ≤ n1 , AddVertex() only concedes loops. If i > n1 Addvertex() concedes only edges (v1 , vj ) with 1 ≤ j ≤ n1 . Increment i by 1, go back to Step 2. See Figure 3 for an illustration of Algorithm ConstructBipartiteRealization(). The following definitions are needed in the sequel. In the process of contraction implemented by the accumulating recursion chain, we observe that the degrees are accumulating on a1 . If we think of recursive contractions of a graph, this is equivalent to saying that the edges are accumulating on v1 as v1 seems to swallow the other vertices one by one. Hence when reversing the contraction operation in ConstructBipartiteRealization(), vertex v1 plays the role of the ’mother that spawns’ all the other vertices one by one and concedes some edges to them according to their degrees. Thus AddVertex() can attach an edge e to a new vertex vs only if e is incident to v1 . This observation prompts the following formal definitions. Let A = (A1 : A2 ) be a bipartite degree sequence, where A1 and A2 have respectively n1 and n2 entries such th that n1 + n2 = n. Up to the n1 iteration of ConstructBipartiteRealization(), an edge is available if it is a loop incident to v1 . An edge e is lost otherwise. th From the (n1 + 1) iteration of ConstructBipartiteRealization() onwards, an edge is available if it is incident to v1 and a vertex vj with 1 ≤ j ≤ n1 . An edge e is lost otherwise. In the obvious way, we say that a vertex is available if it is incident to some available edge. Let Vav , Eav and Ev respectively denote the j


Rejection sampling of bipartite graphs with given degree sequence

257

set of all available vertices, the set of all available edges and the set of available edges that are incident to the vertex vj , for j ≤ n1 . An edge e = (v1 , vj ) is conceded if AddVertex() disconnects it from v1 so that e becomes e = (vj , vk ) for some vertex vk 6= v1 . We then say that v1 (or sometimes Ev or just vj ) j concedes the edge e. A vertex vs having degree as is fully inserted if as edges are conceded to it. A graph G is said to be (re)constructed if it is an output of ConstructBipartiteRealization(). V5

V5

A

(6)

V4

(3,1:2,1,1)

V2

V4

A

V4

V3

V4

V1

V2

V4

V3

V1

V3

V4

V2

V1

(5,2,1)

V1

V1

V2

A(1)

G(3) V1

(8)

G(4) V2

V3

(6,1,1)

(7,1)

V2

V3

V3

A(2)

V3

G(5) V1

V2

V3

A(3)

V2

(4,1,2,1) V1

A(4)

V3

G(6) V1

(5)

V5 V3

V1

V2

V2

G(2) V1

V2

V1

G

(1)

Figure 3: Random reconstruction tree of (3, 1 : 2, 1, 1). Graphs drawn on the same height as the degree sequence A(i) corresponds to all the graphs having A(i) as their (6) degree sequence. Notice that only realizations of A are bipartite.


258

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

The next observation is an obvious consequence of the definition of the algorithm ConstructBipartiteRealization(). We single it out for the sake of clarity as it is used in the sequel. Remark 2 From the (n1 + 1)th iteration of ConstructBipartiteRealization(), the number of available edges is equal to the number of edges left to be inserted until ConstructBipartiteRealization() terminates. It is because the number of available edges at the end of (n1 )th iteration is equal to half the sum of degrees ai ∈ A1 , and by the definition of the bipartite degree sequence, this number is equal to half the sum of degrees aj ∈ A2 . Theorem 2 Let A = (ai , a2 , · · · , an ) = (A1 : A2 ) be a bipartite degree sequence having n entries where A1 and A2 respectively have n1 and n2 entries, such that n1 + n2 = n. Let W be the bipartite recursion chain of A. Then Algorithm ConstructBipartiteRealization() constructs in time linear on a +a +···+an m = i 22 a bipartite graph G having n vertices and m edges such that G is a realization of A. Moreover, every bipartite realization of A can be constructed in this way. Proof. By Algorithm AddVertex(), the graph G(n) output by Algorithm ConstructBipartiteRealization() is assured to be a realization of A. We need only to prove that G(n) is bipartite. Now, since up to the nth iteration of Construct1 BipartiteRealization(), the routine AddVertex() always chooses loops incident to v1 , vertices inserted from the second iteration up to the nth iteration of 1 ConstructBipartiteRealization() (i.e., vertices in V1 ) can never be adjacent to each other. Moreover, from the (n1 + 1)th to the nth iteration, AddVertex() never chooses an edge (v1 , vj ) with j > n1 . Thus all the vertices inserted from the (n1 + 1)th iteration onwards (i.e., vertices in V2 ) are never adjacent to n each other. Thus, we only have to show that (in G ), v1 is not adjacent to any vertex inserted before the nth iteration of AddVertex(). So, suppose that 1 n G contains an edge e = (v1 , vj ) with j ≤ n1 . But, at the beginning of the (n1 + 1)th iteration, the number of all edges incident to v1 is equal to the sum of the degrees of the vertices left to insert until the end of the Algorithm. Thus one vertex vj with j > n1 is not fully inserted. This is a contradiction. It remains to prove that any bipartite realization G of A can be constructed in this way. So, let G be a realization of A and let e = (vi , vj ), where vi ∈ V1 and vj ∈ V2 , be any edge of G such that vertex vi has degree ai and vertex vj has degree aj . Also suppose that vertex vi and vj respectively were inserted at


Rejection sampling of bipartite graphs with given degree sequence

259

the ith and jth iteration of ConstructBipartiteRealization(), with i ≤ n1 and j > n1 . We need to show that at the jth iteration, there is a positive probability to have an edge e that is incident to vi and e is available. Assume to the contrary, that is, at the jth iteration all the edges incident to vi must be lost. Now all the edges incident to vi are lost before that jth iteration only if at some stage of the running of Algorithm ConstructBipartiteRealization(), there are only the edges that are available and these are exhausted before reaching the jth iteration. Thus, at the jth iteration there are no more available edges. That is, there is no edge incident to v1 . But this means that an +1 +an +2 +...+aj−1 ≥ 1 1 m, contradicting Observation 2. As for the running time, Algorithm ConstructBipartiteRealization() calls Algorithm AddVertex() once for every new vertex vk to be inserted. If vk has degree ak , Algorithm AddVertex() has to go through ak iterations to insert the ak edges of vk . Hence the total number of iterations to terminate ConstructBipartiteRealization() is a1 + a2 + ... + an = 2m.

3

Construction of simple bipartite graphs

Till now, ConstructBipartiteRealization() generates any bipartite realization of the bipartite degree sequence A. But, it is easy to modify AddVertex() so that the output of ConstructBipartiteRealization() is always a simple graph. One obvious condition can be stated as follows. (a) If the Algorithm is inserting the jth edge of vertex vs ( with j > 1 and vs ∈ V2 ) and vk (vk ∈ V1 ) is already adjacent to vs , then no more available edge incident to vk should be chosen. This would prevent ConstructBipartiteRealization() from outputting graphs with multiple edges (vs , vk ). Thus this condition is necessary, but it is not sufficient. Indeed, it is easy to see that the following must also apply. (b) While inserting vertex vs and avoiding choosing edges incident to vk so as not to construct multiple edges (vs , vk ), ConstructBipartiteRealization() may fall into a stage where there are more edges incident to vk than there are vertices left to insert, and G, the graph output by ConstructBipartiteRealization() would then have a multiple edge (v1 , vk ). (c) Let A1 and A2 be (separatly) ordered in non decreasing order, where a1 is the largest entry of A1 and an +1 is the largest entry of A2 . Let Mk be the 1 set of the last k entries of A1 and let max(k) = an −k+1 . Let there be an entry 1 as in A2 satisfying the following. (f1). s − n1 ≥ max(k), (i.e., the number of entries of A2 preceding as is


260

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

greater or equal to the maximal entry in Mk ) (f2). as > n1 − k, (i.e., inserting vs would require more neighbours than there are vertices in V1 \Mk ) and, f(3). n1 k−1 s−1 X X X ai + max(0, ai − n + s). aj ≥ i=k

j=n +1 1

i=1

(that is, the number of edges required to insert vertices of V2 prior to vs exceeds the number of edges available on vertices in Mk plus the minimum number of edges that a vertex vi (with vi ∈ V1 \Mk ) has to concede prior to th the s iteration to prevent vi from having more edges than there are vertices th left to be inserted from the s iteration onwards.) If as ∈ A2 satisfies (f1), (f2) and (f3), then as is said to be k-fat. Let Fk denote the set of all the entries that are k-fat. See an illustration in Figure 4. F6 6

5

4

4

4

4

3

1

F7

M6

7

6

5

4

M4 3

2

2

1

1

Figure 4: A=(7,6,5,4,3,2,2,1,1 : 6,5,4,4,4,4,3,1), where A1 = (7, 6, 5, 4, 3, 2, 2, 1, 1) and A2 = (6, 5, 4, 4, 4, 4, 3, 1). Entries are labelled so that the leftmost entry of A1 is a1 and the rightmost entry of A2 is a17 . The entries a14 and a15 are 6-fat while a15 and a16 are 7-fat.

Now, if (a) is to be respected and ConstructBipartiteRealization() chose every vertex in Mk to concede an edge to every one of the s − n1 vertices preceding vs , then ConstructBipartiteRealization() would get stuck at the stage of inserting vertex vs . This is because by (f1) and (f3), no vertex in Mk would have any edge to concede to vs and so there would be a maximum of n1 − k available vertices. But by (f2), vertex vs needs more adjacent neighbors than the only n1 −k available vertices. Hence, ConstructBipartiteRealization() must take some precautionary measures by not exhausting all the edges incident to vertices in Mk prior to the insertion of vs .


Rejection sampling of bipartite graphs with given degree sequence Figure 5 illustrates how the Algorithm would get stuck at its s

th

261

iteration.

F6 6

5

4

4

4

4

3

1

F7

7

6

5

4

M6

3

2

M4

2

1

1

Figure 5: This is a choice of edges that may exhaust all the edges incident to vertices in M6 prior to the 14th iteration. In this choice, vertex v1 , v2 and v3 must concede 3, 2 and 1 edges respectively lest they would have too many edges after the 13th iteration. Still, vertex v14 would not get inserted fully and the Algorithm would stall.

Although (a), (b) and (c) seem to contradict each other, this section defines all these conditions in a formal settings and proves that they can be satisfied simultaneously. Although the analysis seems lengthy, this set of conditions are just inequalities involving the number of edges and vertices already inserted and the number of edges and vertices left to be inserted at each stage of the Algorithm. Moreover, checking these conditions at each iteration of AddVertex() requires checking O(n2 ) inequalities altogether. Thus it does not add to the running time. Let A = (A1 : A2 ) be a bipartite degree sequence of a simple graph, where A1 and A2 have respectively n1 and n2 entries such that n1 + n2 = n. We recall that Eav represents the set of available edges. That is, edges that are incident to v1 and vertices inserted before the nth iteration of ConstructBi1 partiteRealization(), that is, the vertices of V1 . For vj ∈ V1 , we recall that Ev j is the set of available edges incident to vj . That is, the set of parallel edges connecting v1 and vj . Obviously Ev ⊆ Eav for all j. In particular, Ev is the j 1 set of loops incident to v1 . Some of the Algorithms given in the literature, such as in [1], have the disadvantage that it has to restart. The algorithm given here allows to choose only edges such that it never has to restart. In order to be able to do that, the choice of edges at every stage must be such that no vertex is incident to too many edges of the ’wrong type’. If at its sth iteration, Algorithm ConstructBipartiteRealization() is inserting the vertex vs that has degree as , then ConstructBipartiteRealization() has to


262

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

call the routine AddVertex() that has to go through as iterations. We recall that the (s, t)th stage of ConstructBipartiteRealization() is the iteration where AddVertex() inserts the tth edge of the sth vertex. Let Xs,t and |X|s,t denote respectively a set and its cardinality at the (s, t)th stage of ConstructBipartiteRealization(). To help the reader, we first introduce the motivation for the definitions. At each stage of constructing a simple graph, every vertex vj , where vj ∈ V1 , must be connected by at most one edge to any other vk , where vk ∈ V2 . So, if some vertex vj has more available edges than the vertices left to be inserted after its sth iteration, ConstructBipartiteRealization() would never be able to get rid of all these multiple edges, which would then appear in the final graph. This prompts the following definitions. The vertex vj where j ≤ n1 ( i.e., vj ∈ V1 ) is due if |Ev |st = n − (s − 1), j

(1)

that is, Ev has as many edges as there are vertices left to be inserted. The j vertex vj is overdue if |Ev |st > n − (s − 1), j

(2)

that is, there are too many available edges incident to vj and whatever are the future choices, the Algorithm would never output a simple graph. The vertex vj is undue if it is neither due nor overdue. Obviously, a stage is due, undue, overdue if there is a vertex that is due, undue or overdue,respectively. Let Mk be the set of the last k entries of A1 . An entry as in A2 is k-fat if conditions (f1), (f2) and (f3) are satisfied. We let Fk to denote the set of vertices that are k-fat. A bipartite degree sequence A is fat if it contains a k-fat entry for some integer k > 0. Let ri = ai − n1 + k, where k is the largest integer such that ai is k-fat. The (s, t)th stage is ruined if there is an entry ai with i > s (that is, the vertex vi is not inserted yet) that is fat and the number of vertices in Mk that are available is less than ri . It is not ruined otherwise. The next lemma indicates that once ConstructBipartiteRealization() has taken a ‘wrong path’, it is impossible to mend the situation. Lemma 1 Suppose ConstructBipartiteRealization() is inserting the vertex vs such that s > n1 , (i.e., inserting vs into V2 ). Then the following hold. (a) If the vertex vj is due, it is due or overdue at the next stage. If it is overdue, it is overdue at any future stage.


Rejection sampling of bipartite graphs with given degree sequence

263

(b) If the (s, t)th stage is overdue, then the previous stage (the stage inserting the previous edge) is either due or overdue. (c) If the (s, t)th stage is ruined, then the next stage is also ruined. Proof. (a) Suppose vj is due and Addvertex() does not choose an edge from Ev . j Since no edge of Ev is chosen, the left side of Equation 1 remains same while j the right hand side either goes down by one if ConstructBipartiteRealization() moves to a new vertex vs+1 or stays the same if ConstructBipartiteRealization() moves to another edge t + 1 of the same vertex vs . Hence the next stage is due or overdue. On the other hand, if Addvertex() chooses an edge from Ev , the left hand side goes down by 1 and the right one stays the same. But j if Ev concedes only one edge to vs (as we shall see shortly), Ev is still due j j at the insertion of vertex vs+1 . Similar arithmetical arguments as above show that if vj is overdue, it stays overdue. (b) Suppose vj is overdue at the (s, t)th stage but is undue at the stage inserting the previous edge. Then at the previous stage, we have |Ev | < n − (s − 1). j

(3)

Now, either the last edge inserted is chosen from Ev or not. Moreover, in j either case, Algorithm ConstructBipartiteRealization() moves to a new vertex or not. If it stays on the same vertex and the chosen edge is not from Ev , the j right and the left hand sides of Equation 3 are both unchanged. Hence vj is undue at the (s, t)th stage, which is a contradiction. If it stays on the same vertex and the chosen edge is from Ev , the left hand side of Equation 3 goes j down by 1 while the right hand side is unchanged. Hence vj is also undue at the (s, t)th stage and this is again is a contradiction. Suppose ConstructBipartiteRealization() moves to a new vertex. If the chosen edge is not from Ev , the right hand side of Equation 3 goes down by 1 while j

the right hand side is unchanged. Hence vj is due at the (s, t)th stage, a contradiction. If the chosen edge is from Ev , both left hand and right hand j

sides of Equation 3 go down by 1. Hence vj is normal at the (s, t)th stage, a contradiction. (c) Assume that the (s, t)th stage is ruined. That is, there is a fat vertex vi that is not inserted yet, but the number of vertices in Mk which are available is less than ri . But, at the next stage, this number can never increase. Thus it would also be ruined.


264

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

While Lemma 1 says that once ConstructBipartiteRealization() takes a wrong path, it is impossible to mend it, the next routine gives preventive measures to avoid getting into that wrong path in the first place. ChooseCorrectEdge() Let A be not fat and ConstructBipartiteRealization() is at its (s, t)th stage with s > n1 (that is, inserting vertex vs into V2 ). Then, (1) for each vertex vj ∈ V1 , do not choose an edge in Ev if there is already j an edge (vs , vj ). (2) if the vertex vj is due, pick an edge from Ev . If many vertices are due, j pick an edge uniformly at random from the vertices that are due. Now assume that A is fat and for some integer k > 0, Fk is not empty. Then, for every entry ai ∈ Fk choose at random ri = ai − n1 + k different entries in Mk . The only condition imposed on the choice is that an entry aj can be chosen at most once for each fat vertex and at most aj times for all the fat vertices combined. If ai is k-fat, let Ri , called the reserve pool of ai , be the set of vertices in Mk chosen for ai . Let Rij , the reserve matrix, be an n1 by n2 matrix whose columns are indexed from 1 to n1 (indices of entries of A1 ), and rows are indexed from n1 + 1 to n (indices of entries of A2 ), and Rij = 1 if the entry aj ∈ Ri , and zero otherwise. Obviously, the sum of entries in row i is equal to ri and the sum of entries of column j must be less or equal to aj . At the (s, t)th stage, a vertex vj ∈ V1 is exhausted if the sum of row j plus the number of vertices adjacent to vj equals aj . (that is, the number of edges already conceded by vj and the number of edges of vj in the reserve pools equals aj ). (3) If ConstructBipartiteRealization() is at its (s, t)th stage with s > n1 and as is not fat, then apply (1) and (2) subject to not choosing a vertex vj if vj is exhausted. If as is fat, first choose all the vertices in Rs , then apply (1) and (2) if necessary.

Complexity Issues Before proving that the conditions set in routine ChooseCorrectEdge() are necessary and sufficient to sample a simple bipartite graph at random, we observe that, if A = (a1 , a2 , ..., an ) = (A1 : A2 ) where P A1 and A2 have respectively n1 and n2 entries such that n1 +n2 = n and ni=1 ai = 2m, ChooseCorrectEdge() runs altogether in O(n1 n2 ) steps. Indeed, at the sth iteration of ConstructBipartiteRealization(), ChooseCorrectEdge() has to check Equation 1 only once


Rejection sampling of bipartite graphs with given degree sequence

265

for every vertex vj ∈ V1 . But there are n2 iterations and n1 vertices vj with j ≤ n1 . This takes O(n1 n2 ) steps. Constructing the Reserve Matrix R requires O(n1 n2 ) steps as one has to check Conditions (f1), (f2) and (f3) for each of the n2 entries of A2 and writing the n1 n2 entries of the matrix R. Theorem 3 Algorithm ConstructBipartiteRealization() reconstructs a simple graph if and only if AddVertex() calls the routine ChooseCorrectEdge(). In other words, ConstructBipartiteRealization() outputs a simple graph if and only if the choice of edges satisfies Conditions (1), (2) and (3). Proof. Assume to the contrary that Conditions (1) and (2) hold but ConstructBipartiteRealization() outputs a bipartite graph G with multiple edges or loops. By Condition (1) there can not be a multiple edge connecting two vertices vj and vk such that j ≤ n1 and k > n1 . Moreover, by the definition of the routine ConstructBipartiteRealization(), there can not be a double edge (vk , vl ) where k, l > n1 . Hence if G fails to be a simple graph, it must have either a loop or a multiple edge incident to v1 and vj such that j ≤ n1 . So, in G, let the vertex v1 is incident to either a loop e or a multiple edge (v1 , vj ) such that j ≤ n1 . But, by the definition of the bipartition, the number of edges incident to v1 at the end of the nth iteration of ConstructBipartite1 Realization() equals the number of edges left to be inserted until ConstructBipartiteRealization() terminates. Hence, some vertex vk such that k > n1 is not fully inserted. This is a contradiction. Conversely, let the condition (1) or (2) be not satisfied and let G be the realization output by ConstructBipartiteRealization(). If condition (1) is not satisfied at the (s, t)th stage, this would create a double edge (vj , vs ) with j ≤ n1 and s > n1 . Now, since Algorithm Addvertex() can not concede the double edge (vj , vs ) anymore as they are lost, the double edge (vj , vs ) would appear in G. Hence G would not be simple. Assume that the condition (2) is not satisfied. That is, there is a vertex vj with j ≤ n1 that is due at the (s, t)th stage, where s > n1 , but Algorithm Addvertex() does not pick any of the elements of Ev for all the remaining edges conceded to vs . Then vj is j overdue at the insertion of vertex vs+1 , and by Lemma 1(b) it remains overdue until the end of Algorihm 2. Hence G is not simple as it must have a multiple edge (vi , vj ). If condition (3) is not satisfied, Algorithm ConstructBipartiteRealization() may stall. Let a correct edge and vertex be an edge chosen by Algorithm ChooseCorrectEdge and a vertex incident to a correct edge, respectively. So if ConstructBipartiteRealization() terminates, we have shown that it always outputs a


266

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

simple graph. It remains to show that it always terminates by showing that there is always a correct edge so that conditions (1) and (2) can be satisfied at every stage of ConstructBipartiteRealization(). Theorem 4 Algorithm ConstructBipartiteRealization() always terminates. That is, Conditions (1) and (2) are always satisfied at every stage of ConstructBipartiteRealization(). Proof. Suppose A does not contain any fat entry. That is, as long as an edge e = (v1 , vj ) is a correct vertex, it can be chosen. Obviously, Condition (1) can always be forced on AddVertex(). But, while trying hard to satisfy Condition (1), the algorithm may let a vertex vj of V1 , to become overdue. If at the (s, t)th stage the vertex vj is due, we prove that it is always possible to concede an edge from Ev to vs . j So assume to the contrary that vj is due but Addvertex() can not pick an edge from Ev . This is possible only if there are too many vertices that are due. j That is, as < n10 ≤ n1 , where n10 is the number of vertices that are due at the (s, t)th stage. But we also have as ≥ as+1 ≥ ... ≥ an . Moreover, as all these n10 vertices are due, each of them is incident to n − s available edges. Hence we have as + as+1 + · · · + an < n10 (n − s). That is, there are more available edges than there are edges left to be inserted until ConstructBipartiteRealization() terminates. This contradicts Observation 2. Let the entry ai be k-fat. If all the correct edges e = (v1 , vj ) such that vj ∈ Mk are conceded prior to the insertion of the vertex vi , then by definition of fat entry, Algorithm ConstructBipartiteRealization() would stall as there would not be enough edges to connect to vi . But, we assume that the Algorithm reserved ri edges to concede to vi . Hence vi can always be inserted. So, we only need to check (c1), whether putting some edges in reserve would prevent some non-fat vertex vs from being inserted for lack of correct edges and, (c2), whether it is always possible to construct the reserve matrix Rij . (c1) Assume that s < i. That is, vs precedes vi . Let all vertices preceding vs have been inserted but there are not enough correct edges to insert vs . This is possible if reserving edges for vertices in Fk and inserting vertices preceding vs exhausts q vertices of V1 and as > n1 − q. Without loss of generality, we may assume that the last q vertices of V1 are exhausted. So, let the available vertices be vertices v1 , . . . , vn −q+1 . If the number of available edges is less than 1 as , then A1 < A2 . This is a contradiction. So, let the number of available edges be greater or equal to as . Thus the number of available vertices is less than as , so that Condition (1) prevents as edges from being connected to vs . Let H


Rejection sampling of bipartite graphs with given degree sequence

267

be the graph obtained after the insertion of vs−1 by ’fully’ connecting all the vertices in V2 \vs , making sure to connect vertices in Fk with edges that are reserved for them in Rij . Then, by the definition of ri , it is easy to check that every vertex in Fk is adjacent to every vertex in V1 \Mk . Also, since all the vertices in Mq are exhausted after the insertion of vs−1 , one can check that none of the vertices in Mq \Mk is adjacent to a vertex in V2 \(Fk ∪ V≤s ), where V≤s denotes the set of vertices from vn +1 up to vs . ( i.e., V2 \(Fk ∪V≤s ) is the set 1 of vertices between vs and Fk ). Thus, only the vertices in V1 \Mq are adjacent to vertices in V2 \(Fk ∪ V≤s ). Since all the vertices, except for vs are properly connected and |A1 | = |A2 |, the number of available edges is as but the number of available vertices is less than as . Therefore, by the pigeonhole principle, there is an available vertex having at least two available edges. Without loss of generality, we may consider v1 to be the culprit. Now, since only the vertices in V1 \Mq are adjacent to the vertices in V2 \(Fk ∪ V≤s ), either v1 is adjacent to all the vertices in V2 \(Fk ∪ V≤s ) or it is not. If it is, then v1 was due during an iteration prior to or during the insertion of vs−1 and the algorithm did not select it to concede an edge. This is a contradiction. Suppose that it is not adjacent to some vertex vt ∈ V2 \(Fk ∪ V≤s ). Then at < n1 − q, since vt is fully connected. But, by the non decreasing ordering of A2 , we also have at ≥ ai . Moreover, since ai ∈ Fk , we have ai ≥ n1 − k. Hence we have ai ≥ n1 − k > n1 − q > at . This is also a contradiction. Therefore vertex vs can be fully inserted. See Figure 6 which helps to understand notations in part (c1). Fk

vs V2

.........

.........

Mq

.........

V1

.........

.........

.........

Mk

.........

.........

v1

Figure 6: Finally, let ai be k-fat, as be not k-fat and s > i. (that is, vs is to be inserted after vi ). If there are not enough correct edges to connect to vs , then |A1 | < |A2 |. This is a contradiction. (c2) Suppose that it is not possible to built the reserve matrix. But, since


268

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

P P ai ≤ n1 for all entries in Fk , this would imply either F ai > V \M ai + k 1 k P ai = |A1 |, or ai > n1 for some entry ai ∈ Fk . This is a contradiction. M k

It still remains to show that the algorithm constructs all the simple realizations of A. Lemma 2 Let Gn ,n be the n1 , n2 - complete bipartite graph. That is, the bi1 2 partite graph where one part contains n1 vertices each having degree n2 and the second part contains n2 vertices each of degree n1 . Then ConstructBipartiteRealization() satisfying Conditions (1) and (2) can reconstruct Gn ,n as a 1 2 realization of A = (A1 : A2 ) where A1 has n1 entries ai = n2 and A2 has n2 entries aj = n1 . Proof. At the beginning of the (n1 + 1)th iteration, Ev = n1 for each of the j n1 vertices already inserted. Hence each such vertex is due. Now, the vertex vn +1 has degree an +1 = n1 by the definition of A. Hence, by Condition (2), 1 1 AddVertex() chooses one edge from each of the n1 vertices vj with j ≤ n1 and inserts vn +1 completely. By Lemma 1, each vj is still due at the (n1 + 2)th 1 iteration. Again, by Condition (2), AddVertex() chooses one edge from each of the n1 vertices vj with j ≤ n1 and inserts vn +2 completely. And so on, until the 1 insertion of vertex vn , and Algorithm ConstructBipartiteRealization() outputs the graph Gn ,n . 1

2

Let G be a graph, a delete-minor of G 0 = G\e is the graph obtained from G by deleting the edge e. If A = (A1 : A2 ) is a bipartite degree sequence, let A 0 be the degree sequence obtained from A by subtracting 1 from two of its entries ai and aj , where ai ∈ A1 and aj ∈ A2 . Thus, if A is the degree sequence of a bipartite graph G, then A 0 is the degree sequence of some delete-minor of G. Lemma 3 If ConstructBipartiteRealization() satisfying Conditions (1) and (2) can reconstruct G as a realization of A, then it can reconstruct all the delete-minors of G that are realizations of A 0 . Proof. Let G be a bipartite graph output by Algorithm ConstructBipartiteRealization() and let G\e be a delete-minor of G. In the graph G, let the edge e be incident to vertices vj and vk having respectively degrees aj and ak , where j ≤ n1 and k > n1 . Thus in G\e, vertices vj and vk have degrees aj − 1 and ak − 1. Let f be any edge of G\e. Since G is output by ConstructBipartiteRealization(), there is a series of choices of correct edges such that f can be


Rejection sampling of bipartite graphs with given degree sequence

269

inserted. In that series of choices either e is inserted before or after f. If e is inserted after f, the same series of choices would insert f in G\e. If e is inserted before f, the same series of choices, minus the insertion of e, would also lead to the insertion of f in G\e, since Algorithm ConstructBipartiteRealization() does not need to insert any edge incident to vj and vk as their degrees are down by 1. Corollary 1 Let G be a simple bipartite realization of a degree sequence A = (A1 : A2 ) where A1 and A2 have n1 and n2 entries respectively. Then there is a positive probability that G is output by Algorithm ConstructBipartiteRealization() if Conditions (1) and (2) are satisfied. Proof. Every simple bipartite graph having one part of n1 vertices and another of n2 vertices can be obtained from Gn ,n by a series of deletions. 1

3.1

2

Sampling all bipartite realizations uniformly

Although Theorem 2 shows that the routine ConstructBipartiteRealization() can construct a realization of A in time linear on the number of edges of its realizations, we need the next result to show that it can construct any bipartite realization of A with equal probability, provided we define the probability π(e) with which AddVertex() has to insert the edge e. If at its kth iteration ConstructBipartiteRealization() is to insert the vertex vk that has degree ak , then ConstructBipartiteRealization() has to call AddVertex() that has to go through ak iterations. Let the (s, t)th stage of ConstructBipartiteRealization() be the iteration where AddVertex() inserts the tth edge of the sth vertex and let (s,t) G denote the graph obtained at that (s, t)th stage. With this notation, let (s,a ) (s) G be the graph G s . The random reconstruction tree, denoted by T , is a directed rooted tree where the root is the sole realization of the degree sequence A(1) , and the (s, t)th level contains all those possible graphs obtainable after inserting the tth edge of the sth vertex, and there is an arc from a graph H at level i to the graph G at level i + 1 if it is possible to move from H to G by the concession of a single available edge. Realizations of A are thus the leaves of the tree T . With this formalism, sampling a random bipartite realization of the degree sequence A is equivalent to performing a random walk from the root until a leaf is reached, and every step of the random walk consists of walking along a random arc of T . See Figure 7 for an illustration. Rejection sampling


270

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

Let G be a realization of A. That is, G is a leaf of the tree T . Obviously, there are many paths of T leading to G. Let p be such a path and let πp (G) denote the probability to reach G along the path p. Now πp (G) can easily be Q computed on the fly since πp (G) = e∈E(G) π(e), where E(G) denotes the set of 1 edges of G and π(e) is the probability to choose the edge e. Now π(e) = |Vcor |, where Vcor is the set of all correct vertices at the insertion of e. The only problem is that G can be reached from many paths. The next result proves that all these paths have equal probability. Lemma 4 Let G be a realization of A that can be reached through the paths p and q of T . Then πp (G) = πq (G). Proof. Let E(G) denote the set of edges of G. Then, p can be seen as a reordering of a subset of edges chosen along q. Now, since the vertices are added in the same order along q as along p, we may only consider the case where p and q differ on a single vertex and edges e and f are interchanged in p and q. Let Vcor (e) and Vcor (f) denote the sets of correct vertices at the insertion of e and f, respectively. If the Algorithm can choose either the edge e or f, then Vcor (e) = Vcor (f) and the probability to choose either must be the same. Lemma 4 allows to compute π(G) on the fly. For any path p leading to G, we have Y Y 1 π(G) = π(e) = , |Vcorr (e)| e∈G

e∈G

where Vcorr (e) is the set of vertices in V1 that are incident to some correct n edge. Hence, to get π(G) on the fly, one set π(G) = π(G 1 ) = 1. For every n (i) partial realization G from (G 1 ) to G multiply π(G) by |V 1 (e)| . Finally corr output π(G) with G. Now let min(π) be a lower bound of the probabilities to reach of the realizations of A. This lower bound can be calculated using only parameters of A. Indeed, if |Vav (e)| stands for the number of vertices in V1 that are adjacent to v1 at the insertion of edge e, then we have the inequality 1 1 |V (e)| ≤ |V (e)| ≤ π(e) and, for any realization G, we have av

corr

Y e∈G

Y 1 ≤ π(e) ≤ π(G). |Vav (e)| e∈G

Finally, since |Vv (e)| ≤ n1 and every realization of A has m edges, we get 1

Y Y 1 1 ≤ ≤ π(e) ≤ π(G). m n1 |Vav (e)| e∈G

e∈G


Rejection sampling of bipartite graphs with given degree sequence

w4 w3 w2 (4,3,3: 3,3,2,2)

v3 v1

w4 w3 w2

v2

v1

w4 w3 w2

w1

v1

w1

w4 w3 w2

v3 v2 1

w3 w2

w1

v3 v1

v2

v3

v2

v1

v2 1

w3 w2 w 1

w1

(6,3,3, 3,3,2)

w3 w2

v1

v2

v2

w3 w2 w1

w3 w2 w 1

w1

v2 1

v1 2

v3 v1

v3

v2 1

1

w3 w2

w1

v3

v3 v1

v3

v3 v1

v1

v2

v2 2

2

w2

(a ,a ) 1

v1

v2

w4 w3 w2 w 1

v1

v2

v3

v3

v3

( a1 , a7 )

w4 w3 w2 w1

w4 w3 w2 w 1

w1

271

6

w1

v3

(8,3,3, 3,3)

v1

v2 2

( a1 , a5 )

w1

v3

(11,3,3, 3) 3

( a1 , a4 )

v1

v2 v3

(14,3,3) 4

v1

( a1 , a3 )

(17,3) ( a1 , a2 )

v2

7

v1

v2

(20)

10

v1

Figure 7: Random reconstruction tree of (4, 3, 3 : 3, 3, 2, 2). The level of T on the same height as the degree sequence A(i) corresponds to all the graphs having A(i) as their degree sequence. The arrows that are crossed denote the edges that would not lead to a simple realization.


272

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

Algorithm RejectionSampling() Input: Bipartite degree sequence A = (A1 : A2 ), where A1 and A2 have n1 and n2 entries respectively such that n1 + n2 = n and an integers r1 . Output: A sequence of r1 bipartite simple realizations of A where every realization has equal probability. Step 1 Put A1 and A2 in non decreasing order. Step 2 Construct the recursion chain of A by calling the routine ConstructBipartiteRecursionChain(). Step 3 Call ConstructBipartiteRealization() to construct the realization G. Let π(G) be the probability computed on the fly and get u, a random number in (0, 1). If u < min(π) π(G) , accept G and go back to Step 3 until one gets r1 realizations. Else, reject G and go back to Step 3 until one gets r1 realizations. Obviously, Algorithm RejectionSampling() samples every realization of A with the same probability equal to min(π). Now, it is known that Step 1 takes log(n1 ) + log(n2 ) iterations and, as shown earlier, Step 2 takes n1 + n2 iterations. In Step 3, ChooseCorrectEdge() does O(n1 n2 ) inequality checks altogether while AddVertex() needs 2m iterations to insert all the vertices. Thus, the overall running time to get the minimum probability is given by log(n1 ) + log(n2 ) + r1 (n1 n2 + 2m) = O(r1 (n1 n2 + 2m)) O(3r1 m) O(m). Finally, T , the running time of generating a realization of A uniformly, is a 1 geometric random variable with expected running time given by (π(acc) where π(acc) is the acceptance probability for the realization G with the highest probability of being output by ConstructBipartiteRealization(). So π(acc) =

min(π) min(π) . =Q π(G) |Vcorr (e)| e∈G

Now if n2 → ∞, then |Vcorr (e)| →

n1 2

on average. Therefore, 1

nm min(π) 1 1 = = m. π(acc) → 2 m 2 m 2 (n ) (n ) 1

1

Hence T → 2m . For the typical Darwin tables m is about 40 edges. Thus 2m is a manageable running time.


Rejection sampling of bipartite graphs with given degree sequence

273

Acknowledgements The first author would like to express his gratitude to the University of Bristol, the third to University of Kashmir, India, and SERB-DST for continuous support in research.

References [1] M. Bayati, J. H. Kim and A. Saberi, A sequential algorithm for generating random graphs, Algorithmica, 58 (2010), 860–910. [2] E. A. Bender and E. R. Canfield, The assymptotic number of labelled graphs with given degree sequence, J. Combin. Theory, Ser A., 24 (3) (1978), 296–307. [3] J. Blitzstein and P. Diaconis, A sequential importance sampling algorithm for generating random graphs with prescribed degree sequence, Internet Math., 6 (4) (2011), 489–522. [4] F. Boesch and F. Harary, Line removal algorithms for graphs and their degree lists, IEEE Trans. Circuits Syst. CAS, 23 (12) (1976), 778–782. [5] B. Bollobas, A probalistic proof of an assymptotic formula for the number of labelled regular graphs, European J. Combin., 1 (4) (1980), 311–316. [6] R. A. Brualdi, Matrices of zeroes and ones with fixed row and column sum vectors, Linear Algebra Appl., 33 (1980), 159–231 [7] T. Brylawsky and J. Oxley, The Tutte polynomial and its applications, in N. White, ed., Matroid Applications, Encyclopedia of Mathematics and its Applications, Cambridge University Press, (1992) 123–225. [8] Y. Chen, P. Diaconis, S. Holmes and J. S. Liu, Sequential Monte Carlo methods for statistical analysis of tables, J. Amer. Stat. Assoc., 100 (2005), 109–120. [9] G. W. Cobb and Y. Chen, An application of Markov Chains Monte Carlo to community ecology, American Math. Monthly, 110 (2003), 265–288. [10] C. Cooper, M. Dyer and C. Greenhill, Sampling regular graphs and Peerto-Peer network, Combinatorics, Probability and Computing, 16 (2007), 557–594.


274

K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan

[11] P. Diaconis and A. Gangolli, Rectangular arrays with fixed margins. In Discrete Probability and Algorithms (Minneapolis, MN, 1993), IMA Vol. Math. Appl. 72, 15–41, New York Springer, 1995. [12] P. Diaconnis and B. Sturmfels, Algebraic Algorithms for sampling from conditional distributions, Annal. Statist., 26 (11) (1998), 363–3977. [13] P. Erdős and T. G. Gallai, Graphs with prescribed degrees of vertices (Hungarian), Mat. Lapok, 11 (1960), 264–274. [14] P. L. Erdős, I. Miklós and Z. Toroczkai, A simple Havel-Hakimi type algorithm to realize graphical degree sequences of directed graphs, Electron. J. Combin., 17(1) (2010), #R66. [15] M. Gail and N. Mantel, Counting the number of r x c contingency tables with fixed margins, J. American Stat. Assoc., 72 (1977), 859–862. [16] G. Guenoche, Counting and selecting at random bipartite graphs with fixed degrees, Revue francaise d’automatique, d’informatique et de recherche operationelle, 24 (1) (1990), 1–14. [17] J. Guillaume and M. Latapy, Bipartite Graphs as models of complex networks, Preprint. [18] M. Jerrum and A. Sinclair, Approximating the permanent, SIAM J. Comput., 18 (6) (1989) 1149–1178. [19] M. Jerrum and A. Sinclair, Fast uniform generation of regular graphs, Theoretical Computer Science, 73 (1) (1990), 91–100. [20] M. Jerrum and A. Sinclair, The Markov Chain Monte Carlo method: an approach to approximate counting and integration, in Approximation Algorithms for NP-hard problems, (D. S. Hochbaum ed.) PWS Publishing House, 1995, 482–519. [21] R. Kannan, P. Tetali and S. Vempala, Simple Markov-chain algorithms for generating bipartite graphs and tournaments, Random Struct. Algorithms, 14 (4) (1999), 293–308. [22] K. K. Kayibi, S. Pirzada and T. A. Chishti, Sampling contingency tables, AKCE International Journal of Graphs and Combinatorics, in press. [23] H. Kim, Z. Toroczkai, P. L. Erdős, I. Miklós and L. A. Székely, Degreebased graph construction, J. Phys. A: Math. Theor., 42 (2009), 392001.


Rejection sampling of bipartite graphs with given degree sequence

275

[24] M. Luby, D. Randall and A. Sinclair, Markov chain algorithms for planar lattice structures, SIAM J. Comput., 31 (1) (2001), 167–192. [25] B. McKay and N. C. Wormald, Uniform generation of random regular graphs of moderate degree. J. Algorithms, 11 (1) (1990), 52–67. [26] L. McShine, Random sampling of labeled tournaments, Electron. J. Combin., 7 (2000), #R8. [27] I. Miklós, P. L. Erdös and L. Soukup, Towards random uniform sampling of bipartite graphs with given degree sequence, Electron. J. Combin., 20, 1 (2013), P16. [28] M. Molloy and B. Reed, A critical point for random graphs with a given degree sequence, Random Struct. Algorithms, 6 (1995), 161–180. [29] M. E. J. Newman, A. L. Barabasi and D. J. Watts, The structure and Dynamics of networks (Princeton Studies in Complexity, Princeton UP) (2006) pp 624. [30] S. Pirzada, An introduction to Graph Theory, Universities Press, OrientBlackSwan, Hyderabad (2012). [31] A. Roberts and L. Stone, Island-sharing by archipelago species, Oecologica, 83 (1990), 560–567. [32] H. J. Ryser, Combinatorial properties of matrices of zeros and ones, Can. J. Math., 9 (1957), 371–377. [33] A. Sinclair, Convergence rates of Monte Carlo experiments, in Numerical Methods for Polymeric Systems, (S. G. Whittington ed.) IMA volumes, 639–648. [34] V. V. Vazirani, Approximation algorithms, Springer-Verlag, Berlin, Heildelberg, New York, 2003. [35] F. Viger and M. Latapy, Efficient and simple generation of random simple connected graphs with prescribed degree sequence, Lecture Notes in Computer Science 3595 (2005), 440–449. [36] N. Wormald, Models of random regular graphs, In Surveys in Combonatorics, 1999 (Canterbury), Cambridge University Press, London Math. Soc. Lecture Note Ser. 267, 239–298. Received: March 28, 2018


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 276–286 DOI: 10.2478/ausm-2018-0021

Some sufficient conditions for certain class of meromorphic multivalent functions involving Cho-Kwon-Srivastava operator S. K. Mohapatra

T. Panigrahi

Department of Mathematics, School of Applied Sciences, KIIT Deemed to be University, Odisha, India email: susanta.k.mohapatra1978@gmail.com

Department of Mathematics, School of Applied Sciences, KIIT Deemed to be University, Odisha, India email: trailokyap6@gmail.com

Abstract. Making use of a meromorphic analogue of the Cho-KwonSrivastava operator for normalized analytic functions, we introduce below a new class of meromorphic multivalent function in the punctured unit disk and obtain certain sufficient conditions for functions to belong to this class. Some consequences of the main result are also mentioned.

1 Let

Introduction and motivation P p

denote the class of functions of the form: ∞

f(z) =

X 1 + ak−p zk−p p z

(p ∈ N := {1, 2, 3, · · · })

(1)

k=1

which are analytic in the punctured unit disk: U∗ := {z : z ∈ C, 0 < |z| < 1} = U \ {0}, 2010 Mathematics Subject Classification: 30C45 Key words and phrases: meromorphic multivalent function, starlike function, convex function, close-to-convex function, Jack’s lemma, Cho-Kwon-Srivastava operator

276


Some sufficient conditions

277

having the only pole of order p at origin. In particular for p = 1, we write P P := . 1 P P For functions f ∈ p given by (1) and g ∈ p given by ∞

X 1 g(z) = p + bk−p zk−p z

(z ∈ U∗ ),

(2)

k=1

we define f ∗ g by ∞

(f ∗ g)(z) :=

X 1 zp f(z) ∗ zp g(z) ak−p bk−p zk−p = (g ∗ f)(z) =: p + p z z

(z ∈ U∗ ),

k=1

(3) where ∗ denotes the usual Hadamard product( or convolution) of analytic functions. P Pk Pc P Let ∗p (α), p (α) and p (α) be the subclasses of the class p consists of meromorphic multivalent functions which are respectively starlike, convex and close-to-convex funcions ofPorder α (0 ≤ α < p). P Analytically, a function f ∈ p is said to be in the class ∗p (α) if and only if zf0 (z) < − > α (z ∈ U∗ ). (4) f(z) P P Similarly, a function f ∈ p is said to be in the class kp (α) if and only if zf00 (z) < −1 − 0 > α (z ∈ U∗ ). f (z)

(5)

Pc Furthermore, a function f ∈ p (α) if and only if f is of the form (1) and satisfies f0 (z) < − −p−1 > α (z ∈ U∗ ). (6) z P∗ P∗ Pk P Pc P We observe that (α) := (α), (α) := (α), (α) := 1 1 k 1 c (α) P∗ P P P where (α), consisting of meromork (α) and c (α) are subclasses of phic univalent functions which are respectively starlike, convex and close-toconvex of order α (0 ≤ α < 1). For recent expository work on meromorphic functions see([5, 7, 11, 14, 16]). For the purpose of defining transform, Liu and Srivastava [7] studies meromorphic analogue of the Carlson-Shaffer operator [1] by introducing the func-


278

S. K. Mohapatra, T. Panigrahi

tion φp (a, c; z) given by φp (a, c; z) :=

2 F1 (a, 1; c; z) zp

∞

=:

X (a)k 1 + zk−p , p z (c)k

(7)

k=1

∗

(z ∈ U ; a ∈ C, c ∈ C \

− Z− 0 , Z0

:= {0, 1, 2, · · · })

where 2 F1 (a, 1; c; z) is the Gauss hypergeometric series and (λ)n is the Pochhammer symbol (or shifted factorial) given by 1 (n = 0) Γ (λ + n) (λ)n := = Γ (λ) λ(λ + 1)...(λ + n − 1) (n ∈ N). Recently, Mishra et al. [9] (also see [10]) defined the function φ†p (a, c; z), the generalized multiplicative inverse of φp (a, c; z) given by the relation φp (a, c; z) ∗ φ†p (a, c; z) =

1 − z)λ+p

zp (1

∗ (a, c ∈ C \ Z− 0 , λ > −p, z ∈ U ). (8)

If λ = −p + 1, then φ†p (a, c; z) is the inverse of φp (a, c; z) with respect to the Hadamard product P ∗. Using P this function φp (a, c; z), they considered an λ operator Lp (a, c) : p −→ p as follows: Lλp (a, c)f(z) : = φ†p (a, c; z) ∗ f(z) = ∞

=

2 F1 (λ

+ p, c; a; z) zp

X (λ + p)k (c)k 1 + ak−p zk−p p z (a)k (1)k

(9)

(z ∈ U∗ ).

k=1

The holomorphic version of the function φ†p (a, c; z) is given by the relation: zp 2 F1 (a, 1; c; z) ∗ φ†p (a, c; z) :=

zp (1 − z)λ+p

(a, c ∈ C \ Z− 0 , λ > −p; z ∈ U),

and the associated transform Lλp (a, c)f(z) = φ†p (a; c; z) ∗ f(z) were studied by Cho et al. [2]. The transform Lλp (a, c) is popularly known as the Cho-KwonSrivastava operator in literature (see, for details [4, 12, 15]). Recently, Prajapat [13] (also see [3]) introduced a class of analytic and multivalent function B(p, n, µ, α) and investigated some sufficient conditions for this class. Furthermore, Goyal and Prajapat [5] introduced the class Tp (λ, µ, α)


Some sufficient conditions

279

by making use of an extended derivative operator of Ruscheweyh type and investigated some sufficient conditions for a certain function to belong to this class. Motivated by the aforementioned work, in this paper the authors introduce a new class TpÎť,Îą (Âľ, a, c) by making use of a meromorphic analogue of Cho-KwonSrivastava operator LÎťp (a, c) for normalized multivalent analytic function as follows: P Definition 1 A function f ∈ p is said to be in the class TpÎť,Îą (Âľ, a, c) if it satisfies the following condition:

zp+1 LÎť (a, c)f(z) 0

p

Âľâˆ’1 + p < p − Îą

zp LÎť (a, c)f(z)

(10) p

(z ∈ U∗ , p ∈ N, Îť > −p, Âľ ≼ 0, 0 ≤ Îą < p, a, c ∈ C \ Z− 0) The condition (10) implies that zp+1 (LÎťp (a, c)f(z))0 < − Âľâˆ’1 > Îą zp LÎťp (a, c)f(z)

(11)

It is clear from the above P P definition that Tp−p+1,Îą (2, a, a) = ∗p (Îą) and Tp−p+1,Îą (1, a, a) = cp (Îą). In the present paper, we obtain certain sufficient conditions for functions f to be in the class TpÎť,Îą (Âľ, a, c). We need the following lemma for our investigation. Lemma 1 (see [6, 8]) Let the function w(z) be non-constant and regular in U such that w(0) = 0. If |w(z)| attains its maximum value on the circle |z| = r < 1 at a point z0 ∈ U, then z0 w0 (z0 ) = kw(z0 ), where k is real and k ≼ 1.

2

Main results

Unless otherwise stated, we mention throughout the sequel that p ∈ N, Âľ ≼ 0, Îť > −p, a, c ∈ C \ Z− 0 , 0 ≤ Îą < p.


280

S. K. Mohapatra, T. Panigrahi

P Theorem 1 If f ∈ p given by (1) satisfies anyone of the following inequalities:

" 00

zp+1 Lλ (a, c)f(z) 0 z Lλp (a, c)f(z)

p 1+p+

−

zp Lλ (a, c)f(z) µâˆ’1 (Lλp (a, c)f(z))0 p (12) 0 #

z Lλp (a, c)f(z)

−(µ − 1) p +

< p − α,

Lλp (a, c)f(z)

00 0

z(Lλ z(Lλ p (a,c)f(z)) p (a,c)f(z))

1 + p +

p + Lλ (a,c)f(z) 0 − (µ − 1) (Lλ (a,c)f(z))

p p

< p−α , 0

(2p − α)2 p+1 λ z (L (a,c)f(z))

− p λ p µâˆ’1

(z L (a,c)f(z))

(13)

p

00 0

z Lλ (a,c)f(z)) z(Lλ p (a,c)f(z))

1 + p + ( λp

− (µ − 1) p + (Lp (a,c)f(z))0 Lλ

p (a,c)f(z) 1

< , 0

2p − α zp+1 (Lλ p (a,c)f(z))

− µâˆ’1 + p

(zp Lλp (a,c)f(z))

(14)

and   0 00 z(Lλ z(Lλ p (a,c)f(z)) p (a,c)f(z))     0 − (µ − 1) p + 1 + p +  0 (Lλ Lλ  zp+1 Lλp (a, c)f(z)  p (a,c)f(z)) p (a,c)f(z)  < 1, < µâˆ’1 0   p λ p+1 λ z   (Lp (a,c)f(z)) z Lp (a, c)f(z)   + p   µâˆ’1 p λ z L (a,c)f(z) ( ) p (15) 

then f ∈ Tpλ,α (µ, a, c). Proof. Let f(z) ∈

P p

be given by (1). Define the function w(z) by

0 zp+1 Lλp (a, c)f(z) − µâˆ’1 = p + (p − α)w(z). zp Lλp (a, c)f(z)

(16)

Clearly w(z) is analytic in U with w(0) = 0. Taking logarithmic differentiation on both sides of (16) with respect to z, we obtain 00 0 z Lλp (a, c)f(z) z Lλp (a, c)f(z) (p − α)zw0 (z) 1+p+ −(µâˆ’1) p + = . (Lλp (a, c)f(z))0 Lλp (a, c)f(z) p + (p − α)w(z) (17)


Some sufficient conditions

281

From (16) and (17), we have 00 0 " z Lλp (a, c)f(z) zp+1 Lλp (a, c)f(z) φ1 (z) = − µâˆ’1 1 + p + (Lλp (a, c)f(z))0 zp Lλp (a, c)f(z) # 0 z Lλp (a, c)f(z) = (p − α)zw0 (z), −(µ − 1) p + Lλp (a, c)f(z)

00

1+p+ φ2 (z) =

z(Lλ p (a,c)f(z)) 0 (Lλ p (a,c)f(z))

(zp Lλp (a,c)f(z))

z(Lλ p (a,c)f(z)) 0 (Lλ p (a,c)f(z))

−

(19)

嵉ˆ’1

(p − α)zw0 (z) , [p + (p − α)w(z)]2 1+p+

=

zp+1 (Lλ p (a,c)f(z))

00

φ3 (z) =

0

0

− =

− (µ − 1) p +

z(Lλ p (a,c)f(z)) Lλ p (a,c)f(z)

(18)

− (µ − 1) p + 0

zp+1 (Lλ p (a,c)f(z))

嵉ˆ’1

(zp Lλp (a,c)f(z))

z(Lλ p (a,c)f(z)) Lλ p (a,c)f(z)

+p

0

(20)

zw0 (z) , w(z)[p + (p − α)w(z)]

and

00 0 z(Lλ z(Lλ p (a,c)f(z)) p (a,c)f(z)) 0 1+p+ p + Lλ (a,c)f(z) 0 − (µ − 1) (Lλ zp+1 Lλp (a, c)f(z) p (a,c)f(z)) p φ4 (z) = µâˆ’1 0 p+1 λ p λ z (Lp (a,c)f(z)) z Lp (a, c)f(z) µâˆ’1 + p (zp Lλp (a,c)f(z)) zw0 (z) = . (21) w(z) Now we claim that |w(z)| < 1 in U. For otherwise there exists a point z0 ∈ U such that max |w(z)| = |w(z0 )| = 1. (22) |z|<|z0 |

Then from Lemma 1 we find that z0 w0 (z0 ) = kw(z0 )

(k ≥ 1).

(23)

Therefore, letting w(z0 ) = eiθ in each of the equation (18) to (21), we obtain |φ1 (z0 )| = |(p − α)z0 w0 (z0 )| = |(p − α)keiθ | ≥ (p − α),

(24)


282

S. K. Mohapatra, T. Panigrahi

(p − α)z0 w0 (z0 )

= |(p − α)ke | ≥ (p − α) , |φ2 (z0 )| =

[p + (p − α)w(z0 )]2 |p + (p − α)eiθ |2 (2p − α)2

z0 w0 (z0 )

|φ3 (z0 )| =

w(z0 )[p + (p − α)w(z0 )]

1 k

=

, [p + (p − α)eiθ ] 2p − α

z0 w0 (z0 ) = k ≥ 1, < {φ4 (z0 )} = < w(z0 )

(25)

(26)

(27)

which contradicts our assumption (12) to (15), respectively. Therefore, |w(z)| < 1 holds true for all z ∈ U. Then (16) we have

zp+1 (Lλ (a, c)f(z))0

p + p

= |(p − α)w(z)| < (p − α)

zp Lλ (a, c)f(z) µ−1

p

which implies that f ∈ Tpλ,α (µ, a, c).

3

Consequences of main result

Putting a = c, λ = −p + 1, µ = 1 in Theorem 1, we get the following result: P Corollary 1 Let the function f(z) defined by (1) belong to the class p . If f(z) satisfies any one of the following inequalities:

f0 (z) zf00 (z)

z−p−1 1 + p + f0 (z) < p − α,

zf00 (z)

1 + p + f0 (z)

< p−α ,

(2p − α)2 f0 (z)

− z−p−1

zf00 (z)

1 + p + f0 (z)

1

<

2p − α ,

f0 (z)

− −p−1 − p

z

and < then f(z) ∈

Pc

p (α).

 − 

f0 (z) z−p−1

1+p+ 0

zf00 (z) f0 (z)

(z) − zf−p−1 −p

  

< 1,


Some sufficient conditions

283

Letting p = 1 in Corollary 1 we obtain the following result. P Corollary 2 If f(z) ∈ satisfies any one of the following inequalities:

0

f (z) zf00 (z)

z−2 2 + f0 (z) < 1 − α,

zf00 (z)

2 + f0 (z)

< 1−α ,

2 f0 (z)

− z−2 (2 − α)

zf00 (z)

2 + f0 (z)

0

< 1 ,

f (z)

− −2 − 1 2 − α z

   00  f0 (z) 2 + zf0 (z)  f (z)  < − −2  f0 (z) < 1,  z − −2 − 1 

and

then f(z) ∈

P

z

c (α).

Further in the special case when α = 0, Corollary 2 reduces to Corllary 3 stated below: P Corollary 3 If f(z) ∈ satisfies anyone of the following inequalities:

00 (z)

2 0 zf

−z f (z) 2 +

< 1,

f0 (z)

zf00 (z)

2 + f0 (z) 1

z2 f0 (z) < 4 ,

zf00 (z)

2 + f0 (z) ) 1

z2 f0 (z) + 1 < 2 ,

and

then f(z) ∈

P c

z2 f0 (z) zf00 (z) < 2+ 0 < 1, f (z) z2 f0 (z) + 1 P (≡ c (0)).

Letting a = c, λ = −p + 1, µ = 2 in Theorem 12, we obtain the following:


284

S. K. Mohapatra, T. Panigrahi

P Corollary 4 If f ∈ p given by (1) satisfies anyone of the following inequalities:

zf0 (z) zf00 (z) zf0 (z)

f(z) 1 + f0 (z) − f(z) < p − α,

f(z) zf00 (z) zf0 (z)

p−α

− 1 + , − <

zf0 (z) f0 (z) f(z) (2p − α)2

zf00 (z) zf0 (z)

0 1 f(z)

1 + f (z) 0

< 2p − α , zf (z) − f(z) − p

   zf0 (z) 1 +  <  f(z)

and

then f(z) ∈

P∗

zf00 (z) zf0 (z) f0 (z) − f(z) zf0 (z) f(z) + p

   < 1, 

p (α).

By putting p = 1 in Corollary 4, we have P Corollary 5 If f ∈ satisfies anyone of the following inequalities:

zf0 (z) zf00 (z) zf0 (z)

f(z) 1 + f0 (z) − f(z) < 1 − α,

00 (z) 0 (z)

f(z) zf zf 1−α

zf0 (z) 1 + f0 (z) − f(z) < (2 − α)2 ,

zf00 (z) zf0 (z)

1 + f0 (z) − f(z)

< 1 ,

zf0 (z) −1 2−α

− f(z)

   zf0 (z) 1 +  <  f(z)

and

then f(z) ∈

P∗

zf00 (z) zf0 (z) f0 (z) − f(z) zf0 (z) f(z) + 1

   < 1, 

(α).

On further setting α = 0 in Corollary 5, we get: P Corollary 6 If f(z) ∈ satisfies any one of the following inequalities:

zf0 (z) zf00 (z) zf0 (z)

− 1 + − < 1,

f(z) f0 (z) f(z)


Some sufficient conditions

285

f(z) zf00 (z) zf0 (z)

1

zf0 (z) 1 + f0 (z) − f(z) < 4 ,

zf00 (z) zf0 (z)

1 + −

1 f0 (z) f(z)

< 2, zf0 (z)

− f(z) − 1

   00 0  zf0 (z) 1 + zf0 (z) − zf (z)  f (z) f(z)   < 1, < 0 (z) zf  f(z)  +1 then f(z) ∈

P∗

f(z)

.

Acknowledgement The authors would like to thank to the editor and anonymous referees for their comments and suggestions which improve the contents of the manuscript. Further, the present investigation of the second-named author is supported by CSIR research project scheme no: 25(0278)/17/EMR-II, New Delhi, India.

References [1] B. C. Carlson, D. B. Shaffer, Starlike and pre-starlike hypergeometric functions, SIAM J. Math. Anal., 15 (1984) 737–745. [2] N. E. Cho, O. S. Kwon, H. M. Srivastava, Inclusion relationships and argument properties for certain subclasses of multivalent functions associated with a family of linear operators, J. Math. Anal. Appl., 292 (2004) 470– 483. [3] B. A. Frasin, On certain classes of multivalent analytic functions, J. Inequal. Pure Appl. Math., 6 (4) (2005) 1–11. [4] F. Ghanim M. Darus, Some properties on a certain class of meromorphic functions related to Cho-Kwon-Srivastava operator, Asian-European J. Math., 5 (4) (2012), Art. ID 1250052 (9 pages). [5] S. P Goyal, J. K. Prajapat, A new class of meromorphic multivalent functions involving certain linear operator, Tamsui Oxford J. Math. Sci., 25 (2) (2009) 167–176. [6] I. S. Jack, Functions starlike and convex of order α, J. London Math. Soc., 2 (3) (1971) 469–474.


286

S. K. Mohapatra, T. Panigrahi

[7] J. L. Liu, H. M. Srivastava, A linear operator and associated families of meromorphically multivalent functions, J. Math. Anal. Appl., 259 (2001), 566–581. [8] S. S. Miller, P. T. Mocanu, Second order differential inequalities in the complex plane, J. Math. Anal. Appl., 65 (1978), 289–305. [9] A. K. Mishra, T. Panigrahi, R. K. Mishra, Subordination and inclusion theorems for subclasses of meromorphic functions with applications to electromagnetic cloaking, Math. Comput. Modelling, 57 (2013), 945–962. [10] A. K. Mishra, M. M. Soren, Certain subclasses of multivalent meromorphic functions involving iterations of the Cho-Kwon-Srivastava transform and its combinations, Asian-European J. Math., 7 (4) (2014), (21 pages). [11] M. L. Morga, Hadamard product of certain meromorphic univalent functions, J. Math. Anal. Appl., 157 (1991), 10–16. [12] J. Patel, N. E. Cho, H. M. Srivastava, Certain subclasses of multivalent functions associated with a family of linear operators, Math. Comput. Modelling, 43 (2006), 320–338. [13] J. K. Prajapat, Some sufficient conditions for certain class of analytic and multivalent functions, Southeast Asian Bull. Math, 34 (2010), 357–363. [14] R. K. Raina, H. M. Srivastava, A new class of meromorphically multivalent functions with applications of generalized hypergeometric functions, Math. Comput. Modelling, 43 (2006), 350–356. [15] Z.-G. Wang, H.-T. Wang, Y. Sun, A class of multivalent non-Bazelevic functions involving the Cho-Kwon-Srivastava operator, Tamsui Oxford J. Math. Sci., 26 (1) (2010), 1–19. [16] N. Xu, D. Yang, On starlikeness and close to convexity of certain meromorphic function, J. Korean Soc. Math. Edus. Ser. B, Pure Appl. Math., 10 (2003), 566–581.

Received: July 5, 2017


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 287–297 DOI: 10.2478/ausm-2018-0022

Alternative proofs of some formulas for two tridiagonal determinants Feng Qi

Ai-Qi Liu

Institute of Mathematics, Henan Polytechnic University, China College of Mathematics, Inner Mongolia University for Nationalities, China Department of Mathematics, College of Science, Tianjin Polytechnic University, China email: qifeng618@gmail.com, qifeng618@hotmail.com, qifeng618@qq.com https://qifeng618.wordpress.com

Department of Mathematics, Sanmenxia Polytechnic, China email: smxptliu@hotmail.com

Abstract. In the paper, the authors provide five alternative proofs of two formulas for a tridiagonal determinant, supply a detailed proof of the inverse of the corresponding tridiagonal matrix, and provide a proof for a formula of another tridiagonal determinant. This is a companion of the paper [F. Qi, V. Čerňanová, and Y. S. Semenov, Some tridiagonal determinants related to central Delannoy numbers, the Chebyshev polynomials, and the Fibonacci polynomials, Politehn. Univ. Bucharest Sci. Bull. Ser. A Appl. Math. Phys. 81 (2019), in press.

2010 Mathematics Subject Classification: Primary 47B36; Secondary 11B39, 11B83, 11C08, 11C20, 11Y55, 15B36, 26C99 Key words and phrases: alternative proof, tridiagonal determinant, tridiagonal matrix, inverse

287


288

1

F. Qi, A.-Q. Liu

Introduction

For c ∈ C and k ∈ N, define the k × k tridiagonal matrix Mk (c) by   c 1 0 0 ··· 0 0 0 1 c 1 0 · · · 0 0 0   0 1 c 1 · · · 0 0 0     Mk (c) =  ... ... ... ... . . . ... ... ...    0 0 0 0 · · · c 1 0   0 0 0 0 · · · 1 c 1 0 0 0 0 · · · 0 1 c k×k and denote the determinant |Mk (c)| of the k × k tridiagonal matrix Mk (c) by Dk (c). In [7, Remark 4.4], the explicit expression k ` 1 X ` 2` (−1) 6 Dk (−6) = k k−` 6 `=0

was derived from some results in [7, Theorem 1.2] for the Cauchy products of central Delannoy numbers, where qp = 0 for q > p ≥ 0. For information on central Delannoy numbers, please refer to the papers [6, 7] and plenty of references cited therein. In [7, Remar 4.4], the authors guessed that the explicit formula X k k X ` k ` 2`−k m k−2m k − m Dk (c) = (−1) (−1) c = (−1) c (1) k−` m `=0

m=0

should be valid for all c ∈ C and k ∈ N and claimed that the equality (1) can be verified by induction on k ∈ N straightforwardly. In the paper [6], the authors discovered a generating function of the sequence Dk (c), provided an analytic proof of the explicit formula (1), established a simple formula for computing the tridiagonal determinant Dk (c), found a determinantal expression for Dk (c), presented the inverse of the symmetric tridiagonal matrix Mk (c), connected Dk (c) with the Chebyshev polynomials [6, 9, 11] and the Fibonacci numbers and polynomials [1, 6, 8], reviewed computation of general diagonal determinants, supplied two new formulas for computing general diagonal determinants, generalized central Delannoy numbers [6, 7], and represented the Cauchy product of the generalized central Delannoy numbers [6] in terms of Dk (c). In this paper, we pay our attention on the following four conclusions.


Some formulas for two tridiagonal determinants

289

Theorem 1 ([6, Theorem 2.2]) For k ≥ 0 and c ∈ C, the formula (1) is valid. Theorem 2 ([6, Theorem 3.1]) For c ∈ C, α = β1 = the tridiagonal determinant Dk (c) can be computed by  αk+1 − βk+1     α−β , Dk (c) = k + 1,   (−1)k (k + 1),

√ c+ c2 −4 2

, and k ≥ 0,

c 6= ±2; (2)

c = 2; c = −2.

Theorem 3 ([6, Theorem 5.1]) For k ∈ N, the inverse of the symmetric R (c) = , where tridiagonal matrix Mk (c) can be computed by M−1 ij k k×k  i − µi λk−j+1 − µk−j+1 λ  − ,    (λ − µ)(λk+1 − µk+1 )  Rij = (−1)i+j i(k − j + 1) ,  k+1    i(k − j + 1)  − , k+1

c 6= ±2 c=2 c = −2

for i < j, Rij = Rji for i > j, and λ and µ are defined by λ=

1 1 2 = −α = − . =√ µ β c2 − 4 − c

Theorem 4 ([6, Section 8]) For n ∈ N and a, b, c ∈ C, we have

a

c

0

Dn = .

..

0

0

b 0 · · · 0 0

a b · · · 0 0

c a · · · 0 0

.. .. . . . .

. .. ..

. . 0 0 · · · a b

0 0 · · · c a n×n  √ √ 2 − 4bc n+1 − a − a2 − 4bc n+1 a + a    √ , n 2n+1 a2 − 4bc =  a  (n + 1) , 2

a2 6= 4bc; (3) a2 = 4bc.


290

F. Qi, A.-Q. Liu

In Section 2 of this paper, we will supply two alternative proofs of Theorem 1. In Section 3, we will provide three alternative proofs of Theorem 2. In Section 4, we will present a detailed proof of Theorem 3. In Section 5, we will provide a proof of Theorem 4. In the last section of this paper, we will list several remarks.

2

Two alternative proofs of Theorem 1

Now we are in a position to supply two alternative proofs of Theorem 1. Proof. [First alternative proof of Theorem 1] Let D0 (c) = 1. Theorem 2.1 in [6] states that the sequence Dk (c) for k ≥ 0 can be generated by ∞

X 1 Fc (t) = 2 = Dk (c)tk . t − ct + 1

(4)

k=0

By the formula for the sum of a geometric progression, the generating function Fc (t) can be expanded as ∞ ∞ X̀ X ` X ` 2 m ` Fc (t) = (−1) t − ct = (−1) c`−m t`+m (5) m `=0 `=0 m=0

for t2 − ct < 1. Hence, it follows for k ≥ 0 that [Fc (t)]

(k)

∞ X̀ X

` `−m `+m (k) (−1) c t = m `=0 m=0 ∞ X̀ X (k) m ` → (−1) c`−m lim t`+m t→0 m `=0 m=0 k X ` k ` = (−1) k! (−1) c2`−k k−` m

`=0

for t2 − ct < 1 and as t → 0. The formula (1) is thus proved. Proof. [Second alternative proof of Theorem 1] Taking k = ` + m in (5) leads to " k # ∞ X ∞ X X ` k−` 2`−k k Fc (t) = (−1) c t = Dk (c)tk k−` k=0 `=0 k=0

2

for t − ct < 1. The formula (1) is proved again. The proof of Theorem 1 is complete.


Some formulas for two tridiagonal determinants

3

291

Three alternative proofs of Theorem 2

We now start out to provide three alternative proofs of Theorem 2. Proof. [First alternative proof of Theorem 2] It is clear that the generating 1 1 function Fc (t) in (4) can be rewritten as Fc (t) = t−α t−β . By virtue of the Leibniz theorem for the product of two functions, we have (k) X (`) (k−`) k 1 1 k 1 1 (k) = [Fc (t)] = t−αt−β t−α t−β ` `=0 k k X X k (−1)` `! (−1)k−` (k − `)! k (−1)` `! (−1)k−` (k − `)! = → ` (t − α)`+1 (t − β)k−`+1 ` (−α)`+1 (−β)k−`+1 `=0 `=0 k k X 1 1 k! X β ` k! 1 − (β/α)k+1 αk+1 − βk+1 = k! = = k = k! `+1 k−`+1 k α β β α β 1 − β/α α−β `=0

`=0

as t → 0. The formula (2) is thus proved. Proof. [Second alternative proof of Theorem 2] The generating function Fc (t) can also be rewritten as 1 1 1 − . (6) Fc (t) = α−β t−α t−β Then a straightforward computation reveals 1 (−1)k k! (−1)k k! (k) [Fc (t)] = − α − β (t − α)k+1 (t − β)k+1 1 1 αk+1 − βk+1 1 → −k! = k! − α − β αk+1 βk+1 α−β as t → 0. The proof of Theorem 2 is complete. Proof. [Third alternative proof of Theorem 2] The formula for the sum of a geometric progression yields ∞ ∞ X X 1 tk 1 tk =− = − and t−α αk+1 t−β βk+1 k=0

k=0

for |t| < min{|α|, |β|}. Thus, in view of αβ = 1 and (6), we obtain ∞ ∞ ∞ X 1 1 X 1 αk+1 − βk+1 k X k Fc (t) = − t = t = Dk (c)tk α−β βk+1 αk+1 α−β k=0

k=0

k=0

for |t| < min{|α|, |β|}. The formula (2) is thus proved. The proof of Theorem 2 is complete.


292

4

F. Qi, A.-Q. Liu

A detailed proof of Theorem 3

We now present a detailed proof of Theorem 3. In the paper [2], the inverse of the symmetric tridiagonal matrix Mk (c) was discussed. We denote the inverse matrix of Mk (c) by M−1 k (c) = (Rij )k×k . Then, basing on discussions in [2, Eq. (9)], one can see without difficulty that the elements Rij can be represented as

Rij = (−1)i+j

Di−1 (c)Dk−j (c) , Dk (c)

1≤i<j≤k

and Rij = Rji for 1 ≤ j < i ≤ k. Making use of the formula (2) yields  αi−1+1 − βi−1+1 αk−j+1 − βk−j+1     α−β α−β   , (−1)i+j k+1 − βk+1 α Rij =  α−β    i−1 (i − 1 + 1)(±1)k−j (k − j + 1)  (±1) (−1)i+j  , (±1)k (k + 1)  αi − βi αk−j+1 − βk−j+1   (−1)i+j , c 6= ±2 (α − β)(αk+1 − βk+1 ) =  i(k − j + 1)  (−1)i+j (±1)i−j−1 , c = ±2 k+1  (−α)i − (−β)i (−α)k−j+1 − (−β)k−j+1   − ,    [(−α) − (−β)][(−α)k+1 − (−β)k+1 ]  = (−1)i+j i(k − j + 1) ,  k+1     − i(k − j + 1) , k+1  i i λk−j+1 − µk−j+1 λ − µ   − , c 6= ±2    (λ − µ)(λk+1 − µk+1 )  = (−1)i+j i(k − j + 1) , c=2  k+1     − i(k − j + 1) , c = −2 k+1 for 1 ≤ i < j ≤ k. The proof of Theorem 3 is complete.

c 6= ±2

c = ±2

c 6= ±2 c=2 c = −2


Some formulas for two tridiagonal determinants

5

293

A proof of Theorem 4

The determinant Dn satisfies the recurrence relation Dn = aDn−1 − bcD . √ n−2 a+ a2 −4bc 2 Solving the equation x − ax + bc = 0 reaches to two roots α = 2 √

2

and β = a− a2 −4bc . These two roots satisfy α + β = a and αβ = bc. Then by the above recurrence relation one can write Dn − αDn−1 = β[Dn−1 − αDn−2 ] = β2 [Dn−2 − αDn−3 ] = · · · = βn−2 [D2 − αD1 ] = βn−2 [(a2 − bc) − αa] = βn . Similarly, one can deduce that Dn − βDn−1 = αn . Accordingly, when α 6= β, that is, a2 6= 4bc, one finds (α − β)Dn = αn+1 − βn+1 , that is, a+ αn+1 − βn+1 Dn = = α−β

a2 − 4bc

n+1

− a−

√ 2n+1 a2 − 4bc

a2 − 4bc

n+1 .

When α = β, that is, a2 = 4bc, we have Dn = αn + αDn−1 = αn + α(αn−1 + αDn−2 ) = · · · = (n − 1)αn + αn−1 D1 n a n n−1 n = (n − 1)α + α (2α) = (n + 1)α = (n + 1) . 2 The formula (3) is thus proved. The proof of Theorem 4 is complete.

6

Several remarks

Finally, we list several remarks on tridiagonal determinants. Remark 1 The identities

−c 1 0

2 −2c 1

0 6 −3c

.. . .. . Dk (c) , . . .

0 0 0

0 0 0

0 0 0

··· ··· ··· .. . ··· ··· ···

−(k − 2)c 1 0

(k − 1)(k − 2) −(k − 1)c 1

0 k(k − 1) −kc

0 0 0 .. .

0 0 0 .. .

0 0 0 .. .


294

F. Qi, A.-Q. Liu

0 0 0

0 0 0

0 0 0

.. .. ..

. . .

c 1 0

1 c 1

0 1 c k×k  k+1 − µk+1  k! λ  , k  λ−µ ` k! X = = k (−1)` c2` (−1)k (k + 1)!,  c k−`  `=0  (k + 1)!,

c

1

0

= (−1)k k! ...

0

0

0

1 0 0 ··· c 1 0 ··· 1 c 1 ··· .. .. .. . . . . . . 0 0 0 ··· 0 0 0 ··· 0 0 0 ···

are neither trivial nor obvious, where λ = determinant Dk (c) satisfies D0 (c) = 1,

D1 (c) = −c,

1 µ

=

c 6= ±2 c=2 c = −2

2 c2 −4 −c

= −α = − β1 . The

D2 (c) = 2(c2 − 1),

and Dk (c) = −kcDk−1 (c) − k(k − 1)Dk−2 (c), P k Then, if letting Fc (t) = ∞ k=0 Dk (c)t , we have ∞ X k=2 ∞ X

Dk (c)tk = −ct

∞ X

kDk−1 (c)tk−1 − t2

k=0

(7)

k(k − 1)Dk−2 (c)tk−2 ,

k=2

k=2

Dk (c)tk − D0 (c) − D1 (c)t = −ct

∞ X

k ≥ 2.

∞ X

(k + 1)Dk (c)tk

k=1

− t2

∞ X

(k + 2)(k + 1)Dk (c)tk ,

k=0

"∞ # "∞ # 2 X d X d Fc (t) − 1 + ct = −ct Dk (c)tk+1 − t2 2 Dk (c)tk+2 , dt dt k=1 k=0 " ∞ # " ∞ # 2 X X d k 2 d 2 k Fc (t) − 1 + ct = −ct t Dk (c)t − t t Dk (c)t , dt d t2 k=1

k=0

d d2 Fc (t) − 1 + ct = −ct [t(Fc (t) − 1)] − t2 2 t2 Fc (t) , dt dt 4 00 2 0 t Fc (t) + t (4t + c)Fc (t) + 2t2 + ct + 1 Fc (t) − 1 = 0.


Some formulas for two tridiagonal determinants

295

This means that the generating function of the sequence Dk (c) = (−1)k k!Dk (c) is the solution of the second order linear ordinary differential equation t4 f 00 (t) + t2 (4t + c)f 0 (t) + 2t2 + ct + 1 f(t) − 1 = 0 with initial values f(0) = 1 and f 0 (0) = −c. This differential equation is solvable, but its solution is not elementary. Remark 2 The method used in the proof of [6, Theorem 3.1] can not be applied to the sequence Dk (c), since its recurrence relation (7) is not a homogeneous linear recurrence relation with constant coefficients. Remark 3 The central Delannoy numbers D(k) were generalized in [10] as Z 1 b 1 1 p Da,b (k) = d t, k ≥ 0, b > a > 0 π a (t − a)(b − t) tk+1 and, by [7, Lemma 2.4], we find that Da,b (k) can be generated by 1 p

(x + a)(x + b)

=

∞ X

Da,b (k)xk .

k=0

By virtue of conclusions in [4, Section 2.4] and [3, Remark 4.1], the generalized central Delannoy numbers Da,b (k) for k ≥ 0 can be computed by 1 b 1 Da,b (k) = k+1 2 F1 k + 1, ; 1; 1 − , 2a > b > a > 0, k ≥ 0, a 2 a where 2 F1 is the classical hypergeometric function which is a special case of the generalized hypergeometric series p Fq (a1 , . . . , ap ; b1 , . . . , bq ; z)

=

∞ X (a1 )n . . . (ap )n zn n=0

(b1 )n . . . (bq )n n!

for complex numbers ai ∈ C and bi ∈ C \ {0, −1, −2, . . . }, for positive integers p, q ∈ N, and for  `−1   Q (x + k), ` ≥ 1 (x)` = k=0  1, `=0 which is called the rising factorial of x ∈ R. Remark 4 This paper and [6] are extracted from different parts of the preprint [5].


296

F. Qi, A.-Q. Liu

References [1] C.-P. Chen, A.-Q. Liu, and F. Qi, Proofs for the limit of ratios of consecutive terms in Fibonacci sequence, Cubo Mat. Educ. 5 (3) (2003), 23–30. [2] G. Y. Hu and R. F. O’Connell, Analytical inversion of symmetric tridiagonal matrices, J. Phys. A 29 (7) (1996), 1511–1513; Available online at http://dx.doi.org/10.1088/0305-4470/29/7/020. [3] F. Qi and B.-N. Guo, The reciprocal of the weighted geometric mean is a Stieltjes function, Bol. Soc. Mat. Mex. (3) 24 (1) (2018), 181–202; Available online at http://dx.doi.org/10.1007/s40590-016-0151-5. [4] F. Qi and V. Čerňanová, Some discussions on a kind of improper integrals, Int. J. Anal. Appl. 11 (2) (2016), 101–109. [5] F. Qi, V. Čerňanová, and Y. S. Semenov, On tridiagonal determinants and the Cauchy product of central Delannoy numbers, ResearchGate Working Paper (2016), available online at http://dx.doi.org/10.13140/RG.2. 1.3772.6967. [6] F. Qi, V. Čerňanová, and Y. S. Semenov, Some tridiagonal determinants related to central Delannoy numbers, the Chebyshev polynomials, and the Fibonacci polynomials, Politehn. Univ. Bucharest Sci. Bull. Ser. A Appl. Math. Phys. 81 (2019), in press. [7] F. Qi, V. Čerňanová, X.-T. Shi, and B.-N. Guo, Some properties of central Delannoy numbers, J. Comput. Appl. Math. 328 (2018), 101–115; Available online at https://doi.org/10.1016/j.cam.2017.07.013. [8] F. Qi and B.-N. Guo, Expressing the generalized Fibonacci polynomials in terms of a tridiagonal determinant, Matematiche (Catania) 72 (1) (2017), 167–175; Available online at https://doi.org/10.4418/2017.72.1.13. [9] F. Qi, D.-W. Niu, and D. Lim, Notes on the Rodrigues formulas for two kinds of the Chebyshev polynomials, HAL archives (2018), available online at https://hal.archives-ouvertes.fr/hal-01705040. [10] F. Qi, X.-T. Shi, and B.-N. Guo, Some properties of the Schröder numbers, Indian J. Pure Appl. Math. 47 (4) (2016), 717–732; Available online at http://dx.doi.org/10.1007/s13226-016-0211-6.


Some formulas for two tridiagonal determinants

297

[11] F. Qi, Q. Zou, and B.-N. Guo, Some identities and a matrix inverse related to the Chebyshev polynomials of the second kind and the Catalan numbers, Preprints 2017, 2017030209, 25 pages; Available online at https://doi. org/10.20944/preprints201703.0209.v2.

Received: June 19, 2018


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 298–318 DOI: 10.2478/ausm-2018-0023

Computing metric dimension of compressed zero divisor graphs associated to rings S. Pirzada

M. Imran Bhat

Department of Mathematics, University of Kashmir, India email: pirzadasd@kashmiruniversity.ac.in

Department of Mathematics, University of Kashmir, India email: imran bhat@yahoo.com

Abstract. For a commutative ring R with 1 6= 0, a compressed zerodivisor graph of a ring R is the undirected graph ΓE (R) with vertex set Z(RE ) \ {[0]} = RE \ {[0], [1]} defined by RE = {[x] : x ∈ R}, where [x] = {y ∈ R : ann(x) = ann(y)} and the two distinct vertices [x] and [y] of Z(RE ) are adjacent if and only if [x][y] = [xy] = [0], that is, if and only if xy = 0. In this paper, we study the metric dimension of the compressed zero divisor graph ΓE (R), the relationship of metric dimension between ΓE (R) and Γ (R), classify the rings with same or different metric dimension and obtain the bounds for the metric dimension of ΓE (R). We provide a formula for the number of vertices of the family of graphs given by ΓE (R×F). Further, we discuss the relationship between metric dimension, girth and diameter of ΓE (R).

1

Introduction

Beck [7] first introduced the notion of a zero divisor graph of a ring R and his interest was mainly in coloring of zero divisor graphs. Anderson and Livingston [3] studied zero divisor graph of non-zero zero divisors of a commutative ring 2010 Mathematics Subject Classification: 13A99, 05C78, 05C12 Key words and phrases: metric dimension, zero-divisor graph, compressed zero-divisor graph, equivalence classes

298


Metric dimension of compressed zero divisor graphs associated to rings 299 R. For a commutative ring R with 1 6= 0, let Z∗ (R) = Z(R) \ {0} be the set of non-zero zero divisors of R. A zero divisor graph Γ (R) is the undirected graph with vertex set Z∗ (R) and the two vertices x and y are adjacent if and only if xy = 0. This zero divisor graph has been studied extensively and even more the idea has been extended to the ideal based zero divisor graphs in [15, 23] and modules in [20]. Inspired by ideas from Mulay [16], we study the zero divisor graph of equivalence classes of zero divisors of a ring R. Anderson and LaGrange [4] studied this under the term compressed zero divisor graph ΓE (R) with vertex set Z(RE ) \ {[0]} = RE \ {[0], [1]}, constructed by taking the vertices to be equivalence classes [x] = {y ∈ R | ann(x) = ann(y)}, for every x ∈ R \ ([0] ∪ [1]) and each pair of distinct classes [x] and [y] is joined by an edge if and only if [x][y] = 0, that is, if and only if xy = 0. If x and y are distinct adjacent vertices in Γ (R), we note that [x] and [y] are adjacent in ΓE (R) if and only if [x] 6= [y]. It is clear that [0] = {0} and [1] = R \ Z(R) and that [x] ⊆ Z(R) \ {0}, for each x ∈ R \ ([0] ∪ [1]). Some results on the compressed zero divisor graph can be seen in [5]. For example, consider R = Z12 . Here, Z∗ (R) = {2, 3, 4, 6, 8, 9, 10} is the vertex set of Γ (R), see Fig 1(a). For the vertex set of ΓE (R), we have ann(2) = {6}, ann(3) = {4, 8}, ann(4) = {3, 6, 9}, ann(6) = {2, 4, 6, 8, 10}, ann(8) = {3, 6, 9}, ann(9) = {4, 8}, ann(10) = {6}. So, Z(RE ) = {[2], [3], [4], [6]} is the vertex set of ΓE (R), see Fig 1(b).

Figure 1: Γ (Z12 ) and ΓE (Z12 ) We note that the vertices of the graph ΓE (R) correspond to annihilator ideals in the ring and hence prime ideals if R is a Noetherian ring in which case Z(RE ) is called as the spectrum of a ring. Clearly ΓE (R) is connected and diam(ΓE (R)) ≤ 3. Also diam(ΓE (R)) ≤ diam(Γ (R)). Anderson and LaGrange [5] showed that gr(ΓE (R)) ≤ 3 if ΓE (R) contains a cycle and determined the structure of ΓE (R) when it is acyclic and the monoids RE when ΓE (R) is a star ∼ ΓE (S) for a Noetherian or finite graph. In [4], they also show that ΓE (R) =


300

S. Pirzada, M. Imran

commutative ring S. The compressed zero-divisor graph has some advantages over the earlier studied zero divisor graph Γ (R) as seen in [1, 2, 3] or subsequent zero divisor graph determined by ideal of R as seen in [15, 23]. For example, Spiroff and Wickham [[27], Proposition 1.10] showed that there are no finite regular graphs ΓE (R) for any ring R with more than two vertices. Further, they showed that R is a local ring (a ring R is said to be a local ring if it has a unique maximal ideal) if ΓE (R) is a star graph with at least four vertices. Another important aspect of studying graphs of equivalence classes is the connection to associated primes of the ring. In general, all the associated primes of a ring R correspond to distinct vertices in ΓE (R). Through out, R will denote a commutative ring with unity, U(R) its set of units. We will denote a finite field on q elements by Fq , ring of integers modulo n by Zn and all graphs are simple graphs in the sense that there are no loops. For basic definitions from graph theory we refer to [11, 17], and for commutative ring theory we refer to [6, 13]. A graph G is connected if there exists a path between every pair of vertices in G. The distance between two vertices u and v in G, denoted by d(u, v), is the length of the shortest u − v path in G. If such a path does not exist, we define d(u, v) to be infinite. The diameter of a graph is the maximum distance between any two vertices of G. The diameter is 0 if the graph consists of a single vertex. Also, the girth of a graph G, denoted by gr(G), is the length of a smallest cycle in G. Slater [25] introduced the concept of a resolving set for a connected graph G under the term locating set. He referred to a minimum resolving set as a reference set for G and called the cardinality of a minimum resolving set (reference set) the location number of G. Independently, Harary and Melter [12] discovered these concepts as well but used the term metric dimension, rather than location number. The concept of metric dimension has appeared in various applications of graph theory, as diverse as, pharmaceutical chemistry [8, 9], robot navigation [14], combinatorial optimization [24], sonar and coast guard Loran [26]. We adopt the terminology of Harary and Melter. In this paper, we study the notion of metric dimension of ΓE (R). We explore the relationship between metric dimension of ΓE (R) and ΓE (R). We obtain the metric dimension of ΓE (R) whenever it exists. We also classify the rings having the same or different metric dimension and obtain bounds for the metric dimension of ΓE (R). We also provide relationship between the metric dimension, girth and diameter of ΓE (R).


Metric dimension of compressed zero divisor graphs associated to rings 301

2

Metric dimension of some graphs ΓE (R)

Let G be a connected graph with n ≥ 2 vertices. For an ordered subset W = {w1 , w2 , . . . , wk } of V(G), we refer to the k-vector as the metric representation (locating code) of v with respect to W as r(v|W) = (d(v, w1 ), d(v, w2 ), . . . , d(v, wk )) The set W is a resolving set of G if distinct vertices have distinct metric representations (codes) and a resolving set containing the minimum number of vertices is called a metric basis for G and the metric dimension, denoted by dim(G), of G is the cardinality of a metric basis. If W is a finite metric basis, we say that r(v|W) are the metric coordinates of vertex v with respect to W. The only vertex of G whose metric coordinate with respect to W has 0 in its ith coordinate of r(v|W) is {wi }. So the vertices of W necessarily have distinct metric representations. Since only those vertices of G that are not in W have coordinates all of which are positive, it is only these vertices that need to be examined to determine if their representations are distinct. This implies that the metric dimension of G is at most n − 1. In fact for every connected graph G of order n ≥ 2, we have 1 ≤ dim(G) ≤ n − 1. For example, consider the graph G given in Figure 2. Take W1 = {v1 , v3 }. So, r(v1 |W1 ) = (0, 1), r(v2 |W1 ) = (1, 1), r(v3 |W1 ) = (1, 0), r(v4 |W1 ) = (1, 1), r(v5 |W1 ) = (2, 1). Notice, r(v2 |W1 ) = (1, 1) = r(v4 |W1 ), therefore W1 is not a resolving set. However, if we take W2 = {v1 , v2 }, then r(v1 |W2 ) = (0, 1), r(v2 |W2 ) = (1, 0), r(v3 |W2 ) = (1, 1), r(v4 |W2 ) = (1, 2), r(v5 |W2 ) = (2, 1). Since distinct vertices have distinct metric representations, W2 is a minimum resolving set and thus this graph has metric dimension 2. V1

v2

v3

v4

v5

Figure 2: dim(G) = 2 Now, we have the following observation. Lemma 1 A connected graph G of order n has metric dimension 1 if and ∼ Pn , where Pn denotes a path on n vertices of length n − 1. only if G =


302

S. Pirzada, M. Imran

∼ Pn . Let x1 − x2 − · · · − xn be a path on n vertices of G. Proof. Suppose G = Since d(xi , x1 ) = i − 1 for 1 ≤ i ≤ n, it follows {x1 } is a minimum resolving set and therefore metric basis for ΓE (R). So dim(Pn ) = 1. Conversely, let G be not a path. Then either G is a cycle or it contains a vertex v whose degree is at least 3. But, G can not be a cycle as dim(G) = 2, see ([18], Lemma 2.3). Let u1 , u2 , . . . , uk be the vertices adjacent to v. Since dim(G) = 1 and if W = {w} is a metric basis for G, then the metric representation of every vertex has a single coordinate. If d is the length of the shortest path from v to w, the coordinates of each ui with respect to W is one of {d − 1, d, d + 1}, but d(ui , w) = d can not occur for all i (1 ≤ i ≤ k). Therefore, it follows that at least two adjacent vertices of v have the same metric coordinates, which is a contradiction. Hence G is a path. A graph G(V, E) in which each pair of distinct vertices is joined by an edge is called a complete graph. A complete graph of n vertices is denoted by Kn . A graph G is said to be bipartite if its vertex set V can be partitioned into two sets V1 and V2 such that every edge of G has one end in V1 and another in V2 . A bipartite graph is complete if each vertex of one partite set is joined to every vertex of the other partite set. We denote the complete bipartite graph with partite sets of order m and n by Km,n . More generally, a graph is complete r-partite if the vertices can be partitioned into r distinct subsets, but no two elements of the same subset are adjacent. Based on the above definitions, we have the following observations. Proposition 1 The metric dimension of the compressed zero divisor graph ΓE (R) is 0 if and only if the zero divisor graph Γ (R) of R (R Z2 × Z2 ) is a complete graph. ∼ Kn , then either R = ∼ Z2 × Z2 or xy = 0 for all x, y ∈ Z∗ (R). Proof. If Γ (R) = Let v1 , v2 , . . . , vn be the zero divisors of Γ (R), then [v1 ] = [v2 ], · · · = [vn ] implies that all the vertices of Γ (R) would collapse to a single vertex in ΓE (R) and we know the metric dimension of a single vertex graph is 0. Conversely, assume that Γ (R) is not isomorphic to Kn . Then Γ (R) contains at least one vertex not adjacent to all the other vertices. Thus |ΓE (R)| ≥ 2, so that dim(ΓE (R)) ≥ 1. We can also obtain the converse part by letting dim(ΓE (R)) = 0. Then ΓE (R) = {[a]} for some a ∈ Z∗ (R), that is, ΓE (R) is a graph on a single vertex, which then implies Γ (R) is either isomorphic to a single vertex or a complete graph Kn , for all n ≥ 1. If G is a connected graph of order n ≥ 2, we say two distinct vertices u and v are distance similar, if d(u, a) = d(v, a) for all


Metric dimension of compressed zero divisor graphs associated to rings 303 a ∈ V(G) − {u, v}. It can be seen that the distance similar relation (∼) is an equivalence relation on V(G) and two distinct vertices are distance similar if either uv ∈ / E(G) and N(u) = N(v), or uv ∈ E(G) and N[u] = N[v]. Further we can find several results on metric dimension for zero divisor graphs of rings in [18, 19, 21]. Proposition 2 The metric dimension of ΓE (R) is 1 if Γ (R) is isomorphic to a complete bipartite graph Km,n , with m or n ≥ 2. Proof. Let Γ (R) be isomorphic to a complete bipartite graph Km,n with two distance similar classes V1 and V2 . Let V1 = {u1 , u2 , · · · , um } and V2 = {v1 , v2 , · · · , vn } such that ui vj = 0 for all i 6= j. Clearly, each of V1 and V2 is an independent set. We see that [u1 ] = [u2 ] = · · · = [um ] and [v1 ] = [v2 ] = · · · = [vn ], so that V1 and V2 each represents a single vertex in ΓE (R). Since the graph is connected, ΓE (R) is isomorphic to K1,1 , a path on two vertices. Therefore by Lemma 1, we have dim(ΓE (R)) = 1. Remark 1 Note that the converse of this result need not be true, the graph ∼ Z2 × Z2 , then illustrated in Fig.1 being a counter example. However, if R = ∼ ∼ ΓE (R) = K1,1 with metric dimension 1 and Γ (R) = ΓE (R). One of the important differences between Γ (R) and ΓE (R) is that the later can not be complete with at least three vertices, as seen in ([27], Proposition ∼ Kn,1 , 1.5). However, if ΓE (R) is complete r-partite, then r = 2 and ΓE (R) = for some n ≥ 1, see ([27], Proposition 1.7). A second look at the above result allows us to deduce some facts about star graphs. A complete bipartite graph of the form Kn,1 , n ∈ N ∪ {∞} is called a star graph. If n = ∞, we say the graph is an infinite star graph. Corollary 1 If R is a ring such that ΓE (R) is a star graph Kn,1 with n ≥ 2, then dim(ΓE (R)) = n − 1. Proof. First we identify a centre vertex of Kn,1 adjacent to n vertices. Then partition the vertex set V of order n + 1 into two distance similar classes, with centre vertex in one class V1 and the remaining n vertices in another class V2 which is clearly an independent set. Choose a subset of vertices W of V and u ∼ v. Then r(u|W) = r(v|W) whenever both u, v ∈ / W. Hence the metric basis contains all except at most two vertices one from each class Vi , 1 ≤ i ≤ 2. Therefore, dim(Kn,1 ) = |V(ΓE (R))| − 2 = n + 1 − 2 = n − 1. For example the metric dimension of K1,3 is 2, see Figure 3.


304

S. Pirzada, M. Imran

Figure 3: dim(K1,3 = 2) Corollary 2 If R is a commutative ring such that ΓE (R) has at least n ≥ 3 vertices, then dim(ΓE (R)) 6= n − 1. Proof. Suppose dim(ΓE (R)) = n−1, (n ≥ 3). Then, by [18, Lemma 2.2], ΓE (R) is a complete graph on n vertices which is a contradiction to the argument prior to Corollary 1. Therefore dim(ΓE (R)) 6= n − 1. Remark 2 It is not known whether for each positive integer n, the star graph Kn,1 can be realized as ΓE (R) for some ring R. However, there is a ring R = Z2 [x, y, z]/(x2 , y2 ) whose ΓE (R) is a star graph with infinitely many ends, that is, ΓE (R) is an infinite star graph. This ring also shows that the Noetherian condition is not enough to force ΓE (R) to be finite, see [27]. For n = 3, if the local ring R is isomorphic to Z4 [x]/(x2 ) or Z2 [x, y]/(x2 , y2 ) or Z4 [x, y]/(x2 , y2 , xy − ∼ K1 ,3 and therefore dim(ΓE (R)) = 2. For n = 4, if the 2, 2x, 2y), then ΓE (R) = ∼ K1,4 local ring R is isomorphic to Z8 [x, y]/(x2 , y2 , 4x, 4y, 2xy), then ΓE (R) = 2 2 ∼ and therefore dim(ΓE (R)) = 3. For n = 5, if R = Z2 [x, y, z]/(x , y , z2 , xy), ∼ K5,1 and therefore dim(ΓE (R)) = 4. This star graph K1,5 is the then ΓE (R) = smallest star graph that can be realized as ΓE (R), but not as a zero divisor graph. By definition of the compressed zero divisor graph ΓE (R) of a ring R, it is clear that each vertex in ΓE (R) is a representative of a distinct class of zero divisor activity in R. Thus, dim(ΓE (R)) ≤ dim(Γ (R)). However, the strict inequality holds if ΓE (R) has at least 3 vertices. [x,y] Example 1 In the rings R = (xZ22,xy,2x) ,R= easy to find that dim(ΓE (R)) < dim(Γ (R)).

Z4 [x,] , (x2 )

R = Z16 , R =

Z8 [x] , (2x,x2 )

it is

It will be interesting to see the family of rings in which the equality dim(ΓE (R) = dim(Γ (R) occurs.


Metric dimension of compressed zero divisor graphs associated to rings 305 A ring R is called a Boolean ring if a2 = a for every a ∈ R. Clearly a Boolean ring R is commutative with char(R) = 2, where char(R) denotes the characteristic of a ring R. More generally, a commutative ring is von Neumann regular ring if for every a ∈ R, there exists b ∈ R such that a = a2 b, or equivalently, R is a reduced zero dimensional ring, see [13, Theorem 3.1]. A Boolean ring is clearly a von Neumann regular, Q but not conversely. For example, let {Fi }i∈I be a family of fields, then Fi is always von Neumann i∈I

∼ Z2 for all i ∈ I. Also the set regular, but it is Boolean if and only if Fi = B(R) = {a ∈ R | a2 = a} of idempotents of a commutative ring R becomes a Boolean ring with multiplication defined in the same way as in R, and addition defined by the mapping (a, b) 7→ a + b − 2ab. In [13, Lemma 3.1], if r, s ∈ Γ (R), the conditions N(r) = N(s) and [r] = [s] are equivalent if R is a reduced ring, and these are equivalent to the condition rR = sR if R is a von Neumann regular ring. Furthermore, if R is a von Neumann regular ring and B(R) is the set of idempotent elements of R, the mapping defined by e 7→ [e] is isomorphism from the subgraph of Γ (R) induced by B(R) r {0, 1} onto ΓE (R) [13, Proposition 4.5]. In particular, if R is a Boolean ring (i.e., R = B(R)), then ∼ Γ (R). From this discussion, we have the following characterization. ΓE (R) = Proposition 3 Let R be a reduced commutative ring with unity. Then, metric dimension of the zero divisor graph Γ (R) equals to metric dimension of its corresponding compressed zero divisor graph if R is a Boolean ring. Note that the converse of this result is not true in general. For example, the graphs in Figure 4 being a counter example, where dim(Γ (Z6 )) = dim(ΓE (Z6 )), but R is not a Boolean ring.

Figure 4: dim(Γ (Z6 )) = dim(ΓE (Z6 )) = 1 ∼ Corollary 3 Let R and S be commutative reduced rings with unity 1. If Γ (R) = Γ (S), then dim(ΓE (R)) = dim(ΓE (S)). Remark 3 As seen in [21, Theorem 2], for the graph Γ (Πni=1 Z2 ) of a finite Boolean ring dim(Γ (Πni=1 Z2 )) ≤ n, dim(Γ (Πni=1 Z2 )) ≤ n − 1 for n = 2, 3, 4 and dim(Γ (Πni=1 Z2 )) = n for n = 5. This is also true for ΓE (R), follows by Proposition 3. The case n > 5 is still open.


306

3

S. Pirzada, M. Imran

Bounds for the metric dimension of ΓE (R)

In this section, we investigate the role of metric dimension in the study of the structure of the graph ΓE (R). We also obtain metric dimension of some special type of rings that exhibit ΓE (R). Pirzada et al [18] characterized those graphs Γ (R) for which the metric dimension is finite and for which the metric dimension is undefined [18, Theorem 3.1]. The analogous of this result is as follows. Theorem 1 Let R be a commutative ring. Then (i) dim(ΓE (R)) is finite if and only if R is finite. (ii) dim(ΓE (R)) is undefined if and only if R is an integral domain. However, dim(ΓE (R)) may be finite if R is infinite. For example, ∼ K1,3 + e (or paw graph), see Figure 5, and R = Z[x, y]/(x3 , xy) has ΓE (R)) = therefore has dim = 2.

Figure 5: The following lemma will be used to find the metric dimension of finite local rings. Lemma 2 If R is a finite local ring, then |R| = pn , for some prime p and some positive integer n. Now, we have the following results. Proposition 4 If R is a local ring with |R| = p2 and p = 2, 3, 5, then dim(ΓE (R)) is either 0 or undefined. Proof. Consider all local rings of order p2 with p a prime. According to [10, Fp [x] , and Zp2 . If R is a p. 687] local rings of order p2 are precisely Fp2 , (x2 ) ∼ F 2 , then ΓE (R) is an empty graph, which implies field of order p2 , i.e., R = p ∼ Fp [x] , or dim(ΓE (R)) is undefined. If R is not a field and |R| = p2 , i.e., R = (x2 )


Metric dimension of compressed zero divisor graphs associated to rings 307 Zp2 then ΓE (R) is a single vertex, when p = 2, 3 or 5 which then immediately gives that dim(ΓE (R)) = 0. From the above result, we also observe that dim(Γ (R)) = dim(ΓE (R)), if ∼ Z8 , Z2 [x]/(x3 ), Z4 [x]/(2x, x2 − 2). R= Proposition 5 If R is a local ring (not a field) of order (i) p3 with p = 2 or 3, then dim(ΓE (R)) is 0, and dim(ΓE (R)) = 1 only ∼ Z2 [x]/(x3 ), Z8 , Z4 [x]/(2x, x2 − 2), Z3 [x]/(x3 ), Z9 [x]/(3x, x2 − 3), if R = Z9 [x]/(3x, x2 − 6) or Z27 (ii) p4 with p = 2, then dim(ΓE (R)) is 0, 1 or 2. Proof. (i) The following is the list of all the local rings of order p3 . Fp3 ,

Fp [x, y] Fp [x] Zp 2 [x] Zp 2 [x] , , , (x, y)2 (x3 ) (px, x2 ) (px, x2 − p)

Case(a). When p = 2, the equivalence classes of the zero divisors in the local rings Z2 [x, y]/(x, y)2 and Z4 [x]/(2x, x2 ) are same and is given by [a] = {x, y, x + y} for any zero divisor a of the first ring and [b] = {2, x, x + 2} for any zero divisor b of the second ring, that is, they get collapsed to a single vertex. Therefore dim(ΓE (R)) = 0. However, ΓE (R) of the rings Z2 [x]/(x3 ), Z8 and Z4 [x]/(2x, x2 − 2) is isomorphic to the graph K1,1 , which then, by Lemma 1, gives dimE (R) = 1. Case(b). When p = 3 in the above list of local rings, we find that the compressed zero divisor graph structure of the rings Z3 [x]/(x3 ), Z9 [x]/(3x, x2 − 3), Z9 [x]/(3x, x2 − 6) and Z27 is same and is isomorphic to K1,1 . Then, by Lemma Z3 2 [x] Z3 [x, y] 1, we have dimE (R) = 1. Also, in the rings and , the equiva2 (3x, x ) (x, y)2 lence classes of all the zero divisors is same and is given by [a] = {3, 6, x, 2x, x+ 3, x + 6, 2x + 3, 2x + 6} for any non-zero zero divisor a of the first ring and [b] = {x, 2x, y, 2y, x + y, 2x + y, x + 2y, 2x + 2y} for any non-zero zero divisor b of the later ring. Thus, ΓE (R) for both rings is a graph on a single vertex and follows that dim(ΓE (R) = 0. (ii) Consider the local rings of order p4 , when p = 2. Corbas and Williams [10] conclude that there are 21 non-isomorphic commutative local rings with identity of order 16. The rings with dim(ΓE (R)) = 0 are F4 [x]/(x2 ), Z2 [x, y, z]/ (x, y, z)2 and Z4 [x]/(x2 + x + 1). The rings with dim(ΓE (R)) = 1 are Z2 [x]/(x4 ), Z2 [x, y]/(x3 , xy, y2 ), Z4 [x]/(2x, x3 − 2), Z4 [x]/(x2 − 2), Z8 [x]/(2x, x2 ), Z16 , Z4 [x]


308

S. Pirzada, M. Imran

/(x2 − 2x − 2), Z8 [x]/(2x, x2 − 2), Z4 [x]/(x2 − 2x), Z2 [x]/(x4 ) and Z2 [x]/(x4 ). Further the rings with dim(ΓE (R)) = 2 are Z4 [x]/(x2 ), Z2 [x, y]/(x2 , y2 ) and Z2 [x, y]/(x2 − y2 , xy). Now, we find the metric dimension of ΓE (Zn ). Proposition 6 Let p be a prime number. (i) If n = 2p and p > 2, then dim(ΓE (Zn )) = 1. (ii) If n = p2 , then dim(ΓE (Zn )) = 0. Proof. (i) If p = 2, since ΓE (Z4 ) is a graph with single vertex. So, dim(ΓE (Z4 ) = 0. If p > 2, the zero divisor set of Zn is {2, 2.2, 2.3, . . . , 2.(p − 1), p}. Since, char(Zn ) = 2p, it follows that p is adjacent to all other vertices. Thus the equivalence classes of these zero divisors are given by [p] = {2, 2.2, 2.3, . . . , 2.(p − 1)}, [2] = [2.2] = · · · = [2.(p − 1)] = {p}. So, the vertex set of ΓE (Zn ) is Z(RE ) = {[p], [2x]} for any positive integer x = 1, 2, . . . , p − 1. Thus ΓE (Zn ) is a path P2 which then, by Lemma 1, gives dim(ΓE (Zn ) = 1. (ii) If n = p2 and p > 2, the zero divisor set of Zn is {p, p.2, p.3, . . . , p(p − 1)}. Since char(Zn ) = p2 , it follows that the equivalence class of all these zero divisors is same and is {p, p.2, p.3, . . . , p.(p − 1)}. Thus, ΓE (R) in this case is a graph on a single vertex and therefore dim(ΓE (R) = 0. From the above result, we have the following observations. Corollary 4 Let p be a prime number (i) If n = 2p and p > 2, then |ΓE (Zn )| = 2. (ii) If n = p2 , then |ΓE (Zn )| = 1. (iii) If n = pk , k > 3 and p > 2, then |ΓE (Zn )| = k − 1. Proof. (i) and (ii) follow from Proposition 6. (iii) When n = pk , k > 3 and p ≥ 2, the zero divisors of Zn are Z(Zn ) = {upi |u ∈ U(Zn )}, for i = 1, 2, . . . , k − 1. Now the equivalence classes of zero divisors are [up] = {upk−1 }, [up2 ] = {upk−1 , upk−2 }, . . . , [upk−1 ] = {upk−1 , upk−2 , . . . , up2 , up}. In this way, we get k − 1 distinct equivalence classes. Thus, |ΓE (Zn )| = k − 1. Corollary 5 dim(ΓE (Zn )) ≤ 2k − 2, where n = pk , for any prime p > 2 and k > 3.


Metric dimension of compressed zero divisor graphs associated to rings 309 Proof. By [18, Theorem 2.1]. If G is a connected graph with G partitioned into m distance similar classes that consist of a single vertex, then dim(G) ≤ |V(G)| + m. Using part (iii) of Corollary 4, the result follows. The following important lemma, which is used later in the proof of several results, provides a combinatorial formula for the number of vertices of the compressed zero divisor graph ΓE (R × Fq ). Lemma 3 Let R be a finite commutative local ring with unity 1 and |R| = pk and let Fq be a finite prime field. Then |Z∗ ((R × Fq )E )| = 2k or 2(1 + |Z∗ (RE )|. Proof. Let R be a finite commutative local ring with unity and |R| = pk , k ≥ 1. We consider the following three cases. ∼ Fp , for some prime p. Then the zero divisor set of Z∗ (Fp ×Fq ) = Case 1. R = {{(a, 0)}, {(0, x)}}, for every a ∈ U(R) and 0 6= x ∈ Fq . Now, to find the equivalence classes of these zero divisors, the set {(a, 0)} and {(0, x)} respectively correspond to vertices [(a, 0)] and [(0, x)] in ΓE (R × Fq ), for any a ∈ U(R) and for any x ∈ Fq . Therefore, |Z∗ (R × Fq )E | = 2k, where k = 1. ∼ Zk , (k ≥ 2). The equivalence class of each element (a, 0), for Case 2. R = p every a ∈ U(R) is same, since [(a, 0)] = {(0, x)}, for all x ∈ Fq . In this way, we get one vertex of ΓE (R × Fq ). Also, the equivalence classes of each element (0, x), for every 0 6= x ∈ Fq is same, since [(0, x)] = {(a, 0)}. So, this gives another vertex of ΓE (R × Fq ). Moreover, for any unit u in R, we get two zero divisor sets of equivalence classes given by Z1 = {[(up, 0)], [(up2 , 0)], . . . , [(upk−1 , 0)]} Z2 = {[(up, 1)], [(up2 , 1)], . . . , [(upk−1 , 1)]}. We note that there is no other possible equivalence class. Claim [(upk−1 , 1)] = [(upk−1 , xi ], for all 1 ≤ i ≤ q − 2. If [(upk−1 , 1)] 6= [(upk−1 , xi ], there exists some zero divisor in R × Fq , say (a1 , 0) adjacent to (upk−1 , 1) but not adjacent to (upk−1 , xi ), which is a contradiction. The total number of zero divisors is |Z∗ ((R × F)E )| = 2 + |Z1 | + |Z2 | = 2 + k − 1 + k − 1 = 2k or 2 + 2|Z∗ (RE )| = 2(1 + |Z∗ (RE )|). Case 3. R is a local ring other than Fp and Zpk . So, we consider all local rings R with |R| = pk , especially k = 2, 3 or 5 and the rings of order p2 , p3 or p4 are mentioned in proof of Proposition 4 and 5. Then the set of zero divisors of equivalence classes include [(a, 0)], a ∈ U(R)


310

S. Pirzada, M. Imran [(0, xi )], for any i, 1 ≤ i ≤ q − 2 Z1 = {[(a1 , 0)], [(a2 , 0)], . . . , [(ar , 0)]} Z2 = {[(a1 , 1)], [(a2 , 1)], . . . , [(ar , 1)]}. where a1 , a2 , . . . , ar are the non-zero zero divisors of the set Z(RE ).

There is no other possible equivalence class as a zero divisor. Claim [(ai , 1)] = [(ai , xj )], 1 ≤ i ≤ r and 1 ≤ j ≤ q − 2. For if, [(ai , 1)] 6= [(ai , xj )], there exists some zero divisor (ak , 0) adjacent to one of [(ai , 1)] or [(ai , xj )], but not to the other, which is a contradiction. Thus, |Z∗ ((R × Fq )E )| = 2 + 2|Z∗ (RE )| = 2(1 + |Z∗ (RE )|. Example 2 Consider the ring Z8 × Z3 , Here, R = Z23 , k = 3, and U(R) = {1, 3, 5, 7}. For the zero divisors of equivalence classes, we have [(1, 0)] = {(0, 1), (0, 2)}, [(3, 0)] = {(0, 1), (0, 2)}, [(5, 0)] = {(0, 1), (0, 2)}, [(7, 0)] = {(0, 1), (0, 2)}. Also, [(2, 0)] = {(0, 1), (0, 2), (4, 0), (4, 1), (4, 2)}, [(4, 0)] = {(0, 1), (0, 2), (2, 0), (2, 1), (2, 2), (4, 0), (4, 1), (4, 2), (6, 0), (6, 1), (6, 2)}, [(6, 0)] = {(0, 1), (0, 2), (4, 0), (4, 1), (4, 2)}. Moreover, [(0, 1)] = {(1, 0), (2, 0), (3, 0), (4, 0), (5, 0), (6, 0), (7, 0)}, [(0, 2)] = {(1, 0), (2, 0), (3, 0), (4, 0), (5, 0), (6, 0), (7, 0)}, [(2, 1)] = {(4, 0)}, [(4, 2)] = {(2, 0), (4, 0), (6, 0)}, [(4, 1)] = {(2, 0), (4, 0), (6, 0)}, [(2, 2)] = {(4, 0)}, [(6, 1)] = {(4, 0)}, [(6, 2)] = {(4, 0)} Thus, |ΓE (Z8 × Z3 )| = {[(0, 1)], [(1, 0)], [(2, 0)], [(4, 0)], [(2, 1)], [(4, 1)]}. Using Lemma 3, we can directly have, |ΓE (Z8 × Z3 )| = 2 × 3 = 6. Remark 4 Lemma 3 holds if we replace Fq by any finite field F. More gener∼ R1 × R2 , ally, let R be any finite commutative ring with unity 1. We know R = where each Ri , 1 ≤ i ≤ 2, is a local ring. If either R1 or R2 is a field, the number of vertices is always given by the formula 2(1 + |Z∗ (R1E )| or 2(1 + |Z∗ (R2E )|, since the equivalence classes of zero divisors of ΓE (R1 × R2 ) are always of the form {[(0, 1)], [(1, 0)], [(a, 0)], [(0, b)], [(a, 1)], [(1, b)], [(a, b)], where a and b are the non-zero zero divisors and Z∗ (R1E ), Z∗ (R2E ) denote the number of zero divisor equivalence classes of R1 and R2 respectively. The result holds trivially if both R1 and R2 are fields. Theorem 2 Let R be a finite commutative local ring with unity 1 and finite field Fq . Then, dim(ΓE (R × Fq )) = 1 or at most 4k or 4t where k ≥ 2 and t are integers, t = 1 + |Z∗ (RE )|. Proof. Let R be a finite commutative local ring with unity 1. We consider the following three cases.


Metric dimension of compressed zero divisor graphs associated to rings 311 Case 1. R is a field. Then, by Case 1 of Lemma 3, ΓE (R × Fq ) is a path on two vertices. Therefore, by Lemma 2.1, dim(ΓE (R × Fq )) = 1. ∼ Z k , k ≥ 2. In this case, we partition the vertices into distance Case 2. R = p similar classes in ΓE (R) given by V1 = {[(a, 0)]}, for any a ∈ U(R) V2 = {[(0, x)]}, for any x ∈ Fq Z1 = {[(up, 0)]}, Z2 = {[(up2 , 0)]}, . . . , Zk−1 = {[(upk−1 , 0)]} W1 = {[(up, 1)]}, W2 = {[(up2 , 1)]}, . . . , Wk−1 = {[(upk−1 , 1)]} Then, dim(ΓE (R×Fq ) ≤ |Z∗ ((R×Fq )E )+m where m is the number of distance similar classes that consist of a single vertex. Hence by case 2 of Lemma 3, we have dim(ΓE (R × Fq ) ≤ 2k + 2(k − 1) + 2 = 4k. Case 3. R is a local ring other than Zkp and Fkp (k ≥ 1). Then, by Case 3 of Lemma 3, dim(ΓE (R×Fq )) ≤ 2(1+|Z∗ (RE )|+2|Z∗ (RE )|+2 = 4(1+|Z∗ (RE )|) = 4t where t is any integer given by t = 1 + |Z∗ (RE )|. We say that a graph G has a bounded degree if there exists a positive integer M such that the degree of every vertex is at most M. In the next theorems, we obtain an upper bound for the number of zero divisors in a finite commutative ring R with unity 1 with finite metric dimension. The analogous of these results holds in case of ΓE (R). Proposition 7 If Γ (R) is a zero divisor graph with finite metric dimension k, then |Z∗ (R)| ≤ 3k + k. Proof. Let Γ (R) be a zero divisor graph with metric dimension k. We choose two vertices, say w1 and w2 , from the metric basis W. Since the diameter of Γ (R) is at most 3, each coordinate of metric representation is an integer between 0 and 3 and only the vertices of a metric basis have one coordinate 0. The remaining vertices must get a unique code from one of the 3k possibilities. Therefore, |Z∗ (R)| ≤ 3k + k. Proposition 8 Let R be a commutative ring and ΓE (R) be a corresponding compressed zero divisor graph with |Z∗ (R)| ≥ 2. Then dim(ΓE (R)) ≤ |Z∗ (RE )| − d, where d is the diameter of ΓE (R).


312

S. Pirzada, M. Imran

Proof. By [21, Theorem 5.2], if R is a commutative ring and Γ (R) is the corresponding zero divisor graph of R such that |Z∗ (R)| ≥ 2, then dim(Γ (R)) ≤ |Z∗ (R)| − d0 where d0 is the diameter of Γ (R). Since dim(ΓE (R)) ≤ dim(Γ (R)) and |Z∗ (RE )| ≤ |Z∗ (R)|, therefore dim(ΓE (R)) ≤ |Z∗ (RE )| − d, where d is the diameter of ΓE (R).

Proposition 9 If Γ (R) is a finite graph with metric dimension k, then every vertex of this graph has degree at most 3k − 1. Proof. Let W = {w1 , w2 , . . . , wk } be a metric basis of Γ (R) with cardinality k. Consider a vertex v with metric representation (d(v, w1 ), d(v, w2 ), . . . , d(v, wk )). If u is adjacent to v, then r(v|W) 6= r(u|W) and |d(v, wi ) − d(u, wi )| ≤ 1 for all wi ∈ W, 1 ≤ i ≤ k. If d is distance from v to wi , then the distance of u from wi is one of the numbers {d, d − 1, d + 1}. Thus, there are three possible numbers for each of the k coordinates of r(u|W), but d(u, wi ) 6= d(v, wi ) for all 1 ≤ i ≤ k. This implies that there are at most 3k − 1 different possibilities for r(u|W). Since all vertices must have distinct metric coordinates, the degree of v is at most 3k − 1. ∼ ΓE (R) for some ring R. There are A graph G is realizable as ΓE (R) if G = many results which imply that most graphs are not realizable as ΓE (R), like ΓE (R) is not a cycle graph, nor a complete graph with at least three vertices. Proposition 10 The metric dimension of realizable graphs ΓE (R) with 3 vertices is 1. Proof. Spiroff et al. proved that the only one realizable graph ΓE (R) with exactly three vertices as a graph of equivalence classes of zero divisors for some ring R is P3 , see Figure 6. Clearly, its metric dimension is 1. [2]

[4]

[8]

Figure 6: Z16


Metric dimension of compressed zero divisor graphs associated to rings 313 Proposition 11 The metric dimension of realizable graphs ΓE (R) with 4 vertices is either 1 or 2. Proof. All the realizable graphs ΓE (R) on 4 vertices are shown in Figure 7. It is easy to see their metric dimension is either 1 or 2.

Figure 7: (Z4 × F4 )

Z4 [x]/(x2 )

Z[x,y] (x3 ,xy)

Proposition 12 The metric dimension of realizable graphs ΓE (R) with 5 vertices is either 2 or 3. Proof. The only realizable graphs of equivalence classes of zero divisors of a ring R with 5 vertices are shown in Figure 8. It is easy to see the metric dimension of the first three graphs is 2 and for the star graph is 3 (by Corollary 1).

Figure 8: (Z9 [x]/(x2 ), Z64 , Z3 [x, y]/(xy, x3 , y3 , x2 − y2 ), Z8 [x, y]/(x2 , y2 , 4x, 4y, 2xy)

4

Relationship between metric dimension, girth and diameter of ΓE (R)

In this section, we examine the relationship between girth, diameter and metric dimension of ΓE (R). Since gr(ΓE (R)) ∈ {3, ∞}, it is worth to mention that, for a reduced commutative ring R with 1 6= 0, gr(ΓE (R)) = 3 if and only if gr(Γ (R)) = 3 and that gr(ΓE (R)) = ∞ if and only if gr(Γ (R)) ∈ {4, ∞}. However, if R is not reduced, then we may have gr(Γ (R)) = 3 and either gr(ΓE (R)) = 3 or ∞. The following result gives the metric dimension of ΓE (R) in terms of the girth of ΓE (R) of a ring R.


314

S. Pirzada, M. Imran

Theorem 3 Let R be a finite commutative ring with gr(ΓE (R)) = ∞. (i) If R is a reduced ring, then dim(ΓE (R)) = 1. ∼ Z6 , Z8 , Z2 [x]/(x3 ) or Z4 [x]/(2x, x2 −2), then dim(ΓE (R)) = |Z∗ (RE )|− (ii) If R = 1. ∼ Z4 , Z9 , Z2 [x]/(x2 ), then dim(ΓE (R)) = 0. (iii) If R = (iv) dim(ΓE (R)) = 0 or 1 if and only if gr(Γ (R)) ∈ {4, ∞}. ∼ Z2 × A for Proof. If R is a reduced ring and R Z2 × Z2 , then we know R = some finite field A. Therefore, by Remark 4, R has two equivalence classes of zero divisors [(0, 1)] and [(1, 0)], adjacent to each other. Hence, dim(ΓE (R)) = 1. ∼ Z2 × Z2 , then R being a Boolean ring, implies Γ (R) = ∼ ΓE (R). Also, if R = Therefore, by Case 1 of Lemma 3, the result follows. In part (ii), these rings are non reduced and ΓE (R) are isomorphic to K1,1 . Rings listed in part (iii) represents ΓE (R) on a single vertex, part (iv) follows from the above comments. We can also prove the Part (i) by using the fact that if R is reduced and ∼ Z2 × A for some finite field A. Thus Γ (R) is a complete R Z2 × Z2 , then R = ∼ Z2 × Z2 , then bipartite and the result follows from Proposition 2. Now, if R = ∼ K1,1 , whose metric dimension is 1. Since, Z2 × Z2 is a Boolean ring, Γ (R) = therefore by Proposition 3, we have dim(ΓE (R)) = 1. ∼ F1 × F2 × If R is a reduced ring with non-trivial zero divisor graph, then R = · · · × Fk for some integer k ≥ 2 and for finite fields F1 , F2 , . . . , Fk . If R is not a ∼ R1 × R2 × · · · × Rt , for some integer reduced ring, then either R is local or R = t ≥ 2 and local rings R1 , R2 , . . . , Rt , where at least one Ri is not a field. Now, we have the following observations for the finite commutative rings whose zero divisor graphs can be seen in [22]. Corollary 6 If R is a finite commutative ring with unity 1 and gr(ΓE (R)) = ∞, then the compressed zero divisor graph of the reduced rings R × F where F is a finite field, is isomorphic to the compressed zero divisor graph of the following local rings with metric dimension 1, R being any local ring. Z8 , Z2 [x]/(x3 ), Z4 [x]/(2x, x2 − 2), Z2 [x, y]/(x3 , xy, y2 ), Z8 [x]/(2x, x2 ), Z4 [x]/(x3 , 2x2 , 2x), Z9 [x]/(3x, x2 − 6), Z9 [x]/(3x, x2 − 3), Z3 [x]/(x3 ), Z27 . Proof. The reduced rings R × F with gr(ΓE (R)) = ∞, all have compressed zero divisor graph isomorphic to K1,1 , by Case 2 of Lemma 3. Also, the local rings listed above have the same compressed zero divisor graph isomorphic to K1,1 .


Metric dimension of compressed zero divisor graphs associated to rings 315 Proposition 13 Let R be a finite commutative ring with 1 and gr(ΓE (R)) = ∞. The following are the non reduced rings with dim(ΓE (R)) = 1 Z2 × Z4 , Z3 × Z4 , Z4 × F4 , Z2 × Z9 , Z5 × Z4 , Z3 × Z9 , Z2 [x]/(x2 ) × F4 , Z2 ×Z2 [x]/(x2 ), Z3 ×Z2 [x]/(x2 ), Z2 ×Z3 [x]/(x2 ), Z3 ×Z3 [x]/(x2 ), Z5 ×Z2 [x]/(x2 ), Z2 × Z2 [x, y]/(x, y)2 , Z2 × Z4 [x]/(2, x)2 . ∼ R1 ×R2 × · · ·×Rk , where k ≥ 2 Proof. If R is not a local ring, we can write R = ∼ R1 × R2 , where either R1 and each Ri is a local ring. In case of above rings R = or R2 is a field. Therefore, using Remark 4, we have |ΓE (R)| = 4 and it is easy to see that ΓE (R) isomorphic to a path on 3 vertices. Thus, gr(ΓE (R)) = ∞ and dim(ΓE (R)) = 1. ∼ F1 × F2 × F3 , then it is easy to see that the three vertices [(1, 0, 0)], If R = [(0, 1, 0)] and [(0, 0, 1)] are adjacent with ends [(0, 1, 1)], [(1, 0, 1)], and [(1, 1, 0)] respectively and thus |ΓE (R)| = 6. ∼ F1 × F2 × F3 . Proposition 14 Let R be a reduced commutative ring and R = Then, gr(ΓE (R)) = 3 and dim(ΓE (R)) = 2. We now proceed to study the relationship between diameter and metric dimension of compressed zero divisor graphs. Since diam(ΓE (R)) ≤ 3, if ΓE (R) contains a cycle. We have the following results. Theorem 4 Let R be commutative ring and ΓE (R) be its corresponding compressed zero divisor graph. (i) dim(ΓE (R)) = 0 if and only if diam(ΓE (R)) = 0. (ii) dim(ΓE (R)) = 0 if and only if diam(Γ (R)) = 0 or 1, R Z2 × Z2 . ∼ F1 × F2 , where F1 and F2 are (iii) dim(ΓE (R)) = diam(ΓE (R)) = 1 if R = fields. (iv) dim(ΓE (R)) = 1 and diam(ΓE (R)) = 3, if R is non reduced ring isomorphic to the rings given in Proposition 13. (v) dim(ΓE (R)) = 0 if Z(R)2 = 0 and |Z(R)| ≥ 2. Proof. (i) dim(ΓE (R)) = 0 if and only if ΓE (R) is a single vertex graph if and only if diam(ΓE (R)) = 0.


316

S. Pirzada, M. Imran

(ii) Let dim(ΓE (R)) = 0. Then Γ (R) is complete and thus diam(Γ (R)) = 0 or 1. Conversely, let diam(Γ (R)) = 0 or 1, then Γ (R) is complete, thus dim(ΓE (R)) = 0 unless R Z2 × Z2 . ∼ F1 × F2 , then by Case 1 of Lemma 3, |ΓE (R)| = 2, since the only (iii) Let R = ∼ equivalence classes of zero divisors are [(0, 1)] and [(1, 0)]. So, ΓE (R) = K1,1 . Thus, dim(ΓE (R)) = diam(ΓE (R)) = 1. (iv) Rings listed in this case correspond to a path of length 3. (v) Let |Z(R)| ≥ 2 and (Z(R))2 = 0. Hence ann(a) = ann(b), for each a, b ∈ Z(R)∗ , which implies that diam(ΓE (R)) = 0. Therefore, dim(ΓE (R)) = 0.

Acknowledgements This research is supported by the University Grants Commission, New Delhi with research project number MRP-MAJOR-MATH-2013-8034.

References [1] S. Akbari and A. Mohammadian, On zero divisor graphs of a ring, J. Algebra, 274 (2004), 847–855. [2] D. D. Anderson and M. Naseer, Beck’s coloring of a commutative ring, J. Algebra, 159 (1993), 500–514. [3] D. F. Anderson and P. S. Livingston, The zero divisor graph of a commutative ring, J. Algebra, 217 (1999), 434–447. [4] D. F. Anderson and J. D. LaGrange, Commutative Boolean Monoids, reduced rings and the compressed zero-divisor graphs, J. Pure and Appl. Algebra, 216 (2012), 1626–1636. [5] D. F. Anderson and J.D. LaGrange, Some remarks on the compressed zero-divisor graphs, J. Algebra, 447 (2016), 297–321. [6] M. F. Atiyah and I. G. MacDonald, Introduction to Commutative Algebra, Addison-Wesley, Reading, MA (1969). [7] I. Beck, Coloring of commutative rings, J. Algebra, 26 (1988), 208–226.


Metric dimension of compressed zero divisor graphs associated to rings 317 [8] P. J. Cameron and J. H. Van Lint, Designs, Graphs, Codes and their Links, in: London Mathematical Society Student Texts, 24 (5), Cambridge University Press, Cambridge, 1991. [9] G. Chartrand, L. Eroh, M. A. Johnson and O. R. Oellermann, Resolvability in graphs and metric dimension of a graph, Disc. Appl. Math., 105 (1-3) (2000), 99–113. [10] B. Corbas and G. D. Williams, Rings of order p5 . II. Local rings, J. Algebra, 231 (2) (2000), 691–704. [11] R. Diestel, Graph Theory, 4th ed. Vol. 173 of Graduate texts in mathematics, Heidelberg: Springer, 2010. [12] F. Harary and R. A. Melter, On the metric dimension of a graph, Ars Combin., 2 (1976), 191–195. [13] J. A. Huckaba, Commutative Rings with Zero Divisors, Marcel-Dekker, New York, Basel, 1988. [14] S. Khuller, B. Raghavachari and A. Rosenfeld, Localization in graphs, Technical Report CS-TR-3326, University of Maryland at College Park, 1994. [15] H. R. Maimani, M. R. Pournaki and S. Yassemi, Zero divisor graph with respect to an ideal, Comm. Algebra, 34 (3) (2006), 923–929. [16] S. B. Mulay, Cycles and symmetries of zero-divisors, Comm. Algebra, 30 (7) (2002), 3533–3558. [17] S. Pirzada, An Introduction to Graph Theory, Universities Press, Orient Blackswan, Hyderabad, 2012. [18] S. Pirzada, Rameez Raja and S. P. Redmond, Locating sets and numbers of graphs associated to commutative rings, J. Algebra Appl., 13 (7) (2014), 1450047. [19] S. Pirzada and Rameez Raja, On the metric dimension of a zero divisor graph, Comm. Algebra, 45 (4) (2017), 1399–1408. [20] S. Pirzada and Rameez Raja, On graphs associated with modules over commutative rings, J. Korean Math. Soc., 53 (5) (2016), 1167–1182.


318

S. Pirzada, M. Imran

[21] Rameez Raja, S. Pirzada and Shane Redmond, On locating numbers and codes of zero divisor graphs associated with commutative rings, J. Algebra Appl., 15 (1) (2016), 1650014. [22] S. P. Redmond, On zero divisor graph of small finite commutative rings, Disc. Math., 307 (2007), 1155–1166. [23] S. P. Redmond, An Ideal based zero divisor graph of a commutative ring, Comm. Algebra, 31 (9) (2003), 4425–4443. [24] A. Sebo and E. Tannier, On metric generators of graphs, Math. Oper. Research., 29 (2) (2004), 383–393. [25] P. J. Slater, Dominating and reference sets in a graph, J. Math, Phys. Sci., 22 (1988), 445–455. [26] P.J. Slater, Leaves of trees, Congressus Numerantium, 14 (1975), 549– 559. [27] S. Spiroff and C. Wickham, A zero divisor graph determined by equivalence classes of zero divisors, Comm. Algebra, 39 (7) (2011), 2338–2348.

Received: August 6, 2018


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 319–328 DOI: 10.2478/ausm-2018-0024

Θ-modifications on weak spaces E. Rosas

C. Carpintero

Departmento de Ciencias Naturales y Exacta, Universidad de la Costa, Colombia Departamento de Matematicas Universidad de Oriente, Venezuela email: erosas@cuc.edu.co, ennisrafael@gmail.com

Departamento de Matematicas Universidad De Oriente, Venezuela Universidad Autónoma del Caribe, Colombia email: ccarpintero@sucre.udo.edu.ve

J. Sanabria Facultad de Ciencias Básicas, Universidad del Atlántico, Colombia email: jesanabri@gmail.com

Abstract. In this article, we want to study and investigate if it is possible to use the notions of weak structures to develop a new theory of θ - modifications in weak spaces and study their properties, finally we study some forms of weak continuity using this modifications.

1

Introduction

In [5], Császar and Makai Jr. introduced and studied the notions of δµ1 µ2 -open sets and θµ1 µ2 -open sets defined by two generalized topologies µ1 and µ2 on a nonempty set X and they proved that: δµ1 µ2 and θµ1 µ2 are generalized topologies on X and θµ1 µ2 ⊆ δµ1 µ2 ⊆ µ1 . The notions of (θw1 w2 , θσ1 σ2 )-continuous was introduced and characterized by W. K. Min in [6], also introduced and characterized the notions of (δw1 w2 , δσ1 σ2 )-continuous on generalized topological 2010 Mathematics Subject Classification: 54C05, 54C08, 54C10 Key words and phrases: weak structure, w-space, w-closure, (Θw1 w2 , Θv1 v2 )-continuous functions

319


320

E. Rosas, C. Carpintero, J. Sanabria

spaces and (δw1 w2 , θv1 v2 )-continuous. W. K. Min in [7], introduced the notions of mixed weak (µ, ν1 ν2 )-continuity between a generalized topology µ and two generalized topologies ν1 , ν2 , also he introduced and characterized continuity in terms of mixed generalized (ν1 , ν2 )0 -semiopen sets, (ν1 , ν2 )0 -preopen sets, (ν1 , ν2 )-preopen sets [4] ,(ν1 , ν2 ), -β, -open sets and θ(ν1 , ν2 ), -open sets [5] . Ugur Sengul in [12], using the δ and θ-modifications in bigeneralized topologies, introduced the notion of (δµ1 µ2 , θσ1 σ2 )-continuity between two Bi-GTSś. Also he characterized such continuity in terms of mixed generalized open sets: δµ1 µ2 -open sets, θµ1 µ2 -open sets. In this article, we want to study if it is possible, using weak structures to make a new theory related to θ -modifications of weak spaces and study some weak forms of continuity.

2

Preliminaries

Definition 1 [9] Let X be a nonempty set. A subfamily wX of the power set P(X) is called a weak structure on X if it satisfies the following: 1. ∅ ∈ wX and X ∈ wX . 2. For U1 , U2 ∈ wX , U1 ∩ U2 ∈ wX The pair (X, wX ) is called a w-space on X. An element U ∈ wX is called w-open set and the complement of a w-open set is a w-closed set Definition 2 [9] Let (X, wX ) be a w-space. For a subset A of X, T 1. The w-closure of A is defined as wC(A) = {F : A ⊆ F, X \ F ∈ wX }. S 2. The w-interior of A is defined as wI(A) = {U : U ⊆ A, U ∈ wX }. Theorem 1 [9] Let (X, wX ) be a w-space on X and A, B subsets of X. Then the following hold: 1. If A ⊆ B, then wI(A) ⊆ wI(B) and wC(A) ⊆ wC(B). 2. wI(wI(A)) = wI(A) and wC(wC(A)) = wC(A). 3. wC(X \ A) = X \ wI(A) and wI(X \ A) = X \ wC(A). 4. x ∈ wC(A) if and only if U ∩ A 6= ∅, for all U ∈ wX with x ∈ U. 5. x ∈ wI(A) if and only if there exists U ∈ wX with x ∈ U, such that U ⊆ A.


Θ-modifications on weak spaces

321

6. If A is w-closed (resp. w-open), then wC(A) = A(resp. wI(A) = A). Theorem 2 [11] Let (X, wX ) be a w-space on X and A, B subsets of X. Then the following hold: 1. wI(A ∩ B) = wI(A) ∩ wI(B). 2. wC(A ∪ B) = wC(A) ∪ wC(B). Theorem 3 Let (X, wX ) be a w-space on X and A, B subsets of X. Then the following hold: 1. wI(A) ∪ wI(B) ⊆ wI(A ∪ B). 2. wC(A ∩ B) ⊆ wC(A) ∩ wC(B).

3

Modification on weak structures

Throughout this paper if w1 , w2 are two weak structures on a nonempty set X. Then (X, w1 , w2 ) is called a biweak space. Recall that Császar, A. [3], showed that the δ and θ-modifications of topological spaces can be generalized for the case when the topology is replaced by the generalized topologies µ1 , µ2 in the sense of [1]. W. K. Min [6], gave a characterization for (θµ1 µ2 , θσ1 σ2 )-continuity and introduce the concepts of (δµ1 µ2 , δσ1 σ2 )-continuity on generalized topological spaces and investigate the relationship between (δµ1 µ2 , θσ1 σ2 )-continuity, (θµ1 µ2 , θσ1 σ2 )-continuity and (δµ1 µ2 , δσ1 σ2 )-continuity. In our case, we want to study what happen when the generalized topologies are replaced by weak structures. Definition 3 Let (X, w1 , w2 ) be a biweak space. A subset A of X is said to be Υw1 w2 -open (resp. Υw1 w2 -closed) if A = w1 I(w2 C(A)) (resp. A = w1 C(w2 I(A))). Example 1 Let (X, w1 , w2 ) be a biweak space, where X = {a, b, c}, w1 = {∅, X, {a}, {b}} and w2 = {∅, X, {a}, {c}}. Observe that the set A = {b} is Υw1 w2 -open, the set B = {c} is Υw2 w1 -open and the set C = {a, b} is not Υw2 w1 -open set. Definition 4 Let (X, w1 , w2 ) be a biweak space. 1. A ∈ θw1 w2 if and only if for each x ∈ A, there exists an U ∈ w1 such that x ∈ U ⊆ w2 C(U) ⊆ A.


322

E. Rosas, C. Carpintero, J. Sanabria

2. A ∈ δw1 w2 if and only if A ⊂ X and if x ∈ A, there exists a w2 -closed set F such that x ∈ w1 I(F) ⊆ A. Example 2 In Example 1: 1. θw1 w2 = {∅, X, {b}, {a, b}}, 2. δw1 w2 = {∅, X, {b}, {a, b}}, 3. θw2 w1 = {∅, X, {c}, {a, c}}, 4. δw2 w1 = {∅, X, {a, c}, {c}}. Example 3 Let (X, w1 , w2 ) be a biweak space, where X = {a, b, c}, w1 = {∅, X, {a}, {b}} and w2 = {∅, X, {a}, {a, b}, {c}}. Observe that: 1. θw1 w2 = {∅, X, {b}, {a, b}}, 2. δw1 w2 = {∅, X, {a, b}, {b}}, 3. θw2 w1 = {∅, X, {c}, {a, c}}, 4. δw2 w1 = {∅, X, {a, c}, {c}}. Example 4 Let (X, w1 , w2 ) be a biweak space, where X = {a, b, c}, w1 = {∅, X, {a}, {b}, {c}} and w2 = {∅, X, {a}, {a, b}, {c}}. Observe that: 1. θw1 w2 = {∅, X, {b}, {c}, {b, c}, {a, b}}, 2. δw1 w2 = {∅, X, {c}, {a, b}, {b, c}}, 3. θw2 w1 = {∅, X, {a}, {b}, {c}, {a, b}, {a, c}, {b, c}}, 4. δw2 w1 = {∅, X, {a, b}, {c}}. Example 5 Let X = {a, b, c} with weak structures w1 = {∅, X, {b}} and w2 = {∅, X, {a}}. Observe that: 1. θw1 w2 = {∅, X}, 2. δw1 w2 = {∅, X, {b}}, 3. θw2 w1 = {∅, X},


Θ-modifications on weak spaces

323

4. δw2 w1 = {∅, X, {a}}. Remark 1 According with Example 4, δw1 w2 is not necessary a weak structures on X, then first of all, we have an answer. We can not doing similarly modification as [5], if we replace generalized topology by weak structure. Theorem 4 Let (X, w1 , w2 ) be a biweak space. The collection θw1 w2 is a strong generalized topology on X. Proof. It is easy to see that: ∅ and X belong Sto θw1 w2 . Now consider {Ui : i ∈ I} a collection of elements of θw1 w2 and x ∈ i∈I Ui , then for some i ∈ SI, x ∈ Ui and then there is V ∈ w , such that x ∈ U ⊆ w C(V ) ⊆ U ⊆ i 1 i 2 i i i∈I Ui . It S follows that i∈I Ui ∈ θw1 w2 . Theorem 5 Let (X, w1 , w2 ) be a biweak space. The collection θw1 w2 is a weak structure on X. Proof. It is easy to see that: ∅ and X belong to θw1 w2 . Now consider U1 , U2 two elements of θw1 w2 and x ∈ U1 ∩ U2 , then x ∈ Ui for i = 1, 2. Then there exists Vi ∈ wi for i = 1, 2, such that x ∈ Vi and w2 C(Vi ) ⊆ Ui . It follows that x ∈ V1 ∩ V2 and w2 C(V1 ) ∩ w2 C(V2 ) ⊆ U1 ∩ U2 . But V1 ∩ V2 ∈ w1 and V1 ∩ V2 ⊆ w2 C(V1 ∩ V2 ) ⊆ w2 C(V1 ) ∩ w2 C(V2 ) ⊆ U1 ∩ U2 . Hence U1 ∩ U2 ∈ θw1 w2 . Remark 2 Observe that if (X, w1 , w2 ) is a biweak space, θw1 w2 is a topology on X. Theorem 6 Let (X, w1 , w2 ) be a biweak space. The collection δw1 w2 is a strong generalized topology on X. Proof. It is easy to see that: ∅ and X belong S to δw1 w2 . Consider {Vi : i ∈ I} a collection of elements of δw1 w2 and x ∈ i∈I Vi , then for some i ∈ I, x ∈ Vi and then there is S w2 -closed set F such that S x ∈ w1 I(F) ⊆ Vi and hence, x ∈ w1 I(F) ⊆ Vi ⊆ i∈I Vi . In consequence, i∈I Vi ∈ δw1 w2 . Remark 3 According with Example 3, θw1 w2 δw1 w2 and δw1 w2 w1 .

w1 and by Example 4, θw1 w2

Remark 4 Let (X, w1 , w2 ) be a biweak space. There are no relation between θw1 w2 and δw1 w2 , see Examples 4 and 5.


324

E. Rosas, C. Carpintero, J. Sanabria

Remark 5 If we start with a biweak space (X, w1 , w2 ). We obtain that θw1 w2 is a topology on X, see Remark 2. δw1 w2 is a strong generalized topology on X, see Theorem 6 and there are no relation between θw1 w2 and δw1 w2 , see Examples 4 and 5. Definition 5 A ∈ θw1 w2 is called θw1 w2 -open set and its complement is called θw1 w2 -closed. According with Definition 5, we define the θw1 w2 -closure of a subset A of X, as follows: Definition 6 Let (X, w1 , w2 ) be a biweak space. 1. The θw1 w2 -closure T of A is defined as: Cθw1 w2 (A) = {F : A ⊆ F, F is θw1 w2 -closed set in X}. 2. The θw1 w2 -interior of A is defined as: S Iθw1 w2 (A) = {U : U ⊆ A, U is θw1 w2 -open set in X}. 3. γθw1 w2 (A) = {x ∈ X : w2 C(U) ∩ A 6= ∅, for every U ∈ w1 containing x}. Example 6 In Example 2. The Cθw1 w2 (∅) = ∅, Cθw1 w2 (X) = X, Cθw1 w2 ({a}) = {a, c}, Cθw1 w2 ({b}) = X, Cθw1 w2 ({c}) = {c}, Cθw1 w2 ({a, b}) = X, Cθw1 w2 ({a, c}) = {a, c}, Cθw1 w2 ({b, c}) = X. Theorem 7 Let (X, w1 , w2 ) and (X, v1 , v2 ) be two biweak space and A ⊆ X. If w1 ⊆ v1 and w2 ⊆ v2 . Then θw1 w2 ⊆ θv1 v2 Proof. Let A ∈ θw1 w2 and x ∈ A, then there exists an U ∈ w1 such that x ∈ U ⊆ w2 C(U) ⊆ A. Since w1 ⊆ v1 , U ∈ v1 and v2 C(U) ⊆ w2 C(U) ⊆ A. Theorem 8 Let (X, w1 , w2 ) be a biweak space and A ⊆ X. The following are true: 1. A ⊆ γθw1 w2 (A) ⊆ Cθw1 w2 (A). 2. A is θw1 w2 -closed if and only if A = γθw1 w2 (A). 3. x ∈ Iθw1 w2 (A) if and only if there exists a w1 -open set U containing x such that x ∈ U ⊆ w2 C(U) ⊆ A. 4. if A is w2 -open, then w1 C(A) = γθw1 w2 (A).


Θ-modifications on weak spaces

325

Proof. 1. Since θw1 w2 is a weak space, the result follows. 2. If A is θw1 w2 -closed, then A = Cθw1 w2 (A), Now using 1, the result follows. 3. Is a consequence of Definition 6. 4. Let x ∈ w1 C(A) and U any w1 -open set containing x, then U ∩ A 6= ∅, follows that w2 C(U) ∩ A 6= ∅ and then w1 C(A) ⊆ γθw1 w2 (A). Now consider x ∈ γθw1 w2 (A), then for each w1 -open set U containing x, w2 C(U) ∩ A 6= ∅, and then there exists an element z ∈ w2 C(U) ∩ A, since A is w2 -open, z ∈ A, therefore x ∈ w1 C(A).

4

Modification on weak continuous functions

Definition 7 Let (X, w1 , w2 ) and (Y, v1 , v2 ) be two biweak spaces. A function f : X → Y is said to be (Θw1 w2 , Θv1 v2 )-continuous if for every Θv1 v2 -open set V, f−1 (V) is Θw1 w2 -open. Observe that if (X, w1 , w2 ) and (Y, v1 , v2 ) are two biweak spaces, Θw1 w2 and Θv1 v2 are topologies, then the notion of (Θw1 w2 , Θv1 v2 )-continuous functions is similar to the well known concept of continuous functions. Example 7 In Example 4. Observe that: 1. θw1 w2 = {∅, X, {b}, {c}, {b, c}, {a, b}}, 2. θw2 w1 = {∅, X, {a}, {b}, {c}, {a, b}, {a, c}{b, c}}, The identity function f : X → X is (Θw2 w1 , Θw1 w2 )-continuous but is not (Θw1 w2 , Θw2 w1 )-continuous. Theorem 9 Let (X, w1 , w2 ) and (Y, v1 , v2 ) be two biweak spaces; let f : X → Y. Then the following are equivalent: 1. f is (Θw1 w2 , Θv1 v2 )-continuous, 2. For each x ∈ X and each Θv1 v2 -open set V containing f(x), there exists a Θw1 w2 -open set U containing xsuch that f(U) ⊆ V. 3. For each x ∈ X and each Θv1 v2 -open set V containing f(x), there exists a w1 -open set U containing x such that f(w2 C(U) ⊆ V. Proof. The proof follows applying definition.


326

E. Rosas, C. Carpintero, J. Sanabria

Definition 8 Let (X, w1 ) be a weak space and (Y, v1 , v2 ) be a biweak space. A function f : (X, w1 ) → (Y, v1 , v2 ) is said to be faintly (w1 , Θv1 v2 )-continuous if for every Θv1 v2 -open set U, f−1 (U) is w1 -open. Example 8 Let (X, w1 , w2 ) and (Y, v1 , v2 ) be a biweak spaces, where X = Y = {a, b, c}, w1 = {∅, X, {a}, {b}, {c}}, w2 = {∅, X, {a}, {a, b}, {c}}, v1 = {∅, Y, {a}, {b}} and v2 = {∅, Y, {a}, {c}}. Observe that: 1. θw1 w2 = {∅, X, {b}, {a, b}}, 2. θw2 w1 = {∅, X, {c}, {a, c}}, 3. θv1 v2 = {∅, X, {b}, {a, b}}, 4. θv2 v1 = {∅, X, {c}, {a, c}}. Consider a function f : (X, w2 ) → (Y, v1 , v2 ) defined as f(a) = b, f(b) = a, f(c) = c. Then f is faintly (w2 , Θv1 v2 )-continuous but is neither (Θw1 w2 , Θv1 v2 )continuous nor (Θw2 w1 , Θv1 v2 )-continuous Example 9 The function defined in Example 4 is not faintly (w1 , Θw1 w2 )continuous Remark 6 If (X, w1 , w2 ), (Y, v1 , v2 ) are two biweak spaces and f : (X, w1 , w2 ) → (Y, v1 , v2 ) is a function. The concepts of (Θw1 w2 , Θv1 v2 )-continuous and faintly (w1 , Θv1 v2 )-continuous are independent. Theorem 10 Let (X, w1 , w2 ) and (Y, v1 , v2 ) be two biweak spaces. If f : (X, w1 , w2 ) → (Y, v1 , v2 ) is (Θw1 w2 , Θv1 v2 )-continuous, then for every Θv1 v2 -closed set F, f−1 (F) is a Θw1 w2 -closed set. Proof. It follows by duality.

Definition 9 Let (X, w1 ) be a weak space and (Y, v1 , v2 ) be a biweak space. A function f : (X, w1 ) → (Y, v1 , v2 ) is said to be mixed weakly (w1 , v1 v2 )continuous at x ∈ X if for every v1 -open set V, containing f(x), there exists a w1 -open set U containing x such that f(U) ⊆ v2 C(V). Then f is mixed weakly (w1 , v1 v2 )-continuous if it is mixed weakly (w1 , v1 v2 )-continuous at every point x ∈ X.


Θ-modifications on weak spaces

327

Example 10 Let (X, w1 ) be a weak space and (Y, v1 , v2 ) be a biweak space, where X = Y = {a, b, c} and weak structures: w1 = {∅, X, {a}, {b}}, v1 = {∅, X, {b}} and v2 = {∅, X, {a}}. Consider f : (X, w1 ) → (Y, v1 , v2 ), defined as f(a) = b, f(b) = c, f(c) = a. Then f is mixed weakly (w1 , v1 v2 )-continuous. Remark 7 Let (X, w) be a weak space and (Y, v1 , v2 ) be a biweak space. If v1 = v2 , then the notion of mixed weakly (w, v1 v2 )-continuous function is just the notion of weak weakly (w, v1 )-continuous functions, that is, for any v1 -open set V, there exists a w1 -open set U such that f(U) ⊆ v1 C(V). Theorem 11 Let f : X → Y be a function, w1 a weak structure on a nonempty set X, and v1 , v2 be two weak structures on a nonempty set Y. Then: 1. If f is mixed weakly (w1 , v1 v2 )-continuous, then f(w1 C(A)) ⊆ γθv1 v2 (f(A)) for every subset A of X. 2. If f(w1 C(A)) ⊆ γθv1 v2 (f(A)) for every subset A of X, then w1 C(f−1 (v2 I(G)) ⊆ f−1 (v1 C(V)) for every v2 -open set V of Y. Proof. 1. Consider A ⊆ X, x ∈ w1 C(A) and V any v1 -open set containing f(x). By hypothesis f is mixed weakly (w1 , v1 v2 )-continuous, then there exists a w1 -open set U containing x such that f(U) ⊆ v2 C(V). Since x ∈ w1 C(A) and U is a w1 -open set U containing x, A ∩ U 6= ∅. In consequence, ∅ 6= f(A) ∩ f(U) ⊆ v2 C(V) ∩ f(A). Follows that f(x) ∈ γθv1 v2 (f(A)) and hence, f(w1 C(A)) ⊆ γθv1 v2 (f(A)). 2. Clear.

References [1] A. Császar, Generalized topology, generalized continuity, Acta Math. Hungar., 96 (2002), 351–357. [2] A. Császar, Generalized open sets in generalized topologies, Acta Math. Hungar., 96 (2005), 53–66. [3] A. Császar, δ and θ–modifications of generalized topologies, Acta Math. Hungar., 120 (3) (2008), 275–279. [4] A. Császar, Mixed constructions for generalized topologies, Acta Math. Hungar., 122 (2009), 153–159.


328

E. Rosas, C. Carpintero, J. Sanabria

[5] A. Császar and E. Makai Jr., Further remarks on δ and θ–modifications, Acta Math. Hungar., 123 (2009),223–228. [6] W. K. Min, A note on δ and θ–modifications, Acta Math. Hungar., 132 (2011), 107–112. [7] W. K. Min, Mixed weak continuity on generalized topological spaces, Acta Math. Hungar., 132 (2011), 339–347. [8] W. K. Min, On weakly wτ g–closed sets in associated w–spaces, International Journal of Pure an Applied Mathematics, 113 (1) (2017), 181–188. [9] W. K. Min and Y. K. Kim, On weak structures and w–spaces, Far East Journal of Mathematical Sciences, 97 (5) (2015), 549–561. [10] W. K. Min and Y. K. Kim, w–semi open sets and w–semi continuity in weak spaces,International Journal of Pure an Applied Mathematics, 110 (1) (2016), 49–56. [11] W. K. Min and Y. K. Kim, On generalized w-closed sets in w–spaces, International Journal of Pure an Applied Mathematics,110(2), (2016), 327–335. [12] U. Sengul, More on δ- and θ-modifications. Submitted (2017).

Received: May 7, 2018


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 329–339 DOI: 10.2478/ausm-2018-0025

Uniqueness theorems related to weighted sharing of two sets Pulak Sahoo

Himadri Karmakar

Department of Mathematics, University of Kalyani, India email: sahoopulak1@gmail.com

Department of Mathematics, University of Kalyani, India email: himadri394@gmail.com

Abstract. Using the notion of weighted sharing of sets, we study the uniqueness problem of meromorphic functions sharing two finite sets. Our results are inspired from an article due to J. F. Chen (Open Math., 15 (2017), 1244–1250).

1

Introduction, Definitions and Main results

In this paper, a meromorphic function means a function which is meromorphic in the entire complex plane C. Throughout the paper, we adopt the standard notations of Nevanlinna value distribution theory as explained in [6] and [12]. We denote by M(C) the class of all meromorphic functions defined in C and by M1 (C) the class of meromorphic functions which have finitely many poles in C. For convenience, we denote any set of positive real numbers of finite linear measure by E, not necessarily the same at each occurrence. For a nonconstant meromorphic function h, we denote by S(r, h) any quantity satisfying S(r, h) = o{T (r, h)} for r → ∞, r 6∈ E. The order λ(f) of f ∈ M(C) is defined as λ(f) = lim sup r−→∞

log T (r, f) . log r

2010 Mathematics Subject Classification: 30D35, 30D30 Key words and phrases: uniqueness, set sharing, weighted sharing

329


330

P. Sahoo, H. Karmakar

For a meromorphic function f and a set S ⊂ C ∪ {∞}, we define Ef (S) (Ef (S)) to be the set of all a-points of f, where a ∈ S, together with their multiplicities (ignoring their multiplicities). We say that two functions f and g share the set S CM (IM) if Ef (S) = Eg (S) (Ef (S) = Eg (S)). The development of research works related to set sharing problems was broadly initiated due to the following question which was raised by F. Gross [5]. Question 1 Can one find two finite sets Si (i = 1, 2) of C ∪ {∞} such that any two nonconstant entire functions f and g satisfying Ef (Si ) = Eg (Si ) for i = 1, 2 must be identical? In 1994, H. X. Yi [14] proved the following theorem which gives an affirmative answer to Gross’s question. Theorem A Let S1 = {ω | ωn −1 = 0} and S2 = {a}, where n ≥ 5 is an integer, a 6= 0 and a2n 6= 1. If f and g are entire functions such that Ef (Sj ) = Eg (Sj ) for j = 1, 2, then f ≡ g. In [5], F. Gross also pointed that if the answer of Question 1 is affirmative, then it would be interesting to know how large the sets can be. In 1998, H. X. Yi [15] proved the following theorem which deals with the above comment. Theorem B Let S1 = {0} and S2 = {ω | ω2 (ω+a)−b = 0}, where a and b are two nonzero constants such that the algebraic equation ω2 (ω + a) − b = 0 has no multiple roots. If f and g are two entire functions satisfying Ef (Sj ) = Eg (Sj ) for j = 1, 2, then f ≡ g. In this direction, a lot of research works have been devoted during the last two decades (see [4], [9], [10], [13]). We recall the following recent result due to J. F. Chen [2]. Theorem C Let k be a positive integer and let S1 = {α1 , α2 , . . . , αk }, S2 = {β1 , β2 }, where α1 , α2 , . . . , αk , β1 , β2 are k+2 distinct finite complex numbers satisfying (β1 − α1 )2 (β1 − α2 )2 · · · (β1 − αk )2 6= (β2 − α1 )2 (β2 − α2 )2 · · · (β2 − αk )2 . If two nonconstant meromorphic functions f and g in M1 (C) share S1 CM, S2 IM, and if the order of f is neither an integer nor infinite, then f ≡ g. In the same paper, the author also proved another result concerning unique range sets. Before stating the result, we present the definition of unique range sets.


Weighted sharing of two sets

331

Definition 1 For a family of functions G, the subsets S1 , S2 , . . . , Sq of C∪{∞} such that for any f, g ∈ G, f and g share Sj CM for j = 1, 2, . . . , q imply f ≡ g, are called unique range sets (URS, in brief ) for the functions in G. Theorem D Let k be a positive integer and let S1 = {α1 , α2 , . . . , αk }, S2 = {β1 , β2 }, where α1 , α2 , . . . , αk , β1 , β2 are k+2 distinct finite complex numbers satisfying (β1 − α1 )2 (β1 − α2 )2 · · · (β1 − αk )2 6= (β2 − α1 )2 (β2 − α2 )2 · · · (β2 − αk )2 . If the order of f is neither an integer nor infinite, then the sets S1 and S2 are the URS of meromorphic functions in M1 (C). The condition (β1 −α1 )2 (β1 −α2 )2 . . . (β1 −αk )2 6= (β2 −α1 )2 (β2 −α2 )2 . . . (β2 − αk )2 in Theorems C and D can not be dropped as shown by the following example. P zn Example 1 [2] For a positive integer k, let f(z) = ∞ n=1 n3n , g(z) = −f(z), S1 = {−1, 1, −2, 2, . . . , −k, k}, and S2 = {−(k + 1), k + 1}. Then using the result of [3, p. 288] we deduce λ(f) =

1 n log n 1 = lim sup = . 3n 3n log n 3 log n n−→∞ lim inf n−→∞ n log n

Clearly f(z), g(z) ∈ M1 (C), f(z) and g(z) share S1 , S2 CM. But f(z) 6≡ g(z). The assumption “nonconstant meromorphic functions f and g in M1 (C)” in Theorems C and D cannot be relaxed to “nonconstant meromorphic functions f and g in M(C)” as shown by the following example. P∞ zn 1 Example 2 [2] For a positive integer k, let f(z) = n=1 n3n , g(z) = f(z) , 1 1 S1 = 2, 2 , 3, 13 , . . . , k, k1 , S2 = k + 1, k+1 . From Example 1 we note that λ(f) = 13 and, therefore, using the result of [3, p. 293] we see that g(z) has infinitely many poles in C. Moreover, f(z) and g(z) share the sets S1 , S2 CM. But f(z) 6≡ g(z). The following example given in [2] shows the necessity of the assumption in Theorems C and D that the order of f is neither an integer nor infinite.


332

P. Sahoo, H. Karmakar z

z Example 3 For a positive integer f(z) = ee ), g(z) = k, let f(z) = e 1 (resp. 1 1 1 1 f(z) , S1 = 2, 2 , 3, 3 , . . . , k, k , S2 = k + 1, k+1 . Then by Lemma 8 in section 2 we see that λ(f) = 1 (resp. λ(f) = ∞). Though all other conditions of Theorems C and D are satisfied, f(z) 6≡ g(z).

However, the research on set sharing problem gained a new dimension when the idea of weighted sharing, introduced by I. Lahiri in 2001 (see [7], [8]), was incorporated. The necessary definitions are as follows: Definition 2 Let k be a nonnegative integer or infinity. For a ∈ C ∪ {∞} we denote by Ek (a; f) the set of all a-points of f, where an a-point of multiplicity m is counted m times if m ≤ k and k+1 times if m > k. If Ek (a; f) = Ek (a; g), we say that f and g share the value a with weight k. We write f and g share (a, k) to mean that f and g share the value a with weight k. Clearly if f, g share (a, k) then f, g share (a, p) for any integer p where 0 ≤ p < k. In particular, f and g share a CM (IM) if and only if f and g share (a, ∞) ((a, 0)). Definition 3 Let S be a set of distinct elements of C ∪ {∞} and k be a nonnegative integer or infinity. We denote by Ef (S, k) the set ∪a∈S Ek (a; f). We say that f and g share the set S with weight k, or simply f and g share (S, k) if Ef (S, k) = Eg (S, k). Definition 4 Let k be a positive integer and S1 = {α1 , α2 , . . . , αk }, where αi ’s are nonzero complex constants. Suppose that P P zk − ( αi )zk−1 + . . . + (−1)k−1 ( αi1 αi2 ...αik−1 )z P(z) = , (1) (−1)k+1 α1 α2 ...αk where αi ∈ S1 for i = 1, 2, . . . , k. Let m1 be the number of simple zeros of P(z) and m2 be the number of multiple zeros of P(z). Then we define Γ1 := m1 + m2 and Γ2 := m1 + 2m2 . Regarding Theorem C, one may ask the following question: Question 2 Is the conclusion of Theorem C still true if f and g share (S1 , 2) and S2 IM instead of sharing S1 CM and S2 IM? In this paper, we try to find possible answers to the above question and prove the following theorems:


Weighted sharing of two sets

333

Theorem 1 Let f, g ∈ M1 (C) and S1 = {α1 , α2 , . . . , αk }, S2 = {β1 , β2 }, where α1 , α2 , . . . , αk , β1 , β2 are k + 2 distinct nonzero complex constants satisfying k > 2Γ2 . If f, g share (S1 , 2) and S2 IM, then f ≡ g, provided (β1 − α1 )2 (β1 − α2 )2 · · · (β1 − αk )2 6= (β2 − α1 )2 (β2 − α2 )2 · · · (β2 − αk )2 and f is of non-integer finite order. Theorem 2 Let S1 and S2 be stated as in Theorem 1 with k > 2Γ2 . If M2 (C) denote the subclass of meromorphic functions of non-integer finite order in M1 (C), then the sets S1 and S2 are the URS of meromorphic functions in M2 (C), provided (β1 − α1 )2 (β1 − α2 )2 · · · (β1 − αk )2 6= (β2 − α1 )2 (β2 − α2 )2 · · · (β2 − αk )2 . We now state some more definitions (see [7], [8]). Definition 5 For a ∈ C ∪ {∞}, we denote by N(r, a; f| = k) the reduced counting function of the a-points of f whose multiplicities are exactly k. In particular, N(r, a; f| = 1) or N(r, a; f| = 1) is the counting function of the simple a-points of f. Definition 6 For a positive integer m we denote by N(r, a; f| ≤ m) (N(r, a; f| ≥ m)) the counting function of those a-points of f whose multiplicities are not greater (less) than m, where each a-point is counted according to its multiplicity. N(r, a; f| ≤ m) and N(r, a; f| ≥ m) are the corresponding reduced counting functions. Definition 7 We denote by N2 (r, a; f) the sum N(r, a; f) + N(r, a; f| ≥ 2). Definition 8 Let f and g be two nonconstant meromorphic functions such that f and g share (a, 2) for a ∈ C ∪ {∞}. Let z0 be an a-point of f with multiplicity p and an a-point of g with multiplicity q. We denote by NL (r, a; f) (NL (r, a; g)) the reduced counting function of those a-points of f and g where (3 p > q ≥ 3 (q > p ≥ 3). Also we denote by NE (r, a; f) the counting function of (3 (3 those a-points of f and g where p = q ≥ 3. Clearly NE (r, a; f) = NE (r, a; g). Definition 9 Let f, g share the value a IM. We denote by N∗ (r, a; f, g) the reduced counting function of those a-points of f whose multiplicities differ from the multiplicities of the corresponding a-points of g. Clearly N∗ (r, a; f, g) = N∗ (r, a; g, f) and N∗ (r, a; f, g) = NL (r, a; f)+NL (r, a; g).


334

2

P. Sahoo, H. Karmakar

Lemmas

In this section, we present some lemmas which will be needed in the sequel. We denote by H the following function: 00 00 F G 2F 0 2G 0 H= − , − − F0 F−1 G0 G−1 where F and G are two meromorphic functions in M1 (C). Lemma 1 [7] If F, G share (1, 1) and H 6≡ 0, then N(r, 1; F| = 1) ≤ N(r, ∞; H) + S(r, F) + S(r, G). Lemma 2 Let F, G ∈ M1 (C). If F, G share (1, 0) and H 6≡ 0, then N(r, ∞, H) ≤ N(r, 0; F| ≥ 2) + N(r, 0; G| ≥ 2) + N∗ (r, 1; F, G) +N0 (r, 0; F 0 ) + N0 (r, 0; G 0 ) + S(r, F) + S(r, G), where N0 (r, 0; F 0 ) is the reduced counting function of those zeros of F 0 which are not the zeros of F(F − 1). N0 (r, 0; G 0 ) is defined similarly. Proof. Noting that N∗ (r, ∞; F, G) = S(r, F)+S(r, G), this lemma can be proved in a similar manner as in Lemma 4 of [9]. Lemma 3 [1] Let F and G be two nonconstant meromorphic functions sharing (1, 2). Then (3

2NL (r, 1; F) + 3NL (r, 1; G) + 2NE (r, 1; F) + N(r, 1; F| = 2) ≤ N(r, 1; G) − N(r, 1; G). Lemma 4 [11] Let f be a nonconstant meromorphic function and P(f) = a0 + a1 f + a2 f2 + . . . + an fn , where a0 , a1 , a2 , . . . , an are constants and an 6= 0. Then T (r, P(f)) = nT (r, f) + O(1). Lemma 5 [15] If H ≡ 0, then T (r, G) = T (r, F) + O(1). If, in addition, N(r, 0; F) + N(r, ∞; F) + N(r, 0; G) + N(r, ∞; G) < 1, T (r) r→∞,r6∈E lim sup

where T (r) = max{T (r, F), T (r, G)} then either F ≡ G or F.G ≡ 1.


Weighted sharing of two sets

335

Remark 1 We observe that the above lemma holds for F, G ∈ M(C). As our discussion is restricted in M1 (C), we may drop the terms N(r, ∞; F) and N(r, ∞; G) while using this result. Lemma 6 Let F, G ∈ M1 (C). If F and G share (1, 2) and H 6≡ 0, then (3

(i) T (r, F) ≤ N2 (r, 0; F) + N2 (r, 0; G) − m(r, 1; G) − NE (r, 1; F) − NL (r, 1; G) + S(r, F) + S(r, G); (3

(ii) T (r, G) ≤ N2 (r, 0; G) + N2 (r, 0; F) − m(r, 1; F) − NE (r, 1; G) − NL (r, 1; F) + S(r, F) + S(r, G). Proof. The proof of this lemma flows in the line of the proof of Lemma 2.13 in [1]. As we are dealing with functions of class M1 (C), we insist in presenting the proof for the sake of completeness. From the second fundamental theorem of Nevanlinna , we have T (r, F) ≤ N(r, 0; F) + N(r, ∞; F) + N(r, 1; F) − N0 (r, 0; F 0 ) + S(r, F); that is, T (r, F) ≤ N(r, 0; F) + N(r, 1; F) − N0 (r, 0; F 0 ) + S(r, F).

(2)

T (r, G) ≤ N(r, 0; G) + N(r, 1; G) − N0 (r, 0; G 0 ) + S(r, G).

(3)

Similarly,

Combining (2) and (3), we obtain T (r, F) + T (r, G) ≤ N(r, 0; F) + N(r, 0; G) + N(r, 1; F) + N(r, 1; G) −N0 (r, 0; F 0 ) − N0 (r, 0; G 0 ) + S(r, F) + S(r, G). (4) We also see that (3

N(r, 1; F) + N(r, 1 : G) ≤ N(r, 1; F| = 1) + N(r, 1; F| = 2) + NE (r, 1; F) +NL (r, 1; F) + NL (r, 1; G) + N(r, 1; G). Using Lemma 1 and Lemma 2 in (5), we obtain that N(r, 1; F) + N(r, 1; G) ≤ N(r, 0; F| ≥ 2) + N(r, 0; G| ≥ 2) + 2NL (r, 1; F) (3

+2NL (r, 1; G) + N(r, 1; F| = 2) + NE (r, 1; F) +N0 (r, 0; F 0 ) + N0 (r, 0; G 0 ) + N(r, 1; G) +S(r, F) + S(r, G).

(5)


336

P. Sahoo, H. Karmakar

Substituting the value of N(r, 1; G) from Lemma 3, we obtain N(r, 1; F) + N(r, 1; G) ≤ N(r, 0; F| ≥ 2) + N(r, 0; G| ≥ 2) + 2NL (r, 1; F) (3

+2NL (r, 1; G) + N(r, 1; F| = 2) + NE (r, 1; F) +N0 (r, 0; F 0 ) + N0 (r, 0; G 0 ) + N(r, 1; G) (3

−2NL (r, 1; F) − 3NL (r, 1; G) − 2NE (r, 1; F) −N(r, 1; F| = 2) + S(r, F) + S(r, G) ≤ N(r, 0; F| ≥ 2) + N(r, 0; G| ≥ 2) − NL (r, 1; G) (3

−NE (r, 1; F) + T (r, G) − m(r, 1; G) + N0 (r, 0; F 0 ) +N0 (r, 0; G 0 ) + S(r, F) + S(r, G).

(6)

Noting the fact that N2 (r, a; f) = N(r, a; f) + N(r, a; f| ≥ 2), the lemma follows from (4) and (6). Lemma 7 Let f, g ∈ M1 (C). If f, g share the set {β1 , β2 } IM, then λ(f) = λ(g). Proof. Proof of this lemma can be extracted from the first part of the proof of Theorem 1.3 in [2] (see p. 1247). Lemma 8 (see [12, p. 65]) Let h be an entire function and f(z) = eh(z) . Then (i) if h(z) is a polynomial of deg h, then λ(f) = deg h; (ii) if h(z) is a transcendental entire function, then λ(f) = ∞. Lemma 9 (see [12, p. 115]) Let a1 , a2 and a3 be three distinct complex numbers in C ∪ {∞}. If two nonconstant meromorphic functions f and g share a1 , a2 and a3 CM, and if the order of f and g is neither an integer nor infinity, then f ≡ g.

3

Proof of the Theorems

Proof. [Proof of Theorem 1] Let F = P(f) and G = P(g) where P(z) is defined as in (1). Clearly F, G share (1, 2) as f, g share (S1 , 2). From Lemma 4, we obtain T (r, F) = kT (r, f) + S(r, f);

(7)


Weighted sharing of two sets

337

T (r, G) = kT (r, g) + S(r, g).

(8)

Let H 6≡ 0. By Lemma 6, we have T (r, F) ≤ N2 (r, 0; F) + N2 (r, 0; G) + S(r, F) + S(r, G) = N2 (r, 0; P(f)) + N2 (r, 0; P(g)) + S(r, f) + S(r, g) ≤ Γ2 N(r, 0; f) + Γ2 N(r, 0; g) + S(r, f) + S(r, g) ≤ Γ2 {T (r, f) + T (r, g)} + S(r, f) + S(r, g).

(9)

Similarly, T (r, G) ≤ Γ2 {T (r, f) + T (r, g)} + S(r, f) + S(r, g).

(10)

From (7)-(10), we obtain k{T (r, f) + T (r, g)} ≤ 2Γ2 {T (r, f) + T (r, g)} + S(r, f) + S(r, g), which is a contradiction as k > 2Γ2 . Hence H ≡ 0. Let T (r) = max{T (r, F), T (r, G)}. Now, N(r, 0; F) + N(r, 0; G) ≤ Γ1 N(r, 0; f) + Γ1 N(r, 0; g) ≤ Γ1 {T (r, f) + T (r, g)} + S(r, f) + S(r, g) Γ1 = {T (r, F) + T (r, G)} + S(r, F) + S(r, G) k 2Γ1 T (r) + o{T (r)}. (11) ≤ k As k > 2Γ2 ≥ 2Γ1 , from Lemma 5 and (11), we obtain either F ≡ G or F.G ≡ 1. If possible, let F.G ≡ 1. Then P(f).P(g) ≡ 1. As g ∈ M1 (C), we have P(g) ∈ M1 (C). Hence P(f) has at most finitely many zeros. Therefore P(f) = µ1 (z)eφ1 (z) , where µ1 (z) is a rational function and φ1 (z) is an entire function, which is a contradiction by Lemma 8 as the order of f is neither an integer not infinity. Similarly if we consider the case when P(g) has at most finitely many zeros, we arrive at a contradiction as λ(g) = λ(f), by Lemma 7. Hence the case F.G ≡ 1 can not occur. If F ≡ G, we have P(f) ≡ P(g), which gives (f(z) − α1 )(f(z) − α2 ) . . . (f(z) − αk ) ≡ 1. (g(z) − α1 )(g(z) − α2 ) . . . (g(z) − αk ) From (12) and the assumption (β1 − α1 )2 (β1 − α2 )2 · · · (β1 − αk )2 6= (β2 − α1 )2 (β2 − α2 )2 · · · (β2 − αk )2 ,

(12)


338

P. Sahoo, H. Karmakar

we obtain that f(z) = β1 if and only if g(z) = β1 since f and g share S2 IM. Similarly, we see that f(z) = β2 if and only if g(z) = β2 . Consequently, we have f and g share β1 and β2 IM. Again, from (12) we see that f and g share β1 , β2 and ∞ CM. Noting that the order of f is neither an integer nor infinity, the conclusion follows from Lemma 7 and Lemma 9. Proof. [Proof of Theorem 2] If f, g share S1 and S2 CM, then f, g certainly share (S1 , 2) and S2 IM, which satisfies the conditions of Theorem 1 and hence the conclusion follows. Here we omit the details.

Acknowledgment The authors are thankful to the referee for his/her valuable suggestions towards the improvement of the paper. The second author is thankful to UGC-JRF scheme of research fellowship for financial assistance.

References [1] A. Banerjee, Uniqueness of meromorphic functions sharing two sets with finite weight, Portugal. Math. (N. S.), 65 (2008), 81–93. [2] J. F. Chen, Uniqueness of meromorphic functions sharing two finite sets, Open Math., 15 (2017), 1244–1250. [3] J.B. Conway, Functions of One Complex Variable, Springer–Verlag, New York, 1973. [4] M. L. Fang and H. Guo, On meromorphic functions sharing two values, Analysis, 17 (1997), 355–366. [5] F. Gross, Factorization of meromorphic functions and some open problems, Complex Analysis (Proc. Conf. Univ. Kentucky, Lexington, KY, 1976), pp. 51–69, Lecture Notes in Math., Vol 599, Springer, Berlin, 1977. [6] W. K. Hayman, Meromorphic functions, Clarendon Press, Oxford, 1964. [7] I. Lahiri, Weighted value sharing and uniqueness of meromorphic functions, Complex Var. Theory Appl., 46 (2001), 241–253. [8] I. Lahiri, Weighted sharing and uniqueness of meromorphic functions, Nagoya Math. J., 161 (2001), 193–206.


Weighted sharing of two sets

339

[9] I. Lahiri and A. Banerjee, Weighted sharing of two sets, Kyungpook Math. J., 46 (2006), 79–87. [10] P. Li and C. C. Yang, On the unique range sets for meromorphic functions, Proc. Amer. Math. Soc., 124 (1996), 177–185. [11] C. C. Yang, On deficiencies of differential polynomials II., Math. Z., 125 (1972), 107–112. [12] C. C. Yang and H. X. Yi, Uniqueness Theory of Meromorphic functions, Kluwer Academic Publishers, Dordrecht, 2003. [13] B. Yi and Y. H. Li, The uniqueness of meromorphic functions that share two sets with CM, Acta Math. Sin., Chin. Ser., 55 (2012), 363–368 (in Chinese). [14] H. X. Yi, Uniqueness of meromorphic functions and a question of Gross, Sci. China. Ser. A, 37 (1994), 802–813. [15] H. X. Yi, Meromorphic functions that share one or two values, Complex Var. Theory Appl., 28 (1995), 1–11. [16] H. X. Yi, On a question of Gross concerning uniqueness of entire functions, Bull. Austral. Math. Soc., 57 (1998), 343–349. [17] H. X. Yi, Meromorphic functions that share two sets, Acta Math. Sin., Chin. Ser., 45 (2002), 75–82 (in Chinese).

Received: January 28, 2018


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 340–346 DOI: 10.2478/ausm-2018-0026

Scaling functions on the spectrum Abdullah

Firdous A. Shah

Department of Mathematics, ZHDC, University of Delhi, India email: abd.zhc.du@gmail.com

Department of Mathematics, University of Kashmir, India email: fashah79@gmail.com

Abstract. A generalization of Mallat’s classic theory of multiresolution analysis based on the theory of spectral pairs was considered by Gabardo and Nashed [4] for which the translation set Λ = {0, r/N}+2 Z is no longer a discrete subgroup of R but a spectrum associated with a certain onedimensional spectral pair. In this short communication, we characterize the scaling functions associated with such a nonuniform multiresolution analysis by means of some fundamental equations in the Fourier domain.

1

Introduction

Multiresolution analysis (MRA) is an important mathematical tool since it provides a natural framework for understanding and constructing discrete wavelet systems. The concept of an MRA structure has been extended in various setups in recent years. More precisely, they have been generalized to different dimensionalities, to lattices different from Zd , allowing the subspaces of MRA to be generated by Riesz basis instead of orthonormal basis, admitting a finite number of scaling functions, replacing the dilation factor 2 by an integer M ≥ 2 or by an expansive matrix A ∈ GLd (R) as long as A ⊂ AZd (see [1]). All these concepts were developed on regular lattices, that is the translation set is always a group. Recently, Gabardo and Nashed [3, 4] considered a generalization of Mallat’s classical MRA [6] based on the theory of 2010 Mathematics Subject Classification: 42C15, 42C40, 65T60 Key words and phrases: nonuniform multiresolution analysis, scaling function, spectrum, Fourier transform

340


Scaling functions on the spectrum

341

spectral pairs, in which the translation set Λ = {0, r/N} + 2Z, where N ≥ 1 is an integer, 1 ≤ r ≤ 2N − 1, r is an odd integer relatively prime to N, acting on the scaling function related with an MRA to generate the core subspace V0 is no longer a group, but a union of two lattices, which is associated with a famous open conjecture of Fuglede on spectral pairs [2]. They call it nonuniform multiresolution analysis (NUMRA). By an NUMRA, we mean a sequence of embedded closed subspaces {Vj : j ∈ Z} of the Hilbert space L2 (R) that satisfies the following conditions: (a) Vj ⊂ Vj+1 for all j ∈ Z; S (b) j∈Z Vj is dense in L2 (R); T (c) j∈Z Vj = {0}; (d) f(x) ∈ Vj if and only if f(2Nx) ∈ Vj+1 for all j ∈ Z; (e) there exists a function φ ∈ V0 such that {φ(x − λ)}λ∈Λ is an orthonormal basis for V0 . It is worth noticing that, when N = 1, one recovers the standard definition of one dimensional MRA with dyadic dilation 2. When, N > 1, the dilation factor of 2N ensures that 2NΛ ⊂ Z ⊂ Λ. If φ is a scaling function of an NUMRA, then by condition (e) we can express this function in terms of the orthonormal basis {φ(x − λ) : λ ∈ Λ} as X φ(x) = hλ φ 2Nx − λ . (1) λ∈Λ

where the convergence is in L2 (R) and {hλ }λ∈Λ ∈ l2 . Refinement equation (1) can be rewritten in the Fourier domain as ξ ξ ^ ^ φ(ξ) = m0 φ (2) 2N 2N where m0 is the low pass filter associated with the scaling function φ and is of the form m0 (ξ) = m10 (ξ) + e−2πiξr/N m20 (ξ). (3) One of the fundamental problems in the study of wavelet theory is to find conditions on the scaling functions so that they can generate an MRA for L2 (R). Our main purpose in this short communication is to characterize those functions that are scaling functions for an NUMRA of L2 (R). To achieve our goal, we need the following technical results obtained in [4, 5, 7] that will be used in sequel.


342

Abdullah, F. A. Shah

2 Theorem 1 [4] Let {Vj : j ∈ Z} be a sequence T of closed subspaces of L (R) satisfying conditions (a), (d) and (e). Then, j∈Z Vj = {0}.

Theorem 2 [5] Let {Vj : j ∈ Z} be a sequence of closed subspaces of L2 (R) satisfying conditions (a), (d) and (e). Assume that the function φ of condition ^ is continuous at ξ = 0. Then the following two conditions (e) is such that φ are equivalent:

^ (2N)−j ξ = 1 a.e. ξ ∈ R; (i) lim φ j→∞ S (ii) j∈Z Vj = L2 (R). Proposition 1 [7] Let N be a positive integer, and r ∈ {1, 3, . . . , 2N − 1} be an odd integer. Let φ ∈ L2 (R) with φ 2 = 1. Then, (i) For each fixed odd r, the family φ(x − λ) : λ ∈ Λ is an orthonormal system in L2 (R) if and only if 2 X

^ ξ + p

= 2, for a.e. ξ ∈ R and (4)

φ 2 p∈Z

X p

2

^ (5) e−iπrp/N φ ξ+

= 0, for a.e. ξ ∈ R. 2 p∈Z

(ii) The collection φ(x − λ) : λ ∈ Λ is an orthonormal system for every odd integer r ∈ {1, 3, . . . , 2N − 1} if and only if X

φ(ξ ^ − β) 2 = 1, for a.e. ξ ∈ R, (6) β∈ΓN

where ΓN = {nN + j/2 : n ∈ Z, j = 0, 1, 2, . . . , N − 1}.

2

Characterization of scaling functions on the spectrum

In this section we will characterize those functions that are scaling functions for an NUMRA of L2 (R) by means of some basic equations in the Fourier domain. Before formulating our main result, let us clarify what we mean when we say that a function is a scaling function for an NUMRA. Given a function φ ∈ L2 (R), we define the closed subspaces {Vj : j ∈ Z} of L2 (R) as follows:

V0 = span φ(x − λ) : λ ∈ Λ , and Vj = f : f (2N)−j x ∈ V0 , j ∈ Z\{0}.


Scaling functions on the spectrum

343

We say that φ ∈ L2 (R) is a scaling function for an NUMRA of L2 (R) if the sequence of closed subspaces {Vj : j ∈ Z} as defined above forms an NUMRA for L2 (R). Theorem 3 A function φ ∈ L2 (R) is a scaling function for an NUMRA of L2 (R) if and only if X

φ(ξ ^ − β) 2 = 1, texta.e

(7)

β∈ΓN

^ lim φ (2N)−j ξ = 1 a.e. ξ ∈ R

(8)

j→∞

and there exists a periodic function m0 of the form (3) such that ^ φ(ξ) = m0

ξ 2N

^ φ

ξ 2N

,

a.e. ξ ∈ R.

(9)

Proof. Suppose φ is a scaling function for an NUMRA. Then, {φ(x − λ) : λ ∈ Λ} forms an orthonormal system in L2 (R) which is equivalent to equation (7) by Proposition 1. Equality (9) follows from equations (2) and (3). Since {Vj : j ∈ Z} S is an NUMRA for L2 (R), we have j∈Z Vj = L2 (R). Therefore, from Theorem 2, we infer that Z 2

^

lim

φ (2N)−j ξ dξ = 1. j→∞ ΓN

Since m0 (ξ) is of the form (3), so it is easy to compute the following two conditions in terms of the 1/2-periodic functions m10 , m20 as 2N−1 X

p

2

2 p

2

1

m0 ξ +

+ m0 ξ +

= 1, 4N 4N

p=0 2N−1 X p=0

−iπrp/N

e

and

p

2

2 p

2

1

m0 ξ +

+ m0 ξ +

= 0. 4N 4N

(10)

(11)

2

2 If we take M0 (ξ) = m10 (ξ) + m20 (ξ) , then clearly M0 ξ + 14 = M0 (ξ) and |m0 (ξ + N/2)|2 + |m0 (ξ)|2 M0 (ξ) = . (12) 2


344

Abdullah, F. A. Shah

Subsequently, Eqs. (10) and (11) takes the form 2N−1 X p=0

p M0 ξ + = 1, 4N

and

2N−1 X p=0

p = 0. e−iπrp/N M0 ξ + 4N

Hence, M0 (ξ) ≤ 1, a.e. ξ ∈ R, which together with (12) implies ≤ 1

|m0 (ξ)| −j ξ) is ^ a.e. ξ ∈ R. This inequality along with equality (9) shows that φ((2N) non-decreasing for a.e. ξ ∈ R as j → ∞. Let

^ −j

(13) Φ(ξ) = lim φ((2N) ξ) . j→∞

^ ≤ 1 a.e, therefore, Lebesgue’s dominated convergence theorem Since φ(ξ) implies that Z Φ(ξ) dξ = 1. ΓN

We now prove the converse. Assume that (7), (8) and (9) are satisfied. The orthonormality of the system {φ(x − λ) : λ ∈ Λ} follows immediately from (7). This fact alongwith the definition of V0 gives us (e) of the definition of an NUMRA. Moreover, the definition of the subspaces Vj also shows that f(x) ∈ Vj holds if and only if f 2Nx ∈ Vj+1 which is (d) of the definition of −j/2 f((2N)−j x) ∈ V , then there exists an NUMRA. Thus, we say that 0 P if (2N) a sequence {hλ }λ∈Λ satisfying λ∈Λ hλ < ∞ such that X f (2N)−j x = (2N)j/2 hλ φ(x − λ). (14) λ∈Λ

Taking Fourier transform on both sides of (14), we obtain ^ f^ (2N)j ξ = µj (ξ)φ(ξ) where µj (ξ) = as

P

−2πiλξ . λ∈Λ hλ e

(15)

Since Λ = {0, r/N} + 2Z, we can rewrite µj (ξ)

µj (ξ) = µ1j (ξ) + e−2πiξr/N µ2j (ξ)

(16)

where µ1j and µ2j are locally L2 , 1/2-periodic functions. Now, for each j ∈ Z, we claim that

^ Vj = f : f^ (2N)j ξ = µj (ξ)φ(ξ) for some periodic function µj (ξ) . (17)


Scaling functions on the spectrum

345

To prove the inclusion Vj ⊂ Vj+1 , it is enough to show that V0 ⊂ V1 . Assume that f ∈ V0 , then by equation (17), it follows that there exists a ^ = µ0 (ξ)φ(ξ), ^ locally L2 function say µ0 such that f(ξ) where µ0 (ξ) = µ10 (ξ) + e−2πiξr/N µ20 (ξ). Using (9), we obtain ^ ^ ^ f(2Nξ) = µ0 (2Nξ)φ(2Nξ) = µ0 (2Nξ)m0 (ξ)φ(ξ). Moreover, µ0 (2Nξ)m0 (ξ) can be further expressed in the form η1 (ξ) + e−2πiξr/N η2 (ξ), where

η1 (ξ) = µ10 (2Nξ) + e−4πiξr µ20 (2Nξ) m10 (ξ)

η2 (ξ) = µ10 (2Nξ) + e−4πiξr µ20 (2Nξ) m20 (ξ). Using the fact that |m0 (ξ)| ≤ 1 for a.e. ξ ∈ ΓN , we have Z

µ0 (2Nξ) 2 m0 (ξ) 2 dξ ≤

Z

µ0 (2Nξ) 2 dξ < ∞, ΓN

ΓN

which implies that f ∈ V1 . We have already seen that separation property (c) of an NUMRA follows from (a), (d) and (e). Now it remains to prove density property (b) of an NUMRA, that is; L2 (R) = ∪j∈Z Vj . To prove this, we assume that Pj be the orthogonal projection onto the closed subspace Vj of L2 (R), then it suffices to show that

Pj f − f 2 = kfk2 − Pj (f), f → 0 as j → ∞. 2 2 2 Since (2N)j/2 φ (2N)j x − λ λ∈Λ is an orthonormal basis for Vj . Therefore, for any compactly supported function f, we have

Pj f, f 2 =

Z

2

2

^ f (ξ) dξ.

φ (2N)−j ξ ^

(18)

R

Implementing condition (8), it follows that the right hand side of (18) converges to kfk22 as j → ∞. This completes the proof of Theorem 3.


346

Abdullah, F. A. Shah

References [1] L. Debnath and F. A. Shah, Wavelet Transforms and Their Applications, Birkhäuser, New York, 2015. [2] B. Fuglede, Commuting self-adjoint partial different operators and a group theoretic problem. J. Funct. Anal., 16 (1974), 101–121. [3] J. P. Gabardo and M. Z. Nashed, An analogue of Cohen’s condition for nonuniform multiresolution analyses, in: A. Aldroubi, E. Lin (Eds.), Wavelets, Multiwavelets and Their Applications, in: Cont. Math., 216, Amer. Math. Soc., Providence, RI, (1998), 41–61. [4] J. P. Gabardo and M. Z. Nashed, Nonuniform multiresolution analysis and spectral pairs, J. Funct. Anal., 158 (1998), 209–241. [5] J. P. Gabardo and X. Yu, Wavelets associated with nonuniform multiresolution analysis and one-dimensional spectral pairs, J. Math. Anal. Appl., 323 (2006), 798–817. [6] S. G. Mallat, Multiresolution approximations and wavelet orthonormal bases of L2 (R), Trans. Amer. Math. Soc., 315 (1989), 69–87. [7] X. Yu and J. P. Gabardo, Nonuniform wavelets and wavelet sets related to the one-dimensional spectral pairs, J. Approx. Theory., 145 (2007), 133–139.

Received: May 2, 2018


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 347–367 DOI: 10.2478/ausm-2018-0027

Integrals of polylogarithmic functions with negative argument Anthony Sofo Victoria University, Australia email: anthony.sofo@vu.edu.au

Abstract. The connection between polylogarithmic functions and Euler sums is well known. In this paper we explore the representation and many connections between integrals of products of polylogarithmic functions and Euler sums. We shall consider mainly, polylogarithmic functions with negative arguments, thereby producing new results and extending the work of Freitas. Many examples of integrals of products of polylogarithmic functions in terms of Riemann zeta values and Dirichlet values will be given.

1

Introduction and preliminaries

It is well known that integrals of products of polylogarithmic functions can be associated with Euler sums, see [16]. In this paper we investigate the representations of integrals of the type Z1 xm Lit (−x) Liq (−x) dx, 0

for m ≥ −2, and for integers q and t. For m = −2, −1, 0 we give explicit representations of the integral in terms of Euler sums and for m ≥ 0 we give a recurrence relation for the integral in question. We also mention two specific 2010 Mathematics Subject Classification: 11M06, 11M32, 33B15 Key words and phrases: polylogarithm function, Euler sums, zeta functions, Dirichlet functions

347


348

A. Sofo

integrals with a different argument in the polylogarithm. Some examples are highlighted, almost none of which are amenable to a computer mathematical package. This work extends the results given by [16], who examined a similar integral with positive arguments of the polylogarithm. Devoto and Duke [14] also list many identities of lower order polylogarithmic integrals and their relations to Euler sums. Some other important sources of information on polylogarithm functions are the works of [19] and [20]. In [3] and [12] the authors explore the algorithmic and analytic properties of generalized harmonic Euler sums systematically, in order to compute the massive Feynman integrals which arise in quantum field theories and in certain combinatorial problems. Identities involving harmonic sums can arise from their quasi-shuffle algebra or from other properties, such as relations to the Mellin transform Z1 M[f(z)](N) = dz zN f (z) , 0

where the basic functions f(z) typically involve polylogarithms and harmonic sums of lower weight. Applying the latter type of relations, the author in [6], expresses all harmonic sums of the above type with weight w = 6, in terms of Mellin transforms and combinations of functions and constants of lower weight. In another interesting and related paper [17], the authors prove several identities containing infinite sums of values of the Roger’s dilogarithm function. defined on x ∈ [0.1], by   Li2 (x) + 21 ln x ln (1 − x) ; 0 < x < 1 LR (x) = . 0 ; x=0  ζ (2) ; x=1 The Lerch transcendent, Φ (z, t, a) =

∞ X

zm (m + a)t m =0

is defined for |z| < 1 and < (a) > 0 and satisfies the recurrence Φ (z, t, a) = z Φ (z, t, a + 1) + a−t . The Lerch transcendent generalizes the Hurwitz zeta function at z = 1, Φ (1, t, a) =

∞ X

1 (m + a)t m =0


Integrals of polylogarithmic

349

and the polylogarithm, or de-Jonquière’s function, when a = 1, Lit (z) :=

∞ X zm , t ∈ C when |z| < 1; < (t) > 1 when |z| = 1. mt

m =1

Let Hn =

n X 1 r=1

r

Z1 = 0

X 1 − tn n dt = γ + ψ (n + 1) = , 1−t j (j + n)

H0 := 0

j=1

be the nth harmonic number, where γ denotes the Euler-Mascheroni constant, P (m) Hn = nr=1 r1m is the mth order harmonic number and ψ(z) is the digamma (or psi) function defined by ψ(z) :=

d Γ 0 (z) 1 {log Γ (z)} = and ψ(1 + z) = ψ(z) + , dz Γ (z) z

moreover, ψ(z) = −γ +

∞ X n=0

1 1 − n+1 n+z

.

More generally a non-linear Euler sum may be expressed as,   qj Y t r X (±1)n Y m k (αj ) (β )   Hn Jn k np n≥1

j=1

k=1

where p ≥ 2, t, r, qj , αj , mk , βk are positive integers and  q  m n n j+1 q X X m (−1)  1 (β) (α) , Jn = . Hn = α j jβ j=1

j=1

If, for a positive integer λ=

t X j=1

αj qj +

r X

βj mj + p,

j=1

then we call it a λ-order Euler sum. The polygamma function ∞

ψ(k) (z) =

X dk 1 {ψ(z)} = − (−1)k+1 k! k dz (r + z)k+1 r=0


350

A. Sofo

and has the recurrence ψ

(k)

(z + 1) = ψ

(k)

(−1)k k! . (z) + zk+1

The connection of the polygamma function with harmonic numbers is, (α+1)

Hz

= ζ (α + 1) +

(−1)α (α) ψ (z + 1) , z 6= {−1, −2, −3, ...} . α!

(1)

and the multiplication formula is ψ(k) (pz) = δm,0 ln p +

1 pk+1

p−1 X

ψ(k) (z +

j=0

j ) p

(2)

for p a positive integer and δp,k is the Kronecker delta. We define the alternating zeta function (or Dirichlet eta function) η (z) as ∞ X (−1)n+1 η (z) := nz

= 1 − 21−z ζ (z)

(3)

n =1

where η (1) = ln 2. If we put ∞ (p) X (−1)n+1 Hn S (p, q) := , nq n =1

in the case where p and q are both positive integers and p + q is an odd integer, Flajolet and Salvy [15] gave the identity: X p + i − 1 p p ζ (p + i) η (2k) 2S (p, q) = (1 − (−1) ) ζ (p) η (q) + 2 (−1) p−1 i+2k=q X q + j − 1 (−1)j η (q + j) η (2k) , + η (p + q) − 2 (4) q−1 j+2k=p

where η (0) = 12 , η (1) = ln 2, ζ (1) = 0, and ζ (0) = − 12 in accordance with the analytic continuation of the Riemann zeta function. We also know, from the work of [11] that for odd weight (p + q) we have [ p2 ] ∞ (p) X X Hn p + q − 2j − 1 p BW (p.q) = = (−1) ζ (p + q − 2j) ζ (2j) p−1 nq n =1

j =1

(5)


Integrals of polylogarithmic

351

[ p2 ] X 1 p + q − 2j − 1 p+1 p ζ (p + q − 2j) ζ (2j) + 1 + (−1) ζ (p) ζ (q) + (−1) q−1 2 j =1 ζ (p + q) p+1 p + q − 1 p+1 p + q − 1 1 + (−1) + (−1) , + p q 2 where [z] is the integer part of z. It appears that some isolated cases of BW (p.q) , for even weight (p + q) , can be expressed in zeta terms, but in general, almost certainly, for even weight (p + q) , no general closed form expression exits for BW (p.q) . (at least at the time of writing this paper). Two examples with even weight are ∞ ∞ (2) (4) X X Hn 1 Hn 13 2 = ζ (3) − ζ (6) , = ζ (8) . n4 3 n4 12

n =1

n =1

The work in this paper extends the results of [16] and later [25], in which they gave identities of products of polylogarithmic functions with positive argument in terms of zeta functions. Other works including, [1], [4], [8], [10], [13], [18], [22], [23], [24], cite many identities of polylogarithmic integrals and Euler sums, but none of these examine the negative argument case. The following result was obtained by Freitas, [16]. Lemma 1 For q and t positive integers Z1

X Lit (x) Liq (x) (−1)j+1 ζ (t + j) ζ (q − j + 1) + (−1)q+1 EU (t + q) dx = x q−1

j =1

0

where EU (m) is Euler’s identity given in the next lemma. The following lemma will be useful in the development of the main theorem.

Lemma 2 The following identities hold: for m ∈ N. Euler’s identity states EU (m) =

m−2 ∞ X X Hn (m = (m + 2)ζ + 1) − ζ (m − j) ζ (j + 1) . nm

n =1

j=1

(6)


352

A. Sofo

For p a positive even integer, HE (p) =

∞ X n =1

p Hn = (ζ (p + 1) + η (p + 1)) − (ζ (p) + η (p)) ln 2 (2n + 1)p 2 (7)

p

−1 2 1X (ζ (p + 1 − 2j) + η (p + 1 − 2j)) (ζ (2j) + η (2j)) . − 2 j=1

For p a positive odd integer, HO (p) =

∞ X

Hn p (ζ (p + 1) + η (p + 1)) − (ζ (p) + η (p)) ln 2 p = (2n + 1) 4 n =1 p−1 ! 1 1 + (−1) 2 p+1 p+1 − +η ζ (8) 4 2 2 2 1X (ζ (p − 2j) + η (p − 2j)) (ζ (2j + 1) + η (2j + 1)) − 2 b

j=1

where η (z) is the Dirichlet eta function, b =

h

p−1 4

i

1+(−1) 2

p−1 2

and [z] is

the greatest integer less than z. For p and t positive integers we have ∞ X (−1)n+1 F (p, t) = np (n + 1)t n =1 p X p+t−r−1 p−r (−1) = η (r) p−r r =1 t X p+t−s−1 (−1)p+1 (1 − η (s)) , + t−s

(9)

s =1

G (p, t) =

∞ X

1 p+t−1 p+1 (−1) = p np (n + 1)t n =1 p X p+t−r−1 p−r (−1) + ζ (r) p−r r =2 t X p+t−s−1 (−1)p + ζ (s) , t−s s =1

(10)


Integrals of polylogarithmic

353

Hn p+t−2 p+1 = (−1) ζ (2) HG (p, t) = p (n + 1)t p−1 n n =1 p X p+t−r−1 p−r (−1) EU (r) + p−r r =2 t X p+t−s−1 p (−1) (EU (s) − ζ (s + 1)) . + t−s

(11)

and ∞ X

s =2

Proof. The identity (6) is the Euler relation and by manipulation we arrive at (7) and (8). The results (7) and (8) are closely related to those given by Nakamura and Tasaka [21]. For the proof of (9) we notice that p X 1 1 p+t−r−1 p−r (−1) t = r p p − r n n (n + 1) r =1 t X 1 p+t−s−1 p (−1) + t−s (n + 1)s s =1

therefore, summing over the integers n, p ∞ X X (−1)n+1 p+t−r−1 p−r (−1) F (p, t) = = η (r) p (n + 1)t p−r n n =1 r =1 t X p+t−s−1 p (1 − η (s)) (−1) + t−s s =1

and hence (9) follows. Consider, X p (−1)p+1 1 1 p+t−2 p+t−r−1 p−r (−1) + = t p−1 p−r n (n + 1) nr np (n + 1) r =2 t X 1 p+t−s−1 p (−1) + , t−s (n + 1)s s =2

and summing overP the integers n produces the result (10). The proof of (11) folHn lows by summing ∞ n =1 np (n+1)t in partial fraction form. An example, from (8) ∞ X n =1

Hn (2n + 1)

9

=

9207 889 961 2 511 ζ (10) − ζ (5) − ζ (7) ζ (3) − ζ (9) ln 2 2048 1024 512 256


354

A. Sofo

and from (7), ∞ X

Hn (2n + 1)

n =1

8

=

511 381 441 ζ (9) − ζ (7) ζ (2) − ζ (6) ζ (3) 64 256 256 −

465 255 ζ (5) ζ (4) − ζ (8) ln 2. 256 128

2

Summation identity

We now prove the following theorems. Theorem 1 For positive integers q and t, the integral of the product of two polylogarithmic functions with negative arguments Z1

Z0

I0 (q, t) = Lit (−x) Liq (−x) dx = 0

=

Lit (x) Liq (x) dx −1

q−1 X

(12) (−1)j+1 η (q − j + 1) F (t, j)

j =1

+ (−1)q (F (t, q + 1) − (F (t, q) − G (t, q)) ln 2) + (−1)q Wn (q, t) where the sum ∞ X

1 1 1 Wn (q, t) = − Hn q + (2n)t (2n + 1)q nt (n + 1) (2n + 1)t (2n + 2)q n =1 (13) is obtained from (6), (7), (8) and the terms F (·, ·) , G (·, ·) are obtained from (9) and (10) respectively.

Proof. By the definition of the polylogarithmic function we have Z1 I0 (q, t) = Lit (−x) Liq (−x) dx =

n =1 r =1

0

=

∞ X ∞ X (−1)n+r n =1 r =1

∞ X ∞ X

nt

 

q

(−1)n+r nt rq (n + r + 1)

(−1) + (n + r + 1) (n + 1)q

q X j =1

(−1)

j+1

(n + 1)j rq−j+1

 


Integrals of polylogarithmic

355

  X q ∞ j+1 n+r q+1 X (−1) η (q − j + 1)  (−1)  (−1) q 1 H n+1 − 1 H n + = (n + 1) nt 2 2 2 2 (n + 1)j j =1

n =1

=

q−1 X

(−1)

j+1

η (q − j + 1)

j =1

+ (−1)q

∞ X

(−1)n+1

n =1

nt (n + 1)j

+ (−1)

q

∞ X

(−1)n+1

n =1

nt (n + 1)q+1

∞ X (−1)n+1 1 1 n − ln 2 − . H H n+1 nt (n + 1)q 2 2 2 2

n =1

Now we utilize the double argument identity (2) together with (9) we obtain I0 (q, t) =

q−1 X

(−1)j+1 η (q − j + 1) F (t, j) + (−1)q F (t, q + 1)

j =1

+ (−1)q

∞ X (−1)n+1 n H − H − 2 ln 2 , n 2 nt (n + 1)q

n =1

we can use the alternating harmonic number sum identity (4) to simplify the last sum, however we shall simplify further as follows. I0 (q, t) =

q−1 X

(−1)j+1 η (q − j + 1) F (t, j) + (−1)q F (t, q + 1)

j =1 ∞ X (−1)n+1 n+1 n (−1) (1 (−1) ) H + (−1) − + ln 2 n − Hn q [2] nt (n + 1) q

n =1

where [z] is the integer part of z. Now I0 (q, t) =

q−1 X

(−1)j+1 η (q − j + 1) F (t, j) + (−1)q F (t, q + 1)

j =1

− (−1)q (F (t, q) − G (t, q)) ln 2 + (−1)q Wn (q, t) where Wn (q, t) =

∞ X n =1

Hn

1 1 1 t q − nt (n + 1)q + t (2n) (2n + 1) (2n + 1) (2n + 2)q

and the infinite positive harmonic number sums are easily obtainable from (6), (7), (8), hence the identity (12) is achieved. The next theorem investigates the integral of the product of polylogarithmic functions divided by a linear function.


356

A. Sofo

Theorem 2 Let (t, q) be positive integers, then for t + q an odd integer Z1 I1 (t, q) =

Lit (−x) Liq (−x) dx = − x

0

=

Z0

Lit (x) Liq (x) dx x

−1

q−1 X

(14)

(−1)j+1 η (t + j) η (q − j + 1)

j =1

+ (−1)q+1 (ζ (t + q) + η (t + q)) ln 2 + (−1)q+1 2−t−q − 1 EU (q + t) + (−1)q+1 HO (q + t) . For t + q an even integer q−1 X I1 (t, q) = (−1)j+1 η(t + j)η(q − j + 1) + (−1)q+1 (ζ(t + q) j =1

+ η(t + q)) ln 2 + (−1)q+1 2−t−q − 1 EU (q + t)

(15)

+ (−1)q+1 HE (q + t) . Proof. Consider Z1 I1 (t, q) =

X (−1)n Lit (−x) Liq (−x) dx = x nt n≥1

0

Z1 xn−1 Liq (−x) dx, 0

and successively integrating by parts leads to

I1 (t, q) =

q−1 X (−1)n X n≥1

nt+j

η (q − j + 1) +

j =1

1 X (−1)n+q+1 Z n≥1

nt+q−1

xn−1 Li1 (−x) dx.

0

Evaluating the inner integral, Z1

Z1 x

0

n−1

n−1

Li1 (−x) dx = − x 0

1 ln (1 + x) dx = n

1 1 H n − H n−1 − ln 2 , 2 2 2 2


Integrals of polylogarithmic

357

so that

I1 (t, q) =

q−1 X (−1)n X

nt+j

n≥1

+

(−1)j η (q − j + 1)

j =1

X (−1)n+q+1 1 1 n − H H − ln 2 n−1 nt+q 2 2 2 2

n≥1

=

q−1 X

(−1)j+1 η (q − j + 1) η (t + j)

j =1

X (−1)n+q+1 1 1 + H n − H n−1 − ln 2 . nt+q 2 2 2 2 n≥1

If we now utilize the multiplication formula (2) we can write

I1 (t, q) =

q−1 X

(−1)j+1 η (q − j + 1) η (t + j)+(−1)q+1

X (−1)n+1 n≥1

j =1

nt+q

Hn − H n2 .

Now consider the harmonic number sum X (−1)n+1

Hn − H n2

=

X (−1)n+1

n (1 − (−1) ) ln 2

!

+ (−1)n+1 H[ n ] − Hn 2   n X (−1)j X (−1)n+1 (1 − (−1)n ) ln 2 + (−1)n+1  = nt+q j j=1 n≥1 n+1 X X X (−1) 1 Hn Hn n ) (1 (−1) = − ln 2 + − 1 + nt+q 2t+q nt+q (2n + 1)t+q n≥1 n≥1 n≥1 X 1 X Hn Hn = (ζ (t + q) + η (t + q)) ln 2 + − 1 + 2t+q nt+q (2n + 1)t+q n≥1 n≥1

n≥1

nt+q

n≥1

nt+q

where [z] is the integer part of z. Hence


358

A. Sofo

I1 (t, q) =

q−1 X

(−1)j+1 η (q − j + 1) η (t + j) + (−1)q+1 (ζ (t + q) + η (t + q)) ln 2

j =1

+ (−1)

q+1

X 1 X Hn Hn q+1 (−1) − 1 + t+q t+q 2 n (2n + 1)t+q

n≥1

=

q−1 X

n≥1

(−1)j+1 η (q − j + 1) η (t + j) + (−1)q+1 (ζ (t + q) + η (t + q)) ln 2

j =1

+ (−1)

q+1

+ (−1)q+1

1

− 1 EU (q + t) 2t+q   HO (q + t) , for t + q odd 

, HE (q + t) , for t + q even

hence (14) and (15) follow.

Remark 1 It is interesting to note that, for m ∈ R, Z1

Lit (−xm ) Liq (−xm ) 1 dx = x m

0

Z1

Lit (−x) Liq (−x) dx x

0

The next theorem investigates the integral of the product of polylogarithmic functions divided by a quadratic factor. Theorem 3 For positive integers q and t, the integral of the product of two polylogarithmic functions with negative arguments Z1 I2 (t, q) =

Lit (−x) Liq (−x) dx = x2

0

Z0

Lit (x) Liq (x) dx x2

−1

= η (q + 1) +

q−1 X

(16) (−1)j η (q − j + 1) F (j, t)

j =1 q

+ (−1) (F (q, t) + G (q, t)) ln 2 + (−1)q Wn (t, q) where the sum, Wn (t, q) =

∞ X n =1

Hn

1 1 1 − + (2n)q (2n + 1)t nq (n + 1)t (2n + 1)q (2n + 2)t (17)


Integrals of polylogarithmic

359

is obtained from (6), (7), (8) and the terms F (·, ·) , G (·, ·) are obtained from (9) and (10) respectively. Proof. Following the same process as in Theorem 2, we have, Z1 I2 (t, q) =

X (−1)n Lit (−x) Liq (−x) dx = x2 nt n≥1

0

Z1 = − x−1 Liq (−x) dx +

xn−2 Liq (−x) dx 0

X (−1)n n≥2

0

Z1

Z1 xn−2 Liq (−x) dx,

nt 0

and re ordering the summation index n, produces I2 (t, q) = η (q + 1) +

1 X (−1)n+1 Z

(n + 1)t n≥1

xn−1 Liq (−x) dx.

0

Integrating by parts, we have,  I2 (t, q) = η (q + 1) +

X (−1)n+1   t  (n + 1) 

n≥1

Pq−1

R1

   

(−1)j η(q−j+1) j =1 nj

q+1

+ (−1) nq−1

xn−1 Li1 (−x) dx

0

X (−1)n+1 nj (n + 1)t n≥1 j =1 X (−1)n+q 1 1 n − H H + . n−1 − ln 2 q (n + 1)t 2 2 2 2 n n≥1

= η (q + 1) +

q−1 X

(−1)j η (q − j + 1)

Using the multiplication Theorem (2) and following the same steps as in Theorem 2, we have I2 (t, q) = η (q + 1) +

q−1 X

(−1)j η (q − j + 1) F (j, t)

j =1 q

+ (−1) (F (q, t) + G (q, t)) ln 2 + (−1)q Wn (t, q) , and the proof of Theorem 3 is finalized. The following recurrence relation holds for the reduction of the integral of the product of polylogarithmic functions multiplied by the power of its argument.


360

A. Sofo

Lemma 3 For (q, t) ∈ N and m ≥ 0, let Z0

Z1 m

J (m, q, t) =

x

Lit (−x) Liq (−x) dx = (−1)

0

m

xm Lit (x) Liq (x) dx −1

then (m + 1) J (m, q, t) = η (q) η (t) − J (m, q, t − 1) − J (m, q − 1, t) . For q = 1, (m + 1) J (m, 1, t) = η (t) + mJ (m − 1, 1, t) + J (m − 1, 1, t − 1) − J (m, 1, t − 1) − mK (m, t) − K (m, t − 1) where

Z1 xm Lit (−x) dx.

K (m, t) = 0

Proof. The proof of the lemma follows in a straight forward manner after integration by parts. We list some examples of the results of the integrals in Theorems 1, 2 and 3. Example 1 Z1 (Li3 (−x))2 dx =

I0 (3, 3) =

5 3 9 2 ζ (3) + ζ (4) − ζ (2) ζ (3) 16 8 4

0

+ (3ζ (3) − 6ζ (2) − 40) ln 2 + 4ζ (2) + 12 ln2 2 + 20. Z1

I0 (3, 4) =

49 9 3 Li3 (−x) Li4 (−x) dx = η (4) + ζ (3) 2ζ (5) − ζ (6) − ζ2 (3) 4 64 16 0 5 15 7 + ζ (3) − ζ (2) + 10ζ (2) − 6ζ (3) + ζ (4) + 70 ln 2 4 2 4 17 3 3 − ζ (4) − ζ (5) + ζ (2) ζ (3) − 20 ln2 2 − 35. 2 16 2 Z1

I1 (2m, 2m + 1) = 0

Li2m (−x) Li2m+1 (−x) 1 dx = η2 (2m + 1) x 2


Integrals of polylogarithmic

361

for m ∈ N. Z1 I1 (4, 7) =

1 Li4 (−x) Li7 (−x) dx = η (5) η (7) − η2 (6) . x 2

0

Z1

Li3 (−x) Li4 (−x) 49 23 9 dx = 2ζ (5) − ζ (6) + ζ (4) − ζ2 (3) 2 x 64 8 16 0 7 + 10ζ (3) − 10ζ (2) + 6ζ (3) + ζ (4) ln 2 4 21 3 + 10ζ (2) − ζ (2) ζ (3) − 20 ln2 2 − ζ (3) ζ (4) , 2 32 1 Z 9 9 Li3 (−x) 2 dx = ζ (4) − ζ2 (3) + 6ζ (3) I2 (3, 3) = x 8 16 I2 (3, 4) =

0

3 + 6ζ (2) − ζ (2) ζ (3) − (6ζ (2) + 3ζ (3)) ln 2 − 12 ln2 2. 4 These results build on the work of [16] and [25] where they explored integrals of polylogarithmic functions with positive arguments only. Freitas gives many R1 Liq (x) Lit (x) particular examples of identities for dx, but no explicit identity x2 0

of the form (16) is given. Therefore in the interest of presenting a complete record we list the following theorem. Theorem 4 For positive integers q and t, the integral of the product of two polylogarithmic functions with positive arguments, Z1 P (q, t) =

Liq (x) Lit (x) dx = (−1)q HG (q, t) x2

0

+

q−1 X

(−1)j+1 ζ (t + j) G (j, t) ,

j =1

where G (·, ·) and HG (·, ·) are given by (10) and (11) respectively. Proof. The proof follows the same technique as that used in Theorem 3.


362

A. Sofo

Example 2 Z1 P (4, 5) =

Li4 (x) Li5 (x) 114 dx = 70ζ (2) − 35ζ (3) − ζ (4) − 10ζ (5) x2 5

0

31 5 ζ (6) − ζ2 (3) − 5ζ (2) ζ (3) − 3ζ (3) ζ (4) 4 2 7 − ζ (2) ζ (5) − ζ (8) − ζ (3) ζ (5) , 6

− ζ (4) ζ (5) −

Z1

Li4 x3

Li4 x3 2 2 5 dx = ζ (4) ζ (5) + ζ (2) ζ (7) − ζ (9) . x 3 3 3

0

It is interesting to note the degenerate case, that is when t = 0, of theorems 1, 2 and 3. The following results are noted. x Remark 2 For t = 0, Li0 (−x) = − 1+x , hence

Z1 Liq (−x) Li0 (−x) dx = (−1)q (1 − η (q + 1))

I0 (q, 0) = 0

+

q−1 X

(−1)j+1 η (q − j + 1) (1 − η (j)) − (−1)q (2 − ζ (q) − η (q)) ln 2

j =1

+ (−1)q Z1

I1 (q, 0) =

 HO (q) , for q odd 1 q (EU (q) (q (−1) − 1 − ζ + 1)) + .  2q HE (q) , for q even

X Liq (−x) Li0 (−x) (−1)j+1 η (q − j + 1) η (j) dx = x j =1 0 1 + (−1)q+1 − 1 EU (q) + (−1)q+1 (ζ (q) + η (q)) ln 2 2q   HO (q) , for q odd + (−1)q+1 .  HE (q) , for q even q−1


Integrals of polylogarithmic Z1

363

X Liq (−x) Li0 (−x) (−1)j η (q − j + 1) η (j) I2 (q, 0) = dx = x2 j =1 0 1 q − 1 EU (q) + (−1)q (ζ (q) + η (q)) ln 2 + η (q + 1) + (−1) 2q   HO (q) , for q odd q + (−1) .  HE (q) , for q even q−1

Here we notice that I2 (q, 0) = η (q + 1) − I1 (q, 0) . There are some special cases of polylogarithmic integrals which are worthy of a mention and we list two in the following corollary. Corollary 1 Let q, t ∈ N then, Z1

Liq − x1

S1 (q, t) =

Lit (−x)

x

dx =

q−1 X

η (t + j) η (q − j + 1)

j =1

0

(18) + η (q + t + 1) + (η (q + t) + ζ (q + t)) ln 2   HE (q + t) , for q + t even 1 − 1 EU (q + t) + + .  2q+t H0 (q + t) , for q + t odd Z1 S2 (q) =

Li2 (1 − x) Liq (x) dx = ζ (2) ζ (q + 1) − BW (2, q + 1) , x

(19)

0

where BW (2, q + 1) is given by (5). Proof. If we follow the same procedure as in theorem 2, we obtain Z1 S1 (q, t) =

Liq − x1

Lit (−x)

x 0

+

X (−1)n+1 n≥1

nt+q

dx =

q−1 X j =1

Hn − H n2 .

η (q − j + 1) η (t + j) + η (q + t + 1)


364

A. Sofo

simplifying as in Theorem 2, we arrive at the identity (18). From Euler’s reflection formula we now that Li2 (1 − x) + Li2 (x) + ln x ln (1 − x) = ζ (2) so that Z1 S2 (q) =

(−Li2 (x) − ln x ln (1 − x) + ζ (2)) Liq (x) dx. x

0

Integrating term by term as in theorem 2, we obtain (19)

Example 3 Some examples of the corollary follow.

Z1 S1 (2, 5) =

Li2 − x1

Li5 (−x)

x 0 Z1

S1 (q, q) =

Liq − x1

dx =

Liq (−x)

x

2345 ζ (8) − η (3) η (5) , 768

dx = qζ (2q + 1) ,

0

Z1 S1 (9, 5) =

Li9 − x1

x

Li5 (−x)

dx = 7ζ (15) − η (6) η (9) − η (7) η (8) .

0

Z1 S2 (3) = 0 Z1

S2 (8) =

25 Li2 (1 − x) Li3 (x) dx = ζ (6) − ζ2 (3) x 12 Li2 (1 − x) Li8 (x) dx x

0

= 27ζ (11) − 8ζ (2) ζ (9) − 6ζ (4) ζ (7) − 4ζ (6) ζ (5) − 2ζ (8) ζ (3) . Summary In this paper we have developed new Euler sum identities (7) and (8) of general weight p + 1 for p ∈ N. Moreover, we have developed the new identities (16) and (18). In a series of papers [2], [5], [6], the authors explore linear combinations of associated harmonic polylogarithms and nested


Integrals of polylogarithmic

365

harmonic numbers. The multiple zeta value data mine, computed by Blumlein et. al. [7], is an invaluable tool for the evaluation of harmonic numbers. Values with weights of twelve, for alternating sums and weights above twenty for non-alternating sums are presented.

References [1] J. Ablinger, J. Blümlein, Harmonic sums, polylogarithms, special numbers, and their generalizations. Computer algebra in quantum field theory, 1–32, Texts Monogr. Symbol. Comput., Springer, Vienna, 2013. [2] J. Ablinger, J. Blümlein, C. Schneider, Harmonic sums and polylogarithms generated by cyclotomic polynomials, J. Math. Phys., 52 (10) (2011), 102301. [3] J. Ablinger, J. Blümlein, C. Schneider, Analytic and algorithmic aspects of generalized harmonic sums and polylogarithms, J. Math. Phys., 54 (98) (2013), 74 pp. [4] D. H. Bailey, J. M. Borwein, Computation and structure of character polylogarithms with applications to character Mordell-Tornheim-Witten sums, Math. Comp., 85 (297) (2016), 295–324. [5] J. Blümlein, Algebraic relations between harmonic sums and associated quantities, Comput. Phys. Comm., 159 (1) (2004), 19–54. [6] Blümlein, Johannes Structural relations of harmonic sums and Mellin transforms at weight w = 6. Motives, quantum field theory, and pseudodifferential operators, 167–187, Clay Math. Proc., 12, Amer. Math. Soc., Providence, RI, 2010. [7] J. Blümlein, D. J. Broadhurst, J. A. M. Vermaseren, The multiple zeta value data mine, Comput. Phys. Comm., 181 (3) (2010), 582–625. [8] D. Borwein, J. M. Borwein, D. M. Bradley, Parametric Euler sum identities, J. Math. Anal. Appl., 316 (1) (2006), 328–338. [9] J. M. Borwein, D. M. Bradley, D. J. Broadhurst, P. Lison ěk, Special values of multiple polylogarithms, Trans. Amer. Math. Soc., 353 (3) (2001), 907–941.


366

A. Sofo

[10] J. M. Borwein, I. J. Zucker, J. Boersma, The evaluation of character Euler double sums, Ramanujan J., 15 (3) (2008), 377–405. [11] D. Borwein, J. M. Borwein, R. Girgensohn, Explicit evaluation of Euler sums, Proc. Edinburgh Math. Soc., 38 (2) (1995), 277–294. [12] F. Chavez, C. Duhr, Three-mass triangle integrals and single-valued polylogarithms, J. High Energy Phys., 2012, no. 11, 114, front matter + 31 pp. [13] A. I. Davydychev and Kalmykov, M. Yu, Massive Feynman diagrams and inverse binomial sums. Nuclear Phys. B. 699 (1-2) (2004), 3–64. [14] A. Devoto, D. W. Duke, Table of integrals and formulae for Feynman diagram calculations, Riv. Nuovo Cimento, 7 (6) (1984), 1–39. [15] P. Flajolet, B. Salvy, Euler sums and contour integral representations, Experiment. Math., 7 (1) (1998), 15–35. [16] P. Freitas, Integrals of polylogarithmic functions, recurrence relations, and associated Euler sums, Math. Comp., 74 (251) (2005), 1425–1440. [17] A. Hoorfar, Qi Feng, Sums of series of Rogers dilogarithm functions, Ramanujan J., 18 (2) (2009), 231–238. [18] M. Yu. Kalmykov, O. Veretin, Single-scale diagrams and multiple binomial sums, Phys. Lett. B 483 (1–3) (2000), 315–323. [19] K. S. Kölbig, Nielsen’s generalized polylogarithms, SIAM J. Math. Anal., 17 (5) (1986), 1232–1258. [20] R. Lewin, Polylogarithms and Associated Functions, North Holland, New York, 1981. [21] T. Nakamura, K. Tasaka, Remarks on double zeta values of level 2, J. Number Theory, 133 (1) (2013), 48–54. [22] A. Sofo, Polylogarithmic connections with Euler sums. Sarajevo, J. Math., 12 (24) (2016), no. 1, 17–32. [23] A. Sofo, Integrals of logarithmic and hypergeometric functions, Commun. Math., 24 (1) (2016), 7–22.


Integrals of polylogarithmic

367

[24] A. Sofo, Quadratic alternating harmonic number sums, J. Number Theory, 154 (2015), 144–159. [25] Ce Xu, Yuhuan Yan, Zhijuan Shi, Euler sums and integrals of polylogarithm functions, J. Number Theory, 165 (2016), 84–108.

Received: August 8, 2018


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 368–374 DOI: 10.2478/ausm-2018-0028

A note on some relations between certain inequalities and normalized analytic functions Müfit Şan

Hüseyin Irmak

Department of Mathematics, Faculty of Science, Çankırı Karatekin University, Turkey email: mufitsan@karatekin.edu.tr, mufitsan@hotmail.com

Department of Mathematics, Faculty of Science, Çankırı Karatekin University, Turkey email: hirmak@karatekin.edu.tr, hisimya@yahoo.com

Abstract. In this note, an extensive result consisting of several relations between certain inequalities and normalized analytic functions is first stated and some consequences of the result together with some examples are next presented. For the proof of the presented result, some of the assertions indicated in [5], [8] and [11] along with the results in [3] and [4] are also considered.

1

Introduction, definitions and motivation

Firstly, here and throughout this investigation, let C be the complex plane, U be the unit open disc, i.e., {z ∈ C : |z| < 1} and also let H denote the class of all analytic functions in U. Moreover, a function f(z) ∈ H is said to be a convex function (in U) if f(U) is a convex domain. In this respect, let A be the subclass of all functions H such that f(0) = f 0 (0) − 1 = 0, that is, f(z) ∈ A is of the form f(z) = z + a1 z + a2 z2 + · · · , where z ∈ U and ai ∈ C 2010 Mathematics Subject Classification: 30C45, 26A33, 30C45 Key words and phrases: complex plane, domains in the complex plane, differential inequalities in the complex plane, normalized analytic functions, univalent, starlikeness, convexity

368


Certain inequalities and normalized analytic functions

369

for all i = 1, 2, 3, · · · . In general, the subclass of A consisting of all univalent functions is denoted by S. At the same time, f(z) ∈ A is convex function iff <e{1 + zf 00 (z)/f 0 (z)} > 0 for all z ∈ U. Furthermore, f(z) ∈ H is said to be starlike if f(z) is univalent and f(U) is a starlike domain (with respect to z = 0). It is well-known that f(z) ∈ A is starlike iff <e{zf 0 (z)/f(z)} > 0 for all z ∈ U. The classes K and S ∗ denote the normalized functions’ class of the functions f(z) in S, when f(U) is convex and f(U) is starlike, respectively. The class S ∗ (α) denotes the class of all starlike functions f(z) of order α (0 ≤ α < 1) if f(z) ∈ A and <e{zf 0 (z)/f(z)} > α for all z ∈ U. Besides, the class K(α) denotes the class of all sconvex functions f(z) of order α (0 ≤ α < 1) if f(z) ∈ A and <e{1 + zf 0 (z)/f(z)} > α for all z ∈ U. Namely, K(α) is the class of all convex functions f(z) ∈ A satisfying the condition <e{1 + zf 00 (z)/f 0 (z)} > α for all z ∈ U and for some α (0 ≤ α < 1). In addition, let S ∗ := S ∗ (0) and K := K(0), which are the subclasses of starlike and convex functions with respect to the origin (z = 0) in U, respectively. (See, for the details of the related definitions (and also information), [1], [2], and see also (for novel examples) [3], [4], [6], [7].) The literature presents us several works including important or interesting results between certain inequalities and certain classes of the functions which are analytic and univalent in the disc U. For those, one may look over the earlier results presented in [3], [8], [9], [10] and [11]. In particularly, in [8], the problem of finding λ > 0 such that the condition |f 00 (z)|, where f(z) ∈ A and z ∈ U, implies f(z) ∈ S ∗ , was firstly considered by P. T. Mocanu for λ = 2/3. √ Later, in [9], S. Ponnusamy and V. Singh considered the problem for λ = 2/ 3. Afterwards, in [10], M. Obradović focused on the problem for λ = 1 by proving that his result is sharp. In [11], N. Tuneski also obtained certain results dealing with the same problems, which are also generalizations of the results of M. Obradović in [10]. In this investigation, by using a different technique, developed by S. S. Miller and P. T. Mocanu in [5], certain results determined by the functions f(z) ∈ A relating to both condition |f 00 (z)| ≤ λ for some values of λ > 0 and the classes S ∗ (α) and K(α) are restated and then their certain consequences which will be important for (analytic and) geometric function theory are given. In addition, only for the proofs of these consequences of our main results derived in the Section 2 of this paper, both the assertion of S. S. Miller and P. T. Mocanu given in [3] and the results of N. Tuneski given in [11] are also used. The following two assertions (Lemma 1 in [3] and Lemma 2 in [11] below) will be required to prove the main results.


370

M. Şan, H. Irmak

Lemma 1 Let f(z) ∈ A, z ∈ U and 0 ≤ α < 1. Then,

(2 − α) f 00 (z) ≤ 2(1 − α) ⇒ f(z) ∈ S ∗ (α). The result is sharp. Lemma 2 Let f(z) ∈ A, z ∈ U and 0 ≤ α < 1. Then,

(2 − α) f 00 (z) ≤ 1 − α ⇒ f(z) ∈ K(α). The result is sharp. The following important assertion (see, for its details and also example, [3] (p. 33-34 and a = 0)) will be required to prove the main results. Lemma 3 Let Ω ⊂ C and suppose that the function ψ : C2 × U → C satisfies ψ Meiθ , Keiθ ; z ∈ / Ω for all K ≥ Mn , θ ∈ R, and z ∈ U. If the function p(z) is in the class:

H 0, n := p(z) ∈ H : p(z) = an zn + an+1 zn+1 + . . . . (z ∈ U) and ψ p(z), zp0 (z); z ∈ Ω , then |p(z)| < M, where for some M > 0 and for all z ∈ U.

2

The main results, implications and examples

By making use of Lemma 3, we shall firstly give and then prove the main result, which is given by Theorem 1 Let f(z) = z + a1 z + a2 z2 + a3 z3 + · · · ∈ A, z ∈ U, 0 < δ < M, and let f 00 (z) 6= 2a2 − δ. Then,

00

zf000 (z) M (M − δ)

f (z) < M + 2|a2 | . <e < ⇒ δ − 2a2 + f00 (z) δ2 + (δ + M)2 Proof. Let us define p(z) by p(z) = f 00 (z) − 2a2 ,


Certain inequalities and normalized analytic functions

371

where f(z) = z + a1 z + a2 z2 + a3 z3 + · · · ∈ A and z ∈ U. Clearly, p(z) is in the class H[0, 1] (when, of course, a3 6= 0). Then, it immediately follows that zp0 (z) zf000 (z) = δ + p(z) δ − 2a2 + f00 (z) Let ψ(r, s; z) :=

f 00 (z) 6= 2a2 − δ; z ∈ U . s δ+r

and Ω :=

w : w ∈ C and <e{w} <

M (M − δ) δ2 + (δ + M)2

.

Then we have 0

ψ p(z), zp (z); z =

zp0 (z) = δ + p(z)

zf000 (z) ∈Ω δ − 2a2 + f00 (z)

for all z in U. Furthermore, for any θ ∈ R, K ≥ nM ≥ M, and z ∈ U, we obviously obtain that Keiθ M (M − δ) iθ iθ <e ψ Me , Ke ; z = <e , ≥ iθ 2 δ + Me δ + (δ + M)2 i.e., ψ Meiθ , Keiθ ; z ∈ 6 Ω. Therefore, in respect of the Lemma 3, the definition of p(z) easily yields that

p(z) = f00 (z) − 2a2 < M (M > 0; z ∈ U) , which completes the desired proof.

Proposition 1 Let f(z) = z + a1 z + a2 z2 + a3 z3 + · · · ∈ A, z ∈ U, 0 < δ < 1, and let f 00 (z) 6= 2a2 − δ. Then, zf000 (z) <e < Φ (α, δ, a2 ) ⇒ f(z) ∈ S ∗ (α), δ − 2a2 + f00 (z) where Φ (α, δ, a2 ) :=

[2 (1 − α) − 2 (2 − α) |a2 |] [2 (1 − α) − (2 − α) (2 |a2 | + δ)] (2 − α)2 δ2 + [(2 − α) (δ − 2 |a2 |) + 2 (1 − α)]2

.


372

M. Şan, H. Irmak

Proof. If we take M + 2|a2 | :=

2(1 − α) 2−α

(0 ≤ α < 1)

in Theorem 1 and just then use Lemma 1, we easily get the proof.

By letting α := 0 in Proposition 1, we first obtain the following corollary. Corollary 1 Let f(z) = z + a1 z + a2 z2 + a3 z3 + · · · ∈ A, z ∈ U, 0 < δ < 1 , and let f 00 (z) 6= 2a2 − δ. Then, <e

zf000 (z) δ − 2a2 + f00 (z)

<

(1 − 2 |a2 |) (1 − 2 |a2 | − δ) 2

(1 − 2 |a2 | + δ) +

⇒ f(z) ∈ S ∗ .

δ2

By taking δ := 2 |a2 | in Corollary 1, we next have the following corollary. Corollary 2 Let f(z) = z + a1 z + a2 z2 + a3 z3 + · · · ∈ A, z ∈ U, 0 < 2 |a2 | < 1, and let f 00 (z) 6= 0. Then, <e

zf000 (z) f00 (z)

<

1 − 2 |a2 | 2

1 + 4 |a2 |

⇒ f(z) ∈ S ∗ .

For this result (i.e., for Corollary 2), the following example can be easily given. Example 1 Take f(z) = z + 14 z2 + a3 z3 and let |a3 | <

1 18 .

Since

00 1

f (z) = + 6a3 z ≥ 1 − 6|a3 | > 1 − 1 = 1 > 0 ,

2

2 2 3 6 we arrive at f 00 (z) 6= 0. Besides, it is obvious that |a2 | = time, clearly, <e

zf000 (z) f00 (z)

= 1 − <e

1 1 + 12a3 z

<

1 4

< 12 . At the same

1 − 2|a2 | 2 = . 2 1 + 4|a2 | 5

In that case, as a result 2, it is clear that f(z) ∈ S ∗ . We also

00 of

Corollary

1 indicate that, since f (z) = 2 + 6a3 z < 1, Lemma 1 immediately implies that the function f(z) is starlike in U.


Certain inequalities and normalized analytic functions

373

Proposition 2 Let f(z) = z + a1 z + a2 z2 + a3 z3 + · · · ∈ A, z ∈ U, 0 < δ < 21 , and let f 00 (z) 6= 2a2 − δ. Then, zf000 (z) < Φ (α, δ, a2 ) ⇒ f(z) ∈ K(α) , <e δ − 2a2 + f00 (z) where Φ (α, δ, a2 ) :=

[(1 − α) − 2 (2 − α) |a2 |] [(1 − α) − (2 − α) (2 |a2 | + δ)] (2 − α)2 δ2 + [(2 − α) (δ − 2 |a2 |) + (1 − α)]2

.

Proof. If we put 1−α (0 ≤ α < 1) 2−α in Theorem 1 and just then use Lemma 2, we easily arrive at the desired result in Proposition 2. By putting α = 0 in Proposition 2, we then get the following result. M + 2|a2 | :=

Corollary 3 Let f(z) = z + a1 z + a2 z2 + a3 z3 + · · · ∈ A, z ∈ U, 0 < δ < 21 , and let f 00 (z) 6= 2a2 − δ. Then, (1 − 4 |a2 |) (1 − 4 |a2 | − 2δ) zf000 (z) <e < ⇒ f(z) ∈ K . 00 δ − 2a2 + f (z) (1 − 4 |a2 | + 2δ)2 + 4δ2 By setting δ := 2 |a2 | in Corollary 3, we also get the following corollary. Corollary 4 Let f(z) = z + a1 z + a2 z2 + a3 z3 + · · · ∈ A, z ∈ U, 0 < 2 |a2 | < 21 , and let f 00 (z) 6= 0. Then, 000 zf (z) 1 − 4 |a2 | <e ⇒ f(z) ∈ K . < 00 f (z) 1 + 4 |a2 |2 The following can be also given to exemplify the result given above. Example 2 Take f(z) = z + 18 z2 + a3 z3 and let |a3 | < 241√2 . Since

00 1

1

f (z) = + 6a3 ≥ 1 − 6|a3 | > 1 − √ >0,

4

4 4 4 2 we obtain f 00 (z) 6= 0. Furthermore, it is clear that |a2 | = 41 < 21 . At the same time, obviously, 000 zf (z) 1 1 − 4|a2 | 8 <e = 1 − <e < = . f00 (z) 1 + 24a3 z 1 + 4|a2 |2 17 In of Corollary 4, it is clear that f(z) ∈ K. Then, since

00 this case,

as a result

f (z) = 1 + 6a3 < 1, Lemma2 immediately implies that function f(z) is 4 convex in U.


374

M. Şan, H. Irmak

References [1] P. L. Duren, Univalent Functions, in: A Series of Comprehensive Studies in Mathematics, vol. 259, Springer-Verlag, New York, Berlin, Heidelberg, Tokyo, 1983. [2] A. W. Goodman, Univalent Functions, Vols. I and II, Polygonal Publishing House, Washington, New Jersey, 1983. [3] H. Irmak, M. Şan, Some relations between certain inequalities concerning analytic and univalent functions, Appl. Math. Lett., 23 (8) (2010), 897– 901. [4] H. Irmak, T. Bolboaca, N. Tuneski, Some relations between certain classes consisting of α-convex type and Bazilević type functions, Appl. Math. Lett., 24 (12) (2011), 2010–2014. [5] S. S. Miller, P. T. Mocanu, Differential Subordinations, Theory and Applications, Marcel Dekker, New York-Basel, 2000. [6] M. Şan, H. Irmak, Some novel applications of certain higher order ordinary complex differential equations to normalized analytic functions, J. Appl. Anal. Comput., 5 (3) (2015), 479–484. [7] M. Şan, H. Irmak, Some results consisting of certain inequalities specified by normalized analytic functions and their implications, Acta Univ. Apulensis Math. Inform. 46 (2016), 107–114. [8] P. T. Mocanu, Two simple sufficient conditions for starlikenes, Mathematica (Cluj), 34 (57) (1992), 175–181. [9] S. Ponnusamy, V. Singh, Criteria for univalent, starlike and convex functions, Bull. Belg. Math. Soc. Simon Stevin, 9 (4) (2002), 511–531. [10] M. Obradović, Simple sufficient conditions for univalence, Mat. Vesnik, 49 (3–4) (1997), 241–244. [11] N. Tuneski, On some simple sufficient conditions for univalence, Math. Bohem., 126 (1) (2001), 229–236.

Received: June 11, 2017


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 375–377 DOI: 10.2478/ausm-2018-0029

Finite groups with a certain number of cyclic subgroups II Marius Tărnăuceanu Faculty of Mathematics, “Al. I. Cuza” University, Iaşi, Romania email: tarnauc@uaic.ro

Abstract. In this note we describe the finite groups G having |G| − 2 cyclic subgroups. This partially solves the open problem in the end of [3].

Let G be a finite group and C(G) be the poset of cyclic subgroups of G. The connections between |C(G)| and |G| lead to characterizations of certain finite groups G. For example, a basic result of group theory states that |C(G)| = |G| if and only if G is an elementary abelian 2-group. Recall also the main theorem of [3], which states that |C(G)| = |G| − 1 if and only if G is one of the following groups: Z3 , Z4 , S3 or D8 . In what follows we shall continue this study by describing the finite groups G for which |C(G)| = |G| − 2.

(∗)

First, we observe that certain finite groups of small orders, such as Z6 , Z2 ×Z4 , D12 and Z2 × D8 , have this property. Our main theorem proves that in fact these groups exhaust all finite groups G satisfying (∗). Theorem 1 Let G be a finite group. Then |C(G)| = |G| − 2 if and only if G is one of the following groups: Z6 , Z2 × Z4 , D12 or Z2 × D8 . 2010 Mathematics Subject Classification: Primary 20D99; Secondary 20E34 Key words and phrases: cyclic subgroups, finite groups

375


376

M. Tărnăuceanu

Proof. We will use the same technique as in the proof of Theorem 2 in [3]. Assume that G satisfies (∗), let n = |G| and denote by d1 = 1, d2 , . . . , dk the positive divisors of n. If ni = |{H ∈ C(G) | |H| = di }|, i = 1, 2, . . . , k, then k X

ni φ(di ) = n.

i=1

Since |C(G)| =

Pk

i=1 ni

= n − 2, one obtains k X

ni (φ(di ) − 1) = 2,

i=1

which implies that we have the following possibilities: Case 1. There exists i0 ∈ {1, 2, . . . , k} such that ni0 (φ(di0 ) − 1) = 2 and ni (φ(di ) − 1) = 0, ∀ i 6= i0 . Since the image of the Euler’s totient function does not contain odd integers > 1, we infer that ni0 = 2 and φ(di0 ) = 2, i.e. di0 ∈ {3, 4, 6}. We remark that di0 cannot be equal to 6 because in this case G would also have a cyclic subgroup of order 3, a contradiction. Also, we cannot have di0 = 3 because in this case G would contain two cyclic subgroups of order 3, contradicting the fact that the number of subgroups of a prime order p in G is ≡ 1 (mod p) (see e.g. the note after Problem 1C.8 in [1]). Therefore di0 = 4, i.e. G is a 2-group containing exactly two cyclic subgroups of order 4. Let n = 2m with m ≥ 3. If m = 3 we can easily check that the unique group G satisfying (∗) is Z2 × Z4 . If m ≥ 4 by Proposition 1.4 and Theorems 5.1 and 5.2 of [2] we infer that G is isomorphic to one of the following groups: - M2m ; - Z2 × Z2m−1 ; m−2

- ha, b | a2

m−3

= b8 = 1, ab = a−1 , a2

= b4 i, where m ≥ 5;

- Z2 × D2m−1 ; m−2

m−4

- ha, b | a2 = b2 = 1, ab = a−1+2 where m ≥ 5.

m−3

c, c2 = [c, b] = 1, ac = a1+2

i,


Finite groups with a certain number of cyclic subgroups II

377

All these groups have cyclic subgroups of order 8 for m ≥ 5 and thus they do not satisfy (∗). Consequently, m = 4 and the unique group with the desired property is Z2 × D8 . Case 2. There exist i1 , i2 ∈ {1, 2, . . . , k}, i1 6= i2 , such that ni1 (φ(di1 ) − 1) = ni2 (φ(di2 ) − 1) = 1 and ni (φ(di ) − 1) = 0, ∀ i 6= i1 , i2 . Then ni1 = ni2 = 1 and φ(di1 ) = φ(di2 ) = 2, i.e. di1 , di2 ∈ {3, 4, 6}. Assume that di1 < di2 . If di2 = 4, then di1 = 3, that is G contains normal cyclic subgroups of orders 3 and 4. We infer that G also contains a cyclic subgroup of order 12, a contradiction. If di2 = 6, then we necessarily must have di1 = 3. Since G has a unique subgroup of order 3, it follows that a Sylow 3-subgroup of G must be cyclic and therefore of order 3. Let n = 3·2m , where m ≥ 1. Denote by n2 the number of Sylow 2-subgroups of G and let H be such a subgroup. Then H is elementary abelian because G does not have cyclic subgroups of order 2i with i ≥ 2. By Sylow’s Theorems, n2 |3 and n2 ≡ 1 (mod 2), ∼ Zm × Z3 , a implying that either n2 = 1 or n2 = 3. If n2 = 1, then G = 2 ∼ Z6 . If n2 = 3, then group that satisfies (∗) if and only if m = 1, i.e. G = |CoreG (H)| = 2m−1 because G/CoreG (H) can be embedded in S3 . It follows that G contains a subgroup isomorphic with Zm−1 × Z3 . If m ≥ 3 this has 2 more than one cyclic subgroup of order 6, contradicting our assumption. Hence ∼ S3 , a group that does not either m = 1 or m = 2. For m = 1 one obtains G = have cyclic subgroups of order 6, a contradiction, while for m = 2 one obtains ∼ D12 , a group that satisfies (∗). This completes the proof. G=

References [1] I. M. Isaacs, Finite group theory, Amer. Math. Soc., Providence, R. I., 2008. [2] Z. Janko, Finite 2-groups G with |Ω2 (G)| = 16, Glas. Mat. Ser. III, 40 (2005), 71–86. [3] M. Tărnăuceanu, Finite groups with a certain number of cyclic subgroups, Amer. Math. Monthly, 122 (2015), 275–276. Received: December 2, 2017


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 378–394 DOI: 10.2478/ausm-2018-0030

New classes of local almost contractions Mónika Zákány Technical University of Cluj-Napoca, North University Center of Baia Mare, Romania email: zakanymoni@yahoo.com

Abstract. Contractions represents the foundation stone of nonlinear analysis. That is the reason why we propose to unify two different type of contractions: almost contractions, introduced by V. Berinde in [2] and local contractions (Martins da Rocha and Filipe Vailakis in [7]). These two types of contractions operate in different space settings: in metric spaces (almost contractions) and semimetric spaces (for local contractions). That new type of contraction was built up in a new space setting, which is the pseudometric space. The main results of this paper represent the extension for various type of operators on pseudometric spaces, such as: generalized ALC, Ćirić-type ALC, quasi ALC, Ćirić-Reich-Rus type ALC. We propose to study the existence and uniqueness of their fixed points, and also the continuity in their fixed points, with a large number of examples for ALC-s.

1

Introduction

First, we present the concept of almost contraction, following V. Berinde in [2]. Definition 1 (see [2]) Let (X, d) be a metric space. T : X → X is called almost contraction or (δ, L)- contraction if there exist a constant δ ∈ (0, 1) and some L ≥ 0 such that d(Tx, Ty) ≤ δ · d(x, y) + L · d(y, Tx), ∀ x, y ∈ X.

(1)

2010 Mathematics Subject Classification: 47H10,54H25 Key words and phrases: almost local contraction, coefficient of contraction, fixed- point theorem

378


New classes of local almost contractions

379

Remark 1 The term of almost contraction is equivalent to weak contraction, and it was first introduced by V. Berinde in [2]. Because of the simmetry of the distance, the almost contraction condition (1) includes the following dual one: d(Tx, Ty) ≤ δ · d(x, y) + L · d(x, Ty), ∀ x, y ∈ X,

(2)

obtained from (1) by replacing d(Tx, Ty) by d(Ty, Tx) and d(x, y) by d(y, x). Obviously, to prove the almost contactiveness of T , it is necessary to check both (1) and (2). A strict contraction satisfies (1), with δ = a and L = 0, therefore it is an almost contraction with a unique fixed point. Many examples of almost contractions are given in [1]-[3]. Weak contractions represent a generous concept, due to various mappings satisfying the condition (1). Such examples of weak contraction was given by V. Berinde in [2]. Definition 2 [5] Let (X, d) be a metric space. Any mapping T : X → X is called Ćirić-Reich-Rus contraction if it is satisfied the condition: d(Tx, Ty) ≤ α · d(x, y) + β · [d(x, Tx) + d(y, Ty)], ∀ x, y ∈ X,

(3)

where α, β ∈ R+ and α + 2β < 1. Proposition 1 (see [8]) Let (X, d) be a metric space. Any Ćirić-Reich-Rus contraction,i.e., any mapping T : X → X satisfying the condition (3), represent an almost contraction. Theorem 1 A mapping satisfying the contractive condition: there exists 0 ≤ h < 12 such that d(Tx, Ty) ≤ h · max{d(x, y), d(x, Tx), d(y, Ty), d(x, Ty), d(y, Tx)},

(4)

for all x, y ∈ X, is a weak contraction. An operator satisfying (4) with 0 < h < 1 is called quasi-contraction. Remark 2 Theorem 1 prove that quasi-contractions with 0 < h < 12 are always weak contractions. However, there exists quasi-contractions with h ≥ 12 , presented in Example 1 by V. Berinde in [2], as it follows:


380

M. Zákány

Example 1 Let T : [0, 1] → [0, 1] a mapping given by Tx = 32 for x ∈ [0, 1), and T1 = 0. Then T has the following properties: 1) T satisfies (4) with h ∈ [ 23 , 1), i.e., T is quasi-contraction; 2) T satisfies (1), with δ ≥ 23 and L ≥ δ, i.e., T is also weak contraction; 3) T has a unique fixed point, x∗ = 23 . Since we were familiarized with the class of almost contractions, we introduce the concept of local contractions, another interesting type of operators with unexpected applications. The concept of local contraction was presented by Martins da Rocha and Filipe Vailakis in [7]. Definition 3 (see [7]) Let F be a set and let D = (dj )j∈J be a family of semidistances defined on F. We let σ be the weak topology on F defined by the family D. A sequence (fn )n∈N∗ is said to be σ − Cauchy if it is dj -Cauchy, ∀ j ∈ J. A subset A of F is said to be sequencially σ-complete if every σ-Cauchy sequence in A converges in A for the σ-topology. A subset A ⊂ F is said to be σ-bounded if diamj (A) ≡ sup{dj (f, g) : f, g ∈ A} is finite for every j ∈ J. Let r be a function from J to J. An operator T : F → F is called local contraction with respect (D, r) if, for every j, there exists βj ∈ [0, 1) such that ∀ f, g ∈ F,

dj (Tf, Tg) ≤ βj dr(j) (f, g).

Definition 4 The mapping d(x, y) : X × X → R+ is said to be a pseudometric if: 1. d(x, y) = d(y, x); 2. d(x, y) ≤ d(x, z) + d(z, y); 3. x = y implies d(x, y) = 0 (instead of x = y ⇔ d(x, y) = 0 in the metric case). Definition 5 (see [11]) Let r be a function from J to J. An operator T : F → F is an almost local contraction (ALC) with respect (D, r) or (δ, L)- contraction, if there exist a constant δ ∈ (0, 1) and some L ≥ 0 such that dj (Tf, Tg) ≤ δ · dj (f, g) + L · dr(j) (g, Tf), ∀ f, g ∈ F.

(5)

Theorem 2 [11] Assume that the space F is σ- Hausdorff, which means: for each pair f, g ∈ F, f 6= g, there exists j ∈ J such that dj (f, g) > 0.


New classes of local almost contractions

381

If A is a nonempty subset of F, then for each h in F, we let dj (h, A) ≡ {dj (h, g) : g ∈ A}. Consider a function r : J → J and let T : F → F be an almost local contraction with respect to (D, r). Consider a nonempty, σ- bounded, sequentially σ- complete, and T - invariant subset A ⊂ F. (E) If the condition ∀ j ∈ J,

lim βj βr(j) · · · βrn (j) diamrn+1 (j) (A) = 0

n→∞

(6)

is satisfied, then the operator T admits a fixed point f∗ in A. (S) Moreover, if h ∈ F satisfies ∀ j ∈ J,

lim βj βr(j) · · · βrn (j) drn+1 (j) (h, A) = 0,

n→∞

(7)

then the sequence (T n h)n∈N is σ- convergent to f∗ . Example 2 Let X = [0, n] × [0, n] ⊂ R2 , n ∈ N∗ , T : X → X,

x y ( 2 , 2 ) if (x, y) 6= (1, 0) T (x, y) = (0, 0) if (x, y) = (1, 0) The diameter of the subset X = [0, n] × [0, n] ⊂ R2 is given by the diagonal line of the square whose four sides have length n . We shall use the pseudometric: dj (x1 , y1 ), (x2 , y2 ) = |x1 − x2 | · e−j , ∀ j ∈ J, (8) where J is a subset of N. This is a pseudometric, but not a metric, take for example: dj ((1, 4), (1, 3)) = |1 − 1| · e−j = 0, however (1, 4) 6= (1, 3) In this case, we shall use the function r(j) = 2j . By applying the inequality (5) to our mapping T , we get for all x = (x1 , y1 ), y = (x2 , y2 ) ∈ X

x1 x2 −j x1

−j

− · e ≤ θ · |x1 − x2 | · e −j 2 + L · x − ·e 2 , 2 2 2 2 for all j ∈ J, which can be write as the equivalent form −j

|x1 − x2 | · e 2 ≤ 2θ · |x1 − x2 | + L · |2x2 − x1 |, The last inequality became true if we take θ = 12 ∈ (0, 1), L = 4 ≥ 0. Hence T is an almost local contraction, with the unique fixed point (0, 0). T is continuous in the fixed point, at (0, 0) ∈ Fix(T ), but is not continuous at (1, 0) ∈ / Fix(T ).


382

M. Zákány

Example 3 With the assumptions from the previous example and the pseudometric defined by (8) where j ∈ J, and r(j) = 2j , we get another example for almost local contractions. Considering T : X → X,

(x, −y) if (x, y) 6= (1, 1) T (x, y) = (0, 0) if (x, y) = (1, 1) T is not a contraction because the contractive condition: dj (Tx, Ty) ≤ θ · dj (x, y),

(9)

is not valid ∀ x, y ∈ X, and for any θ ∈ (0, 1). Indeed, (9) is equivalent with: | x1 − x2 | · e−j ≤ θ · |x1 − x2 | · e−j , ∀ j ∈ J. The last inequality leads us to 1 ≤ θ, which is obviously false, considering θ ∈ [0, 1). However, T becomes an almost local contraction if: −j

−j

|x1 − x2 | · e−j ≤ θ · |x1 − x2 | · e 2 + L · |x2 − x1 | · e 2 −j

which is equivalent to : e 2 ≤ θ + L. For θ = 31 ∈ [0, 1) , L = 2 ≥ 0 and j ∈ J, the last inequality becomes true, i.e. T is an almost local contraction with many fixed points: FixT = {(x, 0) : x ∈ R}. In this case, we have: ∀ j ∈ J,

lim θ

n→∞

n+1

n+1 1 diamrn+1 (j) (A) = lim · (n − 1)2 = 0 n→∞ 3

This way, the existence of the fixed point is assured, according to condition (E) from Theorem 2. The continuity of T in (0, 0) ∈ Fix(T ) is valid, but we have discontinuity in (1, 1), which is not a fixed point of T . Example 4 Let X be the set of positive functions: X = {f|f : [0, ∞) → [0, ∞)}, which is a subset of the real functions F = {f : R → R}. Let dj (f, g) = |f(0) − g(0)| · e−j , ∀ f, g ∈ X, r(j) = 2j , ∀ j ∈ J. Indeed, dj is a pseudometric, but not a metric, take for example dj (x, x2 ) = 0, but x 6= x2 .


New classes of local almost contractions

383

Considering the mapping Tf = |f|, ∀ f ∈ X, and using the inequality (1) from the definition of almost local contractions: −j

−j

|f(0) − g(0)| · e−j ≤ θ · |f(0) − g(0)| · e 2 + L · |g(0) − f(0)| · e 2

which is equivalent to: e−j/2 ≤ θ + L. This inequality becames true if j > 0, θ = 1 L = 3 > 0. Hence, T is an ALC. However, T is not a contraction, 4 ∈ (0, 1), because the contractive condition (9) leads us again to the false assumption: 1 ≤ θ. The mapping T has infinite number of fixed points: FixT = {f ∈ X} = X, by taking: |f(x)| = f(x), ∀ f ∈ X, x ∈ [0, ∞)

2

Main results

The main results of this paper represent the extension for various type of operators on pseudometric spaces, such as: generalized ALC, Ćirić-type ALC, quasi ALC, Ćirić-Reich-Rus type ALC. a) Generalized ALC Definition 6 Let r be a function from J to J. Let A ⊂ F be a τ-bounded sequencially τ-complete and T - invariant subset of F. A mapping T : A → A is called generalized almost local contraction if there exist a constant θ ∈ (0, 1) and some L ≥ 0 such that ∀ x, y ∈ X, ∀ j ∈ J we have: dj (Tx, Ty) ≤ θ · dr(j) (x, y) + L · min{dr(j) (x, Tx), dr(j) (y, Ty), dr(j) (x, Ty), dr(j) (y, Tx)}

(10)

Remark 3 It is obvious that any generalized almost local contraction is an almost contraction, i.e., it does satisfy inequality (1). Theorem 3 Let T : A → A be a generalized almost local contraction, i.e., a mapping satisfying (10), and also verifying the condition (7) for the unicity of fixed point. Let Fix(T ) = {f}. Then T is continuous at f. Proof. Since T is a generalized almost local contraction, there exist a constant θ ∈ (0, 1) and some L ≥ 0 such that (10) is satisfied. We know by Theorem 7 that T has a unique fixed point, say f. Let {yn }∞ n=0 be any sequence in X converging to f. Then by taking y := yn ,

x := f


384

M. Zákány

in the generalized almost local contraction condition (10), we get dj (Tf, Tyn ) ≤ θ · dr(j) (f, yn ), n = 0, 1, 2, · · ·

(11)

since f is a fixed point for T , we have min{dr(j) (x, Tx), dr(j) (y, Ty), dr(j) (x, Ty), dr(j) (y, Tx)} = dr(j) (f, Tf) = 0. Now, by letting n → ∞ in (11), we get Tyn → Tf, which shows that T is continuous at f. b) Ćirić-type almost local contraction Definition 7 (see Berinde, [4]) Let (X, d) be a complete metric space. The mapping T : X → X is called Ćirić almost contraction if there exist a constant α ∈ [0, 1) and some L ≥ 0 such that d(Tx, Ty) ≤ α · M(x, y) + L · d(y, Tx), for all x,y ∈ X,

(12)

where M(x, y) = max{d(x, y), d(x, Tx), d(y, Ty), d(x, Ty), d(y, Tx)}. From the above definition the following question arises: it is possible to expand it to the case of almost local contractions? The answer is affirmative and is given by the next definition. But first we need to remind the Lemma of Ćirić ([6]), which will be essential in proving our main results. Lemma 1 Let T be a quasi-contraction on X and let n be any positive integer. Then, for each x ∈ X, and all positive integers i, j, where i, j ∈ {1, 2, · · · n} implies d(T i x, T j x) ≤ h · δ[O(x, n)], where we denoted δ(A) = sup{d(a, b) : a, b ∈ A} for a subset A ⊂ X. Remark 4 Observe that, by means of Lemma 1, for each n, there exist k ≤ n such that d(x, T k x) = δ[O(x, n)]. Lemma 2 (see [6]) Let T be a quasi-contraction on X. Then the inequality 1 δ[O(x, n)] ≤ d(x, T k x) 1−h holds for all x ∈ X.


New classes of local almost contractions

385

Definition 8 Under the assumptions of definition 5, the operator T : A → A is called Ćirić-type almost local contraction with respect (D, r) if, for every j ∈ J, there exist the constants θ ∈ [0, 1) and L ≥ 0 such that dj (Tf, Tg) ≤ θ · Mr(j) (f, g) + L · dr(j) (g, Tf), for all f,g ∈ A,

(13)

where Mr(j) (f, g) = max dr(j) (f, g), dr(j) (f, Tf), dr(j) (g, Tg), dr(j) (f, Tg), dr(j) (g, Tf) . Remark 5 Although this class is more wide than the one of almost local contractions, similar conclusions can be stated as in the case of almost local contractions, as it follows: Theorem 4 Consider a function r : J → J, let a nonempty, τ- bounded, sequentially τ- complete, and T - invariant subset A ⊂ X and let T : A → A be Ćirić- type almost local contraction with respect to (D,r). Then 1. T has a fixed point,i.e., Fix(T ) = {x ∈ X : Tx = x} 6= φ; ∗ 2. For any x0 = x ∈ A, the Picard iteration {xn }∞ n=0 converges to x ∈ Fix(T );

3. The following a priori estimate is available: dj (xn , x∗ ) ≤

θn dj (x, Tx), (1 − θ)2

n = 1, 2...

(14)

Proof. For the conclusion of the Theorem, we have to prove that T has at least a fixed point in the subset A ⊂ X. To this end, let x ∈ A be arbitrary, and let {xn }∞ n ∈ N with n=0 be the Picard iteration defined by xn+1 = Txn , x0 = x. Take x := xn−1 , y := xn in (13) to obtain dj (xn , xn+1 ) = dj (Txn−1 , Txn ) ≤ θ · Mr(j) (xn−1 , xn ), since dj (xn , Txn−1 ) = dj (Txn−1 , Txn−1 ) = 0. Continuing in this manner, for n ≥ 1, by Lemma 1 we have dj (T n x, T n+1 x) = dj (TT n−1 x, T 2 T n−1 x) ≤ θ · δ[O(T n−1 x, 2)]. By using Remark 4, we can easily conclude: there exist a positive integer k1 ∈ {1, 2} such that δ[O(T n−1 x, 2)] = dj (T n−1 x, T k1 T n−1 x)


386

M. Zákány

and therefore dj (xn , xn+1 ) ≤ θ · dj (T n−1 x, T k1 T n−1 x). By using once again Lemma 1, we obtain, for n ≥ 2, dj (T n−1 x, T k1 T n−1 x) = dj (TT n−2 x, T k1 +1 T n−2 x) ≤ ≤ θ · δ[O(T n−2 x, k1 + 1)] ≤ θ · δ[O(T n−2 x, 3)]. Continuing in this manner, we get dj (T n x, T n+1 x) ≤ θ · δ[O(T n−1 x, 2)] ≤ θ2 · δ[O(T n−2 x, 3)]. By applying repeatedly the last inequality, we get dj (T n x, T n+1 x) ≤ θ · δ[O(T n−1 x, 2)] ≤ · · · ≤ θn · δ[O(x, n + 1)].

(15)

At this point, by Lemma 2, we obtain δ[O(x, n + 1)] ≤ δ[O(x, ∞)] ≤

1 dj (x, Tx), 1−θ

which by (15) yields dj (T n x, T n+1 x) ≤

θn dj (x, Tx). 1−θ

(16)

The last inequality and the triangle inequality can be merged to obtain the following estimate: dj (T n x, T n+p x) ≤

θn 1 − θp · dj (x, Tx). 1−θ 1−θ

(17)

Let us remind the fact that 0 ≤ θ ≤ 1, then, by using (17), we can conclude that {xn }∞ n=0 is a Cauchy sequence. The subset A is assumed to be sequentially τ-complete, there exists x∗ in A such that {xn } is τ- convergent to x∗ . After simple computations involving the triangular inequality and the Definition (13), we get dj (x∗ , Tx∗ ) ≤ dj (x∗ , xn+1 ) + dj (xn+1 , Tx∗ ) = dj (T n+1 x, x∗ ) + dj (T n x, Tx∗ ) ≤ dj (T n+1 x, x∗ ) + θ max{dj (T n x, u), dj (T n x, T n+1 x), dj (x∗ , Tx∗ ), dj (T n x, Tx∗ ), dj (T n+1 x, x∗ )}+ + L · dj (x∗ , Txn )


New classes of local almost contractions

387

Continuing in this manner, we obtain dj (x∗ , Tx∗ ) ≤ dj (T n+1 x, x∗ ) + θ · [dj (T n x, u) + dj (T n x, T n+1 x) + dj (x∗ , Tx∗ ) + dj (T n+1 x, x∗ )] + L · dj (x∗ , Txn ). These relations leads us to the following inequalities: dj (x∗ , Tx∗ ) ≤

1 [(1 + θ)dj (T n+1 x, x∗ ) 1−θ + (θ + L)dj (x∗ , Txn ) + θdj (T n x, T n+1 x)].

(18)

Letting n → ∞ in (18) we obtain dj (x∗ , Tx∗ ) = 0, which means that x∗ is a fixed point of T . The estimate (14) can be obtained from (16) by letting p → ∞. This completes the proof. Remark 6 1) Theorem 4 represent a very important extension of Banach’s fixed point theorem, Kannan’s fixed point theorem, Chatterjea’s fixed point theorem, Zamfirescu’s fixed point theorem, as well as of many other related results obtained on the base of similar contractive conditions. These fixed point theorems mentioned before ensures the uniqueness of the fixed point, but the Ćirić type almost local contraction need not have a unique fixed point. 2) Let us remind (see Rus [9], [10]) that an operator T : X → X is said to be a weakly Picard operator (WPO) if the sequence {T n x0 }∞ n=0 converges for all x0 ∈ X and the limits are fixed point of T . The main merit of Theorem 4 is the very large class of Weakly Picard operators assured by using it. The uniqueness of the fixed point of a Ćirić type almost local contraction can be assured by imposing an additional contractive condition, quite similar to (13), according to the next theorem. Theorem 5 With the assumptions of Theorem 4, let T : A → A be a Ćirić type almost local contraction with the additional inequality, which actually means the monotonicity of the pseudometric: dr(j) (f, g) ≤ dj (f, g), ∀ f, g ∈ A, ∀ j ∈ J.

(19)

If the mapping T satisfies the supplementary condition: there exist the constants θ ∈ [0, 1) and some L1 ≥ 0 such that dj (Tf, Tg) ≤ θ · dr(j) (f, g) + L1 · dr(j) (f, Tf), for all f,g ∈ A, ∀ j ∈ J, then

(20)


388

M. Zákány

1) T has a unique fixed point, i.e., Fix(T ) = {f∗ }; 2) The Picard iteration {xn }∞ n=0 given by xn+1 = Txn , for any x0 ∈ A;

n ∈ N converges to f∗ ,

3) The a priori error estimate (14) holds; 4) The rate of the convergence of the Picard iteration is given by dj (xn , f∗ ) ≤ θ · dr(j) (xn−1 , f∗ ),

n = 1, 2, . . . , ∀ j ∈ J

(21)

Proof. 1) Suppose, by contradiction, there are two distinct fixed points f∗ and g∗ of T . Then, by using (20), and condition (19) for every fixed j ∈ J with f := f∗ , g := g∗ we get: dj (f∗ , g∗ ) ≤ θ · dr(j) (f∗ , g∗ ) ≤ θ · dj (f∗ , g∗ ) ⇔ (1 − θ) · dj (f∗ , g∗ ) ≤ 0, which is obviously a contradiction with dj (f∗ , g∗ ) > 0. So, we prove the uniqueness of the fixed point. The proof for 2) and 3) is quite similar to the proof from the Theorem 4. 4) At this point, letting g := xn , f := f∗ in (20), it results the rate of convergence given by (21). The proof is complete. The contractive conditions (13) and (20) can be merged to maintain the unicity of the fixed point, stated by the next theorem. Theorem 6 Under the assumptions of definition 8, let T : A → A be a mapping for which there exist the constants θ ∈ [0, 1) and some L ≥ 0 such that for all f, g ∈ A and ∀ j ∈ J dj (Tf, Tg) ≤ θ · Mr(j) (f, g) + L · min{dr(j) (f, Tf), dr(j) (g, Tg), dr(j) (f, Tg), dr(j) (g, Tf)},

(22)

where Mr(j) (f, g) = max{dr(j) (f, g), dr(j) (f, Tf), dr(j) (g, Tg), dr(j) (f, Tg), dr(j) (g, Tf)}. Then 1. T has a unique fixed point,i.e., Fix(T ) = {f∗ }; 2. The Picard iteration {xn }∞ n=0 given by xn+1 = Txn , for any x0 ∈ A;

n ∈ N converges to f∗ ,


New classes of local almost contractions

389

3. The a priori error estimate (14) holds. Particular case 1. The famous Ćirić’ s fixed point theorem for single valued mappings given in [6] can be obtain from Theorems 4, 6, 5 by taking L = L1 = 0 and considering r the identity mapping: r(j) = j. The Ćirić’ s contractive condition represent one of the most general metrical condition that provide a unique fixed point by means of Picard iteration. Despite this observation, the contractive condition given for Ćirić-type almost local contraction (in (13)) possess a very high level of generalisation. Note that the fixed point could be approximated by means of Picard iteration, just like in the case of Ćirić’ s fixed point theorem, although the uniqueness of the fixed point is not ensured by using (13). 2. If the maximum from Theorem 6 becomes: max dr(j) (f, g), dr(j) (f, Tf), dr(j) (g, Tg), dr(j) (f, Tg), dr(j) (g, Tf) = dj (f, g), for all f, g ∈ A, then we can easily obtain Theorem 2 (E) from Theorem 4. Also, by Theorem 5 we obtain Theorem 2 (U) (see Zakany,[11]). In the light of the above informations about the Ćirić-type ALC-s, it is natural to extend it to the Ćirić-type strict almost local contractions. Definition 9 Let X be a set and let D = (dj )j∈J be a family of pseudometrics defined on X. In order to underline the local character of these type of contractions, we let A ⊂ X a subset of X. We let τ be the weak topology on X defined by the family D. Let r be a function from J to J. The operator T : A → A is called Ćirić-type strict almost local contraction with respect (D, r) if it simultaneously satisfies conditions (Ci − ALC) and (ALC − U), with some real constants θC ∈ [0, 1), LC ≥ 0 and θu ∈ [0, 1), Lu ≥ 0, respectively. (Ci − ALC)

dj (Tf, Tg) ≤ θC · Mr(j) (f, g) + LC · dr(j) (g, Tf), for all f,g ∈ A,

for every j ∈ J, where Mr(j) (f, g) = max dr(j) (f, g), dr(j) (f, Tf), dr(j) (g, Tg), dr(j) (f, Tg), dr(j) (g, Tf) . (ALC − U) dj (Tf, Tg) ≤ θu ·dr(j) (f, g)+Lu ·dr(j) (f, Tf), for all f,g ∈ A, ∀ j ∈ J, We end with a few examples that have an illustrative role. They presents Ćirić’ type almost local contractions, without having unique fixed point.


390

M. Zákány

Example 5 Let A be the set of positive functions A = {f|f : [0, ∞) → [0, ∞)}, which is the subset of all real functions X = {f : R → R}, A ⊂ X. We shall use the pseudometric: dj (f, g) = |f(0) − g(0)| · j, ∀ j ∈ J; J ⊂ N, ∀ f, g ∈ A. Indeed, dj is a pseudometric, but not a metric, take for example dj (x3 , x2 ) = 0, but x3 6= x2 . Considering the mapping Tf = |f|, ∀ f ∈ A, r(j) = j + 1. Note that the restrictive condition (19) is also verified. By using condition (5) for almost local contractions: |f(0) − g(0)| · j ≤ θ · |f(0) − g(0)| · (j + 1) + L · |g(0) − f(0)| · (j + 1) which is equivalent to: j ≤ (θ + L)(j + 1). This inequality becames true if j j > 1, θ = 15 ∈ (0, 1), L = 3 > 0, and j−1 ∈ (1, 2). Hence, T is an almost local contraction. However, T is not a contraction, because the contractive condition d(Tx, Ty) ≤ θ · d(x, y) leads us to the false assumption: 1 ≤ θ. The map T is Ćirić-type almost local contraction, because Mr(j) (f, g) = |f(0) − g(0)| · (j − 1), and from (13) we have the equivalent form |f(0) − g(0)| · j ≤ θ · |f(0) − g(0)| · (j − 1) + L · |f(0) − f(0)| · (j − 1). Again, we get the inequality j ≤ (θ + L)(j − 1). The mapping T has infinite number of fixed points: FixT = {f ∈ A} = A, by taking: |f(x)| = f(x), ∀ f ∈ A,

x ∈ [0, ∞).

In fact, the uniqueness condition (20) is not valid, having in view the equivalent form: |f(0) − g(0)| · j ≤ θ · |f(0) − g(0)| · (j − 1) + L1 · |f(0) − f(0)| · (j − 1), which leads us to the contradiction j ≤ θ(j − 1), i.e. the mapping T not satisfy the uniqueness condition (20). In fact, not even (22) is satisfied, by computing Mr(j) (f, g) = |f(0)−g(0)|·(j−1) and min{dr(j) (f, Tf), dr(j) (g, Tg), dr(j) (f, Tg), dr(j) (g, Tf)} = |f(0) − g(0)| · (j − 1) (since j > 1). By replacing these values in (22), we get |f(0) − g(0)| · j ≤ θ · |f(0) − g(0)| · (j − 1) + L · |f(0) − f(0)| · (j − 1), which also lead to the previous contradiction.


New classes of local almost contractions

391

Example 6 By taking the mapping from Example 4, with a small modification, which is: let X be the set of positive functions X = {f|

f : [0, ∞) → [0, ∞)},

which is a subset of the real functions F = {f : R → R}. Let dj (f, g) = |f(x0 ) − g(x0 )| · ej , ∀ f, g ∈ X, r(j) = 2j , ∀ j ∈ Z. We can conclude in the same manner that T is also a Ćirić type almost local contraction, i.e., it satisfy the contractive condition (13). j Indeed, we have Mr(j) (f, g) = |f(x0 ) − g(x0 )| · e 2 . This way, the condition (13) became the contractive condition for almost local contractions (5). By considering L = 0 in the definition 8 of Ćirić-type almost local contraction, we get a new type of ALC, that is the quasi-almost local contraction. c) Quasi-almost local contractions Definition 10 Under the assumptions of definition 5, the operator T : A → A is called quasi-almost local contraction with respect (D, r) if, for every j ∈ J, there exist the constant θ ∈ [0, 1) such that dj (Tf, Tg) ≤ θ · Mr(j) (f, g), for all f,g ∈ A,

(23)

where Mr(j) (f, g) = max{dr(j) (f, g), dr(j) (f, Tf), dr(j) (g, Tg), dr(j) (f, Tg), dr(j) (g, Tf)}. Theorem 7 Consider a function r : J → J, let a nonempty, τ- bounded, sequentially τ- complete, and T - invariant subset A ⊂ X and let T : A → A be quasi-almost local contraction with respect to (D, r). Then 1. T has a fixed point,i.e., Fix(T ) = {x ∈ X : Tx = x} 6= φ; ∗ 2. For any x0 = x ∈ A, the Picard iteration {xn }∞ n=0 converges to x ∈ Fix(T );

3. The following a priori estimate is available: dj (xn , x∗ ) ≤

θn dj (x, Tx), (1 − θ)2

n = 1, 2, . . .

(24)


392

M. Zákány

Proof. Obviously, we have to follow the steps from the proof of Theorem 4, with the only difference that the constant L = 0, as in the case of quasi ALC-s. The uniqueness of the fixed point is also assured by imposing an additional condition, just like in the class of Ćirić-type almost local contraction, as it follows. Theorem 8 With the assumptions of Theorem 4, let T : A → A be a quasialmost local contraction with the additional inequality: dr(j) (f, g) ≤ dj (f, g), ∀ f, g ∈ A, ∀ j ∈ J.

(25)

If the mapping T satisfies the supplementary condition: there exist the constants θ ∈ [0, 1) such that dj (Tf, Tg) ≤ θ · dr(j) (f, g) + L1 · dr(j) (f, Tf), for all f,g ∈ A, ∀ j ∈ J,

(26)

then 1. T has a unique fixed point,i.e., Fix(T ) = {f∗ }; 2. The Picard iteration {xn }∞ n=0 given by xn+1 = Txn , for any x0 ∈ A;

n ∈ N converges to f∗ ,

3. The a priori error estimate (14) holds; 4. The rate of the convergence of the Picard iteration is given by dj (xn , f∗ ) ≤ θ · dr(j) (xn−1 , f∗ ),

n = 1, 2, ..., ∀ j ∈ J

(27)

d) Ćirić-Reich-Rus type almost local contraction Definition 11 Under the assumptions of definition 5, the operator T : A → A is called Ćirić-Reich-Rus type almost local contraction with respect (D, r) if the mapping T : A → A satisfying the condition dj (Tf, Tg) ≤ δ · dr(j) (f, g) + L · [dr(j) (f, Tf) + dr(j) (g, Tg)],

(28)

for all f, g in A, where δ, L ∈ R+ and δ + 2L < 1 Theorem 9 If the pseudometric d satisfy the condition: dr(j) (f, g) < dj (f, g), ∀ j ∈ J, ∀ f, g ∈ A, then any Ćirić- Reich- Rus type almost local contraction, i.e. any mapping T : A → A satisfying the condition (28) with L 6= 1 is an almost local contraction.


New classes of local almost contractions

393

Proof. Using condition (28) and the triangle rule, we get dj (Tf, Tg) ≤ δ · dr(j) (f, g) + L · [dr(j) (f, Tf) + dr(j) (g, Tg)] ≤ δ · dr(j) (f, g) + L · [dr(j) (g, Tf) + dr(j) (Tf, Tg) + dr(j) (f, g) + dr(j) (g, Tf)] The condition for the pseudometric leads us to: dj (f, g) > dr( j) (f, g), dj (Tf, Tg) > dr( j) (Tf, Tg), dj (g, Tf) > dr( j) (g, Tf) From this point, we get after simple computations: (1 − L) · dj (Tf, Tg) ≤ (δ + L) · dj (f, g) + 2L · dr( j) (g, Tf)

(29)

and which implies dj (Tf, Tg) ≤

2L δ+L · dj (f, g) + · d (g, Tf), ∀ f, g ∈ A 1−L 1 − L r(j)

(30)

Considering δ, L ∈ R+ and δ + 2L < 1 , the inequality (28) holds, with 2L δ+L 1−L ∈ (0, 1) and 1−L ≥ 0. Therefore, any Ćirić-Reich-Rus type almost local contraction with the condition for the pseudometric, is an almost local contraction.

References [1] V. Berinde, On the approximation of fixed points of weak contractive mappings, Carpathian J. Math., 19 (1) (2003), 7–22. [2] V. Berinde, Approximating fixed points of weak contractions using the Picard iteration, Nonlinear Analysis Forum, 9 (1) (2004), 43–53. [3] V. Berinde, A convergence theorem for some mean value fixed point iterations in the class of quasi contractive operators, Demonstr. Math., 38 (2005), 177–184. [4] V. Berinde, General constructive fixed point theorems for Ćirić- type almost contractions in metric spaces, Carpathian J. Math., 24 (2) (2008), 10–19.


394

M. Zákány

[5] Lj. B. Ciric, A generalization of Banah‘s contraction principle, Proc. Am. Math. Soc., 45 (1974), 267–273. [6] Lj. B. Ciric, On contraction type mappings, Math. Balkanica, 1 (1971), 52–57. [7] Martins-da-Rocha, Filipe, Vailakis, Yiannis, Existence and uniqueness of a fixed point for local contractions, Econometrica, 78 (3) (May, 2010), 1127–1141. [8] M. Păcurar, Iterative Methods for Fixed Point Approximation, ClujNapoca, Editura Risoprint, 2009. [9] I. A. Rus, Generalized Contractions and Applications, Cluj University Press, Cluj-Napoca, 2001. [10] I. A. Rus, Picard operator and applications, Babeş-Bolyai Univ., 1996. [11] Z. Mónika, Fixed Point Theorems For Local Almost Contractions, Miskolc Mathematical Notes, 18 (1) (2017), 499–506.

Received: June 13, 2017


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 395–401 DOI: 10.2478/ausm-2018-31

Slant helices of (k, m)-type in E4 Münevver Yildirim Yilmaz

Mehmet Bektaş

Department of Mathematics, Firat University, Türkiye email: myildirim@firat.edu.tr

Department of Mathematics, Firat University, Türkiye email: mbektas@firat.edu.tr

Abstract. In the present work, we define new type slant helices called (k,m)-type and we conclude that there are no (1,k) type (1 ≤ k ≤ 4) slant helices. Also we obtain conditions for different type slant helices.

1

Introduction

The curve theory has been one of the most studied subject because of having many application area from geometry to the various branch of science. Especially the characterizations on the curvature and torsion play important role to define special curve types such as so-called helices. The curves of this type have drawn great attention from mathematics to natural sciences and engineering. Helices appear naturally in structures of DNA, nanosprings. They are also widely used in engineering and architecture.The concept of slant helix defined by Izumiya and Takeuchi [6] based on the property that the principal normal lines of an α curve(with non-vanishing curvature) make constant angle with a fixed direction of the ambient space. After this lightening work many researcher have characterized this type of curves in various spaces. For instance in [1] authors extended slant helix concept to En and conclude that there are no slant helices with non-zero constant curvatures in the space E4 . The slant helix subject are also considered in 3-, 4-, and n- dimensional Eucliedan spaces, respectively in [7, 10, 12] different dimensions. Moreover different properties 2010 Mathematics Subject Classification: 53C040, 53A05 Key words and phrases: slant helices, Serret-Frenet frame

395


396

M. Y. Yilmaz, M. Bektaş

of helices are also discussed in [4, 8, 11, 13] . On the other hand in A.T Ali, R. Lopez and M. Turgut extended this study to the k-type slant helix in E41 . In this study they called α curve as k-type slant helix if there exists on (non-zero) constant vector field U ∈ E41 such that hVk+1 , Ui = const, for 0 ≤ k ≤ 3. Here Vk+1 shows the Frenet vectors of this curve [2]. One may easily conclude that 0-type slant helices are general helices and 1-type slant helices correspond just slant helices. They consider k-type slant helices for partially null and pseudo null curves, and in hyperbolic space. In accordance with above studies, in this work we define (k, m)-type slant helices in E4 and we show that there are no (1, m) type slant helices in E4 .

2

Preliminaries

In this section we will present on brief basic tools for the space curves in E4 . A detailed information can be found in [5]. Let α : I ⊆ R → E4 be an arbitrary curve in Euclidean space E4 . The standard scalar product in E4 given by hx, yi =

4 X x i yi i=1

where x, y ∈ E4 (1 ≤ i ≤ 4) , Then the curveD α is said toE be of unit speed 0 0 (or parametrized by arclength) if it satisfies α (s) , α (s) = 1. In addition the norm of an arbitrary vector x in E4 is given by p kxk = hx, xi Let {T, N, B1 , B2 } be the moving frame along the unit speed curve α, where T, N, B1 and B2 denote, the tangent, the principal normal, binormal and trinormal vector fields, respectively. Then the Frenet formulas are given by [3]    0 T 0 k1 0 0 T N0  −k1  N 0 k 0 2  0=  . B   0 −k2 0 k3  B1  1 0 B2 0 0 −k3 0 B2 

(1)

Hence k1 , k2 and k3 are called, the first, the second and the third curvature of α. If k3 6= 0 for each s ∈ I ⊆ R, the curve lies fully in E4 .


Slant helices of (k, m)-type in E4

3

397

(k, m)-type slant helices in E4

In this section, we will define (k, m) type slant helices in E4 . Definition 1 Let α be a regular unit speed curve in E4 with Frenet frame {V1 , V2 , V3 , V4 } . We call α is a (k, m) type slant helix if there exists a nonzero constant vector field U ∈ E4 satisfies hVk , Ui = m (m constant) for 1 ≤ k ≤ 4 k 6= m. The constant vector U is on axis of (k, m)-type slant helix. We decompose U with respect to Frenet frame {T, N, B1 , B2 } as U = u1 T + u2 N + u3 B1 + u4 B2 , where ui = ui (s) are differentiable functions of s. Here we denote V1 = T, V2 = N, V3 = B1 , V4 = B2 . From now on, in sake of easinesss, we will use these notations and assume that ki 6= 0, (1 ≤ i ≤ 3) . Theorem 1 There are no (1, 2) type slant helices in E4 . Proof. Assume that α is a (1, 2) type slant helix. Then for a constant vector field U. hT, Ui = a is const and hN, Ui = b is constant. Differentiating this equation and using Frenet equations, we obtain k1 hN, Ui = 0 means that U is orthogonal to N. Theorem 2 There are no (1, 3) type slant helices in E4 . Proof. Assume that α is a (1, 3) type slant helix. Then we may write hT, Ui = const = a hB, Ui = const = b. Also taking account Theorem 1 we decompose U as follows U = aT + bB1 + u2 B2 (2) Differentiating constant vector U, we get 0

a (k1 N) + b (−k2 N + k3 B2 ) + u2 B2 + u2 (−k3 B1 ) = 0 ak1 N − bk2 = 0

(3)

−u2 k3 = 0

(4)

u2 + bk3 = 0.

(5)

0

From (4) we get u2 = 0 and hence b = 0, which means that there are no (1, 3) type slant helices in E4 . Theorem 3 There are no (1, 4) type slant helix in E4 .


398

M. Y. Yilmaz, M. Bektaş

Proof. Assume that α is a (1, 4) type slant helix. Then we may write U = aT + u1 B1 + bB2 We know that U is constant then we get 0

a (k1 N) + u1 B1 + u1 (−k2 N + k3 B2 ) + b (−k3 B1 ) = 0 ak1 − u1 k2 = 0 0

(6)

u1 − bk3 = 0

(7)

u1 k3 = 0

(8)

means that u1 = 0 and from (7) we get b = 0, hence there are no (1, 4) type slant helix in E4 . Corollary 1 There are no (1, k) type slant helix in E4 . Theorem 4 If α is a (2, 3) type slant helix in E4 ⇐⇒ there exist a constant such that Zs k2 (t) − k1 (t) dt = 0. k1 (t) 0

Proof. Assume that α is a (2, 3) type slant helix in E4 . Then we may write U = u1 T + aN + bB1 + u2 B2 . Differentiating constant vector U, one may get 0

u1 (k1 N) + u1 T + a (−k1 T + k2 B1 ) 0

+b (−k2 N + k3 B2 ) + u2 B2 + u2 (−k3 B1 ) = 0 0

u1 − ak1 = 0

(9)

u1 k1 − bk2 = 0

(10)

ak2 − k3 = 0

(11)

0

bk3 + u2 = 0 Using (9) and (12)

(12)

Zs u1 = a k1 (t) dt 0

(13)


Slant helices of (k, m)-type in E4 and

399

Zs u2 = −b k3 (t) dt

(14)

0

From (10) we get u1 =

b kk21 .

Taking into account this result in (13) we get

Zs a k2 c k1 (t) dt = b = k1 (t) dt. k1 b Zs

(15)

0

0

We assume a and b are constants so we can denote Using this fact in (15) we conclude that

a b

= const.

Zs k2 (t) − k1 (t) dt = 0. k1 (t) 0

This completes the proof.

Theorem 5 There are no (2, 4) type slant helix in E4 . Proof. Assume that α is a (2, 4) type slant helix in E4 . Then we may write U = u1 T + aN + u2 B1 + bB2 . Recall that U is a constant vector, we obtain 0

u1 T + u1 (k1 N) + a (−k1 T + k2 B1 ) 0

+u2 B1 + u2 (−k2 N + k3 B2 ) + b (−k3 B1 ) = 0 0

u1 − ak1 = 0

(16)

u1 k1 − u2 k2 = 0

(17)

0

ak2 + u2 − bk3 = 0

(18)

u2 k3 = 0

(19)

From (19) we get u2 = 0. Using this in (17) we get u1 = 0 and finally we get a = b = 0 which means that there are no (2, 4) type slant helix in E4 . Theorem 6 There are no (3, 4) type slant helix in E4 .


400

M. Y. Yilmaz, M. Bektaş

Proof. Assume that α is a (3, 4) type slant helix in E4 . Then we may write U = u1 T + u2 N + aB1 + bB2 Taking into account of the constant vector U we get 0 0 u1 − u2 k1 T + u1 k1 + u2 − ak2 N + (u2 k2 − bk3 ) B1 + (ak3 ) B2 = 0 0

u1 − u2 k1 = 0 0

u1 k1 + u2 − ak2 = 0 u2 k2 − bk3 = 0 ak3 = 0 ⇒ a = 0 so there is no (3, 4) type slant helix in E4 .

References [1] A. T. Ali, M. Turgut, Some characterizations of slant helices in the Euclidean space En, Hacet. J. Math. Stat. , 39 (3), (2010), 327–336. [2] A. T. Ali, R. Lopez, M. Turgut, k–type partially null and pseudo null slant helices in Minkowski 4-space, Math. Commun., 17 (1) (2012), 93– 103. [3] H. Gluck, Higher curvature of curves in Euclidean space, Amer. Math. Monthly, 73 (1996), 699–704. [4] I. Gök, O. Z. Okuyucu, T. Kahraman, H. H. & Hacisalihoğlu, On the quaternionic B 2-slant helices in the Euclidean space E 4, Adv. in App. Cliff. Alg, 21 (4) (2011), 707–719. [5] H. H. Hacisalihoğlu, Diferensiyel Geometri, İnönü Üniversitesi Fenedebiyat Fakültesı Yayınları (In Turkish), 1983. [6] S. Izumiya, N. Takeuchi, New Special Curves and Developable Surfaces, Turk J Math, 28 (2004), 153–163. [7] L. Kula, N. Ekmekci, Y. Yayli, K. Ilarslan, Characterizations of slant helices in Euclidean 3-space, Turkish J. Math. 33 (2009), 1–13.


Slant helices of (k, m)-type in E4

401

[8] L. Kula, Y. Yayli, On Slant helix and its spherical indicatrix, App. Math. and Comput., 169 (2005), 600–607. [9] R. S. Millman, G. D. Parker, Elements of differential geometry, PrenticeHall Inc., Englewood Cliffs, N. J. USA, 1977. [10] M. Onder, M. Kazaz, H. Kocayigit, O. Hand Kılıc, B2-slant helix in Euclidean 4-space E4, Int. J. Cont. Math. Sci., 3 (29) (2008), 1433–1444. [11] Z. Özdemir, I. Gök, F. N. Ekmekci, & Y. Yaylı, A New Approach on Type-3 Slant Helix in E 4, Gen. Math. Notes, 28 (1) (2015), 40–49. [12] A. Şenol, E. Ziplar, Y. Yayli, İ Gök, A new approach on helices in Euclidean n-space, Math. Commun., 18 (1), (2013), 241–256. [13] M. Turgut, S. Yilmaz, Characterizations of some special helices in E4 , Sci. Magna, 4 (1) (2008), 51–55. [14] M. Turgut, & A. T. Ali, Some characterizations of special curves in the Euclidean space E4, Acta Univ. Sapientiae Math., 2 (1) (2010), 111–122.

Received: June 22, 2017


Acta Univ. Sapientiae, Mathematica, 10, 2 (2018) 402–417 DOI: 10.2478/ausm-2018-32

On extensions of Baer and quasi-Baer modules

Ebrahim Hashemi

Marzieh Yazdanfar∗

Factually of Mathematics, Shahrood University of Technology, Shahrood, Iran email: eb− hashemi@shahroodut.ac.ir

Factually of Mathematics, Shahrood University of Technology, Shahrood, Iran email: m.yazdanfar93@gmail.com

Abdollah Alhevaz Factually of Mathematics, Shahrood University of Technology, Shahrood, Iran email: a.alhevaz@shahroodut.ac.ir

Abstract. Let R be a ring, MR a module, S a monoid, ω : S −→ End(R) a monoid homomorphism and R ∗ S a skew monoid ring. Then M[S] = {m1 g1 + · · · + mn gn | n ≥ 1, mi ∈ M and gi ∈ S for each 1 ≤ i ≤ n} is a module over R ∗ S. A module MR is Baer (resp. quasi-Baer ) if the annihilator of every subset (resp. submodule) of M is generated by an idempotent of R. In this paper we impose S-compatibility assumption on the module MR and prove: (1) MR is quasi-Baer if and only if M[s]R∗S is quasi-Baer, (2) MR is Baer (resp. p.p) if and only if M[S]R∗S is Baer (resp. p.p), where MR is S-skew Armendariz, (3) MR satisfies the ascending chain condition on annihilator of submodules if and only if so does M[S]R∗S , where MR is S-skew quasi-Armendariz. 2010 Mathematics Subject Classification: 16D80; 16S34; 16S36 Key words and phrases: S-compatible module, reduced module, Baer module, Quasi-Baer module, skew monoid ring *Corresponding author

402


On extensions of Baer and quasi-Baer modules

1

403

Introduction and preliminaries

Throughout this paper R denotes an associative ring with identity and MR is a right R-module. According to [16] a ring R is Baer if the right annihilator of every nonempty subset of R is generated by an idempotent. Quasi-Baer rings were initially introduced by Clark [10]. A ring R is quasi-Baer if the right annihilator of every right ideal of R generated by an idempotent. Another generalization of Baer rings is p.p.-rings. Recall that a ring R is called right (resp. left) p.p if right (left) annihilator of every element of R is generated by an idempotent. Birkenmeier et al. in [7] introduced principally quasi-Baer rings. A ring R is called right principally quasi-Baer (or p.q.-Baer for short) if the right annihilator of a principal right ideal of R is generated by an idempotent. In [1] Armendariz studied the behaver of a polynomial ring over Baer ring. He proved for a reduced ring R, R[x] is Baer if and only if R is Baer [1, Theorem B]. Also, he provid an example to show that the “Armendariz” condition is not superfluous. Birkenmeier and park [9] extended this result to monoid ring. We now introduce the definitions and notions used in this paper. If A and B are non-empty subsets of a monoid S, then an element s0 ∈ AB = {ab : a ∈ A, b ∈ B} is said to be a unique product element (u.p. element for short) in the product of AB if it is uniquely presented in the form of s = ab where a ∈ A and b ∈ B. Recall that a monoid S is called unique product monoid (u.p. monoid for short) if for any two non-empty finite subsets A, B ⊆ S there exist a ∈ A and b ∈ B such that ab is u.p. element in the product of AB. The class of u.p. monoids are quite large. For example this class includes the right or left ordered monoid and torsion free nilpotent groups. Every u.p. monoid S is cancellative [9, Lemma 1.1] and has no non-unit element of finite order. Assume that R is a ring, S a monoid and ω : S −→ End(R) a monoid homomorphism. For each g ∈ S we denote P the image of g by ωg (i.e., ω(g) = ωg ). Then all finite formal combinations ni=1 ai gi , with point-wise addition and multiplication induced by (ag)(bh) = (aωg (b))gh form a ring that is called skew monoid ring and it is denoted by R ∗ S. The construction of skew monoid ring generalizes some classical ring construction such as skew polynomial rings, skew Laurent polynomial rings and monoid rings. Hence any result on skew monoid ring has its counterpart in each of the subclasses. As a generalization of monoid rings, we introduce the notion of modules over skew monoid rings. For a module MR , let M[S] = {m1 g1 + · · · + mn gn | n ≥ 1, mi ∈ M and gi ∈ S for each 1 ≤ i ≤ n}. Then M[S] is a right module over R∗ S under the following scaler product operation: for m(s) = m1 g1 +· · ·+mn gn ∈


404

E. Hashemi, M. Yazdanfar, A. Alhevaz

P M[S] and f(s) = a1 h1 + · · · + am hm ∈ R ∗ S, m(s)f(s) := i,j mi ωgi (aj )gi hj . For a nonempty subset X of MR , let annR (X) = {r ∈ R | Xr = 0}. The notion of reduced, Armendariz, Baer, p.p and quasi-Baer module introduced in [18] by Lee and Zhou. A module MR is called reduced if for any m ∈ M and a ∈ R, ma = 0 implies mR ∩ Ma = 0. A module MR is called Baer if, for any nonempty subset X of M, annR (X) = eR where e2 = e ∈ R. A module MR is called p.p if for any element m ∈ M, annR (m) = eR where e2 = e ∈ R. A module MR is called quasi-Baer if, for any right R-submodule X of M, annR (X) = eR where e2 = e ∈ R. Clearly, R is reduced (resp. Baer, right p.p, quasi-Baer) if and only if RR is reduced (resp. Baer, right p.p, quasi-Baer). Lee and Zhou [18] proved that MR is reduced if and only if M[x]R[x] is reduced. Various results of reduced rings were extended to modules in [18, 2]. Recall that from [6] an idempotent e ∈ R is left (resp. right) semicentral in R if exe = xe (resp. exe = ex) for all x ∈ R. Equivalently, e = e2 ∈ R is left (resp. right) semicentral if eR (resp. Re) is an ideal of R. Since the right annihilator of a right R-module is an ideal, then the right annihilator of a right R-module is generated by a left semicemtral idempotent in a quasi-Baer module. We denote the set of all left (resp. right) semiccentral idempotents of R with S` (R) (resp. Sr (R)). A module MR is called principally quasi-Baer (or p.q.-Baer for short) if, for any m ∈ M, annR (mR) = eR where e2 = e ∈ R. Clearly R is a right p.q.-Baer if and only if RR is p.q.-Baer module. In this paper we introduce and study the concept of S-skew Armendariz modules as a generalization of S-Armendariz rings [19]. For a u.p. monoid S and monoid homomorphism ω : S −→ End(R) we show that reduced module MR is S-skew Armendariz. We investigate the quasi-Baer and related conditions on right R ∗ S-module M[S] for a u.p. monoid S and monoid homomorphism ω : S −→ Aut(R). We impose S-compatibility assumption on the module MR and prove: (1) MR is quasi-Baer if and only if M[s]R∗S is quasiBaer, (2) MR is Baer (resp. p.p) if and only if M[S]R∗S is Baer (resp. p.p), when MR is S-skew Armendariz, (3) MR satisfies the ascending chain condition on annihilator of submodules if and only if so does M[S]R∗S , when MR is S-skew quasi-Armendariz. Our results extend Armendariz [1, Theorem B], Groenewald [11, Theorem 2], Birkenmeier, Kim and Park [8, Theorem 1.2], Birkenmeier and Park [9, Theorem 1.2, Corollary 1.3].


On extensions of Baer and quasi-Baer modules

2

405

S-skew Armendariz modules

Let R be a ring with an endomorphism σ. According to [4] for a module MR and an endomorphism σ : R −→ R, we say that MR is σ-compatible if for each m ∈ M and r ∈ R, we have mr = 0 if and only if mσ(r) = 0. For more details on σ-compatible rings refer to [13, 14]. Definition 1 Let R be a ring, S a monoid and ω : S −→ End(R) a monoid homomorphism. We say that a module MR is S-compatible if MR is ωg compatible for each g ∈ S. Notic that R is S-compatible if and only if RR is S-compatible. Now we give some examples of S-compatible modules. Example 1 [4, Example 4.4] Let R0 be a domain of characteristic zero, and R := R0 [t]. Define σ|R0 = idR0 and σ(t) = −t. Let MR := R0 ⊕ R0 ⊕ R0 ⊕ · · · , where t ∈ R acts on MR as follows: for (m0 , m1 , m2 , . . .) ∈ M, we set (m0 , m1 , m2 , . . .).t := (0, m0 k0 , m1 k1 , m2 k2 , . . .) where the ki (i ∈ N) are fixed nonzero integers. We show that M is σ-compatible. For this, it suffices to show that ann(m) = 0 whenever 0 6= m ∈ M. Suppose that (a0 , a1 , a2 , · · · )(br tr + br+1 tr+1 + “higher terms”) = 0, where ai , bi ∈ R0 for every i ∈ N and br 6= 0. First applying tr to (a0 , a1 , a2 , . . .) gives (0, 0, · · · , 0, a0 k0 k1 · · · kr−1 , a1 k1 k2 · · · kr , . . .)(br +br+1 t+“ higher terms”) = 0. Upon computing this expression, we deduce that a0 k0 k1 . . . kr−1 br = 0. Since the characteristic is zero, R is a domain, and k0 k1 . . . kr−1 br 6= 0, we deduce that a0 = 0. Now, we may proceed inductively to show that all ai = 0. From this calculation, we deduce that MR is σ-compatible. Example 2 [14, Example 1.1] Let R1 be a ring, D a domain and R = Tn (R1 )⊕ D[y], where Tn (R1 ) is upper n × n triangular matrix ring over R1 . Let α : D[y] −→ D[y] be a monomorphism which is not surjective. We define an endomorphism α : R −→ R of R by α(A ⊕ f(y)) = A ⊕ α(f(y)) for each A ∈ Tn (R1 ) and f(y) ∈ D[y]. In [14, Example 1.1] it is shown that R is an α-compatible. Example 3 Let R be a ring and σi an endomorphism of R such that R be a σi compatible for each 1 ≤ i ≤ n. Let S be a monoid generated by {x1 , x2 , . . . , xn } and ω : S −→ End(R) a monoid homomorphism such that ωxj = σji . One can i ∼ R[x1 , x2 , . . . , xn ; σ1 , σ2 , . . . , σn ]. show that R is S-compatible and R ∗ S =


406

E. Hashemi, M. Yazdanfar, A. Alhevaz

According to Lee and Zhou [18] a module MR is Armendariz if, for elements m(x) = m0 +m1 x+· · ·+mn xn ∈ M[x] and f(x) = a0 +a1 x+· · ·+am xm ∈ R[x], m(x)f(x) = 0 implies mi aj = 0 for each 1 ≤ i ≤ n, 1 ≤ j ≤ m. In [21] Zhang and Chen, introduced the concept of a σ-skew Armendariz module and studied its properties. A module MR is called σ-skew Armendariz module, if, whenever m(x)f(x) = 0 where m(x) = m0 + m1 x + · · · + mn xn ∈ M[x] and f(x) = a0 +a1 x+· · ·+am xm ∈ R[x; σ], we have mi σi (bj ) = 0 for each 0 ≤ i ≤ n, 0 ≤ j ≤ m. In [19], Liu introduced the concept of a S-Armendariz ring and studied its properties. In the following we introduce the concept of S-skew Armendariz module as a generalization of S-Armendariz rings. Definition 2 Let R be a ring, S a monoid and ω : S −→ End(R) a monoid homomorphism. We say that a module MR is S-skew Armendariz module if, for elements m(s) = m1 g1 +· · ·+mn gn ∈ M[S] and f(s) = a1 h1 +· · ·+at ht ∈ R∗S, m(s)f(s) = 0 implies mi ωgi (aj ) = 0 for each 1 ≤ i ≤ n, 1 ≤ j ≤ t. In the case of ω is identity homomorphism, we say MR is S-Armendariz module. Notice that for a ring R and monid S with monoid homomorphism ω : S −→ End(R), R is S-skew Armendariz (resp. S-Armendariz) if and only if RR is S-skew Armendariz (resp. S-Armendariz). Theorem 1 Let R be a ring, S a monoid and ω : S −→ End(R) a monoid homomorphism. Then MR is S-skew Armendariz if and only if for every elements m(s) = m1 g1 + · · · + mn gn ∈ M[S] and f(s) = a1 h1 + · · · + at ht ∈ R ∗ S, m(s)f(s) = 0 implies mi1 ωgi1 (aj ) = 0 for each 1 ≤ j ≤ t and some 1 ≤ i1 ≤ t. Proof. The forward direction is clear. For the converse, suppose that m(s) = m1 g1 +· · ·+mn gn ∈ M[S] and f(s) = a1 h1 +· · ·+at ht ∈ R∗S with m(s)f(s) = 0. Then there exists 1 ≤ i1 ≤ n such that mi1 ωgi1 (aj ) = 0 for each 1 ≤ j ≤ t. Without loss of generality we can assume that i1 = 1. Thus 0 = m(s)f(s) = (m2 g2 + · · · + mn gn )f(s). Then by induction on n we can conclude that mi ωgi (aj ) = 0 for each 1 ≤ i ≤ n and 1 ≤ j ≤ t. Hence MR is S-skew Armendariz. If S is a monoid generated by {x} and ω : S −→ End(R) such that ωxi = σi for an endomorphism σ of R, then the skew monoid ring R ∗ S is isomorphic to skew polynomial ring R[x; σ] and M[S] is isomorphic to M[x]. Thus we have the following equivalent condition for a module to be σ-skew Armendariz. Corollary 1 Let MR be a module and σ an endomorphism of R. Then MR is σ-skew Armendariz if and only if for every polynomials m(x) = m0 + m1 x + · · · + mn xn ∈ M[x] and f(x) = a0 + a1 x + · · · + at xt ∈ R[x; σ], m(x)f(x) = 0 implies mi1 σi1 (aj ) = 0 for each 0 ≤ j ≤ t and some 0 ≤ i1 ≤ n.


On extensions of Baer and quasi-Baer modules

407

Corollary 2 Let R be a ring and σ an endomorphism of R. Then R is σ-skew Armendariz if and only if for every polynomials f(x) = a0 + a1 x + · · · + an xn , g(x) = b0 + b1 x + · · · + bm xm ∈ R[x; σ], f(x)g(x) = 0 implies ai0 σi0 (bj ) = 0 for each 0 ≤ j ≤ m and some 0 ≤ i0 ≤ n. Recall that a module MR is reduced if, for any m ∈ M and a ∈ R, ma = 0 implies mR ∩ Ma = 0. Lemma 1 The following are equivalent for a module MR . (i) MR is reduced and S-compatible. (ii) The following conditions hold for any m ∈ M, a ∈ R and g ∈ S, (a) ma = 0 implies mRa = 0. (b) ma = 0 if and only if mωg (a) = 0. (c) ma2 = 0 implies ma = 0. Proof. The proof is straightforward.

For an element f(s) = a1 g1 + · · · + an gn ∈ R ∗ S with ai 6= 0 for each i, we say that length (f(s)) = n and denote it by `(f(s)). Similarly, we can define `(m(s)) = t for an element m(s) = m1 h1 + · · · + mt ht ∈ M[S]. Proposition 1 Let R be a ring, S a u.p. monoid and ω : S −→ End(R) a monoid homomorphism. Then S-compatible reduced module MR is S-skew Armendariz. Proof. Assume that m(s) = m1 g1 + · · · + mn gn ∈ M[S] and f(s) = a1 h1 + · · · + at ht ∈ R ∗ S with m(s)f(s) = 0. We proceed by induction on `(m(s)) + `(f(s)) = n + t. If `(m(s)) = 1 or `(f(s)) = 1, then the result is clear Since u.p. monoids are cancellative by [6, Lemma 1.1]. From m(s)f(s) = 0 there exist 1 ≤ i ≤ n, 1 ≤ j ≤ t such that gi hj is u.p. element in the product of two subsets {g1 , . . . , gn } and {h1 , . . . , ht } of S. Without loss of generality we can assume that i = j = 1. Thus m1 ωg1 (a1 ) = 0 and so m1 a1 = 0 since MR is S-compatible. Therefore 0 = m(s)f(s)a1 = (m1 g1 + · · · + mn gn )(a1 ωh1 (a1 )h1 + · · · + at ωht (a1 )ht ). By using of Lemma 1, from m1 a1 = 0 we have m1 ωg1 (aj ωhj (a1 )) = 0 for each 1 ≤ j ≤ t since MR is reduced and S-Compatible. Thus 0 = m(s)f(s)a1 = (m2 g2 +· · ·+mn gn )f(s)a1 = m 0 (s)(f(s)a1 ). Since `(m 0 (s))+`(f(s)a1 ) < n+t satisfying m 0 (s)f(s)a1 = 0, by induction hypothesise mi ωgi (aj ωhj (a1 )) = 0 which implies that mi aj a1 = 0 for each 2 ≤ i ≤ n, 1 ≤ j ≤ t, since MR is S-compatible. Thus mi a21 = 0


408

E. Hashemi, M. Yazdanfar, A. Alhevaz

and so mi a1 = 0 for each 2 ≤ i ≤ n, by Lemma 1. Hence 0 = m(s)f(s) = m(s)(a2 h2 + · · · + at ht ). Then by induction mi ωgi (aj ) = 0 for each 1 ≤ i ≤ n and 1 ≤ j ≤ t. Therefore MR is S-skew Armendariz. If ω is identity homomorphism (i.e. ωg = idR the identity homomorphism of R for each g ∈ S) we deduce the following corollary. Corollary 3 Let MR be a reduced and S a u.p. monoid. Then MR is SArmendariz. Corollary 4 [2, Theorem 2.19] Every reduced module is Armendariz. Corollary 5 Let R be a reduced ring, S a u.p. monoid and ω : S −→ End(R) a monoid homomorphism. Then R is S-skew Armendariz. Proposition 2 Let S be a monoid and MR a S-skew Armendariz module. If m(s) = m1 g1 + · · · + mn gn ∈ M[S] and fi (s) = ai1 hi1 + · · · + aiti hiti ∈ R ∗ S for 1 ≤ i ≤ k are such that m(s)f1 (s) · · · fk (s) = 0, then mj ωgj (a1i1 )ωgj ωh1 (a2i2 ) · · · ωgj ωh1 . . . ωhk−1 (akik ) = 0 i1

i1

ik−1

for each 1 ≤ j ≤ n and 1 ≤ ir ≤ ti , 1 ≤ r ≤ k. Proof. Suppose m(s)f1 (s) · · · fk (s) = 0. Then from m(s)(f1 (s) · · · fk (s)) = 0 we have mj ωgj (a) = 0 for each 1 ≤ j ≤ n and each coefficient a of f1 (s)f2 (s) · · · fk (s), since MR is S-skew Armendariz and S-compatible. Thus (mj gj f1 (s))f2 (s) · · · fk (s) = 0 for each 1 ≤ j ≤ n. Thus mj ωgj (a1i1 )ωgj ωh1 (a 0 ) = i1

0 for each 1 ≤ j ≤ n, 1 ≤ i1 ≤ t1 and each coefficient a 0 of f3 (s) · · · fk (s). By continuing this manner, we see that mj ωgj (a1i1 )ωgj ωh1 (a2i2 ) · · · ωgj ωh1 . . . i1

ωhk−1 (akik ) = 0 for each 1 ≤ j ≤ n and 1 ≤ ir ≤ ti , 1 ≤ r ≤ k. ik−1

i1

As a consequence of Propositions 1 and 2 we have the following result. Corollary 6 Let R be a ring, S a u.p. monoid and ω : S −→ End(R) a monoid homomorphism. Let MR be a S-compatible reduced module. If m(s) = m1 g1 + · · · + mn gn ∈ M[S] and fi (s) = ai1 hi1 + · · · + aiti ∈ R ∗ S for 1 ≤ i ≤ k are such that m(s)f1 (s) · · · fk (s) = 0, then mj ωgj (a1i1 )ωgj ωh1 (a2i2 ) · · · ωgj ωh1 . . . ωhk−1 (akik ) = 0 i1

for each 1 ≤ j ≤ n and 1 ≤ ir ≤ ti , 1 ≤ r ≤ k.

i1

ik−1


On extensions of Baer and quasi-Baer modules

409

It is proved in [18, Theorem 1.6] MR is reduced if and only if M[x]R[x] is reduced. In the following we extend this result to M[S]R∗S . Proposition 3 Let R be a ring, S a u.p. monoid and ω : S −→ End(R) a monoid homomorphism. Then module MR is reduced and S-compatible if and only if M[S]R∗S is reduced. Proof. Assume that MR is reduced and m(s) = m1 g1 + · · · + mn gn ∈ M[S], f(s) = a1 h1 + · · · + at ht ∈ R ∗ S with m(s)f(s) = 0. Let g(s) = b1 k1 + · · · + bm km ∈ R ∗ S and k(s) = n1 s1 + · · · + np sp ∈ M[S] such that m(s)g(s) = k(s)f(s) ∈ m(s)(R ∗ S) ∩ M[S]f(s). From m(s)f(s) = 0 we have mi ωgi (aj ) = 0 = mi aj for each 1 ≤ i ≤ n, 1 ≤ j ≤ t, by Proposition 1 and S-compatibility assumption on MR . Then by Lemma 1 we have mi raj = 0 for each r ∈ R which implies that 0 = m(s)g(s)f(s) = k(s)f2 (s). Therefore ni aj al = 0 for each 1 ≤ i ≤ p and 1 ≤ j, ` ≤ t by Proposition 2. Thus ni a2j = 0 and so ni aj = 0 for each 1 ≤ i ≤ p and 1 ≤ j ≤ t by Lemma 1. Therefore k(s)f(s) = 0 which implies that m(s)(R ∗ S) ∩ M[S]f(s) = 0 and hence M[S]R∗S is reduced. Conversely, assume that M[S]R∗S is reduced and m ∈ M, r ∈ R with mr = 0. Also assume that n ∈ M, a ∈ R such that ma = nr ∈ Mr ∩ mR. Put m(s) = mg and k(s) = nh for some g, h ∈ S. Thus m(s)a = k(s)r ∈ M[S]r ∩ m(s)(R ∗ S). Since M[S]R∗S is reduced M[S]r ∩ m(s)(R ∗ S) = 0 which implies that ma = nr = 0. Hence MR is reduced. Now, assume that mr = 0 for some m ∈ M and r ∈ R. For each g ∈ S we have mgr = mωg (r)g ∈ M[S]r∩m(R∗S). Since M[S]R∗S is reduced, M[S]r ∩ m(R ∗ S) = 0. Thus mωg (r) = 0. Clearly, if mωg (r) = 0 for each g ∈ S we have mr = 0. Therefore MR is S-compatible. Corollary 7 Let R be a ring and σ an endomorphism of R. Then MR is reduced and σ-compatible if and only if M[x]R[x;σ] is reduced. Corollary 8 Let R be a ring and σ an endomorphism of R. Then R is reduced and σ-compatible if and only if R[x; σ] is reduced.

3

Extensions of Baer and quasi-Baer modules

In this section we study on the relationship between the Baerness and p.pproperty of a module MR and right R ∗ S-module M[S]. According to [5] a module MR is called quasi-Armendariz if whenever m(x) R[x]f(x) = 0 for m(x) = m0 + m1 x + · · · + mn xn ∈ M[x] and f(x) = a0 + a1 x + · · · + am xm ∈ R[x], then mi Raj = 0 for all 1 ≤ i ≤ n and 1 ≤ j ≤ m. Let S be


410

E. Hashemi, M. Yazdanfar, A. Alhevaz

a monoid. According to [12] a ring R is called S-quasi Armendariz if for each two elements α = a1 g1 + · · · + an gn , β = b1 h1 + · · · + bm hm ∈ R[S] satisfy αR[s]β = 0, implies that ai Rbj = 0 for each 1 ≤ i ≤ n and 1 ≤ j ≤ m. Definition 3 Let R be a ring, S a monoid and ω : S −→ End(R) a monoid homomorphism. A module MR is called S-skew quasi-Armendariz, if for any m(s) = m1 g1 + · · · + mn gn ∈ M[S] and f(s) = a1 h1 + · · · + at ht ∈ R ∗ S satisfy m(s)(R ∗ S)f(s) = 0 implies that mi gi Rgaj hj = 0 for each 1 ≤ i ≤ n, 1 ≤ j ≤ t and g ∈ S. Clearly a ring R is S-skew quasi-Armendariz if and only if RR is S-skew quasiArmendariz. Birkenmeier and Park in [9, Theorem 1.2] proved that for a u.p. monoid S the monoid ring R[S] is quasi-Baer (resp. right p.q.-Baer) if and only if R is quasi-Baer (resp. right p.q.-Baer). In the following we extend these results to M[S] as a right R ∗ S-module. Theorem 2 Let R be a ring, S a u.p. monoid, ω : S −→ Aut(R) a monoid homomorphism. If MR is S-compatible, then we have the following: (i) MR is right p.q.-Baer if and only if M[S]R∗S is right p.q.-Baer. (ii) MR is quasi-Baer if and only if M[S]R∗S is quasi-Baer. In this case, MR is S-skew quasi-Armendariz. Proof. (i) Assume that R is right p.q.-Baer. Let m(s) = m1 g1 + · · · + mn gn ∈ M[S]. There exists ei ∈ S` (R) such that T annR (mi R) = ei R for 1 ≤ i ≤ n. Then e = e1 e2 · · · en ∈ S` (R) and eR = ni=1 annR (mi R). Since every compatible automorphism is idempotent stabilizing by [3, Theorem 2.14] we have e(R∗S) ⊆ annR∗S (m(s)R∗S). Note that annR∗S (m(s)R∗S) ⊆ annR∗S (m(s)R). Now we show that annR∗S (m(s)R) ⊆ e(R ∗ S). Let g(s) = b1 h1 + · · · + bm hm ∈ annR∗S (m(s)R). Then m(s)Rg(s) = 0. We proceed by induction on n to show that g(s) ∈ e(R ∗ S). Let n = 1. Then m1 g1 R(b1 h1 + · · · + bt ht ) = 0. Thus m1 g1 Rbj hj = 0 for each 1 ≤ j ≤ t, since S is cancellative, by [9, Lemma 1.1]. Since ωg1 is automorphism m1 Rωg1 (bj ) = 0 and so ωg1 (bj ) ∈ annR (m1 R) = e1 R for each 1 ≤ j ≤ t. Thus ωg1 (bj ) = e1 ωg1 (bj ) and so bj = e1 bj for each 1 ≤ j ≤ t, since ωg1 is a compatible automorphism of R. Therefore bj ∈ e1 R = eR. Hence g(s) = eg(s) ∈ e(R ∗ S), as desired. Now assume that (∗)

(m1 g1 + · · · + mn gn )R(b1 h1 + · · · + bt ht ) = 0.


On extensions of Baer and quasi-Baer modules

411

Since S is u.p. monoid there exist 1 ≤ i ≤ n, 1 ≤ j ≤ t such that gi hj is u.p. element in the product of two subsets {g1 , . . . , gn } and {h1 , . . . , ht } of S. Without loss of generality we can assume that i = n, j = t. Thus mn gn Rbt ht = 0. That is ωgn (bt ) ∈ annR (mn R) = en R and ωgn (bt ) = en ωgn (bt ). Since ωgn is a compatible automorphism of R, bt = en bt and bt ∈ en R. Replacing R by Ren in the equation (∗) we have (m1 g1 + · · · + mn−1 gn−1 )R(en b1 h1 + · · · + en bt ht ) = 0. By induction on n we have en bj ∈ e1 R ∩ e2 R ∩ · · · ∩ en−1 R for each 1 ≤ j ≤ t. In particular, Tn bt ∈ e1 R ∩ · · · ∩ en−1 R. Therefore bt = en bt ∈ e1 R ∩ · · · ∩ en R = eR = i=1 annR (mi R). Since ωgi is a compatible automorphism of R for each 1 ≤ i ≤ n we have (∗∗)

(m1 g1 + · · · + mn gn )R(b1 h1 + · · · + bt−1 ht−1 ) = 0.

Since S is u.p. monoid there exist 1 ≤ i ≤ n, 1 ≤ j ≤ t − 1 such that gi hj is u.p. element in the product of two subsets {g1 , . . . , gn } and {h1 , . . . , ht−1 } of S. Without loss of generality we can assume that i = n, j = t − 1. Thus mn gn Rbt−1 ht−1 = 0 which implies that ωgn (bt ) ∈ ann(mn R) = en R and ωgn (bt−1 ) = en ωgn (bt−1 ). Therefore bt−1 = en bt−1 , since ωgn is an idempotent stabilizing automorphism of R. Replacing R by Ren in the equation (∗∗) we have (m1 g1 +· · ·+mn−1 gn−1 )Ren (b1 h1 +· · ·+bt−1 ht−1 ) = 0. Then by induction on n we can conclude that en bj ∈ annR (m1 R) ∩ · · · ∩ annR (mn−1 R) for each 1 ≤ j ≤ t−1 and hence bt−1 = en bt−1 ∈ ∩ni=1 annR (mi R) = eR. Therefore from the equation (∗∗) we have 0 = (m1 g1 + · · · + mn gn )R(b1 h1 + · · · + bt−2 ht−2 ). By continuing this process we can conclude that bj ∈ ∩ni=1 annR (mi R) = eR for each 1 ≤ j ≤ t which implies that g(s) = eg(s). Thus annR (m(s)R) ⊆ e(R ∗ S). So we have annR∗S (m(s)(R ∗ S)) ⊆ annR (m(s)R) ⊆ e(R ∗ S). Hence annR∗S (m(s)R ∗ S) = e(R ∗ S). Therefore M[S]R∗S is p.q.-Baer. Conversely assume that M[S]R∗S is p.q.-Baer. Take m ∈ M. Then annR∗S (m(R ∗ S)) = e(s)(R ∗ S) for some idempotent e(s) = e1 s1 + · · · + en sn in R ∗ S. Let a ∈ annR (mR). Since MR is S-compatible, annR (mR) ⊆ annR∗S (m(R ∗ S)) = e(s)(R ∗ S). Therefore a = e(s)a = (e1 g1 + · · · + en gn )a. Thus there exist 1 ≤ i0 ≤ n such that a = ei0 ωgi0 (a) and so annR (mR) ⊆ ei0 R. Since e(s) ∈ annR∗S (m(R ∗ S)) then 0 = mRe(s) = mR(e1 s1 + · · · + en gn ). Since S is cancellative mRei = 0 for each 1 ≤ i ≤ n. Thus ei0 ∈ annR (mR) and hence annR (mR) = ei0 R. Also, ei0 is idempotent, since ei0 ∈ annR (mR), a = ei0 ωgi0 (a) for each a ∈ annR (mR) and ωgi0 is idempotent stabilizing, we have ei0 = ei0 ωgi0 (ei0 ) = e2i0 . Therefore R is p.q.-Baer. (ii) Assume that MR is quasi-Baer. First we show that MR is S-skew quasiArmendariz. Suppose that m(s) = m1 g1 + · · · + mn gn ∈ M[S] and f(s) =


412

E. Hashemi, M. Yazdanfar, A. Alhevaz

a1 h1 + · · · + at ht ∈ R ∗ S such that m(s)(R ∗ S)f(s) = 0. Thus m(s)rgf(s) = 0 for each r ∈ R, g ∈ S. We proceed by induction on `(m(s)) + `(f(s)) = n + t. If `(m(s)) = 1, then m1 g1 rg(a1 h1 + · · · + at ht ) = 0. Since S is cancellative m1 g1 rgaj hj = 0, as desired. Also if `(f(s)) = 1 the result is clear. From (∗)

(m1 g1 + · · · + mn gn )rg(a1 h1 + · · · + at ht ) = 0

there exist 1 ≤ i ≤ n, 1 ≤ j ≤ t such that gi hj is u.p. element in the product of two subsets {g1 , . . . , gn } and {h1 , . . . , ht } of S. Without loss of generality we can assume that i = n, j = t. Then mn gn rgat ht = 0 and so mn ωgn (r)ωgn ωg (at ) = 0 = mn r 0 ωgn ωg (at ). Thus ωgn ωg (at ) ∈ annR (mn R) = eR such that e2 = e ∈ R and so ωgn ωg (at ) = eωgn ωg (at ). Replacing rg by reg in the equation (∗) we have (m1 g1 + · · · + mn−1 gn−1 )reg(a1 h1 + · · · + at ht ) = 0 since ωg is idempotent stabilizing by [3, Theorem 2.14]. Then by induction we can conclude that mi gi regaj hj = 0 for 1 ≤ i ≤ n − 1, 1 ≤ j ≤ t. Thus mi gi regat ht = 0 and so mi gi reωg (at )ght = 0 for each 1 ≤ i ≤ n − 1. Since ωgn ωg (at ) = eωgn ωg (at ) and ωgn is a compatible automorphism of R, ωg (at ) = eωg (at ). Thus 0 = mi gi reωg (at )ght = mi gi rωg (at )ght for each 1 ≤ i ≤ n−1. On the other hand mn gn regat ht = 0 and hence mi gi rgat ht = 0 for each 1 ≤ i ≤ n. Thus 0 = m(s)rgf(s) = (m1 g1 + · · · + mn gn )rg(a1 h1 + · · ·+at−1 ht−1 ). Then by induction hypothesis mi gi rgaj hj = 0 for each 1 ≤ i ≤ n, 1 ≤ j ≤ t−1. Therefore mi gi Rgaj hj = 0 for each 1 ≤ i ≤ n, 1 ≤ j ≤ t. Hence MR is S-skew quasi-Armendariz. Let V be a submodule of M[S]. Let U be a right R-submodule of M generated by all coefficients of elements of V. Since MR is quasi-Baer annR (U) = eR for some e2 = e ∈ R. Thus e(R ∗ S) ⊆ annR∗S (V), since ωs is compatible automorphism for each s ∈ S. Suppose that g(s) = b1 h1 + · · · + bt ht ∈ annR∗S (V). Thus for each m(s) = m1 g1 + · · · + mn gn ∈ V, m(s)(R ∗ S)g(s) = 0 and hence mi gi Rgbj hj = 0 for each 1 ≤ i ≤ n, 1 ≤ j ≤ t since MR is S-skew quasi-Armendariz. Therefore ωgi ωg (bj ) ∈ annR (U) = eR which implies that ωgi ωg (bj ) = eωgi ωg (bj ) for each 1 ≤ i ≤ n, 1 ≤ j ≤ t. Since ωs is compatible automorphism of R for each s ∈ S, bj = ebj for each 1 ≤ j ≤ t. That is g(s) ∈ e(R ∗ S) and so annR∗S (V) ⊆ e(R ∗ S). Hence M[S]R∗S is quasi-Baer. Conversely, assume that M[S]R∗S is quasi-Bear and U is a right R-submodule of MR . Then as in the proof of the sufficiently of (i), one can show that annR (U) is generated as a right R-submodule, by an idempotent of R.Therefore M is quasi-Baer. Now we obtain the following results as a corollary of Theorem 2.


On extensions of Baer and quasi-Baer modules

413

Corollary 9 Let R be a ring, S a u.p. monoid, ω : S −→ Aut(R) a monoid homomorphism and MR is a S-compatible module. Then we have the following: (i) MR is a reduced p.p.- module if and only if M[S]R∗S is a reduced p.p.module. (ii) MR is a reduced Baer module if and only if M[S]R∗S is a reduced Baer module. Proof. (i) Clearly reduced p.p.- modules are p.q.-Baer. Then the result follows from Theorem 2 and Proposition 3. (ii) The result follows from Theorem 2 and the fact that a reduced quasiBaer module is Baer. Corollary 10 Let R be a ring and S a u.p. monoid. Then we have the following: (i) [6, Theorem 1.2] R is quasi-Baer (resp. right p.q.-Baer) if and only if R[S] is quasi-Baer (resp. right p.q.-Baer). (ii) [6, Corollary 1.3] R is reduced Baer (resp. p.p.- ring) if and only if R[S] is a reduced Baer (resp. p.p.- ring). Corollary 11 Let MR be a module. Then the following are equivalent: (i) MR is quasi-Baer (resp. p.q.-Baer). (ii) M[x]R[x] is quasi-Baer (resp. p.q.-Baer). (iii) M[x, x−1 ]R[x,x−1 ] is quasi-Baer (resp. p.q.-Baer). Corollary 12 Let R be a σ-compatible ring for an automorphism σ of R. Then the following are equivalent: (i) R is quasi-Baer (resp. p.q.-Baer). (ii) R[x; σ] is quasi-Baer (resp. p.q.-Baer). (iii) R[x, x−1 ; σ] is quasi-Baer (resp. p.q.-Baer). (iv) R[x] is quasi-Baer (resp. p.q.-Baer). (v) R[x, x−1 ] is quasi-Baer (resp. p.q.-Baer).


414

E. Hashemi, M. Yazdanfar, A. Alhevaz

Birkenmeier et al. [6, Example 1.5] showed that the “u.p. monoid” condition on S in Theorem 2 is not superfluous. The next example shows that the “S-compatibility” assumption on RR in Theorem 2 is not superfluous. Example 4 [15, Example 2] Let K be a field, A = K[s, t] a commutative polynomial ring, and consider the ring R = A/(st). Then R is reduced. Let s = s + (st) and t = t + (st) in R = A/(st). Define an automorphism σ of R by σ(s) = t and σ(t) = s. Hirano in [15] showed that R[x; σ] is quasi-Baer but R is not quasi-Baer. Since σ(st) = 0 but sσ(t) = s2 6= 0 (since R is reduced), hence σ is not compatible. Therefore the “compatibility” assumption on σ is not superfluous. Theorem 3 Let R be a ring, S a u.p. monoid and ω : S −→ Aut(R) a monoid homomorphism. If MR is a S-compatible and S-skew Armendariz module, then MR is Baer if and only if M[S]R∗S is Baer. Proof. The proof is similar to that of Theorem 2.

Corollary 13 Let R be a ring, S a u.p. monoid and ω : S −→ Aut(R) a monoid homomorphism. Let MR is S-compatible reduced module. Then MR is Baer if and only if M[S]R∗S is Baer. Proof. This follows from Proposition 1 and Theorem 3.

Corollary 14 Let R be a σ-compatible ring for an automorphism σ of R. If R is σ-skew Armendariz, then the following are equivalent: (i) R is Baer. (ii) R[x; σ] is Baer . (iii) R[x, x−1 ; σ] is Baer. (iv) R[x] is Baer. (v) R[x, x−1 ] is Baer. Theorem 4 Let R be a ring, S a monoid and ω : S −→ End(R) a monoid homomorphism. If MR is S-compatible and S-skew quasi-Armendariz, then MR satisfies the ascending chain condition on annihilator of submodules if and only if so does M[S]R∗S .


On extensions of Baer and quasi-Baer modules

415

Proof. Assume that MR satisfies the ascending chain condition on annihilator of submodules. Let V1 ⊆ V2 ⊆ . . . be a chain of annihilator of submodules of M[S]R∗S . Then there exist submodules Ki of M[S]R∗S such that annR∗S (Ki ) = Vi and Ki ⊇ Ki+1 for each i ≥ 1. Let Ui be a submodule of M generated by all coefficients of elements of Ki . Clearly U1 ⊇ U2 ⊇ · · · . Then annR (U1 ) ⊆ annR (U2 ) ⊆ · · · is a chain of annihilator of submodules of MR . Since MR satisfies the ascending chain condition on annihilator of submodules there exists n ≥ 1 such that annR (Un ) = annR (Ui ) for all i ≥ n. We show that annR∗S (Kn ) = annR∗S (Ki ) for all i ≥ n. Let f(s) = a1 h1 + a2 h2 + · · · + at ht ∈ annR∗S (Ki ). For each m(s) = m1 g1 + · · · + mn gn ∈ Ki , m(s)(R ∗ S)f(s) = 0. Therefore mi gi Rgap hp = 0 for each 1 ≤ j ≤ n, 1 ≤ p ≤ t since M[S] is S-skew quasi-Armendariz. Thus mj Rωgj ωg (ap ) = 0 and so mj Rap = 0, since MR is S-compatible. Therefore ap ∈ ann(Ui ) = ann(Un ) for each 1 ≤ p ≤ t and hence f(s) ∈ annR∗S (Kn ). Thus annR∗S (Kn ) = annR∗S (Ki ). Now assume that M[S]R∗S satisfies the ascending chain condition on annihilator of submodules. Let U1 ⊆ U2 ⊆ · · · be a chain of annihilator of submodules of MR . Then there exist submodules Mi of M such that annR (Mi ) = Ui . Thus M1 ⊇ M2 ⊇ · · · . Hence Mi [S] is a submodule of M[S]R∗S , Mi [S] ⊇ Mi+1 [S] and annR∗S (Mi [S]) ⊆ annR∗S (Mi+1 [S]) for all i ≥ 1. Thus annR∗S (M1 [S]) ⊆ annR∗S (M2 [S]) ⊆ · · · is a chain of annihilator of submodules of M[S] and so there exists n ≥ 1 such that annR∗S (Mn [S]) = annR∗S (Mi [S]). We show that annR (Mn ) = annR (Mi ) for i ≥ n. Assume that r ∈ annR (Mi ). Since M is S-compatible, r ∈ annR∗S (Mi [S]) = annR∗S (Mn [S]) for all i ≥ n. For each m(s) ∈ Mn [S] and r ∈ R, m(s)(R ∗ S)r = 0 which implies that mp gp Rgr = 0 for each 1 ≤ p ≤ t, g ∈ S, since MR is S-skew quasi-Armendariz. Thus mp Rωgp ωg (r) = 0 = mp Rr, since MR is S-compatible, and so r ∈ annR (Mn ). Therefore annR (Mi ) = annR (Mn ). Corollary 15 Let MR be a module and σ a compatible automorphism of R. The following are equivalent: (i) MR satisfies the ascending chain condition on annihilator of submodules. (ii) M[x]R[x;σ] satisfies the ascending chain condition on annihilator of submodules. (iii) M[x, x−1 ]R[x,x−1 ;σ] satisfies the ascending chain condition on annihilator of submodules.


416

E. Hashemi, M. Yazdanfar, A. Alhevaz

References [1] E. P. Armendariz, A note on extensions of Baer and p.p.-rings, J. Austral. Math. Soc., 18 (1974), 470–473. [2] A. Alhevaz and A. Moussavi, On skew Armendariz and skew quasiArmendariz modules, Bull. Iranian. Math. Soc., 38 (2012), 55–84. [3] A. Alhevaz, D. Kiani, McCoy property of skew Laurent polynomials and power series rings, J. Algebra Appl., 13 (2014), 1350083. [4] S. Annin, Associated Primes Over ore extension Rings, J. Algebra Appl., 3 (2004), 193–205. [5] M. Baser, On Armendariz and quasi-Armendariz modules, Not. Mat., 26 (1) (2006), 173–177. [6] G. F. Birkenmeier, Idempotents and completely semiprimes ideals, Comm. Algebra, 11 (1983), 567–580. [7] G. F. Birkenmeier, J. Y. Kim and J. K. Park, Principally quasi-Baer rings, Comm. Algebra, 29 (2001), 639–660. [8] G. F. Birkenmeier, J. Y. Kim and J. K. Park, Polynomial extensions of Baer and quasi-Baer rings, J. Pure Appl. Algebra, 159 (2001), 24–42. [9] G. F. Birkenmeier, J. K. Park, Triangular matrix representations of ring extension, J. Algebra, 265 (2003), 457–477. [10] W. E. Clark, Twisted matrix units semigroup algebras, Duke math. J, 34 (1967), 417–424. [11] N. J. Groenewald, A note on extensions of Baer and p.p.-rings, Publ. Inst. Math., 34 (1983), 71–72. [12] E. Hashemi, Quasi-Armendariz rings relative to a monoid, J. Pure Appl. Algebra, 211 (2007), 374–382. [13] E. Hashemi, Compatible ideals and radicals of Ore extensions, New York J. Math., 12 (2006), 349–356. [14] E. Hashemi and A. Moussavi, Polynomial extension of Baer and quasiBaer rings, Acta Math. Hungar., 107 (2005), 207–224.


On extensions of Baer and quasi-Baer modules

417

[15] Y. Hirano, On annihilator ideals of a polynomial ring over a noncommutative ring, J. Pure Appl. Algebra, 168 (2002), 45–52. [16] I. Kapslanky, Rings of operators, Benjamin, New York, 1965. [17] N. H. Kim and Y. Lee, Armendariz rings and reduced rings, J. Algebra, 223 (2000), 477–488. [18] T. K. Lee and Y. Zhou, Reduced Modules in: Rings, Modules, Algebras, and Abelian Groups, Lecture Notes in Pure and Appl. Math., vol. 236, Marcel Dekker, New York, 2004, pp. 365–377. [19] Z. Liu, Armendariz rings relative to a monoid, Comm. Algebra, 33 (2005), 649–661. [20] S. T. Rizvi and C. S. Roman, Baer and quasi-Baer modules, Comm. Algebra, 32 (2004), 103–123. [21] C. P. Zhang and J. L. Chen, α-skew Armendariz modules and α-semicommutative modules, Taiwanese J. Math, 12 (2008), 473–486.

Received: June 12, 2018


Contents Volume 10, 2018

F. Ayant, D. Kumar Fredholm type integral equation with special functions . . . . . . . . . . 5 M. E. Balazs Maia type fixed point theorems for Ćirić-Prešić operators . . . . . . 18 R. S. Batahan, A. A. Bathanya On generalized Laguerre matrix polynomials . . . . . . . . . . . . . . . . . . . 32 H. Dimou, Y. Aribou, A. Chahbi, S. Kabbaj On a quadratic type functional equation on locally compact abelian groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 H. Fukhar-ud-din, V. Berinde Fixed point iterations for Prešić-Kannan nonexpansive mappings in product convex metric spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 H. Ö. Güney, G. Murugusundaramoorthy, J. Sokól Subclasses of bi-univalent functions related to shell-like curves connected with Fibonacci numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 S. Kajántó, A. Lukács A note on the paper “Contraction mappings in b-metric spaces” by Czerwik . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 R. Khaldi, A. Guezane-Lakoud On generalized nonlinear Euler-Bernoulli Beam type equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

418


N. A. Lone, T. A. Chishti Fundamental theorem of calculus under weaker forms of primitive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 P. Najmadi, Sh. Najafzadeh, A. Ebadian Some properties of analytic functions related with Booth lemniscate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 F. Qi, B.-N. Guo On the sum of the Lah numbers and zeros of the Kummer confluent hypergeometric function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 K. Parand, K. Rabiei, M. Delkhosh An efficient numerical method for solving nonlinear Thomas-Fermi equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 E. Peyghan, F. Firuzi Totally geodesic property of the unit tangent sphere bundle with g-natural metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 R. S. Kushwaha, G. Shanker On the L-duality of a Finsler space with exponential metric αeβ/α . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 N. Ravikumar Certain classes of analytic functions defined by fractional q-calculus operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178 L. Bognár, A. Joós, B. Nagy An improvement for a mathematical model for distributed vulnerability assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 A. A. Bouchentouf, A. Guendouzi, A. Kandouci Performance and economic analysis of Markovian Bernoulli feedback queueing system with vacations, waiting server and impatient customers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218

419


C. Carpintero, N. Rajesh, E. Rosas On real valued ω-continuous functions . . . . . . . . . . . . . . . . . . . . . . . . 242 K. K. Kayibi, U. Samee, S. Pirzada, M. A. Khan Rejection sampling of bipartite graphs with given degree sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249 S. K. Mohapatra, T. Panigrahi Some sufficient conditions for certain class of meromorphic multivalent functions nvolving Cho-Kwon-Srivastava operator . 276 F. Qi, A.-Q. Liu Alternative proofs of some formulas for two tridiagonal determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287 S. Pirzada, M. Imran Computing metric dimension of compressed zero divisor graphs associated to rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298 E. Rosas, C. Carpintero, J. Sanabria Θ-modifications on weak spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319 P. Sahoo, H. Karmakar Uniqueness theorems related to weighted sharing of two sets . . 329 Abdullah, F. A. Shah Scaling functions on the spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340 A. Sofo Integrals of polylogarithmic functions with negative argument 347 M. Şan, H. Irmak A note on some relations between certain inequalities and normalized analytic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368

420


M. Tărnăuceanu Finite groups with a certain number of cyclic subgroups II . . . . 375 M. Zákány New classes of local almost contractions . . . . . . . . . . . . . . . . . . . . . . 378 M. Y. Yilmaz, M. Bektaş Slant helices of (k, m)-type in E4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395 E. Hashemi, M. Yazdanfar, A. Alhevaz On extensions of Baer and quasi-Baer modules . . . . . . . . . . . . . . . 402

421



Information for authors Acta Universitatis Sapientiae, Mathematica publishes original papers and surveys in all field of Mathematics. All papers will be peer reviewed. Papers published in current and previous volumes can be found in Portable Document Format (pdf) form at the address: http://www.acta.sapientia.ro. The submitted papers should not be considered for publication by other journals. The corresponding author is responsible for obtaining the permission of coauthors and of the authorities of institutes, if needed, for publication, the Editorial Board disclaims any responsibility. Submission must be made by email (acta-math@acta.sapientia.ro) only, using the LaTeX style and sample file at the address: http://www.acta.sapientia.ro. Beside the LaTeX source a pdf format of the paper is needed too. Prepare your paper carefully, including keywords, 2010 Mathematics Subject Classification (MSC 2010) codes (http://www.ams.org/msc//msc2010.html), and the reference citation should be written in brackets in the text as [3]. References should be listed alphabetically using the following examples: For papers in journals: A. Hajnal, V. T. Sós, Paul Erdős is seventy, J. Graph Theory, 7 (1983), 391–393. For books: D. Stanton, D. White, Constructive combinatorics, Springer, New York, 1986. For papers in contributed volumes: Z. Csörnyei, Compilers in Algorithms of informatics, Vol. 1. Foundations (ed. A. Iványi), mondAt Kiadó, Budapest, 2007, pp. 80–130. For internet sources: E. Ferrand, An analogue of the Thue-Morse sequence, Electron. J. Comb., 14 (2007) #R30, http://www.combinatorics.org/. Illustrations should be given in Encapsulated Postcript (eps) format. Authors are encouraged to submit papers not exceeding 15 pages, but no more than 10 pages are preferable. One issue is offered each author. No reprints will be available.

Printed by Idea Printing House Director: Péter Nagy



Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.