Journal of Applied Mathematics-1

Journal of
Applied Mathematics
ISSN: 1110-757X (Print)
Volume 2015, Isue 1, 2015
About this Journal

Journal of Applied Mathematics is a peer-reviewed, open access journal devoted to the publication
of original research papers and review articles in all areas of applied, computational, and industrial
mathematics.
Journal of Applied Mathematics currently has an acceptance rate of 22%. The average time between
submission and final decision is 57 days and the average time between acceptance and final
publication is 36 days.
Hindawi Publishing Corporation
410 Park Avenue

15th Floor, #287 pmb
New York, NY 10022
USA
This work is licensed under a Creative Commons Attribution License. Which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
Journal of
Applied Mathematics
Abstracting and Indexing

The articles of Journal of Applied Mathematics are included in the following databases/resources:
Academic OneFile
Academic Search Alumni Edition
Academic Search Complete
Advanced Technologies Database with Aerospace
Aerospace Database
Aluminium Industry Abstracts
ANTE: Abstracts in New Technologies and Engineering
Cabell's Directories
CNKI Scholar
Computer and Information Systems Abstracts
Corrosion Abstracts
CSA Civil Engineering Abstracts
CSA Engineering Research Database
CSA Technology Research Database
Current Abstracts
Current Index to Statistics (CIS)
Directory of Open Access Journals (DOAJ)
Earthquake Engineering Abstracts
EBSCO Engineering Source
EBSCO MainFile
EBSCOhost Connection
Electronics and Communications Abstracts
Google Scholar
HighBeam Research
InfoTrac Custom journals
INIST's Catalog of Articles and Monographs
INSPEC
J-Gate Portal
Mathematical Reviews (MathSciNet)
Mechanical and Transportation Engineering Abstracts
ProQuest Advanced Technologies and Aerospace Collection
ProQuest Computer Science Journals
ProQuest Engineering Collection
ProQuest SciTech Collection
ProQuest Technology Collection
Referativnyi Zhurnal (VINITI)
Scopus
Statistical Theory and Method Abstracts (STMA-Z)

The DBLP Computer Science Bibliography
The Electronic Library of Mathematics (EMIS ELibM)
Zentralblatt MATH Database (zbMATH)
All of Hindawis content is archived in Portico, which provides permanent archiving for electronic
scholarly journals, as well as via the LOCKSS initiative.
410 Park Avenue

New York, NY 10022
USA
This work is licensed under a Creative Commons Attribution License. Which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
Editorial Board
Saeid Abbasbandy, Imam Khomeini International University, Iran
Mina B. Abd-El-Malek, American University in Cairo, Egypt
Mohamed A. Abdou, Alexandria University, Egypt
Subhas Abel, Gulbarga University, India
Janos Abonyi, University of Pannonia, Hungary
Sergei Alexandrov, Russian Academy of Sciences, Russia
M. Montaz Ali, University of the Witwatersrand, South Africa
Mohammad R, Aliha, Iran University of Science and Technology, Iran
Carlos J. S. Alves, Technical University of Lisbon, Portugal
Mohamad Alwash, West Los Angeles College, USA
Gholam R. Amin, Sultan Qaboos University, Oman
Igor Andrianov, RWTH Aachen University, Germany
Boris Andrievsky, Institute for Problems of Mechanical Engineering of the RAS, Russia
Whye-Teong Ang, Nanyang Technological University, Singapore
Abul-Fazal M. Arif, King Fahd University of Petroleum & Minerals, Saudi Arabia
Sabri Arik, Isik University, Turkey
Ali R. Ashrafi, Kashan University, Iran
Allaberen Ashyralyev, Fatih University, Turkey
Francesco Aymerich, University of Cagliari, Italy
Jalel Azaiez, University of Calgary, Canada
Seungik Baek, Michigan State University, USA
Hamid Bahai, Brunel University, United Kingdom
Olivier Bahn, HEC Montral, Canada
Antonio Bandera, Universidad de Malaga, Spain
Jean-Pierre Barbot, ECS ENSEA, France
Mostafa Barigou, University of Birmingham, United Kingdom
Roberto Barrio, Universidad de Zaragoza, Spain
Henri Baudrand, Laboratoire LAPLACE-GRE, France
Alfredo Bellen, University of Trieste, Italy
Vasile Berinde, Universitatea de Nord din Baia Mare, Romania
Jafar Biazar, University of Guilan, Iran
Anjan Biswas, King Abdulaziz University, Saudi Arabia
Abdellah Bnouhachem, Universit Ibn Zohr, Morocco
Gabriele Bonanno, University of Messina, Italy
Walter Briec, University of Perpignan, France
James Robert Buchanan, Millersville University, USA
Humberto Bustince, Public University of Navarra, Spain
Xiao Chuan Cai, University of Colorado, USA
Piermarco Cannarsa, University of Rome Tor Vergata, Italy
Jinde Cao, Southeast University, China
Yijia Cao, Hunan University, China
Zhenfu Cao, Shanghai Jiao Tong University, China
Bruno Carpentieri, University of Groningen, The Netherlands
Ana Carpio, Universidad Complutense de Madrid, Spain
Alexandre Carvalho, University of So Paulo, Brazil
Song Cen, Tsinghua University, China
Wei-Der Chang, Shu-Te University, Taiwan
Tai-Ping Chang, National Kaohsiung First University of Science and Technology, Taiwan
Shih-sen Chang, Yunnan University of Finance And Economics, China
Shuenn-Yih Chang, National Taipei University of Technology, Taiwan
Kripasindhu Chaudhuri, Jadavpur University, India
Yuming Chen, Wilfrid Laurier University, Canada
Jianbing Chen, Tongji University, China
Rushan Chen, Nanjing University of Science and Technology, China
Xinkai Chen, Shibaura Institute of Technology, Japan
Ke Chen, University of Liverpool, United Kingdom

Zhang Chen, Shandong University, China
Zhi-Zhong Chen, Tokyo Denki University, Japan
Ru-Dong Chen, Tianjin Polytechnic University, China
Hui Cheng, Northwestern Polytechnical University, China
Chin-Hsiang Cheng, National Cheng Kung University, Taiwan
Jin Cheng, Bridge Engineering, China
Ching-Hsue Cheng, National Yunlin University of Science and Technology, Taiwan
Qi Cheng, Oklahoma State University, USA
Eric Cheng, The Hong Kong Polytechnic University, Hong Kong
Francisco Chiclana, De Montfort University, United Kingdom
Jen-Tzung Chien, National Chiao Tung University, Taiwan
Chongdu Cho, Inha University, Korea
Han H. Choi, Dongguk University, Republic of Korea
Md Sazzad Hossien Chowdhury, International Islamic University Malaysia, Malaysia
Yu-Ming Chu, School of Mathematics and Computation Science, Hunan City University,
Yiyang 413000, China, China
Hung-Yuan Chung, National Central University, Taiwan
Angelo Ciaramella, Parthenope University of Naples, Italy
Pedro J. Coelho, University of Lisbon, Portugal
Carlos Conca, Universidad de Chile, Chile
Vitor Costa, Universidade de Aveiro, Portugal
Lirong Cui, Beijing Institute of Technology, China
Jie Cui, Tennessee Technological University, USA
Livija Cveticanin, University of Novi Sad, Serbia
Gaoliang Dai, Physikalisch-Technische Bundesanstalt, Germany
Hua Dai, Nanjing University of Aeronautics and Astronautics, China
Binxiang Dai, Central South University, China
Michael Defoort, Universit de Valenciennes, France
Hilmi Demiray, Isik University, Turkey
Ming-Cong Deng, Tokyo University of Agriculture and Technology, Japan
Youjun Deng, Ecole Normale Superieure, France
Orazio Descalzi, Universidad de los Andes, Chile
Raffaele Di Gregorio, Universit di Ferrara, Italy
Kai Diethelm, Technische Universitt Braunschweig, Germany
Daniele Dini, Imperial College London, United Kingdom
Urmila Diwekar, Vishwamitra Research Institute, USA
Vit Dolejsi, Charles University, Czech Republic
BoQing Dong, Anhui University, China
Rodrigo W. dos Santos, Federal University of Juiz de Fora, Brazil
Wenbin Dou, Southeast University, China
Rafael Escarela-Perez, Universidad Autonoma Metropolitana, Mexico
Magdy A. Ezzat, Alexandria University, Egypt
Meng Fan, Northeast Normal University, China
Ya Ping Fang, Sichuan University, China
Istvn Farag, Etvs Lornd University, Hungary
Didier Felbacq, University of Montpellier 2, France
Ricardo Femat, Instituto Potosino de Investigacin Cientfica y Tecnolgica, Mexico
Antonio J. M. Ferreira, Universidade do Porto, Portugal
George Fikioris, National Technical University of Athens, Greece
Michel Fliess, Ecole Polytechnique, France
Marco A. Fontelos, Universidad Autnoma de Madrid, Spain
Dimitris Fotakis, Biosense Center, University of Novi Sad, Serbia
Tomonari Furukawa, Virginia Polytechnic Institute and State University, USA
Maria Gandarias, University of Cdiz, Spain
Xiao-wei Gao, Dalian University of Technology, China
Xin-Lin Gao, Southern Methodist University, USA
Huijun Gao, Harbin Institute of Technology, China

Naiping Gao, Tongji University, China
Laura Gardini, Urbino University, Italy
Winston Garira, University of Venda, South Africa
Leonidas N. Gergidis, University of Ioannina, Greece
Bernard J. Geurts, University of Twente, The Netherlands
Sandip Ghosal, Northwestern University, USA
Dibakar Ghosh, Indian Statistical Institute, Kolkata, India
Pablo Gonzlez-Vera, Universidad de La Laguna, Spain
Alexander N. Gorban, University of Leicester, United Kingdom
Keshlan S. Govinder, University of KwaZulu-Natal, South Africa
Said R. Grace, Cairo University, Egypt
Jose L. Gracia, University of Zaragoza, Spain
Maurizio Grasselli, Politecnico di Milano, Italy
Zhi-Hong Guan, Huazhong University of Science & Technology, China
Nicola Guglielmi, Universit Degli Studi di l'Aquila, Italy
Frdric Guichard, McGill University, Canada
Kerim Guney, Nuh Naci Yazgan University, Turkey
Shu-Xiang Guo, Air Force Engineering University, China
Vijay Gupta, Netaji Subhas institute of Technology, India
Saman K. Halgamuge, University of Melbourne, Australia
Ridha Hambli, Universit d'Orlans, France
Abdelmagid S. Hamouda, Qatar University, Qatar
Bo Han, Harbin Institute of Technology, China
Maoan Han, Shanghai Normal University, China
Pierre Hansen, HEC Montreal, Canada
Ferenc Hartung, University of Pannonia, Hungary
Xiao-Qiao He, City University of Hong Kong, China
Yuqing He, China Academy of Sciences, China
Nicolae Herisanu, Universitatea "Politehnica" Timisoara, Romania
Onesimo Hernandez-Lerma, Centro de Investigacin y de Estudios Avanzados del IPN, Mexico
Luis Javier Herrera, Universidad de Granada, Spain
Johannes C. D. Hoenderkamp, Eindhoven University of Technology, The Netherlands
Thomas Hhne, Helmholtz-Zentrum Dresden-Rossendorf, Germany
Wei-Chiang Hong, Oriental Institute of Technology, Taiwan
Sun-Yuan Hsieh, National Cheng Kung University, Taiwan
Ying Hu, Universit Rennes 1, France
Ning Hu, Chiba University, Japan
Zhilong L. Huang, Zhejiang University, China
Jianguo Huang, Shanghai Jiao Tong University, China
Dan Huang, Hohai University, China
Ting-Zhu Huang, University of Electronic Science and Technology of China, China
Zhenkun Huang, Jimei University, China
Asier Ibeas, Universitat Autonoma de Barcelona, Spain
Mustafa Inc, Frat University, Turkey
Gerardo Iovane, University of Salerno, Italy
Anuar Ishak, Universiti Kebangsaan Malaysia, Malaysia
Takeshi Iwamoto, Hiroshima University, Japan
George Jaiani, Tbilisi State University, Georgia
GunHee Jang, Hanyang University, Korea
Zhongxiao Jia, Tsinghua University, China
Daqing Jiang, China University of Petroleum, China
Haijun Jiang, Xinjiang University, China
Jianjun Jiao, Guizhou University of Finance and Economics, China
Xing Jin, Southeast University, China
Zhen Jin, Shanxi University, China
Lucas Jdar, Universitat Politcnica de Valencia, Spain
Zlatko Jovanoski, University of New South Wales, Australia

Tadeusz Kaczorek, Warsaw University of Technology, Poland
Dongsheng Kang, South-Central University for Nationalities, China
Sangmo Kang, University of California, San Francisco, USA
Ido Kanter, Bar-Ilan University, Israel
Abdul Hamid Kara, University of the Witwatersrand, South Africa
Dimitrios A. Karras, Sterea Hellas Institute of Technology, Greece
Ihsan Kaya, Yildiz Technical University, Turkey
Dogan Kaya, Istanbul Commerce University, Turkey
Chaudry Masood Khalique, North-West University, South Africa
Waqar A. Khan, National University of Sciences and Technology, Pakistan
Khalil Khanafer, University of Michigan, USA
Hyunsung Kim, Kyungil University, Korea
Jong Hae Kim, Sun Moon University, Republic of Korea
Younjea Kim, Sungkyunkwan University, Republic of Korea
Svein Kleiven, Royal Institute of Technology, Sweden
Kazutake Komori, Daido University, Japan
Vassilis Kostopoulos, University of Patras, Greece
Jisheng Kou, Hubei Engineering University, China
Roberto A. Kraenkel, Universidade Estadual Paulista, Brazil
Kannan Krithivasan, SASTRA University, India
Vadim . Krysko, Saratov State Technical University, Russia
Jin L. Kuang, Singapore Offshore Technology Center, Singapore
Gennady M. Kulikov, Tambov State Technical University, Russia
Venugopal Kumaran, National Institute of Technology, India
Miroslaw Lachowicz, University of Warsaw, Poland
Heung-Fai Lam, City University of Hong Kong, Hong Kong
Tak-Wah Lam, The University of Hong Kong, Hong Kong
Hak-Keung Lam, Kings College London, United Kingdom
Luciano Lamberti, Polytechnic University of Bari, Italy
Peter G L Leach, University of Cyprus, Cyprus
Jaehong Lee, Sejong University, Republic of Korea
Usik Lee, Inha University, Republic of Korea
Wen-Chuan Lee, Chang Jung Christian University, Taiwan
Myung-Gyu Lee, Pohang University of Science and Technology, Korea
Jinsong Leng, University of Electronic Science and Technology of China, China
Shuai Li, The Hong Kong Polytechnic University, Hong Kong
Wenlong Li, The University of Hong Kong, Hong Kong
Qingdu Li, Chongqing University of Posts and Telecommunications, China
Lixiang Li, Beijing University of Posts and Telecommunications, China
Xiang Li, Beijing University of Chemical Technology, China
Wan-Tong Li, Lanzhou University, China
Shiyong Li, Yanshan University, China
Qin Li, Beijing Technology and Business University, China
Hua Li, Nanyang Technological University, Singapore
Yan Liang, Northwestern Polytechnical University, China
Dongfang Liang, University of Cambridge, United Kingdom
Jin Liang, Shanghai Jiao Tong University, China
Ching-Jong Liao, National Taiwan University of Science and Technology, Taiwan
Teh-Lu Liao, National Cheng Kung University, Taiwan
Yong-Cheng Lin, Central South University, China
Gao Lin, Dalian University of Technology, China
Chong Lin, Qingdao University, China
Yiping Lin, Kunming University of Science and Technology, China
Ignacio Lira, Pontificia Universidad Catlica de Chile, Chile
Sheng Liu, The University of Queensland, Australia
Zhuangjian Liu, Institute of High Performance Computing, Singapore
Zhengrong Liu, South China University of Technology, China

Weiqing Liu, College of Civil Engineering, Nanjing Tech University, Nanjing, China, China
Meiqin Liu, Zhejiang University, China
Peter Liu, Tamkang University, Taiwan
Peide Liu, Shandong University of Finance and Economics, China
Pengfei Liu, Zhejiang University, China
Tao Liu, Lanzhou University of Technology, China
Zhijun Liu, Hubei University for Nationalities, China
Chein-Shan Liu, National Taiwan University, Taiwan
Yansheng Liu, Shandong Normal University, China
Shutian Liu, Dalian University of Technology, China
Kang Liu, University of Missouri-Columbia, USA
Chongxin Liu, Xi'an Jiaotong University, China
Fei Liu, Jiangnan University, China
Jose L. Lpez, Universidad Pblica de Navarra, Spain
Shiping Lu, Anhui Normal University, China
Hongbing Lu, Hubei University, China
Benzhuo Lu, Chinese Academy of Sciences, China
Yuan Lu, Shenyang University, China
Hao Lu, GE Capital, USA
Henry Horng-Shing Lu, National Chiao Tung University, Taiwan
Gilles Lubineau, King Abdullah University of Science and Technology, Saudi Arabia
Zhen Luo, Tianjin University, China
Lifeng Ma, Nanjing University of Science and Technology, China
Changfeng Ma, Fujian Normal University, China
Ruyun Ma, Northwest Normal University, China
Li Ma, Jiangsu University, China
Shuping Ma, Shandong University, China
Krzysztof Magnucki, Poznan University of Technology, Poland
Nazim I. Mahmudov, Eastern Mediterranean University, Turkey
Oluwole D. Makinde, Stellenbosch University, South Africa
Flavio Manenti, Politecnico di Milano, Italy
Francisco J. Marcelln, Universidad Carlos III de Madrid, Spain
Vasile Marinca, Politehnica University of Timisoara, Romania
Giuseppe Marino, Universit della Calabria, Italy
Guiomar Martn-Herrn, Universidad de Valladolid, Spain
Carlos Martn-Vide, Universitat Rovira i Virgili, Spain
Alessandro Marzani, Universit degli Studi di Bologna, Italy
Nikos E. Mastorakis, Technical University of Sofia, Bulgaria
Nicola Mastronardi, Consiglio Nazionale delle Ricerche, Italy
Panayotis Takis Mathiopouloss, National Observatory of Athens, Greece
Gianluca Mazzini, University of Ferrara, Italy
Michael McAleer, Erasmus University Rotterdam, The Netherlands
Stephane Metens, Universit Paris Diderot - Paris 7, France
Michael Meylan, University of Newcastle, Australia
Efren Mezura-Montes, Universidad Veracruzana Department of Artificial Intelligence, Mexico
Fan Min, Southwest Petroleum University, China
Alain Miranville, Universit de Poitiers, France
Ram N. Mohapatra, University of Central Florida, USA
Syed Abdul Mohiuddine, King Abdulaziz University, Saudi Arabia
Gisele Mophou, Universite des Antilles et de la Guayne, France
Jos Morais, Universidade de Trs-os-Montes e Alto Douro, Portugal
Cristinel Mortici, Valahia University of Targoviste, Romania
Emmanuel Moulay, Universit de Poitiers, France
Jaime E. Munoz Rivera, LNCC, Brazil
Javier Murillo, Universidad de Zaragoza, Spain
Roberto Natalini, Consiglio Nazionale delle Ricerche, Italy
Srinivasan Natesan, Indian Institute of Technology, India

Tatsushi Nishi, Osaka University, Japan
Mustapha Nourelfath, Laval University, Canada
Andreas chsner, Griffith University, Australia
Wlodzimierz Ogryczak, Warsaw University of Technology, Poland
Roger Ohayon, Conservatoire National des Arts et Metiers, France
Javier Oliver, Universidad Politcnica de Valencia, Spain
Soontorn Oraintara, University of Texas at Arlington, USA
Donal O'Regan, National University of Ireland, Ireland
Martin Ostoja-Starzewski, University of Illinois, USA
Turgut zi, Ege University, Turkey
Claudio Padra, Centro Atmico Bariloche, Argentina
Vincent Pagneux, Universit du Maine, France
Reinaldo Martinez Palhares, Federal University of Minas Gerais, Brazil
Quanke Pan, Northeastern University, China
Endre Pap, University of Novi Sad, Serbia
Sehie Park, Seoul National University, Korea
Manuel Pastor, Universidad Politcnica de Madrid, Spain
Giuseppe Pellicane, University of Kwazulu-Natal, South Africa
Francesco Pellicano, Universit di Modena e Reggio Emilia, Italy
Juan Manuel Pea, Universidad de Zaragoza, Spain
Jian-Wen Peng, Chongqing Normal University, China
Ricardo Perera, Universidad Politcnica de Madrid, Spain
Malgorzata Peszynska, Oregon State University, USA
Allan C. Peterson, University of Nebraska-Lincoln, USA
Andrew Pickering, Universidad Rey Juan Carlos, Spain
Somyot Plubtieng, Naresuan University, Thailand
Hector Pomares, University of Granada, Spain
Roland Potthast, University of Reading, United Kingdom
Kerehalli V. Prasad, VSK University, India
Radu-Emil Precup, Politehnica University of Timisoara, Romania
Mario Primicerio, Universit degli Studi di Firenze, Italy
Morteza Rafei, Technische Universiteit Delft, The Netherlands
Phadungsak Rattanadecho, Thammasat University, Thailand
Laura Rebollo-Neira, Aston University, United Kingdom
Mahmoud M. Reda Taha, University of New Mexico, USA
Roberto Ren, Universit di Siena, Italy
Bruno G. M. Robert, Universit de Reims-Champagne-Ardenne, France
Juan A. Rodriguez-Velzquez, Universitat Rovira i Virgili, Spain
Ignacio Rojas, University of Granada, Spain
Carla Roque, Universidade do Porto, Portugal
Debasish Roy, Indian Institute of Science, India
Imre J. Rudas, buda University, Hungary
Abbas Saadatmandi, University of Kashan, Iran
Kunihiko Sadakane, University of Tokyo, Japan
Samir Saker, Mansoura University, Egypt
R. Sakthivel, Department of MathematicsSungkyunkwan University, Republic of Korea
Juan Jos Salazar Gonzlez, Universidad de La Laguna, Spain
Miguel A. F. Sanjuan, Universidad Rey Juan Carlos, Spain
Bogdan Sasu, West University of Timisoara, Romania
Richard Saurel, Aix Marseille Universit, France
Wolfgang Schmidt, Frankfurt School of Finance & Management, Germany
Mehmet Sezer, Celal Bayar University, Turkey
Naseer Shahzad, King Abdulaziz University, Saudi Arabia
Pengjian Shang, Beijing Jiaotong University, China
M. Shariyat, K.N. Toosi University of Technology, Iran
Hui-Shen Shen, Shanghai Jiao Tong University, China
Jian Hua Shen, Chinese Academy of Sciences, China

Yong Shi, The University of Nottingham Ningbo China, China
Yasuhide Shindo, Tohoku University, Japan
Patrick Siarry, Universit de Paris 12, France
Fernando Simes, Instituto Superior Tcnico, Portugal
Theodore E. Simos, University of Peloponnese, Greece
Georgios Sirakoulis, Democritus University of Thrace, Greece
Robert Smith?, The University of Ottawa, Canada
Francesco Soldovieri, Italian National Research Council, Italy
Abdel-Maksoud A. Soliman, Suez Canal University, Egypt
Qiankun Song, Chongqing Jiaotong University, China
Xinyu Song, Xinyang Normal University, China
Yuri N. Sotskov, National Academy of Sciences of Belarus, Belarus
Ivanka Stamova, University of Texas at San Antonio, USA
Niclas Strmberg, Jnkping University, Sweden
Ray K.L. Su, The University of Hong Kong, Hong Kong
Housheng Su, Huazhong University of Science and Technology, China
Chengjun Sun, The University of Hong Kong, Hong Kong
Jitao Sun, Tongji University, China
Wenyu Sun, Nanjing Normal University, China
Wai Yuen Szeto, The University of Hong Kong, Hong Kong
Toshio Tagawa, Tokyo Metropolitan University, Japan
Ying Tan, Peking University, China
San-Yi Tang, Shaanxi Normal University, China
XianHua Tang, Central South University China, China
Nasser-Eddine Tatar, King Fahd University of Petroleum and Mineral, Saudi Arabia
Zhidong Teng, Xinjiang Univeristy, China
Engang Tian, Nanjing Normal University, China
Alexander Timokha, Norwegian University of Science and Technology, Norway
Aydin Tiryaki, Izmir University, Turkey
Yiying Tong, Michigan State University, USA
Hossein Torkaman, Shahid Beheshti University, Iran
Mariano Torrisi, Universit di Catania, Italy
Juan J. Trujillo, Universidad de La Laguna, Spain
Jung-Fa Tsai, National Taipei University of Technology, Taiwan
George Tsiatas, National Technical of University of Athens, Greece
Charalampos Tsitouras, Technological Educational Institute of Chalkis, Greece
Antonia Tulino, Alcatel-Lucent, USA
Stefano Ubertini, Universit degli Studi della Tuscia, Italy
Jeong S. Ume, Changwon National University, Korea
Alexandrina Untaroiu, University of Virginia, USA
Sergey Utyuzhnikov, University of Manchester, United Kingdom
Kuppalapalle Vajravelu, University of Central Florida, USA
Alvaro Valencia, University of Chile, Chile
Olaf van der Sluis, Philips Research Laboratories, The Netherlands
Erik Van Vleck, University of Kansas, USA
Ezio Venturino, Universit degli Studi di Torino, Italy
Jesus Vigo-Aguiar, Univesrisy of Salamanca, Spain
Michael N. Vrahatis, University of Patras, Greece
Michael Vynnycky, Royal Institute of Technology, Sweden
Xiang Wang, Nanchang University, China
Yaonan Wang, Hunan University, China
Shuming Wang, National University of Singapore, Singapore
Youqing Wang, Beijing University of Chemical Technology, China
Dongqing Wang, Qingdao University, China
Peiguang Wang, Hebei University, China
Baolin Wang, Harbin Institute of Technology, China
Guangchen Wang, Shandong University, China

Mingxin Wang, Harbin Institute of Technology, China
Yuh-Rau Wang, Saint John's University, Taiwan
Heng Wang, Institute for Infocomm Research (IR), Singapore
Qing-Wen Wang, Shanghai University, China
Junzo Watada, Waseda University, Japan
Wei Wei, Guizhou University, China
Junjie Wei, Harbin Institute of Technology, China
Jinjia Wei, Xi'an Jiaotong University, China
Guoliang Wei, University of Shanghai for Science and Technology, China
Li Weili, Beijing Jiaotong University, China
Martin Weiser, Zuse Institute Berlin, Germany
Frank Werner, Otto von Guericke University of Magdeburg, Germany
Man Leung Wong, Lingnan University, Hong Kong
Xiangjun Wu, Henan University, China
Cheng Wu, Beijing Institute of Technology, China
Shi-Liang Wu, Anyang Normal University, China
Shanhe Wu, Longyan University, China
Wei Wu, Dalian University of Technology, China
Qingbiao Wu, Zhejiang University, China
Yuqiang Wu, Qufu Normal University, China
Min-Hsien Wu, Chang Gung University, Taiwan
Xian Wu, Yunnan Normal University, China
Yonghui Xia, Zhejiang Normal University, China
Tiecheng Xia, Shanghai University, China
Xian Xian-Yong, Sichuan University, China
Xuejun Xie, Qufu Normal University, China
Gongnan Xie, Northwestern Polytechnical University, China
Yuesheng Xu, Syracuse University, USA
Daoyi Xu, Sichuan University, China
Zhiqiang Xu, Chinese Academy of Sciences, China
Wei Xu, Nanjing University, China
Fuzhen Xuan, East China University of Science and Technology, China
Gregory S. Yablonsky, Washington University, USA
Chao Yan, Intel Corporation, USA
Chao Yang, Chinese Academy of Sciences, China
Chunyu Yang, China University of Mining and Technology, China
Bin Yang, Dalian University of Technology, China
Xiao-Jun Yang, China University of Mining and Technology, China
Dar-Li Yang, National Formosa University, Taiwan
Na Yang, Beijing Jiaotong University, China
Guowei Yang, Qingdao University, China
Chao-Tung Yang, Tunghai University, Taiwan
Zhichun Yang, Chongqing Normal University, China
Her-Terng Yau, National Chin-Yi University of Technology, Taiwan
Wei-Chang Yeh, National Tsing Hua University, Taiwan
Guan H. Yeoh, University of New South Wales, Australia
Chih-Wei Yi, National Chiao Tung University, Taiwan
Simin Yu, Guangdong University of Technology, China
Bo Yu, China University of Petroleum, China
Jinyun Yuan, Universidade Federal do Paran, Brazil
Xiaohui Yuan, Huazhong University of Science and Technology, China
Gonglin Yuan, Guangxi University, China
Rafal Zdunek, Wroclaw University of Technology, Poland
Ashraf Zenkour, Kafrelsheikh University, Egypt
Guisheng Zhai, Shibaura Institute of Technology, Japan
Jianming Zhan, Hubei University for Nationalities, China
Meng Zhan, Chinese Academy of Sciences, China

Zhengqiu Zhang, Hunan University, China
Jiangang Zhang, Lanzhou Jiaotong University, China
Ke Zhang, Nanjing University of Aeronautics and Astronautics, China
Long Zhang, Xinjiang University, China
Zhihua Zhang, Beijing Normal University, China
Henggui Zhang, University of Manchester, United Kingdom
Jifeng Zhang, Harbin Engineering University, China
Liang Zhang, Xidian University, China
Heping Zhang, Lanzhou University, China
Jingxin Zhang, Monash University, Australia
Sheng Zhang, Bohai University, China
Chongbin Zhao, Central South University, Australia
Shan Zhao, University of Alabama, USA
Baosheng Zhao, University of Science and Technology Liaoning, China
Yun-Bo Zhao, Imperial College London, United Kingdom
Renat Zhdanov, BIO-Key International, USA
Dong Zheng, Broadcom Corporation, USA
Huaichun Zhou, Huazhong University of Science and Technology, China
Shangbo Zhou, Chongqing University, China
Huamin Zhou, Huazhong University of Science and Technology, China
Bin Zhou, Harbin Institute of Technology, China
Quanxin Zhu, Nanjing Normal University, China
Xinqun Zhu, University of Western Sydney, Australia
William Zhu, Minnan Normal University, China
Goangseup Zi, Korea University, Korea
Zhiqiang Zuo, Tianjin University, China
Editorial Workflow
The following is the editorial workflow that every manuscript submitted to the journal undergoes
during the course of the peer-review process.
The entire editorial workflow is performed using the online Manuscript Tracking System. Once a
manuscript is submitted for publication, the manuscript is checked by the journals editorial office
to ensure that it is suitable to go through the normal peer review process. Once this is done, the
manuscript is sent to an appropriate Editor based on the subject of the manuscript and the
availability of the Editors. All manuscripts shall be handled by an Editor who is not from the same
institute as any of the manuscripts authors, is not from the same country as any of the manuscripts
authors, and has not co-authored any papers with any of the manuscripts authors in the past three
years.
If the Editor finds that the manuscript may not be of sufficient quality to go through the normal peer
review process, or that the subject of the manuscript may not be appropriate for the journals scope,
the manuscript shall be rejected with no further processing.
If the Editor finds that the submitted manuscript is of sufficient quality and falls within the scope of
the journal, they should assign the manuscript to a number of external reviewers, provided that no
conflict of interests exists between these reviewers and the manuscripts authors. The reviewers will
then submit their reports on the manuscripts along with their recommendation of one of the
following actions to the Editor:
Publish Unaltered
Consider after Minor Changes
Consider after Major Changes
Reject: Manuscript is flawed or not sufficiently novel
When all reviewers have submitted their reports, the Editor can make one of the following editorial
recommendations:
Publish Unaltered
Consider after Minor Changes
Consider after Major Changes
Reject
If the Editor recommends Publish Unaltered, the manuscript will undergo a final check by the
journals editorial office in order to ensure that the manuscript and its review process adhere to the
journals guidelines and policies. Once this is done, the authors will be notified of the manuscripts
acceptance, and the manuscript will appear in the Articles in Press section of the journals website.
If the Editor recommends Consider after Minor Changes, the authors are notified to prepare and
submit a final copy of their manuscript with the required minor changes suggested by the reviewers.
The Editor reviews the revised manuscript after the minor changes have been made by the authors.
Once the Editor is satisfied with the final manuscript, the manuscript can be accepted.
If the Editor recommends Consider after Major Changes, the recommendation is communicated
to the authors. The authors are expected to revise their manuscripts in accordance with the changes
recommended by the reviewers and to submit their revised manuscript in a timely manner. Once the
revised manuscript is submitted, the Editor can then make an editorial recommendation which can
be Publish Unaltered, Consider after Minor Changes, or Reject.
If the Editor recommends rejecting the manuscript, the rejection is immediate. Also, if the majority
of the reviewers recommend rejecting the manuscript, the rejection is immediate.
The editorial workflow gives the Editors the authority to reject any manuscript because of
inappropriateness of its subject, lack of quality, or incorrectness of its results. The Editor cannot
assign himself/herself as an external reviewer of the manuscript. This is to ensure a high-quality,
fair, and unbiased peer-review process of every manuscript submitted to the journal, since any
manuscript must be recommended by one or more (usually two or more) external reviewers along
with the Editor in charge of the manuscript in order for it to be accepted for publication in the
journal.
The name of the Editor recommending the manuscript for publication is published with the
manuscript to indicate and acknowledge their invaluable contribution to the peer-review process
and the indispensability of their contributions to the running of the journals.
The peer-review process is single blinded; that is, the reviewers know who the authors of the
manuscript are, but the authors do not have access to the information of who the peer reviewers are.
Every journal published by Hindawi has an acknowledgment page for the researchers who have
performed the peer-review process for one or more of the journal manuscripts in the past year.
Without the significant contributions made by these researchers, the publication of the journal
would not be possible.
Journal of
Applied Mathematics
Table of Contents
Palmprint Authentication System Based on Local and Global Feature
Fusion Using DOST
N. B. Mahesh Kumar and K. Premalatha
01-11
A Comparison of Evolutionary Computation Techniques for IIR

Model Identification
Erik Cuevas, Jorge Glvez, Salvador Hinojosa, Omar Avalos, Daniel Zaldvar,
and Marco Prez-Cisneros
12-20
Analysis of Repairable Geo/G/1 Queues with Negative Customers

Doo Ho Lee and Kilhwan Kim
21-30
Radial Basis Point Interpolation Method with Reordering Gauss Domains

for 2D Plane Problems
Shi-Chao Yi, Fu-jun Chen, and Lin-Quan Yao
31-44
Initial Time Difference Stability of Causal Differential Systems in terms

of Lyapunov Functions and Lyapunov Functionals
Cokun Yakar and Mustafa Bayram Gcen
45-51
Bifurcation of Safe Basins and Chaos in Nonlinear Vibroimpact Oscillator

under Harmonic and Bounded Noise Excitations
Rong Haiwu, Wang Xiangdong, Luo Qizhi, Xu Wei, and Fang Tong
52-61
A General Setting and Solution of Bellman Equation in Monetary Theory

Xiaoli Gan and Wanbo Lu
62-70
1:3 Resonance and Chaos in a Discrete Hindmarsh-Rose Model

Bo Li and Zhimin He
71-80
Exponential Convergence for Numerical Solution of Integral Equations Using

Radial Basis Functions
Zakieh Avazzadeh, Mohammad Heydari, Wen Chen, and G. B. Loghmani
81-89
Working with Missing Data: Imputation of Nonresponse Items in Categorical

Survey Data with a Non-Monotone Missing Pattern
Machelle D. Wilson and Kerstin Lueck
90-98

Journal of Applied Mathematics
Volume 2015, Issue 1, 2015
http://dx.doi.org/10.1155/2014/918376
Research Article
Palmprint Authentication System Based on Local and Global
Feature Fusion Using DOST
N. B. Mahesh Kumar and K. Premalatha
Bannari Amman Institute of Technology, Sathyamangalam, Erode, Tamilnadu 638 401, India
Correspondence should be addressed to N. B. Mahesh Kumar; mknbmaheshkumar@gmail.com
Received 23 September 2014; Revised 16 December 2014; Accepted 16 December 2014; Published 31 December 2014
Academic Editor: Qiankun Song
Copyright 2014 N. B. M. Kumar and K. Premalatha. Tis is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
Palmprint is the region between wrist and fngers. In this paper, a palmprint personal identifcation system is proposed based on
the local and global information fusion. Te local and global information is critical for the image observation based on the results of
the relationship between physical stimuli and perceptions. Te local features of the enhanced palmprint are extracted using discrete
orthonormal Stockwell transform. Te global feature is obtained by reducing the scale of discrete orthonormal Stockwell transform
to infnity. Te local and global matching distances of the two palmprint images are fused to get the fnal matching distance of the
proposed scheme. Te results show that the fusion of local and global features outperforms the existing works on the available three
datasets.
1. Introduction
Te unique physical or behavioural characteristics of human
beings are of broad interest in the biometrics based methods
[1]. In the modern e-world, biometrics has great potentials
because of their high accuracy and convenience to use.
Researchers and scientists have systematically investigated
the use of a number of biometric characteristics like fngerprint [25], face [6], iris [7, 8], palmprint [9, 10], hand
geometry [11], hand vein [12], fnger knuckle-print [13, 14],
voice [15], and ear [16] in the development of computing
techniques. Te use of applications like automatic personal
authentication systems in day-today life results in reliable and
efective security control. Recently, hand-based biometrics
has been attracting signifcant attention in the modern eworld.
Te region between wrist and fngers is referred to as
palmprint. It has features like texture, wrinkles, principle
lines, ridges, and minutiae points that can be used for its
representation. Te critical properties of biometric characteristics such as universality, individuality, stability, and
collectability are satisfed by palmprint compared to other
biometric traits. It has relatively stable and unique features. It
is nonintrusive and data collection is very easy. It needs fewer
cooperations to collect data from persons. It provides high

efciency using low resolution images taken from the low
cost device. Recently palmprint based recognition systems
have received a wide attention from researchers and eforts
have been made to build a palmprint based system based
on structural features of palmprints such as principle lines,
wrinkles, datum points, minutiae points, ridges, and crease
points.
Features extracted from the palmprint image can be
classifed into local and global features. Local features characterize the main topological characteristics of a small part
of the trajectory. Global features characterize the relationship
of diferent line segments within a trajectory. Te overall
content and shape of the objects in the image represent the
global features. Te specifc information in the local regions
is defned as local features. Te advantages of local features are
defned in terms of template size and its discriminability. But
the local features also have inevitable drawbacks in practical
usage. Due to the sensitivity of the palmprint image, it is
difcult to exactly obtain the local features. Te performance
of the local features degrades due to its poor palmprint quality
and capture area. Te global features represent the palmprint
in a global perspective. Most of the global features are
continuous and smooth everywhere expect in some special
Image acquisition
Preprocessing
Local and global

feature extraction
Database
Enroll/
verify/
identify
Decision
Matching
Figure 1: Block diagram of the proposed palmprint based recognition system.
regions. Te global features can be extracted more consistently from the poor quality of the partial palmprints. It is
too space and time consuming to directly store and compare
the features pixel by pixel. Global feature based matching
overlaps two given templates with diferent transformation
parameters and estimates the similarity score between the
corresponding cells. Compared to local features, the global
features have less distinctness and hence been ofen exploited
together with other features or in the preprocessing stage
of palmprint matching. Both the local and global features
are fairly independent and capture current information and
hence it is realistic to improve the discriminating ability of
matching by fusing features. Te local and global features are
combined in the matching stage with accessible feature level
fusing strategies. Te combined local and global features can
improve the performance of the palmprint systems on large
scale databases. How to select features is pivotal for the efect
of feature combination. Te combination of irrelative features
will improve the accuracy or efciency of the palmprint
biometric systems. Te suitable hierarchical approach can
be employed to reduce the additional time or memory cost
required while fusing the local and global features.
In this work, a local-global feature fusion technique for
palmprint biometric system is proposed. Te instantaneous
phase information is extracted as the local feature by DOST.
Te global feature can be obtained by aggregating the scale
of the DOST. Te DOST can be reduced to the Fourier
transform of the whole image if the scale of the DOST is
increased to infnity. Hence, the local features cannot be
obtained but the fnest resolution for the global frequency
analysis of the image can be obtained. Terefore, Fourier
transform coefcients are obviously taken as the global
features. Te arrangement between intraclass palmprint ROIs
can also be refned by the global Fourier features. Te two
matching distances are calculated by matching the local and
the global information separately. Ten the two matching
distances are merged by fusion rule to get the fnal matching
distance. Te block diagram of the proposed system is shown

in Figure 1.
Te rest of this paper is organized as follows. Section 2
describes the review on palmprint authentication. Section 3
describes the overview of discrete orthonormal Stockwell
transform (DOST). Preprocessing and ROI extraction is
discussed in Section 4. Section 5 illustrates the local and
global feature extraction and matching. Section 6 presents
local-global information fusion for palmprint recognition.
Experimental results and discussion are demonstrated in
Section 7. Section 8 reports the conclusion.
2. Literature Review
In [4], a new BioHashing technique is proposed to generate
a cancellable template in the context of unordered set of
noisy minutiae features and shows that the work is efcient in
realistic context. In [17], an intersession variability modelling
(ISV) and joint factor analysis (JFA) for face authentication
are proposed to achieve a least signifcant error rate reduction
for measuring accuracy. Te line features in palmprint are
extracted by using sobel and morphological methods [10].
In [18], a new chaf point generation technique for the fuzzy
vault in biocryptosystems is proposed and shows that the
proposed technique achieves more chaf points compared to
Khalil-Hanis algorithm. In [19], the principle lines along with
isolated points are used as features. Te ridges containing
creases of the palmprint are eliminated and it has been
proposed in [20]. Te features like end points of principle
lines have been used in [9]. Te end points in the palmprint
are found to be directional invariant. Te crease points that
are related to diseases of a person have been suggested in
the palmprint biometrics system. Te directional line energy
features are also used to identify the palmprint of a person.
Further, the palmprint biometric systems that are based
on structural features are not invariant to occlusion of the
palmprint image.

Te feature extraction and pattern matching of palmprint
images are focused mainly. Te features in the palmprint
images can be classifed as local and global features based
on the range of pixels involved in the feature extraction. Te
local feature is a quantity calculated within the local patch and
scrambling the detailed behaviours within this specifc area.
Global information is obtained from all or most of the pixels
in the palmprint image and the universal characteristics of the
observed image can be examined. Hence, palmprint biometrics systems can be classifed into local based approaches and
global based approaches according to these defnitions. Te
local-global combination for palmprint recognition has been
discussed only in few papers. Te local and global features
are critical for the image observation and identifcation of
persons as per the theories in the feld of psychophysics
and neurophysiology [22]. A global feature reproduces the
universal features of the image and is appropriate for rough
representation. A local feature scrambles the more complete
information within the specifc region. Terefore, local feature is suitable for better representation. Hence, the improved
recognition accuracy is obtained if the local and global feature
can be appropriately combined. It is already used in biometrics systems like iris, face, and fngerprint. Te conventional
Daugman-like classifer uses global features enclosed by zerocrossing boundaries in the palmprint biometrics systems.
In two-level palmprint matching scheme [23], the global
features for coarse level fltering can be extracted by Hough
transform and, for fne-level matching, the local features
extracted from the localities and directions of the principal
lines in the palmprint.
Te local and global features are obtained from the
nonnegative matrix factorization technique and principle
component analysis in [24] can be combined for palmprint
biometric system. In face recognition [6], the global and
local information are extracted by principle component
analysis (PCA) and Haar wavelets. Global features in the face
images are extracted by keeping the low-frequency Fourier
coefcients, and the local information is extracted using
Gabor flters. Te local and global information combination
was also utilized in the fngerprint biometrics system [25].
3. Overview of Discrete Orthonormal

Stockwell Transform
Te S-transform is more powerful than other multiresolution
techniques like short time Fourier transform (STFT) and
wavelet transform (WT). Te phase of the S-transform referenced to the time origin provides useful and supplementary
information about spectra that is not available from locally
referenced phase information in the continuous wavelet
transform (CWT).
Te S-transform is advantageous for the analysis of
palmprint images as it preserves the phase information using
linear frequency scaling.
(1) However, the major limitation of S-transform is its
high time and space complexity due to its redundant
nature, which makes it impractical in many cases.
3
(2) Te 2D-ST of an array of size has a computational complexity of O (4 + 4 log ) and storage
requirements of O (4 ).
(3) To eliminate this problem of 2D-ST, we use DOST
which is also a multiresolution technique for extraction of features from the palmprint images and is
based on the S-transform.
(4) DOST have less computational and storage complexity in comparison to the S-transform because it uses
an orthonormal set of basis functions, while retaining
all the advantageous properties of S-transform.
(5) 2D-DOST provides a spatial frequency representation
of an image, with computational and storage complexity as O (2 + 2 log ) and O (2 ), respectively.
With the dyadic sampling scheme in order 0, 1, 2; . . . log2

1, DOST of an palmprint image (, ) is performed
by the following steps.
(1) Two-dimensional Fourier transform (FT) is applied
to the image (, ) to obtain Fourier samples (, V).
(2) Partition the Fourier sample (, V) and multiply it

by the square root of the number of points in the
partition, and perform an inverse FT. Ten the voice
image is calculated as
( , , V , V ) =
2
1 2
2 , 2
( + V , V + V ) ,
=22 V=22
here V = 2 1+2
2 (
+
),
21 21
V = 2 1+2
(1)
2
where V and V are the horizontal and vertical voice

frequencies, where corresponds to -coordinate
in space, where corresponds to -coordinate in
space.
(3) Tus, the features of the palmprint images are
obtained afer the transformation.
4. Preprocessing and ROI Extraction

In this section, the acquired hand image is preprocessed and
the palmprint region is extracted. Te square area inside
the palm region of the hand image is considered as the
palmprint or region of interest (ROI). Normally the extracted
palm region has the rotation and translation error. In order
to minimize the rotation and translation error, the ROI
algorithm identifes gaps between fngers. Te gaps between
the fngers are used as the reference points to align the
image. Te images of the extracted palm region are aligned
rotationally by levelling the gap between the index fnger
(a)
(b)
Figure 2: (a) Sample input image. (b) ROI image for PolyU palmprint database.
(a)
(b)
Figure 3: (a) Sample input image. (b) ROI image for COEP palmprint database.
(a)
(b)
Figure 4: (a) Sample input image. (b) ROI image for IIT Delhi palmprint database.
(IF) and the middle fnger (MF) and that between the MF
and the ring fnger (RF). Among the three gaps, the gap
between MF and RF is the toughest point. In order to
eliminate the translation error, the gap between MF and RF
is used as the reference point. Te maximum square region
extracted from the palm area is selected as the ROI. Te
square region is horizontally centered on the axis running
through the gap between MF and RF. Due to the regular and
controlled uniform illumination conditions during image
capture, the acquired hand image and its background contrast
in colour. Te sample input image is shown in Figures 2(a),
3(a), and 4(a) for PolyU palmprint database, COEP palmprint database, and IIT Delhi palmprint database. Global
thresholding has been applied to extract the hand from the
background. Te image is processed with opening and closing
morphological operations to remove any isolated small blobs
or holes. Contour-tracing algorithm is applied to extract the
contour of the hand image from the palmprint. Te ROI
extraction technique of IIT Delhi palmprint database, PolyU
palmprint database, and COEP palmprint database are same
as in [9, 26]. Te corresponding Figures 2(b), 3(b), and 4(b)
are shown in this paper which gives the details of ROI image.
5. Local-Global Feature
Extraction and Matching
0.1
5.1. Local Feature. Te feature extraction technique in any

palmprint biometric system is used to obtain good interclass
separation in minimum time. Te features of palmprint
are extracted by using 2D-DOST afer the completion of
preprocessing and ROI segmentation from palmprint. Te
local variation of instantaneous phase is used to extract
features from palmprint. Instantaneous phase obtained using
DOST is the resolution of phase with respective to time. It
has the merits of great accurateness, sturdiness to brightness
variation, and fast matching. When matching two palmprint
images, the angular distance based on the normalized Hamming distance is used.
5.2. Global Feature. DOST is used to extract the instantaneous phase information and also it can be considered as a
windowed Fourier transform. Te DOST can be considered
as a windowed Fourier transform. Te phase-only correlation
(POC) and band-limited phase-only correlation (BLPOC)
are utilized to match the global information of the two
palmprint images.
5.2.1. Phase-Only Correlation (POC). Te Fourier transform
coefcients are considered as the global information. Te
two given Fourier transforms are matched using POC. POC
based techniques have been widely used in image registration
tasks. Recently, a POC technique considered as a resemblance
measure in certain biometrics systems [7, 22]. BLPOC is
more efective when compared to the conventional POC.
BLPOC is utilized to estimate the translation parameters
between palmprint ROIs and also measure the resemblance
of the Fourier transforms of the aligned ROIs. POC is an
operational method to estimate the displacement parameters
between two palmprint images in the Fourier transform
domain. Its primary principle is the translation property of
the Fourier transforms.
Consider two 1 2 images (1 , 2 ) and (1 , 2 ),
where the assumption is that the index ranges are 1 =
1 , . . . , 1 (1 > 0) and 2 = 2 , . . . , 2 (2 > 0).
For mathematical simplicity 1 = 21 + 1 and 2 =
22 + 1. Let (1 , 2 ) and (1 , 2 ) denote the 2D discrete
Fourier transforms (2D DFTs) of the two images. (1 , 2 )
and (1 , 2 ) are given by
(1 , 2 ) = (1 , 2 ) 11 1 22 2 = (1 ,2 ) (1 ,2 ) ,
1 ,2
(1 , 2 ) = (1 , 2 ) 11 1 22 2 = (1 ,2 ) (1 ,2 ) ,
1 ,2
(2)
where 1 = 1 , . . . , 1 . 2 = 2 , . . . , 2 , 1 =
(2/1 ) . 2 = (2/2 ) and the operator 1 2 denotes
1
2
1 =
2 =
. (1 ,2 ) and (1 ,2 ) are amplitude com1
2
ponents and (1 , 2 ) and (1 , 2 ) are phase components.
0.08
0.066
0.044
0.022
0
0.02
6000
8800
400
600
400
200
200
0
Figure 5: POC function for PolyU palmprint database.
Te cross spectrum (1 , 2 ) between (1 , 2 ) and

(1 , 2 ) is given by
(1 , 2 ) = (1 , 2 ) (1 , 2 )
= (1 ,2 ) (1 ,2 ) (1 ,2 ) ,
(3)
where (1 , 2 ) denotes the complex conjugate of (1 , 2 )

and (1 , 2 ) = (1 , 2 ) (1 , 2 ). On the other hand,
1 , 2 )
the cross spectrum (or normalized cross spectrum) (
is defned as
(1 , 2 ) (1 , 2 )
(1 , 2 ) =
(1 , 2 ) (1 , 2 )
(4)
= (1 ,2 ) .
Te POC function (1 , 2 ) = is 2D inverse discrete Fourier

1 , 2 ) and is given by
transforms (IDFT) of (
(1 , 2 ) =
1

(1 , 2 ) 1 1 1 2 2 2 ,
1 , 2
(5)
1
2
where 1 2 denotes =
.
=
1
1
2
2
Te most signifcant property of POC is compared with
the normal correlation and it gives accuracy in pattern matching. Te distinct sharp peak is obtained when POC function
is the same. Te peak in the POC drops considerably if the
two palm print images are not the same. Figures 5, 6, and 7
show the POC function of ROI for PolyU palmprint database,
COEP palmprint database, and IIT Delhi palmprint database.
Tus, the POC function shows much higher perception
ability than the normal correlation function. In order to
match the two palmprint images, the height of the peak will
be used as a good resemblance measure.
5.2.2. Band-Limited Phase-Only Correlation (BLPOC). In

the POC based image matching method, all the frequency
components are involved. However, high frequency components can be susceptible to noise. BLPOC is used to
eliminate noisy high frequency components. BLPOC bounds
0.1
0.15
0.08
0.1
0.06
0.04
0.05
0.02
0
0
0.02
6000
0.05
300
800
8
400
400
200
300
200
100
200
0
4400
200
600
100
0 0
Figure 6: POC function for COEP palmprint database.
Figure 8: BLPOC function for PolyU palmprint database.
0.06
0.15
0.044
0.1
0.022
0.05
0
0
0.022
6000
0.05
300
800
8
400
600
200
0
4
400
200
400
200
300
200
100
100
0
Figure 7: POC function for IIT Delhi palmprint database.

Figure 9: BLPOC function for COEP palmprint database.
the range of spectrum of the given palmprint image. BLPOC

function between two ROI images can be measured as the
POC function between their low-pass fltered versions based
on the defnition of BLPOC. Terefore, BLPOC function
can preserve the properties of the POC function. BLPOC
function got a distinct peak if two ROI images are the
same. But the locality of peak in two ROI images gives
the translational displacement. Experimentations specify that
BLPOC function provides a much higher perception ability
than the original POC function in palmprint biometric
system. Terefore, the efcient size of spectrum is given by
1 = 20 + 1 and 2 = 20 + 1. BLPOC function is given by
0 0 (, ) =
0
0
1
(,V)2(/ 1 +V/ 2 ) ,
1 2 = V=
0
0.15
0.1
0.05
0
0.05
300
400
200
300
100
200
100
(6)
where = 0 , . . . , 0 , and = 0 , . . . , 0 , where =

0 , . . . , 0 , and V = 0 , . . . , 0 , where 0 0 0 ,
0 0 0 .
Figures 8, 9, and 10 are BLPOC function for PolyU
palmprint database, COEP palmprint database, and IIT Delhi
palmprint database, respectively. Tese examples indicate
that in the instance of genuine matching (a matching performed between a pair of palmprint images from the same
hand) the BLPOC will show a much sharper peak than POC;
Figure 10: BLPOC function for IIT Delhi palmprint database.
however, for an imposter matching (a matching performed

between a couple of palmprint images from diferent hands),
neither BLPOC nor POC will display a diferent severe peak.
BLPOC function can be adopted to align the translation
between two palmprint ROI images and then to measure the
resemblance between Fourier transforms of the aligned ROIs.
0.9
0.9
Genuine acceptance rate (%)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
105
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
104
103
102
False acceptance rate (%)
101
0
105
100
Figure 11: ROC curve using local features of proposed DOST for
PolyU palmprint database.
101
100
COEP palmprint database.
0.9
0.9
103
102
Proposed work
Ordinal code
Palm code
Proposed work
Ordinal code
Palm code
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.1
0
105
104
104
103
102
101
100
Proposed work
Ordinal code
Palm code
0
105
104
103
102
101
100
Proposed work
Ordinal code
Palm code
Figure 12: ROC curve using global features of proposed DOST for
PolyU palmprint database.
COEP palmprint database.
6. Local-Global Information Fusion (LGIF) for

Palmprint Recognition
function. Ten the mutual regions can be cropped based on

which the pattern matching is completed. Te displacement
parameters between two palmprint ROI images can be calculated from the peak locality of global BLPOC function. Ten
two palmprint ROI images can be aligned based on the displacement parameters and extract the mutual regions. Ratio
between the mutual region area and the area of the original
ROI can be checked in the palmprint system. Afer alignment
and common area cropping, two palmprint ROI images are
constructed. Ten matching distance is obtained by matching
two palmprint ROI images. Te angular distance based on
the normalized Hamming distance is used to match the two
palmprint images for local features in palmprint biometric
systems. Te two palmprint images and are created
afer alignment and common area cropping. Te matching
LGIF based palmprint recognition algorithm is described in

this section. Te palmprint image acquisition device and the
ROI extraction system can decrease the symmetrical transformations between intraclass ROIs. But it is quiet unavoidable
because there is some translation between intraclass ROIs
in the palmprint images. Tese will deteriorate the genuine
matching scores. Tis problem is solved by transforming
the given information in horizontal and vertical directions
numerous times and the smallest of the subsequent similar
distances is measured as the fnal matching distance. Tis
problem can be solved in a dissimilar way by calculating the
displacement parameters between two ROIs using BLPOC

1
0.9
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.1
0
105
0.8
104
103
102
101
0
105
100
IIT Delhi palmprint database.
101
100
Figure 17: ROC curve using both local and global features of
proposed DOST for PolyU palmprint database.
0.9
0.9
103
102
Proposed work
Ordinal code
Palm code
Proposed work
Ordinal code
Palm code
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
105
104
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
104
103
102
101
100
Proposed work
Ordinal code
Palm code
IIT Delhi palmprint database.
distance is obtained by matching the two palmprint images

and . Te peak value of the BLPOC function between
two palmprint images is used to measure the similarity of
their Fourier transforms. Te peak value is indicated as pocS;
therefore, the matching distance is termed as = 1 pocS.
Te two matching distances and are obtained. We
got the fnal matching distance by fusing the two distances
and . In BLPOC function, the peak value between
palmprint ROI images is used to calculate the resemblance
of their Fourier transforms to obtain matching distance. In
this paper, two matching distances are obtained from two
dissimilar matchers; namely, local feature based matcher
(matcher 1) and global feature based matcher (matcher 2)
and the maximum weighted (MW) rule can be adopted.
Te weights are assigned according to the equal error rate
0
105
104
103
102
101
100
Proposed work
Ordinal code
Palm code
proposed DOST for COEP palmprint database.
(EER) obtained on a training dataset by diferent matchers

based on MW fusion rule. It is obvious that the weights are
inversely proportional to the corresponding EERs. Ten, the
fnal matching distance is estimated.
7. Experimental Results and Discussion

Tree diferent sets of hand image databases are used for the
proposed system. Te PolyU palmprint database [27] contains 7752 grayscale images corresponding to 386 diferent
palms in BMP image format. Te twenty samples are collected
in two diferent sessions. 10 samples were collected from frst
session and 10 samples were captured in the second session,
respectively.
9
Table 1: Performance comparison of diferent palmprint verifcation
schemes for both local and global combination.
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
105
104
103
102
101
100
Proposed work
Ordinal code
Palm code
proposed DOST for IIT Delhi palmprint database.
CRR (%)
EER (%)
(a) Performance for PolyU database
Palm code [9]
99.91
0.5338
Ordinal code [21]
100.00
0.0709
Proposed
100.00
0.0035
(b) Performance for COEP database
Palm code [9]
99.61
3.6732
Ordinal code [21]
99.84
1.7544
Proposed
100.00
1.2605
(c) Performance for IIT Delhi database
Palm code [9]
99.68
5.2108
Ordinal code [21]
100.00
1.1762
Proposed
100.00
1.0217
DI
5.5807
6.6785
7.2541
3.0734
3.3966
4.0228
1.800
2.2174
3.0168
Table 2: Computation time for key processes both local and global
combination.
3.5
Proposed
3
Proposed
2.5
Proposed
Extraction (ms)
Matching (ms)
(a) Systems speed for PolyU database
54.56
1.5
(b) Systems speed for COEP database
85.41
3.07
(c) Systems speed for IIT Delhi database
83.34
3.15
Total (ms)
56.06
88.48
86.49
(%)
1.5
0.5
0
0.1
0.2
0.3
0.4
Matching distance
0.5
Imposter
Genuine
Figure 20: Distance distributions of genuine matching and imposter

matching of proposed DOST for PolyU palmprint database.
COEP palmprint database contains [28] 1344 images

corresponding to 168 people. Eight diferent images are
collected from single persons palm.
IIT Delhi palmprint database [29] contains 230 subjects.
Te subjects in the palmprint database are collected from
the age group 1257 years. Seven images per each subject
are collected from the lef and right hand. An identifcation
number is assigned to every user for the acquired images.

Te resolution of these images is 800 600 pixels. Afer
preprocessing, original images, 150 150 pixels, are automatically cropped and considered as ROI.
Verifcation means by evaluating one or more individual
genetic traits of a person can be uniquely identifed. In this
paper, all the classes of palmprint images were involved. Te
metrics such as EER, correct recognition rate (CRR), and
receiver operating characteristics (ROC) are used to measure
the recognition accuracy. Te CRR of a system can be defned
by
CRR =
1
100,
2
(7)
where 1 denotes the number of correct recognition of palmprint images and 2 is the total number of palmprint images.
Performance comparison of diferent palmprint verifcation
schemes is shown in Table 1. Table 2 shows the computation
time for key processes.
ROC curves are used to analyse the performance of
the palmprint biometric system. ROC is a plot of genuine
acceptance rate (GAR) against false acceptance rate (FAR)
and shown in Figures 11, 12, 13, 14, 15, 16, 17, 18, and 19
for PolyU palmprint database, COEP palmprint database,
and IIT Delhi palmprint database. Te FAR and FRR of a
palmprint biometric system can be obtained by adjusting the
various thresholds. It is known that both FAR and FRR are
10

well the genuine and the imposter distributions are separated
which is defned as
2
,
DI = 1
(8)
(12 + 22 ) /2
3.5
2.5
where 1(2 ) is the mean of the genuine (imposter) matching

distances and 1 (2 ) is the standard deviation of the genuine
(imposter) matching distances. Ten performance of any
biometric system can be measured by its score among other
metrics.
(%)
1.5
8. Conclusion
0.5
0
0.1
0.2
0.3
0.4
Matching distance
0.5
Imposter
Genuine

matching of proposed DOST for COEP palmprint database.
3.5
In this work, palmprint images are used for recognition. LGIF

based palmprint features are critical for the image recognition
and observation. Tey play dissimilar and corresponding
roles in a system. In LGIF, the local instantaneous phase
information obtained by the DOST technique is measured
as the local feature. From the perception of time-frequency
analysis, if the scale of the DOST goes to infnity, it converts
to the Fourier transform. Te global information of palmprint
recognition system refnes the alignment of palmprint image
in pattern matching. Extensive investigational outcomes
shown on the palmprint database indicate that the proposed
system could achieve much better performance in terms of
EER.
Conflict of Interests
Te authors declare that there is no confict of interests
regarding the publication of this paper.
2.5
2
(%)
Acknowledgments
Te authors would like to thank Hong Kong Polytechnic
University, IIT-Delhi, and College of Engineering-Pune for
providing the palmprint databases.
1.5
References
0.5
0
0.1
0.2
0.3
0.4
Matching distance
0.5
Imposter
Genuine

matching of proposed DOST for IIT Delhi palmprint database.
inversely related. Tus, at a particular threshold, these two

curves intersect. Tis intersect point is termed as an EER.
Te EER is defned at the FAR which is equal to the FRR.
Te distance distributions of genuine matching and imposter
matching acquired by proposed scheme are plotted in Figures
20, 21, and 22. Te decidability index (DI) is defned as how
[1] A. K. Jain, P. J. Flynn, and A. Ross, Handbook of Biometrics,

Springer, Berlin, Germany, 2010.
[2] D. Maltoni, D. Maio, A. K. Jain, and S. Prabhakar, Handbook of
Fingerprint Recognition, Springer, London, UK, 2009.
[3] N. Ratha and R. Bolle, Automatic Fingerprint Recognition
Systems, Springer, New York, NY, USA, 2011.
[4] R. Belguechi, E. Cherrier, C. Rosenberger, and S. Ait-Aoudia,
Operational bio-hash to preserve privacy of fngerprint minutiae templates, IET Biometrics, vol. 2, no. 2, pp. 7684, 2013.
[5] J. Gu, J. Zhou, and C. Yang, Fingerprint recognition by
combining global structure and local cues, IEEE Transactions
on Image Processing, vol. 15, no. 7, pp. 19521964, 2006.
[6] F. Yuchun, T. Tieniu, and W. Yunhong, Fusion of global
and local features for face verifcation, in Proceedings of the
International Conference on Pattern Recognition, pp. 382385,
Quebec, Canada, August 2002.
[7] K. Miyazawa, K. Ito, T. Aoki, K. Kobayashi, and H. Nakajima,
An efective approach for Iris recognition using phase-based

image matching, IEEE Transactions on Pattern Analysis and
Machine Intelligence, vol. 30, no. 10, pp. 17411756, 2008.
[8] Z. Sun, Y. Wang, T. Tan, and J. Cui, Improving iris recognition
accuracy via cascaded classifers, IEEE Transactions on Systems,
Man and Cybernetics C: Applications and Reviews, vol. 35, no. 3,
pp. 435441, 2005.
[9] D. Zhang, W.-K. Kong, J. You, and M. Wong, Online palmprint identifcation, IEEE Transactions on Pattern Analysis and
Machine Intelligence, vol. 25, no. 9, pp. 10411050, 2003.
[10] C. C. Han, H. L. Cheng, C. L. Lin, and K. C. Fan, Personal
authentication using palm-print features, Pattern Recognition,
vol. 36, no. 2, pp. 371381, 2003.
[11] H. Matos, P. Oliveira, and F. Magalhaes, Hand-geometry based
recognition system, in Image Analysis and Recognition, vol.
7325 of Lecture Notes in Computer Science, pp. 3845, Springer,
2012.
[12] D. Huang, Y. Tang, Y. Wang, L. Chen, and Y. Wang, Hand vein
recognition based on oriented gradient maps and local feature
matching, in Computer Vision, vol. 7727 of Lecture Notes in
Computer Science, pp. 430444, Springer, New York, NY, USA,
2013.
[13] L. Nanni and A. Lumini, A multi-matcher system based on
knuckle-based features, Neural Computing and Applications,
vol. 18, no. 1, pp. 8791, 2009.
[14] A. Kumar and Y. Zhou, Personal identifcation using fnger
knuckle orientation features, Electronics Letters, vol. 45, no. 20,
pp. 10231025, 2009.
[15] H. Hollein, Forensic Voice Identifcation, Academic Press, 2002.
[16] C. Middendorf, Multi-Biometric Approaches to Ear Biometrics
and Sof Biometrics, BiblioBazaar, 2011.
[17] C. McCool, R. Wallace, M. McLaren, L. El Shafey, and S.
Marcel, Session variability modelling for face authentication,
IET Biometrics, vol. 2, no. 3, pp. 117129, 2013.
[18] T. H. Nguyen, Y. Wang, Y. Ha, and R. Li, Improved chaf
point generation for vault scheme in bio-cryptosystems, IET
Biometrics, vol. 2, no. 2, pp. 4855, 2013.
[19] N. Duta, A. K. Jain, and K. V. Mardia, Matching of palmprints,
Pattern Recognition Letters, vol. 23, no. 4, pp. 477485, 2002.
[20] J. Funada, N. Ohta, M. Mizoguchi et al., Feature extraction
method for palmprint considering elimination of creases, in
Proceedings of the International Conference on Pattern Recognition, pp. 18491854, Brisbane, Australia, August 1998.
[21] Z. N. Sun, T. Tan, Y. Wang, and S. Z. Li, Ordinal palmprint
represention for personal identifcation [represention read
representation], in Proceedings of the IEEE Computer Society
Conference on Computer Vision and Pattern Recognition (CVPR
05), vol. 1, pp. 279284, June 2005.
[22] Y. Su, S. Shan, X. Chen, and W. Gao, Hierarchical ensemble of
global and local classifers for face recognition, IEEE Transactions on Image Processing, vol. 18, no. 8, pp. 18851896, 2009.
[23] F. Li, K. H. Leung, and X. Yu, A two-level matching scheme
for speedy and accurate palmprint identifcation, in Advances
in Multimedia Modeling, vol. 4352 of Lecture Notes in Computer
Science, pp. 323332, Springer, Berlin, Germany, 2006.
[24] X. Pan, Q. Ruan, and Y. Wang, Palmprint recognition using
fusion of local and global features, in Proceedings of the
International Conference on Intelligence Signal Processing and
Communication System, pp. 642645, Xiamen, China, November 2007.
11
[25] S. B. Nikam, P. Goel, R. Tapadar, and S. Agarwal, Combining
gabor local texture pattern and wavelet global features for fngerprint matching, in Proceedings of the International Conference on Computational Intelligence and Multimedia Applications
(ICCIMA 07), pp. 409416, Washington, DC, USA, 2007.
[26] G. S. Badrinath and P. Gupta, Stockwell transform based palmprint recognition, Applied Sof Computing, vol. 11, no. 7, pp.
42674281, 2011.
[27] PolyU Palmprint Database, Te Hong Kong Polytechnic
University, July 2010, http://www4.comp.polyu.edu.hk/biometrics.
[28] COEP Palm Print Database (College of Engineering Pune),
http://www.coep.org.in/.
[29] Palmprint Image Database version 1, IIT, Delhi, India, 2007,
http://www4.comp.polyu.edu.hk/csajaykr/IITD/Database
Palm.htm.

http://dx.doi.org/10.1155/2014/918376
Research Article
A Comparison of Evolutionary Computation Techniques
for IIR Model Identification
Erik Cuevas,1,2 Jorge Glvez,1 Salvador Hinojosa,1 Omar Avalos,1
Daniel Zaldvar,1,2 and Marco Prez-Cisneros3
1
Departamento de Electronica, Universidad de Guadalajara, CUCEI, Avenida Revolucion 1500, 44430 Guadalajara, JAL, Mexico
Centro Universitario Azteca, Unidad de Investigacion, Avenida Juarez 340, 44280 Guadalajara, JAL, Mexico
3
CUTONALA, Avenida Nuevo Periferico 555, Ejido San Jose Tateposco, 48525 Tonala, JAL, Mexico
2
Correspondence should be addressed to Erik Cuevas; erik.cuevas@cucei.udg.mx

Received 18 August 2014; Accepted 30 September 2014; Published 25 December 2014
Academic Editor: Xin-She Yang
Copyright 2014 Erik Cuevas et al. Tis is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
System identifcation is a complex optimization problem which has recently attracted the attention in the feld of science and
engineering. In particular, the use of infnite impulse response (IIR) models for identifcation is preferred over their equivalent
FIR (fnite impulse response) models since the former yield more accurate models of physical plants for real world applications.
However, IIR structures tend to produce multimodal error surfaces whose cost functions are signifcantly difcult to minimize.
Evolutionary computation techniques (ECT) are used to estimate the solution to complex optimization problems. Tey are ofen
designed to meet the requirements of particular problems because no single optimization algorithm can solve all problems
competitively. Terefore, when new algorithms are proposed, their relative efcacies must be appropriately evaluated. Several
comparisons among ECT have been reported in the literature. Nevertheless, they sufer from one limitation: their conclusions are
based on the performance of popular evolutionary approaches over a set of synthetic functions with exact solutions and well-known
behaviors, without considering the application context or including recent developments. Tis study presents the comparison of
various evolutionary computation optimization techniques applied to IIR model identifcation. Results over several models are
presented and statistically validated.
1. Introduction
System identifcation is a complex optimization problem
which has recently attracted the attention in the feld of
science and engineering. System identifcation is important
in the disciplines of control systems [1], communication [2],
signal processing [3], and image processing [4].
In a system identifcation confguration, an optimization
algorithm attempts to iteratively determine the adaptive
model parameters to get an optimal model for an unknown
plant by minimizing some error function between the output
of the candidate model and the output of the plant. Te optimal model or solution is attained when such error function
is efectively reduced. Te adequacy of the estimated model
depends on the adaptive model structure, the optimization
algorithm, and also the characteristic and quality of the inputoutput data [5].
Systems or plants can be better modeled through infnite

impulse response (IIR) models because they emulate physical
plants more accurately than their equivalent FIR (fnite
impulse response) models [6]. In addition, IIR models are
typically capable of meeting performance specifcations using
fewer model parameters. However, IIR structures tend to
produce multimodal error surfaces whose cost functions
are signifcantly difcult to minimize [7]. Hence, in order
to identify IIR models, a practical, efcient, and robust
global optimization algorithm is necessary to minimize the
multimodal error function.
Traditionally, the least mean square (LMS) technique and
its variants [8] have been extensively used as optimization
tools for IIR model identifcation. Te wide acceptance of
such gradient based optimization techniques is due to the low
complexity and simplicity of implementation. However, the
13
error surface for the IIR model is mostly multimodal with
respect to the flter coefcients. Tis may result in leading
traditional gradient-descent approaches into local optima [9].
Te difculties associated with the use of gradient based
optimization methods for solving several engineering problems have contributed to the development of alternative
solutions. Evolutionary computation techniques (ECT) such
as the particle swarm optimization (PSO) [10], artifcial bee
colony (ABC) [11], electromagnetism-like method (EM) [12],
cuckoo search (CS) [13], and fower pollination algorithm
(FPA) [14] have received much attention regarding their
potential as global optimization methods in real-world applications. Inspired by the evolution process and survival of
the fttest in the biological world, ECT are search methods
that are diferent from traditional optimization methods.
Tey are based on a collective learning process within a
population of candidate solutions. Te population in ECT is
usually arbitrarily initialized, and each iteration (also called
a generation) evolves towards better and better solution
regions by means of randomized processes where several
operators are applied to each candidate solution. ECT have
been applied to many engineering optimization problems and
have proven to be efective for solving some specifc problems,
including multimodal optimization, dynamic optimization,
noisy optimization, and multiobjective optimization [1517].
Hence, they are becoming increasingly popular tools to solve
various hard optimization problems.
As an alternative to gradient based techniques, the
problem of IIR modelling has also been handled through
evolutionary computation techniques. In general, they have
been demonstrated to yield better results than those based on
gradient algorithms with respect to accuracy and robustness
[9]. Such approaches have produced several robust IIR identifcation systems by using diferent evolutionary computation
techniques such as PSO [18], ABC [19], EM [20], and CS [21],
whose results have been individually reported.
ECT are ofen designed to meet the requirements of
particular problems because no single optimization algorithm can solve all problems competitively [22]. Terefore,
when new alternative algorithms are proposed, their relative
efciency must be appropriately evaluated. Many eforts [23
25] have also been devoted to comparing ECT to each other.
Typically, such comparisons have been based on synthetic
numerical benchmark problems with most studies verifying
if one algorithm outperforms others over a given set of
benchmarks functions overlooking any statistical test. However, few comparative studies of various ECT considering the
application context are available in the literature. Terefore, it
is very important to discuss and compare the performance of
ECT methods from an application point of view.
Tis paper presents the comparison of various evolutionary computation optimization techniques that are applied
to IIR model identifcation. In the comparison, special
attention is paid to recently developed algorithms such as
the cuckoo search (CS) and the fower pollination algorithm
(FPA), including also popular approaches as the particle
swarm optimization (PSO), the artifcial bee colony (ABC)
optimization, and the electromagnetism-like optimization
(EM) algorithm. Results over several models with diferent

ranges of complexity are presented and validated within a
statistically signifcant framework.
Te rest of this paper is organized as follows: Section 2
presents a review of the evolutionary computation techniques
that are employed in the comparison whereas Section 3
discusses the IIR system identifcation problem. In Section 4
all experimental results are depicted with some concluding
remarks being drawn in Section 5.
2. Evolutionary Computation
Techniques (ECT)
In the real world, many optimization problems can be
considered as black box challenges. Ofen, less information
is available about an optimization problem itself unless the
information emerges from function evaluations. In the worst
case, nothing is known about the characteristics of the ftness
function, for example, whether it is unimodal or multimodal.
On the other hand, ECT are used to estimate the solution
to complex optimization problems since they adapt easily to
black-box formulations and extremely ill-behaved functions.
ECT are based on a collective learning process within a
population of candidate solutions. Te population in ECT
is usually arbitrarily initialized while each iteration (also
called a generation) evolves towards better solution regions
by means of randomized processes with several operators
being applied to each candidate solution. ECT have been
applied to many engineering optimization problems ensuring
an efective solution for some specifc problems, including
multimodal optimization, dynamic optimization, noisy optimization, multiobjective optimization, and others [1517].
Terefore, ECT are becoming increasingly popular tools
to solve various hard optimization problems. Tis section
presents a brief description of fve evolutionary computation
techniques: swarm optimization (PSO), artifcial bee colony
(ABC) optimization and electromagnetism-like optimization
(EM), cuckoo search (CS), and fower pollination algorithm
(FPA), which have been all employed in our comparative
study.
2.1. Particle Swarm Optimization (PSO). PSO, proposed by
Kennedy and Eberhart in 1995 [10], is a population-based
stochastic optimization technique that is inspired on the
social behavior of bird focking or fsh schooling. Te algorithm searches for the optimum using a group or swarm
formed by possible solutions of the problem, which are
called particles. From the implementation point of view, in
the PSO operation, a population P ({p1 , p2 , . . . , p}) of
particles (individuals) evolves from the initial point ( =
0) to a total gen number of iterations ( = gen). Each
particle p ( [1, . . . , ]) represents a -dimensional vector
{,1
, ,2
, . . . , ,
} where each dimension corresponds to a
decision variable of the optimization problem at hand. Te
quality of each particle p (candidate solution) is evaluated
by using an objective function (p ) whose fnal result
represents the ftness value of p . During the evolution
process, the best global position g (1 , 2 , . . . ) seen so far
14
is stored with the best position p (,1

, ,2
, . . . , ,
) being
reached by each particle. Such positions are computed by
considering a minimization problem as follows:
g=
arg min
{1,2,...,},{1,2,...,}
( (p )) ,
= arg min ( (p )) .
(1)
{1,2,...,}
In this work, the modifed PSO version proposed by Lin et

al. in [26] has been implemented. Under such approach, the
of each particle p is calculated by using
new position p+1
the following equations:

+1
= V,
+ 1 1 (,
,
) + 2 2 ( ,
);
V,
+1
+1
= ,
+ V,
,
,
(2)
where is called the inertia weight that controls the

impact of the current velocity on the updated velocity
( [1, . . . , ], [1, . . . , ]). 1 and 2 are the positive
acceleration coefcients that rule the movement of each
particle towards the positions g and p , respectively. 1 and
2 are uniformly distributed random numbers that are chosen
within the interval [0, 1].
2.2. Artifcial Bee Colony (ABC). Te artifcial bee colony
(ABC) algorithm, proposed by Karaboga [11], is an ECT
inspired by the intelligent foraging behavior of a honeybee swarm. In the ABC operation, a population L
({l1 , l2 , . . . , l}) of food locations (individuals) is evolved
from the initial point ( = 0) to a total gen number of
iterations ( = gen). Each food location l ( [1, . . . , ])

, ,2 , . . . , ,
} where
represents a -dimensional vector {,1
each dimension corresponds to a decision variable of the
optimization problem to be solved. Afer initialization, an
objective function evaluates each food location to assess
whether it represent an acceptable solution (nectar-amount)
or not. Guided by the values of such an objective function,
the candidate solution l is evolved through diferent ABC
operations (honeybee types). In the main operator, each
food location l generates a new food source t in the
neighborhood of its present position as follows:
t = l + (l l ) ,
, (1, 2, . . . , ) ,
(3)
where l is a randomly chosen food location, satisfying the

condition = . Te scale factor is a random number
between [1, 1]. Once a new solution t is generated, a
ftness value representing the proftability associated with
a particular solution ft(l ) is calculated. Te ftness value
for a minimization problem can be assigned to a candidate
solution l by the following expression:
ft (l )
1
,
{
1
+
(l )
={
{1 + abs ( (l )) ,
if (l ) 0,
if (l ) < 0,
(4)
where () represents the objective function to be minimized.

Once the ftness values are calculated, a greedy selection
process is applied between t and l . If ft(t ) is better
than ft(l ), then the candidate solution l is replaced by t ;
otherwise, l remains.
2.3. Electromagnetism-Like (EM) Algorithm. Te EM algorithm, proposed by Ilker et al. [12] is a simple and populationbased search algorithm which has been inspired by the
electromagnetism phenomenon. In EM, individuals emulate
charged particles which interact to each other based on
the electromagnetism laws of repulsion and attraction. Te
method utilizes , -dimensional points x , = 1, 2, . . . , ,
where each point x is a -dimensional vector containing
, . . . , ,
})
the parameter values to be optimized (x = {,1
whereas denotes the iteration (or generation) number. Te
initial population X = {x1 , x2 , . . . , x

} (being = 0) is
taken from uniformly distributed samples of the search space.
We denote the population set at the th generation by X ,
because members of X change with . Afer the initialization
of X0 , EM continues its iterative process until a stopping
condition (e.g., the maximum number of generations, =
gen) is met. An iteration of EM consists of three steps.
In the frst step each point in X moves to a diferent
location by using the attraction-repulsion mechanism of the
electromagnetism theory. In the second step, points moved by
the electromagnetism principle are further moved locally by
a local search procedure. Finally, in the third step, in order to
generate the new population X+1 , a greedy selection process
selects best points between those produced by the local search
procedure and the originals. Both the attraction-repulsion
mechanism and the local search in EM are responsible for
driving the members x of X to the proximity of the global
optimum.
2.4. Cuckoo Search (CS) Method. CS is one of the latest
nature-inspired algorithms that has been developed by Yang
and Deb [13]. CS is based on the brood parasitism of some
cuckoo species. In addition, this algorithm is enhanced by
the so-called Levy fights [27], rather than by simple isotropic
random walks. From the implementation point of view of
the CS operation, a population E ({e1 , e2 , . . . , e}) of
eggs (individuals) is evolved from the initial point ( = 0)
to a total gen number of iterations ( = 2 gen). Each
egg e ( [1, . . . , ]) represents a -dimensional vector
, ,2
, . . . , ,
{,1
decision variable of the optimization problem to be solved.
Te quality of each egg e (candidate solution) is evaluated
by using an objective function (e ) whose fnal result
represents the ftness value of e . Tree diferent operators
defne the evolution process of CS: (A) Levy fight, (B) the
replacing of nests operator for constructing new solutions,
and (C) the elitist selection strategy.
(A) Te Levy Flight. One of the most powerful features
of cuckoo search is the use of Levy fights to generate
15
new candidate solutions (eggs). Under this approach, a new

candidate solution e+1
( [1, . . . , ]) is produced by
perturbing the current e with a change of position c . In order

to obtain c , a random step s is generated by a symmetric Levy
distribution. For producing s , Mantegnas algorithm [28] is
employed as follows:
s =
|k|1/
(5)
where u ({1 , . . . , }) and k ({V1 , . . . , V }) are -dimensional

vectors and = 3/2. Each element of u and k is calculated by
considering the following normal distributions:
(0, 2 ) ,
= (
V (0, V2 ) ,
(1 + ) sin( /2)
)
((1 + )/2) 2(1)/2
1/
V = 1,
(6)
where () represents the Gamma distribution. Once s

has been calculated, the required change of position c is
computed as follows:
c = 0.01 s (e ebest ) ,
(7)
where the product denotes entrywise multiplications

whereas ebest is the best solution (egg) seen so far in terms
of its ftness value. Finally, the new candidate solution e+1
is
calculated by using
= e + c .
e+1
(8)
(B) Replacing Some Nests by Constructing New Solutions.

Under this operation, a set of individuals (eggs) are probabilistically selected and replaced with a new value. Each
individual e ( [1, . . . , ]) can be selected with a
probability [0, 1]. In order to implement this operation,
a uniform random number 1 is generated within the range
[0, 1]. If 1 is less than , the individual e is selected and
modifed according to (5); otherwise e remains with no
change. Tis operation can be resumed by the following
model:
={
e+1
e + rand (e e ) ,
e ,
with probability ,
with probability (1 ) ,
(9)
where rand is a random number normally distributed

whereas and are random integers from 1 to .
either
(C) Te Elitist Selection Strategy. Afer producing e+1
by the operator A or by the operator B, it must be compared

is better than
with its past value e . If the ftness value of e+1
e , then e+1
is
accepted
as
the
fnal
solution;
otherwise,
e
is retained. Tis procedure can be resumed by the following

statement:
e+1
={
e+1
,
e ,
if (e+1
) < (e ) ,
otherwise.
(10)
Te elitist selection strategy denotes that only high-quality

eggs (best solutions near to the optimal value) which are the
most similar to the host birds eggs have the opportunity to
develop (next generation) and become mature cuckoos.
2.5. Flower Pollination Algorithm (FPA). Te fower pollination algorithm (FPA), proposed by Yang [14], is an
ECT inspired by the pollination process of fowers. In
FPA, individuals emulate a set of fowers or pollen gametes
which behaves based on biological laws of the pollination
process. From the implementation point of view, in the
}) of fower
FPA operation, a population F ({f1 , f2 , . . . , f
positions (individuals) is evolved from the initial point ( =
0) to a total gen number of iterations ( = gen). Each
fower f ( [1, . . . , ]) represents a -dimensional vector
{,1
, ,2
, . . . , ,
decision variable of the optimization problem to be solved.
In FPA, a new population F+1 is produced by considering
two operators: local and global pollination. A probabilistic
global pollination factor is associated with such operators.
In order to decide which operator should be applied to each
current fower position f , a uniform random number is
generated within the range [0, 1]. If is less than , the
global pollination operator is applied to f . Otherwise, the
local pollination operator is considered.
Global Pollination Operator. Under this operator, the original
position f is displaced to a new position f+1 according to
the following model:
f+1 = f + (f g) ,
(11)
where g is the global best position seen so far whereas

controls the length of the displacement. Te value is
generated by a symmetric Levy distribution according to (5)(6).
Local Pollination Operator. In the local pollination operator,
the current position f is perturbed to a new position f+1 as
follows:
f+1 = f + (f f ) ;
, , (1, 2, . . . , ) ,
(12)
where f and f are two randomly chosen fower positions,

satisfying the condition = = . Te scale factor is a
random number between [1, 1].
3. IIR Model Identification

(Problem Formulation)
System identifcation is the mathematical representation of
an unknown system by using only input-output data. In a
system identifcation confguration, an optimization algorithm attempts to iteratively determine the adaptive model
parameters to get an optimal model for the unknown plant
based on minimizing some error function between the output
of the candidate model and the actual output of the plant.
16
d(t) +
Plant
4. Experimental Results
e(t)
In the comparison study, a comprehensive set of experiments

has been used to test the performance of each evolutionary
computation technique. Te set considers the use of IIR
models with diferent orders. Such experimental set has been
carefully selected to assure compatibility between similar
works reported in the literature [1821]. In the comparison,
fve ETC have been considered: PSO, ABC, EM, CS, and FPA.
Te parameter setting for each evolutionary computation
algorithm that is used in the comparison is described as
follows.
u(t)
y(t)
IIR model
Evolutionary computation
technique
Figure 1: Adaptive IIR model for system identifcation.
Te use of infnite impulse response (IIR) models for

identifcation is preferred over their equivalent FIR (fnite
impulse response) models since the former produce more
accurate models of physical plants for real world applications
[6]. In addition, IIR models are typically capable of meeting
performance specifcations using fewer model parameters.
Figure 1 represents an IIR identifcation model of any arbitrary system.
An IIR system can be represented by the transfer function:
() 0 + 1 1 + 2 2 + +
,
=
()
1 + 1 1 + 2 2 + +
(13)
where and are the number of numerator and denominator coefcients of the transfer function, respectively, and,
and are the pole and zero parameters of the IIR model
( [1, . . . , ], [1, . . . , ]). Equation (13) can be written as
diference equation of the form:
=1
=0
() = ( ) + ( ) ,
(14)
where () and () represent the th input and output

of the system, respectively. Terefore, the set of unknown
parameters that models the IIR system is represented by
= {1 , . . . , , 0 , . . . , }. Considering that the number of
unknown parameters of is ( + + 1), the search space S of
feasible values for is R(++1) .
According to the block diagram of Figure 1, the output of
the plant is () whereas the output of the IIR flter is (). Te
output diference between the actual system and its model
yields the error () = () (). Hence, the problem of
IIR model identifcation can be considered as a minimization
problem of the function () stated as the following:
() =
1
2
( () ()) ,
=1
(15)
where is the number of samples used in the simulation.

Te aim is to minimize the cost function () by adjusting
. Te optimal model or solution is attained when the error
function () reaches its minimum value, as follows:
= arg min ( ()) .
S
(16)
(1) PSO: the parameters are set to 1 = 2, 2 = 2; besides,

the weight factor decreases linearly from 0.9 to 0.2
[18].
(2) ABC: the algorithm has been implemented using the
guidelines provided by its own reference [19], using
the parameter = 100.
(3) EM: particle number = 50, = 0.001, = 4,
= 300. Such values, according to [12, 20]
represent the best possible confguration.
(4) CS: according to [13, 21], the parameters are set to =

0.25 and the number of generations gen = 500.
(5) FPA: the probabilistic global pollination factor is set

to 0.8. Under such value, the algorithm presents the
best performance according to [14].
For all algorithms, the population size has been set to 25

( = 25) whereas the maximum iteration number has been
confgured to 3000 generations (gen = 3000).
Te results are divided into two sections. In the frst set,
the performance of each ETC for each identifcation experiment is presented. In the second set, the results are analyzed
from a statistical point of view by using the Wilcoxon test.
4.1. IIR Model Identifcation Results. Te results are reported
considering three experiments that include (1) a secondorder plant with a frst-order IIR model; (2) a second-order
plant with a second-order IIR model; and fnally, (3) a highorder plant with a high-order model. Each case is discussed
below.
(1) A Plant with a Second-Order System and a First-Order IIR
Model (First Experiment). In this experiment, each algorithm
is applied to identify a second-order plant through a frstorder IIR model. Under such conditions, the unknown plant
and the IIR model hold the following transfer
functions:
(1 ) =
0.05 0.41
,
1 1.13141 + 0.252
.
( ) =
1 1
1
(17)
In the simulations, it has been considered a white sequence

of 100 samples ( = 100) for the input (). Since a reduced
order model is employed to identify a plant of a superior
17
5
4 4
3
0.5
8
6
f()
0
4
0.5
2
1
0
b
1
0.5
1
1
1
0.5
0.5
0.5
(a)
(b)
Figure 2: Multimodal error surface () for the frst experiment: (a) tridimensional fgure and (b) contour.
order, () is multimodal [19]. Te error surface () is

shown in Figure 2.
Te performance evaluation over 30 diferent executions
is reported in Table 1 considering the following indexes: the
best parameter values (ABP), the average () value (AV),
and the standard deviation (SD). Te best parameter values
(ABP) report the best model parameters obtained during the
30 executions while the average () value (AV) indicates
the average minimum value of (), considering the same
number of executions. Finally, the standard deviation (SD)
reports the dispersion from the average () value regarding
30 executions.
According to Table 1, the CS algorithm provides better
results than PSO, ABC, and EM. In particular, the results
show that CS maintains a considerable precision (the lowest AV value) and more robustness (smallest SD value).
Nevertheless, the CS performance is similar to the FPA
algorithm. On the other hand, the worst performance is
reached by the PSO algorithm. Such a fact corresponds to its
difculty (premature convergence) to overcome local minima
in multimodal functions.
(2) A Plant with Second-Order System and Second-Order
IIR Model (Second Experiment). In the second experiment,
the performance for each algorithm is evaluated at the
identifcation of a second-order plant through a second-order
IIR model. Terefore, the unknown plant and the IIR
model hold the following transfer functions:
1
(1 ) =
,
1
1 1.4 + 0.492
(1 ) =
1 + 1
+ 2
.
2
(18)
For the simulations, the input () that is applied to the system

and to the IIR model simultaneously has been confgured as a
white sequence with 100 samples. Since the order of the model
is equal to the order of the to-be-identifed system ,
only one global minimum exists in () [19]. Te results of
this experiment over 30 diferent executions are reported in
Table 2.
Table 1: Performance results of the frst experiment.

Algorithms
PSO
ABC
EM
CS
FPA
0.9125
0.1420
0.9034
0.9173
0.9364
ABP
0.3012
0.3525
0.3030
0.2382
0.2001
AV
SD
0.0284
0.0197
0.0165
0.0101
0.0105
0.0105
0.0015
0.0012
3.118 004
5.103 004
Te results in Table 2 show that PSO, ABC, EM, CS,

and FPA have similar values in their performance. Te
evidence shows that evolutionary algorithms maintain a
similar average performance when they face unimodal lowdimensional functions [29, 30]. In particular, the test remarks
that the small diference in performance is directly related to
a better exploitation mechanism included in CS and FPA.
(3) A Superior-Order Plant and a High-Order Model (Tird
Experiment). Finally, the performance for each algorithm
is evaluated at the identifcation of a superior-order plant
through a high-order IIR model. Terefore, the unknown
plant and the IIR model hold the following transfer
functions:
(1 ) =
1 0.42 0.654 + 0.266

,
1 0.772 0.84984 + 0.64866
+ 1 + 2 2 + 3 3 + 4 4
( ) = 0 1 1
.
1 + 1 + 2 2 + 3 3 + 4 4
1
(19)
Since the plant is a sixth-order system and the IIR model a

fourth-order system, the error surface () is multimodal just
as it is in the frst experiment. A white sequence with 100
samples has been used as input. Te results of this experiment
over 30 diferent executions are reported in Tables 3 and 4.
Table 3 presents the best parameter values (ABP) whereas
Table 4 shows the average () value (AV) and its standard
deviation (SD).
According to the AV and SD indexes in Table 4, the CS
algorithm fnds better results than PSO, ABC, EM, and FPA.
18
Table 2: Performance results of the second experiment.
ABP
2
0.4925
0.6850
0.4802
0.4900
0.4900
1
1.4024
1.2138
1.0301
1.400
1.400
Algorithms
PSO
ABC
EM
CS
FPA
0.9706
0.2736
1.0091
1.000
1.000
AV
SD
4.0035 005
0.3584
3.9648 005
0.000
4.6246 32
1.3970 005
0.1987
8.7077 005
0.000
2.7360 31
Table 3: Te best parameter values (ABP) for the second experiment.

Algorithms
PSO
ABC
EM
CS
FPA
1
0.3683
1.1634
0.4950
0.9599
0.0328
2
0.7043
0.6354
0.7049
0.0248
0.1059
3
0.2807
1.5182
0.5656
0.0368
0.0243
4
0.3818
0.6923
0.2691
0.0002
0.7619
Table 4: Te average () value (AV) and the standard deviation

(SD).
Algorithms
PSO
ABC
EM
CS
FPA
AV
5.8843
7.3067
0.0140
6.7515 004
0.0018
SD
3.4812
4.3194
0.0064
4.1451 004
0.0020
Table 5: -values produced by Wilcoxons test comparing CS vs

PSO, ABC, EM and FPA over the Te average () values (AV)
from Tables 1, 2 and 4.
CS vs
First
experiment
Second
experiment
Tird
experiment
PSO
ABC
EM
FPA
6.5455 13 8.4673 13 3.8593 08 0.7870

1.5346 14 1.5346 14 1.5346 14 0.3313
6.5455 13 1.5346 14 4.3234 13
0.1011
ABP
0
0.9939
0.5214
1.0335
0.2377
1.0171
1
0.6601
1.2703
0.6670
0.0031
0.0038
2
0.8520
0.3520
0.4682
0.3579
0.2374
3
0.2275
1.1816
0.6961
0.0011
0.0259
4
1.4990
1.9411
0.0673
0.5330
0.3365
produced by Wilcoxons test for the pairwise comparison of

the the average () value of four groups. Such groups are
formed by CS versus PSO, CS versus ABC, CS versus EM, and
CS versus FPA. As a null hypothesis, it is assumed that there is
no signifcant diference between averaged values of the two
algorithms. Te alternative hypothesis considers a signifcant
diference between the AV values of both approaches.
For the case of PSO, ABC, and EM, all values reported
in Table 5 are less than 0.05 (5% signifcance level) which is a
strong evidence against the null hypothesis. Terefore, such
evidence indicates that CS results are statistically signifcant
and that it has not occurred by coincidence (i.e., due to
common noise contained in the process). On the other hand,
since the values for the case of CS versus FPA are more
than 0.05, there is not statistical diference between both.
Terefore, it can be concluded that the CS algorithm is better
than PSO, ABC, and EM in the application of IIR modeling
for system identifcation. However, CS presents the same
performance as FPA and therefore there is not statistical
evidence that CS surpasses the FPA algorithm.
5. Conclusions
Te results show that CS presents better precision (AV value)
and robustness (SD value). Tese results also indicate that CS,
FPA, and EM are able to identify the sixth-order plant under
diferent accuracy levels. On the other hand, PSO and ABC
obtain suboptimal solutions whose parameters weakly model
the unknown system.
4.2. Statistical Analysis. In order to statistically validate the
results, a nonparametric statistical signifcance-proof which
is known as Wilcoxons rank sum test for independent samples [31, 32] has been conducted over the the average ()
value (AV) data of Tables 1, 2, and 4 with a 5% signifcance
level. Te test has been conducted considering 30 diferent
executions for each algorithm. Table 5 reports the values
Tis paper presents a comparison study between fve evolutionary algorithms for the IIR-based model identifcation.
Under this research, the identifcation task is considered as an
optimization problem. In the comparison, special attention
is paid to recently developed algorithms such as the cuckoo
search (CS) and the fower pollination algorithm (FPA), also
including popular approaches such as the particle swarm
optimization (PSO), the artifcial Bee colony optimization
(ABC), and the electromagnetism-like (EM) optimization
algorithm.
Te comparison has been experimentally evaluated over a
test suite of three benchmark experiments that produce multimodal functions. Te experiment results have demonstrated
that CS outperforms PSO, ABC, and EM in terms of both the
19
accuracy (AV values) and robustness (SD values), within a
statistically signifcant framework (Wilcoxon test). However,
there is not statistical evidence that CS surpasses the FPA
performance.
Te remarkable performance of CS and FPA is explained
by two diferent features: (i) operators (such as Levy fight)
that allow a better exploration of the search space, increasing
the capacity to fnd multiple optima, and (ii) their exploitation operators that allow a better precision of previously
found solutions.
References
[1] X. Zhou, C. Yang, and W. Gui, Nonlinear system identifcation
and control using state transition algorithm, Applied Mathematics and Computation, vol. 226, pp. 169179, 2014.
[2] M. Albaghdadi, B. Briley, and M. Evens, Event storm detection
and identifcation in communication systems, Reliability Engineering and System Safety, vol. 91, no. 5, pp. 602613, 2006.
[3] P. Frank Pai, B.-A. Nguyen, and M. J. Sundaresan, Nonlinearity
identifcation by time-domain-only signal processing, International Journal of Non-Linear Mechanics, vol. 54, pp. 8598, 2013.
[4] H.-C. Chung, J. Liang, S. Kushiyama, and M. Shinozuka,
Digital image processing for non-linear system identifcation,
International Journal of Non-Linear Mechanics, vol. 39, no. 5, pp.
691707, 2004.
[5] J. Na, X. Ren, and Y. Xia, Adaptive parameter identifcation
of linear SISO systems with unknown time-delay, Systems &
Control Letters, vol. 66, no. 1, pp. 4350, 2014.
[6] O. Kukrer, Analysis of the dynamics of a memoryless nonlinear
gradient IIR adaptive notch flter, Signal Processing, vol. 91, no.
10, pp. 23792394, 2011.
[7] T. Mostajabi, J. Poshtan, and Z. Mostajabi, IIR model identifcation via evolutionary algorithmsa comparative study,
Artifcial Intelligence Review, 2013.
[8] C. Dai, W. Chen, and Y. Zhu, Seeker optimization algorithm
for digital IIR flter design, IEEE Transactions on Industrial
Electronics, vol. 57, no. 5, pp. 17101718, 2010.
[9] W. Fang, J. Sun, and W. Xu, A new mutated quantum-behaved
particle swarm optimizer for digital IIR flter design, Eurasip
Journal on Advances in Signal Processing, vol. 2009, Article ID
367465, 7 pages, 2009.
[10] J. Kennedy and R. C. Eberhart, Particle swarm optimization,
in Proceedings of the IEEE International Conference on Neural
Networks, vol. 4, pp. 19421948, December 1995.
[11] D. Karaboga, An idea based on honey bee swarm for numerical optimization, Tech. Rep. TR06, Computer Engineering
Department, Engineering Faculty, Erciyes University, 2005.
[12] B. Ilker, S. Birbil, and F. Shu-Cherng, An electromagnetismlike mechanism for global optimization, Journal of Global
Optimization, vol. 25, no. 3, pp. 263282, 2003.
[13] X.-S. Yang and S. Deb, Cuckoo search via Levy fights, in
Proceedings of the World Congress on Nature and Biologically
Inspired Computing (NABIC 09), pp. 210214, December 2009.

[14] X. S. Yang, Flower pollination algorithm for global optimization, in Unconventional Computation and Natural Computation, vol. 7445 of Lecture Notes in Computer Science, pp. 240
249, Springer, Berlin, Germany, 2012.
[15] C. Ahn, Advances in Evolutionary Algorithms: Teory, Design
and Practice, Springer, New York, NY, USA, 2006.
[16] R. Chiong, T. Weise, and Z. Michalewicz, Variants of Evolutionary Algorithms for Real-World Applications, Springer, New York,
NY, USA, 2012.
[17] M. Oltean, Evolving evolutionary algorithms with patterns,
Sof Computing, vol. 11, no. 6, pp. 503518, 2007.
[18] S. Chen and B. L. Luk, Digital IIR flter design using particle
swarm optimisation, International Journal of Modelling, Identifcation and Control, vol. 9, no. 4, pp. 327335, 2010.
[19] N. Karaboga, A new design method based on artifcial bee
colony algorithm for digital IIR flters, Journal of the Franklin
Institute, vol. 346, no. 4, pp. 328348, 2009.
[20] E. Cuevas and D. Oliva, IIR flter modeling using an algorithm
inspired on electromagnetism, Ingeniera Investigacion y Tecnologa, vol. 14, no. 1, pp. 125138, 2013.
[21] A. P. Patwardhan, R. Patidar, and N. V. George, On a cuckoo
search optimization approach towards feedback system identifcation, Digital Signal Processing, vol. 32, pp. 156163, 2014.
[22] D. H. Wolpert and W. G. Macready, No free lunch theorems for
optimization, IEEE Transactions on Evolutionary Computation,
vol. 1, no. 1, pp. 6782, 1997.
[23] E. Elbeltagi, T. Hegazy, and D. Grierson, Comparison among
fve evolutionary-based optimization algorithms, Advanced
Engineering Informatics, vol. 19, no. 1, pp. 4353, 2005.
[24] D. Shilane, J. Martikainen, S. Dudoit, and S. J. Ovaska, A
general framework for statistical performance comparison of
evolutionary computation algorithms, Information Sciences,
vol. 178, no. 14, pp. 28702879, 2008.
[25] V. Osuna-Enciso, E. Cuevas, and H. Sossa, A comparison of
nature inspired algorithms for multi-threshold image segmentation, Expert Systems with Applications, vol. 40, no. 4, pp. 1213
1219, 2013.
[26] Y.-L. Lin, W.-D. Chang, and J.-G. Hsieh, A particle swarm
optimization approach to nonlinear rational flter modeling,
Expert Systems with Applications, vol. 34, no. 2, pp. 11941199,
2008.
[27] I. Pavlyukevich, Levy fights, non-local search and simulated
annealing, Journal of Computational Physics, vol. 226, no. 2, pp.
18301844, 2007.
[28] R. N. Mantegna, Fast, accurate algorithm for numerical simulation of Levy stable stochastic processes, Physical Review E,
vol. 49, no. 5, pp. 46774683, 2007.
[29] E. Cuevas, M. Gonzalez, D. Zaldivar, M. Perez-Cisneros, and
G. Garca, An algorithm for global optimization inspired by
collective animal behavior, Discrete Dynamics in Nature and
Society, vol. 2012, Article ID 638275, 24 pages, 2012.
[30] E. Cuevas, M. Cienfuegos, D. Zaldvar, and M. Perez-Cisneros,
A swarm optimization algorithm inspired in the behavior of
the social-spider, Expert Systems with Applications, vol. 40, no.
16, pp. 63746384, 2013.
[31] S. Garca, D. Molina, M. Lozano, and F. Herrera, A study on
the use of non-parametric tests for analyzing the evolutionary
algorithms behaviour: a case study on the CEC2005 special
session on real parameter optimization, Journal of Heuristics,
vol. 15, no. 6, pp. 617644, 2009.

[32] D. Shilane, J. Martikainen, S. Dudoit, and S. J. Ovaska, A
general framework for statistical performance comparison of
evolutionary computation algorithms, Information Sciences,
vol. 178, no. 14, pp. 28702879, 2008.
20

http://dx.doi.org/10.1155/2014/918376
Research Article
Analysis of Repairable Geo/G/1 Queues with Negative Customers
Doo Ho Lee1 and Kilhwan Kim2
1
2
Sofware Contents Research Lab, ETRI, Daejeon 305-700, Republic of Korea

Department of Management Engineering, Sangmyung University, Chungnam 330-720, Republic of Korea
Correspondence should be addressed to Doo Ho Lee; enjdhlee@gmail.com

Received 3 September 2014; Accepted 8 December 2014; Published 24 December 2014
Academic Editor: Kannan Krithivasan
Copyright 2014 D. H. Lee and K. Kim. Tis is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
We consider discrete-time Geo/G/1 queues with negative customers and a repairable server. Te server is subject to failure due
to a negative customer arrival. As soon as a negative customer arrives at a system, the server fails and one positive (ordinary)
customer is forced to leave. At a failure instant, the server is turned of and the repair process immediately begins. We construct the
mathematical model and present the probability generating functions of the system size distribution and the FCFS sojourn time
distribution. Finally, some numerical examples are given to show the infuence of negative customer arrival on the performance
measures of the system.
1. Introduction
Afer the introduction in the work of Gelenbe [1], there has
been a rapid increase in the literature on queueing systems
with negative customers due to their applications to neural
networks, communication systems, and manufacturing systems. If a negative customer arrives at a queueing system, it
removes one ordinary customer (called a positive customer)
according to a predetermined removal discipline. Tere are
two typical removal disciplines: (i) a negative customer
removes the customer being served (RCH); (ii) a negative
customer removes the customer who arrived most recently
(RCE). Te negative customers cannot accumulate in the
queue and do not receive services. Te negative customers
have a rich and wide range of real applications such as job canceling signals accompanied by machine breakdowns [2], network routing control [3], and load balancing [4]. For example,
if a certain virus enters a computer system, the virus interrupts system operations, deletes fles, and infects other fles.
In telecommunication systems, the negative customers can be
used to model cell losses caused by the arrival of a corrupted
cell, when the preceding cells of a packet would be discarded.
Te concept of the negative customer has been applied to
the M/M/1 queue [5], the M/G/1 queue [6], and the GI/M/1
queue [7]. Excellent surveys on negative customers have been
provided by Artalejo [8], Do [9, 10], and Gelenbe [1113].
Recently, this topic has been extended to a discrete-time

queueing system. Note that the discrete-time queue is more
suitable to describe the operation of time-slotted digital
communication systems due to the packetized nature of
transport protocols. Atencia and Moreno [14] studied the
Geo/Geo/1 queues with negative customers and various
removal disciplines caused by the negative customers. Zhou
[15] investigated the discrete-time GI/G/1 queue with RCHtype negative customers. Wang and Zhang [16] considered
the Geo/Geo/1 retrial queue with negative customers and
an unreliable server. Park et al. [17] obtained the results
of the Geo/G/1 queue with negative customers and without
repair times. Recently, Chae et al. [18] extended Atencia and
Morenos [14] Geo/Geo/1 queues to the GI/Geo/1 queues,
where the interarrival times are generally distributed. Te
results of the queueing system with negative customers under
the assumption of a Markovian arrival process can be found
in Wu et al. [19, 20].
In most of the literature on queueing systems with negative customers, the server is assumed to be reliable on a permanent basis, regardless of the arrivals of negative customers.
However, in many practical systems, the arrival of a negative
customer can cause server breakdown as well as destruction
of work in a system. For example, when a computer server
is infected with a computer virus, a recovery time is needed
22
to restore the server. Similarly, when a canceling signal for the
job in progress arrives at a database server, a system cleaning
time such as a rollback time for incomplete data operation
may be needed. For this reason, we consider the Geo/G/1
queues with negative customers under the assumption that
the server is repairable. Since the performance measure of
such a system may be infuenced by breakdown and repair
time, the repairable system is well worth analyzing in view of
queueing theory along with reliability theory.
Tere are few researches on repairable queueing systems
with negative customers. Wang and Zhang [16] considered
the Geo/Geo/1 retrial queue with negative customers and an
unreliable server, where server breakdown occurs whenever
negative customers arrive at a system. Lee et al. [21] presented
the results of repairable Geo/G/1 queues with disasters, where
if a disaster arrives at a system, all present customers (i.e., a
customer in service plus customers in queue) are forced to
leave the system. In [21], the authors considered a disaster as
server failure which leads to the destruction of all work in
process in a system. However, neither of them considered the
case of repairable queueing systems with negative customers
which cause server breakdown as well as canceling a single
ordinary customer.
Systems under our study have following features. Positive
customers arrive at a single server queue, according to a
Bernoulli process. A server provides a service to each customer on an FCFS basis. Te service times are independent
and identically distributed (i.i.d.) random variables that are
generally distributed. Negative customers arrive when the
server is busy according to a Bernoulli process. Negative
customers do not have an efect on the system if the server
is idle or under repair. Each time a negative customer arrives
at the system while the server is busy, the server fails and the
customer in service leaves the system. At a failure epoch, the
server is turned of and a repair period immediately begins.
Te server repair times are i.i.d. random variables that are
generally distributed. During the repair period, the stream of
new arrivals continues. Depending on whether new arrivals
are allowed to enter the system during the repair time, there
can be one of the two following systems. In System 1, the
customers newly arriving during the repair period cannot
enter the system and are blocked. Tis is the case when a
virus-infected computer server is isolated from the network
during the repair time without any backup server. In System 2,
they join the queue and wait for the server to be repaired. Tis
is the case when a backup server just stores requests during
the repair time and passes them to the original server when
it is recovered. As soon as the repair period ends, the server
promptly becomes available. Both Systems 1 and 2 have RCH
discipline. For each system, using the probability generating
function (PGF) technique, we present PGFs of the system size
distribution and the sojourn time distribution.
In this paper, we frst describe the mathematical model in
Section 2. In Sections 3 and 4, general results on the system
size and the sojourn time of System 1 and System 2 are
presented. Section 5 deals with numerical experiments which
we conducted to investigate the infuence of the arrival of
negative customers on the mean system size of each system.
2. Model Description
Troughout this paper, we adopt the late arrival system (LAS)
[22]. Let the time axis be marked by = 0, 1, 2, . . .. According
to the LAS model, the potential arrival of a positive customer
takes place during the interval ( , ) and the potential service
completion occurs during the interval (, + ), where + and
represent lim 0 (+||) and lim 0 (||), respectively.
Since the arrivals of a positive customer and a negative customer can occur simultaneously at a slot boundary, the order
of these events must be stated. We assume that the potential
arrival of a negative customer occurs during the interval
( , ) and immediately before the potential arrival of a positive customer. We also assume that the potential completion
of a repair time occurs during the interval ( , ) and immediately before the potential arrival of a positive customer.
We further assume that a negative customer arrival and a
repair completion do not occur at the same slot boundary
simultaneously.
We defne commonly used notations to analyze both Systems 1 and 2. Interarrival times of positive customers { }
=1
are i.i.d. discrete random variables and follow a geometric
distribution with parameter :
Pr{ = } =
= 1 ,
[ ] = 1 .
1;
0 < < 1;
(1)
Interarrival times of negative customers { }

=1 are i.i.d.
discrete random variables and follow a geometric distribution
with parameter :
Pr{ = } = 1 ,
= 1 ,
[ ] = 1 .
1;
0 < < 1;
(2)
Service times { }
=1 are i.i.d. general discrete random variables. A distribution and its PGF are denoted by Pr{ = } =
( 1) and () =
=1 , respectively. Repair times
{ }=1 are i.i.d. general discrete random variables and have

the distribution Pr{ = } = ( 1) and its PGF
() =
=1 . We assume that { }=1 , { }=1 , { }=1 ,
and { }=1 are mutually independent.

Systems 1 and 2 are represented by a Markov chain. Let
(+ ) be the number of customers in the system at + . Let
(+ ) be the server state at + and be defned as follows:
(+ ) = {
0,
1,
Te server is under repair at + ,

Te server is available at + .
(3)
Ten, in both System 1 and System 2, {(+ ), (+ ), (+ ),

(+ ), = 0, 1, . . .} is a Markov chain, where the supplementary variables (+ ) and (+ ), respectively, represent
the remaining service time and the remaining repair time all
23
at + . One of our main purposes is to obtain the stationary

distribution:
= lim Pr{(+ ) = },
0.
() = lim Pr{( ) = , ( ) = 0, ( ) = },
() = lim Pr{(+ ) = , (+ ) = 1, (+ ) = },
=1
(5)
0,
and the normalizing condition is given by

0 + (1, 1) + (1, 1) = 1.
(11)
(, ) = (, + 1) + 1 (, 1).
(12)
(, )(1 1 ) = 1 () (, 1) (, 1).
= lim Pr{(+ ) = , (+ ) = 1} = (),

=1
(, ) =
(, ) = () ,
|| 1,
(, ) = () ,
|| 1,
=0
=1
(, ) = (, ) ,
|| 1,
(, ) = (, ) ,
|| 1.
=1
=1
(, 1) = []1 (, 1).
(15)
(16)
Multiplying (9) and (10) by and then summing over ,

1, together with (8), we obtain
(6)
(, )
= 1 (, + 1)
(17)
+ [0 (, 1) + 1 1 (, 1) 0 (1 )],
where 0 = + and 1 = ( + ). Multiplying (17) by

and summing over , 1, yield
(, )(1 1 1 )
3. System 1
= 1 1 (, 1)(() )
In System 1, newly arriving customers are blocked when the

server is under repair.
3.1. System Size Distribution for System 1. With the limiting
probabilities defned in Section 2, the Kolmogorov equations
for the stationary distribution are given by
() = ( + 1) + +1 ,
(() 1)
(, 1).
( 1)
Letting = 1 in (15), we obtain the following relationship by

the LHospital rule:
1.
To solve the Kolmogorov equations, we introduce PGFs:
(14)
Incorporating (14) back into (13) then yields
(13)
To obtain (, 1), we insert = 1 into (13). Tus,

(, 1) = 1 (, 1).
0 = lim Pr{(+ ) = 0, (+ ) = 1},
2,
Multiplying (12) by and summing over , 1, yield
1, 1,
= lim Pr{(+ ) = , (+ ) = 0} = (),
+ ( + 1) + +1 (1) ),
(10)
Multiplying (7) by and summing over , 0, we

obtain
0, 1,
+ (1 ( + 1) + (1)
(4)
Since we are interested in the steady-state behavior of the systems, we employ the following limiting probabilities throughout this paper to manipulate the Kolmogorov equations of the
Markov chain:
+
() = 1 (1) + (1)
0,
0 = (0 (1) + 0 + 1 (1)),
1 () = 0 (1) + 1 (1) + 0
+ (1 (1) + 1 ( + 1) + 2 (1) ),
(7)
(8)
(9)
(18)
+ ()[0 (, 1) 0 (1 )].
Inserting = 1 into (18) and solving (, 1), we obtain

(, 1) =
(1 )[0 (1 ) 0 (, 1)]
.
1 ((1 ) )
(19)
Utilizing (14) and (19) in (18), (, ) is represented as

(, )
=
(() (1 ))[0 (1 ) 0 (, 1)] (20)

.
( 1 )((1 ) )
24
By inserting = 1 into (20), it follows that
0 (1 )(1 (1 ))
.
0 (1 (1 )) + (1 1 )((1 ) )
(, 1) =
(21)
Let () denote the PGF of the system size distribution. From

(16) and (21), we can obtain
() = 0 + (, 1) + (, 1)
= 0 +
0 (1 )( + [])(1 (1 ))
,
0 (1 (1 )) + (1 1 )((1 ) )
(22)
where 0 = ( (1 ()))/[1 + [](1 ())] by the

normalizing condition (11).
Diferentiating (22) with respect to and taking limit
1 lead to the mean system size given by
=
1 + [](1 ())
[
(23)
(1 + [])(1 () ())
( + ())
where () = ()/ in (23).
],
Remark 1. System 1 is stable if and only if |()| < for

|| 1. Te denominator of () for System 1 can be written
as (1 1 )(() ), where
() =
Pr{ = 1} = Pr{ = 1},

1
Pr{ = } = Pr{ = } + Pr{ }1 ,

=1
() = Pr{ = }
=
1 ()
We further defne the modifed service time as the virtual

service time, including a time to repair. Since a repair period
follows each time a negative customer arrives, from the point
of view of a TC, the modifed service times can be regarded as
service times of the customers who arrive prior to a TC. Let
and (), respectively, denote the modifed service time
and its PGF under our assumption. Ten, we obtain
0 + (1 0 )(1 )
.
1 1
2,
(26)
=1
() + (1 ())()
.
1
Since we assumed that during a slot the potential arrival of

a negative customer and a repair completion occur immediately before the potential arrival of a positive customer, a TCs
arrival which occurs during ( , ) may belong to one of the
following cases.
Case 1. A TC arrives fnding the server is idle. Ten, the TCs
service is immediately started.
Case 2. A TC arrives at a slot during which a repair completion occurs. Ten, the waiting time of the TC is the total service time of the customers who are in the queue on the arrival
of the TC because the repair has just been completed.
(24)
Here, (1 1 ) has no zeros for || 1. By Rouches theorem,

() also has no zeros for || 1 if and only if
()/|=1 < 1. Tus, the necessary and sufcient condition for the stability of System 1 is < (1 ())1 .
Case 3. A TC arrives at a slot during which the server is busy

and a negative customer does not arrive. Ten, the waiting
time of the TC is the total service time of the customers who
are in the queue on the arrival of the TC, plus the remaining
modifed service time of the customer in service.
3.2. Sojourn Time Distribution for System 1. We derive the

PGF of the sojourn time of a test customer (TC) under the
FCFS discipline for System 1, regardless of whether or not
its service is interrupted by a negative customer. Te sojourn
time is defned as the sum of the queue waiting time and
the service time. We do not take blocked customers sojourn
times into consideration because they are equal to 0.
Let us defne the actual service time as the actual amount
of service time that a TC receives before departing the system
either by a service completion or by a negative customer. Let
and (), respectively, denote the actual service time and
its PGF under our assumption. Ten, we obtain
In other cases, a positive customer is blocked, and we do

not take blocked customers into consideration for waiting
time distribution. Let , denote the waiting time of a TC
that arrives in case , = 1, 2, 3, and defne , () =
Pr{Case }[, | Case ]. From Case 1, we have
Pr{ = } = Pr{ = } + Pr{ }
() = Pr{ = } =
=1
1,
+ (1 )()
.
1
,1 ()
= lim Pr{( ) = 0, ( ) = 1}0
= lim Pr{(( 1)+ ) = 0, (( 1)+ ) = 1}
(27a)
= 0 .
Case 2 yields
(25)
,2 ()
= lim Pr{( ) = , ( ) = 0, (+ ) = 1}[ ()]
=0
25
= lim Pr{(( 1)+ ) = , (( 1)+ ) = 1}[ ()]
=0
= (1)[ ()] = ( (), 1).

=0
(27b)
=1 =0
( ) = 1, (+ ) = 1}[ ()]
=1
= lim Pr{(( 1)+ ) = , (( 1)+ ) = + 1,

(( 1)+ ) = 1}[ ()]
= ( + 1)[ ()]
=1 =0

= ()[ ()]
=1 =1
+ ()[ ()]
=1 =1
()
1
[1 ( (), )
()
[ +

[] =
=1
()]
()
(1 ()
1
()].
1
1 ()
(1 + [])(1 () ())
( + ())
(30)
.
Remark 3. Equations (23) and (30) confrm the result of

Littles law: = [], where the efective arrival rate
= (1 ).
Remark 4. Since is the blocking probability of a positive

customer, the service canceling probability of a positive
customer is given by = (1 )(1()) because a servicecanceled positive customer frst enters the system without
blocking and then a negative customer arrives before the
completion of the service of the service-canceled positive
customer.
+ ( ( (), 1) 1 ( (), ))
(29)
which leads to the mean sojourn time given by
+ Pr{ = | ( ) = } ()]
1
(28)
Let be a TCs sojourn time. As and are independent, we fnally obtain

() = () (),
( (), 1)
( +
())],
()
1
Remark 2. Now that we deal with the Bernoulli arrival

process queueing system, BASTA property [22] is employed
to derive , ().
[Pr{ ( ) < | ( ) = }
=1
( (), )
(1
())
()
1
where = [](1 ())/(1 + [](1 ())), which

denotes the probability that the arrival of a positive customer
is blocked.
+ Pr{ = | ( ) = } ()]
=1 =0
1
1
[Pr{ ( ) < | ( ) = }
,1 (1) + ,2 (1) + ,3 (1)
[0 +
= lim Pr{( ) = , ( ) = ,
,1 () + ,2 () + ,3 ()
() =
=
From Case 3, we have

,3 ()
Let denote the unconditional waiting time of a TC. Combining (27a), (27b), and (27c) with (14), we have () given
by
4. System 2
(27c)
In System 2, newly arriving customers enter the system while

the server is under repair. Tey receive their service afer
repair.
26
4.1. System Size Distribution for System 2. With the limiting

probabilities defned in Section 2, the Kolmogorov equations
that govern System 2 can be written as follows:
0 () = 0 ( + 1) + 1 ,
1,
0 = (0 (1) + 0 + 1 (1)),
(32)
(33)
1 () = 0 (1) + 1 (1) + 0
+ (1 (1) + 1 ( + 1) + 2 (1) ),
(34)
+ ( + 1) + +1 (1) ),
(35)
2,
0 + (1, 1) + (1, 1) = 1.
(36)
(, ) = 0 ((, + 1) + 1 (, 1)),
(37)
Multiplying (31) and (32) by and summing over ,

0, we obtain
where 0 = +. Multiplying (38) by and summing over

, 1, yield
(, )(1 1 0 )
= 0 (1 () (, 1) (, 1)).
(38)
Inserting = 0 into (38) and solving (, 1), we obtain

(, 1) = 1 (0 ) (, 1).
(39)
Substituting (39) into (38) results in

(, ) =
0 (() (0 ))
(, 1).
( 0 )
+ ()[0 (, 1) 0 (1 )].
To solve (, 1), we insert = 1 into (43). Tus, we have

(, 1) =
0 (1 (0 ))
(, 1).
(1 0 )
(40)
(41)
(, )
(42)
+ [0 (, 1) + 1 (, 1) 0 (1 )],
1
(44)
(() (1 ))(0 (1 ) 0 (0 ) (, 1))

.
( 1 )((1 ) )
(45)
(, 1)
=
0 (1 )(1 (1 ))
.
0 (0 )(1 (1 )) + (1 1 )((1 ) )
(46)
Let () denote the PGF of the system size distribution. From

(41) and (46), we can obtain
() = 0 + (, 1) + (, 1)
= 0 +
0 (1 (1 ))[(1 ) + 0 (1 (0 ))]
,
0 (0 )(1 (1 )) + (1 1 )((1 ) )
(47)
where 0 = ( (1 + [])(1 ()))/ by the normalizing

condition (36).
Diferentiating (47) with respect to and taking limit
1 lead to the mean system size given by
=
(1 ())
+ ((2 [(2 + 2[] + 2 [( + 1)])
(1 ()) 2(1 + []) ()])
(48)
Multiplying (34) and (35) by and then summing over

, 1, together with (33), we obtain
= 1 (, + 1)
(1 )[0 (1 ) 0 (, 1)]
.
1 ((1 ) )
Using (39) and (44) in (43), (, ) is represented as
Letting = 1 in (40), we obtain the relationship

(, 1) =
(43)
By inserting = 1 into (45), it follows that
and the normalizing condition is given by
= 1 1 (, 1)(() )
(, )
() = 1 (1) + (1)
+ (1 ( + 1) + (1)
(, )(1 1 1 )
(31)
() = 1 ( + 1) + ( + 1) +
+ +1 ,
where 1 = ( + ). Multiplying (42) by and summing

over , 1, yield
(2[ (1 + [])(1 ())]) ).
Remark 5. Te denominator of () for System 2 can be

written as (1 1 )(() ), where
() =
0 (0 ) + (1 1 0 (0 ))(1 )
.
1 1
(49)
Similar to Remark 1, for the stability of System 2, the following

inequality is the necessary and sufcient condition under
which steady-state solution exists: < [(1 + [])(1
())]1 .
27
4.2. Sojourn Time Distribution for System 2. We derive the

PGF of the FCFS sojourn time of a TC for System 2. Te
method used in this section is much simpler than that used
in Section 3.2. Te PGF of the waiting time for the standard
Geo/G/1 queue is known to be [22]
(1 )
,
1 + ()
(50)
where is a normalizing constant. We frst replace () in (50)

with () in (27a), (27b), and (27c). Ten, is determined by
normalization. Applying this procedure, () is given by
()
(51)
(1 )(1 )
=
,
(1 )(1 ) (1 ())(1 ())
where = ( (1 + [])(1 ()))/. Since and are

independent, we have () as
() = () (),
(52)
which leads to the mean sojourn time given by

[]
=
1 ()
For the negative binomial case, the PMF of the service time is
defned by
= (
1 1 4 1 4
)( ) ( ) ,
3
2
2
= 4, 5, . . . .
(55)
For the geometric mixture case, the PMF of the service time
is defned by
4 1 2 1 1 1
27 1
( )( ) + ( )( ) ,
5 3 3
5 28 28
= 1, 2, . . . .
(56)
Te above three service times have a common mean of 8.

Te coefcients of variation are, respectively, 0.94, 0.35, and 2
for the geometric, negative binomial, and geometric mixture
cases.
Similarly, the repair time distributions are assumed to follow one of the three distributions. For the geometric case, the
PMF of the repair time is defned by
1 4 1
= ( )( ) ,
5 5
= 1, 2, . . . .
(57)
For the negative binomial case, the PMF of the repair time is
defned by
+ (([(2 + 2[] + [( + 1)])

2
(53)
(1 ()) 2(1 + []) ()])

1
Remark 6. Equations (48) and (53) confrm the result of

Littles law: = [].
Remark 7. Te service canceling probability of a positive
customer is given by = 1 () because a negative customer arrives before the completion of the service of a service-canceled positive customer.
5. Numerical Examples
Te purpose of this section is to study the infuence of negative customer arrival on the mean system size and the service
canceling probability of both System 1 and System 2. We
also investigate the infuence of the type of the repair time
distribution on the mean system size. In every example,
positive customers arrive at a system, according to a Bernoulli
process at a rate of 0.1. We consider three types of service time
distributions: a geometric distribution, a negative binomial
distribution, and a mixture of two diferent geometric distributions. Specifcally, for the geometric case, the probability
mass function (PMF) of the service time is defned by
= 1, 2, . . . .
1 2 2 3 2
)( ) ( ) ,
1
5
5
= 2, 3, . . . .
(58)
For the geometric mixture case, the PMF of the repair time is
defned by
(2[ (1 + [])(1 ())]) ).
1 7 1
= ( )( ) ,
8 8
= (
(54)
7 3
17 1
( )( )
10 20 20
3 9
1 1
+ ( )( ) ,
10 10 10
= 1, 2, . . . .
(59)
While these three repair times have a common mean of 5,

their coefcients of variation are, respectively, 0.89, 0.55, and
1.15 for the geometric, negative binomial, and geometric mixture cases.
In Figures 1, 2, and 3, the mean system sizes and the
service canceling probabilities of both System 1 and System
2 are shown as functions of when the repair time follows a
negative binomial distribution.
Figures 1 to 3 confrm that the mean system sizes of both
System 1 and System 2 decrease as increases. Furthermore,
for all cases, the mean system size of System 2 is greater than
the mean system size of System 1 as expected. Tis is obvious
because, in System 2, customers newly arriving during the
repair period join the queue and wait for the server to be
repaired without leaving the system. We also verify that the
service canceling probabilities of both System 1 and System 2
increase as increases and the service canceling probability of
System 2 is greater than that of System 1 for all cases. Tis is
because, in System 1, the arrival of a positive customer can be
blocked.
28
1
Service canceling probability
1.5
Mean system size
1.25
1
0.75
0.5
0.25
0.9
0.8
0.7
0.6
0.5
0.4
0
0
0.2
0.4
0.6
0.8
0.2
0.4
0.6
0.8
Geo service
System 1
System 2
Geo service
System 1
System 2
Figure 1: Mean system size and service canceling probability over with geometric service time distribution.
1
2.5
Mean system size
2
1.5
1
0.5
0.9
0.8
0.7
0.6
0.5
0.2
0.4
0.6
0.8
0.2
0.4
0.6
0.8
Negative binomial service

System 1
System 2

System 1
System 2
Figure 2: Mean system size and service canceling probability over with negative binomial service time distribution.
Mean system size
0.8
0.6
0.4
0.2
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.4
0.6
0.8
0.2
0.4
Geometric mixture service

System 1
System 2
0.6
0.8
Geometric mixture service

System 1
System 2
Figure 3: Mean system size and service canceling probability over with geometric mixture service time distribution.
29
References
Mean system size
2.5
2
1.5
1
0.2
0.4
0.6
0.8

Geometric
Negative binomial
Geometric mixture
Figure 4: Mean system size over with various repair time

distributions.
Second, we investigate the tendencies of the mean system

size of System 2 by varying the repair time distributions. In
Figure 4, the mean system size of System 2 is shown as a
function of when the service time follows the negative
binomial distribution.
As shown in Figure 4, for all three diferent distributions
of repair times, as increases, the mean system sizes decrease.
Moreover, for all values of , the higher the coefcient of
variation is, the greater the mean system size is. Te overall
results in this section confrm that the mean system size is a
decreasing function of .
6. Conclusion
We considered two repairable Geo/G/1 queueing systems
with negative customers, in which a negative customer arrival
causes the customer in service to leave the system and the
server to fail. At a failure epoch, the repair process immediately begins. Tose arriving during the repair period are
blocked in System 1, while they remain in System 2. For
these systems, we derived some important results of system
characteristics such as the PGFs of the system sizes and
FCFS sojourn times. Finally, some numerical examples were
performed to illustrate the infuence of the negative customer
arrival rate on the systems. Our research presents an extension of the discrete-time repairable queueing model and the
analysis of the model may provide a decision making tool for
repair policies arising in many practical systems.
For further studies, we can extend this model to more
complex situations such as the discrete-time repairable
queueing systems with the phase type arrival process of either
positive customers or negative customers.
[1] E. Gelenbe, Random neural networks with negative and positive signals and product form solution, Neural Computation,
vol. 1, no. 4, pp. 502510, 1989.
[2] R. J. Boucherie and O. J. Boxma, Te workload in the M/G/1
queue with work removal, Probability in the Engineering and
Informational Sciences, vol. 10, no. 2, pp. 261277, 1996.
[3] C. Morfopoulou, Network routing control with G-networks,
Performance Evaluation, vol. 68, no. 4, pp. 320329, 2011.
[4] J. Wu and X. Yin, An M/G/1 retrial G-Queue with non-exhaustive random vacations and an unreliable server, Computers &
Mathematics with Applications, vol. 62, no. 5, pp. 23142329,
2011.
[5] P. G. Harrison and E. Pitel, Sojourn times in single-server
queues with negative customers, Journal of Applied Probability,
vol. 30, no. 4, pp. 943963, 1993.
[6] P. G. Harrison and E. Pitel, Te M/G/1 queue with negative
customers, Advances in Applied Probability, vol. 28, no. 2, pp.
540566, 1996.
[7] W. S. Yang and K. C. Chae, A note on the GI/M/1 queue with
Poisson negative arrivals, Journal of Applied Probability, vol. 38,
no. 4, pp. 10811085, 2001.
[8] J. R. Artalejo, G-networks: a versatile approach for work
removal in queueing networks, European Journal of Operational Research, vol. 126, no. 2, pp. 233249, 2000.
[9] T. V. Do, An initiative for a classifed bibliography on G-networks, Performance Evaluation, vol. 68, no. 4, pp. 385394, 2011.
[10] T. V. Do, Bibliography on G-networks, negative customers and
applications, Mathematical and Computer Modelling, vol. 53,
no. 1-2, pp. 205212, 2011.
[11] E. Gelenbe, Product-form queueing networks with negative
and positive customers, Journal of Applied Probability, vol. 28,
no. 3, pp. 656663, 1991.
[12] E. Gelenbe, G-networks: a unifying model for neural and
queueing networks, Annals of Operations Research, vol. 48, no.
5, pp. 433461, 1994.
[13] E. Gelenbe, Te frst decade of G-networks, European Journal
of Operational Research, vol. 126, no. 2, pp. 231232, 2000.
[14] I. Atencia and P. Moreno, A single-server G-queue in discretetime with geometrical arrival and service process, Performance
Evaluation, vol. 59, no. 1, pp. 8597, 2005.
[15] W. H. Zhou, Performance analysis of discrete-time queue
GI/G/1 with negative arrivals, Applied Mathematics and Computation, vol. 170, no. 2, pp. 13491355, 2005.
[16] J. Wang and P. Zhang, A discrete-time retrial queue with negative customers and unreliable server, Computers and Industrial
Engineering, vol. 56, no. 4, pp. 12161222, 2009.
[17] H. M. Park, W. S. Yang, and K. C. Chae, Te Geo//1 queue
with negative customers and disasters, Stochastic Models, vol.
25, no. 4, pp. 673688, 2009.
[18] K. C. Chae, H. M. Park, and W. S. Yang, A GI/Geo/1 queue
with negative and positive customers, Applied Mathematical
Modelling, vol. 34, no. 6, pp. 16621671, 2010.
[19] J. Wu, Z. Liu, and Y. Peng, On the BMAP/G/1 G-queues
with second optional service and multiple vacations, Applied
Mathematical Modelling, vol. 33, no. 12, pp. 43144325, 2009.
[20] J. Wu, Z. Liu, and G. Yang, Analysis of the fnite source $MAP/
PH/N$ retrial $G$-queue operating in a random environment,
Applied Mathematical Modelling: Simulation and Computation
for Engineering and Environmental Systems, vol. 35, no. 3, pp.
11841193, 2011.
30
[21] D. H. Lee, W. S. Yang, and H. M. Park, Geo/G/1 queues
with disasters an d general repair times, Applied Mathematical
[22] H. Takagi, Queueing Analysis: A Foundation of Performance
Evaluation. Vol. 3, North-Holland, Amsterdam, Te Netherlands, 1993.

http://dx.doi.org/10.1155/2014/918376
Research Article
Radial Basis Point Interpolation Method with Reordering
Gauss Domains for 2D Plane Problems
Shi-Chao Yi,1 Fu-jun Chen,1 and Lin-Quan Yao2
1
2
School of Mathematical Sciences, Soochow University, Suzhou 215006, China

School of Urban Rail Transportation, Soochow University, Suzhou 215137, China
Correspondence should be addressed to Fu-jun Chen; chenfujun20@163.com

Received 13 June 2014; Revised 28 August 2014; Accepted 28 August 2014; Published 23 December 2014
Academic Editor: Guan H. Yeoh
Copyright 2014 Shi-Chao Yi et al. Tis is an open access article distributed under the Creative Commons Attribution License,
We present novel Gauss integration schemes with radial basis point interpolation method (RPIM). Tese techniques defne
new Gauss integration scheme, researching Gauss points (RGD), and reconstructing Gauss domain (RGD), respectively. Te
developments lead to a curtailment of the elapsed CPU time without loss of the accuracy. Numerical results show that the schemes
reduce the computational time to 25% or less in general.
1. Introduction
In the past two decades, the development and application of
meshfree methods have attracted much attention. One of the
reasons is the versatility of meshfree methods for complex
geometry of solids and fexibility for diferent engineering
problems [1]. Element-free Galerkin (EFG) method, which is
originated by Belytschko et al. [25], is one of the most widely
used meshfree methods. Te key advantage of EFG method is
that only nodal data is required and no element connectivity
is needed, when moving least-squares (MLS) interpolation is
used to construct trial and test functions. It is currently widely
used in computational mechanics and other areas, such as by
Sarboland and Aminataei [6] for nonlinear nonhomogeneous
Burgers Equation and Pirali et al. [7] for crack discontinuities
problem. In the meantime, some new techniques are used
to improve the performance of the MLS, complex variable
moving least-squares [810] and improved complex variable
moving least-squares [11], the moving least-squares with singular weight function [12], and so forth. But shape functions
constructed by MLS interpolation do not possess Kronecker
delta function property; the treatment of essential boundary
conditions is one of the typical drawbacks. Tus, many special
techniques have been proposed to impose essential boundary
conditions [1315], such as point collocation [13], Lagrange
multipliers [2], singular weighting functions [14], and penalty
method [15]. None of these methods is fully satisfactory, as
they still need additional eforts to enforce essential boundary

conditions.
In order to totally eliminate the drawback associated with
EFG method for imposing essential boundary conditions,
Liu and Gu have developed the point interpolation methods (PIM) by using polynomial basis or/and radial basis
function (RBF) [1619]. However, singularity may occur if
arrangement of the nodes is not consistent with the order of
polynomial basis, while the inverse of moment matrix always
exists for arbitrarily scattered nodes with radial basis [20]. In
this paper, we mainly care about the radial point interpolation
method (RPIM) studied by Wang and Liu in [21], which is
based on the global weak form.
On the other hand, high computational cost is still one
of the main drawbacks in meshfree method with Galerkin
weak form. In the RPIM method for a given computational
point, an linear system (the coefcient matrix is called
the moment matrix) should be solved to construct shape
functions if radial basis functions are selected, where is the
number of nodes in the support domain. Furthermore, if
monomial basis functions are added, the linear system will be
extended to ( + ) ( + ). Tis is very time-consuming
especially for the meshless methods based on weak forms,
where integration should be employed and the number of
computational points is much bigger than the number of
nodes.
32
In fact, common meshless method is based on the numerical integration for Gauss domains with RPIM. It is a tedious
process when the integration points are extremely more than
the nodes. In practical work, some Gauss point has the same
nodes as some neighboring computational points. Tus, they
only need one shape function with them. A process will be
needed to fnd these nodes. Tis scheme is called researching
Gauss points method (RGP). But the storage of searching
program will increase quickly with the increasing number of
nodes and Gauss points. Tus, more computational time is
spent. To avoid this, reconstructing Gauss domains (RGD) is
presented. All computational points share one same infuence
domain (the same nodes) in the domain. Tis scheme ofsets
the RGPs disadvantage without the searching program. Te
reason why we call it researching Gauss points method is
that the RGP method needs a searching nodes program same
as in the RPIM. Te two techniques are collectively named
reordering Gauss domain (RG) method.
Te remainder of the paper is arranged as follows.
In Section 2, the radial basis point interpolation method
(RPIM) is presented to get the shape functions. In Section 3,
a Galerkin weak form and its numerical algorithm are
studied for 2D solid mechanics problems. Te reordering
Gauss domains methods are then constructed in Section 4.
In Section 5, several examples are presented to show the
efectiveness of the proposed method and some parameters
performances of the proposed method are also investigated.
Finally, we end this paper with some conclusions in Section 6.
where = ( )2 + ( )2 and and are two

parameters. is defned as
= 0 ,
(4)
where 0 is a dimensionless coefcient and is a parameter

of the nodal distance. For regularly distributed nodal case,
is the shortest distance between the node and its neighbor
nodes. Efects of 0 and have been studied in detail in [22].
In static analysis of 2D solid problems, it has been found that
0 = 1.0 and = 1.03 lead to good results. Hence, these
numbers are used in this paper.
Te second term in (1) consists of polynomials. To ensure
invertible interpolation matrix of RBF, the polynomial added
into the RBF cannot be arbitrary. A low-degree polynomial is
ofen needed to augment RBF to guarantee the nonsingularity
of the moment matrix. In addition, the linear polynomial
added into the RBF can also ensure linear consistency and
improve the interpolation accuracy [22]. Tus, the linear
polynomial is added into the MQ RBF in the following
discussion.
Coefcients and in (1) can be determined by enforcing that (1) and (2) be satisfed at the nodes surrounding
point . Equations (1) and (2) can be rewritten in the matrix
form:
Ga0 [
where
R0 P0 a
u
] [ ] = [ ] ,
0
PT0 0 b
(5)
aT0 = [aT bT ] = {1 , 2 , . . . , , 1 , 2 , 3 } ,
T
2. Radial Basis Point Interpolation

Method (RPIM)
Te RPIM interpolation () of , for all , can be
defned by
=1
=1
() = () + ()
(1)
with the constraint condition
( ) = 0,
=1
= 1, 2, . . . , ,
(2)
where () is the radial basis function (RBF), is the number

of nodes in the neighborhood of , () is the monomial
in the space coordinates pT () = [1, , ], is the number
of polynomial basis functions, and coefcients and are
interpolation constants. Te variable of the radial basis
function () is the distance between the interpolation point
and a node . It is necessary to construct the interpolation
function here to solve the equations.
Tere are a number of types of radial basis functions.
Characteristics of radial basis functions have been widely
investigated in [20, 22]. In this paper, the following multiquadrics (MQ) radial basis function is used:
() =
(2
+ ) ,
(3)
uT = {1 , 2 , . . . , } ,
1 (1 ) 2 (1 )
[1 (2 ) 2 (2 )
[
R0 = [ ..
..
[ .
.
[1 ( ) 2 ( )
(1 )
(2 )]
]
.. ] ,
d
. ]
( )]
(6)
1 1 1
PT0 = [1 2 ] .
[1 2 ]3
Te matrix R0 is symmetric, and thus the matrix G is
symmetric. Ten, the interpolation equation (1) is fnally
expressed as
() = [RT () pT ()] G1 u = () u ,
where the shape function () is defned by
(7)
() = [1 () , 2 () , . . . , ()] = [RT () pT ()] G1 .

(8)
And RT () = [1 (), 2 (), . . . , ()], pT () = [1, , ]. It

can be found from the above discussion that RPIM passes
through the nodal values. Terefore, RPIM shape functions
given in (8) satisfy the Kronecker delta condition. Tus,
( ) = = {
1,
0,
= ,
= .
(9)
33
3. Variational Form of 2D Plane Problem

Consider the 2D problem of the deformation of a linear elastic
medium from an undeformed domain , enclosed by :
+ b = 0,
in ,
(10)
where is the stress tensor corresponding to the displacement feld u and b is a body force vector, and boundary
conditions are as follows:
n=t
on ,
on ,
u=u
4. Reconstructing Gauss Domains Methods

(11)
in which t and u are prescribed tractions and displacements,

respectively, on the traction boundary and on the displacement boundary and n is the unit outward normal matrix
to the boundary .
Using the standard principle of minimum potential
energy for (10)-(11), that is, to fnd u (1 ())3 such that
=
1
T D uT b uT t
2
(12)
is stationary, where () denotes the Sobolev space of order

, and = D are strain-stress vectors, and D is the strainstress matrix. RPIM equation (1) is used to approximate the
displacements in the Galerkin procedure. Ten we can get
0 { }
(13)
u = { } = [ ] { } = u .
=1
=1
{ }
Substituting (13) into (12) leads to the following total potential
energy in the matrix form:
1 T
(14)
u K u uT f,
2
and invoking = 0 results in the following linear system of
u:
=
Ku = f,
(15)
in which the stifness matrix K is built from 2 2 matrices K

and the right-hand side vector f is built from the 2 1 vectors
f . Tese matrices and vectors are defned by
K = BT DB ,
, = 1, 2, . . . , ,
f = b + t ,
where
= 1, 2, . . . , ,
(16)
, 0
B = [ 0 , ] ,
[, , ]
1 ] 0
[] 1 0 ]
D=
[
]
1 ]2 0 0 1 ]
[
2 ]
Whether the RPIM method or other meshless methods

based on global weak form, background cells are necessary
to obtain the numerical integration of (16). Diferent from
the fnite element method (FEM), which uses the same
nodes for both interpolation and numerical integration,
background cells in meshless methods are independent of
the interpolations. In this paper, quadrilateral cells and Gauss
quadrature are used for the numerical integration.
(for plane stress problem) .

(17)
A weak form, in contrast to a strong form (collocation

method in general), requires weaker consistency on the
assumed feld functions. Te consistency requirement on
the assumed functions for feld variables is very diferent
from the strong form. For a 2th-order diferential governing system equation, the strong formulation requires a
consistency of the 2th order, while the weak formulation
requires a consistency of only the th order. But the meshfree
method usually uses the integral representation of feld
variable functions for solving strong form system equations
and the numerical integration is extremely time-consuming.
Numerical accuracy mainly depends on the number of Gauss
points in the corresponding domain; the more the Gauss
points, the better the results in general. Tus, one of the
most time-consuming steps in the meshless method is the
construction of shape functions, since for every point of
interest a linear system should be computed. As discussed
in Section 2, using radial basis functions to construct shape
functions, an linear system should be computed for
every computational point. If monomials are added, an
( + ) ( + ) linear system should be solved. Tis is
very time-consuming especially for meshless methods based
on weak form, where a large number of integration points are
used. In this section, we propose the researching Gauss points
(RGP) and reconstructing Gauss domains (RGD) methods,
which are together named reordering Gauss domains (RG)
methods and can partly reduce the computation cost for the
meshless methods compared with the RPIM.
4.1. Researching Gauss Points (RGP) with the Same Nodes.
In the RPIM approximation, shape function consists of two
parts: [RT () pT ()] and G1 (8). Te time consumption
of [RT () pT ()] is less than that of G1 , because the
computational complexities are ( + ) and (( + )3 ),
respectively.
Every Gauss point has its own [RT () pT ()] and is
diferent from the rest, but it may have the same G1 as the
other. Tus, we need a storage place, containing the public
nodes and the Gauss points information. Te data is obtained
by searching the Gauss points. Tus, the method is called
researching Gauss points (RGP) method.
Te red area represents the nodes set (involved red nodes
) and the blue area represents the Gauss points set (involved
blue gauss points ) in Figure 1(a). It should be pointed out
that one-to-one mapping exists between the Gauss points
sets and the nodes points sets (can be verifed by logical
34

Te corresponding nodes and Gauss points
Te role of the center of the Gauss domain
Nodes
Gauss points
Te center of the Gauss domain
Nodes
Gauss points
(a)
(b)
Figure 1: Te distribution of the nodes and the Gauss points: fnd the 1-1 mapping (a); assign the 1-1 mapping (b).
deduction). Tus, the key work is how to get the relationship.

First of all, we construct matrix = [ ], where
, denote the number of the nodes and the number of the
Gauss points, respectively. Te element = [ ] is function:
1
{
{
= {
{
{0
the th nodes locate in the support domain

of the th gauss point;
otherwise.
(18)
Tus, is written as
[1
[
S=[
[
[1
[
0]
]
]
.
]
0]
]
we present the reconstructing Gauss domains (RGD) method

without the searching process. In the fxed Gauss domain,
all Gauss points are mandatorily contacted with some neighbouring nodes. Tese nodes lie in the support domain of the
Gauss domains centroid. Tat is, every Gauss domain has its
own basis functions or G1 corresponding to the centroid.
A center point () of the Gauss domain is added in
Figure 1(b) compared with Figure 1(a). Without the searching
process, we mandatorily consider that the center point is seen
by not only the barycentre of the Gauss domain, but also
the barycentre of the domain involving some nodes. Tese
nodes locate in the support domain with on the barycentre
point. Tus, the one-to-one mappings are ascertained by the
corresponding barycentre points.
(19)
If the th row elements are the same as the th row

elements, we confrm the th and th Gauss points have
the same basis functions ( ) ( ). It is a rule, which
gives one-to-one mappings between the two types of sets by
comparing the rows of . Some codes (isequal.m, unqiue.m,
etc.) could be used to fnd these mappings in MATLAB. Te
process of one-to-one mapping is also called the searching
program distinguishing the searching nodes program in
advance. For the sake of simplicity, the RGP method is
divided into two parts: Part I (the researching program) and
the rest.
4.2. Reconstructing Gauss Domains (RGD). Te searching
program saves the time consumption in Gauss integration,
but it needs extra time and storage for the searching process.
With the rapid increase of the nodes and the Gauss points,
the searching process costs also grow exponentially. Ten,
5. Numerical Experiments
In this section, several numerical examples are selected to
demonstrate the applicability of the RG meshless methods.
Te numerical results for these examples are compared
with the analytical solutions and the RPIM solution. Square
support domains are used for calculations in the present
paper. stands for the average node distance. Te MQ radial
basis function and the linear polynomial are used to construct
shape functions. All runs are performed in MATLAB 7.0 on
an Intel Pentium 4 (2 GB RAM) Windows XP system.
5.1. Patch Test. Te frst numerical example is the standard
patch test shown in Figure 2. Te patch test consists of thirteen nodes including fve interior nodes. A 2 2 rectangular
background mesh is used for numerical integration and 4 4
Gauss points are used in each mesh. In this patch test, the
displacements are prescribed on all outside boundaries by a
linear function of and on the patches of the dimension.
Te parameters are taken as = 1.0 and ] = 0.3. Te linear
35
Table 1: Results at interior nodes located irregularly in the 13 nodes patch.

Internal node
Coordinates
(1.0, 1.0)
10
(0.65, 1.0)
11
(0.70, 1.5)
12
(1.3, 1.2)
13
(1.2, 0.6)
( , )
( , , )
( , )
( , , )
( , )
( , , )
( , )
( , , )
( , )
( , , )
11
12
2
10
[3]2 ( ) + (4 + 5])
+ (3 ) 2 ] ,
6
4
(20)
where the moment of inertia of the beam is given by

= 3 /12. Te stresses corresponding to the above
displacements are
Figure 2: A patch with 13 nodes.

y
RGD
(0.601, 0.599)
(0.853, 0.853, 0)
(0.391, 0.599)
(0.861, 0.853, 0.002)
(0.421, 0.897)
(0.856, 0.855, 0.001)
(0.779, 0.718)
(0.846, 0.852, 0.002)
(0.720, 0.359)
(0.854, 0.856, 0)
2
[(6 3) + (2 + ]) (2
)] ,
6
4
13
RGP/RPIM
(0.601, 0.599)
(0.853, 0.853, 0)
(0.391, 0.599)
(0.861, 0.853, 0.002)
(0.422, 0.897)
(0.856, 0.855, 0.001)
(0.779, 0.718)
(0.846, 0.852, 0.002)
(0.720, 0.359)
(0.854, 0.856, 0)
beam has a unit thickness, and thus a plane stress problem

is considered here. Te closed-form solution is available for
parabolic traction of force :
Exact
(0.600, 0.600)
(0.857, 0.857, 0)
(0.390, 0.600)
(0.857, 0.857, 0)
(0.420, 0.900)
(0.857, 0.857, 0)
(0.780, 0.720)
(0.857, 0.857, 0)
(0.720, 0.360)
(0.857, 0.857, 0)
x
P
( )
,
displacement functions are = 0.6 and = 0.6. RG

methods of the patch test require that the displacement of
any interior node be given by the same linear function and
that the strains and stresses be constant in the patch. Te
RG methods pass the patch test when linear polynomials
are added ( = 3). However, when polynomial term is not
included ( = 0), the patch test does not easily pass. Te
same is for the RPIM method; see [21]. Computed results at
interior nodes for the RG methods and the RPIM method and
the exact results are listed in Table 1.
5.2. Cantilevered Beam. Te second example is a cantilever
beam problem; see Figure 3. Consider a beam of length
and height subjected to traction at the free end. Te
2
(
2 ) ,
2 4
= 0.
(21)
Te beam parameters are taken as = 3.0107 , ] = 0.3, =

12, = 48, and = 1000 in computation.
To evaluate efects of various parameters, we use the twonorm relative errors and for displacement and stress,
respectively. Tey are defned by
Figure 3: Cantilever beam.
2
u u
,
u2
,
2
(22)
and u are the approximate and exact solution of

where u
and are the approximate and exact
displacements and
value of stresses.
5.2.1. Efect of Irregular Node Distribution. Node distributions
with 325 irregular nodes and 481 irregular nodes are shown
in Figure 4. A background mesh is necessary to obtain the
numerical integrations of (15). Tis mesh is independent of
nodes for interpolations, while FEM uses the same nodes
for both interpolation and numerical integration. For this
problem, 10 6 and 15 6 background cells are used for
the 325 irregular nodes problem and 481 irregular nodes
problem, respectively. In each cell, a 4 4 Gauss quadrature
is used to evaluate the stifness matrix. Figure 5(a) shows a
comparison of the analytical solution and numerical solution
36
(a)
(b)
Figure 4: Meshless models with irregular data points: 325 nodes (a), 481 nodes (b).
103
1
Displacement in y direction along y = 0
Shear stress along x = L/2
20
1
20
2
40
3
4
60
80
6
100
7
120
8
9
0
10
20
30
40
50
140
6
RPIM/RGP
RGD
Analytical
RPIM/RGP
RGD
Analytical
(a)
(b)
Figure 5: Results for 325 irregular nodes: (a) displacement and (b) shear stress.
of along = 0 for this problem with 325 irregular

nodes. Te plot shows an excellent agreement between the
analytical and numerical results for each method when an
irregular node distribution is used. Figure 5(b) is the sectional
distribution of shear stress along the = 24 section
for this problem with 325 irregular nodes. Te closed-form
solution is also plotted for comparison.
Computational results including error and computational
cost are listed in Table 2 for the cantilevered beam problem
with 325 irregular nodes and 481 irregular nodes. From
Figure 5 and Table 2, we can fnd that the RG methods have
better accuracy and are less time-consuming than the RPIM
method.
Table 2: Computational results for irregular node distribution.

Irregular
nodes
RGP
(Part I + the rest)
RGD
325
CPU
3.034
0.0046
0.0013
0.0793
0.0582
2.917 = 1.424 + 1.493
1.162
481
CPU
5.2.2. Efect of the Size of Support Domain of RGD. As

discussed in Section 4.2, the centroid of every Gauss cell has
a support domain. Te range of the support domain is
an important parameter. Te 325 regular nodes distribution
(25 13: 25 nodes in the direction and 13 nodes in the
direction) is used to study the efect of the size of the support
domain for the RG methods.
In Table 3, we list the numerical results for the RGD
method with diferent . To compare with the RPIM method,
RPIM
4.521
0.0080
0.0019
0.0824
0.0472
6.468 = 4.354 + 2.114
1.668
we also list the computational results for the RPIM method

with diferent radiuses of the support domain. From Table 3,
we can see that the RGD method has better steadiness,
computational efciency, and computational accuracy than
the RPIM method. Particularly, with = 4.5, the elapsed
CPU time is reduced to 12.7% (see the data in the boxes)
of the previous one. Tis shows that the size of the support
2.5
3.0
3.5
4.0
4.5
5.0
37
Table 3: Efect of the size of support domain of RGD (Example 5.2).
0.0513
0.0129
0.0109
0.0006
0.0004
0.0002
RGD
0.2033
0.1000
0.0945
0.0285
0.0274
0.0202
CPU
0.413
0.646
0.733
1.182
1.336
2.112
0.0074
0.0031
0.0031
0.0002
0.0021
0.0004
RPIM and RGP

RPIM CPU
1.808
3.003
4.752
7.135
10.303
14.599
0.0522
0.0620
0.0305
0.0533
0.0191
0.0294
Table 4: Convergence study on regular nodes for Example 5.2.

Regular nodes
RGD
CPU
RGP/RPIM
RPIM CPU
RGP CPU (Part I + the rest)
13 4
25 7
37 10
49 13
61 16
0.0384
0.1891
0.440
0.0063
0.0399
0.839
0.0035
0.0676
1.972
0.0025
0.0668
2.730
0.0015
0.0182
5.420
0.0358
0.2057
1.864
0.219 + 0.350
0.0028
0.0522
2.450
1.404 + 0.168
0.0024
0.0559
5.637
5.943 + 2.147
0.0196
0.1434
5.981
9.975 + 3.049
0.0240
0.1738
12.302
59.036 + 6.468
Table 5: Convergence study on irregular nodes for Example 5.2.

Irregular nodes
RGD
CPU
RGP/RPIM
RPIM CPU

2.151 = 1.023 + 1.128
2.537 = 1.131 + 1.406
3.277 = 1.798 + 1.479
4.071 = 2.608 + 1.463
5.195 = 3.720 + 1.475
6.451 = 4.911 + 1.540
52
175
370
637
976
0.0413
0.1836
0.419
0.0062
0.0399
0.831
0.0021
0.0578
2.055
0.0055
0.0832
2.627
0.0175
0.1138
4.387
0.0386
0.1758
1.637
0.203 + 0.332
0.0003
0.0553
2.486
0.832 + 0.734
0.0020
0.0422
5.600
5.989 + 1.667
0.0038
0.0610
6.178
10.018 + 2.480
0.0003
0.0108
12.814
58.798 + 3.672
domain has increased and its CPU time takes up a larger

proportion of the entire time, in which the consumption
surged from 48% to 76% of the entire cost. Te searching
process has become the major expenditure of the entire
program. From Table 3, we can fnd that the RGD method has
a stable convergency with the increase of the , but it does not
appear in the RPIM. In this paper, we take the suitable to be
4.
5.2.3. Convergence Study. For convergence studies, fve different regular node distributions with 52 (13 4: 13 nodes
in the direction and 4 nodes in the direction) nodes, 175
(25 7) nodes, 370 (37 10) nodes, 637 (49 13) nodes,
and 976 (61 16) nodes and fve diferent irregular node distributions are considered. Te size of the support domain is
taken as = 4. Te computational results are listed in Tables 4
and 5. In Tables 4 and 5, we also list the computational results
of the RPIM method. Te convergence curves with diferent

node distributions are plotted in Figure 5. In Figure 5,
is the maximum size of node arrangement. From Tables
4 and 5 and Figure 5 we can see that the present method
possesses good convergence and less computational times
than the RPIM method in general. However, the searching
process consumption overtakes 88% of the entire process in
the 976 regular/irregular nodes, which become cumbersome.
Te oscillatory behavior of the convergence curve of RGD
method exists in regular node, and oscillatory behaviors
appear in the two node distributions of the RPIM method.
Figure 5 shows that RGD method is steadier than the RPIM
method. Te reason of oscillatory appearance is that the
accuracy of most meshless methods is closely related to the
integration scheme, the number of Gauss points, the radiuses of the support domain and infuence domain, and so on.
Tus, it is a difcult task to get the best accuracy for all cases.
38

Convergence study with error of displacement for regular nodes
1.4
Convergence study with error of displacement for irregular nodes

1
1.6
1.5
1.8
2
Log10 (ru )
Log10 (ru )
2
2.2
2.5
2.4
3
2.6
3.5
2.8
3
0.1
0.1
0.2
0.3
0.4
Log10 (h)
0.5
0.6
0.7
RPIM/RGP
RGD
4
0.1
0.1
0.2
0.3
0.4
Log10 (h)
0.5
0.6
0.7
RPIM/RGP
RGD
(a)
(b)
Figure 6: Convergence study with error of displacement: regular nodes (a), irregular nodes (b).
(Tables 4 and 5 and Figure 6). Tus, the RGP method is not
considered in the next section.
(a)
(b)
Figure 7: Displacement 13 7 (a) and 36 13 (b) regular nodes

distribution (scaled 400).
5.2.4. Discussion on the Computational Results. To further

discuss the efectiveness of the RG methods, we discuss the
computational results of the cantilever beam problem in
this subsection. Figure 6 shows a comparison of analytical
solution and the present RGD numerical solution for the
beam defection. Two nodal distributions, 91 (13 7) regular
nodes and 481 (37 13) regular nodes, are used. Te stress
results are also obtained. Figures 8 and 9 illustrate the
analytical solution and the RG methods solutions for the
stress and the shear stress of the beam using 481 regular
nodes, respectively. Figures 68 show the RG methods and
RPIM have good performance for the stress and poor
performance for the stress . But in terms of time, the RGD
method is better than the RGP method and RPIM method
5.2.5. A Special Case of the RD Methods. As a special case of

the RD methods, the vertices of the Gauss domains coincide
with the nodes in regular nodes distribution; see Figure 11.
Figure 11 shows that a 5 5 uniform grid and a 4 4 Gauss
quadrature in each Gauss domain are used. Te Gauss points
in the Gauss domain have the same nodes (red nodes) with
= 2. We consider that the process of constructing shape
function has three parts: the [() ()], solving the inverse
of G, and computing [() ()]G1 . Tese computational
complexities are (( + )), (( + )3 ), and (( + )2 ),
respectively. Table 6 shows the complexity of the integration
for a Gauss domain with 16 Gauss points.
Te efciency ratio is defned as CPURPIM /CPURG . To
discuss the efciency ratio, two nodal distributions, 325 (25
13) regular nodes and 629 (37 17) regular nodes, are used.
Table 7 shows that the efciency ratios are 11.832 and 13.362,
respectively. Te efciency ratios are also expressed as
efciency ratios
= (NG (16 (( + )3 ) + 16 (( + )2 )
+16 ( + ))
+ preprocess + postprocess)
(23)
(NG ( (( + )3 ) + 16 (( + )2 )
+16 ( + ))
+ preprocess + postprocess) .
39
2000
RPIM/RGP solution x
Analytical solution x
15
15
1500
1500
10
1000
500
10
1000
500
500
500
1000
10
1000
10
1500
15
1500
15
2000
2000
0
10
20
30
40
10
20
(a)
30
40
(b)
2000
RGD solution xy
15
1500
10
1000
500
500
10
1000
15
1500
2000
0
10
20
30
40
(c)
Figure 8: Stress : analytical results (a), RGP numerical results (b), and RGD numerical results (c).
To refne the nodes domain, the NG will be a large

number. Te elapsed CPU times of compute shape functions
is far than the rest. Tus,
efciency ratios (16 (( + )3 )
+ 16 (( + )2 )
+16 ( + ))
( (( + )3 ) + 16 (( + )2 )
+16 ( + ))
16.
(24)
Te formula of the efciency ratios fully explains that the

ratio is 11.832 for 325 (25 13) regular nodes (NG = 288)
and 13.362 for 629 (37 17) regular nodes (NG = 576).
Numerical results show that the rates are 11.832 and 13.362
which gradually close to the theory upper 16 with the increase
of NG (the number of the Gauss domains). If we want to

improve the accuracy of numerical results with refning the
nodes domain, the number of the Gauss points should be
added and the efciency ratios will be improved further.
5.3. Plate with a Central Circular Hole. Consider now a plate
with a central circular hole subjected to a unidirectional
tensile load of 1.0 in the direction (see Figure 11(a)). Tis
is a typical plane stress problem.
Te closed-form solution of stresses is
= 1
=
=
2 3
34
[
cos 4,
cos
2
+
cos
4]
+
2 2
24
2 1
34
[
cos 4,
cos
2
cos
4]
2 2
24
2 1
34
[
sin 4,
sin
2
+
sin
4]
+
2 2
24
(25)
where (, ) are polar coordinates, is the measured counterclockwise from the positive -axis, and is the radius of
40

0
Analytical solution xy
RPIM/RGP solution xy
15
15
20
20
10
10
40
40
5
5
0
60
60
80
80
10
100
10
100
15
15
120
120
0
10
20
30
40
10
20
30
(a)
40
(b)
RGD solutionx
15
20
10
40
5
0
60
80
10
100
15
120
0
10
20
30
40
(c)
Figure 9: Shear stress : analytical results (a), RGP numerical results (b), and RGD numerical results (c).
the hole. Te corresponding displacements, in the plain stress

case, are given by
=
1
1
2
{ [
+ cos 2] +
[1 + (1 + ) cos 2]
4
2
Te Gauss points distribution with = 2 in the 5 5 grid
4
cos 2} ,
3
2
1
4
[(1 )
3 ] sin 2,
4
(26)
where = /(2(1 + ])) and = (3 ])/(1 + ]). Te material

properties are = 1.0 103 and ] = 0.3.
By taking advantage of its symmetry, only a quarter of
the model is considered in the analysis (see Figure 11(b)).
Symmetry conditions are imposed on the lef and the bottom
edges, and the inner boundary of the hole is traction-free.
Tese boundary conditions include (i) essential boundary
conditions on the bottom and lef edges on which displacement is computed from the exact displacement given in
(26) and (ii) natural boundary conditions on the right and
top edges on which traction is computed from the exact
stress given in (25). A typical node distribution with 99 nodes
Gauss points
Nodes
Figure 10: Te diagram of 5 5 domains for the special case.
and the corresponding background mesh constructed for the

numerical integration are shown in Figure 12(a). For each
background cell, a 4 4 Gauss quadrature is employed. It
41
Table 6: Te complexity of the integration for a Gauss domain.
Part I
(( + ))
16 times
16 times
Complexity
RPIM
RG
Part II
(( + )3 )
16 times
1 time
Part III
(( + )2 )
16 times
16 times
Sum
16 (( + )3 ) + 16 (( + )2 ) + 16 ( + )
(( + )3 ) + 16 (( + )2 ) + 16 ( + )
Table 7: Efect of the size of support domain for the RG method (Example 4.2).
Schemes
RG
RPIM
Efciency ratio
Grid
25 13
4.7796 004
0.0204
37 17
CPU
1.267
14.991
11.832
7.3559 005
CPU
2.373
31.708
13.362
0.0054
5m
2a
x
o
5m
(a)
(b)
Figure 11: Plate with a central circular hole and its model problem: (a) a plate with a hole, (b) a quarter of the plate and boundary conditions.
Table 8: Efect of the size of support domain for the RG methods

for the 99 nodes (Example 5.3).
1.1
1.2
1.5
1.8
2.0
3.0
0.0149
0.0097
0.0094
0.0056
0.0054
0.0044
RGD
0.1101
0.0881
0.0623
0.0886
0.1154
0.0590
CPU
0.880
0.963
1.175
1.580
2.043
4.600
0.0219
0.0205
0.0271
0.0133
0.0115
0.0057
RPIM
0.1005
0.1004
0.1526
0.2052
0.2226
01500
CPU
0.968
1.052
1.547
2.383
3.052
9.286
is also of great interest to further study the efect of the

nodal infuence domain size and the computational support
domain size. It should be noted that the support domain for
all nodes is a circle with varying radius. Tey are chosen
such that the support is small for nodes near the hole and
bigger for nodes near the edges. Table 8 presents the relative
errors and elapsed CPU times with diferent parameters for
Table 9: Convergence study for Example 5.3.

Scheme
RGD
CPU
RPIM
CPU
99 nodes
289 nodes
625 nodes
1089 nodes
0.0149
0.1101
0.880
0.0053
0.0534
2.030
0.0025
0.0384
3.480
0.0016
0.0343
13.516
0.0219
0.1005
0.968
0.0057
0.0541
4.523
0.0031
0.0343
17.419
0.0022
0.0283
55.530
the 99 nodes. In this case, the computational results show

relatively good accuracy and less elapsed CPU time. For
convergence studies, four diferent node distributions with 99
nodes, 289 nodes, 625 nodes, and 1089 nodes are considered
in Table 9. Te last three node distributions are taken from
[23]. Due to singular property for RPIM, was advised as 1.1
compared with the RGD method. Te computational results
42

5
4.5
4.5
3.5
3.5
2.5
2.5
1.5
1.5
0.5
0.5
0
0
0
1
2
3
4
Nodes and background cells distribution
Analytical
Numerical (scaled 80)
(a)
(b)
Figure 12: Plate with a central circular hole: (a) 99 nodes distribution, (b) analytical and RGD solution of displacement.
Analytical solution xx
Numerical solution xx
5
3
4.5
4.5
2.5
4
3.5
2.5
3.5
2
3
3
1.5
Y 2.5
Y 2.5
1.5
2
1
1
1.5
1.5
1
0.5
0.5
0.5
0.5
0
0
0
0
0
0
(a)
(b)
Figure 13: Stress : analytical results (a) and RGD numerical results (b).
are presented in Table 9, which also shows that the RGD

method possesses high convergence. Te elapsed CPU times
of the four node distributions fall 9%, 55%, 80%, and 76%,
respectively. Te computed displacements and stresses are
also shown for 99 nodes distribution problem in Figures
1013. Te displacements computed by the RGD method
at nodes are plotted and compared to the exact solution
(see Figure 12(b)). Figures 13 and 14 compare the analytical
solution and RGD solution of stress and shear stress ,
respectively. Figure 15(a) compares the stresses at stations
along the lef edge of the plate for the RPIM solution, the RGD
solution, and the closed-form solution. Figure 15(b) plots the
stresses at stations along the bottom edge of the plate for
these methods. From Figures 1113, it is shown that the RGD

has a higher accuracy compared with the RPIM solution. In
particular, the RGD solution of stress at (0, 5) agrees with
the analytical result, but the RPIM result is far away from the
analytical result.
6. Conclusions
In this paper, we have studied researching Gauss points and
reordering Gauss domain technique (GD methods), where
the computational processes are more reasonable. Secondly,
this technique is compatible with most of the common
RPIM methods, and this aspect is worth studying. Finally,
43
Analytical solution xy
Numerical solution xy
5
0.1
4.5
4.5
0
0.1
0.1
3.5
3.5
0.2
0.2
Y 2.5
0.3
3
0.3
Y 2.5
0.4
0.4
0.5
1.5
1.5
0.6
0.5
1
1
0.7
0.6
0.5
0.5
0.8
0
0
X
(a)
(b)
Figure 14: Shear stress : analytical results (a) and RGD numerical results (b).
0.2
3.5
0
Stress y along y = 0 section
Stress x along x = 0 section
2.5
1.5
0.2
0.4
0.6
0.8
1
1
1.2
1
1.5
2.5
3
y
3.5
4.5
1.5
2.5
3
x
3.5
4.5
RGD
Analytical
RPIM
RGD
Analytical
RPIM
(a)
(b)
Figure 15: Comparisons of RGD and analytical solution: stress at the stations along the lef edge of the plate (a) and stress at the stations
along the bottom edge of the plate (b).
3D problems could be solved with this scheme to save the

computation time.
anonymous referees for their hard work which has greatly

improved the original paper.
References

Acknowledgments
Tis work is supported by the National Natural Science
Foundation of China under Grant nos. 11172192 and 11301290.
Finally, the authors wish to thank the editors and the
[1] X. Zhang, K. Z. Song, M. W. Lu, and X. Liu, Meshless

methods based on collocation with radial basis functions,
Computational Mechanics, vol. 26, no. 4, pp. 333343, 2000.
[2] T. Belytschko, Y. Y. Lu, and L. Gu, Element-free Galerkin
methods, International Journal for Numerical Methods in Engineering, vol. 37, no. 2, pp. 229256, 1994.
[3] Z. Zhang, P. Zhao, and K. M. Liew, Improved element-free
Galerkin method for two-dimensional potential problems,
44
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]

Engineering Analysis with Boundary Elements, vol. 33, no. 4, pp.
547554, 2009.
Y. Cao, J.-L. Dong, and L.-Q. Yao, A modifcation of the moving least-squares approximation in the element-free galerkin
method, Journal of Applied Mathematics, vol. 2014, Article ID
528082, 13 pages, 2014.
F. Z. Loua, N. Nat-Sad, and S. Drid, Implementation of
an efcient element-free Galerkin method for electromagnetic
computation, Engineering Analysis with Boundary Elements,
vol. 31, no. 3, pp. 191199, 2007.
M. Sarboland and A. Aminataei, On the numerical solution
of one-dimensional nonlinear nonhomogeneous Burgers equation, Journal of Applied Mathematics, vol. 2014, Article ID
598432, 15 pages, 2014.
H. Pirali, F. Djavanroodi, and M. Haghpanahi, Combined
visibility and surrounding triangles method for simulation of
crack discontinuities in meshless methods, Journal of Applied
Mathematics, vol. 2012, Article ID 715613, 16 pages, 2012.
Y. M. Cheng and J. H. Li, A meshless method with complex
variables for elasticity, Acta Physica Sinica, vol. 54, no. 10, pp.
44634471, 2005.
L. Chen and Y. M. Cheng, Reproducing kernel particle method
with complex variables for elasticity, Acta Physica Sinica, vol. 57,
no. 1, pp. 110, 2008.
M. Peng, P. Liu, and Y. Cheng, Te complex variable elementfree Galerkin (CVEFG) method for two-dimensional elasticity
problems, International Journal of Applied Mechanics, vol. 1, no.
2, pp. 367385, 2009.
B. Funong, L. Dongming, W. Jianfei, and C. Yumin, An
improved complex variable element-free Galerkin method for
two-dimensional elasticity problems, Chinese Physics B, vol. 21,
no. 2, Article ID 020204, 2012.
I. Kaljevic and S. Saigal, An improved delement free Galerkin
formulation, International Journal for Numerical Methods in
G. J. Wagner and W. K. Liu, Application of essential boundary conditions in mesh-free methods: a corrected collocation
method, International Journal for Numerical Methods in Engineering, vol. 47, no. 8, pp. 13671379, 2000.
I. Kaljevic and S. Saigal, An improved element free Galerkin
formulation, International Journal for Numerical Methods in
T. Zhu and S. N. Atluri, A modifed collocation method and a
penalty formulation for enforcing the essential boundary conditions in the element free Galerkin method, Computational
Mechanics, vol. 21, no. 3, pp. 211222, 1998.
G. R. Liu and Y. T. Gu, A point interpolation method for twodimensional solid, International Journal for Numerical Methods
in Engineering, vol. 50, no. 4, pp. 937951, 2001.
Y. T. Gu and G. R. Liu, A local point interpolation method for
static and dynamic analysis of thin beams, Computer Methods
in Applied Mechanics and Engineering, vol. 190, no. 42, pp. 5515
5528, 2001.
G. R. Liu and Y. T. Gu, A local radial point interpolation
method (LRPIM) for free vibration analyses of 2-D solids,
Journal of Sound and Vibration, vol. 246, no. 1, pp. 2946, 2001.
G. R. Liu, G. Y. Zhang, Y. T. Gu, and Y. Y. Wang, A
meshfree radial point interpolation method (RPIM) for threedimensional solids, Computational Mechanics, vol. 36, no. 6,
pp. 421430, 2005.
[20] H. Wendland, Error estimates for interpolation by compactly

supported radial basis functions of minimal degree, Journal of
Approximation Teory, vol. 93, no. 2, pp. 258272, 1998.
[21] J. G. Wang and G. R. Liu, A point interpolation meshless
method based on radial basis functions, International Journal
for Numerical Methods in Engineering, vol. 54, no. 11, pp. 1623
1648, 2002.
[22] J. G. Wang and G. R. Liu, On the optimal shape parameters
of radial basis functions used for 2-D meshless methods,
Computer Methods in Applied Mechanics and Engineering, vol.
191, no. 23-24, pp. 26112630, 2002.
[23] X. Zhang and Y. Liu, Meshless Method, Tsinghua University
Press, Beijing, China, 2004.

http://dx.doi.org/10.1155/2014/918376
Research Article
Initial Time Difference Stability of Causal Differential Systems
in terms of Lyapunov Functions and Lyapunov Functionals
CoGkun Yakar1 and Mustafa Bayram Gcen2
1
2
Department of Mathematics, Gebze Technical University, Gebze, 141-41400 Kocaeli, Turkey

Department of Mathematics, Faculty of Sciences, Yldz Technical University, Istanbul, Turkey
Correspondence should be addressed to Coskun Yakar; cyakar@gyte.edu.tr

Received 9 June 2014; Revised 10 November 2014; Accepted 24 November 2014; Published 22 December 2014
Academic Editor: Zhilong L. Huang
Copyright 2014 C. Yakar and M. B. Gucen. Tis is an open access article distributed under the Creative Commons Attribution
cited.
We investigate the qualitative behavior of a perturbed causal diferential equation that difers in initial position and initial time
with respect to the unperturbed causal diferential equations. We compare the classical notion of stability of the causal diferential
systems to the notion of initial time diference stability of causal diferential systems and present a comparison result in terms
of Lyapunov functions. We have utilized Lyapunov functions and Lyapunov functional in the study of stability theory of causal
diferential systems when establishing initial time diference stability of the perturbed causal diferential system with respect to the
unperturbed causal diferential system.
1. Introduction
Lyapunovs second method is a standard technique used in
the study of the qualitative behavior of causal diferential
systems along with a comparison result [1] that allows the
prediction of behavior of a causal diferential system when
the behavior of the null solution of a comparison system is
known. However, there has been difculty with this approach
when trying to apply it to unperturbed causal diferential
systems and associated perturbed causal diferential systems
with an initial time diference. Te difculty arises because
there is a signifcant diference between initial time diference
(ITD) stability and the classical notion of stability. Te classical notions of stability are with respect to the null solution, but
ITD stability is with respect to unperturbed causal diferential
system where the perturbed causal diferential system and the
unperturbed causal diferential system difer in both initial
position and initial time [24].
In this paper, we have resolved this difculty and have a
new comparison result which again gives the null solution
a central role in the comparison diferential system. Tis
result creates many paths for continuing research by direct
application and generalization.
In Section 2, we present defnitions and necessary background material. In Section 3, we discuss and compare the
diferences between the classical notion of stability and
the recent notion of initial time diference (ITD) stability.
In Section 4, we have comparison theorems via Lyapunov
functions with initial time diference. In Section 5, we have
our main results of initial time diference stability criteria and
asymptotic stability via Lyapunov functions and we fnally
have our main result of initial time diference uniformly
asymptotic stability via Lyapunov functionals.
2. Preliminaries
In order to investigate the theory of stability for causal
diferential equations, we need comparison results in terms
of Lyapunov-like functions.
Consider the causal diferential systems
() = () () ,
() = () () ,
(0 ) = 0
(0 ) = 0
and the perturbed system of (2)

() = () () ,
(0 ) = 0
for 0 0
(1)
for 0 0
(2)
for 0 ,
(3)
46
where , : = ([0 , ], R ) is a causal operator and = [ R : < < ]. A special case of (3) is where ()() = ()() + ()() and
()() is the perturbation term. We assume the existence
and uniqueness of solutions () and () of (1) and (3)
through (0 , 0 ) and (0 , 0 ), respectively. We need to have
some classes of functions to utilize Lyapunov-like functions
for the generalized derivative of Lyapunov functions which
has to satisfy suitable conditions as follows:
= { : (, ()) () (, ()) ()
for 0 }
Defnition 3. For a real-valued function (, ) [R+

R , R+ ] one defnes the Dini derivatives as follows:
(a)
+ (, )
= lim+ sup
1
[ ( + , () + () ()) (, ())]
= lim inf
1
[ ( + , () + () ()) (, ())]
(, )
0
(4)
1 = { : (, ()) (, ()) for 0 } ,

where
(i) () > 0 is continuously diferentiable on R+ and
(0 ) = 1;
+ (, )
= lim+ sup
0
(ii) (, ) [R+ R , R+ ] is a Lyapunov function.
Before we can establish our comparison theorem and Lyapunov stability criteria for initial time diference we need to
introduce the following defnitions.
Defnition 1. Te solution (, 0 , 0 ) of system (3) through
(0 , 0 ) is said to be initial time diference stable with respect
0 , 0 ) = ( , 0 , 0 ), where (, 0 , 0 )
to the solution (,
is any solution of system (1) for 0 0, 0 R+ , and
= 0 0 if and only if given any > 0 there exist 1 =
1 (, 0 ) > 0 and 2 = 2 (, 0 ) > 0 such that
(, 0 , 0 ) ( , 0 , 0 ) <
whenever 0 0 < 1 , 0 0 < 2
for 0 .
(5)
If 1 , 2 are independent of 0 , then the solution (, 0 , 0 )

of system (3) is initial time diference uniformly stable with
respect to the solution (, 0 , 0 ). If the solution (, 0 , 0 )
of system (3) through (0 , 0 ) is initial time diference stable
and there exist 1 (0 ) > 0 and 2 (0 ) > 0 such that
lim (, 0 , 0 ) ( , 0 , 0 ) = 0
(6)
for all (, 0 , 0 ) and ( , 0 , 0 ) with 0 0 <

1 and |0 0 | < 2 for 0 then it is said to
be initial time diference asymptotically stable with respect
to the solution ( , 0 , 0 ). It is initial time diference
uniformly asymptotically stable with respect to the solution
( , 0 , 0 ) if 1 and 2 are independent of 0 .
Defnition 2. A function () is said to belong to the class K
if [(0, ), R+ ], (0) = 0, and () is strictly monotone
increasing in .
(7)
for (, ) R+ R .
(b) One defnes the generalized derivatives (Dini-like
derivatives) as follows:
1
[ ( + , ()
()
+ (() () (
) ()))
(, )
= lim inf
0
(, () ())]
(8)
1
[ ( + , ()
()
+ (() () (
) ()))
(, () ())]
for (, ) R+ R , where (, 0 , 0 ) is the solution

0 , 0 ) = ( , 0 , 0 ), where
of system (3) and (,
(, 0 , 0 ) is any solution of system (1) for 0 0,
0 R+ , and = 0 0 .
(c) Let (, ) [R+ , R+ ] be any Lyapunov
functional. One defnes its generalized derivatives as
follows:
+ (, () ())
= lim+ sup
0
[ ( + , ( + , , ) ( + , , ))
(, () ())]
(, () ())
= lim inf
0
(9)
[ ( + , ( + , , ) ( + , , ))
(, () ())] ,
where ( + , , ) is the solution of IVP (3) through (, )

+ , , )
is the solution of IVP through (, ),
()
=
and (
( ), where () = (, 0 , 0 ) is any solution of the IVP
(1), for 0 .
47
3. Causal Stability and New Notion of ITD

Causal Stability
3.1. Causal Stability. Let (, 0 , 0 ) be any solution of the
causal diferential system (1) with
() = () () ,
(0 ) = 0
for 0 0,
(10)
0 0 ))() = ((0 0 ) + (0 ))()

where ((
((0 ))(). Hence, we can assume, without loss of generality,
that (, 0 , 0 ) 0 is the null solution of system (10) and we
can limit our study of stability to that of the null solution [5].
However, it is not possible to do a similar transformation for
initial time diference stability analysis.
where : and is the set = [ R : < <

].
Assume that ()() = 0 for 0 0 so that 0 is
a null solution of (1) through (0 , 0). Now, we can state the
well-known defnitions concerning the stability of the null
solution.
3.2. New Notion of ITD Causal Stability. Let (, 0 , 0 ) be

0 , 0 ) = ( , 0 , 0 ) where
a solution of (2) and (,
(, 0 , 0 ) is any solution of system (1) for 0 0. Let us
make a transformation similar to that in (15). Set (, 0 , 0
0 ) = (, 0 , 0 ) ( , 0 , 0 ) for 0 . Ten
Defnition 4. Te null solution 0 of (1) is said to be stable

if and only if, for each > 0 and for all 0 R+ , there exists a
positive function = (, 0 ) that is continuous in 0 for each
such that
(17)
implies (, 0 , 0 ) <
for 0 .
(11)
If is independent of 0 , then the null solution 0 of (1) is

said to be uniformly stable.
Defnition 5. Te solution of (, 0 , 0 ) (10) through (0 , 0 )
is said to be stable with respect to the solution (, 0 , 0 ) of
(1) for 0 R+ if and only if given any > 0 there exists a
positive function = (, 0 ) that is continuous in 0 for each
such that
0 0 implies (, 0 , 0 ) (, 0 , 0 ) <
for 0 .
(12)
If is independent of 0 , then the solution of system (10) is

uniformly stable with respect to the solution (, 0 , 0 ).
We remark that, for the purpose of studying the classical
stability of a given solution (, 0 , 0 ) of system (10), it is
convenient to make a change of variable. Let (, 0 , 0 ) and
(, 0 , 0 ) be the unique solutions of (1) and (10), respectively,
and set
(, 0 , 0 0 ) = (, 0 , 0 ) (, 0 , 0 )
for 0 . Ten
(, 0 , 0 0 ) = (, 0 , 0 ) (, 0 , 0 )
(0 0 ) ()
= ( (0 0 ) + (0 )) () ( (0 )) ()
(0 0 )) () .
(0 0 ) () = (
(13)
(14)
(, 0 , 0 0 ) = (, 0 , 0 ) ( , 0 , 0 )
(0 0 )) (; ) .
(0 0 ) () = (
(18)
One can observe that even if 0 = 0 , (, 0 , 0) is not zero

and is not the null solution of the transformed system and the
solution ( , 0 , 0 ) does not correspond to the identically
0 0 ))(; ). Consequently,
zero solution of () = ((
stability properties of null solution cannot be used in order
to fnd ITD stability properties using this approach.
4. Comparison Theorems via Lyapunov

Functions with Initial Time Difference
In our earlier work and in the work of others [1, 2, 5],
the diferences between the classical notion of stability and
ITD stability did not allow the use of the behavior of the
null solution in our ITD stability analysis. Te main result
presented in this section resolves those difculties with a
new approach that allows the use of the stability of the null
solution of the comparison system to predict the stability
properties of (, 0 , 0 ) the solution of (3) with respect to
()
= ( , 0 , 0 ) where (, 0 , 0 ) is any solution of
system (1).
Teorem 6. Assume that
(i) (, ) [R+ R , R+ ] and (, ) is locally
Lipschitzian in ;
(ii) for 0 and 1 ,
(, (, () ())) ,
+ (, )
(19)
where
(15)
(16)
It is easy to observe that (, 0 , 0 0 ) 0 is a solution

of the transformed system if 0 0 = 0 which implies
(0)() = 0 for 0 0. Since () = 0 and (, 0 , 0) = 0
is the null solution, the solution of (, 0 , 0 ) (1) corresponds
0 0 ))()
to the identically null solution of () = ((
+ (, () ())
= lim+ sup
0
1
[ ( + , () ()
)
()))
+ (() () (
(, () ())]
(20)
(, ) R+ R and [R+ R+ , R+ ];
with (, ),
48

(iii) () = (, 0 , 0 ) is the maximal solution of the scalar
diferential equation
= (, ) , (0 ) = 0 0
existing on [0 , ).
(21)
Ten, if ()
= ( , 0 , 0 ), = 0 0 , where
(, 0 , 0 ) is any solution of the causal diferential system (1)
and () = (, 0 , 0 ) is any solution of causal diferential
system (3) existing on [0 , ) such that (0 , 0 0 ) 0
implies
(, () ()) ()
0 .
(22)
= ( , 0 , 0 ), = 0 0 , and let (, 0 , 0 )
Proof. Let ()
be any solution of the causal diferential system (1) and let
() = (, 0 , 0 ) be any solution of causal diferential system
(3) for 0 . Defne
() = (, () ())
for 0
(23)
0 )) 0 . For some
so that (0 ) = (0 , (0 ) (
sufciently small > 0, consider the diferential equation
= (, ) + ,
(0 ) = 0 +
for 0
(24)
whose solutions (, ) = (, 0 , 0 , ) exist as far as (). In

order to prove the conclusion of the theorem, we need to show
that
() = (, () ()) < (, )
for 0 .
(25)
Te standard computation for small enough > 0 implies
( + ) ()
= ( + , ( + ) ( + )) (, () ())
( + , ( + ) ( + ))
)
()))
( + , () () + (() () (
)
()))
+ ( + , () () + (() () (
(, () ())
[ ( + ) ()] [ ( + ) ()]
)
())
(() () (
)
()))
+ ( + , () () + (() () (
(, () ()) .
Since (, ) is locally Lipschitzian in and > 0 is the

Lipschitzian constant, we obtain
( + ) ()
[1 () 2 ()]
)
()))
+ ( + , () () + (() () (
for 1 > 0 ,
(1 ) = (1 , ) .
where 1 and 2 are error terms. Tis shows that

+ ()
1
= lim+ sup [1 () 2 ()]
0
+ lim+ sup
0
(26)
Consequently, ( )()
1 .
for 1 0 .
)
()))
+ (() () (
(31)
(, (, () ()))
= (, ())
for 0 1 <
since (1 () 2 ())/ 0 as 0. Hence, at = 1 , we

have
(27)
By using the assumption on and (, ) 0, the solutions
(, ) are nondecreasing in . Since () = (, () ())

for 0 , we get
(, () ()) (1 , )
1
[ ( + , () ()
(, () ())]
It then follows that

+ (1 ) (1 , ) = (1 , (1 , )) + .
(30)
(, () ()) ,
If this is not true, then there exists a 1 0 such that

() < (, )
(29)
(28)
+ (1 ) + (1 , (1 ) (1 ))
(1 , (, (1 ) (1 ))) = (1 , (1 ))
< (1 , (1 , )) +
(32)
which contradicts (27). Hence () < (, ), which yields the

desired estimate as 0:
() = (, () ()) (, 0 , 0 )
Terefore these complete the proof.
for 0 . (33)
Journal of Applied
AppliedMathematics
Mathematics
14
5
Corollary 7. Let satisfy the conditions of Teorem 6 with

(, ) 0 and () 1 . Ten
(, () ()) (0 , 0 0 )
0 ,
(34)
where () and () are any solutions of the initial value

problems (1) and (2), respectively. Equivalently we have
(2 , (2 ) (2 )) (1 , (1 ) (1 ))
0 1 2 < .
Teorem 8. Assume that
(35)
(, (, () ()) ()) ,
= lim+ sup
0
)
()))
+ (() () (
(37)
(, ) R+ R and [R+ R+ , R+ ];
with (, ),
(iii) () = (, 0 , 0 ) is the maximal solution of the scalar
diferential equation
existing on [0 , ).
(38)
= ( , 0 , 0 ), = 0 0 , where () is any
Ten, if ()
solution of the system (1) and () = (, 0 , 0 ) is any solution
of the system (3) existing on [0 , ) such that (0 ) (0 , 0
0 ) 0 implies
(, () ()) () ()
0 .
(39)
Proof. Defne (, () ())

= (, () ())().
Let
0 and () . Ten it is easy to see that
+ (, () ())
() + (, () ()) + (, () ()) ()
(, (, () ()) ())
= (, (, () ())) .
(40)
Ten, by the application of Teorem 6, it follows that
(, () ()) () = (, () ()) ()
Tese complete the proof.
(42)
for 0 .
1
[ ( + , () ()
)
()))
+ (() () (
(43)
(, () ())]
(, () ())]
(0 ) = 0 0
0,
+ (, )
+ (, () ())
1
[ ( + , () ()
= (, ) ,
(ii) for 0 and 1 ,

where
+ (, () ())
0
(i) Let (, ) [R+ (), R+ ] and let (, ) be locally

Lipschitzian in ;
(36)
where
= lim+ sup
5.1. Initial Time Diference Stability Criteria via Lyapunov

Functions. We will give sufcient conditions for the stability
of the unperturbed systems () = (, 0 , 0 ) of (1) in terms
of Lyapunov functions and we assume that the solutions of
causal diferential systems of (1) exist and are unique for
0 .
Teorem 9. Assume the following.
(i) (, ) [R+ R , R+ ] and (, ) is locally Lipschitzian in ;

(ii) for 0 and ,
() + (, () ()) + (, () ()) ()
5. Initial Time Difference Stability of Causal

Differential Systems
(, ) R+ ();
with (, ),
(iii) let (, ) be positive defnite and decrescent on R+

() : (, ) satisfes
() (, ) () ,
(, ) R+ () ,
, K.
(44)
Ten the solution of perturbed causal diferential system

(, 0 , 0 ) of (3) is initial time diference stable with respect
to the solution of unperturbed causal diferential system
0 , 0 ) = ( , 0 , 0 ), where (, 0 , 0 ) is any solution
(,
of system (1), for 0 .
Proof. Let > 0 and 0 be given. Choose 1 (, 0 ) > 0
and 2 (, 0 ) > 0 such that () < () where (, 0 ) =
max [1 (, 0 ), 2 (, 0 )] > 0. Ten, we claim that, with this
, stability of the solution (, 0 , 0 ) of unperturbed causal
diferential system of (1) follows for 0 . If this is not true,
0 , 0 )
then there would exist solutions (, 0 , 0 ) of (3), (,
for 0 , and 2 > 1 > 0 satisfying
(1 ) (1 ) = ,
() ()
(2 ) (2 ) = ,
for 1 2 .
(45)
Ten, by using assumption (ii) and Corollary 7, we get the

estimate
(41)
(2 , (2 ) (2 )) (1 , (1 ) (1 ))
for 1 2 < .
(46)
50
Terefore, using (45) and assumption (iii), together with

the choice of , implies
() = ( (2 ) (2 )) (2 , (2 ) (2 ))
(1 , (1 ) (1 )) ( (1 ) (1 ))
(47)
= () < ()
which is a contradiction. Hence the solution (, 0 , 0 ) of

unperturbed causal system of (1) is stable for 0 .
If 1 = 1 (, 0 ) > 0 and 2 = 2 (, 0 ) > 0 are independent of 0 , then the solution (, 0 , 0 ) of unperturbed causal
system of (1) is uniformly stable for 0 .
(i) Let (, ) [R+ (), R+ ] and let (, ) be locally
Lipschitzian in ;
(ii) for 0 and ,
+ (, () ()) () + (, () ()) () 0,
(48)
where () is continuously diferentiable for 0 with
(0 ) = 1, () 1, and () as ,
+ (, () ())
= lim+ sup
0
1
[ ( + , () ()
)
()))
+ (() () (
(49)
(, ) R+ ();
with (, ),
(iii) let (, ) be positive defnite and decrescent on R+

() : (, ) satisfes
(, ) R+ () ,
, K.
(50)

(, 0 , 0 ) of (3) is initial time diference asymptotically stable
with respect to the solution of unperturbed causal diferential
0 , 0 ) = ( , 0 , 0 ), where (, 0 , 0 ) is any
system (,
solution of system (1), for 0 .
Proof. By applying Teorem 8, we get
(, () ()) () (0 , 0 0 )
( () ())
(, () ()) (0 , 0 0 ) 1 ()
(0 0 ) 1 () < (0 ) 1 () <
(52)
for 0 + since 1 () does exist and 1 () = 1/()

0 as . Hence the solution (, 0 , 0 ) of unperturbed
causal system of (1) is quasiasymptotically stable for 0 .
Terefore the solution of perturbed causal diferential system
(, 0 , 0 ) of (3) is initial time diference asymptotically
stable with respect to the solution of unperturbed causal
0 , 0 ) = ( , 0 , 0 ) or the
diferential system (,
solution (, 0 , 0 ) of unperturbed causal system of (1) is
asymptotically stable for 0 .
If 1 = 1 (, 0 ) > 0, 2 = 2 (, 0 ) > 0 and = (, 0 ) >
0 are independent of 0 , then the solution (, 0 , 0 ) of
unperturbed causal system of (1) is uniformly asymptotically
stable for 0 .
5.2. Initial Time Diference Stability Criteria via Lyapunov
Functionals. In this section, we employ Lyapunov functional
for discussing initial time diference stability theory in the
context of causal diferential equations.
We will give a result parallel to the original Lyapunov
second method on uniformly asymptotic stability with initial
time diference.
(, () ())]
() (, ) () ,
that, given any > 0 and 0 0, there exists a = (, 0 ) > 0

satisfying
for 0 . (51)
By Teorem 9, we have the solution (, 0 , 0 ) of unperturbed causal system of (1) which is stable for 0 . We
only have to prove quasiasymptotic stability. In order to do
that, let = so that 0 = (, 0 ) where (, 0 ) =
max [1 (, 0 ), 2 (, 0 )] > 0. Choose 0 0 < 0 and
|0 0 | < 2 . Ten, in view of (ii), (iii), and (51), it follows
(i) Let (, ) [R+ , R+ ] and + (, () ())
(() ()),
K;
(ii) let (, ) be positive defnite and decrescent on R+ :

(, ) satisfes
() (, ) () ,
(, ) R+ ,
, K.
(53)

(, 0 , 0 ) of (3) is initial time diference uniformly asymptotically stable with respect to the solution of unperturbed
0 , 0 ) = ( , 0 , 0 ), where
causal diferential system (,
(, 0 , 0 ) is any solution of system (1), for 0 .
Proof. Let > 0 and 0 be given. Choose 1 (, 0 ) =
1 () > 0 and 2 (, 0 ) = 2 () > 0 such that
() < () ,
(54)
where () = min [1 (), 2 ()] > 0. Ten, we have that, with

this and , stability of the solution (, 0 , 0 ) of unperturbed
causal diferential system of (1) follows for 0 . If this is
51
not true, then there would exist solutions (, 0 , 0 ) of (3),

0 , 0 ) for 0 , and 1 > 0 satisfying
(,
(1 , 0 , 0 ) (1 , 0 , 0 ) = ,
(, 0 , 0 ) (, 0 , 0 )
(55)
for 0 1 .
Ten, by using assumption (ii) and Corollary 7, we get the

estimate
(, (, 0 , 0 ) (, 0 , 0 )) (0 , 0 0 )
for 0 1 < .
(1 , (1 , 0 , 0 ) (1 , 0 , 0 ))
(0 , 0 0 ) (0 0 ) = () < ()
(57)
(58)
In view of uniform stability, it is easy to show that there exists

a , 0 < < 0 +, where = 1+(0 )/() and 0 0 <
, 0 , 0 ) < ().
0 and |0 0 | < 0 and ( , 0 , 0 ) (
Here = () corresponds to > 0 for uniform stability. If it
0 , 0 ), [0 , 0 + ]. Ten,
is not, let (, 0 , 0 ) (,
by assumption (i), we get
(, (, 0 , 0 ) (, 0 , 0 ))
(0 , 0 0 ) ( (, 0 , 0 ) (, 0 , 0 ))
(59)
for [0 , 0 + ]. As a result of this
0 (0 + , (0 + , 0 , 0 ) (0 + , 0 , 0 ))
(0 , 0 0 )
0 +
(0 ) () < 0
for 0 + .
Tis completes the proof.

for 0 .
(61)
which is a contradiction. Hence the solution (, 0 , 0 ) of

unperturbed causal system of (1) is uniformly stable for 0
as follows.
To prove uniformly asymptotic stability, set = and
designate 0 = () so that
0 0 < 0 ,
0 0 < 0
implies (, 0 , 0 ) ( , 0 , 0 ) <
0 0 < ,
0 0 < ,
(, 0 , 0 ) ( , 0 , 0 ) <
(56)
Terefore, using (54), (55), (56), and assumption (ii), together

with the choice of , yields
() = ( (1 , 0 , 0 ) (1 , 0 , 0 ))
by the defnition of . Tis contradiction shows that there

exists a > 0 such that ( , 0 , 0 ) ( , 0 , 0 ) < .
Tis implies, by stability, that
( (, 0 , 0 ) (, 0 , 0 ))
(60)
Acknowledgments
Tis work has been supported by Te Yldz Technical University and Yeditepe University Department of Mathematics
and Te Scientifc and Technological Research Council of
Turkey. Te authors also would like to thank V. Lakshmikantham and the referees for their insightful comments and
detailed suggestions which improved the quality of the paper.
References
[1] V. Lakshmikantham, S. Leela, Z. Drici, and F. A. McRae, Teory
of Causal Diferential Equations, vol. 5 of Atlantis Studies in
Mathematics for Engineering and Science, 2010.
[2] V. Lakshmikantham and S. Leela, Diferential and Integral
Inequalities, vol. 1, Academic Press, New York, NY, USA, 1969.
[3] V. Lakshmikantham and A. S. Vatsala, Diferential inequalities
with initial time diference and applications, Journal of Inequalities and Applications, vol. 3, no. 3, pp. 233244, 1999.
[4] M. D. Shaw and C. Yakar, Generalized variation of parameters
with initial time diference and a comparison result in term
Lyapunov-like functions, International Journal of Nonlinear
Diferential Equations, Teory Methods and Applications, vol. 5,
pp. 86108, 1999.
[5] F. Brauer and J. Nohel, Te Qualitative Teory of Ordinary
Diferential Equations, W.A. Benjamin, New York, NY, USA,
1969.

http://dx.doi.org/10.1155/2014/918376
Research Article
Bifurcation of Safe Basins and Chaos in Nonlinear Vibroimpact
Oscillator under Harmonic and Bounded Noise Excitations
Rong Haiwu,1 Wang Xiangdong,1 Luo Qizhi,1 Xu Wei,2 and Fang Tong2
1
2
Department of Mathematics, Foshan University, Foshan 528000, China

Department of Applied Mathematics, Northwestern Polytechnical University, Xian 710072, China
Correspondence should be addressed to Rong Haiwu; ronghw@foshan.net

Received 10 August 2014; Revised 5 October 2014; Accepted 7 October 2014; Published 21 December 2014
Academic Editor: Qingdu Li
Copyright 2014 Rong Haiwu et al. Tis is an open access article distributed under the Creative Commons Attribution License,
Te erosion of the safe basins and chaotic motions of a nonlinear vibroimpact oscillator under both harmonic and bounded random
noise is studied. Using the Melnikov method, the systems Melnikov integral is computed and the parametric threshold for chaotic
motions is obtained. Using the Monte-Carlo and Runge-Kutta methods, the erosion of the safe basins is also discussed. Te sudden
change in the character of the stochastic safe basins when the bifurcation parameter of the system passes through a critical value
may be defned as an alternative stochastic bifurcation. It is founded that random noise may destroy the integrity of the safe basins,
bring forward the occurrence of the stochastic bifurcation, and make the parametric threshold for motions vary in a larger region,
hence making the system become more unsafely and chaotic motions may occur more easily.
1. Introduction
Nonsmooth factors arise naturally in engineering applications, such as impacts, collisions, and dry frictions [1].
A considerable amount of research activities have focused
on nonsmooth dynamical systems, including vibroimpact
systems, collisions dynamics, chattering dynamics, and stickslip motions, in recent years. In practice, engineering structures are ofen subjected to time dependent loadings of
stochastic nature, such as the natural phenomena due to wind
gusts, earthquakes, ocean waves, and random disturbance
or noise which always exists in a physical system. Te
infuence of random excitation on the dynamical behavior of
a vibroimpact dynamical system has caught the attention of
many researchers. Many efective methods have been developed, for example, linearization method used by Metrikyn
[2], quasistatic approach method used by Stratonovich [3],
exponential polynomial ftting method proposed by Zhu
[4], Markov process method used by Jing and Young [5],
stochastic averaging method used by Xu et al. [6, 7], variable
transformation method used by Zhuravlev [8], energy balance method used by Iourtchenko and Dimentberg [9], mean
impact Poincare map method used by Feng and He [10], path
integration method used by Iourtchenko and Song [11], and
numerical simulation method used by Dimentberg et al. [12].

In [13], the authors tried to review and summarize the existing
methods, results, and literature available for solving problems
of stochastic vibroimpact systems. However, most researches
focused on responses statistics, such as statistic moment and
probability density function of the vibroimpact oscillator, and
few are focused on the bifurcations and chaos of the stochastic
vibroimpact dynamical systems.
It is well known that, by calculating the distance between
the stable and unstable manifold, Melnikovs method [14]
gives a powerful approximate tool for investigating chaotic
motions in deterministic smooth system. However, classical
Melnikovs method is not directly appropriate in the nonsmooth system. Some efective Melnikovs methods have been
proposed for deterministic piecewise smooth dynamical systems [15, 16] and nonlinear vibroimpact dynamical systems
[17]. To our knowledge, few Melnikovs methods are well
developed for the stochastic vibroimpact dynamical systems.
In this paper, the bifurcation of safe basins and chaos
of a nonlinear vibroimpact oscillator under both harmonic
and bounded noise excitation are investigated. Te impact
considered here is an instantaneous impact with restitution
factor . Te paper is organized as follows. In Section 2,
Melnikovs method is extended to the analysis of homoclinic
53
bifurcation in the stochastic vibroimpact system, and the

conditions for the onset of chaos are derived in the mean
square value sense. In Section 3, the variations of safe basins
are presented numerically when one changes the amplitude
of the harmonic excitation both in the deterministic and
stochastic cases. Conclusions are presented in Section 4.
2. The Stochastic Melnikov Function

2.1. Teoretical Analysis. Consider a classical Dufng
vibroimpact oscillator with bilateral constrains under both
harmonic and bounded random noise excitations governed
by the following equation:
+ 3 = ( + 1 cos 1 + 2 ()) ,
+ = (1 ) ,
|| = 1,
|| < 1,
(1)
where the dot indicates diferentiation with respect to time

, is a small scale, 0 < 1, represents the
intensity of the nonlinear term, is the damping coefcient,
1 and 1 are amplitude and frequency of the deterministic
excitation, respectively, and 0 < 1 is the restitution
factor to be a known parameter of impact losses, = 1
, whereas subscripts minus and plus refer to value
of response velocity just before and afer the instantaneous
impact. Tus + and are actually rebound and impact
velocities of the mass, respectively. Tey have the same
magnitude whenever = 1; therefore this special case is
that of elastic impacts, whereas in case < 1 some impact
losses are observed. 2 denotes the intensity of the random
excitation and () is a bounded noise process, which was
introduced by Stratonovich [18]:
() = cos (2 + () + ) ,
(2)
where 2 > 0 is the frequency of the random excitation, ()

is a standard Wiener process, and a uniformly distributed
random number in (0, 2). ()

is the random disorder
which describes random temporal deviations of the excitation
frequency from its expectation or mean 2 . Te process ()
has the following power spectral density [19]:
() =
1
1
2
+
].
[
2 4 ( 2 )2 + 4 4 ( + 2 )2 + 4
(3)
Obviously |()| 2 is a bounded random process. Periodicin-time excitation with a random disorder, or random phase
modulation appears in structural dynamics with traveling
loadings and/or structures, having certain imperfect spatial
periodicity in certain problems of aeroelasticity [19]. Tis
process will be assumed to be narrow-band, which is clearly
seen to be in the case provided that 0. We assume
that 2 in this paper. Te bounded noise model
(2) is a suitable model and many researches have been
done on the responses of nonlinear system under bounded
noise excitation [20, 21]. Obviously, bounded noise () has
continuous sample function; hence Melnikovs method may

be used in the stochastic system (1).
Te physical model of (1) can be viewed as the motion
of a mass with harden stifness under both harmonic and
bounded random noise external excitations, while two collision obstacles are placed before and afer the equilibrium
position with distance 1.
When = 0, system (1) reduces to an unperturbed
deterministic vibroimpact system. Using stability analysis,
one collects two center-type fxed points (1/, 0) and
(1/, 0), and a saddle-type fxed point (0, 0). Denotes =
; two homoclinic orbits connecting the two center-type fxed
points and saddle-type fxed point are
2
( , ) = ( sech ( 0 ) ,
2
sech ( 0 ) tanh ( 0 )) ,
2
( , ) = ( sech ( + 0 ) ,
0,
(4)
2
sech ( + 0 ) tanh ( + 0 )) ,
0,
where 0 = arcsech(/2).
Equation (4) is exactly the solution of the unperturbed
deterministic system (1) as = 0 without vibroimpact,
and similar system has been discussed in [17] by using the
Melnikov method. Using the results in [17, 22], one obtains
the stochastic Melnikov function for homoclinic orbits of
system (1):
() = 1 () + 2 ()
+ 1 3 () + 2 () ,
(5)
where 1 (), 2 (), and 3 () are the deterministic

parts of the random process (); they are caused by the
deterministic harmonic excitation, and () represents the
stochastic term which is caused by the bounded noise ():
1 () =
1
2
( 3 ) = 21 ,
2
1 (/2) 0
2 () =
0
+
1
[ 2 () + 2 () ]
1 (/2)
0
2
1 (/2)
sech2 ( 0 ) tanh2 ( 0 )
sech2 ( + 0 ) tanh2 ( + 0 ) ]

=
4
(1 tanh3 0 )
3 1 (/2)
4.5
4
3/2
=
[1 (1 ) ] ,
2
31 (/2)
3.5
0
1
[ cos (1 + 1 ) ()
1 (/2)
cos (1 + 1 ) () ]
1.5
1
0.5
cos (1 + 1 1 0 ) sech () tanh () ]
1
sin (1 )
(2/) 1
() ( + ) ] .
(6)
According to the dynamic theory, the stable and the

unstable manifolds will intersect transversely with each other
which means chaos will occur when there exist simple zeros
in Melnikov function (5) in the deterministic case 2 = 0.
However, in the stochastic case 2 = 0, the Melnikov function
measures the random distance between the stable and the
unstable manifolds. In this case, the threshold value for the
rising of the chaotic motion depends on the property of the
random excitation process and may deviate from the one
for the deterministic case. In order to analyze the simple
zero points of () in the statistics sense, one considers the
following equations:
2
2
,
[ () [ ()]]2 = [2 ()] = 22
[0 [ ()]]2 = [ ()]2 ,
0.4
0.3
(8)
0.5
f2
0.6
0.7
0.8
0.9
|()|2 ()
2 +
1
| ()|2 [
2
2
4 ( 2 ) + 4
() =
= 2
4 ( + 2 ) + 4
() exp () +
] ,
() exp ()
2
sech ( 0 ) tanh ( 0 ) exp ()
2
sech ( + 0 ) tanh ( + 0 ) exp ()
2 +
sech () tanh () sin ( 0 ) .
0
(9)
Ten, from (8), the condition of the occurrence of chaotic

motions of system (1) is
1 > 1 =
1 () + 2 () + 2
3 ()
(7)
where represents the mathematics expectation. Since

[ ()] = 0, the condition for the onset of chaotic motion
in the mean square value sense is
= (1 () + 2 () + 1 3 ()) ,
0.2
Figure 1: Chaotic area of system (1), : theoretical solution and

: numerical solution.
() ( + )
= [ ()]
0.1
sin (1 1 0 ) sech () tanh () ,
where
1
() =
1 (/2)
2
22
cos (1 + 1 + 1 0 ) sech () tanh ()
Nonchaotic area
2.5
2
(2/) 1
Chaotic area
f1
3 () =
54
(10)
2.2. Numerical Results. Now we give some numerical results

to verify the analytic conditions given by (8) and (10). Te
parameters of system (1) are given by
= 1.4,
= 2.4,
1 = 1.5,
= 0.1,
2 = 1.0,
= 1.8,
= 0.4.
(11)
Te variation of the threshold value 1 versus the bounded

noise amplitude 2 is plotted in Figure 1 for the onset of
55

0.6
1
0.8
0.4
0.2
0.4
Velocity y(t)
Displacement x(t)
0.6
0.2
0
0.2
0.4
0
0.2
0.4
0.6
0.6
0.8
1
800
820
840
860
880
900
920
940
960
980 1000
Times
(a) Time history of ()
0.8
1
0.2
0.8 0.6 0.4 0.2 0
Displacement x(t)
0.4
0.6
0.8
(b) Phase plot
= 2.5).
Figure 2: Numerical results of (1) (1 = 2.5, 2 = 0.0, (0) = 0.8, (0)
chaotic monition in system (1), the solid line represents the

analytic results given by (10), and the dashed line represents
the numerical simulations according the criterion of the
largest Lyapunov exponent using the algorithm presented by
Wolf et al. [23]. Usually, calculating the Lyapunov exponents
is regarded as the simplest method to verify the existence
of chaotic behavior. However, the Melnikov method and the
erosion of the safe basins are mainly discussed in this paper,
while the largest Lyapunov exponent is only used to verify
the efciency of the above two methods. Te largest Lyapunov
exponent of system (1) is positive in the area above the dashed
line, which means the occurrence of the chaotic motion,
while the largest Lyapunov exponent of system (1) is negative
in the area below the dashed line, which means no occurrence
of the chaotic motion. It can be seen clearly from Figure 1 that
Melnikovs condition (10) is only a necessary condition for the
occurrence of chaotic motion in the Lyapunov sense. It can
also be seen that the threshold value 1 will decrease when 2
increase thus make the chaotic motions occur more easily.
Tree diferent simulation points in Figure 1 are chosen
to compare the analytic results with the numerical results.
One point is (1 , 2 ) = (2.5, 0.0) in the deterministic
case, which is below the analytic critical value 1 = 3.41,
Melnikovs criterion implies that the motion of system (1) is
regular, and numerical simulations show that there are two
coexisting attractors with the same largest exponent 0.351.
Te corresponding phase portraits and time history portraits
are shown in Figures 2 and 3 for diferent initial values. Te
response time history of system (1) and the phase plot are
shown in Figure 2 in the case (0) = 0.8, (0)

= 2.5.
Clearly, the response is a period one while the phase trajectory
is a limit cycle.
Te response time history of system (1) and the phase plot
= 2.5.
are shown in Figure 3 in the case (0) = 0.4, (0)
Figures 2 and 3 show that the stable steady state solution
of a deterministic vibroimpact system may be diferent for
diferent initial values; such interesting phenomenon also
exists for stochastic vibroimpact system. Another simulation

point is (1 , 2 ) = (2.5, 0.05) in the stochastic case, which
is also below the analytic critical value 1 = 3.38, and
the largest Lyapunov exponent is 0.337. Te corresponding
phase portraits and time history portraits are shown in
Figures 4 and 5 for diferent initial values. Te response time
history of system (1) and the phase plot are shown in Figure 4
= 2.5. Clearly, the response is a
in the case (0) = 0.8, (0)
quasiperiod one while the phase trajectory is a difused limit
cycle.
Te response time history of system (1) and the phase plot
= 2.5.
are shown in Figure 5 in the case (0) = 0.4, (0)
Figures 4 and 5 show that the stationary response of
system (1) may be diferent for diferent initial values in
some parameter area. Te random noise () will change the
steady-state response of system (1) from a period solution
to a quasiperiod solution. In fact, due to the existence of
random noise, the motion may jump from one stable position
to another with some, maybe small, probability. However, it
may take a long time to observe such phenomenon, but the
simulation time is always limited in the practical numerical
simulation, so one may give the conclusion only base on fnite
time observed phenomenon.
It is well known, from the theory of nonlinear oscillation,
that if an oscillator with hardening nonlinear stifness is
subjected to sinusoidal excitation, the response may exhibit
sharp jumps in amplitude. Tis jump behavior is associated
with the fact that, over a range of the values of the ratio
of excitation frequency to the natural frequency of the
degenerated linear oscillator, the response amplitude is triplevalued. Terefore the system should have two stationary
responses which depend on the initial condition. However,
it is a disputable problem whether there are more than one
stationary response if an oscillator with hardening nonlinear
stifness is subjected to random excitations [24]. Here, it is
found that there are more than one stationary response in a
nonlinear stochastic vibroimpact system.
56
0.6
0.8
0.4
0.2
0.4
Velocity y(t)
Displacement x(t)
0.6
0.2
0
0.2
0.4
0
0.2
0.4
0.6
0.6
0.8
1
800
820
840
860
880
900
920
940
960
0.8
1
980 1000
Times
0.8 0.6 0.4 0.2 0

0.2 0.4
Displacement x(t)
0.6
0.8
0.6
0.8
(b) Phase plot
= 2.5).
0.6
1
0.8
0.4
0.2
0.4
0.2
Velocity y(t)
Displacement x(t)
0.6
0
0.2
0.4
0
0.2
0.4
0.6
0.6
0.8
1
800
820
840
860
880
900 920
Times
940
960
980 1000
0.8
1
0.8 0.6 0.4 0.2
0.2
0.4
Displacement x(t)
(b) Phase plot
= 2.5).
Te third simulation point is (1 , 2 ) = (5.5, 0.05) in the

stochastic case, which is above the analytic critical value 1 =
3.38, the largest Lyapunov exponent is 0.062. Te response
time history of system (1) and the phase plot are shown in
Figure 6 in the case (0) = 0.8, (0)

= 2.5. Clearly, the
response is a chaotic one.
3. Bifurcation of Safe Basins and Chaos

Alternative to the Melnikov function and Lyapunov exponent
method, there is another method to identify the rising of
chaos. One of these is to determine the global structures of
the system and one of these global structures is the boundary
of safe basin. Te safe basin boundaries of attractors are
usually fractal and naturally incursive since the coexistence of

period and chaotic attractors. Tey are related to homoclinic
or heteroclinic intersections of stable and unstable manifolds
of the saddle points in the system and chaos ofen arises in
such system. Te decrease of the safe basins area is called
basin erosion and will be discussed in this section.
In some time the limitation of the vibration amplitude
may be more important, since the structure of the system
will be destroyed when the amplitude of the vibration passes
through a critical value and thus leads to the researches of the
safe basins [25, 26]. Tere are some relations between erosion
of safe basins and chaotic motions of the system. When the
safe basin is eroded, the boundary of the sage basin will have
fractal structures, and the motions initial from some points
57

1
0.6
0.8
0.4
0.2
0.4
Velocity y(t)
Displacement x(t)
0.6
0.2
0
0.2
0.4
0
0.2
0.4
0.6
0.6
0.8
1
800
820
840
860
880
900 920
Times
940
960
0.8
1
980 1000
0.8 0.6 0.4 0.2 0

0.2 0.4
Displacement x(t)
0.6
0.8
0.6
0.8
(b) Phase plot
0.8
0.8
0.6
0.6
0.4
0.4
Velocity y(t)
Displacement x(t)
= 2.5).
0.2
0
0.2
0
0.2
0.2
0.4
0.4
0.6
0.6
0.8
0.8
1
800
820
840
860
880
900
920
940
960
980 1000
Times
1
1
0.8 0.6 0.4 0.2 0

0.2 0.4
Displacement x(t)
(b) phase plot
= 2.5).
within the safe basin will be chaotic. According to [25, 26], the
safe basins of the system may be defned using a bounded area
in the space of phase trajectories. Te trajectory starting
from the safe basins will be stay in the area when the time
tends to infnity. Otherwise, the trajectory starting beyond the
safe basins will escape the area ; such trajectory is unstable
and may destroy or collapse the system. Te structure of the
safe basins is similar to some attractor basins. Te acreage and
shape of the safe basins will change when the parameter of the
system changes.
3.1. Bifurcation without Bounded Noises. In this paper,
the evolution of the safe basins of system (1) is studied
numerically when the parameter 1 changes its value in the
deterministic case 2 = 0, frstly. In the numerical simulation,

the parameters in system (1) are chosen as:
= 0.8,
1 = 1.5,
= 0.9,
2 = 1.0,
= 0.1,
= 0.4,
Te bounded area is defned as follows:
= 1.8,
2 = 0.
= {(, ) : 1 1, 0.8 0.8} ;
(12)
(13)
then is divided into 100 100 lattices and the lattice points
are taken as the initial values for the solutions of system (1).
If the solution of system (1) stays in area for a long enough
time up to = 2000, such a solution can be approximately

0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
(b) 1 = 1.00, 2 = 0.00, = 0.7500
0.8
0.8
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.6
0.4
0.4
0.4
0.2
0.2
(h) 1 = 4.50, 2 = 0.00, = 0.1776
Figure 7: Erosion of safe basins in system (1): (2 = 0).
taken as a safe solution, and the corresponding lattice may

be taken as part of the safe basins; if the solution of system
(1) escapes the area , such a solution is taken as an unsafe
solution, and the corresponding lattice is beyond the safe
basins. Governing equation (1) is numerically integrated by
the fourth-order Runge-Kutta algorithm, and the numerical
results are shown in Figure 7(a) to Figure 7(i). Te black
region denotes the safe basins while the blank region
represents the unsafe area in Figures 7 and 8.
Te safe basins shown in Figure 7(a) are a densely packed,
integral one, while the safe basins shown in Figure 7(b) to
Figure 7(i) are eroded ones. Calculation results show that
in the case when 1 = 0.21, the boundary of the
safe basins of system (1) are smooth without any erosion as
0.4
0.6
0.8
0.6
0.4
0.2
0.6
0.4
0.2
0.2
0.4
0.6
0.8
0.6
0.4
0.2
0.2
0.4
0.8
0.6
0.8
0.6
0.8
0.8
0.6
1
0.4
0.6
0.8
0.4
0.6
0.4
0.4
0.2
0.8
0
0.2
0.2
0.2
0.4
0.2
y
0
0.2
0.8
0.6
0.2
0.8
0.8
0.4
0.6
(f) 1 = 4.20, 2 = 0.00, = 0.4199
0.6
(g) 1 = 4.30, 2 = 0.00, = 0.3937
(e) 1 = 4.00, 2 = 0.00, = 0.4450
0.8
0.8
0.8
0.6
0.4
0.2
0.8
0.2
0.4
0.6
0.2
0.8
0.8
0.8
0.6
0.8
0.4
0.8
0.2
0.6
0.6
0.2
0.4
0.6
0.4
0.4
0.8
0.4
0.6
0.2
0.2
(d) 1 = 3.00, 2 = 0.00, = 0.5500
0.2
0.2
y
0.8
0.6
0.2
(c) 1 = 2.00, 2 = 0.00, = 0.7008
0.6
0.4
0.8
0.6
0.8
0.6
0.4
(a) 1 = 0.00, 2 = 0.00, = 0.7654
0.2
0.2
0.4
0.6
0.8
0.2
0.8
0.8
0.6
0.8
0.4
0.8
0.6
0.2
0.4
0.6
0.4
0.4
0.6
0.8
0.2
0.6
0.2
0.4
0
0.2
0.8
58
(i) 1 = 4.53, 2 = 0.00, = 0.0024
shown in Figure 7(a), where = 0.7654 is a proportionality

coefcient of the safe basins acreage to the whole acreage
of area . It can be seen from Figure 7 that will decease
when 1 increase which implies erosion of the safe basins.
In the case when 1 > , the boundary of the safe basins
are eroded more and more with increase of 1 as shown in
Figure 7(b) to Figure 7(i), and in the case when 1 > =
4.54, the safe basins disappear completely. One may call such
phenomena that happen in the sudden change of the safe
basins, which become from integrated one to eroded one or
from eroded one to nothingness when 1 passes through the
critical values = 0.21 and = 4.54, as deterministic safe
basins bifurcation. Ten = 0.21 and = 4.54 are two
bifurcation points if 1 is chosen as bifurcation parameter. In
59
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.2
0.2
y
0.4
0.2
0.2
0.6
0.6
0.4
0.2
0.2
0.4
0.6
0.8
0.6
0.4
0.2
0.2
0.4
0.8
0.6
0.8
0.6
0.8
0.8
0.6
0.8
0.6
0.6
0.4
0.4
0.2
0.4
0.8
0.4
0.2
0.4
0.2
y
0
0.2
0.2
0.8
0.2
0.4
0.4
0.8
0.8
0.6
0.4
(d) 1 = 3.0, 2 = 2.0, = 0.4113
0.6
0.2
(c) 1 = 2.0, 2 = 2.0, = 0.6368
0.8
0.8
(b) 1 = 1.0, 2 = 2.0, = 0.6704
0.6
0.6
0.2
0.8
0.4
(a) 1 = 0.0, 2 = 2.0, = 0.6712
0.8
0.6
0.8
0.6
0.4
0.2
0.2
0.4
0.8
0.6
0.8
0.6
0.4
0.8
0.2
0.8
0.2
0.6
0.8
0.4
0.4
0.6
0.6
0.4
0.6
1
0.4
0.8
0.2
0.2
0.2
0.2
0.8
(e) 1 = 3.2, 2 = 2.0, = 0.2852
(f) 1 = 3.3, 2 = 2.0, = 0.1056
0.6
0.4
0.2
y
0
0.2
0.4
0.8
0.6
0.4
0.2
0.2
0.4
0.8
0.6
0.8
0.6
(g) 1 = 3.4, 2 = 2.0, = 0.0600
Figure 8: Erosion of safe basins in system (1): (2 = 2.0).
the case when 1 , the boundary of the safe basins of

system (1) are smooth; when < 1 the safe basins are
eroded and when 1 > , the safe basins disappear.
When 2 = 0, the threshold value form condition (10) is
1 = 4.93, which is close to the second safe basins bifurcation
point = 4.54; it shows that there are strong relations
between the erosion of safe basin and chaotic motions in the
deterministic system.
3.2. Bifurcation with Bounded Noises. Next, we consider the
efect of the random noise on the safe basins, the parameters
in system (1) are chosen as
= 0.8,
1 = 1.5,
= 0.9,
2 = 1.0,
= 0.1,
= 0.4,
= 1.8,
2 = 2.0,
(14)
which are the same as in deterministic case for comparison

except that 2 changes from 0 to 2. Ten, the erosion of
the safe basin of system (1) is discussed numerically. For the
method of numerical simulation, the reader is referred to Shinozuka and Jan [27], and the method of Monte-Carlo is used
to generate random samples. Here, only 10 random samples
are used in this paper due to the limitation of calculation
capacity. If the solution of system (1) stays in area for a long
enough time not less than 2000 in all the 10 random samples,
such a solution can be approximately taken as a safe solution,
and then the corresponding lattice may be taken as a part of
the safe basins, which is defned in a similar way as for the
deterministic one; if the solution of system (1) escapes from
area , such a solution is taken as an unsafe solution, and
the corresponding lattice is not belonging to the safe basins.

One may call such safe basins as stochastic safe basins. Te
governing equation (1) is numerically integrated by the fourth
order Runge-Kutta algorithm, and the numerical results are
shown in Figure 8(a) to Figure 8(g).
Figures 8(a)8(g) show that the stochastic safe basins are
eroded more and more with increase of 1 , which is similar
to the deterministic case as shown in Figures 7(a)7(i), yet
with signifcant diferences in and . Calculation results
show that, in the case when 1 > = 0.00, the boundary of
the stochastic safe basins begins to be eroded more and more
with increase of 1 as shown in Figure 8(b) to Figure 8(g)
and, in the case when 1 > = 3.43, the stochastic safe
basins disappear completely. One may call such phenomena
that happen in the sudden change of the stochastic safe
basins, which become from integrated one to eroded one or
from eroded one to nothingness when 1 passes through the
critical values = 0.00 and = 3.43, as stochastic safe
basins bifurcation. Ten = 0.00 and = 3.43 are two
bifurcation points if 1 is chosen as bifurcation parameter.
In the case when 1 , the boundary of the stochastic
safe basins of system (1) are smooth; when < 1
the stochastic safe basins are eroded and when 1 > , the
safe basins disappear. Te random disturbance 2 () causes
and decrease from = 0.21, = 4.54 to = 0.00,
= 3.43 and makes the system more unsafe in comparison
with the deterministic case.
When 2 = 2.0, the threshold value form condition
(10) is 1 = 3.46, which is close to the second safe basins
bifurcation point = 3.43; it also shows that there are
strong relations between the erosion of safe basin and chaotic
motions in stochastic dynamical system, which is similar in
the deterministic system. Te incursive fractal fngers are also
observed in the stochastic system, which means that chaotic
responses still exist in the stochastic system (1) when 2 = 0.
From Figure 8, one knows that random noise can aggravate
the erosion of the safe basin.
Overall, random noise may destroy the integrity of the
safe basins boundary, bring forward the occurrence of the
stochastic bifurcation, and hence make the system become
more unsafely. Te threshold value of the stochastic dynamical system form condition (10) is 1 = 3.46, which is smaller
than the threshold value 1 = 4.93 of the deterministic
dynamical system. Obviously, random noises make the parametric threshold for chaotic motions vary in a larger region,
hence making the chaotic motions occur more easily.
4. Conclusions and Discussion

Te erosion of the safe basins and chaotic motions of
a nonlinear vibroimpact oscillator under harmonic and
bounded random noise is studied. Melnikovs method in the
deterministic vibroimpact system is extended to the analysis
of homoclinic bifurcations and chaos in the stochastic case.
Te results reveal that the threshold amplitude 1 for the
onset of chaos decreases as the noise amplitude 2 increase.
Although the theory of stochastic bifurcation has been
advanced to a new level in the last decade, there remain a
lot of problems to be solved. Even the defnition of stochastic
bifurcation needs to be improved. In this paper, we suggest
60
an alternative defnition for stochastic bifurcation based on
the analysis of the safe basins of a sofening Dufng oscillator
subject to deterministic harmonic and bounded random excitations, which focuses on a sudden change in the character
of the safe basins of the dynamical system as the bifurcation
parameter passes through a critical value. Tis defnition
applies equally well either to the stochastic bifurcation or to
the deterministic bifurcation. However, the application of the
defnition for real systems needs more efect. Te analysis
shows that the random noise causes the two bifurcation
points and decrease; the parametric threshold for
chaotic motions varies in a larger region, therefore making
the system more unsafely and making the chaotic motions
occur more easily. From physical point of view, the results
of this paper can help one to better design the system, such
that the system operates in a nonchaotic state and then can
be controlled more easily, meanwhile reducing the erosion of
the safe basins which makes the system more secure.
In the paper, Melnikovs methods and bifurcation of safe
basins are the main research methods. In the fact, there
are other efective method to verify the chaos, for instance,
topological horseshoes method which has been successfully
applied in many works [2830].
Acknowledgments
Te work reported in this paper was supported by the
National Natural Science Foundation of China under Grant
no. 11401096 and 11326123, the Natural Science Foundation of
Guangdong Province under Grant no. S2013010014485 and
S201310012463, and Special fund of the Guangdong College
discipline construction under Grant nos. 2013KJCX0189 and
2013B020314020.
References
[1] B. Brogliato, Non-smooth Mechanics: Models, Dynamics and
Control, Springer, London, UK, 1999.
[2] V. S. Metrikyn, On the theory of vibro-impact devices with
randomly varying parameters, Izvestiya Vysshikh Uchebnykh
Zavedenii, Radiofzika, vol. 13, pp. 412, 1970 (Russian).
[3] R. L. Stratonovich, Topics in the Teory of Random Noise, vol.
1-2, Gordon and Breach, New York, NY, USA, 1963.
[4] H. T. Zhu, Stochastic response of vibro-impact Dufng oscillators under external and parametric Gaussian white noises,
Journal of Sound and Vibration, vol. 333, no. 3, pp. 954961, 2014.
[5] H.-S. Jing and M. Young, Random response of a single-degreeof-freedom vibro-impact system with clearance, Earthquake
Engineering & Structural Dynamics, vol. 19, no. 6, pp. 789798,
1990.
[6] M. Xu, Y. Wang, X. L. Jin, Z. L. Huang, and T. X. Yu, Random response of vibro-impact systems with inelastic contact,
International Journal of Non-Linear Mechanics, vol. 52, pp. 26
31, 2013.
61
[7] C. Li, W. Xu, J. Feng, and L. Wang, Response probability density
functions of Dufng-Van der Pol vibro-impact system under
correlated Gaussian white noise excitations, Physica A, vol. 392,
no. 6, pp. 12691279, 2013.
[8] V. F. Zhuravlev, A method for analyzing vibration-impact
systems by means of special functions, Mechanics of Solids, vol.
11, pp. 2327, 1976.
[9] D. V. Iourtchenko and M. F. Dimentberg, Energy balance for
random vibrations of piecewise-conservative systems, Journal
of Sound and Vibration, vol. 248, no. 5, pp. 913923, 2001.
[10] Q. Feng and H. He, Modeling of the mean Poincare map
on a class of random impact oscillators, European Journal of
Mechanics, A: Solids, vol. 22, no. 2, pp. 267281, 2003.
[11] D. V. Iourtchenko and L. L. Song, Numerical investigation of a
response probability density function of stochastic vibroimpact
systems with inelastic impacts, International Journal of NonLinear Mechanics, vol. 41, no. 3, pp. 447455, 2006.
[12] M. F. Dimentberg, O. Gaidai, and A. Naess, Random vibrations with strongly inelastic impacts: response PDF by the
path integration method, International Journal of Non-Linear
[13] M. F. Dimentberg and D. V. Iourtchenko, Random vibrations
with impacts: a review, Nonlinear Dynamics, vol. 36, no. 24,
pp. 229254, 2004.
[14] J. Guckenherner and P. Holmes, Nonlinear oscillations, in
Dynamical Systems and Bifurcations of Vector Fields, Springer,
New York, NY, USA, 1983.
[15] S.-N. Chow and S. W. Shaw, Bifurcations of subharmonics,
Journal of Diferential Equations, vol. 65, no. 3, pp. 304320,
1986.
[16] S. W. Shaw and R. H. Rand, Te transition to chaos in a
simple mechanical system, International Journal of Non-Linear
[17] Z. Du and W. Zhang, Melnikov method for homoclinic
bifurcation in nonlinear impact oscillators, Computers and
Mathematics with Applications, vol. 50, no. 3-4, pp. 445458,
2005.
[18] R. Stratonovich, Topics in the Teory of Random Noise, vol. 1,
Gordon and Breach Science Publishers, New York, NY, USA,
1963.
[19] Y. K. Lin and G. Q. Cai, Probabilistic Structural Dynamics:
Advanced Teory and Applications, Mc-Graw-Hill, New York,
NY, USA, 1995.
[20] M. F. Dimentberg, D. V. Iourtchenko, and O. van Ewijk,
Subharmonic response of a quasi-isochronous vibroimpact
system to a randomly disordered periodic excitation, Nonlinear
Dynamics, vol. 17, no. 2, pp. 173186, 1998.
[21] H. Rong, X. Wang, W. Xu, and T. Fang, Subharmonic response
of a single-degree-of-freedom nonlinear vibroimpact system to
a randomly disordered periodic excitation, Journal of Sound
and Vibration, vol. 327, no. 1-2, pp. 173182, 2009.
[22] E. Simiu, Chaotic Transitions in Deterministic and Stochastic
Systems, Princeton Series in Applied Mathematics, Princeton
University Press, Princeton, NJ, USA, 2002.
[23] A. Wolf, J. B. Swif, H. L. Swinney, and J. A. Vastano, Determining Lyapunov exponents from a time series, Physica D, vol. 16,
no. 3, pp. 285317, 1985.
[24] H. Rong, X. Wang, W. Xu, and T. Fang, Resonant response of
a non-linear vibro-impact system to combined deterministic
harmonic and random excitations, International Journal of
Non-Linear Mechanics, vol. 45, no. 5, pp. 474481, 2010.

[25] R. P. Eason, A. J. Dick, and S. Nagarajaiah, Numerical investigation of coexisting high and low amplitude responses and
safe basin erosion for a coupled linear oscillator and nonlinear
absorber system, Journal of Sound and Vibration, vol. 333, no.
15, pp. 34903504, 2014.
[26] C. B. Gan, Noise-induced chaos and basin erosion in sofening
Dufng oscillator, Chaos, Solitons and Fractals, vol. 25, no. 5,
pp. 10691081, 2005.
[27] M. Shinozuka and C.-M. Jan, Digital simulation of random
processes and its applications, Journal of Sound and Vibration,
vol. 25, no. 1, pp. 111128, 1972.
[28] X.-S. Yang, Topological horseshoes and computer assisted
verifcation of chaotic dynamics, International Journal of Bifurcation and Chaos, vol. 19, no. 4, pp. 11271145, 2009.
[29] Q. Li and X.-S. Yang, A simple method for fnding topological
horseshoes, International Journal of Bifurcation and Chaos, vol.
20, no. 2, pp. 467478, 2010.
[30] Q. Li, S. Tang, and X.-S. Yang, New bifurcations in the simplest
passive walking model, Chaos, vol. 23, no. 4, Article ID 043110,
2013.

http://dx.doi.org/10.1155/2014/918376
Research Article
A General Setting and Solution of
Bellman Equation in Monetary Theory
Xiaoli Gan and Wanbo Lu
School of Statistics, Southwestern University of Finance and Economics, Chengdu 610074, China
Correspondence should be addressed to Wanbo Lu; luwanbo.swufe@gmail.com
Received 28 September 2014; Revised 22 November 2014; Accepted 22 November 2014; Published 21 December 2014
Academic Editor: Zhihua Zhang
Copyright 2014 X. Gan and W. Lu. Tis is an open access article distributed under the Creative Commons Attribution License,
As an important tool in theoretical economics, Bellman equation is very powerful in solving optimization problems of discrete time
and is frequently used in monetary theory. Because there is not a general method to solve this problem in monetary theory, it is
hard to grasp the setting and solution of Bellman equation and easy to reach wrong conclusions. In this paper, we discuss the rules
and problems that should be paid attention to when incorporating money into general equilibrium models. A general setting and
solution of Bellman equation in monetary theory are provided. Te proposed method is clear, is easy to grasp, is generalized, and
always leads to the correct results.
1. Introduction
In recent years, many economists applied business cycle
approaches to macroeconomic modeling so that monetary
factors could be modeled into dynamic general equilibrium
models. As an important method of monetary economics
modeling, infnite horizon representative-agent models provide a close link between theory and practice; its research
framework can guide practice behavior and be tested by
actual data. Tis method can link monetary economics and
other popular models for studying business cycle phenomena
closely. Tere are three basic monetary economics approaches
introducing money to economic general equilibrium models
in the infnite horizon representative-agent framework. First,
it is assumed that utility could be yielded by money directly
so that the money variable has been incorporated into utility
function of the representative-agent models [1]. Second,
assuming asset exchanges are costly, transaction costs of some
form give rise to a money demand [2, 3]. Clower [4] considers
that money is used for some types of transactions. Brock [5],
McCallum and Goodfriend [6], and Croushore [7] assume
that time and money can be combined to yield transaction
services. Tird, money is treated as an asset that can be used
to transfer resources between generations [8].
Dynamic optimization is the main involved issue during

the modeling process. It is represented and solved by Bellman
equation method, namely, the value function method. Te
method will obtain a forward-looking households path to
maximize lifetime utility through the optimal behavior and
further relevant conclusions. Te setting of Bellman equation
is the frst and crucial step to solve dynamic programming
problems. It is hard to grasp the setting and solution of
Bellman equation and easy to reach wrong conclusions since
there is not a general method to set Bellman equation or the
settings of Bellman equation are excessively fexible. Walsh
[9] used Bellman equation to set and solve dynamic general
equilibrium models of money. However he does not give
a general law of Bellman equation setting. Te setting and
solution of the equation in his book are ambiguous and not
clear. To the best of the authors knowledge, there have been
no studies of uniform setting of Bellman equation.
In this paper, we provide a set of general setting and
solution methods for Bellman equation with multipliers. It
is very clear and easy to grasp. Te most important thing is
that the proposed method can always lead to correct results.
We apply our method to monetary general equilibrium
models that are in the framework of the frst two basic
monetary economic approaches which incorporate money
63
into economic general equilibrium model and make some

extensions. At the same time, we compare our method with
other current methods of setting and solution for Bellman
equation to display the clarity and correctness of the proposed
method. It is assumed that all the relevant assumptions of
applying Bellman equation are satisfed. Tis is to ensure the
feasibility of analysis and solution. For the convenience of
discussion and suitable length of the paper, we mainly discuss
the certainty linear programming problems. Te Bellman
equations setting and solution of uncertainty problems are
similar to those with certainty problems essentially.
Step 2. Set up Bellman equation with multipliers to express

dynamic optimization problem in Step 1:
() ()
()
, 2 , . . . ,
)
(1
= max { (1 , 2 , . . . , )
()
()
()
, 2,+1
, . . . , ,+1
)}
+ (1,+1
+ 1 1 (1 , . . . , , +1, , . . . , )
+ + (1 , . . . , , +1, , . . . , ) ,
where () is the value function and is the multiplier of

the th constraint (), = 1, 2, . . . , .
2. A New Method of Setting and Solution of

Bellman Equation
First of all, we provide the theoretical details of the new
general method and steps when applying Bellman equation
to solve problems. It has many advantages. Te relationship of
the results is very clear through the connection of multipliers.
Tere are no tedious expansions during the derivations. It
difers from the expansion method because it does not need
to consider which control variable should be replaced. Te
technical details about the equivalence of Bellman equations
and dynamic programming problems and the solvability of
set problems can be found in [10]. Actually, it is easy to
reach wrong conclusions using other settings and solutions
of Bellman equation with multipliers, or it has to fnd some
particular skills to get the right results. But our method
does not meet the problem. Te proposed method can be
described as the following fve steps.
Step 3. Compute the partial derivatives of all control variables

on the right side of the equation at Step 2 to derive frst-order
conditions:
() ()
()
(1
, 2 , . . . ,
)
()
(1 , 2 , . . . , )
()
+ [
[
= max (1 , 2 , . . . , )
=0
1 (1 , 2 , . . . , , +1, , . . . , ) = 0,
{
{
{
..
s.t. {
.
{
{
{ (1 , 2 , . . . , , +1, , . . . , ) = 0,
(1)
,
,
(2)
where the objective function (1) represents the maximum

sum of present value of a forward-looking behavioral agents
every period objective function (1 , 2 , . . . , ). In monetary theory, () usually represents the utility function of
a family and is a subjective rate of discount. Equation
(2) is constraints; 1 (), . . . , () are dynamic constraint
functions and usually are intertemporal budget constraints.
() ()
()
Let control variables {1
, 2 , . . . ,
} {1 , 2 , . . . , ,
+1, , . . . , }; the remaining variables {1 , 2 , . . . , ,
() ()
() ()
()
()
+1, , . . . , } {1
, 2 , . . . ,
} = {1
, 2 , . . . ,
} are
state variables.
()
()
()
(1,+1
, . . . , ,+1
) 1,+1
()
()
1,+1
+ +
Step 1. List the expression of target problem.

Consider the dynamic programming problem as
(3)
+ 1
()
()
()
(1,+1
, . . . , ,+1
) ,+1
]
()
()
,+1
1 (1 , . . . , )
()
(1 , . . . , )
()
= 0,
= 1, 2, . . . , .
(4)
During the derivation, it should be taken into account

that next period state variables can be represented by other
control variables according to the constraints, that is, to
()
()
()
, 2,+1
, . . . , ,+1
. Because the constraints are
expand 1,+1
dynamic and are usually intertemporal budget constraints,
()
()
()
those control variables 1,+1
, 2,+1
, . . . , ,+1
at time +
1 can ofen be represented as expressions of the previous period control variables. Attention should be paid to
compute the derivatives of control variables hiding in
()
()
()
, 2,+1
, . . . , ,+1
). But there is no need to expand
(1,+1
()
()
()
1,+1
, 2,+1
, . . . , ,+1
afer the multipliers during the derivation. Other than that, no particular attention should be paid
to other current skills; for example, some control variables
()
()
()
should be replaced afer 1,+1
, 2,+1
, . . . , ,+1
represented
()
()
()
by other control variables in (1,+1 , 2,+1 , . . . , ,+1 ) and so
on.
64
Step 4. By the envelope theorem, take the partial derivatives

of control variables at time on both sides of Bellman
equation to derive the remaining frst-order conditions:
()
()
, . . . ,
)
(1
3. Setting and Solution of Bellman Equation in

Basic Money-In-Utility Model
()
= 1
(1 , . . . , )
()
and solution method and compare with current popular

methods.
+ +
(1 , . . . , )
()
= 1, 2, . . . , .
(5)
()
()
()
Still, it does not need to expand 1,+1
, 2,+1
, . . . , ,+1
afer the multipliers during the derivations. Requirements are
the same as Step 3.
Step 5. Obtain the new relevant results about target problem

through recursion and substitution according to the above
results.
We could combine several state variables to one as
you need according to specifc economic signifcance and
constraints when there is no need to describe the relevant
economic signifcance of state variables one by one. For exam()
()
() ()
, 2
could be combined as (1
, 2 ) = . is a
ple, 1
() ()
()
()
state variable in value function ((1 , 2 ), 3
, . . . ,
),
()
()
that is, ( , 2 , . . . , ). And then we follow the above
fve steps to solve specifc problem. Te rest of state variables
could be operated similarly. But combinations should not
be overlapped over state or combined state variables. For
()
()
, 2
have been combined as
instance, suppose that 1
() ()
()
as an independent
(1 , 2 ) = . It is infexible to take 2
()
()
state variable in value function ( , 2 , . . . ,
) or put the
()
()
() ()
combination of 2 , 3 , (1 , 2 ) = in value function
()
( , , . . . ,
) as a state variable.
Te steps above give a general setting and solution of
Bellman equation. Tese can be summarized as follows: frst,
set Bellman equation with multipliers of target dynamic
optimization problem under the requirement of no overlaps
of state variables; second, extend the late period state variables
in () on the right side of Bellman equation and there is
no need to expand these variables afer the multipliers; third,
let the derivatives of state variables of time equal zero and
take the partial derivatives of these variables on both sides of
Bellman equation to derive frst-order conditions; fnally, get
more needed results for analysis from these conditions.
Diferent from some current settings which allow overlap
of state variables in value function, our method does not
permit overlaps. In fact, overlap of state variable is easy to
reach wrong conclusions or it has to fnd some particular
skills to get the right results [9]. Tese methods are hard to
generalize. More details will be seen in the following sections.
Now we will take several discrete time dynamic optimization problems that are under framework of the basic two
methods of incorporating money into general equilibrium
models as examples to show the applications of our setting
3.1. Model and Solution Using Previous Approach. Te basic

Money-In-Utility model has few features. Te labor-leisure
options of families are ignored temporarily in utility function.
Only family consumption and real money balances are
involved in utility function; that is, real money balances yield
utility directly. We ignore the uncertainty of currency impact
and technological changes temporarily for convenience.
Te total present utility value of family life cycle is
=
=0 ( , ). Usually the per capita version of
intertemporal budget constraint is
(
1
1
) + + (
)
1+
1 + 1
(1 + 1 ) 1 + 1
+
= + + + ,
(1 + ) (1 + )
(6)
where , , , , , 1 , and are the per capita

consumption, bonds, stock of capital, money balances, net
lump-sum transfer received from the government, nominal
interest rate, and infation rate, respectively, at time . is the
rate of depreciation of physical capital, is the population
growth rate (assumed to be constant), and (1 /(1 +
)) is the production function assumed to be continuously
diferentiable and to satisfy the usual Inada conditions (see
[11]). Note that control variables can be changed and afect
the total discounted value of utility through consumption
in period ; state variables cannot be changed. Determination
of state and control variables is crucial to derive the frst-order
conditions and other results. Te control variables for this
problem are , , and, and the state variables are 1 ,
1 , 1 , , and , the households initial level of resources.
Tis problem can be solved by the expansion method
below according to Walsh [9]. It is started with setting up
value function
( ) = max { ( , ) + (+1 )} .
(7)
Using (6), value function can be expanded as

( )
= max { ( , )
+ ( (
)
1+
+ +1 + (
+
(8)
1
)
1+
(1 + ) +
)} .
(1 + +1 ) (1 + )
65
Replace as ; according to (6), (8) can be

written as an expansion equation as follows:
( )
= ( , )
( ) = max { ( , )
+ ( (

) + +1
1+
+(
+
1
) ( )
1+
+ (+1 )
(9)
(1 + ) +
)} .
(1 + +1 ) (1 + )
( ) + 1
(+1 ) [
] = 0,
1+
(10)
3.2. Our Solving Approach. Now, we use our proposed steps

of setting and solution of Bellman equation to solve the
above basic Money-In-Utility problem. First, let the Bellman
equation with multiplier be
(15)
( )
= ( , ) = ,
(11)
( ) = max { ( , )
1
) + +1 + (
)
1+
1+
(1 + ) +
)} .
(1 + +1 ) (1 + )
(14)
Second, computing the partial derivatives for the control

variables, we obtain the frst-order conditions as
Another expansion method will replace as

; according to (6), (8) can be written as
(+1 )
= 0.
(1 + +1 ) (1 + )
+ ( ) .
and the envelope theorem yields
+ ( (
+ ( , )
( ) = max { ( , ) + (+1 )}
( ) + 1
] = 0,
1+
( ) = ( , ) .
( )
= ( , )
( ) = ( , ) .
(+1 )
( )
= ( , ) +
(1 + +1 ) (1 + )
(+1 ) [
(13)
Using the envelope theorem and computing the derivative

with respect to state variable , we get
[ ( ) + 1 ] = 0,
1+
( )
1 +
= (+1 )
(1 + +1 ) (1 + )
[ ( ) + 1 ] = 0,
1+
( )
= ( , ) (+1 )
1 +
= 0,
(1 + +1 ) (1 + )
( )
= ( , ) + (+1 )
Compute the partial derivatives of control variables to

derive frst-order conditions:
Compute the partial derivatives of control variables to

derive frst-order conditions as
(12)
( )
1 +
= ,
= (+1 )
(1 + +1 ) (1 + )
( )
1
= (+1 )
[ ( ) + 1 ] = , (16)
1+
( )
= ( , ) + (+1 )
1
= .
(1 + +1 ) (1 + )
Tird, using the envelope theorem and computing the

derivatives of both sides of Bellman equation with respect to
state variable , we get
( ) = .
(17)
66
Finally, based on the above results, it is easy to get

( , ) = = ( , )
+ (+1 )
1
.
(1 + +1 ) (1 + )
(18)
Tis expression indicates that the marginal beneft of

increased money holdings should be equal to the marginal
utility of consumption on period . Similarly, we can easily
obtain the allocation of initial level of resources among
consumption, capital, bonds, and money balances, and the
same marginal beneft must be yielded by each use at an
optimum allocation.
Tese results with multiplier are open-and-shut, and it is
easy to fnd the economic signifcance of marginal utility of
consumption. Comparing with current approach mentioned
above, during the process of solving problem, there is no need
to consider which variable should be replaced. Tere are no
tedious expanded expressions. Using expansion method, if
we replace , the derived frst-order conditions are seen to
be messy and it is not easy to fnd the relationships of results.
If we replace , the derived results are a bit more clear, but we
might not think of replacing at the beginning. So expansion
methods are not easy to operate, and it is not clear to replace
variable.

Shopping-Time Model
4.1. Model and Solution Using Previous Approach. In Shopping-Time Model, shopping time is a function of consumption and money balances. Because consumption needs shopping time, leisure is reduced. Household utility is assumed to
depend on consumption and leisure. Consumption can not
only yield utility directly but also decrease utility indirectly.
In this section, is time spent in market employment and
is time spent shopping. is the function of consumption
and money balances; that is, = (, ), > 0, 0.
Total time available is normalized to equal 1. Growth rate of
population is assumed to be 0 for convenience. Let be the
leisure time. Te utility function is (, , ) = V[, 1
(, )]. It donates utility as a function of consumption, labor
supply, and money holdings.
Te households intertemporal objective is maximum
discounted utility subject to resource constraint as
max
=0
s.t.
(19)
(1 , ) + + (1 ) 1
+
(1 + 1 ) 1 + 1
= + + + .
1 +
Because controllable factors, labor supply, and consumption afect labor supply, output is afected not only by state
(20)
According to Walsh [9], afer replacing +1 , , value

function can be expanded as
( , 1 )
= max {V [ , 1 ( , )]
+ [+1 +
(1 + ) +
, (1 , )
1 + +1
+ (1 ) 1 + ]} .
(21)
Computing the partial derivatives with respect to control
variables, we get
( , 1 )
= V [ , 1 ( , )]
V [ , 1 ( , )] ( , )
(+1 , ) = 0,
( , 1 )
= V [ , 1 ( , )]
+ (+1 , ) (1 , ) = 0,
(22)
( , 1 )
= V [ , 1 ( , )] ( , )
+ (+1 , )
1
1 + +1
(+1 , ) = 0.
Computing the partial derivatives of both sides of the

value equation with respect to the state variables , 1 , we
get
( , 1 ) = (+1 , ) [ (1 , ) + 1 ] .
= max V (+ , + )
( , 1 ) = max {V ( , ) + (+1 , )} .
( , 1 ) = (+1 , )
V (+ , 1 + (+ , + ))
=0
variable 1 but also by these controllable factors. Te lef

side of the budget constraint is unsuitable to be combined as
a state variable . Combining the money holdings, bonds,
and transfers to = + (((1 + 1 )1 + 1 )/(1 + )), the
households fnancial assets, value function can be written as
(23)
4.2. Our Solving Approach. Now, we use our proposed

method to solve the above Shopping-Time Model problem.
First, let the Bellman equation with multiplier be
( , 1 ) = max {V ( , ) + (+1 , )}
+ [ + (1 , ) + (1 ) 1
] ,
(24)
67
where state variable +1 = +1 + (((1 + ) + )/(1 + +1 )).

Ten, compute the partial derivatives with respect to control
variables to derive frst-order conditions,
( , 1 )
= V ( , ) V ( , ) ( , ) = ,
( , 1 )
1 +
= (+1 , )
= ,
1 + +1
( , 1 )
= (+1 , ) = ,
( , )
( , 1 )
= V ( , ) ( , ) + +1 = .
1 + +1
(25)
Using the envelope theorem and computing the derivatives with respect to the state variables , 1 , we get
( , 1 ) = ,
( , 1 ) = [ (1 , ) + 1 ] .
(26)
Now, we use the above results to compute the opportunity

cost of holding money. Since the utility function is
(, , ) = V [, (, )] ,
(27)
and based on (6), (25), and (26), we obtain

V ( , ) ( , )
( , , )
=
( , , )
V ( , ) V ( , ) ( , )
=
(( (+1 , )) / (1 + +1 ))
=1
=1
(( (+1 , )) / (1 + +1 ))
(+1 , ) ((1 + ) / (1 + +1 ))
1
= .
1 + 1 +
household utility value

=0 ( ). More complicated utility
function will be discussed in the following sections. Take
Cash-In-Advance constraint form as (1 /(1 + )) +
due to Svensson [12]. Tis means that the agent enters the
period with money holdings 1 and receives a lump-sum
transfer (in real currency terms) for consumption goods.
Bonds and capital may not be purchased by currency. If
capital is assumed to be purchased by money, the Cash-InAdvance constraint will become (1 /(1 + )) + + .
Te budget constraint is rewritten in real terms as
(1 ) + + (1 ) 1
+
(1 + 1 ) 1 + 1
+ + + .
1 +
In monetary theory, constraints are expressed as inequality frequently. Tis constraint describes that the representative agents time real resources should be more than
or equal to the use of it, that is, purchasing consumption,
capital, bonds, and money holdings that are then carried into
period + 1. Because of the assumption of rational agents,
the certainty problems we are discussing, and the positive
opportunity cost of money holdings, the constraints become
equations in equilibriums.
Output is only afected by state variable 1 without
considering the efect of labor supply. Following the solution
methods of Walsh [9], let the lef side of the budget constraint
be a state variable = (1 )+ +(1)1 +(((1+1 )1 +
1 )/(1 + )). It will be very tedious adopting expansion
method with two constraints. Setting Bellman equation with
multipliers will be better. Let state variable 1 in value
function get the economic signifcance of money in frstorder conditions. However, there is an overlapped setting
with ,
( , 1 ) = max { ( ) + (+1 , )}
+ ( )
+ ( 1 + ) .
1 +
(28)
Obviously, the expressions of the derived frst-order

conditions by previous method seem to be tedious and messy,
and it is not so easy to compute the relevant results such as the
opportunity cost of holding money. Our proposed method
is comparatively neat and can easily obtain relevant results
correctly.

Cash-In-Advance Model
5.1. Model and Solution Using Previous Approach. In basic
Cash-In-Advance model, money is used to purchase goods.
Money cannot yield utility itself, but the consumption of
future can yield utility. Svensson [12] assumed that agents
are available for spending only the cash carried over from
the previous period. Tis is essentially diferent from MoneyIn-Utility model. Consider a simple form of discounted
(29)
(30)
5.2. Our Solving Approach. Let = (1 ) + + (1 )1 +

(((1 + 1 )1 )/(1 + )). Tere is no overlap with in the
value function. Te Bellman equation is set as
( , 1 ) = max { ( ) + (+1 , )}
+ ( +
+ (
1
)
1 +
(31)
1
+ ) .
1 +
Te solution of this problem is similar to the application

of the proposed method in Sections 3 and 4. We will not give
unnecessary details here.
Note that Walsh [9] puts overlapping state variables 1
and in value function; it is easy to get wrong results in those
68
problems with Cash-In-Advance constraints although it is no

problem setting ( , 1 ) with no more results required
now. Tis will be seen in the following example of Section 6.
Using the envelope theorem and computing the derivatives with respect to state variables, we get
( , 1 , 1 )
= ( , 1 , 1 ) = + ,
6. Setting and Solution of Bellman Equation

in Model considering Both Labor Time and
Cash-In-Advance Constraints
6.1. Model and Solution Using Previous Approach. Assuming that money is used to purchase consumption goods
and investments, the Cash-In-Advance constraint becomes
(1 /(1 + )) + + , where is investment and
= (1 )1 . In considering of the efects of labor
supply on output and leisure on utility, we have = 1 .
Te agents objective becomes
max
s.t.
( , )
=0
(1 ) + + (1 ) 1
+
(1 + 1 ) 1 + 1
+ + + ,
1 +
(32)
1
+ + .
1 +
Output is afected not only by state variable 1 but also

by controllable factor ; the lef side of the budget constraint
is unsuitable to be combined as a state variable. Tinking
of the setting of the Shopping-Time Model and Cash-InAdvance model above, it is plausible to put overlapped state
variables = + (((1 + 1 )1 + 1 )/(1 + )) and 1 in
value function following the solution methods of Walsh [9].
Te Bellman equation will be
( , 1 , 1 )
+ [ + (1 ) 1 ] .
= [ (1 , ) + (1 )]
+ (1 ) ,
( , 1 , 1 )
1
= ( , 1 , 1 ) = ( + )
.
1
1 +
(35)
When the question is to derive the efect of infation rate
on the steady-state capital-labor ratio, that is, the steadystate relationship of / and , assume that the aggregate
production takes the form = (1 , ) = 1

1 .
Using (34) and (35),
( , )
( , )
=
(1 , )
+
= ([ (+1 , , ) + (+1 , , )
(1 , ) ) ( + )
= ([ ( +1 + +1 )
= max { ( , ) + (+1 , , )}
+ [ + (1 , ) + (1 ) 1
( , 1 , 1 )
= ( , 1 , 1 )
1
(33)
Te frst-order conditions are

( , 1 , 1 )
= ( , ) = + ,
( , 1 , 1 )
1 +
= (+1 , , )
= ,
1 + +1
( , 1 , 1 )
= (+1 , , ) = + ,
( , 1 , 1 )
= (+1 , , )
1
+ (+1 , , )
= ,
1 + +1
( , 1 , 1 )
= ( , ) + (1 , ) = 0.
(34)
1
1
+ ( +1 + +1 )
]
1 + +1
1 + +1
(1 , ) ) ( + )
= 2
Ten
1
]
1 + +1
+ +1
1
(1 , ) +1
.
1 + +1
+
+1 + +1
1
= 2
.
+
1 + +1 +
From (35),
(1 , ) =
( , 1 , 1 ) (1 ) ( + )
+
1
= [ (1 )]
.
(36)
(37)
(38)
Rewriting the aggregate production as = (1 /

)1 , we have
1
1
= ()1 ( 1 ) [ (1 )] .
+
(39)
69

Using (37), we obtain
()1 (
Second, compute the derivatives of state variables and

derive frst-order conditions:
1 1 1
) [ (1 )]
(40)
+1 + +1
1
.
= 2
1 + +1 +
2 1
1
=[
( 1 + )
1 + +1
+1 + +1
]
+
1/(1)
. (41)
In steady-state, = +1 = , = +1 = , +1 =
, 1 = , = , from (41), we have
1/(1)
1 + 1
=[
.
( 1 + )]
(42)
Tis steady-state capital-labor ratio is derived by using

current prevailing methods. However, it is a wrong result. We
will provide the correct result by using the proposed method
of this paper.
6.2. Our Solving Approach. Output is afected by state variable 1 and control variable ; it is unsuitable to let the lef
side of the budget constraint be combined as a state variable.
We should separate 1 from it. Let state variable 1 in
value function alone get the economic signifcance of money
in frst-order conditions. Because overlaps of state variables
are not allowed according to our proposed method, we put
= + (((1 + 1 )1 )/(1 + )) in value function as a state
variable.
First, set up value function with multipliers:
( , 1 , 1 )
= max { ( , ) + (+1 , , )}
+ [ +
1
+ (1 , )
1 +
+ (1 ) 1 ]
+ [
(1 + 1 ) 1
+ 1
1 +
1 +
+ (1 ) 1 ] .
( , 1 , 1 )
1 +
= (+1 , , )
= ,
1 + +1
( , 1 , 1 )
= (+1 , , ) = + ,
Rewriting this equation, we get

1
( , 1 , 1 )
= ( , ) = + ,
(43)
( , 1 , 1 )
= (+1 , , ) = ,
( , 1 , 1 )
= ( , ) + (1 , ) = 0.
(44)
Tird, compute the partial derivatives with respect to ,
1 , and 1 and the envelope theorem yields
( , 1 , 1 )
= ( , 1 , 1 ) = + ,
( , 1 , 1 )
1
= ( , 1 , 1 )
= [ (1 , ) + (1 )] + (1 ) ,
( , 1 , 1 )
1
= ( , 1 , 1 ) = ( + )
.
1
1 +
(45)
Finally, derive the steady-state capital-labor ratio by the
results above:
1/(1)
1 + 1
=[
.
( 1 + )]
(46)
Te process of deriving this ratio is similar to the

deriving process of (42). We will not give unnecessary
details here. Compare (46) with (42); the result of (42) is
(1/2)1/(1) times that of (46). Equation (46) is the correct
result. It is the diferent partial derivatives with respect
to in (34) and (44) that cause this deviation. In (34),
(+1 , , ) + (+1 , , ) (1/(1 + +1 )) = . In
(44), (+1 , , ) = . Fundamentally speaking, the
reason for the deviation is the overlapped setting of state
variables and 1 versus the nonoverlapped setting of state
variables , 1 , and 1 . One has to use the particular skills
in [9] that is hard to think about and grasp to obtain the right
answer or get a wrong result adopting the overlapped setting
( , 1 , 1 ). For this reason, the setting of value function
by Walsh could not be generalized. In this model considering
both Labor Time and Cash-In-Advance constraints, the basic
methods of incorporating money into general equilibrium
models are generalized, and our proposed method of setting

and solution of Bellman equation demonstrates its clarity
and validity and corrects the defect that some other current
methods of setting and solution are easy to get wrong results.
7. Conclusions
As an important tool in theory economics, Bellman equation
is very powerful in solving optimization problems of discrete
time and is frequently used in monetary theory. It is hard to
grasp the setting and solution of Bellman equation and easy to
reach wrong conclusions since there is no general method to
set Bellman equation or the settings of Bellman equation are
excessively fexible. In this paper, we provide a set of general
setting and solution methods for Bellman equation with
multipliers. In the processes of solving monetary problems,
comparing with other current methods in classic reference,
our proposed method demonstrates its features of clarity,
validity, correct results, easy operation, and generalization.
Bellman equation is used not only in monetary problems
but also in almost every dynamic programming problem
associated with discrete time optimization. Our future work
is to study the applicability of the proposed method in this
paper in other areas.
Acknowledgments
Wanbo Lus research is sponsored by the National Science
Foundation of China (71101118) and the Program for New
Century Excellent Talents in University (NCET-13-0961) in
China.
References
[1] M. Sidrauski, Rational choice and patterns of growth in a monetary economy, American Economic Review, vol. 57, no. 2, pp.
534544, 1967.
[2] W. Baumol, Te transactions demand for cash, Quarterly Journal of Economics, vol. 67, no. 4, pp. 545556, 1952.
[3] J. Tobin, Te interest elasticity of the transactions demand for
cash, Review of Economics and Statistics, vol. 38, no. 3, pp. 241
247, 1956.
[4] R. W. Clower, A reconsideration of the microfoundations of
monetary theory, Western Economic Journal, vol. 6, no. 1, pp.
19, 1967.
[5] W. A. Brock, Money and growth: the case of long run perfect
foresight, International Economic Review, vol. 15, pp. 750777,
1974.
[6] B. T. McCallum and M. S. Goodfriend, Demand for money:
theoretical studies, in Te New Palgrave Dictionary of Economics, pp. 775781, Palgrave MacMillan, Houndmills, UK,
1987.
[7] D. Croushore, Money in the utility function: functional equivalence to a shopping-time model, Journal of Macroeconomics,
vol. 15, no. 1, pp. 175182, 1993.
70
[8] P. A. Samuelson, An exact consumption-loan model of interest
with or without the social contrivance of money, Journal of
Political Economy, vol. 66, no. 6, pp. 467482, 1958.
[9] C. E. Walsh, Monetary Teory and Policy, Te MIT Press,
Cambridge, Mass, USA, 3rd edition, 2010.
[10] L. Gong, Optimization Methods in Economics, Peking University
Press, Beijing, China, 2000.
[11] D. Romer, Advanced Macroeconomics, McGraw-Hill, New York,
NY, USA, 4th edition, 2011.
[12] L. E. O. Svensson, Money and asset prices in a cash-in-advance
economy, Journal of Political Economy, vol. 93, no. 5, pp. 919
944, 1985.

http://dx.doi.org/10.1155/2014/918376
Research Article
1 : 3 Resonance and Chaos in a Discrete Hindmarsh-Rose Model
Bo Li and Zhimin He
School of Mathematics and Statistics, Central South University, Changsha, Hunan 410083, China
Correspondence should be addressed to Zhimin He; hezhimin@mail.csu.edu.cn
Received 10 July 2014; Accepted 30 November 2014; Published 17 December 2014
Academic Editor: Zhidong Teng
Copyright 2014 B. Li and Z. He. Tis is an open access article distributed under the Creative Commons Attribution License,
1 : 3 resonance of a two-dimensional discrete Hindmarsh-Rose model is discussed by normal form method and bifurcation theory.
Numerical simulations are presented to illustrate the theoretical analysis, which predict the occurrence of a closed invariant circle,
period-three saddle cycle, and homoclinic structure. Furthermore, it also displays the complex dynamical behaviors, especially the
transitions between three main dynamical behaviors, namely, quiescence, spiking, and bursting.
1. Introduction
When the bifurcation problem of a system
(, ) ,
R , R ,
(1)
is studied by normal form method, researchers are ofen

likely to compute the following two conditions: nondegeneracy conditions and transversality conditions, in which the
derivatives of (, ) with respect to the variables and
the parameters are involved, respectively. If some of the
nondegeneracy and transversality conditions for the oneparameter bifurcations would be violated, the two-parameter
bifurcations can also happen [13]. In this case, one can
obtain cusp, generalized fip, and Chenciner bifurcation
[2]. In the other case, extra multipliers can approach the
unit circle for discrete dynamical system, thus changing the
dimension of the center manifold . Tere are eleven kinds
of two-parameter bifurcations for discrete dynamical system
listed by Kuznetsov (see Section 4 in [2]).
In this paper, we focus on 1 : 3 resonance, which is
less discussed in the existing papers. When Neimark-Sacker
bifurcation is considered, the case that 0 = 2/3 can lead to
1 : 3 resonance. One can fnd more information in [13] and
references cited therein.
In 1982, Hindmarsh and Rose [4] described a two-variable
model of the action potential which is a modifcation of
Fitzhughs B.v.P model in [5] and explained how the close
proximity of the nullclines can be exploited to give a qualitative explanation for burst generation. Te Hindmarsh-Rose
model is known to reproduce all dynamical behaviors, such
as quiescence, spiking, bursting, irregular spiking, and irregular
bursting [4, 6]. Bifurcation analysis is examined once more in
the past, with respect to one or two bifurcation parameters
[711]. Local bifurcations and global bifurcations are also
analysed and these bifurcation phenomena can be used to
explain the transitions between the dynamical behaviors. For
example, the transition between spiking and bursting in the
model can be understood by homoclinic bifurcations [12, 13].
More information on bifurcation can be found in [1, 4, 8
11, 1422].
Recently, X. Liu and S. Liu [8] discussed the codimension2 bifurcations of the following two-dimensional HindmarshRose model:
= 3 + 2 ,
= 2 ,
(2)
where represents the membrane, is recovery variable, and

, , , and are positive parameters. Model (2) can describe
the transitions between the above fve dynamical behaviors,
that is, quiescence, spiking, bursting, irregular spiking, and
irregular bursting. More related works can be found in [4, 7, 9
13, 1721, 2326].
72
Applying the forward Euler method to model (2), we

obtain the following discrete-time Hindmarsh-Rose system:
+ ( + )
),
( )
(
( + 2 + )
3
Te fxed points of map (3) satisfy the following equations:
(4)
So is the root of the following equation:

3 + ( ) 2 + = 0,
2
= .
(5)
Lemma 1. (1) If 272 4( )3 > 0, then map (3) has a

unique fxed point 11 (11 , 11 ), where 11 < min{0, 2(
)/3}.
(2) If 272 4( )3 = 0, then map (3) has two fxed
points 21 (21 , 21 ) and 22 (22 , 22 ), where 21 < 0 < 22 =
2( )/3.
(3) If 272 4( )3 < 0, then map (3) has three diferent
fxed points, 3 (3 , 3 ) ( = 1, 2, 3), where 31 < 0 < 32 <
2( )/3 < 33 .
Te stability of these fxed points can be found in [21]. In
this paper, we focus on the existence and bifurcation analysis
of 1 : 3 resonance. Here, we would like to give the bifurcation
set of 1 : 3 resonance.
Te Jacobian matrix of map (3) at the fxed point ( , )
is given by
2
1 + (3 + 2 )
),
2
1
= 3 + 2 1,
= 3 2 + 2 .
(8)
It is easy to get that two eigenvalues of ( , ) are

1,2 = 1 +
( 2 4) .
2
(9)
Further, if = 3/ and 2 = 3, then we have 1,2 =

(3 1)/2.
Here, we present the bifurcation set of 1 : 3 resonance as
follows:
= {(, , , , ) :
3
(2 + 3)2 + 3
= , =
, , , , , > 0} .
24
(10)
It is obvious to fnd that < 0 and > 0 from the

bifurcation set. Hence, the 1 : 3 resonance only can occur at
11 (11 , 11 ), 21 (21 , 21 ), 31 (31 , 31 ), and 33 (33 , 33 ). In
the following, we present our discussions for 11 (11 , 11 ).
Te similar arguments can be undertaken at the fxed points
21 (21 , 21 ), 31 (31 , 31 ), and 33 (33 , 33 ).
3. 1 : 3 Resonance
Using the Cardan formula (see [28]), we get the following

results (see also [8, 21]).
( , ) = (
(7)
where
2. Local Dynamics for Fixed Points of Map (3)
+ 2 + = 0.
() := 2 (2 + ) + (1 + + 2 ) = 0,
(3)
where > 0 is the step size. In [21, 27], we proved that map
(3) possesses fip bifurcation, Neimark-Sacker bifurcation,
and 1 : 1, 1 : 2, and 1 : 4 resonance. Te aim of this paper is to
prove that this discrete model possesses the 1 : 3 resonance.
Te method we used is based on the normal form method
and bifurcation theory of discrete dynamical system (see
Kuznetsov, Sections 4 and 9 in [2]).
Tis paper is organized as follows. In Section 2, we present
the existence and local stability of fxed points for map (3). In
Section 3, we show that there exist some values of parameters
such that map (3) undergoes 1 : 3 resonance. In Section 4, we
present numerical simulations, which not only illustrate our
results with the theoretical analysis but also exhibit complex
dynamical behaviors. Finally, a brief discussion is given in
Section 5.
3 + 2 = 0,
and the corresponding characteristic equation of the Jacobian

matrix ( , ) can be written as
(6)
In this section, we show that there exist some values of

parameters such that map (3) undergoes 1 : 3 resonance by
using bifurcation theory [13]. Here, the step sizes and are
considered as bifurcation parameters to present bifurcation
analysis at the fxed point 11 (11 , 11 ).
We discuss the 1 : 3 resonance of map (3) at 11 (11 , 11 )
when the parameters vary in a small neighborhood of
. Taking parameters (, , , 0 , 0 ) arbitrarily from , we
consider map (3) with (, , , 0 , 0 ), which is described by
+ 0 ( 3 + 2 )
( )
).
(
0 ( + 0 2 + )
(11)
+ ( + )
),
( ) (
( + 2 + )
(12)
Te eigenvalues of map (11) at the fxed point 11 (11 , 11 ) are

1,2 = (3 1)/2.
Now, we consider a perturbation of map (11) as follows:
3
where | 0 |, | 0 | 1 which are small perturbation

parameters.
Let = 11 and V = 11 . Ten we transform the
fxed point 11 (11 , 11 ) to the origin and map (12) becomes
(1 + 11 ) + V + 13 2 3
( ) (
),
V
21 + (1 ) V 2
(13)
73
where = (3 1)/2, we get
where
2
+ 211 ,
11 = 311
13 = 311 + ,
21 = 211 .
(1
(14)
) (0 , 0 ) , (0 , 0 ) = 0,
(23)
which implies that
Map (13) can be denoted as
( ) (, ) ( ) + (, V, , ) ,
V
V
(0 , 0 ) , (0 , 0 ) = 0.
(15)
(24)
Using (19), (21), and (24), we have
where
(, ) = (
1 + 11
),
21 1
2 3
(, V, , ) = ( 13
).
2
= (0 , 0 ) , .
(16)
3 1
(0 , 0 ) .
2
3 + 1
(0 , 0 ) (0 , 0 ) =
(0 , 0 ) ,
2
(18)
(0 , 0 ) , (0 , 0 ) = 1,
(19)
which is normalized according to
where , means the standard scalar product in C2 : , =

1 1 + 2 2 .
Now any vector = (, V) R2 can be represented in
the form
C.
= (0 , 0 ) , (0 , 0 ) + ( , 0 , 0 )
= (0 , 0 ) , (0 , 0 ) ( (0 , 0 ) + (0 , 0 ))
+ ( (0 , 0 ) + (0 , 0 ), 0 , 0 )
(17)
Here, we also introduce the adjoint eigenvector

(0 , 0 ) C2 , satisfying
= (0 , 0 ) + (0 , 0 ),
From (21) and (25), we get

+1 = (0 , 0 ) , +1
In the following, we will present our analysis in the critical

case. It is easy to fnd the eigenvalues of (0 , 0 ) and their
corresponding eigenvector (0 , 0 ) C2 as follows:
(0 , 0 ) (0 , 0 ) =
= (0 , 0 ) , (0 , 0 ) (0 , 0 ) + (0 , 0 )
+ ( (0 , 0 ) + (0 , 0 ), 0 , 0 )
3 1
+ (0 , 0 ) ,
2
( (0 , 0 )+ (0 , 0 ), 0 , 0 ) .
(26)
Afer calculation, we can choose

(0 ,
(20)
(0 , 0 ) ,
= (0 , 0 ) , (0 , 0 ) + (0 , 0 )
= (0 , 0 ) , (0 , 0 ) + (0 , 0 ) , (0 , 0 ) .
(21)
1
(0 , 0 ) , (0 , 0 ) = (0 , 0 ) , (0 , 0 )
(0 , 0 ) , (0 , 0 )
(0 , 0 ) , (0 , 0 ) ,
(22)
3 3
0 11 ) ,
2
3(3 + 20 11 ) + 3 3
(
,
)
60
3
From the above equation, we have
Since
(25)
(27)
as (0 , 0 ) and (0 , 0 ), respectively.
By (26), map (15) can be transformed into the complex
form

3 1
1
+
(0 , 0 ) ,
2
!!
2+3
(28)
where
20 = 11 = 02
= 02 13 +
3 2
(20 313 20 11 13 ) ,
3 0
30 = 21 = 12 = 03
= 303 + 303 (3 + 20 11 ) .
(29)
74
Here, we denote (0 , 0 ) by , with + = 2, 3. And

(0 , 0 ) would be denoted by , with + = 2, 3 in the
introduced transformation.
Now, we introduce the following transformation to annihilate some second order terms:
1
1
= + 20 2 + 11 + 20 2 ,
2
2
(30)
where coefcients with + = 2 will be confrmed in the

following, and we can obtain
1
1 2
1
= 20 2 11 20 2 + (20
+ 11 02 ) 3
2
2
2
3
2 2
+ ( 20 11 + 11 + 02 ) 2
2
1
1
2
+ ( 11 20 + 11
+ 02 11 + 20 02 ) 2
2
2
1
( + 11 02 ) 3 + (||4 ) .
2 02 20
03 =
3
(1 + 3) 02 20 + 311 02 + 320 02
2
+
3
3
(1 + 3) 11 02 + 03 + (3 3) 02 20 .
2
2
By setting
20 =
3
,
3 20
30 =
3 3
2
+ 30 ,
11 02 + 320
2
12 =
3 1
1
+
+ (||4 ) ,
2
!!
2+3
(32)
where
20 = 20 + 320 ,
11 = 211 + (3 3) 11 ,
02 = 02 ,
3
30 = (3 3) 20 20 + 311 02
2
21 =
3
3
2
+ 30 ,
( 3 1) 11 02 + (3 + 3) 20
2
2
1
(5 + 3) 11 20 + (2 3) 20 11
2
+ 211 11 + (1 3) 11 11 + 02 02
+
+
1
(1 + 3) 02 02 + 21
2
1
(5 3 3) 11 20 + (3 3) 11 11 ,
2
12 = 20 02 +
1
(1 3) 11 20
2
+ (20 + (3 + 3) 11 ) 11 + (3 + 1) 02 11 + 12
1
+ ( (1 3) 20 + 211 ) 02
2
1
2
(3 + 3) 11 20 (3 + 3) 11
2302 11 ,
2
3 + 3
11 ,
6
02 = 0,
3 + 23
3 3 2
20 11 +
11 + 21 ,
3
3
3 + 3
3 3
11 02
20 02 +
6
3
+
Tus, using (30) and its inverse transformation, map (28) is

changed into the following form:
11 =
(34)
then we have 20 = 11 = 0, 02 = 02 and 30 , 21 , 12 , 03

can be simplifed in the following. Hence, the transformation
(30) is defned and
21 =
(31)
(33)
(35)
3 + 3 2 3
+ 12 ,
11
3
3 20 11
03 = 311 02 320 02 + 03 .
To further simplify map (32), we introduce the following

transformation:
2
3
1
1
1
1
= + 30 3 + 12 + 21 2 + 03 .
6
2
2
6
(36)
Afer using (36) and its inverse transformation, map (32) is

changed into the following form:
3 1
2
1
4
+ 02 +
+ ( ) , (37)
2
2
!!
+=3
where
30 = 30 +
3 3
30 ,
2
12 = 12 + 312 ,
By setting
30 =
3 + 3
30 ,
6
03
21 = 21 ,
3 3
03 = 03 +
03 .
2
21 = 0,
12 =
3 + 3
=
03 ,
6
3
,
3 12
(38)
(39)
then we have 30 = 21 = 03 = 0. Hence, the transformation

(36) is defned. Using transformation (36), map (32) fnally
becomes the following normal form of the bifurcation with
1 : 3 resonance:

3 1
2
(0 , 0 ) 2 + (4 ) ,
+ (0 , 0 ) +

2
(40)
75
1.4
1.4
1.3
1.3
1.2
1.2
1.1
1.1
x
1
0.9
0.9
0.8
0.8
0.7
1.4
1.5
1.6
1.7
1.8
1.9
0.7
1.4
1.5
1.6
(a)
1.7
1.8
1.9
(b)
1.3
1.2
1.1
x
1
0.9
0.8
0.7
0.16
2
0.15
1.9
1.8
0.14
1.7
0.13
1.6
1.5
(c)
Figure 1: 1 : 3 resonance bifurcation diagram at with = 2, = 3, = 0.8, = 0.12428002, and = 1.6771238. (a) In (, ) plane with
= 0.12428002. (b) In (, ) plane with = 0.14. (c) In (, , ) plane.
where
(0 , 0 ) =
(0 , 0 ) =
20 11 (3 + 23)
6
02
,
2
2
(3 3) 11
6
(c) there is a homoclinic structure formed by the stable and

unstable invariant manifolds of the period-three cycle
intersecting transversally in an exponentially narrow
parameter region.
+
21
.
2
(41)
If 1 (0 , 0 ) = (3/2)(3 + 1)(0 , 0 ), 1 (0 , 0 ) =
3|(0 , 0 )|2 (3(1 + 3)/2)(0 , 0 ), a similar argument
as in Lemma 9.13 in [2] can be obtained.
Teorem 2. Let (0 , 0 ) . If 1 (0 , 0 ) = 0 and
Re(1 (0 , 0 )) = 0, then map (3) has the the following complex
dynamical behaviors:
(a) there is a Neimark-Sacker bifurcation curve at the
trivial fxed point 0 of map (40);
(b) there is a saddle cycle of period-three corresponding to

the saddle fxed point ( = 1, 2, 3) of map (40);
Remark 3. Here, the intersection of these manifolds, which

form a homoclinic tangency, implies the existence of Smale
horseshoes and therefore an infnite number of long-period
orbits appear. It illustrates a route from period-3 to chaos.
4. Numerical Simulations
In this section, the 2-dimensional and 3-dimensional bifurcation diagrams show that the 1 : 3 resonance is the degenerate
case of Neimark-Sacker bifurcation. So there exists a closed
invariant circle near the 1 : 3 resonance. Here, we provide the
following case to illustrate the dynamic behaviors of map (3).
Take parameters = 2, = 3, = 0.8, = 0 =
0.12428002, and = 0 = 1.6771238 in map (3). We know
that map (3) has a fxed point (1.1176267, 0.95523687).
Te eigenvalues of the corresponding Jacobian matrix ()
76
0.2
Maximum Lyapunov exponents
0.4
0.2
0.4
0.2
0
0.2
0.4
0.6
0.6
1.4
1.5
1.6
1.7
1.8
1.9
0.8
1.4
1.5
1.6
1.7
1.8
1.9
(a)
(b)
0.14
0.2
0.138
0.2
0.136
0.4
0
0.134
0.2
0.2
0.2
d 0.132
0.4
0.13
0.4
0.2
0.6
0.128
0.6
0.4
0.14
0.135
0.126
1.8
0.13
d
1.7
0.125
1.6
0.12 1.5
0.8
0.8
(c)
0.124
1.5
1.55
1.6
1.65
1.7
1.75
1.8
(d)
Figure 2: Maximum Lyapunov exponents of map (3) near 1 : 3 resonance as and vary. (a) and (b) are maximum Lyapunov exponents
corresponding to (a) and (b) in Figure 1. (c) and (d) are 3-dimensional maximum Lyapunov exponents in [1.5, 1.82] [0.124, 0.14] and
2-dimensional projection onto (, ).
are 1,2 = 0.8660254347 0.5000000165 (3

1)/2. Afer calculating, we further have Re(1 (0 , 0 )) =
284.6634292 = 0. Terefore, from Teorem 2, we see that
fxed point is a 1 : 3 resonance point.
Figures 1(a) and 1(b) show the 2-dimensional bifurcation
diagrams when = 0.12428002 and = 0.14, respectively,
and varies in a neighborhood of 0 = 1.6771238. Figure 1(c)
shows the 3-dimensional bifurcation diagram when , vary
in a neighborhood of (0 , 0 ) = (1.6771238, 0.12428002).
From Figure 1(c), we can observe the relations between
1 : 3 resonance and Neimark-Sacker bifurcation. In fact, the
1 : 3 resonance is the degenerate case of Neimark-Sacker
bifurcation when 0 = 2/3. Here, 0 is the eigenvalues of
the Jacobian matrix (6). Moreover, the fip bifurcation occurs
afer the Neimark-Sacker bifurcation and 1 : 3 resonance.
Te Lyapunov exponents corresponding to the bifurcation
diagram in Figure 1 are computed and plotted in Figure 2.
We easily see that there are the positive Lyapunov exponents
and negative Lyapunov exponents. It means that map (3) has

chaotic and periodic behaviors near the 1 : 3 resonance. Te
3-dimensional maximum Lyapunov exponents are given in
Figure 2(c).
Figures 3(a)3(o) show the phase portraits of map (3)
near for diferent and . Furthermore, as varies around
= 0 = 0.12428002 and 0 = 1.6771238, from Figures
3(a)3(g), we can see that period-3 orbits and period-6 orbits,
eventually leading to chaos when decreases to 0.11325. Tis
is the classical route to chaos. Besides, as increases and
= 0 = 0.12428002 from Figures 3(h)3(j), we can observe
that there exists a fxed point connecting to three saddles,
chaos, and more new complex phenomena in certain regions
near . Here, in Figure 3(k), the diferent colours are chosen
to demonstrate the diferent orbits near the 1 : 3 resonance,
which shows the homoclinic structure near a 1 : 3 resonance.
Furthermore, both Smale horseshoes and an infnite number
of long-period orbits occur. Finally, from Figures 3(m)3(o),
77
0.9548
0.9585
0.959
0.955
0.9595
0.9552
0.9554
0.96
0.9556
0.9605
y
0.9558
0.961
0.956
0.9615
0.9562
0.962
0.9564
0.9625
0.9566
0.963
1.106
1.108
1.11
1.112
0.9568
1.1155 1.116 1.1165 1.117 1.1175 1.118 1.1185
1.114
(a)
(b)
0.955
0.86
0.9551
0.88
0.9552
0.9
0.9553
0.92
0.9554
y
y
0.9555
0.94
0.9556
0.96
0.9557
0.98
0.9558
0.9559
1.1166
1.117
1.1174
1.1178
1
0.95
1.1182
1.05
(c)
1.1
1.15
1.2
1.25
1.1
1.15
1.2
1.25
(d)
0.84
0.82
0.86
0.84
0.88
0.86
0.88
0.9
0.9
y
0.92
y
0.92
0.94
0.94
0.96
0.96
0.98
0.98
1
0.95
1.05
1.1
1.15
1.2
1
0.95
1.25
(e)
1.05
(f)
0.8
0.86
0.82
0.88
0.84
0.9
0.86
0.92
0.88
y
y
0.9
0.94
0.92
0.96
0.94
0.98
0.96
0.98
0.85 0.9 0.95
1
1
1.05 1.1 1.15 1.2 1.25 1.3
1.05
1.1
1.15
x
(g)
(h)
Figure 3: Continued.
1.2
1.25
78

0.8
0.82
0.84
0.85
0.86
0.88
0.9
0.9
y
0.92
0.95
0.94
0.96
0.98
1
1.02
0.85
0.9 0.95
1.05
1.1
1.15
1.2
1.05
0.7
1.25
0.8
0.9
(i)
1.1
1.2
1.3
1.4
(j)
0.8
0.88
0.9
0.85
0.92
0.94
y
0.9
0.96
y
0.95
0.98
1
1
1.02
1.05
0.85 0.9 0.95
1.04
1
1.05
1.1
1.15
1.2
(k)
1.05 1.1 1.15 1.2 1.25
(l)
0.9762
0.96
0.9763
0.965
0.9764
0.9765
0.97
0.9766
y
0.9767
0.975
0.9768
0.98
0.9769
0.977
0.985
0.9771
0.9772
1.0848
1.0852
1.0856
1.086
0.99
1.06
1.0864
(m)
1.07
1.08
1.09
1.1
1.11
(n)
0.92
0.93
0.94
0.95
0.96
y 0.97
0.98
0.99
1
1.01
1.02
1
1.02 1.04 1.06 1.08
1.1
1.12 1.14 1.16
x
(o)
Figure 3: Phase portraits corresponding to Figure 1 for diferent and . (a) = 1.6771238, = 0.13; (b) = 1.6771238, = 0.125; (c)
= 1.6771238, = 0.12457; (d) = 1.6771238, = 0.12; (e) = 1.6771238, = 0.115; (f) = 1.6771238, = 0.1134; (g) = 1.6771238,
= 0.09; (h) = 1.685, = 0.12428002; (i) = 1.78, = 0.12428002; (j) = 1.83, = 0.12428002; (k) = 1.75, = 0.14; (l) = 1.9,
= 0.14; (m) = 1.76, = 0.15; (n) = 1.77, = 0.15; (o) = 1.83, = 0.15.

the phase portraits show that Neimark-Sacker bifurcation
occurs.
5. Conclusion
In this paper, we investigated the 1 : 3 resonance of a discrete
Hindmarsh-Rose model. Here, we examined the fxed points
of the model in detail and showed that the model exhibits the
1 : 3 resonance. Furthermore, near 1 : 3 resonance point, the
Neimark-Sacker bifurcation and the homoclinic bifurcation
can occur. Te onset of 1 : 3 resonance means that, in some
cases, there is a region such that the model will have an
invariant circle from three-saddle cycle.
Here, we want to note that 1 : 3 resonance involves the
bifurcations of Z3 symmetric system, which are not discussed
in this paper. From the presented phase portraits, some
symmetric phenomena can be observed. Te homoclinic loop
can explain the transition between spiking and bursting.
Acknowledgments
Te authors thank the editor and the referees for their valuable suggestions and comments which led to the improvement of the paper.
References
[1] J. Guckenheimer and P. Holmes, Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields, Springer, Berlin,
Germany, 1983.
[2] Y. A. Kuznetsov, Elements of Applied Bifurcation Teory,
Springer, New York, NY, USA, 3rd edition, 2004.
[3] S. Wiggins, Introduction to Applied Nonlinear Dynamical Systems and Chaos, vol. 2, Springer, New York, NY, USA, 2nd
edition, 2003.
[4] J. L. Hindmarsh and R. M. Rose, A model of the nerve impulse
using two frst-order diferential equations, Nature, vol. 296,
no. 5853, pp. 162164, 1982.
[5] R. FitzHugh, Impulses and physiological state in theoretical
models of nerve membrane, Biophysical Journal, vol. 1, pp. 445
467, 1961.
[6] J. Nagumo, S. Arimoto, and S. Yoshizawa, An active pulse
transmission line simulating nerve axon, Proceedings of the IRE,
vol. 50, pp. 20612070, 1962.
[7] J. M. Gonzalez-Miranda, Complex bifurcation structures in
the Hindmarsh-Rose neuron model, International Journal of
Bifurcation and Chaos, vol. 17, no. 9, pp. 30713083, 2007.
[8] X. Liu and S. Liu, Codimension-two bifurcation analysis in
two-dimensional Hindmarsh-Rose model, Nonlinear Dynamics, vol. 67, no. 1, pp. 847857, 2012.
[9] M. Storace, D. Linaro, and E. de Lange, Te HindmarshRose neuron model: bifurcation analysis and piecewise-linear
approximations, Chaos, vol. 18, no. 3, Article ID 033128, 2008.
79
[10] S. Nikolov, An alternative bifurcation analysis of the RoseHindmarsh model, Chaos, Solitons and Fractals, vol. 23, no. 5,
pp. 16431649, 2005.
[11] S. Tsuji, T. Ueta, H. Kawakami, H. Fujii, and K. Aihara, Bifurcations in two-dimensional Hindmarsh-Rose type model, International Journal of Bifurcation and Chaos, vol. 17, no. 3, pp. 985
998, 2007.
[12] D. Linaro, A. Champneys, M. Desroches, and M. Storace,
Codimension-two homoclinic bifurcations underlying spike
adding in the Hindmarsh-Rose burster, SIAM Journal on
Applied Dynamical Systems, vol. 11, no. 3, pp. 939962, 2012.
[13] A. Shilnikov and M. Kolomiets, Methods of the qualitative
theory for the Hindmarsh-Rose model: a case study. A tutorial,
International Journal of Bifurcation and Chaos, vol. 18, no. 8, pp.
21412168, 2008.
[14] Q. Chen, Z. Teng, L. Wang, and H. Jiang, Te existence of
codimension-two bifurcation in a discrete SIS epidemic model
with standard incidence, Nonlinear Dynamics. An International
Journal of Nonlinear Dynamics and Chaos in Engineering Systems, vol. 71, no. 1-2, pp. 5573, 2013.
[15] Z. He and X. Lai, Bifurcation and chaotic behavior of a discretetime predator-prey system, Nonlinear Analysis. Real World
Applications, vol. 12, no. 1, pp. 403417, 2011.
[16] Z. He and J. Qiu, Neimark-Sacker bifurcation of a third-order
rational diference equation, Journal of Diference Equations
and Applications, vol. 19, no. 9, pp. 15131522, 2013.
[17] J. L. Hindmarsh and R. M. Rose, A model of neuronal
bursting using three coupled frst order diferential equations,
Proceedings of the Royal Society of London Series B: Biological
sciences, vol. 221, no. 1222, pp. 87102, 1984.
[18] A. L. Hodgkin and A. F. Huxley, A quantitative description
of membrane current and its application to conduction and
excitation in nerve, Te Journal of Physiology, vol. 117, no. 4, pp.
500544, 1952.
[19] G. Innocenti, A. Morelli, R. Genesio, and A. Torcini, Dynamical phases of the Hindmarsh-Rose neuronal model: studies of
the transition from bursting to spiking chaos, Chaos, vol. 17, no.
4, Article ID 043128, 2007.
[20] Z. Jing, Y. Chang, and B. Guo, Bifurcation and chaos in discrete
FitzHugh-Nagumo system, Chaos, Solitons & Fractals, vol. 21,
no. 3, pp. 701720, 2004.
[21] B. Li and Z. He, Bifurcations and chaos in a two-dimensional
discrete Hindmarsh-Rose model, Nonlinear Dynamics, vol. 76,
no. 1, pp. 697715, 2014.
[22] X. Liu and D. Xiao, Complex dynamic behaviors of a discretetime predator-prey system, Chaos, Solitons & Fractals, vol. 32,
no. 1, pp. 8094, 2007.
[23] S.-S. Chen, C.-Y. Cheng, and Y.-R. Lin, Application of a
two-dimensional Hindmarsh-Rose type model for bifurcation
analysis, International Journal of Bifurcation and Chaos in
Applied Sciences and Engineering, vol. 23, no. 3, Article ID
1350055, 2013.
[24] S. Ma and Z. Feng, Fold-Hopf bifurcations of the RoseHindmarsh model with time delay, International Journal of
Bifurcation and Chaos, vol. 21, no. 2, pp. 437452, 2011.
[25] D. Terman, Te transition from bursting to continuous spiking
in excitable membrane models, Journal of Nonlinear Science,
vol. 2, no. 2, pp. 135182, 1992.
[26] X.-J. Wang, Genesis of bursting oscillations in the HindmarshRose model and homoclinicity to a chaotic saddle, Physica D:
Nonlinear Phenomena, vol. 62, no. 14, pp. 263274, 1993.
80
[27] B. Li and Z. M. He, 1 : 2 and 1 : 4 resonances in a twodimensional discrete Hindmarsh-Rose model, Nonlinear
Dynamics, 2014.
[28] A. D. Polyanin and A. I. Chernoutsan, A Concise Handbook of
Mathematics, Physics, and Engineering Science, CRC Press, New
York, NY, USA, 2011.

http://dx.doi.org/10.1155/2014/918376
Research Article
Exponential Convergence for Numerical Solution of
Integral Equations Using Radial Basis Functions
Zakieh Avazzadeh,1 Mohammad Heydari,2 Wen Chen,1 and G. B. Loghmani3
1
State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, College of Mechanics and Materials,
Hohai University, Nanjing 210098, China
2
Young Researchers and Elite Club, Islamic Azad University, Ashkezar Branch, Ashkezar 8941613695, Iran
3
Department of Mathematics, Yazd University, P.O. Box 89195-741, Yazd, Iran
Correspondence should be addressed to Zakieh Avazzadeh; z.avazzadeh@yahoo.com
Received 4 July 2014; Revised 31 October 2014; Accepted 1 November 2014; Published 10 December 2014
Academic Editor: Xiao-wei Gao
Copyright 2014 Zakieh Avazzadeh et al. Tis is an open access article distributed under the Creative Commons Attribution
cited.
We solve some diferent type of Urysohn integral equations by using the radial basis functions. Tese types include the linear
and nonlinear Fredholm, Volterra, and mixed Volterra-Fredholm integral equations. Our main aim is to investigate the rate of
convergence to solve these equations using the radial basis functions which have normic structure that utilize approximation in
higher dimensions. Of course, the use of this method ofen leads to ill-posed systems. Tus we propose an algorithm to improve
the results. Numerical results show that this method leads to the exponential convergence for solving integral equations as it was
already confrmed for partial and ordinary diferential equations.
1. Introduction
Integral equations have been solved by many diferent methods [1, 2]. In [3] integral equations and methods of their
solving are classifed. Tis reference includes some traditional
methods for solving integral equations. But some recent
methods are Adomian decomposition method (ADM) [4],
homotopy perturbation method (HPM) [5], Hes variational
iteration methods [6], optimal control [7], wavelets [811],
neural networks [12], simulation method [13], block-pulse
method [14], and so forth. Also, there are many other articles
which contain new approaches in solving integral equations
[1518]. It is necessary to recall that most of the mentioned
methods are not easy for solving integral equations in higher
dimensions [1924] and also for solving the mixed VolterraFredholm cases [6, 2527]. However, in the present paper,
we restrict ourselves to the method of radial basis functions
(RBFs).
Te RBF methodology was introduced by Hardy [28].
At frst, it was popular in multivariate interpolation [2934].
In 1990, Kansa introduced a way to use these functions for
solving parabolic, hyperbolic, and elliptic partial diferential

equations [35]. Afer that, radial basis functions have been
widely applied in numerous felds. In spite of many other
applications of RBFs, we only focus on the use of RBFs for
solving integral equations.
In 1992, Makroglou [36] applied the collocation technique
to solve various linear and nonlinear integral equations. In
2002, Galperin and Kansa [37] applied RBFs for solution
of weakly singular Volterra integral equations by global
optimization. In 2007, Alipanah and Dehghan [38] used RBFs
for solving one-dimensional nonlinear Fredholm integral
equations without optimization technique and via quadrature
integration methods. In [39], this method is generalized for
two dimensions problems and the accuracy of the method
is compared with the traditional spectral method in [40].
Also, Avazzadeh et al. used the RBFs for solving partial
integrodiferential equations [4143].
In [38, 39], the method was applied for Fredholm integral
equations. In this paper, we describe the method for solving
more diferent types of integral equations such as Volterra and
82
types of integral equations can be solved by this method.

Terefore, the method can solve linear and nonlinear Fredholm, Volterra, and mixed Volterra-Fredholm equations even
in higher dimensions.
Te paper is organized in the following way. In Section
2, the radial basis functions as a tool for approximation are
introduced. In Section 3, we recall the method of the solution of Fredholm integral equation [38, 39] and then the
Volterra and mixed Volterra-Fredholm integral equations
will be solved by using radial basis functions. In Section 4,
some illustrating examples are presented. Te last section
includes conclusion and some ideas for future work.
2. Radial Basis Functions

Defnition 1. Consider a given set of distinct data points
{ }=0 and the corresponding data values { }=0 , the basic
RBF interpolant is given by
() = ( ) ,
(1)
c = f,
(2)
=0
where is the Euclidean norm, , R ( is a positive

fnite integer), and is scalar. Also (), 0, is some
radial basis functions. Te coefcient is determined from
the interpolation ( ) = , = 0, 1, . . . , , which leads to
the following symmetric linear system:
where the entries of c, f, and are given by
c = [0 , . . . , ] ,
= ( ) ,
f = [0 , . . . , ] ,
, = 0, 1, . . . , .
(3)
Sometimes, { }=0 are called center points. Every basis is

directly related to one center point. Since these points are
chosen arbitrarily, we have a mesh-free method [4446].
Te sufcient conditions for () in (3) to guarantee nonsingularity of the matrix are given in [47]. Also, Micchelli [33]
showed that a larger class of functions could be considered,
and thus the RBF method is uniquely solvable.
Tere are two kinds of radial basis functions, the piecewise smooth and the infnitely smooth radial functions. For
infnitely smooth radial functions, we have a shape parameter
. Te parameter is a free parameter for controlling the
shape of functions. As 0 the radial functions become
fatter [48, 49].
Some piecewise smooth RBFs are 3 (Cubic) and 2 log
(Tin plate spline) and some common infnitely smooth
examples of the () that lead to a uniquely solvable method
are in the following forms:
linear: ,
Gaussian (GA): () ,
Multiquadric (MQ): (1 + ()2 )/2 , ( = 0, = 2N),
2
inverse multiquadric (IMQ): (1 + ()2 )1/2 ,
inverse quadric (IQ): (1 + () ) .

2 1
Madych and Nelson have proved exponential convergence property of multiquadratic approximation [31, 50]. He
has shown that under certain conditions the interpolation
error is = (/ ) (note that MQ RBF has been redefned
from Hardys original defnition by the transformation =
1/), is the mesh size, and 0 < < 1 is a constant. As is
said in [51], this implies that there are two ways to improve
the approximation: either by reducing the size of or by
increasing the magnitude of . It means that if then
0. While reducing leads to the heavy computations,
increasing is without the extra computational cost. However, according to uncertainty principle of Schaback [52],
as the error becomes smaller, the matrix becomes more illconditioned; hence the solution will break down as becomes
too large. Nevertheless, there exists a wide range of that high
accurate results can be produced. So, if we could solve the illconditioned system, we could increase and obtain the best
approximation [50]. Tere are some experimental trials about
the shape parameter, ill-conditioning, and convergence [53
55].
Tere are some methods for trade-of between and
error [56, 57]. Te golden section algorithm [56] as a new
method for fnding a good shape parameter can be efective
but ofen it is expensive. Baxter [58] investigated the preconditioned conjugate gradient technique. Casciola et al. [59]
regularized the solutions with changing the Euclidean norm
to the anisotropic norm. Karageorghis et al. [60] applied the
matrix decomposition algorithm for improving 3D elliptic
problems. Also, there are the regularization techniques for
solving ill-conditioned systems such as truncated singular
value decomposition (TSVD) and Tikhonov regularization
method. Reader can see details in [5860] and the references
there in.
3. Integral Equation
3.1. Fredholm Integral Equation. Consider the following
Fredholm integral equation of the Urysohn form:
() (, , ()) = () ,
(4)
where is constant; () and (, , ()) are assumed to be

defned on the interval , . Let () be a radial
basis function and we approximate () with the following
interpolant function:
() ( ) = () ,
=0
(5)
where = [0 , 1 , . . . , ] and () = [( 0 ), (
1 ), . . . , ( )] . Now, by replacing (5) in (4) we obtain
() (, , ()) () .
(6)
83
In the above equation, , = 0, 1, . . . , , are unknown. For

computing them, we collocate the points , = 0, 1, . . . , , as
follows:
( ) ( , , ()) ( ) .
(7)
By applying the Legendre quadrature integration formula

[61], (5) can be changed to the following form:
( ) ( , , ( )) ( ) .
=0
(8)
Tis is a nonlinear system of equations that can be solved by

Newtons iterative method to obtain the unknown vector .
Similarly, for the two-dimensional integral equation,
consider the Fredholm integral equation as follows:
(, ) (, , , , (, ))
= (, ) ,
(, ) [, ] [, ] ,
(9)
where (, , , , (, )) and (, ) are given analytic

functions. According to (1) the function (, ) may be
represented by approximate series as
() ( ) = () ,
=0
(10)
where is any natural number, = (, ) R , and =

( , ) R2 . Noting the previous section, it is clear that
the collocation points { }=0 can be chosen as the centers.
However, the selection process of the center points can afect
accuracy; sometimes the uniform points or random points
are preferred.
Replacing (10) in (9) we have
(, ) (, , , , (, ))
(, ) ,
(, ) [, ] [, ] .
(11)
In the above equation, only ( = 0, 1, . . . , ) are unknowns

and it is an interesting technical advantage in using of RBFs. It
means the process of solving is no more complicated in spite
of increasing the dimension of the given problem.
Now we substitute the given collocation points in the
above equation. Collocation points can be the same center
points or any other arbitrary points:
( , ) ( , , , , (, ))
( , ) .
(12)
By applying the quadrature integration formula, (12) can be

changed to the following form:

( , ) ( , , , , ( , ))
=0 =0
( , ) .
(13)
Tis is a nonlinear system of equations that can be solved

by Newtons iterative method to obtain the unknown vector
. We recall that the obtained linearized system by Newtons
method is ill-conditioned and the use of regularization
methods is efcient. Also, we can apply some other iterative
regularization methods for solving an ill-conditioned nonlinear system of equations [62].
Te mentioned method for solving Fredholm integral
equation was discussed in [38, 39] and in the current paper
it is generalized for solving Volterra and Volterra-Fredholm
integral equations.
3.2. Volterra Integral Equation. Consider the following
Urysohn Volterra integral equation:
() (, , ()) = () ,
(14)
where is constant; () and (, , ()) are assumed to be

defned on the interval , . Similar to the previous
section, we substitute (5) in (14) and collocate the points ,
= 0, 1, . . . , . So we have
( ) ( , , ()) ( ) .
(15)
Using the linear transform as follows,

= () =
+

+
,
2
2
(16)
reduces (15) to the following integral equation:

( )
1
( , () , ( ())) ( ) .
2
1
(17)
Now, by applying a quadrature integration formula, we

approximate the integral in (17) and thus we have a nonlinear
system of equations that can be solved by Newtons iterative
method to obtain the unknown vector .
It is easy to use that the proposed method for solving the
two-dimensional Volterra integral equation gives
( (, )) (, , , , (, ))
= (, ) ,
(, ) [, ] [, ] ,
(18)
where (, , , , (, )) and (, ) are given analytic functions. Similarly, we substitute (10) in (18) and collocate the
points ( , ), , = 0, 1, . . . , . So we have
( ( , )) ( , , , , (, ))
( , ) .
(19)
84

Table 1: Numerical results of diferent RBFs for Example 1. Te roots
of Legendre polynomial are chosen as center points.
In the above equation, let the linear transforms

+
+
,
2
2
+

= ] () =
+
.
2
2
= () =
(20)
Terefore, (19) is reduced to the integral equation which can

be solved easily:
( ( , ))
1
( ) ( )
(21)
] () , ( () , ] ()))
( , ) .
Now, by applying a quadrature integration formula, we

approximate the obtained integrals; then the nonlinear system of equations can be solved by Newtons iterative method
to obtain the unknown vector .
3.3. Te Mixed Volterra-Fredholm Integral Equation. In general, Volterra-Fredholm integral equations can be classifed
into diferent types [6, 26]. We will only investigate the mixed
Volterra-Fredholm integral equations in the following form:
( (, )) (, , , , (, ))
= (, ) ,
(, ) [, ] [, ] ,
(22)
where (, , , , ) and (, ) are given analytic functions.

Similarly, we must substitute (10) in (22) and collocate the
points ( , ), , = 0, 1, . . . , . Afer that, let the linear
transforms
+
= () =
+
,
2
2
+

+
,
= ] () =
2
2
(23)
and apply the quadrature integration formula corresponding

to [1, 1]. Similar to the previous section, we get the following
system of equations:
( ( , ))
( , , ( ) ,
=0 =0
( , ) .
9
16
25
36
49
( , , () ,
1 1
] ( ) , ( ( ) , ] ( )))
(24)
Te obtained nonlinear system of equations can be solved by

Newtons iterative method to fulfll the unknown vector .
GA
7.6 10
MQ
9.0 10
2.1 10
1.6 105
3.0 107
2.0 10
1.1 105
7.0 107
6.6 108
2.5 1010
4.9 108
3.7 1010
IQ
7.5 103
1.6 104
1.4 105
2.6 107
6.2 108
2.8 1010
4. Numerical Example
In this section, some examples are given to show validity and
efciency of the mentioned method. In this paper, the computations have been done using the Maple 13 with 100
digits precision. Note that digits are important factor because
the obtained systems are ill-conditioned. For improving the
results we have two ways: increasing digits and applying a
regularization method. Since decreasing of digits leads to
intensive decreasing of accuracy [54], we preferred to use
high digits instead of applying complicated regularization
methods. It can be a trade-of between increasing of complexity of mathematical operations in applying a regularization
algorithm and increasing digits.
In our practical experiments with 100 digits, even applying SVD, QR, and iterative refnement methods did not
afect the results for solving the linear systems. It must be
mentioned that although it occurred in our experiments,
it can be related to the rate of ill-conditioning. Hence,
we reported the result of solving the obtaining nonlinear
systems by Newtons iteration method without applying any
regularization method. In this study, the criterion of accuracy
is the value of infnity norm of the error function.
Example 1. Consider the following Volterra integral equation
[63]:
(, ) (2 + cos ) (, )
0
1
1
1
= sin 5 + 5 cos 2 sin (2 ) ,
4
4
4
(25)
where (, ) [0, 1]2 and the exact solution is (, ) =

sin . Te error for some radial basis functions and for
diferent values of is given in Table 1.
As it is mentioned, the interpolation error is = (/ ),
is the mesh size, = 1/, and 0 < < 1 is a constant [50].
However the proof is for an interpolant function [50] and
we are involved with a nonlinear integral equation; since we
apply the collocation method, the error of the approximation
and interpolation techniques are nearly the same.
Now we investigate how afects error. Note that is
directly related to because is inversely related to mesh size:
1
.
(26)
85
Example 1
10
103
104
105
106
10
7
108
8
109
9
1010
10
20
30
40
10
20
30
40
n
Example 1
Example 2
Example 3
(b)
(a)
Example 3
Example 2
2
3
4
3
5
6
7
4
8
9
10
11
10
20
30
40
10
20
30
40
n
(c)
(d)
Figure 1: Te horizontal axis is related to and the vertical axis shows the corresponding error values. It is perceived that the relation between
and log (error) is linear. It confrms can afect error exponentially. We emphasize the spectral accuracy is the most important feature of
the method. Te relevant data is presented in Tables 1, 2, and 3.
So, we expect increasing of decreases exponentially [44,

50]. In Figure 1, we show how afects error. Also, we
show the efect of on error in Figure 2. We must note the
obtained system is nonlinear. Terefore, although we expect
the exponential trend, the nature of nonlinearity has negative
efect on the rate of convergence which is diferent for each
problem.
Example 2. Consider the following Volterra nonlinear integral equation [20]:
(, ) [(, )]2
0
1
2
1
= + 5 + 3 3 5 ,
5
9
5
2
(, ) [0, 1]2 ,
(27)
86
10
Table 2: Numerical results of diferent RBFs for Example 2. Te

roots of Legendre polynomial are chosen as center points.
Error
104
9
16
25
106
10
36
49
1010
GA
2.2 10
2.5 10
3.2 107
7.1 108
6.0 1011
1.5 1011
MQ
2.6 10
1.2 10
3.0 107
7.4 108
4
5.1 1011
2.1 1011
IQ
2.2 102
5.2 104
4.4 107
8.0 108
5.9 1011
2.0 1011
Table 3: Numerical results of diferent RBFs for Example 3. Te

roots of Legendre polynomial are chosen as the center points.
0.1
0.5
5
c
10
50 100
Figure 2: Te horizontal axis is related to or shape parameter and

the vertical axis shows the corresponding error values where = 49.
Te relation between and log (error) is linear in the interval (0, 10).
Although the trend is changed, the graph confrms can afect error
exponentially. In brief, the error function is sensitive to intensively.
Te data is related to Example 1.
with the exact solution (, ) = 2 +2 . Error for some radial

basis functions for diferent values of is given in Table 2. Te
efect of on error is shown in Figure 1.
Example 3. Consider the following Volterra-Fredholm integral equation of Urysohn type [24]:
(, )
0
= ln (1 +
(1 2 )
(1 + ) (1 + 2 )
(1 exp(,) )
)+
,
2
1+
8 (1 + ) (1 + 2 )
where
(, ) = (, )
(, , , , (, )) (, ) ,
4
9
16
25
36
49
(28)
(29)
(30)
such that is the approximated solution, is a large number,

and ( , ) are points uniformly distributed over the domain.
Tis criterion shows tolerance of error in the solution region.
GA
MQ
IQ
1.2 102
2.1 103
8.0 104
1.7 102
2.4 104
7.7 104
1.2 102
2.4 103
6.5 104
5.2 106
4.9 106
7.2 106
1.0 104
1.6 105
3.1 104
2.8 105
3.4 104
1.4 105
Table 4: Value of the root of mean square of residual errors that is

computed by (29). Te results for Examples 1, 2, and 3 are reported
for Gaussian radial basis functions and = 400. Comparison
between Tables 1, 2, and 3 and following results confrms the strong
correlation between absolute maximum of error and the root of
mean square of residual errors. In brief, the estimation error of
approximation is possible by (29) even as the exact solution is not
given.
where (, ) [0, 1]2 and the exact solution is (, ) =

ln(1 + /(1 + 2 )). Error for some radial basis functions
for diferent values of is given in Table 3. Te exponential
efect of on error is shown in Figure 1.
So far we compute the errors in the infnity norm that are
shown in Tables 1, 2, and 3. Now, we compute the root mean
square residual errors by the formula
=1 ,
4
9
16
25
36
49
Example 1
Example 2
Example 3
1.4 107
3.1 108
8.3 104
7.5 106
1.0 10
1.8 105
9.3 10
8.9 105
8.5 108
6.8 109
2.0 1011
5.2 1012
1.8 105
1.4 106
1.2 103
2.1 104
5.9 106
8.2 107
Te results for Examples 1, 2, and 3 are reported in Table 4

for the Gaussian radial basis function with = 400. In fact,
if we have the exact solution, we can compute the absolute
maximum of error or infnity norm of error function; otherwise, we must compute the root mean square residual error
which is defned in (29). Te comparison between Tables
1, 2, 3, and 4 confrms the strong correlation between the
absolute maximum of errors and the root of mean square
of residual errors. So, even without having the exact solution we still will be able to estimate the error of approximation.
5. Conclusion
First, we exploit some technical methods that can be used to
improve results.

Supplementary Techniques
Quadrature Integration Methods. we applied the generalized Gauss-Lobatto quadrature on interval [, ] where the
boundaries of the interval coincide with the collocation
points. Since 0 = and = in Gauss-Lobatto quadrature,
we can improve the approximation on boundaries and the
results are stable on boundary.
Regularization. We suggest applying the regularization methods for solving the resulting linear and particularly nonlinear
systems. Considering there is no guarantee that Newtons
method leads to the convergent solution, using of the regularization methods such as Tikhonov or Landweber is defnitely
recommended for ill-conditioned nonlinear system [62].
Partitioning. As mentioned the obtained systems by the
collocation points are ill-conditioned. Moreover, the illconditioning is worse by increasing the number of collocation points. Terefore, to avoid the ill-conditioning we
can partition domain of problems to the smaller area. Tis
technique gives the smaller systems that can be solved easier.
Te numerical experiments show that performance of this
technique can be efective when it is not possible to improve
the tools of computation.
Finding the accurate solution of the two-dimensional
integral equations is usually difcult. In this work, the linear
and nonlinear, Fredholm, Volterra, and mixed VolterraFredholm integral equations of the second kind are solved
by applying the radial basis functions (RBFs) method. Te
illustrative examples confrm the exponential convergence
rate for integral equations similar to rate of convergence for
solving partial and ordinary deferential equations using RBF
method reported in [53].
Acknowledgments
Te work described in this paper was supported by the
National Basic Research Program of China (973 Project
no. 2010CB832702), the National Science Funds for Distinguished Young Scholars of China (1077403), NSFC Funds
(no. 11372097), the 111 project under Grant B12032. Also,
the authors greatly appreciate the precious time of Professor
Mahdi Dehghan (from Amirkabir University, Iran) who help
them generously.
References
[1] K. E. Atkinson, Te Numerical Solution of Integral Equations of
the Second Kind, Cambridge University Press, Cambridge, UK,
1997.
[2] A. J. Jerri, Introduction to Integral Equations with Applications,
John Wiley & Sons, New York, NY, USA, 1999.
[3] A. D. Polyanin and A. V. Manzhirov, Handbook of Integral Equations, Chapman & Hall/CRC, 2008.
87
[4] A.-M. Wazwaz, A First Course in Integral Equations, World Scientifc, Singapore, 1997.
[5] S. Abbasbandy, Application of Hes homotopy perturbation
method to functional integral equations, Chaos, Solitons &
Fractals, vol. 31, no. 5, pp. 12431247, 2007.
[6] S. A. Yousef, A. Lotf, and M. Dehghan, Hes variational iteration method for solving nonlinear mixed Volterra-Fredholm
integral equations, Computers & Mathematics with Applications, vol. 58, no. 11-12, pp. 21722176, 2009.
[7] S. A. Belbas, A new method for optimal control of Volterra
integral equations, Applied Mathematics and Computation, vol.
189, no. 2, pp. 19021915, 2007.
[8] Y. Mahmoudi, Wavelet Galerkin method for numerical solution of nonlinear integral equation, Applied Mathematics and
Computation, vol. 167, no. 2, pp. 11191129, 2005.
[9] K. Maleknejad, R. Mollapourasl, and M. Alizadeh, Numerical
solution of Volterra type integral equation of the frst kind with
wavelet basis, Applied Mathematics and Computation, vol. 194,
no. 2, pp. 400405, 2007.
[10] G. Vainikko, A. Kivinukk, and J. Lippus, Fast solvers of integral
equations of the second kind: wavelet methods, Journal of
Complexity, vol. 21, no. 2, pp. 243273, 2005.
[11] S. Yousef and M. Razzaghi, Legendre wavelets method for the
nonlinear Volterra-Fredholm integral equations, Mathematics
and Computers in Simulation, vol. 70, no. 1, pp. 18, 2005.
[12] A. Golbabai, M. Mammadov, and S. Seifollahi, Solving a system
of nonlinear integral equations by an RBF network, Computers
& Mathematics with Applications, vol. 57, no. 10, pp. 16511658,
2009.
[13] M. H. Kalos and P. A. Whitlock, Monte Carlo Methods, WileyBlackwell, Weinheim, Germany, 2nd edition, 2004.
[14] K. Maleknejad, M. Shahrezaee, and H. Khatami, Numerical
solution of integral equations system of the second kind by
block-pulse functions, Applied Mathematics and Computation,
vol. 166, no. 1, pp. 1524, 2005.
[15] Z. Avazzadeh, M. Heydari, and G. B. Loghmani, Numerical
solution of Fredholm integral equations of the second kind
by using integral mean value theorem, Applied Mathematical
[16] K. Maleknejad and H. Derili, Numerical solution of integral
equations by using combination of spline-collocation method
and Lagrange interpolation, Applied Mathematics and Computation, vol. 175, no. 2, pp. 12351244, 2006.
[17] E. Mammen and K. Yu, Nonparametric estimation of noisy
integral equations of the second kind, Journal of the Korean
Statistical Society, vol. 38, no. 2, pp. 99110, 2009.
[18] H. K. Pathak, M. S. Khan, and R. Tiwari, A common fxed point
theorem and its application to nonlinear integral equations,
Computers & Mathematics with Applications, vol. 53, no. 6, pp.
961971, 2007.
[19] S. A. Belbas, Optimal control of Volterra integral equations in
two independent variables, Applied Mathematics and Computation, vol. 202, no. 2, pp. 647665, 2008.
[20] H. Guoqiang, K. Hayami, K. Sugihara, and W. Jiong, Extrapolation method of iterated collocation solution for twodimensional nonlinear Volterra integral equations, Applied
Mathematics and Computation, vol. 112, no. 1, pp. 4961, 2000.
[21] G. Han and J. Wang, Extrapolation of Nystrom solution for two
dimensional nonlinear Fredholm integral equations, Journal of
Computational and Applied Mathematics, vol. 134, no. 1-2, pp.
259268, 2001.
88
[22] M. Heydari, Z. Avazzadeh, H. Navabpour, and G. B. Loghmani,
Numerical solution of Fredholm integral equations of the
second kind by using integral mean value theorem II. High
dimensional problems, Applied Mathematical Modelling, vol.
37, no. 1-2, pp. 432442, 2013.
[23] K. Maleknejad, S. Sohrabi, and B. Baranji, Application of
2D-BPFs to nonlinear integral equations, Communications in
Nonlinear Science and Numerical Simulation, vol. 15, no. 3, pp.
527535, 2010.
[24] A. M. Wazwaz, A reliable treatment for mixed VolterraFredholm integral equations, Applied Mathematics and Computation, vol. 127, no. 2-3, pp. 405414, 2002.
[25] M. A. Abdou, K. I. Mohamed, and A. S. Ismail, On the numerical solutions of Fredholm-Volterra integral equation, Applied
Mathematics and Computation, vol. 146, no. 2-3, pp. 713728,
2003.
[26] A. A. Badr, Numerical solution of Fredholm-Volterra integral
equation in one dimension with time dependent, Applied
Mathematics and Computation, vol. 167, no. 2, pp. 11561161,
2005.
[27] A. Yildirim, Homotopy perturbation method for the mixed
Volterra-Fredholm integral equations, Chaos, Solitons and
Fractals, vol. 42, no. 5, pp. 27602764, 2009.
[28] R. L. Hardy, Multiquadratic equation of topology and other
irregular surface, Journal of Geophysical Research, vol. 76, pp.
19051915, 1971.
[29] M. D. Buhmann, Multivariate interpolation in odd-dimensional Euclidean spaces using multiquadrics, Constructive
Approximation, vol. 6, no. 1, pp. 2134, 1990.
[30] M. D. Buhmann and C. A. Micchelli, Multiquadric interpolation improved, Computers & Mathematics with Applications,
vol. 24, no. 12, pp. 2125, 1992.
[31] W. R. Madych and S. A. Nelson, Multivariate interpolation
and conditionally positive defnite functions, II, Mathematics
of Computation, vol. 54, no. 189, pp. 211230, 1990.
[32] W. R. Madych and S. A. Nelson, Multivariate interpolation
and conditionally positive defnite functions, Approximation
Teory and its Applications, vol. 4, no. 4, pp. 7789, 1988.
[33] C. A. Micchelli, Interpolation of scattered data: distance matrices and conditionally positive defnite functions, Constructive
Approximation, vol. 2, no. 1, pp. 1122, 1986.
[34] G. B. Wright, Radial basis function interpolation: numerical and
analytical developments [Ph.D. Dissertation], University of Colorado, 2003.
[35] E. J. Kansa, Multiquadricsa scattered data approximation
scheme with applications to computational fuid-dynamicsII
solutions to parabolic, hyperbolic and elliptic partial diferential
equations, Computers & Mathematics with Applications, vol. 19,
no. 8-9, pp. 147161, 1990.
[36] A. Makroglou, Radial basis functions in the numerical solution
of Fredholm integral and integro-diferential equations, in
Advances in Computer Methods for Partial Diferential Equations, VII, R. Vichnevetsky, D. Knight, and G. Richter, Eds., pp.
478484, Karen Hahn Imacs, New Brunswick, NJ, USA, 1992.
[37] E. A. Galperin and E. J. Kansa, Application of global optimization and radial basis functions to numerical solutions of weakly
singular Volterra integral equations, Computers & Mathematics
with Applications, vol. 43, no. 35, pp. 491499, 2002.
[38] A. Alipanah and M. Dehghan, Numerical solution of the nonlinear Fredholm integral equations by positive defnite functions, Applied Mathematics and Computation, vol. 190, no. 2,
pp. 17541761, 2007.

[39] A. Alipanah and S. Esmaeili, Numerical solution of the twodimensional Fredholm integral equations using Gaussian radial
basis function, Journal of Computational and Applied Mathematics, vol. 235, no. 18, pp. 53425347, 2011.
[40] Z. Avazzadeh, M. Heydari, and G. B. Loghmani, A comparison
between solving two dimensional integral equations by the
traditional collocation method and radial basis functions,
Applied Mathematical Sciences, vol. 5, no. 2124, pp. 11451152,
2011.
[41] Z. Avazzadeh, M. Heydari, W. Chen, and G. Loghmani,
Smooth solutio of partial integro-diferential equations using
radial basis functions, Te Journal of Applied Analysis and Computation, vol. 4, no. 2, pp. 115127, 2014.
[42] Z. Avazzadeh, Z. Beygi Rizi, F. M. Maalek Ghaini, and G.
B. Loghmani, A semi-discrete scheme for solving nonlinear hyperbolic-type partial integro-diferential equations using
radial basis functions, Journal of Mathematical Physics, vol. 52,
no. 6, Article ID 063520, 15 pages, 2011.
[43] Z. Avazzadeh, Z. Beygi Rizi, F. M. Maalek Ghaini, and G. B.
Loghmani, A numerical solution of nonlinear parabolic-type
Volterra partial integro-diferential equations using radial basis
functions, Engineering Analysis with Boundary Elements, vol.
36, no. 5, pp. 881893, 2012.
[44] M. D. Buhmann, Radial Basis Functions, Cambridge University
Press, 2003.
[45] M. D. Buhmann, Multivariable interpolation using radial basis
functions [Ph.D. Dissertation], University of Cambridge, 1989.
[46] W. Cheney and W. Light, A Course in Approximation Teory,
William Allan, New York, NY, USA, 1999.
[47] I. J. Schoenberg, Metric spaces and completely monotone functions, Annals of Mathematics, vol. 39, no. 4, pp. 811841, 1938.
[48] T. A. Driscoll and B. Fornberg, Interpolation in the limit of
increasingly fat radial basis functions, Computers & Mathematics with Applications, vol. 43, no. 35, pp. 413422, 2002.
[49] B. Fornberg, G. Wright, and E. Larsson, Some observations
regarding interpolants in the limit of fat radial basis functions,
Computers & Mathematics with Applications, vol. 47, no. 1, pp.
3755, 2004.
[50] W. R. Madych, Miscellaneous error bounds for multiquadric
and related interpolators, Computers & Mathematics with
Applications, vol. 24, no. 12, pp. 121138, 1992.
[51] J. Li, A. H.-D. Cheng, and C.-S. Chen, A comparison of efciency and error convergence of multiquadric collocation method
and fnite element method, Engineering Analysis with Boundary
Elements, vol. 27, no. 3, pp. 251257, 2003.
[52] R. Schaback, Error estimates and condition numbers for
radial basis function interpolation, Advances in Computational
Mathematics, vol. 3, no. 3, pp. 251264, 1995.
[53] A. H. Cheng, M. A. Golberg, E. J. Kansa, and G. Zammito,
Exponential convergence and H-c multiquadric collocation
method for partial diferential equations, Numerical Methods
for Partial Diferential Equations, vol. 19, no. 5, pp. 571594,
2003.
[54] C.-S. Huang, C.-F. Lee, and A. H.-D. Cheng, Error estimate,
optimal shape factor, and high precision computation of multiquadric collocation method, Engineering Analysis with Boundary Elements, vol. 31, no. 7, pp. 614623, 2007.
[55] Z.-C. Li and H.-T. Huang, Study on efective condition number
for collocation methods, Engineering Analysis with Boundary
Elements, vol. 32, no. 10, pp. 839848, 2008.

[56] C. H. Tsai, J. Kolibal, and M. Li, Te golden section search algorithm for fnding a good shape parameter for meshless collocation methods, Engineering Analysis with Boundary Elements,
vol. 34, no. 8, pp. 738746, 2010.
[57] J. G. Wang and G. R. Liu, On the optimal shape parameters of
radial basis functions used for 2-D meshless methods, Computer Methods in Applied Mechanics and Engineering, vol. 191,
no. 23-24, pp. 26112630, 2002.
[58] B. J. Baxter, Preconditioned conjugate gradients, radial basis
functions, and Toeplitz matrices, Computers & Mathematics
with Applications, vol. 43, no. 35, pp. 305318, 2002.
[59] G. Casciola, L. B. Montefusco, and S. Morigi, Te regularizing
properties of anisotropic radial basis functions, Applied Mathematics and Computation, vol. 190, no. 2, pp. 10501062, 2007.
[60] A. Karageorghis, C. S. Chen, and Y. S. Smyrlis, Matrix decomposition RBF algorithm for solving 3D elliptic problems, Engineering Analysis with Boundary Elements, vol. 33, no. 12, pp.
13681373, 2009.
[61] P. K. Kythe and M. R. Schaferkotter, Handbook of Computational
Method for Integration, Chapman & Hall/CRC Press, 2005.
[62] B. Kaltenbacher, A. Neubauer, and O. Scherzer, Itera tive
Regularization Methods for Nonlinear Ill-Posed Problems, Walter
de Gruyter, 2008.
[63] A. Tari, M. Y. Rahimi, S. Shahmorad, and F. Talati, Solving a
class of two-dimensional linear and nonlinear Volterra integral
equations by the diferential transform method, Journal of
Computational and Applied Mathematics, vol. 228, no. 1, pp. 70
76, 2009.
89

http://dx.doi.org/10.1155/2014/918376
Research Article
Working with Missing Data: Imputation of
Nonresponse Items in Categorical Survey Data with
a Non-Monotone Missing Pattern
Machelle D. Wilson1 and Kerstin Lueck2,3
1
Department of Public Health Sciences, Division of Biostatistics, University of California, Davis, Davis, CA 95616, USA
Social Psychology, Te University of Adelaide, Adelaide, SA 5005, Australia
3
Department of Integration and Confict, Max Planck Institute, 06017 Halle, Germany
2
Correspondence should be addressed to Machelle D. Wilson; mdwilson@phs.ucdavis.edu

Received 9 June 2014; Accepted 16 October 2014; Published 7 December 2014
Academic Editor: Jin Liang
Copyright 2014 M. D. Wilson and K. Lueck. Tis is an open access article distributed under the Creative Commons Attribution
cited.
Te imputation of missing data is ofen a crucial step in the analysis of survey data. Tis study reviews typical problems with missing
data and discusses a method for the imputation of missing survey data with a large number of categorical variables which do not
have a monotone missing pattern. We develop a method for constructing a monotone missing pattern that allows for imputation
of categorical data in data sets with a large number of variables using a model-based MCMC approach. We report the results of
imputing the missing data from a case study, using educational, sociopsychological, and socioeconomic data from the National
Latino and Asian American Study (NLAAS). We report the results of multiply imputed data on a substantive logistic regression
analysis predicting socioeconomic success from several educational, sociopsychological, and familial variables. We compare the
results of conducting inference using a single imputed data set to those using a combined test over several imputations. Findings
indicate that, for all variables in the model, all of the single tests were consistent with the combined test.
1. Introduction
Te problem of bias due to missing data has received a good
deal of attention over the last 20 years and the correction of
bias due to item and unit nonresponse remains an important
problem for investigators using survey data [19]. For data
missing because of item nonresponse, imputation of the
missing data is ofen the best solution. However, methods
for imputing categorical data are still experimental in some
sofware releases. Many sofware packages will automatically
remove cases with missing values from the analysis, greatly
reducing the sample size, ofen causing a drastic loss of information. Additionally, if the data are not missing completely
at random, removing cases with missing items will result in
biased parameter estimates in subsequent analyses.
Durrant [10] conducted an extensive review of various
imputation methods. She showed that parameter estimates
can vary considerably with diferent methods and noted

their advantages and disadvantages. She noted that with the
regression model approach problems arise with failures in
model assumptions, but in the case where model assumptions hold the modelling approach works well. Regression
modelling is superior to mean imputation or similar methods in that it makes use of the information in the entire
sample to impute the missing value for each observation.
However, as with mean imputation, regression imputation
leads to underestimation of the variability in the data, because
predicted values from the regression model are treated in
the substantive analysis as if they were random observations
from the sample population, leading to biased estimates
of the population variance and subsequently the standard
errors used to conduct inference about the model parameters.
Tere does not appear to be a consensus as to the best
method, as much depends on the nature of the data and
91
the missing data process. However, in many situations, the
use of model-based approaches such as Markov chain Monte
Carlo methods (MCMC) is superior. Tese methods defne
a model or distribution for the missing data (the missing
data model) and sample from this distribution to impute the
missing values [9], hence simulating or mimicking random
sampling from the population of interest. Tese methods
have been shown via theory and simulation to converge to
the true target distribution. Sampling from the distribution
of the missing variable reduces the underestimation of the
population variance, as well as the standard errors of the
parameter estimates for the substantive model, which would
result from most other single-imputation methods such
as mean imputation or regression imputation alone. Ofen
the model-based approaches are combined with regression
modelling to perform the imputation so as to also more
fully exploit the information in the data. For categorical
data, logistic regression is a natural choice and has the
advantage of accurately modelling the distribution of the
missing data given the observed data. Te parameters are
easily estimated via the incomplete observed data [11, 12].
However, the model-based approaches in conjunction with
logistic regression become problematic for data sets with a
large number of variables. Tis is because the relationships
between the variables are modelled via cross-tabulation and
the size of the contingency table grows exponentially with
additional variables, ofen creating a situation which exceeds
the limitations of the sofware package [13]. Te limitation
can be circumvented if the missing data pattern is monotone.
However, assessing the existence of a monotone missing
pattern is equally problematic for a large number of variables.
Furthermore, for the typical researcher in education and
the social sciences, ease of implementation of the imputation
so as to devote time and energy to the substantive model
is of more interest than programming advanced statistical
algorithms. Hence, our goal here is to provide a simple
method for exploiting the modelling capabilities of SAS or
other sofware packages and circumventing the difculties for
the case of a large number of categorical variables with a nonmonotone missing data pattern.
In this paper we will review the degrees of randomness
and the implications for imputation. We will discuss MCMC
implementation of multivariate normal data and MCMC
combined with logistic regression for imputation of categorical data, the monotone missing requirement, creation of
a monotone missing pattern, and how to perform multiple
imputations using this method. Finally, we will apply the
method to data drawn from the National Latino and Asian
American Study (NLAAS) and show the use of multiple
imputations for an example substantive model using these
data.
1.1. Degrees of Randomness in Missing Data. A missing data
process is said to be missing completely at random (MCAR)
if the probability that a subject is missing a variable is
completely independent of the value of the variable and of
the values of any other variables. For example, missing data
from lost survey pages are MCAR because, presumably, the
probability that a page was lost is not in any way related

to the value of any of the variables measured on that page,
nor to any other possible variables related to the missing
data. To restate, a missing variable is said to be MCAR if
the probability that the variable is missing from a subject is
neither related to the value of the missing variable nor to the
value of any other variable for that subject.
A missing variable is said to be missing at random (MAR)
when the probability that a variable is missing from a given
subject depends on the value of another variable for that
subject but not on the value of the missing variable itself. For
example, if we have a variable income that is more likely to be
missing from respondents with higher levels of education but
is no more likely to be missing for higher or lower incomes,
then both the missing and the observed income values in a
survey are a random sample of the population of income at
a given level of education but are not a random sample of
the population of all possible income. Tat is, the conditional
distribution of income given education is unbiased and is
representative of the distribution of income for any given
level of education. Hence, the income variables are missing
at random given the education level. However, the sample
of income may be biased for the unconditional distribution
of income because any relationship between income and
education will cause bias in the observed sample. For example,
if higher education is correlated with higher income, then
the sample mean of all incomes in the data set will be
biased downwards because the higher incomes associated
with higher education were more likely to be missing.
If the probability that a particular question is not
answered is dependent on the answer itself, then the missing
data process is nonrandom and the resulting bias in the
parameter estimates cannot be corrected without information from outside the sample. For example, if low-income
respondents are more likely to refuse to answer a question
about their income level, then the estimates for income will
be biased, since lower levels of income were more likely to
be excluded. If we have other sources of information about
income, we may be able to correct this, but the sample itself
is biased and by itself will produce biased parameter estimates. Similarly, the model parameter estimates containing
statistics based on the sample values for income may also
be biased. Tis issue is similar to bias that can result from
unit nonresponse to surveys and similar corrective measures
may be possible. In any case, certainly, avoidance of both unit
and item nonresponse is the best solution, when possible [6
8, 14].
For most survey data, including the National Latino and
Asian American Study, we cannot assume that the missing
data are MCAR. Some respondents may be more likely
to refuse to answer certain questions depending on their
understanding of the question, their education level, their
cultural identity, or other characteristics. However, it can
usually be argued for surveys with a large number of variables
that the missing data in the survey can be assumed to be
MAR because we have a large number of variables with which
to model the missing data process. Tat is, the larger the
number of variables we have, the more likely that there is a
variable (or a combination of variables) in our data set for
92
which the conditional distribution of the missing variable is

unbiased [11, 15, 16].
It should be noted here that if the number of continuous
variables in the data set is small, we are more likely to
encounter problems with the MAR assumption. Under the
MAR assumption, unbiased estimates of the missing data
values can be obtained by conditioning the imputed value
on the observed variables that model the missing data
process. Imputation of missing data using small data sets can
increase the risk of violating the MAR assumption in that the
missingness may depend on a variable we did not include
in the imputation model. If auxiliary variables (variables not
intended for the substantive model, but may be correlated
with those that are) can be collected, it is expedient to do
so [17, 18]. Care should be taken and expert knowledge
employed to consider possible relationships between variables and the probability of missingness when building the
imputation model.
1.2. Imputation of Missing Data Using Bayesian Methods. One
useful approach to imputation is to use a Bayesian modelbased method. In this method, a posterior distribution for
the parameters of the missing data distribution given the
observed data is obtained using Bayes Rule. Te posterior
distribution is based on the maximum likelihood estimates
for the population parameters for the data and a prior
distribution for the parameters. Te prior distribution for the
parameters models our uncertainty about the true parameter
values. In general, the posterior distribution of the parameter
vector, , of the distribution of a given random variable, , is
expressed as
Pr { | } =
{ | } ()
,
{ | } ()
(1)
where () is the prior distribution of , { | } is the

likelihood function for the data, and the denominator is the
sum (or integral for continuous sets) over all possible values
of . Te denominator is a normalizing constant (once the
data have been observed) that ensures that the posterior
distribution is a valid probability distribution [19]. In the
missing data context, once we have obtained the posterior
distribution of the parameters of the missing data distribution, we sample the missing data population parameter values
from their posterior distribution using simulation, and then
we impute the missing data by sampling via simulation of
the missing data values from their distribution given the
previously sampled parameter values and the observed data.
Finding the posterior distribution ofen involves the use of
MCMC methods, most ofen the Gibbs sampler. Note that, in
the Bayesian framework, the parameters are random variables
and come from a probability distribution. Tis probability
distribution is meant to model our uncertainty in the model
parameter values. In practice, if we have no prior information
about the distribution of the parameters, we can specify
a non-informative prior. With non-informative prior, the
Bayesian point estimate for the parameters will be equal to the
maximum likelihood estimator in most situations. Te beneft of the Bayesian approach, especially for the imputation of
missing values, is that, unlike with regression imputation, we

are not limited to point estimation in building the imputation
model but can model our uncertainty in the parameter
estimates for each imputed variable by sampling them from
their posterior distribution for each imputation and thus
each imputation uses a diferent parameter estimate in the
imputation model to impute the missing value, thus more
realistically incorporating variability due to uncertainty in the
parameter estimates into the imputed data set.
In non-model-based approaches, such as mean imputation, hot-deck imputation, and regression imputation, the
variability in the imputed data will be less than the variability in the population. Tat is, variance estimates based
on the complete data (the observed data with imputed
values replacing the missing data) will be biased downwards
because the imputed data does not contain information
about uncertainty in the parameter estimates used in the
imputation [11, 2022]. Likewise, standard errors based on
the variance estimates will be biased downwards, possibly
afecting inference in the substantive model. Hence, with
the Bayesian approach, the downward bias of the population
variance estimates for the complete data set is reduced
because we are modelling the uncertainty in the imputation
model parameters.
1.3. Imputation of Multivariate Normal Data. In the method
we discuss here we need to frst impute any multivariate
normal data that we may have in our survey before imputing
the categorical data. Having at least a few complete variables
(either observed or imputed) will help us in establishing a
monotone missing pattern for the categorical data. Some
sofware packages, such as SAS, can perform model-based
imputations, such as the Bayesian method described above,
within a canned procedure so that the investigator does not
need to have mastered advanced statistical computing.
In the context of missing multivariate normal data
we aim to sample from the posterior distribution, ( |
obs , mis ), to obtain estimates for the population parameters,
the mean vector, , and the covariance matrix . Once we
have population parameter estimates we sample from the
distribution of the missing data, given the parameters and
the observed data, (mis | , , obs ), in order to impute
the missing data. Tese methods produce a Markov chain
whose stationary distribution is the target distribution. For
data from an approximately multivariate normal distribution,
the imputation process involves the following steps.
(1) Te Imputation Step. At step , given the current estimates for the mean vector, t = (1 , 2 , 3 , . . . , ),
and covariance matrix, t , the I-step simulates the
missing values for each missing observation independently. Tat is, if the variables with missing values
for the th observation are denoted by (mis) and the
variables with observed values by (obs) , then the Istep draws independent values for (mis) from the
current conditional distribution for (mis) given (obs)
and t = (t , t ).
(2) Te Posterior Step. Given a complete sample, that
is, with all missing values provisionally imputed,
93

the P-step simulates the mean vector and covariance
matrix from their respective posterior distributions.
Tese new estimates are then used in the next I-step,
+ 1.
Tese two steps constitute a Markov chain whose equilibrium

distribution is the true distribution of (mis) given (obs)
and the parameter estimates that have been simulated from
their respective posterior distributions. Tat is, with a current
parameter estimate t at the th iteration, the I-step draws
(+1)
from (mis | obs , t ) and the P-step draws t+1
mis
+1
from ( | obs , mis
). Tis produces a Markov chain
+1 B+1
+2 B+2
, . . . , (mis , ), (mis , ), . . ., where is a large number
such that the chain has converged to the target distribution
(mis | obs , ). Once the chain has converged, each
simulation is an independent draw from this distribution, the
values of which are then used to impute the missing data [11].
1.4. Imputation of Categorical Data. For ordinal or nominal

data, we can use a logistic regression model to impute
missing data once a monotone missing data pattern had been
established and the posterior distribution of the parameters
for the regression imputation model has been found. Once
a model has been ftted, the missing values can be imputed
using the predicted values from the model [4, 23].
For a missing binary class variable with possible
outcomes 0 and 1, we ft a logistic regression model using the
observed data for and its covariates 1 , 2 , 3 , . . . , and
a vector of sampled from their posterior distribution. We
have
logit ( ) = 0 + 1 1 + 2 2 + + ,
(2)
= Pr ( = 1 | 1 , 2 , . . . , )
(3)
where 1 , 2 , 3 , . . . , are the covariates for , where

and where
logit ( ) = log (
).
(4)
Te imputed values are simulated algorithmically using the

following steps.
(1) At step , randomly draw a new parameter vector,

= (1 , 2 , . . . , ), from the current posterior
predictive distribution, where = t1 + Vhj Z,
where Vhj is the upper triangular matrix of the
Cholesky decomposition and Z is a vector of + 1
independent random normal variates. Te posterior
predictive distribution is updated at each step, , given
the observed data and the imputed data from the last
step.
(2) For each observation with missing given covariates
1 , 2 , 3 , . . . , and t fnd the expected probability that = 1 given by .
(3) Draw , a uniform (0, 1) random variable. If < ,
impute = 1; else impute = 0.
Te above algorithm produces a Markov chain whose

stationary distribution converges to the true distribution of
. Te imputed are our best estimates of the true value of
the missing variable for each observation given the observed
data and the covariates. Furthermore, sampling from the
posterior distribution of the parameter vector, , models
our uncertainty in the imputation model parameters, thus
providing a more realistic variability in the imputed data. Tis
algorithm can be extended for ordinal or categorical variables
with more than two categories.
1.5. Te Monotone Missing Pattern Requirement. For the
imputation of categorical data, if the missing data pattern is
non-monotone, this can cause difculties in the imputation
in a variety of situations [12, 2426]. A data set is said to have
a monotone missing pattern when it is possible to arrange the
variables in order such that if an individual is missing variable
then that individual is also missing all subsequent variables
, > . Te data set below has a monotone missing pattern:
Obs
1
2
3
4
1
2.1
3.0
1.9
3.2
2 3 4
0 10.2 0.5
1 11.0
1
(5)
Because survey data are ofen categorical with a large number

of variables, fnding such an ordering of the variables may
be prohibitively time consuming or impossible. However, if
we can achieve a monotone missing data pattern, we can
use the automated model-based capabilities of many sofware
packages to impute our missing data and avoid many of the
pitfalls of other types of imputation.
1.5.1. Creating a Monotone Missing Pattern. To construct a
monotone missing data pattern, the frst step is to use the
model-based approach described above for multivariate normal data in the data set to simultaneously impute the missing
data for the continuous variables. Once this step is completed,
we can impute the incomplete categorical variables, one at
a time, using the model-based approach for categorical data
described above, which requires a monotone missing pattern
for implementation. Once a variable is complete it can be
used in the imputation of the next variable. Hence, at each
step in the process only one variable is incomplete, creating
by default a monotone missing data pattern. We recommend
starting with the variable with the fewest missing values and
ending with the variable with the most missing values. Tis
procedure is repeated until all variables are complete.
1.6. A Word about Multiple Imputations. Some investigators argue that multiple imputation is necessary to obtain
unbiased estimates of the standard errors and hence for
conducting inference [1, 4, 5, 14, 27]. However, most multiple
imputation procedures work in tandem with the procedure
for the substantive analysis. For example, SASs Proc MI
works in tandem with Proc MIanalyze, which performs the
substantive analysis afer each of the multiple imputations.
Because we must frst build a monotone missing pattern,
94
we must frst impute each missing variable for each case

before building the substantive logistic regression model and
we cannot exploit the multiple aspects of Proc MI and
other similar sofware implementations. Furthermore, we
must impute the normal variables using diferent methods
than the categorical variables and hence these need to be
imputed in a separate step. Furthermore, in our case study as
well as in many studies involving ordinal data, we construct
indices from item responses to measure constructs such as
socioeconomic status, family cohesion, and language profciency. We need to have complete data to build these indices
and building constructs is not an imbedded part of any sofware implementation. Multiple imputation procedures work
in conjunction with the substantive analysis by repeating
the imputation several (up to 10 usually) times, each time
estimating the parameters of the substantive model and their
standard errors. Less biased estimates of the standard errors
can then be obtained based on changes in the parameter
estimates across diferent imputations. Inference about the
parameters is then conducted using this improved standard
error.
Tis cannot be implemented within the canned procedure
if the data need to be imputed variable by variable. Hence
confdence intervals and values could retain some downward bias when performing single imputation even with the
model-based approach.
We can, of course, perform our own procedure for the
imputation and the data analysis several times and calculate
the variance of the diferent parameter estimates of interest
across diferent analyses with diferent imputations. Here we
show an example of this procedure which involved imputing
the data, calculating indices based on the complete data,
ftting the substantive logistic regression model, repeating the
entire process several times, and calculating the total variance
as a weighted sum of the within- and between-imputation
variance estimates. Te resulting standard error could then
be used to conduct inference. While it may sound tedious,
in practice once the code is written it is quite simple and
straightforward.
Let be the number of imputations performed, producing diferent point estimates for the parameters and
their standard errors. Te combined point estimate for the
parameter, , is given by the mean over all imputations:
=
1
,
=1
(6)
1
.
=1
(7)
where is the estimate from the th imputation. Let

be the variance estimate from the th imputation; then the
within-imputation variance is given by the mean over all
imputations:
2
1
( ) .
1 =1
= + (1 +
1
) .
(9)
Te statistic has a distribution given by
1/2 ( ) (V ) ,
(10)
where is the value of the parameter under the null

hypothesis and the degrees of freedom are given by
V = ( 1) [1 +
] .
(1 + 1/)
(11)
Inference can then be conducted via the construction of

confdence intervals or hypothesis testing [4, 28].
1.7. Assessing Imputation. Assessing how well an imputation
worked is somewhat problematic. If the MCAR assumption
holds, then we would expect only small changes in the
means of the diferent variables and no change in the basic
shape of the distribution for quantitative variables. Hence,
small changes in the histograms of variables between the
incomplete and complete data (before and afer imputation)
indicate MCAR. For quantitative MAR data, we would expect
small changes in the shape of the conditional distribution of
given each level of or combination of s. If there are a large
number of variables it may be impossible to check. However,
it can be instructive to plot histograms of the incomplete
data for the missing variable by diferent categories of a few
categorical variables and compare these histograms to the
same for the complete data. If there are no drastic changes,
then this is evidence for the data being at least MAR. For
categorical data, assessing the imputation is even harder. In
practice, usually we can only examine summary statistics
and look for any problematic data or data patterns. Tis
is a difcult theoretical problem and until it is resolved by
theorists, the investigator must rely on common sense and
reasonable care with checking of assumptions. If, in expert
opinion and experience, respondents are likely to refuse
to answer certain types of questions based on the answer
to the question itself, and no number of other participant
characteristics can be used to model this probability of
refusal, then methods to correct this bias using information
from outside the sample are indicated.
In general, though, in the assessment of imputation using
model-based approaches, if the algorithm converges and
produces no anomalous values, then we have no reason to
question the results, as MCMC methods have been shown by
strong theory and simulation to produce samples from the
target distribution.
2. Methods
Te between-imputation variance is given by

=
Te estimate for the total variance for is given by
(8)
We tested the method on a case study using the National

Latino and Asian American Study (NLAAS). Te NLAAS
core sampling procedure resulted in a nationally representative sample of 4649 Latino and Asian Americans and
95
immigrants who resided in the contiguous United States.
Regarding the Latino sample, there were 577 Cubans, 868
Mexicans, and 614 other Latinos. Te subcategory other
Latinos included immigrants from Colombia, the Dominican Republic, Ecuador, El Salvador, Guatemala, Honduras,
Nicaragua, and Peru. Te Asian sample consisted of 600
Chinese, 508 Filipino, and 520 Vietnamese participants and
467 other Asians. Te subcategory other Asians consisted
of Koreans, Japanese, Asian Indians, and individuals of other
Asian backgrounds.
Te NLAAS data set had one of the most comprehensive
and advanced designs ever developed. A detailed description
of the NLAAS methods of data collection has been previously
documented [2932].
Te sampling techniques consisted of three major
approaches. First, core and secondary sampling units were
selected according to probability proportionate to size, from
which household members in the continental United States
were sampled. While primary sampling units were defned as
metropolitan statistical areas, secondary sampling units were
formed from contiguous groupings of census blocks. Second,
high-density supplemental sampling was applied, using a
greater than 5% density criterion, in which Asian and Latino
groups were oversampled. Asian and Latino individuals who
did not belong to the target groups under which these
geographical areas were classifed were still eligible to take
part in NLAAS. For example, Vietnamese individuals living
in a Chinese high-density census block were eligible. Tird,
secondary respondents were recruited from households in
which one eligible participant had already been recruited and
interviewed. Secondary respondents sampling was used to
further increase the number of study participants. In all three
sampling procedures explained above weighting corrections
were applied to take into account joint probabilities of
selection.
Te NLAAS instruments were available in Cantonese,
Mandarin, Tagalog, Vietnamese, Spanish, and English. Tey
were translated using standard translation as well as backtranslation techniques. All participants received an introductory letter and the study brochure in their preferred
language. Tose who gave their consent to take part in the
study were screened and interviewed by professionals who
had linguistic and cultural backgrounds similar to those
of the sample population. Interviews were conducted with
computer-assisted interviewing sofware in the preferred
language of the participants. Face-to-face interviews with
the participants were administered in the core and highdensity samples. Exceptions were made when respondents
specifcally requested a telephone interview or when faceto-face interviewing was prohibitive. Te average length of
each interview was 2.4 hours. As a measure of quality control,
a randomly selected sample of participants with completed
interviews was contacted to validate the data.
Written consent was obtained for all study participants,
protocols, and procedures. Human subject approval was
given by the Cambridge Health Alliance, Harvard University,
the University of Michigan, and the University of Washington.

2.1. Imputing Missing Values for the NLAAS Data Set. All but
a few variables in the NLAAS data set had missing observations. We selected a total of 75 out of approximately 3000
items available in the NLAAS dataset for the imputation, with
68 items of interest in the substantive model. Te 75 items
include both single variables such as sex, race, participants
education, spouses education, mothers education, fathers
education, child labor, economic resources, and multivariable
constructs such as social networks, family cohesion, language
preference, ethnic or native language profciency, English
language profciency, and socioeconomic success. Out of the
75 items used for imputation, 5 were either approximately
normal or could be normalized using transformations. Te
variable SE2 (spouses education) was transformed using the
square root transformation and EM2 (child labor/age at
employment) was transformed using exponentiation to 3/2.
Te remaining 70 variables were binary or ordinal with 2 to 5
categories. Additionally, four variables had no missing data.
Hence we had 9 variables with which to build the frst imputation model and 75 variables with which to build the fnal
imputation model. Te extent of the missing data can be visualized in the histogram and box-plot shown in Figures 1 and 2.
We used the SAS procedure Proc MI to perform the imputations using the MCMC model-based method described
above. First the continuous variables were normalized by a
suitable transformation, if necessary. Ten the multivariate
normal imputation, as described above, was performed on
these. Next, the categorical variable with the fewest missing
values was imputed using all completely observed variables
and the normal variables imputed in the frst stage of
imputation. We used the monotone discrim model for binary
and 3-category variables and monotone logistic for variables
with more than 3 categories. Tese methods implement
the methods model-based MCMC procedure described and
require the monotone missing pattern.
Once all missing values were imputed, we developed
indices to measure various abstract concepts such as English
language profciency, ethnic or native language profciency,
language preference, social networks, and family cohesion.
We developed a model to predict socioeconomic success in
Latino and Asian immigrants based on constructs such as
language preference and profciency, economic resources, social
networks, family cohesion, and child labor. Te constructs
were built from responses to items in the survey using the
complete data.
For illustration of the multiple imputation procedure,
we performed the procedure 10 times for data used in a
substantive model to predict socioeconomic success based
on several constructs. We estimated the total variance and
performed the -test for the null hypothesis that the true
parameter value is zero versus the alternative that is it not
equal to zero for each parameter in the model.
To assess the imputation, we checked for extreme or nonsensical values afer each imputation and graphed histograms
of the continuous variables. Examples of the histograms are
shown in Figures 3, 4, 5, and 6. All tables and fgures were
produced in SPSS.
96
250
60
Frequency
Frequency
200
40
150
100
20
50
0
0
100
200
300
Missing
400
500
600
Figure 1: Histogram of the number of missing observations in 75

items used from the NLAAS data.
0.00
10.00
15.00
20.00
25.00
5.00
Number of years mother attended school
30.00
Figure 4: Histogram afer imputation of missing values of the

number of years the participants mother attended school.
Frequency
Missing
400
300
Frequency
200
100
Variable
Figure 2: Bar chart of the extent of missingness showing each

variable.
Frequency
?1
?4
?7
?10
?13
?16
?19
?22
?25
?28
?31
?34
?37
?40
?43
?46
?49
?52
?55
?58
?61
?64
?67
?72
?75
Poor
Good
Fair
How well do you write in English?
500
200
150
100
50
0
200
150
100
50
0
200
150
100
50
0
200
150
100
50
0
Excellent
Frequency
600
0.00
1.00
2.00
3.00
4.00
5.00
6.00
Square root of fathers education
Figure 5: Histograms of raw data of the square root of fathers

education by writing profciency in English.
400
Frequency
300
3. Results and Discussion

200
100
0
0
10
20
30
Number of years mother attended school
Figure 3: Histograms of raw data for the number of years the

participants mother attended school.
3.1. Imputation. For this study, the imputation appeared to

work well. Tere were no problems with convergence and no
implausible values were observed. Figure 3 shows the small
changes in the distribution of the variable Mothers Ed, the
number of years of education for the respondents mother.
Tese results are typical of the quantitative variables imputed.
Figures 5 and 6 show the histograms for a continuous variable
which is the square root of the number of years of education
for the participants father. Te square root was necessary
to achieve approximate normality. Te small changes in the
conditional distribution lend evidence that the missing data
process for this variable is MAR.
97
Table 1: Results of logistic regression substantive analyses on 10 imputed data sets. Te far right column reports the percentage of the imputed
data sets which resulted in rejecting the null hypothesis that the parameter is equal to zero versus the two-sided alternative. Te other columns
represent the parameter estimates and values for the -test described in Section 1.6. = 4649, 9 (0.05) = 2.26.
Index
Child labor
English language profciency
Parents education
Race
Sex
4.40 01
1.98 01
6.83 02
3.67 01
2.25 01
Good
Fair
How well do you write English?
Poor
120
100
80
60
40
20
0
120
100
80
60
40
20
0
120
100
80
60
40
20
0
120
100
80
60
40
20
0
Excellent
Frequency
Frequency
Frequency
Frequency
Dependent variable: socioeconomic success (0, 1).
1.17 02
1.65 04
8.86 05
1.26 03
1.11 03
0.000
1.000 2.000 3.000 4.000 5.000

Square root of fathers education
6.000
Figure 6: Histograms of complete data of the square root of fathers

education by writing profciency in English.
3.2. Multiple Imputation. Te results for the multiple imputations are shown in Table 1. Findings indicate that, for all
variables in the model, all of the single tests were consistent
with the combined test.
Our approach represents a simple and very efective
method for imputation of survey data, which are ofen
ordinal or nominal. Our method combines the capability of
modelling the missing data distribution of the automated
model-based procedures, such as Bayesian MCMC methods,
commonly available in many sofware packages, while circumventing the current limitations in many of these packages
for the imputation of a large number of categorical variables.
Te authors declare there is no confict of interests regarding
the publication of this paper.
Acknowledgment
Te project described was supported in part by the National
Center for Advancing Translational Sciences (NCATS),
3.15 03
4.00 04
2.59 04
3.72 04
3.13 03
1.51 02
6.05 04
3.73 04
1.67 03
4.55 03
3.57
8.07
3.54
8.98
3.33
% Reject
100
100
100
100
100
National Institutes of Health (NIH), through Grant no. UL1

TR000002.
References
[1] P. D. Allison, Multiple imputation for missing data: a cautionary tale, Sociological Methods & Research, vol. 28, no. 3, pp. 301
309, 2000.
[2] P. D. Allison, Missing Data, Sage, Tousand Oaks, Calif, USA,
2001.
[3] P. A. Gimotty and M. B. Brown, Imputation procedures
for categorical data: their efects on the goodness-of-ft chisquare statistic, Communications in Statistics: Simulation and
Computation, vol. 19, no. 2, pp. 681703, 1990.
[4] D. B. Rubin, Multiple Imputation for Nonresponse in Surveys,
John Wiley & Sons, New York, NY, USA, 1987.
[5] D. B. Rubin, Multiple imputation afer 18+ years, Journal of the
American Statistical Association, vol. 91, no. 434, pp. 473489,
1996.
[6] U. Kohler, Surveys from inside: an assessment of unit nonresponse bias with internal criteria, Survey Research Methods, vol.
1, no. 2, pp. 5567, 2007.
[7] T. W. Smith, Survey non-response procedures in crossnational perspective: the 2005 ISSP non-response survey,
Survey Research Methods, vol. 1, no. 1, pp. 4554, 2007.
[8] A. W. Hoogendoorn and J. Daalmans, Nonresponse in the
recruitment of an internet panel based on probability sampling,
Survey Research Methods, vol. 3, no. 2, pp. 5972, 2009.
[9] G. Chen and T. Astebro,

How to deal with missing categorical
data: test of a simple Bayesian method, Organizational Research
Methods, vol. 6, no. 3, pp. 309327, 2003.
[10] G. B. Durrant, Imputation methods for handling itemnonresponse in practice: methodological issues and recent debates,
International Journal of Social Research Methodology, vol. 12, no.
4, pp. 293304, 2009.
[11] J. L. Schafer, Analysis of Incomplete Multivariate Data, Chapman
& Hall, New York, NY, USA, 1997.
[12] J. K. Vermunt, J. R. van Ginkel, L. A. van der Ark, and K. Sijtsma,
Multiple imputation of incomplete categorical data using latent
class analysis, Sociological Methodology, vol. 38, no. 1, pp. 369
397, 2008.
[13] N. J. Horton and K. P. Kleinman, Much ado about nothing:
a comparison of missing data methods and sofware to ft
incomplete data regression models, Te American Statistician,
vol. 61, no. 1, pp. 7990, 2007.
[14] R. M. Groves, Non-response rates and nonresponse bias in
household surveys, Public Opinion Quarterly, vol. 70, no. 5, pp.
646675, 2006.

[15] J. L. Schafer, Multiple imputation: a primer, Statistical Methods
in Medical Research, vol. 8, no. 1, pp. 315, 1999.
[16] S. van Buuren, H. C. Boshuizen, and D. L. Knook, Multiple
imputation of missing blood pressure covariates in survival
analysis, Statistics in Medicine, vol. 18, pp. 681694, 1999.
[17] P. D. Allison, Handling missing data by maximum likelihood, SAS Global Forum: Statistics Data Analysis 312-2012,
2012, http://www.statisticalhorizons.com/wp-content/uploads/
MissingDataByML.pdf.
[18] Statistical Computing Seminars: Multiple Imputation in
Stata, Institute for Digital Research and Education, UCLA,
http://www.ats.ucla.edu/stat/stata/seminars/missing data/mi
in stata pt1.htm.
[19] A. Gelman, J. B. Carlin, H. S. Stern, and D. B. Rubin, Bayesian
Data Analysis, Chapman & Hall, New York, NY, USA, 1995.
[20] K. H. Li, Imputation using Markov chains, Journal of Statistical
Computation and Simulation, vol. 30, no. 1, pp. 5779, 1988.
[21] R. J. A. Little and D. B. Rubin, Statistical Analysis with Missing
Data, John Wiley & Sons, New York, NY, USA, 2002.
[22] C. Liu, Bartletts decomposition of the posterior distribution
of the covariance for normal monotone ignorable missing data,
Journal of Multivariate Analysis, vol. 46, no. 2, pp. 198206, 1993.
[23] M. A. Tanner and W. H. Wong, Te calculation of posterior
distributions by data augmentation, Journal of the American
Statistical Association, vol. 82, pp. 528540, 1987.
[24] P. Minini and M. Chavance, Sensitivity analysis of longitudinal
binary data with non-monotone missing values, Biostatistics,
vol. 5, no. 4, pp. 531544, 2004.
[25] P. J. Lavrakas, Encyclopedia of Survey Research Methods, Sage,
Tousand Oaks, Calif, USA, 2008.
[26] D. Zhang, A Monte Carlo Investigation of Robustness to Nonnormal Incomplete Data of Multilevel Modeling, Texas A&M
University Press, College Station, Tex, USA, 2005.
[27] K. Lueck, Evaluationsmethoden der Bildungsforschung, LISUM,
Ludwigsfelde, Germany, 2006.
[28] J. Barnard and D. B. Rubin, Small-sample degrees of freedom
with multiple imputation, Biometrika, vol. 86, no. 4, pp. 948
955, 1999.
[29] M. Alegra, D. Takeuchi, G. Canino et al., Considering context,
place and culture: the National Latino and Asian American
Study, International Journal of Methods in Psychiatric Research,
vol. 13, no. 4, pp. 208220, 2004.
[30] S. G. Heeringa, J. Wagner, M. Torres, N. Duan, T. Adams,
and P. Berglund, Sample designs and sampling methods of
the Collaborative Psychiatric Epidemology Studies (CPES),
International Journal of Methods in Psychiatric Research, vol. 13,
no. 4, pp. 221240, 2004.
[31] K. Lueck and M. Wilson, Acculturative stress in Asian immigrants: the impact of social and linguistic factors, International
Journal of Intercultural Relations, vol. 34, no. 1, pp. 4757, 2010.
[32] K. Lueck and M. Wilson, Acculturative stress in Latino Immigrants: the impact of social, socio-psychological and migrationrelated factors, International Journal of Intercultural Relations,
vol. 35, no. 2, pp. 186195, 2011.
98
Author Guidelines
Submission
Manuscripts should be submitted by one of the authors of the manuscript through the online
Manuscript Tracking System. Regardless of the source of the word-processing tool, only electronic
PDF (.pdf) or Word (.doc, .docx, .rtf) files can be submitted through the MTS. There is no page
limit. Only online submissions are accepted to facilitate rapid publication and minimize
administrative costs. Submissions by anyone other than one of the authors will not be accepted. The
submitting author takes responsibility for the paper during submission and peer review. If for some
technical reason submission through the MTS is not possible, the author can contact
jam@hindawi.com for support.
Terms of Submission
Papers must be submitted on the understanding that they have not been published elsewhere and are
not currently under consideration by another journal published by Hindawi or any other publisher.
The submitting author is responsible for ensuring that the articles publication has been approved by
all the other coauthors. It is also the authors responsibility to ensure that the articles emanating
from a particular institution are submitted with the approval of the necessary institution. Only an
acknowledgment from the editorial office officially establishes the date of receipt. Further
correspondence and proofs will be sent to the author(s) before publication unless otherwise
indicated. It is a condition of submission of a paper that the authors permit editing of the paper for
readability. All inquiries concerning the publication of accepted papers should be addressed to
jam@hindawi.com
Peer Review
All manuscripts are subject to peer review and are expected to meet standards of academic
excellence. If approved by the editor, submissions will be considered by peer-reviewers, whose
identities will remain anonymous to the authors.
Concurrent Submissions
In order to ensure sufficient diversity within the authorship of the journal, authors will be limited to
having two manuscripts under review at any point in time. If an author already has two manuscripts
under review in the journal, he or she will need to wait until the review process of at least one of
these manuscripts is complete before submitting another manuscript for consideration. This policy
does not apply to Editorials or other non-peer reviewed manuscript types.
Article Processing Charges
Journal of Electrical and Computer Engineering is an open access journal. Open access charges
allow publishers to make the published material available for free to all interested online visitors.
For more details about the article processing charges of Journal of Electrical and Computer
Engineering, please visit the Article Processing Charges information page.
Units of Measurement
Units of measurement should be presented simply and concisely using System International (SI)
units.
Title and Authorship Information

The following information should be included
Paper title
Full author names
Full institutional mailing addresses
Email addresses
Abstract
The manuscript should contain an abstract. The abstract should be self-contained and citation-free
and should not exceed 200 words.
Introduction
This section should be succinct, with no subheadings.
Materials and Methods
This part should contain sufficient detail so that all procedures can be repeated. It can be divided
into subsections if several methods are described.
Results and Discussion
This section may each be divided by subheadings or may be combined.
Conclusions
This should clearly explain the main conclusions of the work highlighting its importance and
relevance.
Acknowledgments
All acknowledgments (if any) should be included at the very end of the paper before the references
and may include supporting grants, presentations, and so forth.
References
Authors are responsible for ensuring that the information in each reference is complete and
accurate. All references must be numbered consecutively and citations of references in text should
be identified using numbers in square brackets (e.g., as discussed by Smith [9]; as discussed
elsewhere [9, 10]). All references should be cited within the text; otherwise, these references will
be automatically removed.
Preparation of Figures
Upon submission of an article, authors are supposed to include all figures and tables in the PDF file
of the manuscript. Figures and tables should not be submitted in separate files. If the article is
accepted, authors will be asked to provide the source files of the figures. Each figure should be
supplied in a separate electronic file. All figures should be cited in the paper in a consecutive order.
Figures should be supplied in either vector art formats (Illustrator, EPS, WMF, FreeHand,
CorelDraw, PowerPoint, Excel, etc.) or bitmap formats (Photoshop, TIFF, GIF, JPEG, etc.). Bitmap
images should be of 300 dpi resolution at least unless the resolution is intentionally set to a lower
level for scientific reasons. If a bitmap image has labels, the image and labels should be embedded
in separate layers.
Preparation of Tables
Tables should be cited consecutively in the text. Every table must have a descriptive title and if
numerical measurements are given, the units should be included in the column heading. Vertical
rules should not be used.
Proofs
Corrected proofs must be returned to the publisher within 2-3 days of receipt. The publisher will do
everything possible to ensure prompt publication. It will therefore be appreciated if the manuscripts
and figures conform from the outset to the style of the journal.
Copyright
Open Access authors retain the copyrights of their papers, and all open access articles are
distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution and reproduction in any medium, provided that the original work is
properly cited.
The use of general descriptive names, trade names, trademarks, and so forth in this publication,
even if not specifically identified, does not imply that these names are not protected by the relevant
laws and regulations.
While the advice and information in this journal are believed to be true and accurate on the date of
its going to press, neither the authors, the editors, nor the publisher can accept any legal
responsibility for any errors or omissions that may be made. The publisher makes no warranty,
express or implied, with respect to the material contained herein.
Disclosure Policy
A competing interest exists when professional judgment concerning the validity of research is
influenced by a secondary interest, such as financial gain. We require that our authors reveal any
possible conflict of interest in their submitted manuscripts.
If there is no conflict of interest, authors should state that The author(s) declare(s) that there is no
conflict of interest regarding the publication of this paper.
Contact Information
Email Addresses
Editorial Office:jam@hindawi.com
Postal Addresses
410 Park Avenue
New York, NY 10022
USA
Cairo Free Zone
Road 5, Block E
Nasr City, Cairo 11816
Egypt
Fax Numbers
New York, USA
Philadelphia, USA
Amsterdam, The Netherlands
Brussels, Belgium
Paris, France
Milan, Italy
Zurich, Switzerland
London, England
Germany
Sydney, Australia
Tokyo, Japan
+1-212-409-8818
+1-215-893-4392
+31-20-5248282
+32-2-7062424
+33-1-53013737
+39-02-95441264
+41-44-2742823
+44-2076917066
+49-3221-10899-1460
+61-29-4751375
+81-3-44965310

Journal of Applied Mathematics-1

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Journal of Applied Mathematics-1

Transféré par

Droits d'auteur :

Formats disponibles

Journal of

About this Journal

Hindawi Publishing Corporation

410 Park Avenue

Abstracting and Indexing

Statistical Theory and Method Abstracts (STMA-Z)

Hindawi Publishing Corporation

410 Park Avenue

Ke Chen, University of Liverpool, United Kingdom

Huijun Gao, Harbin Institute of Technology, China

Zlatko Jovanoski, University of New South Wales, Australia

Zhengrong Liu, South China University of Technology, China

Srinivasan Natesan, Indian Institute of Technology, India

Jian Hua Shen, Chinese Academy of Sciences, China

Guangchen Wang, Shandong University, China

Meng Zhan, Chinese Academy of Sciences, China

A Comparison of Evolutionary Computation Techniques for IIR

Analysis of Repairable Geo/G/1 Queues with Negative Customers

Radial Basis Point Interpolation Method with Reordering Gauss Domains

Initial Time Difference Stability of Causal Differential Systems in terms

Bifurcation of Safe Basins and Chaos in Nonlinear Vibroimpact Oscillator

A General Setting and Solution of Bellman Equation in Monetary Theory

1:3 Resonance and Chaos in a Discrete Hindmarsh-Rose Model

Exponential Convergence for Numerical Solution of Integral Equations Using

Working with Missing Data: Imputation of Nonresponse Items in Categorical

Hindawi Publishing Corporation

cooperations to collect data from persons. It provides high

Journal of Applied Mathematics

Local and global

Figure 1: Block diagram of the proposed palmprint based recognition system.

distance. Te block diagram of the proposed system is shown

Journal of Applied Mathematics

3. Overview of Discrete Orthonormal

With the dyadic sampling scheme in order 0, 1, 2; . . . log2

(2) Partition the Fourier sample (, V) and multiply it

where V and V are the horizontal and vertical voice

4. Preprocessing and ROI Extraction

Journal of Applied Mathematics

Journal of Applied Mathematics

5.1. Local Feature. Te feature extraction technique in any

ponents and (1 , 2 ) and (1 , 2 ) are phase components.

Figure 5: POC function for PolyU palmprint database.

Te cross spectrum (1 , 2 ) between (1 , 2 ) and

where (1 , 2 ) denotes the complex conjugate of (1 , 2 )

Te POC function (1 , 2 ) = is 2D inverse discrete Fourier

5.2.2. Band-Limited Phase-Only Correlation (BLPOC). In

Journal of Applied Mathematics

Figure 6: POC function for COEP palmprint database.

Figure 8: BLPOC function for PolyU palmprint database.

Figure 7: POC function for IIT Delhi palmprint database.

the range of spectrum of the given palmprint image. BLPOC

where = 0 , . . . , 0 , and = 0 , . . . , 0 , where =

Figure 10: BLPOC function for IIT Delhi palmprint database.

however, for an imposter matching (a matching performed

Genuine acceptance rate (%)

Journal of Applied Mathematics

Genuine acceptance rate (%)

6. Local-Global Information Fusion (LGIF) for

function. Ten the mutual regions can be cropped based on

LGIF based palmprint recognition algorithm is described in

Journal of Applied Mathematics

Genuine acceptance rate (%)

Genuine acceptance rate (%)

distance is obtained by matching the two palmprint images