The development of ARPACK began as a research code written in Matlab and then in Fortran77 in 1990. Initially, the code was developed to study and verify the properties of the Implicitly Restarted Arnoldi Method described in [44]. Preliminary experience with that code showed considerable promise in performance and also seemed to provide a solid foundation for the development of serious mathematical software for large structured eigenvalue problems.
During the academic year 1991-92, Dr. Phuong Vu (at that time with Cray Research) was granted permission to work in a half time appointment to the NSF Center for Research on Parallel Computation at Rice University on the development of ARPACK. At the outset, we attempted to design the software to be efficient and portable on conventional high performance computing architectures. Of course, our design was also intended to be easily modified to effectively utilize a variety of parallel architectures (resulting in P_ARPACK). Phuong's experience with users at Cray Research suggested that a reverse communication interface was essential. We are deeply indebted to Phuong for the design and implementation of this interface. He set the coding and documentation style and developed the initial implementation of all of the basic computational routines for real (single and double precision) matrices. His fundamental design has served us well as we have improved and expanded upon the package over the past few years.
We wish to thank the numerous users with applications and also our fellow numerical analysts who worked with initial ``alpha" and then ``beta" versions of the code. Their feedback and patience throughout this development has been invaluable. This interaction has often given us a wonderful sense of community and the words ``it solved my problem!" always managed to brighten up the drudgery of developing and maintaining the software. In particular, we would like to mention Jean-Philippe Brunet, Daniela Calvetti, Lawrence Cowsar, Olivier Daube, David Day, Stewart Edwards, Ralph T Goodwin III, Ed Hayes, Lennart Johnsson, Michiel Kooper, Kristi Maschhoff, Karl Meerbergen, Frank Milde, Seymour Parter, Phil Pendergast, George Phillips, John Red-Horse, Lothar Reichel, Tod Romo, Will Sawyer, Jennifer Scott, Rajesh Kumar Singh, Allison Smith, Allister Spence, Zdenko Tomasic, Henk Van der Vorst.
ARPACK is freely available through the world wide web and by anonymous ftp (See Chap. 1). It relies heavily upon the LAPACK software [1] and upon the BLAS [21,11,10]. Portability with performance, accuracy, and robustness is a direct consequence. We are greatly indebted to the authors of that software and more generally to the larger numerical analysis community that has contributed in many ways to its development.
Finally, we would like to thank the National Science Foundation, DARPA, and the Department of Energy for their generous support of this project (See Chap. 1 for full citation).