MINRES-QLP Pack and Reliable Reproducible Research via Staunch Scientific Software

2013-09-07T02:51:03Z (GMT) by Sou-Cheng (Terrya) Choi
<p>MINRES-QLP Pack is a suite of (extended) Krylov subspace methods for solving large linear systems and linear least-squares problems possibly with singular or ill-conditioned matrices or linear operators with special symmetries. Our purpose is to create and develop robust open-source implementations of the associated algorithms in MATLAB, Fortran, and PETSc that are faithful to the theory, following the philosophy of reproducible research and practicing development principles of staunch scienti fic software (SSS) that promote reliable reproducible research (RRR).</p> <p><br>In this paper, we review the key features in the ongoing theoretical and software development of our algorithms in MINRES-QLP Pack. We highlight the most eff ective software engineering tools known to us, that are potentially useful to other scienti c research areas. We contend that the principles of RRR and SSS should be propagated to computational science students in advanced courses of scienti fic computing. To this end, we have started an experimental seminar course, "Reliable Mathematical Software" (IIT MATH-573) in our institution. We will share our research practice and pedagogic experiences in the Workshop on Sustainable Software for Science.</p>