Interval Subroutine Library Mission

We propose the collection, standardization, and distribution of a full-featured, production quality library for reliable scientific computing with routines using interval techniques for use by the wide community of applications developers. 1 Vision – Why are we doing this? The interval/reliable computing research community has long worked to attract practicing scientists and engineers to use its results. We use any of the terms interval, reliable, verified computation in the sense of producing rigorous bounds on true results, e.g., [1, 2]. The Interval Subroutine Library (ISL) is a project to place interval tools into the hands of people we believe will benefit from their use by gathering and refining existing tools from many interval authors. We acknowledge that intervals carry a steep learning curve, and that they sometimes have been over-promised. The winning strategy for widespread adoption of interval technologies is the development of “killer applications” that are so much better (in some sense) than current practice that practicing scientists and engineers have no choice but to adopt the new technology. The ISL team wants to see such killer applications appear, but producing them is not our mission. The routine use of interval techniques by practicing scientists and engineers is hampered by a lack of widely-used, comprehensive, quality interval software that is available on all major platforms (Linux, Mac, Unix, Windows). Once such software is available, use of interval techniques is likely to grow along at least three paths: small-scale applications by scientists/engineers in the course of their daily work; professionally built applications in a specific area, such as global optimization or curve graphing; and the almost invisible embedding of verified computing as a tool in commodity software such as spreadsheets or scientific data analysis and document preparation. ? This work was supported in part by EPSRC Grant D033373/1. Submitted to Reliable Implementation of Real Number Algorithms: Theory and Practice, in the LNCS series of Springer Verlag, P. Hertling, C. Hoffmann, W. Luther, and N. Revol, eds. Dagstuhl Seminar Proceedings 06021 Reliable Implementation of Real Number Algorithms: Theory and Practice http://drops.dagstuhl.de/opus/volltexte/2006/712 ISL can provide the infrastructure for such developments. ISL targets application developers, those who are developing the significant applications. Intervalbased tools tailored for specific end-practitioner applications are developed by applications developers with expertise in applications areas, but those developers are not finding interval tools they perceive as attractive for their applications. Currently, if we talk to a group of scientists or engineers about intervals and convince them of the value of interval techniques, when they ask, “Great! What software can I use?” there is a long, painful pause. We have many tools, packages, and research codes, but we have no CD that solves their problems with rigorous bounds. The ISL project itself does not author software; contributing authors do that. The goal of the ISL project is to gather and disseminate a library of high quality interval-based tools. The fundamental requirements is, “Thou shalt not lie.” Routines are expected to return an enclosure of the correct mathematical result or else provide a suitable indication of failure. The qualities of interest for the ISL project include • correctness, • comprehensiveness, • reliability, • performance, • robustness, • maintainability, and • usability, • portability. To achieve these qualities, the ISL project encourages its contributing authors to use software engineering sound principles, including documentation, good architecture, thorough testing, and coding standards. The documentation produced and the process of assembling the library also support the goal of achieving high quality. Documentation should be complete, consistent, correct, usable, verifiable, maintainable, and reusable. The development process should have the qualities of productivity, timeliness, and transparency. The authors are embarking upon a plan for the cooperative development of such a library. This paper lays out the broad scope of the project. 1.1 Short-term goals By the end of 2007, we expect to offer interval BLAS levels 1 and 2 and a collection of problem-solving packages, mostly chosen from existing software, including linear systems, optimization, and differential equations. The collection may include utilities for automatic differentiation, Taylor models, and constraint propagation. For our plan to achieve this, see § 3.1.