< back to main site

Publications

Testing the numerical correctness of software.

Cox, M G; Harris, P M; Johnson, E G; Kenward, P D; Parkin, G I (2004) Testing the numerical correctness of software. NPL Report. CMSC 34/04

[img] Text
cmsc34.pdf

Download (330kB)

Abstract

We describe the application of a general methodology for testing the numerical correctness of scientific software to functions for the calculations of sample (arithmetic) mean and sample standard deviation, straight-line (ordinary) regression and polynomial (ordinary) regression. The functions tested are taken from a number of proprietary software packages and libraries, including the NAG and IMSL (FORTRAN) libraries, Microsoft Excel, LabVIEW, S-PLUS, Matlab and various Java numerical libraries.
Each stage of the methodology, from documenting the specifications of the functions tested through the definition of performance parameters and performance measures to the presentation and interpretation of the results of the testing, is described. In this way, and by stating any assumptions made in the application of the methodology, the testing undertaken is made as objective as possible given the (“black-box”) nature of the testing. A web-based facility is used to generate the reference data sets and corresponding reference results used in the testing, and sufficient information is provided for readers to reproduce these if they wish. In this way, the testing reported here is made transparent, repeatable and traceable, and may be extended to other functions for these calculations.
This report constitutes one of the deliverables of Project 2.1 Numerical Software Testing within the UK Department of Industry’s National Measurement System Software Support for Metrology Programme 2001-2004.

Item Type: Report/Guide (NPL Report)
NPL Report No.: CMSC 34/04
Subjects: Mathematics and Scientific Computing
Mathematics and Scientific Computing > Numerical Computation
Last Modified: 02 Feb 2018 13:16
URI: http://eprintspublications.npl.co.uk/id/eprint/2889

Actions (login required)

View Item View Item