Jim Wilkinson and NAG

2019 | Celebrating 100 years since his birth

James Wilkinson
James Wilkinson. CS/22997/8. 
Crown copyright.
Central Photographic Section, NPL.
Image courtesy of Sven Hammarling,
and Nicholas J. Higham, 
School of Mathematics, 
The University of Manchester.

James Hardy Wilkinson, known to his friends as ‘Jim’, was arguably the world’s foremost authority on matrix computations for three decades starting in the 1950s. He spent most of his career at the National Physical laboratory where he worked with, amongst others, Alan Turing. He was a pioneer of floating-point arithmetic and developed backwards error analysis, a fundamental technique for determining how stable a numerical algorithm is. His FRS, gained for his work in numerical analysis, helped make the subject respectable, and pioneered the way for many others. In 1971 he edited, with Christian Reinsch, the Handbook for Automatic Computation, which collected together a number of state-of-the-art numerical algorithms along with source code in Algol 60. This was great for NAG’s Algol 60 Library but a nuisance for the Fortran. NAG’s David Sayers recalls that “Oxford University performed the translation but a little too literally, so array accesses were by row instead of by column; until we could put this right, I had to argue that this kept the matrix locked in memory and was therefore advantageous for the majority of (small) problems”.

NAG was founded in 1970 as an inter-university project to develop the world’s first portable library of numerical algorithms. Jim Wilkinson was an enthusiastic supporter of the project and, in the early days of the NAG Library, much of the linear algebra software was either contributed by him or based on his work. His books, Rounding Errors in Algebraic Processes and The Algebraic Eigenvalue Problem, were NAG’s bibles in those early years and are still referred to today. Jim also introduced the NAG team to researchers throughout the world, in particular the team at Argonne National Laboratory (ANL) in the US, who at the time were developing a high-performance library for solving matrix eigenvalue problems, EISPACK.

Following Jim’s death in 1986, NAG, ANL and NPL decided to establish The Wilkinson Prize for Numerical Software, to be awarded every four years to an outstanding package or application, judged on:

  • the clarity of the software implementation and documentation;
  • the importance of the application(s) addressed by the software;
  • the portability, reliability, efficiency and usability of the software implementation;
  • the clarity and depth of analysis of the algorithms and the software in the submission;
  • the quality of the test software.

Unlike other prizes in the field of numerical analysis, this prize was focussed on the software, not the underlying algorithms. Under the stewardship of NAG, ANL and NPL there have been seven sets of winners:

  • 1991: Linda Petzold for DASSL
  • 1995: Chris Bischof and Alan Carle for ADIFOR
  • 1999: Matteo Frigo and Steven Johnson for FFTW
  • 2003: Jonathan Shewchuch for Triangle
  • 2007: Wolfgang Bangerth for deal.II.
  • 2011: Andreas Waechter and Carl D. Laird for Ipopt
  • 2015: P.E. Farrell, S.W. Funke, D.A. Ham and M.E. Rognes for dolfin-adjoint

To ensure the long-term future of the prize beyond 2015, NAG, ANL and NPL created an endowment to allow SIAM to adopt the prize, and the 2019 winners are Jeff Bezanson, Stefan Karpinski, and Viral B. Shah for the Julia Programming Language.

Jim Wilkinson inspired much of the modern field of numerical analysis and his legacy lives on in the books and papers that he authored, and in the NAG Library and many other high-quality pieces of numerical software. NAG is privileged to have had his support and guidance in our formative years.