Todd, John. "The Problem of Error in Digital Computation." Offprint in mimeo of Todd's contribution to L.B. Rall's Error in Digital Computation, (1965). 11x8, 38pp, offset printed. Staple bound. Very good condition. $125
Obituary of Prof. Todd (1911-2007, professor at the California Institute of Technology) from Caltech (2007): 'Todd's legacy and impact on modern-day mathematics can be seen in his contributions to analysis, linear algebra, and computation. His work was a precursor to and helped shape the foundation for today's computer science field. An emeritus professor at the California Institute of Technology, Todd developed the first undergraduate courses at Caltech in numerical analysis and numerical algebra, which play a key role in scientific computing..."
Review from The Computer Journal, 1966: " 'The Problem of Error in Digital Computation', by John Todd.
Starting with the assertion that it does not appear feasible to perform rigorous error analyses of the traditional kind
for all algorithms, or to accompany all computations by automatic error analyses, Professor Todd argues that we need to change our standards of acceptability. He surveys interpolation, square roots, eigenvalues of symmetric matrices, Monte Carlo methods, good error estimates and controlled numerical experiments. He prefers "cheap" error estimates, i.e. those which can be had fo r a fraction of the work needed to produce the result itself. The controlled numerical experiments are to be carried out on new algorithms using specially prepared sets of test data. Only algorithms which perform well in the experiments or are otherwise interesting need be subjected to more complete error analysis."
Comments