vsnyder@jato.jpl.nasa.gov (Van Snyder) (03/21/91)
Archive-name: math/num-analysis/numerical-recipes/1991-03-20 Original-posting-by: vsnyder@jato.jpl.nasa.gov (Van Snyder) Original-subject: Re: Numerical Recipes code copyrighted Reposted-by: emv@msen.com (Edward Vielmetti, MSEN) In article <1991Mar20.173427.17693@ux1.cso.uiuc.edu> andreess@mrlaxs.mrl.uiuc.edu (Marc Andreessen) writes: > >The problem I have with this is that it eliminates the niche Numerical >Recipes could otherwise hold. I can't point computationally-naive users >to NR as I'd like to... Why do you want to screw the "computationally naive user?" He's exactly the kind of user who should be steered AWAY from NR, beacuse he doesn't have the tools to tell the difference between the good and the bad. The authors certainly didn't do it for him! They tell you it's all great. But in random excursions into NR (based on requests for consultation), we found the treatment of five out of five sections to be inadequate: Initial value problem for ordinary differential equations -- presented from viewpoint of 1970; Quadrature -- presented from viewpoint of 1860; Nonlinear least squares -- presented from viewpoint of 1963, and carelessly done (overflows and underflows of damping parameter could silently cause bogus answers); "robust estimation" -- exceedingly badly done, with no apparent understanding of the ramifications of the fact that the approximating function is not differentiable where it inter- polates the data; Bessel functions -- presented from viewpoint of 1959. We think this sample is statistically significant. We don't have the time to dig in the same detail into every section. Recommend that computationally naive users refer to good numerical analysis texts, e.g. Kendall Atkinson's, or Kahaner, Moler and Nash (which comes with code). Then, get REFEREED code from netlib@research.att.com by sending email consisting of "send index". I know that if a computationally naive user educated in physics walks across the quad of a typical university to the applied math department, and asks a simple, elementary question, he'll be taken into Sobolev space and left there to rot. Try the library instead. Look for ACM (Association for Computing Machinery) Transactions on Mathematical Software. -- vsnyder@jato.Jpl.Nasa.Gov ames!elroy!jato!vsnyder vsnyder@jato.uucp