🚀 go-pugleaf

In article <C0pFE0.D8n@sci.kun.nl> alexp@sci.kun.nl (Alex Priem) writes:

>Better yet, try 'Numerical Recipes'. 

I left that out on purpose. It is not a bad book per se, but it implicitly
encourages people to use the included code samples as if they were polished
production code. These are _not_ the state of the art, and the otherwise
uninformed reader can get into some peculiar misconceptions. It is
generally better to either use a widely available public domain code
written by professionals (also known as 'your tax dollar at work') or
to write your own after having studied the literature a bit if you are
aware of some of the issues.

_Numerical Recipes_ has a place when you're after something quick and
dirty and you haven't been around the block before. However, many of the
subjects covered have been classic questions of scientific computing
for decades, and it really isn't necessary to settle for less than first
rate. This casts no aspersion on the value of NR as a teaching tool,
where the code samples provide an opportunity to teach simplified versions
of the often much less penetrable 'no-holds-barred' production codes.

In optimization in particular, the clooge (and _yes_, that's the correct
spelling) of using LINMIN may reduce the clutter and length of the book,
but it does not bear much resemblance to the _correct_ way to do it,
on the other hand, it's pedagogically much nicer since the student can
see quite clearly how the one-D minimization is parlayed into the
multidimensional case.

Finally, NR is not a comprehensive treatment of optimization, especially
when it comes to nonlinear constraints, which (if memory serves) were
mentioned in the original request.

So better yet, don't stop with just NR on this question.

Later,
Andrew Mullhaupt