Re: Cholesky factorisation

[Craig Markwardt <craigmnet(at)cow.physics.wisc.edu>]

Pavel Romashkin <pavel@netsrv1.cmdl.noaa.gov> writes:
>
> I can't help hearing some sarcasm regarding IDL in David McClain's
> postings.  Why is that? After all, accepting computer's numerical
> precision is a matter of faith, too, especially in life-critical
> applications Raytheon is developing. I find using upper-level
> language like IDL very convenient, and availability of "lower-level"
> tools like pointers is nice. C is great, but so is IDL I think.  If
> we don't trust functions written by others, we might as well wright
> everything ourselves in C...
> 

I think McClain's point is that the implementation of built-in
functions in IDL is strictly black-box.  But in IDL's case, it's more
like black magic... :-)

Seriously though, many of the routines *are* implemented in IDL
itself, and thus inspectable.  That's good.

For the built-in routines, however, one has to take it on faith that
RSI did a robust implementation, and Numerical Recipes is not always
the right choice.  Floating point precision on a computer is
well-defined these days by the IEEE, and thus I have more faith in it.

Craig


-- 
--------------------------------------------------------------------------
Craig B. Markwardt, Ph.D.         EMAIL:    craigmnet@cow.physics.wisc.edu
Astrophysics, IDL, Finance, Derivatives | Remove "net" for better response
--------------------------------------------------------------------------