[Shootout-list] Implicitode test
Simon Geard
simon@whiteowl.co.uk
Thu, 14 Apr 2005 07:32:56 +0100
Brent Fulgham wrote:
>--- Sebastien Loisel <sloisel@gmail.com> wrote:
>
>
>>Hello Simon,
>>
>>Thanks a lot for this implementation, I was dreading
>>writing it. I feel guilty for your hardship, I knew
>>this test was difficult to implement in Fortran and
>>many other languages. You have also implemented it
>>better than I would have. Your criticisms are
>>accurate, this is not how I use AD in my real code.
>>
>>
>
>Is this a 'same way' test? Why can't a Fortran
>programmer solve the problem using the normal Fortran
>idioms and language features?
>
>I think we may just have a lack of communication on
>this test that has caused Simon to waste a lot of his
>time.
>
>
Please don't think I'm complaining about the time I spent. I tried the
implementation because I like f90 and I like Numerical Analysis and
thought that this test should be a good vehicle for f90.
The point I'm trying to make is that this program is more than 'Hello
World' and most implementers are going to spend some time debugging
their code. The use of AD makes this much harder (because of the
indirection) whilst not providing (at least in the Newton case) any
numerical benefit. I'd suggest on the basis of my experience that
implementers use a finite difference approximation to the first
derivative in the first instance and when the numbers are about right
add in the AD code.
Simon