On Thu, Apr 08, 2021 at 07:17:42PM +0200, Kurt Pagani wrote:
> Maybe you are already aware of
> 
> COMPUTATION OF BESSEL AND AIRY FUNCTIONS AND OF RELATED GAUSSIAN QUADRATURE
> FORMULAE by WALTER GAUTSCHI
> https://www.cs.purdue.edu/homes/wxg/selected_works/section_02/169.pdf

Thanks for reference.

> The code is in Fortran though.
> https://www.cs.purdue.edu/archives/2001/wxg/codes/
> 
> IMO it would be easier, however, if we took it from GSL:
> https://www.gnu.org/software/gsl/doc/html/specfunc.html
> Bill  wrote an interface: 
> https://github.com/billpage/gsla/blob/master/gsl.lisp

Both ilutrate what I wrote: Gautschi has 0<p<1 and real x. GSL
has real arguments.  As I wrote, there are good looking
implementations for real p (and complex x).  A have a toy
multiple precision implementation, it is fast enough
if arguments are resonably small.  OTOH our Bessel J and
H01 also seem to havve resonable accuracy when arguments
are small.  The problem is large arguments.  Also, it
would be silly to use multiple precision implementation
to get double accuracy, but do not provide real multiple
precision implementation.  But with multiple precision
results problem of large arguments gets worse...

Just to be clear: in case of Airy we can try to hide
bug, that is provide low accuracy version.  That
would be easy, but the accuracy problem would come back
later.  The point is to sort out the mess so that
it works.  And ATM it seem that nobody else did this
in really satisfactory way, so appearently we should
be first to do this. 

-- 
                              Waldek Hebisch

-- 
You received this message because you are subscribed to the Google Groups 
"FriCAS - computer algebra system" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/fricas-devel/20210408180649.GA27175%40math.uni.wroc.pl.

Reply via email to