🚀 go-pugleaf

In article <C0K1Hy.BMJ@watserv2.uwaterloo.ca>, kvetzal@watarts.uwaterloo.ca (Ken Vetzal) writes:
> Does anyone know if NAG has an e-mail address which accepts reports of 
> suspected bugs?  
Certainly:    infodesk@nag.co.uk
              infodesk@nag.com   for North American customers
> I find that calling one of their random number 
> generator routines works fine if I request an even number of random
> numbers but blows up (produces Infinity - NaN) if an odd number of
> random numbers is requested.  A sample program which does this is
> included below:
> 
> 
> 	implicit double precision(a-h,o-z) 
> 	parameter(maxobs=365)
> 	double precision e(maxobs)
> 	external g05cbf, g05fdf
> 	call g05cbf(170565)
> 	do 100 n=1,2
> 	call g05fdf(0.d0,1.d0,maxobs,e)
> 	do 101 i=1,maxobs
> 	write(8,*)i,e(i)
> 101	continue
> 100	continue
> 	stop
> 	end
> 
> This will blow up with maxobs set to 365 or 367 or any other odd
> number I've tried > 100, but seems to work fine at 364 or 366 or
> any other even number > 100.  Thanks in advance for any advice.
> 
> Ken Vetzal (kvetzal@watarts.uwaterloo.ca)
This error in G05FDF is known to us. More precisely, the error only occurs
if two or more calls are made to G05FDF without an intervening call of G05CBF
or G05CCF, and if at least one of the values of N in those calls is odd.
The corrected version of the routine will be in the next release of
the NAG Fortran Library and in a future error bulletin.

Our apologies for the inconvenience that this has caused.

Richard Brankin
-- 
** R.W. Brankin -- Numerical Algorithms Group Ltd. -- nagrwb@vax.ox.ac.uk **
**   suggestions for sig. welcome   --   will need   include<std.discl>   **