The ABINIT Users Mailing List ( CLOSED )

[Matthieu Verstraete <mjv500@york.ac.uk>]

Hello Xavier - prompt as ever when your code is on the line. The problem 
is actually not with the routine, but it is being called on a negative
argument:

 rho,rhomtrd,rhom1 =  -2.796187555105125E-19 NaN  -3.5762980139663923E+18
 del =  1.808955175498057
 NaN NaN
 NaN NaN
...

So the initial rho should be filtered: I suspect that the actual routine 
which does **(1/3.) uses exponentials and logs (obvious) and that someone 
forgot to check something there. The problem with taking the cube root of 
a negative appears for pgf90 and gfortran as well (just checked). Python 
can do it!

Should this be done at all? If so I can commit my patch to 5.5.3, with the 
absolute value taken and the sign restituted at the end of the loop. The 
call to invcb comes from the poissonsolver part of the code.

Matthieu




================================================================
Dr. Matthieu Verstraete                 mailto:mjv500@york.ac.uk
Dept. of Physics, University of York,     tel: +44 1904 43 22 08
Heslington, YO10 5DD York, United Kingdom fax: +44 1904 43 22 14