The ABINIT Users Mailing List ( CLOSED )

[prendergast2@llnl.gov]

On large simulations (i.e. big acell)

Note that there is a bug (at least in v4.0.3) which appears when the size of 
the simulation cell is quite large (e.g. 50 a.u.). A numerical test in the 
generation of Gvectors (kpgsph.f) fails for large cell dimensions.

The tolerance used as zero for this test is tol10 (i.e. 1.d-10). However, one 
of the numbers that is compared with tol10 is proportional to the cell 
dimension to the -6th power. So, for 50 a.u. we would already end up below 
tol10 for this test. This causes the program to halt at this point 
unnecessarily. A quick fix is to replace tol10 in this subroutine with tol14 
for example and recompile. However, some other fix possibly related to 
machine precision might be more appropriate.

For small cell sizes this isn't going to be a problem.