I am experiencing NWChem crashes in the TDDFT module that occur when the MAXVECS variable is set too low, which is calculation dependent. I haven't attempted any source debugging just yet, but I conjecture that this occurs when the number of trial vectors reaches MAXVECS, which would necessitate some kind of truncation that isn't being handled properly. This is occurring for multiple molecules with a variety of settings on several different machines.
The simplest demonstration of the bug that I've been able to come up with is:
geometry
Ne 0.0 0.0 0.0
end
basis
* library 6-31G
end
tddft
nroots 1
maxvecs 2
print timings
end
task tddft energy
When run on one processor, the crash is reported in the output file as:
0:0:nga_copy_patch:g_b indices out of range :: 0
for ALGORITHM =1 or =2 and
0:0:ndra_write_sect: d_a dim error:: 2
for ALGORITHM =3 or =4. In both cases, the error is reported between the TDDFT timing reports for "Diagon" and "Residual".
|