| Re: Job dies after last "Num. of grids overlapping" ( No.1 )|
- Date: 2005/09/06 13:06
- Name: T.Ozaki
I tried your input file, and found that the fault is due to quite large
memory requirement in this job. The scf.energycutoff of 255 Ryd you gave
is adjusted to 349.0, 350.2, and 347.2 Ryd for a-, b-, and c-axes, respectively,
so that the difference can be minimized as few as possible, as you can find
the adjusted cutoff energies in the standard output.
In my case where I used 6 nodes(2cpus/node) of Xeon with 2 GByte memory/node,
zsm.memory0 shows that an array, Orbs_Grid, requires 1547 MByte to be allocated,
and in fact my job also died due to the large memory requirement.
When I decreased the scf.energycutoff to 130 Ryd, my job seems to run normally
on my machine.
Also, it would be better to use a k-space mixing scheme with tuned mixing
parameters for this system to avoid charge sloshing, since this system is
large enough. (P.S. I was able to obtain the SCF convergence without changing
any parameters for SCF mixing. It seems that this system is easy to handle
with respect to SCF convergence)