MBUILD -- Too many Neighbors; Increase MAXELST

[CallumFrEllis]

Hi Tinker-hp team,

I am running a simulation of ~80000 atoms and am getting MBUILD -- Too many Neighbors; Increase MAXELST errors. It seems that the default number is 2500, in MOD_sizes.f, should this be sufficent? I know large systems are run on Tinker HP so is there perhaps another reason I am getting this error?

The simulation was minimized and NVT and NPTed to convergence fine on Tinker9 on a GPU machine, so don't think it is a simulation issue?

The arc_box tests run fine.

Thanks,

Callum

[louislagardere]

Hi Callum,
The default MAXELST value should be fine for default electrostatic cutoffs, could you share your inputs so that I can try them on my side ?

Best,
Louis

[CallumFrEllis]

VQIVYK_H2O_box_prod_start.txt

prod.txt
prod_opt.txt

Hi Louis,

Thanks for getting back to me please find attached prod.txt (key file) prod_opt.txt (slurm file) and VQIVYK_H2O_box_prod_start.txt is the starting xyz file.

Best wishes,

Callum

[louislagardere]

Hi again,
so everything works with default maxelst value if you comment the "cutoff 12" line in the keyfile which corresponds to a default 7A value for ewald real space cutoff (which is fine with default parameters for the PME grid)

[CallumFrEllis]

Hi Louis, Thanks for looking into this, working now. Another quick question, do I have to include ARCHIVE in the key file to write to an .arc? In tinker gpu i think this is the default but it appears to be writing every 1000 steps, whereas I would prefer a .arc or .dcd.

Thanks again,
Callum

[louislagardere]

Yes you have to include it to produce an *arc file, the keyword to add in the keyfile to produce dcd outputs is "DCDIO"