Memory allocation of lammps

Dear LAMMPS users,

I am running LAMMPS 17Nov16 on a linux machine and simulate the thermal equilibration of a polymer, using KOKKOS version of reax/c.

When the simulation starts, it says 80.1 Mb memory usage/processor, after roughly 24000 steps it quits, saying: Error on Proc0: Unable to allocate 2100Mb/processor…

I don’t know where that comes from, because:

-The number of atoms in the simulation stays the same, in fact it is slightly decreasing because of thermal losses

-The size of the simulation box is fixed, no shrink wrapping or things like this

-The only fixes applied to the enselmble are qeq/reax/kk (because I have to) and nvt/kk

-The temperature should start at 20K and should increase quite slowly, 2.8K/1ps to 300K (in fact, when I print out the temperature every ps, it fluctuates around 290-310 K, could that be the reason?, where does this high temperature comes from if I just apply fix nvt 20 300 50 ?)

Please find my input file in the attached,

best regards


in.pyrolysis.txt (3.17 KB)

3524-68-3-struktur.txt (3.05 KB)

161182-73-6-struktur.txt (5.59 KB)

What kind of device are you running this on? It worked for me on 1 MPI with 1 GPU…. 44500 steps and running.


Okay, the problem is with “run N every M”. Are you really sure what the “every” keyword does and this is what you intend to do?

From your last bullet point you seem to want to run “one long” simulation that is 200000 steps…. Instead with the every keyword you are running 200 simulations each with 1000 steps. Each of the 200 simulations is a NVT integration from 20K to 300K within 1000 steps. That is why your heating rate (and all thermodynamic variables) are messed up. If you want to print the temperature every 1000 steps, just use the “temp” keyword within the “thermo_style custom” command (actually the default thermo_style already has the temp keyword)…


Okay, the problem is with “run N every M”. Are you really sure what the
“every” keyword does and this is what you intend to do?

​the use of "run" with ​"every" may also trigger memory leaks. i recall,
that USER-REAXC had issues in that regard, also with loops in general. if i
remember correctly, those all have been addressed except for the case,
where "clear" was called in each loop iteration, which still resulted in -
extremely difficult to find the reason for - memory leaks. it is quite
possible, that the increased complexity of using KOKKOS introduced new
issues or re-introduced old issues in this regard.


Hi all,

thank you for the fast response!, I now removed the “every” keyword because my intention actually was to run one big simulation. It now worked well for the desired 200000 steps, thank you!

Best regards,