cp2k_shell stucked at sending positions

[Geng Sun]

Hello, everyone,

I am encountering a strange problem using python ase cp2k calculators.

 I am using CP2K calculator to optimize strucutres, but many of my calculations got stucked at some point. I swithed on the debug=True option within the CP2K calculator class. The output shows that all the calculations stoped at a line "Sending: *END" after sending the goemtries.

Which looks like:

----

Sending: -1.781344305128055749e+00 6.164699427899241080e+00 1.045976295437978898e+01
Sending: 9.389194690638786867e-01 6.532143743769908140e-02 9.348241640577274580e+00
Sending: 1.570355323807899106e+00 2.132461060074572590e+00 1.060578631428835550e+01
Sending: 2.831839588942477892e+00 3.801566033085470586e+00 9.188823229078074917e+00
Sending: -2.388545699875365136e-01 3.935826158081394688e+00 1.050017090366437422e+01
Sending: 8.885877016843556353e-01 6.213307231606743031e+00 9.739227808743871506e+00
Sending: -2.366576032546644281e+00 4.062879394492279417e+00 9.149917213016674822e+00
Sending: 1.016987309042416632e+00 -7.640199617120106979e-01 1.066762903969025267e+01
Sending: -2.957585080366882391e+00 3.314360406155627548e+00 1.037902374314034759e+01
Sending: *END
-------

My experience is that this problem is more frequent for large system (~ 250 atoms), but less frequent for small examples.

So what is the possible reason for this error? I will appreciate any suggesstions.

Thank you very much.

Best Wishes,

Geng

[Maxime Van den Bossche]

Dear Geng,

I remember having the same problem before, but it was quite

some time in the past, so I hope my memory is correct.

I think I solved it by switching from IntelMPI to OpenMPI

in compiling (and running) the CP2K binaries. So at least,

if you got this behavior with IntelMPI, I think it would be worth

giving OpenMPI a try.

Note that, in my use of the ASE interface with CP2K+OpenMPI,

I need to suppress certain MPI-related messages, which would

otherwise crash the ASE interface. You can do this by adding

the "-mca btl ^openib" flags to your command for running the

CP2K shell, e.g.:

mpirun -np 16 -mca btl ^openib cp2k_shell.popt

Hope it helps,

Maxime

[Geng Sun]

Hello Maxime,

Thank you very much  for your suggestions.

I am using the CP2K binaries compiled by the cluster administrators. Indeed I am not sure of the version of MPI.

I will give it a try to use openmpi to see what will happen.

Best,

Geng

在 2018年11月10日星期六 UTC-8上午8:14:16，Maxime Van den Bossche写道：

[Maxime Van den Bossche]

I'm glad to help out a bit, as I remember this was

a very annoying (and puzzling) problem when I first

encountered it!

If you know the path to your cp2k_shell.popt binary,

you can look at the output of the "ldd" command, and

that should contain references to either IntelMPI or

OpenMPI (or yet other) libraries, e.g.

ldd <path to binary>

or (if the directory with the CP2K binaries is in your $PATH):

ldd `which cp2k_shell.popt`

--Maxime