problem of using lammps + ipi

[Bryukhanov Ilya]

Hi!

I have recently installed i-pi in lammps by adding USER-MISC package. Howeveer, when using it with lammps I stucked on the problem

ERROR on proc 0: Error opening UNIX socket: server may not be running or the path to the socket unavailable (../fix_ipi.cpp:121)

I tried both serial lammps compilation or mpi cluster run.

Could you tell me what I'm doing wrong?

Regards

Ilya

[Michele Ceriotti]

You need to give some more detail here. Are you trying to run one of the examples in i-pi or is this your own input? 

Can you run i-PI without problems using the test driver that you can compile in driver/ ?

[Bryukhanov Ilya]

I run i-PI for some tutorials in tutorials/ folder by calling

"i-pi tutorial.xml" in one terminal and "i-pi-driver -m sg -h localhost -o 15 -p 31415" in another.

They seem to work: "tut.*" files are created.

However, I failed to run tutorials in lammps/ folder. I use Mar20 version of Lammps. 

"lmp_mpi_serial -i in.lmp" command turned to the error message

[Michele Ceriotti]

Which tutorial? 

If you go to examples/lammps/h2o-pimd.4

run
 i-pi input.xml 

and then in another  terminal 

lmp_whateve < in.lmp

does it work?

[Bryukhanov Ilya]

Thanks a lot. Now it seems working. I ran "graphene" tutorial instead.

Could you tell me if it is possible to run i-pi + lammps on cluster in serial or mpi mode?

[Michele Ceriotti]

examples/lammps/graphene should also run if you run i-pi before you start LAMMPS.

should be possible to run i-PI + lammps on a cluster, but I stongly recommend you first give a good read to the manual, because it'll be hard to get it to work if you don't understand the basic mechanism of functioning of the code.

[Somayeh Ahmadkhani]

Dear users,

 I am facing the same Error but using examples and all othere atill couldn't solve it


in log.driver I am facing the following error using lammps as a client

LAMMPS (24 Oct 2018)

OMP_NUM_THREADS environment is not set. Defaulting to 1 thread. (src/comm.cpp:87)

  using 1 OpenMP thread(s) per MPI task

Reading data file ...

  orthogonal box = (-2.5 -2.5 -1.5) to (2.5 4.2 67.5417)

  1 by 1 by 1 MPI processor grid

  reading atoms ...

  75 atoms

  scanning bonds ...

  2 = max bonds/atom

  scanning angles ...

  1 = max angles/atom

  reading bonds ...

  50 bonds

  reading angles ...

  25 angles

Finding 1-2 1-3 1-4 neighbors ...

  special bond factors lj:   0          0          0

  special bond factors coul: 0          0          0

  2 = max # of 1-2 neighbors

  1 = max # of 1-3 neighbors

  1 = max # of 1-4 neighbors

  2 = max # of special neighbors

11 atoms in group left

52 atoms in group middle

12 atoms in group right

75 atoms in group all_atoms

11 atoms in group left

12 atoms in group right

52 atoms in group middle

75 atoms in group all_atoms

WARNING: Fix ipi always uses group all (src/USER-MISC/fix_ipi.cpp:190)

PPPM initialization ...

  extracting TIP4P info from pair style

  using 12-bit tables for long-range coulomb (src/kspace.cpp:321)

  G vector (1/distance) = 0.87081

  grid = 12 15 216

  stencil order = 5

  estimated absolute RMS force accuracy = 0.0304782

  estimated relative force accuracy = 9.17843e-05

  using double precision FFTs

  3d grid and FFT values/proc = 130962 38880

ERROR on proc 0: Error opening INET socket: wrong port or server unreachable (src/USER-MISC/fix_ipi.cpp:109)

Last command: run 5000000

application called MPI_Abort(MPI_COMM_WORLD, 1) - process 0

[unset]: write_line error; fd=-1 buf=:cmd=abort exitcode=1

:

system msg for write_line failure : Bad file descriptor

"log.driver.1" 50L, 1764C 

and I have  attached lammps, i-pi and scrtips

do you have any idea how can I solve it?

[Michele Ceriotti]

Since you seem to be running on a cluster, it depends on how it is set up. 

This https://docs.ipi-code.org/distributed.html can maybe help a bit. 

One possible issue I can see is that the i-pi server and the clients are dispatched

to different nodes, so the two cannot find each other. IF they run on the same node,

then it should work, unless the `host` file in the network configuration does not 

define localhost to be an alias which is unlikely but not impossible.

you may want to try using a unix domain socket, that is also faster.

if they're indeed dispatched to different nodes, then it's tricky - best is to

run i-pi on the login node if they allow you, and to specify the IP address

a side note - be careful when you use LLMs in your work, this comment below 

<!-- i-PI MUST be in NVE mode because LAMMPS is handling temperature controls -->

is false - LAMMPS only evaluates forces so you most definitely need to take care about

temperature on the i-pi side.

M