Segmentation fault without further information

User discussions
1

GROMACS version: 2025.4-conda_forge
GROMACS modification: No
Hi,

I am trying to simulate a Self-Assembly Monolayer. I have followed what I understand are standard steps:

Box creation, Energy minimisation, NVT process and finally NPT. The process works until it reaches the NPT. I works properly the first 100000 steps. Then I stop and try to run it another 100000 steps, and a segmentation fault is created.

The reason to do two NPT steps is because it gave me an error associated to the number of CPUs so I have tried to minimise it making it work in shorter steps. I run the program in a 16CPU x64 server (32 threads).

The same procedure works with other molecules.

Is there any way to get more information from grimacss about the cause of the segmentation fault?

Thanks in advance

2

You should never get a segmentation fault.

If you got a core file, the following should give some information:

gdb path-to-gmx-binary core-file

If not, you need to run gmx from gdb:

gdb –args gmx mdrun …

3

Hi, after using the gdb alternative (I have no core-file), the message I get is:

Program received signal SIGSEGV, Segmentation fault.

[Switching to Thread 0x7fffccff9700 (LWP 31715)]

0x00007ffff73728d1 in spread_on_grid(gmx_pme_t const*, PmeAtomComm*, PmeAndFftGrids*, bool, bool, bool) [clone ._omp_fn.1] ()

   from /opt/anaconda3/envs/gmx_env/bin.AVX2_256/../lib.AVX2_256/libgromacs.so.10

Any ideas about the reason behind?

Thanks another time

4

That likely means that at least one atom has very large coordinates, which causes out of range PME grid indexing. This is one of the few issues we do not catch, because of the cost of the checks. This indicates that your initial conformation is not stable or your time step is too large.