Hi everyone,
I have found a segfault in liggghts (with atom_style granular). Files to reproduce are joined. To reproduce, use mpirun with at least 2 processors (mpirun -np 2 liggghts -in agg.in_.txt); segfault will occur between 15'000 and 20'000 time step.
About system: Linux; result of uname -a command: Linux PC-GUESNET 3.16.0-38-generic #52~14.04.1-Ubuntu SMP Fri May 8 09:43:57 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux; mpirun version is 1.6.5.
About Liggghts: last version in github (3.3.0), compiling with default option (make fedora), no any modifications.
More information: to cancel the segfault (make this script working), you can:
- Use an other normal model (here hooke/stiffness, e.g. hertz/stiffness does not segfault).
- Use an other tangential model (here tangential history, e.g. no any tangential model does not segfault).
- Erase the tree lines about thermo (default thermo style does not segfault).
- Add "unfix" before the second fix (while documentation says "If you specify a new fix with the same ID and style as an existing fix, the old fix is deleted [...]. This is the same as if an "unfix" command were first performed on the old fix").
- Run this script without mpirun, or with mpirun and only one processor.
I don't know what's append. If you need more information, just demand me.
Regards,
Étienne.
| Attachment | Size |
|---|---|
 agg.in_.txt | 1.28 KB |
| agg.coord_.txt | 116.81 KB |
ckloss | Tue, 12/22/2015 - 19:48
Hi Étienne,
Hi Étienne,
thanks - we'll try to have a look before the next release in mid Jan
However, one thing caught my eye:
set group all density 1e18
1e18 seems to be quite high?
Best wishes
Christoph
Guesnet Etienne | Mon, 01/04/2016 - 16:32
Hi,
Hi,
10^18 is not so high. I work with little particles (~10 nm), so I cannot consider realistic values (because timestep will be something like 10^-12).
As I work in quasi-static, I am trying mass renormalization (I prefer real Young modulus than real masses). It means each nano-particles weights some grammes.
I am working to make this correctly, and confirm results are true.
Best,
Étienne.
ckloss | Wed, 01/13/2016 - 22:31
Hi Étienne,
Hi Étienne,
thanks again for the note. I tried my best - that's a quite tricky one. On my end I could resolve the issue but I am not 100% sure if the root cause is eliminated. Please check the new version (probably tomorrow) and let me know in case you still have problems!
Christoph
Guesnet Etienne | Mon, 01/18/2016 - 10:20
Hi,
Hi,
No problems with LIGGGHTS-PUBLIC 3.3.1. Thanks!
Étienne.