Dear
my system works for a temperature of 10K. However for T=300K I have the following error
ERROR: Lost atoms: original 43688 current 43609 (src/thermo.cpp:438)
What could I do?
Please note that you have an extremely high pressure. That is usually an indication that you have a bad initial geometry or a bad model with unsuitable potential parameters or both.
The fact that you can run at 10K doesn’t indicated anything, because you a) have set a tiny timestep
(for elements like C O N you should be able to use 1fs instead of 0.1fs even at 300K) and b) not much happens at a low temperature of 10K, so your system can remain in some meta stable state.
A bad geometry with close contacts can often be relaxed by running a minimization before starting an MD. If that does not help you likely have problems with your model.
In case of problems it is also advisable to output thermodynamic data and trajectory more frequently, so you can observe which properties are at unusual values during minimization or MD and which atoms move unexpectedly.
1 Like
The problem was that it used repulsive LJ-type walls and the wall was too close to the edge of the box. The reflective walls solved the problem. It would be possible to use the Fix Wall with the repulsive potential of LJ, but I would have to CORRECTLY adjust these walls.
Thank you so much, Axel.
Just today I have received error and all of the atom lost. Here is how bing chat helps me understand Axel’s comment from another discussion:
LAMMPS is a program that simulates how atoms and molecules move and interact with each other. It can run on many computers at the same time to speed up the calculations. To do this, it divides the space where the atoms are into smaller regions called sub-domains, and assigns each sub-domain to a different computer.
Each computer needs to know the positions and velocities of the atoms in its own sub-domain, as well as some of the atoms in the neighboring sub-domains. These are called local atoms and ghost atoms, respectively. However, to save time and memory, each computer only communicates the ghost atoms that are within a certain distance from its sub-domain boundary. This distance is called the communication cutoff2.
Sometimes, an atom can move so fast or so far that it goes beyond the communication cutoff of its original sub-domain. This can happen for different reasons, such as when an atom leaves the simulation box (if there are no periodic boundaries), or when an atom experiences a very strong force (e.g. due to a collision), or when the sub-domains change their shape or size (e.g. due to shrinkwrap boundaries). When this happens, the original computer will lose track of that atom, and will not be able to communicate it to the new computer that owns it. This is called a lost atom2.
Lost atoms are a problem because they can cause errors or crashes in the simulation. To avoid them, one needs to choose appropriate parameters for the simulation, such as the time step, the communication cutoff, the neighbor list frequency, and the boundary conditions2.
So, I changed the number of the MPI - OMP threads combination in my job file from 24 CPU to 16. Previously, it maps this 24 CPU into 2 x 2 x 6 grids. This gave me the lost atom error. Also, the potential energy, the pressure are very huge, up to the power of 45. Now the 16 maps into 2 x 2 x 4 grids and it give no error and the energy and pressure are acceptable.
Regards
Nur
If the results differ this much between using different domain decompositions, there is something seriously wrong in your input, either geometry or potential parameters. That problem is just hidden by the one decomposition and exposed by the other. You cannot trust either result until you have fully understood why you get the bad results in the one case. No chat bot will be able to tell you.
2 Likes