Loop cannot be run normally in python lammps module

LAMMPS LAMMPS General Discussion
1

Hi LAMMPS developers and users,

I encountered two problems when assigning a series of time-dependent values for each atom.

LAMMPS version: LAMMPS 64-bit 2Aug2023-MSMPI
python version: 3.10.14
System: windows 10

Problem 1
The following script could be run normally via lmp command lmp -in test.in:

variable cur_step loop 2000
label loop_start
print "${cur_step}"
next cur_step
jump SELF loop_start

Output:

1
2
…
2000

But could only run for one cycle in python lammps module:

from lammps import lammps

lmp = lammps()
lmp.commands_string(
    """
    variable cur_step loop 2000
    label loop_start
    print "${cur_step}"
    next cur_step
    print "after loop"
    jump SELF loop_start
    """
)

Output:

1

This is very weird. Why could it be normal in lammps but abnormal in python?

Problem 2
I used the following script for the assignment:

variable cur_step loop ${n_step}
label loop_start
run 1
variable FP delete
variable FP atomfile ${case_path}/csv/force_${cur_step}.csv
next cur_step
jump SELF loop_start

It ran as expected via the command lmp -in, but merely too slow. I guess the for_loop limited the performance, in which the per-atom values were deleted and then read at every timestep. Is there a faster and more efficient way to implement this function?

Any comment or advice will be appreciated.

2

It is not weird but the consequence of documented behavior.

In general, looping in LAMMPS is based on reading the input from a file.
Please read the documentation of the “jump” command. The “SELF” keyword is a special case and only works in some cases. The normal version of this command would require to provide a file name and the “jump” command would close the current file, open the provided file name and then read it line by line until it encounters the “label loop_start” line.

Expecting this for a block of strings to work is not logical.
That said, if you upgrade to the 29 August 2024 version, this block of code should work, because extra code has been added for the special case of a self-contained block of strings that use the “SELF” pseudo file in the “jump” command.

What you are doing here makes no sense. Why reopen and read the file in every step? Why not read it outside the loop and be done?

3

@akohlmey
Thanks for your kind reply. I am now clear about Problem 1.

For Problem 2,

Why reopen and read the file in every step? Why not read it outside the loop and be done?

They are different files. At a certain timestep, I should open the corresponding file.

What I want to assign is a time-dependent force for each atom i, i.e., F(i, t).
To assign them via variable atomfile, I stored these values in N_\mathrm{step} csv files. At each timestep t_i, I read the corresponding forces from force_{step}.csv and assigned them to each atom.
image

4

You can speed up the process a little bit by combining the files into a single file and use the next command to loop through the data set, but the general overhead of reading and parsing a text file and broadcasting its information across all processors remains and may be dominant versus the computation of the forces.

The only way I can think to speed this up would be to not read forces from files, but compute them on-the-fly.

1 Like
5

Got it, thank you very much!