Multibinit spin dynamics inconsistencies

Running Abinit Other tools and postprocessing in ABINIT
1

Dear Abinit users,

While I was testing Multibinit for spin dynamics simulations (example nr.1, tmulti5_1 from the test folder) I noticed how values for the final magnetic moment differ as the number of processors increases. The results are listed below (examples of full results attached). I used ABINIT v.9.2.2 compiled with Intel2018 compilers on a local AMD-based cluster (each node having two 24-core processors).

nproc | Ms

1 | 4.10E-01
2 | 4.11E-01
4 | 4.03E-01
8 | 4.09E-01
16 | 4.02E-01
24 | 3.39E-01
32 | 4.06E-01
48 | 2.52E-01
96 | 6.44E-02

I am not sure what I am doing wrong as I just ran the ready-made examples…has anyone encountered anything similar?

Cheers,
Aleks
0001_tmulti5_1.out (23.7 KB)
0004_tmulti5_1.out (23.7 KB)
0007_tmulti5_1.out (23.7 KB)
0008_tmulti5_1.out (23.7 KB)

2

Dear Alex,
The parameters in tmulti5_1 is not suitable for a realistic run (the cell size, the time steps are too small). It is for testing whether the spin dynamics can run (on some test farm machines within reasonable time), not for testing the end result. When different number of cores are used, the random numbers used in the dynamics differ and the fluctuation could be large.

Here the Mst is just from a final snapshot so the fluctuation is large when the cell is small. Indeed this can be improved by averaging over the snapshots.

You can try to increase the size of the cell and the time of steps to see if the Cv, chi, and Mst can converge.

Best regards,
HeXu