Results differs in different versions

User discussions
1

GROMACS version:Gromacs-2023 , Gromacs-2026
GROMACS modification: Yes/No
Hello,

I am working on a project to study the dielectric constant of the system. I used the gmx current tool to calculate the rotational and translational part of the system. We installed the gromacs 2026 version on our cluster. We previously used gromacs-2023 version. We compared the results from gmx current from both the version and both are differ significantly.

The simulation were tested using the same force field, same input and same machines.

Attached is the input script, the auto correlation function of rotational part.

This difference affect on the dielectric constant calculations.

Any suggestion why the results are different.

gromacs_dipole.txt (6.1 KB)

gromcas_2026
gromcas_2026640×480 3.4 KB

gromcas_2023
gromcas_2023640×480 3.54 KB

2

The results have not converged, so they will be different. MD simulations are chaotic, so they diverge exponentially. Running exactly the same tpr with the same binary can produce different results, depending on whether the exact run conditions are binary reproducible or not.

3

Thanks for the reply. The system was annealing for 5 ns from 700k to 300k and equilibration was performed for 20ns.