1/viscosity output is wrong in gmx-2018.2, but OK in gmx-5.1.5
When calculating viscosity with periodic perturbation, the output average value of 1/viscosity in 2018.2 is around zero. The NEMD is set up in NVT ensemble. Only "cos-acceleration" is set, instead of "acc-grps" and "acceleration" in .mdp file. The same tpr file works with gmx-5.1.5, i.e. gives value in reasonable range.
Both versions of gromacs are compiled on CentOS 7, with gcc-7.3.1/or gcc-4.8.5, cuda-9.2, with the following cmake command:
cmake .. -DGMX_BUILD_OWN_FFTW=ON -DGMX_GPU=on
Missing factor of 2 in cos-acceleration
The bug was introduced in commit 69470fc4e9fd990eb53b22c17dc7a699c583d126
Last working version is 2016.5
Affected 2018 and 2019
#1 Updated by Jiannan Liu 2 months ago
I am having exactly the same issue as Sun with 2018.4. I am calculating viscosity of water. 2016.5 gives reasonable results, but 2018.4 does not. In contrast to Sun's GPU version, I am using 2018.4 with CPU version only.
Please let me know if you need more information. Hope this can be fixed soon.
#6 Updated by Jiannan Liu 2 months ago
Here are the files and commands to reproduce.
# to run simulation, a0.gro is a pre-equilibrated water configuration gmx pdb2gmx -f a0.gro -ff opls -water spce gmx grompp -f nvt_vis.mdp gmx mdrun -pin on -ntmpi 1 # get results, expected ~1700 for 1/Viscosity when \lambda is 0.12 # In 2018, the result is near zero gmx energy -b 1000