Green-Kubo method for calculating viscosity

Dear all,

Dear all,

Viscosity estimates are known to have difficulty converging. You may need to run multiple short runs once the system is at equilibrium and compute the average viscosity from these instead of a single correlation measurement.


Yong Zhang, Akihito Otani, and Edward J. Maginn. “Reliable Viscosity Calculation from Equilibrium Molecular Dynamics Simulations: A Time Decomposition Method” JCTC

for suggested best-practicies

Dear Smith,

Thanks for your answer.
It seems that the calculated viscosity is quite dependent on the value of correlation length. Of course, correlation length should affect convergence of the computed viscosity. But I think that it’s effect on my results is abnormal. for example, if I reduce correlation length from 400 to 100, the calculated viscosity reduces from about 0.00082 pa.s to about 0.00037 pa.s. It is while the correlation values obtained for off-diagonal elements of the stress tensor do not show a significant change. Is it normal?
It should be noted that I have reduced my time step to 2.0 fs now, that is the appropriate time step in presence of water molecules.


Hi Majid,

2fs is fairly standard; what matters for your correlation function is the frame saving rate, which should be between 5fs and 20fs . You need to be saving frequently in order to resolve the high frequency correlations (short times) and you need to run the system “long enough” (which is hard to say a priori) to get enough sampling in the of the low-frequency correlations (long times)