I am trying to calculate solvation free energy in gromacs using the free-energy option.
When I run gmx bar command on the output files, there are two tables output on the screen.
The first table has two columns s_A and s_B among others. I want to understand the meaning of these two quantities.
In the documentation it says that it is the relative entropy of one state in the phase space of the other, which is the measure of the overlap of phase space of the two states.
It also says, that, the value closer to zero indicates that the overlap is better.
However, what values of s_A and s_B are required for the phase spaces to be suffiently overlapping?
Also, since going away from zero, decreases the overlap, does the same happen when the values of s_A and s_B go in the negative direction?
Are small negative values (closer to 0) acceptable, or negative values are to be avoided altogether?
I don’t understand what you mean with negative. These are entropies, so always positive.
It is difficult to say what is acceptable. The error estimates should tell you what is acceptable. Errors can be higher either do to little overlap (high entropy) and/or due to long correlation times. In general it is good to add lambda points in intervals where the entropy is high.
What I meant by negative is the s_A and s_B values for some windows come out to be negative for me with a warning that this violates the second law of thermodynamics. I read the reference paper for the gmx bar and one thing I understood is it cannot be negative.
I understand your point regarding the acceptability of values. Thank you for that explanation.
If you get negative values you like have serious sampling issues. Then you can not trust the results. Either something is wrong in your setup or you need to run much longer.
I would like to ask a more specific question in this regard.
I did a free energy calculation with a delta_lambda value of 0.005. The overlap in this case is not satisfactory to me. So, I introduced intermediate lambda value, so the new delta_lambda is now 0.0025, half of what earlier was. But I now the deltaU_fwd and deltaU_bwd are even more apart in comparison to the earlier scenario.
The basis of comparing the overlaps is the difference between the delta_H values in the .xvg files. For the scenario where delta_lambda is 0.005, the delta_H values (forward jump and backward jump) on an average appear to have a 2 to 3 units difference, whereas this difference is 5 to 6 units in the scenario where delta_lambda is 0.0025.
I do not understand how this can be. Can you help me in this regard?
I hope my question is clear. I am attaching a link to my input and output files, if you want to have a look.