Free Energy Calculation Using Blue-moon Method

[mhl980]

Dear Everyone,

I'm trying to use blue-moon method. I follow a simple instruction in the forum (The link below).
https://cms.mpi.univie.ac.at/vasp-forum ... =4&t=17118

The result looks very good since the average of free energy gradient change (average every 100 MD steps) smoothly. However, when I tested on my own molecule I found that my free energy gradient fluctuated in a wide range even i averaged them every 200 MD steps.

1. So Is there any way to improve the result ?.
2. I simulate the reaction at very high temperature should I choose small or large timestep (POTIM) ? and the same question with growth rate (INCREM).

Here is my result.


And my INCAR
System = NSO
!General setting
PREC = Normal
ISPIN = 1
ICHARG = 2
ENCUT = 300
LREAL = Auto
ISMEAR = 0
SIGMA = 0.2
NPAR = 4
NELM = 500
LCHARG = .FALSE.
!LWAVE = .TRUE.
ALGO = Very Fast

!Slow-growth approach
IBRION = 0
TEBEG = 1000
TEEND = 1000
POTIM = 0.5
NSW = 5000
INCREM = 0.001
MDALGO = 2
SMASS = 0
NBLOCK = 50
NWRITE = 0
LBLUEOUT=.TRUE.

Best Regards
MHL

[admin]

Your results look nice.
Calculated curves are more smooth when the simulation time of one
point on the collective variable is longer,
timestep shorter, and temperature lower (do you really need T=1000K?).
Sensitive element is H. You should use such timestep, that during
one stretching vibration of the C-H bond at least ten simulation
steps are performed.

[mhl980]

Dear Admin,

Thank you for your reply,

I have one more question.
Intead of using slow-growth approach, I can sample points along the reaction coordinate and calculate average gradient energy (constrained force) at those sample points (then do the integration).
The gradient energy can be calculated from the output below (the last term "|z|^(-1/2)*(lambda+GkT)" divide to the second one "|z|^(-1/2)"). In my case, it is equal to lambda since the value of GkT is very small.

>Blue_moon
lambda |z|^(-1/2) GkT |z|^(-1/2)*(lambda+GkT)
b_m> -0.181209E+02 0.970143E+00 -0.450497E-14 -0.175799E+02

if I perform a simulation of 5000 steps at each sample point and I want to calculate the average of gradient energy. Should I
1. Average all the values of lambda (or [|z|^(-1/2)*(lambda+GkT)]/[|z|^(-1/2)]) for all the 5000 steps ?.
2. Average all the values of <|z|^(-1/2)> and <|z|^(-1/2)*(lambda+GkT)> first then divide ?.

Best Regards
MHL

[admin]

Average all values of lambda