Free Energy Calculation from Advanced MD Techniques

Message

vinchy · #1 Post by **vinchy** » Tue May 08, 2012 8:28 pm

Hello,

I am using the new feature in VASP 5.5.12---advanced MD to calculate the free energy of liquid phase reactions. I tried the thermodynamic integration of free energy gradient method, as explained in Bucko's paper (J. Phys.: Condens. Matter 20 (2008) 06421).

My system is a liquid phase reaction, so I put the reactant molecule and about 30 H2O molecules in a 10*10*10 angstrom cube box. After about ten picosceond MD calculation, the standard deviation of free energy gradient is in the range 0.2-0.7, which is huge and approximately 10 times of the average free energy gradient value. It seems a common way to decrease the standard deviation is to run in longer time, which I cannot afford: to let the standard deviation lies in 0.02-0.07, I will need 100 times longer time, which is unreasonable. I repeated this kind of calculation for different reactions, and always got the standard deviation in this large scale. My INCAR file is shown below, and is there anything wrong?
ALGO=Fast
PREC = Medium
NELM = 40
LREAL = Auto
NSW = 30000
ISYM = 0
MDALGO = 2
SMASS = 2.0
ISIF = 0
IBRION = 0
POTIM = 0.5
TEBEG = 320
LBLUEOUT=.TRUE.
INCREM = -1E-6
SHAKEMAXITER = 2000
ISMEAR = 0
SIGMA = 0.05

On the other hand, I am looking at another method of the advanced MD technique---biased molecular dynamics. Is umbrella sampling method implemented in VASP? since it is not very clear in the manual about how to perform this calculation, what tags should be used in INCAR?

Am I looking at the right way to do a free energy calculation? Any comments is welcome! Thanks in advance!

Best,

Ping

#2 Post by **admin** » Wed May 09, 2012 1:21 pm

The equilibration time for liquid phase is long. Even dimer molecules require several nanoseconds (cf. e.g. JPCB112(2008)242) no matter which biased MD technique is used. Slight improvement could be achieved by making hydrogens heavier (mass of H = 3) and using longer timestep (POTIM = 1).

vinchy · #3 Post by **vinchy** » Wed May 09, 2012 5:43 pm

Thank you so much for reply. Actually I already did that in my calculation by using tritium isotope for hydrogen and 1fs as time step. But in the length of my calculation (about 10 ps), I did not see improvement of the standard deviation. So what else can I do?

#4 Post by **admin** » Thu May 10, 2012 10:31 am

Only longer simulation time will help. If you really must deal with the liquid phase then you should switch to faster techniques (force fields) which allow you to do simulations on the scale of hundreds of nanoseconds.