Constant-pressure molecular dynamics in VASP.5.3.2

Message

fanghz · #16 Post by **fanghz** » Tue Oct 23, 2012 10:33 pm

In NPT simulation, I found the SMASS keyword does not work any more, so the LANGEVIN_GAMMA is supposed to monitor the MD temperature, is that right? In my case, I set TEBEG=900 and TEEND=900, but I got an average temperature of 800K, as observed from OSZICAR file. Any ideas to bring the MD temperature to be consistent with the target temperature? Thanks!
[ Edited Tue Oct 23 2012, 11:13PM ]

fanghz · #17 Post by **fanghz** » Tue Oct 23, 2012 11:26 pm

According to the manual, the default value of PMASS is 1000, while the suggested value here is between 1 and 10. I'm confused by the huge difference.

[quote author=0, POTIM, TEBEG) plus add he following flags:

ISIF=3 #variable cell shape and volume
MDALGO = 3
PSTRESS = [scalar]

LANGEVIN_GAMMA = [real array] # friction coeficients (in ps^-1) for atomic degrees of freedom, one number for each species defined in POSCAR must be provided

LANGEVIN_GAMMA_L= [real] #friction coeficient (in ps^-1) for lattice degrees of freedom (one number)

PMASS=[real] . # mass (in amu) for lattice degrees of freedom (one number)

The optimal values for the friction coefficients are system dependent - we don't have easy rule to determine the optimal values (typical values are from interval between 0 ps^-1 (no thermostating) and 100 ps^-1 ) - please check the vast literature on Langevin dynamics (good starting point is certainly the book of Allen and Tildesley: "Computer simulations of liquids"). As to the value for mass of lattice degrees of freedom (PMASS) - this is again very much system dependent, the optimal value is a compromise between two opposing factors: too large value leads to very slow motion of lattice degrees of freedom (and hence the sampling is inefficient) while too small POMASS (without adjusting POTIM to very small value) could cause numerical problems. In our experience, reasonable values of POMASS are from interval 1 and 10 - but once again we recommend to check literature and make careful tests for your particular system of interest before you perform "production run".
[ Edited Tue Sep 25 2012, 01:51PM ]
[ Edited Tue Sep 25 2012, 01:56PM ]
[ Edited Tue Sep 25 2012, 03:35PM ]
[ Edited Fri Sep 28 2012, 10:03AM ][/quote]

pblanko · #18 Post by **pblanko** » Wed Oct 24, 2012 9:37 am

[quote author=900 and TEEND=900, but I got an average temperature of 800K, as observed from OSZICAR file. Any ideas to bring the MD temperature to be consistent with the target temperature? Thanks!
[/quote]

I also noted that. The initial average temperature for several thousand time steps is under the desired value. The developers told me the cause is the stocastic algorithm. If you keep the simulation you will notice the average temperature gets to the right value after about ~5,000 - 10,000 steps.

fanghz · #19 Post by **fanghz** » Thu Oct 25, 2012 12:03 am

As running meaningful NPT simulation requires a high ENCUT value, its extremely expensive to run 5000-10000 MD steps using VASP. I think there must be some other solutions to adjust the temperature.
[ Edited Thu Oct 25 2012, 12:05AM ]

#20 Post by **admin** » Mon Oct 29, 2012 3:00 pm

SMASS is used in NVT simulations with Nose-Hoover thermostat and is inactive in Langevine dynamics used in the present NpT implementation. As to the temperature - it is not guaranteed that the average temperature from N steps equals the temperature of thermostat unless N approaches infinity. In practice, proper setting of friction parameters, and atomic and lattice masses can "accelerate" the equilibration.

pblanko · #21 Post by **pblanko** » Wed Nov 28, 2012 3:57 am

[quote="HIGH as recommended. PSTRESS=10 !1GPa[/quote"]

I solved the problems with instability of the NPT for long runs using ALGO=Fast instead of IALGO=48. It is slightly slower but worked well for simulations with over > 25,000 MD steps.

maartendft · #22 Post by **maartendft** » Wed Dec 19, 2012 8:37 pm

Hi all,

I'm trying to run NPT-MD in VASP, but for some reason, the cell geometry does not change. I think I compiled VASP correctly for NPT-MD so maybe the problem is in my INCAR? Can somebody please post an INCAR that works for NPT-MD so that I can trace down the problem? My INCAR is posted below.

system = Ti
ISTART = 1
ICHARG = 1
ISMEAR = 2
SIGMA = 0.1
LREAL =AUTO
PREC = Accurate
MDALGO = 3
ENCUT=450
PSTRESS= 0
EDIFFG = -0.001
EDIFF = 0.00001
LWAVE = .TRUE.
#optimizations mentioned for linux clusters in the vasp manual
#LPLANE = .TRUE.
#NPAR = 4
#LSCALU = .FALSE.
#NSIM = 4
#Molecular Dynamics:
IALGO =48
ISYM =0
NELMIN = 4
NSW = 40000
SMASS =-1
MAXMIX =45
IBRION = 0
IWAVPR =11
ISIF =3
POTIM =2
#DATA:
NPACO =256 # pair correlation function
APACO =16 # pair correlation function
NBLOCK =1
KBLOCK =10000 #same as NSW
TEBEG = 20
TEEND = 20
#REAL SPACE INTEGRATION:
LREAL =A
LANGEVIN_GAMMA = 50
LANGEVIN_GAMMA_L = 50
PMASS=1000

maartendft · #23 Post by **maartendft** » Wed Dec 19, 2012 8:39 pm

PS

The cell vectors in CONTCAR are the same as those in POSCAR for every MD-step, so that's why I think nothing gets updated. Or does it get written out somewhere else?