How does the VASP MD stop running?

[luckylin]

I met a problem with MD running. I use Nose to control the temperature. And the temperature I used is fixed at 473.15 k. Here is my INCAR file:

SYSTEM = XX
ENCUT = 480
PREC = Normal
ISTART = 0
ICHARG = 2
ISMEAR = 0
SIGMA = 0.05
ISYM = 0

#MD settings
IBRION = 0
ISIF = 0
POTIM = 0.5
NSW = 5000
NELM = 100
TEBEG = 473.15
TEEND = 473.15
MDALGO = 2
SMASS = 1.0
POMASS = 26.981 28.085 16.000 14.001 3.000 63.546
ALGO = Very Fast
INCREM = -1e-8
SHAKEMAXITER = 2000

LBLUEOUT = .TRUE.

#Don't write WAVECAR and CHGCAR
LWAVE = .FALSE.
LCHARG = .FALSE.

NWRITE = 2

But it stops at 1452 step and the last temperature seems absurd (over 10000k). I start MD running with the optimised structure from vasp geometry optimisation calculation.
Do I need to preheat the system? Is it that the temperature I used here for MD running a bit high?
Could anyone explain this problem? Many thanks!

[admin]

There is a check of the temperature in the code, and when it is rapidly
increased, like in your case, the calculation stops.
I recommend you to display structures produced my the simulation
and have a look what is going on.

[luckylin]

Thank you for your reply. Now it is running and give me the REPORT to calculate the free energy difference. But I have some puzzles on that.

'The unit of lambda depends on the constrained coordinate. When constrained coordinate is Angstroem,
then the unit is eV/A. The integration of the gradient of the free energy (last number in the b_M> line)
along Angstroems (eq. 6.30) then gives the change of the free energy in eV.'

This is what you answered before. But I found a bit confused when I calculate the free energy difference from the REPORT.
The last number in the b_M> line is |z|^(-1/2)*(lambda+GkT), which I think is not the free energy gradient. Because from eq. 6.29, free energy gradient
is equal to |z|^(-1/2)*(lambda+GkT) divided by |z|^(-1/2).

Am I right? Looking forward to your reply. Thanks!

[admin]

You can calculate the change of the free energy as a path integral (6.30).
For this purpose you need to define the path (ICONST file). The path is sampled
in a series of points. In each point the gradient of the free energy is calculated
over the MD run using Lagrange multipliers lambda (6.29). The unit of the gradient depends
on the definition of the path. When the path is a distance measured in A, then
the unit of the gradient is eV/A. The integration over the whole path then
gives the free energy change in eV.

[luckylin]

Dear Admin, 
Thank you very much for your response. I still have one question and it is related to the response you posted Wed Nov 26, 2014 2:06 pm (topic: ‘Help please: constrained MD does not work’). In the response, it is stated that the last column in the REPORT file gives the free energy difference. However, according to the description of the columns in the REPORT file and the source code, I assume it is the last column divided by |z|^(-1/2) (second column) that gives the free energy difference (eq. 6.29). Am I right on this? Looking forward to your reply.

[admin]

Yes, correct. Note this is not the free energy difference, but gradient of the free energy.
The integration of gradients gives the free energy difference.