relax at 0 K

I want to relax the sample at 0K and 0 Pa, how can I achieve that? It seems that the fix npt command can not be used at 0 K.

Lisa

Try fix box/relax combined with minimization ?

Cheers
Yi

Thank you. But I want the sample to be equilibrated at 0 K and 0 Pa. I tried that but the target pressure can not reach 0 Pa at the end of the minimize.

fix 1 all box/relax aniso 0.0
minimize 1e-10 1e-10 1000 10000

And another concern is that I am afraid that the minimize just reach the matastable state and the final state is not stable. Is there a better way to do this? I want to check that after the sample been equilibrated at 0 K, the pressure should be also 0, and the energy should be conserved during the NVE run.

Lisa

2011/8/18 Ian <yidong80@…24…>

Thank you. But I want the sample to be equilibrated at 0 K and 0 Pa. I tried

there is no "equilibration" at 0K. at 0K _nothing_ moves.

that but the target pressure can not reach 0 Pa at the end of the minimize.

fix 1 all box/relax aniso 0.0
minimize 1e-10 1e-10 1000 10000

And another concern is that I am afraid that the minimize just reach the
matastable state and the final state is not stable. Is there a better way to
do this? I want to check that after the sample been equilibrated at 0 K, the
pressure should be also 0, and the energy should be conserved during the NVE
run.

you cannot have dynamics at 0K. full stop.
that is a physics limitation not a software or simulation issue.

now, how you reach the absolute minimum energy conformation
and how validate that, is a tricky problem. you have far too many
degrees of freedom to make an exhaustive search, so you can only
approximate it. expecting a simple minimization to find the absolute
minimum right away is indeed optimistic, but the important question
is what do you need the absolute minimum energy structure for or
rather, how close do you need to be. sometimes just randomizing
positions and then doing some smart annealing followed by a thorough
minimization can get you close enough. sometimes, you have to
come up with something smarter. you'll only find that through a
careful study of the literature, though.

cheers,
    axel.

Hi All ,

I just couldnt stop to give a feedback to the following :

axel wrote:

you cannot have dynamics at 0K. full stop.
that is a physics limitation not a software

I wish to note that in the QM approach there is alaways movement (even at 0K) . I just wanted to give a clarification.
Anyways Axel is correct . In Classical MD the absolute zero is a Full Stop. Then again the sentence "that is a physics limitation "
Its debatable. I’m of the opinion its a sofware limitation.

Best Regards
OG

Hi All ,

I just couldnt stop to give a feedback to the following :

axel wrote:

you cannot have dynamics at 0K. full stop.
that is a physics limitation not a software

I wish to note that in the QM approach there is alaways movement (even at
0K) . I just wanted to give a clarification.
Anyways Axel is correct . In Classical MD the absolute zero is a Full Stop.
Then again the sentence "that is a physics limitation "
Its debatable. I'm of the opinion its a sofware limitation.

if you see it this way, then you have to say it is
a limitation of the model (classical mechanics).

you cannot blame lammps for not implementing
the most general description of particles (where
should we start? quarks?)... :wink:

cheers,
    axel.

Actually, I want to check if the implemented potential has the right pressure calculation. One way to check is that P=0 when T=0 for the relaxed sample. This is why I want to do it. Do you have better way to check if the pressure is right?

Usually, I equilibrating the system at 1 K and 0 Pa standing for 0K using NPT. Then I run a NVE to check if the system has reached the equilibration. I am confused that how exactly do you define the energy conservation. Do you mean that the total energy just stay one value or can also change a little, such as change from 44459.345 to 44459.351? When I equilibrating the sample at 1 K, I use NPT to set the temperature to be 300 K, however, When I just use
units metal
timestep 0.0001
fix 1 all npt temp 300.0 300.0 0.01 aniso 0.0 0.0 0.05
run 300000
unfix 1
fix 2 all nve
run 100000
The total energy changes a little, and the crystal structure destroyed, but if I use fix 1 all npt temp 1.0 300.0 0.01 aniso 0.0 0.0 0.05, the energy stay at one value and the crystal structure keeps. I don’t know if the energy or pressure has problems.

Lisa

2011/8/18 Axel Kohlmeyer <[email protected]>

Actually, I want to check if the implemented potential has the right
pressure calculation. One way to check is that P=0 when T=0 for the relaxed
sample. This is why I want to do it. Do you have better way to check if the
pressure is right?

just take two or more atoms, compute it manually and compare.

getting to p=0 is very difficult and doesn't say much
about the correctness of your potential.

axel.

I have checked the force and energy for two and three atoms. But for so few atoms, how can I check the pressure?

Another question is I am confused that how exactly do you define the energy conservation in NVE. Do you mean that the total energy just stay one value or can also change a little, such as change from 44459.345 to 44459.351?

Lisa

2011/8/18 Axel Kohlmeyer <[email protected]>

you said you have checked the force and energy, right?
Why don’t you calculate pressure from them?

Am … Ok …Here is my feedback :

assuming your using a LJ potential or a multibody potentiual e.g EAM , then you may want to look at the equation of state EOS ( Energy vs volume) If the system is at equilibrium, then the crystal should have “minimal volume (lattice parameter)” & "minimal Energy (Ecohesive). This should correspond also to the minimal pressure. You can not fix both the Volume, Pressure, at T=0 and at the same time be at the minimum of Energy. The calculation of a EOS is relative easy task when T = 0 . When T not equal zero , then things get messy, and a integrator must be use to evaluate the trayectory of the atoms. The atoms in this sence move in the (NVE) essembke then energy should be conserve, meaning E = constant , but it fluctuate a little bit from the equlibrium . You can practicatly plot E vs Time . You must understand the importance of the tine step. If your time step is to high then the integrator fails. Also you must be careful in knowing how must time the simulation need to equilibrate the system. etc etc etc etc … Many things came to play when running a MD sim.

OG

Er, I still have confustions. Because dU=TdS-PdV, if dU=0 and T=0, then we have P=0. Why it is impossible? When the system equilibrated and I run a NVE, the potential energy fluctuate, but the total energy should stay constant, I am not sure the total energy should stay exactly one value or can change a little? We should look into potential energy or total energy to check if the energy is conserved?

Lisa

2011/8/18 Oscar Guerrero <oguerrero@…2636…>

If you run minimize with fix box/relax you can equiliibrate
the system to (near) 0 pressure. And it will be at 0K by
definition. See the doc pages for the two commands.
What you can't guarantee is that the system will come to
rest at exactly zero. Esp since pressures in a solid are huge
and change dramatically for small displacements..

Steve

If you are at the minimun in the EOS , then P = 0 , E = Ecohesive and V = lattice parameter at T = 0 .
If you move out of the minimum then the volume change and therefore the pressure change. you can plot (E vs V) or in your
case (E vs P) . If you do dU = Tds - PdV and T = 0 and P=0 then The system doesn’t chane at all . You could try do this analyticaly by grabing a LJ potential . then plot IT . That;s All .

Thank you for your replies.

I used
fix 1 all box/relax aniso 0.0
minimize 1e-10 1e-10 1000 10000
for 10800 atoms, and the final result of pressure is 209 bar. This is of big difference with the 0 pressure. Does this within the reasonable pressure fluctuation from 0 pressure?

Lisa

2011/8/19 Oscar Guerrero <oguerrero@…2636…>

Thank you for your replies.

I used
fix 1 all box/relax aniso 0.0
minimize 1e-10 1e-10 1000 10000
for 10800 atoms, and the final result of pressure is 209 bar. This is
of big difference with the 0 pressure. Does this within the reasonable
pressure fluctuation from 0 pressure?

why don't you randomize your atom coordinates
a bit and re-minimize and see what happens.

you can also experiment with different minimizer
algorithms. some work better for certain systems
than others.

overall, it is not easy to get to the perfect minimum
in just one run of a normal minimization algorithm.

cheers,
   axel.

Ok, I see the problem. Thank you very much. I have so many things to learn, do you have any recommentation for studying the theory and basic examples of MD simulations?

Lisa

2011/8/19 Axel Kohlmeyer <[email protected]>

Ok, I see the problem. Thank you very much. I have so many things to learn,
do you have any recommentation for studying the theory and basic examples of
MD simulations?

my "classic" text book canon on general MD and related topics are:

- Frenkel / Smit, "understanding molecular simulations"
- McQuarrie, "Statistical Mechanics" (very "old school")
- Chandler, "Introduction to Modern Statistical Mechanics"
- Hansen, McDonald, "Theory of Simple Liquids"
- Allen / Tildesley, "computer simulations of liquids"
  (quite old - published in 1987 or so -, but many MD codes
  are based on the ideas in this book)

in addition, you may want to find a text book on simulations
specific to your area of research. i worked on water, electrolytes,
and electrified interfaces when i was a grad student, hence the
focus on liquids...

cheers,
    axel.