Reason for sudden jump in stress optimization

[AsifShah]

Dear All,
I am running optimization for a metal-semiconductor structure. I ran it earlier for 24 hours then it got interrupted & I took last minus 2 image from trajectory & rerun the calculation. However, I observed the following sudden increase in stress as shown in image below.
Also the simulation is converging very slowly.

Kindly provide any reason for why it is converging slowly? & why this sudden increase in stress?

[Anders Blom]

Sorry, one cannot explain that by just looking at the log. I suggest you view the trajectory to see if the structure changed suddenly. Also note that optimizations can be slow for many reasons, incl. starting far from the minimum, multiple or very flat minima that you keep hopping around. If the structure is simple enough it might make more sense to scan the lattice parameters first, to get a sense of where the minimum is, and if it's a device you can approach it stepwise with rigid regions, but again without knowing more about your system, one can only guess.

[AsifShah]

Can I get your email where I can send further details?

Also, Could you explain what you mean by "scan lattice parameters..."?

Thanks

[Anders Blom]

Compute total energy as function of lattice constants to find the minimum. We do no have email support, sorry.

[AsifShah]

Here is an attached pic that shows sudden jumps in AIMD calculations as well. What could be possible reasons? This happened in many simulations?

[Anders Blom]

Hard to tell from just a picture but my gut feeling is that some of those SCF calculations do not converge.

[AsifShah]

Thankyou Anders Blom

[Anders Blom]

The good thing is that it seems it can recover, so the steps after converge ok and the calculation seems to proceed OK. However, looking into the convergence might speed it up quite a bit, so I would take out the geometry of one of those "weird" steps, run it separately, and see if you can make it converge better/faster by tuning some parameters; maybe you need more k-points, higher temperature, different mixing. Even if more k-points is slower nominally, if it uses considerably fewer SCF steps it's faster overall anyway (and more accurate).

[AsifShah]

Yes I tried increasing K point also. On the positive it is converging towards tolerance value however, the speed of convergence is very very slow.

Apart from damping factor are there any other crucial parameters that can help increase speed of convergence?
Also, talking of increase in temperature I read from QATK manual that increasing it too much can give incorrect results. (I use 1000K)

[Anders Blom]

Temperature, depends on whether it's a metal or semiconductor. 1000 K is high for insulators but ok for metals.
Other convergence tricks are usually problem-specific so I cannot say without knowing the system. Vacuum? Magnetic?

[AsifShah]

The temperature I am talking of is the broadening temperature (1000K) & the system(Semiconductor) is in the vacuum.
Also, the calculator is NPT- Maryna Tobias Klein.

Also, if i run simulation for say 1500fs I see "grep -rni converge file.log | wc" gives me 1502. Should I be worried about these sudden jumps in temperature or just consider it some glitch due to numerical error as output file generates no warning or error sign?

Thanks

[Anders Blom]

You could grep for "not converge" or "warning" instead to check. Yeah unless your energies or MD temperature go crazy, I would not worry much if a single calculation here and there don't converge or behave strange.

[AsifShah]

Dear Anders Blom,
Yes, I did grep -rni Warning/not converge but found no results which show I should not then worry about a single calculation jump as you said.

Also, I would like to ask, about restarting an MD simulation due to power shutdown. I use the following method:
1. I extract the last minus 1 image from the trajectory, send it to the script generator then use same calculators & MD block with same parameters as in the original script that was interrupted.
2. Further I send it to the editor where I replace the default initial velocity using Boltzmann (300K) by initial_velocity = Configuration_velocity(remove_center_of_mass=True)

I have attached the file below also.

My concern is whether I should keep equilibration time for pressure & temperature again in this restart or not as the trajectory had already reached equilibration (temperature, pressure) in the previous run. Is there any better method to restart the MD simulation?
Kindly help in this regard.
Thanking you in anticipation.

[Anders Blom]

MD is very short-sighted, it basically only knows what happened in the previous step. So if you read positions and velocities from the last step, it's really  the same as if the previous run had continued from that point, and there will be no need for a new equilibration.

[AsifShah]

Thank you for all the help.