Dear Maximo,

The Number of ionic relaxation steps (NSW) should be increased if your convergence criteria are satisfied. Check for maximum force on each atom and the rms value. The maximum force should be small (in the order of~0.00 eV/Angst). I normally use NSW=200 and often restart the calculation until force is small enough. If your system is too large then you can reduce KPOINTS to achieve faster convergence. If required, you can restart calculation with higher points after convergence.      

You can change the  convergence criteria using EDIFFG and EDIFF tag in vasp.

The stoping of VASP after given NSW (ionic steps) does not mean that your calculation reached the convergence. 

You should also look carefully at your electronic structure to validate the correct ground state with the lowest energy (or experimental) magnetic ordering, especially in case of TMOs.  

Let me know if you need further info.

Thanks,

Dear Mudit,

I want to ask here if we are not getting convergence till NSW supplied in INCAR  then we should rerun the calculation after CONTCAR to POSCAR.

Am I right ? Then we should read the WAVECAR (ISTART=1) or no need to read WAVECAR. When we should read WAVECAR and what thing affected when we read WAVECAR file.

Thank you

Shilendra

Dear Shailendra,

Yes, you are correct that one should restart the calculation until you reach the convergence. About the way of restarting the calculation, the calculation will be faster if you use wavecar (mv CONTCAR POSCAR, ISTART=1), but for some reason,  if my previous calculation was diverging (or showing fairly wrong magnetic moments) then I would not use the old wavecar.   There is nothing wrong in restarting the calculation from scratch (say for crashed calculations).

Thanks Mudit,

NSW = 200 is a big number, but suppose that we choose this value for our calculation and after the calculation end we don't get reached required accuracy message if I understand your answer to Shailendra's question we can rerun from CONTCAR and WAVECAR, even the reached accuracy haven't obtained, yet, is this correct?

If this is correct, can I use less NSW value and balance doing more rerun?

We need to do rerun unto we get reached required accuracy message, or we should be get his message in first run?

Thanks again.

Setting up NSW with big or small values doesn't make any difference. You can use NSW=200 (or even more) safely. If your calculation still doesn't converge in 200 iterations means that your starting geometry is far away from the final geometry. Hence, you have to re-run from CONTCAR.

Thanks for the clarification.

Dear Maximo, 

As Jorge said NSW is the number of maximum ionic steps will be performed during relaxation. If your calculation converges before these many steps, vasp will stop. Please find the attached script for printing out the maximum force of each step. "./ouscar.sh -f " will print the force  of each iteration to give you the idea of convergence.   

Hi Maximo,

an alternative, iterative, way, if you are worried about a too large value of NSW (you should not in any case, NSW value as stated is "for free". I usually adopt values between 500 and 2000 in order to get a "safe" convergence...) is to start from very loose values of the convergence parameters in your INCAR (say, EDIFF, EDIFFG and so on): you should usually get the first converged structure in very few steps (if the initial POSCAR reasonably describes your system).

Then reduce the convergence parameters progressively, run after run, as stated by copying CONTCAR in POSCAR and setting ISTART=1 in the INCAR. Also, start from gamma point and increase the mesh of the k-points in the KPOINTS file. Still you will get convergence in a reasonably short time. 

Sometimes indeed specifying very tight parameters from scratch in the INCAR makes the calculations a lot more time consuming and in this case you risk you have to specify very large values of NSW.

I hope it helps

cheers

Thanks Mudit again for the script.

Thanks to Giacomo, you suggest Gamm point after first run?

I ever  use Monkhorst-pack. Have you test performance using  Monkhorst-pack and Gamma? If your answer is yes. If your answer is yes, Do you get less time with the same accuracy?

I use default values of EDIFF, EDIFFG.

Dear Giacomo Giorgi thank you for nice answer,

Here I want to clear one thing generally I give EDIFF=10e-5 and EDIFFG=-0.01. SO for supercell contains 128 atoms my system does not converge till 200 or sometimes 500 NSW. Then I do CONTCAR to POSCAR and rerun calculation without (ISTART=1) then it converge upto 100 NSW.

I want to ask two thing here first my way of optimizing is right and what happens when we read WAVECAR (ISTART=1) after CONTCAR to POSCAR and run calculation. What will happen when we do CONTCAR to POSCAR and rerun calculation without ISTART=1.

Second as you told first give loose convergence criteria how much loose like EDIFF=10e-3 and EDIFFG=-0.5 etc and first run calculation after convergence we have to do CONTCAR to POSCAR and change criteria EDIFF=10e-5 and EDIFFG=-0.01 and rerun the calculation. Am I right ?

Would second way as you told take less time comparable to my way as I mentioned . Which way of optimizing structure is suitable reasonable.

Shilendra

Dear Mudit

How will run outcar.sh script, I tried but could not do.

Shilendra

Dear Shailendra,

Please make it executable (chmod 777) and then execute as said in my last comment. 

Thanks,

Thank You Dear Mudit,

I got result as-

Maximum FORCE is : 0.009744  ATOM=88 x

Is it right ? What is unit here ? I gave EDIFF=10e-5 and EDIFFG=-0.01

Shilendra

Dear Shailendra, 

Your forces are converged. Now you may look at the electronic structure.

Thanks, 

Dear Mudit,

I get the message reached required accuracy - stopping structural energy minimisation in the first run, and when I run ./outcar.sh -f, I get

                                Maximum FORCE is : 0.023352 ATOM=8 z

I used the the default values EDIFF = 0.1E-03 and EDIFFG = 0.1E-02

Do I need run again?

Maximum FORCE refer to EDIFFG value?

Thanks!

Hi,

I have seen two different EDIFFG values (-0.01 and 0.1E-02) from the above two posts for atomic relaxation. May I know how will the sign (+,-) affect the atomic relaxation and which value is more reliable for atomic relaxation (-0.01 or 0.001)? Often I have seen 0.01 ev/Angstrom in the papers. Could any one elaborate this point? 

Hi Muhammad,

Based from the VASP manual (see link below), my understanding is this.

• Positive EDIFFG - stops relaxation if the difference in free energy between last two ionic steps is less than the EDIFFG value set.
• Negative EDIFFG - stops relaxation when all  forces are less than EDIFFG value set.

In papers, I often see this phrase "The ions are allowed to relax until all atomic forces are less than 0.01eV/Angstrom." In this case, I am assuming that this meant that the settings use is EDIFFG = -0.01.

In my case, I also use negative EDIFFG. In fact, the manual said that this is really the more convenient setting, rather than the positive EDIFFG setting.

http://cms.mpi.univie.ac.at/vasp/guide/node107.html

Thank you for making it clear @ Shaira Narido.

Dear all,

All I understand is when we got  reached required accuracy - stopping structural energy minimisation in the OUTCAR, it means our calculation is converged.

But I still don't understand how Mudit's script define the convergence. I run it to my converged calculation (NSW = 300, EDIFFG = -0.02, number of atom= 13) it gave me:

Maximum FORCE is : 0.016860 ATOM=9 z

And I run it to my uncoverged calculation (NSW = 300, EDIFFG = -0.02, number of atom= 36), and it gave me:

Maximum FORCE is : 0.053219 ATOM=21 y

Please someone explain it :)

Hi, Rizalina Tama Saragi 

As I am getting, you gave EDIFFG=0.02 so in your converged result Mudit's script is showing FORCE is : 0.016860 which is less than EDIFFG value but in your unconverged case it is showing FORCE is : 0.053219 which is more than EDIFFG value.

So I think your point is that what is the need of this script because in converged case it shows forces less than EDIFFG value when we get message reached required accuracy - stopping structural energy minimisation.

I am not sure sometimes it happens when we get massage reached required accuracy - stopping structural energy minimisation but our force is more than EDIFFG value. If  so than the importance of this script is only to know exact converged force on any system. Am I right or wrong, more suggestions needed ?

Shilendra

Thanks Shilendra for your answer.

I rerun my uncinverged calculation, by copying CONTCAR to POSCAR, and finally the calculation is converged. but the final geometry is far away from what I expected to be. What should I do?

I also found, my other calculation (has more atoms) which  is converged after 66 steps, but once again the final geometry has completely different "shape". What do you think happened here?

Hi Rizalina,

I think you should try different optimization method like constant volume (ISIF=4) or varying volume (ISIF=3) etc. If you are doing surface calculations then ISMEAR=2 is suitable.

Shilendra

Hi Shilendra, thanks for your reply

It turned out, that the center of mass of my system was different, so it looked like different "shape" but the geometry actually was fine (I'm using jmol). So I didn't put the keyword you suggested. But thank you for the responses.

Actually, I was telling about VASP software. I thought you are using VASP software.

Shilendra

Oh of course, I am using vasp for the calculation. I mean I'm using jmol for the visualization -_-'

Haha, thanks anyway

Hi Mudit Dixit,

How can i activate the script (attached by you) on my system to print the force for each iteration? As i have no concept of python language. So i am facing difficulty to activate the script attached by you.

Thanks alot

Copy the script at your $home. Change the permission to execute: chmod 777 ~/out car.sh. Then run the script from the run directory : ~/outcar.sh -f OUTCAR

Hello Mudit,

I am a little bit confused by your script. My EDIFFG is -0.01. However, when I run the ionic relaxation it ends after NSW steps but when running your script it shows me that the Maximum FORCE is : 0.000767 ATOM=7 x. Hence, it should have finished based on the fact that this is below the required minimal force. Looking at all the individual forces in the OUTCAR file shows me several forces that are higher than that value which would explain why it has not converged yet.

Thank you for your help!

I guess one should start with a lower NSW value. I would suggest using ISIF2 and make sure that you are using the lowest possible k-point mesh for your system. It would cost a lower computation time. After the completion, it can be started again and again doing copy CONTCAR to POSCAR until it gives the message ' reached required accuracy - stopping structural energy minimization'.

Regardless of ISIF, if you are doing structural optimization in VASP, you have to get the line "reached required accuracy - stopping structural energy minimization" in the OUTCAR. So, what I suggest is just do trial and error with different values of NSW. Because this value can be different for different materials as well as the type of calculations. Also, it doesn't matter with higher values of NSW in the calculation, but it should be sufficient also. So, you can try with 100 and keep increase until the structure is not fully relaxed.