I have been trying to do geometrical optimization for ZnSe Quantum Dots using Quantum ESPRESSO.

The convergence is very slow, so slow that even after 200 iterations I could not get convergence. I tried reducing mixing_beta to 0.3 like it says in the manual for QE(things got worse) and then I tried raising it to 0.8, this actually gave better results. Raising the cutoff energy also did not have much effect.

The weird thing is that once one decimal place is obtained it just keeps going a bit up and down not being able to converge the next digits easily. I think this could be solved by reducing the step size in the algorithm, is that somehow possible?

Convergence graph(energy) enter image description here

Input file

    calculation   = "relax"
    forc_conv_thr =  1.00000e-03
    max_seconds   =  1.72800e+05
    nstep         = 100
    outdir        = ".\outdir"
    prefix        = "me"
    pseudo_dir    = "C:\Users\aman\.burai\.pseudopot"

    a           =  2.50000e+01
    b           =  2.50000e+01
    c           =  2.50000e+01
    degauss     =  1.00000e-02
    ecutrho     =  2.49895e+02
    ecutwfc     =  2.75032e+01
    ibrav       = 8
    nat         = 44
    ntyp        = 2
    occupations = "fixed"
    smearing    = "gaussian"

    conv_thr         =  1.00000e-06
    electron_maxstep = 200
    mixing_beta      =  8.00000e-01
    startingpot      = "atomic"
    startingwfc      = "atomic+random"

    ion_dynamics = "bfgs"


K_POINTS {gamma}

Zn     65.39000  Zn.pbe-van.UPF
Se     78.96000  Se.pbe-n-rrkjus_psl.1.0.0.UPF

Zn      7.349204   5.983991  13.367822
Zn     11.402731  13.004906  13.367822
Zn      7.349204   5.983991   6.705404
Zn     11.402731  13.004906   6.705404
Zn     13.429495  11.834754  10.036613
Zn      9.375968   4.813839  10.036613
Zn      5.322440   9.494449  13.367822
Zn      9.375967   9.494448  13.367822
Zn      7.349204  13.004906  13.367822
Zn      5.322440   9.494449   6.705404
Zn      9.375967   9.494448   6.705404
Zn      7.349204  13.004906   6.705404
Zn     15.456259   8.324296  10.036613
Zn     11.402731   8.324296  10.036613
Zn     13.429494   4.813839  10.036613
Zn     13.429495   9.494449  13.367822
Zn     11.402731   5.983991  13.367822
Zn     13.429495   9.494449   6.705404
Zn     11.402731   5.983991   6.705404
Zn      5.322440  11.834754  10.036613
Zn      7.349204   8.324296  10.036613
Zn      9.375968  11.834754  10.036613
Se     13.429495  11.834754   5.868711
Se      9.375968   4.813839   5.868711
Se      7.349204   5.983991   9.199920
Se     11.402731  13.004906   9.199920
Se     13.429495  11.834754  12.531128
Se      9.375968   4.813839  12.531128
Se     11.402731   8.324296   5.868711
Se      9.375967   9.494448  15.862337
Se      5.322440   9.494449   9.199920
Se      9.375967   9.494448   9.199920
Se      7.349204  13.004906   9.199920
Se     15.456259   8.324296  12.531128
Se     11.402731   8.324296  12.531128
Se     13.429494   4.813839  12.531128
Se      7.349204   8.324296   5.868711
Se      9.375968  11.834754   5.868711
Se     15.456259   5.983991   9.199920
Se     13.429495   9.494449   9.199920
Se     11.402731   5.983991   9.199920
Se      5.322440  11.834754  12.531128
Se      7.349204   8.324296  12.531128
Se      9.375968  11.834754  12.531128

Any other advice on convergence is also very much optimized, Thanks! P.S the structure is neutral and I don't think there is anything wrong with the structure.

  • $\begingroup$ Just to clarify, you are now using the structure from your answer to mattermodeling.stackexchange.com/questions/6107/…? Has this undergone any geometry optimization steps or is it still just failing at the initial SCF? If its just the initial step, its likely there is still an issue with the geometry. Prior to the advice about mixing_beta, the QE manual makes the same point about slow SCF convergence: "In most cases: your input data is bad, or else your system is metallic and you are treating it as an insulator." $\endgroup$
    – Tyberius
    Jun 23 at 17:45
  • $\begingroup$ So I guess my advice would be to perturb the geometry a couple different ways and see if it can at least complete a single SCF cycle. Also, I don't know QE very well, but I don't see anything in the input about the spin of your system. If it is automatically picking a strange value for the spin, that may also cause issues. $\endgroup$
    – Tyberius
    Jun 23 at 17:48
  • $\begingroup$ @Tyberius I am not actually running calculations on that one , but it faced a similar issue of slow convergence, like it converged after 130 iterations the first time and with each step(in the geometrical optimization) it increased. But the thing is I have tried a lot of structures which do not converge. There was only I structure which sort of converged in a few iterations and I do not know what was special about it(maybe I did something by chance) and can you please elaborate about changing the geometry a couple of different ways . P.S looking into the spin thing. Thanks.. $\endgroup$ Jun 24 at 19:45
  • $\begingroup$ @Tyberius and as you can see the calculation is converging of sort as in not going up or down continuously, so can I say my structure is correct?? And there is only a minor mistake in parameters somewhere $\endgroup$ Jun 25 at 7:45

My suggestion is that you first get an SCF calculation to converge for this structure before you try relaxing, especially if you aren't sure what you're going to get. Then you can proceed to converging the numerical settings like the cutoff energies for your properties of interest.

I was able to get an SCF calculation to converge using your input file by switching the occupations = "smearing" instead of "fixed". It seems like this may have been what you intended based on your choices for "smearing" and "degauss", which seem to be ignored if you have occupations = "fixed".

  • $\begingroup$ You seem to be correct. But can you please elaborate a little about why is your answer correct? As far as I know occupations = "smearing" is used for metals and ZnSe is a semiconductor for which we use occupations = "fixed". Actually, I am supposed to be researching properties of semicundocoting quantum dots so that is why I did not switch from fixed to smearing. Thanks for your answer $\endgroup$ Jun 28 at 10:20
  • $\begingroup$ Are you sure that the structure you created and that DFT “sees” is a semiconductor? DFT is known to underestimate band gaps. Gaussian smearing usually improves convergence numerically and it’s safe to use it regardless of whether the system is metallic or not. In the limit of infinitely narrow (zero) smearing width, the result should be correct. $\endgroup$ Jun 28 at 11:18
  • $\begingroup$ could you please explain once a bit about what actually is the difference between "smearing" and "fixed". I have studied the difference but I am still a bit confused. And really really thanks for your reply I have been stuck for the past couple of days by using occupations = "fixed" $\endgroup$ Jun 28 at 21:34
  • $\begingroup$ This is probably better suited for another question. For example, this one for smearing. For fixed occupations, some QE documentation says: "occupations="fixed" occupies the lowest (N electrons)/2 states". $\endgroup$ Jun 28 at 23:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.