search

shankar1729 / jdftx

JDFTx: software for joint density functional theory
http://jdftx.org
84 stars 54 forks source link

Lattice optimization #94

Closed khyeoh closed 5 years ago

khyeoh commented 5 years ago

Dear Dr. Shankar,

I tried to carry out lattice optimization on a 2D system and ended up with the following:

LatticeMinimize: Step failed along negative gradient direction. LatticeMinimize: Probably at roundoff error limit. (Stopping)

I grep the Lattice Minimize in the log file which gave us the following info:

LatticeMinimize: Iter: 0 F: -821.729011961387414 |grad|_K: 2.968e-02 t[s]: 3815.51 LatticeMinimize: Wolfe criterion not satisfied: alpha: 1 (E-E0)/|gdotd0|: 0.0871758 gdotd/gdotd0: -1.28844 (taking cubic step) LatticeMinimize: Wolfe criterion not satisfied: alpha: 0.474271 (E-E0)/|gdotd0|: 0.0386057 gdotd/gdotd0: -1.03225 (taking cubic step) LatticeMinimize: Wolfe criterion not satisfied: alpha: 0.210804 (E-E0)/|gdotd0|: 0.0157544 gdotd/gdotd0: 0.0611815 (taking cubic step) LatticeMinimize: Wolfe criterion not satisfied: alpha: 0.0594357 (E-E0)/|gdotd0|: 0.00500995 gdotd/gdotd0: -0.0660324 (taking cubic step) LatticeMinimize: Undoing step. LatticeMinimize: Step failed along negative gradient direction. LatticeMinimize: Probably at roundoff error limit. (Stopping) LatticeMinimize: Iter: 0 F: -821.729011961360470 |grad|_K: 2.968e-02 t[s]: 3821.14 LatticeMinimize: Wolfe criterion not satisfied: alpha: 1 (E-E0)/|gdotd0|: 0.0871753 gdotd/gdotd0: -1.3203 (taking cubic step) LatticeMinimize: Wolfe criterion not satisfied: alpha: 0.478064 (E-E0)/|gdotd0|: 0.0357162 gdotd/gdotd0: -0.52913 (taking cubic step) LatticeMinimize: Wolfe criterion not satisfied: alpha: 0.176991 (E-E0)/|gdotd0|: 0.0154759 gdotd/gdotd0: 0.396534 (taking cubic step) LatticeMinimize: Wolfe criterion not satisfied: alpha: 0.0422058 (E-E0)/|gdotd0|: 0.00282226 gdotd/gdotd0: 0.240041 (taking cubic step) LatticeMinimize: Undoing step. LatticeMinimize: Step failed along negative gradient direction.

May I know how to overcome this issue?

Thanks Yeoh

shankar1729 commented 5 years ago

Check the convergence of the electronic minimize / SCF: this sort of issue can crop up particularly if the electronic iterations are not converging.

If that looks fine, try adjusting the dirUpdateScheme in command lattice-minimize to FletcherReeves or PolarkRibiere (see http://jdftx.org/CommandLatticeMinimize.html)

Best, Shankar

khyeoh commented 5 years ago

Dear Dr Shankar,

Correct me if I am wrong, the electronic minimize SCF seems converge. (Please see attached files) I try both scheme but it did not work. I attached the output file here for your reference.

Outputfile.zip

Hope you can help.

Thanks! Yeoh

shankar1729 commented 5 years ago

Indeed, the electronic minimization seems fine. The other point might just be that your lattice already happens to be quite close to converged. I'm not sure exactly what the issue is, but you can try a few things:

  1. Switch to the primitive cell of the surface for lattice optimization if that's an option. This will reduce internal degrees of freedom and overall make your calculation much faster anyway.

  2. Ionically minimize the system more throughly before starting a lattice minimize.

If all this fails, simply do calculations at a few different strains and fit for the equilibrium. We've had to do this in some problematic cases.

Best, Shankar

khyeoh commented 5 years ago

Dear Dr. Shankar,

Previously, the input coordinate that I used are those already relaxed using Quantum Espresso. I re-do the calculations, by using the un-optimized input coordinate and it works. I am using the FletcherReeves scheme.

Thanks!