Dear Trinh, considering that the bulk modulus of silicon is about 1 Mbar (1000 Kbar) a convergence error in the calculated pressure of about 3 Kbar implies an error in the volume of about 0.3 % which mean 0.1 % for the lattice parameter which is what you have and is usually considered a good accuracy. Other factors like XC functional and fitting procedure give larger variations If you want a convergence accuracy better that 0.1 % you need to converge the pressure below a fraction of Kbar.
stefano PS: as for the difference between the stress calculated by vc-relax at the end of the relaxation and an scf calculation done at the converged structure this is due to the fact that during vc-relax the calculation is performed at fixed number of plane waves so that the two calculations are not done with the same basis. vtmtrinh at caltech.edu wrote: > Dear Stefano, > > I have read the link you show. Thank you very much. I actually used high > Ecut (35Ry). The results I sent in the previous email are for Ecut-35Ry. > Let me check everything again to see if I made any mistake. > > Best, > > Trinh > > >> Dear Trinh Vo, >> this has been explained a number of times in this forum. >> The way stress is calculated assumes that the number on plane >> waves does not change during the deformation. >> This is NOT what you are doing when calculating energy vs alat >> curves at fixed CUTOFF ... which explains the difference between the two >> results. >> Stress converges slowly with cutoff. To be reliable calculations >> involving stress should use higher cutoff than usual. >> >> Have a look at the attached slides that try to clarify the issue. >> These were presented in a Summer School in Santa Barbara in 2009 >> whose material can be accessed at >> http://media.quantum-espresso.org/santa_barbara_2009_07/index.php >> >> best >> stefano >> >> vtmtrinh at caltech.edu wrote: >> >>> Dear All, >>> >>> I am running a test calculation on a cell with 64 Si atoms. I first run >>> a >>> series of scf calculations of different cell sizes (the atoms position >>> coordinates are also generated accordingly with cell dimensions by my >>> code). I got the following results: >>> >>> a (A) Total E(Ry) >>> 10.775 -508.1653034 >>> 10.770 -508.1653391 >>> 10.76932 -508.1653368 >>> 10.765 -508.1653067 >>> 10.76 -508.1652012 >>> >>> From the above results, it looks like to me I would have a min E close >>> to/at a=10.77A (plotting E vs a give me a nice parabola curve). I then >>> rerun the structure at min E (i.e. a=10.77) using vc-relax. I obtained >>> the following: >>> >>> CELL_PARAMETERS (alat) >>> 1.000000000 0.000000000 0.000000000 >>> 0.000000000 1.000000000 0.000000000 >>> 0.000000000 0.000000000 1.000000000 >>> >>> >>> alat = 20.3289 au (or 10.75760138 A), E=-508.1650690 Ry >>> >>> Comparing to the results by scf calculations, I see the result from >>> vc-relax and scf calculations are not consistent at all. I thought they >>> should. However, the total stress computed for the scf calculation at >>> min >>> energy is still large (P = -3.36 kbar), while the stress calculated by >>> the >>> vc relax is small P = -0.04 kbar. I also noted that the cell size at >>> which the system has a min stress (i.e. almost zero stress) given by scf >>> calculations does not coincide with the cell size at which the the >>> system >>> has a minimum energy (min of the parabola). I thought that at the >>> optimized cell size, both stress and energy have to be minimum. I don't >>> understand why there is such a difference. Am I wrong in the way I was >>> doing? >>> >>> Thank you very much, >>> >>> Trinh Vo >>> JPL/Caltech >>> >>> >>> >>> >>> >>> _______________________________________________ >>> Pw_forum mailing list >>> Pw_forum at pwscf.org >>> http://www.democritos.it/mailman/listinfo/pw_forum >>> >>> >> _______________________________________________ >> Pw_forum mailing list >> Pw_forum at pwscf.org >> http://www.democritos.it/mailman/listinfo/pw_forum >> >> > > > _______________________________________________ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum >
