cam-B3LYP with pobTZVP SCF convergence
-
Thank you for the forum and responses. I have a typical problem for a rather simple mixed ionic/covalent system comprised of Na, Si and O only. Simple system but SCF convergence is not. Of note, I really like the presence of built-in basis sets and use them exclusively. I am running EOS (and not only) with cam-B3LYP/potTZVPrev2 built-in basis set. SCF starts converging to some 10-6 and then it starts oscillating. It is very typical of my medium-level CRYSTAL user attempts and expertise. it is an insulating state, with a bandgap of 10 eV, so no funny conductive state convergence problems. None of my attempts to use LEVSHIFT, NODIIS and FMIXING make it to converge to TOLDEE of 7 (or in EOS case to 8). Raising LEVSHIFT just leads to conducting state eventually and calculation does not converge. I understand that this is a triple zeta basis set but since it is built in and presumably balanced and robust, I was hoping it would lead to convergence somehow.
-
I should say I understand it is basis set problem. Part of the problem I always found difficult to work with CRYSTAL (untill these built in basis sets appeared) that one had to tinker with basis sets so much. So I tried for the input above built in pob-DZVPP basis set and did not stand a chance, calculations right away entered conducting state and did not converge. Should I still tinker with the built in basis set to remove diffuse functions? To what extent?
BASISSET POB-DZVPP DFT cam-B3LYP XLGRID END -
One thing worth trying before changing the basis set is increasing the
TOLINTEGkeyword values. This can be particularly important when dealing with hybrid functionals, especially CAM-B3LYP, as it has a high fraction of EXX at long range. Suggested values to start with could be8 8 8 15 30. You can further increase these values if needed to improve SCF convergence, but make sure that the fifth threshold is at least double the fourth one.If you want, you could also share the output files so I can take a closer look.
-
Thank you, this one converged the SCF
-
Hi,
Find below a modified input file for which the SCF converges in 25 iterations:
Na2Si2O5, Rakic Physics and Chemistry of Minerals 29 (2002) 477-484 - Ale mod CRYSTAL 0 0 0 14 4.725582 23.296569 7.936255 90.15 18 11 .2299 .26138 .5247 11 .2743 .51408 .3548 11 .2534 .47281 -.1019 11 .7548 .28108 .2783 14 .2961 .36325 .1966 14 .6843 .34077 .6429 14 .1831 .40821 .5373 14 .7948 .38941 -.0185 8 .1210 .3844 .0352 8 .2339 .3015 .2523 8 .6209 .3690 .1429 8 .2440 .4105 .3392 8 .7431 .3414 -.1590 8 .7456 .2819 .5655 8 .3573 .3561 .6162 8 -.1423 .3920 .5597 8 .2452 .4667 .6166 8 .7336 .4511 -.0767 EOS RANGE 0.95 1.05 8 PREOPTGEOM MAXCYCLE 500 END BASISSET POB-TZVP-REV2 DFT cam-B3LYP XLGRID END TOLINTEG 10 10 10 10 20 SHRINK 6 6 BIPOSIZE 11202400 EXCHSIZE 11202400 MAXCYCLE 200 ENDWe have increased the values of TOLINTEG, used an isotropic shrinking factor, and re-activated the DIIS accelerator.
-
Rectified with 8 8 8 15 30. I have a related question - what is the rationale of using isotropic shrinking factor in the lattice that clearly calls for asymmetric one? 4.725582 23.296569 7.936255 can't be having the same shrinking factor simply since lattice parameters are so dissimilar. In fact, I see that all the time in CRYSTAL - isotropic factor. I converged the energy of the system and isotropic factor was not the one that lead to the overall energy convergence.
-
job314 Good point! About the shrinking factor: the anisotropic shrinking factor in CRYSTAL does not work properly for those calculations where the symmetry of the system may change (for instance in frequency calculations, FREQCALC, where displaced nuclear configurations are explored, or elastic calculations, ELASTCON, where the lattice is strained, etc.). So in general, I personally tend to avoid using an anisotropic shrinking factor.
However, for symmetry-preserving calculations (such as SCF, OPTGEOM, EOS) the use of an anisotropic shrinking factor should be fine.
