Differences between electron density calculating using DZP and SZP

[wot19920302]

Dear Quantumwise staffs:
        I calculated spin density of one system using SZP and DZP (ATK ver.2015) and found the difference exists in left and right edges. Fig1 is spin density using SZP, Fig2 using DZP. I can't understand what leads to such results. Could you please help me?
       Best
       Attachments are scripts. script1 uses SZP  and script2 uses DZP.
       

[lknife]

The only thing I know is that DZP is more accurate than SZP. And, as it is said in the tutorial, OMX basis set is more accurate than FHI (for ATK2015). Maybe you can refer to this technical note: http://docs.quantumwise.com/technicalnotes/atk_pseudo_basis/atk_pseudo_basis.html, where you can get some ideas about which basis set is appropriate for you system.

[wot19920302]

Thanks for your help, sir . I have browsed the link you provided and noticed FHI DZP (Fig.1)  can properly describe C and H elements.  And there are many papers investigating graphene nanoribbon even using SZP.  In fact, spin density only differs in left and right edges. Others sides are the same.  It confuses me very much.

[Ulrik G. Vej-Hansen]

It seems that you use the same geometry in both cases, so it could be because one of the systems is effectively strained.

[lknife]

Sorry I failed to notice that you are dealing with graphene. In that case, FHI-DZP or FHI-SZP is enough. 

[wot19920302]

You are right, sir, before perform NEGF,  I have used DZP to optimize central region at first.  the script attached below is optimization script. Is the script wrong?
best

[wot19920302]

anyone can help me，please？

[Jess Wellendorff]

I am in principle not surprised that FHI-SZP calculations can give different results than with FHI-DZP. Sometimes the SZP basis set is just not large enough. I would check using a 3rd combination of pseudopotential+basis set, for example HGH or OMX.

[Anders Blom]

The differences or rather the results for DZP do not look nice from a physical standpoint. You may need to try a few things, like more k-points in the C direction, or making the device longer. Strain should not matter since the electrode is strained by the same amount as the layers closest to it, and it's not a matter of accuracy - the "electrode copy" should resemble the electrode, and I can't imagine the electrode has that spin density in the middle. So it's more a matter of symmetry/consistency than anything else.

That said, the spin state where all atoms are equal is not the ground state, so maybe it's also an effect of the isovalue being very small? Try the more common spin setup with opposite spins on the two edges, maybe there the difference is not a problem.

[wot19920302]

Thanks for your guide！ I have solved the problem by extending chentral region as you said.