I know LAN2DZ is one of the standards for using heavier atoms. But it should take a while to run DFT with 100+ atoms.

As said Fernando, 100 atoms like Ge could be computational expensive for DFT. However, i think that Ge can be computed with DFT with a reasonable computational time

Are you working with Ge sheet like graphene? In this case you can work with periodic boundary conditions (PBC) and thus you could reduce the number of atoms.

It also depends of the package you are using. Gaussian?

You could start with a standar level: b3lyp/6-31g(d)

Maybe, siesta or DMol3? Level such as pbe/DZP or pbe/DNP could be correct (these programs are recomended for PBC calculations)

Besides, the accurancy and computational time is not only related with the basis set, it also depends on the functional (for the same system). Obviously, hybrid functional (e.g. b3lyp) is more expensive than pure functionals (pbe).

I

Thanks Fernando for suggesting LANL2DZ.

Thanks Gregorio for your answer. Actually I am performing cluster calculations and not PBE. Gaussian program seems a good choice for big molecule calculations.

All depens on your sytem and theoretical level, which could also be dependent on the selected program.

If you are working with cluster of moderated size, you can use more exigent functional and basis sets

Ge is a bad beast... I'm working on germanate, particularly on vibrational spectroscopy of either molecules and crystals. I found out a large systematic error in the vibrational frequencies of the Ge-O stretching modes that appears to be related to dispersion effects not accounted for at the DFT level, no matter what functional you employ. For Ge(OH)4 I had to work at least with MP2 to recover a decent structure and the Raman spectrum, and also I had to employ a relatively large basis set: the correlation consistent CCT (ACCT doesn't improve results; CCQ is as good as CCT; CCD or ACCD are just terrible...). Grimme parameter's can be optimized, to be used in B3LYP-D, to reproduce the MP2 results, but then they are very hardly transferable in other Ge-based systems like crystals containing germanium... That's a little bit frustrating... I got no problems like those, in silicates.     

TZ+(md,np) is a better option over the Pople basis sets, with m=1,2 or 3 and n=less than or equal to m -- depends on computational affordability. For heavier atoms, use effective core potentials such as LANL2DZ or better ones (I have no experience with ecps)

Dear Mauro Prencipe

Thanks for the details of Ge calculations. In fact I am also working on Ge molecules vibration (hydrogenated molecules). It seems that as the number of Ge atoms increases the errors become to rise in comparison with experiment. The DFT version that I use right now is PBE.

Dear Prashant Uday Manohar

Thanks for your suggestion. Ge is a heavy element that become to be very expensive computationally after 30 atoms so that I have to compromise between accuracy and execution time. I hope your suggestion works with the high number of Ge atoms.

What physical quantity do you want to calculate? If core potential and the effect of the electrons of lower shells (i.e., relativistic effects) are important for calculating the quantity, big basis such as LANL2DZ should be used. Otherwise you can use other basis sets. May be your should change the method (DFT, TDDFT,CI,...)

Best Regards

H.Simchi

Hi ,

I think there is no definite rule for this. From my experience  I  can say that LANL2DZ is good for heavier atoms but not always. When LANL2DZ fails, it's hard to find anything more  appropriate for heavier atoms.

Ironically, instead having an answer , I am presenting a question . What's the best source from where I can know what's the best BASIS SET  and HYBRID FUNCTIONAL for a particular chosen system consisting of a particular number of atoms and their atomic numbers ?

It really depends on the properties you are calculating and the level of accuracy you are expecting!

STO3G is minimal and it gives you a rough result.

631G is definitely better, but if you want more accuracy, you need to add polarization.

For heavy atoms, you do not want to do an all-electron calculation and want to get rid of the core electrons. You need to use a pseudo potential.

What code are you using?

I would recommend SIESTA. 

The solution of the present problem lies between good results that can be obtained using sophisticated basis sets at one side and the other side of acceptable computational time and sometimes convergence criteria that can be reached through smaller basis sets that discard core electrons.

Dear All

Please could some one reply me that actually what is basis set?

Regards

Dear Azam this definition is taken from the english wikipedia:

A basis set in theoretical and computational chemistry is a set of functions (called basis functions) which are combined in linear combinations (generally as part of a quantum chemical calculation) to create molecular orbitals. For convenience these functions are typically atomic orbitals centered on atoms, but can theoretically be any function; plane waves are frequently used in materials calculations.