I think OpenMX is one of the best packages for the DFT calculations.  follow the link:  www.openmx-square.org/   

Stick to GGA and avoid LDA. LDA gives the wrong ground-state of iron. Also more modern functionals suffer this dilemma, PBSol comes to mind and perhaps AM05.

Basis: For metal-metal systems, codes that use a plane-wave basis are fine. In molecular crystals, codes with localised orbital basis states (the one I use is called FPLO) are more reliable.

Functional: I would also tend to prefer GGA over LDA for this kind of system.

As Anthony mentioned, plane wave basis set is best for periodically repeating solids. For finite (molecular) systems, localized basis sets are better. You do have to choose among a plethora of these localized basis sets.

I think meta-GGA's would be worth to look at. The performance has been benchmarked for 3d transition metals here: http://dx.doi.org/10.1063/1.2162161

If you're interested in molecular properties, my suggestion is to use the mPW1PW functional, that generally provides correct geometric parameters. If you're interested in spectroscopic absorption/emission properties, the PBE0 functional could be a good choice. On the contrary, the B3LYP functional, largely used in the literature, often overestimates metal-metal bond distances.

As for BSs, it depends again on your aims. A computational cost effective starting choice could be using double-z Ahlrichs bs plus polarization functions for all atomic species. If heavier metal species are involved instead, the use of ECP is recommended. 

my suggestion is to use WIEN2k code( http://www.wien2k.at/reg_user/textbooks/) based on FPLAPW (all electrons methode can be more accurate but need a huge memory and calculation time) and for your system GGA better than LDA.

Fe in wien2k you can read cottenier introduction: http://www.wien2k.at/reg_user/textbooks/DFT_and_LAPW-2_cottenier.pdf

It would seem from you question that you wish to perform some kind of molecular calculation. There is, unfortunately, no correct answer here, It depends on what properties you are interested in. GGA will almost certainly improve on LDA structural results, but GGA can perform quite poorly when it comes to stabilising the correct spin-state. You may find that you need a hybrid exchange-correlation functional with a significant proportion of Hartree-Fock exchange (e.g. B3LYP, PBE0, even BHLYP) in order to stabilise the correct ground state of your system. This type of calculation is only really practical with an atom-based basis-set (although some periodic codes do allow you to use hybrid functionals). However, while hybrid functionals may give you reasonable geometries, they may not describe e.g. excited states very well, where you may be better off employing a range-separated functional (e.g CAM-B3LYP). It really does depend on what you properties you need to evaluate. 

Regarding basis set, this will depend on the size of the system, but ideally you want to use at least a polarised triple-zeta basis set. The def-TZVP basis sets available in TURBOMOLE and from the basis set exchange (https://bse.pnl.gov/bse/portal) are relatively inexpensive sets of this type.

Finally, if your systems are small enough and you want detailed information about the electronic structure of complexes containing open-shell transition metallic centres, you should think about wavefunction based methods, such as CASSCF.

I think, you need to use a hybrid DFT methods such as B3LYP and B3P86 with suitable basis sets such as pseudo potentials for Fe like LANL2TZ+ or defuse function Lan2tz-f if you are looking to optimize the structural parameters. If you wank to calculate the thermodynamics and binding energies then go to use pure DFT methods with pseudo potential for Fe

This recent paper (2013) may be informative; their study was on Mn though:  http://pubs.acs.org/doi/abs/10.1021/ct4002398

well i suggest if you are using gaussian03 with dft try b3lyp with lanl2dz or b3lyp with lanl2mb they work good for metal based systems and even metal doped organic systems where metals belong to 3d series

GGA (PBE) is feasible and included in all codes.

I support the question. In the DFT we can get very strange results with metals.

If you are particularly interested in transition metal-metal complexes, you can go for B3LYP functional (with some decent basis set) as it gives quite reliable geometries that has transition metals.

I'm doing research by the semi-empirical method PM3 (with the accuracy 0.0001 kcal/mol) with HyperChem 7.0 software (Hypercube, Inc) (RHF function, Polak-Ribiere alghoritm), the accuracy of energy measurements was 0.001 kcal/mol.

Regards

Hello Daria, the B3LYP functional is the first which many people think, but in the case of transition molecules i think that M06 fuctional is the best, it is parametrized with metals and it presents good results. If you want publish, you could use the 6-31G(d) basis set for the organic part and LANL2DZ for the metals. The above was studied by J. Baldenebro-López and you can search his articles.

Regards

 Hi Daria,

The best overall performance is observed for WB97XD/6-31G(d), which offers relatively small statistical errors when considering the overall structure as well as selected distances.

MO6-L is particularly suitable for transition metals.

Try a bunch of these ones suggested, see which ones give you the best results. Try to sample one or two from each type if you have time. Trial and error is the only way, all DFT functionals are wrong anyway.

In my experience B3LYP is the best choice, but M06 shows good results too.

Which basis set is appropriate for polarization function for any atom or Mn(II) atom?