The same way you optimize a neutral molecule. One problem you may encounter is that usually Zwiterions are not stable in gas phase, so the optimization goes back to the neutral form.  In that case you may need to simulate the media to stabilize it. A typical example is glycine, which in gas phase exists as neutral, but in solution the Zwitterion form is more stable. 

Dear Antonio,

Thanks for your answer. The main question is how can I localize the charge on an atom prior to optimization.

You should not localize. Charges shall be such as they are.

You cannot do that. The charges are calculated from the electronic density after the SCF is performed. There are several methods to calculate charges on a given atom, but this is a different story.

Glycine has proton transfer, so it can be zwitterion. Localizing charge cannot reproduce a zwitterion. And zwitterion can be stable under the influence of polar group like metal ion and water.

If you follow what Saman has stated and then optimize the geometry in a polarized medium (e.g. PCM keyword in gaussian with the choice of your solvent) it should avoid the neutralization which is bound to happen as Saman and Antonio have already stated if you optimize in the gas phase.

A possible solution can be constrained DFT, where you can localize the charge or spin density in  different parts of your system. It is implemented in NWChem and Q-Chem.

http://www.nwchem-sw.org/index.php/Density_Functional_Theory_for_Molecules

Dear Soumen, That's exactly what I was looking for.

You could also consider taking a look at the excited states. One of them could be your Zwitterion.

Hi! I am trying to optimize the GABA molecule. I started with "extended" zwitterionic structure. Gaussian with the following settings:

#n B3LYP/6-31G(d,p) Opt

returns change to neutral form, however if I add diffusion:

#n B3LYP/6-31++G(d,p) Opt

zwitterion is stable.

Do you have any idea why zwitterion gains stability? Or which level of theory/ basis set shall I choose to have accurate results?