Hi Oleg

Thanks for your suggestion.I tried as you suggested with opt(maxstep=3) but still after optimization, it is changing to enol form.(seen in Gaussview). [check attached files]

Strangely, it worked well when HF theory was used (all basis sets except 3-21g).

But the problem is when I use DFT theory.

Thank you.

Are you sure that keto form is stable (minimum). What I mean is do you expect a barrier between the keto and enol form? If there is no barrier then you will always end up with an enol form (which has a minimum). In order to confirm this point  you can run a potential energy scan along the O-H distance. In Gaussian you can do this with relaxed scan option (opt=modredundant).

I would second what Tuhin says above, since the enol form is aromatic and the keto form is not.  I would guess that if you can't optimize the keto form, you haven't checked to see if the O-H-O vibrational frequency is positive (after all, why would you do a frequency calculation on an non-converged structure).

I also have a suggestion for generating a "false" stable keto form (which I will refer to as the "exo" keto form) for comparison: if you direct the OH bond away from keto carbonyl, you will probably obtain a local minimum (especially if you start from a perfectly planar geometry).  If the keto form exists in solution, it is probably by hydrogen bonding between the solvent and the OH, so the endo keto form may be relevant to you.  Additionally, if there is no barrier to H transfer between the two forms you indicate in your images (the "endo" keto and enol forms), there might still be a barrier to conversion of the exo keto form to the enol (via an endo keto form-like geometry).  I would guess inversion of the OH is more likely than rotation, but that is just a guess and (if you care to know) you should investigate both.