Could you please post more than one line from the Gaussian output (at least a few last lines)?

Are You sure that the order of the atoms (both QM and MM) in both geometries is the same?

RedQX1 iteration 16 Try 4 RMS(Cart)= 0.00002335 RMS(Int)= 3.53742867 XScale= 0.99900090

RedQX1 iteration 16 Try 5 RMS(Cart)= 0.00002335 RMS(Int)= 3.53743521 XScale= 0.99899876

RedQX1 iteration 17 Try 1 RMS(Cart)= 0.00002335 RMS(Int)= 3.53743521 XScale= 0.99899876

RedQX1 iteration 18 Try 1 RMS(Cart)= 0.00000467 RMS(Int)= 3.53742998 XScale= 0.99900047

RedQX1 iteration 18 Try 2 RMS(Cart)= 0.00000467 RMS(Int)= 3.53743129 XScale= 0.99900004

RedQX1 iteration 18 Try 3 RMS(Cart)= 0.00000467 RMS(Int)= 3.53743260 XScale= 0.99899961

RedQX1 iteration 19 Try 1 RMS(Cart)= 0.00000467 RMS(Int)= 3.53743260 XScale= 0.99899961

RedQX1 iteration 20 Try 1 RMS(Cart)= 0.00000093 RMS(Int)= 3.53743155 XScale= 0.99899996

Old curvilinear step not converged, using linear step:

SCX= 6.09D+01 DXMaxT= 1.77-317 SCLim= 0.00D+00 Fact= 0.00D+00

RedCar/ORedCr failed for GTrans.

Error termination via Lnk1e in /opt/SC/aplicaciones/g09-D.01/l101.exe at Tue May 12 08:45:53 2015.

Job cpu time: 0 days 0 hours 14 minutes 39.4 seconds.

File lengths (MBytes): RWF= 5 Int= 0 D2E= 0 Chk= 1 Scr= 1

I think they have the same order. I just check the QM atoms because it seems to me pretty much the same.

OK, looks like atom ordering is not the problem.

In your case gaussian can't converge the SCF in the first (probably) geometry optimization step. That tells that the geometry of the guessed TS taken from QST2 is not that good. The're unfortunately no easy solution to this, but you can try guessing the geometry of the TS on your own (using chemical intuition) and then trying opt(ts,noeigentest).

Ok, just a little doubt: I need to use QST3, and put mi own idea of the transition state, or directly try to fin the transition state TS(Berry) on my own? 

Thanks

Well, you can try both. I had little success with QST3 and more success with TS, but that requires a really good guess of the ts geometry. 

Ok...i will do the two options.

Bartosz, thank you very much,

I only try TS and it works very good.