TPSS - differences with cp2k/gamess

[ires]

Dear all,

I did some test calculations on acetylene, using different density
functional approximations and electronic-structure programs. There I
noticed that the TPSS results obtained with cp2k are different from
the ones obtained with gamess or gaussian. This was the case for both
the geometry and the vibrational frequencies of acetylene:

The experimental values are: CC bond length - 1.2027 A; CH bond length
- 1.062 A.

With gamess-us (TPSS/TZ2P), I obtain r(CC) = 1.2028 A, r(CH) = 1.0648
A, corresponding to a mean relative error (MRE) of 0.13%.

With gaussian09 (TPSS/6-311G*), I obtain r(CC) = 1.2049 A, r(CH) =
1.0656 A, MRE = 0.26%.

With cp2k (TPSS/GTH-PBE+TZ2P-MOLOPT, version 2.1.79), I obtain r(CC) =
1.2101 A, r(CH) = 1.0739 A, MRE = 0.87%

Both, the CC and the CH bond are significantly longer in the cp2k
optimized geometry compared with the other two geometries.

For the harmonic frequencies, I obtain deviations from the
experimental harmonic frequencies of:
with gamess mean absolute deviation (MAD) =  14 cm-1, RMSD = 17 cm-1
with gaussian MAD = 13 cm-1, RMSD = 18 cm-1
with cp2k MAD = 46 cm-1, RMSD = 52 cm-1

For the anharmonic frequencies, I obtain deviations from experimental
anharmonic frequencies of:
with gamess MAD = 36 cm-1, RMSD = 41 cm-1
with cp2k MAD = 52 cm-1, RMSD = 66 cm-1

The disagreement at the harmonic level is much more pronounced than at
the anharmonic level.

For other functionals, like PBE, the agreement of the results from
cp2k and other electronic structure programs is very good.

The calculations are performed fully periodic, using a 10x10x14 A^3
unit cell, which leaves 10 A distance between the repeated C2H2
molecules. I also tested larger unit cells, but the results do not
change.

My question is: Am I doing something wrong? Is there something I need
did not take into account?
This is the input I used for the optimization (and similar ones for
the harmonic frequency analysis and PES generation):

&FORCE_EVAL
  METHOD Quickstep
  &DFT
      BASIS_SET_FILE_NAME  /Users/cp2k/BASIS_MOLOPT
      POTENTIAL_FILE_NAME /Users/cp2k/GTH_POTENTIALS
    &MGRID
      CUTOFF 300
    &END MGRID
    &QS
      EPS_DEFAULT 1.0E-8
    &DISTRIBUTION
      SKIP_OPTIMIZATION TRUE
    &END DISTRIBUTION
    &END QS
    &SCF
      EPS_SCF 1.0E-7
      SCF_GUESS RESTART
      MAX_SCF 500
      &OT
        PRECONDITIONER FULL_ALL
         MINIMIZER DIIS
         STEPSIZE 0.08
         N_DIIS 10
       &END OT
       &OUTER_SCF
         OPTIMIZER DIIS
         MAX_SCF 30
       &END OUTER_SCF
    &END SCF
&XC
     DENSITY_CUTOFF 1.0e-9
     &XC_GRID
      XC_SMOOTH_RHO SPLINE3
      XC_DERIV SPLINE3_SMOOTH
     &END XC_GRID
     &XC_FUNCTIONAL
     &TPSS
     &END TPSS
     &END XC_FUNCTIONAL
   &END XC
  &END DFT
  &SUBSYS
    &CELL
      ABC 10.0 10.0 14.0
    &END CELL
    &COORD
  C         5.0000000000        5.0000000000        6.3854890654
  C         5.0000000000        5.0000000000        7.5956054689
  H         5.0000000000        5.0000000000        5.3115675669
  H         5.0000000000        5.0000000000        8.6694973010
    &END COORD
    &KIND C
      BASIS_SET TZV2P-MOLOPT-GTH
      POTENTIAL GTH-PBE-q4
    &END KIND
    &KIND H
      BASIS_SET TZV2P-MOLOPT-GTH
      POTENTIAL GTH-PBE-q1
    &END KIND
  &END SUBSYS
&END FORCE_EVAL
&GLOBAL
  PROJECT c2h2_tpss
  PREFERRED_FFT_LIBRARY FFTSG
  RUN_TYPE GEO_OPT
  PRINT_LEVEL LOW
&END GLOBAL
&MOTION
  &GEO_OPT
    MAX_ITER 300
    MAX_FORCE 1.E-6
    OPTIMIZER BFGS
  &END GEO_OPT
&END MOTION

I also tried using a higher cutoff and different interpolation
schemes, but the results did not change.
I am grateful for any comments.
Thanks
Inga