Weird behaviour of Ti pseudo with core correction

[andreaschiaf]

Hi Abinit Users,

I'm trying to generate a pseudopotetnial for Ti with FHI code.
In particular I'm focusing on LDA (Perdew-Wang 92) norm-conserving Troullier-Martins pseudos.
I would like to have as valence configuration: 3d24s2.
I'm testing my pseudos simulating bulk Ti-hcp.

I noticed that when the core correction is included, then I get a better estimation of the
lattice parameter. However if I study the total energy trend as function
of energy cutoff using pseudos with core correction, then I don't get a monotonic
decreasing behaviour.

Actually I always thought the energy minimization is variation problem respect to
energy cutoff, no matter is bad you pseudo. Thus I don't understand why here is not the case.

I'm using:
Pseudo generation input (using with Vloc l=1):
22.00 5 2 8 1.10 : z nc nv iexc rnlc
1 0 2.00 : n l f
2 0 2.00
2 1 6.00
3 0 2.00
3 1 6.00
3 2 2.00
4 0 2.00
2 t : lmax s_pp_def
0 2.30 0.00 t l,rcut,et,pptype
1 2.30 25.00 t l,rcut,et,pptype
2 2.20 0.00 t l,rcut,et,pptype

Abinit input:
occopt 4
tsmear 0.005
ngkpt 16 16 10
shiftk 0.0 0.0 0.5
acell 5.45 5.45 5.45
rprim 1.0 0.0 0.0
0.5 0.86602540378443864676 0.0
0.0 0.0 1.58


and I'm getting:
ecut1 30.0 ---> ETOT -16.448271670434
ecut2 40.0 ---> ETOT -16.583513157166
ecut3 50.0 ---> ETOT -16.569911201610
ecut4 60.0 ---> ETOT -16.530812894014

Could you tell me where I'm wrong?!
Thanks a lot!