Zhen,
10e-4 and phonon is 10e-5. The structure comes out fine. (2.80 Ang and
4.62 Ang and 90. 89.999 119.9722).
I do not advise to use the values obtained in variable cell calculation
as finals.
By symmetry it should be 90, 90, 120, so put it strictly like that and
use the lattice constants suggested by variable cell calculation.
Then you can try to optimize them a little bit manually.
(All this should be done after you've checked the convergence vs.
EnergyCutoff).
Remember that FORCES IN THE UNIT CELL SHOULD BE STRICTLY ZERO!
Otherwise, it does not make any sense to compute phonons.
Your forces are not. I assume that the problem is not only in the k-grid
(it may be already good enough, but checking the total energy
convergence vs k-grid would be desirable)
the problem in your case is likely in incorrect c/a ratio, or positions
of atoms in the unit cell (I have experience only with wurtzite GaN
structure but assume that in Ti the situation is similar, i.e. the atoms
deviate from the ideal-sphere close-packing values). In these cases
optimization of the cell is really a pain in the neck. But anyhow you
have to achieve zero forces first of all.
2nd step: The energy grid should be very fine for phonons.
You have to be sure that you do not have the egg-box effect (see the
manual), i.e. slight shift of atoms changes forces drastically due to
bad mesh.
!!! You may not see deviations in total energy but this does not mean
that you mesh is good enough for forces.
Using soft-confined basis should help here.
As well as using AT LEAST 300 Ry cutoff (preferably with
GridCellSampling to increase the effective cutoff even more)
As Andrei Postnikov always suggests:
try to add a constant shift to each coordinate in the unit cell (say
0.1A) and see if your forces (and total energy) change.
If they do, then you still have an egg-box effect and you need a denser
Energy grid.
And of course SZP basis is far away from the optimal one.
DZP would be much better.
Ideally you should find (or generate with Utils from siesta3) an
optimized basis.
