alm CV.in process is very slow, are there any strategies to speed it up?

[chunweizhu]

Dear developers
In the cross validation process, to determine the optimal amplitude of regularization ($\alpha$) is very slow. Is there any acceleration strategy?
Best regards

[ttadano]

How slow is it? Did you use the default value of STOP_CRITERION?

[ttadano]

Yes, this is a large problem. Can you make the cutoff radii smaller, check the validation score, and gradually increase the radii if necessary?

[chunweizhu]

Hi @ttadano,
I'm going to do a test like this, Thank you for your patient answer.😊
Best regards

[chunweizhu]

Hi @ttadano
I'm sorry, I would like to ask you another small question.
When using the extract.py script：

$ python ${ALAMODE_ROOT}/tools/extract.py --VASP SPOSCAR --offset vasprun0.xml vasprun_harm*.xml > DFSET_harmonic

To obtain the correct DFSET file, does the atomic order in POSCAR (used to obtain vasprun_harm*.xml) and the atomic order in the SPOSCAR file have to maintain a one-to-one correspondence?

---

SPOSCAR
.........

$x_1$ $y_1$ $z_1$ atom1
$x_2$ $y_2$ $z_2$ atom2
$x_3$ $y_3$ $z_3$ atom3
$x_4$ $y_4$ $z_4$ atom4
if the SPOSCAR file atomic order is atom1 atom2 atom3 atom4

---

POSCAR
.........

$x_2$ $y_2$ $z_2$ atom2
$x_1$ $y_1$ $z_1$ atom1
$x_3$ $y_3$ $z_3$ atom3
$x_4$ $y_4$ $z_4$ atom4
So the POSCAR file atomic order is atom2 atom1 atom3 atom4 that can not obtain the correct DFSET file?

[ttadano]

Yes, the order of atoms in the supercell is important and must be consistent throughout the calculations.

[chunweizhu]

Hi @ttadano
Thank you for your answer it helps me a lot.😊
Best regards