scf calculation is too slow

[rkarkee101]

Hi,

I am new user of Phoebe. I am running a HfTe5 (12 atoms) calculations and the scf calculation is too slow from the quantum espresso patch.

Is this expected or did I miss something?

On regular quantum espresso, the scf calculations takes about 2 min. But this is taking longer ~ 50 mins.

It also complained about "some processors have no G-vectors for symmetrization" after printing FFT dimension and RAM estimation (this was not present in the regular QE)

Can you please help?

Thanks

Rijan

[AnishDas-2025]

Hi, " "some processors have no G-vectors for symmetrization" means that your number of cores are more your number of k points. How many nbands you have used in scf? Number of nbands must be complied with the number of bands used in wannier90 later. If everything is right, then you can try with mpirun/srun -n pw.x -npool -n option. Check if it might work

[jcoulter12]

Thank you Anish! I'm traveling and you got here first. :)

@rkarkee101

Anish is right, improper parallelism could be a reason for a slowdown, for sure -- there are a few other possible reasons.

I could see a system with Hf and Te, 12 atoms, requiring quite a lot of plane wave coefficients and having a lot of electrons. The patched version of QE has basically one modification -- it's writing this plane-wave coefficient information to file, so that each piece of the el-ph calculation can be done with the same wavefunction gauge. If a system has a lot of plane wave coefficients, is very low symmetry, or you're on a filesystem system where read-write actions are very slow, this could all cause a slowdown.

How many electrons are in this system, for reference? And, assuming some phoebe.scf files may have been written, how big are they in terms of memory? Do you have a very high plane wave cutoff?

With more info, I'll try to assess the best option for you. I have one other DFT code interface which will not have this trouble, which is useful in some cases -- but, maybe for a large unit cell may not be as helpful.
Ideally I would also like to use the recently developed VASP interface when it's finalized as well, but it's not quite ready yet.

Best,
Jenny

[rkarkee101]

Hi Jenny and Anish,

Thanks for your reply.

There are 184 electrons and my cutoff is 75 Ry.
I think thats fine (I was more worried about Phonon grid calculation taking too long but it was fine).
In my grid, I have 21 q-points and I got the following error at the end of (0 0 0) calculations (which only took about 2.5 hours which is similar in regular calculations).

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Error in routine phoebe (1):
k and q grids are incommensurate

Here is my scf and ph input files

&CONTROL
calculation = 'scf'
pseudo_dir = '/users/rkarkee/pseudo'
outdir = './'
prefix = 'relax'
tprnfor = .true.
tstress = .true.
! etot_conv_thr=1d-5
! forc_conv_thr=1d-4

/

&SYSTEM

ecutwfc     =  75,
ecutrho= 300,
ibrav       = 0,
nat         = 12,
ntyp        = 2,
nbnd=132

/

&ELECTRONS
conv_thr = 1.00000e-10
electron_maxstep = 200
mixing_beta = 0.3
diagonalization = "david"

/
&IONS
/

&CELL
/

K_POINTS {automatic}
6 6 4 0 0 0

ATOMIC_SPECIES
Hf 178.49 Hf.upf
Te 127.60 Te.upf

CELL_PARAMETERS (angstrom)
1.961309481 7.120031888 0.000000000
-1.961309875 7.120031780 0.000000000
0.000000000 0.000000000 13.476144159

ATOMIC_POSITIONS (crystal)
Hf 0.3139670092 0.3139670362 0.7499999470
Hf 0.6860327178 0.6860327618 0.2499999640
Te 0.6604894418 0.6604894298 0.7499999470
Te 0.3395103962 0.3395103892 0.2499999640
Te 0.9274294965 0.9274295525 0.8526497923
Te 0.0725702375 0.0725702385 0.3526498633
Te 0.9274294965 0.9274295525 0.6473500297
Te 0.0725702375 0.0725702385 0.1473500827
Te 0.7923128047 0.7923128277 0.4340762718
Te 0.2076869613 0.2076869643 0.9340761648
Te 0.7923128047 0.7923128277 0.0659236882
Te 0.2076869613 0.2076869643 0.5659236212

phonons for HfTe5
&inputph
tr2_ph=1.0d-16
prefix="relax"
ldisp=.true.
nq1= 4
nq2= 4
nq3=4
fildyn="hfte.dyn"
fildvscf="hfte.dvscf"
electron_phonon="epa"
nmix_ph=20
/

[jcoulter12]

Hi @rkarkee101,

Glad you were able to get through the last difficulty.
This one is pretty easy -- it is just as the error says, your k and q grids are not commensurate. Basically all el-ph codes ask that your k and q grids be aligned so that your q grid is a factor of the k grid (as in, each dimension of k divided by the same dimension of q is an integer). This is because quantum espresso needs to calculate points where two ks are connected by a q which is only possible if the grids align this way.

You can see your q grid = 4,4,4 and your kgrid = 6,6,4.
You can do either qgrid = 3,3,2 or qgrid = 6x6x4, which would both be commensurate.

Indeed, this could be a long phonon calculation because it has a lot of electrons -- typically qpt #1 is the fastest one by a fair bit.

Best,
Jenny

[rkarkee101]

Hi Jenny,

I am trying to compute Linewidth along a Band Path (following the same tutorial).

I have the rta_ph_relaxation_times.json and path_ph_bandstructure.json.

When I do tau_path.py rta_ph_relaxation_times.json and path_ph_bandstructure.json

I get this complaint INPUT INPUT2 calcIndex,
error: the following arguments are required: calcIndex.

I dont see anything regarding calcIndex in the inputs. Did I miss something?

Rijan