yambo5.0 the SAVE file(dipoles and wf) didn't read in LOG

[Quxiao]

Dear all：
I have a question about a wanning when I try use the yambo5.0 to calculate the GW with PPA.
I read the LOG file,vi the "l-gw.out_HF_and_locXC_gw0_dyson_em1d_ppa_CPU_1",
The wanning is that "<---> P1: [DIP] Checking dipoles header
<---> P1: [WARNING] DIPOLES database not correct or missing"
I don't konw what cause that question.
When I use the yambo4.4 to calculate the GW with PPA,vi the "l- gw.out_HF_and_locXC_gw0_dyson_em1d_ppa_CPU_1",
I could see that "<02s> P0001: [WF-Oscillators/G space] Performing Wave-Functions I/O from ./SAVE",however,when I use the yambo5.0,I find that it didn't read the SAVE file，for example，the "<02s> P0001: Reading wf_fragments_1_1" and the "<03s> P0001: Dipoles: P and iR (T): | | [000%] --(E) --(X)<03s> P0001: Reading kb_pp_pwscf_fragment_1".
Thanks
QX

[Daniele Varsano]

Dear Quxiao,
this is completely harmless, it has just changed the messaging in the LOG/report.
Yambo check if already exists the dipole databases, if not or if different from what is asked in input it calculates them.
The reading wf message has been dropped.
Best,
Daniele

[Quxiao]

Dear Daniele：
I find some difference when I calclulate from yambo4. 4 to yambo5.0
I calculate the G0W0 with the PPA,when I use the yambo4.4,after test convergence,I got the gap is 3.7ev,
however,when I use the 5.0 with the same parameters,the results become:2.2ev.
so big a change,I want to konw what's wrong with that? I ensure I use the same systerm to calculate.
I will continue to calculate other systerm to solve that.Please give me some help
Thanks
QX
BIT

[Daniele Varsano]

Dear Quxiao,
what you get with the 5.0 it is weird.
Can you post the report file for both calculation?

Best,
Daniele

[Quxiao]

Dear Daniele：
This is the calculation of the MoS2,which I calculate it with the 4.4 and the 5.0.
Please give me some advice
Thanks
Quxiao

[Daniele Varsano]

Dera Quixiao,

it is not easy to spot the problem. The problem exists already at HF level!
Note the Exc value:
4.4
[xc] E_xc : -12.34436 [Ha]
E_xc : -24.68873 [Ry]

5.0
[xc] E_xc : -1.716654 [Ha]
E_xc : -3.433307 [Ry]

The 5.0 is totally off, this can be compared with the Exc in the QE output.

May ask you to do the following tests:

1) Start for the same QE ground-state calculation if you d not have already done, here looking at the Fermi energy and the Eo value it seems they are different but it could be due to a shift (use ElecTemp=0eV) as in the 5.0 it is finite.
2) Perform the p2y conversion with the respective executables (4.4 and 5.0)
3) Generate the input file accordingly for 4.4 and 5.0 (do not use the same file).
4) Do the setup using ElecTemp=0.0 eV
5) Do an HF calculation (no need to do a full GW) for few states using the same parameters (here you have different FFT). Use the default FFT and set a large cutoff for HF and Vxc, in particular, set VXCRLvcs to its maximum: 144393

Let's see if we can spot the problem, otherwise, it is possible there could be a problem in the compilation.

Best,
Daniele

[Quxiao]

Dear Daniele：
I recalculated the MoS2 system and set the temperature to 0. In both calculations, gw.in is the same except for parallel parameter settings .However, I found that there are still problems with the exchange correlation energy. I don't quite understand. Can you give me some help?
My calculation process is to complete the SCF and nscf calculation with QE, P2Y get the database file, and then calculate the GW of yambo，I know that QE calculate the DFT,and the FFTGvecs(in gw.in) which I set it is 80Ry,however, I found it changed automatically
in r-gw.out_HF_and_locXC_gw0_dyson_em1d_ppa. It became 42Ry in 4.4 and 66Ry in 5.0.
4.4:[xc] E_xc : -12.34985 [Ha]
E_xc : -24.69970 [Ry]

5.0:[xc] E_xc : -1.717292 [Ha]
E_xc : -3.434583 [Ry]
Therefore, how can I set parameters to ensure that FFT does not change automatically and that the calculation of exchange correlation energy is accurate？I promise that parameters I set is compeletely same，however got the different results,maybe I have neglected something，please give me some help.Thank you for your selfless dedication！
Thanks
Quxiao

[Daniele Varsano]

Dear Quxiao,

I don't think the problem here is to be ascribed to the FFT value.
here are my suggestions:
1) Check if there are relevant differences in the r_setup files between 4.4 and 5.0.
2) Check if the PSP you are using have non linear core correction (it is reported in the PSP file)
3) I start to suspect that it could be a problem with the compilation environment for 5.0 . Do you have the possibility to repeat the calculation on a different machine?

Best,
Daniele

[Quxiao]

Dear Daniele：
Firstly，thanks for your help!
In fact，I calculate the calculations on two machines，one machines installed the 4.4,another installed 5.0.
I have two questions ,
1. Although the FFT do not affect the results,I still want to konw,Why does it change automatically?
one is the 42Ry,another is the 66Ry,when I set it 80Ry
2. The parameter （Exchange RL components ）I use is automatically generated by yambo ，however the 4.4 get is the 4991RL,and the 5.0 is the 9851RL.When I calculate that,I set them all to 4991RL.
however,I found that did not change the results.The E_xc which calculated by 5.0 is still different with the 4.4 whatever I set them 4991RL or 9851RL.
The question is that why the parameters (max number of G-vectors planned to use) could change?

Another point is that my input commands are yambo - X - P P - g n - v all - F gw.in,Will it have any impact？
With your help, I did find the change of this exchange-correlation, but I checked my r_setup,no obvious changes were found in the setup file. What do you mean by PSP file? I didn't find this file in my calculation. Is it a log file?
And I will calculate in the same machines with the 4.4 and 5.0 soon.
Thanks for your help again!
Best wishes
Quxiao
BIT

[Daniele Varsano]

Dear Quixiao,

1) This should not happen, anyway check the ecut you set in the nscf calculations, is it possible you considered two different cutoff? The FFTGVecs cannot be set larger than the wfc cutoff you used in the nscf calculation. As a suggestion, do not change it from the default unless you have memory issues.

2) Probably passing from 4.4 to 5.0 the default values it is changed. If I remember well in the previous version the VXCRLvcs was related to the EXXRLvcs while in the new one it is not. For VXCRLvcs it is important to set to the max gvector components of the density when using PBE. In any case I'm pretty sure this is not your problem as the E_xc you find for 5.0 is totally off and it is not a matter of convergence. You can also compare that value with the exchange-correlation energy reported in the QE output (they should be the same). Importantly when repeating a calculation delete the ndb.HF* database otherwise it is not recommputed.

The correct syntax is:
yambo -X p -g n -V all

anyway yambo -h provides you help on the syntax

PSP--> pseudopotentials

And I will calculate in the same machines with the 4.4 and 5.0 soon.

Yes, this is a check I would do, possibly in the machine where you installed the 4.4 as this the calculation that seems to provide reasonable results. Importantly you can do the two calculations starting from the very same nscf run.


Best,
Daniele