2D kpoint grid being interpreted as 3D

[arb83@cam.ac.uk]

Dear Daniele and Claudio

I used the NoDiagSC to initialise yambo and everything was running fine. I then added the line:

"
GTermKind= "BG"
"

to my input file and the code crashed at the Newton solver. Might this be related to using NoDiagSC? I'm also neglecting the Vnl part of the potential but I don't see why this should have an effect.

Best wishes,

Alan

[claudio]

Dear Alan

please provide us an example of this crash.

I know that some bug has been fixed for the BG terminator,
but I think your version should be fine

best
Claudio

[Daniele Varsano]

Dear Alan,
NoDiagSC and the commutator of Vnl should not have any effect on the terminator.
Can you provide us more details about the error you are facing? e.g. input file and report/log?
Note that the use of GTermKind has an impact on the memory requirements, and it is not yet implemented in GPU machines (in case you are running on GPU).

Best,
Daniele

[arb83@cam.ac.uk]

Dear Daniele and Claudio

I'm working on quite a large system so it may just be that memory is the issue. The only error message I see is in three of the LOG files, all read the same:

"
<03m-19s> P0017: [07] Dyson equation: Newton solver


<03m-19s> P0017: [PARALLEL distribution for [G_m_G] G-vectors] Loaded/Total(Percentual):21/3279(1%)
P0017: [ERROR] STOP signal received while in :[07] Dyson equation: Newton solver
P0017: [ERROR] Error search for G=G1-G2 !!
"

I'm now trying with extra memory available in case that helps. I'm not running on a GPU machine.

With best wishes,

Alan

[Daniele Varsano]

Dear Alan,
the message is rather clear and it is not related to memory. The code in unable to find G1-G2 difference among the G at your disposal, and this is a quantity you need to apply the BG technique. It is possible it is just a problem of roundness and you can try to raise the threshold to find the Gvectors difference in:
./src/common/G_index.F

Code: Select all

 65  do ig=ng1,ng2
 66    if (v_is_zero(v1-g_vec(ig,:),zero_=G_iku_zero)) then
 67      G_index=ig
 68      exit
 69    endif
 70  enddo

you can change the condition in e.g.:

Code: Select all

 66    if (v_is_zero(v1-g_vec(ig,:),zero_=G_iku_zero*5.)) then

or even higher value. Then recompile the code.

Even better if you print the results of v_is_zero(v1-g_vec(ig,:) after the loop to check that actually that condition is just slightly larger than the default definition of G_iku_zero=10^-5

Best,
Daniele

[arb83@cam.ac.uk]

Dear Daniele

Thank you for this, it has allowed the code to run again. I'm now finding that the report file includes a new section entitled 'Dipoles' which contains the warning: " [WARNING] DIPOLES database not correct or not present". I've ignored Vnl in these calculations, is this the reason for this warning or is there something else I'm missing? I just want to check it's not as a result of re-compiling yambo.

All the best,

Alan

[Daniele Varsano]

Dear Alan,
this means that either the dipoles were not present or they were calculated in a previous run with a different number of bands.
In both cases, the dipoles are recalculated and it is not related to the change you did in the source.

Best,
Daniele

[arb83@cam.ac.uk]

Dear Daniele

I am now seeing an equivalent error message when running BSE calculations:

"
<02s> P82-cpu-e-698: X indexes | | [000%] --(E) --(X)
P82-cpu-e-698: [ERROR] STOP signal received while in :[03] Transferred momenta grid
P82-cpu-e-698: [ERROR] Error search for G=G1-G2 !!
"

I note this time the stop signal is received in the Transferred momenta grid. The re-compiled version of Yambo solved the equivalent GW error discussed above. Should I recompile Yambo editing the './src/common/G_index.F' again, but increase the tolerance further (beyond 5*, which worked in the other case), or do I need to edit a different file before recompiling?

Best wishes,

Alan

[Daniele Varsano]

Dear Alan,
the file to be modified is the same as before.
Probably you can solve just increasing the tolerance, maybe it is safe to look if it is just a matter of rounding, i.e. looking if
there is an ig where v1-g_vec(ig,:) is actually small and raise the tolerance accordingly.

Best,
Daniele

[arb83@cam.ac.uk]

Dear Daniele

Thanks for getting back to me. I'm not sure which file to look in to find the values of v1 and ig, could you let me know?

Best wishes,

Alan