Page 2 of 2

Re: Question on |g|(qx,qy) distribution

Posted: Thu Nov 02, 2017 3:07 pm
by andreyl
TL;DR: To obtain adequate |g_ijnu| distribution for the whole BZ, fsthick must include not only all the E(k), but also E(k+q) of bands you want to build distribution for. In my case fsthick of 3eV was not enough. Degenerate states need special treatment, since non diagonal electron-phonon matrix elements may be important.

Hi,

after several attempts to find the source of the problem I found that my data were affected by:

1) Too small fsthick. This was the main troublemaker in my case.
http://epw.org.uk/Documentation/Inputs#fsthick
Looks like it limits not only k bands considered, but also of k+q.
I can't completely wrap my head around it, since I expected, that if k
band is into the calculation, corresponding k+q should also be, seems like it is not.
Still, it explains the black lines on the plot completely: the setup didn't allowed the code to work
with scatterings to bands outside of fsthick:)

The way out in this case is to start with fsthick which includes all (in my case 5) bands,
obtain the results and slowly start to reduce it step by step, till black lines appear.

here is an example:
Image
I think it speak for itself.

2) My Wannierization was not good enough. Although the decay looked reasonable, I had
unphysical spreads and centers for some MLWFs. Well, seems like sometimes good band structure
is not enough.

3) Averaging in generate point. Well, this is a though one. I still don't understand it completely.
We have a twofold degeneracy in k = K for graphene, a famous Dirac point, that means that we have four |g ijnu| for every mode. To illustrate this, here are unavenged |g ijnu| for all acoustic modes:
Image

And here are averaged ones (sum of contributions from one column divided by 4):

Image

It is still not 100% as in references I've given in first post, but, so far so good. I am mostly worried about absolute values: mine a 2-3 times smaller, that may not give the correct values of deformation potential. At the same time, there are still ways of increasing the accuracy.

I haven't tested the proposed truncation technique yet, since it turned out to be a little more computational demanding, than I assumed initially. Maybe this will improve the results.

p.s.
I see that links to figures, that I posted previously are no more valid, in case somebody want to see everything, check out the "paper" file:
https://paper.dropbox.com/doc/Graphene- ... kTLb8i7E1R

Re: Question on |g|(qx,qy) distribution

Posted: Fri Nov 03, 2017 1:54 pm
by sponce
Hello Andreyl,

Thank you for sharing the solution !

The EPW code was not initially intended to print the electron-phonon matrix element g.

If you care about the self-energy, then k+q states that are energetically far from the state you are considering (here around Fermi level) do not matter much (hence the fstick thing to decrease computational load).

It look much better now indeed.

Best,
Samuel

Re: Question on |g|(qx,qy) distribution

Posted: Fri Apr 19, 2019 8:28 am
by SaraFiore
Dear experts,
I am occurring in the same problem more or less for a monolayer of MoS2
I tried to use
k = 21 21 7 and q = 7 7 1
and
k = 16 16 2 and q = 16 16 2
but the decay files give in both cases bad behaviour when I plot decay.epmat_wanep vs Rp (i.e. 2nd and 3rd column)

I reported my results and input files here
https://paper.dropbox.com/doc/DECAY-COMPARISON--AbhHkS6boK8gxJS1ueBVnn~jAQ-dAvIjE8XtsbMWJQDpvmyK

This bad behaviour obviously translates into a very bad behaviour of the g(k,q) (shortly reported here https://paper.dropbox.com/doc/ML-MoS2-EPW5.1-EFVB--AbggM14PjJG9GQ9zb_h8d6eBAQ-LDJdFth7UUfYm8f9E0zw0)

Any help is very appreciated

Thank you in advance

Sara Fiore
PhD student ETH Zurich