[Wannier] disentanglement convergence and frozen window

[Giovanni Pizzi]

Dear Elias,
I don't know it is useful in your case, one should probably look deeper into the band structure, but you may want to give a look to the attached file: it is a new feature (contributed by Gabriele Sclauzero, ETH Zurich) for disentangling only within spherical regions of k space (typically, near specific k-points).

The attached PDF is the documentation of the new feature, that is still only in the private development branch and not yet released. If you think it could be useful for you, we can send you the source code.

Best,
Giovanni


--
Giovanni Pizzi
Post-doctoral Research Scientist
EPFL STI IMX THEOS
MXC 340 (Bâtiment MXC)
Station 12
CH-1015 Lausanne (Switzerland)
Phone: +41 21 69 31124




On 19 Feb 2015, at 18:12, Elias Assmann wrote:

> Hi,
>
> I have a calculation where I find that disentanglement does not
> converge when I use a frozen window.  Without the frozen window,
> disentanglement converges quickly:
>
> +---------------------------------------------------------------------+<-- DIS
> |  Iter     Omega_I(i-1)      Omega_I(i)      Delta (frac.)    Time |<-- DIS
> +---------------------------------------------------------------------+<-- DIS
>       1      15.18420341      13.87841728       9.409E-02      1.42  <-- DIS
>       2      14.48064415      13.51066622       7.179E-02      1.55  <-- DIS
>       3      13.96026696      13.24672414       5.387E-02      1.69  <-- DIS
>       4      13.57678297      13.04599162       4.069E-02      1.82  <-- DIS
> …
>     116      12.23570367      12.23570367       8.045E-11     16.87  <-- DIS
>     117      12.23570367      12.23570367       6.777E-11     17.01  <-- DIS
>     118      12.23570367      12.23570367       5.708E-11     17.14  <-- DIS
>     119      12.23570367      12.23570367       4.809E-11     17.27  <-- DIS
>
> (The complete ‘wout’ file is attached.)  But when I turn the frozen
> window on, the procedure seems to get stuck in a loop until it runs
> out of iterations:
>
> +---------------------------------------------------------------------+<-- DIS
> |  Iter     Omega_I(i-1)      Omega_I(i)      Delta (frac.)    Time |<-- DIS
> +---------------------------------------------------------------------+<-- DIS
>       1      12.67123899      12.48798995       1.467E-02      1.37  <-- DIS
>       2      12.54971488      12.44729442       8.228E-03      1.47  <-- DIS
>       3      12.49131486      12.42411347       5.409E-03      1.57  <-- DIS
>       4      12.45357352      12.41102843       3.428E-03      1.68  <-- DIS
> …
>     997      12.39231118      12.39299287      -5.501E-05    106.04  <-- DIS
>     998      12.39231119      12.39299288      -5.501E-05    106.15  <-- DIS
>     999      12.39231118      12.39299287      -5.501E-05    106.26  <-- DIS
>    1000      12.39231119      12.39299288      -5.501E-05    106.36  <-- DIS
>
> I tried making the frozen window smaller, but the end result is the
> same.
>
> Also, I have a little trouble understanding the output.  I would have
> expected “Omega_I(i)” in one row to be equal to “Omega_I(i-1)” in the
> next row.  And if the spread oscillates, why is Delta always negative?
>
> I have actually posted about this calculation before:
> <http://mailman.qe-forge.org/pipermail/wannier/2014-June/000806.html>.
> The problem was that I cannot get a good Wannier projection on an
> N×N×N k-mesh with N even.
>
> I am revisiting it now because for my current application I need the
> even-N (specifically, power-of-2) k-mesh.  I noticed that an extrinsic
> band touches my target bands just at the R-point of the BZ (0.5, 0.5,
> 0.5); and that R is included in even-N but not odd-N k-meshes.  So I
> tried to use disentanglement with that one extra band.
>
> But disentanglement does not help.  The Wannierization seems to have
> the same problems in all the cases.
>
> Finally, I tried including another band in the disentanglement.  In
> this case, disentanglement does converge nicely, but the
> Wannierization is not satisfactory either.
>
>
>        Elias
>
>
> PS: The immediate motive for this post was that I was the puzzling
> behavior of the disentanglement.  But obviously I would also be
> grateful for any other hints on how to converge the even-N
> calculation.
>
> --
> Elias Assmann
> Institute of Solid State Physics
> Vienna University of Technology
> <http://www.ifp.tuwien.ac.at/cms/>
> <W14_k8.wout>_______________________________________________
> Wannier mailing list
> Wannier at quantum-espresso.org
> http://mailman.qe-forge.org/cgi-bin/mailman/listinfo/wannier

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/wannier/attachments/20150220/8b482795/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: tutorial_20.pdf
Type: application/pdf
Size: 393303 bytes
Desc: tutorial_20.pdf
URL: <http://lists.quantum-espresso.org/pipermail/wannier/attachments/20150220/8b482795/attachment.pdf>