[Wannier] get the conduction bands right

[Butler Butler]

Hi verybody,

I am new to wannier functions. As I hate to guess the initial WFs case by case, I just designed my own automatic projection scheme and implemented it into our own code. Then I tested whether my projection works. So here is what I got:

1. By my projection scheme, for semiconductors, the valence bands are very well reproduced when num_bands=num_wann. That is, with no disentanglement, the interpolated bands are good. 

For simple systems such as GaAs, I need like 10 wannierise loops to get the smooth interpolated bands.

2. Including disentanglement (that is for num_bands>num_wann), the interpolating quality drastically drops. Not only the conduction bands are extremely bad, the quality of the valence bands are also ruined. 

3. Setting the inner energy window (setting dis_froz_max to the top of the valence bands), the interpolating quality of the valence bands indeed improves. But never at the level when without disentanglement. Besides, the interpolated conduction bands are also improved. But the quality of both are still unacceptable.

So I have the following questions:

1. From the above description, is there a hint of whether my projection scheme works or not? Or maybe there are just some bugs in my .amn coding?

I have no idea why it seems to work very well for valence bands but not for conduction bands. From such sharp contrast, is there a hint of how I shall improve my projection?

As a code developer, the above question is my biggest concern. If my scheme does not really work, then I have to implement some other schemes.

2. From the theory viewpoint, it seems there is a distinction on what are the best WFs for the valence and for the conduction. Then what is that?

Basically I want know what are the best way to design the initial WFs, which works well for both the valence and conduction.

3. Disentanglement seems to be very sensitive to the .amn implementation, while the wannierise process is very insensitive to it (I note that we can even use Bloch phases for wannierise but not for disentanglement). Is that right?

4. Can I expect a good, unified projection scheme for most systems at all?  Or maybe we just have to stay at the present status: different projections for different systems? I really do not want to guess every time.

5. Is it practical to expect to get the correct conduction & valence bands for really large systems (c.f. for hundreds of bands altogether)?

Thanks for all reading this,

Dr. Lin-Hui Ye
Department of Electronics
Peking University, China
 		 	   		  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/wannier/attachments/20101104/77761bd4/attachment.html>