From lanhaiping at gmail.com Fri Nov 3 21:27:04 2006 From: lanhaiping at gmail.com (lan haiping) Date: Sat, 4 Nov 2006 04:27:04 +0800 Subject: [Wannier] How to construct Hamiltonian matrix with MLWFs Message-ID: Hi, As far as i know, one of main applications of MLWFs is model Hamiltionian construction, which has already been used in quantum-transport calculations under TB-Green function framework. But there is no word about such hamiltonian construction in wannier90's mannual. Does such feature be oimplemented ? Regards, Hai-Ping -------------- next part -------------- An HTML attachment was scrubbed... URL: From janietwen at gmail.com Thu Nov 9 13:25:34 2006 From: janietwen at gmail.com (Janiet Wen) Date: Thu, 9 Nov 2006 20:25:34 +0800 Subject: [Wannier] Several questions about distentangle process Message-ID: <392e95900611090425i5392d9efy9d12d151a62cdb45@mail.gmail.com> Dear developers : I have several questions about distentangle process giving in SMV paper. We should set an energy window for distentangle process , while my question is about this energy window's setting. Should all bands in the energy window be included in the distentangle process ? The nscf calculation includes 'nbnd' bands, which cannot insure all bands included in the energy window. So, how to confirm oneself that all bands are included in this energy window ? Just count bands in this window , and make sure there are several bands above this window ? Another question is about wannierise process for metal cnt. I just wanna repeat yongsu's work on pristine cnt , and obtain the minimal basis for cnt. But there are some problems i donot understand clearly. Just with occupied bands for wannierise, there is no problem both fo semi/metal cnt to reach convergence. One of my confusion is why could wannierise reach convergence without distentangle process in metal cnt since i just set 'num_wan' to the number of valence bands ? Why fermi surface effect in metal cnt donot affect this wannierise process ? What we need to consider is the degenerate bands in distentangle process , is it right ? Next i consider '2.5 orbitals per atom' to distentangle WFs. Just with exact 'nat X 2.5' bands for 'nscf' calculation , it is difficult to reach convergence in wannierise process , Does this problem relate to 'not all bands inculded in the energy window ' ? Thanks for your patience and time ! Best Wishes Janiet -------------- next part -------------- An HTML attachment was scrubbed... URL: From lee0su at MIT.EDU Thu Nov 9 22:16:33 2006 From: lee0su at MIT.EDU (Young-Su Lee) Date: Thu, 09 Nov 2006 16:16:33 -0500 Subject: [Wannier] Several questions about distentangle process In-Reply-To: References: Message-ID: <45539AB1.9040008@mit.edu> Dear Janiet, > > I have several questions about distentangle process giving in SMV > paper. > We should set an energy window for distentangle process , while my > question > is about this energy window's setting. Should all bands in the energy > window > be included in the distentangle process ? You can exclude certain bands: that's one of the features in wannier90. > Just count bands in this window , and make > sure there are several bands above this window ? I think that will work. > > Another question is about wannierise process for metal cnt. I just > wanna > repeat yongsu's work on pristine cnt , and obtain the minimal basis > for cnt. > But there are some problems i donot understand clearly. Just with > occupied > bands for wannierise, there is no problem both fo semi/metal cnt to reach > convergence. One of my confusion is why could wannierise reach > convergence without distentangle process in metal cnt since i just set > 'num_wan' to the number of valence bands ? CNT is a zero gap semicondutor and below the Fermi level, the number of bands is the same regardless of a k-vector you choose. However, for a truly metallic system you will never(?) find this situation, e.g. in SMV paper, number of bands below E_F varies when you sweep through the k-space:so there is no hope of getting WFs without disentanglement. For CNT, with a bit of fortune, you can still have localized WFs without disentanglement. Here are two things: First of all, it's not always the case. I once tried (5,0), but it didn't work. Second, those WFs are not good for band structure interpolation. If you do band structure interpolation with those WFs, you will find an extremely poor agreement with one you get from PWSCF code, except the k-points you sampled to get WFs. In a word, you can get WFs from valence band only, but they are useless (well, it also depends on what you want to do with them). > What we need to consider > is the degenerate bands in distentangle process , is it right ? This question is not clear enough for me. > Next i > consider '2.5 orbitals per atom' to distentangle WFs. Just with exact > 'nat > X 2.5' bands for 'nscf' calculation , it is difficult to reach > convergence > in wannierise process , > Does this problem relate to 'not all bands inculded in the energy > window ' ? > It depends on many sorts of things. First, you have to plot your band structure from nscf calculation before conducting any Wannier-related calculations. Make sure that your 'nbnd' is big enough to cover whole bonding+antibonding pi-bands. If you've done this step, I don't know the answer without looking at the output files. Hope this answers some of your questions. Young-Su Lee From janietwen at gmail.com Fri Nov 10 19:53:39 2006 From: janietwen at gmail.com (Janiet Wen) Date: Sat, 11 Nov 2006 02:53:39 +0800 Subject: [Wannier] Several questions about distentangle process In-Reply-To: <45539AB1.9040008@mit.edu> References: <45539AB1.9040008@mit.edu> Message-ID: <392e95900611101053l4026bd1av49d42af1be33f1f8@mail.gmail.com> Thank you , Young-Su. 2006/11/10, Young-Su Lee : > > Dear Janiet, > > > > > I have several questions about distentangle process giving in SMV > > paper. > > We should set an energy window for distentangle process , while my > > question > > is about this energy window's setting. Should all bands in the energy > > window > > be included in the distentangle process ? > > You can exclude certain bands: that's one of the features in wannier90. Does it mean that we could do distentangle process in certain energy window with as minimal bands as possible ? I think the bands exclude of coz affect results, what we should take care is that the bands of same characteristics should not be excluded . Is it right for my understanding this exclude feature ? > > Just count bands in this window , and make > > sure there are several bands above this window ? > > I think that will work. > > > > > Another question is about wannierise process for metal cnt. I just > > wanna > > repeat yongsu's work on pristine cnt , and obtain the minimal basis > > for cnt. > > But there are some problems i donot understand clearly. Just with > > occupied > > bands for wannierise, there is no problem both fo semi/metal cnt to > reach > > convergence. One of my confusion is why could wannierise reach > > convergence without distentangle process in metal cnt since i just set > > 'num_wan' to the number of valence bands ? > > CNT is a zero gap semicondutor and below the Fermi level, the number of > bands is the same regardless of a k-vector you choose. > > However, for a truly metallic system you will never(?) find this > situation, e.g. > in SMV paper, number of bands below E_F varies when you sweep through > the k-space:so there is no hope of getting WFs without disentanglement. > > For CNT, with a bit of fortune, you can still have localized WFs without > disentanglement. > > Here are two things: > > First of all, it's not always the case. I once tried (5,0), but it > didn't work. > > Second, those WFs are not good for band structure interpolation. > If you do band structure interpolation with those WFs, you will find an > extremely poor agreement > with one you get from PWSCF code, except the k-points you sampled to get > WFs. > > In a word, you can get WFs from valence band only, but they are useless > (well, it also depends on what you want to do with them). > > > > What we need to consider > > is the degenerate bands in distentangle process , is it right ? > > This question is not clear enough for me. Sorry for my expression. To my knowledge, besides bandgap, the difference between insulator/semiconductor and metal is fermi surface effect in metal systems. there are entangled bands near fermi surface in metal systems. So for distentangle process , it is not due to fermi surface effect , but to bands degenerate . Is it right for my understanding ? > Next i > consider '2.5 orbitals per atom' to distentangle WFs. Just with exact > 'nat > X 2.5' bands for 'nscf' calculation , it is difficult to reach > convergence > in wannierise process , > Does this problem relate to 'not all bands inculded in the energy > window ' ? > It depends on many sorts of things. First, you have to plot your band structure from nscf calculation before conducting any Wannier-related calculations. Make sure that your 'nbnd' is big enough to cover whole bonding+antibonding pi-bands. If you've done this step, I don't know the answer without looking at the output files. a poor question is about antibonding pi bands counting , Does it equal to the number of atoms ? I appreciate your kind reply. Best wishes. Hai-Ping -------------- next part -------------- An HTML attachment was scrubbed... URL: From lee0su at MIT.EDU Mon Nov 13 23:06:42 2006 From: lee0su at MIT.EDU (Young-Su Lee) Date: Mon, 13 Nov 2006 17:06:42 -0500 Subject: [Wannier] Several questions about distentangle process In-Reply-To: <392e95900611101053l4026bd1av49d42af1be33f1f8@mail.gmail.com> References: <45539AB1.9040008@mit.edu> <392e95900611101053l4026bd1av49d42af1be33f1f8@mail.gmail.com> Message-ID: <4558EC72.4050308@mit.edu> > > Does it mean that we could do distentangle process in certain energy > window with as minimal bands as possible ? I think the bands exclude > of coz affect results, what we should take care is that the bands of > same characteristics should not be excluded . Is it right for my > understanding this exclude feature ? > I would start from including all bands and check projection information of disentangled subspace to the original space. If certain bands have very very small contribution, you may exclude these bands. I'm not totally sure whether it's going to work. In my opinion, exclude feature will be useful when some bands are hampering disentanglement/minimization procedure while you are sure that they should not be included in the disentangled subspce by symmetry. > <> > > To my knowledge, besides bandgap, the difference between > insulator/semiconductor and metal is fermi surface effect in metal > systems. there are entangled bands near fermi surface in metal > systems. So for distentangle process , it is not due to fermi surface > effect , but to bands degenerate . Is it right for my understanding ? > I think disentanglement will be useful when gap is small even if the system is not metallica. Suppose the system has very very small gap at Gamma (G afterwards). Then it is likely that HOMO(G+delta) has a character of LUMO(G). In that case, by including LUMO(G) in the disentangled subspace, we can better describe HOMO(G+delta). I may not the right person to explain this, but this is the best answer I can give. > > a poor question is about antibonding pi bands counting , > Does it equal to the number of atoms ? > Yes. But nbnd must be set higher than 2.5*nat, since sp2 anti-bonding or free electron-like states mix with anti-bonding pi-bands. -------------- next part -------------- An HTML attachment was scrubbed... URL: From lanhaiping at gmail.com Thu Nov 23 13:21:58 2006 From: lanhaiping at gmail.com (lan haiping) Date: Thu, 23 Nov 2006 20:21:58 +0800 Subject: [Wannier] large spreeds for some WFs Message-ID: Dear All, I am trying to study disentanglement of wannier with nanotube (10 0) by including several conduction bands. There are some problems that i didnot come to for the cases of only valence bands included. First , it takes me long time to reach the convergence thread for disentangle process. I just tried to tune dis_mix_ratio , but there is no much effect. In my calculation, I had included 130 bands for disentangling 100 WFs, it would take about 4000 iterations to reach convergence no matter what dis_mix_ratio i set to. This value is of course seriouly larger than the default value (200). Though there are a few parameters for disentanglement, i donot have much idea about energy window settings for disentanglement and frozen energy window. Could we just set dis_win_min to conduction bands' bottom in my semiconducting cnt case ? I also tried to set frozen energy window , but there was no much effect for improving convergence. Could the proper frozen energy window improve convergence ? and would you please tell how to set up frozen energy window properly ? Another question is about large spreads for some WFs after wannierise.Afterwannierise process, there are mainly 3 groups of spreads of WFs. Spreads of first group are around 0.6- 0.7, while 1.6-2.4 are for second group. There are 2 WFs for third group, of spreads around 7.0 ,which are great larger than that of other two group. Does such result have proper physical meaning ? Any comments will be appreciated. As far as i know, large spreads of WFs may not be suitable for tight binding calculation . Are there any methods or settings to obtain smaller spreads of WFs ? Best Wishes, hai-ping -------------- next part -------------- An HTML attachment was scrubbed... URL: From mostofi at MIT.EDU Fri Nov 24 17:37:11 2006 From: mostofi at MIT.EDU (Arash A Mostofi) Date: Fri, 24 Nov 2006 11:37:11 -0500 (EST) Subject: [Wannier] large spreeds for some WFs In-Reply-To: References: Message-ID: Dear Hai-Ping There are two things that might help: (i) increase the number of conduction bands. If you're trying to obtain partially-occupied, out-of-plane p_z orbitals for your nanotube, you probably need to include bands that are at least 5eV or so above the Fermi level. (ii) use the projections to give the dientanglement algorithm a good starting guess, eg, an s-type orbital on each bond and a p_z orbital on each atom. Arash On Thu, 23 Nov 2006, lan haiping wrote: > Dear All, > > I am trying to study disentanglement of wannier with nanotube (10 0) by > including several conduction bands. > There are some problems that i didnot come to for the cases of only valence > bands included. > First , it takes me long time to reach the convergence thread for > disentangle process. I just tried to tune dis_mix_ratio , but there is no > much effect. In my calculation, I had included 130 bands for disentangling > 100 WFs, it would take about 4000 iterations to reach convergence no matter > what dis_mix_ratio i set to. This value is of course seriouly larger than > the default value (200). > Though there are a few parameters for disentanglement, i donot have much > idea about > energy window settings for disentanglement and frozen energy window. Could > we just set dis_win_min to conduction bands' bottom in my semiconducting cnt > case ? I also tried to set frozen energy window , but there was no much > effect for improving convergence. Could the proper frozen energy window > improve convergence ? and would you please tell how to set up frozen energy > window properly ? > > Another question is about large spreads for some WFs after > wannierise.Afterwannierise process, there are mainly 3 groups of > spreads of WFs. Spreads of > first group are around 0.6- 0.7, while 1.6-2.4 are for second group. There > are 2 WFs for third group, of spreads around 7.0 ,which are great larger > than that of other two group. Does such result have proper physical meaning > ? > Any comments will be appreciated. As far as i know, large spreads of WFs > may not be suitable for tight binding calculation . Are there any methods or > settings to obtain smaller spreads of WFs ? > > Best Wishes, > > hai-ping >