2014-11-18 18:05 GMT+00:00 Seyed Mohammad Tabatabaei <[email protected]>:
> I am deeply thankful for your sincere support. Hopefully, my last > questions are: > > You said: > "Yes, the idea as I said is that you want the electrodes to behave "as > much as bulk as you can". > So I would put in a big vacuum" > > I think you mean a "big vacuum" full of buffer unit cells at the two > ends of the scattering region plus a short vaccum to avoid direct > contact between left and right electrodes. Am I right? > No. Not a short vacuum. Well it could be that it works. But... > > Also, I am doubtful on what value should be used for TS.UpdateDMCROnly > and why? Would you please elaborate this point for me a little bit > more? > Again, I would encourage you to draw a sketch of what that flag means. > > Best regards, > Mohammad, > > > > > > On 11/18/14, Nick Papior Andersen <[email protected]> wrote: > > 2014-11-18 14:40 GMT+01:00 Seyed Mohammad Tabatabaei <[email protected]>: > > > >> Dear Nick, > >> > >> I am very happy that you are here and thank you for your guidance. I > >> would be indebted to you if you help me better understand the point. I > >> have tried to put my questions in a yes/no format so that it takes you > >> a short time to answer them. In case I was wrong or misunderstood, I > >> would be thankful if you provide me with brief explanations. > >> > >> Even if I add buffer unit cells to the left and right electrodes, say > >> 4 buffer unit cells on each side, again a VACUUM is required because > >> the left and right electrodes are still different and cannot form a > >> continuous system. Then, am I getting you right? I mean a short vacuum > >> is inevitable in such a system. > >> > > Yes, the idea as I said is that you want the electrodes to behave "as > much > > as bulk as you can". > > So I would put in a big vacuum. > > > > > >> If the first .DM is only an initial guess for TranSIESTA, then, it > >> should not affect the final results. Its function is to help > >> TranSIESTA converge more rapidly to the final result. Am I right? I > >> mean the .DM would not enter any other calculations except the first > >> guess. > >> > > In principle yes, in reality not so much... Your initial starting guess > can > > certainly have an impact on your final converged system. You may end up > in > > a local minima... > > > > > >> I read the SIESTA manual and, as you noted in your response, it says > >> it does not include buffer atoms in the TranSIESTA run. I also added > >> the entry for TS.UpdateDMCROnly here: > >> > >> "TS.UpdateDMCROnly (logical): During the TranSiesta (Green’s > >> functions) self consistent cycle, it updates only the density matrix > >> elements of the contact region, if set to true. The electrodes and > >> coupling terms are kept as the ones obtained in a first Siesta run. If > >> set as false, the coupling terms are also updated by the Green’s > >> functions density matrix. If a larger number of electrode layers > >> (metallic systems) are included in the contact region, the coupling > >> terms may not need to be updated. If set to false, however, may result > >> in larger number of iterations to converge." > >> > >> I think this flag should be set to "false" to account for the worst > >> case. :) I would be really grateful if you further explain it to me. > >> Particularly, I can not undestand "If a larger number of electrode > >> layers (metallic systems) are included in the contact region, the > >> coupling terms may not need to be updated." As far as I can > >> understand, the geometry for the scattering region calculation is > >> composed of the left electrode + contact region + right electrode. > >> Then, no electrode layers would exist in the contact region. Moreover, > >> I do not understand why updating the coupling terms is not needed in > >> this case. > >> > > Again, you want the electrodes to behave "as much as bulk as you can". > If > > they are not, how can you couple a bulk semi infinite region to your > device > > region. > > This is really important! > > Try and sketch how siesta perceives things, then how transiesta does, > then > > overlay those two scenarios. This should make it clear how to engage > > calculations you want to perform. > > > >> > >> Finally, I have exprimented one more thing with SIESTA and wanted to > >> know your notion about that. Since my first intention was to add a > >> vacuum between scattering regions, I tried to do this in a > >> continuation-like run scheme. That is, for the system I presented in > >> my question, I first run a Transiesta calculation from scratch in the > >> case with no vacuum which generated new .DM and .TSDE files. Then I > >> copied the electrode .TSHS file along with the .TSDE file to another > >> folder and increased the vector in the z direction by 2 Angs. I did > >> the same thing five times, that is, I inserted a 10 Angs vacuum in the > >> fifth run. Fortunately, all the transiesta calculations converged. Is > >> it a correct practice for inserting a vacuum? Whats wrong with it? It > >> is important to note that the calculations did not converged in cases > >> where I also copied the .DM file maybe because the .DM from the > >> previous calculations were a bad guess. > >> > > It will probably restart from a clean sheet. It will only use the DM > > information in cases where the sparsity pattern does not change. > > I would definitely advice you NOT to re-use the DM file from another > > geometry. Only if you increase the vacuum for non-interacting cells > (which > > will only happen for large vacuum) is this allowed/sensible. > > > >> > >> Best wishes, > >> > >> > >> > >> > >> > >> > >> > >> > >> > >> On 11/18/14, Nick Papior Andersen <[email protected]> wrote: > >> > 2014-11-17 8:31 GMT+00:00 Seyed Mohammad Tabatabaei <[email protected] > >: > >> > > >> >> Dear SIESTA Users, > >> >> > >> >> I have a few conceptual questions about TranSIESTA and would be > really > >> >> grateful if you help me with them. In fact, the answer to these > >> >> questions is quite important for my calculations. I have used > >> >> TranSIESTA > >> >> to study the current in an 8-AGNR system. My system consists of > >> >> 16X7=112 > >> >> carbon atoms. 16 atoms make the left electrode, 5X16=80 atoms make > the > >> >> scattering region, and 16 atoms make the right electrode. As the > >> >> transport > >> >> direction is along the z direction, my system would look like this > >> >> along > >> >> the z direction: > >> >> > >> >> ... + Left 16 atoms + Scat. region 80 atoms + Right 16 atoms + Left > 16 > >> >> atoms + Scat. region 80 atoms + Right 16 atoms + ... > >> >> > >> >> I successfully calculated the current in this case for a particular > >> bias. > >> >> Then, I increased the length of the vector along the z direction by > >> about > >> >> 3 > >> >> Angs and repeated the calculation. Note that I only increased the > >> >> length > >> >> of > >> >> the vector and did not add any atoms. All the other settings in the > >> >> .fdf > >> >> file was kept intact. In this case, my system looked like the > >> >> following: > >> >> > >> >> ... + Vacuum of about 3 Angs + Left 16 atoms + Scat. region 80 atoms > + > >> >> Right 16 atoms + Vacuum of about 3 Angs + Left 16 atoms + Scat. > region > >> >> 80 atoms + Right 16 atoms + ... > >> >> > >> >> Interestingly, I found almost identical current values as the ones > >> >> obtained in the case with no vacuum. Finally, I increased the void to > >> >> 10 > >> >> Angs without modifying any other settings. In this case, although the > >> >> DM > >> >> did not converged in 1000 iterations, the code calculated currents > >> >> similar > >> >> to the ones obtained for no vacuum but with about 40 percent > >> >> deviation. > >> >> My > >> >> questions are the following: > >> >> > >> > I would not recommend that you _only_ increase the vacuum. Consider > >> > that > >> > the initial guess for transiesta is that of a SIESTA calculation. When > >> you > >> > enter transiesta the first iterations "sees" a bulk hamiltonian where > >> > the > >> > electrodes exists, however the initial guess from SIESTA is that of a > >> > vacuum just of the electrode. > >> > That is very unphysical to start converging a system with such > >> > deviating > >> > boundary conditions. > >> > > >> > The reason that the first one converges may well be that the initial > >> guess > >> > is closer than the 10 ang case, and hence it will converge to the same > >> > Hamiltonian. > >> > > >> >> > >> >> 1- Does the neighboring scattering regions affect each other in the > >> >> scattering region calculation? What is the difference between > >> >> calculations > >> >> which do not include any vacuum with calculations which consider, say > >> >> a > >> >> 10 > >> >> Angs, vacuum? > >> >> > >> > No. They do not per see effect each other, however the periodicity > >> > allows > >> > the initial guess to be somewhat close to the open boundary conditions > >> > imposed by transiesta. > >> > Hence if they do mean something, you have set up your system > >> > incorrectly. > >> > > >> >> > >> >> 2- How should I make the code converge when I increase the void to 10 > >> >> Angs? How should I modify my .fdf code in the case with vacuum so > that > >> >> I > >> >> can obtain identical currents to the case with no vacuum. > >> >> > >> > You should add buffer atoms. > >> > The reasoning and guidelines for this are quite simple: > >> > 1. Your device region electrodes should behave as bulk electrodes. > >> > 2. This means that you need enough layers (on BOTH sides of the > >> electrode) > >> > to satisfy a "bulk"-like Hamiltonian. > >> > 3. If you have different types of electrodes, add buffer atoms. > >> > 4. Buffer atoms does not cost anything in transiesta (however, they do > >> cost > >> > something in terms of internal data structures), so you can be > generous > >> > here if your system is not that big. > >> > 5. If you do add buffer atoms you are almost certainly (I would highly > >> > recommend that you use it!) required to utilise the flag > >> TS.UpdateDMCROnly. > >> > Please read about this flag in the manual, with the above information > >> > you > >> > should be able to understand which value it should be set to, AND why > >> > it > >> is > >> > important. :) > >> > > >> >> > >> >> Sorry for my bother-making. I am trying to simulate a system with > >> >> TranSIESTA which contains quite different left and right electrodes > >> >> that > >> >> cannot come into direct contact without the scattering region in > >> between. > >> >> The only way for me is to introduce vacuum between neighboring > >> scattering > >> >> regions. In brief, I would be really grateful if you tell me how can > I > >> >> determine the size of the sufficient vacuum > >> >> between neighboring scattering regions so that my results can be > >> >> credible. > >> >> > >> >> Best wishes, > >> >> Mohammad, > >> >> > >> > > >> > > >> > > >> > -- > >> > Kind regards Nick > >> > > >> > > > > > > > > -- > > Kind regards Nick > > > -- Kind regards Nick
