Dear Artem 2013/4/26 Akshu <[email protected]>
> Dear nick > I have a question about one of your notes > > 4) I also played with relaxation of the geometry of the SR. So I used several > CG steps to equilibrate the structure. I expected, >> >> >> >> SIESTA first takes over and does the relaxation and then Transiesta >> calculates the Gree's functions upon the relaxed geometry. >> >> Instead, Transiesta calculates its stuff at each relaxation step. Is it ok? >> Why is it so? >> >> Then you just get the transiesta forces instead of the siesta forces. So > you relax against transiesta forces. You could try and do siesta relaxation > and compare. But don't relax the electrodes. > > What happens if the siesta forces are zero but the transiesta forces are > not, though the transiesta cycle converges. Is it necessary to relax > transiesta forces as well. And what do you mean by don't relax the > electrodes...please explain > I should have been more specific on the electrodes. The requirement of transiesta is to have the electrodes fixed. I.e. when you apply the self-energies they must "match" the structure where they are put, hence if you do geometric optimization without the constraint of the electrode atoms they could move away from where the self-energy "lives". As for the forces, I have never really done this my-self. But in principle you could do siesta/transiesta cycle geometry relaxation (but often this is overkill). As a rule-of thumb (if you need transmission for the relaxed structure): 1. relax geometry with siesta (separate CG-run) 2. Do transiesta on the geometry obtained from 1. (i.e. do not further relax the structure) 3. Carefully go through your out-file and check the energies, forces, etc. 4. If the forces are excessive, then you need to think about what you want to achieve. Note especially that if the forces are excessive with a bias, then possibly (note that I am not saying "only if the forces are excessive") other time and bias-dependent effects may be governing the motion/position of the atoms. In this case you need to do further studies of your system and relaxing in siesta/transiesta might not provide the full picture. Kind regards Nick > > > On 26-Apr-2013, at 1:16 AM, Nick Papior Andersen <[email protected]> > wrote: > > Dear Artem, > > I will give some short notes on your questions. > > 2013/4/25 Artem Baskin <[email protected]> > >> Dear SIESTA/TRANSIESTA users, >> >> I have recently read the paper (PRB 81, 205437) and found it very >> interesting and enlightening. Since I want to reproduce these results, >> I faced with some questions that I want to address here, some of them are >> technical >> >> >> >> >> >> whereas others are conceptual. From now on I will refer to the mentioned >> above paper and the >> materials that I found >> http://unam.bilkent.edu.tr/mt2/transiesta/agnr8-transiesta/README.txt >> >> >> >> >> >> 1) What is (are) the particular reason(-s) to choose 8AGNR as >> an electrode? In this case, the scattering region is defined in a dummy >> way since the only difference between scattering and electrode regions is >> >> >> >> >> >> for which area the charge neutrality is preserved and how many atoms are >> treated out of equilibrium. Am I right? >> >> This is a structural choice, you want the electrodes to represent a > semi-infinite system which you put in contact with the scattering region. > The electrode atoms should be as close (in electronic structure) to the > bulk semi-infinite electrode (as well explained in Brandbyge et al. paper: > DOI: 10.1103/PhysRevB.65.165401) > I dont understand what you mean by "scattering region is defined in a > dummy way". > >> >> 2) In the example >> (http://unam.bilkent.edu.tr/mt2/transiesta/agnr8-transiesta/README.txt) I >> found >> >> TS.NumUsedAtomsLeft/Right 16 >> TS.BufferAtomsLeft/Right 16 >> >> Is there any reason why to use these values? >> >> It is the authors of the papers choice. Buffer atoms are also well > explained in the paper of Brandbyge et al. They are simply not considered > in the transiesta SCF. > >> >> 3) I was trying to reproduce some of the results. I took 17AGNR as an >> electrode with a standard >> >> >> >> >> unit cell (38 atoms). As a scattering region, I choose the same 17AGNR of 4 >> and 6 unit cells long. I have arranged carefully >> >> the atoms in the scattering regions so that first and last 38 atoms >> correspond to the electrode regions. To my surprise, >> >> >> >> >> for the scattering region with 4 unit cells I got a pretty decent resulting >> transmission spectrum (step like) whereas for >> the scattering region with 6 unit cells the transmission spectra is just a >> set of delta-functions. Does anybody have a clue >> >> >> >> >> >> why is it so? Moreover, when transiesta takes over and calculates the >> Green's functions I see a very strange charge population >> on atomic orbitals (H atoms at the edges get zero or even negative >> charge!!!!) How it can be ok for the smaller SR and >> >> >> >> >> >> terribly weird for just a two unit cells bigger SR? Do I need to worry about >> parameters defining complex contour integration >> options (I double checked the TS.ComplexContour.Emin to make sure that it is >> below the min eigenvalue in both cases)? >> >> >> >> >> >> I cannot really comment on this case as I haven't made any of these > calculations myself, have you read Ref. 24 in the paper you mention? > You could also try to increase the electrode size in the z-direction, take > one more dimer, for instance, on both sides (in fact I would probably make > it two dimers bigger). > >> 4) I also played with relaxation of the geometry of the SR. So I used >> several CG steps to equilibrate the structure. I expected, >> >> >> >> SIESTA first takes over and does the relaxation and then Transiesta >> calculates the Gree's functions upon the relaxed geometry. >> >> Instead, Transiesta calculates its stuff at each relaxation step. Is it ok? >> Why is it so? >> >> Then you just get the transiesta forces instead of the siesta forces. So > you relax against transiesta forces. You could try and do siesta relaxation > and compare. But don't relax the electrodes. > >> >> 5) Did anyone try to evaluate the Fermi wavelength for such a system as >> compared to the width of the ribbon to make sure that we deal with a >> >> >> >> >> >> truly 1D system? >> >> >> 6) Due to the specific choice of the electrode for the set up (the middle >> >> >> >> >> part coincides with the leads) one may interpret the transmission spectra >> using the band structure of the electrode region only. From that >> >> perspective it is unclear for me how one would explain the conductance >> profile for the non zero bias. Moreover, what if the electrodes were of >> the different material, say, gold or nikel? >> >> 7) Another technical issue arises when I try to apply this "approach" to >> >> >> >> >> >> the case when the electrode are made of different material as compared to >> the scattering region. The Siesta Manual says: "it is also crucial that >> the atomic positions specified at the left (right) EL calculation must be >> >> >> >> >> >> equivalent to the left (right) electrode part of the SR set up. Here, >> equivalent means that they can be made equal by a simple translation in >> space." In the examples provided in Siesta and in the case of paper the EL >> and SR >> >> >> >> >> >> are the same, that is why it is easy to provide the equivalence of the >> atoms of the electrodes and those of the SR just multiplying the first two >> lattice vectors for the EL by integer number(s). But what if the lattice >> >> >> >> >> >> constants for EL and SR are not commensurable, how can I get the >> equivalence of the atomic positions for EL and SR? What is the "electrode >> part of the SR"? >> >> I am not quite sure about your question, the electrode coordinates have > to be the same. If I understand you correct, then you have to multiply your > system until the unit-cell vectors are equivalent for some integer values. > >> >> 8) By definition, the whole set up in reality should not have the >> >> translation symmetry, the scattering region does not have translational >> invariance. Yet, in TRANSIESTA set up we have to apply the periodic >> boundary conditions in the xy-directions, moreover, we need to apply the >> >> >> >> >> >> periodic boundary conditions along the current direction (z). Can anyone >> comment on this? Why do we need them? If I want to analyse (and visualise) >> the wave functions of the scattering region then I have a problem to >> >> >> >> >> >> decide for which k-point in 3D Brillouin-zone I need to construct the wf >> of a given eigenvalue. It is very counter-intuitive to consider the >> scattering region as a chain (or even bulk) rather than a "super-molecule" >> >> >> >> >> >> with molecular orbitals as it is should be. >> >> Transiesta k-points are only considered in the transverse direction. So > kz=0. for transiesta SCF. > >> >> 9) Could you generally comment the meaning of >> >> TS.NumUsedAtomsLeft/Right >> TS.BufferAtomsLeft/Right >> TS.ReplicateA1(A2)Left/Right >> >> >> >> parameters in terms of their physical meaning and geometry of the set up? >> >> The geometry setup follow this order > > 1. Left buffer atoms (not used in transiesta) > 2. atoms used in the left electrode TSHS x ReplicateA1 x ReplicateA2 > 3. Scattering region > 4. atoms used in the right electrode TSHS x ReplicateA1 x ReplicateA2 > 5. right buffer atoms (not used in transiesta) > > I believe I have made a reasonable explanation of the replicate feature in > the manual. > In short, the buffer atoms can be used to match non-equal electrodes in > the left/right side. > >> >> >> 10) It is known that <SystemLabel>.TSDE contains the info about the >> non-equilibrium density matrix. Manual says that .TSDE file can be used to >> analyse the non equilibrium charge density with the aid of DENCHAR. >> >> >> >> >> >> However, Denchar needs .DM file but not .TSDE. I renamed .TSDE file by .DM >> file to see the difference but I did not see any difference between >> equilibrium and non equilibrium charge density even when I applied bias of 2 >> or >> >> >> >> >> >> more Volts. Have you tried to do this, if so, please, share with me what >> you got and what we should get. >> >> This is system dependent. If you have only tried one system, then try on > another system. > >> >> >> If anyone could comment on at least one of the issues listed above, I would >> be very obliged. >> >> >> >> >> >> Best, >> >> Artem Baskin >> >> >> > Kind regards Nick > >
