Dear David 2012/1/6 David Egger <[email protected]>: > Hi Riccardo, > > > On 01/05/2012 08:43 PM, Riccardo Rurali wrote: >> >> >> As far as I've understood if you have 2 you are sure things are ok, but >> having 3 does not mean things are wrong. I think (again, as far as I've >> understood) that if you have 2 the matrix is strictly tridiagonal and >> certain numerical routine can be used to do the "Green functions stuff". If >> you have 3 the matrix is strictly not tridiagonal, but this could simply >> mean that you have a few 0.000001 where you are supposed to have 0.0000. If >> you manage to converge numercially your results this should been that the >> tridiagonal routines could do their job is spite of the few 0.00001. >> In other words, 2 is a recomendation, not a strcit requirement. Now, 4 >> seems a bit too much...:) >> >> Riccardo > > > thank you very much for answering. > > So it is more a thing of numerical convenience, okay. I'm going to check if > the length of the electrode has any influence on T(E) in my setup. > > > Again, thanks for helping! > > > David
My personal experience is that before fixing this point in a calculation with some CNT system, I had really unphysical results but still convergent calculation. I don't know the transiesta implementation really in details so I cannot say what goes really wrong (wrong surface Green's function and more I guess), but if you have a "3" and you don't take care that this is really reflecting small numerical differences, you can get very wrong numbers. In other words what Riccardo says is correct, but I recommend you to be very careful on this point otherwise your results can be completely wrong! Gabriele
