Dear Andrei, thank you for quick reply.
I have some experience with AIMPRO which is also based on atom-centered basis functions, but it does not have parameters like EnergyShift, EnergySplit, nothing similar. What we usually did is we calculated theoretical bulk lattice constant and then used it in our slab calculations. Not like you suggested: vary EnergyShift so that theoretical and experimental lattice constant coinside. You say that I should care about total energy differences convergence. Does it mean that the choice of exact EnergyShift is not so important? Does the total energy difference only weakly depend on EnergyShift? You see, my question is practical, how should I choose EnergyShift so that I could calculate total energy difference correctly? Kind regards Ruslan 2008/12/1 <[EMAIL PROTECTED]> > > Dear SIESTA users, > > I have difficulties trying to converge calculation with respect to > > PAO.EnergyShift. > > The system is 8 layers silicon slab, hydrogenated. > > The total energy varies strongly with PAO.EnergyShift parameter: > > PAO.EnergyShift Etot > > 300 meV -21007.7361 eV > > 250 meV -21012.1582 eV > > 200 meV -21017.8688 eV > > 150 meV -21019.6457 eV > > 100 meV -21022.2932 eV > > 75 meV -21022.4298 eV > > 50 meV -21021.1673 eV > > 25 meV -21020.9682 eV > > > > and for values less than 75 meV Etot begins to growth. > > Is this behaviour normal? > > Should I choose the PAO.EnergyShift which minimizes Etot? > > The change in Etot vs PAO.EnergyShift is much stronger than when I > > change other technical parameters, > > for example MeshCutoff. I am afraid if this can cause large calculation > > error in my results? > > > > Regards > > Ruslan > > > > Dear Ruslan, > I dared to produce a lengthy comment to the subject which seems important > and often misunderstood. > > 1. The role of PAO.EnergyShift is to tune the localization > of your basis functions. In principle - as it probably is in any other code > with atom-centered basis functions - their choice is completely > up to the user. You can draw them by hand and feed into Siesta, > or construct in any other way. Of course the procedure is important, > and the Siesta default - pseudoatomic orbitals - is, on the average, > an extremely good choice, but not the a priori unique one. > Speaking about this particular choice, the quality of the basis > (i.e., the number of functions you use) is probably more important > than the exact value of PAO.EnergyShift. Deciding on whether your basis > (its dimension, localization) is good by just looking at the absolute total > energy values makes relatively few sense. > What you should look at are THE PROPERTIES you are ultimately after, > which properties are typically total energy DIFFERENCES. > About the convergence of these differences you should care, > to the accuracy which is relevant to your particular problem in question. > > 2. The behaviour your obtain (of total energy vs. PAO.EnergyShift) is > a correct textbook behaviour; however, you won't probably learn much > useful from it. What happens: you set your atoms at a certain fixed > distance. > The chemical bonding will mostly play at a certain range between > the atoms. As you vary your PAO.EnergyShift, your basis functions > with their nodes smoothly drift through this region, yielding more > and then again less variational freedom at distances where the chemistry > mostly needs it, and more variational freedom results in lower total > energy. > So in a bonded system at fixed interatomic distance, > as you vary PAO.EnergyShift, you'll probably always find a minimum. > (Note that for an isolated atom you won't have it: the smaller > PAO.EnergyShift > the easier the confinement the lower the total energy.) > > 3. Why I say that you won't probably learn much useful from it: you do > this simple test with a fixed interatomic distance. However, in real life > you'd probably like at some point to optimize your structure. So yes, > it might seem a good idea first to tune your basis to the experimental > interatomic distance and then "predict" a good lattice constant > with the thus tuned basis. But don't be too enchanted by this possibility. > > 4. A BIG difference between changing PAO.EnergyShift and changing > "other technical parameters, for example MeshCutoff" - I'd explicitly > use k-mesh to it - is the following. I repeat that the choice of basis > is your free choice: you can make it based on your experience and > common sense; in some cases you'd need ultimate precision, in other you'd > need very large basis extension, in yet others you would trade some > precision > to reasonably minimize your basis, or whatever. Whereas MeshCutoff and > k-mesh cutoff allow SYSTEMATIC verification of their convergence (within > actual necessary limits of the accuracy of total energy differences), > whatever your choice of basis, and this verification must be always done, > as a matter of routine hygiene. > > Best regards, > > Andrei > > --- Dr. Andrei Postnikov ---- > Paul Verlaine University - Institute de Chimie, Physique et Mat'eriaux, > Laboratoire de Physique des Milieux Denses, 1 Bd Arago, F-57078 Metz, > France > ---------------------------------------------------------------------- >
