> dr. Roberto Pasianot, dr. Andrei Posnikov, > > Thanks for the replies.I have a question about the following: > > *"You may wish to start calculation with its value of say 300 - 600 K > but then take care to reduce it to a small enough value that > the total energy does not vary with it any more."* > > Does this mean in CG type of run and MD, once I have optimized my > structure > with low band-gap and 300K smearing, I should reduce the "electronic > temperature" to go on calculating other properties, such as polarization?
Dear Pedram: yes, you got the idea. You can look at ElectronicTemperature as at yet another "cutoff", to be used with care and checked against. For example: you can converge your calculation with one k-point, but before extracting the final result you may (should) wish to increase this number and look how the "meaningful" results are affected. Similarly with ElectronicTemperature: its elevated value helps the convergence but spoils the results. It has any effect at all if you have a metal or narrow-band semiconductor, i.e., when occupied and unoccupied states are close enough, within your smearing value. "Other properties" are affected inasmuch as they depend on the integration over the Brillouin zone. You ask about polarization; this normally means that you have a substantial band gap, probably much broader than the ElectronicTemperature you will use. Summarizing: normally you calculate the polarization (or, other "final" properties) on a well converged case, when the convergence is no more an issue and you can (should) set the ElectronicTemperature to very small value without any problem. Best regards Andrei Postnikov > Regards, > Pedram > > > On Thu, Sep 5, 2013 at 7:48 PM, Roberto Pasianot > <[email protected]>wrote: > >> Hi Pedram, >> >> MD temperature is "reaĺ", thermodynamic temperature. >> Electronic temperature is just a smearing parameter for the >> Fermi surface, that easies the computation of integrals in >> reciprocal space involving occupied states. You could use >> the MP scheme instead of the FD one, thus loosing connection >> with any statistical distribution. Or even choose zero, with >> possible horrible effects on SCF convergence. >> >> Bye, bye, >> >> Roberto >> >> >> >> On 09/05/2013 11:41 AM, pedram heidari wrote: >> >>> Hi, >>> >>> I have problem with temperatures in MD Anneal in Si.fdf that is present >>> in >>> "tests" directory in siesta code. (I have enclosed the .fdf) >>> >>> There, I can see the electronic temperature is 300K and two other >>> parameters as: >>> >>> MD.InitialTemperature 100 K >>> MD.TargetTemperature 600 K >>> >>> To my understanding, the electronic temperature comes from the >>> fermi-dirac >>> distribution and deals with the valence "electrons" but MD temperatures >>> come >>> from Maxwell-Boltzmann distribution for "atoms" and should consider the >>> role of >>> the nuclei as well, if I am not mistaken. >>> >>> What is exactly the point of mentioning the temperature in two >>> different >>> distributions in one MD simulation? >>> >>> Thanks, >>> Pedram >>> >>> >> >
