Dear Igor, your phonons at Gamma are sane (you may check the frequencies, but otherwise they seem OK). This hints that a priori there is no problem with the equilibrium geometry.
Now, you run a single cell > SuperCell_1 0 > SuperCell_2 0 > SuperCell_3 0 so I don't understand how you expect to get phonon dispersions from this. > (a supercell would otherwise do band folding and band structure would > have proportionally more bands). - right, this is how it is supposed to be. In "SIESTA way" you get force constants from supercell calculation, and then VIBRA Fourier transforms them and helps you to trace the phonon branches as if in a single cell. Note that the supercell size is the user's responsibility, and nothing will automatically prevent you from using a too small one... If in doubt (and in order to test things), try to make a doubled cell and calculate "exactly" a zone-boundary phonon... Best regards Andrei -- prof. Andrei Postnikov -- tel. +33-372749149 -- University of Lorraine - Laboratoire de Chimie/Physique - A2MC ICPM, 1 Bd Arago - BP 95823, F-57078 Metz Cedex 03, France ------------------------------------------------------------------------ ----- popov <po...@ipb.ac.rs> a écrit : > Hi, > > My Siesta calculations of phonon band structure of a spinel crystal > gives negative-energy > translational (the lowest two) bands along the Brillouin zone (see the > fig. in attachment). > > Suspecting that geometry optimized by CG in Siesta is at a saddle point, > before FC run I > optimized the geometry with various other algorithms, including those > from ASE (preconditioned > LBFGS, mdMin, GPMin) but I could not get rid of negative frequencies for > any of the optimization algorithms. The force tolerance for any of these > optimizers was 0.005 eV/Ang (default is 0.04). > > The primitive cell is rhombohedral and I use vibra for postprocessing of > the FC calculations. > Input for Vibra is similar to fdf for FC calculations, with addition of: > > SuperCell_1 0 > SuperCell_2 0 > SuperCell_3 0 > AtomicDispl 0.06 Ang > Eigenvectors True > > and I added masses of atoms in %block AtomicCoordinatesAndAtomicSpecies > in 5th column > and specified AtomicCoordinatesFormat NotScaledCartesianAng. > (should I consider InitMesh instead?) > corresponding to the real space grid step division of 5.9114/130 = 0.045 > Ang, where 5.9114 Ang > is the module of lattice vectors. I used MD.FCDispl 0.06 Ang. > > I use a primitive cell with 14 atoms for the FC calculations, i.e. I did > not make a supercell > in order to reconstruct 14*3=42 bands in band structure that is common > in literature for spinels > (a supercell would otherwise do band folding and band structure would > have proportionally more bands). > > Now I am running some additional calculations with finer MeshCutoff and > MD.FCDispl (1400 Ry, 0.02 Ang). > > I am out of options. Do you have a clue what could be cause of imaginary > frequencies of the > translational modes throughout the BZ and how to get rid of them? > > Thanks in advance, > Igor > >
-- SIESTA is supported by the Spanish Research Agency (AEI) and by the European H2020 MaX Centre of Excellence (http://www.max-centre.eu/)