On Tue, 21 Nov 2006, SuiYang wrote: | Dear SIESTA users: | I have used SIESTA to study the property of FePt systems.Here I have a question about the atomic forces SIESTA calculated: | When I am calculating the FePt unit cell(including 4 atoms) with its lattice constant around 4 ang.The corresponding atomic forces are around 0.02 eV/ang.When I am using the 2x2x2 supercell(including 32 atoms) as unit cell, with relatively the same atomic positions,and the lattice constant is changed to 8 ang to maintain it is the same bulk as the previous one,however ,the atomic forces are increased to around one hundred eV/ang. | | I also calculated another system using 3x3x3 supercell as unit cell, the forces are much even higher ,it seems the atomic force will increase with the number of atoms and the lattice constant. | | I am rather confused with its behavior,why there is such a drastic change in atomic forces and stress tensor while the bulk is the same, only the expression of it is different. | | I've attached both 1x1x1 and 2x2x2's input and output files,if needed. | I really need your help. |
Dear SuiYang, As Riccardo pointed out, you are doing calculation with the default setting for k-points, that is just one k-point. This is no good when calculating bulk - in any case, metals. You can see that Siesta fails to converge your system - which is in fact metal with partially filled bands, - when using just one k-point and hence no dispersion; it just desperately swaps the Fermi energy back and forth: siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax Ef(eV) ... siesta: 23 -3027.3747 -2953.1394 -2953.2110 0.8298 2.5907 ... siesta: 98 -3034.0919 -2952.4118 -2952.4146 0.8258 -2.0656 siesta: 99 -3027.3736 -2953.3750 -2953.4466 0.8258 2.5911 Normally you should be wondered by this behavior before going any further... After you manage to get reasonable convergence of electronic properties, take into account that the forces are especially sensitive to the k-mesh, and (if you really need good forces, i.e. for phonons) you should check how they converge as you increase the number of k-points. Your forces are zero in your 4-atom cell, because the positions of atoms are symmetric with respect to the real-space mesh (defined by the Mesh cutoff, 200 Ry in your case), and the errors cancel. In order to see how good/bad the forces REALLY are, you can displace the atoms a bit from the symmetric positions, imposing the uniform shift (AtomicCoordinatesOrigin). In your 2x2x2-cell you have apparently the atoms situated "at random" with respect to the mesh along the Y-direction, and you see huge fluctuations of forces, which - in principle, i.e. apart from numerical errors - should be zero by symmetry: siesta: Atomic forces (eV/Ang): 1 0.000001 94.720964 0.000006 2 -0.000004 -94.720975 0.000003 3 0.000004 -117.510866 -0.000002 4 -0.000005 117.510859 0.000000 I must confess I don't understand why X and Y behave differently, avan as they are completely equivalent with respect of atomic positions and mesh construction. I'd be glad to see somebody's comment on that. You did't need to provide .fdf separately because they are dumped at the beginning of each Siesta output anyway. I don't know what you are after in your study, but if I remember correctly, in real life FePt is magnetic, so that when you switch on the spin, this will affect your forces as well. Hope this helps, Andrei Postnikov +-- Dr. Andrei Postnikov ---- Tel. +33-387315873 ----- mobile +33-666784053 ---+ | Paul Verlaine University - Institute de Physique Electronique et Chimie, | | Laboratoire de Physique des Milieux Denses, 1 Bd Arago, F-57078 Metz, France | +-- [EMAIL PROTECTED] ------------ http://www.home.uni-osnabrueck.de/apostnik/ --+