<div dir="ltr"><div>Dear Evgeny, <br></div><br><div class="gmail_quote"><div><pre>1. Let's consider a supercell with odd number of electron with
nspin=1 (non-polarized calculation). The result should probably be
incorrect, but how would the error manifest? I'm specifically
interested in errors for electron state structure and total energy.
<br>A/ The calculation may be good if the system is not magnetic.<br>There are some other important parameters: k-points and smearing.<br>If you have a nk1 x nk2 x nk3 k-point mesh, the calculation if equivalent to<br>a calculation of the supercell of size nk1 x nk2 xnk3 with only the<br>gamma point. Hence, if the number of k-points is even, you have an equivalent even number of electrons. <br>Even if you choose an odd k-points grid, the smearing may cause the HOMO to be half occupied.<br>This is the case for a simple non-magnetic metal, e.g. Al, Cu, where you need a dense k-points grid, <br>and there wide conduction band, and the lowest half of the conduction band gets occupied. <br><br>In my experience, the incorrect solutions appear when the Fermi level intersects a narrow, non-dispersive conduction band, <br>or have an isolated HOMO level half-occupied. In these cases, applying nspin=2, and breaking the symmetry <br>with starting_magnetization, the half-occupied <br>level or band can split into an occupied and an empty band. This the case of defects in a big supercell (usually gamma point), <br>or the case of a dissociating H2 molecule. The same happens for some crystals where the Fermi level is within a narrow d-band, <br>nspin=1 gives a metallic system, and nspin=2 gives an insulating state. <br>Well, if d-levels are involved the system may still be incorrectly metallic due <br>to the self-interaction error, see the LDA+U method. <br></pre><pre>In case of doubt, and I would say, always, do a test spin-polarized calculation and see if you get a smaller total energy. <br></pre><pre><br><br>2. Let's consider a supercell A which is a half of supercell B. A
calculation with supercell B reveals a band with electron population
of 1 electron /per B (the system is paramagnetic and the calculation
uses nspin=2). Would a band with close energy manifest for a run with
supercell A ?<br><br>A/ If supercell B is the double of A, then you have an even number of electrons, unless you charge the system. <br>I think that you cannot have such band with just one electron in supercell B, <br>assuming the supercell B is not relaxed after duplication of A. <br><br>Also, note that to make A and B equivalent, for supercell A you need to duplicate the number of k-points, therefore you end <br>with the same number of occupied energy levels and the same energies, just distributed in a different fashion. <br><br></pre><pre>Best regards<br></pre><pre>Eduardo Menendez<br></pre><pre><a href="http://www.gnm.cl/emenendez">www.gnm.cl/emenendez</a> <br></pre> </div></div></div>