SIGMA effect on the total energy

[zhuanghl]

I want to calculate the energy difference between the supercell(with the same shape and calculation parameters) with and without grain boundary.There is some problem I met with the SIGMA tag.As we know that there are 3 different energies in the OUTCAR file,we should take the second energy(i.e. energy without entropy)for use for the metallic system.But what the problem is that energy difference between the first energy and the second energy is different. What is worse is that some is larger than the free energy,while others are less(though the average value for each atom are both very small).Has anyone seen such problems before?In this case,I am not sure,which oneof the energy to be used?
ps:I want to get the accurate energy

[admin]

the E(sigma --> 0) should be taken if you use the Methfessel-Paxton method (ISMEAR=1 or 2). Moreover, SIGMA should be chosen carefully, | F-E | should be negligible (in the meV-range).
concerning the "larger" energies: these most probabaly are due to the N=1 or 2 polynomial fit used by the MP method. Please have a look at the k-points.pdf file (p16) of the vasp-workshop,
http://cms.mpi.univie.ac.at/vasp-workshop
for further details