If your system is not too large, I'd advise that you determine the relaxed lattice constants from a series of fixed-cell calculations, especially if the system is more or less symmetrical. Although your stresses seem quite low to me, I think that the only way to improve them might be, indeed, to lower the stress tolerance. Note that since you are using a finite integration grid, no one can guarantee that you will ever get zero stresses. So, check for the integration grid cutoff as well, you might have to increase it at some point.
2007/3/27, Yurko Natanzon <[EMAIL PROTECTED]>:
Dear siesters, I'm thinking on increasing the accuracy of calculating elastic constants (i'm interested in shear ones but it doesn't matter). First I relax a lattice to get a stress tensor zero, and then make deformations to get a new stress tensor and calculate elastic constants. But in fact the stress tensor of relaxed lattice is not zero, it has a form like this one: -0.000020 -0.000056 0.000233 -0.000056 0.000053 -0.000033 0.000233 -0.000033 0.001350 the values of this tensor in fact influence the results of elastic constant calculations. In fact, if I get +1% deformation in one direction and -1% - I may get totally different values of stress tensor. Is there a way to take the influence of zero stress tensor into account? If I'll be able to know, which strain corresponds to this stress tensor, I can find the lattice vectors of ideally relaxed lattice, and this will solve the problem. I would be grateful for any advices regarding this. Of course, I may try to set StressTolerance and Force Tolerance to some low values, but this will take much more CG steps to fully relax the lattice and much more time. Also in some cases I can't fully relax the lattice because of constraints, but need to relax only several components of stress tensor. -- Yurko Natanzon PhD Student Henryk Niewodniczański Institute of Nuclear Physics Polish Academy of Sciences ul. Radzikowskiego 152, 31-342 Kraków, Poland Email: [EMAIL PROTECTED], [EMAIL PROTECTED]
