[Smeagol-discuss] Convergence problem

[Ivan Rungger]

Hello Zeila,
> I'm always asking smeagol to converge the DM down to 1e-6 and - in 
> most of the cases -  the convergence is achieved.
That's interesting. Well OK so it seems that it is doable.
> Saying that 1e-6 is too small for smeagol do you mean that 
> - in smeagol there are weaker tolerances and, hence, it doesn't make 
> sense to ask for 1e-6?
> - or that it is too difficoult, while still makes sense numerically, 
> to achieve it? 
Numerically it should be OK, but it depends on the system. Generally the 
achievable tolerance will just be the same as in a siesta DFT 
calculation. The potential on the real space mesh is stored as single 
precision in siesta, so I think 1e-6 is probably very close to the 
numerically achievable accuracy. From the physical point of view once 
you get to 1e-4 usually the results (i.e. transmission, pdos, current) 
don't change when you converge further.
> When I had convergence problems, I had oscillations of the dDmax among 
> something like 0.00n and 0.000n
Once it is down to 0.0001 the results should be OK (0.00n is too high I 
think). What one should do is to calculate the transmission at 0.000n 
and then converge it further to 0.0000n, and see if it is different. 
Usually it should not change significantly. Usually if it oscillates at 
0.000n it helps to stop and restart the calculation with a bigger mixing 
(because it is already close to convergence).

Cheers,

 Ivan