submatrix_energy_cutoff
A.K.A. – The importance of the initial guess for the wavefunctions
Use 1/5 of the total energy cutoff with a maximum of about 10 or 12.
Examples
Ex.1 using the direct minimization method
Ex.1 using the direct minimization method
This is the C-8layer system. Fig. 1 shows that for this sytem the value of submatrix_energy_cutoff
is very important. The value of submatrix_energy_cutoff = 10 works well for all three methods. As
seen in other plots, the “dir.gcg” method does the best. If submatrix_energy_cutoff = 10 is used, the convergence degrades so much that this setting is unusable. The problem seems to be that the initial guess for the wavefunctions is so far off. As one proceeds with the minimization, the path falls into a valley (in some but not all directions) from which it is hard to escape.
Fig. 1 convergence of direct minimization
methods
Ex. 2 SCF methods
Using a Carbon surface as the system, we again can see the importance of the 1/5th criteria.
Fig. 1 convergence of direct minimization methods
