Improved diagonalisation

As was noted above, the matrix diagonalisation step is not usually the ratedetermining step in an SCF calculation. However, methods of improved diagonalisation and interpolation are still important since for many systems the calculation is not nearly so well-behaved as our testbench and the iterative procedure is often either very poorly convergent or even oscillatory after many cycles. In these cases it is wise to retain the interpolation procedure for the whole calculation. 

Efficient methods of selection of admixed configurations in this approach are described in [12]. The computer code for generating the reduced configuration state list is presented in [13]. A special diagonalisation method is developed to reduce the orders of matrices to be diagonalised [14]. A more detailed overview of the methods considered may be found in [15]. All these improvements considerably speed up the calculations and widen the domains of applicability of the methods developed. 

What is more important from the point of view of the general design of codes for the SCF method is the way in which eigen (or its replacement) is used. If we can make an improvement in the design of the SCF process, this will make the use of any diagonalisation method more effective. 

---