Under relaxation
The PCG solver can optionally use an under-relaxation factor between 0.0 and 1.0 to improve convergence. In general a low value will lead to convergence, but at a slower convergence rate (i.e. with many SZ iterations). Higher values will increase the convergence rate, but at the risk of non-convergence.
Automatic (dynamic) estimation of under-relaxation factors
If the automatic estimation of the under-relaxation factor is allowed, the under-relaxation factor is calculated automatically as part of the outer iteration loop in the PCG solver. The algorithm determines the factors based on the minimum residual-2-norm value found for 4 different factors. To avoid numerical oscillations the factor is set to 90% of the factor used in the previous iteration and 10% of the current optimal factor.
The time used for automatic estimation of relaxation factors may be significant when compared to subsequently solving the equations and the option is only recommended in steady-state cases.
Under-relaxation by user-defined constant factor
This option allows the user to define a constant relaxation factor between 0.0 and 1.0. In general a value of 0.2 has been found suitable for most simulations.