gradient-GLO

Gradient

The Gradient parameter (applicable to non-linear balances) gives the rate of change of the convergence, and so gives the rate at which a balance converges. It has a value between 0 and 1. A value of 1.0 gives the fastest convergence, but may cause instability. A steeper gradient (closer to 0) creates more stability in solving the convergence; however, this also reduces the step size.

If you specify a gradient that is too steep, you may reduce the step size below the minimum set value, causing Sigmafine to stop before it has solved the problem. A small step size may be needed for highly nonlinear constraints.

Sigmafine has the flexibility to deal with some step size problems. For example, if a step is larger than the previous step, Sigmafine increases the steepness of the gradient to stabilize convergence. If a step is much smaller than the previous step, Sigmafine decreases the gradient to accelerate convergence. When these situations occur Sigmafine does not use the gradient specified in the Convergence Gradient parameter.