Iterations Panel

Relaxation

Relaxation slows iterative progress toward a solution reducing the destabilizing effect of non-linear systems. By default, the Solver implements an adaptive relaxation scheme that reduces the relaxation dynamically in response to highly non-linear features of the problem. Relaxation can be directly specified in the Iterations panel.

When applied to the flow solution, relaxation r is used as follows:

A relaxation of 1 is the same as no relaxation. Each iteration's prediction is used directly in the next iteration.

A relaxation of 0 is not allowed, as this would prevent the parameter from changing.

  • Flow Rate Relaxation - For highly non-linear systems, a flow rate relaxation of 0.1 is advised. Experience shows that using flow rate relaxation factors between 0.5 and 1 has limited usefulness. If a flow rate relaxation of 1 does not work, you should generally try flow rate relaxation parameters between 0.1 and 0.5. In more extreme cases, a flow rate relaxation factor smaller than 0.1 may be required.

  • Pressure Relaxation - Because of the nature of the matrix solver, experience indicates that if pressure relaxation is used, it should always be set to 0.5 or 1 to maintain stability.

Iteration Limit

  • Maximum Iterations - 50000 by default, sufficient for the vast majority of models. Most models converge in a relatively small number of iterations. This limit does not directly affect the solution.

  • Iteration Span for AutoDetect - If the Solver detects that no meaningful progress has been made to convergence within this many iterations, relaxation is automatically lowered. This also automatically adjusts tolerance.

Matrix Method

Network analysis requires solving complex matrices potentially thousands of times. Different methods may solve faster or slower, but have no bearing on results. The default method of Gaussian Elimination is very robust and usually provides the fastest convergence.

Iteration History

Iteration History shows a history of which pipe or junction is most out of tolerance, which is helpful for troubleshooting.

Related Blogs

Having model convergence problems? Start here!