I am using dynamic assignment for a network and have performed several runs, but the model never achieves convergence. Do you have any suggestions?

Here are two general tips for facilitating convergence.

1) Click Traffic > Dynamic Assignment > Parameters.
- On the Files tab, scale the total volume to e.g. 20%.
- On the Choice tab, activate 'Avoid long detours' and 'Limit number of paths', and for the max. number of paths per parking lot relation, decrease the default value of 999 to a value that make sense for the spatial scope of your network.

2) Click Simulation > Parameters.
- The random seed increment should be set to 0 for convergence runs. If the model has already converged and only evaluations are performed, the random seed increment should be > 0.
- Change the dynamic assignment volume increment to e.g. 5%. For more information on the dynamic assignment volume increment, click Help > PTV Vissim Help and refer to the section 'Defining simulation parameters'.

Other useful tips:
- Select only one of the three convergence criteria; preferably, choose 'Travel time on paths'.
- If you select multiple convergence criteria, there is a chance that convergence will never be reached due to the increasing requirements the convergence criteria could become too strict.
- If you select the convergence criterion 'Volume on edges', there is a chance that convergence will never be reached because the absolute number of vehicles on the highest volume links fluctuate more than on links with less volume although the percentage deviation is the same.
- Use longer evaluation intervals (>= 15 minutes): A path is converged if it converges in all time intervals.
- The default setting for convergence criterion is 95%, but depending on the network and especially the frequency of low path volumes (per interval), lower values may be necessary to meet the condition. Longer time intervals usually allow for higher percentages.
- Heavily overloaded networks typically never converge. Therefore, it may be reasonable to assume that the paths and distribution achieved from convergence runs at 70% traffic volume is similar to those at 100% traffic volume. Rather than assigning 100% of demand and never achieving convergence in a congested network, the paths could be developed using incremental increases in demand up to e.g. 70%. At this point, the demand could be fixed at 70% for convergence runs. Assuming that convergence is achieved and the paths are reviewed for suitability, the demand could be increased back up to 100% for evaluation runs.