Go to www.awrcorp.com
Back to search page Click to download printable version of this guide.

5.1.2. Using Adaptive Mesh Refinement (AMR)

The intent of adaptive mesh refinement (AMR) is to allow the simulator to create a more finely discretized mesh in more "interesting" areas of the structure in order to generate a more accurate solution. At the same time, areas that are less interesting can be discretized more coarsely to reduce the necessary resources. During AMR, a solution is computed for the current mesh and that solution is then used to refine the mesh. After iterating several times, the solution converges and triggers the final solve (if necessary) and stops.

Use of an AMR convergence criterion is enabled by default. You can disable the convergence criterion for AMR by setting the value of Enable under Convergence to False in the simulation properties. If Enable is set to True, the solution is considered to have converged when the error value for the metric indicated in Target falls below the value specified in Tolerance. If Enable is set to False, no error is considered; the solution stops iterating when the number of specified iterations is reached.

Although AMR is very powerful, the criteria it uses to refine the mesh cannot be perfect for every possible system. As a result, you should always check the AMR convergence behavior to ensure that the mesh is being refined as expected. During AMR, the simulator outputs tables that indicate the current state of the refinement process. The convergence target should decrease more or less monotonically, as the mesh is refined (this may not always be the case, but it is a good rule of thumb). In addition, a mesh annotation from a late iteration should show mesh refinement (many small elements) in sensible parts of the structure (where fields are strong, or rapidly varying, for instance). In the final mesh, the element size will likely vary substantially over the mesh.

If AMR terminates before the results have converged to your requirements, reduce the convergence Tolerance in the simulation properties AMR Sequence tab. This will typically cause AMR to run longer and produce a more accurate result. You may manually force AMR to perform more iterations, by increasing the Minimum Iterations parameter that must execute before termination. Once the minimum number of iterations has been executed, AMR will terminate after the next iteration where the convergence tolerance is met. Alternately, you may change the value in Minimum Converged Iterations, to force AMR to perform additional iterations after convergence. AMR will always terminate if it performs the number of iterations in Maximum Iterations. This protects against run-away simulations, or simulations that become much too large after excessive iteration. Adjust this value to allow the simulation to run longer, or to force it to terminate more quickly.

By default, when you save a model all of the results of each AMR iteration are compressed into the *.apz file created by the save process. You can choose to save only the results of the final ARM iteration by setting Save Only Results of Last AMR Iteration to True.

For details on AMR of a particular simulation type, see the AMR documentation sections for that simulation type in the following sections.

Please send email to awr.support@cadence.com if you would like to provide feedback on this article. Please make sure to include the article link in the email.

Legal and Trademark Notice