Automated decision support for compilation and simulation flags

[Francesco Casella]

Small models can be simulated with default choices and the DASSL solver with a tight enough tolerance, producing correct results in a reasonable amount of time.

As the problem size increases, selecting the correct code generation strategy and solution strategy might become crucial to obtain good (or even acceptable) performance.

Features like stiffness, size, presence of many and/or large nonlinear strong components, presence and quantity of events, sparsity of the DAEs and of the ODEs jacobian are all elements that need to be considered for a correct choice. Most of these features could be detected automatically by running a diagnostic simulation with DASSL, then a report could be generated suggesting better flag settings.

This seems a nice topic for a thesis. We have a large number of test cases in our coverage tests, that could be used as a benchmark. AFAIK, no other Modelica tool has anything like this.

Comments?