Abstract
When using gradient-based optimization methods for nonlinear dynamical systems, the required computational time primarily depends on the underlying gradient computation. Optimization toolboxes usually provide default gradients by numerical differentiation, whereby in most cases finite differences are utilized. The perturbation by finite differences must not be chosen too large or too small. To avoid problems with a too small perturbation, the finite difference step can be carried out in the direction of the complex axis alternatively. However, both methods have the disadvantage that they require a new solution of the dynamical system for each gradient entry. Analytically derived gradients overcome this issue, but require more effort in terms of derivation and implementation. Two of the most commonly applied analytical approaches are outlined, which are direct differentiation and the adjoint method.
| Originalsprache | Englisch (Amerika) |
|---|---|
| Publikationsstatus | Veröffentlicht - 2026 |
| Veranstaltung | The 8th International Conference on Multibody System Dynamics - IMSD 2026 - University of Sevilla, Sevilla, Spanien Dauer: 16 Juni 2026 → 19 Juni 2026 https://www.imsd2026.es/ |
Konferenz
| Konferenz | The 8th International Conference on Multibody System Dynamics - IMSD 2026 |
|---|---|
| Land/Gebiet | Spanien |
| Ort | Sevilla |
| Zeitraum | 16.06.2026 → 19.06.2026 |
| Internetadresse |
Fingerprint
Untersuchen Sie die Forschungsthemen von „On the Comparison of Gradient Computations for Optimal Control Problems in Multibody Dynamics“. Zusammen bilden sie einen einzigartigen Fingerprint.Zitieren
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver