A semi-algebraic look at Sequential Quadratic Programming (SQP)
Event details
| Date | 09.11.2015 |
| Hour | 15:15 |
| Speaker | Jérôme Bolte |
| Location | |
| Category | Conferences - Seminars |
The study of the behavior of seqeunces generated by Sequential Quadratic Programming (SQP) or Sequential Convex Programming method (SCP) is a difficult matter: sequences can oscillate, active constraints seem to have an unpredictable behaviour, cluster points are most of the time KKT points but oscillations can still occur, value convergence can be arbitratril slow.. All these counter-intuitive phenomena can occur with indefinitely smooth data and nice qualifcation conditions, yet they do not occur in real life perhaps because of some numerical obstruction.
How can we explain this discrepancy between theory and practice? The object of this talk is to show that semi-algebraicity (subanalyticity / definability) gives an elegant answer to this apparent contradiction.
Joint work with E.Pauwels (Toulouse, France).
How can we explain this discrepancy between theory and practice? The object of this talk is to show that semi-algebraicity (subanalyticity / definability) gives an elegant answer to this apparent contradiction.
Joint work with E.Pauwels (Toulouse, France).
Practical information
- Expert
- Free