Brief note on equality constraints in pure dual SAO settings
- 71 Downloads
We reflect on using equality constraints in a pure dual setting, to augment the predominant use of inequality constraints only in pure dual statements in structural optimization. Notwithstanding the fact that equalities have previously not been used in pure dual statements, their inclusion is sound from a theoretical point of view. This was already shown by Falk as early as 1965, but not really exploited in the structural optimization community. Hence, we herein mainly elaborate on some possible interesting numerical results. To overcome the difficulties associated with dual variables going to infinity if a subproblem is infeasible, we rely on the so-called bounded dual approach to achieve convergence. While the theme of this brief note is a pure dual method which can accommodate equality constraints, we depart with a sequential approximate optimization (SAO) algorithm, which provides for both equality and inequality constraints. The SAO algorithm in essence solves a sequence of approximate quadratic programs (QPs), but the approximate Hessian is diagonal. This means that the separable QP subproblems can then easily be solved using a pure dual method, which is attractive if the number of design variables n ≫ m, with m the total number of constraints. When m > n, the subproblems can efficiently be solved using a sequential quadratic programming (SQP)-like approach with an approximate diagonal Hessian matrix, but the resulting iteration paths are identical to that of the dual formulation; only the cost of solving the subproblems themselves is affected.
KeywordsDuality Equality constraints Simultaneous analysis and design (SAND) Sequential approximate optimization (SAO)
We would like to thank the anonymous reviewers of a first draft of our note for their helpful and insightful comments.
- Etman LFP, Groenwold AA, Rooda JE (2009) On diagonal QP subproblems for sequential approximate optimization. In: Proceedings 8th world congress on structural and multidisciplinary optimization, Lisboa, Portugal. Paper 1065Google Scholar
- Falk JE (1965) A constrained Lagrangian approach to nonlinear programming. PhD dissertation, University of Michigan, Ann Arbor, Department of MathematicsGoogle Scholar
- Fletcher R, Leyffer S, Toint PL (1998) On the global convergence of an SLP-filter algorithm. Technical report 00/15, Department of Mathematics, University of Namur, Namur, BelgiumGoogle Scholar
- Haftka RT, Gürdal Z (1991) Elements of structural optimization, volume 11 of Solid Mechanics and its applications, 3rd. Kluwer Academic Publishers Dordrecht, DordrechtGoogle Scholar
- Svanberg K (2011) Equality constraints and the globally convergent version of the method of moving asymptotes. In: Book of abstracts, 9th world congress on structural and multidisciplinary optimization, Shizuoka, Japan. Paper 232Google Scholar
- Toropov VV (2008) Personal discussionGoogle Scholar
- Vanderplaats GN (2001) Numerical optimization techniques for engineering design. Vanderplaats R&D, Inc. Collorado SpringsGoogle Scholar
- Wood DW, Groenwold AA, Etman LFP (2011) Bounding the dual of Falk to circumvent the requirement of relaxation in globally convergent SAO algorithms. Draft technical computing report SORG/DTCR/04/11, Department of Mechanical and Mechatronic Engineering, University of Stellenbosch, Stellenbosch,South AfricaGoogle Scholar
- Wood DW (2012) Dual sequential approximation methods in structural optimisation. PhD dissertation, University of Stellenbosch, South Africa, Department of Mechanical EngineeringGoogle Scholar
- Zhu C, Byrd RH, Lu P, Nocedal J (1994) l-BFGS-b: FORTRAN subroutines for large scale bound constrained optimization. Technical Report NAM-11, Northwestern University, EECS DepartmentGoogle Scholar