Pre-processing sum-of-squares
A new sum-of-squares example has been made available. This example concentrates on the pre-processing capabilities of YALMIPs sum-of-squares module.
New example
Thomas Besselmann has contributed a new example on MPC for a chaotic nonlinear system, taking an LPV approach.
New release R20090505
Finally! Not much new, mainly refinements and bug fixes...
Fixed bug in problems involving several logdet terms, causing erroneous solutions when the model had to be converted to a geomean problem. Fixed bug in mixed-integer representations of absolute value. Fixed bug in mixed-integer representation of 1-norm. Fixed bug in mixed-integer representation of sign. Fixed bug in mixed-integer representation of sort. Fixed bug related to free variables in LPSOLVE. Fixed some bugs related to mixed-integer nonconvex nonlinear programming in bmibnb. Fixed bug causing a crash in some trivially infeasible geometric programming problems. Fixed a bug related to SDP constraints in PENNON. Fixed a bug related to ordering of parametric variables in multiparametric programs. Massive performance improvment in .* (element-by-element multiplication .)
Significant performance improvement in robustification of linear inequalities with uncertainties described with simple bounds. The command dual can now be applied directly to lists of constraints (generated without the set operator.). This essentially means that set is obsolete. New syntax supported for tagging (naming) constraints. Generalization of the supported uncertainty classes in uncertain models. Parametrized sum-of-squares constraints are now supported in the constraint plot functionality. Added the logistic operator. Added the convexity aware logsumexp operator. Fixed a bug in the max operator that caused problems with arguments involving Inf. Cleaned up some of the logic in the logic programming support to improve consistency (this may cause some old code to fail) Added some automatic reformulations to improve models and convexity propagation. Note that one should not rely on these conversions, but try to code the model using the structured operators from the beginning.x*log(x) is converted to an entropy expression.
log(sum(exp(x)) is converted to a logsumexp.
Robust optimization poster and paper
A poster and a related paper (Löfberg:2008b) presenting the robust optimization module has been uploaded to the Wiki.