Skip to main content
European Commission logo print header

Continuous Constraints: Updating the Technology

Deliverables

Seventeen modules are have been developed so far: Basic Splitter, the BCS Inference Engine, Best Point, Check Box, Check Infeasibility, Check Number, a wrapper for ILOG CPLEX, interval constraint propagation, a wrapper for DONLP2-INTV, Exclusion boxes using Karush-John conditions, Hessians, Karush-John conditions generator, Linear relaxation generator using slopes, LP solve, Simple Convexity module for bound-constrained minimization STOP, a wrapper for DASH Xpress, Point Verifier.
The COCONUT environment is a modular solver environment for nonlinear global optimisation problems with an open-source kernel, which can be expanded by commercial and open-source solver components (inference engines). The open design of the solver architecture, and its extensibility to include both open source modules and commercial programs, was chosen in the hope that the system will be a unique platform for global optimisation in the future, serving the major part of the community, bringing their members closer together.
The benchmarking suite is a comprehensive collection of over three hundred constraint satisfaction and around one thousand optimisation problems from academic, industrial and real-life domains. Executable versions of these test problems, as well as information on their sources, are publicly available on the COCONUT web site. The problems range in difficulty from easy to very challenging, their sizes vary from a few variables to over 1000 variables. These test problems come from both the research literature and a wide spectrum of applications, including chemical engineering (pooling and blending, separation, heat exchanger network design, phase and chemical equilibrium, reactor network synthesis, etc.); computational chemistry (including molecular design); civil engineering; robotics; operations research; economics (including Nash equilibrium, Walrasian equilibrium, and traffic assignment problems); finance (portfolio optimisation); multi-commodity network flow problems; process design; stability analysis; VLSI chip design. The problem suite also incorporates as integral part most problems from the CUTE test collection (covering among others the Argonne test set, the Hock and Schittkowski collection, the Dembo network problems, the Gould quadratic programs, etc.), from the handbook of test problems in local and global optimisation, from the GLOBAL Library, and from the Numerica test problem collection; in addition, it contains many other constraint satisfaction problems from the literature and the world wide Web. Some problems (e.g. linear or convex quadratic programs) are known to be solvable by local optimisation alone. They were retained in the benchmark to be able to measure the overhead, which global solvers have in order to prove optimality, compared with local solvers who assume the convex structure from the outset and hence are usually significantly faster. As far as reasonable, all problems were checked for correctness; inconsistencies with information available from other sources were removed if possible. Information about (approximate) solutions and putative global minimizers and minima were provided in all but a few cases (where runtime constraints became active). All test problems are coded and can be downloaded in the AMPL modelling language.

Searching for OpenAIRE data...

There was an error trying to search data from OpenAIRE

No results available