Comparison of Interval C/C++ Libraries in Global Optimization Rafal Dąbrowski * and Bartlomiej Jacek Kubica *1 * Warsaw University of Technology, Institute of Control and Computation Engineering, Warsaw, Poland, e-mail: R.J.Dabrowski@stud.elka.pw.edu.pl Abstract. This article contains results of examination and comparison of some most popular interval libraries. Comparative analysis was based on investigating the efficiency of computations of simple functions and efficiency of the Interval-Branch-And-Bound global optimization method. 1 Introduction Interval methods (see e. g. [2], [3], [6]) are a robust tool of global optimization. Thanks to the use of outward rounding when computing interval enclosures on the objective, the interval branch- and-bound method is guaranteed to find all global minima (and distinguish them from local ones unless they are epsilon-optimal). This power comes at a price – interval methods are in general very computationally intensive, which results in long computation times and high memory demands. Because of that the efficiency is crucial for successful use of interval algorithms. And efficiency is affected not only by the implementation of the branch-and-bound method, but also low-level underlying procedures. As there are several publicly available interval libraries nowadays (e. g. [4], [5], [7], [8], [9]), it is useful to investigate and compare their efficiency to ba able to choose a proper one for a specific problem. The paper is organized as follows. First we describe four publicly available interval libraries – PROFIL/BIAS, FILIB++, C-XSC and MPFI presenting their basic features. Then we compare the results of a branch-and-bound algorithm implemented in all of them for a few typical nonconvex optimization problems. Obviously, examined functions have many variables and contain in some cases many local minimizers. 2 Interval libraries Interval libraries provide basic interval operations and – optionally – some more advanced procedures. In this paper four of such libraries are considered: 1. PROFIL/BIAS 2. FILIB++ 1 The research has been supported by the Polish Ministry of Science and Higher Education under grant N N514 416934.