Portal de Periódicos da CAPES

Concurrent Cuba

2016; Elsevier BV; Volume: 207; Linguagem: Inglês

10.1016/j.cpc.2016.05.012

1879-2944

Thomas Hahn,

Mathematical Approximation and Integration

The parallel version of the multidimensional numerical integration package Cuba is presented and achievable speed-ups discussed. The parallelization is based on the fork/wait POSIX functions, needs no extra software installed, imposes almost no constraints on the integrand function, and works largely automatically. Program title: Cuba Catalogue identifier: ADVH_v4_2 Program summary URL:http://cpc.cs.qub.ac.uk/summaries/ADVH_v4_2.html Program obtainable from: CPC Program Library, Queen's University, Belfast, N. Ireland Licensing provisions: yes No. of lines in distributed program, including test data, etc.: 30720 No. of bytes in distributed program, including test data, etc.: 729941 Distribution format: tar.gz Programming language: C. Computer: Designed for: all platforms with an ISO C99 C Compiler. Tested on: x86-Linux/gcc, x86-MacOS/clang, x86-Windows/Cygwin-gcc. Operating system: Linux, Mac OS, Windows(Cygwin). Has the code been vectorized or parallelized?: Yes, Parallelized RAM: 1M words Classification: 4.11. Does the new version supersede the previous version?: Yes Nature of problem: Multidimensional numerical integrations, e.g. of phase spaces. Solution method: The Cuba library contains the four algorithms Vegas, Suave, Divonne, and Cuhre with the following characteristics: RoutineBasic integration methodAlgorithm typeVariance reductionVegasSobol quasi-random sampleMonte CarloImportance samplingSuaveSobol quasi-random sampleMonte CarloGlobally adaptive subdivisionDivonneKorobov quasi-random sample or Sobol quasi-random sample or cubature rulesMonte Carlo Monte Carlo deterministicDeterministic stratified sampling, aided by methods from numerical optimizationCuhreCubature rulesDeterministicGlobally adaptive subdivision Reasons for new version: Parallelization Summary of revisions: Version 4.2 adds parallelization based on the fork/wait POSIX functions. This means that no extra software needs to be installed, almost no constraints are imposed on the integrand function, and the parallelization works largely automatically. Unusual features: Coherent interface in Fortran, C/C++, and Mathematica. Can integrate vector integrands. Running time: Varies greatly depending on the integrand and the chosen accuracy. Can range from seconds to days.