NLopt release notes

From AbInitio

Revision as of 15:58, 17 June 2010; Stevenj (Talk | contribs)
(diff) ←Older revision | Current revision | Newer revision→ (diff)
Jump to: navigation, search
Release notes
NLopt manual
License and Copyright

Here, we describe what has changed between releases of the NLopt package. You can also refer to the NEWS file in the Meep package (or the ChangeLog file for a more detailed listing).


NLopt 2.0.2

17 June 2010

  • Fixed compilation failure in Microsoft Visual Studio, due to incorrect usage of __stdcall keyword; thanks to Dave Katz for the bug report.

NLopt 2.0.1

16 June 2010

  • Bug fix in Fortran API (for nlo_get_ functions returning arrays).
  • Fixed buggy compilation with MinGW.

NLopt 2.0

15 June 2010

  • New C API, that works by creating an nlopt_opt "object" and then calling functions to set the optimization parameters -- much more extensible than the old API (which is preserved for backwards compatibility). (Updated Fortran, Matlab, and GNU Octave wrappers as well.)
  • C++ nlopt.hpp wrapper around C API, allowing namespaces, object constructors/destructors, std::vector<double>, and exceptions to be exploited.
  • New nlopt wrappers callable from Python and GNU Guile, generated with the help of SWIG.
  • New man nlopt manual page documenting new API.
  • New AUGLAG algorithm(s) implementing an augmented-Lagrangian method proposed by Birgin and Martinez (2008), which supports nonlinear equality and inequality constraints "wrapped around" other local/global optimization methods.
  • Added API for nonlinear equality constraints (currently only supported by AUGLAG and ISRES algorithms).
  • Support inequality constraints directly in ORIG_DIRECT algorithms (no need to return NaN when constraint is violated).
  • Inequality/equality constraints now have optional tolerances that are used as conditions in stopping criteria.
  • Pseudo-randomize simplex steps in COBYLA algorithm, improving robustness by avoiding accidentally taking steps that don't improve conditioning (which seems to happen sometimes with active bound constraints). The algorithm remains deterministic (a deterministic seed is used), however.
  • Allow COBYLA to increase the trust-region radius if the predicted improvement was approximately right and the simplex is OK, following a suggestion in the SAS manual for PROC NLP that seems to improve convergence speed.
  • Added nlopt_force_stop function to force a (graceful) halt to the optimization, and corresponding NLOPT_FORCED_STOP return code.
  • Improved thread-safety in random-number generation: thread-local storage is used for random-number state, on compilers that support it (e.g. gcc, Intel, Microsoft), to make the generation thread-safe. In this case, the random-number seed must be set per-thread.
  • Return an error in global-search algorithms if the domain is not finite.
  • Use stdcall convention on Windows; thanks to Alan Young for the suggestion.
  • Added missing absolute-tolerance criteria in Luksan algorithms; thanks to Greg Nicholas for the bug report.
  • Fixed compilation under C++, and use C++ compiler for everything in --with-cxx mode; thanks to Greg Nicholas for the bug report.
  • In MMA, only stop at minf_max/stopval if the point is feasible.
  • Fix Matlab mex file to not include unnecessary nlopt-util.h file, simplifying Windows compilation.

NLopt 1.2

18 November 2009

  • Added Powell's BOBYQA algorithm for box-constrained optimization without derivatives, an improvement on NEWUOA.
  • Added ISRES evolutionary algorithm, supporting nonlinearly constrained global optimization.
  • New functions nlopt_{set/get}_stochastic_population to provide optional greater control over the random "population" sizes in stochastic algorithms (although it still has a sensible default).
  • Bug fix: remove extraneous text accidentally included in nlopt.f Fortran include file.
  • Bug fix: configure script now correctly handles Matlab installation when MEX_INSTALL_DIR is specified manually by the user.

NLopt 1.1

12 November 2008

  • configure script detects whether --enable-shared is required in order to compile Matlab and Octave plugins (as is the case on x86_64), and disables compilation of those plugins if --enable-shared is not used.
  • Added --without-octave and --without-matlab configure options to disable Octave and Matlab plugins, respectively.
  • Modified COBYLA algorithm to have better support for bound constraints.
  • Added new NLOPT_ROUNDOFF_LIMITED failure code to indicate cases in which optimization breaks down due to roundoff errors, in which case it is possible that useful results were obtained.
  • Experimental support for nonlinear equality constraints via augmented-Lagrangian method.
  • Support for compiling under Windows (ideally with MinGW) as a DLL, although you have to manually add #define NLOPT_DLL to nlopt.h after installing (after compiling NLopt).
  • Added several checks for roundoff-related breakdown to NEWUOA code.
  • When only a relative error tolerance is specified, no longer fails to halt when exact convergence to zero is obtained.
  • Workaround for incompatible qsort_r functions in BSD and GNU libc by always using my own version; thanks to Wendy Vandoolaeghe and Philippe Preux for the bug report and explanation.
  • Workaround for gcc 3.4.x conflict with HUGE_VAL definition in Solaris (gcc bug 19933).
  • Better identification of Matlab-plugin installation directory.
  • Fixed identification of Octave-plugin installation directory for recent Octave versions.

NLopt 1.0.1

13 November 2008

  • Allow user to override Matlab-plugin installation directory with MEX_INSTALL_DIR.
  • Bug fix in my DIRECT code that prevented convergence (DIRECT-L unaffected).
  • MLSL needs a nonzero default ftol_rel and/or xtol_rel to ensure that its local searches terminate; use roughly machine precision as defaults.

NLopt 1.0

11 November 2008

  • Initial public release.
Personal tools