1 NLopt 2.0.2 (17 June 2010)
3 * Fixed compilation failure in Microsoft Visual Studio, due to
4 incorrect usage of __stdcall keyword; thanks to Dave Katz for the
7 NLopt 2.0.1 (16 June 2010)
9 * Bug fix in Fortran API (for nlo_get_ functions returning arrays).
11 * Fixed buggy compilation with MinGW.
13 NLopt 2.0 (15 June 2010)
15 * New C API, that works by creating an nlopt_opt "object" and then calling
16 functions to set the optimization parameters -- much more extensible
17 than the old API (which is preserved for backwards compatibility).
18 (Updated Fortran, Matlab, and GNU Octave wrappers as well.)
20 * C++ nlopt.hpp wrapper around C API, allowing namespaces, object
21 constructors/destructors, std::vector<double>, and exceptions
24 * New nlopt wrappers callable from Python and GNU Guile, generated
25 with the help of SWIG.
27 * New 'man nlopt' manual page documenting new API.
29 * New AUGLAG algorithm(s) implementing an augmented-Lagrangian method
30 proposed by Birgin and Martinez (2008), which supports nonlinear
31 equality and inequality constraints "wrapped around" other
32 local/global optimization methods.
34 * Added API for nonlinear equality constraints (currently only
35 supported by AUGLAG and ISRES algorithms).
37 * Support inequality constraints directly in ORIG_DIRECT algorithms
38 (no need to return NaN when constraint is violated).
40 * Inequality/equality constraints now have optional tolerances that
41 are used as conditions in stopping criteria.
43 * Pseudo-randomize simplex steps in COBYLA algorithm, improving robustness
44 by avoiding accidentally taking steps that don't improve conditioning
45 (which seems to happen sometimes with active bound constraints). The
46 algorithm remains deterministic (a deterministic seed is used), however.
48 * Allow COBYLA to increase the trust-region radius if the predicted improvement
49 was approximately right and the simplex is OK, following a suggestion
50 in the SAS manual for PROC NLP that seems to improve convergence speed.
52 * Added nlopt_force_stop function to force a (graceful) halt to
53 the optimization, and corresponding NLOPT_FORCED_STOP return code.
55 * Improved thread-safety in random-number generation: thread-local
56 storage is used for random-number state, on compilers that support
57 it (e.g. gcc, Intel, Microsoft), to make the generation thread-safe.
58 In this case, the random-number seed must be set per-thread.
60 * Return an error in global-search algorithms if the domain is not finite.
62 * Use stdcall convention on Windows; thanks to Alan Young for the suggestion.
64 * Added missing absolute-tolerance criteria in Luksan algorithms; thanks
65 to Greg Nicholas for the bug report.
67 * Fixed compilation under C++, and use C++ compiler for everything in
68 --with-cxx mode; thanks to Greg Nicholas for the bug report.
70 * In MMA, only stop at minf_max/stopval if the point is feasible.
72 * Fix Matlab mex file to not include unnecessary nlopt-util.h file,
73 simplifying Windows compilation.
75 NLopt 1.2 (18 November 2009)
77 * Added Powell's BOBYQA algorithm for box-constrained optimization
78 without derivatives, an improvement on NEWUOA.
80 * Added ISRES genetic algorithm, supporting nonlinearly constrained
83 * New functions nlopt_{set/get}_stochastic_population to provide
84 optional greater control over the random "population" sizes in
85 stochastic algorithms (although it still has a sensible default).
87 * Bug fix: remove extraneous text accidentally included in nlopt.f Fortran
90 * Bug fix: configure script now correctly handles Matlab installation
91 when MEX_INSTALL_DIR is specified manually by the user.
93 NLopt 1.1 (12 November 2009)
95 * configure script detects whether --enable-shared is required
96 in order to compile Matlab and Octave plugins (as is the case
97 on x86_64), and disables compilation of those plugins if
98 --enable-shared is not used.
100 * Added --without-octave and --without-matlab configure options to
101 disable Octave and Matlab plugins, respectively.
103 * Modified COBYLA algorithm to have better support for bound
106 * Added new NLOPT_ROUNDOFF_LIMITED failure code to indicate
107 cases in which optimization breaks down due to roundoff errors,
108 in which case it is possible that useful results were obtained.
110 * Experimental support for nonlinear equality constraints via
111 augmented-Lagrangian method.
113 * Support for compiling under Windows (ideally with MinGW) as a
114 DLL, although you have to manually add #define NLOPT_DLL
115 to nlopt.h *after* installing (after compiling NLopt).
117 * Added several checks for roundoff-related breakdown to NEWUOA code.
119 * When only a relative error tolerance is specified, no longer
120 fails to halt when exact convergence to zero is obtained.
122 * Workaround for incompatible qsort_r functions in BSD and GNU libc
123 by always using my own version; thanks to Wendy Vandoolaeghe
124 and Philippe Preux for the bug report and explanation.
126 * Workaround for gcc 3.4.x conflict with HUGE_VAL definition in Solaris
129 * Better identification of Matlab-plugin installation directory.
131 * Fixed identification of Octave-plugin installation directory for
132 recent Octave versions.
134 NLopt 1.0.1 (13 Nov. 2008)
136 * Allow user to override Matlab-plugin installation directory with
139 * Bug fix in my DIRECT code that prevented convergence (DIRECT-L unaffected).
141 * MLSL needs a nonzero default ftol_rel and/or xtol_rel to ensure that
142 its local searches terminate; use roughly machine precision as defaults.
144 NLopt 1.0 (11 Nov. 2008)
146 * Initial public release.