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