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