1 NLopt 2.2.1 (6 September 2010)
3 * If you compile nlopt.h with the NLOPT_DLL_EXPORT symbol #defined,
4 it now uses the dllexport directive (under Windows), useful for
5 compiling an NLopt DLL under Microsoft compilers; thanks to Benoit
6 Scherrer for the suggestion.
8 * Handle case where copysign function is missing, e.g. on Windows;
9 thanks to Benoit Scherrer for the bug report.
11 * Remove C99-style mixed declaration and code in a couple files, so
12 that code compiles in C89; thanks to Benoit Scherrer for the bug report.
14 * Removed a few compiler warnings under Microsoft compilers; thanks
15 to Benoit Scherrer for the bug report.
17 * Export nlopt_get_algorithm_name function on Windows; thanks to Ofek
18 Shilon for the bug report.
20 * Don't use dllimport directive with lcc on Windows (which doesn't
21 support it); thanks to Laurent Vanbeylen for the bug report.
23 * Update Nodedal README directory to indicate that Nocedal's LBFGS code
24 is now available under the GPL, and therefore may be distributed with
25 a future NLopt version (although Luksan's LBFGS code already works well).
27 * Bug fix in set/get_xtol_abs; thanks to David Rivest-Henault for the report.
29 NLopt 2.2 (15 July 2010)
31 * Added SLSQP algorithm for gradient-based local optimization with
32 nonlinear constraints via sequential quadratic programming, based
33 on the implementation by Dieter Kraft that was adapted for SciPy.
35 * Modified BOBYQA and COBYLA algorithms to support unequal initial
36 step sizes in different directions; thanks to Tom Fiddaman for pointing
37 out the need for this in the case where different directions have
38 very different scales.
40 * Added Python module docstring; thanks to Sebastian Walter for the suggestion.
42 * Added GUILE_INSTALL_DIR variable to allow the user to change the
43 Guile installation directory.
45 * Added Fortran interface for vector-valued constraints.
47 * Throw correct exceptions in Python for the add_*constraint functions;
48 thanks to Dmitrey Kroshko for the bug report.
50 * Support forced stop and exceptions in ORIG_DIRECT algorithm.
52 * Remove arbitrary 1e20 upper bound on function values from ORIG_DIRECT
55 * Bugfix in C++ interface (and some other language front-ends) when
56 deallocating the nlopt_opt object in cases like MLSL where
57 local_optimizer is used; thanks to Jurgen Werner for the bug report.
59 NLopt 2.1.2 (8 July 2010)
61 * The Python mconstraint (vector-valued constraint) functions
62 now pass a 2-dimensional array for the gradient argument, rather
63 than a flattened 1d array.
65 * Improved handling of exceptions and forced stops for constrained
66 optimization, making sure that no constraints are evaluated after
69 * Return an NLOPT_INVALID_ARGS error if more than n equality constraints
70 are added in an n-dimensional problem.
72 * Fix bug that could cause spurious NLOPT_INVALID_ARGS errors when
73 adding constraints under rare circumstances.
75 * Eliminate a few small memory leaks that could occur under error conditions.
77 NLopt 2.1.1 (7 July 2010)
79 * More robust configure check for Python include directories, via
80 distutils.sysconfig.get_python_inc() and numpy.get_include().
81 Thanks to Nathaniel Smith for the tip.
83 * Bug fix in Guile interface: added missing prefix to nlopt-version-major
86 NLopt 2.1 (6 July 2010)
88 * New vector-valued constraint feature; thanks to Dmitrey Kroshko of OpenOpt
91 * COBYLA now accepts equality constraints.
93 * Guard against multiple inclusion in nlopt.hpp; thanks to Saul
94 Thurrowgood for the suggestion.
96 NLopt 2.0.2 (17 June 2010)
98 * Fixed compilation failure in Microsoft Visual Studio, due to
99 incorrect usage of __stdcall keyword; thanks to Dave Katz for the
102 NLopt 2.0.1 (16 June 2010)
104 * Bug fix in Fortran API (for nlo_get_ functions returning arrays).
106 * Fixed buggy compilation with MinGW.
108 NLopt 2.0 (15 June 2010)
110 * New C API, that works by creating an nlopt_opt "object" and then calling
111 functions to set the optimization parameters -- much more extensible
112 than the old API (which is preserved for backwards compatibility).
113 (Updated Fortran, Matlab, and GNU Octave wrappers as well.)
115 * C++ nlopt.hpp wrapper around C API, allowing namespaces, object
116 constructors/destructors, std::vector<double>, and exceptions
119 * New nlopt wrappers callable from Python and GNU Guile, generated
120 with the help of SWIG.
122 * New 'man nlopt' manual page documenting new API.
124 * New AUGLAG algorithm(s) implementing an augmented-Lagrangian method
125 proposed by Birgin and Martinez (2008), which supports nonlinear
126 equality and inequality constraints "wrapped around" other
127 local/global optimization methods.
129 * Added API for nonlinear equality constraints (currently only
130 supported by AUGLAG and ISRES algorithms).
132 * Support inequality constraints directly in ORIG_DIRECT algorithms
133 (no need to return NaN when constraint is violated).
135 * Inequality/equality constraints now have optional tolerances that
136 are used as conditions in stopping criteria.
138 * Pseudo-randomize simplex steps in COBYLA algorithm, improving robustness
139 by avoiding accidentally taking steps that don't improve conditioning
140 (which seems to happen sometimes with active bound constraints). The
141 algorithm remains deterministic (a deterministic seed is used), however.
143 * Allow COBYLA to increase the trust-region radius if the predicted improvement
144 was approximately right and the simplex is OK, following a suggestion
145 in the SAS manual for PROC NLP that seems to improve convergence speed.
147 * Added nlopt_force_stop function to force a (graceful) halt to
148 the optimization, and corresponding NLOPT_FORCED_STOP return code.
150 * Improved thread-safety in random-number generation: thread-local
151 storage is used for random-number state, on compilers that support
152 it (e.g. gcc, Intel, Microsoft), to make the generation thread-safe.
153 In this case, the random-number seed must be set per-thread.
155 * Return an error in global-search algorithms if the domain is not finite.
157 * Use stdcall convention on Windows; thanks to Alan Young for the suggestion.
159 * Added missing absolute-tolerance criteria in Luksan algorithms; thanks
160 to Greg Nicholas for the bug report.
162 * Fixed compilation under C++, and use C++ compiler for everything in
163 --with-cxx mode; thanks to Greg Nicholas for the bug report.
165 * In MMA, only stop at minf_max/stopval if the point is feasible.
167 * Fix Matlab mex file to not include unnecessary nlopt-util.h file,
168 simplifying Windows compilation.
170 NLopt 1.2 (18 November 2009)
172 * Added Powell's BOBYQA algorithm for box-constrained optimization
173 without derivatives, an improvement on NEWUOA.
175 * Added ISRES genetic algorithm, supporting nonlinearly constrained
178 * New functions nlopt_{set/get}_stochastic_population to provide
179 optional greater control over the random "population" sizes in
180 stochastic algorithms (although it still has a sensible default).
182 * Bug fix: remove extraneous text accidentally included in nlopt.f Fortran
185 * Bug fix: configure script now correctly handles Matlab installation
186 when MEX_INSTALL_DIR is specified manually by the user.
188 NLopt 1.1 (12 November 2009)
190 * configure script detects whether --enable-shared is required
191 in order to compile Matlab and Octave plugins (as is the case
192 on x86_64), and disables compilation of those plugins if
193 --enable-shared is not used.
195 * Added --without-octave and --without-matlab configure options to
196 disable Octave and Matlab plugins, respectively.
198 * Modified COBYLA algorithm to have better support for bound
201 * Added new NLOPT_ROUNDOFF_LIMITED failure code to indicate
202 cases in which optimization breaks down due to roundoff errors,
203 in which case it is possible that useful results were obtained.
205 * Experimental support for nonlinear equality constraints via
206 augmented-Lagrangian method.
208 * Support for compiling under Windows (ideally with MinGW) as a
209 DLL, although you have to manually add #define NLOPT_DLL
210 to nlopt.h *after* installing (after compiling NLopt).
212 * Added several checks for roundoff-related breakdown to NEWUOA code.
214 * When only a relative error tolerance is specified, no longer
215 fails to halt when exact convergence to zero is obtained.
217 * Workaround for incompatible qsort_r functions in BSD and GNU libc
218 by always using my own version; thanks to Wendy Vandoolaeghe
219 and Philippe Preux for the bug report and explanation.
221 * Workaround for gcc 3.4.x conflict with HUGE_VAL definition in Solaris
224 * Better identification of Matlab-plugin installation directory.
226 * Fixed identification of Octave-plugin installation directory for
227 recent Octave versions.
229 NLopt 1.0.1 (13 Nov. 2008)
231 * Allow user to override Matlab-plugin installation directory with
234 * Bug fix in my DIRECT code that prevented convergence (DIRECT-L unaffected).
236 * MLSL needs a nonzero default ftol_rel and/or xtol_rel to ensure that
237 its local searches terminate; use roughly machine precision as defaults.
239 NLopt 1.0 (11 Nov. 2008)
241 * Initial public release.