3 * New genetic algorithm ESCH, thanks to Carlos Henrique da Silva Santos.
5 * Fix swig dir for VPATH builds, thanks to Sandro Vitenti for the bug report.
7 * Use python-config program in the configure script to find the include
8 directories for Python, if possible (may be overridden by PYTHON_CONFIG
11 * Bugfix in copy constructor for C++ and Python interfaces.
13 NLopt 2.3.1 (16 September 2012)
15 * Bug fix: COBLYA should return ROUNDOFF_LIMITED rather than XTOL_REACHED
16 if the trust-region radius underflows to zero; thanks to David Liu.
18 * Bug fix: incorrect return value from set min/max objective.
20 * Handle case of negative rescalings (from negative dx) in COBYLA and BOBYQA;
21 thanks to Alexander Law for the bug report
23 NLopt 2.3 (20 July 2012)
25 * In Matlab/Octave interface, make returning NaN from the objective
26 equivalent to nlopt_force_stop; thanks to Norman Violet for the suggestion.
28 * Added CCSA-quadratic (NLOPT_LD_CCSAQ), similar to MMA.
30 * Added interface for supplying a preconditioner (approximate Hessian);
31 currently only supported in CCSAQ.
33 * When adding mconstraints, allow tol==NULL as synonym for zero tolerances.
35 * Added missing NLOPT_LD_SLSQP constant in Matlab/Octave.
37 * Lower tolerance for dual optimization in MMA/CCSAQ; thanks to
38 Christophe Leruste for the problem report
40 * Fixed bug in timer, thanks to William Vaughn for the patch.
42 * Bug fix to convergence test in sbplx; thanks to Douglas Bates.
44 NLopt 2.2.4 (9 June 2011)
46 * Fixed linking problem for C++ and Python shared libraries on MacOS X;
47 thanks to Volker Lorrmann for the bug report.
49 NLopt 2.2.3 (8 June 2011)
51 * Fixed additional re-entrancy problem in Luksan algorithms missed
52 in the previous version; thanks to Gert Wollny for the bug report.
54 * Fixed set/get vector_storage in Fortran interface for Fortran
55 compilers with all-uppercase linkage.
57 NLopt 2.2.2 (26 May 2011)
59 * Added set_vector_storage API to modify the memory usage and the
60 subspace dimension for low-storage quasi-Newton methods.
62 * Fixed missing support for maxtime stopping criteria in Luksan and
63 ORIG_DIRECT algorithms; thanks to Jurgen Werner for the bug report.
65 * Fixed algorithms to support the case where the lower and upper bounds
66 are equal for some variables (which effectively eliminates those
67 variables from optimization).
69 * Added missing xtol check to SLSQP, which caused erroneous ROUNDOFF_LIMITED
70 error codes to be returned; thanks to Alexander Riess for the bug report.
72 * Fixed slight overcounting of function evaluations for maxeval check
75 * Fixed deprecated API to support xtol_abs == NULL for backward
76 compatibility (thanks to Francesco Biscani for the bug report).
78 * Made Luksan algorithms (e.g. LBFGS and other quasi-Newton methods)
79 re-entrant; thanks to Gert Wollny for the bug report. (Fixes
82 * Fixed bug in DIRECT (not ORIG_DIRECT), where a typo caused suboptimal
83 convergence in some cases; thanks to Sinisa Hristov for the bug report.
85 NLopt 2.2.1 (6 September 2010)
87 * If you compile nlopt.h with the NLOPT_DLL_EXPORT symbol #defined,
88 it now uses the dllexport directive (under Windows), useful for
89 compiling an NLopt DLL under Microsoft compilers; thanks to Benoit
90 Scherrer for the suggestion.
92 * Handle case where copysign function is missing, e.g. on Windows;
93 thanks to Benoit Scherrer for the bug report.
95 * Remove C99-style mixed declaration and code in a couple files, so
96 that code compiles in C89; thanks to Benoit Scherrer for the bug report.
98 * Removed a few compiler warnings under Microsoft compilers; thanks
99 to Benoit Scherrer for the bug report.
101 * Export nlopt_get_algorithm_name function on Windows; thanks to Ofek
102 Shilon for the bug report.
104 * Don't use dllimport directive with lcc on Windows (which doesn't
105 support it); thanks to Laurent Vanbeylen for the bug report.
107 * Update Nodedal README directory to indicate that Nocedal's LBFGS code
108 is now available under the GPL, and therefore may be distributed with
109 a future NLopt version (although Luksan's LBFGS code already works well).
111 * Bug fix in set/get_xtol_abs; thanks to David Rivest-Henault for the report.
113 NLopt 2.2 (15 July 2010)
115 * Added SLSQP algorithm for gradient-based local optimization with
116 nonlinear constraints via sequential quadratic programming, based
117 on the implementation by Dieter Kraft that was adapted for SciPy.
119 * Modified BOBYQA and COBYLA algorithms to support unequal initial
120 step sizes in different directions; thanks to Tom Fiddaman for pointing
121 out the need for this in the case where different directions have
122 very different scales.
124 * Added Python module docstring; thanks to Sebastian Walter for the suggestion.
126 * Added GUILE_INSTALL_DIR variable to allow the user to change the
127 Guile installation directory.
129 * Added Fortran interface for vector-valued constraints.
131 * Throw correct exceptions in Python for the add_*constraint functions;
132 thanks to Dmitrey Kroshko for the bug report.
134 * Support forced stop and exceptions in ORIG_DIRECT algorithm.
136 * Remove arbitrary 1e20 upper bound on function values from ORIG_DIRECT
139 * Bugfix in C++ interface (and some other language front-ends) when
140 deallocating the nlopt_opt object in cases like MLSL where
141 local_optimizer is used; thanks to Jurgen Werner for the bug report.
143 NLopt 2.1.2 (8 July 2010)
145 * The Python mconstraint (vector-valued constraint) functions
146 now pass a 2-dimensional array for the gradient argument, rather
147 than a flattened 1d array.
149 * Improved handling of exceptions and forced stops for constrained
150 optimization, making sure that no constraints are evaluated after
153 * Return an NLOPT_INVALID_ARGS error if more than n equality constraints
154 are added in an n-dimensional problem.
156 * Fix bug that could cause spurious NLOPT_INVALID_ARGS errors when
157 adding constraints under rare circumstances.
159 * Eliminate a few small memory leaks that could occur under error conditions.
161 NLopt 2.1.1 (7 July 2010)
163 * More robust configure check for Python include directories, via
164 distutils.sysconfig.get_python_inc() and numpy.get_include().
165 Thanks to Nathaniel Smith for the tip.
167 * Bug fix in Guile interface: added missing prefix to nlopt-version-major
170 NLopt 2.1 (6 July 2010)
172 * New vector-valued constraint feature; thanks to Dmitrey Kroshko of OpenOpt
175 * COBYLA now accepts equality constraints.
177 * Guard against multiple inclusion in nlopt.hpp; thanks to Saul
178 Thurrowgood for the suggestion.
180 NLopt 2.0.2 (17 June 2010)
182 * Fixed compilation failure in Microsoft Visual Studio, due to
183 incorrect usage of __stdcall keyword; thanks to Dave Katz for the
186 NLopt 2.0.1 (16 June 2010)
188 * Bug fix in Fortran API (for nlo_get_ functions returning arrays).
190 * Fixed buggy compilation with MinGW.
192 NLopt 2.0 (15 June 2010)
194 * New C API, that works by creating an nlopt_opt "object" and then calling
195 functions to set the optimization parameters -- much more extensible
196 than the old API (which is preserved for backwards compatibility).
197 (Updated Fortran, Matlab, and GNU Octave wrappers as well.)
199 * C++ nlopt.hpp wrapper around C API, allowing namespaces, object
200 constructors/destructors, std::vector<double>, and exceptions
203 * New nlopt wrappers callable from Python and GNU Guile, generated
204 with the help of SWIG.
206 * New 'man nlopt' manual page documenting new API.
208 * New AUGLAG algorithm(s) implementing an augmented-Lagrangian method
209 proposed by Birgin and Martinez (2008), which supports nonlinear
210 equality and inequality constraints "wrapped around" other
211 local/global optimization methods.
213 * Added API for nonlinear equality constraints (currently only
214 supported by AUGLAG and ISRES algorithms).
216 * Support inequality constraints directly in ORIG_DIRECT algorithms
217 (no need to return NaN when constraint is violated).
219 * Inequality/equality constraints now have optional tolerances that
220 are used as conditions in stopping criteria.
222 * Pseudo-randomize simplex steps in COBYLA algorithm, improving robustness
223 by avoiding accidentally taking steps that don't improve conditioning
224 (which seems to happen sometimes with active bound constraints). The
225 algorithm remains deterministic (a deterministic seed is used), however.
227 * Allow COBYLA to increase the trust-region radius if the predicted improvement
228 was approximately right and the simplex is OK, following a suggestion
229 in the SAS manual for PROC NLP that seems to improve convergence speed.
231 * Added nlopt_force_stop function to force a (graceful) halt to
232 the optimization, and corresponding NLOPT_FORCED_STOP return code.
234 * Improved thread-safety in random-number generation: thread-local
235 storage is used for random-number state, on compilers that support
236 it (e.g. gcc, Intel, Microsoft), to make the generation thread-safe.
237 In this case, the random-number seed must be set per-thread.
239 * Return an error in global-search algorithms if the domain is not finite.
241 * Use stdcall convention on Windows; thanks to Alan Young for the suggestion.
243 * Added missing absolute-tolerance criteria in Luksan algorithms; thanks
244 to Greg Nicholas for the bug report.
246 * Fixed compilation under C++, and use C++ compiler for everything in
247 --with-cxx mode; thanks to Greg Nicholas for the bug report.
249 * In MMA, only stop at minf_max/stopval if the point is feasible.
251 * Fix Matlab mex file to not include unnecessary nlopt-util.h file,
252 simplifying Windows compilation.
254 NLopt 1.2 (18 November 2009)
256 * Added Powell's BOBYQA algorithm for box-constrained optimization
257 without derivatives, an improvement on NEWUOA.
259 * Added ISRES genetic algorithm, supporting nonlinearly constrained
262 * New functions nlopt_{set/get}_stochastic_population to provide
263 optional greater control over the random "population" sizes in
264 stochastic algorithms (although it still has a sensible default).
266 * Bug fix: remove extraneous text accidentally included in nlopt.f Fortran
269 * Bug fix: configure script now correctly handles Matlab installation
270 when MEX_INSTALL_DIR is specified manually by the user.
272 NLopt 1.1 (12 November 2009)
274 * configure script detects whether --enable-shared is required
275 in order to compile Matlab and Octave plugins (as is the case
276 on x86_64), and disables compilation of those plugins if
277 --enable-shared is not used.
279 * Added --without-octave and --without-matlab configure options to
280 disable Octave and Matlab plugins, respectively.
282 * Modified COBYLA algorithm to have better support for bound
285 * Added new NLOPT_ROUNDOFF_LIMITED failure code to indicate
286 cases in which optimization breaks down due to roundoff errors,
287 in which case it is possible that useful results were obtained.
289 * Experimental support for nonlinear equality constraints via
290 augmented-Lagrangian method.
292 * Support for compiling under Windows (ideally with MinGW) as a
293 DLL, although you have to manually add #define NLOPT_DLL
294 to nlopt.h *after* installing (after compiling NLopt).
296 * Added several checks for roundoff-related breakdown to NEWUOA code.
298 * When only a relative error tolerance is specified, no longer
299 fails to halt when exact convergence to zero is obtained.
301 * Workaround for incompatible qsort_r functions in BSD and GNU libc
302 by always using my own version; thanks to Wendy Vandoolaeghe
303 and Philippe Preux for the bug report and explanation.
305 * Workaround for gcc 3.4.x conflict with HUGE_VAL definition in Solaris
308 * Better identification of Matlab-plugin installation directory.
310 * Fixed identification of Octave-plugin installation directory for
311 recent Octave versions.
313 NLopt 1.0.1 (13 Nov. 2008)
315 * Allow user to override Matlab-plugin installation directory with
318 * Bug fix in my DIRECT code that prevented convergence (DIRECT-L unaffected).
320 * MLSL needs a nonzero default ftol_rel and/or xtol_rel to ensure that
321 its local searches terminate; use roughly machine precision as defaults.
323 NLopt 1.0 (11 Nov. 2008)
325 * Initial public release.