1 NLopt 2.4.1 (19 November 2013)
3 * Use cdecl calling convention instead of stdcall on Win32, to
4 simplify shared-library usage and for consistency with Win64.
6 NLopt 2.4 (2 November 2013)
8 * New genetic algorithm ESCH, thanks to Carlos Henrique da Silva Santos.
10 * Fix swig dir for VPATH builds, thanks to Sandro Vitenti for the bug report.
12 * Use python-config program in the configure script to find the include
13 directories for Python, if possible (may be overridden by PYTHON_CONFIG
14 environment variable).
16 * Bugfix in copy constructor for C++ and Python interfaces.
18 * Bugfix in return value for setting min/max objective.
20 * Handle negative rescalings in COBYLA and BOBYQA.
22 * Plugin installation honors the configure --prefix, if any
23 (thanks to xantares@github).
25 * Various compilation fixes for compatibility with recent software.
27 NLopt 2.3.1 (16 September 2012)
29 * Bug fix: COBLYA should return ROUNDOFF_LIMITED rather than XTOL_REACHED
30 if the trust-region radius underflows to zero; thanks to David Liu.
32 * Bug fix: incorrect return value from set min/max objective.
34 * Handle case of negative rescalings (from negative dx) in COBYLA and BOBYQA;
35 thanks to Alexander Law for the bug report
37 NLopt 2.3 (20 July 2012)
39 * In Matlab/Octave interface, make returning NaN from the objective
40 equivalent to nlopt_force_stop; thanks to Norman Violet for the suggestion.
42 * Added CCSA-quadratic (NLOPT_LD_CCSAQ), similar to MMA.
44 * Added interface for supplying a preconditioner (approximate Hessian);
45 currently only supported in CCSAQ.
47 * When adding mconstraints, allow tol==NULL as synonym for zero tolerances.
49 * Added missing NLOPT_LD_SLSQP constant in Matlab/Octave.
51 * Lower tolerance for dual optimization in MMA/CCSAQ; thanks to
52 Christophe Leruste for the problem report
54 * Fixed bug in timer, thanks to William Vaughn for the patch.
56 * Bug fix to convergence test in sbplx; thanks to Douglas Bates.
58 NLopt 2.2.4 (9 June 2011)
60 * Fixed linking problem for C++ and Python shared libraries on MacOS X;
61 thanks to Volker Lorrmann for the bug report.
63 NLopt 2.2.3 (8 June 2011)
65 * Fixed additional re-entrancy problem in Luksan algorithms missed
66 in the previous version; thanks to Gert Wollny for the bug report.
68 * Fixed set/get vector_storage in Fortran interface for Fortran
69 compilers with all-uppercase linkage.
71 NLopt 2.2.2 (26 May 2011)
73 * Added set_vector_storage API to modify the memory usage and the
74 subspace dimension for low-storage quasi-Newton methods.
76 * Fixed missing support for maxtime stopping criteria in Luksan and
77 ORIG_DIRECT algorithms; thanks to Jurgen Werner for the bug report.
79 * Fixed algorithms to support the case where the lower and upper bounds
80 are equal for some variables (which effectively eliminates those
81 variables from optimization).
83 * Added missing xtol check to SLSQP, which caused erroneous ROUNDOFF_LIMITED
84 error codes to be returned; thanks to Alexander Riess for the bug report.
86 * Fixed slight overcounting of function evaluations for maxeval check
89 * Fixed deprecated API to support xtol_abs == NULL for backward
90 compatibility (thanks to Francesco Biscani for the bug report).
92 * Made Luksan algorithms (e.g. LBFGS and other quasi-Newton methods)
93 re-entrant; thanks to Gert Wollny for the bug report. (Fixes
96 * Fixed bug in DIRECT (not ORIG_DIRECT), where a typo caused suboptimal
97 convergence in some cases; thanks to Sinisa Hristov for the bug report.
99 NLopt 2.2.1 (6 September 2010)
101 * If you compile nlopt.h with the NLOPT_DLL_EXPORT symbol #defined,
102 it now uses the dllexport directive (under Windows), useful for
103 compiling an NLopt DLL under Microsoft compilers; thanks to Benoit
104 Scherrer for the suggestion.
106 * Handle case where copysign function is missing, e.g. on Windows;
107 thanks to Benoit Scherrer for the bug report.
109 * Remove C99-style mixed declaration and code in a couple files, so
110 that code compiles in C89; thanks to Benoit Scherrer for the bug report.
112 * Removed a few compiler warnings under Microsoft compilers; thanks
113 to Benoit Scherrer for the bug report.
115 * Export nlopt_get_algorithm_name function on Windows; thanks to Ofek
116 Shilon for the bug report.
118 * Don't use dllimport directive with lcc on Windows (which doesn't
119 support it); thanks to Laurent Vanbeylen for the bug report.
121 * Update Nodedal README directory to indicate that Nocedal's LBFGS code
122 is now available under the GPL, and therefore may be distributed with
123 a future NLopt version (although Luksan's LBFGS code already works well).
125 * Bug fix in set/get_xtol_abs; thanks to David Rivest-Henault for the report.
127 NLopt 2.2 (15 July 2010)
129 * Added SLSQP algorithm for gradient-based local optimization with
130 nonlinear constraints via sequential quadratic programming, based
131 on the implementation by Dieter Kraft that was adapted for SciPy.
133 * Modified BOBYQA and COBYLA algorithms to support unequal initial
134 step sizes in different directions; thanks to Tom Fiddaman for pointing
135 out the need for this in the case where different directions have
136 very different scales.
138 * Added Python module docstring; thanks to Sebastian Walter for the suggestion.
140 * Added GUILE_INSTALL_DIR variable to allow the user to change the
141 Guile installation directory.
143 * Added Fortran interface for vector-valued constraints.
145 * Throw correct exceptions in Python for the add_*constraint functions;
146 thanks to Dmitrey Kroshko for the bug report.
148 * Support forced stop and exceptions in ORIG_DIRECT algorithm.
150 * Remove arbitrary 1e20 upper bound on function values from ORIG_DIRECT
153 * Bugfix in C++ interface (and some other language front-ends) when
154 deallocating the nlopt_opt object in cases like MLSL where
155 local_optimizer is used; thanks to Jurgen Werner for the bug report.
157 NLopt 2.1.2 (8 July 2010)
159 * The Python mconstraint (vector-valued constraint) functions
160 now pass a 2-dimensional array for the gradient argument, rather
161 than a flattened 1d array.
163 * Improved handling of exceptions and forced stops for constrained
164 optimization, making sure that no constraints are evaluated after
167 * Return an NLOPT_INVALID_ARGS error if more than n equality constraints
168 are added in an n-dimensional problem.
170 * Fix bug that could cause spurious NLOPT_INVALID_ARGS errors when
171 adding constraints under rare circumstances.
173 * Eliminate a few small memory leaks that could occur under error conditions.
175 NLopt 2.1.1 (7 July 2010)
177 * More robust configure check for Python include directories, via
178 distutils.sysconfig.get_python_inc() and numpy.get_include().
179 Thanks to Nathaniel Smith for the tip.
181 * Bug fix in Guile interface: added missing prefix to nlopt-version-major
184 NLopt 2.1 (6 July 2010)
186 * New vector-valued constraint feature; thanks to Dmitrey Kroshko of OpenOpt
189 * COBYLA now accepts equality constraints.
191 * Guard against multiple inclusion in nlopt.hpp; thanks to Saul
192 Thurrowgood for the suggestion.
194 NLopt 2.0.2 (17 June 2010)
196 * Fixed compilation failure in Microsoft Visual Studio, due to
197 incorrect usage of __stdcall keyword; thanks to Dave Katz for the
200 NLopt 2.0.1 (16 June 2010)
202 * Bug fix in Fortran API (for nlo_get_ functions returning arrays).
204 * Fixed buggy compilation with MinGW.
206 NLopt 2.0 (15 June 2010)
208 * New C API, that works by creating an nlopt_opt "object" and then calling
209 functions to set the optimization parameters -- much more extensible
210 than the old API (which is preserved for backwards compatibility).
211 (Updated Fortran, Matlab, and GNU Octave wrappers as well.)
213 * C++ nlopt.hpp wrapper around C API, allowing namespaces, object
214 constructors/destructors, std::vector<double>, and exceptions
217 * New nlopt wrappers callable from Python and GNU Guile, generated
218 with the help of SWIG.
220 * New 'man nlopt' manual page documenting new API.
222 * New AUGLAG algorithm(s) implementing an augmented-Lagrangian method
223 proposed by Birgin and Martinez (2008), which supports nonlinear
224 equality and inequality constraints "wrapped around" other
225 local/global optimization methods.
227 * Added API for nonlinear equality constraints (currently only
228 supported by AUGLAG and ISRES algorithms).
230 * Support inequality constraints directly in ORIG_DIRECT algorithms
231 (no need to return NaN when constraint is violated).
233 * Inequality/equality constraints now have optional tolerances that
234 are used as conditions in stopping criteria.
236 * Pseudo-randomize simplex steps in COBYLA algorithm, improving robustness
237 by avoiding accidentally taking steps that don't improve conditioning
238 (which seems to happen sometimes with active bound constraints). The
239 algorithm remains deterministic (a deterministic seed is used), however.
241 * Allow COBYLA to increase the trust-region radius if the predicted improvement
242 was approximately right and the simplex is OK, following a suggestion
243 in the SAS manual for PROC NLP that seems to improve convergence speed.
245 * Added nlopt_force_stop function to force a (graceful) halt to
246 the optimization, and corresponding NLOPT_FORCED_STOP return code.
248 * Improved thread-safety in random-number generation: thread-local
249 storage is used for random-number state, on compilers that support
250 it (e.g. gcc, Intel, Microsoft), to make the generation thread-safe.
251 In this case, the random-number seed must be set per-thread.
253 * Return an error in global-search algorithms if the domain is not finite.
255 * Use stdcall convention on Windows; thanks to Alan Young for the suggestion.
257 * Added missing absolute-tolerance criteria in Luksan algorithms; thanks
258 to Greg Nicholas for the bug report.
260 * Fixed compilation under C++, and use C++ compiler for everything in
261 --with-cxx mode; thanks to Greg Nicholas for the bug report.
263 * In MMA, only stop at minf_max/stopval if the point is feasible.
265 * Fix Matlab mex file to not include unnecessary nlopt-util.h file,
266 simplifying Windows compilation.
268 NLopt 1.2 (18 November 2009)
270 * Added Powell's BOBYQA algorithm for box-constrained optimization
271 without derivatives, an improvement on NEWUOA.
273 * Added ISRES genetic algorithm, supporting nonlinearly constrained
276 * New functions nlopt_{set/get}_stochastic_population to provide
277 optional greater control over the random "population" sizes in
278 stochastic algorithms (although it still has a sensible default).
280 * Bug fix: remove extraneous text accidentally included in nlopt.f Fortran
283 * Bug fix: configure script now correctly handles Matlab installation
284 when MEX_INSTALL_DIR is specified manually by the user.
286 NLopt 1.1 (12 November 2009)
288 * configure script detects whether --enable-shared is required
289 in order to compile Matlab and Octave plugins (as is the case
290 on x86_64), and disables compilation of those plugins if
291 --enable-shared is not used.
293 * Added --without-octave and --without-matlab configure options to
294 disable Octave and Matlab plugins, respectively.
296 * Modified COBYLA algorithm to have better support for bound
299 * Added new NLOPT_ROUNDOFF_LIMITED failure code to indicate
300 cases in which optimization breaks down due to roundoff errors,
301 in which case it is possible that useful results were obtained.
303 * Experimental support for nonlinear equality constraints via
304 augmented-Lagrangian method.
306 * Support for compiling under Windows (ideally with MinGW) as a
307 DLL, although you have to manually add #define NLOPT_DLL
308 to nlopt.h *after* installing (after compiling NLopt).
310 * Added several checks for roundoff-related breakdown to NEWUOA code.
312 * When only a relative error tolerance is specified, no longer
313 fails to halt when exact convergence to zero is obtained.
315 * Workaround for incompatible qsort_r functions in BSD and GNU libc
316 by always using my own version; thanks to Wendy Vandoolaeghe
317 and Philippe Preux for the bug report and explanation.
319 * Workaround for gcc 3.4.x conflict with HUGE_VAL definition in Solaris
322 * Better identification of Matlab-plugin installation directory.
324 * Fixed identification of Octave-plugin installation directory for
325 recent Octave versions.
327 NLopt 1.0.1 (13 Nov. 2008)
329 * Allow user to override Matlab-plugin installation directory with
332 * Bug fix in my DIRECT code that prevented convergence (DIRECT-L unaffected).
334 * MLSL needs a nonzero default ftol_rel and/or xtol_rel to ensure that
335 its local searches terminate; use roughly machine precision as defaults.
337 NLopt 1.0 (11 Nov. 2008)
339 * Initial public release.