1 NLopt 2.2.3 (8 June 2011)
3 * Fixed additional re-entrancy problem in Luksan algorithms missed
4 in the previous version; thanks to Gert Wollny for the bug report.
6 * Fixed set/get vector_storage in Fortran interface for Fortran
7 compilers with all-uppercase linkage.
9 NLopt 2.2.2 (26 May 2011)
11 * Added set_vector_storage API to modify the memory usage and the
12 subspace dimension for low-storage quasi-Newton methods.
14 * Fixed missing support for maxtime stopping criteria in Luksan and
15 ORIG_DIRECT algorithms; thanks to Jurgen Werner for the bug report.
17 * Fixed algorithms to support the case where the lower and upper bounds
18 are equal for some variables (which effectively eliminates those
19 variables from optimization).
21 * Added missing xtol check to SLSQP, which caused erroneous ROUNDOFF_LIMITED
22 error codes to be returned; thanks to Alexander Riess for the bug report.
24 * Fixed slight overcounting of function evaluations for maxeval check
27 * Fixed deprecated API to support xtol_abs == NULL for backward
28 compatibility (thanks to Francesco Biscani for the bug report).
30 * Made Luksan algorithms (e.g. LBFGS and other quasi-Newton methods)
31 re-entrant; thanks to Gert Wollny for the bug report. (Fixes
34 * Fixed bug in DIRECT (not ORIG_DIRECT), where a typo caused suboptimal
35 convergence in some cases; thanks to Sinisa Hristov for the bug report.
37 NLopt 2.2.1 (6 September 2010)
39 * If you compile nlopt.h with the NLOPT_DLL_EXPORT symbol #defined,
40 it now uses the dllexport directive (under Windows), useful for
41 compiling an NLopt DLL under Microsoft compilers; thanks to Benoit
42 Scherrer for the suggestion.
44 * Handle case where copysign function is missing, e.g. on Windows;
45 thanks to Benoit Scherrer for the bug report.
47 * Remove C99-style mixed declaration and code in a couple files, so
48 that code compiles in C89; thanks to Benoit Scherrer for the bug report.
50 * Removed a few compiler warnings under Microsoft compilers; thanks
51 to Benoit Scherrer for the bug report.
53 * Export nlopt_get_algorithm_name function on Windows; thanks to Ofek
54 Shilon for the bug report.
56 * Don't use dllimport directive with lcc on Windows (which doesn't
57 support it); thanks to Laurent Vanbeylen for the bug report.
59 * Update Nodedal README directory to indicate that Nocedal's LBFGS code
60 is now available under the GPL, and therefore may be distributed with
61 a future NLopt version (although Luksan's LBFGS code already works well).
63 * Bug fix in set/get_xtol_abs; thanks to David Rivest-Henault for the report.
65 NLopt 2.2 (15 July 2010)
67 * Added SLSQP algorithm for gradient-based local optimization with
68 nonlinear constraints via sequential quadratic programming, based
69 on the implementation by Dieter Kraft that was adapted for SciPy.
71 * Modified BOBYQA and COBYLA algorithms to support unequal initial
72 step sizes in different directions; thanks to Tom Fiddaman for pointing
73 out the need for this in the case where different directions have
74 very different scales.
76 * Added Python module docstring; thanks to Sebastian Walter for the suggestion.
78 * Added GUILE_INSTALL_DIR variable to allow the user to change the
79 Guile installation directory.
81 * Added Fortran interface for vector-valued constraints.
83 * Throw correct exceptions in Python for the add_*constraint functions;
84 thanks to Dmitrey Kroshko for the bug report.
86 * Support forced stop and exceptions in ORIG_DIRECT algorithm.
88 * Remove arbitrary 1e20 upper bound on function values from ORIG_DIRECT
91 * Bugfix in C++ interface (and some other language front-ends) when
92 deallocating the nlopt_opt object in cases like MLSL where
93 local_optimizer is used; thanks to Jurgen Werner for the bug report.
95 NLopt 2.1.2 (8 July 2010)
97 * The Python mconstraint (vector-valued constraint) functions
98 now pass a 2-dimensional array for the gradient argument, rather
99 than a flattened 1d array.
101 * Improved handling of exceptions and forced stops for constrained
102 optimization, making sure that no constraints are evaluated after
105 * Return an NLOPT_INVALID_ARGS error if more than n equality constraints
106 are added in an n-dimensional problem.
108 * Fix bug that could cause spurious NLOPT_INVALID_ARGS errors when
109 adding constraints under rare circumstances.
111 * Eliminate a few small memory leaks that could occur under error conditions.
113 NLopt 2.1.1 (7 July 2010)
115 * More robust configure check for Python include directories, via
116 distutils.sysconfig.get_python_inc() and numpy.get_include().
117 Thanks to Nathaniel Smith for the tip.
119 * Bug fix in Guile interface: added missing prefix to nlopt-version-major
122 NLopt 2.1 (6 July 2010)
124 * New vector-valued constraint feature; thanks to Dmitrey Kroshko of OpenOpt
127 * COBYLA now accepts equality constraints.
129 * Guard against multiple inclusion in nlopt.hpp; thanks to Saul
130 Thurrowgood for the suggestion.
132 NLopt 2.0.2 (17 June 2010)
134 * Fixed compilation failure in Microsoft Visual Studio, due to
135 incorrect usage of __stdcall keyword; thanks to Dave Katz for the
138 NLopt 2.0.1 (16 June 2010)
140 * Bug fix in Fortran API (for nlo_get_ functions returning arrays).
142 * Fixed buggy compilation with MinGW.
144 NLopt 2.0 (15 June 2010)
146 * New C API, that works by creating an nlopt_opt "object" and then calling
147 functions to set the optimization parameters -- much more extensible
148 than the old API (which is preserved for backwards compatibility).
149 (Updated Fortran, Matlab, and GNU Octave wrappers as well.)
151 * C++ nlopt.hpp wrapper around C API, allowing namespaces, object
152 constructors/destructors, std::vector<double>, and exceptions
155 * New nlopt wrappers callable from Python and GNU Guile, generated
156 with the help of SWIG.
158 * New 'man nlopt' manual page documenting new API.
160 * New AUGLAG algorithm(s) implementing an augmented-Lagrangian method
161 proposed by Birgin and Martinez (2008), which supports nonlinear
162 equality and inequality constraints "wrapped around" other
163 local/global optimization methods.
165 * Added API for nonlinear equality constraints (currently only
166 supported by AUGLAG and ISRES algorithms).
168 * Support inequality constraints directly in ORIG_DIRECT algorithms
169 (no need to return NaN when constraint is violated).
171 * Inequality/equality constraints now have optional tolerances that
172 are used as conditions in stopping criteria.
174 * Pseudo-randomize simplex steps in COBYLA algorithm, improving robustness
175 by avoiding accidentally taking steps that don't improve conditioning
176 (which seems to happen sometimes with active bound constraints). The
177 algorithm remains deterministic (a deterministic seed is used), however.
179 * Allow COBYLA to increase the trust-region radius if the predicted improvement
180 was approximately right and the simplex is OK, following a suggestion
181 in the SAS manual for PROC NLP that seems to improve convergence speed.
183 * Added nlopt_force_stop function to force a (graceful) halt to
184 the optimization, and corresponding NLOPT_FORCED_STOP return code.
186 * Improved thread-safety in random-number generation: thread-local
187 storage is used for random-number state, on compilers that support
188 it (e.g. gcc, Intel, Microsoft), to make the generation thread-safe.
189 In this case, the random-number seed must be set per-thread.
191 * Return an error in global-search algorithms if the domain is not finite.
193 * Use stdcall convention on Windows; thanks to Alan Young for the suggestion.
195 * Added missing absolute-tolerance criteria in Luksan algorithms; thanks
196 to Greg Nicholas for the bug report.
198 * Fixed compilation under C++, and use C++ compiler for everything in
199 --with-cxx mode; thanks to Greg Nicholas for the bug report.
201 * In MMA, only stop at minf_max/stopval if the point is feasible.
203 * Fix Matlab mex file to not include unnecessary nlopt-util.h file,
204 simplifying Windows compilation.
206 NLopt 1.2 (18 November 2009)
208 * Added Powell's BOBYQA algorithm for box-constrained optimization
209 without derivatives, an improvement on NEWUOA.
211 * Added ISRES genetic algorithm, supporting nonlinearly constrained
214 * New functions nlopt_{set/get}_stochastic_population to provide
215 optional greater control over the random "population" sizes in
216 stochastic algorithms (although it still has a sensible default).
218 * Bug fix: remove extraneous text accidentally included in nlopt.f Fortran
221 * Bug fix: configure script now correctly handles Matlab installation
222 when MEX_INSTALL_DIR is specified manually by the user.
224 NLopt 1.1 (12 November 2009)
226 * configure script detects whether --enable-shared is required
227 in order to compile Matlab and Octave plugins (as is the case
228 on x86_64), and disables compilation of those plugins if
229 --enable-shared is not used.
231 * Added --without-octave and --without-matlab configure options to
232 disable Octave and Matlab plugins, respectively.
234 * Modified COBYLA algorithm to have better support for bound
237 * Added new NLOPT_ROUNDOFF_LIMITED failure code to indicate
238 cases in which optimization breaks down due to roundoff errors,
239 in which case it is possible that useful results were obtained.
241 * Experimental support for nonlinear equality constraints via
242 augmented-Lagrangian method.
244 * Support for compiling under Windows (ideally with MinGW) as a
245 DLL, although you have to manually add #define NLOPT_DLL
246 to nlopt.h *after* installing (after compiling NLopt).
248 * Added several checks for roundoff-related breakdown to NEWUOA code.
250 * When only a relative error tolerance is specified, no longer
251 fails to halt when exact convergence to zero is obtained.
253 * Workaround for incompatible qsort_r functions in BSD and GNU libc
254 by always using my own version; thanks to Wendy Vandoolaeghe
255 and Philippe Preux for the bug report and explanation.
257 * Workaround for gcc 3.4.x conflict with HUGE_VAL definition in Solaris
260 * Better identification of Matlab-plugin installation directory.
262 * Fixed identification of Octave-plugin installation directory for
263 recent Octave versions.
265 NLopt 1.0.1 (13 Nov. 2008)
267 * Allow user to override Matlab-plugin installation directory with
270 * Bug fix in my DIRECT code that prevented convergence (DIRECT-L unaffected).
272 * MLSL needs a nonzero default ftol_rel and/or xtol_rel to ensure that
273 its local searches terminate; use roughly machine precision as defaults.
275 NLopt 1.0 (11 Nov. 2008)
277 * Initial public release.