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