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