1 NLopt 2.2 (15 July 2010)
3 * Added SLSQP algorithm for gradient-based local optimization with
4 nonlinear constraints via sequential quadratic programming, based
5 on the implementation by Dieter Kraft that was adapted for SciPy.
7 * Modified BOBYQA and COBYLA algorithms to support unequal initial
8 step sizes in different directions; thanks to Tom Fiddaman for pointing
9 out the need for this in the case where different directions have
10 very different scales.
12 * Added Python module docstring; thanks to Sebastian Walter for the suggestion.
14 * Added GUILE_INSTALL_DIR variable to allow the user to change the
15 Guile installation directory.
17 * Added Fortran interface for vector-valued constraints.
19 * Throw correct exceptions in Python for the add_*constraint functions;
20 thanks to Dmitrey Kroshko for the bug report.
22 * Support forced stop and exceptions in ORIG_DIRECT algorithm.
24 * Remove arbitrary 1e20 upper bound on function values from ORIG_DIRECT
27 * Bugfix in C++ interface (and some other language front-ends) when
28 deallocating the nlopt_opt object in cases like MLSL where
29 local_optimizer is used; thanks to Jurgen Werner for the bug report.
31 NLopt 2.1.2 (8 July 2010)
33 * The Python mconstraint (vector-valued constraint) functions
34 now pass a 2-dimensional array for the gradient argument, rather
35 than a flattened 1d array.
37 * Improved handling of exceptions and forced stops for constrained
38 optimization, making sure that no constraints are evaluated after
41 * Return an NLOPT_INVALID_ARGS error if more than n equality constraints
42 are added in an n-dimensional problem.
44 * Fix bug that could cause spurious NLOPT_INVALID_ARGS errors when
45 adding constraints under rare circumstances.
47 * Eliminate a few small memory leaks that could occur under error conditions.
49 NLopt 2.1.1 (7 July 2010)
51 * More robust configure check for Python include directories, via
52 distutils.sysconfig.get_python_inc() and numpy.get_include().
53 Thanks to Nathaniel Smith for the tip.
55 * Bug fix in Guile interface: added missing prefix to nlopt-version-major
58 NLopt 2.1 (6 July 2010)
60 * New vector-valued constraint feature; thanks to Dmitrey Kroshko of OpenOpt
63 * COBYLA now accepts equality constraints.
65 * Guard against multiple inclusion in nlopt.hpp; thanks to Saul
66 Thurrowgood for the suggestion.
68 NLopt 2.0.2 (17 June 2010)
70 * Fixed compilation failure in Microsoft Visual Studio, due to
71 incorrect usage of __stdcall keyword; thanks to Dave Katz for the
74 NLopt 2.0.1 (16 June 2010)
76 * Bug fix in Fortran API (for nlo_get_ functions returning arrays).
78 * Fixed buggy compilation with MinGW.
80 NLopt 2.0 (15 June 2010)
82 * New C API, that works by creating an nlopt_opt "object" and then calling
83 functions to set the optimization parameters -- much more extensible
84 than the old API (which is preserved for backwards compatibility).
85 (Updated Fortran, Matlab, and GNU Octave wrappers as well.)
87 * C++ nlopt.hpp wrapper around C API, allowing namespaces, object
88 constructors/destructors, std::vector<double>, and exceptions
91 * New nlopt wrappers callable from Python and GNU Guile, generated
92 with the help of SWIG.
94 * New 'man nlopt' manual page documenting new API.
96 * New AUGLAG algorithm(s) implementing an augmented-Lagrangian method
97 proposed by Birgin and Martinez (2008), which supports nonlinear
98 equality and inequality constraints "wrapped around" other
99 local/global optimization methods.
101 * Added API for nonlinear equality constraints (currently only
102 supported by AUGLAG and ISRES algorithms).
104 * Support inequality constraints directly in ORIG_DIRECT algorithms
105 (no need to return NaN when constraint is violated).
107 * Inequality/equality constraints now have optional tolerances that
108 are used as conditions in stopping criteria.
110 * Pseudo-randomize simplex steps in COBYLA algorithm, improving robustness
111 by avoiding accidentally taking steps that don't improve conditioning
112 (which seems to happen sometimes with active bound constraints). The
113 algorithm remains deterministic (a deterministic seed is used), however.
115 * Allow COBYLA to increase the trust-region radius if the predicted improvement
116 was approximately right and the simplex is OK, following a suggestion
117 in the SAS manual for PROC NLP that seems to improve convergence speed.
119 * Added nlopt_force_stop function to force a (graceful) halt to
120 the optimization, and corresponding NLOPT_FORCED_STOP return code.
122 * Improved thread-safety in random-number generation: thread-local
123 storage is used for random-number state, on compilers that support
124 it (e.g. gcc, Intel, Microsoft), to make the generation thread-safe.
125 In this case, the random-number seed must be set per-thread.
127 * Return an error in global-search algorithms if the domain is not finite.
129 * Use stdcall convention on Windows; thanks to Alan Young for the suggestion.
131 * Added missing absolute-tolerance criteria in Luksan algorithms; thanks
132 to Greg Nicholas for the bug report.
134 * Fixed compilation under C++, and use C++ compiler for everything in
135 --with-cxx mode; thanks to Greg Nicholas for the bug report.
137 * In MMA, only stop at minf_max/stopval if the point is feasible.
139 * Fix Matlab mex file to not include unnecessary nlopt-util.h file,
140 simplifying Windows compilation.
142 NLopt 1.2 (18 November 2009)
144 * Added Powell's BOBYQA algorithm for box-constrained optimization
145 without derivatives, an improvement on NEWUOA.
147 * Added ISRES genetic algorithm, supporting nonlinearly constrained
150 * New functions nlopt_{set/get}_stochastic_population to provide
151 optional greater control over the random "population" sizes in
152 stochastic algorithms (although it still has a sensible default).
154 * Bug fix: remove extraneous text accidentally included in nlopt.f Fortran
157 * Bug fix: configure script now correctly handles Matlab installation
158 when MEX_INSTALL_DIR is specified manually by the user.
160 NLopt 1.1 (12 November 2009)
162 * configure script detects whether --enable-shared is required
163 in order to compile Matlab and Octave plugins (as is the case
164 on x86_64), and disables compilation of those plugins if
165 --enable-shared is not used.
167 * Added --without-octave and --without-matlab configure options to
168 disable Octave and Matlab plugins, respectively.
170 * Modified COBYLA algorithm to have better support for bound
173 * Added new NLOPT_ROUNDOFF_LIMITED failure code to indicate
174 cases in which optimization breaks down due to roundoff errors,
175 in which case it is possible that useful results were obtained.
177 * Experimental support for nonlinear equality constraints via
178 augmented-Lagrangian method.
180 * Support for compiling under Windows (ideally with MinGW) as a
181 DLL, although you have to manually add #define NLOPT_DLL
182 to nlopt.h *after* installing (after compiling NLopt).
184 * Added several checks for roundoff-related breakdown to NEWUOA code.
186 * When only a relative error tolerance is specified, no longer
187 fails to halt when exact convergence to zero is obtained.
189 * Workaround for incompatible qsort_r functions in BSD and GNU libc
190 by always using my own version; thanks to Wendy Vandoolaeghe
191 and Philippe Preux for the bug report and explanation.
193 * Workaround for gcc 3.4.x conflict with HUGE_VAL definition in Solaris
196 * Better identification of Matlab-plugin installation directory.
198 * Fixed identification of Octave-plugin installation directory for
199 recent Octave versions.
201 NLopt 1.0.1 (13 Nov. 2008)
203 * Allow user to override Matlab-plugin installation directory with
206 * Bug fix in my DIRECT code that prevented convergence (DIRECT-L unaffected).
208 * MLSL needs a nonzero default ftol_rel and/or xtol_rel to ensure that
209 its local searches terminate; use roughly machine precision as defaults.
211 NLopt 1.0 (11 Nov. 2008)
213 * Initial public release.