7 * New AGS global solver ([#194]), thanks to Vladislav Sovrasov.
9 * New `nlopt_get_numevals` function providing a built-in evaluation counter ([#160]).
11 * New `nlopt_get_errmsg` function for more descriptive error messages.
13 * Build system is converted to `cmake` ([#49]), thanks to Julien Schueller
15 * Plugins updated for recent Octave and Guile versions.
17 * Various other build fixes and minor bug fixes.
23 * Fix CRS for empty dimensions (where lower `==` upper bound) (issue [#13]).
25 * Improvements to CMake (thanks to @xantares) and Windows builds (issue [#12]).
27 * Fix guile2 compatibility (issue [#21]).
33 * Use `cdecl` calling convention instead of stdcall on Win32, to
34 simplify shared-library usage and for consistency with Win64.
40 * New genetic algorithm ESCH, thanks to Carlos Henrique da Silva Santos.
42 * Fix swig dir for `VPATH` builds, thanks to Sandro Vitenti for the bug report.
44 * Use `python-config` program in the `configure` script to find the include
45 directories for Python, if possible (may be overridden by `PYTHON_CONFIG`
46 environment variable).
48 * Bugfix in copy constructor for C++ and Python interfaces.
50 * Bugfix in return value for setting min/max objective.
52 * Handle negative rescalings in COBYLA and BOBYQA.
54 * Plugin installation honors the `configure --prefix`, if any
55 (thanks to xantares@github).
57 * Various compilation fixes for compatibility with recent software.
63 * Bug fix: COBLYA should return `ROUNDOFF_LIMITED` rather than `XTOL_REACHED`
64 if the trust-region radius underflows to zero; thanks to David Liu.
66 * Bug fix: incorrect return value from set min/max objective.
68 * Handle case of negative rescalings (from negative dx) in COBYLA and BOBYQA;
69 thanks to Alexander Law for the bug report
75 * In Matlab/Octave interface, make returning NaN from the objective
76 equivalent to `nlopt_force_stop`; thanks to Norman Violet for the suggestion.
78 * Added CCSA-quadratic (`NLOPT_LD_CCSAQ`), similar to MMA.
80 * Added interface for supplying a preconditioner (approximate Hessian);
81 currently only supported in CCSAQ.
83 * When adding mconstraints, allow `tol==NULL` as synonym for zero tolerances.
85 * Added missing `NLOPT_LD_SLSQP` constant in Matlab/Octave.
87 * Lower tolerance for dual optimization in MMA/CCSAQ; thanks to
88 Christophe Leruste for the problem report
90 * Fixed bug in timer, thanks to William Vaughn for the patch.
92 * Bug fix to convergence test in sbplx; thanks to Douglas Bates.
98 * Fixed linking problem for C++ and Python shared libraries on MacOS X;
99 thanks to Volker Lorrmann for the bug report.
105 * Fixed additional re-entrancy problem in Luksan algorithms missed
106 in the previous version; thanks to Gert Wollny for the bug report.
108 * Fixed set/get `vector_storage` in Fortran interface for Fortran
109 compilers with all-uppercase linkage.
115 * Added `set_vector_storage` API to modify the memory usage and the
116 subspace dimension for low-storage quasi-Newton methods.
118 * Fixed missing support for maxtime stopping criteria in Luksan and
119 `ORIG_DIRECT` algorithms; thanks to Jurgen Werner for the bug report.
121 * Fixed algorithms to support the case where the lower and upper bounds
122 are equal for some variables (which effectively eliminates those
123 variables from optimization).
125 * Added missing xtol check to SLSQP, which caused erroneous `ROUNDOFF_LIMITED`
126 error codes to be returned; thanks to Alexander Riess for the bug report.
128 * Fixed slight overcounting of function evaluations for maxeval check
131 * Fixed deprecated API to support `xtol_abs == NULL` for backward
132 compatibility (thanks to Francesco Biscani for the bug report).
134 * Made Luksan algorithms (e.g. LBFGS and other quasi-Newton methods)
135 re-entrant; thanks to Gert Wollny for the bug report. (Fixes
138 * Fixed bug in DIRECT (not `ORIG_DIRECT`), where a typo caused suboptimal
139 convergence in some cases; thanks to Sinisa Hristov for the bug report.
145 * If you compile `nlopt.h` with the `NLOPT_DLL_EXPOR`T symbol `#defined`,
146 it now uses the `dllexport` directive (under Windows), useful for
147 compiling an NLopt DLL under Microsoft compilers; thanks to Benoit
148 Scherrer for the suggestion.
150 * Handle case where `copysign` function is missing, e.g. on Windows;
151 thanks to Benoit Scherrer for the bug report.
153 * Remove C99-style mixed declaration and code in a couple files, so
154 that code compiles in C89; thanks to Benoit Scherrer for the bug report.
156 * Removed a few compiler warnings under Microsoft compilers; thanks
157 to Benoit Scherrer for the bug report.
159 * Export `nlopt_get_algorithm_name` function on Windows; thanks to Ofek
160 Shilon for the bug report.
162 * Don't use `dllimport` directive with `lcc` on Windows (which doesn't
163 support it); thanks to Laurent Vanbeylen for the bug report.
165 * Update Nodedal README directory to indicate that Nocedal's LBFGS code
166 is now available under the GPL, and therefore may be distributed with
167 a future NLopt version (although Luksan's LBFGS code already works well).
169 * Bug fix in `set`/`get_xtol_abs`; thanks to David Rivest-Henault for the report.
175 * Added SLSQP algorithm for gradient-based local optimization with
176 nonlinear constraints via sequential quadratic programming, based
177 on the implementation by Dieter Kraft that was adapted for SciPy.
179 * Modified BOBYQA and COBYLA algorithms to support unequal initial
180 step sizes in different directions; thanks to Tom Fiddaman for pointing
181 out the need for this in the case where different directions have
182 very different scales.
184 * Added Python module docstring; thanks to Sebastian Walter for the suggestion.
186 * Added `GUILE_INSTALL_DIR` variable to allow the user to change the
187 Guile installation directory.
189 * Added Fortran interface for vector-valued constraints.
191 * Throw correct exceptions in Python for the `add_*constraint` functions;
192 thanks to Dmitrey Kroshko for the bug report.
194 * Support forced stop and exceptions in `ORIG_DIRECT` algorithm.
196 * Remove arbitrary `1e20` upper bound on function values from `ORIG_DIRECT`
199 * Bugfix in C++ interface (and some other language front-ends) when
200 deallocating the nlopt_opt object in cases like MLSL where
201 local_optimizer is used; thanks to Jurgen Werner for the bug report.
207 * The Python mconstraint (vector-valued constraint) functions
208 now pass a 2-dimensional array for the gradient argument, rather
209 than a flattened 1d array.
211 * Improved handling of exceptions and forced stops for constrained
212 optimization, making sure that no constraints are evaluated after
215 * Return an `NLOPT_INVALID_ARGS` error if more than n equality constraints
216 are added in an n-dimensional problem.
218 * Fix bug that could cause spurious `NLOPT_INVALID_ARGS` errors when
219 adding constraints under rare circumstances.
221 * Eliminate a few small memory leaks that could occur under error conditions.
227 * More robust configure check for Python include directories, via
228 `distutils.sysconfig.get_python_inc()` and `numpy.get_include()`.
229 Thanks to Nathaniel Smith for the tip.
231 * Bug fix in Guile interface: added missing prefix to nlopt-version-major
238 * New vector-valued constraint feature; thanks to Dmitrey Kroshko of OpenOpt
241 * COBYLA now accepts equality constraints.
243 * Guard against multiple inclusion in `nlopt.hpp`; thanks to Saul
244 Thurrowgood for the suggestion.
250 * Fixed compilation failure in Microsoft Visual Studio, due to
251 incorrect usage of `__stdcall` keyword; thanks to Dave Katz for the
258 * Bug fix in Fortran API (for `nlo_get_` functions returning arrays).
260 * Fixed buggy compilation with MinGW.
266 * New C API, that works by creating an `nlopt_opt` "object" and then calling
267 functions to set the optimization parameters — much more extensible
268 than the old API (which is preserved for backwards compatibility).
269 (Updated Fortran, Matlab, and GNU Octave wrappers as well.)
271 * C++ `nlopt.hpp` wrapper around C API, allowing namespaces, object
272 constructors/destructors, `std::vector<double>`, and exceptions
275 * New nlopt wrappers callable from Python and GNU Guile, generated
276 with the help of SWIG.
278 * New `man nlopt` manual page documenting new API.
280 * New AUGLAG algorithm(s) implementing an augmented-Lagrangian method
281 proposed by Birgin and Martinez (2008), which supports nonlinear
282 equality and inequality constraints "wrapped around" other
283 local/global optimization methods.
285 * Added API for nonlinear equality constraints (currently only
286 supported by AUGLAG and ISRES algorithms).
288 * Support inequality constraints directly in `ORIG_DIRECT` algorithms
289 (no need to return NaN when constraint is violated).
291 * Inequality/equality constraints now have optional tolerances that
292 are used as conditions in stopping criteria.
294 * Pseudo-randomize simplex steps in COBYLA algorithm, improving robustness
295 by avoiding accidentally taking steps that don't improve conditioning
296 (which seems to happen sometimes with active bound constraints). The
297 algorithm remains deterministic (a deterministic seed is used), however.
299 * Allow COBYLA to increase the trust-region radius if the predicted improvement
300 was approximately right and the simplex is OK, following a suggestion
301 in the SAS manual for PROC NLP that seems to improve convergence speed.
303 * Added `nlopt_force_stop` function to force a (graceful) halt to
304 the optimization, and corresponding `NLOPT_FORCED_STOP` return code.
306 * Improved thread-safety in random-number generation: thread-local
307 storage is used for random-number state, on compilers that support
308 it (e.g. gcc, Intel, Microsoft), to make the generation thread-safe.
309 In this case, the random-number seed must be set per-thread.
311 * Return an error in global-search algorithms if the domain is not finite.
313 * Use `stdcall` convention on Windows; thanks to Alan Young for the suggestion.
315 * Added missing absolute-tolerance criteria in Luksan algorithms; thanks
316 to Greg Nicholas for the bug report.
318 * Fixed compilation under C++, and use C++ compiler for everything in
319 `--with-cxx` mode; thanks to Greg Nicholas for the bug report.
321 * In MMA, only stop at minf_max/stopval if the point is feasible.
323 * Fix Matlab mex file to not include unnecessary `nlopt-util.h` file,
324 simplifying Windows compilation.
330 * Added Powell's BOBYQA algorithm for box-constrained optimization
331 without derivatives, an improvement on NEWUOA.
333 * Added ISRES genetic algorithm, supporting nonlinearly constrained
336 * New functions `nlopt_{set/get}_stochastic_population` to provide
337 optional greater control over the random "population" sizes in
338 stochastic algorithms (although it still has a sensible default).
340 * Bug fix: remove extraneous text accidentally included in `nlopt.f` Fortran
343 * Bug fix: `configure` script now correctly handles Matlab installation
344 when `MEX_INSTALL_DIR` is specified manually by the user.
350 * `configure` script detects whether `--enable-shared` is required
351 in order to compile Matlab and Octave plugins (as is the case
352 on x86_64), and disables compilation of those plugins if
353 `--enable-shared` is not used.
355 * Added `--without-octave` and `--without-matlab` configure options to
356 disable Octave and Matlab plugins, respectively.
358 * Modified COBYLA algorithm to have better support for bound
361 * Added new `NLOPT_ROUNDOFF_LIMITED` failure code to indicate
362 cases in which optimization breaks down due to roundoff errors,
363 in which case it is possible that useful results were obtained.
365 * Experimental support for nonlinear equality constraints via
366 augmented-Lagrangian method.
368 * Support for compiling under Windows (ideally with MinGW) as a
369 DLL, although you have to manually add `#define NLOPT_DLL`
370 to nlopt.h *after* installing (after compiling NLopt).
372 * Added several checks for roundoff-related breakdown to NEWUOA code.
374 * When only a relative error tolerance is specified, no longer
375 fails to halt when exact convergence to zero is obtained.
377 * Workaround for incompatible `qsort_r` functions in BSD and GNU libc
378 by always using my own version; thanks to Wendy Vandoolaeghe
379 and Philippe Preux for the bug report and explanation.
381 * Workaround for gcc 3.4.x conflict with `HUGE_VAL` definition in Solaris
384 * Better identification of Matlab-plugin installation directory.
386 * Fixed identification of Octave-plugin installation directory for
387 recent Octave versions.
393 * Allow user to override Matlab-plugin installation directory with
396 * Bug fix in my DIRECT code that prevented convergence (DIRECT-L unaffected).
398 * MLSL needs a nonzero default `ftol_rel` and/or `xtol_rel` to ensure that
399 its local searches terminate; use roughly machine precision as defaults.
405 * Initial public release.
407 <!--- issue/PR links -->
408 [#12]: https://github.com/stevengj/nlopt/issues/12
409 [#13]: https://github.com/stevengj/nlopt/issues/13
410 [#21]: https://github.com/stevengj/nlopt/issues/21
411 [#49]: https://github.com/stevengj/nlopt/issues/49
412 [#160]: https://github.com/stevengj/nlopt/issues/160
413 [#194]: https://github.com/stevengj/nlopt/issues/194