Release notes
What’s new in Knitro 14.2 ?
Knitro 14.2 offers significant performance improvements on nonconvex problems both in terms of efficiency and solution quality.
Knitro 14.2 offers significant improvements in robustness and efficiency when solving models with conic constraints using the specialized conic algorithm (enabled via
bar_conic_enable
=1).Knitro 14.2 offers the ability to retrieve the best feasible iterate found during the optimization via the new
KN_get_best_feasible_iterate()
API function or by setting the new user optionsoltype
=1.Knitro 14.2 improves the efficiency of the
ncvx_qcqp_init
user option, which can be used to improve the initial points when solving nonconvex QPs/QCQPs. This also has an effect when solving nonconvex QPs/QCQPs with the multi-start procedure (ms_enable
=1).Knitro 14.2 introduces new user options to set time limits on optimization solves. The new
maxtime
option can be used to set a global time limit (using real/wall clock time) that applies to single solves as well as multi-start, multi-algorithm, tuner or MIP solves. The following new options are also available:mip_sub_maxtime
can set a time limit on MIP subproblems;ma_sub_maxtime
can set a time limit on multi-algorithm subproblems;ms_sub_maxtime
can set a time limit on multi-start subproblems; andtuner_sub_maxtime
can set a time limit on tuner subproblems. All previous maxtime options have been deprecated.Knitro 14.2 adds a new
KN_add_compcons()
API function to replace the previousKN_set_compcons()
function for adding complementarity constraints. The new function allows adding complementarity constraints in batches through multiple function calls (instead of requiring that they all be added at once).Knitro 14.2 adds a new
initptfile
option that allows the user to specify a file from which to read the initial point for the Knitro algorithms.
Bug Fixes in Knitro 14.2.0
Fixed bug when running the MISQP algorithm (
mip_method
=3) and the MIP local search heuristic is enabled.Fixed bug when running the MISQP algorithm (
mip_method
=3) with multi-start.Fixed bug that could lead to a segfault when solving an over-constrained problem (i.e. a problem with more equality constraints than variables).