![]() |
ИСТИНА |
Войти в систему Регистрация |
ИПМех РАН |
||
This work continues the development of Newton-type methods for MCP, based on identification of active indices. These methods possess local superlinear convergence under very mild assumptions. However, the algorithm of this kind developed so far may generate trajectories outside of the feasible box. At the same time, in real-world applications one often faces situations when the mapping is defined only on the feasible box, or even on a smaller set not containing (some parts of) the boundary of the feasible box, and hence, the methods which produce infeasible trajectories may be unusable. In this work, we propose new globalization strategy, generating only strictly feasible trajectories.