ready for CUTEst profiling and code review

BNCG updated with the latest active set estimation and Fischer-Braumeister based convergence test

Logic errors in LS fallback and TR loops fixed

Fixed bug in initial trust radius interpolation

Copying of Fortran function handles to embedded TAO debugged

BNCG accelerator added into BNLS, works for TAO examples, segfaults on CUTEst probs

propagated reduced hessian changes into the initial trust radius estimation

LMVM matrix can now deal with reduced systems that change in size based on # of active variables

BNLS back to using trust radius

Pseudo-code algorithms added to in-code documentation

reduced Hessian matrix is now a separate clone

some active set fixes applied

some more refactoring to eliminate copy-pasted code

Documentation work on the new algorithms

Wrapping up for a pull request

- New active set estimation added based on Bertsekas' method
- Active set estimation split off into the utilities folder so that it can later be applied to BLMVM and B

Wrapping up for a pull request

- New active set estimation added based on Bertsekas' method
- Active set estimation split off into the utilities folder so that it can later be applied to BLMVM and BNCG
- Convergence of BNK methods exhaustively profiled on CUTEst problems

show more ...

all bounded Newton methods are solving the entire CUTEst test set...still need improvements to active set estimation

New Tao monitor added that prints detailed globalization info alongside convergence

fixed bug in trust region initialization

trust region update based on predicted reduction changed to leave radius untouched for Newton steps on the interior

initial trust radius estimation improved with active-set Hessian masking

Fixed LMVM step calculation in line-search fallback

initial trust radius guess modified VecMedian, test outputs updated

Function-level descriptive comments

BNTL added and tested with TAO examples

Trust region projection improved in BNTR