link to my thesis
link to my thesis
link to my thesis
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
236 CHAPTER 12. CONCLUSIONS & DIRECTIONS FOR FURTHER RESEARCH<br />
k = 1 case, it is convenient <strong>to</strong> impose the technical condition that all the poles of<br />
the original given system H(s) are distinct. We have imposed a similar constraint for<br />
ease of exposition for all co-orders k in this <strong>thesis</strong>. Extensions may be developed in<br />
the future <strong>to</strong> handle the case with poles of multiplicities larger than 1 <strong>to</strong>o.<br />
Moreover, the algebraic techniques presented in this part of the <strong>thesis</strong> may be<br />
combined with numerical local search techniques <strong>to</strong> arrive at practical algorithms,<br />
for instance <strong>to</strong> save on hardware requirements, computation time, or <strong>to</strong> deal with<br />
numerical accuracy issues. To deal with numerical conditioning issues is an important<br />
research <strong>to</strong>pic, <strong>to</strong> be able <strong>to</strong> deal with problems that are of a somewhat larger size.<br />
A disadvantage of the co-order k = 2 and k = 3 case, from the perspective of<br />
computational efficiency, is that one can only decide about global optimality of a<br />
solution G(s) when all the solutions of the polynomial eigenvalue problem have been<br />
computed and further analyzed. Currently it is investigated for the co-order k =2<br />
case whether the third order polynomial V H can be employed by iterative polynomial<br />
eigenproblem solvers <strong>to</strong> limit the number of eigenvalues that require computation.<br />
Possibly, one could modify the available Jacobi–Davidson methods for a polynomial<br />
eigenvalue problem in such a way that it iterates with the involved matrix but targets<br />
on small positive real values of the criterion function V H first (in the same spirit as<br />
the JDCOMM method).<br />
When we compute the Kronecker canonical forms of the singular matrix pencils<br />
in the co-order k = 2 and k = 3 case, we use exact arithmetic <strong>to</strong> split off the singular<br />
blocks of these pencils. This is only possible when the given transfer function H(s) is<br />
given in exact format. Exact computations are computationally hard and, moreover,<br />
they become useless when the transfer function H(s) is given in numerical format.<br />
To overcome this computational limitation it is advisable <strong>to</strong> use numerical algorithms<br />
<strong>to</strong> compute the Kronecker canonical forms of the singular matrix pencils as described<br />
for example in [34].<br />
Finally, H 2 model-order reduction of the multi-input, multi-output case along<br />
similar lines as the approach followed here will lead <strong>to</strong> further generalizations <strong>to</strong>o.<br />
Furthermore, the discrete time case can be handled <strong>to</strong>o when using the isometric<br />
isomorphism mentioned in Remark 8.2 which transforms the continuous time case<br />
in<strong>to</strong> the discrete time case.