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.
5.3. EFFICIENCY OF THE ND-SYSTEMS APPROACH 59<br />
whereas the eigenvalues of the matrices A T x i<br />
correspond <strong>to</strong> the i-th coordinate of the<br />
stationary points (for which it is usually far less clear in advance in which range<br />
the value at the global optimum will be located). Therefore, when dealing with the<br />
matrix A T p λ<br />
one may focus on the calculation of only a few eigenvalues, as one is<br />
interested in the smallest real eigenvalue that corresponds <strong>to</strong> a real solution point.<br />
Iterative methods allow the user <strong>to</strong> ‘zoom in’ on a few eigenvalues, whereas in case<br />
of A T x i<br />
all the real eigenvalues need <strong>to</strong> be computed, for each i =1, 2,...,n, and all<br />
resulting real critical points have <strong>to</strong> be substituted in<strong>to</strong> the criterion function <strong>to</strong> find<br />
the global optimum.<br />
In this <strong>thesis</strong>, approach (ii) of computing only the smallest real eigenvalue of the<br />
matrix A T p λ<br />
is investigated because only one eigenvalue is required when searching for<br />
the global minimum.<br />
5.3 Efficiency of the nD-systems approach<br />
The following research questions are addressed in this section: (i) For a given multidimensional<br />
time instant (t 1 ,...,t n ), what is the most efficient way <strong>to</strong> compute the<br />
value of y t1,...,t n<br />
from a given initial state w 0,...,0 using the n difference equations<br />
(5.3)? And as a closely related question, what is the most efficient way <strong>to</strong> compute<br />
the whole state vec<strong>to</strong>r w t1,...,t n<br />
? (ii) Can we design a suboptimal heuristic procedure<br />
that computes the state vec<strong>to</strong>r w t1,...,t n<br />
at acceptable computational costs?<br />
5.3.1 A linear complexity result for computing y t1 ,...,t n<br />
The first research question raised above is especially important when the efficiency<br />
of just one iteration of the iterative eigenvalue solver is under consideration. In this<br />
single iteration the state vec<strong>to</strong>r w t1,...,t n<br />
has <strong>to</strong> be computed in an efficient way. This<br />
in contrast <strong>to</strong> the case where we study the efficiency of solving the eigenvalue problem<br />
as a whole.<br />
The following result addresses the computational complexity that can be achieved<br />
by an optimal algorithm <strong>to</strong> compute y t1,...,t n<br />
from w 0,...,0 .<br />
Theorem 5.1. Consider a set of n multidimensional recursions of the form (5.3)<br />
and let an initial state w 0,...,0 be given. Then every algorithm that computes the value<br />
of y t1,...,t n<br />
, using only the recursions (5.3), has a computational complexity which<br />
increases at least linearly with the <strong>to</strong>tal time |t|.<br />
Proof. Each recursion from the set (5.3) allows one <strong>to</strong> compute the value of<br />
y t1,...,t n<br />
from a set of values for which the <strong>to</strong>tal times are all within the range<br />
|t| −m, |t| −m +1,...,|t| −1. The largest <strong>to</strong>tal time among the entries of the<br />
initial state w 0,...,0 corresponds <strong>to</strong> y m−1,...,m−1 and is equal <strong>to</strong> n(m − 1). Therefore,<br />
<strong>to</strong> express y t1,...,t n<br />
in terms of the quantities contained in the initial state requires at<br />
least ⌈(|t|−n(m − 1))/m⌉ applications of a recursion from the set (5.3). Hence, the<br />
computational complexity of any algorithm along such lines increases at least linearly