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.
7.1. COMPUTING THE MINIMUM OF A POLYNOMIAL OF ORDER 8 111<br />
required opera<strong>to</strong>r actions/matrix-vec<strong>to</strong>r products (MVs), and the running time. For<br />
these computations the default parameters of the various methods are used.<br />
Table 7.2: Minimal eigenvalue of the opera<strong>to</strong>r A p1<br />
(one eigenvalue calculated)<br />
Method Eigenvalue Error #MVs Time (s)<br />
Eigs 4.09516474405595 3.03 × 10 −10 3040 173<br />
JDQR 4.09516474449668 1.38 × 10 −10 312 21<br />
JDQZ 4.09516474427060 8.86 × 10 −11 302 21<br />
Eigs with balancing 4.09516468968153 5.47 × 10 −8 2752 162<br />
JDQR with balancing 4.09516473924322 5.12 × 10 −9 421 28<br />
JDQZ with balancing 4.09516473593991 8.42 × 10 −9 389 28<br />
The global minimum computed by the JDQZ method produces the critical value<br />
that is the closest <strong>to</strong> the ‘true’ critical value computed by the local search method.<br />
For this setting, the nD-systems approach uses the fewest opera<strong>to</strong>r actions. Using<br />
the corresponding eigenvec<strong>to</strong>r, the coordinates of the stationary point corresponding<br />
<strong>to</strong> this global minimum are computed as:<br />
x 1 =+0.876539213107485<br />
x 2 = −0.903966282291641<br />
x 3 =+0.862027936168838<br />
x 4 = −0.835187476763094<br />
(7.5)<br />
For the problem of finding the global minimum of a polynomial there are several<br />
other specialized software packages available, which employ different approaches. To<br />
put the performance of the nD-systems approach in<strong>to</strong> perspective, we will briefly<br />
discuss the outcomes of the computation of the global minimum of polynomial (7.1)<br />
when using the software packages SOSTOOLS, GloptiPoly and SYNAPS.<br />
SOSTOOLS is a Matlab <strong>to</strong>olbox for formulating and solving sum of squares (SOS)<br />
problems (see [86], [90]). This <strong>to</strong>olbox uses the Matlab solver SeDuMi (see [96]) <strong>to</strong><br />
solve the involved semi-definite programs (SDP). To compute the global minimum,<br />
SOSTOOLS searches for the largest possible γ for which p 1 (x 1 ,x 2 ,x 3 ,x 4 ) −γ is still a<br />
sum of squares. This γ may be the global minimum p ⋆ we are looking for, depending<br />
on whether the polynomial p 1 (x 1 ,x 2 ,x 3 ,x 4 ) − p ⋆ can be written as a sum of squares<br />
of polynomials (see [87]). Note that the nD-systems approach does not suffer from<br />
such a limitation.<br />
GloptiPoly (see [54]) solves a multivariable polynomial optimization problem by<br />
building and solving convex linear matrix inequality (LMI) relaxations of the problem<br />
using the solver SeDuMi. It produces a series of lower bounds which converge <strong>to</strong> the<br />
global optimum we are looking for. The theory of moments and positive polynomials<br />
is used in the implementation of this software (see [74] and [75]).<br />
SYNAPS [91] is a C++ environment for symbolic and numeric computations. It<br />
provides a routine <strong>to</strong> search for the real solutions of a polynomial system of equations<br />
like system (7.2) within a given domain (based on a subdivision method). All the