10.03.2015 Views

v2009.01.01 - Convex Optimization

v2009.01.01 - Convex Optimization

v2009.01.01 - Convex Optimization

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

4.2. FRAMEWORK 267<br />

minimizes an affine function on an intersection of the elliptope with<br />

hyperplanes. Although the same Boolean solution is obtained from<br />

this approximation (631) as compared with semidefinite program (626),<br />

when given that particular data from Example 4.2.3.1.1, Singer confides a<br />

counter-example: Instead, given data<br />

[ ] [ ]<br />

1 0 √2 1<br />

1<br />

A =<br />

1<br />

, b =<br />

(633)<br />

0 1 √2 1<br />

then solving approximation (631) yields<br />

⎛⎡<br />

1 − √ 1<br />

2<br />

y ⋆ ⎜⎢<br />

= round⎝⎣<br />

1 − √ 1<br />

2<br />

1<br />

⎤⎞<br />

⎥⎟<br />

⎦⎠ =<br />

⎡<br />

⎢<br />

⎣<br />

0<br />

0<br />

1<br />

⎤<br />

⎥<br />

⎦ (634)<br />

(infeasible, with or without rounding, with respect to original problem (614))<br />

whereas solving semidefinite program (626) produces<br />

⎡<br />

⎤<br />

1 1 −1 1<br />

round(G ⋆ ) = ⎢ 1 1 −1 1<br />

⎥<br />

⎣ −1 −1 1 −1 ⎦ (635)<br />

1 1 −1 1<br />

with sorted eigenvalues<br />

λ(G ⋆ ) =<br />

⎡<br />

⎢<br />

⎣<br />

3.99999965057264<br />

0.00000035942736<br />

−0.00000000000000<br />

−0.00000001000000<br />

⎤<br />

⎥<br />

⎦ (636)<br />

Truncating all but the largest eigenvalue, from (622) we obtain (confer y ⋆ )<br />

⎛⎡<br />

⎤⎞<br />

⎡ ⎤<br />

0.99999999625299 1<br />

x ⋆ = round⎝⎣<br />

0.99999999625299 ⎦⎠ = ⎣ 1 ⎦ (637)<br />

0.00000001434518 0<br />

the desired minimum cardinality Boolean result.<br />

4.2.3.1.3 Exercise. Minimum cardinality Boolean art.<br />

Assess general performance of standard-practice approximation (631) as<br />

compared with the proposed semidefinite program (626).

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!