25.11.2015 Views

The MOSEK Python optimizer API manual Version 7.0 (Revision 141)

Optimizer API for Python - Documentation - Mosek

Optimizer API for Python - Documentation - Mosek

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.

142 CHAPTER 11. THE OPTIMIZERS FOR CONTINUOUS PROBLEMS<br />

Tolerance<br />

ɛ p<br />

ɛ d<br />

ɛ g<br />

ɛ i<br />

Parameter name<br />

dparam.intpnt tol pfeas<br />

dparam.intpnt tol dfeas<br />

dparam.intpnt tol rel gap<br />

dparam.intpnt tol infeas<br />

Table 11.1: Parameters employed in termination criterion.<br />

Finally, if<br />

ɛ i b T y k ><br />

‖b‖ ∞<br />

∥<br />

∥A T y k + s k∥ ∥ ∞<br />

max(1, ‖c‖ ∞ )<br />

then y k is reported as a certificate of primal infeasibility.<br />

It is possible to adjust the tolerances ɛ p , ɛ d , ɛ g and ɛ i using parameters; see table 11.1 for details. <strong>The</strong><br />

default values of the termination tolerances are chosen such that for a majority of problems appearing<br />

in practice it is not possible to achieve much better accuracy. <strong>The</strong>refore, tightening the tolerances<br />

usually is not worthwhile. However, an inspection of (11.5) reveals that quality of the solution is<br />

dependent on ‖b‖ ∞ and ‖c‖ ∞ ; the smaller the norms are, the better the solution accuracy.<br />

<strong>The</strong> interior-point method as implemented by <strong>MOSEK</strong> will converge toward optimality and primal<br />

and dual feasibility at the same rate [10]. This means that if the <strong>optimizer</strong> is stopped prematurely<br />

then it is very unlikely that either the primal or dual solution is feasible. Another consequence is that<br />

in most cases all the tolerances, ɛ p , ɛ d and ɛ g , has to be relaxed together to achieve an effect.<br />

<strong>The</strong> basis identification discussed in section 11.2.2.2 requires an optimal solution to work well; hence<br />

basis identification should turned off if the termination criterion is relaxed.<br />

To conclude the discussion in this section, relaxing the termination criterion is usually is not worthwhile.<br />

11.2.2.2 Basis identification<br />

An interior-point <strong>optimizer</strong> does not return an optimal basic solution unless the problem has a<br />

unique primal and dual optimal solution. <strong>The</strong>refore, the interior-point <strong>optimizer</strong> has an optional<br />

post-processing step that computes an optimal basic solution starting from the optimal interior-point<br />

solution. More information about the basis identification procedure may be found in [11].<br />

Please note that a basic solution is often more accurate than an interior-point solution.<br />

By default <strong>MOSEK</strong> performs a basis identification. However, if a basic solution is not needed, the<br />

basis identification procedure can be turned off. <strong>The</strong> parameters<br />

• iparam.intpnt basis,<br />

• iparam.bi ignore max iter, and<br />

• iparam.bi ignore num error<br />

controls when basis identification is performed.

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

Saved successfully!

Ooh no, something went wrong!