Interior-point method: Difference between revisions

Content deleted Content added
spelling fix: changed performane to performance
m Idea: Typo fixing, replaced: suffiicent → sufficient
Line 63:
* The solver is Newton's method, and a ''single'' step of Newton is done for each single step in ''t''.
 
They proved that, in this case, the difference ''x<sub>i</sub>'' - ''x''*(''t<sub>i</sub>'') remains at most 0.01, and f(''x<sub>i</sub>'') - f* is at most 2*''m''/''t<sub>i</sub>''. Thus, the solution accuracy is proportional to 1/''t<sub>i</sub>'', so to add a single accuracy-digit, it is suffiicentsufficient to multiply ''t<sub>i</sub>'' by 2 (or any other constant factor), which requires O(sqrt(''m'')) Newton steps. Since each Newton step takes O(''m n''<sup>2</sup>) operations, the total complexity is O(''m<sup>3/2</sup> n''<sup>2</sup>) operations for accuracy digit.
 
[[Yurii Nesterov|Yuri Nesterov]] extended the idea from linear to non-linear programs. He noted that the main property of the logarithmic barrier, used in the above proofs, is that it is [[self-concordant]] with a finite barrier parameter. Therefore, many other classes of convex programs can be solved in polytime using a path-following method, if we can find a suitable self-concordant barrier function for their feasible region.<ref name=":0" />{{Rp|___location=Sec.1}}