Content deleted Content added
→enumeration: explain Corliss more |
|||
Line 47:
::I looked through Corliss' article and added some details to the article. The roots are enumerated according to their order. I don't know what happens if there are infinitely many roots; it makes sense to assume they all have probability zero, but perhaps the axioms of probability theory do not work in this case. -- [[User:Jitse Niesen|Jitse Niesen]] ([[User talk:Jitse Niesen|talk]]) 15:51, 20 September 2008 (UTC)
Corliss' paper is rather poorly worded. He takes a probability space of functions with the property that the distribution of zeros is the same as the distribution of an odd number of independent random variables with uniform distribution. (This implies, for example, that multiple zeros don't happen.) Applying bisection to a random function in that probability space, the probability that the ''i''-th smallest zero is found is 0 if ''i'' is even (but this is always true for functions with only simple zeros), and independent of ''i'' if ''i'' is odd. Corliss' result says nothing at all about the behaviour of bisection on particular functions. Also, I don't know any real-life class of functions which satisfies Corliss' assumptions, so the practical significance is uncertain. I hope my new version of the text is clear enough. [[User:McKay|McKay]] ([[User talk:McKay|talk]]) 02:16, 12 June 2009 (UTC)
== Convergence ==
Hi, I have a question on the "convergence linearly" . Since the absolute error of bisection convergence is |b-a|/2^n, so the rate of convergence is O(1/2^n) which means the growth of error is exponential. I think it is more precise to use this information instead of the phrase "convergence linearly"! <span style="font-size: smaller;" class="autosigned">—Preceding [[Wikipedia:Signatures|unsigned]] comment added by [[Special:Contributions/132.235.37.166|132.235.37.166]] ([[User talk:132.235.37.166|talk]]) 13:24, 3 October 2008 (UTC)</span><!-- Template:UnsignedIP --> <!--Autosigned by SineBot-->
| |||