Formula: complexity estimates for divide-and-conquer algorithms

Let $ T(n)$ be the running time estimate of a divide-and-conquer algorithm. Assume that $ T(n)$ satisfies a relation of the form:

$\displaystyle T(n) \leq a \, T(n/b) + S(n)$ (5)

where $ S(n)$ is the cost of the combine-part, $ a \geq 3$ is the number of recursive calls and $ n/b$ is the size of a sub-problem, with $ b$ an integer greater than $ 1$ . Assume that the following holds for all $ n$ :

$\displaystyle S(b \, n) \ \geq \ b S(n) \ \ {\rm and} \ \ S(n) \ \geq \ n.$    

Then we have $ T(n) \ \in \ {\cal O}(S(n) \, n^{{{\log}_b}(a) - 1})$ .

Marc Moreno Maza
2008-01-31