Question

This may be a silly question, but does anyone know of a proof that binary search is asymptotically optimal? That is, if we are given a sorted list of elements where the only permitted operation on those objects is a comparison, how do you prove that the search can’t be done in o(lg n)? (That’s little-o of lg n, by the way.) Note that I’m restricting this to elements where the only operation permitted operation is a comparison, since there are well-known algorithms that can beat O(lg n) on expectation if you’re allowed to do more complex operations on the data (see, for example, interpolation search).

Answer 1

From here:

• The number of possible outcomes should be at least O(n).
• You can represent the decisions done by the algorithm by nodes of a “decision tree”: if the items compare greater then it goes on way, if not it goes the other way. The nodes of the tree are the states of the algorithm and the leaves are the outcomes. So there should be at least O(n) nodes in the tree.
• Each tree on O(n) nodes has at least O(log N) levels.