Each computable function has an infinite number of different program representations in a given programming language. In the theory of algorithms one often strives to find a program with the smallest complexity for a given computable function and a given complexity measure (such a program could be called optimal). Blum's speedup theorem shows that for any complexity measure, there exists a computable function, such that there is no optimal program computing it, because every program has a program of lower complexity. This also rules out the idea there is a way to assign to arbitrary functions their computational complexity, meaning the assignment to any f of the complexity of an optimal program for f. This does of course not exclude the possibility of finding the complexity of an optimal program for certain specific functions.
Given a Blum complexity measure and a total computable function with two parameters, then there exists a total computable predicate (a boolean valued computable function) so that for every program for , there exists a program for so that for almost all
is called the speedup function. The fact that it may be as fast-growing as desired (as long as it is computable) means that the phenomenon of always having a program of smaller complexity remains even if by "smaller" we mean "significantly smaller" (for instance, quadratically smaller, exponentially smaller).
- Blum, Manuel (1967). "A Machine-Independent Theory of the Complexity of Recursive Functions" (PDF). Journal of the ACM. 14 (2): 322–336. doi:10.1145/321386.321395.
- Van Emde Boas, Peter (1975). Bečvář, Jiří (ed.). "Ten years of speedup". Proceedings of Mathematical Foundations of Computer Science, 4th Symposium, Mariánské Lázně, September 1-5, 1975. Lecture Notes in Computer Science. Springer-Verlag. 32: 13–29. doi:10.1007/3-540-07389-2_179..