Notes

Chapter 12: The Principle of Computational Equivalence

Section 8: Undecidability and Intractability


P versus NP questions

Most programs that are explicitly constructed to solve specific problems tend at some level to have rather simple behavior—often just repetitive or nested, so long as appropriate number representations are used. And it is this that makes it realistic to estimate asymptotic growth rates using traditional mathematics, and to determine whether the programs operate in polynomial time. But as the pictures on page 761 suggest, arbitrary computational systems—even Turing machines with very simple rules—can exhibit much more complicated behavior with no clear asymptotic growth rate. And indeed the question of whether the halting times for a system grow only like a power of input size is in general undecidable. And if one tries to prove a result about halting times using, say, standard axioms of arithmetic or set theory, one may find that the result is independent of those axioms. So this makes it far from clear that the general P=NP question has a definite answer within standard axiom systems of mathematics. If one day someone were to find a provably polynomial time algorithm that solves an NP-complete problem then this would establish that P=NP. But it could well be that the fastest programs for NP-complete problems behave in ways that are too complicated to prove much about using the standard axioms of mathematics.


From Stephen Wolfram: A New Kind of Science [citation]