I'm interested in examples of theorems that employ the proof techniques that are utilized in the proof of the undecidability of Kolmogorov Complexity.

Definition:(Sipser) Let x be a binary string. We say that the minimal description of x, written as d(x), is the shortest string $\langle$M,w$\rangle$ where TM M on input w halts with x on its tape. So, the Kolmogorov Complexity K(x) is written as, K(x)=|d(x)|. K(x) is defined to be the length of minimal description of x.

Theorem: K(x) is not a computable function.

Proof/Sketch of Proof (attributed to Chor): Proof of negation. $\forall$n, let $y_{n}$ be the lexicographical first string y that satisfies n < K(y). Consider the following TM M: On input n (encoded in binary), M generates one by one all binary strings $x_{0}$, $x_{1}$, $x_{2}$, $x_{3}$... in lexicographic order.

For each $x_{i}$ it produces, M computes K($x_{i}$).

If K($x_{i}$) > n, then the TM M, outputs $x_{i}$ and halts. Else, the TM M, continues to examine the next lexicographical string $x_{i+1}$.

Since the function K is unbounded, it is guaranteed that M will eventually come across a string x satisfying K(x) $>$ n.

Question: what will the TM M output on input n?

By definition on input n TM M outputs $y_{n}$ (the lexicographical first string whose Kolmogorov complexity exceeds n, K(x) > n), but the length of n is $log_{2}$(n). So we have $K_{M}$($y_{n}$) $\leq$ $log_{2}$(n). There is a constant $c_{M}$ such that $\forall$y, K(y) $\leq$ $K_{M}$(y) + $c_{M}$, so $\forall$n K($y_{n}$) $\leq$ $log_{2}$(n) + $c_{M}$.

By definition of $y_{n}$ for all n, n < K($y_{n}$). By combining the two inequalities we get: n < $log_{2}$(n) + $c_{M}$, but for large enough n this is false. Thus a contradiction.

Question: What other theorems utilize a similar proof technique in their proofs?

For example: The proof that the set of incompressible strings is undecidable is very similar with some slight modifications.

  • 3
    $\begingroup$ This looks like an application of the Berry paradox (en.wikipedia.org/wiki/Berry_paradox), in the same way that Gödel's incompleteness theorem and the halting problem are applications of the liar paradox. $\endgroup$ – George Lowther Jun 16 '11 at 21:52
  • 1
    $\begingroup$ It's not a proof by contradiction. It's a proof of negation. $\endgroup$ – Andrej Bauer Mar 25 '17 at 21:18

Using the same technique, one can construct infinitely many statements which are true with probability arbitrarily close to 1, but are nonetheless unprovable. See lemma 4 in http://theory.stanford.edu/~trevisan/cs172/notek.pdf

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.