English

The normalized algorithmic information distance can not be approximated

Information Theory 2020-02-18 v1 Computational Complexity math.IT

Abstract

It is known that the normalized algorithmic information distance NN is not computable and not semicomputable. We show that for all ϵ<1/2\epsilon < 1/2, there exist no semicomputable functions that differ from NN by at most~ϵ\epsilon. Moreover, for any computable function ff such that limtf(x,y,t)N(x,y)ϵ|\lim_t f(x,y,t) - N(x,y)| \le \epsilon and for all nn, there exist strings x,yx,y of length nn such that tf(x,y,t+1)f(x,y,t)Ω(logn)\sum_t |f(x,y,t+1) - f(x,y,t)| \ge \Omega(\log n). This is optimal up to constant factors. We also show that the maximal number of oscillations of a limit approximation of NN is Ω(n/logn)\Omega(n/\log n). This strengthens the ω(1)\omega(1) lower bound from [K. Ambos-Spies, W. Merkle, and S.A. Terwijn, 2019, Normalized information distance and the oscillation hierarchy], see arXiv:1708.03583 .

Keywords

Cite

@article{arxiv.2002.06683,
  title  = {The normalized algorithmic information distance can not be approximated},
  author = {Bruno Bauwens and Ilya Blinnikov},
  journal= {arXiv preprint arXiv:2002.06683},
  year   = {2020}
}

Comments

14 pages, 2 figures

R2 v1 2026-06-23T13:43:20.055Z