2

According to Ryan Williams's answer $ALT\text{-}SPACE(a(n),\log n)\subseteq NSPACE(a(n)\log n)$.

Does there exist any better bound (for example something like $ALT\text{-}SPACE(a(n),\log n)\subseteq NSPACE(\log (a(n))\log n)$), or is it tight or an open problem?

Yuval Filmus
  • 276,994
  • 27
  • 311
  • 503
Mohsen Ghorbani
  • 537
  • 4
  • 12

1 Answers1

2

It's open, at least for the inclusion $ALTSPACE(\alpha(n),\log n)\subseteq NSPACE(f(n))$ with $\Omega(\log n)\leq f(n)\leq \alpha(n)^{o(1)}$. According to Chandra, Kozen, and Stockmeyer, we know $$DTIME(n^k)\subseteq ALTSPACE(n^{2k},\log n)\subseteq \mathsf{P}.$$

Therefore if it is true, then $\mathsf{P}\neq \mathsf{NPSPACE}$ which is not known to be true. On the other hand, rejecting it will yield $\mathsf{P}\neq \mathsf{NL}$, which is also not known to hold.

Wei Zhan
  • 1,183
  • 7
  • 16
  • Excuse me . I can't understand how $ALTSPACE(n^{2k},\log n)\subseteq P$. only things I know is $ASPACE(\log n)\subseteq P$ and the number of alternations is constant but if we replace the constant with a polynomial function how we could show that equation? can you explain a little more? Thank you again. – Mohsen Ghorbani Jul 24 '18 at 09:34
  • 2
    @MohsenGhorbani The proof of $ASPACE(\log n)\subseteq P$ works with any number of alternations, not only for constants. On the other hand with $s(n)=O(\log n)$ space there are at most $2^{O(s(n))}$ possible configurations, and thus $2^{O(s(n))}=n^{O(1)}$ alternations. – Wei Zhan Jul 24 '18 at 15:12