几乎催化计算
Almost-catalytic Computation
AI总结:
本文提出几乎催化计算模型,证明若问题及其补集均有几乎催化算法则可转化为 ZPP 算法,并借助纠错码构造高复杂度催化集,使 DSPACE(n^k) 和 DSPACE(log^k n) 类可被相应几乎催化类接受。
摘要:
为空间受限模型设计算法,并要求算法所用空间满足恢复条件,是 Buhrman 等人(2014)提出的催化计算模型所面临的一项重要挑战。受“除非恢复空间有用,否则无需恢复”这一场景的启发,我们定义 $ACL(A)$ 为能够被关于 $A$(我们称之为催化集)的几乎催化图灵机接受的语言类,该图灵机至多使用 $c\log n$ 个工作空间和 $n^c$ 个催化空间。我们证明,如果某个问题分别存在以 $A \subseteq Σ^*$ 及其补集为催化集的几乎催化算法,那么该问题可以由 ZPP 算法解决。利用这一结果,我们推导出:要设计催化算法,只需设计催化集为奇权重字符串集($PARITY$)的几乎催化算法即可。为此,我们考虑集合 $A$ 的两个复杂度度量,它们在 $PARITY$ 上取得最大值——随机投影复杂度(${\cal R}(A)$)和子立方体划分复杂度(${\cal P}(A)$)。借助纠错码,我们证明对所有 $k \ge 1$,存在语言 $A_k \subseteq Σ^*$,使得 $DSPACE(n^k) \subseteq ACL(A_k)$,其中对每个 $m \ge 1$,$\mathcal{R}(A_k \cap \{0,1\}^m) \ge \frac{m}{4}$ 且 $\mathcal{P}(A_k \cap \{0,1\}^m)=2^{m/4}$。这与催化机器模型形成对比,在催化机器模型中,尚不清楚它能否对任意 $ε> 0$ 接受 $DSPACE(\log^{1+ε} n)$ 中的所有语言。进一步改进催化集 $A$ 的划分复杂度,我们证明对所有 $k \ge 1$,存在 $A_k \subseteq \{0,1\}^*$,使得 $\mathsf{DSPACE}(\log^k n) \subseteq ACL(A_k)$,其中对每个 $m \ge 1$,$\mathcal{R}(A_k \cap \{0,1\}^m) \ge \frac{m}{4}$ 且 $\mathcal{P}(A_k \cap \{0,1\}^m)=2^{m/4+Ω(\log m)}$。
英文摘要:
Designing algorithms for space bounded models with restoration requirements on the space used by the algorithm is an important challenge posed about the catalytic computation model introduced by Buhrman et al. (2014). Motivated by the scenarios where we do not need to restore unless is useful, we define $ACL(A)$ to be the class of languages that can be accepted by almost-catalytic Turing machines with respect to $A$ (which we call the catalytic set), that uses at most $c\log n$ work space and $n^c$ catalytic space. We show that if there are almost-catalytic algorithms for a problem with catalytic set as $A \subseteq Σ^*$ and its complement respectively, then the problem can be solved by a ZPP algorithm. Using this, we derive that to design catalytic algorithms, it suffices to design almost-catalytic algorithms where the catalytic set is the set of strings of odd weight ($PARITY$). Towards this, we consider two complexity measures of the set $A$ which are maximized for $PARITY$ - random projection complexity (${\cal R}(A)$) and the subcube partition complexity (${\cal P}(A)$). By making use of error-correcting codes, we show that for all $k \ge 1$, there is a language $A_k \subseteq Σ^*$ such that $DSPACE(n^k) \subseteq ACL(A_k)$ where for every $m \ge 1$, $\mathcal{R}(A_k \cap \{0,1\}^m) \ge \frac{m}{4}$ and $\mathcal{P}(A_k \cap \{0,1\}^m)=2^{m/4}$. This contrasts the catalytic machine model where it is unclear if it can accept all languages in $DSPACE(\log^{1+ε} n)$ for any $ε> 0$. Improving the partition complexity of the catalytic set $A$ further, we show that for all $k \ge 1$, there is a $A_k \subseteq \{0,1\}^*$ such that $\mathsf{DSPACE}(\log^k n) \subseteq ACL(A_k)$ where for every $m \ge 1$, $\mathcal{R}(A_k \cap \{0,1\}^m) \ge \frac{m}{4}$ and $\mathcal{P}(A_k \cap \{0,1\}^m)=2^{m/4+Ω(\log m)}$.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。