在网格上计算永久变量
Computing Permanents on a Trellis
摘要:
自Ryser(1963)和Valiant(1979)的工作以来,计算矩阵的积和式的问题就引起了人们的兴趣。另一方面,自20世纪60年代以来,网格在编码理论中得到了广泛的研究。在这项工作中,我们建立了两个域之间的连接。我们介绍了典型的网格$T_n$,代表所有的排列,并表明,永久的$n$由$n$矩阵$A$可以计算为流在这个网格。在一定的规范化下,基于网格的方法调用的操作比最知名的精确方法略少。此外,如果$A$具有结构,则$T_n$变得适合于顶点合并,从而显著降低其复杂性。- 重复行:假设$A$只有$t<n$不同的行。由于Clifford和Clifford(2020),计算$per(A)$的最佳已知方法具有复杂性$O(n^{t+1})$。合并$T_n$中的顶点,我们得到一个简化的网格,具有复杂性$O(n^t)$。- 订单统计:使用网格,我们计算的联合分布的$t$顺序统计的$n$独立,但不相同的分布,随机变量在时间$O(n^{t+1})$。以前,多项式时间方法仅在变量来自两个不相同的分布时才为人所知。- 稀疏矩阵:假设$A$中的每个条目都是非零的,概率为$d/n$,其中$d$是常数。我们发现,$T_n$可以修剪成指数级更少的顶点,导致复杂性$O(ϕ^n)$与$ϕ<2$。- TSP:将$T_n$与另一个表示遍历的网格相交,我们得到一个表示循环排列的网格。使用后一个网格来解决旅行销售员问题恢复了著名的Held-Karp算法。值得注意的是,在所有情况下,使用网格理论中的已知技术来获得简化网格。我们希望其他的网格理论的结果适用于其他结构矩阵。
英文摘要:
The problem of computing the permanent of a matrix has attracted interest since the work of Ryser(1963) and Valiant(1979). On the other hand, trellises were extensively studied in coding theory since the 1960s. In this work, we establish a connection between the two domains. We introduce the canonical trellis $T_n$ that represents all permutations, and show that the permanent of a $n$ by $n$ matrix $A$ can be computed as a flow on this trellis. Under certain normalization, the trellis-based method invokes slightly less operations than best known exact methods. Moreover, if $A$ has structure, then $T_n$ becomes amenable to vertex merging, thereby significantly reducing its complexity. - Repeated rows: Suppose $A$ has only $t<n$ distinct rows. The best known method to compute $per(A)$, due to Clifford and Clifford (2020), has complexity $O(n^{t+1})$. Merging vertices in $T_n$, we obtain a reduced trellis that has complexity $O(n^t)$. - Order statistics: Using trellises, we compute the joint distribution of $t$ order statistics of $n$ independent, but not identically distributed, random variables in time $O(n^{t+1})$. Previously, polynomial-time methods were known only when the variables are drawn from two non-identical distributions. - Sparse matrices: Suppose each entry in $A$ is nonzero with probability $d/n$ with $d$ is constant. We show that $T_n$ can be pruned to exponentially fewer vertices, resulting in complexity $O(ϕ^n)$ with $ϕ<2$. - TSP: Intersecting $T_n$ with another trellis that represents walks, we obtain a trellis that represents circular permutations. Using the latter trellis to solve the traveling salesperson problem recovers the well-known Held-Karp algorithm. Notably, in all cases, the reduced trellis are obtained using known techniques in trellis theory. We expect other trellis-theoretic results to apply to other structured matrices.
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。