使用 n 次算术运算进行矩阵乘法
Multiplying matrices using n arithmetic operations
AI总结:
本文提出一种对自然数矩阵仅用n次算术运算(一次乘法加n-1次加法)的矩阵乘法算法,通过将输入编码为两个大数实现,虽受限于实际场景,但开辟了新的研究路径。
摘要:
众所周知,方阵乘法算法复杂度的下界至少需要 $n^2$ 次算术运算。这一结论基于以下推理:由于输入矩阵中共有 $2n^2$ 个数字,任何算法必然至少对每个数字操作一次。在本文中,我们表明对于某些特定实例,例如具有自然数条目的矩阵,情况并非一定如此。我们提出了一种算法,该算法执行一次乘法和 $(n-1)$ 次加法,因此总共使用 n 次算术运算。该方法巧妙之处在于将原始的 $2n^2$ 个元素编码为两个数量级更大的数字。因此,尽管对每个输入至少处理一次,但它依赖于较少的算术运算次数。在用于分析该问题的计算模型中,这种编码操作不可用,因此目前尚不清楚这项工作是否会影响矩阵乘法已接受的复杂度结果,但新算法的复杂度(在考虑编码的情况下)为 $3n^2 + 2n - 1$ 次运算。此外,由于乘法操作数数量级的指数增长,其实际使用仅限于问题的某些实例。尽管如此,这项工作提出了一种新颖的、受数学启发的算法,同时指向了一条替代研究路径,这为新颖算法以及矩阵乘法及其相关复杂度的分类开辟了可能性。
英文摘要:
It is widely known that the lower bound for the algorithmic complexity of square matrix multiplication resorts to at least $n^2$ arithmetic operations. The justification builds upon the following reasoning: given that there are $2 n^2$ numbers in the input matrices, any algorithm necessarily must operate on each at least once. In this paper, we show that this is not necessarily the case for certain instances of the problem, for instance matrices with natural number entries. We present an algorithm performing a single multiplication and $(n - 1)$ sums, therefore using n arithmetic operations. The ingenuity of the approach relies on encoding the original $2n^2$ elements as two numbers of much greater magnitude. Thus, though processing each of the inputs at least once, it relies on a lower count of arithmetic operations. In the computational model used to analyze this problem, such encoding operation is not available, thus it is not clear this work affects the currently accepted complexity results for matrix multiplication, but the new algorithm complexity (when taking into account the encodings) is $3n^2 + 2n - 1$ operations. In addition, given the exponential increase in multiplication operands magnitude, its practical usage is constrained to certain instances of the problem. Nonetheless, this work presents a novel mathematically inspired algorithm while pointing towards an alternative research path, which opens the possibility of novel algorithms and a taxonomy of matrix multiplications and associated complexities.
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。