{"title":"在矩阵乘法时间内计算克雷洛夫迭代","authors":"Vincent Neiger, Clément Pernet, Gilles Villard","doi":"arxiv-2402.07345","DOIUrl":null,"url":null,"abstract":"Krylov methods rely on iterated matrix-vector products $A^k u_j$ for an\n$n\\times n$ matrix $A$ and vectors $u_1,\\ldots,u_m$. The space spanned by all\niterates $A^k u_j$ admits a particular basis -- the \\emph{maximal Krylov basis}\n-- which consists of iterates of the first vector $u_1, Au_1, A^2u_1,\\ldots$,\nuntil reaching linear dependency, then iterating similarly the subsequent\nvectors until a basis is obtained. Finding minimal polynomials and Frobenius\nnormal forms is closely related to computing maximal Krylov bases. The fastest\nway to produce these bases was, until this paper, Keller-Gehrig's 1985\nalgorithm whose complexity bound $O(n^\\omega \\log(n))$ comes from repeated\nsquarings of $A$ and logarithmically many Gaussian eliminations. Here\n$\\omega>2$ is a feasible exponent for matrix multiplication over the base\nfield. We present an algorithm computing the maximal Krylov basis in\n$O(n^\\omega\\log\\log(n))$ field operations when $m \\in O(n)$, and even\n$O(n^\\omega)$ as soon as $m\\in O(n/\\log(n)^c)$ for some fixed real $c>0$. As a\nconsequence, we show that the Frobenius normal form together with a\ntransformation matrix can be computed deterministically in $O(n^\\omega\n\\log\\log(n)^2)$, and therefore matrix exponentiation~$A^k$ can be performed in\nthe latter complexity if $\\log(k) \\in O(n^{\\omega-1-\\varepsilon})$, for\n$\\varepsilon>0$. A key idea for these improvements is to rely on fast\nalgorithms for $m\\times m$ polynomial matrices of average degree $n/m$,\ninvolving high-order lifting and minimal kernel bases.","PeriodicalId":501033,"journal":{"name":"arXiv - CS - Symbolic Computation","volume":"25 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-02-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Computing Krylov iterates in the time of matrix multiplication\",\"authors\":\"Vincent Neiger, Clément Pernet, Gilles Villard\",\"doi\":\"arxiv-2402.07345\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Krylov methods rely on iterated matrix-vector products $A^k u_j$ for an\\n$n\\\\times n$ matrix $A$ and vectors $u_1,\\\\ldots,u_m$. The space spanned by all\\niterates $A^k u_j$ admits a particular basis -- the \\\\emph{maximal Krylov basis}\\n-- which consists of iterates of the first vector $u_1, Au_1, A^2u_1,\\\\ldots$,\\nuntil reaching linear dependency, then iterating similarly the subsequent\\nvectors until a basis is obtained. Finding minimal polynomials and Frobenius\\nnormal forms is closely related to computing maximal Krylov bases. The fastest\\nway to produce these bases was, until this paper, Keller-Gehrig's 1985\\nalgorithm whose complexity bound $O(n^\\\\omega \\\\log(n))$ comes from repeated\\nsquarings of $A$ and logarithmically many Gaussian eliminations. Here\\n$\\\\omega>2$ is a feasible exponent for matrix multiplication over the base\\nfield. We present an algorithm computing the maximal Krylov basis in\\n$O(n^\\\\omega\\\\log\\\\log(n))$ field operations when $m \\\\in O(n)$, and even\\n$O(n^\\\\omega)$ as soon as $m\\\\in O(n/\\\\log(n)^c)$ for some fixed real $c>0$. As a\\nconsequence, we show that the Frobenius normal form together with a\\ntransformation matrix can be computed deterministically in $O(n^\\\\omega\\n\\\\log\\\\log(n)^2)$, and therefore matrix exponentiation~$A^k$ can be performed in\\nthe latter complexity if $\\\\log(k) \\\\in O(n^{\\\\omega-1-\\\\varepsilon})$, for\\n$\\\\varepsilon>0$. A key idea for these improvements is to rely on fast\\nalgorithms for $m\\\\times m$ polynomial matrices of average degree $n/m$,\\ninvolving high-order lifting and minimal kernel bases.\",\"PeriodicalId\":501033,\"journal\":{\"name\":\"arXiv - CS - Symbolic Computation\",\"volume\":\"25 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-02-12\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - CS - Symbolic Computation\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2402.07345\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Symbolic Computation","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2402.07345","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Computing Krylov iterates in the time of matrix multiplication
Krylov methods rely on iterated matrix-vector products $A^k u_j$ for an
$n\times n$ matrix $A$ and vectors $u_1,\ldots,u_m$. The space spanned by all
iterates $A^k u_j$ admits a particular basis -- the \emph{maximal Krylov basis}
-- which consists of iterates of the first vector $u_1, Au_1, A^2u_1,\ldots$,
until reaching linear dependency, then iterating similarly the subsequent
vectors until a basis is obtained. Finding minimal polynomials and Frobenius
normal forms is closely related to computing maximal Krylov bases. The fastest
way to produce these bases was, until this paper, Keller-Gehrig's 1985
algorithm whose complexity bound $O(n^\omega \log(n))$ comes from repeated
squarings of $A$ and logarithmically many Gaussian eliminations. Here
$\omega>2$ is a feasible exponent for matrix multiplication over the base
field. We present an algorithm computing the maximal Krylov basis in
$O(n^\omega\log\log(n))$ field operations when $m \in O(n)$, and even
$O(n^\omega)$ as soon as $m\in O(n/\log(n)^c)$ for some fixed real $c>0$. As a
consequence, we show that the Frobenius normal form together with a
transformation matrix can be computed deterministically in $O(n^\omega
\log\log(n)^2)$, and therefore matrix exponentiation~$A^k$ can be performed in
the latter complexity if $\log(k) \in O(n^{\omega-1-\varepsilon})$, for
$\varepsilon>0$. A key idea for these improvements is to rely on fast
algorithms for $m\times m$ polynomial matrices of average degree $n/m$,
involving high-order lifting and minimal kernel bases.