arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2408.16906math.FAmath.DGmath.MG
收藏

酉矩阵片段特征值之和的凸性

Convexity of sums of eigenvalues of a segment of unitaries

Gabriel Larotonda, Martin Miglioli

AI总结:

本文证明酉矩阵路径前m个特征值角度之和在单位矩阵邻域内是凸函数,并刻画了最优邻域,进而将结果推广到特殊酉群中的Ad-不变Finsler范数。

摘要:

对于一个n×n酉矩阵u=e^z,其中z为反厄米矩阵,u的角度是其谱的辐角,即-iz的谱。对于1≤m≤n,我们证明s_m(t),即酉矩阵路径t↦e^{tx}e^y的前m个角度之和,是t的凸函数(前提是该路径保持在单位矩阵的一个邻域内)。该邻域用矩阵的算子范数描述,并且是最优的。我们证明当所有映射t↦s_m(t)都是线性时,x与y可交换。文中给出了在酉群中酉不变范数方面的若干应用。然后我们将这些应用推广到特殊酉矩阵群中的Ad-不变Finsler范数。最后这个结果是通过证明紧半单李群K中任何Ad-不变Finsler范数都是一族我们称之为轨道范数的上确界而得到的,这些轨道范数由K的Killing型诱导。

英文摘要:

For a $n\times n$ unitary matrix $u=e^z$ with $z$ skew-Hermitian, the angles of $u$ are the arguments of its spectrum, i.e. the spectrum of $-iz$. For $1\le m\le n$, we show that $s_m(t)$, the sum of the first $m$ angles of the path $t\mapsto e^{tx}e^y$ of unitary matrices, is a convex function of $t$ (provided the path stays in a vecinity of the identity matrix). This vecinity is described in terms of the opertor norm of matrices, and it is optimal. We show that the when all the maps $t\mapsto s_m(t)$ are linear, then $x$ commutes with $y$. Several application to unitarily invariant norms in the unitary group are given. Then we extend these applications to $Ad$-invariant Finsler norms in the special unitary group of matrices. This last result is obtained by proving that any $Ad$-invariant Finsler norm in a compact semi-simple Lie group $K$ is the supremum of a family of what we call orbit norms, induced by the Killing form of $K$.

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。

↑