矩阵迹的性质

一些常用的矩阵期望计算准则

Posted by Joker Hook on September 11, 2021

基础性质

设$r$为标量,则有

\[tr(\pmb{A}+\pmb{B}) = tr(\pmb{A}) + tr(\pmb{B})\] \[tr(r\pmb{A}) = r \cdot tr(\pmb{A})\] \[tr(\pmb{A}) = tr(\pmb{A}^T)\]

矩阵乘积的迹

\[tr(\pmb{A}\pmb{B}) = tr(\pmb{B}\pmb{A})\] \[tr(\pmb{A}\pmb{B}\pmb{C}) = tr(\pmb{B}\pmb{C}\pmb{A}) = tr(\pmb{C}\pmb{A}\pmb{B})\]

当矩阵$\pmb{A}, \pmb{B}, \pmb{C}$为$n \times n$的对称矩阵时,如下矩阵乘积转换成立:

\[\begin{align} tr(\pmb{A}\pmb{B}\pmb{C}) &= tr(\pmb{B}\pmb{C}\pmb{A}) = tr(\pmb{C}\pmb{A}\pmb{B}) \\ &=tr(\pmb{A}\pmb{C}\pmb{B}) = tr(\pmb{C}\pmb{B}\pmb{A}) = tr(\pmb{B}\pmb{A}\pmb{C}) \end{align}\]