A pair of square matrices $X$ and $Y$ are called similar if there exists a nonsingular matrix $T$ such that $T^{-1}XT=Y$ holds. It is known that the transformation matrix $T$ is not unique for given $X$ and $Y$. I'm just wondering whether those non-unique transformation matrices would have any relation among themselves, like having column vectors with same directions...

What I want to mean is: Given $X$ and $Y$ a pair of similar matrices, if $S$ and $T$ are two possible transformation matrices satisfying $S^{-1}XS = T^{-1}XT = Y$, is there any generic (apart from scaling) relation between $T$ and $S$ (e.g., direction of column vectors)?

For a specific example, consider $X = \begin{bmatrix} A & BK\\ C & 0 \end{bmatrix}$ and $Y = \begin{bmatrix} A+A^{-1}BKC & -A^{-1}BKCA^{-1}B\\ KC & -KCA^{-1}B \end{bmatrix}$. Assuming $K$ to be invertible it can be shown that $X$ and $Y$ are similar with transformation matrix $T = \begin{bmatrix} I & -A^{-1}B\\ 0 & K^{-1} \end{bmatrix}$. Can there be any other matrix $S$ which will be independent of $K$, and would result $S^{-1}XS=Y$?

You can always use $S=\alpha\,T$ with $\alpha$ some non-zero scalar not equal to one. Another way of finding more solutions is by using the Kronecker product, which allows you to write the transformation as

\begin{align} X\,T &= T\,Y \\ (I \otimes X)\,\text{vec}\,T &= (Y^\top \otimes I)\,\text{vec}\,T \\ (I \otimes X - Y^\top \otimes I)\,\text{vec}\,T &= 0 \end{align}

Using any (non-zero) vector which lies in the null space of $I \otimes X - Y^\top \otimes I$ for $\text{vec}\,T$ will satisfy the equation (however I am not sure if every $T$ would be invertible). If the dimension of the null space is one then you can only do the scalar multiplication. If the dimension is bigger then one you can also take a linear combinations of linear independent solutions (this both holds for $\text{vec}\,T$ and $T$, since the vectorization transformation and its inverse are also linear).

Your Answer

 

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.