Eigenvectors of matrix equation $AX+XB^\text{T}=\lambda (CX+XD^\text{T})$

1k Views Asked by At

We have the matrix eigenvalue problem

$$AX+XB^\text{T}=\lambda (CX+XD^\text{T})$$

Where $\lambda$ is the eigenvalue, $X$ is $m\times n$ and plays the role of the eigenvector, and $A$, $B$, $C$, and $D$, are square matrices of appropiate size, and for my purposes, $C$ and $D$ are diagonal. Rearranging we get to

$$(A-\lambda C)X=-X(B-\lambda D)^\text{T}$$

Now, trying to solve this through separation of variables, choose $X=uv^\text{T}$

$$(A-\lambda C)uv^\text{T}=-uv^\text{T}(B-\lambda D)^\text{T}=muv^\text{T}$$

where $m$ is the separation constant. Now we separate

$$(A-\lambda C)u=mu$$ $$(B-\lambda D)v=-mv$$

My question is, how can I solve this system of eigenproblems? Is there a numerical algorithm for this?

Background: all this came from the matrix analogy of solving a PDE of the form

$$ \frac{1}{f(x)+g(y)}(L_x u+ L_y u)=\lambda u $$

Where $L_x$ and $L_y$ are univariate linear differential operators.