If I start with matrix

$$A = \begin{bmatrix}a & b \\ c & d \end{bmatrix}$$

and I express it as a sum

$$A = \begin{bmatrix} w & x \\ y & z \end{bmatrix} + \begin{bmatrix}(a/w-1)w & (b/x-1)x \\ (c/y-1)y & (d/z-1)z \end{bmatrix}$$

I'll now call the first and second matrix on the RHS $B$ and $C$, respectively. For clarity,

$$B = \begin{bmatrix} w & x \\ y & z \end{bmatrix}$$

$$C = \begin{bmatrix}(a/w-1)w & (b/x-1)x \\ (c/y-1)y & (d/z-1)z \end{bmatrix}$$

My question is on the resulting eigenvectors and eigenvalues. Would the original eigenvalues and eigenvectors of $A$ be preserved in the sum of $B$ and $C$? In other words, would

$$\mbox{eig} (A) = \mbox{eig} (B) + \mbox{eig} (C)$$

since $A=B+C$ in the example above?

Also, does matrix $C$ have a special name? It looks like a "relative" matrix?

I should also note, there is nothing special about the matrix other than it is square.

  • $\begingroup$ If $B$ and $C$ commute, at best one can say 'eig($A$)$\subseteq$eig($B$)+eig($C$)'. $\endgroup$ – Arundhathi Jun 24 '16 at 6:26
  • $\begingroup$ @M.Vinay You are correct. I will edit that to prevent any confusion. $\endgroup$ – ThatsRightJack Jun 24 '16 at 8:13

Suppose we have a rather dull matrix

$$A=\begin{bmatrix} 1&2\\ 2&1 \end{bmatrix}\quad B=\begin{bmatrix} 1&1\\ 1&1 \end{bmatrix} \quad C=\begin{bmatrix} 0&1\\ 1&0 \end{bmatrix}$$

The eigenvalues of $A$ are $\lambda_1=3, \lambda_2=-1$ with eigenvectors $v_1=\begin{bmatrix} 1\\ 1 \end{bmatrix}, v_2=\begin{bmatrix} -1\\ 1 \end{bmatrix}$

The eigenvalues of $B$ are $\lambda_1=2,\lambda_2=0$ with eigenvectors $v_1=\begin{bmatrix} 1\\ 1 \end{bmatrix}, v_2=\begin{bmatrix} -1\\ 1 \end{bmatrix}$

The eigenvalues of $C$ are $\lambda_1=-1,\lambda_2=1$ with eigenvectors $v_1=\begin{bmatrix} -1\\ 1 \end{bmatrix}, v_2=\begin{bmatrix} 1\\ 1 \end{bmatrix}$

While it seems that $v_1+v_2$ are equal for all three matrices, there is no similar conclusion about the relationships among the eigenvectors as you have described.

  • $\begingroup$ Thanks for the answer. I'm able to follow your example, but if the matrix A is given by block elements a,b,c,d, where none of them are equal, does that make a difference in your example? $\endgroup$ – ThatsRightJack Jun 24 '16 at 5:46
  • $\begingroup$ Also, if I'm not mistaken, your example above seems to show that eig(A)=eig(B)+eig(C) holds? I obviously don't want to draw any generic conclusions, but if it is true under certain conditions, then I can at least check if it holds for my specific application. $\endgroup$ – ThatsRightJack Jun 24 '16 at 5:53
  • $\begingroup$ OP, sorry for not being clear. It seems that the example shows that eig(A)$\neq$eig(B)+eig(C) as you can see from the eigenvectors. Thus, since we have a counterexample, we are done. $\endgroup$ – Claire Jun 24 '16 at 5:59
  • $\begingroup$ I'm obviously not an expert in eigen problems, but is seems that all vectors are the same and the eigenvalues of B and C add up to A? Is there no significance to that? $\endgroup$ – ThatsRightJack Jun 24 '16 at 6:06
  • 1
    $\begingroup$ @Erica I thought that might be the cause of confusion. I will change that wording to prevent confusion. It sounds like M.Vinay is deferring to you for presenting the correct answer? About 5 comments up, M.Vinay gives a good mathematical explanation. Would you mind integrating that into your answer and making the necessary edits so that I can accept your answer. Also, do you have any comments about the idea of a "relative" matrix? Is that something that is well known or of any significance? (These last two questions, I'm just curious). Thanks for your help and time Erica. $\endgroup$ – ThatsRightJack Jun 24 '16 at 21:23

Based on the comments following the answer given by @Erica, I decided to post a solution which I feel is more correct. I asked @Erica to make the changes, but she never took the opportunity to. I copied the following statement from @M. Vinay in the discussion comments following the answer of @Erica. I feel this is the correct answer and the example from @Erica supports it if you change the eigenvalue and eigenvector labels for $C$.

@M. Vinay states:

If B and C have a common eigenvector, then A=B+C also has the same eigenvector and a corresponding eigenvalue that is the sum of the corresponding eigenvalues of B and C. For example, let v be an eigenvector of B corresponding to eigenvalue λ, and also of C corresponding to eigenvalue μ. Then Av=(B+C)v=Bv+Cv=λv+μv=(λ+μ)v. That is, v is an eigenvector of A, with eigenvalue λ+μ.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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