Let $A\in\mathbb{F}^{n×n}$ and define $L: \mathbb{F}^{n×n}\rightarrow \mathbb{F}^{n×n}$ by $L(X)=AX$. If $\lambda$ is an eigenvalue of $A$ with geometric multiplicity 1, show that the geometric multiplicity of $\lambda$ as an eigenvalue of $L$ is at least $n$.

I have already proven that $\sigma (A)=\sigma (L)$. This connects with the problem above.

Any hint will do and is a great help. Thanks.


I don't think too much spectral theory is needed to solve this problem, to wit:

If $\lambda \in \Bbb F$ is an eigenvalue of $A$ of geometric multiplicity $1$, then the eigenspace corresponding to $\lambda$ is a $1$-dimensional subspace of $\Bbb F^n$; thus there is a vector

$0 \ne \vec x \in \Bbb F^n \tag 1$


$A\vec x = \lambda \vec x; \tag 2$

furthermore, any non-zero vector

$0 \ne \vec y \in \Bbb F^n \tag 3$


$A \vec y = \lambda \vec y \tag 4$

must be collinear with $\vec x$:

$\exists 0 \ne \alpha \in \Bbb F, \; \vec y = \alpha \vec x. \tag 5$

Next, consider the $n$ matrices

$X_i \in \Bbb F^{n \times n}, \; 1 \le i \le n, \tag 6$

where the $i$-th column of $X_i$ is $\vec x$, and all other columns are $0$; that is, the $X_i$ take the form

$X_i = [0 \; 0 \; \ldots \; \vec x \; \ldots \; 0 \; 0]. \tag 7$

Now any

$Y \in \Bbb F^{n \times n} \tag 8$

may be written

$Y = [\vec y_1 \; \vec y_2 \; \ldots \; \vec y_n], \tag 9$

with each

$\vec y_i \in \Bbb F^n; \tag{10}$

the action of $L$ on $Y$ is thus given by

$LY = [A\vec y_1 \; A\vec y_2 \ldots \; A\vec y_n]; \tag{11}$

it follows that

$LX_i = [0 \; 0 \; \ldots \; A\vec x \; \ldots \; 0 \; 0] = [0 \; 0 \; \ldots \; \lambda \vec x \; \ldots \; 0 \; 0] = \lambda [0 \; 0 \; \ldots \; \vec x \; \ldots \; 0 \; 0] = \lambda X_i; \tag {12}$

that is, each $X_i$ is a $\lambda$-eigenvector of $L$ in $\Bbb F^{n \times n}$; furthermore, the $X_i$ are linearly independent over $\Bbb F$, for given any

$a_i \in \Bbb F, \; 1 \le i \le n, \tag{13}$

we have

$\displaystyle \sum_1^n a_iX_i = [a_1 \vec x \; a_2\vec x \; \ldots \; a_n \vec x] \ne 0 \tag{14}$

provided at least one $a_i \ne 0$.

We have thus demonstrated the existence of $n$ linearly independent $\lambda$-eigenvectors of $L$ in $F^{n \times n}$, that is, that the geometric multiplicity of $\lambda$ as an eigenvalue of $L:\Bbb F^n \to \Bbb F^n$ is at least $n$. $OE\Delta$.

Nota Bene: Based upon what we have done above, we can, with only a little extra work, show that in fact the geometric multiplicity of $\lambda$ as an eigenvalue of $L$ is in fact precisely $n$; for with $Y$ as in (8)-(11),

$LY = \lambda Y \Longrightarrow A \vec y_i = \lambda \vec y_i, \; 1 \le i \le n; \tag{15}$

then as noted above in (3)-(5) we have

$\vec y_i = \alpha_i \vec x, \; 1 \le i \le n, \tag{16}$

and so

$Y = \displaystyle \sum_1^n \alpha_i X_i; \tag{17}$

that is, every eigenvector $Y$ of $L$ lies in $\text{span}\{ X_i, \; 1 \le i \le n \}$; this proves the dimension of the $\lambda$-eigenspace of $L$ is precisely $n$, and hence the geometric multiplicity of $\lambda$ is exactly $n$ as well. End of Note.


Hint: If $Ax = \lambda x$ for a non-zero $x \in \Bbb F^n$, then $A(xy^T) = \lambda(xy^T)$ for any vector $y \in \Bbb F^n$.


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.