Let $T: V \rightarrow V$ be a linear operator.

I need to demonstrate that if all nonzero vectors of $V$ are eigenvectors of $T$, then there is one specific $\lambda \in K$ such that $T(v) = \lambda v$, for all $v \in V$.

I understand that, if all nonzero vectors of $V$ are eigenvectors of $T$, then $T$ must be a scaling transformation. It just stretch or shrinks vectors, but doesn't change their directions.

So, the statement says that if it happens, then, there is a single $\lambda$ such that $T(v) = \lambda v$. In other words, if there is such transformation, then it scales all vectors by the same scalar $\lambda$.

Applying the transformation to our standard basis vectors, we have: $$ T(e_1) = \lambda_1 e_1 \\ T(e_2) = \lambda_2 e_2 \\ \vdots \\ T(e_n) = \lambda_n e_n $$

I understand I need to prove that $\lambda_1 = \lambda_2 = \dots = \lambda_n$, but I can't see how!


$$ v = c_1e_1 + c_2e_2 + \dots + c_ne_n \\ T(v) = \mu v = \lambda_1c_1e_1 + \lambda_2c_2e_2 + \dots + \lambda_nc_ne_n \\ $$

Since what's multiplying $v$ coordinates is $\lambda_i$, then all of them must be $\mu$. I'm not sure how to 'mathematize' this. Is this idea correct?

EDIT 2 Extending the left hand side of EDIT 1, we have: $$ \mu v = \lambda_1c_1e_1 + \dots + \lambda_nc_ne_n \\ \mu(c_1e_1 + \dots + c_ne_n) = \lambda_1c_1e_1 + \dots + \lambda_nc_ne_n \\ \mu c_1e_1 + \dots + \mu c_ne_n = \lambda_1c_1e_1 + \dots + \lambda_nc_ne_n \\ $$

And since $e_i$ are linearly independent, $\mu = \lambda_1 = \lambda_2 = \dots = \lambda_n$. Is this proof correct?

  • 2
    $\begingroup$ Hint: consider $e_1+e_2$. $\endgroup$ – Chris Eagle May 17 '12 at 18:23
  • $\begingroup$ I believe I've come to something. If ALL vectors of V are eigenvectors, then $T(v)$ is also an eigenvector, since T maps $V \rightarrow V$. Then it can be written as $\mu v$, where $\mu$ is another eigenvalue. Than $\mu = \lambda_1 = \lambda_2 = \dots = \lambda_n$. Is it correct? $\endgroup$ – João Daniel May 17 '12 at 18:28
  • $\begingroup$ Your edit shows that $\mu = \lambda_i c_i$, $\forall i$, since the $e_i$ are linearly independent. $\endgroup$ – copper.hat May 17 '12 at 18:40
  • $\begingroup$ But $\mu$ is a scalar multiplying $v$, and $c_i$ are $v$ coordinates on standard basis, i.e., $c_ie_i$ are $v$ coordinates. If $\mu$ is multiplying $v$, why does $c_i$ comes to $\mu$ equation? $\endgroup$ – João Daniel May 17 '12 at 18:44
  • $\begingroup$ You have now asked THREE different questions in the same post. What happens to the answers to your FIRST question, posted while this was the only one? $\endgroup$ – Did May 17 '12 at 19:13

Since all $v\in V$ are eigenvectors, we can choose $e_i$, the $i$th unit vector. Then by assumption we have $T e_i = \lambda_i e_i$ for some $\lambda_i$. It follows that $T$ is diagonal, with elements $\lambda_1,...,\lambda_n$ on the diagonal.

Now choose $v=e_1+...+e_n$, again for some $\lambda$, we have $Tv=\lambda v$, so we have $$T v = T(e_1+...+e_n) = \lambda_1 e_n +... + \lambda_n e_n = \lambda (e_1+...+e_n).$$ Since the $e_i$ are linearly independent, it follows that $\lambda = \lambda_1 = ... = \lambda_n$. Hence $Tx = \lambda x$, $\forall x$.

  • 1
    $\begingroup$ When you write $v = e_1 + \dots + e_n$, you're considering it for unit vectors. Shouldn't you consider $v = c_1e_1 + \dots + c_ne_n$ with $c_i \in K$, in order to extend it to all vectors of $V$? $\endgroup$ – João Daniel May 17 '12 at 18:51
  • 1
    $\begingroup$ What needs to be extended? If you know $T$ on a basis, you know $T$ everywhere, by linearity. First I show $T$ is diagonal, since $T(\sum x_i e_i) = \sum \lambda_i x_i e_i$, then I show that all diagonal elements are equal. $\endgroup$ – copper.hat May 17 '12 at 19:02
  • $\begingroup$ Sorry, I got confused mixing it with my try of demonstration. I'm still trying to assimilate your answer ;). $\endgroup$ – João Daniel May 17 '12 at 19:11
  • 3
    $\begingroup$ This is a very nice proof. +1 $\endgroup$ – Michael Joyce May 17 '12 at 19:43
  • $\begingroup$ @copper.hat $v$ is just a specific vector that sets $v = e_1+...+e_n$, you shouldn't show that it works for all $u \in V$? $\endgroup$ – Avishay28 Sep 25 '17 at 7:46

Hint: Assume that $Tu=\lambda u$ and $Tv=\mu v$ for some nonzero vectors $u$ and $v$ and some $\lambda$ and $\mu$.

  • Show that $\{u,v\}$ is a linearly independent family.
  • Show that $\{u+v,au+bv\}$ is a linearly independent family, for every $a\ne b$.
  • Show that $\{T(u+v),u+v\}$ is not linearly independent.
  • Conclude that $\lambda=\mu$.
  • $\begingroup$ What is a linearly free family? $\endgroup$ – copper.hat May 17 '12 at 18:38
  • $\begingroup$ @copper.hat: A Gallicism. $\endgroup$ – Did May 17 '12 at 18:44
  • $\begingroup$ A straight couple from the 60's maybe? $\endgroup$ – copper.hat May 17 '12 at 18:49
  • $\begingroup$ @copper.hat: Nice... :-) $\endgroup$ – Did May 17 '12 at 18:58

Assume for a contradiction that $v,w$ are eigenvectors for $\lambda\neq\mu$, respectively, (in particular they are nonzero) and that $T(v+w)=\nu(v+w)$ for some $\nu\in K$. Since any nonzero scalar multiple of an eigenvector is an eigenvector for the same eigenvalue, $v$ and $w$ cannot be linearly dependent. Then by this linear independence $\nu v+\nu w=T(v+w)=T(v)+T(w)=\lambda v+\mu w$ implies $(\nu,\nu)=(\lambda,\mu)$, which contradicts $\lambda\neq\mu$.

So if all nonzero vectors are eigenvectors, then all of them must be so for the same eigenvalue$~\lambda$, and one has $T=\lambda I$. (Pedantically, if $\dim V=0$ there are no eigenvectors at all, and one is free to choose$~\lambda$; "specific" in the question is not justified in this case.)


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.