Problem. A be an $m\times n$ matrix of rank $m$ with $n>m$. If for some nonzero real $\alpha$ we have $$x^tA A^tx=\alpha x^tx ~~~~ \forall x \in \Bbb{R}^m$$ then $A^tA$ has

  1. exactly two distinct eigen values

  2. $0$ as an eigen value with multiplicity $n-m$

  3. $\alpha$ is a non zero eigen value.

  4. exactly two non zero eigen values.

My attempt.

We know, $rank(A^tA)=rank(A)=m$. So, $Nullity(A^tA)=n-m>0$ So, $0$ is an eigen value of $A^tA$ with geometric multiplicity $n-m$. But $A^tA$ is symmetric so is diagonalizable i.e. $0$ is regular. Hence $0$ is an eigen value of (algebraic) multiplicity $n-m$ i.e. Option 2 is correct. (without using the given equation)

Now we have to conclude about the rest. I think for this we have to use the given equation.

I start by checking for the non zero eigen values of $A^tA$ (In fact, $A^tA$ should have $m$ non zero real eigen values)

Let $\lambda$ be a non zero eigen value of $A^tA$. Then there exists a non-null vector $v \in \Bbb{R}^n$ such that $A^tAv=\lambda v$. Now to use the given equation I think we need a vactor $x \in \Bbb{R}^m$. So, I try to choose $x=Av$ in that equation.....But I can't get any conclusion from there...!!

Any help is appreciated. Thank you.

N.B: This question has an answer here. But I cannot find out how to use $AA^t=\alpha I$ further.

EDIT: Well, I have found a way to proceed further:

Let $\lambda$ be a non-zero eigen value of $A^tA$. Then it can be proved it is also an eigen value of $AA^t$ (see here). Therefore, there exits a non-zero vector $v \in \Bbb{R}^m$ such that $AA^tv=\lambda v$. Again from the given equation we have, $v^tAA^tv=\alpha v^tv$, i.e., $v^t(AA^tv)=\alpha v^tv$, or, $\lambda v^tv=\alpha v^tv$ this implies, $\lambda = \alpha$.

Hence the only non-zero eigen value of $A^tA$ is $\alpha$.

So, finally Correct options are $1,2$ and $3$.

Please tell me whether I am right or made any mistake?


Your logic looks good to me, though it can be done more simply than you have it. You were on the right track with setting $x = Av$. Then $$A^tx = A^tAv = \lambda v\\AA^tx = \lambda Av = \lambda x\\x^tAA^tx = \lambda x^tx\\\alpha x^tx = \lambda x^tx$$ Since $\|x\| \ne 0$, we have $\lambda = \alpha$.


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.