1

Given matrices $A$ and $C$ such that $ACA=0$, show that characteristic polynomial of $AB$ and $A(B+C)$ is same for any matrix B.This question was in a worksheet for Cayley Hamilton Theorem. The order of the matrices was not mentioned so I assume I have to prove it in general $A, B, C \in M_{n \times n}(\mathbb{R})$
If $r(A)=0$ then $A=0$ and the proof is trivial.
If $r(A)=n$ then $A$ is invertible so $C=0$ and the proof is trivial again.
So for the non trivial case, $1<r(A)<n$ which also implies $A$ is a matrix with $det(A)=0$
$r(AB)\leq r(A)$ and $r(A(B+C))\leq r(A)$ so both matrices must also have a zero determinant.

My attempt was to show charpoly $AB$ is satisfied by $AB+AC$ and vice-versa to conclude the charpoly must be equal (which now that I think about it, will not prove their characteristic polynomials are equal!!)
$(AB+AC)^n = (AB)^{n-1}A(B+C)$ for $n\in \mathbb{N}$
Now let $det(xI-AB)=f(x) = c_0 + c_1x + ... + c_{n-1}x^{n-1} + x^n$
Evaluate $f(AB+AC) = c_0I + c_1(AB+AC) + ... + c_{n-1}(AB+AC)^{n-1} + (AB+AC)^n$
$f(AB+AC) = c_0I + c_1(AB+AC) + ... + c_{n-1}(AB)^{n-2}A(B+C) + (AB)^{n-1}A(B+C)$
$f(AB+AC) = f(AB) + c_1AC + c_2(AB)AC + ... + c_{n-1}(AB)^{n-2}AC + (AB)^{n-1}AC$

And this is where I was stumped. Any hints on how to approach this question? Or just the solution straight up also works. Thank you.

Nikhil
  • 21
  • $A':= AB$ and $N':= AC$, then apply the below duplicate link – user8675309 Aug 13 '24 at 16:12
  • @user8675309 I follow everything in that link but I'm unsure how the charpoly must be same given that $tr((A+N)^k) = tr(A^k)$ for all k – Nikhil Aug 13 '24 at 17:11
  • There are 4 proofs in that link -- I gave 2 of them. If you aren't familiar with the fact that for $n\times n$ matrices, if $\text{trace}\big(X^k\big) = \text{trace}\big(Y^k\big)$ for all $k\in \mathbb N$ then they have the same eigenvalues over $\mathbb C$ [e.g. by Newton's Identities or the simpler argument I gave here https://math.stackexchange.com/questions/3623345/products-of-matrices-in-either-order-have-the-same-characteristic-polynomial/3624060#3624060 ) then use one of the other 3 proofs. – user8675309 Aug 13 '24 at 17:19
  • actually the easiest trace identity argument is: if the $\text{trace}\big(X^k\big) = \text{trace}\big(Y^k\big)$ for $1\leq k\leq 2n$ then they have the same eigenvalues over $\mathbb C$ -- consider some eigenvalue $\lambda$ of $X$ with algebraic multiplicity of $m$. There is a degree $2n$ polynomial (e.g. using Vandermonde matrix or Lagrange interpolation) such that $p(\lambda) =1$ and $p$ sends all other eigenvalues for $X$ and $Y$ to $0\implies m=\text{trace}\big(p(X)\big)=\text{trace}\big(p(Y)\big)$ so $Y$ has $\lambda$ with algebraic multiplicity of $m$ as well. – user8675309 Aug 13 '24 at 17:37
  • For a more pedestrian argument (hence easier to follow), see my recent answer in that link: https://math.stackexchange.com/a/4958116 – Anne Bauval Aug 13 '24 at 18:50

0 Answers0