5

I got some homework in my school about matrix. These questions are seem so easy to solve but I always get stuck. Here they are:

  1. Let $A,B \in \mathbb{R}^{2017\times2017}$ matrices which satisfy the following equation. $$A^{-1} = (A+B)^{-1}-B^{-1}$$ and $\det(A^{-1})=2017.$ Find $\det(B)$.

My attempt:

\begin{equation*} \begin{split} (A+B)A^{-1} &= (A+B)\left[(A+B)^{-1}-B^{-1}\right] \quad \quad \text{multiplying both sides by (A+B)} \\ A^{-1}A +BA^{-1} &= (A+B)(A+B)^{-1}-(A+B)B^{-1} \\ I+BA^{-1} &= I - AB^{-1}-I\\ I+BA^{-1} &=-AB^{-1}\\ BA^{-1} +AB^{-1} +I&= O \end{split} \end{equation*} then I don't know how to continue.

2.Let $A,B\in \mathbb{R}^{2017 \times 2017}$ matrices which satisfy the equation $$AB^{2}-2BAB+B^{2}A=O$$ What is the largest eigenvalue of $AB-BA?$

$ABB+BBA=2BAB$

$ABB+BBA=BAB+BAB$

$ABB-BAB=BAB-BBA$

$(AB-BA)B=B(AB-BA)$

what is this means? I really need your thoughts, thanks in advance.

SutMar
  • 417

3 Answers3

7

Note that this is $A^{-1} +B^{-1}=(A+B)^{-1}$ i.e we have $$(A+B)(A^{-1} +B^{-1}) = I \implies I + AB^{-1}+BA^{-1}+I = I$$ or equivalently $$I + AB^{-1} + BA^{-1} = 0 \implies \begin{cases}B+A+BA^{-1}B = 0 \\ A + B + AB^{-1}A = 0\end{cases}$$

So $$BA^{-1}B = AB^{-1}A \implies (\det B)^3 = (\det A)^3 \implies \det B = \det A.$$

Zain Patel
  • 17,200
  • 5
  • 29
  • 59
2

Pertaining to 1, interestingly enough, I don't think such $A,B$ are possible for real matrices of odd dimension. We have $A^{-1} +B^{-1}=(A+B)^{-1}$, so $$(A+B)(A^{-1} +B^{-1}) = I,$$ $$I + AB^{-1}+BA^{-1}+I = I,$$ $$AB^{-1}+I+BA^{-1} = O.$$ Multiplying by $AB^{-1}$ on the left, we obtain $$(AB^{-1})^2+AB^{-1}+I=O.$$ Thus the minimal polynomial for $AB^{-1}$ divides $p(x) = x^2+x+1$. However, $p$ is irreducible over $\mathbb{R}$, so $p$ must the minimal polynomial. As the minimal polynomial is irreducible quadratic, the characteristic polynomial for $AB^{-1}$ must be a power of $p$, but this isn't possible as the degree of the characteristic polynomial is $2017$ which is odd.

Edit: If we assume the dimensions of $A$ and $B$ are even, this does lead to another solution to 1. The roots of $p$ and thus the eigenvalues of $AB^{-1}$ are $\lambda_1=-\frac{1}{2}+\frac{\sqrt3}{2}i, \lambda_2=-\frac{1}{2}-\frac{\sqrt3}{2}i$. Then we have $\det AB^{-1}=\lambda_1\lambda_2=1$, so $\det A = \det B$.

  • You are very sharp-eyed! +1 I guess the original question was set in an even-numbered year, but some lazy instructor just reused the same question and modified the size to 2017. – user1551 Jun 06 '17 at 17:18
  • @user1551 I really just got lucky! :-D I was trying to solve this problem by playing around with minimal polynomials and stumbled upon the contradiction. – Christian Sykes Jun 06 '17 at 17:26
  • 1
    @user1551 As to the laziness of the instructor, having been one myself, I can't really judge too harshly. I've certainly been guilty of similar crimes. – Christian Sykes Jun 06 '17 at 17:31
1

For your second question, we are given

$$p: AB^2 - 2BAB + B^2A=0.$$

We want to show that $AB-BA$ is nilpotent (i.e all eigenvalues of $AB-BA$ are zero.)

We will do this in four parts:

1.) show that for any linear operators $X, Y$, then $tr(XY-YX)=0;$

2.) show that, for the spectrum $\left \{\lambda_1,...,\lambda_n \right \}$ of any operator $X $, then, for all $k \in \Bbb{N},$ we have that $$tr(X^k) = {\lambda_1^k+...+\lambda_n^k};$$ 3.) show that $p \implies B $ commutes with $AB - BA$;

4.) show (by induction) that for our particular $A,B $ under $p $, there is an operator $X$ for each $k$ such that $$(AB-BA)^k = XB-BX$$

With all this we can say that for eigenvalues $\lambda_1,...,\lambda_n$ of $AB-BA$, then for all $k$ we have

$$\begin{equation}\begin{split}\lambda_1^k+...+\lambda_n^k&=tr[(AB-BA)^k] \\&= tr(XB-BX)\\& = 0\end{split}\end{equation}$$

Hence, $\lambda_i =0\: \forall i\in \left\{1,..., n \right\} $.

I'll leave you to figure out 1.) and 2.). Below are 3.) (which you've already shown) and 4.).

Part 3.)

\begin{equation}\begin{split} 0 &= AB^2-2BAB+B^2A\\&= AB^2 - BAB - BAB + B^2A\\&= (AB-BA)B - B(AB-BA) \end{split} \\ \end{equation}$ \implies (AB-BA)B=B(AB-BA)$

Part 4.)

$\text{At }k = 1, \text{then } X_1=A. \\ \text{Assume } (AB-BA)^k=X_kB-BX_k. \text{Then} $ \begin{align}\begin{split} (AB-BA)^{k+1}& = (AB-BA)^k(AB-BA) \\ & = (X_kB-BX_k)(AB-BA) \\&=X_kB(AB-BA)-BX_k (AB-BA)\\&=X_k(AB-BA)B-BX_k(AB-BA)\\&=X_{k+1}B-BX_{k+1} \end{split} \end{align}

bloomers
  • 1,158
  • first, thank you for giving me an answer to my post. But, how do you see that it should be nilpotent? and should I show it all (4 parts) ? – SutMar Jun 06 '17 at 12:21
  • 1
    An operator is nilpotent if and only if all of its eigenvalues are zero. This can be shown rigorously, but it may be easier to convince yourself of this by multiplying any upper-triangular matrix with $0's$ on the diagonal by itself. And then again, etc. It is easily observable that the product will eventually be $0$. Then do the same thing but change one of the diagonals to a non-zero quantity. You will observe that this will never go to zero no matter what power you raise it to. – bloomers Jun 06 '17 at 12:28
  • You're welcome. And there may be a more concise way to prove this, but these four parts are the most direct I could produce. So I would show all four. – bloomers Jun 06 '17 at 21:17