6
$\begingroup$

Problem: Let $A$ and $B$ be real orthogonal matrices, $n$x$n$, where $n$ is an odd number. Prove that, at least one of the matrices $A+B$ and $A-B$ has to be singular.

What have I done so far:

-Since matrices $A$ and $B$ are real orthogonal, it means that their determinants are $-1$ or $+1$

First I observed matrix $A+B$: $$A+B=AI+BI=ABB^T+BAA^T=AB(B^T+A^T)=AB(A+B)^T\Rightarrow$$

$$det(A+B)=det(AB)det(A+B)^T=det(A)det(B)det(A+B)$$

So, if $detA=detB$ we have $det(A+B)=det(A+B)$ which tells me nothing. But, if $detA=-detB$ we have $det(A+B)=-det(A+B)$ which can only hapen if $det(A+B)=0$ making $A+B$ singular.

Then, I tried the same for $A-B$: $$A-B=AI-BI=ABB^T-BAA^T=AB(B^T-A^T)=AB(A-B)^T\Rightarrow$$

$$det(A-B)=det(AB)det(A-B)^T=det(A)det(B)det(A-B)$$

So, if $detA=-detB$ we have $det(A-B)=-det(A-B)$ which can happen if $det(A-B)=0$ making $A-B$ singular.

Is this correct or should I do matrix $A-B$ differently? I am also a little confused why does it have to be said that $n$ is an odd number? What should that tell me?

Any help is greatly appreciated.

$\endgroup$
4
  • 2
    $\begingroup$ See here: math.stackexchange.com/questions/491344/… $\endgroup$ May 30 '16 at 14:32
  • $\begingroup$ Hint: the fact that $n$ is odd tells you that either $\det A = - \det B$ or $\det A = -\det (-B)$. $\endgroup$
    – lulu
    May 30 '16 at 14:32
  • 1
    $\begingroup$ How do you justify the third equal sign you wrote? $\endgroup$ May 30 '16 at 14:34
  • 1
    $\begingroup$ Every real eigenvalue of an orthogonal matrix has absolute value 1; if $n$ is odd, then the orthogonal matrices must have a real eigenvalue, since complex eigenvalues come in complex-conjugate pairs $\endgroup$ May 30 '16 at 14:36
8
$\begingroup$

(1) If $X$ is real $n\times n$ skew symmetric matrix, i.e., $X+X^T=0$, and if $T$ means transpose, then $$ {\rm det}\ X={\rm det}\ -X^T={\rm det}\ (-I)X={\rm det}\ X (-1)^n $$ where $I$ is an identity. So if $n$ is odd then ${\rm det}\ X=0$

(2) Assume that $A,\ B$ are orthogonal Then since $AA^T=I=BB^T$, $$ (A+B)(A-B)^T= BA^T - AB^T $$

Note that if $Y=BA^T$, then $(A+B)(A-B)^T=Y-Y^T$ is skew symmetric. So ${\rm det}\ (A+B)(A-B)^T =0$. So at leat one of $A+B,\ A-B$ is singular.

$\endgroup$
2
  • $\begingroup$ How do you conclude that $Y-Y^T$ is skew symmetric? $\endgroup$ May 27 '17 at 16:53
  • $\begingroup$ It is followed from direct computation $(Y-Y^T)^T=-Y + Y^T=-(Y-Y^T)$ $\endgroup$
    – HK Lee
    May 27 '17 at 16:57
5
$\begingroup$

You can write $$ A\pm B=A(\mathbf 1 \pm A^T B)=A(\mathbf 1 \pm Q), \quad Q\textrm{ orthogonal} $$ Then $A\pm B $ is singular iff $\mathbf 1 \pm Q $ is singular. But $Q$ has always an eigenvector $v$ with eigenvalue $\pm 1$ when $n$ is odd, therefore $\mathbf 1 -Q $ or $\mathbf 1 + Q$ has to be singular.

$\endgroup$
0
$\begingroup$

Hint Since $A$ is invertible, $A \pm B$ is singular iff $A^{-1} (A \pm B) = I \pm A^{-1} B$ is. On the other hand $A^{-1} B$ is orthogonal (the set of orthogonal matrices is closed under inverses and multiplication), and since $n$ is odd, it has $\mp 1$ as an eigenvalue (every eigenvalue of an orthogonal matrix has unit modulus, and every odd-sized real matrix has a real eigenvalue). Now, use the fact that if $\lambda$ is an eigenvalue of $C$, then $\lambda + \mu$ is an eigenvalue of $A + \mu I$.

$\endgroup$

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.