Let,$A,B$ are two $n\times n,(n\ge 2)$ non-singular matrices with real entries $(a)$ If $A^{-1}+B^{-1}=(A+B)^{-1}$ , then show that $\text{det(A)}=\text{det(B)}.$ I am getting $$B^{-1}A+A^{-1}B=-I_n.$$ Then how to proceed
-
2Welcome to MSE. You haven't actually asked a question - you've just stated something. – Feb 08 '18 at 16:17
-
2Welcome to MSE, this site helps in solving mathematical questions and it is not dedicated to solve homeworks. for that it is highly recommended to share us your own thoughts, and what have you tried, so that others can help you. this will also avoid you taking downvotes. – Nizar Feb 08 '18 at 16:21
2 Answers
The answer by @TsemoAristide has an obvious gap. Since my comment under the answer was not replied, I'm just posting it here so that people can see it. In the answer, when taking the determinant of the right-hand side, the inverse on $I+AB^{-1}$ was omitted. What should be correct is $$\det(A^{-1})\det(I+AB^{-1})=\det(B^{-1})\det(I+AB^{-1})^{-1},$$ which does not immediately imply $\det(A^{-1}) = \det(B^{-1})$.
I give an answer below, the complexity of which exceeds what I have expected. So any comments to simplify are welcome.
As the op has derived, we have (in slightly different way but basically the same): $$AB^{-1} + BA^{-1} + I = \mathbf{0}.$$ Setting $C = AB^{-1}$ one realizes $C^{-1} = BA^{-1}$, so the above equation becomes $$C + C^{-1} + I = \mathbf{0},\quad\text{or}\hspace{1.5ex}C^2 + C + I = \mathbf{0}.$$ Let $\lambda$ be any eigenvalue of $C$, then $\lambda$ satifies $$\lambda^2 + \lambda + 1 = 0,$$ so $\lambda$ is either of the two conjugate roots $\omega$, $\bar\omega$ of the quadratic, whose product $\omega\bar\omega=1$. Since $C$ is real matrix its eigenvalues appear in conjugate pairs (counting multiplicity), so it must be that $$\det C = \prod\lambda_i = (\omega\bar\omega)^\frac{n}{2} = 1,$$ i.e., $$\det\left(AB^{-1}\right) = 1 \implies \det(A) = \det(B).$$ (By the proof we also showed that $n$ must be even.)
Edit: also by the proof, the result holds if $$A^{-1} + B^{-1} = r(A+B)^{-1},\text{ for some }0<r<4.$$
- 172
$A^{-1}+B^{-1}=(A+B)^{-1}$ implies that $A^{-1}(I+AB^{-1})=((AB^{-1}+I)B)^{-1}=B^{-1}(AB^{-1}+I)^{-1}$ we deduce that $det(A^{-1})det(I+AB^{-1})=det(B^{-1})det(I+AB^{-1})$ and $det(A^{-1})=det(B^{-1})$ and this implies that $det(A)=det(B)$.
- 87,475
-
-
-
-
@ Tsemo Aristide No, $I+AB^{-1}$ is not necessarily invertible when $A$ and $B$ are invertible : consider for example the case $A=-B$. – Jean Marie Feb 08 '18 at 16:30
-
-
-
All right, then take $A=2C$ and $B=-\dfrac12 C$ for a certain invertible matrix $C$... – Jean Marie Feb 08 '18 at 16:33
-
$det(I+AB^{-1})=det(A+B)det(B^{-1})$ so if $det(B^{-1})$ and $det(A+B)$ are not zero, so is $det(I+AB^{-1})$. – Tsemo Aristide Feb 08 '18 at 16:38
-
1I was only able to get $\det(A^{-1})\det(I+AB^{-1})=\det(B^{-1})\det(I+AB^{-1})^{-1}$. Could you explain a little bit how to derive $\det(A^{-1})=\det(B^{-1})$ from this? – Mathematical Feb 08 '18 at 16:45
-
1The step from $A^{-1}(I+AB^{-1})=B^{-1}(AB^{-1}+I)^{-1}$ to $\det(A^{-1})\det(I+AB^{-1})=\det(B^{-1})\det(I+AB^{-1})$ overlooks the inverse taken on the second matrix factor. – hardmath Feb 08 '18 at 21:01