Consider some invertible $N\times N$ matrix $A$, let $B=A^{-1}$ and let $A_{\left[11,nn\right]}$ be the $\left(N-2\right)\times\left(N-2\right)$ matrix resulting from A after removing row and column 1 and row and column $n \neq 1$. I have seen that in some special case the following equality holds for $n \neq 1$: $$ b_{11}b_{nn}-b_{n1}b_{1n}=\frac{\det A_{\left[11,nn\right]}}{\det A} $$ Is this a general result for all invertible matrices?
-
I checked the $N=3$ case (the simplest one) and it's true in general. You can probably show the full result by using the minor expansion of the det. – Quillo Sep 08 '21 at 14:38
1 Answers
Yes, this result is correct in general. Here is a proof.
Without loss of generality, we can take $n=2$. Partition $A$ and $B$ as $$ A = \pmatrix{A_{11} & A_{12}\\ A_{21} & A_{22}}, \quad B = \pmatrix{B_{11} & B_{12}\\ B_{21} & B_{22}}, $$ where $A_{11}$ and $B_{11}$ have size $2 \times 2$. Because $BA = I$, we can deduce that $$ \pmatrix{B_{11} & B_{12}\\ 0 & I_{N-2}} \pmatrix{A_{11} & A_{12}\\ A_{21} & A_{22}} = \pmatrix{I_2 & 0\\A_{21} & A_{22}}. $$ Taking the determinant of both sides yields $$ \det\pmatrix{B_{11} & B_{12}\\ 0 & I_{N-2}}\det \pmatrix{A_{11} & A_{12}\\ A_{21} & A_{22}} =\det \pmatrix{I_2 & 0\\A_{21} & A_{22}} \implies\\ \det(B_{11})\det(A) = \det(A_{22}) \implies\\ \det(B_{11}) = \frac{\det(A_{22})}{\det(A)}, $$ which is what we wanted to show.
- 225,327
-
Really cool, +1 ! I agree that you can set $n=2$... but when you write $I_{n-2}$, do you mean $I_{N-2}$? In any case, I checked the case $n=2,N=3$ and it works. – Quillo Sep 08 '21 at 14:44
-
1
-
@BenGrossmann Thanks for this. Do you know what is an extension of this result to an expression for $b_{ik} b_{kj} - b_{ij} b_{kk}$ in terms of elements of $A$? I have asked this question here: https://math.stackexchange.com/q/4251300/165163 – Andres Sep 16 '21 at 11:51
-
-
@BenGrossmann Do you think that the result for case $N=3$ does not have some natural extension to $N>3$? – Andres Sep 16 '21 at 12:09
-
@Andres The expression for $N = 3$ is interesting. I'll think about it. – Ben Grossmann Sep 16 '21 at 12:22
-
@Andres I don't think that there's anything simple, but I suspect that there is a result that can be seen with the help of alternating tensor products. – Ben Grossmann Sep 16 '21 at 12:36
-