Consider the following theorem:

Let $\mathrm{\mathbf{A}}\in\mathbb{C}^{n\times n}$ be an upper triangular block matrix with $2\times2$ blocks ($p=2$), i. e.,

$\mathrm{\mathbf{A}}=\left(\begin{array}{cc} \mathrm{\mathbf{A}_{11}} & \mathrm{\mathbf{A}_{12}}\\ 0 & \mathrm{\mathbf{A}_{22}} \end{array}\right)$.

Then $\mathrm{\mathbf{A}}$ is invertible if and only if $\mathbf{A}_{11}$ and $\mathbf{A}_{22}$ are invertible matrices.

(a) Prove the theorem.

(b) Generalize the theorem to $\mathrm{\mathbf{A}}\in\mathbb{C}^{n\times n}$ upper triangular block matrix with $p\times p$ blocks, i. e.,

$\mathrm{\mathbf{A}}=\left(\begin{array}{cccc} \mathrm{\mathbf{A}_{11}} & \mathrm{\mathbf{A}_{12}} & \cdots & \mathrm{\mathbf{A}}_{1n}\\ & \mathrm{\mathbf{A}_{22}} & \cdots & \mathrm{\mathbf{A}}_{2n}\\ & & \ddots & \vdots\\ & & & \mathrm{\mathbf{A}}_{nn} \end{array}\right)$,

and prove. (Note that it's not necessary to calculate the inverse matrix)


(a) In this case $\mathrm{\mathbf{A}^{-1}}\in\mathbb{C}^{n\times n}$ is

$\mathrm{\mathbf{A}}^{-1}=\left(\begin{array}{cc} \mathrm{\mathbf{A}_{11}^{-1}} & \mathrm{-\mathrm{\mathbf{A}_{11}^{-1}}\mathbf{A}_{12}}\mathrm{\mathbf{A}_{22}^{-1}}\\ 0 & \mathrm{\mathbf{A}_{22}^{-1}} \end{array}\right)$. Therefore, $\mathrm{\mathbf{A}}$ is invertible if and only if $\mathbf{A}_{11}$ is invertible and $\mathbf{A}_{22}$ is invertible. $\square$

(b) I think I have to use induction on the number of blocks ($p$), but I don't know how to do

Can you help me? Thanks.


1 Answer 1


a/ -- You have only shown "if $\mathbf{A}_{11}$ and $\mathbf{A}_{22}$ are invertible then $\mathbf{A}$ is invertible". It still remains to show the other implication.

b/ A matrix $\mathbf{A}$ is invertible if and only if the equation $$ \mathbf{A}x = 0$$ has a unique (zero) solution.

$\mathbf{A}_{kk}$ is invertible for all $k = 1 ... n$

We are solving the equation $$\mathbf{A}x = 0.$$

As $\mathbf{A}_{nn}$ is invertible then $x[(n-1)p + 1:np] = 0$ is the unique solution of the equation $$\mathbf{A}_{nn} x[(n-1)p +1: np] = 0$$ i.e. the last $p$ elements of $x$ are zero. Thus, the rightmost column $[\mathbf{A}_{1n};...;\mathbf{A}_{nn}]$ has no effect (it is multiplied by 0). It means that $x[1:(n-1)p]$ is determined by the equation $$ \left(\begin{array}{ccc} \mathbf{A}_{11} & \ldots & \mathbf{A}_{1,n-1} \\ & \ddots & \vdots \\ \mathbf{0} & & \mathbf{A}_{n-1,n-1} \end{array}\right) x[1:(n-1)p] = \mathbf{0}. $$

Now repeat the same argument inductively for matrices $\mathbf{A}_{n-1,n-1}, ..., \mathbf{A}_{11}$.

$\mathbf{A}$ is invertible

Suppose that $\mathbf{A}_{11}$ is not invertible and there exists $z$ such that $\mathbf{A}_{11} z = 0$. Then a vector $y = [z;0;0;...;0]$ is a solution to $\mathbf{A} y = 0$ while $y \neq 0$. This is a contradiction to assumption that $\mathbf{A}$ is invertible and thus $\mathbf{A}_{11}$ is invertible too. That means you can express $$x[1:p] = - \mathbf{A}_{11}^{-1}[\mathbf{A}_{12},...,\mathbf{A}_{1n}]x[p+1:np],$$ and the rest of $x$ is solution to $$\left(\begin{array}{ccc} \mathbf{A}_{22} & \ldots & \mathbf{A}_{2,n} \\ & \ddots & \vdots \\ \mathbf{0} & & \mathbf{A}_{n,n} \end{array}\right) x[p+1:np] = \mathbf{0}.$$ You can now proceed recursively.


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.