1

I would like to show the set $O_n(\Bbb R)$ of orthogonal $n\times n$ matrices is closed in $M_n(\Bbb R)$ by showing every convergent sequence in $O_n(\Bbb R)$ has a limit in $O_n(\Bbb R)$ and not associating matrices as linear operators like it is done here or concluding $O_n(\Bbb R)$ is the preimage of a closed set under a continuous function, but using only sequences in the most elementary way possible.

My attempt:

We can identify $M_n$ with $\Bbb R^{n+\ldots+n}=\Bbb R^{n\cdot n}$ viewing $A\in M_n(\Bbb R), A=(a_{ij}), i,j\in\{1,\ldots,n\}$ as $$A=(a_{11},\ldots,a_{1n},\ldots,a_{j1},\ldots,a_{jn},\ldots,a_{n1},\ldots,a_{nn})\tag 1$$ in order to apply the following proposition:

Let $a_k=(a_k^1,\ldots,a_k^n)\in\Bbb R^n.$ It holds: $$\lim_{k\to\infty}a_k=L=(\ell_1,\ldots,\ell_n)\iff\lim_{k\to\infty} a_k^i=\ell_i,\forall i\in\{1,\ldots,n\}.$$

Let $M\in O_n, M=(m_{ij})$. Then the condition $MM^\tau=I$ gives us $$\Sigma_{ij}:=\sum_{\ell=1}^nm_{i\ell}m_{j\ell}=\begin{cases}1,&\text{ if } i=j,\\0,&\text{ if } i\ne j\end{cases}, i,j\in\{1,\ldots,n\}.$$

Let's rewrite $$O_n=\{M\in M_n(\Bbb R)\mid \Sigma(M)_{ij}=\delta_{ij},i,j\in\{0,\ldots,n\}\}.$$ From $(1),$ we have $M_k\to M\iff ({m_k}_{ij})\to (m_{ij}),\forall i,j\in\Bbb N.$

Now, using the following theorem:

Let $(x_k)_k$ and $(y_k)_k$ be convergent sequences in $\Bbb R$ . Then:

  1. $\lim\limits_{k\to\infty}(x_k\pm y_k)=\lim\limits_{k\to\infty}x_k\pm\lim\limits_{k\to\infty}y_k.$
  2. $\lim\limits_{k\to\infty}x_k\cdot y_k=\lim\limits_{k\to\infty}x_k\cdot\lim\limits_{k\to\infty}y_k.$

From $(1)$ and the theorem above, for an arbitrary sequence $(M_k)_k$ of orthogonal matrices converging to $M\in M_n(\Bbb R)$, it follows $$\delta_{ij}=\lim_{k\to\infty}\Sigma_{ij}(M_k)=\Sigma_{ij}(M)\implies M\in O_n.$$

Therefore, since the sequence $(M_k)_k$ in $O_n$ was arbitrary, we conclude that the limit of any convergent sequence of orthogonal matrices is again an orthogonal matrix, therefore, $O_n$ is closed.


Can someone verify my proof? Is there anything wrong/any room for improvement?

PinkyWay
  • 4,565
  • 2
    I've only skimmed your proof, but it seems that you are basically using the fact that the individual conditions for being an orthogonal matrix are each continuous functions of the entries of the matrix, so if the matrices in the sequence satisfy the conditions and if the matrices converge element-wise, then the limit matrix also satisfies the conditions. This seems ok to me. – angryavian Oct 29 '21 at 17:09

1 Answers1

4

You proof seems good to me.

Using matrix formalism can make it simpler: Let $\{M_p\}_{p\in\mathbb N}$ be the sequence of orthogonal matrices (so $M^T_pM_p=I$) and let $$M=\lim_{p\rightarrow+\infty}M_p$$ Then $$\begin{split} \|M^TM-I\| &= \|(M-M_p)^T M +M_p^T(M-M_p) +\underbrace{M_p^TM_p-I}_{=0}\|\\ &\leq \|(M-M_p)^T M\| + \|M_p^T(M-M_p)\| \\ &\leq \underbrace{\|(M-M_p)^T\|}_{\rightarrow 0}\| M\| + \|M_p^T\|\underbrace{\|(M-M_p)\|}_{\rightarrow 0} \\ \end{split}$$ Thus $M^TM=I$ and $M$ is orthogonal.

In general, when you can express a property (e.g. orthogonality) via a continuous function $f$, that is having the property is equivalent to $f(M_p)=0$, then limits preserve that property: $$0 = \lim_{p\rightarrow+\infty}f(M_p) = f(\lim_{p\rightarrow+\infty}M_p)$$

In your case, $f(M_p)=M_p^T M_p - I$.

Stefan Lafon
  • 12,256
  • 11
  • 29