1
$\begingroup$

The question is:

Let $U$ be an $n \times n$ orthogonal matrix. Show that the rows of U form an orthonormal basis of $\mathbb R ^n $.

So far I have stated: Since $U$ is orthogonal its column vectors are linearly independent and by the Invertible Matrix theorem $U$ is invertible and $U^T$ is invertible.

I know that I need to somehow show that $U^T U=I$, but I don't know how to get there from where I left off.

$\endgroup$
  • $\begingroup$ What is the definition of an orthogonal matrix? $\endgroup$ – Ameet Sharma Apr 25 '14 at 6:45
  • $\begingroup$ A set is orthogonal if each pair of distinct vectors within it are orthogonal to each other. An orthogonal matrix is a matrix whose column vectors form an orthogonal set. $\endgroup$ – Victoria Potvin Apr 25 '14 at 6:48
  • $\begingroup$ Hmm... I'd double check the definition. this is the definition of orthogonal matrix that I learned: mathworld.wolfram.com/OrthogonalMatrix.html. So you'd start knowing that $U^{T}U=I$. then use that to prove the rows form an orthonormal basis. $\endgroup$ – Ameet Sharma Apr 25 '14 at 6:56
  • $\begingroup$ The thing is, these 3 statements can be shown to be equivalent definitions of orthogonal matrix: 1) $U^{T}U=I$. 2) The rows of matrix U form an orthonormal set(they are orthogonal, plus have magnitude one). 3) The columns of matrix U form an orthonormal set. So the question is... is the problem asking you to use 1) to prove 2)... or to use 3) to prove 2)... $\endgroup$ – Ameet Sharma Apr 25 '14 at 7:10
1
$\begingroup$

Hint: This is long (and can be very much shortened), but I think it will be a good learning curve for you to make it long. It became this long, because we start with the way you have understood orthogonal matrices (as seen in comments). I am reproducing it here "A set is orthogonal if each pair of distinct vectors within it are orthogonal to each other. An orthogonal matrix is a matrix whose column vectors form an orthogonal set". This is infact one of the many equivalent ways to define a orthogonal matrix. We will start with this definition without any other assumptions.

We will individually denote both columns and rows of $U$. Let columns of $U$ be given by $N\times 1$ vectors, $c_1,\dots,c_N$. Let its rows be given by $N\times 1$ vectors $r_1,\dots,r_N$. So you have $$U=[c_1|\dots|c_N]$$ and $$U=\begin{bmatrix}r_1^T \\ \vdots \\ r_N^T \end{bmatrix}$$

(notice how I used the transpose). Now note that $$A=U^TU$$ can be viewed upon as a matrix whose $(i,j)$th entry is given as $$A_{ij}=c_i^Tc_j$$ Similarly $$B=UU^T$$ is a matrix whose $(i,j)$th entry is given as $$B_{ij}=r_i^Tr_j$$

Now notice the following things

  • prove $U^TU=A=I$ (why?)
  • use the fact that $rank{(PQ)}\leq \min\{rank(P),rank(Q)\}$ to argue that $U$ and $U^T$ are invertible. $P$ and $Q$ are any arbitrary $N\times N$ matrices. Substitute $P$ and $Q$ with suitable matrices.
  • Use above two facts to prove that $U^{-1}=U^T$. Hint: Perhaps a right multiplication on the LHS of first fact with a suitable matrix.
  • Use above fact to prove $UU^T=I$. Hint: For this you also need that if $P$ are $Q$ are inverses of each other $PQ=I$ and $QP=I$
  • Now using the above fact and definition of $B$, look at $r_i^Tr_j$
$\endgroup$

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.