2

Given an nxn matrix A with mutually orthonormal columns, how does one show that A has mutually orthonormal rows WITHOUT assuming that A is orthogonal (A^T = A^-1)? I can show that A has mutually orthogonal rows by using the orthogonality between the row space and null space of a matrix, as well as the fact that the columns of A span R^n, but I need to show that the rows also have unit length. That is where I am getting stuck.

A similar question was asked here: Orthonormal columns and rows ; however, in trying to prove that A is orthogonal iff A has orthonormal columns, they don't show that AA^T = I given orthonormal columns, so their proof seems incomplete (if A is orthogonal, then A^T *A = I AND AA^T = I). To finish the proof, I believe they would need to answer my question.

Rodrigo de Azevedo
  • 1
Marcel Mazur
  • 31

2 Answers2

3

The dot product of the $i$'th and $j$'th columns of $A$ is $(A^T A)_{ij}$, so $A^T A = I$ is equivalent to the statement that the columns are orthonormal. Now this implies that $A$ has rank $n$, and therefore is invertible, and therefore $A^T = A^{-1}$, so $A A^T = I$ also, which is equivalent to the statement that the rows are orthonormal.

Robert Israel
  • 448,999
3

The canonical answer is Robert's. Here's another argument:

You know that $A^TA=I$. Multiplying by $A$ on the left and $A^T$ on the right, you get $(AA^T)^2=AA^T$. So $AA^T$ is an idempotent. In particular its eigenvalues can only be $0$ and $1$. Let $m$ be the multiplicity of $1$ as an eigenvalue of $AA^T$. Then $$ m=\operatorname{Tr}(AA^T)=\operatorname{Tr}(A^TA)=\operatorname{Tr}(I)=n. $$ This shows that all eigenvalues of $AA^T$ are $1$. Thus it is invertible; as the only invertible idempotent is $I$ we get that $AA^T=I$.

Martin Argerami
  • 205,756