Decomposition of a matrix with orthogonal columns

221 Views Asked by At

Note that this question is also posted in mathoverflow. I'm new to the community so I'm also posting this here if it is more appropriate.


I am given a square matrix $A = [A_1, A_2, ..., A_n]$, where columns $A_i$ are orthogonal to each other. In another word, $A = OD$, where $O$ is an orthogonal matrix and $D$ is a diagonal matrix. The target is to decompose $A$ into a product of two matrices of the same type. That is, $A = X_1X_2$, where $X_1$ and $X_2$ are both named 'xxx'.

('The same type' here just means that we can classify them into one specific category. I know this is pretty vague, but I didn't come up with a better explanation. E.g, we can decompose $A = O \lambda$, where $\lambda$ is a scalar, as a product of two Jacobian matrices of conformal maps. But this is still a special case of $A$.)

For example, if we consider a special case of $A$ as an orthogonal matrix (i.e., $D = I$), one can simply decompose $A$ as a product of two orthogonal matrices. However, this is a very limited case. I am wondering whether there is a general decomposition of $A$?

Any suggestion would be greatly appreciated.


Thanks to @BenGrossmann for making the question clearer: I'm looking for some "type of matrix" such that if $P$, $Q$ are matrices of this type, then $PQ$ will necessarily have orthogonal columns.