Kindly help me in the following:
- I have two Matrices, $A$ of size $(n\times m)$; and $B$ of size $(n\times n)$, where $n>m$.
- $A$ is unknown, but $B$ is known.
- $(A^TA)$ is invertible
- $B$ is symmetric and invertible.
- $B$ is not an Identity Matrix.
The two Matrices are related to each other as follows:
$A\times (A^TA)^{-1}\times A^T = B$
Is there a closed-form solution to finding $A$ from $B$ ?
Thank you in advance!
Not hard to see
1) $BA=A$
2) $\mathrm{rank}A=m$
So all $m$ columns of matrix $A$ are linearly independent eigenvectors which eigenvalues are equal to 1. The set of all eigenvectors with eigenvalues=1 is a space of dimension $k$, $m\le k<n$ (since $B\ne I$ then $k\ne n$) Let $C_1,...,C_k$ be a basis of this space, and $C=(C_1,...,C_k)$ be the $n\times k$ matrix. So, the set of answers (appropriate matrix $A$) is $$A=CD,$$ where $D$ is any $k\times m$ matrix of the rank $m$.