Conjectured \textit{DXZ} decompositions of a unitary matrix (Q2168924)

The authors consider the following problem. Let \(U\) be a unitary \(mr\times mr \) matrix written as an \(r\times r\) matrix \([U_{ij}]\) of \(m\times m\) blocks. Is it possible to find unitary matrices \(D\), \(X\) and \(Z\) written in the same block form such that \[U=DXZ,\tag{*} \] where \(D\) and \(Z\) are block diagonal matrices with \( D=\mathrm{diag}(D_{11},\dots,D_{rr})\) and \(Z=\mathrm{diag}(Z_{11},\dots,Z_{rr})\) and \(X\) is the block matrix \([X_{ij}]\) such that the block row sums \(\sum_{j}X_{ij}=I_{m}\) for \(i=1,\dots,r\) (it can be shown that all block column sums of \(X\) are then also equal to the identity matrix \(I_{m}\)). If \(U\) has a decomposition of the form (*) then we can write \[UV=W, \tag{**}\] where \(V\) and \(W\) are \(n\times m\) matrices which can be written in block form as columns of \(m\times m\) unitary blocks with the first block in \(W\) equal to \(I_{m}\). It is known that a decomposition of the form (*) exists if \(m=1\), see [\textit{M. Idel} and \textit{M. M. Wolf}, Linear Algebra Appl. 471, 76--84 (2015; Zbl 1307.15014)] and when \(m=n/2\), see [\textit{H. Führ} and \textit{Z. Rzeszotnik}, Linear Algebra Appl. 484, 86--129 (2015; Zbl 1330.15033)]. The authors conjecture that the decompositions (*) and (**) always exist. They give proofs for some further special cases, and suggest approaches which they believe could help in other cases, but the general conjectures remains open. If the decomposition (*) holds for \(U\), the authors describe how numerical approximations to \(D\) and \(Z\) can be computed.