Existence of DODDs for arbitrary matrices in sufficiently large dimensions

About 6 years old · traced to

Let X∈Rm×n\mathbf{X}\in\mathbb{R}^{m\times n}, and for an integer d≥m,nd\geq m,n let its 0-inflation be the d×dd\times d matrix obtained by placing X\mathbf{X} in the top-left corner and filling the remaining entries with zeros. A DODD is a decomposition involving diagonal matrices and an orthogonal matrix; write the corresponding rectangular block as Q‾\overline{\mathbf{Q}}. Equivalently, let C∈Rm×d\mathbf{C}\in\mathbb{R}^{m\times d} have orthonormal rows and let F∈Rd×n\mathbf{F}\in\mathbb{R}^{d\times n} have orthonormal columns.

DODD existence conjecture. The d×dd\times d 0-inflation of any matrix X∈Rm×n\mathbf{X}\in\mathbb{R}^{m\times n} admits a DODD for some sufficiently large d≥m,nd\geq m,n. Thus every such matrix has a decomposition

X=ΛQ‾M,\mathbf{X}=\boldsymbol\Lambda\overline{\mathbf{Q}}\mathbf{M},

where Λ∈Rm×m\boldsymbol\Lambda\in\mathbb{R}^{m\times m} and M∈Rn×n\mathbf{M}\in\mathbb{R}^{n\times n} are diagonal, and Q‾∈Rm×n\overline{\mathbf{Q}}\in\mathbb{R}^{m\times n} is the top-left block of an orthogonal matrix Q∈Rd×d\mathbf{Q}\in\mathbb{R}^{d\times d}. Equivalently,

X=ΛC⊤FM.\mathbf{X}=\boldsymbol\Lambda\mathbf{C}^{\top}\mathbf{F}\mathbf{M}.

The conjecture would establish existence of such orthogonal-diagonal decompositions for every rectangular matrix after sufficiently large zero inflation. The source supports the conjecture only with numerical experiments, so its resolution remains open.

References

Primary source

Karim Halaseh, Tommi Muller and Elina Robeva, “Orthogonal Decomposition of Tensor Trains”, arXiv:2010.04202 (2021).

Progress summary

Never refreshed

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Solutions 0

No solutions have been posted yet.