On the rank of a Latin tensor (Q1638003)

The author defines an \(n\times n\) matrix \(A\) to be permative if its rows are distinct permutations of \(n\) unequal elements; \(A\) is a generic permative matrix if the entries of the rows are \(n\) indeterminants \(x_{1},\dots,x_{n}\). Let \(r\) be the rank of an \(n\times n\) permative matrix. Clearly, \(r=n\) for \(n=1\) or \(2\). The author proves that \(r\geq3\) for each \(n\geq3\) and the bound \(r=3\) can be attained for each \(n\). Moreover, the minimal rank of a generic permative \(n\times n\) matrix is the least \(r\) such that \(r!\geq n\). These results answer questions raised in [\textit{X. Hu} et al., Spec. Matrices 4, 233--246 (2016; Zbl 1338.05030)]. A tensor \(T\) of the shape \(n\times n\times \cdots \times n\) (\(d\) factors) is called a Latin tensor if there are \(n\) distinct elements \(x_{1},\dots,x_{n}\) such that whenever \(d-1\) indices are fixed the set of entries of \(T\) obtained as we run over the \(n\) values of the remaining index equals \(\{x_{1},\dots,x_{n}\}\). The author shows that if \(n\geq3\) and the entries of \(T\) are real and nonnegative, then the real tensor rank of a Latin tensor \(T\) is at least \(3\).