The linear interpolation problem for totally positive matrices (Q703628)

A matrix \(A\) is called totally positive or TP if every minor of \(A\) is positive. It has long been known that there are substantial restrictions on the possible vector pairs \(x\) and \(Ax\), as \(A\) ranges over all TP matrices. In this paper the authors address the problem of characterizing vector pairs \(x\), \(y\) such that there is a TP matrix \(A\) with \(y= Ax\). If \(x\) has no zero entries, the variation \(v(x)\) in signs of the entries of \(x\) is the number of times the sign of \(x_{i+1}\) differs from that of \(x_i\), for all possible \(i\). When some entries of \(x\) are \(0\), different notions of sign variation are possible. Let \(x_m\) arise out of \(x\) by replacing its \(0\)-entries by non-zero ones so that \(v(x_m\)) is a minimum, and let \(v_m(x)= v(x_m)\); define \(v_M(x)\) similarly with maximum replacing minimum. Then the main result of the paper is: There is a TP matrix \(A\) such that \(y= Ax\) if and only if \(v_M(y)\leq v_m(x)\). Moreover, in the case of equality the sign of the first entry of \(x_m\) is the same as that of the first entry of \(y_M\). The necessity of the condition was already known.