Real root conjecture fails for five- and higher-dimensional spheres (Q2571325)

The \(f\)-polynomial \(f_X\) of a simplicial complex \(X\) is defined by \(f_X(t) = \sum_{\sigma \in X} \, t^{\#\sigma}\). If \(X\) is a generalized homology sphere of dimension \(n-1\) (GHS\(^{n-1}\)), or even just Eulerian, then the \(h\)-polynomial \(h_X\) of \(X\) can be defined by \[ (1+t)^nh_X\left(\frac1{1+t}\right) = t^nf_X\left(\frac1t\right). \] The Dehn-Sommerville equations then say that \(h_X\) is reciprocal, meaning that \(h_X(t) = t^nh_X(1/t)\). A reciprocal polynomial \(h\) can be written in the form \[ h(t) = (1+t)^n\gamma\left(\frac t{(1+t)^2}\right), \] with \(\gamma\) a unique polynomial of degree at most \(n/2\). The complex \(X\) is said to be flag if every clique in its edge-graph determines a face. The author conjectures that, if \(X\) is a flag GHS, then the coefficients of \(\gamma_X\) are non-negative (integers). This is stronger than the \textit{R. Charney} and \textit{M. Davis} conjecture of [Pac. J.~Math. 171, 117--137 (1995; Zbl 0865.53036)], which proposes that, if \(X\) is a flag GHS\(^{2n-1}\), then \((-1)^nh_X(-1) \geq 0\), but weaker than the real-root conjecture, which suggested that the zeros of the \(h\)-polynomial \(h_X\) of a flag GHS \(X\) should be real. The author shows here that the real-root conjecture does hold for a GHS of dimension at most \(4\), but fails in higher dimensions. This leaves his own conjecture as the next candidate to be tested.