Rationality does not specialize among terminal fourfolds (Q1686482)

Recent years saw a considerable advance in the understanding of rationality questions for algebraic varieties, originated from the seminal work of \textit{C. Voisin} [Invent. Math. 201, No. 1, 207--237 (2015; Zbl 1327.14223)]. \par One long standing open problem is whether rationality is a deformation invariant. Now we have a fairly good understanding of the problem. (Stable) rationality specializes among smooth projective varieties [\textit{M. Kontsevich} and \textit{Yu. Tschinkel}, ``Specialization of birational types'', Preprint, \url{arXiv:1708.05699}; \textit{J. Nicaise} and \textit{E. Shinder}, ``The motivic nearby fiber and degeneration of stable rationality'', Preprint, \url{arXiv:1708.02790}]. Given a smooth projective family, there could be a countable union of proper subvarieties (which is dense in the euclidean topology) parameterizing rational ones, while a very general fiber outside this union is not stably rational [\textit{B. Hassett} et al., Acta Math. 220, No. 2, 341--365 (2018; Zbl 1420.14115)]. \par Totaro gave examples of a family of projective varieties with terminal singularities, such that general fibers are rational but the central fiber is not stably rational. His idea is to take the projective cone of a non-stably rational hypersurface of Fano index 2. It is straight forward to check that such a cone has terminal singularities and a deformation of such a cone (which is again a hypersurface) contains a singular point of multiplicity equal to the degree of the hypersurface minus one, which is well-known to be rational (consider the projection from the singular point). To get such an example, one has to assume the dimension of the hypersurface is at least $4$ [\textit{B. Totaro}, J. Am. Math. Soc. 29, No. 3, 883--891 (2016; Zbl 1376.14017)]. Thus Totaro's example appears in every dimension greater or equal to $5$. \par The paper under review give an example of a family of terminal $4$-folds such that rationality does not specialize in this family. The example is similar. But instead of using hypersurfaces, the author uses a family of weighted hypersurfaces, which allows him to bring the dimension down to $3$. \par One should mention that rationality does specialize in Kawamata log terminal $3$-folds (see the discussion after Question 1 in the current paper).