On realization of conformally-projectively flat statistical manifolds and the divergences (Q1265208)

A statistical manifold is a pseudo-Riemannian manifold \( (M,h) \) together with a torsion-free connection \( \nabla \) such that \( \nabla h \) is symmetric. The motivation stems from statistics [see \textit{S. Amari}, `Differential-geometrical methods in statistics' (Springer-Verlag, Berlin) (1985; Zbl 0559.62001)]. The author discusses this property for immersions \( f:M \to \mathbb R^{n+2} \) of codimension two in the realm of centroaffine differential geometry. In fact, \( f \) is always considered together with a transversal vector function \( \xi \) such that \( D\xi \) has no component in the \(\xi\)-direction (Peterson normalization). So the associated metric \(h\) (assumed to be regular) and the tangential connection \( \nabla \) depend on the choice of \( \xi \). It is an easy consequence of the integrability conditions that such pairs \( (f,\xi) \) define a statistical manifold \( (M,\nabla,h) \). Moreover, it is conformally-projectively flat. The main theorem asserts this and its converse (additionally assuming that \( M \) is simply connected). Finally dual statistical manifolds and their realization in the sense above are studied, and an application to a certain function on \( M \times M \), the geometric divergence, is given.