Improper Poisson line process as SIRSN in any dimension (Q317480)

Scale-invariant random spatial networks (SIRSNs) have been introduced by \textit{D. Aldous} [Electron. J. Probab. 19, Paper No. 15, 41 p. (2014; Zbl 1305.90104)] as models for road networks. They describe routes between pairs of points in Euclidean space \({\mathbb R}^d\) with a limitation of the length and a concentration on main roads. An important property is the invariance under rotations, translations and change of scale. In the only known example of a SIRSN, a binary hierarchy model in two dimensions, the invariance properties were imposed at the end of the construction by a randomization. It was therefore suggested that a more natural model could be based on the Poisson line process. The author confirms this conjecture by considering a stationary and isotropic Poisson line process, where the lines are marked with random speed limits following a power law. The route between two points is then the minimum-time path made of segments on the lines. Although, for \(d\geq 3\), the lines almost surely do not intersect, such a path exists since slower lines are dense. In fact, a major step in establishing the SIRSN properties is the proof that for two points \(x\) and \(y\) in \({\mathbb R}^d\) almost surely only one route (geodesic) \(L_{xy}\) exists. This fact was confirmed for \(d=2\) by \textit{W. S. Kendall} [Ann. Probab. 45, No. 1, 469--517 (2017; Zbl 1379.60012)] and it is shown here for \(d\geq 3\) (with a proof which does not work for \(d=2\)). As a second step, it is shown that \(L_{xy}\) has finite expectation. The final and crucial property of SIRSN is then shown: In each compact set \(C\subset {\mathbb R}^d\) the total length of all the geodesics (truncated by deleting balls around their endpoints) in \(C\) is finite. Thus the network contains main roads in each region which are largely used by the routes.