Convergence of random polygon sequences (Q2681298)

From the Introduction: ``Let \(m \ge 3\) be a natural number and \(\Pi^{(0)}\) a closed \(m\)-sided polygon in \(\mathbf {C}\) with vertices labeled consecutively \(v_1^{(0)},\dots,v_m^{(0)}\). Denote by \(v_1^{(1)},\dots,v_m^{(1)}\) the midpoints of the edges \(v_1^{(0)} v_2^{(0)}\), \(v_2^{(0)} v_3^{(0)},\dots,v_m^{(0)} v_1^{(0)}\), respectively. Connecting \(v_1^{(1)},\dots,v_m^{(1)}\) in the same order as above, we derive a new polygon, and denote it by \(\Pi^{(1)}\). Apply the same procedure inductively to derive a sequence of polygons \(\Pi^{(k)}\), \(k = 0, 1, \dots\)''. ``Identifying the polygon \(\Pi^{(k)}\) with the column vector, whose components are successive vertices of \(\Pi^{(k)}\), we have \(\Pi^{(k)} = A^k \Pi^{(0)}\), in which \(A\) denotes the \(m\times m\) circulant matrix with leading row \((1/2, 1/2, 0, \dots, 0)\)''. The matrix A is called a transition matrix. ``It is clear that the midpoint iteration keeps the centroid of vertices of the polygons unchanged. Thus, one naturally expects that polygon sequences of midpoint iteration converge to their common centroid as \(k\) approaches infinity''. \textit{E. Hintikka} and \textit{X. Sun} [Involve 9, No. 5, 751--764 (2016; Zbl 1347.15029)] ``studied a generalization of the problem of midpoint iteration of polygons, in which the polygon-generating schemes have the formulation \(\Pi^{(k)} = \left(\prod_{j=1}^k A_j\right) \Pi^{(0)}\) where \(A_1,\dots,A_k\) are certain types of stochastic matrices. The transition matrices studied therein are not necessarily circulant, see [\textit{I. Kra} and \textit{S. R. Simanca}, Notices Am. Math. Soc. 59, No. 3, 368--377 (2012; Zbl 1246.15030)], and may vary from one step to another of the polygon-generating process. Under some mild restrictions on the transition matrices, the authors of [loc. cit.] proved that a polygon sequence thus generated converges to a point. Many computer simulations have shown that centroids of the successive polygons form a spiral orbit. In these general settings, the finite Fourier transform technique is no longer applicable. The problem is essentially modeled as a special type of inhomogeneous Markov chain''. ``The main study of the current paper departs from the above deterministic approaches and embraces a stochastic viewpoint. If the participating matrices \(A_1,\dots,A_k\) in the polygon-generating schemes are random stochastic matrices, then the result is a sequence of random polygons''. In the current article, the authors provide comprehensive answer to the following question: Under what conditions imposed on the prescribed probability distributions does a sequence of random polygons thus generated converge almost surely to a random limit point? If all the random variables on the edges obey Dirac delta distributions, the polygon-generating schemes become deterministic, as does the convergence of a polygon sequence to a fixed point in the convex hull of \(\Pi^{(0)}\). Note that the current paper does not require the initial polygon \(\Pi^{(0)}\) to be convex. The layout of the current article is as follows. In Section 2, the authors give background information concerning the setup of the problems and introduce the main mathematical tools needed. Section 3 gives several clearly identifiable conditions under which a random polygon sequence converges to a random point almost surely. In Section 4, the authors investigate the distribution of the limit points to which some suitably designed polygon sequences converge.