Open problems for -skeletons in R 2 and R 3. Miroslaw Kowaluk University of Warsaw EuroGIGA meeting, Lugano 2011
= 0,8 For a given set P containing n points in R m and the parameter we define -skeleton as a graph (P, E), in which xy E iff any point from P\{x,y} doesn’t belong to an area R(x,y, ), where 2. For 0 < < 1, R(x,y, ) is the intersection of two spheres with the radius d(x,y)/2 , which boundaries contain the both points x i y. 1. For = 0, R(x,y, ) is the segment xy. yx
= 1 3. For 1 < , R(x,y, ) is a intersection of two spheres with radius d(x,y)/2 and centered in points (1- /2)x+( /2)y and ( /2)x+(1- /2)y resp. 4. For = , R(x,y, ) is an unbounded strip between two lines containing x and y resp. And perpendicular to the segment xy. = yx = 2 lune-based definition
4. For = , R(x,y, ) is the whole plane. yx = = sphere-based definition 3. For 1 < , R(x,y, ) is the union of two spheres, whose boundaries contain the both points x i y. = 1,25 yx
Properties of -skeletons. The -skeleton for a set of points P and = 1 is called Gabriel Graph (GG(P)) (Gabriel,Sokal 69), and for = 2 is called Relative Neighbourhood Graph (RNG(P)) (Toussaint 80). Theorem (Kirkpatrick,Radke 85). GG(P)MST(P) RNG(P) DT(P) x y z x y z w
Theorem. For 0 < 1 the –skeleton can have (n 2 ) edges. Theorem (Hurtado, Liotta, Meijer, 2003) For 0 < 1 the –skeleton can be computed in the optimal time O(n 2 ). Theorem (Jaromczyk, Kowaluk, Yao) For 1 2 the –skeleton in R 2 in L p (for 1 < p < ) can be computed from DT(P) in linear time.
Open problem. Is there algorithm computing –skeleton for 2 < that requires o(n 2 ) time ?
The –spectrum defines for each pair of points in P the maximum value of for which area R(x,y, ) is empty. Theorem (Hurtado, Liotta, Meijer, 2003) The –spectrum for all pairs of points in P can be found in time O(n 2 ).
Open problem. Is there an algorithm finding –spectrum only for edges which belong to Delaunay triangulation of the given set of points P that requires o(n 2 ) time ?
Let L(u,v) be a length of the shortest path between vertices u and v in a connected graph G in R 2, and D(u,v) be a distance between u i v. Spanning ratio S of the graph G is defined as follows S = max (u,v) G L(u,v)/D(u,v). Theorem (Keil,Gutwin 92). Spanning ratio of the DT(P), where |P| = n, is O(1). Theorem (Bose,Devroye,Evans,Kirkpatrick 02). Spanning ratio of the RNG(P), where |P| = n, is (n). Spanning ratio GG(P), where |P| = n, is (n 1/2 ). L(u,v) D(u,v) vu
Open problem. Find a better estimation of the spanning ratio for the – skeleton where 1 2.
Theorem (Agarwal, Matoušek 92). RNG(P) in R 3 in general position of points in L p (for 1 < p < ) has O(n 4/3 ) edges and can be found in expected time O(n 3/2+ ). Theorem (Chazelle,Edelsbrunner,Guibas,Hershberger,Seidel,Sharir 90). GG(P) in R 3 can have (n 2 ) edges.
Open problem. What is the graph complexity for the –skeleton in R 3 where 1 2 ?
Thank you for your attention