Bar-Visibility Graphs, Generalized an d Specialized Alice Dean Math. & Comp. Sci. Dept. Skidmore College SMC/UVM adean@skidmore.edu www.skidmore.edu/~adean Combinatorics Seminar, October 29, 2003 Talk Outline 1. Introduction: Definitions, motivation, history 2. BVGs: Characterization using stnumbering 3. Generalization: BVGs on surfaces 4. Specialization: Unit BVGs 5. Other variations and open problems Warning There is a definite prejudice in topic selection toward those I’ve been personally involved in! Page 2 of 31 Introduction: Visibility Drawings of Graphs Vertices are represented on a geometric surface by objects such as bars, arcs, rectangles, boxes Edges are represented by visibilities in specified directions (horizontal, vertical, radial) Page 3 of 31 Motivations and Applications Modeling tool for digital circuit design Edges are implied, rather than drawn, simplifying complex displays of complex networks Bars or boxes can have data labels written in them Some visibility models can display nonplanar graphs in the plane Page 4 of 31 Early Work on Visibility Graphs 1976, Garey, Johnson, and So: Use visibility graph models to design a method of testing printed circuits Reduces potential of O(n2) tests to maximum of 12. 1985, Schlag, Luccio, Maestrini, Lee, Wong: Study bar-visibility graphs as models of stick diagrams used in VLSI design 1985-86, Wismath, and independently, Tamassia and Tollis, also Rosenstiehl and Tarjan: Give simple, linear-time characterization bar-visibility graphs in the plane Page 5 of 31 BVGs in the Plane Vertex = horizontal bar in R2 Edge IFF unobstructed vertical visibility band (width > 0) Vertical lines (bands of width 0) do not produce edges Bars may be either open or closed segments Easy to see that all BVGs are planar Page 6 of 31 Characterizing BVGs in the Plane Theorem [TT, W, RT]. A graph G is a BVG IFF it has a planar embedding with all cutpoints on a common face. Necessity proof not hard: Every BVG induces a planar graph with all cutpoints on outer face. Sufficiency proof: Not difficult to paste layouts of 2-connected blocks together. 2-connected sufficiency proof: Requires idea of st-numbering Definition: If |V(G)| = n and s and t are distinct vertices, an st-numbering of G numbers the vertices 1, …, n so that 1. #(s) = 1 and #(t) = n Page 7 of 31 2. Every other vertex has a lowernumbered neighbor and highernumbered neighbor. Page 8 of 31 Constructing an st-numbering Facts: If G is 2-connected and s-t is an edge, then G has an st-numbering [LEC] In this case there is an O(n+m) algorithm to construct numbering [ET] Heuristic for plane 2-connected graphs: Can always do it provided s, t lie on common face t Embed G so that s, t are on outer face, all faces are convex (always possible if G is 2connected), and all s vertices at unique heights with s at bottom, t at top Number vertices from bottom to top Page 9 of 31 2-Connected Proof, Stages 1&2: st-number G and G* 1. Number vertices; direct edges: a) Choose any s and t on outer face; b) Give G an stnumbering c) Direct edges upward, and add two infinite rays, one into s, one out of t. 2. Direct dual edges; number faces a) G*-edge crosses G-edge in right turn t 7 6 5 4 3 2 s 1 5 2 4 1 3 Page 10 of 31 6 b) Directed edges induce stnumbering of G* Page 11 of 31 2-Connected Proof, Stage 3: Construct BVG Layout t 7 6 3 5 2 5 4 1 4 6 3 2 s 1 Vertex numbers correspond to rows Bar for vertex k is in row k Face numbers correspond to columns Bar for vertex k spans columns of faces that it borders 1 2 3 4 5 6 7 6 5 4 3 2 1 Page 12 of 31 Generalization: BVGs on Surfaces 2D surfaces have natural notions of parallel bars and orthogonal visibility bands Cylinder (86,89,91) Sphere (1989) Torus (1998) Möbius Band (2001) Projective Plane (2001) “Flat” Cylinder, Torus, Möbius Band: Page 13 of 31 BVGs for Planar Graphs on the Cylinder Bars parallel to cylinder axis Applications in VLSI: Linear systolic arrays, bit-slice architectures 2-connected case: Very similar to plane, except there are more “split” faces, besides outer face 1-connected case: Hiding blocks is more difficult now. Theorem [TT]. A planar graph G is a cylindrical BVG IFF it has an embedding whose block-cutface tree is a path IFF G’s block-cutpoint tree is a caterpillar. Planar BVG: Block-cutface tree is a star Page 14 of 31 Other “Planar” Surfaces SPHERE: Bars are latitudinal segments; visibility longitudinal Equivalent to BVGs on cylinders with bars perpendicular to axis Closed: Bars can be full circles, v. Open Theorem [TT]. G is an open BVG on the sphere IFF G has a plane embedding with all cutpoints on boundary of at most two faces IFF G has an embedding whose block cutface tree is a double star. [TT Closed Spherical BVGs ] Open Spherical BVGs Planar BVGs Cylindrical BVGs Page 15 of 31 Non-planar BVGs on Surfaces: The Möbius Band Characterized by D using generalized techniques of TT Each face corresponds to two columns in layout Blocks of “base graph” laid out in breadth-first order Theorem. A 2-cell embedding of G on the Möbius band is a BVG (with bars parallel to axis) IFF (1) the blockcutpoint tree of G is a caterpillar; (2) at most one block is non-planar; and (3) Page 16 of 31 the non-planar block is the “head” of the caterpillar. Page 17 of 31 BVGs on Projective Plane Bars are concentric arcs; visibility is radial As with sphere, can have open or closed BVG Characterized by Hutchinson Theorem [D]. G is a BVG on the projective plane IFF G is a BVG on Möbius band with bars orthogonal to the axis. Theorem [Hutchinson]. G is an open BVG on the projective plane IFF either (1) G has plane embedding with all but at most one cutpoint on a common face, or (2) G has projective plane embedding with all cutpoints on a common face. y1 y a b z b a z1 z y y1 z z1 b a x x1 x x1 Page 18 of 31 9 8 7 6 5 4 3 2 1 0 -1 -2 -3 -4 -5 -6 -7 -8 -9 Specialization: Unit BVGs My current main topic of study MOTIVATION: Circuits in which components have a single, fixed size. Definition: A unit bar-visibility graph (UBVG) is a BVG in which all bars have length 1. Theorem. Kn is a UBVG IFF n 3. Kp,q is a UBVG IFF p = 1 and q 3, or p = q = 2. Span(U) = # distinct left xcoordinates E. g., Span = 3: Span = 1 IFF G is a path Page 19 of 31 Span = 2 G is outerplanar Page 20 of 31 Cube, Circular Ladders, and Trees Theorem. The graph K2 Cn is a UBVG IFF n = 3, or n 6 and even Q3 = K2 C4 is not a UBVG UBVG is not a hereditary property. Theorem A tree is a UBVG IFF it has maximum degree 3 and is a subdivision of a caterpillar. Sufficiency was noted by Hutchinson and she conjectured necessity Necessity requires showing that (a) T is a UBVG if a subdivision of T is a UBVG; (b) a certain forbidden Page 21 of 31 subgraph for caterpillars is not a UBVG Page 22 of 31 UBVG Results: Outerplanar Graphs For a plane graph G, G*I denotes the dual graph with the outer face deleted. Theorem. If G is an outerplanar UBVG, then G*I has max degree 6, and if G is triangle-free, then G*I has max degree 4. Theorem. If G is an outerplanar, triangle-free UBVG, then G*I is a subdivision of a caterpillar with max degree 4. 3 1 2 3 4 5 1 2 3 4 5 4 1 2 2 3 5 6 7 4 1 1 2 3 4 5 6 7 Page 23 of 31 Theorem. If G is a simple, 2-connected, outerplanar (embedded) graph such that G*I is a path, then G is a UBVG. More on structure of the Finite Dual If all faces or no faces of G*I are triangles, structure of G*I is actually restricted beyond being a caterpillar of maximum degree 4: The neighbors of each face can be categorized as ‘up,’, ‘down,’ ‘left,’ and ‘right,’ Each face has at most one neighbor of each type. If G*I is drawn on the integer grid, then two nodes that lie on same vertical grid line must lie on a path of faces: Page 24 of 31 Possible structure for finite dual NOT possible structure for finite dual Page 25 of 31 Current Work: Triangulated Polygons Joint work with E. Gethner (CU Denver) and J. P. Hutchinson (Macalester C.) Big goal: Characterize all subdivided polygons that are UBVGs Smaller goal: Characterize all triangulated polygons = 2-connected, outerplanar, near-triangulations Theorem. If a triangulated polygon G is a UBVG then its finite dual G*I has maximum degree 3 and is a subdivision of a caterpillar. Page 26 of 31 Necessity: Forbidden Configurations Every TPUBVG can be represented combinatorially in terms of the ‘spine sequence’ of its finite dual: Theorem. If G is a TPUBVG, all subsequences below (and the corresponding forms with the roles of A and B interchanged) are ‘essentially’ forbidden in its spine sequence: A A(XB A)iA, i 0, where each XB = B or NB Page 27 of 31 There are several other forbidden subsequences, but we believe we are very close to a complete list. Page 28 of 31 Proof: AAA is forbidden u2 U2 a2 u1 U1 A1 a3 u3 U3 A2 A3 a1 a4 v Each triangle Ui is an ‘up-neighbor’ of Ai, meaning its bars can’t protrude to left or right of Ai; All bars b(ai) are neighbors of bar b(v), so must lie in corridor of width < 3; The bars b(u1), b(u2), b(u3) need width 3 or more; Contradiction. Page 29 of 31 Sufficiency: Layout Algorithm Step 1: Scan spine sequence for forbidden subsequences. Step 2: If OK, lay out UBVG by picturing graph as lying on triangulated plane, with some ‘nudges’ (fattened triangles) and ‘wing-splitters’ (subdivided triangles) as needed We are hopeful that we will soon: Complete the characterization of triangulated polygons Extend the results to quadrangulated polygons without too much more difficulty Page 30 of 31 Extend results to all subdivided polygons? To all outerplanar graphs? Page 31 of 31