On Bend-Minimized Orthogonal Drawings of Planar 3-Graphs

On Bend-Minimized Orthogonal Drawings of Planar 3-Graphs Yi-Jun Chang∗1 and Hsu-Chun Yen†2 1 2 Department of EECS, University of Michigan, Ann Arbor, MI, USA Department of Electrical Engineering, National Taiwan University, Taipei, Taiwan Abstract An orthogonal drawing of a graph is a planar drawing where each edge is drawn as a sequence of horizontal and vertical line segments. Finding a bend-minimized orthogonal drawing of a planar graph of maximum degree 4 is NP-hard. The problem becomes tractable for planar graphs of maximum degree 3, and the fastest known algorithm takes O(n5 log n) time. Whether a faster algorithm exists has been a long-standing open problem in graph drawing. In this paper we present an algorithm that takes only Õ(n17/7 ) time, which is a significant improvement over the previous state of the art. 1998 ACM Subject Classification F.2.2 Nonnumerical Algorithms and Problems Keywords and phrases Bend minimization, graph drawing, orthogonal drawing, planar graph Digital Object Identifier 10.4230/LIPIcs.SoCG.2017.29 1 Introduction An orthogonal drawing of a graph is a planar drawing where each edge is composed of a sequence of horizontal and vertical line segments with no crossings. Orthogonal drawings appear in many applications such as the automation of VLSI circuit layout and the drawing of diagrams in information systems. Variants of orthogonal drawings have been introduced in the literature to cope with different constraints or to improve the readability and aesthetic feel: smoothing the drawing [2, 1], requiring orthogonal convexity [5], accommodating vertices of more than 4 incident edges [16, 8], and restricting directions of vertices [11, 13]. Refer to [12] for a survey on orthogonal drawings. Bend-minimization is one of the most classical optimization problems on orthogonal drawings. Given a planar (or plane) graph, the problem asks for an orthogonal drawing with the total number of bends minimized. However, the problem is NP-hard for planar 4-graphs [15].1 To obtain polynomial time algorithms, one has to relax the problem one way or another. For example, it is known that a polynomial time algorithm exists when the first bend on an edge does not incur any cost [3]. Much research effort has been made on bend-minimization for subclasses of planar 4graphs [9, 7, 14, 18, 17]. The two most natural subclasses are planar 3-graphs (reducing ∗ Yi-Jun Chang was supported in part by NSF grant CCF-1514383. Hsu-Chun Yen was supported in part by Ministry of Science and Technology, Taiwan, under grant MOST 103-2221-E-002-154-MY3. 1 We write k-graphs to denote graphs of maximum degree k. Note that the degree of each vertex cannot exceed 4 in an orthogonal drawing, and hence planar 4-graphs are the most general graph class that can be drawn orthogonally. † © Yi-Jun Chang and Hsu-Chun Yen; licensed under Creative Commons License CC-BY 33rd International Symposium on Computational Geometry (SoCG 2017). Editors: Boris Aronov and Matthew J. Katz; Article No. 29; pp. 29:1–29:15 Leibniz International Proceedings in Informatics Schloss Dagstuhl – Leibniz-Zentrum für Informatik, Dagstuhl Publishing, Germany 29:2 On Bend-Minimized Orthogonal Drawings of Planar 3-Graphs the maximum degree by 1) and plane 4-graphs (fixing the planar embedding). For plane 4-graphs, a seminal work by Tamassia [20] demonstrates a reduction from bend-minimization to a computation of a min-cost flow. Following this approach, the runtime has been reduced √ to O(n7/4 log n) [14], and subsequently to O(n1.5 ) [7].2 For planar 3-graphs, the fastest known algorithm for bend-minimization is the O(n5 log n) time algorithm designed by Di Battista, Liotta, and Vargiu, which dates back to 1998 [9]. As stated as an open problem in [4], further improving the time complexity is identified as an important issue in the field of graph drawing. Problem 15: Let G be a planar graph whose vertices have degree at most three. Is there an algorithm to compute a planar bend-minimum orthogonal drawing of G in o(n5 log n) time? In this paper, we answer the problem affirmatively by demonstrating an Õ(n17/7 ) time3 algorithm. Precisely, our algorithm takes O(n · T (n)) time, where T (n) is the time complexity of constructing a bend-minimized orthogonal drawing of a plane 3-graph subject to the constraint that some designated edges have no bend. We will later see that bend-minimization of a plane 3-graph can be reduced to min-cost flow of constant capacity, and a recent breakthrough on unit-capacity min-cost flow in sparse graphs [6] implies T (n) = Õ(n10/7 ). The main challenge of designing a bend-minimization algorithm for planar 3-graphs is to handle the transition from the variable embedding setting to the fixed embedding setting. The naïve approach of testing all possible planar embeddings is very inefficient, as there can be an exponential number of different planar embeddings. A natural way to approach this problem is to devise a dynamic programming procedure on an SPQR-tree, which is a tree structure capable of storing all possible embeddings of a planar graph using linear space. The approach is briefly sketched as follows. By “contracting” all subgraphs of the planar graph G that can be flipped, a graph G0 that has a fixed combinatorial embedding is obtained. An optimal drawing of G0 can be computed quickly using a bend-minimization algorithm for plane 3-graphs. The optimal drawings of the contracted subgraphs are computed recursively. Merging these drawings yields a drawing of G. Using the terminology of SPQR-trees, if G is the pertinent graph of a node µ in the SPQR-tree, then the graph G0 is (a subdivision of) the skeleton of µ, and the contracted subgraphs are the pertinent graphs of some descendants of µ. See Fig. 1 for a conceptual example: (1) G0 = H2 is resulting from contracting the subgraph H1 into a vertex v in G. (2) Merging the drawings of H1 and H2 yields an orthogonal drawing of G (treating each white dot in the figure as a bend). The above strategy does not immediately give us a bend-minimization algorithm. Observe that the outer boundary of G needs to have 4 convex corners. Thus we need an additional constraint which requires that the drawing of G0 and the drawings computed by recursive calls jointly supply 4 convex corners in the outer boundary. To ensure that the final drawing of G is optimal, one has to compute multiple drawings of each contracted subgraph H subject to different constraints on the number of convex corners in the outer boundary of G that H can supply, and to examine all possible combinations of these constraints. In [9] the notion of spirality, which measures how an orthogonal drawing is “rolled up”, is developed to serve as the aforementioned constraints. In this paper, we utilize some tools developed by Rahman et al. in [19]. They characterized the condition for the existence of a no-bend orthogonal drawing based on the number of 2 Throughout the paper, we (...truncated)