Graph Algorithms and Applications

(Dagstuhl–Seminar 9620)

Organizers:

Takao Nishizeki (Tohoku University Sendai, Japan)

Roberto Tamassia (Brown University, USA)
Dorothea Wagner (Universität Konstanz, Germany)

May 13 - 17, 1996

In many fields of applications graphs or networks play an important role for understanding
a concrete situation and modelling problems. The development of algorithms for commu-
nication problems, traffic optimization, scheduling or VLSI-design is for example based
on graphs. Algorithmic graph theory is a classical area of research by now and has been
rapidly expanding during the last three decades. Especially, the interplay between theory
and application gave research in this area again and again new impetus.

While the complexity of today’s “real world” problems is increasing the design of “so-
phisticated” algorithms is an ambitious task. Numerous newly invented problems coming
from new fields of applications have attracted the attention of researchers in algorithmic
graph theory. But, in recent years also classical problems were studied again. For several
fundamental graph problems efficient and simple algorithms were designed in view of the
computer implementation of such algorithms.

This seminar was intended to bring together researchers from different areas in algorithmic
graph theory. Particular emphasis was placed on applications, experimental research and
aspects of the implementation of graph algorithms. The participants had the opportunity
to exchange ideas and discuss new trends in algorithmic graph theory. Main topics of
interest were efficient graph algorithms, graph drawing, algorithm animation with graphs,
implementation of graph algorithms and applications in VLSI-design, traffic optimization,
and CAD.

We had 36 participants from different European countries, Australia, USA, Canada, Japan
and Taiwan. During the workshop 28 lectures have been presented and two software
demonstrations. There was also an open-problem-session on Tuesday evening and a lively
discussion on problems from different fields of application. Schloß Dagstuhl and its staff
provided a very convenient and stimulating environment. All participants appreciated the
cordial atmosphere. The organizers wish to thank all those who helped make the workshop
a fruitful research experience.

3
 Participants

Takao Asano, Chuo University

Tetsuo Asano, Osaka Electro-Communication University
Franz J. Brandenburg, Universität Passau
Ulrik Brandes, Universität Konstanz
Yefim Dinitz, Technion — Haifa
Peter Eades, University of Newcastle
Wen-Lian Hsu, Academica Sinica Taipei
Toshihide Ibaraki, Kyoto University
Michael Jünger, Universität zu Köln
Goos Kant, Utrecht University
Naoki Katoh, Kobe University of Commerce
Michael Kaufmann, Universität Tübingen
Philip Klein, Brown University
Bernhard Korte, Universität Bonn
Han La Poutré, Leiden University
Annegret Liebers, Universität Konstanz
Giuseppe Liotta, Brown University
Rolf Möhring, Technische Universität Berlin
David W. Matula, Southern Methodist University
Kurt Mehlhorn, Max-Planck-Institut für Informatik
Bojan Mohar, University of Ljubljana
Petra Mutzel, Max-Planck-Institut für Informatik
Stefan Näher, Universität Halle
Hiroshi Nagamochi, Kyoto University
Takao Nishizeki, Tohoku University
Stephen North, AT&T Bell Labs
Andras Recski, Technical University of Budapest
Maciej M. Syslo, University of Wroclaw
Roberto Tamassia, Brown University
Ioannis Tollis, University of Texas at Dallas
Dorothea Wagner, Universität Konstanz
Frank Wagner, Freie Universität Berlin
Toshimasa Watanabe, Hiroshima University
Karsten Weihe, Universität Konstanz
Sue Whitesides, McGill University
Christos Zaroliagis, Max-Planck-Institut für Informatik

4
 Program

Monday, May 13, 1996

Morning Session Chair: Dorothea Wagner

9:00 Welcome
9:15 – 10:00 Takao Nishizeki: Decompositions to degree-constrained subgraphs
are simply reducible to edge-colorings
10:30 – 11:00 Hiroshi Nagamochi: Faster edge-splitting algorithms in undirected
graphs
11:15 – 11:45 Takao Asano: Approximation algroithms for the maximum satisfi-
ability problem

Afternoon Session Chair: Peter Eades

14:45 – 15.30 Wen-Lian Hsu: Graph recognition algorithms
16:15 – 16:45 Kurt Mehlhorn: A simple linear time algorithm to find Kuratowski
subgraphs of non-planar graphs
16:45 – 17:30 Bojan Mohar: Embedding graphs in an arbitrary surface in linear
time
17:30 – 18:00 Sue Whitesides: Geometric representations of graphs

Tuesday, May 14, 1996

Morning Session Chair: Michael Jünger

9:00 – 9:30 Karsten Weihe: Reconstructing a surface from its polygonal pieces
9:30 – 10:00 Rolf Möhring: Mesh generation using flow methods
10:30 – 10:50 Tetsuo Asano: Space efficient algorithms for image segmentation
10:50 – 11:20 Goos Kant: Vehicle routing problems with time windows
11:30 – 12:00 Yefim Dinitz: Modelling all bottlenecks (edge cuts) of a network,
with incremental maintenance

Afternoon Session Chair: Takao Nishizeki

14:45 – 15.30 Philip Klein: Approximation algorithms for semidefinite programs
arising from MAX CUT and COLORING
16:15 – 16:45 Naoki Katoh: Computing subgraphs of minimum weight triangu-
lations based on LMT-skeletons

17:00 Application Problems Chair: Ioannis Tollis

19:30 Open Problems Chair: Roberto Tamassia

5
Wednesday, May 15, 1996

Morning Session Chair: Stephen North

9:00 – 9.30 Franz J. Brandenburg: On drawing planar angle graphs
9:30 – 10:00 Peter Eades: Drawing clustered graphs
10:30 – 11:00 Ioannis Tollis: Advances in orthogonal graph drawing
11:00 – 11:45 Michael Kaufmann: On bend-minimum orthogonal graph
drawings

Afternoon Excursion to Trier or hike in the rain

Thursday, May 16, 1996

Morning Session Chair: Toshimasa Watanabe

9:00 – 9.30 Stefan Näher: Implementation of graph algorithms with LEDA
9:30 – 10:00 Giuseppe Liotta: The proximity drawability problem
10:30 – 11:00 Toshihide Ibaraki: Two arc disjoint paths in Eulerian digraphs
11:00 – 11:45 David W. Matula: Two results on search and edge connectivity

Afternoon Session Chair: Sue Whitesides

15:00 – 15.30 Michael Jünger: Practical performance of MIN CUT algorithms
16:15 – 17:00 Petra Mutzel: Optimization on hierarchical graphs
17:00 – 17:30 Maciej Syslo: The bandwith problem of clique caterpillars and tol-
erance graphs
17:30 – 17:45 Dorothea Wagner: PlaNet: A demonstration package for algo-
rithms on PLAnar NETworks
17:45 – 18:00 David W. Matula: Painting graph algorithms

Evening Software Demonstrations

Friday, May 17, 1996

Morning Session Chair: Frank Wagner

9:00 – 9.30 Bernhard Korte: L1 -Steiner trees
9:30 – 10:10 Christos Zaroliagis: All-pairs MIN CUT in sparse networks
10:45 – 11:30 Andras Recski: Some linear time solvable subcases of the
multilayer routing problem

6
 Decompositions to degree-constrained subgraphs are simply
reducible to edge-colorings

Takao Nishizeki
Graduate School of Information Sciences
Tohoku University

The degree-constrained subgraphs decomposition problem, such as an f -coloring, f -factorization

and [g, f ]-factorization, is to decompose a given graph G = (V, E) to edge-disjoint sub-
graphs degree-constrained by integer-valued functions f and g on V . In this talk we show
that the problem can be simply reduced to the edge-coloring problem in polynomial-time.
That is, for any positive integer k, we give a polynomial-time transformation of G to a new
graph such that G can be decomposed to at most k degree-constrained subgraphs if and
only if the new graph can be edge-colored with k colors.

Faster edge-splitting algorithms in undirected graphs

Hiroshi Nagamochi
Graduate School of Engineering
Kyoto University

This paper presents a deterministic O(n(m + n log n) log n) = Õ(nm) time algorithm for
splitting off all edges incident to a vertex s of even degree in a multigraph G, where n and
m are the numbers of vertices and links (= vertex pairs between which G has an edge) in
G, respectively. Based on this, many graph algorithms using edge-splitting can run faster.
For example, the edge-connectivity augmentation problem in an undirected multigraph can
be solved in Õ(nm) time, which is an improvement over the previously known randomized
Õ(n3 ) bound and deterministic Õ(n2 m) bound.

Approximation algorithms for the maximum satisfiability

problem

Takao Asano
Department of Information and System Engineering
Chuo University

The maximum satisfiability problem (MAX SAT) is : given a set of clauses with weights,
find a truth assignment that maximizes the sum of the weights of the satisfied clauses.

7
In this talk, we present approximation algorithms for MAX SAT, including a 0.76544-
approximation algorithm. The previous best approximation algorithm for MAX SAT was
proposed by Goemans-Williamson and has a performance guarantee of 0.7584. Our algo-
rithms are based on semidefinite programming and the 0.75-approximation algorithms of
Yannakakis and Goemans-Williamson.

Graph recognition algorithms

Wen-Lian Hsu
Institute of Information Science
Academica Sinica Taipei

We shall give a brief survey of our recent recognition algorithms for planar graphs, interval
graphs and the consecutive ones property. We shall also talk about algorithms that can
tolerate a small percentage of errors in the input data.

A simple linear time algorithm to find Kuratowski subgraphs of

non-planar graphs

Kurt Mehlhorn
Max-Planck-Institut für Informatik

We extend the Lempel-Even-Cederbaum planarity test such that it yields Kuratowski

subgraphs of non-planar graphs. The extension runs in linear time.

This is joint work with Christoph Hundack and Stefan Näher.

Embedding graphs in an arbitrary surface in linear time

Bojan Mohar
Department of Mathematics
University of Ljubljana

For an arbitrary fixed surface S, a linear time algorithm is presented that for a given graph
G either finds an embedding of G in S or identifies a subgraph of G that is homeomorphic
to a minimal forbidden subgraph for embeddability in S. A side result of the proof of the
algorithm is that minimal forbidden subgraphs for embeddability in S cannot be arbitrarily

8
large. This yields a constructive proof of the result of Robertson and Seymour that for each
closed surface there are only finitely many minimal forbidden subgraphs. The results and
methods of this work can be used to solve more general embedding extension problems.

Geometric representations of graphs

Sue Whitesides
School of Computer Science
McGill University

Traditionally we represent graphs in a visual way by mapping vertices to points and edges
to curves. Graphs may also be represented geometrically, for example as intersection or
contact graphs, as proximity graphs, as visibility graphs.

We survey several results on a 3-dimensional visibility representation that maps vertices to

polygons floating parallel to the xy-plane and edges to visibility lines parallel to the z-axis.

Reconstructing a surface from its polygonal pieces

Karsten Weihe
Fakultät für Mathematik und Informatik
Universität Konstanz

In the computer aided design of vehicles, engines, and machines of all kinds, the surface of a
workpiece is usually modeled by a set of openly disjoint polygons in the three–dimensional
space. These polygons do not fit exactly together, but incident polygons are only placed
(more or less) close to each other. The input describes solely the geometries of all polygons,
and the neighborhoods are not part of the input. These neighborhoods must be computed
by CAD packages automatically, and the results are by far not sufficient.

In this talk we present an algorithm that seems to yield better results. In contrast to the
other approaches that we know of, we abstract from numerical and geometrical aspects as
far as possible and focus on structural aspects.

This is joint work with Thomas Willhalm.

9
 Mesh generation using flow methods

Rolf Möhring
Fachbereich Mathematik
Technische Universität Berlin

Network flow techniques are applied to a problem arising in the computer aided design
of cars, planes, ships or components of them: Refine a coarse mesh of spheric polygons
that approximates the surface of a workpiece such that the resulting mesh is suitable for a
numerical analysis (in particular, we are asked to achieve a specified mesh density and to
generate meshes consisting of conforming quadrilaterals only).

This turns out to be a difficult discrete problem (strongly N P-hard). We show how to
formulate it as a two-step problem consisting of a choice among refinement templates
and a network flow problem to realize a refinement for the given template choice. More
specifically, the network flow problem is a bidirected flow problem on an undirected graph G
with upper and lower capacities on the edges and some additional node balance conditions.
For a given choice of templates, the problem is then reduced to finding a feasible flow in that
graph that satisfies all these constraints. We describe this model and a generic refinement
algorithm based on it, and discuss the encouraging results of a first implementation.

This is joint work with Matthias Müller-Hannemann and Karsten Weihe.

Space efficient algorithms for image segmentation

Tetsuo Asano
Department of Engineering Informatics
Osaka Electro-Communication University

This paper studies the space complexity of a traditional algorithm for region segmentation
which first grows homogeneous regions and then builds its associated region adjancy graph
which reflects a structure of segmentation results. A naive method in which the region
number is kept at each pixel needs O(n log k) bits in total, where n and k are numbers
of pixels and regions obtained, respectively. Application of on-line graph coloring tech-
niques can save the space complexity of the region growing process to Θ(log log k) bits per
pixel with some difficulty to build a region adjacency graph. Finally, we demonstrate the
advantage of a geometric approach which requires only some constant (4 or 5) bits per
pixel for region growing with reasonable time for establishing region adjacency under some
assumption. We also study several algorithmic issues associated with the implementation
of a region growing algortihm.

10
 Vehicle routing problems with time windows

Goos Kant
Department of Computer Science
Utrecht University

In the Vehicle Routing Problem with Time Windows (VRPTW) there are 2 basic heuristic
algorithms: the sequential insertion algorithm and the Savings Method. First I gave a brief
introduction of this field.

Then I explained different data structures for implementing the Savings Method. The ma-
trix, (partial) heap, (recursive) grid and array implementation are explained, and time and
memory requirements are given. From experimental evaluation the array implementation
turned out to be most efficient in practice. In the last part of the talk I presented an
algorithm for computing the shortest duration of a given tour. This algorithm is based
on computational geometry and can be used as a subprocedure in insertion and savings
methods. It is also explained (and experimentally evaluated) that storing shortest path
information at nodes really helps for dynamic testing of savings and insertion costs.

This is joint work with A. van Vliet and C. de Jong.

Modelling all bottlenecks (edge cuts) of a network, with

incremental maintenance

Yefim Dinitz
Departement of Computer Science
Technion, Haifa

Assume we worry about the (near) minimum edge cuts of a network. Those can be traffic
bottlenecks or most probable communication disconnections due to link failures. In various
cases, there can be a polynomial or exponential number of such bottlenecks. The first
question: can one be shown all the bottlenecks in a single picture (model), which size is
linear in the number of nodes, with simple visual rules to identify bottlenecks? Now assume
we add an edge to the network; clearly, exactly the bottlenecks affected change their status.
The second question: can the picture discussed, if exists, visually support those dynamics
for an edge insertion? for several insertions? We answer these questions positively, more
or less, for the cases of the minimum cuts, of the minimum and subminimum cuts, and
of the cuts that are minimum among dividing a set of poles (distinguished nodes) of the
network. The main graph types used for modeling are cactus tree and, for the third case,
also DAG. Open questions: How to draw adequately the models presented? When can

11
we keep, in the picture, certain relations to geographic positions of nodes? Can such a
drawing be maintained undergoing discussed dynamics and remain recognizable?

Approximation algorithms for semidefinite programs arising

from MAX CUT and COLORING

Philip Klein
Department of Computer Science
Brown University

Linear programming has proved a useful tool in the design of approximation algorithm.
Goemans and Williamson showed (1994) that a more general kind of mathematical pro-
gramming, namely semidefinite programming, is similarly useful; they gave an approxima-
tion algorithms for MAX CUT that depends on solving a semidefinite program. Karger,
Motwani, and Sudan then showed (1994) a similar approach yields an approximation al-
gorithm for graph coloring. In each of these algorithms, the computational bottleneck is
solving the semidefinite program. We show that for constant ǫ a (1 + ǫ)-approximate solu-
tion to these programs can be obtained in O(nm log 3 n), where n is the number of nodes
and m is the number of edges.

This is joint work with Hsueh-I Lu.

Computing subgraphs of minimum weight triangulations based

on LMT-skeletons

Naoki Katoh
Department of Management Science
Kobe University of Commerce

We present improvements in finding the LMT-skeleton, which is a subgraph of all mini-

mum weight triangulation. Our improvements consist of: (1) A criterion is proposed to
identify edges in all minimum weight triangulation (2) A faster algorithm is presented for
performing one pass of the Dickenson and Montague method. (3) Improvements in the im-
plementation that may lead to substantial space reduction for uniformly distributed point
sets.

This is joint work with Manabu Sugai and Sin-Wing Chang.

12
 On drawing planar angle graphs

Franz J. Brandenburg
Lehrstuhl für Theoretische Informatik
Universität Passau

Recently, some advanced algorithms for straight line drawings of planar graphs have been
introduced. However, these algorithms produce drawings with many small angles. Their
drawings are not aesthetically nice.

Our goal is to resolve these problems by a beautifier. We consider the angles of planar
straight line drawings. First, we compute maximal angles by repeatedly solving a linear
program. However, the existence of a planar straight line drawing with these angles is not
guaranteed. We approach drawings either by a linear program for the edge length or by a
new spring embedder, which emphasizes on angles.

Our experiments are very promising. The approach outperforms other advanced algorithms
for straight line drawings and unfolds drawings with too many small angles.

This is joint work with H. Steine and A. Stübinger.

Drawing clustered graphs

Peter Eades
Department of Computer Science
University of Newcastle

We show that every c-planar clustered graph has a straight-line c-planar drawing. To prove
this result for clustered graphs we prove a similar result about hierarchical graphs.

This is joint work with Q.W. Feng.

13
 Advances in orthogonal graph drawing

Ioannis Tollis
Computer Science Department
University of Texas at Dallas

An orthogonal drawing of a graph is a drawing such that nodes are placed on grid points
and edges are drawn as sequences of vertical and horizontal segments. In this talk we
present the following results:

(1) An algorithm for constructing orthogonal drawings with at most 2n + 2 bends and
0.76n2 area.

(2) Interactive scenaria for constructing orthogonal drawings such that the drawing re-
mains almost unchanged after an insertion or deletion of an edge or vertex.

The algorithms run in linear time on graphs with n vertices and of maximum degree four.

This is joint work with Achilleas Papakostas.

On bend-minimum orthogonal graph drawings

Michael Kaufmann
Institut für Informatik
Universität Tübingen

In this talk we present old and new techniques to minimize the number of bends in an
orthogonal drawing for a given embedded planar graph. First we review Tamassia’s algo-
rithm which solves the problem for 4-planar graphs. Then we discuss how to get a suitable
model for orthogonal drawings of vertices of higher degree. We adopt a model used in a
drawing of a graph from astrophysics and present an extension of Tamassia’s network flow
method that achieves the optimum while keeping the vertices small. Finally we discuss
two approaches for 0-bend drawings (so-called 2D-visibility drawings) where the vertices
are represented by rectangles of different sizes. Besides of some motivating pictures we also
give two simple theoretical results on this model.

This is joint work with Uli Fößmeier.

14
 Implementation of graph algorithms with LEDA

Stefan Näher
Fachbereich Mathematik und Informatik
Universität Halle

LEDA is a Library of Efficient Datatypes and Algorithms that can be used as a platform for
various kinds of combinatorial and geometric computing. In this talk we demonstrate that
LEDA is particularly useful for implementing graph algorithms by providing an efficient
and comfortable graph data type and many basic graph algorithms and data structures
related to graphs. In particular, the programs for computing a minimum spanning tree
(Kruskal), shortest paths in non-negative edge weights (Dijkstra) and st-numberings for
biconnected graphs are presented. In the second part of the talk some details of the
implementation of the Kuratowski-Algorithm (Hundack, Mehlhorn, Näher) are discussed.

The proximity drawability problem

Giuseppe Liotta
Department of Computer Science
Brown University

A proximity drawing of a graph G is a straight-line drawing of G such that pairs of adjacent

vertices are deemed “close” according to some proximity measure. For example, in a
Gabriel drawing two vertices are adjacent if and only if the circle having the two vertices
as antipodal points is empty (i.e. it does not contain any other third vertex).

Aim of the talk is to survey different types of proximity drawings and describe some of the
main combinatorial results and drawing algorithms that can be found in the literature.

Two arc disjoint paths in Eulerian digraphs

Toshihide Ibaraki
Graduate School of Engineering
Kyoto University

Let G be an Eulerian digraph, and {x1 , x2 }, {y1 , y2} be two pairs of vertices in G. An
instance (G; {x1 , x2 } {y1 , y2}) is called feasible if it contains two arc disjoint x ′ x′′ – and
y ′y ′′ –paths, where {x′ , x′′ } = {x1 , x2 } and {y ′ , y ′′} = {y1 , y2 }. An O(m + n log n) time

15
algorithm is presented to decide whether G is feasible, where n and m are the numbers of
vertices and arcs in G, respectively. The algorithm is based on a structural characterization
of minimal infeasible instances.

This is joint work with Hiroshi Nagamochi and Andras Frank.

Two results on search and edge connectivity

David W. Matula
Depart. of Computer Science & Engineering
Southern Methodist University

We give two results employing iterated search to provide information on the maximum
number of edge disjoint u − v paths, λ(u, v), for all u, v in an n vertex m edge graph. Here
each unit flow augmentation is determined by a breath first search (BFS).

Our first result is that in time O(nm) we can confirm either:

(1) λ(u, v) = min{deg(u), deg(v)} for all vertex pairs u, v; or

(2) for some x, y there is a non trivial x, y min cut (A, Ā),
x ∈ A, y ∈ Ā, with size |(A, Ā)| ≤ min{deg(x), deg(y)} − 1.

In particular our algorithm needs to find at most [n − max degree G] BFS’s with additional
work only O(m). In case (1) we note the Gomory-Hu cut tree must then be a star with
center vertex a max degree vertex of the graph G.

Our second result employs maximum adjacency search vertex ordering (MAS) which can be
determined in linear time O(n+m). An MAS ordering requires for each i that v i have max-
imum adjacency with {v 1 , v2 , . . . , vi−1 } among the remaining vertices V −{v 1 , v2 , . . . , vi−1 }.
Ibaraki and Nagamochi have shown an edge labeling associated with an MAS that parti-
tions the edges onto nested spanning forests F1 , F2 , . . . Fdeg(v\ ) such that an edge labeled
k has k edge-disjoint paths between its endpoints, one each in F 1 , F2 , . . . Fk . We describe
an associated minimum acyclic arc labeling (MA 2 L) procedure determining two partitions
of the edges into nested spanning forests F1 , F2 , . . . Fdeg(v\ ) and F1′ , F2′ , . . . , Fdeg(v
′
1)
. Each
edge having two labels, k from the first forest and j from the second, indicates there are
k +j −1 mutually edge disjoint paths of G between its endpoints, one each in F 1 , F2 , . . . , Fk
and F1′ , F2′ , . . . , Fj−1
′
. The MA2 L can be found in time only slightly greater than linear
by the inverse Ackermann-function (embedded Union-Find Problem). The MA 2 L results
generally provide paths giving a tight lower bound on the λ(u, v) values providing the path
dual to the Gomory-Hu cut tree.

16
 Practical performance of MIN CUT algorithms

Michael Jünger
Institut für Informatik
Universität Köln

The computation of minimum weight cuts in simple graphs with nonnegative edge weights
has direct applications in connectivity problems, and is used as a subroutine in algorithms
for some difficult combinatorial optimization problems. E.g., branch and cut algorithms
for the traveling salesman problem make thousands of calls to mincut subroutines in their
separation of valid inequalities. In the recent years, a lot of progress has been made, and
some interesting new methods have been proposed. We give a computational comparison of
several of them on various data sets. The study includes implementations of algorithms by
Gomory & Hu (Gusfield), Padberg & Rinaldi, Karger & Stein, Hao & Orlin, Nagamochi,
Ono & Ibaraki, Stoer & Wagner. The outcome of our experiments could not be predicted
by the theoretical performance analyses.

This is joint work with Giovanni Rinaldi and Stefan Thienel.

Optimization on hierarchical graphs

Petra Mutzel
Max-Planck-Institut für Informatik

We study optimization problems in hierarchical graphs that have applications in Computa-

tional Biology and Automatic Graph Drawing. Methods that score the similarity between
various sequences of letters are called Multiple Alignment Problems. The Maximum Weight
Trace Problem (MWT) is one of the various models. Given an alignment graph (arising
from the given sequences) with edge weights, find a set of edges that can be realized by
an alignment (called Trace) of maximum weight. We give a characterization in terms of
forbidden subgraphs that enables us to give an integer linear programming formulation for
the MWT-problem. We report on first computational results on solving the LP-relaxation.

The problems we are considering in Automatic Graph Drawing help to improve the draw-
ings of hierarchical graphs. We are focussing on the Straightline Crossing Minimization
Problem and on the L-Planarization Problem on two layers. Depending on the number of
layers, in which the vertices can be permuted freely (zero, one or two) different versions of
the problems arise.

All of these problems are tightly connected to each other. E.g., the MWT-problem for
two sequences is equivalent to L-planarization on two layers with both layers fixed. The

17
knowledge of the latter problem and of the straight-line crossing minimization problem on
two layers with one layer fixed, will give us information about the L-planarization problem
on two layers with one layer fixed.

The MWT-problem is joint work with J. Kececioglu (University of Georgia), H.-P. Lenhof,
K. Mehlhorn and K. Reinert (MPI Saarbrücken), the crossing minimization problem is
joint work with M. Jünger (Universität zu Köln). Thanks to T. Ziegler and R. Weißkircher
(MPI Saarbrücken) who provided most of the software for the graph drawing problems.

The bandwidth problem of clique caterpillars and tolerance

graphs

Maciej Syslo
Institute of Computer Science
University of Wroclaw

We consider the bandwidth problem restricted to clique caterpillars and tolerance graphs.
The bandwidth of a graph G is the minimum of the maximum absolute differece between
adjacent labels when vertices are labelled with integers. We show that the bandwidth of
cliques with hairs and the bandwidth of clique caterpillars with hairs of length 1 depends
only on local density, that is, it is equal to max H⊆G (|V (H)| − 1)/diam(H), where |V (H)| is
the number of vertices of H and diam(H) is the largest length of all shortest paths between
any two vertices of H.

We also present algorithms which label optimally these graphs. Then, we show that the
bandwidth problem for clique caterpillars with hairs of length at most 2 is N P-complete.
Moreover, it is shown that the problem remains N P-complete for tolerance graphs, a
generalization of interval graphs (for which there exists a polynomial-time algorithm).

PlaNet: A demonstration package for algorithms on PLAnar

NETworks

Dorothea Wagner
Fakultät für Mathematik und Informatik
Universität Konstanz

PlaNet is a package for algorithms on planar networks. It comes with a graphical user
interface, which may be used for demonstrating and animating algorithms. Our focus so

18
far has been on disjoint path problems. However, the package is intended to serve as a
general framework, where n algorithms for various problems on planar networks may be
integrated and visualized. For this aim, the structure of the package is designed so that
integration of new algorithms and even new algorithmic problems amounts to applying
a short “recipe”. The package has been used to develop new variations on well-known
disjoint path algorithms, which heuristically optimize additional objections such as the
total length of all paths.

A demo version is accessible through the WWW:

http://www.informatik.uni-konstanz.de/Research/Projects/PlaNet/

This is joint work with Dagmar Handke, Karsten Weihe, Gabriele Neyer, and Wolfram
Schlickenrieder.

Painting graph algorithms

David W. Matula
Depart. of Computer Science & Engineering
Southern Methodist University

Painting Graph Algorithms (PGA) is an interactive program written in Tcl/Tk for use in
X Window systems. The program was created to serve as a graph algorithm research tool.
It has also proved to be useful for teaching graph algorithms. Currently, PGA is primarily
used to refine and extend graph and network search algorithms.

PGA allows the user to draw and edit graph diagrams on a canvas. The user can add and
delete vertices and edges as well as move vertices. PGA supports directed, undirected, and
weighted graphs. Undirected graphs can be displayed with straight edges or pairs of curved
arcs. The adjacency list data structure is dynamically updated as the graph is drawn and
edited. The graph data structures can be saved in a file format suitable for input to other
graph programs.

The painting process animates graph search algorithms by dynamically coloring vertices
and edges as they are reached and visited. Color, and varying brush widths, are applied to
edges and vertices to display information visually, with labels added to elements to indicate
the current state of computation. PGA can be executed in the wish shell, allowing the user
to directly manipulate the data structures as the painting progresses. Painted elements can
also be moved on the canvas, for improved visual display. By experimenting with artistic
variations of painting the graph, we have realized painted graphs that convey the multiple
computational results of search algorithms such as MAS. Painted graphs can be printed or
saved as postscript files.

19
For information on program availability please contact Michael Coming at coming@seas.smu.edu.

L –Steiner trees
1

Bernhard Korte
Institut für Ökonometrie und OR
Universität Bonn

To construct minimal Steiner trees is an old mathematical problem, which in its simplest
form goes back to Fermat. In some sense it is wrongly attributed to the swiss geometer
Jacob Steiner. The first modern formulation as a shortest network problem was given by
the czech mathematicians Jarnik and Kössler in 1934. Today minimal L 1 -Steiner trees are
a central issue in VLSI design.

Compared to other combinatorial optimization problems like the TSP the Steiner tree
problem was considered much harder. Only very small instances could be solved optimally.

Based on some decomposition result which states that each optimal L 1 -Steiner tree consists
only of four types of elementary trees (so called firs) and based on powerful reductions,
we could develop a branch-and-bound type algorithm which can optimally construct L 1 -
Steiner trees for up to 100 terminals in reasonable time. This algorithm enables us to do
the complete routing of up to a million nets on a chip as optimal L 1 -Steiner trees.

All-pairs MIN CUT in sparse networks

Christos Zaroliagis
Max-Planck-Institut für Informatik

Algorithms are presented for the all-pairs min-cut problem in bounded tree-width, planar
and sparse networks. The approach used is to preprocess the input n-vertex network so
that, afterwards, the value of a min-cut between any two vertices can be efficiently com-
puted. A tradeoff is shown between the preprocessing time and the time taken to compute
min-cuts subsequently. In particular, after an O(n log n) preprocessing of a bounded tree-
width network, it is possible to find the value of a min-cut between any two vertices in
constant time. This implies that for such networks the all-pairs min-cut problem can be
solved in time O(n 2 ). This algorithm is used in conjunction with a graph decomposition
technique of Frederickson to obtain algorithms for sparse and planar networks. The run-
ning times depend upon a topological property, γ, of the input network. The parameter
γ varies between 1 and Θ(n); the algorithms perform well when γ = o(n). The value of a

20
min-cut can be found in time O(n + γ 2 log γ) and all-pairs min-cut can be solved in time
O(n2 + γ 4 log γ) for sparse networks. The corresponding running times for planar networks
are O(n + γ log γ) and O(n2 + γ 3 log γ), respectively. The latter bounds depend on a result
of independent interest: outerplanar networks have small “mimicking” networks which are
also outerplanar.

This is joint work with S. Arikati and S. Chaudhuri.

Some linear time solvable subcases of the multilayer routing

problem

Andras Recski
Faculty of Electrical Engineering and Information
University of Budapest

Most of the interesting special cases of the detailed routing problem are N P-complete, for
example to determine whether a given channel routing problem can be solved with given
width in the 2-layer Manhattan model.

However, there are some polynomially solvable subcases as well. For example, if all the
terminals to be interconnected are on a single boundary of the rectangular routing area
(the so called single row routing problem) then the 2-layer Manhattan routing can always
be performed, in fact in linear time, by a classical algorithm of Tibor Gallai. Similarly, the
solvability of the switchbox routing problem with edge disjoint paths in a single layer can
always be decided in linear time by an algorithm of Dorothea Wagner.

In this talk we present linear time solutions for two problems, namely the channel routing
problem in the 2-layer unconstrained model and the switchbox routing problem in the
multilayer unconstrained model. Unfortunately, none of the algorithms are optimal (the
first one does not realize minimum width and the second one does not minimize the number
of layers).

This is joint work with Frank Strzyzewski, Endre Boros, and Ferenc Wettl.

21
 Application Problems
Chair: Ioannis Tollis

Goos Kant: Dijkstra’s shortest path algorithm using less memory There are several
implementations of Dijkstra’s single source shortest path algorithm. In practical instances
of route maps at the ORTEC company, the number of vertices and edges is very large
compared to the maximum edge weight. As an example, a road map of Europe has about
300,000 nodes, and the travel time of a link/edge is at most 60 minutes. If we represent
every edge weight by a bucket and a doubly linked list for all vertices, currently labeled
with the corresponding weight, we need 2 pointers at every vertex. This together yields a
datastructure for the algorithm.
However, 2 pointers is too much memory in practice. To decrease the memory requirements
we don’t want to shrink the network, and we don’t want to apply a considerably larger
running time. The solution we propose is based on the following observation: during
the algorithm only a few vertices are ‘interesting’ at any step. At any step we store the
’interesting’ vertices together with their 2 pointers in a hash table. Of course the hash
table requires memory, but since the number of ’interesting’ vertices during the algorithm
is very small compared with the total number of vertices, this saves a lot of memory.
Is this approach already known, and are there any good alternatives to apply Dijkstra’s
algorithm on such networks?

Stephen North: Dynamic graph layout Our group has several years’ experience in imple-
menting graph-drawing systems and helping software developers to apply these systems to
problems such as software process (work flow) modeling and finite state machine synthesis
and analysis. Although users are pleased with automatic graph layout, they say that stable
incremental layout is essential. Lack of this prevents them from fully adopting our systems.
One of our current projects is to design good heuristic algorithms, interfaces and systems
for effective incremental layout. In fact, dynamic graph layout is highly relevant in many
settings, as most real-world systems are subject to change. Another strong motivation is
to browse large graphs by incrementally adjusting views of small subgraphs.

Stephen North: Efficient algorithms for graphs in external memory One project in
our lab involves analyzing massive volumes of telephone network customer data. AT&T
has more than 80,000,000 customers and on a recent day set handled a record of more than
239,000,000 telephone calls. We would like to solve graph pattern matching problems on
this database (e.g. finding highly connected subgraphs). Besides being large, the graphs
are dynamic and only approximate. We are examining the relevance of recent theoretical
contributions on external graph algorithms and I/O parallelism.

Ioannis Tollis: Design of survivable networks Find better algorithms for designing cost
efficient telecommunications networks, based on the survivable ring architecture model.

22
Ioannis Tollis: Algorithms for developing lower bounds on the cost of Self-Healing-
Ring architectures When solving a difficult optimization problem using heuristics, it is
very often the case that we have no idea about the quality of our solution. This is the case
when designing Self-Healing-Ring architectures. It is an interesting problem to develop
algorithms that for a given network produce a total cost which is a good lower bound for
any (optimal) solution.

Goos Kant: Assigning orders to inspectors In several practical environments, the fol-
lowing problem arises: given n orders and k (k < n) inspectors distributed in an area,
assign the n orders to the k inspectors such that every inspector has roughly the same
working time. The working time of an inspector is defined by the travelling salesman tour
from his location through all assigned orders back to his location + the visiting time of
the assigned orders. Of course, this problem is NP-hard.
Most algorithms in the text books are based on the following technique: Compute for every
order the extra cost for assigning this order to every inspector, say cost c 1 , . . . , ck . Let cα
and cβ be the smallest and one but smallest cost, resp. Then this order is assigned to
a customer, for which cβ − cα is maximum. After updating the costs for the remaining
orders, the procedure is repeated, until all orders are assigned. Variants are: looking at
the best p (p > 2) inspectors, and special scaling values for popular and impopular areas
of the location of the inspector.
However, in almost all cases, it is very difficult to deal with popular (and impopular) areas.
In other words: in some areas the number of orders is relatively very large (small) com-
pared with the number of inspectors. This implies that inspectors from impopular areas
have to do orders in the direction of the popular regions. Steering this is quite compli-
cated in combination with balancing the working times. Are there any papers or practical
algorithms that can deal with these problems?

23
 Open Problems
Chair: Roberto Tamassia

Roberto Tamassia: Robust Steinitz’ Theorem Steinitz’s Theorem says that a graph G
is the skeleton of a convex 3-polytope if and only if it is planar and triconnected. Is there
a “robust” version of this theorem? Namely, what is the set of values for constants h, k,
and l such that given a planar triconnected graph G with N vertices, we can construct a
convex 3-polytope P with skeleton G that verifies the following properties:

• P satisfies the vertex resolution rule, i.e., any two vertices of P are at Euclidean
distance at least 1;

• the coordinates of the vertices of P are rational numbers with O(N h log N) bits;
• P has O(N k ) volume;

• P has O(N l ) aspect ratio (ratio of the largest to the smallest of length, width, and
height of P ).

Ideally, we would like to characterize the set of all minimal triplets (h, k, l) and give efficient
constructions for them.
Trivial lower bounds are h ≥ 0, k ≥ 1 and l ≥ 0. However, no construction for these values
of the constants is known. An O(N 1.2 )-time construction for h = 1, k = 1 and l = 1 is
presented in [*]. While it achieves optimal volume, its disadvantages are the high number
of bits for the vertex coordinates and the poor aspect ratio.
[*] M. Chrobak, M.T. Goodrich and R. Tamassia, “Convex Drawings of Graphs in Two
and Three Dimensions,” Proc. ACM Symp. on Computational Geometry, 319-328 (1996).

Peter Eades: The K7-3D-2B Problem Is there an orthogonal grid drawing of K 7 in 3

dimensions with no edge crossings and at most 2 bends per edge?
(An orthogonal grid drawing of a graph represents vertices as points on the integer grid
and edges as polylines whose segments lie on the integer grid lines.)

Ioannis Tollis: “Better” hierarchical drawings Find algorithms for hierarchical drawings
of graphs that take into account the global shape of the graph: For example, draw a graph
by removing edges until the remaining (sub)graph is a series-parallel, or an upward planar
graph. Draw the subgraph and then insert the edges one at a time.

Petra Mutzel: Separation Problem of Kuratowski Subgraphs Given a nonplanar

graph G = (V, E) with weights we for all edges e ∈ E, 0 ≤ ce ≤ 1 (most of them are 0, or
between 0 and 0.2). Find a subset K ⊆ E that is a subdivision (homeomorph graph) of
P
K5 or K3,3 with minimum weight e∈K we .

24
 The above question arises as “separation problem” in the maximum planar subgraph prob-
lem. Given a solution of a Linear Program, x e , 0 ≤ xe ≤ 1, find a Kuratowski subgraph
P
(i.e., a subdivision of K 3,3 or K5 ) violating the inequality e∈K xe ≤ |K| − 1 or prove
that none such subgraph exists. Rewriting the inequality and setting w e = 1 − xe gives
P
e∈K we ≥ 1. If the minimum weight Kuratowski subgraph satisfies the inequality, then it
is satisfied by all the Kuratowski subgraphs. Otherwise, we found a Kuratowski subgraph
P
that violates the inequality, i.e. e∈K (1 − xe ) < 1.

Takao Nishizeki and X. Zhou: Can the vertex-coloring problem be “simply” re-
duced to the edge-coloring problem? The opposite direction is easy: the edge-coloring
problem can be simply reduced to the vertex-coloring problem, because the chromatic in-
dex of a graph G is equal to the chromatic number of the line graph of G. Since both
problems are NP-complete, either can be reduced to the other plausibly through 3-SAT,
due to the theory of NP-completeness. Thus the open problem asks, given a graph G and a
positive integer k, whether one can find in polynomial time a new graph G’ and a positive
integer k’ such that G can be vertex-colored with at most k colors if and only if G’ can be
edge-colored by at most k’ colors.

Sue Whitesides: How quickly can we determine if a 4-regular graph G is the union
of two Hamilton cycles? Let G be a connected graph on n vertices such that each vertex
has degree 4. How quickly can we determine whether the edge set E(G) of G can be parti-
tioned into two sets, each of which forms a Hamiltonian cycle for G? More generally, how
quickly can we determine whether the edges of a 2k-regular graph G can be partitioned
into k edge-disjoint Hamilton cycles on G? Are these problems NP-complete?
(These questions arose in discussions with Tom Shermer of Simon Fraser University, Canada.)

Giuseppe Liotta: The planar 2D-visibility representability problem Given a set S of

axis-aligned rectangles in the plane, two rectangles of S are horizontally (vertically) visible
if they can be connected by an horizontal (vertical) straight-line segment l that does not
intersect any other rectangle of S. Segment l is called segment of visibility.
A 2D-visibility representation of a graph G is a drawing of G on the plane such that every
vertex is represented by an axis-aligned rectangle and two vertices are adjacent if and only
if their corresponding rectangles are either horizontally or vertically visible. An edge of
a 2D-visibility representation is either an horizontal or a vertical straight-line segment of
visibility. A 2D-visibility representation is planar if no two segments of visibility intersect.
Question: Which planar graphs admit a planar 2D-visibility representation?
Context for the question: graph drawing, visibility representations, zero-bend drawings.

Kurt Mehlhorn: Robustness of network algorithms There are simple examples where
network algorithms when implemented with floating point arithmetic produce wrong re-
sults. For shortest path algorithms which only add numbers it is fairly easy to show that
the computed result is close to the true result. More complicated network algorithms add

25
 and subtract and errors may accumulate. Can one still show a relation between computed
result and true result?

Ioannis Tollis: Nontrivial properties of planar bipartite graphs The ultimate aim is to
partition the vertices of a planar graph into two parts such that each part is bipartite. This
is related to the 4-coloring theorem, and would give an independent constructive proof of
the theorem. So, find properties of planar bipartite graphs that will allow one to discover
algorithms to partition the vertices as mentioned above.

26
Dagstuhl-Seminar 9620 List of participants

Takao Asano Yefim Dinitz

Chuo University Technion – Haifa
Dept. of Information and System Engineer- Department of Computer Science
ing Technion
Bunkyo-ku Haifa 32000
Tokyo 112 Israel
Japan dinitz@cs.technion.ac.il
asano@ise.chuo-u.ac.jp tel: +972-4-8294-368
tel: +81-3-3817-1686
Peter Eades
Tetsuo Asano University of Newcastle
Osaka Electro-Communication University Department of Computer Science
Dept. of Engineering Informatics University Drive
Faculty of Information & Computer Science Callaghan NSW 2308
18-8 Hatsu-cho/Neyagawa Australia
Osaka 572 eades@cs.newcastle.edu.au
Japan tel: +61-49-216034
asano@amlab.osakac.ac.jp
tel: +81-720-24-1131/ext. 2450 Wen-Lian Hsu
Academica Sinica – Taipei
Franz J. Brandenburg Institute of Information Science
Universität Passau Nankang
Lehrstuhl für Theoretische Informatik Taipei 115
94030 Passau Taiwan
Germany hsu@iis.sinica.edu.tw
brandenb@fmi.uni-passau.de
tel: +49-851-509-3030 Toshihide Ibaraki
Kyoto University
Ulrik Brandes Graduate School of Engineering
Universität Konstanz Department of Applied Math. and Physics
Fakultät für Mathematik und Informatik Kyoto 606
Postfach 55 60 D 188 Japan
78434 Konstanz ibaraki@kuamp.kyoto-u.ac.jp
Germany tel: +81-75-753-5504
Ulrik.Brandes@uni-konstanz.de
tel: +49-7531-88-4263 Michael Jünger
Universität zu Köln
Institut für Informatik
Pohligstr. 1
D–50969 Köln
Germany
mjuenger@informatik.uni-koeln.de
tel: +49-221-470-5313
Dagstuhl-Seminar 9620 List of participants

Goos Kant Han La Poutré

Utrecht University Leiden University
Dept. of Computer Science Dept. of Computer Science
P.O. Box 80-089 Niels Bohrweg 1
NL–3508TB Utrecht NL–2300 RA Leiden
The Netherlands The Netherlands
goos@cs.ruu.nl han@wi.leidenuniv.nl
tel: +31-71-5277112
Naoki Katoh
Kobe University of Commerce Annegret Liebers
Department of Management Science Universität Konstanz
8-2-1 Gakuennishi Fakultät für Mathematik und Informatik
Nishi-ku Kobe 651-21 Postfach 55 60 D 188
Japan 78434 Konstanz
naoki@kucgw.kobeuc.ac.jp Germany
tel: +81-78-794-6161 Annegret.Liebers@uni-konstanz.de
tel: +49-7531-88-4263
Michael Kaufmann
Universität Tübingen Giuseppe Liotta
Institut für Informatik Brown University
Sand 13 Department of Computer Science
D–72076 Tübingen 115 Waterman Street
Germany Providence RI 02912
mk@informatik.uni-tuebingen.de USA
tel: +49-7071-29-7404 gl@cs.brown.edu

Philip Klein David W. Matula

Brown University Southern Methodist University
Department of Computer Science Depart. of Computer Science & Engineering
P.O. Box 1910 75275 Dallas TX
Providence RI 02912 USA
USA matula@seas.smu.edu
pnk@cs.brown.edu
tel: +1-401-863-76 44 Kurt Mehlhorn
Max-Planck-Institut für Informatik
Bernhard Korte Im Stadtwald
Universität Bonn 66123 Saarbrücken
Institut für Ökonometrie und OR Germany
Nassestr. 2 mehlhorn@mpi-sb.mpg.de
D–53113 Bonn tel: +49-681-9325-100
Germany
dm@or.uni-bonn.de
tel: +49-228-738770
Dagstuhl-Seminar 9620 List of participants

Rolf Möhring Takao Nishizeki

TU Berlin Tohoku University
Fachbereich Mathematik, MA 6-1 Graduate School of Information Sciences
Straße des 17. Juni 136 Sendai 480-77
10623 Berlin Japan
Germany nishi@ecei.tohoku.ac.jp
moehring@math.tu-berlin.de tel: +81-22-263-9301
tel: +49-30-314-24594
Stephen North
Bojan Mohar AT&T Bell Labs
University of Ljubljana 600 Mountain Avenue
Dept. of Mathematics Murray Hill NJ 07974
Jadranska 19 USA
1111 Ljubljana north@research.att.com
Slovenia
bojan.mohar@uni-lj.si Andras Recski
Technical University of Budapest
Petra Mutzel Faculty of Elec. Eng. and Information
Max-Planck-Institut für Informatik Dept. of Mathematics and Computer Science
Im Stadtwald H–1521 Budapest
66123 Saarbrücken Hungary
Germany recski@vma.bme.hu
mutzel@mpi-sb.mpg.de tel: +36-1-463-2585
tel: +49-681-9325-105
Maciej M. Syslo
Stefan Näher University of Wroclaw
Universität Halle Institut of Computer Science
Fachbereich Mathematik und Informatik Przesmyckiego 20
Kurt-Mothes-Str. 1 PL–51–151 Wroclaw
D–06099 Halle Poland
Germany syslo@ii.uni.wroc.pl
naeher@informatik.uni-halle.de tel: +48-71-247-382
tel: +49-345-552-4712 fax: +48-71-251271

Hiroshi Nagamochi Roberto Tamassia

Kyoto University Brown University
Graduate School of Engineering Department of Computer Science
Dept. of Applied Mathematics and Physics P.O. Box 1910
Kyoto 606-01 Providence RI 02912
Japan USA
naga@kuamp.kyoto-u.ac.jp rt@cs.brown.edu
tel: +81-75-753-5494 tel: +1-401-863-7639
Dagstuhl-Seminar 9620 List of participants

Ioannis Tollis Karsten Weihe

University of Texas at Dallas Universität Konstanz
Computer Science Department Fakultät für Mathematik und Informatik
EC 31 Postfach 55 60 D188
P.O. Box 830688 78434 Konstanz
Richardson TX 75083-0688 Germany
USA karsten.weihe@uni-konstanz.de
tollis@utdallas.edu tel: +49-7531-88-4375
tel: +1-214-8832180
Sue Whitesides
Dorothea Wagner McGill University
Universität Konstanz School of Computer Science
Fakultät für Mathematik und Informatik 3480 University Street
Postfach 55 60 D188 Montreal PQ H3A 2A7
78434 Konstanz Canada
Germany sue@cs.mcgill.ca
Dorothea.Wagner@uni-konstanz.de tel: +1-514-398-7071
tel: +49-7531-88-2893
Christos Zaroliagis
Frank Wagner Max-Planck-Institut für Informatik
Freie Universität Berlin Im Stadtwald
Fachbereich Mathematik und Informatik 66123 Saarbrücken
Takustr. 9 Germany
D–14195 Berlin-Dahlem zaro@mpi.sb.mpg.de
Germany tel: +49-681-9325-116
wagner@inf.fu-berlin.de
tel: +49-30-838-75159/75103

Toshimasa Watanabe
Hiroshima University
Faculty of Engineering
Department of Circuits and Systems
Kagamiyama 1-4-1
Higashi-Hiroshima City 739
Japan
watanabe@huis.hiroshima-u.ac.jp
tel: +81-824-24-7662