DOKK Library

Algorithms for Intersection Graphs for $t$-Intervals and $t$-Pseudodisks

Authors Chandra Chekuri, Tanmay Inamdar,

License CC-BY-3.0

Plaintext
                          T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19
                                       www.theoryofcomputing.org




     Algorithms for Intersection Graphs of
        ๐‘ก-Intervals and ๐‘ก-Pseudodisks
                          Chandra Chekuriโˆ—                          Tanmay Inamdarโ€ 
                   Received December 19, 2019; Revised June 8, 2022; Published June 23, 2022




       Abstract. Intersection graphs of planar geometric objects such as intervals, disks,
       rectangles and pseudodisks are well-studied. Motivated by various applications,
       Butman et al. (ACM Trans. Algorithms, 2010) considered algorithmic questions
       in intersection graphs of ๐‘ก-intervals. A ๐‘ก-interval is a union of ๐‘ก intervalsโ€”these
       graphs are also referred to as multiple-interval graphs. Subsequent work by Kammer
       et al. (APPROX-RANDOM 2010) considered intersection graphs of ๐‘ก-disks (union
       of ๐‘ก disks), and other geometric objects. In this paper we revisit some of these
       algorithmic questions via more recent developments in computational geometry.
       For the minimum-weight dominating set problem in ๐‘ก-interval graphs, we obtain a
       polynomial-time ๐‘‚(๐‘ก log ๐‘ก)-approximation algorithm, improving upon the previously
       known polynomial-time ๐‘ก 2 -approximation by Butman et al. (op. cit.). In the same
       class of graphs we show that it is NP-hard to obtain a (๐‘ก โˆ’ 1 โˆ’ ๐œ–)-approximation for
       any fixed ๐‘ก โ‰ฅ 3 and ๐œ– > 0.
           The approximation ratio for dominating set extends to the intersection graphs of
       a collection of ๐‘ก-pseudodisks (nicely intersecting ๐‘ก-tuples of closed Jordan domains).
       We obtain an ฮฉ(1/๐‘ก)-approximation for the maximum-weight independent set in the
   โˆ— Chandra Chekuri is supported in part by NSF grants CCF-1526799 and CCF-1910149.
   โ€  Tanmay Inamdar is supported in part by NSF grant CCF-1615845.



ACM Classification: F.5.2.2, F.6.2
AMS Classification: 68W25
Key words and phrases: approximation algorithms, computational geometry, multiple interval
graphs, geometric set cover, dominating set, independent set


ยฉ 2022 Chandra Chekuri and Tanmay Inamdar
c b Licensed under a Creative Commons Attribution License (CC-BY)                 DOI: 10.4086/toc.2022.v018a018
                                   C HANDRA C HEKURI AND TANMAY I NAMDAR

       intersection graph of ๐‘ก-pseudodisks in polynomial time. Our results are obtained
       via simple reductions to existing algorithms by appropriately bounding the union
       complexity of the objects under consideration.


1    Introduction
A number of interesting optimization problems can be modeled as packing and covering
problems involving geometric objects in the plane such as intervals, disks, rectangles, triangles,
convex polygons and pseudodisks1. Some of these problems can be studied via properties
of the intersection graph2 defined by the objects under consideration. Geometric intersection
graphs are interesting for both theoretical and practical reasons. For instance, the well-known
Koebeโ€“Andreevโ€“Thurston theorem shows that every planar graph can be represented as the
intersection graph of interior-disjoint disks in the plane (see Sections 13.6 and 13.7 of [34]).
    Interval graphs are another well-studied class of geometric intersection graphs that are
defined by a finite set of intervals on the real line. Algorithmic problems on interval graphs have
been motivated by applications in scheduling, resource allocation, and computational biology.
    Several papers [8, 9, 10, 26] have studied geometric intersection graphs in the setting where
each composite object is now the union of a set of base geometric objects. To make the discussion
concrete we first discuss ๐‘ก-interval graphs. For an integer parameter ๐‘ก โ‰ฅ 1, a ๐‘ก-interval is the
union of ๐‘ก intervals. A ๐‘ก-interval graph is the intersection graph of a set of ๐‘ก-intervals. These
graphs are also called multiple-interval graphs. They have been well-studied from graph
theoretic and algorithmic points of view. For instance, every graph with maximum degree ฮ” can
be represented as a ๐‘ก-interval graph for ๐‘ก = d(ฮ” + 1)/2e [21]. This demonstrates the modeling
power obtained by considering unions of simple geometric objects. Butman et al. [10], building
on [8, 9] (which primarily studied the maximum independent set problem), considered several
optimization problems in ๐‘ก-interval graphs such as minimum vertex cover, dominating set
and maximum clique. Unlike the case of interval graphs, where these problems are tractable,
the corresponding problems in ๐‘ก-interval graphs are NP-hard even for small values of ๐‘ก and
unweighted instances; this can be seen from the preceding comment on the modeling power
of ๐‘ก-interval graphs, and the NP-hardness of several problems on bounded-degree graphs.
Butman et al. describe polynomial-time approximation algorithms for these problems, and the
approximation ratios they obtain depend on ๐‘ก. We refer the reader to [10] and references therein
for a more detailed discussion on applications of multiple-interval graphs. In subsequent work,
Kammer et al. [26, 25] studied (among other models) intersection graphs of ๐‘ก-disks (a ๐‘ก-disk is a
union of ๐‘ก disks) and ๐‘ก-fat objects3. They obtained approximation algorithms for optimization
problems on these graphs such as maximum independent set, minimum vertex cover and
    1A collection of pseudodisks is a set of closed Jordan domains in which the boundaries of any two in the set intersect
in at most two points, counting a point of tangency as a double intersection. Note that the property is defined by the
entire set. See also Def. 2.4.
    2The intersection graph of a set of sets has a node for each set and two nodes are adjacent if the corresponding
sets intersect (their intersection is not empty).
    3There are several equivalent definitions of fatness in the plane. We say that a geometric object ๐‘‚ is ๐›ผ-fat for some
๐›ผ โ‰ฅ 1, if the ratio of the radius of the smallest enclosing disk of ๐‘‚ to the radius of the largest disk enclosed by ๐‘‚, is at


                          T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                                           2
              A LGORITHMS FOR I NTERSECTION G RAPHS OF ๐‘ก-I NTERVALS AND ๐‘ก-P SEUDODISKS

minimum dominating set. We also refer the reader to [38] for connections between geometric
intersection graphs and approximation algorithms.
    We now formally define the three problems that are central to this paper. Let ๐บ = (๐‘‰ , ๐ธ)
be an undirected graph with non-negative rational weights on the nodes. ๐‘† โІ ๐‘‰ is said to be
a dominating set if for any node ๐‘ข โˆˆ ๐‘‰, ๐‘ข โˆˆ ๐‘† or ๐‘ข has a neighbor in ๐‘†. The Minimum-Weight
Dominating Set (MWDS) problem is to find a dominating set of minimum weight in a given
weighted graph. A subset of nodes, ๐ผ โІ ๐‘‰, is said to be an independent set if no two nodes in ๐ผ are
adjacent in ๐บ. The Maximum-Weight Independent Set (MWIS) problem is to find an independent
set of maximum weight in a given weighted graph. Consider a set system (๐‘‹ , ๐’ฎ), where ๐‘‹ is a
finite ground set, and ๐’ฎ is a set of subsets of ๐‘‹. Each set in ๐’ฎ has a non-negative rational weight.
The Minimum-Weight Set Cover (MWSC) problem is to find a set โ„› โˆ— โІ ๐’ฎ of minimum weight
such that ๐‘‹ = ๐‘†๐‘– โˆˆ๐’ฎ ๐‘† ๐‘– . We denote the unweighted versions (that is, all weights are unit) of the
               ร
MWDS, MWIS, and MWSC problems by MDS, MIS, and MSC, respectively.

Convention 1.1 (rational weights). In all our results, the weights are rational numbers.

Remark 1.2. The focus in this paper is on polynomial-time approximation algorithms for
a class of NP-hard optimization problems. As common in this literature, we use the term
โ€œapproximation algorithmโ€ to refer to a deterministic polynomial-time approximation algorithm unless
we explicitly mention the running time or additional aspects such as the use of randomness or
specific oracles.
    Our assumption that the weights are rational (as opposed to real) is necessitated by the fact
that our algorithms rely on solving linear programming relaxations. The emphasis in this paper
is on the approximation ratio achievable via the LP relaxation approach. For these reasons we
do not emphasize the precise polynomial running times of the algorithms.

1.1    Motivation and our contribution
In this paper we utilize powerful techniques from computational geometry [11, 12, 35] to
provide algorithmic results for ๐‘ก-interval graphs, ๐‘ก-disks and other geometric objects, in a unified
fashion. For some problems we obtain substantially improved approximation bounds that
are near-optimal. Our results extend to ๐‘ก-pseudodisks while techniques in earlier work that
exploited properties of intervals [10], or fatness properties of the underlying objects [26], do not
apply.
    Before stating our results in full generality, we discuss the following geometric covering
problem to illustrate some of the basic ideas. Given a set of points on the line, and a set of
weighted intervals, find a minimum-weight subset of the given intervals that cover all the points.
This is a special case of MWSC, and can be solved efficiently via dynamic programming or via
mathematical programming. The natural LP relaxation in the interval case happens to yield
an integer polytopeโ€”the incidence matrix between intervals and points is totally unimodular
(TUM) (it has the consecutive ones property) [33]. Now consider the same problem where
we need to cover a given set of points by (weighted) ๐‘ก-intervals. Approximation algorithms
most ๐›ผ. We say that โ„› is a set of fat objects, if there exists a constant ๐›ผ such that every object in โ„› is ๐›ผ-fat.


                           T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                                    3
                               C HANDRA C HEKURI AND TANMAY I NAMDAR

for this problem were first given by Hochbaum and Levin [23] (in the more general setting of
multicover)โ€”they derived a ๐‘ก-approximation for this problem by reducing it, via the natural LP
relaxation, to the case of ๐‘ก = 1. As far as we are aware, this was the best known approximation
to this problem until our work. A natural question is whether the approximation ratio of ๐‘ก can
be improved. In this paper we show that an ๐‘‚(log ๐‘ก)-approximation can be obtained via tools
from computational geometry such as shallow-cell complexity and quasi-uniform sampling.
It is an easy observation that an MWSC instance in which each set has at most ๐‘ก elements can
be captured as a special case of covering points by ๐‘ก-intervals. Thus, for large values of ๐‘ก, via
known hardness results for MSC [30, 18], one obtains NP-hardness of (๐‘ log ๐‘ก)-approximation
for some universal constant ๐‘ > 0 for covering points by ๐‘ก-intervals. The geometric machinery
allows us to derive an ๐‘‚(log ๐‘ก)-approximation in the much more general setting of covering
points by ๐‘ก-pseudodisks.
     We state our results for ๐‘ก-pseudodisks which capture several shapes of interest, including
intervals and disks. The geometric approach applies in greater generality but here we confine
our attention to pseudodisks.

   โ€ข An ๐‘‚(log ๐‘ก)-approximation for minimum-weight cover of points by ๐‘ก-pseudodisks, in
     other words, the instances of MWSC defined by points and ๐‘ก-pseudodisks in the plane
     (Theorem 3.9).

   โ€ข An ๐‘‚(๐‘ก log ๐‘ก)-approximation for MWDS in ๐‘ก-pseudodisk graphs (Theorem 3.7). Even
     for ๐‘ก-intervals, the best known previous approximation factor was ๐‘ก 2 [10]. We observe
     that, via a simple reduction from Hypergraph Vertex Cover, it is NP-hard to approximate
     MDS within a factor better than (๐‘ก โˆ’ 1 โˆ’ ๐œ–) for any fixed ๐œ– > 0 in ๐‘ก-interval graphs. We
     also show UG-hardness4 for reaching the slightly improved approximation ratio of (๐‘ก โˆ’ ๐œ–)
     (Corollary 3.13).

   โ€ข An ฮฉ(1/๐‘ก)-approximation for MWIS in ๐‘ก-pseudodisk graphs (Theorem 4.4). A 1/(2๐‘ก)-
     approximation has been known for ๐‘ก-intervals [9]. For more general shapes such as
     disks and fat objects, the approximation bounds in [25] depend on ๐‘ก and on the fatness
     parameters. (See [25] for the actual definition.) Our approach also works for packing
     weighted ๐‘ก-pseudodisks into capacitated points (Theorem 4.5).

    Our results are obtained via simple reductions to existing algorithms in geometric packing
and covering, however, the known algorithms use fairly sophisticated ideas. Consequently, in
some cases, the constants in the approximation guarantees we obtain may be worse compared
to the known results for special cases. On the other hand, our results are applicable to a much
more general class of geometric objects. It may be possible to improve the constants for various
special cases by examining the details more closely.
    We describe the necessary geometric background in the next subsection.

  4 UG refers to Khotโ€™s โ€œUnique Gamesโ€ problem [28, 29].


                       T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                   4
           A LGORITHMS FOR I NTERSECTION G RAPHS OF ๐‘ก-I NTERVALS AND ๐‘ก-P SEUDODISKS

1.2   Background from geometric approximation via LP relaxations
The approximability of MWDS and MWIS in general graphs is well understood and essentially
tight upper and lower bounds are known. For MWDS there is an approximation ratio of
(1 + ln ๐‘›) where ๐‘› is the number of nodes [36, 37], and moreover, it is NP-hard to obtain an
((1 โˆ’ ๐‘œ(1)) ln ๐‘›)-approximation [30, 18] even in the unweighted setting. In general graphs, it is
NP-hard to approximate MIS to within a factor of 1/๐‘› 1โˆ’๐œ– for any fixed ๐œ– > 0 [22, 39], and the
best known approximation ratio is ฮฉ((log๐‘ ๐‘›)/๐‘›) for some small integer constant ๐‘ (see [19, 6]).
MWDS and MWSC are closely related and their approximability is essentially the same.
    The preceding results rule out constant-factor approximations for MWDS and MWIS in
graphs, even in the unweighted case. However, in various geometric settings it is possible
to obtain substantially improved algorithms including approximation schemes (PTASes and
QPTASes) [12, 20, 11, 2, 32, 1] and constant-factor approximations [12, 11, 20]. In this paper we
are interested in LP-based approximations for MWSC and MWIS that have been established
via techniques relying on the union complexity of the underlying geometric objects. Union
complexity measures the worst-case representation size of the union of a given set of objects of
a particular type or shape. In the setting of planar objects, the typical measure is the number of
vertices in the arrangement that appear on the boundary of the union. It is well known that
many geometric objects such as intervals (on a line), disks, and squares (in the plane) have linear
union complexity. In fact, this holds for an even larger class of sets of geometric objects, namely
collections of pseudodisks [3, 27].
    Bounds on union complexity have been used to obtain constant-factor and sublogarithmic
approximations for geometric MWSC and its variants (see [15, 13, 35, 11], and [31] for a survey).
Chan and Har-Peled [12] showed that upper bounds on union complexity can also be used to
obtain improved approximations for the MWIS problem. They give an LP rounding algorithm
with approximation guarantee ฮฉ(๐‘›/๐‘ข(๐‘›)) for computing an MWIS of ๐‘› objects where ๐‘ข(๐‘›) is an
upper bound on the worst-case union complexity of ๐‘› objects under consideration. We use this
result to give an ฮฉ(1/๐‘ก)-approximation for computing MWIS of ๐‘ก-pseudodisks. This implies
ฮฉ(1/๐‘ก)-approximation for MWIS of ๐‘ก-intervals, ๐‘ก-disks and ๐‘ก-squares; however, these results
have already been known [8, 26] but these previous results are based on using certain โ€œfatnessโ€
properties of the underlying geometric objects, and hence do not apply to arbitrary pseudodisks.
Shallow-cell complexity (SCC) of a set system provides quantitative bounds on a certain hereditary
sparsity property [11]. A formal definion is given later in the paper. Upper bounds on SCC
have been used to obtain improved approximations for MWSC and MWDS in geometric settings
as well as some combinatorial settings. The general approach here is to round a feasible LP
solution using a framework called quasi-uniform sampling introduced by Varadarajan [35] for
geometric settings. This framework was refined and improved by Chan et al. [11]. We use this
framework, and some known results on SCC for disks and pseudodisks [20, 5], to derive our
results for MWDS and MWSC on ๐‘ก-pseudodisks.
Organization. Section 2 introduces some relevant notation and definitions. Section 3 describes
algorithms for covering problems MWDS and MWSC. Section 4 describes algorithms for MWIS
and a generalization.

                     T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                       5
                                C HANDRA C HEKURI AND TANMAY I NAMDAR


                D                    C                             D                   C




                    A                                                  A
                                           B                                                     B


Figure 1: Left: ๐’ฎ 0 = {๐ด, ๐ต, ๐ถ, ๐ท} is a set of closed Jordan domains that intersect nicely, and
hence is a collection of pseudodisks. Right: ๐’ฎ = {๐ด โˆช ๐ถ, ๐ต โˆช ๐ท} is a collection of 2-pseudodisks.


2       Preliminaries
Let ๐’ž be a family of sets we call โ€œbasic objects.โ€ Given ๐’ž, a ๐‘ก-object is defined as the union of
at most ๐‘ก basic objects, where ๐‘ก is a positive integer. We do not assume that the ๐‘ก basic objects
that define a specific ๐‘ก-object are pairwise non-intersecting; such an assumption may be helpful
in some settings but can be restrictive in others. Without loss of generality, we assume that
each ๐‘ก-object is a union of exactly ๐‘ก basic objects. We are typically interested in the case when
the basic objects come from a specific class of geometric shapes such as intervals, disks, and
pseudodisks (closed Jordan domains).

Definition 2.1 (closed Jordan domain). A subset of the plane is a closed Jordan domain if it is a
compact subset of โ„2 whose boundary is a simple closed curve.

Definition 2.2 (nice intersection). We say that two closed Jordan domains intersect nicely if their
boundaries are either (i) disjoint or (ii) properly intersect exactly twice or (iii) are tangent exactly
once.

Definition 2.3 (๐‘ก-pseudodisk). For ๐‘ก โ‰ฅ 1, a ๐‘ก-pseudodisk is the union of at most ๐‘ก closed Jordan
domains that pairwise intersect nicely.

Definition 2.4 (collection of pseudodisks and ๐‘ก-pseudodisks). Let ๐’ฎ 0 be a set of closed Jordan
domains in the plane. We call ๐’ฎ 0 a collection of pseudodisks [5] if the boundaries of any two
distinct pseudodisks in ๐’ฎ 0 intersect nicely. Let ๐’ฎ 0 be a collection of pseudodisks, and let ๐’ฎ be a
set of objects where each object in ๐’ฎ is the union of at most ๐‘ก objects from ๐’ฎ 0. We say that ๐’ฎ is a
collection of ๐‘ก-pseudodisks.

Remark 2.5. When speaking of a collection ๐’ฎ of ๐‘ก-pseudodisks, we understand that there is an
underlying collection ๐’ฎ 0 of pseudodisks.

       The definition allows two different ๐‘ก-pseudodisks in ๐’ฎ to share a common pseudodisk from
๐’ฎ 0.
                                          (1)   (2)     (๐‘ก)
       Consider a ๐‘ก-object ๐‘‚ ๐‘– . We use ๐‘œ ๐‘– , ๐‘œ ๐‘– , . . . , ๐‘œ ๐‘– to denote the objects whose union equals
                                                                           (1)   (2)       (๐‘ก)
๐‘‚ ๐‘– . Slightly abusing the notation, we also denote by ๐‘‚ ๐‘– the set {๐‘œ ๐‘– , ๐‘œ ๐‘– , . . . , ๐‘œ ๐‘– }. Next, we
formally define the notion of union complexity.

                         T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                        6
            A LGORITHMS FOR I NTERSECTION G RAPHS OF ๐‘ก-I NTERVALS AND ๐‘ก-P SEUDODISKS




                                    D                            C




                                     A                                    B


Figure 2: The vertices, elementary arcs, and faces of dimension two formed by the collection
๐’ฎ 0 = {๐ด, ๐ต, ๐ถ, ๐ท} of pseudodisks are shown as points (filled squares/circles), arcs, and solid
regions, respectively. For โ„› 0 = {๐ด, ๐ถ, ๐ท} the vertices appearing on ๐’ฐ(โ„› 0) are shown as blue
squares.


2.1   Union complexity

The following definitions are adapted from [3, pp. 13-14]. Let ๐’ฎ 0 be a collection of pseudodisks.
Let ๐’ฐ(๐’ฎ 0) = ๐‘†โˆˆ๐’ฎ0 ๐‘† denote their union and let ๐œ•๐’ฐ(๐’ฎ 0) denote the boundary of the union.
                ร
Each face of ๐œ•๐’ฐ(๐’ฎ 0) is a maximal connected (relatively open) subset of ๐œ•๐’ฐ(๐’ฎ 0) that lies on the
boundaries of a fixed subset โ„› 0 โІ ๐’ฎ 0, and avoids all other pseudodisks in in ๐’ฎ 0. The faces of
dimension 0 and 1 are called vertices and elementary arcs, respectively. We say that ๐’ฎ 0 has union
complexity ๐‘ข(ยท) for a function ๐‘ข : โ„ค+ โ†’ โ„ค+ if the following condition is true: for any subset
โ„› 0 โІ ๐’ฎ 0, the number of vertices on ๐œ•๐’ฐ(โ„› 0) is at most ๐‘ข(|โ„› 0 |).
    In the rest of the paper we assume that ๐‘ข(๐‘›) โ‰ฅ ๐‘› for any ๐‘› โ‰ฅ 1 and that ๐‘ข is non-decreasing.
These are natural assumptions for the settings we consider.
    For example, a set of disksโ€”and more generally, a collection of pseudodisksโ€”in the plane
have linear union complexity [3, 27]. The union complexity of ๐‘› fat triangles is ๐‘‚(๐‘› logโˆ— ๐‘›) [4].
The union complexity of ๐‘› axis-aligned rectangles in the plane can be ฮฉ(๐‘› 2 ).



3     Minimum-Weight Dominating Set and Set Cover
As in Butman et al. [10], we consider the following slight generalization of the MWDS that is
called the Red-Blue Dominating Set Problem. The input is an undirected bipartite intersection
graph ๐บ = (โ„› t โ„ฌ, ๐ธ), where โ„› = {๐‘… 1 , . . . , ๐‘… ๐‘ } and โ„ฌ = {๐ต1 , . . . , ๐ต ๐‘€ } are sets of red and blue
๐‘ก-objects, respectively. There is an edge between ๐‘… ๐‘– โˆˆ โ„› and ๐ต ๐‘— โˆˆ โ„ฌ if ๐‘… ๐‘– โˆฉ ๐ต ๐‘— โ‰  โˆ…. Since
each node corresponds to a (red or blue) ๐‘ก-object, we will use the term โ€˜nodeโ€™ and โ€˜๐‘ก-objectโ€™
interchangeably. Each ๐‘… ๐‘– โˆˆ โ„› has a non-negative weight ๐‘ค ๐‘– . The goal is to find a subset โ„› 0 โІ โ„›
of minimum weight such that every blue node ๐ต ๐‘— has a neighbor in the set โ„› 0. Note that the
standard MWDS problem is equivalent to the setting where โ„› = โ„ฌ.

                      T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                            7
                                            C HANDRA C HEKURI AND TANMAY I NAMDAR

3.1     LP relaxation and rounding
Let โ„ = (โ„›, โ„ฌ) denote the given instance of the (generalized) MWDS. In an integer programming
formulation we have a {0, 1} variable ๐‘ฅ ๐‘– corresponding to a red ๐‘ก-object ๐‘… ๐‘– which is intended
to be assigned 1 if ๐‘… ๐‘– is selected in the solution, and is assigned 0 otherwise. We relax the
integrality constraints and describe the natural LP relaxation below.

                                                            ร•
                                           minimize                  ๐‘ค๐‘– ๐‘ฅ๐‘–
                                                            ๐‘… ๐‘– โˆˆโ„›
                                                             ร•
                                           subject to                      ๐‘ฅ ๐‘– โ‰ฅ 1,            โˆ€๐ต ๐‘— โˆˆ โ„ฌ                                     (3.1)
                                                        ๐‘… ๐‘– :๐ต ๐‘— โˆฉ๐‘… ๐‘– โ‰ โˆ…

                                                                           ๐‘ฅ ๐‘– โˆˆ [0, 1], โˆ€๐‘… ๐‘– โˆˆ โ„›                                           (3.2)

    The LP-relaxation can be solved in polynomial time assuming that we are given the
intersection graph defined by โ„. Let ๐‘ฅ be an optimal LP solution ร•     for the given instance
โ„ = (โ„›, โ„ฌ). Since ๐‘ฅ is feasible, for every blue ๐‘ก-object ๐ต ๐‘— , we have       ๐‘ฅ ๐‘– โ‰ฅ 1. For each
                                                                                                               ๐‘… ๐‘– :๐ต ๐‘— โˆฉ๐‘… ๐‘– โ‰ โˆ…
                                                                                             ร•
๐ต ๐‘— โˆˆ โ„ฌ, let ๐‘ 0๐‘— denote an object maximizing the quantity                                                  ๐‘ฅ ๐‘– , where the ties are broken
                                                                                        ๐‘… ๐‘– :๐‘ 0๐‘— โˆฉ๐‘… ๐‘– โ‰ โˆ…
arbitrarily. Let โ„ฌ 0 B {๐‘ 0๐‘— : ๐ต ๐‘— โˆˆ โ„ฌ}. We emphasize that โ„ฌ 0 is a set of (1-)objects, whereas โ„› is a
                                                                   ร•
set of ๐‘ก-objects. Note that for any ๐‘ 0๐‘— โˆˆ โ„ฌ 0,                                   ๐‘ฅ ๐‘– โ‰ฅ 1/๐‘ก.
                                                              ๐‘… ๐‘– :๐‘ 0๐‘— โˆฉ๐‘… ๐‘– โ‰ โˆ…
             ๐‘… ๐‘– โˆˆ โ„›, let ๐‘ฅ 0๐‘– = min{๐‘ก๐‘ฅ ๐‘– , 1}, and let ๐‘ฅ 0 denote the resulting solution. Then, for any
      For anyร•
๐‘ 0๐‘— โˆˆ โ„ฌ 0,                       ๐‘ฅ 0๐‘– โ‰ฅ 1. The following observation follows from the definition of ๐‘ฅ 0.
              ๐‘… ๐‘– :๐‘ 0๐‘— โˆฉ๐‘… ๐‘– โ‰ โˆ…


Observation 3.1. The solution ๐‘ฅ 0 is feasible for the instance โ„ 0 = (โ„›, โ„ฌ 0), and we have that                                       ๐‘… ๐‘– โˆˆโ„› ๐‘ค ๐‘– ๐‘ฅ ๐‘– โ‰ค
                                                                                                                                                   0
                                                                                                                                  ร
๐‘ก ยท ๐‘… ๐‘– โˆˆโ„› ๐‘ค ๐‘– ๐‘ฅ ๐‘– .
   ร

      The preceding step to reduce to โ„ 0 is essentially the same as in [10].

3.1.1    Shallow-cell complexity
Definition 3.2 (Shallow-cell complexity). Let ๐ด โˆˆ {0, 1} ๐‘€ร—๐‘ denote an ๐‘€ ร— ๐‘ (0, 1)-matrix. Let
1 โ‰ค ๐‘˜ โ‰ค ๐‘› โ‰ค ๐‘. Let ๐‘† be any set of ๐‘› columns and let ๐ด๐‘† be the matrix restricted to the columns
of ๐‘†. If the number of distinct rows in ๐ด๐‘† with at most ๐‘˜ ones is bounded by ๐‘“ (๐‘›, ๐‘˜) for any
choice of ๐‘›, ๐‘˜ and ๐‘†, then ๐ด is said to have the shallow-cell complexity ๐‘“ (๐‘›, ๐‘˜).

    Now, we describe how to round ๐‘ฅ 0 to an integral solution via quasi-uniform sampling
technique. The standard LP relaxation for an MWSC instance (๐‘‹ , ๐’ฎ 0) is as follows.



                                     T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                                                          8
            A LGORITHMS FOR I NTERSECTION G RAPHS OF ๐‘ก-I NTERVALS AND ๐‘ก-P SEUDODISKS


                                               ร•
                                minimize                 ๐‘ค๐‘– ๐‘ฅ๐‘–
                                              ๐‘† ๐‘– โˆˆ๐’ฎ 0
                                              ร•
                                subject to             ๐‘ฅ ๐‘– โ‰ฅ 1,           โˆ€๐‘— โˆˆ ๐‘‹                      (3.3)
                                              ๐‘† ๐‘– 3๐‘—

                                                         ๐‘ฅ ๐‘– โˆˆ [0, 1],   โˆ€๐‘† ๐‘– โˆˆ ๐’ฎ 0                   (3.4)

    Let ๐ด โˆˆ {0, 1} ๐‘€ร—๐‘ denote the constraint matrix in the LP relaxation above. This is the
incidence matrix of the set system {๐‘† ๐‘– | 1 โ‰ค ๐‘– โ‰ค ๐‘ } : each row of ๐ด corresponds to an element,
and each column corresponds to a set. The entry ๐ด ๐‘–๐‘— = 1 if the element ๐‘— is contained in ๐‘† ๐‘– ,
otherwise ๐ด ๐‘–๐‘— = 0.
    The following crucial definition is from [11].

Definition 3.3 (Shallow-cell complexity of MWSC instance). Let ๐ด be the constraint matrix
of the MWSC instance โ„. We define the shallow-cell complexity of โ„ to be the shallow-cell
complexity of ๐ด.

    Using bounds on the shallow-cell complexity of an MWSC instance, it is possible to round a
feasible LP solution using a technique known as quasi-uniform sampling [11, 35].

Theorem 3.4 (Chan et al. [11], Varadarajan [35]). Consider an MWSC instance with shallow-cell
complexity ๐‘“ (๐‘›, ๐‘˜) = ๐‘›๐œ™(๐‘›) ยท ๐‘˜ ๐‘ , where ๐œ™(๐‘›) = ๐‘‚(๐‘›), and ๐‘ โ‰ฅ 0 is a constant. Then, there exists a
polynomial-time algorithm to round a given fractional solution to the LP relaxation for this instance to
an integral solution whose cost is within an ๐‘‚(max{log ๐œ™(๐‘), 1}) factor of the cost of the fractional
solution, where the constant hidden in the big-Oh notation depends on the exponent ๐‘, and ๐‘ is the
number of sets in the given instance.

    Usually, ๐œ™(๐‘›) is a function of ๐‘› such that ๐œ™(๐‘›) = ๐‘‚(๐‘›). However, the same guarantee holds
even when ๐œ™ is independent of ๐‘› . When we apply this theorem, we will set ๐œ™(๐‘›) = ๐‘‚(๐‘ก 4 ),
which is independent of ๐‘›.
    Now, we consider an instance โ„ 0 = (โ„›, โ„ฌ 0) of the MWDS problem obtained in the previous
                                                                                              (๐‘˜)
section. Recall that โ„› is a set of ๐‘ก-objects and โ„ฌ is a set of 1-objects. Now, let โ„› 0 = {๐‘Ÿ ๐‘– โˆˆ ๐‘… ๐‘– :
๐‘… ๐‘– โˆˆ โ„›} denote the set of constituent red 1-objects from โ„›. Let โ„ 00 = (โ„› 0 , โ„ฌ 0) denote the MWDS
instance thus obtained. Notice that an MWDS instance can also be thought of as an MWSC
instance. We first prove the following simple lemma.

Lemma 3.5. Let โ„ 0 , โ„ 00 be MWDS instances as defined above. If the shallow-cell complexity of โ„ 00 is
๐‘“ (๐‘›, ๐‘˜), then the shallow-cell complexity of the corresponding instance โ„ 0 is ๐‘”(๐‘›, ๐‘˜) โ‰ค ๐‘“ (๐‘›๐‘ก, ๐‘˜๐‘ก) for any
1 โ‰ค ๐‘˜ โ‰ค ๐‘› โ‰ค ๐‘.
                                                                                                  0
Proof. We prove this fact from the definition of the shallow-cell complexity. Let ๐ดโ„ denote
the constraint matrix corresponding to the instance โ„ 0. Fix some positive integers ๐‘›, ๐‘˜ such
                                                                                                0
that 1 โ‰ค ๐‘˜ โ‰ค ๐‘› โ‰ค |โ„›|, and fix a set ๐‘† โІ โ„› of columns (i. e., ๐‘ก-objects), where |๐‘†| = ๐‘›. Let ๐ดโ„๐‘†

                       T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                                  9
                             C HANDRA C HEKURI AND TANMAY I NAMDAR

denote the constraint matrix restricted to the columns corresponding to ๐‘†. Let ๐‘ƒ denote the set
                                             0
of distinct rows (i. e., blue objects) in ๐ดโ„๐‘† with at most ๐‘˜ ones. We seek to bound |๐‘ƒ|.
               (๐‘˜)
     Let ๐‘†0 = {๐‘Ÿ ๐‘– โˆˆ ๐‘… ๐‘– : ๐‘… ๐‘– โˆˆ ๐‘†} be the corresponding constituent 1-objects. Note that ๐‘†0 โІ โ„› 0,
                        00
and |๐‘†0 | โ‰ค ๐‘›๐‘ก. Let ๐ดโ„ denote the constraint matrix corresponding to the instance โ„ 00 and let
    00            00                                                                         00
๐ดโ„๐‘†0 denote ๐ดโ„ restricted to the columns of ๐‘†0. Let ๐‘ƒ 0 denote the set of rows in ๐ดโ„๐‘†0 with at
                                                                  0
most ๐‘˜๐‘ก ones. Note that, if a row has at most ๐‘˜ ones in ๐ดโ„๐‘† , then it corresponds to exactly one
row in ๐‘ƒ 0. Therefore, |๐‘ƒ| โ‰ค |๐‘ƒ 0 | โ‰ค ๐‘“ (๐‘›๐‘ก, ๐‘˜๐‘ก), where the last inequality follows from the definition
of the shallow-cell complexity of the instance โ„ 00.                                                  

    Now, we state the known bounds on the shallow-cell complexity of an MWDS instance
defined by red and blue sets of pseudodisks.

Theorem 3.6 (Aronov et al. [5]). The shallow-cell complexity of an MWDS instance defined by a
collection of pseudodisks is at most ๐‘“ (๐‘›, ๐‘˜) = ๐‘‚(๐‘› ๐‘˜ 3 ).

   Combining Theorem 3.4, Theorem 3.6, and Lemma 3.5, we obtain the following result.

Theorem 3.7. There exists a polynomial-time ๐‘‚(๐‘ก log ๐‘ก)-approximation algorithm for the Red-Blue
Dominating Set problem defined by a collection of ๐‘ก-pseudodisks.

Proof. Let โ„ = (โ„›, โ„ฌ) be the original instance and let ๐‘ฅ be an optimal LP solution. Define
instance โ„ 0 = (โ„›, โ„ฌ 0) and the corresponding LP solution ๐‘ฅ 0 as before. From Observation 3.1,
we have that ๐‘… ๐‘– โˆˆโ„› ๐‘ค ๐‘– ๐‘ฅ 0๐‘– โ‰ค ๐‘ก ยท ๐‘… ๐‘– โˆˆโ„› ๐‘ค ๐‘– ๐‘ฅ ๐‘– . By Lemma 3.5 the shallow-cell complexity of โ„ 0 is
              ร                   ร
๐‘”(๐‘›, ๐‘˜) โ‰ค ๐‘“ (๐‘›๐‘ก, ๐‘˜๐‘ก), where ๐‘“ (๐‘›, ๐‘˜) = ๐‘‚(๐‘› ๐‘˜ 3 ) by Theorem 3.6. Therefore, ๐‘”(๐‘›, ๐‘˜) โ‰ค ๐‘‚(๐‘›๐‘ก 4 ๐‘˜ 3 ).
Now, using the algorithm from Theorem 3.4 with ๐œ™(๐‘›) = ๐‘‚(๐‘ก 4 ), we can round ๐‘ฅ 0 to an integral
solution of cost at most ๐‘‚(log ๐‘ก) ยท ๐‘… ๐‘– โˆˆโ„› ๐‘ค ๐‘– ๐‘ฅ 0๐‘– โ‰ค ๐‘‚(๐‘ก log ๐‘ก) ยท ๐‘… ๐‘– โˆˆโ„› ๐‘ค ๐‘– ๐‘ฅ ๐‘– , where the inequality
                                        ร                          ร
follows from Observation 3.1.                                                                          

Remark 3.8. Suppose we have an instance of (โ„›, โ„ฌ) of generalized MWDS where each red object
is a ๐‘ก ๐‘… -pseudodisk and each blue object is a ๐‘ก ๐ต -pseudodisk. Then the preceding analysis can be
extended to obtain an ๐‘‚(๐‘ก ๐ต log ๐‘ก ๐‘… )-approximation. We note that for the analogous version of
intervals, a (๐‘ก ๐ต ยท ๐‘ก ๐‘… )-approximation was obtained in [10].

3.1.2   Geometric MWSC with ๐‘ก-objects.
Let ๐’ฎ 0 denote a set of (1-)objects. Consider a Geometric MWSC instance (๐‘‹ , ๐’ฎ), where ๐‘‹ is
a set of points, and ๐’ฎ is a set of ๐‘ sets, each of which is obtained by taking the union of at
most ๐‘ก (1-)objects from ๐’ฎ 0. Each set in ๐’ฎ has a non-negative weight. The goal is to find a
minimum-weight set โ„› โІ ๐’ฎ that covers the set of points ๐‘‹. Furthermore, suppose the set system
(๐‘‹ , ๐’ฎ 0) has shallow-cell complexity ๐‘“ (๐‘›, ๐‘˜) = ๐‘› ๐‘˜ ๐‘ , for some fixed constant ๐‘. This includes
geometric objects such as intervals on a line and (pseudo-)disks in the plane. Then, using similar
arguments as we did for MWDS, one can obtain a bound of ๐‘‚(๐‘› ๐‘˜ ๐‘ ๐‘ก ๐‘+1 ) on the shallow-cell
complexity of the MWSC instance (๐‘‹ , ๐’ฎ). Then Theorem 3.4 implies an ๐‘‚(log ๐‘ก)-approximation
for the MWSC instance (๐‘‹ , ๐’ฎ).

                      T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                          10
            A LGORITHMS FOR I NTERSECTION G RAPHS OF ๐‘ก-I NTERVALS AND ๐‘ก-P SEUDODISKS

   When ๐’ฎ 0 is a set of fat triangles, the shallow-cell complexity of the system (๐‘‹ , ๐’ฎ 0) can
be bounded by ๐‘“ (๐‘›, ๐‘˜) = ๐‘› logโˆ— ๐‘› ยท ๐‘˜ ๐‘ , for some fixed constant ๐‘ [4, 11]. We can then bound
the shallow-cell complexity of the MWSC instance (๐‘‹ , ๐’ฎ) by ๐‘“ (๐‘›๐‘ก, ๐‘˜๐‘ก) = ๐‘› logโˆ— (๐‘›๐‘ก) ยท ๐‘ก ๐‘+1 ๐‘˜ ๐‘ .
Then, Theorem 3.4 implies an ๐‘‚(log(๐‘ก ๐‘+1 ยท logโˆ— (๐‘๐‘ก))) = ๐‘‚(log ๐‘ก + log logโˆ— (๐‘))-approximation.
Therefore, we obtain the following result.

Theorem 3.9. There is a polynomial-time ๐‘‚(log ๐‘ก)-approximation algorithm for MWSC instances
defined by covering points by ๐‘ก-pseudodisks. For covering points by ๐‘ก-fat-triangles, there is an ๐‘‚(log ๐‘ก +
log logโˆ— (๐‘))-approximation where ๐‘ is the number of triangles in the instance.

    The preceding result improves the ๐‘ก-approximation for covering points by ๐‘ก-intervals [23].
The approximation ratio of ๐‘‚(log ๐‘ก) is tight up to constant factors; it is NP-hard to obtain an
๐‘œ(log ๐‘ก)-approximation for covering points by ๐‘ก-intervals. This follows via a reduction from
MWSC [10].

3.2   Integrality of MWDS LP for intervals
Let โ„ = (โ„›, โ„ฌ) be an MWDS instance where โ„› and โ„ฌ are set of intervals on a line. In this
subsection, we prove that the MWDS LP for intervals is integral. Butman et al. [10] proved this
fact using a primal-dual algorithm that constructs an integral solution of the same cost as the LP.
Here we give a simpler proof via the structure of the constraint matrix.
    First, we preprocess blue intervals such that no blue interval is completely contained inside
another blue interval. To understand this, suppose ๐ต1 , ๐ต2 โˆˆ โ„ฌ are two intervals such that
๐ต1 โŠ‚ ๐ต2 . Then, any feasible integral solution must include a red interval ๐‘… that intersects ๐ต1 ,
and thus ๐ต2 . Furthermore, the LP constraint corresponding to ๐ต2 is implied by the constraint
corresponding to ๐ต1 . Therefore, the feasible regions of the LPโ€™s corresponding to โ„ and the
instance obtained by removing ๐ต2 are the same. This justifies removing ๐ต2 from โ„. Repeating
this removal step as along as one blue interval is contained in another blue interval, we obtain
an instance โ„ 0.
    In the following theorem we show that the constraint matrix of the LP corresponding to โ„ 0
satisfies the consecutive ones property, and thus it is totally unimodular. This implies that the
LP corresponding to โ„ 0 (and therefore โ„) is integral (see, e.g., [33]).

Theorem 3.10. Let โ„ = (โ„›, โ„ฌ) be an MWDS instance, where โ„› and โ„ฌ are sets of (1-)intervals. Then,
the MWDS LP is integral.

Proof. Let us denote the left (resp. right) endpoint of an interval ๐ผ by Left(๐ผ) (resp. Right(๐ผ)). Let
๐ด denote the constraint matrix corresponding to the preprocessed instance, where the rows
(i. e., blue intervals) are sorted in the non-decreasing order of their left endpoints. We show that
๐ด has the consecutive ones property for any column.
     Consider any column (i. e., a red interval) ๐‘…. Let ๐ต ๐‘— and ๐ต ๐‘˜ denote the first (leftmost) and
last (rightmost) blue intervals intersecting ๐‘… in this order respectively. Note that ๐‘— โ‰ค ๐‘˜. Now,
suppose for contradiction that there is an interval ๐ตโ„“ with ๐‘— < โ„“ < ๐‘˜ that does not intersect ๐‘….
Let us consider two cases.

                      T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                            11
                              C HANDRA C HEKURI AND TANMAY I NAMDAR

    If Right(๐ตโ„“ ) < Left(๐‘…), then we have that Left(๐ต ๐‘— ) < Left(๐ตโ„“ ) โ‰ค Right(๐ตโ„“ ) < Left(๐‘…) โ‰ค Right(๐ต ๐‘— ).
Here, the third and fourth inequalities follow from the assumptions that ๐‘… โˆฉ ๐ตโ„“ = โˆ… and
๐‘… โˆฉ ๐ต ๐‘— โ‰  โˆ… respectively. However, this implies that ๐ตโ„“ โŠ‚ ๐ต ๐‘— , which is a contradiction.
    On the other hand, if Right(๐ตโ„“ ) > Left(๐‘…), then it must be the case that Left(๐ตโ„“ ) > Right(๐‘…);
otherwise ๐‘… โˆฉ ๐ตโ„“ โ‰  โˆ…, which is a contradiction. However, this implies that Left(๐‘…) โ‰ค Right(๐‘…) <
Left(๐ตโ„“ ) โ‰ค Left(๐ต ๐‘˜ ) โ‰ค Right(๐ต ๐‘˜ ), where the third inequality follows from the fact that โ„“ < ๐‘˜.
However, this contradicts the assumption that ๐ต ๐‘˜ โˆฉ ๐‘… โ‰  โˆ….
    Thus, the column corresponding to ๐‘… satisfies the consecutive ones property.                       

3.3   Hardness results for MWDS
We show hardness of approximation for MWDS in ๐‘ก-interval graphs via a simple reduction. We
first consider the Red-Blue Dominating Set problem for ๐‘ก-intervals which illustrates the basic
idea. We then modify it slightly to prove hardness for the MWDS problem.
     Let (๐‘‹ , ๐’ฎ) be an ๐‘“ -uniform instance of MWSC. That is, any element in ๐‘‹ is contained in
exactly ๐‘“ sets of ๐’ฎ. We reduce this to the Red-Blue Dominating Set problem as follows. For every
set ๐‘† ๐‘– โˆˆ ๐’ฎ, we add a red interval ๐‘… ๐‘– = [๐‘ฆ ๐‘– , ๐‘ง ๐‘– ] on the line. The red intervals are non-overlapping;
concretely, it suffices to choose ๐‘ฆ ๐‘– = 2๐‘– and ๐‘ง ๐‘– = 2๐‘– + 1 for each ๐‘–. We set the weight of ๐‘… ๐‘– to be
that of ๐‘† ๐‘– . For any element ๐‘’ ๐‘— โˆˆ ๐‘‹, we add a exactly ๐‘“ blue points coinciding with points ๐‘ง ๐‘– ,
where ๐‘’ ๐‘— โˆˆ ๐‘† ๐‘– . Note that a blue ๐‘“ -point is covered iff we select a red interval that covers one of
its constituent ๐‘“ points. Thus, there is a one-to-one correspondence between feasible solutions
to the original MWSC instance, and feasible solutions to the Red-Blue dominating set instance
that is created. The following hardness results are known for the ๐‘“ -uniform MWSC problem
even when sets have unit weights.

Theorem 3.11.

   1. It is NP-hard to obtain a ( ๐‘“ โˆ’ 1 โˆ’ ๐œ–)-approximation for ๐‘“ -uniform Set Cover, for any fixed ๐‘“ โ‰ฅ 3
      and ๐œ– > 0 (Dinur et al. [16]).

   2. Furthermore, it is UG-hard to obtain an ( ๐‘“ โˆ’ ๐œ–)-approximation for any fixed ๐‘“ โ‰ฅ 2 and ๐œ– > 0
      (Khot and Regev [29]).

   Therefore, from the preceding reduction, we get the following hardness results for the
Red-Blue Dominating Set problem, even when all weights are unit.

Theorem 3.12. It is NP-hard to obtain a (๐‘ก โˆ’ 1 โˆ’ ๐œ–)-approximation for the special case of Red-Blue
Dominating Set problem where โ„› is a set of intervals and โ„ฌ is a set of ๐‘ก-points. Moreover, in the same
setting, it is UG-hard to obtain an (๐‘ก โˆ’ ๐œ–)-approximation, for any ๐‘ก โ‰ฅ 2 and ๐œ– > 0.

Corollary 3.13. It is NP-hard to obtain a (๐‘ก โˆ’ 1 โˆ’ ๐œ–)-approximation for MWDS in ๐‘ก-interval graphs.
Moreover, in the same setting, it is UG-hard to obtain an (๐‘ก โˆ’ ๐œ–)-approximation, for any ๐‘ก โ‰ฅ 2 and ๐œ– > 0.

Proof. We slightly modify the preceding reduction from ๐‘ก-uniform MWSC to the Red-Blue
Dominating Set problem. For each red interval ๐‘… ๐‘– we also add a green point ๐บ ๐‘– that coincides

                      T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                             12
            A LGORITHMS FOR I NTERSECTION G RAPHS OF ๐‘ก-I NTERVALS AND ๐‘ก-P SEUDODISKS

with ๐‘ฆ ๐‘– , the left end point of ๐‘… ๐‘– , and we set its weight to 0. We set the weight of each blue ๐‘ก-point
to be twice the total weight of all the sets in the MWSC instance (conceptually the weight is โˆž).
The MWDS instance now consits of the red intervals, the blue ๐‘ก-points, and the green points.
The green points have zero weight and hence picking all of them will ensure that all green
points and red intervals are dominated. The green points cannot cover any of the blue points.
Since the blue ๐‘ก-points have very large weight, it is cheaper to pick all the red intervals than pick
any of them. Thus we can restrict attention to dominating sets that include all the green points
and a subset of the red intervals which together cover the blue ๐‘ก-points. Thus, ๐‘ก-uniform MWSC
reduces in an approximation preserving fashion to MWDS in ๐‘ก-interval graphs.                           


4    Maximum-Weight Independent Set
We consider MWIS of ๐‘ก-objects. Here, we are given a set ๐’ฎ = {๐‘†1 , . . . , ๐‘†๐‘› }, where each ๐‘† ๐‘– โˆˆ ๐’ฎ is
a ๐‘ก-object, and has a non-negative weight ๐‘ค ๐‘– . In this section, we assume that we are given the
geometric representation of the ๐‘ก-objects in ๐’ฎ. The goal is to find maximum-weight independent
set ๐’ฎ 0 โІ ๐’ฎ in the intersection graph of ๐’ฎ.
    We confine attention to the case when the base objects are closed Jordan domains. Let ๐’ฑ(๐’ฎ)
denote the set of vertices on ๐œ•๐’ฐ(๐’ฎ). First, we describe a natural LP relaxation for this problem
(see for instance [12]).

                                            ร•
                              maximize               ๐‘ค๐‘– ๐‘ฅ๐‘–
                                            ๐‘† ๐‘– โˆˆ๐’ฎ
                                            ร•
                              subject to             ๐‘ฅ ๐‘– โ‰ค 1,         โˆ€๐‘ โˆˆ ๐’ฑ(๐’ฎ)                  (4.1)
                                            ๐‘† ๐‘– 3๐‘

                                                     ๐‘ฅ ๐‘– โˆˆ [0, 1],         โˆ€๐‘† ๐‘– โˆˆ ๐’ฎ              (4.2)

    We will assume that the preceding LP relaxation can be solved in polynomial time for
instances of interest. This is easy to see for geometric settings in which the vertices of the
arrangement ๐’ฎ can be explicitly computed in polynomial time.
    We have a simple observation that relates the union complexities of ๐‘ก-objects and the
corresponding (1-)objects.

Observation 4.1. Let ๐’ฎ be a set of ๐‘ก-objects, and suppose the set of underlying (1-)objects has union
complexity ๐‘ข(ยท). Then, the union complexity of any ๐‘˜ objects in ๐’ฎ is at most ๐‘ข(๐‘˜๐‘ก).
                                                                     (๐‘˜)
Proof. Let โ„› โІ ๐’ฎ be any subset of ๐‘ก-objects. Let โ„› 0 = {๐‘Ÿ ๐‘– โˆˆ ๐‘… ๐‘– : ๐‘… ๐‘– โˆˆ ๐‘…} denote the underlying
set of (1-)objects. Note that |๐‘…0 | โ‰ค ๐‘ก ยท |๐‘…|. Since the underlying set of (1-)objects has union
complexity ๐‘ข(ยท), the number of arcs on the boundary of ๐‘…0 is at most ๐‘ข(|๐‘…0 |) โ‰ค ๐‘ข(๐‘ก ยท |๐‘…|).      

   Chan and Har-Peled [12] describe an algorithm to round a feasible fractional solution to the
LP relaxation which leads to the following guarantee.

                      T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                              13
                              C HANDRA C HEKURI AND TANMAY I NAMDAR

Theorem 4.2 (Chan and Har-Peled [12]). There is a polynomial-time ฮฉ(๐‘›/๐‘ข(๐‘›))-approximation
algorithm for MWIS in the intersection graph of a set of ๐‘› closed Jordan domains with union complexity
๐‘ข(ยท).

    In particular, this implies an ฮฉ(1)-approximation for MWIS in the intersection graph of
pseudodisks, since pseudodisks have linear union complexity [3, 27]. However, we cannot use
the theorem directly, since a ๐‘ก-pseudodisk is not necessarily a closed Jordan domain. Although
the result in [12] easily extends to our setting via Observation 4.1, for the sake of completeness,
we briefly sketch the main reason. The following following technical lemma from [12] is the
main ingredient in the proof of Theorem 4.2.

Lemma 4.3 (Lemma 4.1 from [12]). Let ๐’ฎ be a set of closed Jordan domains, and let {๐‘ฅ ๐‘– : ๐‘† ๐‘– โˆˆ ๐’ฎ} be a
feasible solution to the MWIS LP relaxation. Furthermore, let โ„› โІ ๐’ฎ be any subset. Then, there is a
universal constant ๐‘ such that (๐‘,๐‘–,๐‘—)โˆˆ๐’ฑ(โ„›) ๐‘ฅ ๐‘– ๐‘ฅ ๐‘— โ‰ค ๐‘ ยท ๐‘ข(โ„ฐ(โ„›)), where โ„ฐ(โ„›) = ๐‘†๐‘– โˆˆโ„› ๐‘ฅ ๐‘– , and (๐‘, ๐‘–, ๐‘—)
                              ร                                                ร
denotes a vertex ๐‘ formed by the intersection of closed Jordan domains ๐‘† ๐‘– and ๐‘† ๐‘— in the arrangement
๐’ฑ(โ„›).

    Lemma 4.3 is the only place in the proof of Theorem 4.2 which uses the assumption that ๐’ฎ
is a set of closed Jordan domains. In other words, a bound on (๐‘,๐‘–,๐‘—)โˆˆ๐’ฑ(โ„›) ๐‘ฅ ๐‘– ๐‘ฅ ๐‘— in terms of the
                                                                 ร
union complexity, as in the statement of the lemma, readily implies an ฮฉ(๐‘›/๐‘ข(๐‘›))-approximation
for the respective set of objects. We need the generalization of Lemma 4.3 which works when
๐’ฎ is a collection of ๐‘ก-pseudodisks. The proof of the lemma is based on a simple yet clever
random sampling argument and does not rely on the objects being closed Jordan domains.
Thus, the lemma holds when ๐’ฎ is a collection of ๐‘ก-pseudodisks; indeed the lemma holds for
๐‘ก-objects where each object is a closed Jordan domain. We can therefore infer an ฮฉ(๐‘›/๐‘ข(๐‘›))
approximation for MWIS in the intersection graph of ๐‘› ๐‘ก-pseudodisks, where ๐‘ข(ยท) denotes the
union complexity of ๐‘ก-pseudodisks.
    Finally, recall that pseudodisks have linear union complexity [3, 27]. Therefore, Observa-
tion 4.1, implies that a collection of ๐‘› ๐‘ก-pseudodisks have ๐‘‚(๐‘›๐‘ก) union complexity. Therefore,
we get the following theorem.

Theorem 4.4. There is a polynomial-time ฮฉ(1/๐‘ก)-approximation algorithm for MWIS in the intersection
graph of ๐‘ก-pseudodisks.

Packing ๐‘ก-objects. We consider a related problem called Maximum-Weight Region Packing. We
are given a set ๐’ฎ of ๐‘ก-objects, and a set of points ๐‘ƒ. Each ๐‘ก-object ๐‘† ๐‘– has a weight ๐‘ค ๐‘– and each
point ๐‘ โˆˆ ๐‘ƒ has a capacity ๐‘(๐‘), which is a positive integer. The goal is to find a maximum-weight
set ๐’ฎ 0 of ๐‘ก-objects, such that for each point ๐‘ โˆˆ ๐‘ƒ, the number of regions of ๐’ฎ 0 that contain
๐‘ is at most ๐‘(๐‘). Note that MWIS is a special case when ๐‘ƒ = ๐’ฑ(๐‘†) is the set of all points in
arrangement and ๐‘(๐‘) = 1 for each ๐‘ โˆˆ ๐‘ƒ.
    Extending the LP rounding algorithm of Chan and Har-Peled [12] for the MWIS problem, Ene
                         ๐‘› 1/๐ถ
et al. [17] gave an ฮฉ(( ๐‘ข(๐‘›) ) )-approximation for Maximum-Weight Region Packing problem,
where ๐ถ is the minimum capacity of any point. Using similar arguments as in MWIS, we obtain
the following result.

                      T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                           14
           A LGORITHMS FOR I NTERSECTION G RAPHS OF ๐‘ก-I NTERVALS AND ๐‘ก-P SEUDODISKS

Theorem 4.5. There is a polynomial-time ฮฉ(1/๐‘ก 1/๐ถ )-approximation algorithm for Maximum-Weight
Region Packing with ๐‘ก-pseudodisks.


5   Concluding remarks
In geometric settings, quasi-uniform sampling has been used for Set Multicover [7], and other
related ideas have been used for Partial Set Cover [24, 14]. Our results for MWDS and MWSC
for ๐‘ก-objects can be extended to these settings. We briefly sketch the ideas.
    The results in [24, 14] establish a generic high-level reduction: an ๐›ผ-approximation via the
natural LP relaxation for MWSC can be translated into an ๐‘‚(๐›ผ)-approximation for Partial Set
Cover for hereditary instances. The geometric instances considered here are hereditary, and
hence our results for MWSC extend to Partial Set Cover.
    Set Multicover is a generalization of Set Cover. In the geometric setting of interest here,
each point ๐‘ has integer demand ๐‘‘ ๐‘ โ‰ฅ 1, and the goal is to choose a minimum-weight subset
of objects so that each point is contained in at least ๐‘‘ ๐‘ objects. Bansal and Pruhs [7] extended
the shallow-cell complexity based framework from Set Cover to Set Multicover. Since we prove
our results by establishing shallow-cell complexity for ๐‘ก-objects, the results in [7] extend to Set
Multicover with ๐‘ก-pseudodisks.
    Finally, we note that even for the special case of MWDS of ๐‘ก-intervals, there is a gap between
the upper and lower bounds on polynomial-time approximation ratios that we established in
this paper: ๐‘‚(๐‘ก log ๐‘ก) and ฮฉ(๐‘ก). Resolving this gap is an interesting open question. Obtaining
smaller leading constants in the approximation ratios, and avoiding the somewhat complicated
machinery of union complexity, for special cases such as ๐‘ก-intervals, is also of interest.


Acknowledgements: We thank Laci Babai for several suggestions and comments that improved
the clarity of the paper.


References
 [1] Anna Adamaszek, Sariel Har-Peled, and Andreas Wiese: Approximation schemes
     for independent set and sparse subsets of polygons. J. ACM, 66(4):29:1โ€“40, 2019.
     [doi:10.1145/3326122] 5

 [2] Anna Adamaszek and Andreas Wiese: Approximation schemes for maximum weight
     independent set of rectangles. In Proc. 54th FOCS, pp. 400โ€“409. IEEE Comp. Soc., 2013.
     [doi:10.1109/FOCS.2013.50] 5

 [3] Pankaj K. Agarwal, Jรกnos Pach, and Micha Sharir: State of the union (of geometric
     objects): A review. In Jacob E. Goodman, Jรกnos Pach, and Richard Pollack, editors,
     Surveys in Discrete and Computational Geometry: Twenty Years Later, pp. 9โ€“48. Amer. Math.
     Soc., 2007. Volume DOI. 5, 7, 14

                    T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                       15
                           C HANDRA C HEKURI AND TANMAY I NAMDAR

 [4] Boris Aronov, Mark de Berg, Esther Ezra, and Micha Sharir: Improved bounds for
     the union of locally fat objects in the plane. SIAM J. Comput., 43(2):543โ€“572, 2014.
     [doi:10.1137/120891241] 7, 11

 [5] Boris Aronov, Anirudh Donakonda, Esther Ezra, and Rom Pinchasi: On pseudo-disk
     hypergraphs. Comput. Geom., 92:101687, 2021. [doi:10.1016/j.comgeo.2020.101687] 5, 6, 10

 [6] Nikhil Bansal, Anupam Gupta, and Guru Guruganesh: On the Lovรกsz theta func-
     tion for independent sets in sparse graphs. SIAM J. Comput., 47(3):1039โ€“1055, 2018.
     [doi:10.1137/15M1051002] 5

 [7] Nikhil Bansal and Kirk Pruhs: Weighted geometric set multi-cover via quasi-uniform
     sampling. J. Comput. Geom., 7(1):221โ€“236, 2016. [doi:10.20382/jocg.v7i1a11] 15

 [8] Reuven Bar-Yehuda, Magnรบs M. Halldรณrsson, Joseph Naor, Hadas Shachnai,
     and Irina Shapira: Scheduling split intervals. SIAM J. Comput., 36(1):1โ€“15, 2006.
     [doi:10.1137/S0097539703437843] 2, 5

 [9] Reuven Bar-Yehuda and Dror Rawitz: Using fractional primal-dual to sched-
     ule split intervals with demands.       Discr. Optimization, 3(4):275โ€“287, 2006.
     [doi:10.1016/j.disopt.2006.05.010] 2, 4

[10] Ayelet Butman, Danny Hermelin, Moshe Lewenstein, and Dror Rawitz: Optimiza-
     tion problems in multiple-interval graphs. ACM Trans. Algorithms, 6(2):40:1โ€“18, 2010.
     [doi:10.1145/1721837.1721856] 2, 3, 4, 7, 8, 10, 11

[11] Timothy M. Chan, Elyot Grant, Jochen Kรถnemann, and Malcolm Sharpe: Weighted
     capacitated, priority, and geometric set cover via improved quasi-uniform sampling. In
     Proc. 23rd Ann. ACMโ€“SIAM Symp. on Discrete Algorithms (SODAโ€™12), pp. 1576โ€“1585. SIAM,
     2012. [doi:10.1137/1.9781611973099.125] 3, 5, 9, 11

[12] Timothy M. Chan and Sariel Har-Peled: Approximation algorithms for Maximum Inde-
     pendent Set of pseudo-disks. Discr. Comput. Geom., 48(2):373โ€“392, 2012. [doi:10.1007/s00454-
     012-9417-5] 3, 5, 13, 14

[13] Chandra Chekuri, Kenneth L. Clarkson, and Sariel Har-Peled: On the set mul-
     ticover problem in geometric settings. ACM Trans. Algorithms, 9(1):9:1โ€“17, 2012.
     [doi:10.1145/2390176.2390185] 5

[14] Chandra Chekuri, Kent Quanrud, and Zhao Zhang: On approximating partial set cover
     and generalizations, 2019. [arXiv:1907.04413] 15

[15] Kenneth L. Clarkson and Kasturi R. Varadarajan: Improved approximation algorithms
     for geometric set cover. Discr. Comput. Geom., 37(1):43โ€“58, 2007. [doi:10.1007/s00454-006-
     1273-8] 5

                    T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                     16
           A LGORITHMS FOR I NTERSECTION G RAPHS OF ๐‘ก-I NTERVALS AND ๐‘ก-P SEUDODISKS

[16] Irit Dinur, Venkatesan Guruswami, Subhash Khot, and Oded Regev: A new multilayered
     PCP and the hardness of hypergraph vertex cover. SIAM J. Comput., 34(5):1129โ€“1146, 2005.
     [doi:10.1137/S0097539704443057] 12

[17] Alina Ene, Sariel Har-Peled, and Benjamin Raichel: Geometric packing under nonuniform
     constraints. SIAM J. Comput., 46(6):1745โ€“1784, 2017. [doi:10.1137/120898413] 14

[18] Uriel Feige: A threshold of ln ๐‘› for approximating set cover. J. ACM, 45(4):634โ€“652, 1998.
     [doi:10.1145/285055.285059] 4, 5

[19] Uriel Feige: Approximating maximum clique by removing subgraphs. SIAM J. Discr. Math.,
     18(2):219โ€“225, 2004. [doi:10.1137/S089548010240415X] 5

[20] Matt Gibson and Imran A. Pirwani: Algorithms for dominating set in disk graphs: Breaking
     the log ๐‘› barrier. In Proc. 18th Eur. Symp. Algorithms (ESAโ€™10), pp. 243โ€“254. Springer, 2010.
     [doi:10.1007/978-3-642-15775-2_21] 5

[21] Jerrold R. Griggs and Douglas B. West: Extremal values of the interval number of a graph.
     SIAM J. Algebr. Discr. Methods, 1(1):1โ€“7, 1980. [doi:10.1137/0601001] 2

[22] Johan Hรฅstad: Clique is hard to approximate within ๐‘› 1โˆ’๐œ€ . Acta Math., 182(1):105โ€“142, 1999.
     [doi:10.1007/BF02392825] 5

[23] Dorit S. Hochbaum and Asaf Levin: Cyclical scheduling and multi-shift scheduling:
     Complexity and approximation algorithms. Discr. Optimization, 3(4):327โ€“340, 2006.
     [doi:10.1016/j.disopt.2006.02.002] 4, 11

[24] Tanmay Inamdar and Kasturi R. Varadarajan: On partial covering for geometric set
     systems. In Proc. 34th Internat. Symp. Comput. Geom. (SoCGโ€™18), pp. 47:1โ€“14. Schloss
     Dagstuhlโ€“Leibniz-Zentrum fuer Informatik, 2018. [doi:10.4230/LIPIcs.SoCG.2018.47] 15

[25] Frank Kammer and Torsten Tholey: Approximation algorithms for intersection graphs.
     Algorithmica, 68(2):312โ€“336, 2014. Preliminary version in [26]. [doi:10.1007/s00453-012-
     9671-1] 2, 4

[26] Frank Kammer, Torsten Tholey, and Heiko Voepel: Approximation algorithms for
     intersection graphs. In Proc. 13th Internat. Workshop on Approximation Algorithms for Combinat.
     Opt. Probl. (APPROXโ€™10), pp. 260โ€“273. Springer, 2010. [doi:10.1007/978-3-642-15369-3_20]
     2, 3, 5, 17

[27] Klara Kedem, Ron Livne, Jรกnos Pach, and Micha Sharir: On the union of Jordan regions
     and collision-free translational motion amidst polygonal obstacles. Discr. Comput. Geom.,
     1:59โ€“70, 1986. [doi:10.1007/BF02187683] 5, 7, 14

[28] Subhash Khot: On the power of unique 2-prover 1-round games. In Proc. 34th STOC, pp.
     767โ€“775. ACM Press, 2002. [doi:10.1145/509907.510017] 4

                     T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                       17
                            C HANDRA C HEKURI AND TANMAY I NAMDAR

[29] Subhash Khot and Oded Regev: Vertex cover might be hard to approximate to within 2 โˆ’ ๐œ–.
     J. Comput. System Sci., 74(3):335โ€“349, 2008. [doi:10.1016/j.jcss.2007.06.019] 4, 12

[30] Dana Moshkovitz:        The Projection Games Conjecture and the NP-hardness
     of ln ๐‘›-approximating set-cover.      Theory of Computing, 11(7):221โ€“235, 2015.
     [doi:10.4086/toc.2015.v011a007] 4, 5

[31] Nabil H. Mustafa and Kasturi R. Varadarajan: Epsilon-approximations and epsilon-nets.
     In Jacob E. Goodman, Joseph Oโ€™Rourke, and Csaba D. Tรณth, editors, Handbook of Discrete
     and Computational Geometry. Chapman and Hall/CRC, 3rd edition, 2017. Taylorโ€“Francis.
     [arXiv:1702.03676] 5

[32] Aniket Basu Roy, Sathish Govindarajan, Rajiv Raman, and Saurabh Ray: Packing
     and covering with non-piercing regions. Discr. Comput. Geom., 60(2):471โ€“492, 2018.
     [doi:10.1007/s00454-018-9983-2] 5

[33] Alexander Schrijver: Combinatorial Optimization: Polyhedra and Efficiency. 2003. Springer. 3,
     11

[34] William P. Thurston: The Geometry and Topology of Three-Manifolds. 1979. Princeton U.
     Press. 2

[35] Kasturi R. Varadarajan: Weighted geometric set cover via quasi-uniform sampling. In
     Proc. 42nd STOC, pp. 641โ€“648. ACM Press, 2010. [doi:10.1145/1806689.1806777] 3, 5, 9

[36] Vijay V. Vazirani: Approximation Algorithms. Springer, 2013. 5

[37] David P. Williamson and David B. Shmoys: The Design of Approximation Algorithms. Cam-
     bridge Univ. Press, 2011. LINK. 5

[38] Yuli Ye and Allan Borodin: Elimination graphs. ACM Trans. Algorithms, 8(2):14:1โ€“23, 2012.
     [doi:10.1145/2151171.2151177] 3

[39] David Zuckerman: Linear degree extractors and the inapproximability of Max Clique
     and chromatic number. Theory of Computing, 3(6):103โ€“128, 2007. Preliminary version in
     STOCโ€™06. [doi:10.4086/toc.2007.v003a006] 5




                    T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                      18
             A LGORITHMS FOR I NTERSECTION G RAPHS OF ๐‘ก-I NTERVALS AND ๐‘ก-P SEUDODISKS

AUTHORS

       Chandra Chekuri
       Professor
       Department of Computer Science
       University of Illinois at Urbana-Champaign
       Urbana, Illinois, United States
       chekuri illinois edu
       https://chekuri.cs.illinois.edu/


       Tanmay Inamdar5
       Researcher
       Department of Informatics,
       University of Bergen
       Bergen, Norway
       Tanmay Inamdar uib no


ABOUT THE AUTHORS

       Chandra Chekuri received his B. Tech in Computer Science and Engineering from
         Indian Institute of Technology, Madras (now Chennai) in 1993 and received
         his Ph. D. in Computer Science from Stanford University in 1998 under the
         supervision of the late Rajeev Motwani. After finishing his Ph. D. he worked
         eight formative years at Lucent Bell Labs before moving to the University of
         Illinois. He mainly works in algorithm design, combinatorial optimization,
         graphs, and mathematical programming.


       Tanmay Inamdar received his Ph. D. in Computer Science from The University of
          Iowa in 2020, where he was advised by Kasturi Varadarajan. Currently he is
         working as a Researcher at The University of Bergen in Norway. His research
          interests include Approximation Algorithms, Computational Geometry, and
          Parameterized Algorithms.




    5This work was done when the author was a Ph. D. student at The University of Iowa, during a visit to University
of Illinois, Urbana-Champaign.


                        T HEORY OF C OMPUTING, Volume 18 (18), 2022, pp. 1โ€“19                                    19