lemmatheorem \aliascntresetthelemma \newaliascntpropositiontheorem \aliascntresettheproposition \newaliascntobservationtheorem \aliascntresettheobservation \newaliascntcorollarytheorem \aliascntresetthecorollary
Finding irrelevant vertices in linear time on bounded-genus graphs††thanks: The first author was supported by the Research Council of Norway via the project BWCA (grant no. 314528). The two last authors were supported by the French-German Collaboration ANR/DFG Project UTMA (ANR-20-CE92-0027). The third author was also supported by the project BOBR that is funded from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme with grant agreement No. 948057. The first and the last author where also supported by the Franco-Norwegian project PHC AURORA 2024. ,††thanks: Emails: [email protected], [email protected], [email protected], [email protected].
Abstract
The irrelevant vertex technique provides a powerful tool for the design of parameterized algorithms for a wide variety of problems on graphs. A common characteristic of these problems, permitting the application of this technique on surface-embedded graphs, is the fact that every graph of large enough treewidth contains a vertex that is irrelevant, in the sense that its removal yields an equivalent instance of the problem. The straightforward application of this technique yields algorithms with running time that is quadratic in the size of the input graph. This running time is due to the fact that it takes linear time to detect one irrelevant vertex and the total number of irrelevant vertices to be detected is linear as well. Using advanced techniques, sub-quadratic algorithms have been designed for particular problems, even in general graphs. However, designing a general framework for linear-time algorithms has been open, even for the bounded-genus case.
In this paper we introduce a general framework that enables finding in linear time an entire set of irrelevant vertices whose removal yields a bounded-treewidth graph, provided that the input graph has bounded genus. Our technique consists in decomposing any surface-embeddable graph into a tree-structured collection of bounded-treewidth subgraphs where detecting globally irrelevant vertices can be done locally and independently. Our method is applicable to a wide variety of known graph containment or graph modification problems where the irrelevant vertex technique applies. Examples include the (Induced) Minor Folio problem, the (Induced) Disjoint Paths problem, and the -Minor-Deletion problem.
Keywords: Graph Minors, Treewidth, Disjoint Paths Problem, Planar Graphs, Surface-Embeddable Graphs, Irrelevant Vertex Technique.
20(12.2, 2.4)
![[Uncaptioned image]](extracted/5708140/logo-erc.jpg){kind=logo}
{textblock}20(12.2, 3.0)
![[Uncaptioned image]](extracted/5708140/logo-eu.jpg){kind=logo}
1 Introduction
The irrelevant vertex technique was introduced by Robertson and Seymour in [57] for deriving a polynomial algorithm for the Disjoint Paths problem. This technique has nowadays evolved to a standard algorithmic paradigm for solving graph containment or graph modification problems [28, 36, 2, 49, 40, 29, 37, 43, 46, 33]. The general idea behind the technique is to exploit certain structural characteristics of the input that make possible to find a vertex of the input graph that is problem-irrelevant: its removal from an instance creates a new instance that is equivalent to the old one.
The archetypical example of the applicability of this technique is the Planar Disjoint Paths problem where, given a planar graph and pairs of terminals , the question is whether has pairwise-disjoint paths where is a path from to , for . For this problem, the core structural characteristic is the “linkage theorem” proved in [59, 60]. This result implies the existence of a function (called the linkage function) such that the existence of a sequence of pairwise disjoint nested cycles in a plane embedding of where all terminals are outside the “outer cycle” of this sequence implies that all paths of a solution of the Planar Disjoint Paths problem can be rerouted away from the “inner cycle” of this sequence. This in turn permits to safely discard from all vertices inside the inner cycle and create an equivalent instance of Planar Disjoint Paths. Intuitively, this sequence of nested cycles “insulates” the terminals from the vertices declared irrelevant. Given that every graph of large enough treewidth contains the subdivision of a wall and, in planar graphs, such a wall certifies the aforementioned insulation sequence, as long as the treewidth of the input graph is above a certain threshold, we detect and discard irrelevant vertices until the treewidth becomes bounded. Then one may solve the problem in linear time using dynamic programming techniques. While this idea appears to be quite simple and intuitive, to prove that the above rerouting is indeed possible is non-trivial even in the case of planar graphs (see [3, 30, 4]). When it comes to problems on general graphs, the planarity condition is substituted by suitable notions of flatness that still permit a rerouting argument based on sufficiently big “insulation” [57, 59, 60], while additional machinery is required in order to find irrelevant vertices for “non-flat” instances.
In problems such as Minor Containment and “induced” or “rooted” variants of it, the above insulation property is not enough for declaring the vertices within the inner cycle irrelevant. Additionally, a “big enough” flow between vertices of the inner and the outer cycle is required. In this paper, we call the union of the insulating cycles and the paths of a flow traversing them a railed nest. Our results concern all problems where insulation by a railed nest is sufficient for declaring irelevance. We refer to this problem property as the insulation property. A third family of problems with the insulation property concerns graph modification problems where the question is to find a set of vertices whose removal may yield some particular minor-closed property, such as planarity (see [46, 49, 7, 63, 52]).
Actually, there are families of problems where for the applications of the irrelevant vertex technique the insulation property is not enough. The reason is that the existence of a railed nest implies that only a subset of the vertices inside the inner cycle is irrelevant and extra algorithmic effort is required in order to detect them. The most representative problem of this category is Topological Minor Containment [33, 37] (see also [24, 31, 28, 30] for other results of this kind). In fact, there are situations where the irrelevant vertex argument relies on combinatorial conditions that extend the insulation argument and the method is combined with other advanced techniques in graph algorithm design [25, 49, 33, 45, 2, 21, 24]. Finally, there were recent attempts to derive general meta-algorithmic conditions for the applicability of the irrelevant vertex technique [32, 22, 23, 66, 65].
The sub-quadratic issue.
As already indicated above, the straightforward application of the irrelevant vertex argument requires quadratic time. One typically needs time in order to detect an irrelevant vertex and remove it; this should be repeated as long as an irrelevant vertex can be detected, that is times. We stress that in general, the detection of the irrelevant vertex in time is not always a straightforward task. Even in the classic case of Disjoint Paths and Minor Containment, the original results of Robertson and Seymour in [57] required time to detect the irrelevant vertex. This was later improved in linear time by Kawarabayashi, Kobayashi, and Reed in [42]. Nowadays, most algorithmic or meta-algorithmic applications of the irrelevant vertex technique require time (see [65, 32, 22]). However achieving sub-quadratic implementations of the irrelevant vertex technique remains an open challenge. This has been achieved for particular problems and, in all of them, the challenge that was met was to detect “many” irrelevant vertices at once, instead of finding them one by one. The first problem for which a linear-time algorithm was derived was Planar Disjoint Paths and its extension Planar Disjoint Connected Subgraphs by Reed, Robertson, Schrijver, and Seymour in [54] (see also [53]). Also, the results of [54, 53] are applicable for every class of surface-embeddable graphs. Next, Mohar, in [51], used the irrelevant vertex technique in order to check whether a graph is embeddable in some particular surface in linear time. The algorithm of [51] was also one of the main ingredients of the linear algorithm of [45] for checking whether a graph can be embedded in the plane with at most crossings. The algorithm of [45] made use of the irrelevant vertex technique by combining the algorithm of [51] with results from discrete geometry. Later, a simpler algorithm for surface-embeddability was proposed by Kawarabayashi, Mohar, and Reed in [44], where irrelevant vertices where found by successively “shrinking” the problem instance to an equivalent one by contracting (big) induced matchings. Kawarabayashi in [38] used a similar approach in order to solve, in linear time, the Planarizer problem, asking whether the removal of at most vertices can make a graph planar. To our knowledge, [38] is the only result that presents a linear-time implementation of the irrelevant vertex technique for a graph modification problem.
Recently Korhonen, Mi. Pilipczuk, and Stamoulis [48] gave an almost linear algorithm for solving the Rooted Minor Folio problem, asking for the set of all rooted minors of some specific size in a graph. The techniques in [48] combined the irrelevant vertex technique with successive instance shrinking and the use of advanced dynamic algorithm techniques from [47]. As a result of [48], the quadratic bound from [42] for the Minor Containment problem has been reduced for the first time to almost linear.
Our results.
In this paper we provide an algorithmic framework for surface-embeddable graphs that, when applicable, can find, in linear time, an entire set of irrelevant vertices whose removal yields a bounded-treewidth graph. We now informally introduce the property that conditions the applicability of our results (for the formal definitions, see Section 2). We deal with problems on rooted graphs of bounded Euler genus. The input graph comes with a bounded-size set of distinguished vertices, called roots. We say that such a problem has the insulation property if for every instance of such that
-
•
contains a sequence of sufficiently many nested and pairwise-disjoint contractible cycles,
-
•
contains a collection of sufficiently many pairwise-disjoint paths between the outer cycle and the inner cycle of , and
-
•
all roots in are embedded outside the outer cycle of
and with being the set of vertices inside the inner cycle of , then for every and are equivalent instances of . This expresses that fact that all vertices in are irrelevant for the instance of . We refer to the above union of the cycles in and the paths in as a railed nest insulating from the roots in .
Typical problems that have the insulation property and for which our results apply include (Induced) Disjoint Paths, (Rooted/Induced) Minor Containment, and Planarizer. This list covers only a small sample. In order to demonstrate the potential of our technique, we dedicate Section 8 to an abstract description of problems that satisfy the insulation property.
Our main result is that for every problem on rooted surface-embeddable graphs that has the insulation property there exists a linear-time algorithm that reduces every instance to an equivalent one whose graph has bounded treewidth. As a corollary we obtain for the first time linear-time algorithms for a wide family of well-studied graph containment/modification problems on surface-embeddable graphs.
Our technique.
In this paper, we follow an approach that deviates significantly from the “shrinking” techniques applied in [44, 48, 38]. Our approach is self-contained and can be black-box applied to every problem that has the insulation property.
Our strategy is to detect in the input graph many different railed nests and discard globally-irrelevant vertices for all of them by processing them locally. To make this possible, we decompose, in linear time, the graph into a tree-structured collection of subgraphs, called slices, each of bounded treewidth. The algorithm processes each slice sequentially. To identify the slices in linear time, we introduce the concept of radial distance decomposition. For each slice we certify locally all railed nests in it and we prove that this permits the detection and elimination of all globally-irrelevant vertices relative to this slice. This certification is implemented in linear time using Courcelle’s theorem on some enhanced version of the slice that, as we prove, is of bounded treewidth as well. We prove that after all slices have been locally processed, an equivalent instance of bounded treewidth is produced and that this procedure can be implemented in linear time. If in addition the problem can be expressed in Counting Monadic Second Order logic (CMSO), it immediately follows that it can be solved in linear time. A formal description of our results is given in Section 2 and an outline of our technique is presented in Subsection 2.3.
2 Formal presentation of the result
We use for the set of all nonnegative integers. Given two integers with , we use to denote the set and to denote the set of all even integers in . Given a we set and and we use to denote the set of all even integers in . If is a collection of objects where the operation is defined, then we use to denote .
2.1 Basic concepts on graphs
All graphs in this paper are simple, undirected, and finite. Given a graph , we denote its vertex and edge set by and respectively. We set . Given some , we denote by the graph obtained if we remove from the vertices in , along with their incident edges. For , we use to denote the graph . Also, the subgraph of induced by , denoted by , is the graph . Given a set , we denote by the graph where is the set of all endpoints of the edges in . If is a graph where and then we say that is a subgraph of .
Given a graph and a set (resp. ) we say that (resp. ) is connected if (resp. ) is a connected graph. Given a connected set of edges we denote by the graph obtained from after the contraction of all edges of to a single vertex.
Embeddings.
In this paper we deal with graphs embedded in a surface . For this, when we refer to a -embedded graph , we agree that is accompanied with an embedding of it in . For notational convenience, we do not distinguish a vertex/edge of from the points corresponding to its embedding. For instance, given a -embedded graph , if is a subgraph of then we denote by the set of points of that are not points of the embedding of . Also, if , then consists of all vertices of that are embedded to points of . Given a cycle of we say that is contractible if one of the two (arcwise) connected) components of is an open disk. We use to denote the sphere. Therefore, -embedded graphs are planar graphs. We use to denote the Euler genus of the surface and for the minimum Euler genus of a surface where can be embedded.
Treewidth and CMSO.
A graph is -chordal if it has no cycle of length as induced subgraph and does not contain a complete graph on vertices. The treewidth of a graph , denoted by is the minimum for which is a subgraph of some -chordal graph. The parameter of treewidth is important for our algorithms, however we do not make any use of its original definition (given in [55]). Instead we use only results around treewidth and, for this reason, we resort to the above definition for brevity.
Rooted graphs.
A rooted graph is a triple where is a graph, and is a bijection. We refer to as the boundary of and to its vertices as the roots of and we denote them by . We say that and are isomorphic if there is an isomorphism from to that extends the bijection A rooted graph is a -rooted graph if . Notice that the notion of a rooted graph is extending the one of a graph as every graph can be seen as a -rooted graph. A rooted graph is -embedded if is -embedded, i.e., is accompanied with a -embedding of it. We also use instead of and we define . Also, if we use to denote the rooted graph . Similarly, we define for some . We set and, in general, if is a graph parameter (such as or ) and , we define its rooted graph extension so that .
In this paper, we see every problem on rooted graphs as a set of rooted graphs, i.e., the set of yes-instances of the problem, whose number of roots is bounded by some fixed number. Also we insist that all problems that we considered concern rooted graphs with at least one root. In the case of simple graphs, i.e., -rooted graphs, we arbitrarily choose some vertex as a root. This convention is useful for the uniformity of our presentation.
Railed nests.
Let be a -embeddable rooted graph and let . An -railed nest of is the union of a collection of cycles of and a collection of pairwise vertex-disjoint paths of such that
-
•
for every the cycle bounds an open disk of the surface where all the cycles in are embedded,
-
•
for every , the graph is a (possibly trivial) path,
-
•
all roots of are embedded in , i.e., and
-
•
at least one vertex of is embedded in .
We refer to (resp. ) as the outer (resp. inner) cycle of . The interior (resp. exterior) of is the open disk (resp. the open set ) and is denoted by (resp. ).
The combinatorial importance of treewidth resides in the fact that it certifies the existence of a railed nest (see also Subsection 4.1).
Proposition \theproposition.
There is a universal constant such that, for every and , if is a -rooted graph where then contains an -railed nest. Moreover, such an -railed nest can be found in time .
The above proposition is a consequence of the results of [18], as explained in Subsection 4.1.
An annotated graph is a tuple for some where each is either a vertex (resp. edge) subset of (resp. ). Monadic Second Order logic (MSO) is a basic tool to express properties in (rooted and/or annotated) graphs. The syntax of MSO includes logical connectives , , , , , variables for vertices, edges, vertex sets, and edge sets, quantifiers , over these variables, the relations when is a vertex variable and is a vertex set variable; when is a vertex variable and is a vertex set variable; when and are vertex variables, with the interpretation that and are adjacent; when is a vertex variable and is an edge variable, with the interpretation that is incident to ; and equality of variables representing vertices, edges, vertex sets, and edge sets. Counting Monadic Second Order logic (CMSO) extends MSO by including atomic sentences testing whether the cardinality of a set is equal to , where and .
The algorithmic importance of treewidth resides, among others, in the next result, known as Courcelle’s theorem. It has appeared in different versions and proofs in [14, 16, 15, 8, 5]. In this paper we make use of the following optimization version (see [5]).
Proposition \theproposition.
For every CMSO-sentence on annotated graphs there is an algorithm that, given an -vertex graph , outputs a maximum size set where , in time .111Given two functions we write to denote that there exists a computable function such that
The decision version of Subsection 2.1 asserts that for every CΜSO sentence on rooted graphs, there is an algorithm deciding whether in -time. A rooted graph problem is CMSO-expressible if there is a CMSO-sentence on rooted graphs such that iff . Notice that Subsection 2.1, readily implies that every CMSO-expressible rooted graph problem with the insulation property, can be solved in quadratic time on bounded-genus bounded-boundary rooted graphs: indeed, to see this we apply Subsection 2.1 in order to detect some big enough railed nest and we discard from all vertices inside its inner cycle. After applying this step a linear number of times, the treewidth will become bounded by some function of and and it will be possible to apply the decision version of Courcelle’s theorem in order to solve the problem in quadratic time. The main contribution of our paper is that any such problem can be solved in linear time.
2.2 Formal description of our result
The insulation property.
Let be a rooted graph problem. Given a rooted graph , a vertex is -irrelevant if if and only if .
We say that a rooted graph problem has the insulation property if
there exists some constant such that, for every -embedded rooted graph with a -railed nest , if then every vertex in is -irrelevant.
Notice that the constant does not depend on the choice of the surface . Given two rooted graphs and we say that they are -equivalent if .
Our main result is the following.
Theorem 2.1.
For every rooted graph problem with the insulation property, there is an algorithm that, given a -vertex rooted graph where , outputs a set such that
-
•
and
-
•
and are -equivalent.
Moreover, the algorithm runs in time .
Using Theorem 2.1 and the decision version of Subsection 2.1 we get the following corollary.
Corollary \thecorollary.
For every CMSO-expressible rooted graph problem that has the insulation property, there is an algorithm that, given an -vertex rooted graph where , decides whether belongs in in time .
2.3 Outline of the proof of Theorem 2.1
We provide here the main ideas of the proof of Theorem 2.1. The proof requires several concepts on embedded graphs that we only present informally here. All formal definitions, statements, and proofs are available in Sections 3, 4, 5, 6 and 7.
Radial distance decompositions
We first present how the algorithm of Theorem 2.1 works in the special case where is a planar graph, i.e., it is -embedded and where there is only one root vertex . We use in order to denote problem instances of this type. The root is useful in order to orient cycles of that do not meet it: each cycle defines two open disks and its exterior is the open disk that contains while its interior is the open disk that does not contain . We also need the concept of a radial graph of that is a bipartite graph whose vertices are the vertices and the faces of and where adjacency expresses incidence between the corresponding faces and vertices. The definition of naturally defines an embedding of in and can be extended for graph embeddings on any fixed surface. A path in defines a radial path in between the corresponding vertices or faces. Accordingly, the distance in between two vertices defines the radial distance in between the corresponding vertices or faces.
We next introduce the radial distance decomposition of as the triple where is a rooted tree, the “bags” form a partition of where , for every in distance from , in , all vertices in are in radial distance from in . We also require that if two vertices or faces have the same, say , radial distance from and there is a radial path between them all whose vertices/faces have radial distance from , then both (resp. ) belong in the same bag. See Fig. 1 for an illustration. The construction of the radial distance decomposition of can be done based on a BFS transversal of , in linear time. We see all edges of the rooted tree as being directed away from the root. A vertex-face edge of is an edge where consists of vertices and consists of faces of . Radial distance decompositions offer the guiding underlying structure for our algorithm. To our knowledge, this is the first time this structure is defined for the algorithmic study of planar graphs.
In Section 3 we prove that each vertex-face edge of corresponds to a cycle of (Subsection 3.2). Moreover, an important property of a radial distance decomposition is that the vertex-face edges of a (directed) path of length in correspond to a sequence of nested and pairwise-disjoint cycles of (Subsection 3.2). These cycles are called aligned and will be the “vital space” where our algorithm will look for disjoint paths that will form the railed nests where the insulation property will be applied. Another property of radial tree decompositions that we need is that every two aligned cycles, i.e., cycles corresponding to vertex-face edges of , are laminar, i.e., no two vertices of the one cycle are separated by the other.
Slices.
Let be some vertex-face edge of where . For some , we define as the set containing every cycle of corresponding to a vertex-face edge where the distance in between an is . As all cycles in are pairwise laminar we may consider the pseudo-disk defined if we remove from the closed interior of all open interiors of the cycles in . We define as the part of the embedding of that is embedded inside this pseudo-disk (see Fig. 2).
By considering the set of vertex-face edges of where , we may decompose into the set of slices . Notice that the union of these slices is the graph without the root vertex . Let us recall at this point that our plan is to sketch how to solve the problem for instances consisting of planar single-rooted graphs. Later in the end of this section we describe how to reduce to this special case the general case where we have many roots and surface-embeddable graphs. In that approach, a general instance is viewed as a planar single-rooted instance that is the “contracted version” of the original rooted graph where represents the contracted territory.
A key idea of the algorithm for the planar single-rooted case is to process each of the slices separately and certify the existence inside them of big-enough railed nests that justify the elimination of (globally) irrelevant vertices from each individual slice. For a visualization of six slices and the way they correspond to subtrees of the radial tree decomposition, see Fig. 2.
Flows and railed nests.
Assume that the problem that we want to solve has the insulation property for some constant . In our arguments the choice of will depend on . We set . Let be a vertex-face edge of and . In Fig. 2 is depicted in red and the cycles in are depicted in blue. Let and recall that contains a nested sequence of cycles where and . If additionally contains a collection of -many pairwise-disjoint paths from to , then forms a -railed nest justifying that all vertices in the open interior of are -irrelevant. Our plan is to find, if it exists, such a collection of -many pairwise-disjoint paths for every cycle . Clearly, these paths are not necessarily paths of and it would cost too much time to look for these disjoint paths in the whole of . Instead, we prove that it is enough to look for disjoint paths in where and is the linkage function (see [3, 59, 60] and Subsection 5.2). Indeed, here we make use of the fact that between (denoted in bold blue) and the cycles of (denoted in brown) there is always a nested sequence of pairwise-disjoint cycles that, because of the linkage theorem, permits a rerouting of every set of -flow from to in to paths inside . Let be the set of cycles in that send a -flow to .
For every individual , we immediately know that all vertices in the interior of are -irrelevant and we certainly could safely remove them. However, discarding only one such set of -irrelevant vertices is not enough for deriving a sub-quadratic algorithm. At this point, one might be tempted to remove the interiors of all cycles in . However, this is unsafe as the vertices that we remove from the interior of some particular might be vertices of the -flow from to certifying the inclusion of some in . In other words, discarding the interior of might entail that the vertices in the interior of lose the property of being -irrelevant.
Our way out of this is to discard the vertices in the union of the open interiors of all cycles in , for every . This is safe as all aforementioned collections of paths, as argued above, avoid . We will argue that this elimination is “massive” enough to produce a bounded treewidth graph when applied to the slices corresponding to the vertex-face edges of the set defined above.
Discarding irrelevant vertices in slices.
Let . We just argued that all collections of -many pairwise-disjoint paths in between and the cycles of also exist in and that this permits us to safely discard the interiors (drawn in yellow in Fig. 2) of the cycles in for every . Our next step is to find in time that is linear in the size of the slice . For this, we exploit the fact that is a planar graph of bounded diameter, therefore its treewidth is . We next add in , for every , a new vertex adjacent to all vertices of and, in Section 4, we prove that this enhancement does not significantly increase the treewidth of the resulting graph . Then we ask for the maximum size of a set such that sends a flow of -many pairwise-disjoint paths to all the vertices of . As this question can be expressed in CMSO, Subsection 2.1 implies that one may compute the set in time (Subsection 5.2). Using this , we also infer and we can safely discard from all vertices that are embedded inside .
Bounding the treewidth.
We apply the above procedure for all slices of depth corresponding to the vertex-face edges of the above-defined set , starting from the edges of that are closer to the root. Recall that a vertex of a slice can appear multiple times only if it belongs in the “boundaries” of different slices. An elementary use of Euler’s formula implies that the total running time for processing all the slices is linear in the size of . This procedure produces a set of irrelevant vertices whose removal produces a -equivalent instance . We use for the (remaining) radial distance decomposition. Our next step is to observe that the new instance satisfies the following property: for every directed path of with vertices there are no disjoint paths between and . The proof of this property is in Subsection 6.4 and, intuitively, is based on the fact that such paths would contain vertex-face edges corresponding to cycles that should have already been eliminated according to the above procedure. Next, again in Subsection 6.4, we prove that this property implies the desired bound on the treewidth of .
From single-rooted and planar to multi-rooted and surface embeddable.
What we just sketched is the proof that Theorem 2.1 holds for single-rooted and planar instances. Assume now that is a -rooted graph embedded in some surface where . We next prove that one may find, in linear time, a collection of shortest paths in the radial graph of meeting all roots of and such that if we contract them to a single vertex in , the resulting graph is planar. This is proved in Subsection 7.2 using as a departure point the algorithmic results of Cabello, Colin de Verdière, and Lazarus in [11]. Using these radial paths of , we find, also in linear time, a connected set of edges of such that if we contract them in to a single vertex the resulting graph is planar. Next we apply the aforementioned irrelevant-vertex elimination procedure on and we prove that the set of irrelevant vertices for is also a set of irrelevant vertices for . Intuitively, this holds because all railed-nests that justified the inclusion of vertices in , while processing , are also present in . What remains is to prove that the bound on the treewidth of also implies a bound to the treewidth of . For this, we use the fact (guaranteed by the choice of ) that all edges in that have been contracted towards creating have endpoints that are within radial distance two from some of the vertices of the radial shortest paths in . Based on this fact and making use of the results of Demaine, Hajiaghayi, and Kawarabayashi in [17], we prove in Subsection 7.2 that the contraction of the edges does not decrease the treewidth of by more than a constant, depending on and . This implies the bound of Theorem 2.1 for general surface-embeddable rooted graphs.
3 Radial distance decompositions and nested cycle sequences
In this section we give the formal definition of radial distance decompositions of pairs and show that each vertex-face edge of the underlying tree corresponds to a cycle of , which we call -aligned.
3.1 Radial distance decompositions
We start by giving the formal definition of radial graphs of surface-embedded graphs. In the rest of the paper, we always assume that the considered embedding of every given -embedded connected graph is a -cell embedding, i.e., every face is homeomorphic to an open disk. A bridge in a graph is an edge such that has more connected components than . A graph is -edge connected if it has no bridges.
Radial graphs.
Let be a surface and let be a -embedded connected graph. We define the radial graph of , denoted by , as the graph whose vertex set is and whose edge set is defined as follows: for every we consider the closed walk of defined by the boundary of and we make adjacent to all the vertices in this walk (we permit multiple edges as a vertex can appear many times in the walk). Note that is a -embedded graph that is bipartite and connected. A radial path in is a path between two vertices in . Notice that each endpoint of such a path may correspond either to a vertex or a face of the embedding of . The radial distance of two disjoint cycles of is the length of the shortest radial path with one enpoint in and one endpoint in .
We give the following straightforward observation for radial graphs of connected and bridgeless graphs embedded on .
Observation \theobservation.
Let be a -embedded graph. If is connected and bridgeless, then every face of is incident to exactly four vertices.
Tree distance decompositions.
Before presenting the definition of tree distance decompositions and radial distance decompositions, let us introduce some additional notation and definitions. Given two vertices of a graph , we denote by the minimum number of edges of a path in with endpoints and ; if there is no such path does, we set . Also, a rooted directed tree is a directed graph and is a vertex of such that has in-degree zero and each vertex different than has in-degree exactly one. In other words, is a tree with some distinguished vertex where all edges of are “oriented away” from . For every edge , we set .
Let be a connected graph and . A tree distance decomposition of is a triple , where is a rooted directed tree and , that satisfies the following conditions.
-
•
is a partition of , with ,
-
•
for every , if , there is a so that , and
-
•
for every , if and there is a -path in , then there is a such that .
Tree distance decompositions (using a different formulation) have been introduced in [67]. Let us note that tree distance decompositions are obtained by the BFS-layering of graphs (with respect to some fixed root) after “grou**” vertices in every layer corresponding to connected components in a bottom-up fashion.
Observe that for every connected graph and every vertex , there is a unique triple satisfying the above conditions. This allows to refer to as the (unique) tree distance decomposition of .
Lemma \thelemma.
Given a connected graph and a vertex , we can compute the tree distance decomposition of in linear time.
Proof.
The distance decomposition of can be computed by producing a BFS-tree for rooted at and recursively storing which vertices of the same level of the search tree are connected in the graph induced by the vertices of the current subtrees. It is easy to observe that the recursive storage of this information maintains linearity of BFS. ∎
Rooted embeddings and their radial distance decompositions.
A rooted embedding is every pair , where is a -edge-connected graph, accompanied with an embedding of it in the sphere, and is a vertex of . We refer to as the “reference point” of . The -edge-connected assumption in rooted embeddings facilitates our presentation and we explain why it does not hurt generality in Subsection 6.1.
From now on, we will only consider tree distance decompositions of where is a rooted embedding. If is the distance decompositon of , we refer to it as the radial distance decomposition of the embedding pair . See Fig. 1 for an illustration of the radial distance decomposition of a rooted embedding.
Let us mention that in [17], the authors define radial colorings of graphs that correspond to BFS-layerings of their radial graphs. Their definition differs from the definition of radial distance decomposition in the fact that in radial colorings we do not “group” vertices in the same “bag” when they are connected in the “suffix” of the decomposition, i.e., radial colorings do not demand the last condition of the definition of a tree distance decomposition.
Vertex and face layers of a radial distance decomposition.
Let be the radial distance decomposition of a rooted embedding . Note that since is bipartite, for every that is in odd (resp. even) distance from in , it holds that (resp. ). For this reason we call the set of nodes of with odd distance from a face layer of and the set of nodes of with even distance from a vertex layer of . We say that and edge of is a vertex-face edge of if is in a vertex layer of . A vertex-face edge of that is incident to the root is called root edge. In Appendix A we show that there is exactly one root edge, but this is only used in the proofs in Appendix A.
3.2 Aligned cycles
In this subsection we show that, given a rooted embedding together with its radial distance decomposition , every non-root vertex-face edge of corresponds to a cycle of . Moreover, this cycle bounds an open disk containing all vertices and faces of that are in bags of nodes in the subtree of rooted at this edge. In order to state this result formally, let us give some additional definitions.
Let be a -embedded graph and let . Let also be a cycle of disjoint from . Note that consists of two open disks. The interior of , denoted by , is the open disk bounded by that does not contain . The open exterior of , denoted by , is the open disk bounded by that contains . In other words serves as a “reference vertex” that orients all cycles that it does not intersect, that way is the disk that is “away from ” and is the disk containing .
Let be the radial distance decomposition of a rooted embedding . Given an edge of (in this case is a child of ), we use to denote the union of all , where is either or a descendant of in .
Lemma \thelemma.
Let be the radial distance decomposition of a rooted embedding . For every non-root vertex-face edge of , there is a unique cycle in such that and is the set of vertices of in the interior of .
The proof of Subsection 3.2 follows from the fact that each (non-root) vertex-face edge of partitions the set of faces of to two parts, so that the union of (the closure of) all faces in each part is a disk. This can be shown by induction on the tree , using the definition of the radial distance decompositions as well as some easy observations on -embedded graphs; a full proof of Subsection 3.2 is given in Appendix A.
Root-aligned cycles.
Let be the radial distance decomposition of a rooted embedding . Given a non-root vertex-face edge of , we denote by the unique cycle of corresponding to it, given by Subsection 3.2. Also, we say that a cycle of is -aligned if there is a vertex-face edge of such that .
The next observation about the relation between -aligned cycles of will be used later in Section 6. It follows directly from Subsection 3.2 and the fact that for every two nodes of the radial distance decomposition of a rooted embedding, the sets and are disjoint.
Observation \theobservation.
Let be a rooted embedding and let be two -aligned cycles of . If intersects the interior of , then is embedded in .
Next, we show that in linear time we can dispose the set of all -aligned cycles of of a given rooted embedding .
Lemma \thelemma.
Let be the radial distance decomposition of a rooted embedding . There is a function that maps each non-root vertex-face edge of to the cycle . Moreover, this function can be computed in linear time.
Proof.
Due to Subsection 3.2, for every non-root vertex-face edge of , the cycle corresponds to the graph induced by the edges of that are incident to one face in and one face in the union of , for all nodes of that are neighboring to ; every edge is incident to exactly two faces because is -edge-connected. Therefore, the function that associates each non-root vertex-face edge of to the cycle can be computed by local queries on the neighborhood of the faces in inside the the union of , for all nodes of that are neighboring to , which in total can be done in linear time. ∎
Nested cycles.
Let be a -embedded graph and let be a sequence of cycles in . We call nested, if its cycles are pairwise-disjoint and every bounds an open disk such that (see Fig. 3 for an example). We call the inner cycle of and the outer cycle of .
By repeatedly applying Subsection 3.2, we show that a sequence of vertex-face edges appearing in a path of the radial distance decomposition gives rise to a sequence of nested cycles.
Lemma \thelemma.
Let be the radial distance decomposition of a rooted embedding . Let and be two non-root vertex-face edges of where is a descendant of . If , for some integer , then there is a nested sequence of many -aligned cycles of where the inner cycle of is and the outer cycle of is .
Proof.
Let be the vertices of the path connecting and in , ordered in the ancestor-descendant relation, i.e., and . For every , let denote the vertex-face edge of . We use to denote the cycle and to denote the open disk . Due to Subsection 3.2, for every it holds that and is the set of vertices of in the interior of . Since for every we have , we get that . Therefore, is a nested sequence of -aligned cycles of where and . ∎
4 Bounding the treewidth of laminar stellations
An important element of our approach, as explained in Subsection 2.3, is to show that the treewidth of a -embedded graph does not increase too much after extending it to its laminar stellation. This corresponds to the graph obtained from a given graph and a collection of strongly laminar cycles after introducing one new vertex per cycle and making this vertex adjacent to all vertices of the corresponding cycle. Laminar stellations are defined in Subsection 4.2, where we also show that their treewidth is a linear function of the treewidth of the original graph. Before this, in Subsection 4.1, we provide some preliminary facts about treewidth and surface-embedded graphs. We will use to denote the path on vertices and to denote the star graph, i.e., the tree consisting of one vertex that is adjacent to leaves.
4.1 Treewidth and and walls in surfaces
We use to denote the Cartesian product of the graphs and . We need the following known result about treewidth products (see e.g., [35, Lemma 19]).
Proposition \theproposition.
If and are graphs, then
The elementary -wall, for , is obtained from after deleting every odd edge in every odd column and every even edge in every even column, and then deleting all degree-one vertices. A -wall is a graph isomorphic to a subdivision of the elementary -wall. Notice that every -wall has a unique (up to topological isomorphism) embedding in the sphere and its perimeter is the boundary of the unique face of this embedding that has more than six degree-3 vertices.
All results on treewidth in this paper are based on the following result that can be seen as a surface-embeddable version of the “grid-exclusion theorem” of Robertson and Seymour [55, 61] (see also [13]).
Proposition \theproposition.
There is a universal constant such that if a -embedded graph has treewidth bigger than , then contains as a subgraph an -wall whose perimeter is a contractible cycle bounding a closed disk where the entire is embedded.
The above result follows from [18, Theorem 4. 8] (see also [27, Lemma 4]). Notice also that Subsection 4.1 readily implies a proof of Subsection 2.1: the -overhead permits us to consider a partition of wall to pairwise-disjoint walls and then pick one that is embedded in a closed disk bounded by its perimeter that does not contain any of the roots of . Next it remains to observe that a big enough wall can give rise to the required -railed nest. The algorithmic part of Subsection 2.1 follows by applyng the linear-time algorithm of [1].
In the next section, we will also use the fact that the treewidth of a graph is upper-bounded by a linear function of its radial radius, which is defined as follows. Given a graph and a vertex , the eccentricity of in is the maximum distance between and some vertex of . The radius of is the minimum eccentricity of its vertices. Given a -embedded graph, we define its radial radius as the radius of its radial graph. We need the following result that follows easily from Subsection 4.1. For a proof with improved constants, see [55, 2.1].
Proposition \theproposition.
If is a -embedded graph of radial radius , then
A graph is called sub-cubic if it has no vertex of degree bigger than three. We say that is a contraction of if can be obtained from after contracting edges. A vertex is a cut-vertex of if has more connected components than . A graph is 2-connected if it has no cut-vertices. A block of a graph is a maximal 2-connected subgraph of .
Lemma \thelemma.
Every -embeddable graph is the contraction of a planar sub-cubic graph where .
Proof.
We examine the non-trivial case where is not a forest. We assume that is 2-connected, otherwise we consider all its blocks of separately and join them by adding edges, an operation that does not increase the treewidth of a graph that is not a forest. The parameter of branch-width was defined in [56] and can be seen as an alternative for treewidth. We do not give the definition of branch-width here. We only need two properties of it. The first is that for every graph , because of [56, (5.2)]. The second is that every 2-connected graph is a contraction of some sub-cubic graph of the same branch-width, because of [26, Lemma 3.4]. It is now easy to derive the bound of the lemma from these two properties. ∎
4.2 Laminar stellations
Let be a -embedded graph and let be a collection of cycles of . We say that is laminar if for every , if and are open disks bounded by the embeddings of and respectively, then either one of the disks in is a subset of the other or the two disks in are disjoint. We call strongly laminar if there is no cycle such that both closed disks bounded by contain some cycle in embedded in them. Given a -embedded graph and a strongly laminar collection of cycles of , the laminar stellation of with respect to , which we denote by , is the graph obtained if, for every , we introduce a new vertex and we make it adjacent with all the vertices of .
Lemma \thelemma.
Let be a -embeddable graph and be a strongly laminar collection of cycles of . Then .
Proof.
We first apply Subsection 4.1 on and consider a -embedded sub-cubic graph where , as in Subsection 4.1. The fact that is a contraction of implies that each cycle corresponds to a cycle in , which we denote by . Notice that the set is a strongly laminar collection of cycles in . We now consider the graph . Clearly, is also sub-cubic and -embedded. Also, as is strongly laminar, each cycle of corresponds to some face of . Also keep in mind that because is a contraction of it also holds that is a contraction of . We denote by the map graph of that is the graph whose vertices are the faces of and where two vertices are adjacent if the corresponding faces have some common incident vertex. Map graphs have been introduced by Chen, Grigni, and Papadimitriou in [12] and in general, map graphs of planar graphs are not necessarily planar. However, it is the case that map graphs of sub-cubic planar graphs are also planar. Using the four-color theorem, we may color the vertices of using four colors. This means that the faces of , and therefore also the cycles of , can be colored with four colors in a way that no two faces (resp. cycles of ) with a common incident vertex have the same color. We partition to four same-colored classes of cycles . We next consider the graph and keep in mind that, from Subsection 4.1 . Clearly contains five disjoint copies of : a copy corresponding in the non-leaf vertex of and four copies . Notice now that for each color , the copies of the cycles in inside do not have common vertices. We now consider a minor of as follows: for every , we contract in every copy of a cycle of in to a single vertex and remove all vertices of the copies of the cycles of in . Observe that the resulting graph is the graph : for every , colored , the vertex is the vertex created because of the contraction of the copy of in . Recall now that is a contraction of which, in turn, is a minor of and that and . This implies that , as required. ∎
5 Finding flows inside slices in linear time
As explained in Subsection 2.3, the key idea of our algorithm is to split the given rooted embedding to slices and then process each slice separately in order to detect big enough railed nests inside them. In this section, and in particular in Subsection 5.1, we give the definition of slices and we use the results of the previous section in order to bound the treewidth of laminar stellations of slices. Also, in Subsection 5.2 we state a consequence of the linkage theorem and we show that in sufficiently large slices, in order to detect flows between particular cycles in each slice, it suffices to search locally inside the slice.
5.1 Slices and their treewidth
Slices are graphs that are embedded in pseudo-disks. We next define pseudo-disks and associated notions and we prove that the treewidth of graphs embedded in pseudo-disks is upper-bounded by a linear function of their radial depth.
Graphs embedded in pseudo-disks.
A pseudo-disk is obtained from a closed disk by removing a finite collection of pairwise disjoint open disks in the interior of . We call the boundary of outer boundary of and the boundary of each of the disks of an inner boundary of . A boundary of is either the outper boundary or some of its inner boundaries. We say that a graph is -embedded for some pseudo-disk if it is accompanied with an embedding in where each boundary of is the embedding of some cycle of . The radial depth of a pseudo-disk embedding of a graph is the minimum for which there is a sequence of vertices and faces of , where
-
•
faces and vertices alternate, i.e., are vertices and are faces,
-
•
is a vertex in the outer boundary of and is a vertex in some inner boundary of ,
-
•
if or are two consecutive elements of this sequence, then is incident to the face , i.e., is embedded in the boundary of .
It is straightforward to observe that every graph embedded in a pseudo-disk that has bounded radial depth also has bounded radial radius. Thus, using Subsection 4.1, we can upper-bound its treewidth.
Lemma \thelemma.
Let be a pseudo-disk and let be a -embedded graph of radial depth at most . Then .
Proof.
Notice that has a -embedding where one, say of its faces is the outer boundary of . Notice also that all vertices or faces of are within radial distance from . This means that the radial radius of is at most and the result follows from Subsection 4.1. ∎
Slices.
Let be a rooted embedding. Let be a cycle of that is -aligned and let . We denote by the set of all cycles of that satisfy the following properties:
-
•
is -aligned,
-
•
, and
-
•
the radial distance of and is equal to .
We also denote by the set Keep in mind that is an open set. We also define as the graph obtained from if we remove all vertices and edges that are disjoint from the closure of and all vertices and edges that are embedded in . The construction of immediately implies the following.
Observation \theobservation.
If , every has a pseudo-disk embedding of radial depth at most .
As a direct consequence of Subsections 5.1, 5.1 and 4.2, we get the following result.
Lemma \thelemma.
Let be a rooted embedding and let be -aligned cycle of . Let also and . If and , then .
Proof.
Due to Subsection 5.1, has a pseudo-disk embedding of radial depth at most . Then, Subsection 5.1 implies that . Moreover, observe that the definition of implies that is strongly laminar. Therefore, Subsection 4.2 implies that . ∎
5.2 Flows in slices
We next give the necessary definitions on linkages that are used in order to state the consequence of the linkage theorem obtained from [3, Lemma 10].
Rerouting linkages.
A linkage in a graph is a subgraph of whose connected components are paths of at least one edge. The paths of a linkage are its connected components and we denote them by . The size of , denoted by , is the number of its paths, i.e., . The terminals of , denoted by , are the endpoints of the paths of and the pattern of is the set . Two linkages of are equivalent if they have the same pattern.
The next result can be extracted by the proof of [3, Lemma 10]. It shows that it is possible to reroute linkages away from a well-insulated part of the graph.
Proposition \theproposition.
There is a function such that for every -embedded graph and every the following holds. If contains a nested sequence of cycles, then for every linkage of of size whose terminals are in the open exterior of the outer cycle of , there is a linkage of that is equivalent to and is disjoint from the interior of the inner cycle of .
Let us note that the function from Subsection 5.2 can be assumed to be single-exponential by the results of [3]. Using Subsection 5.2, we show how to reroute linkages away from the interior of a collection of well-insulated aligned cycles of a rooted embedding.
Lemma \thelemma.
Let be a rooted embedding, let be a -aligned cycle of , and let . For every linkage of with no terminal in , there is an linkage of that is equivalent to and is disjoint from , where .
Proof.
Let be the radial distance decomposition of and let be the vertex-face edge of such that . Also, let be the set of all vertex-face edges of such that and observe that .
Consider an arbitrary ordering of the edges in . We set and for every , we set to be the graph obtained from after removing all vertices of in the interior of . We show that for every , there is a linkage of that is equivalent to ; the claimed linkage in the statement of the lemma is the linkage .
We fix some and we assume that there is a linkage of that is equivalent to . We aim to prove that there is a linkage of that is disjoint from the interior of . We use to denote the vertex-face edge and to denote the edge of where is an ancestor of and . Due to Subsection 3.2, there is a nested sequence of many -aligned cycles of where the inner one is and the outer one is . Note that , since . Also, the terminals of (which are the same as the terminals of ) are disjoint from . By Subsection 3.2, and therefore the terminals of are in the open exterior of . Invoking Subsection 5.2 for and , we get a linkage of that is equivelent to and does not intersect the interior of , as claimed. ∎
Big flows between cycles.
Given a graph and subgraphs of , we use to denote the maximum number of pairwise vertex-disjoint paths in with one endpoint in and one endpoint in . Let be a rooted embedding and let be a -aligned cycle of . Let also and . We denote by the set containing every cycle of for which .
As a direct consequence of Subsection 5.2, we get that in order to compute it suffices to check for which cycles in there is a -flow from to in . This reduces the flow question in the general graph to a “local” question in the corresponding slice.
Lemma \thelemma.
Let be a rooted embedding and let be a -aligned cycle of . Let also and . Then for every cycle , it holds that , where .
Proof.
Note that for every cycle , there is a linkage of of size such that every path of has one endpoint in and one endpoint in . Also, since and all cycles in have disjoint interiors, we get that no terminal of is in . Therefore, by Subsection 5.2, there is a linkage of that is equivalent to and is disjoint from , where . The latter implies that is also a linkage in and therefore since it has the same pattern as , . ∎
We conclude this section by showing that we can compute the set of all aligned cycles in a slice (in an appropriate radial distance) that certify the existence of sufficiently large railed nests. The ingredients used in this proof are the following. First, Subsection 5.2 allows to restrict our attention to the given slice. The existence of large flows can be expressed in MSO in the appropriate laminar stellation of the slice, which has bounded treewidth because of Subsection 5.1. Therefore, we can compute these flows in total linear time using Subsection 2.1.
Lemma \thelemma.
There is an algorithm that given a rooted embedding , a -aligned cycle of , two integers and , and the set of cycles , computes , in time , where and .
Proof.
Our aim is to compute using Subsection 2.1. By Subsection 5.2, it suffices to check for which cycles of in it holds that . We use to denote . In linear time, we construct the graph (by introducing a vertex for every cycle and making it adjacent to all vertices of ). Also, we color the set of all newly introduced vertices by a color, say . Notice that there is an MSO-formula that given a set , expresses, for every , the existence of a collection of paths in from to the vertices of that have as a common endpoint and no other common vertex. A maximum size such set corresponds to the collection of cycles . Therefore, using Subsection 2.1, we can compute the maximum size set such that , from which we can also get (by taking the neighborhood of each in ). Following Subsection 2.1, this can be done in time . Due to Subsection 5.1, and therefore we get the claimed running time. ∎
6 Removing all irrelevant vertices in linear time
In this section, we show how to split a planar single-rooted instance into slices and apply the algorithm of Subsection 5.2 in order to find railed nests inside each slice that certify the irrelevance of sets of vertices.
Since we want to argue about irrelevance in the original instances, the results in this section are stated for general surface-embedded multiple-rooted graphs. However, radial distance decompositions and slices are defined for rooted embeddings, which are single-rooted -edge-connected -embedded graphs. We will next work with reduced embedding pairs, which are rooted embeddings obtained from a surface-embedded graph after the contraction of a connected edge set that make it planar and then adding some edges in the obtained -embedded graph to make it -edge-connected. The corresponding definitions and results about achieving -edge-connectivity and the notion of reduced embedding pairs are given in Subsection 6.1. The specifications of the set to be contracted as well as the algorithm that finds it is presented in Section 7.
The set of vertices to be declared irrelevant for each slice as well as the set of all irrelevant vertices are described in Subsection 6.2, where we also show that their removal from the instance indeed gives an equivalent instance. Then, in Subsection 6.3, we give the algorithm that actually detects this set. We conclude this section with Subsection 6.4, where we show that, after the removal of the set of irrelevant vertices, the (equivalent) instance has bounded treewidth.
6.1 Making a sphere-embedded graph -edge-connected
In order to show how to get a -edge-connected supergraph of a given -embedded graph, we define the operation of digon completion.
Digon completions.
A -edge-connected component of a graph is a vertex-maximal -edge-connected subgraph of . Note that if and are two -edge-connected components of a graph then and if there is an edge with one point in and one point in , then is a bridge of .
Let be a -embedded graph with . The operation of digon completion in constructs a new graph as follows. Let be two distinct -edge-connected components of that share a face of . We distinguish two cases, depending whether and are connected by a bridge, or not. If and are connected by a bridge, say with endpoints , then is obtained by introducing a new edge between and , which we embed inside . If there is no bridge between and , we pick arbitrary vertices and and we construct by introducing two new edges between and , both embedded inside . We say that is obtained by a digon completion from . We call the newly introduced edges digon edges. Note that is also a -embedded graph.
Let be a -embedded graph. We say that a -embedded graph is a -edge-connected digon-completion of if is -edge-connected and either or there is a sequence of graphs, where and such that for every , is not -edge-connected and is obtained by a digon completion from .
Lemma \thelemma.
Let be a -embedded graph and let be a -edge-connected digon-completion of . Also, let be a vertex of . Then for every -aligned cycle of , we have that either contains no digon edge or the interior of is a face of .
Proof.
Assume that , since otherwise contains no digon edges. Let be the sequence of graphs where and certifying that is a -edge-connected digon-completion of .
Let be the radial distance decomposition of and let be the vertex-face edge of such that . Suppose towards a contradiction that contains a digon edge and the interior of is not a face of . Let be the maximum integer in for which there is a digon edge introduced when obtaining from . Observe that maximality of implies that is a cycle in and therefore both endpoints of should belong to the same connected component of . Therefore, by definition of the digon completion operation, should contain only one edge which should also be a bridge in . Since consists of the two edges and , again by the definition of the digon completion operation we have that the interior of bounds a face of , a contradiction. ∎
Reduced embedding pairs.
Let be a surface, let be a -embedded graph, and let be a connected set of edges of such that is planar. We consider a rooted embedding where is a -edge-connected digon-completion of , accompanied with the resulting embedding of in , and is the result of the contraction of the edges in in . We refer to as the reduced embedding pair of . We say that a cycle of is -aligned if is a -aligned cycle of (w.r.t. ).
6.2 Deducing irrelevance from flows
Let be a rooted embedding and let be a -aligned cycle of . Given a , we define the -irrelevant set for as the vertex set
where and is the linkage function (from Subsection 5.2). In words, the -irrelevant set for is the set of all vertices of that are embedded in the interior of all cycles that are in radial distance from , for all aligned cycles that are in radial distance from and send a -flow to .
We next show that given an instance of a rooted graph problem that has the insulation property, if is a connected set of edges that spans the roots and whose contraction makes planar, then for every -aligned cycle of the removal of the -irrelevant set of yields an equivalent instance for .
Lemma \thelemma.
Let be a rooted graph problem that has the insulation property. Also, let be a -embedded rooted graph and let be a connected set of edges of such that is planar and . Then, for every -aligned cycle of , we have that and are -equivalent.
Proof.
Let be a reduced embedding pair of and let be the radial distance decomposition of . By definition, the fact that is -aligned implies that there is a non-root vertex-face edge of such that . Also, note that because of Subsection 6.1, all -aligned cycles of that correspond to vertex-face edges of for which is not a leaf, they are also cycles of .
Consider an arbitrary ordering of the cycles in . We set and for every , we set to be the rooted graph obtained from after the removal of all vertices in . Keep also in mind that is a subset of the interior of .
Because of Subsection 5.2, for every and every , . In other words, given that contains pairwise-disjoint paths between and , such paths also exist in . This allows to consider, for every , a collection of pairwise-disjoint paths from to in . Moreover, since for every , , by Subsection 3.2, there is a nested sequence of cycles of such that the inner cycle of is and the outer cycle of is . Notice now that for every , is a -railed nest of such that belongs to and is the interior of .
Observe also that the fact that is connected implies that is also a -railed nest of , where the endpoints of all edges in belong to and is the interior of . Furthermore, the fact that implies that is a subset of . Therefore, by the insulation property, we have that for every , and are -equivalent. Thus, and are -equivalent for every , which in particular means that and are -equivalent. ∎
Global irrelevant sets.
Let be a rooted embedding and let . We set . Let be the set of all -aligned cycles of whose radial distance from is equal to and we set . We call the global -irrelevant set of . It is easy to see that -edge-connectivity of implies that is also -edge-connected.
By repeatedly applying Subsection 6.2 for all cycles in , we can show that the removal of the global -irrelevant set from an instance gives an equivalent one.
Lemma \thelemma.
Let be a rooted graph problem that has the insulation property. Also, let be a -embedded rooted graph and let be a connected set of edges of such that is planar and . If is the global -irrelevant set for the reduced embedding pair , then and are -equivalent.
Proof.
Let be the radial distance decomposition of . Keep in mind that, because of Subsection 6.1, all -aligned cycles of that correspond to vertex-face edges of where is not a leaf are also cycles of .
We set . Let be the set of all vertex-face edges of so that the distance between and is equal to . Also, consider a linear ordering of such that if , , and is a descendant of , then . The latter property implies that if is in the interior of , then .
We set and for every , we set be the rooted graph obtained from after the removal of all vertices in .
In order to show that and are -equivalent, we distinguish two cases. If there is a such that contains a vertex of , then by Subsection 3.2 . This in turn implies and therefore . If for every , we have that is disjoint from then the interior of is also disjoint from for every because of the definition of the ordering. Therefore, the -irrelevant set of in and in is the same, i.e., . By applying Subsection 6.2 for the rooted graph and the cycle , we get that and are -equivalent. Thus, and are -equivalent for every , which in particular means that and are -equivalent. ∎
6.3 Finding the irrelevant vertices
We next show how to compute the -irrelevant set for each slice and the global -irrelevant set in time that is linear in the size of the split and in the size of the instance, respectively.
Lemma \thelemma.
There is an algorithm that given a rooted embedding , a -aligned cycle of , an integer , and the set of cycles , where , computes the -irrelevant set for in time , where .
Proof.
The algorithm first computes in time using the algorithm of Subsection 5.2. Then, it outputs the union of sets of vertices of in for every . ∎
Let us now describe an algorithmic procedure that, given a rooted embedding and an integer , outputs a vertex set . After presenting this procedure, we show that the output is the -global irrelevant set and that it runs in linear time.
| Algorithm global_trim | ||||||
| Input: | a rooted embedding and a . | |||||
| Output: | a vertex set . | |||||
| 1. | Compute the radial distance decomposition of and | |||||
| a function that maps every non-rooted vertex-face edge of to the cycle . | ||||||
| 2. | Let and let . | |||||
| 3. | Let be the set containing every vertex-face edge of | |||||
| where the distance in between and is . | ||||||
| 4. | Consider a linear ordering of such that | |||||
| if , and is a descendant of then . | ||||||
| 5. | Let be the first edge of . | |||||
| 6. | Let . | |||||
| 7. | Remove from the edge and every edge where . | |||||
| 8. | Let . | |||||
| 9. | If is non-empty, go to Step 5, otherwise output . |
We now show that the above algorithm outputs in linear time the global -irrelevant set.
Lemma \thelemma.
The algorithm global_trim outputs the global -irrelevant set in time .
Proof.
Let be the subset of consisting of all edges for which was added in Step 8. We claim that the union of for all is the -irrelevant set for , i.e., .
To show this, we argue that for every , there is an such that . Indeed, note that an edge was not considered in Step 8 because it was discarded in Step 7 from . Therefore, there is some such that . Since , we have that intersects the interior of a cycle . By Subsection 3.2, every vertex in the interior of is also in the interior of . Therefore every vertex of in the interior of is contained in . Since is a subset of the vertices in the interior of , we get .
We next argue that the algorithm global_trim terminates in linear time. Note that because of Subsection 3.1, we can compute the radial distance decomposition of in linear time. Also, because of Subsection 3.2, in linear time we can get a function that maps every non-rooted vertex-face edge of to the cycle . Now, for each , we set . Observe that the set can be computed in time , using the algorithm of Subsection 6.3. The claimed bound on the running time of global_trim follows by observing that . To see why the latter inequality holds, note that for every two distinct , the cycles do not intersect and and are disjoint sets. Thus . ∎
6.4 Bounding the treewidth
As mentioned in Subsection 2.3, we will show that the removal of the global irrelevant set gives a graph of bounded treewidth. In fact, we show that in a rooted embedding where there is no large nested sequence of aligned cycles with a large flow from the inner one to the outer one, the treewidth is bounded. Let us formally define the aformentioned property.
Let be a rooted embedding and let and . We say that is -thin if for every two -aligned cycles of where is embedded in the interior of and their radial distance is at least it holds that .
Lemma \thelemma.
For every rooted embedding , if is -thin, then .
Proof.
Before proceeding to the proof, let us sketch its main ideas. Supposing towards a contradiction that the treewidth of is large enough, we get a large railed nest. While the cycles of the railed nest are not guaranteed to be -aligned, we show that -aligned cycles are “roughly parallel” to the cycles of . This is based on the observation that if a -aligned cycle crosses many cycles of , then the same will happen to an other -aligned cycle in an appropriate radial distance. In this case, (sub)paths of cycles of will give a large flow between and , contradicting thinness. Therefore, many -aligned cycles are “roughly parallel” to the cycles of the railed nest, and in this case the paths of the nest certify the existence of a large flow, contradicting thinness.
We set . Assume towards a contradiction that has treewidth at least , where is the constant of Subsection 2.1. Then, from Subsection 2.1, applied on the rooted graph and for , we have that contains some -railed nest whose cycle collection is , whose path collection is , and such that the interior of contains some vertex . For , we denote by the open disk bounded by where is embedded.
Clearly, the radial distance between and should be at least as there are at least cycles separating and . Let be the radial distance decomposition of . Let such that and consider the path in between and . Clearly, this path has length at least and contains at least vertex-face edges. Among them let , for some , be those for which (1) the cycle is not embedded in , (2) are as close as possible to the vertex in , and (3) are ordered so that those that are closer to appear earlier. Also we consider be the sequence of nested cycles so that . By the definition of the radial distance decomposition, if , then the radial distance between and is .
For every cycle we denote by the set and by the set . Intuitively, encodes the set of all cycles of that intersects and is the difference (in indices) between the first and the last cycle of that intersects. Observe the following
| (1) | |||||
| (2) | |||||
| (3) |
Let us comment how to get these observations: (1) follows because the radial distance between and is . (2) follows because is embedded in and . To see (3) it is enough to observe that for every vertex of there is a vertex of within radial distance from and this, in turn, implies that is upper-bounded by the half of the radial distance between and .
We next show that the range of cycle in is small.
Claim: for every , it holds that .
Proof of claim: Suppose that . We set . Due to (1), . Observe that because of (2) and (3), for every , it holds that . This in turn implies that each of the cycles in intersect each of the pairwise disjoint cycles in . This certifies that there is a flow of size from to , a contradiction to the fact that is -thin, as the radial distance between and is .
We next set to be minimum such that . This means that either , or . From the above Claim, we have that , which implies that the whole is embedded outside the open disk . Consider the cycles and observe that, from (3), . By the above Claim, we also have . Therefore,
We conclude that is embedded inside the closure of the disk . Then any of the paths of should intersect all cycles . Therefore there is a flow of size from to , a contradiction to the fact that is -thin, as the radial distance between and is . ∎
We next show that the removal of the global -irrelevant set from yields an -thin graph, where . Then, due to Subsection 6.4, it follows that is a linear function of ; recall that is the linkage function from Subsection 5.2.
Lemma \thelemma.
Let be a rooted embedding and let . If is the global -irrelevant set for , then .
Proof.
We set . We show that is -thin, which together with Subsection 6.4 implies that . To see why is -thin, suppose towards a contradiction that there exist two -aligned cycles of with the following properties: , their radial distance is at least , and . Then, there exist two -aligned cycles of such that and the radial distance from both and to is equal to . Moreover, note that implies that , a contradiction to the assumption that is a cycle of . ∎
7 Dealing with surface-embeddable multi-rooted graphs
In this section, we wrap-up the proof of Theorem 2.1 by showing how to deal with the general case of surface-embeddable multi-rooted graphs. We first show how to find a connected set of edges of a surface-embeddable whose contraction yields a planar graph. We then show how to enhance this set in order to guarantee that the roots of are spanned by , while maintaining the property that if the treewidth of is bounded then the treewidth of is also bounded. We conclude this section by assembling the proof of Theorem 2.1.
7.1 From surfaces to spheres
We will use the following result that is implicit in the proof of [11, Lemma 8].
Proposition \theproposition.
There is an algorithm that, given a connected graph embedded in a surface and a vertex , outputs, in linear time, a collection of shortest paths of with the following properties:
-
•
-
•
the paths in have as a common endpoint, and
-
•
is an open disk.
While [11, Lemma 8] does not explicitly state that is an open disk, this can be easily derived by its proof. In fact, in the proof of [11, Lemma 8], the authors construct by picking a subgraph of a cut graph that is, in turn, a subgraph of , where is the tree–cotree decomposition of Eppstein [19]. Since when cutting open the surface along , they obtain a disk, it is easy to observe that the removal of from also creates an open disk.
We next enhance Subsection 7.1 by showing that by taking extra shortest paths that span a prescribed set of vertices, we can get a collection of shortest paths whose contraction makes the graph planar.
Lemma \thelemma.
There is an algorithm that, given a -embedded connected graph and a set , outputs, in linear time, a collection of shortest paths of with the following properties:
-
•
,
-
•
is a connected subgraph of whose vertex set contains , and
-
•
is planar.
Proof.
The algorithm works as follows. It first finds a collection of shortest paths among the vertices of , by fixing an arbitrary vertex and computing a BFS tree with as root. Then, it invokes the algorithm of Subsection 7.1 with the graph and the vertex as inputs, which gives a collection of shortest paths with as a common endpoint. The latter algorithm guarantees that and that is an open disk. It is therefore easy to see that is a collection of shortest paths that satisfies the first two properties in the statement of the lemma. Let be the set of edges of all paths in ; we claim that is planar. Indeed, consider the graph obtained from by subdividing once each edge of that is incident to a vertex of some path in . Since is an open disk, is embedded in a closed isk with all the subdivision vertices, i.e., the vertices in lying on the boundary of . Therefore, by introducing a new vertex that is adjacent to all these vertices in the boundary of , we get a graph that is planar and isomorphic to . Planarity of implies that if we further contract the edges of all paths in we still get a planar graph. ∎
As we explain in the next subsection, the above result (Subsection 7.1) will be applied for radial graphs. Therefore, we need to relate the contraction of paths in the radial to the contraction of edges in the original graph . For this, we have to associate edges of the radial graph have to be contracted to edges of that have to be contracted in order to achieve planarity in , while not changing significantly the radial distances in .
Let be a -embeddable graph and let be a subgraph of its radial graph . Let be the set of all vertices of whose distance from some vertex of is at most two. We denote by the set of all edges of with both endpoints in . Also we denote by the set of all edges of where both their endpoints belong in . See Fig. 4 for an illustration of the above notions.
Lemma \thelemma.
Let be -embeddable graph and let be a subgraph of its radial graph . Then the following hold:
-
1.
If is connected then is a connected set of edges in .
-
2.
If is planar, then is planar.
-
3.
Let be a vertex in and let be a vertex in . Then the radial distance between and is at least four.
Proof.
We use to denote the set and to denote the set of all vertices of that are in distance two from a vertex of in ; note that is a partition of . Also, keep in mind that .
We first show that is the radial graph of . Let be the subgraph of induced by the edges of . Observe that when contracting all edges of in , every connected component of becomes a single vertex. Each such vertex is incident to the faces of that were adjacent to some vertex in . Similarly, the contraction of the edges of in the radial graph gives a set of new vertices, each adjacent to the vertices of in the neighborhood of the vertices of . In order to derive that is the radial graph of , observe that there is a bijection between the vertices of that are result of the contraction of the edges from and the vertices of that are result of the contraction of the edges from .
Property (1) follows from the fact that and that if is connected, then is connected.
Property (2) follows by observing that is a minor of , combined with the fact that, as shown above, is the radial graph of and that if the radial graph of a surface-embedded graph is planar, then the graph itself is planar.
For Property (3), note that since each vertex in is in radial distance at least two from every vertex in and every vertex in is in radial distance at least two from any vertex in , we get that the radial distance between and is at least four. ∎
7.2 Maintaining the bound on treewidth
In order to prove that the contraction of the edges of the shortest radial paths given in the previous subsection does not yield big changes in the treewidth of the graph, we will use a result obtained from [17]. This states that in a surface-embedded graph, if it contains a large enough wall whose perimeter bounds a closed disk where is embedded, every collection of shortest radial paths with terminals outside this disk is guaranteed to not touch a sufficiently central part of the wall. Let us give formal definitions of layers and central subwalls of walls.
Layers of walls and central subwalls.
Let be an odd positive integer. The layers of an -wall are recursively defined as follows. The first layer of is its perimeter. For , the th layer of is the th layer of the wall obtained from after removing from its perimeter and removing recursively all consequent vertices of degree one. Notice that each wall has layers. Given an , a wall of layers is a -wall.
Given an -wall and an odd where , we define the central -subwall of to be the -wall obtained from after removing its first layers and all consequent vertices of degree one.
Given a graph that contains as a subgraph a wall whose boundary is a contractible cycle bounding a closed disk where the entire is embedded, we call compass of in the graph .
The following result can be easily extracted from the proof of [17, Theorem 6].
Proposition \theproposition.
There is a function such that for every the following holds. Let be a -embedded graph that contains as a subgraph an -wall whose perimeter is a contractible cycle bounding a closed disk where the entire is embedded. If is a collection of shortest radial paths in whose endpoints are disjoint from the compass of , then the vertex set of each path in is disjoint from the compass of the -central wall of . Moreover, is a linear function.
The additional assumption that is a linear function follows from [17, Theorem 6] and [18, Lemma 4.7].
We are now ready to show the next result which gives a linear-time reduction from planar single-rooted to surface-embeddable multi-rooted graphs.
Lemma \thelemma.
There is an algorithm that, given a -embedded graph and a set , outputs, in linear time, a collection of many shortest radial paths in such that, given that , the following hold:
-
•
,
-
•
is connected in ,
-
•
is planar, and
-
•
.
Proof.
The algorithm invokes the algorithm of Subsection 7.1 with the radial graph of and the vertex set as input; note that the radial graph of every -embedded graph is a connected graph. Therefore, in linear time, we get a collection of many shortest paths of such that if we set then is a connected subgraph of whose vertex set contains and is planar. Since and (by definition), we get that . By (1) of Subsection 7.1 we get that connectivity of implies that is a connected set of edges in . Also, since is planar, we have that is also planar (due to (2) of Subsection 7.1).
To show that if , then , we argue as follows. Let and , where is the function from Subsection 7.2.
Assume towards a contradiction that , where is the constant of Subsection 4.1. Then, contains as a subgraph a -wall whose perimeter is a contractible cycle bounding a closed disk where the entire is embedded such that does not contain any of the endpoints of the paths in . Also, since , by Subsection 7.2, we get that no vertex of is a vertex of the compass of the -central wall of . By (3) of Subsection 7.1, we have that is disjoint from the vertex set of the compass of the -central subwall of . Therefore, since , is a -wall in , which implies that , a contradiction. ∎
7.3 Wrap-up: proof of Theorem 2.1
We are now ready to prove Theorem 2.1.
Proof of Theorem 2.1.
We first apply Subsection 7.2 and compute a collection of shortest radial paths in such that if and , then is a connected set of edges in such that is planar, , and . We next consider the reduced embedding pair of obtained from by contracting all edges in (this way is defined as the result of these contractions) and then taking the -edge-connected digon-completion of . We compute the global -irrelevant set of using the algorithm global_trim, which runs in linear time as showed in Subsection 6.3. Due to Subsection 6.2, and are -equivalent. Also, by Subsection 6.4, the treewidth of depends only on and therefore is constant. Since is a subgraph of , then and therefore . ∎
8 Problems with the insulation property
Instead of enumerating problems that have the insulation property, we prefer to give two general families that contain most of them. They express several containment and/or modification problems and can be all solved in linear time for graphs of bounded genus using our results. We need first some definitions.
A minor model of a rooted graph in a rooted graph is a set of pairwise-disjoint connected vertex sets of where , for every , and for every two vertices of
| (4) |
If we replace by in (4) then we say that is an induced minor model of in . We say that is a minor (resp. induced minor) of if there is a minor (resp. induced minor) model of in . For every finite set of rooted graphs and every integer , we define the following problem.
-Minor -Deletion
Input: a rooted graph
Question: is there a set where and such that none
of the rooted graphs in is a minor of ?
We also denote by -Induced Minor -Deletion the problem defined as above if we use induced minors instead of minors. When restricted to (rooted) graphs of bounded genus, both above problems, and all problems that can be seen as special cases of them, have the insulation property. We next give a brief explanation of how this follows from known combinatorial results on the existence of irrelevant vertices.
We first consider -Minor -Deletion and we start with the case where . This expresses the Rooted Minor Containment problem, asking whether contains as a minor some of the rooted graphs in . The insulation property for this problem on surface-embeddable graphs follows from [7, Theorem 5.9], which in case of planar graphs, can be seen as a direct consequence of [7, Theorem 5.6]. We stress that [7, Theorem 5.6] is, in turn, making use of the “combing lemma” [30, Theorem 1.4], which is based on the linkage theorem [54, 53] (see [3] for the case of planar graphs and in [50] for the case of surface-embeddable graphs). For surface-embeddable graphs, a version of [6, Theorem 5.6] can be proven for the induced minor setting as well by using the “scattered version” of the combing lemma [30, Corollary 3.6], which in turn is based on the ideas of the “induced linkage” results of Kawarabayashi and Kobayashi in [39]. This implies that also the Rooted Induced Minor Containment problem, which is -Induced Minor -Deletion for , has the insulation property. When , the insulation property for -Minor -Deletion follows from [64, Lemma 16], specialized on surface embeddable graphs. By using the same arguments and having as departure point the induced minor version of [7, Theorem 5.6], the insulation property follows for -Induced Minor -Deletion as well.
Notice that -Minor -Deletion contains most of the problems where an (almost) linear time algorithm implementation of the irrelevant technique is known. For instance, Disjoint Paths is derived when and contains only the graph consisting of roots with a matching between them. Also, Rooted Minor Containment follows if and consists of the minor we are looking for. Moreover, Planarization follows if contains the two Kuratowski graphs. For the general case where is a set of graphs, algorithms for the -Minor -Deletion have been proposed in [62] (running in cubic time) and [52] (running in quadratic time). For bounded genus graphs, this problem as well as its rooted and/or induced version can be solved in linear time, because of our results.
We wish to stress that in the induced minor case few results for the applicability of the irrelevant vertex technique are known. To our knowledge the only problem of this kind that has been examined is the Induced Cycle problem for surface embeddable graphs in [41, 40]. Interestingly, the authors of [20] proved that there is a planar graph where asking for the induced minor containment of is an NP-complete problem. As this is the -Induced Minor -Deletion problem, no polynomial algorithm may be expected in the induced setting in general graphs.
At this point we would like to mention a family of problems with the insulation property that extends further the above framework. It is the -Minor -Elimination Distance problem, introduced by Bulian and Dawar [9, 10] that is defined as -Minor -Deletion where is a finite set of graphs and with the difference that instead of looking for a set where we ask that the tree-depth of is at most .222For a vertex set , the torso of in , denoted by , is the graph obtained from the induced subgraph by turning , the set of vertices in adjacent to each connected component of , into a clique. The tree-depth of a graph is the minimum height of a forest with the property that every edge of connects a pair of nodes that have an ancestor-descendant relationship to each other in . Using similar arguments as those in [64, Lemma 16] in the case of -Minor -Deletion, in [52] it was proved that -Minor -Elimination Distance also has the insulation property. In fact, the same arguments can be applied in the more general setting where, instead of tree-depth, we consider any minor-monotone graph parameter that is functionally bigger than treewidth.
9 Open problems
In this paper we gave an algorithmic framework for the linear-time implementation of the irrelevant vertex technique for a wide family of problems, namely the problems that have the insulation property. As we already mentioned in the introduction, the applicability of the irrelevant vertex technique extends to more general problems that have the following “weak” version of the insulation property:
There is a constant so that for every -embedded rooted graph with a -railed nest , if then some vertex in is -irrelevant.
A typical problem with the above weak insulation property is the -Topological Minor Containment problem where, given a graph , we ask whether it contains a subdivision of as a subgraph. This problem can be extended to the -Rooted Topological Minor Containment problem if we correspond a bounded number of vertices of to each of the roots of . More generally, most known problems with the above weak insulation property are captured by the meta-algorithmic frameworks developed in [32, 22, 23, 66, 65]. These algorithmic meta-theorems apply for graphs of fixed Hadwiger number333The Hadwiger number of a graph is the maximum size of a clique minor in it. and they are based on an extended notion railed-nest that is applicable to flat territories of the input graph instead of disk-embeddable ones. However, even in planar graphs, the fastest algorithm for such problems runs in quadratic time and is based on the repetitive finding (in linear-time) of an irrelevant vertex. To improve the algorithms produced by [32, 22, 23, 66, 65] to sub-quadratic ones is an open challenge even for the case of planar graphs. An important step in this direction would be to prove that -Rooted Topological Minor Containment can be solved in linear time in planar graphs.
Another direction for improvement is to extend Theorem 2.1 beyond the class of bounded-genus graphs. The target here would be again graphs of bounded Hadwiger number. In this direction we believe that our technique is applicable in linear time if we are given the decomposition of the Graph Minors Structural Theorem (GMST) which exists for such graphs [58]. However the currently faster algorithm for the computation of this decomposition is the one of Grohe and Kawarabayashi [34] that runs in quadratic time.
References
- [1] Isolde Adler, Frederic Dorn, Fedor V. Fomin, Ignasi Sau, and Dimitrios M. Thilikos. Fast minor testing in planar graphs. Algorithmica, 64(1):69–84, 2012. doi:10.1007/S00453-011-9563-9.
- [2] Isolde Adler, Martin Grohe, and Stephan Kreutzer. Computing excluded minors. In Proc. of the 19th Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 641–650, 2008. URL: https://dl.acm.org/doi/10.5555/1347082.1347153.
- [3] Isolde Adler, Stavros G. Kolliopoulos, Philipp Klaus Krause, Daniel Lokshtanov, Saket Saurabh, and Dimitrios M. Thilikos. Irrelevant vertices for the planar disjoint paths problem. Journal of Combinatorial Theory, Series B, 122:815–843, 2017. doi:10.1016/j.jctb.2016.10.001.
- [4] Isolde Adler and Philipp Klaus Krause. A lower bound on the tree-width of graphs with irrelevant vertices. Journal of Combinatorial Theory, Series B, 137:126–136, 2019. doi:10.1016/J.JCTB.2018.12.008.
- [5] Stefan Arnborg, Jens Lagergren, and Detlef Seese. Easy Problems for Tree-Decomposable Graphs. Journal of Algorithms, 12(2):308–340, 1991. doi:10.1016/0196-6774(91)90006-K.
- [6] Julien Baste, Ignasi Sau, and Dimitrios M. Thilikos. Hitting minors on bounded treewidth graphs. I. General upper bounds. SIAM Journal on Discrete Mathematics, 34(3):1623–1648, 2020. doi:10.1137/19M1287146.
- [7] Julien Baste, Ignasi Sau, and Dimitrios M. Thilikos. Hitting minors on bounded treewidth graphs. IV. An optimal algorithm. SIAM Journal on Computing, 52(4):865–912, 2023. doi:10.1137/21m140482x.
- [8] Richard B. Borie, R. Gary Parker, and Craig A. Tovey. Automatic generation of linear-time algorithms from predicate calculus descriptions of problems on recursively constructed graph families. Algorithmica, 7(5-6):555–581, 1992. doi:10.1007/BF01758777.
- [9] Jannis Bulian and Anuj Dawar. Graph isomorphism parameterized by elimination distance to bounded degree. Algorithmica, 75(2):363–382, 2016. doi:10.1007/s00453-015-0045-3.
- [10] Jannis Bulian and Anuj Dawar. Fixed-parameter tractable distances to sparse graph classes. Algorithmica, 79(1):139–158, 2017. doi:10.1007/s00453-016-0235-7.
- [11] Sergio Cabello, Éric Colin de Verdière, and Francis Lazarus. Algorithms for the edge-width of an embedded graph. Computational Geometry, 45(5):215–224, 2012. Special issue: 26th Annual Symposium on Computation Geometry at Snowbird, Utah, USA. doi:10.1016/j.comgeo.2011.12.002.
- [12] Zhi-Zhong Chen, Michelangelo Grigni, and Christos H. Papadimitriou. Map graphs. Journal of the ACM, 49(2):127–138, 2002. doi:10.1145/506147.506148.
- [13] Julia Chuzhoy and Zihan Tan. Towards tight(er) bounds for the excluded grid theorem. Journal of Combinatorial Theory, Series B, 146:219–265, 2021. doi:10.1016/J.JCTB.2020.09.010.
- [14] Bruno Courcelle. The monadic second-order logic of graphs. I. recognizable sets of finite graphs. Information and Computation, 85(1):12–75, 1990. doi:10.1016/0890-5401(90)90043-H.
- [15] Bruno Courcelle. The monadic second-order logic of graphs III: tree-decompositions, minors and complexity issues. RAIRO - Theoretical Informatics and Applications, 26:257–286, 1992. doi:10.1051/ita/1992260302571.
- [16] Bruno Courcelle. The expression of graph properties and graph transformations in monadic second-order logic. In Handbook of Graph Grammars and Computing by Graph Transformations, Volume 1: Foundations, pages 313–400. World Scientific, 1997.
- [17] Erik D. Demaine, MohammadTaghi Hajiaghayi, and Ken-ichi Kawarabayashi. Contraction decomposition in h-minor-free graphs and algorithmic applications. In Proc. of the 43rd ACM Symposium on Theory of Computing (STOC), pages 441–450. ACM, 2011. doi:10.1145/1993636.1993696.
- [18] Erik D. Demaine, MohammadTaghi Hajiaghayi, and Dimitrios M. Thilikos. The bidimensional theory of bounded-genus graphs. SIAM Journal on Discrete Mathematics, 20(2):357–371, 2006. doi:10.1137/040616929.
- [19] David Eppstein. Dynamic generators of topologically embedded graphs. In Proc. of the 14th Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), page 599–608. SIAM, 2003. doi:10.5555/644108.644208.
- [20] Michael R. Fellows, Jan Kratochvíl, Matthias Middendorf, and Frank Pfeiffer. The complexity of induced minors and related problems. Algorithmica, 13(3):266–282, 1995. doi:10.1007/BF01190507.
- [21] Fedor V. Fomin, Petr A. Golovach, Tuukka Korhonen, and Giannos Stamoulis. Computing paths of large rank in planar frameworks deterministically. In Proc. of the 34th International Symposium on Algorithms and Computation (ISAAC), volume 283 of LIPIcs, pages 32:1–32:15. Schloss Dagstuhl - Leibniz-Zentrum für Informatik, 2023. doi:10.4230/LIPICS.ISAAC.2023.32.
- [22] Fedor V. Fomin, Petr A. Golovach, Ignasi Sau, Giannos Stamoulis, and Dimitrios M. Thilikos. Compound Logics for Modification Problems. In Proc. of the 50th International Colloquium on Automata, Languages, and Programming (ICALP), volume 261 of Leibniz International Proceedings in Informatics (LIPIcs), pages 61:1–61:21, Dagstuhl, Germany, 2023. Schloss Dagstuhl – Leibniz-Zentrum für Informatik. doi:10.4230/LIPIcs.ICALP.2023.61.
- [23] Fedor V. Fomin, Petr A. Golovach, Giannos Stamoulis, and Dimitrios M. Thilikos. An algorithmic meta-theorem for graph modification to planarity and FOL. ACM Transactions on Computation Theory, 14(3-4):1–29, 2022. doi:10.1145/3571278.
- [24] Fedor V. Fomin, Daniel Lokshtanov, Fahad Panolan, Saket Saurabh, and Meirav Zehavi. Hitting topological minors is FPT. In Proc. of the 52nd Annual ACM SIGACT Symposium on Theory of Computing (STOC), pages 1317–1326, 2020. arXiv version: 1904.02944. doi:10.1145/3357713.3384318.
- [25] Fedor V. Fomin, Daniel Lokshtanov, Saket Saurabh, and Dimitrios M. Thilikos. Linear kernels for (connected) dominating set on H-minor-free graphs. In Proc. of the Twenty-Third Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 82–93, 2012. doi:10.1137/1.9781611973099.7.
- [26] Fedor V. Fomin and Dimitrios M. Thilikos. Dominating sets in planar graphs: Branch-width and exponential speed-up. SIAM Journal on Computing, 36(2):281–309, 2006. doi:10.1137/S0097539702419649.
- [27] James F. Geelen, R. Bruce Richter, and Gelasio Salazar. Embedding grids in surfaces. European Journal of Combinatorics, 25(6):785–792, 2004. doi:10.1016/J.EJC.2003.07.007.
- [28] Petr A. Golovach, Marcin Kaminski, Spyridon Maniatis, and Dimitrios M. Thilikos. The parameterized complexity of graph cyclability. SIAM Journal on Discrete Mathematics, 31(1):511–541, 2017. doi:10.1137/141000014.
- [29] Petr A. Golovach, Marcin Kaminski, Daniël Paulusma, and Dimitrios M. Thilikos. Induced packing of odd cycles in planar graphs. Theoretical Computer Science, 420:28–35, 2012. doi:10.1016/j.tcs.2011.11.004.
- [30] Petr A. Golovach, Giannos Stamoulis, and Dimitrios M. Thilikos. Combing a linkage in an annulus. SIAM Journal on Discrete Mathematics, 37(4):2332–2364, 2023. doi:10.1137/22M150914X.
- [31] Petr A. Golovach, Giannos Stamoulis, and Dimitrios M. Thilikos. Hitting topological minor models in planar graphs is fixed parameter tractable. ACM Transactions on Algorithms, 19(3):23:1–23:29, 2023. doi:10.1145/3583688.
- [32] Petr A. Golovach, Giannos Stamoulis, and Dimitrios M. Thilikos. Model-checking for first-order logic with disjoint paths predicates in proper minor-closed graph classes. In Proc. of the 2023 ACM-SIAM Symposium on Discrete Algorithms, (SODA 2023), pages 3684–3699. SIAM, 2023. doi:10.1137/1.9781611977554.ch141.
- [33] Martin Grohe, Ken-ichi Kawarabayashi, Dániel Marx, and Paul Wollan. Finding topological subgraphs is fixed-parameter tractable. In Proc. of the 43rd ACM Symposium on Theory of Computing (STOC), pages 479–488. ACM, 2011. doi:10.1145/1993636.1993700.
- [34] Martin Grohe, Ken-ichi Kawarabayashi, and Bruce A. Reed. A simple algorithm for the graph minor decomposition - logic meets structural graph theory. In Proc. of the 24th Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 414–431. SIAM, 2013. doi:10.1137/1.9781611973105.30.
- [35] Robert Hickingbotham and David R. Wood. Structural properties of graph products. Journal of Graph Theory, 2023. doi:10.1002/jgt.23023.
- [36] Bart M. P. Jansen, Daniel Lokshtanov, and Saket Saurabh. A near-optimal planarization algorithm. In Proc. of the 25th Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 1802–1811. SIAM, 2014. doi:10.1137/1.9781611973402.130.
- [37] Marcin Kamiński and Dimitrios M. Thilikos. Contraction checking in graphs on surfaces. In Proc. of the 29th International Symposium on Theoretical Aspects of Computer Science (STACS), volume 14 of LIPIcs, pages 182–193, 2012. doi:10.4230/LIPIcs.STACS.2012.182.
- [38] Ken-ichi Kawarabayashi. Planarity allowing few error vertices in linear time. In Proc. of the 50th Annual IEEE Symposium on Foundations of Computer Science (FOCS), pages 639–648, 2009. doi:10.1109/FOCS.2009.45.
- [39] Ken-ichi Kawarabayashi and Yusuke Kobayashi. The induced disjoint paths problem. In Proc. of the 13th International Conference on Integer Programming and Combinatorial Optimization (IPCO), volume 5035 of Lecture Notes in Computer Science, pages 47–61. Springer, 2008. doi:10.1007/978-3-540-68891-4\_4.
- [40] Ken-ichi Kawarabayashi and Yusuke Kobayashi. Algorithms for finding an induced cycle in planar graphs and bounded genus graphs. In Proc. of the Twentieth Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 1146–1155. SIAM, 2009. doi:10.1137/1.9781611973068.124.
- [41] Ken-ichi Kawarabayashi and Yusuke Kobayashi. Algorithms for finding an induced cycle in planar graphs. Combinatorica, 30(6):715–734, 2010. doi:10.1007/S00493-010-2499-X.
- [42] Ken-ichi Kawarabayashi, Yusuke Kobayashi, and Bruce Reed. The disjoint paths problem in quadratic time. Journal of Combinatorial Theory, Series B, 2011. doi:10.1016/j.jctb.2011.07.004.
- [43] Ken-ichi Kawarabayashi, Stephan Kreutzer, and Bojan Mohar. Linkless and flat embeddings in 3-space. Discrete & Computational Geometry, 47(4):731–755, 2012. doi:10.1007/s00454-012-9413-9.
- [44] Ken-ichi Kawarabayashi, Bojan Mohar, and Bruce A. Reed. A simpler linear time algorithm for embedding graphs into an arbitrary surface and the genus of graphs of bounded tree-width. In Proc. of the 49th Annual IEEE Symposium on Foundations of Computer Science (FOCS), pages 771–780, 2008. doi:10.1109/FOCS.2008.53.
- [45] Ken-ichi Kawarabayashi and Bruce A. Reed. Computing crossing number in linear time. In Proc. of the 39th annual ACM symposium on Theory of computing (STOC), pages 382–390, 2007. doi:10.1145/1250790.1250848.
- [46] Tomasz Kociumaka and Marcin Pilipczuk. Deleting Vertices to Graphs of Bounded Genus. Algorithmica, 81(9):3655–3691, 2019. doi:10.1007/s00453-019-00592-7.
- [47] Tuukka Korhonen, Konrad Majewski, Wojciech Nadara, Michał Pilipczuk, and Marek Sokołowski. Dynamic treewidth. In Proc. of the 64th IEEE Annual Symposium on Foundations of Computer Science (FOCS), pages 1734–1744. IEEE, 2023. doi:10.1109/FOCS57990.2023.00105.
- [48] Tuukka Korhonen, Michał Pilipczuk, and Giannos Stamoulis. Minor containment and disjoint paths in almost-linear time, 2024. arXiv:2404.03958.
- [49] Dániel Marx and Ildikó Schlotter. Obtaining a planar graph by vertex deletion. Algorithmica, 62(3-4):807–822, 2012. doi:10.1007/s00453-010-9484-z.
- [50] Frédéric Mazoit. A single exponential bound for the redundant vertex theorem on surfaces. arXiv preprint arXiv:1309.7820, 2013. arXiv:1309.7820.
- [51] Bojan Mohar. A linear time algorithm for embedding graphs in an arbitrary surface. SIAM Journal on Discrete Mathematics, 12(1):6–26, 1999. doi:10.1137/S089548019529248X.
- [52] Laure Morelle, Ignasi Sau, Giannos Stamoulis, and Dimitrios M. Thilikos. Faster Parameterized Algorithms for Modification Problems to Minor-Closed Classes. In Proc. of the 50th International Colloquium on Automata, Languages, and Programming (ICALP), volume 261 of Leibniz International Proceedings in Informatics (LIPIcs), pages 93:1–93:19, Dagstuhl, Germany, 2023. Schloss Dagstuhl – Leibniz-Zentrum für Informatik. doi:10.4230/LIPIcs.ICALP.2023.93.
- [53] Bruce A. Reed. Rooted routing in the plane. Discrete Applied Mathematics, 57(2-3):213–227, 1995. doi:10.1016/0166-218X(94)00104-L.
- [54] Bruce A. Reed, Neil Robertson, Alexander Schrijver, and Paul D. Seymour. Finding disjoint trees in planar graphs in linear time. In Neil Robertson and Paul D. Seymour, editors, Graph Structure Theory, Proceedings of a AMS-IMS-SIAM Joint Summer Research Conference on Graph Minors held June 22 to July 5, 1991, at the University of Washington, Seattle, USA, volume 147 of Contemporary Mathematics, pages 295–301. American Mathematical Society, 1991.
- [55] Neil Robertson and Paul D. Seymour. Graph minors. III. Planar tree-width. Journal of Combinatorial Theory, Series B, 36(1):49–64, 1984. doi:10.1016/0095-8956(84)90013-3.
- [56] Neil Robertson and Paul D. Seymour. Graph minors. X. Obstructions to tree-decomposition. Journal of Combinatorial Theory, Series B, 52(2):153–190, 1991. doi:10.1016/0095-8956(91)90061-N.
- [57] Neil Robertson and Paul D. Seymour. Graph Minors. XIII. The Disjoint Paths Problem. Journal of Combinatorial Theory, Series B, 63(1):65–110, 1995. doi:10.1006/jctb.1995.1006.
- [58] Neil Robertson and Paul D. Seymour. Graph minors. XVI. Excluding a non-planar graph. Journal of Combinatorial Theory, Series B, 89(1):43–76, 2003. doi:10.1016/S0095-8956(03)00042-X.
- [59] Neil Robertson and Paul D. Seymour. Graph Minors. XXI. Graphs with unique linkages. Journal of Combinatorial Theory, Series B, 99(3):583–616, 2009. doi:10.1016/j.jctb.2008.08.003.
- [60] Neil Robertson and Paul D. Seymour. Graph Minors. XXII. Irrelevant vertices in linkage problems. Journal of Combinatorial Theory, Series B, 102(2):530–563, 2012. doi:10.1016/j.jctb.2007.12.007.
- [61] Neil Robertson, Paul D. Seymour, and Robin Thomas. Quickly excluding a planar graph. Journal of Combinatorial Theory, Series B, 62(2):323–348, 1994. doi:10.1006/jctb.1994.1073.
- [62] Ignasi Sau, Giannos Stamoulis, and Dimitrios M. Thilikos. An FPT-Algorithm for Recognizing -Apices of Minor-Closed Graph Classes. In Proc. of the 47th International Colloquium on Automata, Languages, and Programming (ICALP), volume 168 of LIPIcs, pages 95:1–95:20, 2020. doi:10.4230/LIPIcs.ICALP.2020.95.
- [63] Ignasi Sau, Giannos Stamoulis, and Dimitrios M. Thilikos. -apices of minor-closed graph classes. II. Parameterized algorithms. ACM Transactions on Algorithms, 18(3):21:1–21:30, 2022. doi:10.1145/3519028.
- [64] Ignasi Sau, Giannos Stamoulis, and Dimitrios M. Thilikos. -apices of minor-closed graph classes. I. Bounding the obstructions. Journal of Combinatorial Theory, Series B, 161:180–227, 2023. doi:10.1016/j.jctb.2023.02.012.
- [65] Ignasi Sau, Giannos Stamoulis, and Dimitrios M. Thilikos. Parameterizing the quantification of CMSO: model checking on minor-closed graph classes, 2024. arXiv:2406.18465.
- [66] Nicole Schirrmacher, Sebastian Siebertz, Giannos Stamoulis, Dimitrios M. Thilikos, and Alexandre Vigny. Model checking disjoint-paths logic on topological-minor-free graph classes. In Proc. of the 39th Annual ACM/IEEE Symposium on Logic in Computer Science (LICS), pages 68:1–68:12. ACM, 2024. doi:10.1145/3661814.3662089.
- [67] Koichi Yamazaki, Hans L. Bodlaender, Babette de Fluiter, and Dimitrios M. Thilikos. Isomorphism for graphs of bounded distance width. Algorithmica, 24(2):105–127, 1999. doi:10.1007/PL00009273.
Appendix A Proof of Subsection 3.2
In the proof of Subsection 3.2 we will use the fact that there is a exactly one root edge in the tree of the radial distance decomposition of every rooted embedding.
Lemma \thelemma.
Let be the radial distance decomposition of a rooted embedding . Then the root of has degree one in .
Proof.
Recall that . We show that the graph is connected, which by the definition of the distance decomposition will imply that has exactly one child in . Since is a -edge-connected -embedded graph, by Subsection 3.1 we have that every face of is incident to exactly four vertices. Let be the set of all edges of that are contained in faces of incident to , except from the edges that are incident to . Observe that since is planar and every face of is incident to exactly four vertices of , there is a cycle in whose edge set is a subset of . It is now easy to see that every path with vertices in can be rerouted via to a path from to that avoids . ∎
Duals of surface embedded graphs.
Let be a surface and let be a -embedded graph. The dual graph of , denoted by , is the graph whose vertex set is the set of faces of and where two faces are adjacent if they share an edge of . For every , we denote by the edge dual to , which connects the two faces adjacent to in the embedding of . For every , we use to denote the vertex of corresponding to the face of . The dual graph has a natural embedding in , where each vertex is assigned to a point in the interior of the face and for each edge , incident to faces and , the dual edge is assigned to a curve that connects the points and and crosses precisely at the edge . Given a set of edges , we use to denote the set .
Connected sets of faces.
Let be a -embedded graph and let be a subset of its faces. We say that is connected in if for every two faces in there is a -avoiding arc starting from a point in and finishing to a point in and not containing points from a face outside . Observe that is connected in if and only if is connected. Keep in mind that the empty subset of is connected in (as the definition is vacuously satisfied).
Prefixes and suffixes.
Let be the radial distance decomposition of a rooted embedding . Given a node , we define as the set of all nodes that are descendants of (including ), while we define as the subtree of that is obtained from after removing the children of and all their descendants. We also define (resp. ) as the union of all where is a node in (resp. ). Given an edge of (in this case is a child of ), recall that we use to denote the union of all , where is either or a descendant of in . Also, we use to denote the union of all , where belongs to the connected component of that contains .
In order to prove Subsection 3.2, we start by showing that the set of faces in the suffix of a vertex-face edge is connected.
Lemma \thelemma.
Let be the radial distance decomposition of a rooted embedding . For every vertex-face edge , the set is connected in .
Proof.
We use as a shortcut for . Keep in mind that is a node in a face layer of , the set of all vertices of that are adjacent to vertices from but do not belong to are vertices of , and is connected. We set and show that is connected in . Let be two faces of . It is enough to prove that there is a path connecting and in . Since is connected, there exists a path in whose vertices are , starting from and finishing at ; here, correspond to faces of and to vertices of . Also, since , for every , both and all neighbors of in are in . Observe that for every , the set of all faces of incident to is connected in . Thus, there exists a path in between and for every and therefore also a path connecting and in . ∎
We next show that the set of faces in the prefix of a vertex-face edge is connected.
Lemma \thelemma.
Let be the radial distance decomposition of a rooted embedding . For every vertex-face edge , the set is connected in .
Proof.
Let be the set of all vertex-face edges of such that is not connected in . Consider such an edge such that has minimum distance (among all other edges in ) from the root of . We can assume that , since otherwise by Appendix A is the only child of and the empty set is trivially a connected set of faces in .
Let be the vertex-face edge of such that is the parent of in . By minimality of , we have that is connected in . We will argue that is also connected in , arriving to a contradiction to the choice of . We prove this gradually, by showing that , then , and finally is connected in .
Claim 1: The set is connected in .
Proof of Claim 1: We use to denote . Note that . To show that is connected in , it suffices to prove that for every face , there is a path in from the vertex of to some vertex of the dual corresponding to a face .
Let be a vertex of that is in and is incident to both the face and some face in . Notice every face incident to is in . Therefore the set of faces incident to is partitioned into two sets, one consisting of faces in and the other consisting of faces in . This implies the existence of the claimed path.
We next show that is also connected. Before proceeding to the proof, we note that for every edge of , .
Claim 2: The set is connected in .
Proof of Claim 2: For every vertex-face edge of where is a child of , we denote by the set of all faces of in .
For every such , we aim to show that is connected in . To see this, first observe that both and are connected in (because of Claim 1 and Appendix A, respectively). Therefore to show that is connected in , it suffices to show that given faces of where belongs to and belongs to , there is a path in the dual consisting only of vertices of that correspond to faces in . Let be a vertex of that is in and is incident to a face in and a face in . Notice that the faces, call them , of that are incident to are also faces of . Therefore the set is partitioned into two sets, one consisting of faces of and one consisting of faces in . This implies the existence of a path in between any two faces in using only vertices corresponding to faces in , which together with the connectivity of and implies that is connected in .
The proof of the claim follows by observing that the fact that for every vertex-face edge of where is a child of , is connected in implies that is connected in , thus proving that is connected in .
We conclude the proof of the lemma by showing that is connected in . Let be a child of in . We use to denote the set , to denote the set , and to denote the set . Keep in mind that and .
Suppose towards a contradiction that is not connected in . Let be the connected components of and let be the sets of faces corresponding to the vertex sets of , i.e., for every , . Observe that since is -edge-connected, is connected. Therefore, for every , there is an and an such that and are incident faces in . Let be an edge of that is incident to both and and let be some endpoint of . Notice now that and therefore, by definition of the distance decomposition, there is a face of that belongs to and is incident to . The latter implies that for every , there is a , a contradiction to the assumption that is connected in . ∎
We are now ready to prove Subsection 3.2 using Appendices A and A.
Proof of Subsection 3.2.
Let be a non-root vertex-face edge of . By Appendix A and Appendix A, the sets and are both connected in . Let and . Notice that is a partition of the faces of . Since is -edge-connected, for every edge of , there are exactly two faces of that are incedent to this edge. Let be the set of all edges of that are incident to a face from and a face from and let be the set of endpoints of the edges of . It is easy to see that . Connectivity of both and in and planarity of imply that the graph is isomorphic to a cycle. ∎