Abstract 1 Introduction 2 The Proof Modulo Three Critical Lemmas 3 The Gadget Graphs 4 Linear Leaf Powers 5 Conclusion References

k-Leaf Powers Cannot Be Characterized by a Finite Set of Forbidden Induced Subgraphs for 𝒌𝟓

Max Dupré la Tour McGill University, Montreal, Canada Manuel Lafond ORCID Université de Sherbrooke, Canada Ndiamé Ndiaye ORCID McGill University, Montreal, Canada Adrian Vetta ORCID McGill University, Montreal, Canada
Abstract

A graph G=(V,E) is a k-leaf power if there is a tree T whose leaves are the vertices of G, with the property that a pair of distinct leaves u and v share an edge in G if and only if they are distance at most k apart in T. For k4, it is known that there exists a finite set k of graphs such that the class (k) of k-leaf power graphs is characterized as the set of strongly chordal graphs that do not contain any graph in k as an induced subgraph. We prove no such characterization holds for k5. That is, for any k5, there is no finite set k of graphs such that (k) is equivalent to the set of strongly chordal graphs that do not contain as an induced subgraph any graph in k.

Keywords and phrases:
Leaf Powers, Forbidden Graph Characterizations, Strongly Chordal Graphs
Category:
Track A: Algorithms, Complexity and Games
Copyright and License:
[Uncaptioned image] © Max Dupré la Tour, Manuel Lafond, Ndiamé Ndiaye, and Adrian Vetta; licensed under Creative Commons License CC-BY 4.0
2012 ACM Subject Classification:
Mathematics of computing Graph algorithms
Related Version:
Full Version: https://arxiv.org/abs/2407.02412
Acknowledgements:
This research collaboration was instigated at the Montreal Graph Theory Workshop.
Editors:
Keren Censor-Hillel, Fabrizio Grandoni, Joël Ouaknine, and Gabriele Puppis

1 Introduction

A fundamental question in graph theory concerns whether or not a graph G=(V,E) can be represented (or approximated) by a simpler graph, for instance a tree T, while preserving the desired information from the original graph. The pairwise distances of G often need to be summarized into sparser structures, with notable examples including graph sparsifiers [16, 1, 25] and graph spanners [11, 2, 15, 22], which ask for a subgraph of G, or distance emulators [30, 9, 31] which ask for another graph that approximates the distances of G. If the distance information to preserve only concerns “close together” versus “far apart” then this can take the following form: given a graph G and an integer k, does there exists a tree T whose leaves are the vertices of G, such that distinct vertices u and v are adjacent in G if and only if the distance dT(u,v) from u to v in T is at most k? If the answer is affirmative then G is dubbed a k-leaf power of T (and T is dubbed a k-leaf root of G).

The study of k-leaf powers and roots were instigated by Nishimura, Ragde and Thilikos [28]. On the applied side, these graphs are of significant interest in the field of computational biology with respect to phylogenetic trees, which aim to explain the distance relationships observed on available data between species, genes, or other types of taxa. Indeed, k-leaf powers can be used to represent and explain pairs of genes that underwent a bounded number of evolutionary events in their evolution [26, 17], or that have conserved closely related biological functions during evolution [21]. On the theory side, despite their simplicity, several fundamental graph theoretic problems concerning k-leaf powers remain open. The purpose of this research is to resolve one such long-standing open problem. Specifically, we prove that the class (k) of k-leaf power graphs cannot be characterized via a finite set of forbidden induced subgraphs for k5. In contrast, for k4 such finite characterizations were previously shown to exist [12, 6, 4].

1.1 Background

Let (k) denote the class of all k-leaf power graphs, for k2. The class of all leaf power graphs is then denoted by =k(k). The literature on leaf power graphs has primarily focused on two major themes. One, obtaining graphical characterizations for both the class and the classes (k), for fixed values of k. Two, designing efficient algorithms to recognize graphs that belong to these classes.

Let’s begin with the former theme. Here important roles are played by chordal and strongly chordal graphs. A graph is chordal if every cycle of length four or more has a chord, an edge connecting two non-consecutive vertices of cycle. A graph is strongly chordal if it is chordal and all its even cycles of length 6 or more have an odd chord, a chord connecting two vertices an odd distance apart along the cycle. Now, it is known that every graph in and (k) is strongly chordal.111In particular they do not contain, as induced subgraphs, chordless cycles of length greater than three, nor sun graphs. To see this, first note that a leaf power graph is an induced subgraph of a power of a tree. Second, note that trees are strongly chordal, and taking powers and induced subgraphs both preserve this property [29]). However, the reciprocal is not true: there exist strongly chordal graphs that are not leaf powers. The first such example was discovered by Brandstädt et al. [5]. Subsequently, six additional examples were identified by Nevries and Rosenke [27] who conjectured that any strongly chordal graph not containing any of these seven graphs as an induced subgraph is a leaf power. However, a weaker version of this conjecture, that there are only a finite number (rather than seven) of obstructions was was disproved by Lafond [19]. The author constructed an infinite family of minimal strongly chordal graphs that are not leaf powers (i.e., removing any vertex results in a leaf power).

For fixed k, the conjecture that (k) may be characterized by a finite set of obstructions remained open. Indeed, for k4, the classes (k) can be characterized as chordal graphs that do not contain any graph from k as induced subgraphs, where k is a finite set. Specifically:

  • k=2: A graph is in (2) if and only if it is a disjoint union of cliques. That is, (2) is precisely the set of graphs that forbid P3, the chordless path with three vertices, as an induced subgraph. Thus |2|=1.

  • k=3: Dom et al. [12] gave the first characterization of (3): a graph is in (3) if and only if it is chordal and does not contain a bull, a dart or a gem as induced subgraph. Thus |3|=3. Other characterizations of (3) were later discovered [6]

  • k=4: Brandstädt, Bang Le and Sritharan [4] proved that a graph is in (4) if and only if it is chordal and does not contain as induced subgraph one of a finite set 4 of graphs222Formally, they show that the set of basic 4-leaf power, where no two leaves of the leaf root share a parent, can be characterized by chordal graphs which do not have one of 8 graphs as induced subgraphs. 4 can be deduced from this set..

Given this, the aforementioned conjecture naturally arose: for every k, is the class (k) equivalent to the set of chordal graphs that do not contain as induced subgraphs any of a finite set k of graphs?

For k=5, Brandstädt, Bang Le and Rautenbach [7] proved this is true for a special subclass of (5). Specifically, the distance hereditary333A graph G is distance hereditary if for all pairs of vertices (u,v) in all subgraphs of G either the distance is the same as in G or there is no path from u to v. 5-leaf power graphs are chordal graphs that do contain a set of 34 graphs as induced subgraphs. However, for the general case, they state

For k5, no characterization of k-leaf powers is known despite considerable effort. Even the characterization of 5-leaf powers appears to be a major open problem.” [7]

The contribution of this paper is to disprove the conjecture: for all k5, it is impossible to characterize the set of k-leaf powers as the set of chordal graphs which are k-free for |k| finite. In fact, we show that even for the more restrictive class of strongly chordal graphs it is impossible to characterize the set of k-leaf powers as the set of strongly chordal graphs which are k-free for finite |k|.

Let us conclude this section by discussing the second major theme in this area, namely, efficient recognition algorithms. The computational complexity of deciding whether or not a graph is in is wide open. We remark, however, that some graphs in have a leaf rank that is exponential in the number of their vertices, where the leaf rank of a graph G is the minimum k such that G(k) [18]. The question of computing the leaf rank of subclasses of in polynomial time was recently initiated in [23].

For fixed values of k, though, progress has been made in designing polynomial-time algorithms for the (k) recognition problem. For (2),(3) and (4), this immediately follows from the above characterizations because 2,3 and 4 are finite. In fact, all these three recognition problems can be solved in linear time; see [6, 4]. Using a dynamic programming approach, Chang and Ko [10] described a linear-time algorithm for the (5) recognition problem, and Ducoffe [13] proposed a polynomial-time algorithm for the (6) recognition problem. Recently, Lafond [20] designed a polynomial-time algorithm for the (k) recognition problem, for any constant k2. The algorithm is theoretically efficient albeit completely impractical: the polynomial’s exponent depends only on k but is Ω(kk), that is, a tower of exponents kkk of height k. We remark that the algorithm does not rely on specific characterizations of k-leaf power graphs aside from the fact that they are chordal. It appears difficult to significantly improve its running time without a better understanding of the graph theoretical structure of graphs in (k). Our work assists in this regard by improving our knowledge of k-leaf powers in terms of forbidden induced subgraphs.

1.2 Overview and Results

We now present an overview of the paper and our results. In Section 2 we present our main theorem:

Theorem 1.

For k5, the set of k-leaf powers cannot be characterized as the set of strongly chordal graphs which are k-free, where k is a finite set of graphs.

There we discuss the three types of gadgets we need. These gadgets can be combined to form an infinite family of pairwise incomparable graphs which are not k-leaf powers. We prove the main theorem modulo three critical lemmas on the gadgets. In Section 3 we present proofs of the three critical lemmas. Finally, in Section 4 we show how to modify our proof to derive a similar theorem for linear k-leaf powers:

Theorem 2.

For k5, the set of linear k-leaf powers cannot be characterized as the set of strongly chordal graphs which are k-free where k is a finite set of graphs.

Here, a linear k-leaf power is a graph that has a k-leaf root which is the subdivision of a caterpillar. We remark that that the class of linear leaf powers can be recognised in linear time, as shown by Bergougnoux et al. [3].

2 The Proof Modulo Three Critical Lemmas

In this section we prove our main theorem, Theorem 1, assuming the validity of three critical lemmas. The proofs of these lemmas form the main technical contribution of the paper and are deferred to Section 3.

2.1 Preliminaries

Before presenting the proof of Theorem 1, we present necessary definitions and notations. Let’s start with a formal definition of k-leaf powers. Let G=(V,E) be a simple finite graph, and k2 be an integer. Then G is called a k-leaf power if there exists a tree T, known as a k-leaf root of G, with the following properties:

  • V is the set of leaves of T.

  • For any pair of distinct vertices u,vV, there is an edge uvE if and only if dT(u,v)k.

Here dT is the distance metric induced by the tree T when two adjacent vertices are a distance of 1 apart. To simplify the notation, we will use d instead of dT when the context is clear. We will use the notation distG to denote distance within the graph G and thus distinguish it from the distance dT induced by a leaf root T.

2.2 The Proof of the Main Theorem

To prove Theorem 1, for any k5, we will construct a collection of arbitrarily large strongly chordal graphs that are minimal non k-leaf powers. Specifically, these graphs have the property that any proper induced subgraph is a k-leaf power.

To accomplish this goal, we fix k5. We then begin by designing a graph Hn, for all n0, built using three gadget graphs joined in series. First will be the top gadget and last the bottom gadget. In between will be exactly n copies of the interior gadget. We denote these gadget graphs by Top,Bot and I, respectively. These gadget graphs will satisfy a set of critical properties. To formalize these properties we require the following definition. Consider a graph G=(V,E) and T a k-leaf root of G. For vV, let mT(v)=minuV{v}dT(u,v). That is, mT(v) is the shortest distance in the tree T from the leaf v to any other leaf u.

The aforementioned properties of Top,Bot and I are stated in the subsequent three critical lemmas.

Lemma 3.

For all k4, there exists a gadget graph Top that contains a vertex tV(Top) such that:

  1. 1.

    For any k-leaf root T of Top, mT(t)=3.

  2. 2.

    There exists a k-leaf root TTop of Top.

Lemma 4.

For all k4, there exists a gadget graph Bot that contains a vertex bV(Bot) such that:

  1. 1.

    For any k-leaf root T of Bot, mT(b)k1.

  2. 2.

    There exists a k-leaf root TBot such that mTBot(b)=k1

Lemma 5.

For all k5, there exists a gadget graph I that contains two distinct vertices tI,bIV(I) such that:

  1. 1.

    For all k-leaf roots T of I, mT(tI)kmT(bI)=3.

  2. 2.

    There exists a k-leaf root TI of I such that mTI(tI)=k and mTI(bI)=3.

  3. 3.

    There exists a k-leaf root RI of I such that mRI(tI)=k1 and mRI(bI)=4.

We will prove the existence of gadget graphs Top,Bot and I required to verify the three lemmas in Section 3. For the rest of the section, we will assume these lemmas and use them to prove our main result.

First, as alluded to above, we then combine our three gadgets to create an intermediary graph Hn. In particular, Hn is the graph obtained by connecting in series one copy of Top, then n copies of I: I1,,In and finally one copy of Bot. This construction is illustrated in Figure 1(a). The vertices tI and bI, mentioned in Lemma 5, of the j-th copy Ij are denoted tIj and bIj, respectively. Notice that to connect the gadgets within Hn, we identify the vertices described in Lemmas 3, 4, and 5 as follows. We identify t with tI1, for all j<n, bIj with tIj+1, and finally, bIn with b. As a special case when n=0, the graph H0 is obtained by taking Top and Bot and identifying t with b.

In order to prove Theorem 1 we must study the structure of Hn. We denote by HnTop (resp. HnBot) the graph obtained from Hn by deleting the top gadget Top (resp. the bottom gadget Bot), i.e. removing all vertices of Top (resp. Bot) except for the common vertex t=tI1 (resp. b=bIn). Of importance is the next lemma.

Lemma 6.

The graph Hn has the following properties:

  1. 1.

    distHn(b,t)n.

  2. 2.

    Hn is strongly chordal.

  3. 3.

    HnTop and HnBot are both k-leaf powers.

  4. 4.

    Hn is not a k-leaf power.

Proof.

In order to prove 1, the distance between tI and bI in I is at least 1 because the two vertices are distinct. Hence distHn(b,t)n, because there are n copies of the interior gadget I.

For the proof of 2, the three gadgets are k-leaf powers and, therefore, are strongly chordal. The construction of Hn does not introduce additional cycles; thus, Hn remains strongly chordal.

To prove 3, we combine the leaf-root properties provided by the gadgets TTop, TBot, TI, and RI, as illustrated in Figure 1(b).

Refer to caption
(a) The construction of Hn.
Refer to caption
(b) The k-leaf roots of HnBot and HnTop.
Figure 1: The construction of Hn and the leaf roots obtained by removing one gadget.

The left tree in Figure 1(b) is obtained by merging TTop with n copies of TI, denoted as TI1,,TIn. We identify t with tI1, and we identify the parent of t in TTop with the parent of tI1 in TI1. Similarly, for all jn1, we identify bIj and its parent with tIj+1 and its parent, respectively. We now prove that the resulting tree is a k-leaf root of HnBot. Top and each copy of the interior gadget I are the k-leaf power of the corresponding subtree: TTop for Top and TIj for the j-th copy of I. It remains to show that we do not introduce any additional, unwanted edges. If v1 is a leaf of TTop different from t, and v2 is a leaf of TI1 different from tI1 then, using Lemma 3, we conclude d(v1,t)mTTop(t)=3. Furthermore, using the second point of Lemma 5, we conclude d(v2,tI1)mTI(tI)=k. Therefore, d(v1,v2)d(v1,t)+d(v2,tI1)2k+32=k+1>k, and there is no edge between v1 and v2 in the k-th power of the tree. Similarly, if v1 is a leaf of TIj different from bIj for some jn1 and v2 is a leaf of TIj+1 different from tIj+1, we have d(v1,bIj)3 and d(v2,tIj+1)k. Thus, d(v1,v2)d(v1,bIj)+d(v2,tIj+1)2k+32=k+1>k, and there is no edge between v1 and v2 in the k-th power of the tree.

The right tree in Figure 1(b) is formed by merging n copies of RI, denoted RI1,,RIn, with TBot. For all jn1, we identify bIj with tIj+1, and we identify the parent of bIj with the parent of tIj+1. Finally, we identify bIn and its parent in RIn with b and its parent in TBot, respectively. Similar to the left tree, we must prove that no additional, unwanted edges are created. If v1 is a leaf of RIj different from bIj for some jn1 and v2 is a leaf of RIj+1 different from tIj+1 then, using Lemma 5, we conclude d(v1,bIj)mRI(bI)=4 and d(v2,tIj+1)mRI(tI)=k1. Thus, d(v1,v2)d(v1,bIj)+d(v2,tIj+1)2k1+42=k+1>k, and there is no edge between v1 and v2 in the k-th power of the tree. Similarly if v1 is a leaf of RIn different from bIn and v2 is a leaf of TBot different from b then we have d(v1,bIn)4 and, using Lemma 4, d(v2,b)k1. Therefore d(v1,v2)d(v1,bIn)+d(v2,b)2k+1>k and v1,v2 are not connected in the k-th power of the tree. This completes the proof of 3, the third point of the lemma.

It remains to prove 4, the final point of the lemma. We start by proving by induction that for any integer n, in any k-leaf root T of HnBot, there is a leaf at distance 3 of b in T. When n=0, there are no gadgets I between Bot and Top, so b=t and the property holds by property 1 of Lemma 3. Turning to the induction step, assume that the property holds for some n0 and consider a k-leaf root T of Hn+1Bot. Since HnBot is an induced subgraph of Hn+1Bot, some induced subgraph of T is a k-leaf root of HnBot. By the induction hypothesis, there exists a vertex v1 in HnBot at a distance exactly 3 from bIn=tIn+1 in an induced subgraph of T. Adding vertices will not alter the distance, so d(v1,bIn)=3 in T. We claim that every vertex of In+1 is at distance at least k from bIn=tIn+1 in T (except bIn itself). Assume, by contradiction, that there exists a vertex v2 in the last copy In+1, distinct from bIn such that that d(v2,bIn)k1 . This assumption would imply that d(v1,v2)d(v1,bIn)+d(v2,bIn)23+(k1)2=k, meaning that v1 and v2 are connected in the k-th power of T, contradicting the fact that T is a k-leaf root of Hn+1Bot. Therefore, all vertices in In+1, distinct from bIn, are at a distance of at least k from bIn in T. We can now apply property 1 of Lemma 5, which concludes the induction.

Now assume by contradiction that there exists a k-leaf root T of Hn. Then T induces a k-leaf root of HnBot. In particular, there exists a vertex v1 in HnBot such that d(bIn,v1)=3. Moreover, property 1 of Lemma 4 implies that there exists a vertex v2 in Bot, distinct from b, such that d(v2,b)k1. Combining these equations, we get d(v1,v2)k1+32=k, contradicting the fact that there is no edge between v1 and v2. This proves that Hn is not a k-leaf power, as desired.

As stated Hn is an intermediate graph in proving the main result. We will actually show the existence of an induced subgraph Gk,n of Hn that is strongly chordal and minimal non k-leaf power. More precisely, we have the following lemma.

Lemma 7.

For all k5 and n0, there exists a graph Gk,n such that:

  1. 1.

    Gk,n is strongly chordal and contains at least n vertices.

  2. 2.

    Gk,n is not a k-leaf power.

  3. 3.

    If GGk,n is an induced subgraph of Gk,n then G is a k-leaf power.

Proof.

Let Gk,n be a minimal induced subgraph of Hn that is not a k-leaf power. It is strongly chordal because, by Lemma 6, Hn is strongly chordal. By definition, Gk,n is not in (k), but every induced subgraph G of Gk,n is in (k) if GGk,n. It remains to prove that |Gk,n|n. By Lemma 6, both HnBot and HnTop are in (k). Therefore, Gk,n must contain a vertex from Top and a vertex from Bot. Moreover, Gk,n is connected by its minimality, since the disjoint union of k-leaf powers is a k-leaf power. Hence, Gk,n must contain a path from Top to Bot. By Lemma 6, distHn(b,t)n, and therefore |Gk,n|n.

Our main result follows directly from Lemma 7

Proof of Theorem 1.

If (k) is the set of strongly chordal graphs which are k-free, then k must contain Gk,n for all n because it is strongly chordal and a minimal non k-leaf power by Lemma 7. But the set {Gk,n,n0} is infinite because Gk,n has more than n vertices. Therefore k must be infinite, which concludes the proof of Theorem 1.

3 The Gadget Graphs

So we have proven the main theorem modulo the three critical lemmas. Recall to prove these lemmas we must construct the appropriate three gadget graphs, namely Top, Bot and I. We present these constructions and give formal proofs of Lemmas 34 and 5 in this section.

We start with a general observation. In a tree T, if a pair of leaves are a distance of 2 apart, they share the same parent. Consequently, their distances to every other leaf are identical. A consequence of this is that if two vertices are not connected by an edge, or if they have different neighborhoods in a graph, they must be at a distance of at least 3 in any leaf root of that graph. In the gadgets we describe in this section, any two vertices connected by an edge always have distinct neighborhoods. Therefore, we assume that for any pair of vertices x and y and any leaf root T, we have dT(x,y)3.

3.1 The Top Gadget

We begin by showing the existence of an appropriate top gadget, Top.

See 3

Proof.

Let P be the path on the 2k3 vertices v1,,v2k3. The top gadget Top is defined as Pk2 with t=vk2. Two examples of Top are shown in Figure 2, for the cases k{5,6}. First, note that Top is a k-leaf power. In particular, a k-leaf root of Top is the caterpillar TTop, which is constructed by attaching a leaf to every vertex of P. For the first property, we use Lemma 2 from [32], which states that if Pk2 is an induced subgraph of Tk, where T is a tree and k1, then dT(vk3,vk2)3 (note that the lemma is more general, as P can have length 2p+1 for any p2, and the result applies to Pp and dT(vp1,vp)kp+1; here we took p=k2). Thus, if T is any k-leaf root of Top, then Tk does contain Pk2 as an induced subgraph and the property holds.

Figure 2: The Top Gadget for k=5 and k=6.

3.2 The Bottom Gadget

Next, we construct the bottom gadget, Bot. A key technical tool we require is the 4-Point Condition. This is the following classical characterization of tree metrics.

Theorem 8 (4-Point Condition).

[8] Let d be a distance on a finite set V, then there exists a tree T whose leaves are V such that u,vV, dT(u,v)=d(u,v) if and only if the following condition is true for all (u,v,w,t)V:

d(u,v)+d(w,t)max{d(u,w)+d(v,t),d(v,w)+d(u,t)}.

Our bottom gadget Bot will simply be a diamond, the complete graph on 4 vertices minus one edge. Consequently, we begin by proving the following corollary of the 4-Point Condition when applied to a diamond.

Corollary 9.

In any k-leaf root T of a diamond with vertex set {b,v1,v2,v3} where (v1,v3)E, d(b,v2)k.

Proof.

Assume for contradiction that d(b,v2)=k. Then since d(v1,v3)>k, we get d(v1,v3)+d(b,v2)>2k.

On the other hand, since we have a leaf root of the diamond, we must have:

max{d(v1,v2),d(v2,v3),d(v3,b),d(b,v1)}k.

This implies that d(v1,v2)+d(v3,b)2k and d(v2,v3)+d(b,v1)2k.

So, d(v1,v3)+d(b,v2)>max{d(v1,v2)+d(v3,b),d(v2,v3)+d(b,v1)} which contradicts Theorem 8

Corollary 9 allows us to prove our critical lemma for the bottom gadget.

See 4

Proof.

Let the graph Bot be the diamond with vertex set {b,v1,v2,v3} and non-edge (v1,v3). By Corollary 9, d(b,v2)k1 in any leaf root. Thus the first property holds. For the second property, there are two cases, illustrated in Figure 3, depending upon the parity of k.

(a) TBot for k odd
(b) TBot for k even
Figure 3: The k-leaf roots of the diamond with minvV(D){b}d(b,v)=k1. Here the bold edges denote paths of the described length; the dotted edges are the edges of the diamond.
  • If k is odd, start with b and v2 at distance k1. Let O be the midpoint of the two at distance k12 from both. Set v1 and v3 to be each at distance k+12 from O. Then b and v2 will both be at distance exactly k from both v1 and v3, but v1 and v3 are at distance k+1 from each other. Thus, the only distance which is greater than k is d(v1,v3) and the closest vertex to b is v2, as desired.

  • If k is even, start with b and v2 at distance k1. Set O1 to be the point at distance k21 from b and O2 the point at distance k21 from v2. Add v1 at distance k2 from O1 and v3 at distance k2 from O2. Then b is at distance k1 from v1 and k from v3, while v2 is at distance k from v1 and k1 from v3. Note that v1 and v3 are at distance k+1 from each other. Thus, the only distance which is greater than k is d(v1,v3) and the closest vertex to b is v2, as desired.

This lemma follows.

3.3 The Interior Gadget

Lastly, we have the most complex construction, that of the interior gadget, I. Now we require the following lemma which, again, is a consequence of the 4-Point Condition.

Lemma 10.

Suppose that d is the distance function between leaves of a tree T. If d(t,x1)min{d(t,x2),d(t,x3)} and d(y,x1)>max{d(y,x2),d(y,x3)}, then:

d(t,x1)+d(x2,x3)<d(t,x2)+d(x1,x3)=d(t,x3)+d(x1,x2)

Proof.

Assume that d(t,x1)+d(x2,x3)max{d(t,x2)+d(x1,x3),d(t,x3)+d(x1,x2)}. Then, by using this assumption and our bound on d(t,x1), we get:

d(x2,x3) max{d(x1,x3)+(d(t,x2)d(t,x1)),d(x1,x2)+(d(t,x3)d(t,x1))}
max{d(x1,x3),d(x1,x2)}

Then, by combining this bound with our bound on d(y,x1), we get:

d(y,x1)+d(x2,x3)>max{d(y,x2)+d(x1,x3),d(y,x3)+d(x1,x2)}

This contradicts Theorem 8. This implies that the assumption is wrong, that is, we must have:

d(t,x1)+d(x2,x3)<max{d(t,x2)+d(x1,x3),d(t,x3)+d(x1,x2)}.

Now, assume without loss of generality that d(t,x2)+d(x1,x3)d(t,x3)+d(x1,x2). Then, by Theorem 8, we must have:

d(t,x2)+d(x1,x3)max{d(t,x1)+d(x2,x3),d(t,x3)+d(x1,x2)}.

Since d(t,x2)+d(x1,x3)>d(t,x1)+d(x2,x3), this implies that d(t,x2)+d(x1,x3)d(t,x3)+d(x1,x2). So we get that d(t,x2)+d(x1,x3) is bounded above and below by d(t,x3)+d(x1,x2) so they must be equal.

So we get:

d(t,x1)+d(x2,x3)<d(t,x2)+d(x1,x3)=d(t,x3)+d(x1,x2).

This completes the proof.

We will also use the following simple lemma:

Lemma 11.

Suppose that d is the distance function between leaves of a tree T. Then for any 3 leaves u,v,w of T, d(u,v)+d(u,w)+d(v,w) is even.

Proof.

Since we have a tree, there is a unique vertex O which is simultaneously in the path from u to v, the path from v to w and the path from u to w.

Hence d(u,v)+d(u,w)+d(v,w)=2(d(u,O)+d(v,O)+d(w,O)) which must be even.

We now have all the tools needed to prove our critical lemma for the interior gadget.

See 5

Before proving this lemma, let’s discuss the requirement that k5. First observe that no such graph can exist for k2 because if mT(bI)=3 then bI is an isolated vertex in I. Thus its distance to other leaves does not matter as long as it’s large enough, so the lemma could not hold. Similarly, if k=3, the existence of a 3-leaf root TI implies that bI is not an isolated vertex in I. But the existence of a 3-leaf root RI implies that bI is an isolated vertex, a contradiction. Finally, for k=4, while there is no direct simple proof that the statement does not hold for any graph, the existence of a characterization of 4-leaf powers implies that no such graph can exist.

Proof.

For convenience, we denote tI and bI by t and b, respectively. To prove the lemma, we will explicitly construct I for any value of k5. In order to do so, we consider two cases depending, again, upon the parity of k.

For k odd, set q=k12. In particular, for k5 we must have q2.

We construct the graph I using the following sets of vertices:

  • t and b

  • X={x1,,xq}

  • Y={y2,,yq}

The edge set is defined as follows (note that for clarity, we use ordered pairs to represent edges, with the understanding that the edges are undirected):

  • For i=1,q, (t,xi) and (b,xi) are edges. That is, t and b are adjacent to all vertices in X.

  • For i=1,,q, for j=i+1,q, (xi,xj) is an edge. That is, X forms a clique.

  • For i=2q, for j=iq, (yi,xj) is an edge.

  • (b,yq) is an edge.

Equivalently, it will be helpful to define the set of edges using the neighborhood of each vertex:

  • t is adjacent to X={x1,,xq}.

  • b is adjacent to X and to yq.

  • For i=1,,q, xi is adjacent to t, to b, to X{xi} and to yj for j=2,,i (with x1 having no neighbor in Y).

  • For i=2,,q, for j=i,,q, yi is adjacent to xj. If i=q, then yq is also adjacent to b.

That is, we take I=(V,E) to be defined by:

V= {t,b}(i=1q{xi})(i=2q{yi}) (1)
E= (i=1q{(t,xi),(b,xi)})(1i<jq{(xi,xj)})(2ijq{(yi,xj)}){(b,yq)}.

This construction is illustrated in Figure 4. We remark that this construction only makes sense for k5 because if k=3 or k=1 then Y is an empty set.

Figure 4: The interior gadget I for odd k. The bold edges signify all possible connections are made.

We will prove that this graph satisfies the lemma by proving three claims.

Claim 12.

For I as defined in (1):

For all k-leaf roots T of I, mT(t)=kmT(b)=3.

Proof.

Assume that mT(t)=k, then i[q], d(t,xi)=k. Then for all distinct i,j[q], by Lemma 11 with t, xi and xj, d(xi,xj) must be even (and, thus, at most k1).

Assume 1i<j<k. Then d(t,xi)=k=min{d(t,xj),d(t,x)}. Furthermore, d(yj,xi)>kmax{d(yj,xj),d(yj,x)} because yj is adjacent to xj and x but not xi (nor t). So, by Lemma 10, we get:

d(t,xi)+d(xj,x)<d(t,x)+d(xi,xj)=d(t,xj)+d(xi,x)

But because d(t,xi)=d(t,xj)=d(t,x)=k. This is true if and only if

d(xj,x)<d(xi,xj)=d(xi,x)

This implies that for every i[q1], there exists some integer λi such that d(xi,xi+1)=d(xi,xi+2)==d(xi,xq)=λi. Moreover, as shown, the λi must be even. Thus k1λ1>>λq1>2. By definition, 2q=k1. In particular, there are only q1 even numbers greater that 2 and at most k1. Therefore, λi=k+12i. Specifically, we have shown that d(xi,xj)=k+12i, for 1i<jq.

Recall d(t,xi)=k=min{d(t,xq),d(t,b)} and d(yq,xi)>kmax{d(yq,xq),d(yq,b)}, for i<q. So, by Lemma 10, we obtain:

d(xq,b)+d(xi,t)<d(xq,xi)+d(t,b)=d(xq,t)+d(xi,b).

Consider i=1. Recall k5 and q2. So q1 implying that xqx1. It follows that

d(x1,xq)+d(t,b)=d(xq,t)+d(x1,b)
k1+d(t,b)=k+d(x1,b)
d(t,b)=1+d(x1,b)

However, we must have d(t,b)>k and d(x1,b)k. So it must be that d(t,b)=k+1 and d(x1,b)=k.

Next consider i=q1. Because q2, we have i1 and so xq1 exists. Then d(xq1,xq)=k+12(q1)=k+3(k1)=4. Therefore, because d(xi,t)=k for all 1iq, we have

d(xq1,xq)+d(t,b)=d(xq,t)+d(xq1,b)
4+k+1=k+d(xq1,b)
d(xq1,b)=5

Finally, recall that d(xq,b)+d(xq1,t)<d(xq,xq1)+d(t,b). This implies that d(xq,b)+k<4+(k+1). In particular, d(xq,b)<5. Moreover, by Lemma 11, we know d(xq,b)+d(xq,xq1)+d(xq1,b) is even. But d(xq,xq1) is even and d(xq1,b) is odd. Hence d(xq,b) must be odd. In particular, it must be odd and less than 5. Hence d(xq,b)=3, which is what we wanted to show.

Claim 13.

For I as defined in (1):

There exists a k-leaf root TI of I such that mTI(t)=k and mTI(b)=3.

Proof.

We will prove this by explicitly constructing a leaf root. Recall k is odd and k=2q+1.

  1. 1.

    Take a path of length k+1=2q+2 from t to b.

  2. 2.

    Label the vertices along the path from t to b which are at distance q+i of t as Oi for i=1,,q+1.

  3. 3.

    Add a path of length qi+1 from Oi to xi for i=1,,q.

  4. 4.

    Add a path of length k2 from Oq+1 to yq.

  5. 5.

    If k7, add a path of length q+i from Oi to yi for i=2,,q1. (For k=5 we have q=2, so these yi do not exist.)

Figure 5: The k-leaf root TI of the interior gadget for odd k=2q+1. (Recall these will be connected in series below the leaf root of the top gadget; see Figure 1(b).)

This construction is shown in Figure 5. It remains to verify that this is a valid k-leaf root of I, that is, the leaf vertices at distance at most k in the tree are exactly the edges of I. The required case analysis follows.

  • The path from t to b has length k+1>k and t and b are not neighbors, as desired.

  • For i=1,,q, the path from t to xi goes through Oi. That is, it is the path from t to Oi which has length q+i followed by the path from Oi to xi which has length qi+1. So, the path from t to xi has length (q+i)+(qi+1)=k.

  • The path from t to yq goes through Oq+1, so it has length (2q+1)+(k2)>k.

  • For i=1,,q, the path from b to xi goes through Oq and Oi so it has length 2+(qi)+(qi+1)k. In particular, for i=q, the path from b to xq has length 3.

  • The path from b to yq goes through Oq+1 so it has length 1+(k2)k

  • For 1i<jq, the path from xi to xj goes through Oi and Oj so it has length (qi+1)+(ji)+(qj+1)k

  • For i=1,,q and j=2,,q1, the path from xi to yj goes through Oi and Oj (possibly the same vertex) so it has length (qi+1)+(|ij|)+(q+j)=k+(ji)+|ij| which is at most k if and only if ij.

  • For i=1,,q, the path from xi to yq goes through Oi and Oq+1 so it has length (qi+1)+(q+1i)+(k2) which is equal to k if i=q and strictly greater than k otherwise.

For k=5 the case analysis is complete. If k7 then q3 and so |Y|>1. Thus we have four more cases to verify:

  • For i=2,,q1, the path from t to yi goes through Oi, so it has length (q+i)+(q+i)>k.

  • For i=2,,q1, the path from b to yi goes through Oq and Oi so it has length 2+(qi)+(q+i)>k.

  • For 2i<jq1, the path from yi to yj goes through Oi and Oj, so it has length (q+i)+(ji)+(q+j)>k.

  • For i=2,,q1, the path from yi to yq goes through Oi and Oq+1 so it has length (q+i)+(q+1i)+(k2)>k.

Hence the desired k-leaf root exists.

It remains to prove the final property to conclude that the claim holds when k is odd:

Claim 14.

For I as defined in (1):

There exists a k-leaf root RI of I such that mRI(t)=k1 and mRI(b)=4.

Proof.

To construct RI we make two minor modifications to the tree TI used to prove Claim 13. First we start with a path of length k+2 from t to b. To do this we simply place b at distance 2 from Oq+1 instead of 1. All the remaining vertices are then placed using the same process except for x1 which is now at distance q1 from O1 instead of at distance q. The resultant tree is shown in Figure 6.

Figure 6: The k-leaf root RI of the interior gadget for odd k=2q+1. (Recall these will be connected in series above the leaf root of the bottom gadget; see Figure 1(b).)

It suffices to verify that all the vertices are still at a correct distance from x1 and from b.

  • The distance from x1 to all other vertices except b has decreased by 1 so we need to verify that the yi and yq are still at distance at least k+1. For i=2,,q1, the distance from x1 to yi is (q1)+(i1)+(q+i)>k (since i2). The distance from x1 to yq is (q1)+q+(k2)=2q+k3>k (since q2).

  • The distance from b to all other vertices except x1 has increased by 1 so we need to verify that the xi’s and yq are still at distance at most k. For i2, the distance from b to xi is 3+(qi)+(qi+1)k. The distance from b to yq is 2+(k2)=k. Moreover, all distances from b to an xi vertex are now at least 4 and not 3 (including x1, which is at distance k=2q+1 from b). In particular, the distance between b and xq is exactly 4.

Observing that t is now at distance k1 from x1 and is still at distance k to its other neighbors, we get mRI(t)=k1 and mRI(b)=4. Hence the desired k-leaf root exists.

We have now proven the result holds for k odd. A similar but slightly more intricate proof can be used to prove the result also holds for k even to complete the proof of the lemma. The proof for that case is deferred to the full version of the paper. With our three critical lemmas proven, the main theorem now holds by the method shown in Section 2.

4 Linear Leaf Powers

A caterpillar is a graph which has a central path and a set of leaves whose neighbor is on the central path. A graph is said to be a linear leaf power if it has a leaf root which is the subdivision of a caterpillar. Such a leaf root is called a linear leaf root [3]. Our results apply not only to general leaf powers but also to this variant. Indeed, even if we restrict to having a subdivision of a caterpillar as a leaf root, it is impossible to get a simple forbidden subgraph characterization of linear k-leaf powers for k5.

Theorem 15.

For k5, the set of linear k-leaf powers cannot be written as the set of strongly chordal graphs which are k-free where k is a finite set of graphs.

As in the general case, we prove this theorem using three gadgets. However, we must now add a condition to ensure that merging the gadgets preserves having a subdivision of a caterpillar as a leaf root. In particular, we will be able to use the same interior and bottom gadgets but the top gadget needs to be modified. The proof is given in the full version of the paper.

We remark that this result does not immediately imply that there is no characterization for the entire class of linear leaf powers using chordal graphs and a finite number of forbidden induced subgraphs. We have proven that such a characterization is impossible for each k5, but not necessarily for the union over all k. It was proved by Bergougnoux et al. [3] that linear leaf powers are also co-threshold tolerance. Further work on this could include verifying whether co-threshold tolerance graphs can be characterized using a finite number of obstructions. Furthermore, it is worth noting that, despite Theorem 15, linear leaf powers are recognizable in polynomial time [3].

5 Conclusion

We have shown that k-leaf powers require a deeper characterization than strong chordality with a finite set of forbidden induced subgraphs. Several directions to explore remain in order to gain a more comprehensive understanding of k-leaf power graphs. First, is it possible to construct and/or characterize minimal, strongly chordal graphs that are not k-leaf powers? We were able to construct graphs Hn that contain such minimal examples as induced subgraphs; but we did not construct those examples explicitly. Following this line of reasoning, it may be possible to characterize k-leaf powers as strongly chordal graphs that also forbid an additional infinite, but easy-to-describe family of forbidden subgraphs. A famous example of this are interval graphs, which are the chordal graphs containing no asteroidal triples [24].

Second, are there relevant subclasses of k-leaf powers that can be characterized by strong chordality and a finite set of forbidden induced subgraphs? For example, the k-leaf powers whose k-leaf roots admit a subdivision of a star should be easy enough to characterize. What about subdivisions of a tree with a small number, say two or three, of non-leaf vertices? One may also consider the k-leaf powers of caterpillars (not subdivided). Based on the midpoint arguments of Brandstädt et al. [5, Theorem 6], it would appear that, for even k, these coincide with the unit interval graphs whose intervals have length k2 and integer endpoints. Such (twin-free) graphs were shown to admit a finite set of forbidden induced subgraphs in [14]. If this characterization extends to caterpillar k-leaf powers, this would show that taking subdivisions is necessary for our result on caterpillar graphs. It may also be interesting to characterize k-leaf powers for other graph classes that are known to be contained in ; for instance, k-leaf powers that are also ptolemaic graphs, interval graphs, rooted directed path graphs, and others.

References

  • [1] Ittai Abraham, David Durfee, Ioannis Koutis, Sebastian Krinninger, and Richard Peng. On fully dynamic graph sparsifiers. In 2016 IEEE 57th Annual Symposium on Foundations of Computer Science (FOCS), pages 335–344, 2016. doi:10.1109/FOCS.2016.44.
  • [2] R. Ahmed, G. Bodwin, F. Sahneh, K. Hamm, M. Jebelli, S. Kobourov, and R. Spence. Graph spanners: A tutorial review. Computer Science Review, 37:100253, 2020. doi:10.1016/J.COSREV.2020.100253.
  • [3] B. Bergougnoux, S. Høgemo, J. Telle, and M. Vatshelle. Recognition of linear and star variants of leaf powers is in p. In Michael A. Bekos and Michael Kaufmann, editors, Graph-Theoretic Concepts in Computer Science, pages 70–83, Cham, 2022. Springer International Publishing.
  • [4] A. Brandstädt, V. Le, and R. Sritharan. Structure and linear-time recognition of 4-leaf powers. ACM Trans. Algorithms, 5(1), December 2008. doi:10.1145/1435375.1435386.
  • [5] A. Brandstädt, C. Hundt, F. Mancini, and P. Wagner. Rooted directed path graphs are leaf powers. Discrete Mathematics, 310(4):897–910, 2010. doi:10.1016/j.disc.2009.10.006.
  • [6] A. Brandstädt and V. Le. Structure and linear time recognition of 3-leaf powers. Information Processing Letters, 98(4):133–138, 2006. doi:10.1016/j.ipl.2006.01.004.
  • [7] A. Brandstädt, V. Le, and D. Rautenbach. A forbidden induced subgraph characterization of distance-hereditary 5-leaf powers. Discrete Mathematics, 309(12):3843–3852, 2009. doi:10.1016/j.disc.2008.10.025.
  • [8] P. Buneman. A note on the metric properties of trees. Journal of Combinatorial Theory, Series B, 17(1):48–50, 1974. doi:10.1016/0095-8956(74)90047-1.
  • [9] H-C. Chang, R. Krauthgamer, and Z. Tan. Almost-linear ε-emulators for planar graphs. In Proceedings of the 54th Annual ACM SIGACT Symposium on Theory of Computing, pages 1311–1324, 2022.
  • [10] M-S. Chang and M-T. Ko. The 3-steiner root problem. In Andreas Brandstädt, Dieter Kratsch, and Haiko Müller, editors, Graph-Theoretic Concepts in Computer Science, pages 109–120, Berlin, Heidelberg, 2007. Springer Berlin Heidelberg.
  • [11] V. Cohen-Addad, A. Filtser, P. Klein, and H. Le. On light spanners, low-treewidth embeddings and efficient traversing in minor-free graphs. In 2020 IEEE 61st Annual Symposium on Foundations of Computer Science (FOCS), pages 589–600. IEEE, 2020.
  • [12] M. Dom, J. Guo, F. Hüffner, and R. Niedermeier. Error compensation in leaf root problems. In Algorithms and Computation, pages 389–401, December 2004. doi:10.1007/978-3-540-30551-4_35.
  • [13] G. Ducoffe. The 4-steiner root problem. In International Workshop on Graph-Theoretic Concepts in Computer Science, pages 14–26. Springer, 2019.
  • [14] G. Durán, F. Slezak, L. Grippo, F. Oliveira, and J. Szwarcfiter. On unit interval graphs with integer endpoints. Electronic Notes in Discrete Mathematics, 50:445–450, 2015. doi:10.1016/J.ENDM.2015.07.074.
  • [15] A. Filtser, M. Kapralov, and N. Nouri. Graph spanners by sketching in dynamic streams and the simultaneous communication model. In Proceedings of the 2021 ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 1894–1913. SIAM, 2021.
  • [16] Wai Shing Fung, Ramesh Hariharan, Nicholas J.A. Harvey, and Debmalya Panigrahi. A general framework for graph sparsification. In Proceedings of the Forty-Third Annual ACM Symposium on Theory of Computing, STOC ’11, pages 71–80, New York, NY, USA, 2011. Association for Computing Machinery. doi:10.1145/1993636.1993647.
  • [17] M. Hellmuth, C. Seemann, and P. Stadler. Generalized fitch graphs ii: Sets of binary relations that are explained by edge-labeled trees. Discrete Applied Mathematics, 283:495–511, 2020. doi:10.1016/J.DAM.2020.01.036.
  • [18] S. Høgemo. Lower bounds for leaf rank of leaf powers. CoRR, abs/2402.18245, 2024. doi:10.48550/arXiv.2402.18245.
  • [19] M. Lafond. On strongly chordal graphs that are not leaf powers. In H. Bodlaender and G. Woeginger, editors, Graph-Theoretic Concepts in Computer Science, pages 386–398, Cham, 2017. Springer International Publishing.
  • [20] M. Lafond. Recognizing k-leaf powers in polynomial time, for constant k. ACM Trans. Algorithms, 19(4), September 2023. doi:10.1145/3614094.
  • [21] M. Lafond and N. El-Mabrouk. Orthology and paralogy constraints: satisfiability and consistency. BMC genomics, 15:1–10, 2014.
  • [22] H. Le and S. Solomon. Near-optimal spanners for general graphs in (nearly) linear time. In Proceedings of the 2022 Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 3332–3361. SIAM, 2022.
  • [23] V. Le and C. Rosenke. Computing optimal leaf roots of chordal cographs in linear time. In International Symposium on Fundamentals of Computation Theory, pages 348–362. Springer, 2023.
  • [24] C. Lekkerkerker and J Boland. Representation of a finite graph by a set of intervals on the real line. Fundamenta Mathematicae, 51:45–64, 1962.
  • [25] Jiayu Li, Tianyun Zhang, Hao Tian, Shengmin Jin, Makan Fardad, and Reza Zafarani. Sgcn: A graph sparsifier based on graph convolutional networks. In Hady W. Lauw, Raymond Chi-Wing Wong, Alexandros Ntoulas, Ee-Peng Lim, See-Kiong Ng, and Sinno Jialin Pan, editors, Advances in Knowledge Discovery and Data Mining, pages 275–287, Cham, 2020. Springer International Publishing. doi:10.1007/978-3-030-47426-3_22.
  • [26] Y. Long and P. Stadler. Exact-2-relation graphs. Discrete Applied Mathematics, 285:212–226, 2020. doi:10.1016/J.DAM.2020.05.015.
  • [27] R. Nevries and C. Rosenke. Towards a characterization of leaf powers by clique arrangements. Graphs and Combinatorics, 32:2053–2077, 2016. doi:10.1007/S00373-016-1707-X.
  • [28] N. Nishimura, P Ragde, and D. Thilikos. On graph powers for leaf-labeled trees. Journal of Algorithms, 42(1):69–108, 2002. doi:10.1006/jagm.2001.1195.
  • [29] A. Raychaudhuri. On powers of strongly chordal and circular arc graphs. Ars Combin., 34:147–160, 1992.
  • [30] M. Thorup and U. Zwick. Spanners and emulators with sublinear distance errors. In Proceedings of the seventeenth annual ACM-SIAM symposium on Discrete algorithm, pages 802–809, 2006.
  • [31] J. Van Den Brand, S. Forster, and Y. Nazari. Fast deterministic fully dynamic distance approximation. In 2022 IEEE 63rd Annual Symposium on Foundations of Computer Science (FOCS), pages 1011–1022. IEEE, 2022.
  • [32] P. Wagner and A. Brandstädt. The complete inclusion structure of leaf power classes. Theoretical Computer Science, 410(52):5505–5514, 2009. Combinatorial Optimization and Applications. doi:10.1016/j.tcs.2009.06.031.