This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

\plparsep

=1pt

Hamiltonian cycles in annular decomposable Barnette graphs

Saptarshi Bej Author for correspondence: saptarshibej24@gmail.com, saptarshi.bej@uni-rostock.de Institute of Computer Science, University of Rostock, Germany
Abstract

Barnette’s conjecture is an unsolved problem in graph theory. The problem states that every 33-regular (cubic), 33-connected, planar, bipartite (Barnette) graph is Hamiltonian. Partial results have been derived with restrictions on number of vertices, several properties of face-partitions and dual graphs of Barnette graphs while some studies focus just on structural characterizations of Barnette graphs. Noting that Spider web graphs are a subclass of Annular Decomposable Barnette (ADB graphs) graphs and are Hamiltonian, we study ADB graphs and their annular-connected subclass (ADB-AC graphs). We show that ADB-AC graphs can be generated from the smallest Barnette graph (B0B_{0}) using recursive edge operations. We derive several conditions assuring the existence of Hamiltonian cycles in ADB-AC graphs without imposing restrictions on number of vertices, face size or any other constraints on the face partitions. We show that there can be two types of annuli in ADB-AC graphs, ring annuli and block annuli. Our main result is, ADB-AC graphs having non singular sequences of ring annuli are Hamiltonian.

Keywords: Barnette conjecture, Hamiltonian graphs, Annular Decomposition, Planar cubic graphs, Bipartite polyhedral graphs

1 Introduction

Barnette’s conjecture is a long standing unsolved problem in mathematics proposed by David W. Barnette in 1969. The conjecture states:

Conjecture 1.1.

Every 33-regular, 33-connected planar bipartite graph is Hamiltonian. We will henceforth use the term Barnette graphs to refer to 33-regular, 33-connected planar bipartite graphs.

The development towards the Barnette’s conjecture started as early as 1880, when P.G. Tait proposed a weaker statement:

Conjecture 1.2.

Every 33-regular, 33-connected planar graph is Hamiltonian.

This conjecture was ultimately disproved by W.T. Tutte, when he constructed a couter-example to the conjecture, a graph built on 4646 vertices [13]. With further attempts to find a smaller counterexamples for Tait’s conjecture, Holton and McKay found a counterexample on 3838 vertices [8]. However, none of these counterexamples found were bipartite. Tutte thus proposed a conjecture stating:

Conjecture 1.3.

Every 33-regular, 33-connected bipartite graph is Hamiltonian.

Interestingly, a counterexample of this conjecture was found in 1976 by J.D. Horton which is a graph on 9696 vertices, known as the Horton graph [9]. The Barnette conjecture was thus proposed by combining Conjecture 1.2 and Conjecture 1.3 and remains unsolved.

There are several studies that addressed Conjecture 1.1. There have been attempts to understand the structure of Barnette graphs in general [3, 4]. One of the well known approaches is the proof by Holton, Manvel and McKay in 1984, showing that all 33-regular, 33-connected planar bipartite graphs with no more than 64 vertices are Hamiltonian [7]. In the same paper, they also present an interesting approach to generate all possible Barnette graphs from the smallest Barnette graphs, using simple operations defined by adding small structures to existing graphs. Other attempts have been made by studying the duals of Barnette graphs [11, 5, 1]. We conclude from the existing literature, that most of the studies, aiming to find Hamiltonian subclasses of Barnette graphs, focus on studying the dual graphs or imposing restriction on the faces of Barnette graphs.

In this article, we study properties of Annular Decomposable Barnette (ADB graphs) graphs, a sublclass of Barnette graphs. The idea of annular graphs is not novel in itself [2, 12]. Another article derives necessary and sufficient conditions when cubic plane graphs have a rectangular-radial drawing [6]. Interestingly, ADB graphs can be thought of as Barnette graphs that have a rectangular-radial drawing. One related work, that inspired us particularly to study ADB graphs in context to a Hamiltonian problem, is the result that Spider-web graphs have a Hamiltonian cycle excluding any edge in such graphs [10]. The Spider-web graphs can be characterised as ADB-graphs(see Figure 2) in [10]) and thus, can be viewed as a subclass of graphs we have studied. We show how Annular Decomposable Barnette Annular Connected (ADB-AC) graphs, a certain subclass of ADB-graphs can be built recursively from the smallest Barnette graph using edge operations. We derive some sufficient conditions under which, ADB-AC graphs are Hamiltonian. In our results we do not impose restrictions on number of vertices, face size or any other constraint on the face partitions. We show that there can be two types of annuli in ADB-AC graphs, ring annuli and block annuli. Our main result is, ADB-AC graphs having non singular sequences of ring annuli are Hamiltonian.

2 Preliminary definitions and notations

As per the Jordan curve theorem, a non-self-intersecting continuous loop in the 2-D plane has an interior and an exterior. For a planar embedding of a Barnette graph we can view a cycle or face, CC in GG as a Jordan curve and naturally induce a sense of interior, exterior and a boundary to CC. Note that a Hamiltonian cycle HH, if it exists, in a planar embedding of a Barnette graph GG would have the following properties:

  1. 1.

    The interior of HH is path connected

  2. 2.

    The exterior of HH is path connected

  3. 3.

    All vertices in GG lie in the boundary of HH.

The idea is illustrated in Figure 1(a).

Refer to caption
Figure 1: (a) A Hamiltonian cycle in the smallest Barnette graph B0B_{0} (b) An annulus is defined as two non-intersecting Jordan curves on a 2-D plane, topologically equivalent to two non-intersecting concentric circles. The figure shows an annulus and an edge ee, lying in the interior of an annulus.

Note that, two non intersecting Jordan curves on a 2-D plane can always be drawn on the plane such that they are topologically equivalent to two non-intersecting concentric circles. We will refer to such a structure as an annulus (See Figure 1(b)). Intuitively, an annulus will have an interior and the exterior boundary once drawn on a plane as shown in Figure 1(b). The enclosed region between the two Jordan curves in a given planar projection of an annulus will be referred to as the interior of the annulus.

Definition 2.1.

We call a Barnette graph GG to be Annular Decomposable, if there exists a planar embedding of GG such that:

  1. 1.

    GG can be partitioned into sequence of annuli A1,,AxA_{1},\dots,A_{x}

  2. 2.

    The interior boundary of A1A_{1} is a face called interior face fif_{i}

  3. 3.

    The exterior boundary of AxA_{x} is a face called exterior face fef_{e}

  4. 4.

    Interior boundary of some annulus AiA_{i}, xi>1x\geq i>1, is the exterior boundary for some annulus Ai1A_{i-1}

  5. 5.

    Every edge in vivjE(G)v_{i}v_{j}\in E(G) are such that, either viv_{i} and vjv_{j} lie on the boundary of some annulus, or viv_{i} lies in outer boundary of some annulus AkA_{k} and vjv_{j} lies on the inner boundary of AkA_{k} (1kx1\leq k\leq x).

If GG can be partitioned into xx annuli, we call GG an xx-Annular Decomposable Barnette (xx-ADB graph) graph.

Definition 2.2.

Let GG be a xx-ADB graph. If in a planar embedding of GG, A1,,AxA_{1},\dots,A_{x}, fif_{i} and fef_{e} are realized, we call it a planar annular embedding of GG.

Definition 2.3.

Let GG be a xx-ADB graph, x>2x>2. If for every annulus AkA_{k}, 1kx1\leq k\leq x, deleting all edges (along with respective vertices) in the interior of AkA_{k} produces either two separate Barnette graphs or a Barnette graph and a vertex-less Jordan curve, then we call GG an ADB annular connected (xx-ADB-AC graph) graph.

In this article we prove that ADB-AC graphs having non singular sequences of ring annuli are Hamiltonian.

Lemma 2.4.

There exists ADB graphs that are not annular-connected.

Proof.

Let GG be an nn-ADB graph. Then HH be a copy of GG. Consider two consecutive edges eGe_{G} and eGe_{G}^{\prime} that lie in the interior of the outer annuals of GG, that are adjacent to vertices vG1v_{G}^{1} and vG2v_{G}^{2} of GG. Consider the two analogous edges eHe_{H} and eHe_{H}^{\prime} in HH, that are adjacent to vertices vH1v_{H}^{1} and vH2v_{H}^{2} of GG. Let fGf_{G} and fHf_{H} be the edges on the outer boundary of the outer annuli of GG and HH, connecting vG1v_{G}^{1} with vG2v_{G}^{2} and vH1v_{H}^{1} with vH2v_{H}^{2} respectively. Note that each of the four vertices, vG1v_{G}^{1}, vG2v_{G}^{2}, vH1v_{H}^{1} and vH2v_{H}^{2}, there is yet another edge adjacent to the vertices which we will denote as, evG1e_{v_{G}^{1}}, evG2e_{v_{G}^{2}}, evH1e_{v_{H}^{1}} and evH2e_{v_{H}^{2}} respectively. We now form a new graph JJ by the following steps:

  1. 1.

    Join fGf_{G} and fHf_{H} by drawing two edges across the existing edges

  2. 2.

    Join evG1e_{v_{G}^{1}} and evH1e_{v_{H}^{1}} by drawing one edge across the existing edges

  3. 3.

    Join evG2e_{v_{G}^{2}} and evH2e_{v_{H}^{2}} by drawing one edge across the existing edges

The resulting graph KK has 2x+12x+1 annuli and it is easy to see that KK is an ADB graph, where xx is the number of annuli in GG. Note that GG can be viewed as a subgraph of KK. Now we delete all the edges in the interior of the outer annulus of GG, to obtain two graphs KK^{\prime} and K′′K^{\prime\prime} with x+1x+1 and x1x-1 annuli respectively. Note that, KK^{\prime} is not a Barnette graph since it is not bipartite. (See Figure 2) ∎

Refer to caption
Figure 2: The graph shown in the figure is an ADB graph. Note that if we delete the edges in the interior of the outer annulus of the right hand side graph as shown by the dotted lines then the graph is decomposed into a Barnette AD graph (left hand side) and a non-Barnette AD graph (right hand side). The 5-cycles are marked in grey.
Definition 2.5.

Let GG be ADB-AC graph drawn on a plane. We denote the outer annulus of GG as OGO_{G}.

Definition 2.6.

Let GG be ADB-AC graph drawn on a plane. Let GG have xx annuli. For some x>k>0x>k>0, we define the graph formed by considering restricting GG only until the kk-th annulus, to be the kk-annular restriction of GG, denoted by GkG_{k}. Note that we clarify here that in OGkO_{G_{k}}, we do not consider any vertex adjacent to the edges in the interior of OGk+1O_{G_{k+1}}.

3 Main Results

Lemma 3.1.

Let GG be a Barnette graph. Then GG is 2-edge connected.

Proof.
Refer to caption
Figure 3: Figure showing operations that can be used to build all possible Barnette graphs from the smallest Barnette graph shown in Figure 1-(a), proved by Holton, Manvel and McKay [7].

Holton, Manvel and McKay proved that every Barnette graph can be constructed from the smallest Barnette graph B0B_{0} (shown in Figure 1-(a)), by using the operations shown in Figure 3. Given that B0B_{0} is 22-edge connected, it is easy to see by induction that any Barnette graph GG, would be 22-edge connected. ∎

Note that given a Hamiltonian Barnette Graph GG, if we apply the operation described in Figure 3(b), which we will henceforth call a β\beta-operation, any arbitrary number of times, the result is still a Hamiltonian graph. However, this is not true if we apply the operation described in Figure 3(a), which we will henceforth call a α\alpha-operation, any arbitrary number of times. Later, in Theorem 3.6, we show that ADB-AC graphs can be constructed from B0B_{0} by using arbitrary number of α\alpha-operations only.

Lemma 3.2.

If every annulus of an ADB graph GG has a 44-Cycle, then GG is Hamiltonian. Note that this statement is true for any AD-Planar graph.

Proof.

The proof of the Lemma is clear from Figure 4. Note that in Figure 4, the illustrated graph has an odd number of annuli. If we consider a graph with even number of annuli, the same strategy will still work. ∎

Refer to caption
Figure 4: Figure showing strategy of constructing a Hamiltonian cycle given an ADB graph with a 44-cycle in every annulus.

Let us consider an ADB-AC graph GG. Let us consider the outer annulus of the graph GG. There are even number of faces in the outer annulus of GG since, the outer boundary of the outer annulus of GG, is an even cycle. Thus, the set of faces in the outer annulus of GG, denoted by F(OG)F(O_{G}) can be partitioned into subsets F1(OG)F^{1}(O_{G}) and F2(OG)F^{2}(O_{G}), such that no two faces in Fi(OG)F^{i}(O_{G}), i{1,2}i\in\{1,2\} are adjacent to each other.

Lemma 3.3.

Let us consider an ADB-AC graph GG with x3x\geq 3 annuli. Either F1(OG)F^{1}(O_{G}) or F2(OG)F^{2}(O_{G}) consists of only 44-cycles.

Proof.

Let us create a graph HH, by deleting all the edges in the interior of the outer annulus (along with their respective vertices). HH would be an ADB-AC graph, since GG is an ADB-AC graph. Thus, F(OH)F(O_{H}) can be partitioned into subsets F1(OH)F^{1}(O_{H}) and F2(OH)F^{2}(O_{H}), such that no two faces in Fi(OH)F^{i}(O_{H}), i{1,2}i\in\{1,2\} are adjacent to each other.

We claim that, the edges in the interior of the outer annulus of OGO_{G} are all attached to the outer boundary of faces in OHO_{H}, such that all such faces belong either in F1(OH)F^{1}(O_{H}) or in F2(OH)F^{2}(O_{H}), but not both. To prove this claim, let us assume otherwise. Then, there exits two edges e1e_{1} and e2e_{2} in the interior of OGO_{G}, such that e1e_{1} is attached to the outer boundary of some face f1F1(OH)f^{1}\in F^{1}(O_{H}) and e2e^{2} is attached to some face f2F2(OH)f^{2}\in F^{2}(O_{H}). Let, v1v_{1} and v2v_{2} be the two vertices by which e1e_{1} and e2e_{2} are attached to the outer boundary of OHO_{H}. Then there must be an odd number of vertices between v1v_{1} and v2v_{2}, considering the fact that f1f^{1} and f2f^{2} are in different partitions. Thus, there is an odd cycle in GG, which is impossible, since GG is bipartite.

Let us assume, without loss of generality, that all edges in interior of OGO_{G} are attached to outer boundary of faces in F1(OH)F^{1}(O_{H}). Note that for an arbitrary face fF1(OH)f\in F^{1}(O_{H}), to which some of such edges are attached, we can assert that there are only even number of edges attached to ff, Otherwise, ff would transform into a odd cycle (face) ff^{\prime} in GG. Thus, there would always be a sequence of odd number of 44-cycles, c1,,c2k+1c_{1},\dots,c_{2k+1} (for some integer k1k\geq 1) formed in OGO_{G} adjacent to ff^{\prime}. Starting from c1c_{1}, if we consider all alternate faces in OGO_{G}, they will all be 44-cycles and will form a face partition of 44-cycles in OGO_{G}. Thus the statement holds. ∎

As per convention, we will assume F1(OG)F^{1}(O_{G}) to be the face partition containing 44-cycles only in OGO_{G}. Note here that, both the inner and outer annulus in any planar annular embedding of an 22-ADB graph GG must have at least two 44-cycles each. This along with Lemma 3.2 gives the following corollary.

Corollary 3.4.

Every 22-ADB graph is Hamiltonian.

Refer to caption
Figure 5: Figure showing the recursive pyramid strategy to find a Hamiltonian cycle for Statement 1 in Theorem 3.7. The grey region is the path connected region that is extended recursively through the annuli. The dots show possibility of other 44-faces in the kk-th annular restriction.
Corollary 3.5.

Every GG be an xx-ADB graph. If there exists a planar annular projection of GG such that the kk-th annulus 2kx2\leq k\leq x has more edges in its interior than the number of faces in the interior of the k1k-1th annulus, then GG is Hamiltonian.

Proof.

The proof of this statement also follows from Lemma 3.2. Note that if there exists such a graph GG, then each of its annulus will have a 44-cycle. ∎

Theorem 3.6.

Let us consider an ADB-AC graph GG. Then GG can be constructed from B0B_{0} by using arbitrary number of α\alpha-operations only.

Proof.

Let GG have xx annuli for a particular planar annular embedding. For x=1x=1, the result is obviously true.

Recall that, as per our convention, the 44-face-only partition in the outer annulus of an ADB-AC graph GG^{\prime} is assumed to be F1(OG)F^{1}(O_{G^{\prime}}). Let x>1x>1. By Lemma 3.3,say F1(OG)F^{1}(O_{G}) consists of only 44-cycles. Let us select any two such 44-cycles mm and mm^{\prime}. All other edges in OGO_{G} can be removed by performing a inverse of alphaalpha-operations. Then, we perform an inverse α\alpha-operation on the edges e1e_{1} and e2e_{2} in mm such that, e1e_{1} and e2e_{2} are not in the interior of OGO_{G}. Now, we perform another inverse of α\alpha-operation by deleting edges e1e^{\prime}_{1} and e2e^{\prime}_{2} in mm^{\prime} such that, e1e^{\prime}_{1} and e2e^{\prime}_{2} were in the interior of OGO_{G} in GG. This gives us the (x1)(x-1)-annular restriction of GG. By continuing this process we can ultimately reduce the graph to 11-annular restriction of GG. Thus it is clear that GG can be constructed from B0B_{0} by using arbitrary number of α\alpha-operations only. ∎

Theorem 3.7.

Let us consider an ADB-AC graph GG, with x3x\geq 3 annuli.

  1. 1.

    For any kk, 2kx2\leq k\leq x, if the edges interior to kk-th annulus in GG are attached to faces that are in the 44-cycle-only face partition in O(Hk1)O(H_{k-1}), where HH is the k1k-1 annular restriction of GG , then GG is Hamiltonian.

  2. 2.

    For any kk, 2kx2\leq k\leq x, if the edges interior to kk-th annulus in GG are attached to faces that are not in 44-cycle-only face partition in O(Hk1)O(H_{k-1}), where HH is the k1k-1 annular restriction of GG, then GG is Hamiltonian.

Proof.

For the first statement, it is easy to construct the Hamiltonian cycle in GG, in a recursive manner. We start by a planar annular projection of GG. We include the interior face fif_{i} in the interior of a path connected region RR. Recalling our convention that, the 44-face-only partition in the outer annulus of an ADB-AC graph GG^{\prime} is assumed to be F1(OG)F^{1}(O_{G^{\prime}}), we can describe this strategy by extending HH through kk-th annulus by coloring the faces in F1(OGk)F^{1}(O_{G_{k}}) in grey, for every kk-recursively. Thus we gradually extend RR following the strategy shown in Figure 5. It is easy to see that, when we recursively apply this strategy for all the annuli in GG, the boundary of the final path connected region would manifest into a Hamiltonian cycle. We will call this strategy of constructing Hamiltonian cycles henceforth as the pyramid strategy.

Recall that, as per our convention, the 44-face-only partition in the outer annulus of an ADB-AC graph GG^{\prime} is assumed to be F1(OG)F^{1}(O_{G^{\prime}}). Let x>1x>1. For the second statement, note that for any kk, 2kx2\leq k\leq x, if the edges interior to kk-th annulus in GG are attached to faces that are not in 44-cycle-only face partition in O(Hk1)O(H_{k-1}), F2(OH)F^{2}(O_{H}), where HH is the k1k-1 annular restriction of GG, then for the kk-th annulus, there is a 44-cycle. The reason is no edges in the k+1k+1-th annulus are attached to the face partition with 44-cycles of the kk-th annulus. Thus by Lemma 3.2, GG has to be Hamiltonian. The Hamiltonian cycle can be constructed using the same strategy as depicted in Figure 3.2. We will call this strategy of constructing Hamiltonian cycles henceforth as the ring strategy. ∎

Let GG be ADB-AC graph drawn on a plane. From Theorem 3.7, it is evident that the annuli of GG can be one of two possible types depending on how the edges in the interior of a given annulus are attached to the previous annulus. For the kk-th annulus AA, in GG if the edges in the interior of the annulus AA are attached to the k1k-1-th annulus such that the edges are attached to the faces in the face partition of OGk1O_{G_{k-1}} consisting of only 44-cycles, then we call AA a block annulus, otherwise we call AA a ring annulus. We can thus restate Theorem 3.7 as: Given an ADB-AC graph GG, if GG consists of only block annuli or only ring annuli, then GG is Hamiltonian. Note that the spider-web graphs that motivated the work are a subclass of ADB graphs that consist of only block annuli. Also note that given a planar representation ADB-AC graph GG, the first and second annulus can be considered both a block annulus and a ring annulus. By convention, we will consider them as of the same category as the third annulus.

Corollary 3.8.

Let GG be an ADB-AC graph consisting only of ring annuli, A1,,AxA_{1},\dots,A_{x}. Let |FA||F_{A}| denote the number of faces in an annulus. Then GG must have at least 12xΠi=1x(|FAi|)\frac{1}{2^{x}}\Pi_{i=1}^{x}(|F_{A_{i}}|) Hamiltonian cycles.

Proof.

Let GG be an ADB-AC graph consisting only of ring annuli A1,,AxA_{1},\dots,A_{x}. Any arbitrary annulus AiA_{i} must have |FAi||F_{A_{i}}| 44-cycles. One 44-cycle from every annuli can be used to create a Hamiltonian cycle. Thus, it is clear that GG has at least 12xΠi=1x(|FAi|)\frac{1}{2^{x}}\Pi_{i=1}^{x}(|F_{A_{i}}|) Hamiltonian cycles. Given that |FAi||F_{A_{i}}| is at least 44, we can derive that any such graph will have at least 2x2^{x} Hamiltonian cycles. ∎

In Theorem 3.7, we have already discussed a strategy of finding Hamiltonian cycles in an ADB-AC graph GG, given that GG consists of only block annuli. Now we will explore some more strategies to find a Hamiltonian cycle in GG.

Lemma 3.9.

Let us consider an ADB-AC graph GG, with x3x\geq 3 annuli, containing only block annuli. There are at least two more strategies for finding Hamiltonian cycles in GG, in addition to the one described in Theorem 3.7.

Proof.

Recall that, as per our convention, the 44-face-only partition in the outer annulus of an ADB-AC graph GG^{\prime} is assumed to be F1(OG)F^{1}(O_{G^{\prime}}). Let x>1x>1. Let us consider a planar projection of GG such that fif_{i} and fof_{o} are fixed and well defined. Note that such a graph GG must contain a specific substructure described as follows. This substructure contains one face from every annulus such that the faces are stacked on each other (xx faces stacked on one another). Let us denote these faces as g1,,gxg_{1},\dots,g_{x} and observe that gkF1(OGk)g_{k}\in F^{1}(O_{G_{k}}) for the kk annular restriction of GG, 3kx3\leq k\leq x. Also the face gxg_{x} in the xx-th annulus is a 44-cycle. Existence of such a structure is ensured due to the fact that GG has only block annuli. We call such a structure a pyramid. Now, note that, since GG is 22-edge connected, there has to be at least two pyramids in GG. We denote them as B1B_{1} and B2B_{2}.

Note that, in the first annulus of GG, there must exist a partition of faces with all 44-cycles. We choose one from them and call it mm^{\prime}. Our strategy to build a second Hamiltonian cycle in GG is based on these pyramids and the 44-cycle mm^{\prime}. The strategy is as follows:

Refer to caption
Figure 6: Figure showing the pyramid-ring strategy of finding Hamiltonian cycle in ADBAC graph GG with all block annuli.
  1. 1.

    Build a Hamiltonian cycle H1H_{1} in GG using the strategy mentioned in Theorem 3.7. For the sake of understanding let us assume that we color the interior of the Hamiltonian cycle in grey as shown in Figure 5.

  2. 2.

    Note that all the faces in B1B_{1} and B2B_{2} are colored in grey, i.e. are interior to the cycle. Also, mm^{\prime} is colored in white, i.e. is exterior to the cycle.

  3. 3.

    Remove fif_{i} from HH, i.e color fif_{i} in white.

  4. 4.

    Color all faces in annulus 11 except for mm^{\prime}. Call the new grey region H2H_{2}.

  5. 5.

    Remove coloring from all faces in annulus x1x-1 except for the face in annulus x1x-1 that belongs in B1B_{1}.

  6. 6.

    Color all faces in annulus xx except for the 44-cycles attached to the face h1h_{1}, such that h1B1O(Gx1)h_{1}\in B_{1}\cap O(G_{x-1}) and any one 44-cycle mm attached to the face h2h_{2}, such that h2B2O(Gx1)h_{2}\in B_{2}\cap O(G_{x-1}). Also, we choose mm such that the mF1(OG)m\in F^{1}(O_{G}).

  7. 7.

    Among the uncolored 44-cycles mentioned in the step above, color the ones that are in F2(OG)F^{2}(O_{G}).

After Step 33, the grey region gets path-disconnected but retains path-connectivity after Step 44. Step 44 also gives a Hamiltonian cycle H2H_{2} in GG. After Step 55, note that only some vertices in the outer boundary of the x1x-1-th annulus are in the interior of the white region. After Step 66 these vertices again lie in the boundary of the grey-white region. Note both the grey and white regions still remain path connected (the existence of mm ensures the path connectedness of the white region). After Step 77 all the vertices lie in the boundary of the grey-white coloring and both the white and the grey region remain path connected. Thus the boundary of the grey-white path connected regions is a new Hamiltonian cycle H3H_{3}. We will call this strategy of constructing a Hamiltonian cycle in GG as pyramid-ring strategy or pr-strategy. The pr-strategy have been demonstrated in Figure 6. ∎

Theorem 3.10.

Let us consider a planar annular-embedding of an ADB-AC graph GG, with x+yx+y annuli, containing a sequence of x1x\geq 1 block annuli followed by a sequence of y1y\geq 1 ring annuli. Then GG is Hamiltonian.

Proof.

Recall that, as per our convention, the 44-face-only partition in the outer annulus of an ADB-AC graph GG^{\prime} is termed as F1(OG)F^{1}(O_{G^{\prime}}). When x=1,2x=1,2 we treat all the annuli as ring annuli. For x3x\geq 3 block annuli, we first employ the pr-strategy on GxG_{x}. Since the x+1x+1-th annulus and all annuli afterwards are ring annuli, there must exist a 44-cycle in the x+1x+1-th annulus such that the edges of the 44 face are attached to some face fF2(OGx)f\in F^{2}(O_{G_{x}}). Since all the faces in F2(OGx)F^{2}(O_{G_{x}}) are marked in grey as per the pr-strategy, we can extend the grey color to ff and follow the ring strategy for the rest of the y1y-1 annuli after that to obtain a Hamiltonian cycle (whose interior is marked in grey). ∎

Theorem 3.11.

Let us consider a planar annular embedding of an ADB-AC graph GG, with x+yx+y annuli, containing a sequence of x1x\geq 1 ring annuli followed by a sequence of y1y\geq 1 block annuli. Then GG is Hamiltonian.

Proof.

If x=1,2x=1,2, then we can consider all the annuli as block annuli and use the pr-strategy for finding a Hamiltonian cycle. Recall that, as per our convention, the 44-face-only partition in the outer annulus of an ADB-AC graph GG^{\prime} is termed as F1(OG)F^{1}(O_{G^{\prime}}). If x>2x>2, first we can create a Hamiltonian cycle in Gx1G_{x-1} using the ring strategy, and color it’s interior grey such that only one 44-face in O(Gx1)O(G_{x-1}) is colored in white. Until now, for xx-th annulus in Gx1G_{x-1}, the 44-face-only face partition has been conventionally F1(OGx1)F^{1}(O_{G_{x-1}}). We alter this convention from this annulus and call them F2(OGx1)F^{2}(O_{G_{x-1}}). Note that, this allows us to extend the grey region (Hamiltonian cycle) from x1x-1th annulus to the x+yx+yth annulus, using the pr-strategy as described in Lemma 3.9. ∎

From Theorems 3.10 and 3.11, it is clear that:

Theorem 3.12.

Every ADB-AC graph GG, with arbitrary sequence of ring and block annuli is Hamiltonian, if every sequence ring annuli that lies between two sequences of block annuli, has length at least 22, that is, there are only non singular sequences of ring annuli in GG.

Open questions

We propose the following open questions from our study:

  1. 1.

    Are all ADB-AC graphs Hamiltonian?

  2. 2.

    Are all ADB graphs Hamiltonian?

Acknowledgements

I thank my PhD supervisor Olaf Wolkenhauer for his constant encouragement and unfaltering support for this research. This work was in part supported by funds from Bioinformatics Infrastructure (de.NBI) and Establishment of Systems Medicine Consortium in Germany e:Med , as well as the German Federal Ministry for Education and Research (BMBF) programs (FKZ 01ZX1709C). I thank Miss Shukla Sarkar for her patient support.

References

  • [1] Behrooz Bagheri Gh, Tomas Feder, Herbert Fleischner, and Carlos Subi. Hamiltonian cycles in planar cubic graphs with facial 2-factors, and a new partial solution of barnette’s conjecture. Journal of Graph Theory, pages 1–20, 2020.
  • [2] Jean-Claude Bermonf, Johny Bond, Carole Martin, Aleksandar Pekec, and Fred S. Roberts. Optimal orientations of annular networks. Journal of Interconnection Networks, 01(01):21–46, 2000.
  • [3] Jan Florek. On barnette’s conjecture. Discrete Mathematics, 310(10):1531 – 1535, 2010.
  • [4] Jan Florek. On barnette’s conjecture and h+-h+- property. Electron. Notes Discret. Math., 43:375–377, 2013.
  • [5] Jan Florek. Remarks on barnette’s conjecture. Journal of Combinatorial Optimization, 39(1):149 – 155, 2020.
  • [6] M. Hasheminezhad, S. Mehdi Hashemi, B. McKay, and M. Tahmasbi. Rectangular-radial drawings of cubic plane graphs. Computational Geometry, 43(9):767 – 780, 2010.
  • [7] D. A. Holton, B. Manvel, and B. D. McKay. Hamiltonian cycles in cubic 33-connected bipartite planar graphs. J. Combin. Theory Ser. B, 38(3):279–297, 1985.
  • [8] D.A Holton and B.D McKay. The smallest non-hamiltonian 3-connected cubic planar graphs have 38 vertices. Journal of Combinatorial Theory, Series B, 45(3):305 – 319, 1988.
  • [9] J.D. Horton. On two-factors of bipartite regular graphs. Discrete Mathematics, 41(1):35 – 41, 1982.
  • [10] Shin-Shin Kao and Lih-Hsing Hsu. Spider web networks: a family of optimal, fault tolerant, hamiltonian bipartite graphs. Applied Mathematics and Computation, 160(1):269 – 282, 2005.
  • [11] Xiaoyun Lu. A note on barnette’s conjecture. Discrete Mathematics, 311(23):2711 – 2715, 2011.
  • [12] Grace Misereh and Yuri Nikolayevsky. Annular and pants thrackles. Discret. Math. Theor. Comput. Sci., 20, 2018.
  • [13] W. T. Tutte. On Hamiltonian circuits. J. London Math. Soc., 21:98–101, 1946.