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

The Distinguishing Index of Mycielskian Graphs

Rowan Kennedy , Lauren Keough , Mallory Price , Nick Simmons, , Sarah Zaske wilsonan3@mail.gvsu.edu, Grand Valley State University, Allendale Charter Township, MIkeoulaur@gvsu.edu, Grand Valley State University, Allendale Charter Township, MIpricemal@gvsu.edu, Grand Valley State University, Allendale Charter Township, MIsimmonni@mail.gvsu.edu, Grand Valley State University, Allendale Charter Township, MIzaskes@mail.gvsu.edu, Grand Valley State University, Allendale Charter Township, MI
Abstract

The distinguishing index gives a measure of symmetry in a graph. Given a graph GG with no K2K_{2} component, a distinguishing edge coloring is a coloring of the edges of GG such that no non-trivial automorphism preserves the edge coloring. The distinguishing index, denoted Dist(G)\operatorname{Dist^{\prime}}(G), is the smallest number of colors needed for a distinguishing edge coloring. The Mycielskian of a graph GG, denoted μ(G)\mu(G), is an extension of GG introduced by Mycielski in 1955. In 2020, Alikhani and Soltani conjectured a relationship between Dist(G)\operatorname{Dist^{\prime}}(G) and Dist(μ(G))\operatorname{Dist^{\prime}}(\mu(G)). We prove that for all graphs GG with at least 3 vertices, no connected K2K_{2} component, and at most one isolated vertex, Dist(μ(G))Dist(G)\operatorname{Dist^{\prime}}(\mu(G))\leq\operatorname{Dist^{\prime}}(G), exceeding their conjecture. We also prove analogous results about generalized Mycielskian graphs. Together with the work in 2022 of Boutin, Cockburn, Keough, Loeb, Perry, and Rombach this completes the conjecture of Alikhani and Soltani.

Keywords: distinguishing number, graph distinguishing, graph automorphism, Mycielskian graph

MSC 2020: 05C15

1 Introduction

The Mycielskian of a graph GG, denoted μ(G)\mu(G) was first introduced by Jan Mycielski in 1955 to show that there exist triangle-free graphs of arbitrarily large chromatic numbers [11]. Since then, there have been many counting parameters, such as variations of chromatic numbers and domination numbers studied about Mycielskian graphs, e.g., [8, 10, 7, 4, 13, 6]. For a graph GG with vertices v1,,vnv_{1},\dots,v_{n}, the Mycielskian of G, denoted μ(G)\mu(G), has vertices v1,,vn,u1,,un,wv_{1},\dots,v_{n},u_{1},\dots,u_{n},w. For each edge vivjv_{i}v_{j} in GG, the graph μ(G)\mu(G) has edges vivj,viujv_{i}v_{j},v_{i}u_{j}, and uivju_{i}v_{j}. For each uiu_{i} with 1in1\leq i\leq n, μ(G)\mu(G) has edge uiwu_{i}w. We call v1,,vnv_{1},\dots,v_{n} the original vertices, u1,,unu_{1},\dots,u_{n} the shadow vertices, and ww the root. The Mycielskian of K1,3K_{1,3}, μ(K1,3)\mu(K_{1,3}) is shown in Figure 1.

Figure 1: The Mycielskian of K1,3K_{1,3}, μ(K1,3)\mu(K_{1,3}). The blue vertex is the unique vertex of maximum degree in the original graph, the pink vertex is its shadow, and the red vertex is the root.

The distinguishing number and distinguishing index, which are ways of measuring symmetries in a graph, are two counting parameters that have been studied in relation to Mycielskian graphs [2, 5]. To define these terms we first need to define a graph automorphism. A graph automorphism is a bijective function ϕ:V(G)V(G)\phi:V(G)\xrightarrow{}V(G) such that xx is adjacent to yy if and only if ϕ(x)\phi(x) is adjacent to ϕ(y)\phi(y) for all x,yV(G)x,y\in V(G). A distinguishing vertex coloring of a graph GG is a coloring of the vertices of GG such that no non-trivial automorphism preserves the vertex coloring. The distinguishing number, denoted Dist(G)\operatorname{Dist}(G) is the smallest number of colors needed for a distinguishing vertex coloring of GG. Babai introduced this idea, calling it an asymmetric coloring, in [3]. Albertson and Collins independently introduced the same idea and the name distinguishing coloring in [1].

Similarly, a distinguishing edge coloring of a graph GG is a coloring of the edges of GG such that no non-trivial automorphism preserves the edge coloring and the distinguishing index, denoted Dist(G)\operatorname{Dist^{\prime}}(G) is the smallest number of colors needed for a distinguishing edge coloring of GG. The distinguishing index was defined by Kalinowski and Pils´\acute{\text{s}}niak in [9]. Note that graphs with a connected component that is K2K_{2} do not have a distinguishing index, since any edge coloring of a K2K_{2} has a nontrivial color-preserving automorphism. Similarly, graphs with more than one isolated vertex do not have a distinguishing index. Therefore, throughout the paper we assume all graphs GG do not have a K2K_{2} component, and have at most one isolated vertex.

In 2020 Alikhani and Soltani proved that if GG has no vertices uu and vv such that N(u)=N(v)N(u)=N(v) then Dist(μ(G))Dist(μ(G))+1\operatorname{Dist}(\mu(G))\leq\operatorname{Dist}(\mu(G))+1 whenever GG has at least 22 vertices, and Dist(μ(G))Dist(μ(G))+1\operatorname{Dist^{\prime}}(\mu(G))\leq\operatorname{Dist^{\prime}}(\mu(G))+1 whenever GG has at least 33 vertices and no connected component K2K_{2}. They then proposed Conjecture 1.

Conjecture 1.

[2] Let GG be a connected graph of order n3n\geq 3. Then Dist(μ(G))Dist(G)\operatorname{Dist}(\mu(G))\leq\operatorname{Dist}(G) and Dist(μ(G))Dist(G)\operatorname{Dist^{\prime}}(\mu(G))\leq\operatorname{Dist^{\prime}}(G), except for a finite number of graphs.

Notably, their conjecture does not require the graph to avoid vertices that have identical neighborhoods and tightens the inequality by removing the plus one. Boutin, Cockburn, Keough, Loeb, Perry, and Rombach proved and exceeded the distinguishing vertex coloring part of Conjecture 1 in 2022 [5]. This paper completes the proof of the conjecture by showing the edge distinguishing inequality.

Since our proofs will rely on showing automorphisms are trivial based on facts about the graph and its coloring, we introduce definitions and notations about graphs. Given a vertex vv, the vertex neighborhood of vv, denoted NG(v)N_{G}(v), is the set of all vertices adjacent to vv in GG. The degree of a vertex vv is |NG(v)||N_{G}(v)| and is denoted degG(v)\deg_{G}(v). When the graph GG is clear, we omit the subscript and simply write N(v)N(v) or deg(v)\deg(v). The distance between two vertices uu and vv in GG, denoted d(x,y)d(x,y) is the length of a shortest path between vertices xx and yy.

By definition, automorphisms preserve adjacencies and non-adjacencies between vertices. As a result, automorphisms also preserve vertex degrees and distances between vertices.

2 Distinguishing Index of Mycielskian Graphs

In a graph GG, uu and vv are twins if N(u)=N(v)N(u)=N(v). For any graph GG that contains twins uu and vv, there is an automorphism ϕ\phi of GG such that ϕ(u)=v\phi(u)=v and ϕ(v)=u\phi(v)=u while ϕ(x)=x\phi(x)=x for all xV(G){u,v}x\in V(G)\setminus\{u,v\}. In a vertex distinguishing coloring of GG, this implies twins must receive different colors. In an edge distinguishing coloring of GG with twins, we must prevent the automorphism that switches twins using a coloring of the edges. In [2] Alikhani and Soltani consider graphs without twins to avoid this complication. Lemma 2 establishes a fact about distinguishing edge colorings of graphs with twins that will aid us in proving Conjecture 1.

Lemma 2.

Let GG be a graph with vertices x1x_{1} and x2x_{2} such that N(x1)=N(x2)N(x_{1})=N(x_{2}). If cc is a distinguishing edge coloring of GG, then there exists vN(x1)=N(x2)v\in N(x_{1})=N(x_{2}) such that c(vx1)c(vx2)c(vx_{1})\neq c(vx_{2}).

Proof.

Given a graph GG, label the vertices of V(G)V(G) such that for x1,x2V(G)x_{1},x_{2}\in V(G), N(x1)=N(x2)N(x_{1})=N(x_{2}). We will prove the contrapositive: that is, if for all vN(x1)=N(x2)v\in N(x_{1})=N(x_{2}) we have c(vx1)=c(vx2)c(vx_{1})=c(vx_{2}), then cc is not a distinguishing edge coloring of GG. We assume that for all vN(x1)=N(x2)v\in N(x_{1})=N(x_{2}) we have c(vx1)=c(vx2)c(vx_{1})=c(vx_{2}). Let ϕ\phi be the automorphism that swaps twins x1x_{1} and x2x_{2}, so that ϕ(x1)=x2\phi(x_{1})=x_{2}, ϕ(x2)=x1\phi(x_{2})=x_{1}, and ϕ(x)=x\phi(x)=x for all other vertices. For all vN(x1)=N(x2)v\in N(x_{1})=N(x_{2}) we have c(vx1)=c(vx2)c(vx_{1})=c(vx_{2}), and so c(ϕ(vx1))=c(vx2)=c(vx1)c(\phi(vx_{1}))=c(vx_{2})=c(vx_{1}) and c(ϕ(vx2)=c(vx1)=c(vx2)c(\phi(vx_{2})=c(vx_{1})=c(vx_{2}). Since all other vertices are fixed, ϕ\phi is a nontrivial color-preserving automorphism and cc is not a distinguishing edge coloring. ∎

The strategy for proving Conjecture 1 is to separate out cases where the root ww is fixed in any automorphism. In Figure 1 we can see that there is an automorphism of μ(K1,3)\mu(K_{1,3}) such that ϕ(w)=u0\phi(w)=u_{0}, where u0u_{0} is the shadow vertex of the unique vertex of maximum degree in K1,3K_{1,3}. In [5], the authors prove that for any graph GG and for any automorphism ϕ\phi of μ(G\mu(G) the root ww is fixed unless G=K1,mG=K_{1,m} for some m0m\geq 0. Since we rely on this result for our proofs, we state it precisely below.

Lemma 3.

[5] If there is an automorphism ϕ\phi of μ(G)\mu(G) that takes the root ww to any other vertex, then G=K1,mG=K_{1,m} for some m0m\geq 0. Additionally, if G=K1,mG=K_{1,m} for m2m\geq 2 then ϕ(w){w,u0}\phi(w)\in\{w,u_{0}\} where u0u_{0} is the shadow vertex of the unique vertex of maximum degree in GG.

Note that K1,0K_{1,0} and K1,1=K2K_{1,1}=K_{2} do not have any edge-distinguishing colorings. Thus, for the edge distinguishing problem, we only consider K1,mK_{1,m} for m2m\geq 2. We call the graphs K1,mK_{1,m} for m2m\geq 2 star graphs and consider them in Section 2.1. We consider all other graphs in Section 2.2.

2.1 Star Graphs

In this section we will show that Dist(μ(K1,m))Dist(K1,m)\operatorname{Dist^{\prime}}(\mu(K_{1,m}))\leq\operatorname{Dist^{\prime}}(K_{1,m}) proving Conjecture 1 for star graphs. In fact, Theorem 4 shows that Dist(K1,m)\operatorname{Dist^{\prime}}(K_{1,m}) and Dist(μ(K1,m))\operatorname{Dist^{\prime}}(\mu(K_{1,m})) are only equal when m=2m=2 and can be arbitrarily far apart.

u1u_{1}u2u_{2}u3u_{3}v1v_{1}v2v_{2}v3v_{3}v0v_{0}wwu0u_{0}
Figure 2: A distinguishing edge coloring on μ(K1,3)\mu(K_{1,3}) as described in the proof of Theorem 4.
Theorem 4.

For m2m\geq 2

Dist(μ(K1,m))=r\operatorname{Dist^{\prime}}\left(\mu(K_{1,m})\right)=r

where rr is the minimum natural number such that r2m+1r^{2}\geq m+1. In particular,

Dist(μ(K1,m))m=Dist(K1,m).\operatorname{Dist^{\prime}}\left(\mu(K_{1,m})\right)\leq m=\operatorname{Dist^{\prime}}(K_{1,m}).
Proof.

Let V(K1,m)={v0,v1,v2,,vm}V(K_{1,m})=\{v_{0},v_{1},v_{2},\dots,v_{m}\} be the vertex set of K1,mK_{1,m} such that deg(v0)=m\deg(v_{0})=m. Let u0,u1,u2,,umu_{0},u_{1},u_{2},\dots,u_{m} be the shadow vertices in μ(K1,m)\mu(K_{1,m}), and let ww be the root. Additionally, let I={1,2,,m}I=\{1,2,\dots,m\}. Then we have the vertex set

V(μ(K1,m))={v0,v1,,vm,u0,u1,,um,w}V(\mu(K_{1,m}))=\{v_{0},v_{1},\dots,v_{m},u_{0},u_{1},\dots,u_{m},w\}

and the edge set

E(μ(K1,m))={v0vi,v0ui,u0vi,wui,wu0:iI}.E(\mu(K_{1,m}))=\{v_{0}v_{i},v_{0}u_{i},u_{0}v_{i},wu_{i},wu_{0}:i\in I\}.

We consider two important subsets of the edges - let L1L_{1} and L2L_{2} be sets of ordered pairs of edges such that L1={(u0vi,viv0):iI}L_{1}=\{(u_{0}v_{i},v_{i}v_{0}):i\in I\} and L2={(wui,uiv0):iI}L_{2}=\{(wu_{i},u_{i}v_{0}):i\in I\}. Note that each edge in E(μ(K1,m))E(\mu(K_{1,m})) is in an ordered pair in L1L_{1} or L2L_{2} or is the edge u0wu_{0}w. In Figure 2, L1L_{1} represents the "elbows" drawn on the left, and L2L_{2} represents the "elbows" drawn on the right.

Choose rr least such that r2m+1r^{2}\geq m+1. We will color μ(K1,m)\mu(K_{1,m}) with rr colors. Given rr colors, there are r2r^{2} ordered pairs of colors. Let P={p1,p2,pr2}P=\{p_{1},p_{2},\dots p_{r^{2}}\} be the set of all ordered pairs of rr colors. First, we color the edges in L1L_{1}. For iIi\in I, we assign the color pair pip_{i} to the ordered pair (u0vi,viv0)(u_{0}v_{i},v_{i}v_{0}) so that u0viu_{0}v_{i} gets the first color in the ordered pair pip_{i} and viv0v_{i}v_{0} gets the second color in the ordered pair pip_{i}. Similarly, for the ithi^{th} ordered pair in L2L_{2}, (wui,uiv0)(wu_{i},u_{i}v_{0}), we assign the color pair pi+1p_{i+1}. Since r2m+1r^{2}\geq m+1, there are enough color pairs to do so. Finally, color the edge u0wu_{0}w any of the rr colors.

We will now show this coloring is distinguishing. Suppose ϕ\phi is a color-preserving automorphism on μ(K1,m)\mu(K_{1,m}). By Lemma 3, ϕ(w){w,u0}\phi(w)\in\{w,u_{0}\}. By construction, every pair pjp_{j} for 1jr21\leq j\leq r^{2} is distinct so ϕ(vi)vα\phi(v_{i})\neq v_{\alpha} for any αi\alpha\neq i and ϕ(ui)uα\phi(u_{i})\neq u_{\alpha} for any αi\alpha\neq i. Because the color pair p1p_{1} colors an edge pair in L1L_{1} and does not color any edge pair in L2L_{2}, there is no βI\beta\in I such that ϕ(v1)=uβ\phi(v_{1})=u_{\beta}. So ϕ(vi)=vi\phi(v_{i})=v_{i} and ϕ(ui)=ui\phi(u_{i})=u_{i} for every iIi\in I. It follows that ϕ(w)=w\phi(w)=w and ϕ(u0)=u0\phi(u_{0})=u_{0}. So we have a distinguishing edge coloring for μ(K1,m)\mu(K_{1,m}) with rr colors and Dist(μ(K1,m))r\operatorname{Dist^{\prime}}(\mu(K_{1,m}))\leq r.

To show that Dist(μ(K1,m))=r\operatorname{Dist^{\prime}}(\mu(K_{1,m}))=r, suppose that we have a distinguishing coloring with ss colors such that s<rs<r. Let P={p1,p2,ps2}P^{\prime}=\{p_{1},p_{2},\dots p_{s^{2}}\} be the set of all ordered pairs of ss colors. Since s<rs<r, and rr is least such that r2m+1r^{2}\geq m+1, s2<m+1s^{2}<m+1 and PP^{\prime} contains at most mm color pairs. Note that for every iIi\in I, N(vi)={u0,v0}N(v_{i})=\{u_{0},v_{0}\} and hence for all i,jIi,j\in I, viv_{i} and vjv_{j} are twins. By Lemma 2, every edge pair in L1L_{1} must be colored with a different color pair. Since there are mm edge pairs in L1L_{1} we must have s2ms^{2}\geq m. Since s2ms^{2}\leq m by assumption, suppose s2=ms^{2}=m. In this case all color pairs must be used to color the edge pairs of L1L_{1} and all color pairs must be used to color the edge pairs of L2L_{2}. Because L1L_{1} and L2L_{2} use all of the color pairs of PP^{\prime}, we can construct a non-trivial automorphism ϕ\phi that is color preserving where ϕ(w)=u0\phi(w)=u_{0}. So we have found that we cannot make a distinguishing edge coloring with fewer than rr colors and Dist(μ(K1,m))=r\operatorname{Dist^{\prime}}(\mu(K_{1,m}))=r.

Note that Dist(K1,m)=m\operatorname{Dist}^{\prime}(K_{1,m})=m for if any two edges were the same color, their degree 11 vertices could be switched in a color preserving automorphism. For m2m\geq 2, we know m2m+1m^{2}\geq m+1 and hence r=mr=m satisfies r2m+1r^{2}\geq m+1. Since rr is least such that r2m+1r^{2}\geq m+1, rmr\leq m and we conclude

Dist(μ(K1,m))=rm=Dist(K1,m).\operatorname{Dist^{\prime}}\left(\mu(K_{1,m})\right)=r\leq m=\operatorname{Dist^{\prime}}(K_{1,m}).

2.2 Not Star Graphs

In this section we complete the proof of Conjecture 1 by proving Dist(μ(G))Dist(G)\operatorname{Dist^{\prime}}(\mu(G))\leq\operatorname{Dist^{\prime}}(G) for all non-star graphs for which Dist(G)\operatorname{Dist^{\prime}}(G) is defined.

Theorem 5.

Let GG be a graph with no connected K2K_{2} component and at most one isolated vertex GK1,mG\neq K_{1,m} for any mm. Then Dist(μ(G))Dist(G)\operatorname{Dist^{\prime}}(\mu(G))\leq\operatorname{Dist^{\prime}}(G).

Proof.

Define the vertices of μ(G)\mu(G) to be v1,,vnv_{1},\dots,v_{n}, their shadows to be u1,,unu_{1},\dots,u_{n} and the root to be ww. Let cc be a distinguishing coloring of E(G)E(G) with colors 1,,q1,\dots,q. We define a coloring c¯\overline{c} of E(μ(G))E(\mu(G)) that “mimics" the coloring on E(G)E(G). If xy{vjvk,vjuk,vkuj}xy\in\{v_{j}v_{k},v_{j}u_{k},v_{k}u_{j}\} for any 1k,jn1\leq k,j\leq n then c¯(xy)=c(vjvk)\overline{c}(xy)=c(v_{j}v_{k}). Lastly, c¯(wui)=1\overline{c}(wu_{i})=1 for all 1in1\leq i\leq n.

Let ϕ\phi be a color-preserving automorphism. Since GK1,mG\neq K_{1,m} we know by Lemma 3 that for the root ww of μ(G)\mu(G), ϕ(w)=w\phi(w)=w. For all 1in1\leq i\leq n, d(ui,w)=1d(u_{i},w)=1 and d(vi,w)=2d(v_{i},w)=2. Therefore, for all 1in1\leq i\leq n, ϕ(ui)=uj\phi(u_{i})=u_{j} for some 1jn1\leq j\leq n and ϕ(vi)=vj\phi(v_{i})=v_{j} for some 1jn1\leq j\leq n. That is, the levels of μ(G)\mu(G) are fixed set-wise. Moreover, c¯\overline{c} fixes viV(μ(G))v_{i}\in V(\mu(G)) for 1in1\leq i\leq n because c¯\overline{c} restricts to a distinguishing edge coloring of GG. Let uku_{k} be the shadow of a non-twin vertex in μ(G)\mu(G). Then uku_{k} has a unique and fixed neighborhood, and ϕ(uk)=uk\phi(u_{k})=u_{k} for all such uku_{k}.

We will show that twins ui,ujV(μ(G))u_{i},u_{j}\in V(\mu(G)) are fixed using a proof by contradiction. So, we assume that ui,uju_{i},u_{j} are not fixed by ϕ\phi, so ϕ(ui)=uj\phi(u_{i})=u_{j} and ϕ(uj)=ui\phi(u_{j})=u_{i}. Since ui,uju_{i},u_{j} are twins, there are twins viv_{i} and vjv_{j} in GG where N(vi)=N(vj)N(v_{i})=N(v_{j}), so by Lemma 2, there exists some v0N(vi)=N(vj)v_{0}\in N(v_{i})=N(v_{j}) such that c(vkvi)c(v0vj)c(v_{k}v_{i})\neq c(v_{0}v_{j}). Since c(vkvi)=c¯(vkui)c(v_{k}v_{i})=\overline{c}(v_{k}u_{i}) and c(vkvj)=c¯(vkuj)c(v_{k}v_{j})=\overline{c}(v_{k}u_{j}) we know c¯(vkui)c¯(vkuj)\overline{c}(v_{k}u_{i})\neq\overline{c}(v_{k}u_{j}). Then ϕ\phi cannot be a color-preserving automorphism on μ(G)\mu(G) which is a contradiction. Thus uiu_{i} and uju_{j} are fixed as desired. ∎

Theorems 4 and 5 prove Conjecture 1. In fact, we exceed the conjecture as the proofs do not require GG to be connected and there are no exceptions besides those for which Dist(G)\operatorname{Dist^{\prime}}(G) is undefined.

Corollary 6.

For all graphs GG with |V(G)|3|V(G)|\geq 3, no connected K2K_{2} component, and at most one isolated vertex

Dist(μ(G))Dist(G).\operatorname{Dist^{\prime}}(\mu(G))\leq\operatorname{Dist^{\prime}}(G).

3 Generalized Mycielskian Graphs

The generalized Mycielskian, sometimes called the cone over a graph was introduced by Steibitz [12]. Let GG be a graph with vertices v1,,vnv_{1},\dots,v_{n}. Then the generalized Mycielskian of GG with tt layers, denoted μt(G)\mu_{t}(G), has vertex set

V(μt(G))={u10,,un0,u11,,un1,,u1t,,unt,w}.V(\mu_{t}(G))=\{u_{1}^{0},\dots,u_{n}^{0},u_{1}^{1},\dots,u_{n}^{1},\dots,u_{1}^{t},\dots,u_{n}^{t},w\}.

The vertices u10,,un0u_{1}^{0},\dots,u_{n}^{0} represent the "original" vertices v1,,vnv_{1},\dots,v_{n} which can be thought of as the zeroth layer. For each edge vivkv_{i}v_{k} in GG with 1i,kn1\leq i,k\leq n, μt(G)\mu_{t}(G) has edges ui0uk0u_{i}^{0}u_{k}^{0}, uijukj+1u_{i}^{j}u_{k}^{j+1}, and ukjuij+1u_{k}^{j}u_{i}^{j+1} for 0j<t0\leq j<t. Lastly, μt(G)\mu_{t}(G) has the edges uitwu_{i}^{t}w for 1in1\leq i\leq n. In Figure 3, μ(K1,2)\mu(K_{1,2}), μ2(K1,2)\mu_{2}(K_{1,2}) and μ3(K1,2)\mu_{3}(K_{1,2}) are shown.

Figure 3: From left to right, μ(K1,2)\mu(K_{1,2}), μ2(K1,2)\mu_{2}(K_{1,2}) and μ3(K1,2)\mu_{3}(K_{1,2}).

In [5] it is shown that a result analogous to Lemma 3 holds for generalized Mycielskian graphs. We will rely on this result, so we repeat it here, insisting m2m\geq 2.

Lemma 7.

[5] Let GG be a graph, let t1t\geq 1, and let ϕ\phi be an automorphism of μt(G)\mu_{t}(G). If GK1,mG\neq K_{1,m} for any m2m\geq 2 then, for the root vertex ww, we have ϕ(w)=w\phi(w)=w. If G=K1,mG=K_{1,m} for m2m\geq 2 then ϕ(w){w,u0t}\phi(w)\in\{w,u_{0}^{t}\} where u0tu_{0}^{t} is the top-level shadow vertex of the vertex of degree mm in K1,mK_{1,m}.

Motivated by Lemma 7, we again consider star graphs separately.

3.1 Star Graphs

By Lemma 7, the star graphs K1,mK_{1,m} for m2m\geq 2 have an automorphism where, for the root ww, ϕ(w)w\phi(w)\neq w. In particular, for the root ww, ϕ(w){w,u0t}\phi(w)\in\{w,u_{0}^{t}\}. In other words, ww is fixed or is mapped to the shadow vertex of the unique vertex of maximum degree in GG on level tt. The automorphisms that do not fix the root ww are seen in the horizontal symmetry shown in Figure 4.

Theorem 8.

Let m2m\geq 2 and tt\in\mathbb{N}. Then

Dist(μt(K1,m))=r\operatorname{Dist^{\prime}}\left(\mu_{t}(K_{1,m})\right)=r

where rr is the minimum natural number such that r2m+1r^{2}\geq m+1. In particular, for m2m\geq 2.

Dist(μt(K1,m))Dist(K1,m).\operatorname{Dist^{\prime}}(\mu_{t}(K_{1,m}))\leq\operatorname{Dist^{\prime}}(K_{1,m}).
Proof.

Let V(μt((K1,m))={u0j,u1j,u2j,,umj:0jt}{w}V(\mu_{t}((K_{1,m}))=\{u^{j}_{0},u^{j}_{1},u^{j}_{2},\dots,u^{j}_{m}:0\leq j\leq t\}\cup\{w\} where u00u^{0}_{0} is the center of K1,mK_{1,m} and ww is the root.

To define the edge set, we start by defining LiL_{i} for 0it0\leq i\leq t as sets of ordered pairs of edges. Let L0={(u01ui0,ui0u00):iI}L_{0}=\{(u^{1}_{0}u^{0}_{i},u^{0}_{i}u^{0}_{0}):i\in I\}, Lt={(wu0t,uitu0t1):iI}L_{t}=\{(wu^{t}_{0},u^{t}_{i}u^{t-1}_{0}):i\in I\}, and

Lα={(u0α+1uiα,uiαu0α1):1in}L_{\alpha}=\{(u^{\alpha+1}_{0}u^{\alpha}_{i},u^{\alpha}_{i}u^{\alpha-1}_{0}):1\leq i\leq n\}

where 1α<t1\leq\alpha<t. The edges in ordered pairs in each LαL_{\alpha} plus u0twu_{0}^{t}w make up E(μt(K1,m))E(\mu_{t}(K_{1,m})). In Figure 4 the sets LiL_{i} for 0i50\leq i\leq 5 are shown for μ5(K1,3)\mu_{5}(K_{1,3}).

L1L_{1}L5L_{5}L3L_{3}L1L_{1}L2L_{2}L0L_{0}L4L_{4}u05u_{0}^{5}u03u_{0}^{3}u01u_{0}^{1}u00u_{0}^{0}u02u_{0}^{2}u04u_{0}^{4}ww
Figure 4: The sets LiL_{i} for 0i50\leq i\leq 5 in μ5(K1,3)\mu_{5}(K_{1,3}).

Choose rr least such that r2m+1r^{2}\geq m+1. Given rr colors, there are r2r^{2} ordered pairs of colors. Let P={p1,p2,pr2}P=\{p_{1},p_{2},\dots p_{r^{2}}\} be the set of all ordered pairs of rr colors. For each LαL_{\alpha} with 1α<t1\leq\alpha<t, we assign the color pair pip_{i} to the ithi^{th} ordered pair in LαL_{\alpha}, (u0α+1uiα,uiαu0α1)(u^{\alpha+1}_{0}u^{\alpha}_{i},u^{\alpha}_{i}u^{\alpha-1}_{0}) so that u0α+1uiαu^{\alpha+1}_{0}u^{\alpha}_{i} gets the first color in the ordered pair pip_{i} and uiαu0α1u^{\alpha}_{i}u^{\alpha-1}_{0} gets the second color in the ordered pair pip_{i}. Similarly, for the ithi^{th} ordered pair in L0L_{0}, (u01ui0,ui0u00)(u_{0}^{1}u^{0}_{i},u^{0}_{i}u^{0}_{0}), we assign the color pair pip_{i}. Finally, for the ithi^{th} ordered pair in LtL_{t}, (wuit,uitu0t1)(wu^{t}_{i},u^{t}_{i}u^{t-1}_{0}), we assign the color pair pi+1p_{i+1}. Since r2m+1r^{2}\geq m+1, there are enough color pairs to do so. Finally, color the edge u0wu_{0}w any of the rr colors. Examples of this coloring on μ5(K1,3)\mu_{5}(K_{1,3}) and μ2(K1,5)\mu_{2}(K_{1,5}) are shown in Figure 5.

We’ll now prove this coloring is distinguishing. Suppose ϕ\phi is a color-preserving automorphism on μt(K1,m)\mu_{t}(K_{1,m}) and suppose ϕ(w)=u0t\phi(w)=u_{0}^{t}. Since umtu_{m}^{t} has degree 22 and is distance 11 from ww, ϕ(umt)\phi(u_{m}^{t}) must have degree 22 and be distance 11 from ϕ(w)=u0t\phi(w)=u_{0}^{t}. Thus ϕ(umt)=uit1\phi(u_{m}^{t})=u_{i}^{t-1} for some 1it1\leq i\leq t. Similarly, by distance and degree, ϕ(u0t1)=u0t2\phi(u_{0}^{t-1})=u_{0}^{t-2}. Note that (wumt,umtu0t1)Lt(wu_{m}^{t},u_{m}^{t}u_{0}^{t-1})\in L_{t} has color pair pm+1p_{m+1}, and no pair of edges in Lt1={(u0tuit1,uit1u0t2):1in}L_{t-1}=\{(u^{t}_{0}u^{t-1}_{i},u^{t-1}_{i}u^{t-2}_{0}):1\leq i\leq n\} has this color pair. So umtu_{m}^{t} can’t map to uit1u_{i}^{t-1} because then the pair of edges (wumt,umtu0t1)(wu_{m}^{t},u_{m}^{t}u_{0}^{t-1}) would map to the pair of edges (u0tuit1,uit1u0t2)(u_{0}^{t}u_{i}^{t-1},u_{i}^{t-1}u_{0}^{t-2}) and this wouldn’t be color-preserving. So ϕ(w)u0t\phi(w)\neq u_{0}^{t}. Therefore, by Lemma 7, ϕ(w)=w\phi(w)=w.

Since ww is fixed, by distance and degree, the levels of μt(K1,m)\mu_{t}(K_{1,m}) are fixed setwise by ϕ\phi. The construction of our coloring means that on a given level jj with 0jt0\leq j\leq t, we used a distinct color ordered pair for the edges incident to any degree 22 vertex. So, ϕ(uij)ukj\phi(u^{j}_{i})\neq u^{j}_{k} for any kik\neq i. Thus, ϕ(uij)=uij\phi(u^{j}_{i})=u^{j}_{i} for 0jt0\leq j\leq t and 1im1\leq i\leq m. Since we also know ϕ(u0j)=u0j\phi(u^{j}_{0})=u^{j}_{0} for 0jt0\leq j\leq t and ϕ(w)=w\phi(w)=w, we conclude that ϕ\phi is the trivial automorphism and Dist(μt(K1,m))r\operatorname{Dist^{\prime}}(\mu_{t}(K_{1,m}))\leq r.

To show that Dist(μ(K1,m))=r\operatorname{Dist^{\prime}}(\mu(K_{1,m}))=r, we assume for the sake of a contradiction that there’s a distinguishing coloring with ss colors such that s<rs<r. Let P={p1,p2,ps2}P^{\prime}=\{p_{1},p_{2},\dots p_{s^{2}}\} be the set of all ordered pairs of ss colors. Since s<rs<r, and rr is least such that r2m+1r^{2}\geq m+1, we know as2ms^{2}\leq m.

First we’ll show s2ms^{2}\geq m. For each 1jt1\leq j\leq t and 1in1\leq i\leq n, N(uij)={u0j+1,u0j1}N(u^{j}_{i})=\{u^{j+1}_{0},u^{j-1}_{0}\}. Similarly, for 1in1\leq i\leq n, N(ui0)={u01,u00}N(u^{0}_{i})=\{u^{1}_{0},u^{0}_{0}\} and N(uit)={w,u0t1}N(u^{t}_{i})=\{w,u^{t-1}_{0}\}. Hence, for a fixed jj, and all 1a,bn1\leq a,b\leq n, uaju_{a}^{j} and ubju_{b}^{j} are twins. By Lemma 2, every edge pair in LjL_{j} must be colored with a different color pair. There are mm edge pairs in LtL_{t} and s2s^{2} color pairs, so s2ms^{2}\geq m.

Suppose s2=ms^{2}=m. In this case all color pairs must be used to every color the edge pairs of each LjL_{j} for 0jt0\leq j\leq t. Let j\mathcal{L}_{j} be the set of vertices {uij1in}\{u^{j}_{i}\mid 1\leq i\leq n\}. for every jj, each vertex in j\mathcal{L}_{j} is the same distance from ww so we define d(j,w)d(\mathcal{L}_{j},w) to be this distance. We will consider the cases for tt even and tt odd separately.

When tt is odd, for each j\mathcal{L}_{j} that is distance dd from ww there is exactly one other set, call it j\mathcal{L}_{j}^{\prime}, that is distance dd from u0tu^{t}_{0}. Let ψ\psi be the automorphism which swaps ww and u0tu^{t}_{0} and maps the ithi^{th} vertex in j\mathcal{L}_{j} to the vertex in j\mathcal{L}_{j}^{\prime} that has the appropriate colors on its incident edges and complete the automorphism so that distances are preserved. Then ψ\psi is a color preserving automorphism.

If tt is even, then, as seen in Figure 5, 0=0\mathcal{L}_{0}=\mathcal{L}_{0}^{\prime}. Since all mm colorings were used, for any edge pair ϵ\epsilon in L0L_{0} with an ordered pair of different colors, the reverse ordered pair is also assigned to an edge pair δ\delta in L0L_{0}. Let ψ\psi be the automorphism that swaps ww and u0tu^{t}_{0} and for j1j\geq 1, maps the ithi^{th} vertex in j\mathcal{L}_{j} to the vertex in j\mathcal{L}_{j}^{\prime} that has the appropriate colors on its incident edges. Let ψ\psi^{\prime} be the automorphism which swaps the vertices in 0\mathcal{L}_{0} such that ϵ\epsilon maps to δ\delta and vice versa. Then ψψ\psi^{\prime}\psi is a color preserving automorphism.

So we have shown that we can always construct a non-trivial automorphism ϕ\phi that is color preserving when s2=ms^{2}=m. Therefore, we cannot make a distinguishing edge coloring with fewer than rr colors and Dist(μ(K1,m))=r\operatorname{Dist^{\prime}}(\mu(K_{1,m}))=r.

For m2m\geq 2, we know m2m+1m^{2}\geq m+1 and hence r=mr=m satisfies r2m+1r^{2}\geq m+1. Since rr is least such that r2m+1r^{2}\geq m+1, rmr\leq m and we conclude that

Dist(μt(K1,m))rm=Dist(K1,m).\operatorname{Dist^{\prime}}\left(\mu_{t}(K_{1,m})\right)\leq r\leq m=\operatorname{Dist^{\prime}}(K_{1,m}).

u05u_{0}^{5}u03u_{0}^{3}u01u_{0}^{1}u00u_{0}^{0}u02u_{0}^{2}u04u_{0}^{4}wwu00u^{0}_{0}u01u^{1}_{0}wwu02u^{2}_{0}
Figure 5: Distinguishing colorings of μ5(K1,3)\mu_{5}(K_{1,3}) and μ2(K1,5)\mu_{2}(K_{1,5}) using method described in proof of Theorem 8. On the left, m=3m=3 so r=2r=2 and the set P={(red,red),(blue,blue),(red,blue),(blue,red)}P=\{(\text{red},\text{red}),(\text{blue},\text{blue}),(\text{red},\text{blue}),(\text{blue},\text{red})\}. On the right, m=5m=5 so r=3r=3 and the set P={(red,red),(blue,black),(black,blue),(blue,blue),(black,black)}P=\{(\text{red},\text{red}),(\text{blue},\text{black}),(\text{black},\text{blue}),(\text{blue},\text{blue}),(\text{black},\text{black})\}.

3.2 Not Star Graphs

In this section we will show that Dist(μt(G))Dist(G)\operatorname{Dist^{\prime}}(\mu_{t}(G))\leq\operatorname{Dist^{\prime}}(G) for all graphs that are not star graphs and for which Dist(G)\operatorname{Dist^{\prime}}(G) and Dist(μt(G))\operatorname{Dist^{\prime}}(\mu_{t}(G)) are defined. Note that when t2t\geq 2, an isolated vertex in GG will result in 22 isolated vertices in μt(G)\mu_{t}(G). Since isolated vertices can be swapped in any color-preserving automorphism, Dist(μt(G))\operatorname{Dist^{\prime}}(\mu_{t}(G)) would be undefined. So in this section, we insist GG has no connected K2K_{2} component and no isolated vertices.

Theorem 9.

For any graph GG with |V(G)|3|V(G)|\geq 3, no connected K2K_{2} component, no isolated vertices, and GK1,mG\neq K_{1,m} for any mm, then Dist(μt(G))Dist(G)\operatorname{Dist^{\prime}}(\mu_{t}(G))\leq\operatorname{Dist^{\prime}}(G).

Proof.

Let |V(G)|=n|V(G)|=n and define V(μt((G))={u1j,u2j,,unj:0jt}{w}V(\mu_{t}((G))=\{u^{j}_{1},u^{j}_{2},\dots,u^{j}_{n}:0\leq j\leq t\}\cup\{w\} where ww is the root. Let cc be a distinguishing coloring of E(G)E(G) with colors 1,,q1,\dots,q. We define a coloring of E(μt(G))E(\mu_{t}(G)) called c¯\overline{c}. For an edge ui0uk0E(G)u_{i}^{0}u_{k}^{0}\in E(G), we let c¯(ui0uk0)=c¯(uijukj1)=c¯(ukjuij1)=c(ui0uk0)\overline{c}(u_{i}^{0}u_{k}^{0})=\overline{c}(u_{i}^{j}u_{k}^{j-1})=\overline{c}(u_{k}^{j}u_{i}^{j-1})=c(u_{i}^{0}u_{k}^{0}) for 1jt1\leq j\leq t. That is, we mimic the coloring cc so that edges between layers are colored with the same color as the edge that produced them.

Let ϕ\phi be a color-preserving automorphism. We will show ϕ\phi is trivial. To start, since GK1,mG\neq K_{1,m} for any mm we know by Lemma 7 that ww is fixed. Assume for the sake of a contradiction that ϕ(uij)uij\phi(u_{i}^{j})\neq u_{i}^{j} for some 1in1\leq i\leq n and 0jt0\leq j\leq t. Since ww is fixed each level of μt(G)\mu_{t}(G) is fixed setwise. Because the levels are fixed setwise and c¯\overline{c} restricts to a distinguishing edge coloring of GG, ϕ(ui0)=ui0\phi(u_{i}^{0})=u_{i}^{0} for 1in1\leq i\leq n. That is, the vertices on the zeroth level are fixed. Without loss of generality, assume j1j\geq 1 is least such that a vertex on level jj is not fixed. That is, assume if 0<j0\leq\ell<j then ϕ(ui)=ui\phi(u_{i}^{\ell})=u_{i}^{\ell} for all 1in1\leq i\leq n. Each non-twin vertex on level jj in μt(G)\mu_{t}(G) has a unique neighborhood on level j1j-1 and that neighborhood is fixed by the minimality of jj. Hence uiju_{i}^{j} must be a twin vertex and must map to a twin on the same level.

Suppose ϕ(uij)=ukj\phi(u_{i}^{j})=u_{k}^{j}. So N(uij)=N(ukj)N(u_{i}^{j})=N(u_{k}^{j}) and, by the construction of a Mycielskian, N(uk0)=N(uj0)N(u_{k}^{0})=N(u_{j}^{0}). By Lemma 2 we know that there exists some u0N(ui0)=N(uk0)u_{\ell}^{0}\in N(u_{i}^{0})=N(u_{k}^{0}) such that c(ui0u0)c(uk0u0)c(u_{i}^{0}u_{\ell}^{0})\neq c(u_{k}^{0}u_{\ell}^{0}). Moreover, because of the choice of coloring we know c¯(uj1uij)c¯(uj1ukj)\overline{c}(u_{\ell}^{j-1}u_{i}^{j})\neq\overline{c}(u_{\ell}^{j-1}u_{k}^{j}). Since vertices on level j1j-1 are fixed we cannot have ϕ(uij)=ukj\phi(u_{i}^{j})=u_{k}^{j}. Thus ϕ\phi must be the trivial automorphism and there exists a distinguishing coloring of μt(G)\mu_{t}(G) with Dist(G)\operatorname{Dist^{\prime}}(G) colors. So, Dist(μt(G))Dist(G)\operatorname{Dist^{\prime}}(\mu_{t}(G))\leq\operatorname{Dist^{\prime}}(G). ∎

Theorems 8 and 9 show that Conjecture 1 remains true for generalized Mycielskians. We state this result in the following theorem.

Theorem 10.

For all graphs GG with |V(G)|3|V(G)|\geq 3, no isolated vertices, and no connected K2K_{2} component and for all tt\in\mathbb{N}

Dist(μt(G))Dist(G).\operatorname{Dist^{\prime}}(\mu_{t}(G))\leq\operatorname{Dist^{\prime}}(G).

As mentioned in the introduction, the Mycielskian of a graph was first introduced to show that there exist triangle-free graphs of arbitrarily large chromatic numbers. To do so, Mycielski showed that the construction increases the chromatic number while preserving the property of being triangle-free. Iterating the construction on K2K_{2} gives the result. We use μp(G)\mu^{p}(G) to denote the Mycielskian of GG iterated pp times and μtp(G)\mu_{t}^{p}(G) to represent the generalized Mycielskian of GG with tt levels, iterated pp times. More precisely, μt1(G)=μt(G)\mu_{t}^{1}(G)=\mu_{t}(G) and μtp(G)=μt(μtp1(G))\mu_{t}^{p}(G)=\mu_{t}(\mu_{t}^{p-1}(G)). Corollary 11 follows from Theorem 10.

Corollary 11.

For all graphs GG with |G|3|G|\geq 3, t1t\geq 1, and p1p\geq 1,

Dist(μp(G))Dist(G)\displaystyle\operatorname{Dist^{\prime}}(\mu^{p}(G))\leq\operatorname{Dist^{\prime}}(G)

and

Dist(μtp(G))Dist(G).\displaystyle\operatorname{Dist^{\prime}}(\mu^{p}_{t}(G))\leq\operatorname{Dist^{\prime}}(G).

References

  • [1] Michael O. Albertson and Karen L. Collins, Symmetry breaking in graphs, Electron. J. Combin. 3 (1996), no. 1, Research Paper 18, approx. 17. MR 1394549
  • [2] Saeid Alikhani and Samaneh Soltani, The distinguishing number and the distinguishing index of line and graphoidal graph(s), AKCE Int. J. Graphs Comb. 17 (2020), no. 1, 1–6. MR 4145384
  • [3] L. Babai, Asymmetric trees with two prescribed degrees, Acta Math. Acad. Sci. Hungar. 29 (1977), no. 1-2, 193–200. MR 453572
  • [4] R. Balakrishnan and S. Francis Raj, Connectivity of the Mycielskian of a graph, Discrete Math. 308 (2008), no. 12, 2607–2610. MR 2410467
  • [5] Debra Boutin, Sally Cockburn, Lauren Keough, Sarah Loeb, K. E. Perry, and Puck Rombach, Distinguishing generalized Mycielskian graphs, Australas. J. Combin. 83 (2022), 225–242. MR 4435045
  • [6]   , Determining number and cost of generalized Mycielskian graphs, Discuss. Math. Graph Theory 44 (2024), no. 1, 127–149. MR 4686194
  • [7] Xue-gang Chen and Hua-ming Xing, Domination parameters in Mycielski graphs, Util. Math. 71 (2006), 235–244. MR 2278836
  • [8] David C. Fisher, Patricia A. McKenna, and Elizabeth D. Boyer, Hamiltonicity, diameter, domination, packing, and biclique partitions of Mycielski’s graphs, Discrete Appl. Math. 84 (1998), no. 1-3, 93–105. MR 1626550
  • [9] R Kalinowski and M Pils´\acute{\text{s}}niak, Distinguishing graphs by edge-colourings, European J. Combin. 45 (2015), 124–131. MR 3286626
  • [10] Wensong Lin, Jianzhuan Wu, Peter Che Bor Lam, and Guohua Gu, Several parameters of generalized Mycielskians, Discrete Appl. Math. 154 (2006), no. 8, 1173–1182. MR 2219419
  • [11] Jan Mycielski, Sur le coloriage des graphs, Colloq. Math. 3 (1955), 161–162.
  • [12] M. Stiebitz, Beiträge zur theorie der färbungskritschen graphen, Ph.D. thesis, Technische Universität Ilmenau, 1985.
  • [13] Vinny Susan Prebhath and V. Sangeetha, Tadpole domination number of graphs, Discrete Math. Algorithms Appl. 15 (2023), no. 8, Paper No. 2250173, 12. MR 4627856