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

Minimum numbers of Dehn colors of knots and \mathcal{R}-palette graphs

Eri Matsudo The Institute of Natural Sciences, Nihon University, 3-25-40 Sakurajosui, Setagaya-ku, Tokyo 156-8550, Japan matsudo.eri@nihon-u.ac.jp Kanako Oshiro Department of Information and Communication Sciences, Sophia University, Tokyo 102-8554, Japan oshirok@sophia.ac.jp  and  Gaishi Yamagishi g-yamagishi-3c9@eagle.sophia.ac.jp
Abstract.

In this paper, we consider minimum numbers of colors of knots for Dehn colorings. In particular, we will show that for any odd prime number pp and any Dehn pp-colorable knot KK, the minimum number of colors for KK is at least log2p+2\lfloor\log_{2}p\rfloor+2. Moreover, we will define the \mathcal{R}-palette graph for a set of colors. The \mathcal{R}-palette graphs are quite useful to give candidates of sets of colors which might realize a nontrivially Dehn pp-colored diagram. In Appendix, we also prove that for Dehn 55-colorable knot, the minimum number of colors is 44.

Key words and phrases:
Knots, Dehn colorings, Minimum numbers of colors, R{R}-palette graphs
2020 Mathematics Subject Classification:
57K10, 57K12

Introduction

In knot theory, minimum numbers of colors for arc colorings have been studied in many papers (see [1, 2, 4, 5, 9, 10, 11] for example). We denote by mincolpFox(K)\text{mincol}^{\text{Fox}}_{p}(K) the number for a Fox pp-colorable knot KK. As one of properties for mincolpFox(K)\text{mincol}^{\text{Fox}}_{p}(K), the following result is obtained in [8].

  • For any odd prime number pp and any Fox pp-colorable knot KK,

    mincolpFox(K)log2p+2.\text{mincol}^{\text{Fox}}_{p}(K)\geq\lfloor\log_{2}p\rfloor+2.

Note that for the cases when pp is a composite number or KK is a link with multiple components, the similar results are known under some condition (see [6]).

On the other hand, a Dehn pp-coloring is well-known as one of region colorings of knot diagrams which is corresponding to a Fox pp-coloring. However, minimum numbers of colors for Dehn pp-colorings almost have not been studied yet.

In this paper, we will consider minimum numbers of colors of Dehn pp-colorable knots for an odd prime number pp. The number for a Dehn pp-colorable knot KK is denoted by mincolpDehn(K){\rm mincol}^{\rm Dehn}_{p}(K). We show in Theorem 1.5 that

  • for any odd prime number pp and any Dehn pp-colorable knot KK,

    mincolpDehn(K)log2p+2.{\rm mincol}^{\rm Dehn}_{p}(K)\geq\lfloor\log_{2}p\rfloor+2.

Moreover, we will define the \mathcal{R}-palette graph for a set of colors. The \mathcal{R}-palette graphs are quite useful to give candidates of sets of colors which might realize a nontrivially Dehn pp-colored diagram (see Theorem 5.1). In particular, in Theorem 5.3, we give the sets of log2p+2\lfloor\log_{2}p\rfloor+2 colors which might realize a nontrivially Dehn pp-colored diagram for an odd prime number pp with p<25p<2^{5}. Furthermore, we show in Proposition 5.4 that there exists a knot KK with mincolpDehn(K)=log2p+2{\rm mincol}^{\rm Dehn}_{p}(K)=\lfloor\log_{2}p\rfloor+2 for some pp.

This paper is organized as follows: In Section 1, we review the definitions and some basic properties of Dehn colorings and minimum numbers of colors for Dehn colorings. Besides, our main result (Theorem 1.5) is stated in this section. Section 2 is devoted to setting up some extended coloring matrices which are used in Section 3. Theorem 1.5 is proven in Section 3. In Section 4, the \mathcal{R}-palette graph of a set of colors and that of a Dehn pp-colored diagram are defined. In Section 5, we give candidates of sets of colors each of which might realize a nontrivially Dehn pp-colored diagram by using the \mathcal{R}-palette graphs. Moreover in Appendix A, we show that mincol5Dehn(K)=4{\rm mincol}^{\rm Dehn}_{5}(K)=4 for any Dehn 55-colorable knot KK.

1. Dehn colorings and minimum numbers of colors

In this paper, for a prime number pp, we denote by p\mathbb{Z}_{p} the cyclic group /p\mathbb{Z}/p\mathbb{Z}. When p=0p=0, read all the parts of this paper by replacing all p\mathbb{Z}_{p} and pp with \mathbb{Z}.

Let pp be an odd prime number or p=0p=0. Let DD be a diagram of a knot KK and (D)\mathcal{R}(D) the set of regions of DD. A Dehn pp-coloring of DD is a map C:(D)pC:\mathcal{R}(D)\to\mathbb{Z}_{p} satisfying the following condition:

  • for each crossing cc with regions x1,x2,x3x_{1},x_{2},x_{3}, and x4x_{4} as depicted in Figure 1,

    C(x1)+C(x3)=C(x2)+C(x4)C(x_{1})+C(x_{3})=C(x_{2})+C(x_{4})

    holds, where the region x2x_{2} is adjacent to x1x_{1} by an under-arc and x3x_{3} is adjacent to x1x_{1} by the over-arc.

We call C(x)C(x) the color of a region xx by CC. In this paper, as shown in the right of Figure 1, we represent a Dehn pp-coloring CC of a knot diagram DD by assigning the color C(x)C(x) to each region xx. We mean by (D,C)(D,C) a diagram DD given a Dehn pp-coloring CC, and call it a Dehn pp-colored diagram. We denote by 𝒞(D,C)\mathcal{C}(D,C) the set of colors assigned to a region of DD by CC, that is 𝒞(D,C)=ImC\mathcal{C}(D,C)={\rm Im}\,C. The set of Dehn pp-colorings of DD is denoted by Colp(D){\rm Col}_{p}(D). We remark that the number #Colp(D)\#{\rm Col}_{p}(D) is an invariant of the knot KK.

Refer to caption
Figure 1. A crossing on DD and the one on (D,C)(D,C) with C(x1)=a1C(x_{1})=a_{1}, C(x2)=a2C(x_{2})=a_{2}, C(x3)=a3C(x_{3})=a_{3}, C(x4)=a1a2+a3C(x_{4})=a_{1}-a_{2}+a_{3}

Let cc be a crossing of DD with regions x1,x2,x3x_{1},x_{2},x_{3}, and x4x_{4} as depicted in Figure 1. We say that cc of (D,C)(D,C) is trivially colored if

C(x1)=C(x4), and C(x3)=C(x2)C(x_{1})=C(x_{4}),\mbox{ and }C(x_{3})=C(x_{2})

hold, and nontrivially colored otherwise. A Dehn pp-coloring CC of DD is trivial if each crossing of (D,C)(D,C) is trivially colored and nontrivial otherwise. We note that trivial pp-colorings are separated into two types: monochromatic colorings and checkerboard colorings with two colors. We call the former 11-trivial colorings and denote them by C1TC^{\rm 1T}. We call the latter 22-trivial colorings and denote them by C2TC^{\rm 2T}.

A knot KK is Dehn pp-colorable if KK has a Dehn pp-colored diagram (D,C)(D,C) such that CC is nontrivial.

For a semiarc uu of a Dehn pp-colored diagram (D,C)(D,C), when the two regions x1x_{1} and x2x_{2} which are on the both side of uu have C(x1)=a1C(x_{1})=a_{1} and C(x2)=a2C(x_{2})=a_{2} for some a1,a2pa_{1},a_{2}\in\mathbb{Z}_{p}, we call uu an {a1,a2}\{a_{1},a_{2}\}-semiarc, where {a1,a2}\{a_{1},a_{2}\} is regarded as a multiset {a1,a1}\{a_{1},a_{1}\} when a1=a2a_{1}=a_{2}. Moreover, we call an arc uu including an {a1,a2}\{a_{1},a_{2}\}-semiarc an a1+a2¯\overline{a_{1}+a_{2}}-arc, where we note that the value a1+a2¯\overline{a_{1}+a_{2}} does not depend on the choice of a semiarc included in uu.

Remark 1.1.

Fox colorings of knot diagrams have been well-studied for a long time. There is the pp-to-11 correspondence, given in the upper picture of Figure 2, from Dehn colorings to Fox colorings, where a1,a2pa_{1},a_{2}\in\mathbb{Z}_{p} represent the colors of regions of a Dehn pp-colored diagram, a1+a2¯p\overline{a_{1}+a_{2}}\in\mathbb{Z}_{p} represents the colors of arcs of the corresponding Fox pp-colored diagram, and we distinguish colors of regions for Dehn pp-colorings and those of arcs for Fox pp-colorings by putting a bar over a color of an arc as in Figure 2. Then a trivial (resp. nontrivial) Dehn pp-coloring corresponds to a trivial (resp. nontrivial) Fox pp-coloring as depicted in the lower picture of Figure 2. We also note that a knot KK is Dehn pp-colorable if and only if KK is Fox pp-colorable.

Refer to caption
Figure 2. A correspondence between Dehn colorings and Fox colorings

The set Colp(D){\rm Col}_{p}(D) can be regarded as a p\mathbb{Z}_{p}-module with the scalar product sCsC and the addition C+CC+C^{\prime} with

(sC):(D)p;(sC)(x)=sC(x),\displaystyle(sC):\mathcal{R}(D)\to\mathbb{Z}_{p};(sC)(x)=sC(x),
(C+C):(D)p;(C+C)(x)=C(x)+C(x).\displaystyle(C+C^{\prime}):\mathcal{R}(D)\to\mathbb{Z}_{p};(C+C^{\prime})(x)=C(x)+C^{\prime}(x).

Moreover, since for any tpt\in\mathbb{Z}_{p}, the constant map Ct1T:(D)p;Ct1T(x)=tC^{\rm 1T}_{t}:\mathcal{R}(D)\to\mathbb{Z}_{p};C^{\rm 1T}_{t}(x)=t is a kind of Dehn pp-coloring, the map

(sC+t):(D)p;(sC+t)(x)=(sC+Ct1T)(x)=sC(x)+t,(sC+t):\mathcal{R}(D)\to\mathbb{Z}_{p};(sC+t)(x)=(sC+C^{\rm 1T}_{t})(x)=sC(x)+t,

is also a Dehn pp-coloring, and thus, the transformations CsC+tC\mapsto sC+t are closed in Colp(D){\rm Col}_{p}(D). In particular, a transformation CsC+tC\mapsto sC+t with sp×s\in\mathbb{Z}_{p}^{\times} is called a regular affine transformation on Colp(D){\rm Col}_{p}(D). Two Dehn pp-colorings C,CColp(D)C,C^{\prime}\in{\rm Col}_{p}(D) are affine equivalent (written CCC\sim C^{\prime}) if they are related by a regular affine transformation on Colp(D){\rm Col}_{p}(D), that is, C=sC+tC^{\prime}=sC+t for some sp×s\in\mathbb{Z}_{p}^{\times} and tpt\in\mathbb{Z}_{p}.

A regular affine transformation on the p\mathbb{Z}_{p}-module p\mathbb{Z}_{p} is defined by asa+ta\mapsto sa+t for some sp×s\in\mathbb{Z}_{p}^{\times} and tpt\in\mathbb{Z}_{p}. Two subsets S,SpS,S^{\prime}\subset\mathbb{Z}_{p} are affine equivalent (written SSS\sim S^{\prime}) if they are related by a regular affine transformation on p\mathbb{Z}_{p}, that is, S=sS+tS^{\prime}=sS+t for some sp×s\in\mathbb{Z}_{p}^{\times} and tpt\in\mathbb{Z}_{p}. We note that since any regular affine transformation is a bijection, #S=#S\#S=\#S^{\prime} holds if SSS\sim S^{\prime}.

Lemma 1.2.
  • (1)

    Let C,CColp(D)C,C^{\prime}\in{\rm Col}_{p}(D). Then we have

    CC𝒞(D,C)𝒞(D,C).C\sim C^{\prime}\Longrightarrow\mathcal{C}(D,C)\sim\mathcal{C}(D,C^{\prime}).

    Hence we have

    CC#𝒞(D,C)=#𝒞(D,C).C\sim C^{\prime}\Longrightarrow\#\mathcal{C}(D,C)=\#\mathcal{C}(D,C^{\prime}).
  • (2)

    Let S,SpS,S^{\prime}\subset\mathbb{Z}_{p}, and we assume that SSS\sim S^{\prime}. Then there exists CColp(D)C\in{\rm Col}_{p}(D) such that 𝒞(D,C)=S\mathcal{C}(D,C)=S if and only if there exists CColp(D)C^{\prime}\in{\rm Col}_{p}(D) such that 𝒞(D,C)=S\mathcal{C}(D,C^{\prime})=S^{\prime}.

Proof.

(1) A regular affine transformation CC=sC+tC\mapsto C^{\prime}=sC+t on Colp(D){\rm Col}_{p}(D) induces the regular affine transformation xsx+tx\mapsto sx+t on p\mathbb{Z}_{p} with 𝒞(D,C)=s𝒞(D,C)+t\mathcal{C}(D,C^{\prime})=s\mathcal{C}(D,C)+t.

(2) Assume that S=sS+tS^{\prime}=sS+t holds for some regular affine transformation xsx+tx\mapsto sx+t on p\mathbb{Z}_{p}. When there exists CColp(D)C\in{\rm Col}_{p}(D) such that 𝒞(D,C)=S\mathcal{C}(D,C)=S, we have the Dehn pp-coloring C=sC+tC^{\prime}=sC+t. Then 𝒞(D,C)=s𝒞(D,C)+t=sS+t=S\mathcal{C}(D,C^{\prime})=s\mathcal{C}(D,C)+t=sS+t=S^{\prime} holds.

Remark 1.3.

(1) of Lemma 1.2 implies that when we focus on the number of colors used for each nontrivial Dehn pp-coloring of a diagram DD, we may consider only the number of colors used for a representative of each affine equivalence class for nontrivial Dehn pp-colorings of DD.

(2) of Lemma 1.2 implies that when we focus on the sets of colors each of which might be used for a nontrivial Dehn pp-coloring of a diagram, we may consider only representatives of the affine equivalence classes on the subsets of p\mathbb{Z}_{p}.

In this paper, we focus on the minimum number of colors.

Definition 1.4.

The minimum number of colors of a knot KK for Dehn pp-colorings is the minimum number of distinct elements of p\mathbb{Z}_{p} which produce a nontrivially Dehn pp-colored diagram of KK, that is,

min{#𝒞(D,C)|(D,C){ nontrivially Dehn p-colored diagrams of K }}.\min\Big{\{}\#\mathcal{C}(D,C)~{}\Big{|}~{}(D,C)\in\left\{\begin{minipage}{142.26378pt} nontrivially Dehn $p$-colored\\ diagrams of $K$ \end{minipage}\right\}\Big{\}}.

We denote it by mincolpDehn(K){\rm mincol}^{\rm Dehn}_{p}(K).

The next theorem is our main result, which will be proven in Section 3.

Theorem 1.5.

Let pp be an odd prime number. For any Dehn pp-colorable knot KK, we have

mincolpDehn(K)log2p+1,{\rm mincol}^{\rm Dehn}_{p}(K)\geq\log_{2}p+1,

that is,

mincolpDehn(K)log2p+2.{\rm mincol}^{\rm Dehn}_{p}(K)\geq\lfloor\log_{2}p\rfloor+2.

2. Extended coloring matrices

From now on, for an integer matrix MM, rankpM{\rm rank}_{p}M means the rank of MM regarded as a matrix on p\mathbb{Z}_{p}. Note again that rank0M{\rm rank}_{0}M is replaced with rankM{\rm rank}_{\mathbb{Z}}M.

Let DD be a diagram of a knot KK. Let x1,,xn+2x_{1},\ldots,x_{n+2} be the regions, and c1,,cnc_{1},\ldots,c_{n} the crossings of DD. We set an arbitrary orientation for DD.

For each crossing cic_{i}, we set the equation

xi1xi2+xi3xi4=0,x_{i_{1}}-x_{i_{2}}+x_{i_{3}}-x_{i_{4}}=0,

where xi1,xi2,xi3,xi4x_{i_{1}},x_{i_{2}},x_{i_{3}},x_{i_{4}} are the regions around cic_{i} such that xi1x_{i_{1}} is in the right side of both of over- and under-arcs, xi2x_{i_{2}} is adjacent to xi1x_{i_{1}} by an under-arc and xi3x_{i_{3}} is adjacent to xi1x_{i_{1}} by the over-arc (see the left of Figure 3), and where we regard x1,,xn+2x_{1},\ldots,x_{n+2} as variables of the equation. We then have the simultaneous equations

{x11x12+x13x14=0,xi1xi2+xi3xi4=0,xn1xn2+xn3xn4=0,\left\{\begin{array}[]{l}x_{1_{1}}-x_{1_{2}}+x_{1_{3}}-x_{1_{4}}=0,\\ \hskip 28.45274pt\vdots\\ x_{i_{1}}-x_{i_{2}}+x_{i_{3}}-x_{i_{4}}=0,\\ \hskip 28.45274pt\vdots\\ x_{n_{1}}-x_{n_{2}}+x_{n_{3}}-x_{n_{4}}=0,\end{array}\right.

that is

MDcol(x1xn+2)=𝟎.\displaystyle M^{\rm col}_{D}\begin{pmatrix}x_{1}\\ \vdots\\ x_{n+2}\end{pmatrix}=\boldsymbol{0}. (2.1)

The coefficient matrix MDcolM^{\rm col}_{D} is called the Dehn coloring matrix of DD as an unoriented diagram since the p\mathbb{Z}_{p}-module of solutions of (2.1) for an odd prime pp or p=0p=0 is isomorphic to the p\mathbb{Z}_{p}-module Colp(D){\rm Col}_{p}(D) by the isomorphism which sends a solution (a1an+2)t{}^{t}(a_{1}\cdots a_{n+2}) to the Dehn coloring CC such that C(xi)=aiC(x_{i})=a_{i} for each i{1,,n+2}i\in\{1,\ldots,n+2\}.

Refer to caption
Figure 3.

It is known that the knot group GKG_{K} of KK, which is the fundamental group of 3K\mathbb{R}^{3}\setminus K, has the presentation

GK=x1,,xn+2|r1,,rn,x1,\displaystyle G_{K}=\langle x_{1},\ldots,x_{n+2}~{}|~{}r_{1},\ldots,r_{n},x_{1}\rangle, (2.2)

where for each crossing cic_{i} (1in)(1\leq i\leq n) as in the left of Figure 3, we set the relator rir_{i} by

ri=xi3xi11xi2xi41.r_{i}=x_{i_{3}}x_{i_{1}}^{-1}x_{i_{2}}x_{i_{4}}^{-1}.

Let AD(t)A_{D}(t) denote the Alexander matrix of GKG_{K} obtained from the above presentation (2.2) by using the Fox calculus (see [3]), that is AD(t)=(α~π~(rixj))\displaystyle A_{D}(t)=\left(\tilde{\alpha}\circ\tilde{\pi}\left(\frac{\partial r_{i}}{\partial x_{j}}\right)\right), where xj\frac{\partial}{\partial x_{j}} is the Fox derivative FF\mathbb{Z}F\rightarrow\mathbb{Z}F with F=x1,,xn+2F=\langle x_{1},\ldots,x_{n+2}\rangle, π~\tilde{\pi} is the projection FGK\mathbb{Z}F\rightarrow\mathbb{Z}G_{K}, and α~\tilde{\alpha} is the abelianization GKt=[t,t1]\mathbb{Z}G_{K}\rightarrow\mathbb{Z}\langle t\rangle=\mathbb{Z}[t,t^{-1}] such that a meridian element is sent to tt. For the iith row vector of AD(1)A_{D}(-1) (1in)(1\leq i\leq n), the jjth entry is 11 if j=i1j=i_{1} or i3i_{3}, 1-1 if j=i2j=i_{2} or i4i_{4}, and 0 otherwise, which follows from

rixi1=xi3xi11α~π~t(k+1)tk=t2k1t11,\displaystyle\frac{\partial r_{i}}{\partial x_{i_{1}}}=-x_{i_{3}}x_{i_{1}}^{-1}\overset{\tilde{\alpha}\circ\tilde{\pi}}{\mapsto}-t^{-(k+1)}t^{-k}=-t^{-2k-1}\overset{t\mapsto-1}{\mapsto}1,
rixi2=xi3xi11α~π~t(k+1)tk=t2k1t11,\displaystyle\frac{\partial r_{i}}{\partial x_{i_{2}}}=x_{i_{3}}x_{i_{1}}^{-1}\overset{\tilde{\alpha}\circ\tilde{\pi}}{\mapsto}t^{-(k+1)}t^{-k}=t^{-2k-1}\overset{t\mapsto-1}{\mapsto}-1,
rixi3=1α~π~1t11,\displaystyle\frac{\partial r_{i}}{\partial x_{i_{3}}}=1\overset{\tilde{\alpha}\circ\tilde{\pi}}{\mapsto}1\overset{t\mapsto-1}{\mapsto}1,
rixi4=xi3xi11xi2xi41α~π~1t11\displaystyle\frac{\partial r_{i}}{\partial x_{i_{4}}}=-x_{i_{3}}x_{i_{1}}^{-1}x_{i_{2}}x_{i_{4}}^{-1}\overset{\tilde{\alpha}\circ\tilde{\pi}}{\mapsto}-1\overset{t\mapsto-1}{\mapsto}-1

(refer to the right of Figure 3), where we here suppose that xi1,,xi4x_{i_{1}},\ldots,x_{i_{4}} differ from each other. Thus we have

AD(1)=(MDcol1𝟎).A_{D}(-1)=\begin{pmatrix}&\mbox{\Large{$M_{D}^{\rm col}$}}\\ 1&\boldsymbol{0}\end{pmatrix}.

The greatest common divisor of all the (n+1n+1)-minors of AD(1)A_{D}(-1) is called the determinant of KK, which is denoted by det(K){\rm det}(K).

For an odd prime number pp, we suppose that there exists a nontrivial Dehn pp-coloring, say C0C_{0}, satisfying that for some i,j{1,,n+2}i,j\in\{1,\ldots,n+2\} such that i<ji<j, C0(xi)=C0(xj)C_{0}(x_{i})=C_{0}(x_{j}), C2T(xi)C2T(xj)C^{\rm 2T}(x_{i})\not=C^{\rm 2T}(x_{j}) holds. Then we set

BD,C0;i,j=(MDcol𝟎1𝟎1𝟎),B_{D,C_{0};i,j}=\begin{pmatrix}&&\mbox{\Large{$M_{D}^{\rm col}$}}&&\\ \boldsymbol{0}&1&\boldsymbol{0}&-1&\boldsymbol{0}\end{pmatrix},

where for the last row vector, the iith and jjth entries are 11 and 1-1, respectively, and the other entries are 0.

Let us consider the ranks of the matrices MDcolM_{D}^{\rm col}, AD(1)A_{D}(-1) and BD,C0;i,jB_{D,C_{0};i,j}.

For an odd prime pp or p=0p=0, we have at least two linearly independent solutions of (2.1) on p\mathbb{Z}_{p}: one, denoted by 𝒂1T\boldsymbol{a}^{\rm 1T}, is corresponding to the 1-trivial Dehn p\mathbb{Z}_{p}-coloring C1TC^{\rm 1T} with C1T(x1)=1C^{\rm 1T}(x_{1})=1, and the other, denoted by 𝒂2T\boldsymbol{a}^{\rm 2T}, is corresponding to a 2-trivial Dehn p\mathbb{Z}_{p}-coloring C2TC^{\rm 2T} with C2T(x1)=0C^{\rm 2T}(x_{1})=0 and ImC2T={0,1}{\rm Im}C^{\rm 2T}=\{0,1\}. Thus we can see that rankpMDcoln{\rm rank}_{p}M^{\rm col}_{D}\leq n.

Besides, it is well-known that det(K)0{\rm det}(K)\not=0, which gives rankAD(1)=n+1{\rm rank}_{\mathbb{Z}}A_{D}(-1)=n+1, and hence, rankMDcoln{\rm rank}_{\mathbb{Z}}M_{D}^{\rm col}\geq n. Considering this property together with the fact that rankMDcoln{\rm rank}_{\mathbb{Z}}M_{D}^{\rm col}\leq n, we have rankMDcol=n{\rm rank}_{\mathbb{Z}}M_{D}^{\rm col}=n, and indeed, the \mathbb{Z}-module of solutions of (2.1) on \mathbb{Z} is spanned by two vectors 𝒂1T\boldsymbol{a}^{\rm 1T} and 𝒂2T\boldsymbol{a}^{\rm 2T}.

On the other hand, for an odd prime pp, if DD admits a nontrivial Dehn pp-coloring CC, then there exists a solution 𝒂\boldsymbol{a} of (2.1) on p\mathbb{Z}_{p} such that 𝒂1T\boldsymbol{a}^{\rm 1T}, 𝒂2T\boldsymbol{a}^{\rm 2T} and 𝒂\boldsymbol{a} are linearly independent. Hence, for this case, rankpMDcoln1{\rm rank}_{p}M^{\rm col}_{D}\leq n-1.

For the matrix AD(1)A_{D}(-1), rankAD(1)=n+1{\rm rank}_{\mathbb{Z}}A_{D}(-1)=n+1 as mentioned above. For an odd prime pp or p=0p=0, note that the p\mathbb{Z}_{p}-module of solutions of the simultaneous equations

AD(1)(x1xn+2)=𝟎\displaystyle A_{D}(-1)\begin{pmatrix}x_{1}\\ \vdots\\ x_{n+2}\end{pmatrix}=\boldsymbol{0} (2.3)

on p\mathbb{Z}_{p} is a submodule of that of (2.1) on p\mathbb{Z}_{p}. Any vector k𝒂1T+𝒂2Tk\boldsymbol{a}^{\rm 1T}+\ell\boldsymbol{a}^{\rm 2T} (k,k,\ell\in\mathbb{Z}, k0k\not=0) can not be a solution of (2.3) on p\mathbb{Z}_{p} since any solution 𝒂=t(a1an+2)\boldsymbol{a}=\,^{t}(a_{1}\cdots a_{n+2}) of (2.3) needs to satisfy a1=0a_{1}=0, while 𝒂2T\boldsymbol{a}^{\rm 2T} is a linearly independent solution of (2.3) on p\mathbb{Z}_{p}. This implies that the \mathbb{Z}-module of solutions of (2.3) on \mathbb{Z} is spanned by 𝒂2T\boldsymbol{a}^{\rm 2T}.

On the other hand, for an odd prime pp, if DD admits a nontrivial Dehn pp-coloring CC such that C(x1)=0C(x_{1})=0, then there exists a solution 𝒂\boldsymbol{a} of (2.3) on p\mathbb{Z}_{p} such that 𝒂2T\boldsymbol{a}^{\rm 2T} and 𝒂\boldsymbol{a} are linearly independent. Hence, for this case, rankpAD(1)n{\rm rank}_{p}A_{D}(-1)\leq n.

For the matrix BD,C0;i,jB_{D,C_{0};i,j}, we note that it is defined for a given odd prime pp, a Dehn pp-coloring C0C_{0}, and i,j{1,,n+2}i,j\in\{1,\ldots,n+2\}. Note that the \mathbb{Z}- (resp. p\mathbb{Z}_{p}-)module of solutions of the simultaneous equations

BD,C0;i,j(x1xn+2)=𝟎\displaystyle B_{D,C_{0};i,j}\begin{pmatrix}x_{1}\\ \vdots\\ x_{n+2}\end{pmatrix}=\boldsymbol{0} (2.4)

on \mathbb{Z} (resp. p\mathbb{Z}_{p}) is a submodule of that of (2.1) on \mathbb{Z} (resp. p\mathbb{Z}_{p}).

Any vector k𝒂1T+𝒂2Tk\boldsymbol{a}^{\rm 1T}+\ell\boldsymbol{a}^{\rm 2T} (k,k,\ell\in\mathbb{Z}, 0\ell\not=0) can not be a solution of (2.4) on \mathbb{Z} (resp. p\mathbb{Z}_{p}) since any solution 𝒂=t(a1an+2)\boldsymbol{a}=\,^{t}(a_{1}\cdots a_{n+2}) of (2.4) needs to satisfy ai=aja_{i}=a_{j}, while 𝒂1T\boldsymbol{a}^{\rm 1T} is a linearly independent solution of (2.4) on \mathbb{Z} (resp. p\mathbb{Z}_{p}). Hence the \mathbb{Z}-module of solutions of (2.4) on \mathbb{Z} is spanned by 𝒂1T\boldsymbol{a}^{\rm 1T}, which implies that rankBD,C0;i,j=n+1{\rm rank}_{\mathbb{Z}}B_{D,C_{0};i,j}=n+1.

On the other hand, since C0C_{0} is a nontrivial pp-coloring, there exists a solution 𝒂0=t(C0(x1)C0(xn+2))\boldsymbol{a}_{0}=\,^{t}(C_{0}(x_{1})\cdots C_{0}(x_{n+2})) of (2.4) on p\mathbb{Z}_{p}, and 𝒂1T\boldsymbol{a}^{\rm 1T} and 𝒂0\boldsymbol{a}_{0} are linearly independent. Hence, rankpBD,C0;i,jn{\rm rank}_{p}B_{D,C_{0};i,j}\leq n.

As a summary, we have the following properties:

Lemma 2.1.

Let pp be an odd prime number.

  • (1)

    Suppose that DD admits a nontrivial Dehn pp-coloring CC. Then we have

    rankMDcol=n, and rankpMDcoln1.{\rm rank}_{\mathbb{Z}}M_{D}^{\rm col}=n,\mbox{ and }{\rm rank}_{p}M_{D}^{\rm col}\leq n-1.

    In particular, the module of solutions of (2.1) on \mathbb{Z} is spanned by 𝒂1T\boldsymbol{a}^{\rm 1T} and 𝒂2T\boldsymbol{a}^{\rm 2T}, and that on p\mathbb{Z}_{p} is spanned by at least three linearly independent solutions: 𝒂1T\boldsymbol{a}^{\rm 1T}, 𝒂2T\boldsymbol{a}^{\rm 2T} and 𝒂=t(C(x1)C(xn+2))\boldsymbol{a}=\,^{t}(C(x_{1})\cdots C(x_{n+2})).

  • (2)

    Suppose that DD admits a nontrivial Dehn pp-coloring CC with C(x1)=0C(x_{1})=0. Then we have

    rankAD(1)=n+1, and rankpAD(1)n.{\rm rank}_{\mathbb{Z}}A_{D}(-1)=n+1,\mbox{ and }{\rm rank}_{p}A_{D}(-1)\leq n.

    In particular, the module of solutions of (2.3) on \mathbb{Z} is spanned by 𝒂2T\boldsymbol{a}^{\rm 2T}, and that on p\mathbb{Z}_{p} is spanned by at least two linearly independent solutions: 𝒂2T\boldsymbol{a}^{\rm 2T} and 𝒂=t(C(x1)C(xn+2))\boldsymbol{a}=\,^{t}(C(x_{1})\cdots C(x_{n+2})).

  • (3)

    Suppose that DD admits a nontrivial Dehn pp-coloring CC satisfying that C(xi)=C(xj)C(x_{i})=C(x_{j}), C2T(xi)C2T(xj)C^{\rm 2T}(x_{i})\not=C^{\rm 2T}(x_{j}) for some i,ji,j with 1i<jn+21\leq i<j\leq n+2. Then we have

    rankBD,C;i,j=n+1, and rankpBD,C;i,jn.{\rm rank}_{\mathbb{Z}}B_{D,C;i,j}=n+1,\mbox{ and }{\rm rank}_{p}B_{D,C;i,j}\leq n.

    In particular, the module of solutions of (2.4) on \mathbb{Z} is spanned by 𝒂1T\boldsymbol{a}^{\rm 1T}, and that on p\mathbb{Z}_{p} is spanned by at least two linearly independent solutions: 𝒂1T\boldsymbol{a}^{\rm 1T} and 𝒂=t(C(x1)C(xn+2))\boldsymbol{a}=\,^{t}(C(x_{1})\cdots C(x_{n+2})).

3. Proof of Theorem 1.5

Lemma 3.1.

Let MM be an integer square matrix of order kk which satisfies the following condition:

  • (\star)

    for each i{1,,k}i\in\{1,\ldots,k\}, the multiset of nonzero entries of the iith row vector of MM is {2}\{-2\}, {1}\{-1\}, {1}\{1\}, {2}\{2\}, {2,1}\{-2,1\}, {2,2}\{-2,2\}, {1,1}\{-1,-1\}, {1,1}\{-1,1\}, {1,2}\{-1,2\}, {1,1}\{1,1\}, {2,1,1}\{-2,1,1\}, {1,1,1}\{-1,-1,1\}, {1,1,2}\{-1,-1,2\}, {1,1,1}\{-1,1,1\} or {1,1,1,1}\{-1,-1,1,1\}.

Then we have

|detM|2k.|{\rm det}M|\leq 2^{k}.
Proof.

We prove this lemma by the induction of kk and the cofactar expantion of a matrix. In the case that k=1k=1, since |detM|=0,1|\det M|=0,1 or 22, it holds that |detM|2|\det M|\leq 2.

Let k>1k>1. We assume that any square matrix MM^{\prime}, of order k1k-1, satisfying the condition ()(\star) has |detM|2k1|\det{M^{\prime}}|\leq 2^{k-1}.

We first note that the following transformations of matrices do not affect whether the condition ()(\star) is satisfied.

  • Compose a matrix (𝒘1,,𝒘j1,Σ=1k𝒘,𝒘j+1,,𝒘k)\displaystyle\left(\boldsymbol{w}_{1},\dots,\boldsymbol{w}_{j-1},-\Sigma_{\ell=1}^{k}\boldsymbol{w}_{\ell},\boldsymbol{w}_{j+1},\dots,\boldsymbol{w}_{k}\right) from a matrix (𝒘1,,𝒘k)\left(\boldsymbol{w}_{1},\dots,\boldsymbol{w}_{k}\right) satisfying the condition (\star).

  • Remove a row and a column from a matrix satisfying the condition (\star).

Put M=(𝒘1,,𝒘k)M=\left(\boldsymbol{w}_{1},\dots,\boldsymbol{w}_{k}\right). We will show by cases.

(i) In the case that MM has a row whose multiset of nonzero entries is {1}\{-1\} or {1}\{1\}, the cofactor expantion of the row induces |detM|12k1<2k|\det M|\leq 1\cdot 2^{k-1}<2^{k}.

(ii) In the case that MM has a row whose multiset of nonzero entries is {2}\{-2\} or {2}\{2\}, the cofactor expantion of the row induces |detM|22k1=2k|\det M|\leq 2\cdot 2^{k-1}=2^{k}.

(iii) In the case that MM has a row whose multiset of nonzero entries is {1,1}\{-1,-1\}, {1,1}\{-1,1\} or {1,1}\{1,1\}, the cofactor expansion of the row induces |detM|2k1+2k1=2k|\det M|\leq 2^{k-1}+2^{k-1}=2^{k}.

(iv) In the case that MM has a row 𝒗i\boldsymbol{v}_{i} whose multiset of nonzero entries is {2,1}\{-2,1\}, {1,2}\{-1,2\}, {2,1,1}\{-2,1,1\} or {1,1,2}\{-1,-1,2\}, assume that the (i,j)(i,j)-entry is 2-2 or 22 for some jj. Then we can see the determinant as follows.

|detM|\displaystyle|\det M| =\displaystyle= |det(𝒘1,,𝒘j1,Σ=1k𝒘,𝒘j+1,,𝒘k)|\displaystyle\left|\det\left(\boldsymbol{w}_{1},\dots,\boldsymbol{w}_{j-1},-\Sigma_{\ell=1}^{k}\boldsymbol{w}_{\ell},\boldsymbol{w}_{j+1},\dots,\boldsymbol{w}_{k}\right)\right|
=\displaystyle= |det(±1±100𝒘1𝒘2𝒘3𝒘k)|.\displaystyle\left|\det\begin{pmatrix}\pm 1&\pm 1&0&\cdots&0\\ \boldsymbol{w}^{\prime}_{1}&\boldsymbol{w}^{\prime}_{2}&\boldsymbol{w}^{\prime}_{3}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix}\right|.

Thus, this case comes down to the case (iii), and hence, we have |detM|2k|\det M|\leq 2^{k}.

(v) In the case that MM has a row 𝒗i\boldsymbol{v}_{i} whose multiset of nonzero entries is {2,2}\{-2,2\}, assume that the (i,j)(i,j)-entry is 2-2 or 22 for some jj. Then we can see the determinant as follows.

|detM|\displaystyle|\det M| =\displaystyle= |det(𝒘1,,𝒘j1,Σ=1k𝒘,𝒘j+1,,𝒘k)|\displaystyle\left|\det\left(\boldsymbol{w}_{1},\dots,\boldsymbol{w}_{j-1},-\Sigma_{\ell=1}^{k}\boldsymbol{w}_{\ell},\boldsymbol{w}_{j+1},\dots,\boldsymbol{w}_{k}\right)\right|
=\displaystyle= |det(±200𝒘1𝒘2𝒘k)|.\displaystyle\left|\det\begin{pmatrix}\pm 2&0&\cdots&0\\ \boldsymbol{w}^{\prime}_{1}&\boldsymbol{w}^{\prime}_{2}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix}\right|.

Thus, this case comes down to the case (ii), and hence, we have |detM|2k|\det M|\leq 2^{k}.

(vi) In the case that MM has a row 𝒗i\boldsymbol{v}_{i} whose multiset of nonzero entries is {1,1,1}\{-1,-1,1\} or {1,1,1}\{-1,1,1\}, we can see the determinant as follows.

|detM|\displaystyle|\det M| =\displaystyle= |det(±11100𝒘1𝒘2𝒘3𝒘4𝒘k)|\displaystyle\left|\det\begin{pmatrix}\pm 1&1&-1&0&\cdots&0\\ \boldsymbol{w}^{\prime}_{1}&\boldsymbol{w}^{\prime}_{2}&\boldsymbol{w}^{\prime}_{3}&\boldsymbol{w}^{\prime}_{4}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix}\right|
=\displaystyle= |det(±11000𝒘1𝒘2𝒘2+𝒘3𝒘4𝒘k)|\displaystyle\left|\det\begin{pmatrix}\pm 1&1&0&0&\cdots&0\\ \boldsymbol{w}^{\prime}_{1}&\boldsymbol{w}^{\prime}_{2}&\boldsymbol{w}^{\prime}_{2}+\boldsymbol{w}^{\prime}_{3}&\boldsymbol{w}^{\prime}_{4}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix}\right|
\displaystyle\leq |det(𝒘2𝒘2+𝒘3𝒘4𝒘k)|\displaystyle\left|\det\begin{pmatrix}\boldsymbol{w}^{\prime}_{2}&\boldsymbol{w}^{\prime}_{2}+\boldsymbol{w}^{\prime}_{3}&\boldsymbol{w}^{\prime}_{4}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix}\right|
+|det(𝒘1𝒘2+𝒘3𝒘4𝒘k)|\displaystyle+\left|\det\begin{pmatrix}\boldsymbol{w}^{\prime}_{1}&\boldsymbol{w}^{\prime}_{2}+\boldsymbol{w}^{\prime}_{3}&\boldsymbol{w}^{\prime}_{4}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix}\right|
=\displaystyle= |det(𝒘2𝒘3𝒘4𝒘k)|\displaystyle\left|\det\begin{pmatrix}\boldsymbol{w}^{\prime}_{2}&\boldsymbol{w}^{\prime}_{3}&\boldsymbol{w}^{\prime}_{4}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix}\right|
+|det(𝒘1Σ=1k𝒘𝒘4𝒘k)|\displaystyle+\left|\det\begin{pmatrix}\boldsymbol{w}^{\prime}_{1}&-\Sigma_{\ell=1}^{k}{\boldsymbol{w}^{\prime}_{\ell}}&\boldsymbol{w}^{\prime}_{4}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix}\right|

Since the matrices (𝒘2𝒘k)\begin{pmatrix}\boldsymbol{w}^{\prime}_{2}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix} and (𝒘1Σ=1k𝒘𝒘4𝒘k)\begin{pmatrix}\boldsymbol{w}^{\prime}_{1}&-\Sigma_{\ell=1}^{k}{\boldsymbol{w}^{\prime}_{\ell}}&\boldsymbol{w}^{\prime}_{4}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix} satisfy the condition ()(\star), we obtain |detM|2k1+2k1=2k|\det M|\leq 2^{k-1}+2^{k-1}=2^{k}.

(vii) Lastly we consider the case that MM has a row 𝒗i\boldsymbol{v}_{i} whose multiset of nonzero entries is {1,1,1,1}\{-1,-1,1,1\}. We then have

|detM|\displaystyle|\det M| =\displaystyle= |det(111100𝒘1𝒘2𝒘3𝒘4𝒘5𝒘k)|\displaystyle\left|\det\begin{pmatrix}1&1&-1&-1&0&\cdots&0\\ \boldsymbol{w}^{\prime}_{1}&\boldsymbol{w}^{\prime}_{2}&\boldsymbol{w}^{\prime}_{3}&\boldsymbol{w}^{\prime}_{4}&\boldsymbol{w}^{\prime}_{5}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix}\right|
=\displaystyle= |det(111000𝒘1𝒘2𝒘3Σ=1k𝒘𝒘5𝒘k)|.\displaystyle\left|\det\begin{pmatrix}1&1&-1&0&0&\cdots&0\\ \boldsymbol{w}^{\prime}_{1}&\boldsymbol{w}^{\prime}_{2}&\boldsymbol{w}^{\prime}_{3}&-\Sigma_{\ell=1}^{k}\boldsymbol{w}^{\prime}_{\ell}&\boldsymbol{w}^{\prime}_{5}&\cdots&\boldsymbol{w}^{\prime}_{k}\end{pmatrix}\right|.

This case comes down to the case (vi), and hence, we have |detM|2k|\det M|\leq 2^{k}.

This completes the proof. ∎

Let pp be an odd prime integer. Let DD be a diagram of a knot KK. Let x1,,xn+2x_{1},\ldots,x_{n+2} be the regions, and c1,,cnc_{1},\ldots,c_{n} the crossings of DD. Let C1TC^{\rm 1T} and C2TC^{\rm 2T} be the 11- and 22-trivial Dehn pp-coloring, respectively, such that C1T(x1)=1C^{\rm 1T}(x_{1})=1, C2T(x1)=0C^{\rm 2T}(x_{1})=0 and ImC2T={0,1}{\rm Im}C^{\rm 2T}=\{0,1\}. Let CC be a nontrivial Dehn pp-coloring of DD.

Lemma 3.2.

Suppose that for any i,j{1,,n+2}i,j\in\{1,\ldots,n+2\} such that C(xi)=C(xj)C(x_{i})=C(x_{j}), C2T(xi)=C2T(xj)C^{\rm 2T}(x_{i})=C^{\rm 2T}(x_{j}) holds. Then we have

#𝒞(D,C)log2p+1.\#\mathcal{C}(D,C)\geq\log_{2}p+1.
Proof.

Put =#𝒞(D,C)\ell=\#\mathcal{C}(D,C) and M1=AD(1)M_{1}=A_{D}(-1), where AD(1)A_{D}(-1) is the matrix constructed in Section 2 by setting an arbitrary orientation for DD. By Lemma 1.2 and Remark 1.3, we may assume that C(x1)=0C(x_{1})=0 by replacing the Dehn pp-coloring CC by a regular affine transformation if necessary. Put 𝒂1=t(C(x1)C(xn+2))\boldsymbol{a}_{1}=\,^{t}(C(x_{1})\cdots C(x_{n+2})) and 𝒂12T=t(C2T(x1)C2T(xn+2))\boldsymbol{a}_{1}^{\rm 2T}=\,^{t}(C^{\rm 2T}(x_{1})\cdots C^{\rm 2T}(x_{n+2})). Then by (2) of Lemma 2.1,

rankM1=n+1 and rankpM1n,{\rm rank}_{\mathbb{Z}}M_{1}=n+1\mbox{ and }{\rm rank}_{p}M_{1}\leq n,

and the simultaneous equations M1𝒙=𝟎M_{1}\boldsymbol{x}=\boldsymbol{0} have the linearly independent solution 𝒂12T\boldsymbol{a}_{1}^{\rm 2T} on \mathbb{Z}, and two linearly independent solutions 𝒂12T\boldsymbol{a}_{1}^{\rm 2T} and 𝒂1\boldsymbol{a}_{1} on p\mathbb{Z}_{p}.

We put M1=(𝒎1𝒎n+2)M_{1}=(\boldsymbol{m}_{1}\cdots\boldsymbol{m}_{n+2}) and 𝒙=t(x1xn+2)\boldsymbol{x}=\,^{t}(x_{1}\cdots x_{n+2}). We transform M1M_{1}, 𝒙\boldsymbol{x}, 𝒂1\boldsymbol{a}_{1} and 𝒂12T\boldsymbol{a}_{1}^{\rm 2T} as follows. Suppose that i,j{1,,n+2}i,j\in\{1,\dots,n+2\} satisfy i<ji<j and C(xi)=C(xj)C(x_{i})=C(x_{j}). For M1M_{1}, we add 𝒎j\boldsymbol{m}_{j} to 𝒎i\boldsymbol{m}_{i} and delete 𝒎j\boldsymbol{m}_{j}. For 𝒙\boldsymbol{x}, 𝒂1\boldsymbol{a}_{1} and 𝒂12T\boldsymbol{a}_{1}^{\rm 2T}, we delete xjx_{j}, C(xj)C(x_{j}) and C2T(xj)C^{\rm 2T}(x_{j}), respectively. We then have the new equation

(𝒎1𝒎i+𝒎j𝒎j^𝒎n+2)t(x1xixj^xn+2)=𝟎,(\boldsymbol{m}_{1}\cdots\boldsymbol{m}_{i}+\boldsymbol{m}_{j}\cdots\widehat{\boldsymbol{m}_{j}}\cdots\boldsymbol{m}_{n+2})^{t}(x_{1}\cdots x_{i}\cdots\widehat{x_{j}}\cdots x_{n+2})=\boldsymbol{0},

and the solutions

(C(x1)C(xi)C(xj)^C(xn+2))t{}^{t}(C(x_{1})\cdots C(x_{i})\cdots\widehat{C(x_{j})}\cdots C(x_{n+2}))

on p\mathbb{Z}_{p}, and

(C2T(x1)C2T(xi)C2T(xj)^C2T(xn+2))t{}^{t}(C^{\rm 2T}(x_{1})\cdots C^{\rm 2T}(x_{i})\cdots\widehat{C^{\rm 2T}(x_{j})}\cdots C^{\rm 2T}(x_{n+2}))

on \mathbb{Z} and p\mathbb{Z}_{p}, where x^\widehat{x} means that the entry xx is removed. For the resultant matrix obtained from M1M_{1} and the resultant column vectors obtained from 𝒙\boldsymbol{x}, 𝒂1\boldsymbol{a}_{1} and 𝒂12T\boldsymbol{a}_{1}^{\rm 2T}, we repeat the same transformations unless the number of column vectors for the resultant matrix originally from M1M_{1} is \ell. Then, as a consequence, we have the n×n\times\ell-matrix M2M_{2}, the column vectors 𝒚=t(y1,,y)\boldsymbol{y}=\ ^{t}(y_{1},\dots,y_{\ell}), 𝒂2=t(C(y1),,C(y))\boldsymbol{a}_{2}=\ ^{t}(C(y_{1}),\dots,C(y_{\ell})) and 𝒂22T=t(C2T(y1),,C2T(y))\boldsymbol{a}_{2}^{\rm 2T}=\ ^{t}(C^{\rm 2T}(y_{1}),\dots,C^{\rm 2T}(y_{\ell})), where 𝒚\boldsymbol{y} is the column vector originally from 𝒙\boldsymbol{x}.

Remark that =#𝒞(D,C)\ell=\#\mathcal{C}(D,C). We can easily see that rankM2=1{\rm rank}_{\mathbb{Z}}M_{2}=\ell-1 and rankpM22{\rm rank}_{p}M_{2}\leq\ell-2, and 𝒂22T\boldsymbol{a}^{\rm 2T}_{2} is a linearly independent solution of M2𝒚=𝟎M_{2}\boldsymbol{y}=\boldsymbol{0} on \mathbb{Z}, and 𝒂2\boldsymbol{a}_{2} and 𝒂22T\boldsymbol{a}^{\rm 2T}_{2} are linearly independent solutions on p\mathbb{Z}_{p}. Hence, there exists an (1)×(1)(\ell-1)\times(\ell-1)-submatrix M3M_{3} of M2M_{2} such that detM30\det M_{3}\neq 0\in\mathbb{Z} and detM3=0p\det M_{3}=0\in\mathbb{Z}_{p}, which implies that p|detM3|p\leq|\det M_{3}|.

Furthermore, since we have |detM3|21|\det M_{3}|\leq 2^{\ell-1} by Lemma 3.1, it holds that p21p\leq 2^{\ell-1}. This gives the conclusion #𝒞(D,C)=log2p+1\#\mathcal{C}(D,C)=\ell\geq\log_{2}{p}+1. ∎

Lemma 3.3.

Suppose that there exists i,j{1,,n+2}i,j\in\{1,\ldots,n+2\} with i<ji<j such that C(xi)=C(xj)C(x_{i})=C(x_{j}) and C2T(xi)C2T(xj)C^{\rm 2T}(x_{i})\not=C^{\rm 2T}(x_{j}). Then we have

#𝒞(D,C)log2p+1.\#\mathcal{C}(D,C)\geq\log_{2}p+1.
Proof.

We prove this lemma in much the same way as Lemma 3.2. Put =#𝒞(D,C)\ell=\#\mathcal{C}(D,C) and M1=BD,C;i,j(1)M_{1}=B_{D,C;i,j}(-1), where BD,C;i,j(1)B_{D,C;i,j}(-1) is the matrix constructed in Section 2 by setting an arbitrary orientation for DD. Put 𝒂1=t(C(x1)C(xn+2))\boldsymbol{a}_{1}=\,^{t}(C(x_{1})\cdots C(x_{n+2})) and 𝒂11T=t(C1T(x1)C1T(xn+2))=t(1,,1)\boldsymbol{a}_{1}^{\rm 1T}=\,^{t}(C^{\rm 1T}(x_{1})\cdots C^{\rm 1T}(x_{n+2}))=\,^{t}(1,\dots,1). Then by (3) of Lemma 2.1,

rankM1=n+1 and rankpM1n,{\rm rank}_{\mathbb{Z}}M_{1}=n+1\mbox{ and }{\rm rank}_{p}M_{1}\leq n,

and the simultaneous equations M1𝒙=𝟎M_{1}\boldsymbol{x}=\boldsymbol{0} have the linearly independent solution 𝒂11T\boldsymbol{a}_{1}^{\rm 1T} on \mathbb{Z}, and two linearly independent solutions 𝒂11T\boldsymbol{a}_{1}^{\rm 1T} and 𝒂1\boldsymbol{a}_{1} on p\mathbb{Z}_{p}.

We put M1=(𝒎1𝒎n+2)M_{1}=(\boldsymbol{m}_{1}\cdots\boldsymbol{m}_{n+2}) and 𝒙=t(x1xn+2)\boldsymbol{x}=\,^{t}(x_{1}\cdots x_{n+2}). We transform M1M_{1}, 𝒙\boldsymbol{x}, 𝒂1\boldsymbol{a}_{1} and 𝒂11T\boldsymbol{a}_{1}^{\rm 1T} as follows. Suppose that i,j{1,,n+2}i,j\in\{1,\dots,n+2\} satisfy i<ji<j and C(xi)=C(xj)C(x_{i})=C(x_{j}). For M1M_{1}, we add 𝒎j\boldsymbol{m}_{j} to 𝒎i\boldsymbol{m}_{i} and delete 𝒎j\boldsymbol{m}_{j}. For 𝒙\boldsymbol{x}, 𝒂1\boldsymbol{a}_{1} and 𝒂11T\boldsymbol{a}_{1}^{\rm 1T}, we delete xjx_{j}, C(xj)C(x_{j}) and C1T(xj)C^{\rm 1T}(x_{j}), respectively. We then have the new equation

(𝒎1𝒎i+𝒎j𝒎j^𝒎n+2)t(x1xixj^xn+2)=𝟎,(\boldsymbol{m}_{1}\cdots\boldsymbol{m}_{i}+\boldsymbol{m}_{j}\cdots\widehat{\boldsymbol{m}_{j}}\cdots\boldsymbol{m}_{n+2})^{t}(x_{1}\cdots x_{i}\cdots\widehat{x_{j}}\cdots x_{n+2})=\boldsymbol{0},

and the solutions

(C(x1)C(xi)C(xj)^C(xn+2))t{}^{t}(C(x_{1})\cdots C(x_{i})\cdots\widehat{C(x_{j})}\cdots C(x_{n+2}))

on p\mathbb{Z}_{p}, and

(C1T(x1)C1T(xi)C1T(xj)^C1T(xn+2))t{}^{t}(C^{\rm 1T}(x_{1})\cdots C^{\rm 1T}(x_{i})\cdots\widehat{C^{\rm 1T}(x_{j})}\cdots C^{\rm 1T}(x_{n+2}))

on \mathbb{Z} and p\mathbb{Z}_{p}. For the resultant matrix obtained from M1M_{1} and the resultant column vectors obtained from 𝒙\boldsymbol{x}, 𝒂1\boldsymbol{a}_{1} and 𝒂11T\boldsymbol{a}_{1}^{\rm 1T}, we repeat the same transformations unless the number of column vectors for the resultant matrix originally from M1M_{1} is \ell. Then, as a consequence, we have the n×n\times\ell-matrix M2M_{2}, the column vectors 𝒚=t(y1,,y)\boldsymbol{y}=\ ^{t}(y_{1},\dots,y_{\ell}), 𝒂2=t(C(y1),,C(y))\boldsymbol{a}_{2}=\ ^{t}(C(y_{1}),\dots,C(y_{\ell})) and 𝒂21T=t(C1T(y1),,C1T(y))\boldsymbol{a}_{2}^{\rm 1T}=\ ^{t}(C^{\rm 1T}(y_{1}),\dots,C^{\rm 1T}(y_{\ell})), where 𝒚\boldsymbol{y} is the column vector originally from 𝒙\boldsymbol{x}.

We can easily see that rankM2=1 and rankpM22{\rm rank}_{\mathbb{Z}}M_{2}=\ell-1\mbox{ and }{\rm rank}_{p}M_{2}\leq\ell-2, and 𝒂21T\boldsymbol{a}_{2}^{\rm 1T} is a linearly independent solution of M2𝒚=𝟎M_{2}\boldsymbol{y}=\boldsymbol{0} on \mathbb{Z}, and 𝒂2\boldsymbol{a}_{2} and 𝒂21T\boldsymbol{a}_{2}^{\rm 1T} are linearly independent solutions on p\mathbb{Z}_{p}.

Hence, there exists an (1)×(1)(\ell-1)\times(\ell-1)-submatrix M3M_{3} of M2M_{2} such that detM30\det M_{3}\neq 0\in\mathbb{Z} and detM3=0p\det M_{3}=0\in\mathbb{Z}_{p}, which implies that p|detM3|p\leq|\det M_{3}|. Furthermore, since we have |detM3|21|\det M_{3}|\leq 2^{\ell-1} by Lemma 3.1, it holds that p21p\leq 2^{\ell-1}. This gives the conclusion #𝒞(D,C)=log2p+1\#\mathcal{C}(D,C)=\ell\geq\log_{2}{p}+1.

Proof of Theorem 1.5..

From Lemmas 3.2 and 3.3, we have

#𝒞(D,C)log2p+1\#\mathcal{C}(D,C)\geq\log_{2}p+1

for any nontrivially Dehn pp-colored diagram (D,C)(D,C) of a knot. Therefore, we have

mincolpDehn(K)log2p+1{\rm mincol}^{\rm Dehn}_{p}(K)\geq\log_{2}p+1

for any Dehn pp-colorable knot KK. ∎

4. \mathcal{R}-palette graphs

Let pp be an odd prime number, and let SpS\subset\mathbb{Z}_{p} with SS\not=\emptyset. Set μ(S)={{a1,a2}|a1,a2S}\mu(S)=\{\{a_{1},a_{2}\}~{}|~{}a_{1},a_{2}\in S\}, where {a1,a2}\{a_{1},a_{2}\} is regarded as the multiset {a1,a1}\{a_{1},a_{1}\} when a1=a2a_{1}=a_{2}.

For {a1,a2},{a3,a4}μ(S)\{a_{1},a_{2}\},\{a_{3},a_{4}\}\in\mu(S), we set

{a1,a2}{a3,a4} if a1+a2=a3+a4 in p.\{a_{1},a_{2}\}\sim\{a_{3},a_{4}\}\mbox{ if $a_{1}+a_{2}=a_{3}+a_{4}$ in $\mathbb{Z}_{p}$}.

We note that this relation \sim on μ(S)\mu(S) is an equivalence relation. We denote by a1+a2¯\overline{a_{1}+a_{2}} the equivalence class of {a1,a2}μ(S)\{a_{1},a_{2}\}\in\mu(S).

Definition 4.1.

The \mathcal{R}-palette graph of SS is the simple graph GS=(VS,ES)G_{S}=(V_{S},E_{S}) composed of the vertex set

VS=μ(S)/={a1+a2¯|{a1,a2}μ(S)}V_{S}=\mu(S)/\sim=\{\mathbin{\overline{a_{1}+a_{2}}}~{}|~{}\{a_{1},a_{2}\}\in\mu(S)\}

and the edge set ESE_{S} satisfying that

e=b1¯b2¯ESthere exist {a1,a2}b1¯ and {a3,a4}b2¯ such that a1+a3=a2+a4 or a1+a4=a2+a3 in p,e=\mathbin{\overline{b_{1}}}\,\mathbin{\overline{b_{2}}}\in E_{S}\iff\begin{array}[]{l}\mbox{there exist $\{a_{1},a_{2}\}\in\mathbin{\overline{b_{1}}}$ and $\{a_{3},a_{4}\}\in\mathbin{\overline{b_{2}}}$ such }\\ \mbox{that $a_{1}+a_{3}=a_{2}+a_{4}$ or $a_{1}+a_{4}=a_{2}+a_{3}$ in $\mathbb{Z}_{p}$,}\end{array}

where e=b1¯b2¯e=\mathbin{\overline{b_{1}}}\,\mathbin{\overline{b_{2}}} means that ee is an edge connecting the vertices b1¯\mathbin{\overline{b_{1}}} and b2¯\mathbin{\overline{b_{2}}}. We attach the label 21(b1+b2)¯p\mathbin{\overline{2^{-1}(b_{1}+b_{2})}}\in\mathbb{Z}_{p} to the edge ee between b1¯\mathbin{\overline{b_{1}}} and b2¯\mathbin{\overline{b_{2}}} (see Figure 4 for example).

Refer to caption
Figure 4. \mathcal{R}-palette graphs

Let GSG_{S} be the \mathcal{R}-palette graph of SS. A graph G=(V,E)G=(V,E) is an \mathcal{R}-subgraph of GSG_{S} if GG is a subgraph of GSG_{S}, and

eEbe¯Ve\in E\Longrightarrow\mathbin{\overline{b_{e}}}\in V
(i.e., e=b1¯b2¯E21(b1+b2)¯V)(\mbox{i.e., }e=\mathbin{\overline{b_{1}}}\,\mathbin{\overline{b_{2}}}\in E\Longrightarrow\mathbin{\overline{2^{-1}(b_{1}+b_{2})}}\in V)

holds, where be¯\mathbin{\overline{b_{e}}} is the label of ee (see Figure 5 for example).

Refer to caption
Figure 5. \mathcal{R}-palette graphs

Let (D,C)(D,C) be a nontrivially Dehn pp-colored diagram of a knot. The \mathcal{R}-palette graph of (D,C)(D,C) is an \mathcal{R}-subgraph G(D,C)=(V(D,C),E(D,C))G_{(D,C)}=(V_{(D,C)},E_{(D,C)}) of G𝒞(D,C)G_{\mathcal{C}(D,C)} composed of the vertex set

V(D,C)={b¯|there exists a b¯-arc on (D,C)}V_{(D,C)}=\{\mathbin{\overline{b}}~{}|~{}\mbox{there exists a $\overline{b}$-arc on $(D,C)$}\}

and the edge set E(D,C)E_{(D,C)} satisfying that

e=b1¯b2¯E(D,C)there exists a crossing on (D,C) consisting of both the b1¯- and b2¯-under-arcs.e=\mathbin{\overline{b_{1}}}\,\mathbin{\overline{b_{2}}}\in E_{(D,C)}\iff\begin{minipage}{227.62204pt} there exists a crossing on $(D,C)$ consisting of both the $\mathbin{\overline{b_{1}}}$- and $\mathbin{\overline{b_{2}}}$-under-arcs. \end{minipage}

As in the case of GSG_{S}, we attach the label 21(b1+b2)¯p\mathbin{\overline{2^{-1}(b_{1}+b_{2})}}\in\mathbb{Z}_{p} to the edge ee between b1¯\mathbin{\overline{b_{1}}} and b2¯\mathbin{\overline{b_{2}}} (see Figure 6 for example).

We remark that b¯\bar{b} for a b¯\bar{b}-arc was defined just a symbol, but b¯\bar{b} is regarded as an equivalence class as in Definition 4.1 when it represents a vertex.

We also remark that this \mathcal{R}-palette graph of (D,C)(D,C) is isomorphic to the palette graph of (D,C¯)(D,\bar{C}) defined in [8], where C¯\bar{C} is the Fox pp-coloring corresponding to CC as shown in Remark 1.1. Furthermore, it was shown in [8] that the palette graph of (D,C¯)(D,\bar{C}) is connected and has at least log2p+2\lfloor\log_{2}p\rfloor+2 vertices.

Refer to caption
Figure 6. The \mathcal{R}-palette graph of (D,C)(D,C)

5. Evaluations of mincolpDehn(K){\rm mincol}^{\rm Dehn}_{p}(K) by \mathcal{R}-palette graphs

Let pp be an odd prime number.

The next theorem can be used to evaluate the minimum number of colors of knots or select some sets of colors, as candidates, each of which might be the set of colors for a nontrivial Dehn pp-coloring of a knot diagram.

Theorem 5.1.

Let SpS\subset\mathbb{Z}_{p}. If S=𝒞(D,C)S=\mathcal{C}(D,C) for some nontrivially Dehn pp-colored diagram (D,C)(D,C) of a knot, the \mathcal{R}-palette graph GSG_{S} includes a connected \mathcal{R}-subgraph with at least log2p+2\lfloor\log_{2}p\rfloor+2 vertices.

Proof.

This is evident because GSG_{S} includes G(D,C)G_{(D,C)} and as remarked in the previous section, G(D,C)G_{(D,C)} is connected and has at least log2p+2\lfloor\log_{2}p\rfloor+2 vertices. ∎

Example 5.2.

For any subset S7S\subset\mathbb{Z}_{7} such that #S3\#S\leq 3, the \mathcal{R}-palette graph GSG_{S} includes no connected \mathcal{R}-subgraph with at least log27+2\lfloor\log_{2}7\rfloor+2 vertices. This and Theorem 5.1 imply that mincol7Dehn(K)4{\rm mincol}^{\rm Dehn}_{7}(K)\geq 4 for any Dehn 77-colorable knot KK.

Any subset of 7\mathbb{Z}_{7} with four elements is affine equivalent to either {0,1,2,3}\{0,1,2,3\} or {0,1,2,4}\{0,1,2,4\}. We note that by Lemma 1.2 and Remark 1.3, when we select candidates of sets of four colors each of which might be the set of colors for a nontrivially Dehn pp-colored diagram of a knot, it suffices to discuss only the representatives of the affine equivalence classes, that is {0,1,2,3}\{0,1,2,3\} and {0,1,2,4}\{0,1,2,4\}. As shown in the right of Figure 4, the \mathcal{R}-palette graph G{0,1,2,4}G_{\{0,1,2,4\}} clearly has a connected \mathcal{R}-subgraph with at least log27+2\lfloor\log_{2}7\rfloor+2 vertices, which is G{0,1,2,4}G_{\{0,1,2,4\}} itself for example. On the other hand, the \mathcal{R}-palette graph G{0,1,2,3}G_{\{0,1,2,3\}} depicted in the left of Figure 4 does not include a connected \mathcal{R}-subgraph with at least log27+2\lfloor\log_{2}7\rfloor+2 vertices. Therefore, if there exists a nontrivially Dehn 77-colored diagram (D,C)(D,C) for a knot with #𝒞(D,C)=4(=log27+2)\#\mathcal{C}(D,C)=4(=\lfloor\log_{2}7\rfloor+2), then 𝒞(D,C)\mathcal{C}(D,C) is affine equivalent to {0,1,2,4}\{0,1,2,4\}, and moreover, there exists a Dehn 77-coloring CC^{\prime} of DD with CCC\sim C^{\prime} and 𝒞(D,C)={0,1,2,4}\mathcal{C}(D,C^{\prime})=\{0,1,2,4\}.

Similarly as Example 5.2, for each odd prime pp with p<25p<2^{5}, we can give the candidates of log2p+2\lfloor\log_{2}p\rfloor+2 colors by using \mathcal{R}-palette graphs, where log2p+2\lfloor\log_{2}p\rfloor+2 is the lower bound of the evaluation formula of Theorem 1.5. The next theorem gives the results.

Theorem 5.3.

Let pp be an odd prime number with p<25p<2^{5}.

  • (1)

    There is no subset SpS\subset\mathbb{Z}_{p} with at most log2p+1\lfloor\log_{2}p\rfloor+1 elements such that the \mathcal{R}-palette graph GSG_{S} includes a connected \mathcal{R}-subgraph with at least log2p+2\lfloor\log_{2}p\rfloor+2 vertices. This implies that

    mincolpDehn(K)log2p+2{\rm mincol}^{\rm Dehn}_{p}(K)\geq\lfloor\log_{2}p\rfloor+2

    for any Dehn pp-colorable knot KK.

  • (2)

    If there exists a nontrivially Dehn pp-colored diagram (D,C)(D,C) of a knot such that #𝒞(D,C)=log2p+2\#\mathcal{C}(D,C)=\lfloor\log_{2}p\rfloor+2, then

    • (i)

      𝒞(D,C){0,1,2}\mathcal{C}(D,C)\sim\{0,1,2\} when p=3p=3,

    • (ii)

      𝒞(D,C){0,1,2,3}\mathcal{C}(D,C)\sim\{0,1,2,3\} when p=5p=5,

    • (iii)

      𝒞(D,C){0,1,2,4}\mathcal{C}(D,C)\sim\{0,1,2,4\} when p=7p=7,

    • (iv)

      𝒞(D,C){0,1,2,3,6}\mathcal{C}(D,C)\sim\{0,1,2,3,6\} or {0,1,2,4,7}\{0,1,2,4,7\} when p=11p=11,

    • (v)

      𝒞(D,C){0,1,2,4,7}\mathcal{C}(D,C)\sim\{0,1,2,4,7\} when p=13p=13,

    • (vi)

      𝒞(D,C){0,1,2,3,5,9}\mathcal{C}(D,C)\sim\{0,1,2,3,5,9\}, {0,1,2,3,5,10}\{0,1,2,3,5,10\}, {0,1,2,3,5,12}\{0,1,2,3,5,12\}, {0,1,2,3,6,9}\{0,1,2,3,6,9\}, {0,1,2,3,6,10}\{0,1,2,3,6,10\}, {0,1,2,3,6,11}\{0,1,2,3,6,11\}, {0,1,2,3,6,13}\{0,1,2,3,6,13\}, {0,1,2,3,7,11}\{0,1,2,3,7,11\}, {0,1,2,4,5,9}\{0,1,2,4,5,9\}, {0,1,2,4,5,10}\{0,1,2,4,5,10\}, {0,1,2,4,5,12}\{0,1,2,4,5,12\}, or {0,1,2,4,10,13}\{0,1,2,4,10,13\} when p=17p=17,

    • (vii)

      𝒞(D,C){0,1,2,3,5,10}\mathcal{C}(D,C)\sim\{0,1,2,3,5,10\}, {0,1,2,3,6,10}\{0,1,2,3,6,10\}, {0,1,2,3,6,11}\{0,1,2,3,6,11\}, {0,1,2,3,6,12}\{0,1,2,3,6,12\}, {0,1,2,3,6,13}\{0,1,2,3,6,13\}, {0,1,2,3,6,14}\{0,1,2,3,6,14\}, {0,1,2,3,7,12}\{0,1,2,3,7,12\}, {0,1,2,4,5,10}\{0,1,2,4,5,10\}, {0,1,2,4,5,14}\{0,1,2,4,5,14\}, {0,1,2,4,7,12}\{0,1,2,4,7,12\}, or {0,1,2,4,7,15}\{0,1,2,4,7,15\} when p=19p=19,

    • (viii)

      𝒞(D,C){0,1,2,3,6,12}\mathcal{C}(D,C)\sim\{0,1,2,3,6,12\}, {0,1,2,4,7,12}\{0,1,2,4,7,12\}, {0,1,2,4,7,13}\{0,1,2,4,7,13\}, {0,1,2,4,7,14}\{0,1,2,4,7,14\}, {0,1,2,4,9,14}\{0,1,2,4,9,14\}, or {0,1,2,4,10,19}\{0,1,2,4,10,19\} when p=23p=23,

    • (ix)

      𝒞(D,C){0,1,2,4,8,15}\mathcal{C}(D,C)\sim\{0,1,2,4,8,15\} when p=29p=29, and

    • (x)

      𝒞(D,C){0,1,2,4,8,16}\mathcal{C}(D,C)\sim\{0,1,2,4,8,16\} when p=31p=31.

Proof.

This can be shown by Mathematica computations. In particular, we found a connected \mathcal{R}-subgraph with at least log2p+2\lfloor\log_{2}p\rfloor+2 vertices for the \mathcal{R}-palette graph of each candidate of the sets of colors given in (2) of this theorem as depicted in Figures 711. ∎

Refer to caption
Figure 7. Connected \mathcal{R}-subgraphs for p{3,5,7,11,13}p\in\{3,5,7,11,13\}
Refer to caption
Figure 8. Connected \mathcal{R}-subgraphs for p{17}p\in\{17\}
Refer to caption
Figure 9. Connected \mathcal{R}-subgraphs for p{17}p\in\{17\}
Refer to caption
Figure 10. Connected \mathcal{R}-subgraphs for p{19}p\in\{19\}
Refer to caption
Figure 11. Connected \mathcal{R}-subgraphs for p{23,29,31}p\in\{23,29,31\}

We have the following proposition.

Proposition 5.4.

For each odd prime number pp with p<25p<2^{5} and p{13,29}p\not\in\{13,29\}, there exists a Dehn pp-colorable knot KK with mincolpDehn(K)=log2p+2{\rm mincol}^{\rm Dehn}_{p}(K)=\lfloor\log_{2}p\rfloor+2.

Proof.

For p=3p=3 (resp. 5,7,11,17,19,23,315,7,11,17,19,23,31), the diagram illustrated in Figure 13 (resp. 13, 15, 15, 17, 17, 19, 19) admits the Dehn pp-coloring with log2p+2\lfloor\log_{2}p\rfloor+2 colors. Hence, from these properties and Theorem 1.5, we can see that there exists a knot KK with mincolpDehn(K)=log2p+2{\rm mincol}^{\rm Dehn}_{p}(K)=\lfloor\log_{2}p\rfloor+2 for each p{3,5,7,11,17,19,23,31}p\in\{3,5,7,11,17,19,23,31\}.

Refer to caption
Figure 12. A Dehn 33-colored diagram
Refer to caption
Figure 13. A Dehn 55-colored diagram
Refer to caption
Figure 14. A Dehn 77-colored diagram
Refer to caption
Figure 15. A Dehn 1111-colored diagram
Refer to caption
Figure 16. A Dehn 1717-colored diagram
Refer to caption
Figure 17. A Dehn 1919-colored diagram
Refer to caption
Figure 18. A Dehn 2323-colored diagram
Refer to caption
Figure 19. A Dehn 3131-colored diagram

Remark 5.5.

For p=13p=13 or 2929, there is no Dehn pp-colorable knot KK with mincolpDehn(K)=log2p+2{\rm mincol}^{\rm Dehn}_{p}(K)=\lfloor\log_{2}p\rfloor+2. This result will be shown in our next paper [7].

Remark 5.6.

The following properties hold.

  • (1)

    When p=3p=3, any pp-colorable knot KK has mincolpDehn(K)=3{\rm mincol}^{\rm Dehn}_{p}(K)=3.

  • (2)

    When p=5p=5, any pp-colorable knot KK has mincolpDehn(K)=4{\rm mincol}^{\rm Dehn}_{p}(K)=4.

The proof of (1) is left as an exercise to the reader. We give the proof of (2) in Appendix A.

Appendix A

Proof of Remark 5.6 (2).

In this proof, an nn-gon means a region bounded by nn semiarcs.

Let (D,C)(D,C) be any nontrivially Dehn 55-colored diagram of a knot. We may construct a Dehn 55-colored diagram (D,C)(D^{\prime},C^{\prime}) with four colors by removing regions with the color 44 from (D,C)(D,C) in the following steps.

  • (Step 1) Remove crossings that have four or two regions colored by 44.

  • (Step 2) Remove nn-gons colored by 44 for n4n\geq 4.

  • (Step 3) Remove 33-, and 22-gons colored by 44.

In this proof, suppose that a,b,ca,b,c are colors with aba\neq b, aca\neq c, a,b,c4a,b,c\not=4.

(Step 1) First, we remove crossings that have only the color 44 by the move depicted by Figure 20, where we note that there exists a region colored by some aa.

Refer to caption
Figure 20.

Next, we remove {4,4}\{4,4\}-semiarcs by the moves as depicted in Figures 2225, where the situation around each {4,4}\{4,4\}-semiarc is represented by one of Figures 2225.

Refer to caption
Figure 21.
Refer to caption
Figure 22.
Refer to caption
Figure 23.
Refer to caption
Figure 24.
Refer to caption
Figure 25.

Here, we note that

2a442a-4\neq 4 for Figure 22,
ab+44a-b+4\neq 4 for Figure 22,
3b,ab+443-b,a-b+4\neq 4 for Figure 24,
3a,2a+1,2a+243-a,2a+1,-2a+2\neq 4 for Figure 24,
3a,3b,ab+4,a+b+443-a,3-b,a-b+4,-a+b+4\neq 4 for Figure 25

from aba\not=b and a,b4a,b\not=4.

Next we delete crossings that have two regions colored by 44s diagonally as the following move depicted in Figure 26. Here we note that there exists a crossing which has four regions colored by a,b,ca,b,c and 44 near an {a,4}\{a,4\}-semiarc.

Refer to caption
Figure 26.

Here we obtain the color a+b+4-a+b+4 that is not equal to 44 from aba\neq b. Thus, for the resultant Dehn 55-colored diagram, there exist no crossings that have four or two regions colored by 44.

(Step 2) For any nn-gon, say xx, colored by 44 with n4n\geq 4, we focus on an {a,4}\{a,4\}-semiarc for some a4a\not=4, say uu, on the boundary of xx. As in Figure 27, we move the semiarc uu by RII-moves with each semiarc on the boundary of xx which is not an {a,4}\{a,4\}-semiarc, where we perform the RII-moves as uu goes under each semiarc. Here, the 2-gons with no label in the right of Figure 27 are not colored by 44. Thus, we can delete the nn-gon with the color 44.

Refer to caption
Figure 27.

We note that the obtained new regions colored by 44 are 44-, 33-, 22-gons such that the color 44 is not appeared in the neighborhood of the regions. We also note that each 44-gon colored by 44 is as depicted in the left of Figure 28.

Next we delete 44-gons colored by 44 as depicted in Figure 28 if bcb\neq c, Figure 29 if b=c,a2b+1b=c,a\not=2b+1 and Figure 30 if b=c,a=2b+1b=c,a=2b+1.

Refer to caption
Figure 28.
Refer to caption
Figure 29.
Refer to caption
Figure 30.

Here, we note that

bc+44 for Figure 28,\displaystyle b-c+4\neq 4\mbox{ for Figure~{}\ref{app12}},
a+b+4,a+2b,2a+2b+44 for Figure 29, and\displaystyle-a+b+4,-a+2b,-2a+2b+4\neq 4\mbox{ for Figure~{}\ref{app13}, and}
b+3,2b+1,3b+24 for Figure 30\displaystyle-b+3,2b+1,3b+2\neq 4\mbox{ for Figure~{}\ref{app14}}

from bcb\not=c, aba\not=b; a2b+1a\not=2b+1, and b4b\not=4, respectively, and the 2-gons with no label in the right of Figures 29 and 30 are not colored by 44.

Thus, for the resultant Dehn 55-colored diagram, each region colored by 44 is a 33- or 22-gon around which the color 44 is not appeared.

(Step 3) Now we consider 33-gons colored by 44, where we note that such 33-gons are as depicted in the left of Figure 31, the left of Figure 32 or the mirror images for some distinct a,b,ca,b,c.

We first remove 33-gons as depicted in the left of Figure 31 by the move shown in Figure 31.

Refer to caption
Figure 31.

Here, we note that

a+b+4,b+c+4,ac+4,a+c+44\displaystyle-a+b+4,-b+c+4,a-c+4,-a+c+4\not=4

because a,b,ca,b,c are distinct. Although the new 33-gon labeled by ab+ca-b+c might be colored by 44, by considering the case of 3-gons in the left of Figure 32 or the mirror image, the 33-gon with the color 44 can be removed. As for the mirror images of the 33-gons in the left of Figure 31, they are also removed by taking the mirror of the same move.

Next we remove 33-gons as depicted in the left of Figure 32 by the move shown in Figure 32 if ab+c4a-b+c\not=4 and Figure 33 if ab+c=4a-b+c=4.

Refer to caption
Figure 32.
Refer to caption
Figure 33.

Here, we note

a+b+4,b+c+4,a+c+44 for Figure 32,\displaystyle-a+b+4,-b+c+4,-a+c+4\neq 4\mbox{ for Figure~{}\ref{app23}},
a+c+4,a+c+1,a+3,2a+14 for Figure 33\displaystyle-a+c+4,a+c+1,-a+3,2a+1\neq 4\mbox{ for Figure~{}\ref{app23-1}}

because a,b,ca,b,c are distinct and a,b,c4a,b,c\neq 4. As for the mirror images of the 33-gons in the left of Figure 32, they are also removed by taking the mirror of the same moves.

Next let us consider the case of 22-gons colored by 44, where we note that such 22-gons are as depicted in the left of Figure 34, the left of Figure 35 or the mirror image for some a,ba,b.

The 22-gons as depicted in the left of Figure 34 can be deleted by the move in Figure 34.

Refer to caption
Figure 34.

Here, note that

a+b+44\displaystyle-a+b+4\not=4

from aba\neq b. As for the 22-gons as depicted in the left of Figure 35, we remove them by the move as shown in Figure 35 if a+2b4-a+2b\not=4 and 2a+3b4-2a+3b\not=4, Figure 36 if a+2b=4-a+2b=4 (and 2a+3b4-2a+3b\not=4), and Figure 37 if 2a+3b=4-2a+3b=4 (and a+2b4-a+2b\not=4).

Refer to caption
Figure 35.
Refer to caption
Figure 36.
Refer to caption
Figure 37.

Here, we note that

a+b+4,ab+4,2a+2b+44 for Figure 35,\displaystyle-a+b+4,a-b+4,-2a+2b+4\neq 4\mbox{ for Figure~{}\ref{app25}},
2a+1,3a+2,4a+34 for Figure 36, and\displaystyle 2a+1,3a+2,4a+3\neq 4\mbox{ for Figure~{}\ref{app26}, and}
3a+2,4a+3,2a+14 for Figure 37\displaystyle 3a+2,4a+3,2a+1\neq 4\mbox{ for Figure~{}\ref{app27}}

because a,b,ca,b,c are distinct and a,b,c4a,b,c\neq 4. As for the mirror images of the 22-gons in the left of Figure 35, they are also removed by taking the mirror of the same moves.

The resultant diagram has no region colored by 44. This completes the proof. ∎

Acknowledgments

The authors wish to express their thanks to Professor Shin Satoh for several helpful comments on Theorem 5.1 and the property mincol5Dehn(K)=4{\rm mincol}^{\rm Dehn}_{5}(K)=4. The second author was supported by JSPS KAKENHI Grant Number 21K03233.

References

  • [1] H. Abchir, M. Elhamdadi and S. Lamsifer, On the minimum number of Fox colorings of knots, Grad. J. Math. 5 (2020), no. 2, 122–137.
  • [2] F. Bento and P. Lopes, The minimum number of Fox colors modulo 13 is 5, Topology Appl. 216 (2017), 85–115.
  • [3] R. H. Crowell and R. H. Fox, Introduction to knot theory, Graduate Texts in Mathematics, No. 57. Springer-Verlag, New York-Heidelberg, 1977. x+182 pp.
  • [4] Y. Han and B. Zhou, The minimum number of coloring of knots, J. Knot Theory Ramifications 31 (2022), no. 2, Paper No. 2250013, 55 pp.
  • [5] F. Harary and L. H. Kauffman, Knots and graphs. I. Arc graphs and colorings, Adv. Appl. Math. 22(3) (1999), 312–337.
  • [6] K. Ichihara and E. Matsudo, A lower bound on minimal number of colors for links, Kobe J. Math. 33 (2016), no. 1-2, 53–60.
  • [7] E. Matsudo, K. Oshiro and G. Yamagishi, Minimum numbers of Dehn colors of knots and symmetric local biquandle cocycle invariants, preprint.
  • [8] T. Nakamura, Y. Nakanishi, and S. Satoh, The pallet graph of a Fox coloring, Yokohama Math. J. 59 (2013), 91–97.
  • [9] T. Nakamura, Y. Nakanishi, and S. Satoh, 11-colored knot diagram with five colors, J. Knot Theory Ramifications 25 (2016), no. 4, Paper No. 1650017, 22 pp.
  • [10] K. Oshiro, Any 7-colorable knot can be colored by four colors, J. Math. Soc. Japan 62(3) (2010) 963–973.
  • [11] S. Satoh, 5-colored knot diagram with four colors, Osaka J. Math. 46 (2009), no. 4, 939–948.