1: \documentclass[11pt]{article}

2: \usepackage{amssymb}

3: \usepackage{geometry}

4: \usepackage{amsmath}

5: \usepackage{amsfonts}

6: \usepackage{graphicx}

7: 

8: %\setlength{\textwidth}{150mm} \setlength{\textheight}{20cm}

9: %\setlength{\parskip}{2mm} \setlength{\headheight}{0cm}

10: %\setlength{\topmargin}{-0.5cm} \setlength{\oddsidemargin}{0.5cm}

11: %\setlength{\evensidemargin}{0.0cm}

12: 

13: \begin{document}

14: \newcommand{\2}{\vspace{0.2 cm}}

15: \newcommand{\dist}{{\rm dist}}

16: \newcommand{\diam}{{\rm diam}}

17: \newcommand{\rad}{{\rm rad}}

18: \newcommand{\dom}{\mbox{$\rightarrow$}}

19: \newcommand{\ndom}{\mbox{$\not\rightarrow$}}

20: \newcommand{\sdom}{\mbox{$\Rightarrow$}}

21: \newcommand{\nsdom}{\mbox{$\not\Rightarrow$}}

22: \newcommand{\qed}{\hfill$\diamond$}

23: \newcommand{\pf}{{\bf Proof: }}

24: \newtheorem{theorem}{Theorem}[section]

25: \newcommand{\ra}{\rangle}

26: \newcommand{\la}{\langle}

27: \newtheorem{lemma}[theorem]{Lemma}

28: \newtheorem{corollary}[theorem]{Corollary}

29: \newtheorem{proposition}[theorem]{Proposition}

30: \newtheorem{conjecture}[theorem]{Conjecture}

31: \newtheorem{problem}[theorem]{Problem}

32: \newtheorem{remark}[theorem]{Remark}

33: \newtheorem{example}[theorem]{Example}

34: \newcommand{\beq}{\begin{equation}}

35: \newcommand{\eeq}{\end{equation}}

36: \newcommand{\argmax}{{\rm argmax}}

37: \newcommand{\MiP}{MinHOMP($H$) }

38: \newcommand{\MaP}{MaxHOMP($H$) }

39: \newcommand{\vecc}[1]{\stackrel{\leftrightarrow}{#1}}

40: 

41: 

42: \title{Minimum Cost and List Homomorphisms to Semicomplete Digraphs}

43: \author{G. Gutin, A. Rafiey and A. Yeo\\

44: Department of Computer Science\\

45: Royal Holloway University of London\\

46: Egham, Surrey TW20 OEX, UK\\

47: {\tt gutin(arash,anders)@cs.rhul.ac.uk}}

48: 

49: \date{}

50: 

51: \maketitle

52: 

53: 

54: 

55: 

56: \begin{abstract}

57: For digraphs $D$ and $H$, a mapping $f:\ V(D)\dom V(H)$ is a {\em

58: homomorphism of $D$ to $H$} if $uv\in A(D)$ implies $f(u)f(v)\in

59: A(H).$ Let $H$ be a fixed directed or undirected graph. The

60: homomorphism problem for $H$ asks whether a directed or undirected

61: graph input digraph $D$ admits a homomorphism to $H.$ The list

62: homomorphism problem for $H$ is a generalization of the

63: homomorphism problem for $H$, where every vertex $x\in V(D)$ is

64: assigned a set $L_x$ of possible colors (vertices of $H$).

65: 

66: The following optimization version of these decision problems was

67: introduced in \cite{gutinDAM}, where it was motivated by a

68: real-world problem in defence logistics. Suppose we are given a

69: pair of digraphs $D,H$ and a positive cost $c_i(u)$ for each $u\in

70: V(D)$ and $i\in V(H)$. The cost of a homomorphism $f$ of $D$ to

71: $H$ is $\sum_{u\in V(D)}c_{f(u)}(u)$. For a fixed digraph $H$, the

72: minimum cost homomorphism problem for $H$, MinHOMP($H$), is stated

73: as follows: For an input digraph $D$ and costs $c_i(u)$ for each

74: $u\in V(D)$ and $i\in V(H)$, verify whether there is a

75: homomorphism of $D$ to $H$ and, if it exists, find such a

76: homomorphism of minimum cost.

77: 

78: We obtain dichotomy classifications of the computational

79: complexity of the list homomorphism problem and MinHOMP($H$), when

80: $H$ is a semicomplete digraph (a digraph in which every two

81: vertices have at least one arc between them). Our dichotomy for

82: the list homomorphism problem coincides with the one obtained by

83: Bang-Jensen, Hell and MacGillivray in 1988 for the homomorphism

84: problem when $H$ is a semicomplete digraph: both problems are

85: polynomial solvable if $H$ has at most one cycle; otherwise, both

86: problems are NP-complete. The dichotomy for \MiP is different: the

87: problem is polynomial time solvable if $H$ is acyclic or $H$ is a

88: cycle of length 2 or 3; otherwise, the problem is NP-hard.

89: \end{abstract}

90: 

91: \section{Introduction}

92: 

93: For excellent introductions to homomorphisms in directed and

94: undirected graphs, see \cite{hell2003,hell2004}. In this paper,

95: directed (undirected) graphs have no parallel arcs (edges) or

96: loops. The vertex (arc) set of a digraph $G$ is denoted by $V(G)$

97: ($A(G)$). The vertex (edge) set of an undirected graph $G$ is

98: denoted by $V(G)$ ($E(G)$). For a digraph $G$, if $xy\in A(G)$, we

99: say that $x$ {\em dominates} $y$ and $y$ is {\em dominated} by

100: $x$. A $k$-{\em cycle}, denoted by $\vec{C}_k$, is a directed

101: simple cycle with $k$ vertices. A digraph is {\em acyclic} if it

102: has no cycle. A digraph $D$ is {\em semicomplete} if, for each

103: pair $x,y$ of distinct vertices either $x$ dominates $y$ or $y$

104: dominates $x$ or both. A {\em tournament} is a semicomplete

105: digraph with no 2-cycle. Semicomplete digraphs and, in particular,

106: tournaments are well-studied in graph theory and algorithms

107: \cite{bang2000}. A digraph $G'$ is the {\em dual} of a digraph $G$

108: if $G'$ is obtained from $G$ by changing orientations of all arcs.

109: 

110: For digraphs $D$ and $H$, a mapping $f:\ V(D)\dom V(H)$ is a {\em

111: homomorphism of $D$ to $H$} if $uv\in A(D)$ implies $f(u)f(v)\in

112: A(H).$ A homomorphism $f$ of $D$ to $H$ is also called an {\em

113: $H$-coloring} of $G$, and $f(x)$ is called  {\em color} of $x$ for

114: every $x\in V(D).$ We denote the set of all homomorphisms from $D$

115: to $H$ by $HOM(D,H)$. Let $H$ be a fixed digraph. The {\em

116: homomorphism problem for} $H$, ${\rm HOMP}(H)$, asks whether there

117: is a homomorphism of an input digraph $D$ to $H$ (i.e., whether

118: $HOM(D,H)\neq \emptyset$). In the {\em list homomorphism problem

119: for} $H$, ${\rm LHOMP}(H)$, we given an input digraph $D$ and a

120: set (called a {\em list}) $L_v\subseteq V(H)$ for each $v\in

121: V(D)$. Our aim is to check whether there is a homomorphism $f\in

122: HOM(D,H)$ such that $f(v)\in L_v$ for each $v\in V(D).$

123: 

124: The problems ${\rm HOMP}(H)$ and ${\rm LHOMP}(H)$ have been

125: studied for several families of directed and undirected graphs

126: $H$, see, e.g., \cite{hell2003,hell2004}. A well-known result of

127: Hell and Ne\v{s}et\v{r}il \cite{hellJCT48} asserts that ${\rm

128: HOMP}(H)$ for undirected graphs is polynomial time solvable if $H$

129: is  bipartite and it is NP-complete, otherwise. Feder, Hell and

130: Huang \cite{federC19} proved that ${\rm LHOMP}(H)$ for undirected

131: graphs is polynomial time solvable if $H$ is a bipartite graph

132: whose complement is a circular arc graph (a graph isomorphic to

133: the intersection graph of arcs on a circle), and ${\rm LHOMP}(H)$

134: is NP-complete, otherwise. Such a dichotomy classification for all

135: digraphs is unknown and only partial classifications have been

136: obtained; see \cite{hell2004}. For example, Bang-Jensen, Hell and

137: MacGillivray \cite{bangSIAMJDM1} showed that ${\rm HOMP}(H)$ for

138: semicomplete digraphs $H$ is polynomial time solvable if $H$ has

139: at most one cycle and ${\rm HOMP}(H)$ is NP-complete, otherwise.

140: Nevertheless, Bulatov \cite{bulatovACMTCL} managed to prove that

141: for each directed graph $H$, ${\rm LHOMP}(H)$ is either polynomial

142: time solvable or NP-complete. The same result for ${\rm HOMP}(H)$

143: is conjectured, see, e.g., \cite{hell2003,hell2004}. If this

144: conjecture holds, it will imply that the well-known Constraint

145: Satisfaction Problem Dichotomy Conjecture of Feder and Vardi also

146: holds \cite{federSIAMJC28}.

147: 

148: The authors of \cite{gutinDAM} introduced an optimization problem

149: on $H$-colorings for undirected graphs $H$, MinHOMP($H$). The

150: problem is motivated by a problem in defence logistics. Suppose we

151: are given a pair of digraphs $D,H$ and a positive cost $c_i(u)$

152: for each $u\in V(D)$ and $i\in V(H)$. The {\em cost} of a

153: homomorphism $f$ of $D$ to $H$ is $\sum_{u\in V(D)}c_{f(u)}(u)$.

154: For a fixed digraph $H$, the {\em minimum cost homomorphism

155: problem} \MiP is stated as follows: For an input digraph $D$ and

156: costs $c_i(u)$ for each $u\in V(D)$ and $i\in V(H)$, verify

157: whether $HOM(D,H)\neq \emptyset$ and, if $HOM(D,H)\neq \emptyset$,

158: find a homomorphism in $HOM(D,H)$ of minimum cost. The problem

159: \MiP generalizes ${\rm LHOMP}(H)$ (and, thus, HOMP($H$)): assign

160: $c_i(u)=1$ if $i\in L_u$ and $c_i(u)=2,$ otherwise.

161: 

162: In this paper, we obtain dichotomy classifications for ${\rm

163: LHOMP}(H)$ and \MiP when $H$ is a semicomplete digraph. Our

164: classification for ${\rm LHOMP}(H)$ coincides with that for ${\rm

165: HOMP}(H)$ \cite{bangSIAMJDM1} described earlier. However, for \MiP

166: the classification is different: the problem is polynomial time

167: solvable when $H$ is either an acyclic tournament or a 2-cycle or

168: a 3-cycle. Otherwise, \MiP is NP-hard. This implies that even when

169: $H$ is a unicyclic semicomplete digraph on at least four vertices,

170: \MiP is NP-hard (unlike HOMP($H$) and LHOMP($H$)).

171: 

172: The {\em maximum cost homomorphism problem} \MaP is the same

173: problem as MinHOMP($H$), but instead of minimization we consider

174: maximization. Let $M$ be a constant larger than any cost $c_i(u)$,

175: $u\in V(D),\ i\in V(H)$. Then the cost $c'_i(u)=M-c_i(u)$ is

176: positive for each $u\in V(D),\ i\in V(H)$. Due to this

177: transformation, the problems \MiP and \MaP are equivalent. Notice

178: that allowing negative or zero costs would not make \MiP and \MaP

179: more difficult: we can easily transform this more general case to

180: the positive costs one by adding a large constant $M'$ to each

181: cost. This transformation does not change optimal solutions.

182: 

183: 

184: 

185: The rest of the paper is organized as follows. In Section

186: \ref{homsec}, we introduce the homomorphic product of digraphs and

187: relate it to a maximum cost homomorphism. We prove that LHOMP($H$)

188: and \MiP are polynomial time solvable when $H$ is an acyclic

189: tournament. The dichotomy classifications LHOMP($H$) and \MiP when

190: $H$ is a semicomplete digraph are proved in Sections \ref{lhclass}

191: and \ref{semsec}, respectively. We conclude the paper by posing

192: some open problems.

193: 

194: \section{Products and Homomorphisms of Digraphs}\label{homsec}

195: 

196: In this section, we describe an approach for proving that \MaP is

197: polynomial time solvable for certain digraphs $H$. Using our

198: approach, we prove that \MaP is polynomial time solvable for

199: acyclic tournaments. To the best of our knowledge this approach,

200: which is of interest also for ${\rm HOM}(H)$ and ${\rm LHOM}(H)$,

201: has not been studied earlier.

202: 

203: For $H$ belonging to a special family $\cal H$ of digraphs, we can

204: transform \MaP into the problem of finding a maximum cost

205: independent set in a special family ${\cal F}({\cal H})$ of

206: undirected graphs. If the last problem is polynomial time solvable

207: (when, for example, ${\cal F}({\cal H})$ consists of perfect

208: graphs, $2P_2$-free graphs, claw-free graphs or graphs of other

209: special classes, see

210: \cite{alekseev1991,alekseevDAM145,balasN19,bertolazziDAM76,grotschelADM21,nakamuraJORSJ44}),

211: then our approach is useful.

212: 

213: The {\em homomorphic product}  of digraphs $D$ and $H$ is an

214: undirected graph $D\otimes H$ defined as follows: $V(D\otimes

215: H)=\{u_i:\ u\in V(D),\ i\in V(H)\}$, $E(D\otimes H)=\{u_iv_j:\

216: uv\in A(D),\ ij\notin A(H)\}\cup \{u_iu_j: u\in V(D), i\neq j\in

217: V(H)\}.$ Let $\mu=\max\{c_j(v):\ v\in V(D),\ j\in V(H)\}.$ We

218: define the cost of $u_i$, $c(u_i)=c_i(u)+\mu |V(D)|.$ For a set

219: $X\subseteq V(D\otimes H)$, we define $c(X)=\sum_{x\in X}c(x).$

220: 

221: 

222: \begin{theorem}\label{maint}

223: Let $D$ and $H$ be digraphs. Then there is a homomorphism of $D$

224: to $H$ if and only if the number of vertices in a largest

225: independent set of $D\otimes H$ equals $|V(D)|$. If $HOM(D,H)\neq

226: \emptyset$, then $h\in HOM(D,H)$ is of maximum cost if and only if

227: $I=\{x_{h(x)}:\ x\in V(D)\}$ is an independent set of maximum

228: cost.

229: \end{theorem}

230: \pf Let $h:\ D \dom H$ be a homomorphism. Consider $I=\{x_{h(x)}:\

231: x\in V(D)\}$. Suppose that $x_{h(x)}y_{h(y)}$ is an edge in

232: $D\otimes H$. Then either $xy\in A(D)$ and $h(x)h(y)\notin A(H)$

233: or $yx\in A(D)$ and $h(y)h(x)\notin A(H)$. Either case contradicts

234: the fact that $h$ is a homomorphism. Thus, $I$ is an independent

235: set in $D\otimes H.$

236: 

237: Observe that each independent set in $D\otimes H$ contains at most

238: one vertex in each set $S_x=\{x_i:\ i\in V(H)\},$ $x\in V(D)$. Let

239: $I=\{x_{f(x)}:\ x\in V(D)\}$ be an independent set in $D\otimes H$

240: with $|V(D)|$ vertices. Consider the mapping $f:\ x\mapsto f(x).$

241: Assume $xy\in A(D).$ Since $I$ is independent, $f(x)f(y)\in A(H)$.

242: Thus, $f\in HOM(D,H).$

243: 

244: Let $HOM(D,H)\neq \emptyset$ and let $n=|V(D)|.$ Let $X$ and $Y$

245: be subsets of $V(D\otimes H)$ and $|X|=|Y|+1\le n.$ Then

246: $$ c(X)-c(Y)\ge |X|n\mu -(|X|-1)(n+1)\mu\ge \mu

247: >0.$$

248: Thus, in particular, every maximum cost independent set of

249: $D\otimes H$ is a largest independent set. Observe that the cost

250: of the homomorphism $f$ defined above equals the cost of vertices

251: in the independent set $I$ minus $n^2\mu,$ which is a constant.

252: Thus, every maximum cost independent set of $D\otimes H$

253: corresponds to a maximum cost homomorphism of $D$ to $H$ and vise

254: versa.\qed

255: 

256: \begin{remark}

257: Observe that Theorem \ref{maint} holds when $H$ has loops.

258: \end{remark}

259: 

260: 

261: \begin{remark}

262: In applications of Theorem \ref{maint}, we may need to replace a

263: pair $D,H$ by another pair $D',H'$ such that $HOM(D,H)=HOM(D',H')$

264: and the costs of the homomorphisms remain the same.

265: \end{remark}

266: 

267: 

268: 

269: Consider the following corollary of Theorem \ref{maint}. A digraph

270: $D$ is {\em transitive} if $xy,yz\in A(D)$ implies $xz\in A(D)$

271: for all pairs $xy,yz$ of arcs in $D.$ A graph is a {\em

272: comparability graph} if it has an orientation, which is

273: transitive. Bang-Jensen, Hell and MacGillivray \cite{bangSIAMJDM1}

274: proved that if $H$ is an acyclic tournament, then ${\rm HOMP}(H)$

275: is polynomial time solvable. We extend this result to \MiP and

276: MaxHOMP($H$).

277: 

278: \begin{theorem}\label{acyclicth}

279: If $H$ is an acyclic tournament, then \MaP and \MiP are polynomial

280: time solvable.

281: \end{theorem}

282: \pf Let $H$ be an acyclic tournament with $V(H)=\{1,2,\ldots,p\}$

283: and $A(H)=\{ij: 1\le i<j\le p\}.$

284: 

285: Observe that $H$ is transitive. Also observe that $HOM(D,H)=

286: \emptyset$ unless $D$ is acyclic. Since we can verify that $D$ is

287: acyclic in polynomial time (for example, by deleting vertices of

288: indegree 0), we may assume that $D$ is acyclic. Since $H$ is

289: transitive, we have $HOM(D,H)=HOM(D^+,H)$, where $D^+$ is the

290: transitive closure of $D$, i.e., if there is a path from $x$ to

291: $y$ in $D$, then $xy\in D^+.$ One can find the transitive closure

292: of a digraph in polynomial time using DFS or BFS \cite{bang2000},

293: so we may assume that $D$ is transitive.

294: 

295: Let $G=D\otimes H$. Let $G'$ be an orientation of $G$ such that

296: $$A(G')=\{x_iy_j:\ j\leq i, xy\in A(D)\}\cup \{x_ix_j:\ x\in

297: V(D),\ j<i\}.$$ We will prove that $G'$ is a transitive digraph.

298: Let $x_iy_j,y_jz_k\in A(G')$. Observe that $i\ge j\ge k$ and

299: consider three cases covering all possibilities.

300: 

301: \2

302: 

303: {\bf Case 1:} $x=y=z$. Then $x_ix_j,x_jx_k\in A(G')$ and, thus,

304: $i>j>k$ and $x_iz_k=x_ix_k \in A(G').$

305: 

306: \2

307: 

308: {\bf Case 2:}  $x=y=z$ does not hold, but not all vertices $x,y,z$

309: are distinct. Without loss of generality, assume that $x=y\neq z.$

310: Then $x_ix_j,x_jz_k\in A(G')$ and, thus, $i>k$ and $x_iz_k\in

311: A(G').$

312: 

313: \2

314: 

315: {\bf Case 3:} $x,y,z$ are all distinct. Then $xy,yz\in A(D^+)$

316: and, thus, $xz\in A(D^+)$. Since $i\ge k$, we conclude that

317: $x_iz_k\in A(G').$

318: 

319: \2

320: 

321: So, we have proved that $G$ is a comparability graph and, thus, it

322: is perfect. Therefore, a maximum cost independent set in $D\otimes

323: H$ can be found in polynomial time \cite{grotschelADM21}. It

324: remains to apply Theorem \ref{maint}. If $D\otimes H$ has an

325: independent set with $|V(D)|$ vertices, $HOM(D,H)\neq \emptyset$

326: and a maximum cost independent set corresponds to a maximum cost

327: $H$-coloring.\qed

328: 

329: \begin{corollary}

330: If $H$ is an acyclic tournament, then LHOMP($H$) is polynomial

331: time solvable.

332: \end{corollary}

333: 

334: 

335: 

336: \section{Dichotomy for LHOMP($H$)}\label{lhclass}

337: 

338: Recall that $\vec{C}_k$ denotes a directed cycle on $k$ vertices,

339: $k\ge 2$; let $V(\vec{C}_k)=\{1,2,\ldots,k\}.$  One can check

340: whether $HOM(D,\vec{C}_k)\neq \emptyset$ using the following

341: algorithm $\cal A$ from Section 1.4 of \cite{hell2004}. First, we

342: may assume that $D$ is connected (i.e., its underlying undirected

343: graph is connected) as otherwise $\cal A$ can be applied to each

344: component of $D$ separately. Choose a vertex $x$ of $D$ and assign

345: it color 1. Assign every out-neighbor of $x$ color 2 and each

346: in-neighbor of $x$ color $k$. For every vertex $y$ with color $i$,

347: we assign every out-neighbor of $y$ color $i+1$ modulo $k$ and

348: every in-neighbor of $y$ color $i-1$ modulo $k$. We have

349: $HOM(D,\vec{C}_k)\neq \emptyset$ if and only if no vertex is

350: assigned different colors.

351: 

352: 

353: 

354: M. Green \cite{green} was the first to prove Theorem \ref{P3}, but

355: his proof uses polymorphisms (for the definition and results on

356: polymorphisms, see, e.g., \cite{bulatovACMTCL}). Our proof below

357: is elementary and does not require polymorphisms.

358: 

359: \begin{theorem}\label{P3}

360: Let $H$ be a semicomplete digraph with a unique cycle, then ${\rm

361: LHOMP}(H)$ is polynomial time solvable.

362: \end{theorem}

363: \pf It is well-known \cite{bang2000} that a semicomplete digraph

364: with a unique cycle contains a cycle with two or three vertices.

365: Assume that $H$ has a cycle with three vertices (the case of

366: 2-cycle can be treated similarly). Let the vertex set of $H$ be

367: $\{1,2,\ldots,p\}$ and $A(H)=\{ij:\ i<j, (i,j)\neq (a,b)\}\cup

368: \{ba\}$, where $b=a+2.$

369: 

370: We use the following recursive procedure. If $V(D)= \{v\}$ and

371: $L_v \neq \emptyset $, then the solution is trivial. Now suppose

372: that $|V(D)| \ge 2$ and consider the following two properties of a

373: vertex $x$ in $D$:

374: 

375: \begin{itemize}

376: \item[(a)] $x$ has in-degree zero, and $L_x$ has an element

377: smaller than $a$. \item[(b)] $x$ has out-degree zero, and $L_x$

378: has an element greater than

379:   $a+2$.

380: \end{itemize}

381: 

382: 

383: If there is a vertex $x$ with property $(a)$, then define

384: $f(x)=i$, where $i$ is the minimum number in $L_x$, and delete all

385: $j \le i$ from the lists of all out-neighbors of $x$. Run the

386: procedure for $D-x$ with changed lists.  If there is a vertex $x$

387: with property $(b)$, then define $f(x)=i$, where $i$ is the

388: maximum number in $L_x$, and delete all $j \ge i$ from the lists

389: of all in-neighbors of $x$. Run the procedure for $D-x$ with

390: changed lists.

391: 

392: If no vertex with either property exists, then run the algorithm

393: $\cal A$ described in the beginning of this section to find all

394: homomorphisms from $HOM(D,\vec{C}_3),$ where $\vec{C}_3$ has

395: vertices $a,a+1,a+2$. If $HOM(D,\vec{C}_3)\neq \emptyset$, there

396: there are three homomorphism, and it suffices to verify that at

397: least one of them is compatible with the lists.

398: 

399: Clearly, if our procedure succeeds, then we have found a required

400: homomorphism. It remains to see that if the procedure fails, then

401: no required homomorphism exists. This is equivalent to proving

402: that, after all vertices satisfying (a) or (b) have been deleted,

403: every remaining vertex $y$ must have color $a,a+1$ or $a+2.$

404: 

405: Let $D'$ be obtained from $D$ by deleting all vertices satisfying

406: (a) or (b) and let $y\in V(D').$ We prove that $y$ must be colored

407: $a,a+1$ or $a+2.$ Assume that $y$ is in a directed cycle $C$ of

408: $D'$. Observe that any homomorphism of $D$ to $H$ maps $C$ into a

409: directed walk. Thus, $y$ can be colored $a,a+1$ or $a+2$ only.

410: Assume that $y$ is isolated in $D'$. Since $y$ does not satisfy

411: (a) or (b), its list contains only $a$, $a+1$ or $a+2.$ Now

412: consider the case when $y$ is not isolated and it is not in any

413: cycle of $D'.$ Let $P$ be a path in $D'$ containing $y$ such that

414: the initial vertex of $P$ is either in a cycle of $D'$ or its

415: in-degree in $D'$ is zero, and the terminal vertex of $P$ is

416: either in a cycle of $D'$ or its out-degree in $D'$ is zero.

417: Observe that, by the arguments above, the initial vertex of $P$

418: must have color $a$ or larger and the terminal vertex of $P$ must

419: have color $a+2$ or smaller. This implies that every vertex of $P$

420: must have color $a$, $a+1$ or $a+2.$ \qed

421: 

422: \2

423: 

424: Recall that HOMP($H$) is NP-complete when $H$ is a semicomplete

425: digraph with at least two cycles. This result and Theorems

426: \ref{acyclicth} and \ref{P3} imply the following:

427: 

428: \begin{theorem}

429: Let $H$ be a semicomplete digraph. Then LHOMP($H$) is polynomial

430: time solvable if $H$ has at most one cycle, and LHOMP($H$) is

431: NP-complete, otherwise.

432: \end{theorem}

433: 

434: \section{Classification for \MiP and \MaP}\label{semsec}

435: 

436: To solve \MiP for $H=\vec{C}_k$, choose an initial vertex $x$ in

437: each component $D'$ of $D$ (a component of its underlying

438: undirected graph). Using the algorithm $\cal A$ from the previous

439: section, we can check whether each $D'$ admits an

440: $\vec{C}_k$-coloring. If the coloring of $D'$ exists, we compute

441: the cost of this coloring and compute the costs of the other $k-1$

442: $\vec{C}_k$-colorings when $x$ is colored $2,3,\ldots,k,$

443: respectively. Thus, we can find a minimum cost homomorphism in

444: $HOM(D',\vec{C}_k).$  Thus, in polynomial time, we can obtain a

445: $\vec{C}_k$-coloring of the whole digraph $D$ of minimum cost. In

446: other words, we have the following:

447: 

448: \begin{lemma}\label{cycle}

449: For $H=\vec{C}_k$, \MiP and \MaP are polynomial time solvable.

450: \end{lemma}

451: 

452: 

453: 

454: Addition of an extra vertex to a cycle may well change the

455: complexity of \MaP and MinHOMP($H$).

456: 

457: \begin{lemma}\label{cycle+}

458: Let $H'$ be a digraph obtained from $\vec{C}_k$, $k\ge 2$, by

459: adding an extra vertex dominated by the vertices of the cycle, and

460: let $H$ be $H'$ or its dual. Then \MiP and \MaP are NP-hard.

461: \end{lemma}

462: \pf Without loss of generality we may assume that $H=H'$ and that

463: $V(H)=\{1,2,3,\ldots , k, k+1\}$, $123 \ldots k1$ is a $k$-cycle,

464: and the vertex $k+1$ is dominated by the vertices of the cycle.

465: 

466: We will reduce the maximum independent set problem to

467: MinHOMP($H$). Let $G$ be a graph. Construct a digraph $D$ as

468: follows:

469: 

470: 

471: $$V(D)=V(G) \cup \{v_i^e :\ e\in E(G) \ i\in V(H) \},\

472: A(D)=A_1\cup A_2, \mbox{ where}$$

473: $$A_1=\{v_1^e v_2^e, v_2^ev_3^e, \ldots v_{k-1}^ev_k^e,v_k^ev_1^e :\ e\in

474: E(G)\}$$ and

475: $$A_2=\{v_1^{uv}u, v_{k+1}^{uv}u, v_2^{uv}v, v_{k+1}^{uv}v :\ uv\in E(G)\}.$$

476: 

477: 

478: Let all costs $c_i(t)=1$ for $t\in V(D)$ apart from $c_{k+1}(p)=2$

479: for all $p\in V(G).$

480: 

481: 

482: Consider a minimum cost homomorphism $f\in HOM(D,H)$. By the

483: choice of the costs, $f$ assigns the maximum possible number of

484: vertices of $G$ (in $D$) a color different from $k+1$. However, if

485: $pq$ is an edge in $G$, by the definition of $D$, $f$ cannot

486: assign colors different from $k+1$ to both $p$ and $q$. Indeed, if

487: both $p$ and $q$ are assigned colors different from $k+1$, then

488: the existence of $v_{k+1}^{pq}$ implies that they are assigned the

489: same color, which however is impossible by the existence of

490: $\{v_i^{pq} :\  i\in \{1,2,\ldots ,k\} \}$. Observe that $f$ may

491: assign exactly one of the vertices $p,q$ color $k+1$ and the other

492: a color different from $k+1$. Also $f$ may assign both of them

493: color $k+1$. Thus, a minimum cost $H$-coloring of $D$ corresponds

494: to a maximum independent set in $G$ and vise versa (the vertices

495: of a maximum independent set are assigned color $2$ and all other

496: vertices in $V(G)$ are assigned color $k+1$). \qed \2

497: 

498: Interestingly, the problem ${\rm HOMP}(H')$  for $H'$ (especially,

499: with $k=3$) defined in Lemma \ref{cycle+} is well known to be

500: polynomial time solvable (see, e.g.,

501: \cite{bangSIAMJDM1,gutjahr1991,hell2004}). The following lemma

502: allows us to prove that \MaP and \MiP are NP-hard when

503: MaxHOMP($H'$) and MinHOMP($H'$) are NP-hard for an induced

504: subdigraph $H'$ of $H.$

505: 

506: \begin{lemma}\label{reductionl}

507: Let $H'$ be an induced subdigraph of a digraph $H$. If

508: MaxHOMP($H'$) is NP-hard, then \MaP is also NP-hard.

509: \end{lemma}

510: \pf Let $D$ be an input digraph with $n$ vertices and let $c_i(u)$

511: be the costs, $u\in V(D)$, $i\in V(H')$. Let all costs $c_i(u)$ be

512: bounded from above by $\beta(n)$. For each $i\in V(H)-V(H')$ and

513: each $u\in V(D)$, set costs $c_i(u):=n\beta(n)+1.$ Observe that

514: there is an $H$-coloring of $D$ of cost at most $n\beta(n)$ if and

515: only if $HOM(D,H')\neq \emptyset$ and if $HOM(D,H')\neq

516: \emptyset$, then the cost of minimum cost $H$-coloring equals to

517: that of minimum cost $H'$-coloring.\qed

518: 

519: 

520: \2

521: 

522: As a corollary of Theorem \ref{acyclicth} and Lemmas \ref{cycle},

523: \ref{cycle+} and \ref{reductionl}, we obtain the following

524: theorem.

525: 

526: \begin{theorem}\label{classmd}

527: For a semicomplete digraph $H$, \MiP and \MaP are polynomial time

528: solvable if $H$ is acyclic or $H=\vec{C}_k$ for $k=2$ or 3, and

529: NP-hard, otherwise.

530: \end{theorem}

531: \pf By Theorem \ref{acyclicth} and since ${\rm HOMP}(H)$ is

532: NP-complete when a semicomplete digraph $H$ has at least two

533: cycles \cite{bangSIAMJDM1}, we may restrict ourselves to the case

534: when $H$ has a unique cycle. Observe that this cycle has two or

535: three vertices. If no other vertices are in $H$, \MaP and \MiP are

536: polynomial time solvable by Lemma \ref{cycle}. Assume that $H$ has

537: a vertex $i$ not in the cycle. Observe that $i$ is dominated by or

538: dominates all vertices of the cycle, i.e., $H$ contains, as an

539: induced subdigraph one of the digraphs of Lemma \ref{cycle+}. So,

540: we are done by Lemmas \ref{cycle+} and \ref{reductionl}.\qed

541: 

542: \section{Discussions}

543: 

544: In this paper we have obtained dichotomy classifications for the

545: list and minimum cost $H$-coloring problems when $H$ is a

546: semicomplete digraph.

547: 

548: It would be interesting to find out whether there exists a

549: dichotomy classification for the minimum cost $H$-coloring problem

550: (for an arbitrary digraph $H$) and if it does exist, to obtain

551: such a classification. Since these problems seem to be far from

552: trivial, one could concentrate on establishing dichotomy

553: classifications for special classes of digraphs such as

554: semicomplete multipartite digraphs (digraphs obtained from

555: complete multipartite graphs by replacing every edge with an arc

556: or the pair of mutually opposite arcs) and oriented paths (it was

557: proved in \cite{gutjahrDAM35} that the homomorphism problem for

558: oriented paths is polynomial time solvable).

559: 

560: We have obtained some partial results on MinHOMP($H$) for

561: semicomplete multipartite digraphs $H$. To find a complete

562: dichotomy for the case of semicomplete bipartite digraphs, one

563: would need, among other things, to solve an open problem from

564: \cite{gutinDAM}: establish a dichotomy classification for the

565: complexity of \MiP when $H$ is a bipartite (undirected) graph.

566: Indeed, let $B$ be a semicomplete bipartite digraph with partite

567: sets $U,V$ and arc set $A(B)=A_1\cup A_2$, where $A_1=U\times V$

568: and $A_2\subseteq V\times U$. Let $B'$ be a bipartite graph with

569: partite sets $U,V$ and edge set $E(B')=\{uv:\ vu\in A_2\}.$

570: Observe that MinHOMP($B$) is equivalent to MinHOMP($B'$).

571: 

572: It was proved in \cite{gutinDAM} that \MiP is polynomial time

573: solvable when $H$ is a bipartite graph whose complement is an

574: interval graph. It follows from the main result of \cite{federC19}

575: that \MiP is NP-hard when $H$ is a bipartite graph whose

576: complement is not a circular arc graph. This leaves the obvious

577: gap in the classification for \MiP when $H$ is a bipartite graph.

578: 

579: \2

580: 

581: \2

582: 

583: 

584: {\bf Acknowledgement} We are thankful to Dave Cohen, Mike Green,

585: Michael Krivelevich, Alek Vainshtein and others for useful

586: discussions on the topic of the paper.

587: 

588: {\small

589: 

590: \begin{thebibliography}{99}

591: %%%%%%%%%%%%%%%%%%%%%%%%%%

592: 

593: \bibitem{alekseev1991} V.E. Alekseev, On the number of maximal

594: independent sets in graphs from hereditary classes. In: {\em

595: Combinatorial-algebraic methods in discrete optimization}, Univ.

596: of Nizhny Novgorod (1991) 5-8 [in Russian].

597: 

598: \bibitem{alekseevDAM145} V.E. Alekseev and V.V.

599: Lozin, Augmenting graphs for independent sets. {\em Discrete Appl.

600: Math.} 145 (2004) 3-10.

601: 

602: \bibitem{balasN19} E. Balas and Ch.Y. Yu, On graphs with

603: polynomially solvable maximum-weight clique problem. {\em

604: Networks} 19 (1989) 247-253.

605: 

606: \bibitem{bang2000} J. Bang-Jensen and G. Gutin, {\em Digraphs: Theory,

607: Algorithms and Applications}, Springer-Verlag, London, 2000.

608: 

609: \bibitem{bangSIAMJDM1} J. Bang-Jensen, P. Hell and G. MacGillivray,

610: The complexity of colouring by semicomplete digraphs. {\em SIAM J.

611: Discrete Math.} 1 (1988) 281-298.

612: 

613: \bibitem{bertolazziDAM76} P. Bertolazzi, C. De Simone and A.

614: Galuccio, A nice class for vertex packing problem. {\em Discrete

615: Appl. Math.} 76 (1997) 3-19.

616: 

617: \bibitem{bulatovACMTCL} A.A. Bulatov, Tractable conservative

618: constraint satisfaction problems. To appear in {\em ACM Trans.

619: Comput. Logic}.

620: 

621: \bibitem{federC19} T. Feder, P. Hell and J. Huang. List homomorphisms

622: and circular arc graphs. {\em Combinatorica} 19 (1999) 487-505.

623: 

624: \bibitem{federSIAMJC28} T. Feder and M.Y. Vardi, The computational

625: structure of monotone monadic SNP and constraint satisfaction: a

626: study through Datalog and group theory. {\em SIAM J. Comput.} 28

627: (1998) 57-104.

628: 

629: \bibitem{green} Mike Green, Private communication, June, 2005.

630: %\bibitem{chudnovsky} M. Chudnovsky, N. Robertson, P. Seymour and R.

631: %Thomas, The Strong Perfect Graph Theorem. Submitted, see {\tt

632: %http://arxiv.org/abs/math/0212070}.

633: 

634: \bibitem{grotschelADM21} M. Gr\"otschel, L. Lov\'asz, and A. Schrijver,

635: Polynomial algorithms for perfect graphs. {\em Ann. Discrete

636: Math.} 21 (1984), 325--356.

637: 

638: \bibitem{gutinDAM} G. Gutin, A. Rafiey, A. Yeo and M. Tso, Level of repair

639: analysis and minimum cost homomorphisms of graphs. To appear in

640: {\em Discrete Appl. Math.}

641: 

642: \bibitem{gutjahr1991} W. Gutjahr, Graph colourings, PhD Thesis, Free

643: University Berlin, 1991.

644: 

645: \bibitem{gutjahrDAM35} W. Gutjahr, E. Welzl and G. Woeginger,

646: Polynomial graph-colorings. {\em Discrete Appl. Math.} 35 (1992)

647: 29-45.

648: 

649: \bibitem{hell2003} P. Hell,

650: Algorithmic aspects of graph homomorphisms, in `Survey in

651: Combinatorics 2003', London Math. Soc. Lecture Note Series 307,

652: Cambridge University Press, 2003, 239 - 276.

653: 

654: 

655: \bibitem{hellJCT48} P. Hell and J. Ne\v{s}et\v{r}il, On the complexity of

656: $H$-colouring. {\em J. Combin. Theory B} 48 (1990) 92-110.

657: 

658: \bibitem{hell2004} P. Hell and J. Ne\v{s}et\v{r}il, {\em Graphs and

659: Homomorphisms.} Oxford University Press, Oxford, 2004.

660: 

661: \bibitem{nakamuraJORSJ44} D. Nakamura and A. Tamura, A revision of

662: Minty's algorithm for finding a maximum weight stable set of a

663: claw-free graph. {\em J. Oper. Res. Soc. Japan} 44 (2001) 194-204.

664: 

665: %\bibitem{imrich2000} W. Imrich and S. Klavzar. {\em Product Graphs:

666: %Structure and Recognition.} Wiley,

667: %2000.

668: 

669: %\bibitem{west1996} D. West, {\em Introduction to Graph Theory}, Prentice

670: %Hall, Upper Saddle River, N.J., 1996.

671: 

672: \end{thebibliography}

673: }

674: 

675: \end{document}

676: