1: \documentclass{article}

2: \emergencystretch=2em

3: 

4: \usepackage{times}

5: \usepackage{eepic,epic}

6: \usepackage{latexsym}

7: \usepackage[latin1]{inputenc}    

8: 

9: %

10: %

11: %

12: %

13: %

14: %

15: %

16: 

17: %

18: 

19: %

20: \makeatletter%

21: \def\nottoobig#1{{\hbox{$\left#1\vcenter to1.111\ht\strutbox{}\right.\n@space$}}}

22: \makeatother%

23: 

24: %

25: %

26: %

27: %

28: %

29: %

30: \makeatother%

31: 

32: %

33: %

34: %

35: %

36: 

37: %

38: 

39: 

40: 

41: %

42: %

43: %

44: %

45: %

46: %

47: %

48: %

49: \makeatletter \@beginparpenalty=10000 \makeatother

50: %

51: %

52: %

53: %

54: %

55: %

56: %

57: %

58: %

59: %

60: %

61: %

62: %

63: %

64: %

65: %

66: %

67: %

68: 

69: 

70: %

71: %

72: 

73: \def\underl#1 {\leavevmode\let\first=\relax\underli #1 }

74: \def\underli#1 {\ifx&#1\let\next=\relax\unskip

75:                 \else\let\next=\underli\first\ulinebox{#1}\fi\let\first=\undersp\next}

76: \def\undersp{\penalty50\ulinebox{\space}\penalty50}

77: \def\ulinebox#1{\vtop{\hbox{\strut#1}\hrule}}%

78: \def\unice#1 {\underl #1 & }

79: %

80: %

81: %

82: %

83: %

84: %

85: %

86: %

87: %

88: %

89: %

90: %

91: %

92: %

93: %

94: %

95: %

96: %

97: %

98: 

99: 

100: 

101: %

102: %

103: %

104: %

105: %

106: %

107: %

108: \def\desclabel#1{\bf #1\hfil}

109: \def\desc{\list{}{%

110: \labelwidth=\leftmargin

111: \advance \labelwidth by -\labelsep

112: \let \makelabel=\desclabel}}

113: \let\enddesc=\endlist

114: 

115: 

116: 

117: 

118: %

119: 

120: \makeatletter %

121: \renewcommand{\theequation}{\thesection.\arabic{equation}}

122: %

123: %

124: %

125: \newcommand{\band}{\bigwedge}

126: \newcommand{\Band}[3]{(\bigwedge#1\!\!:\,#2\!\!:\,#3)}

127: \newcommand{\bor}{\bigvee}

128: \newcommand{\Bor}[3]{(\bigvee#1\!\!:\,#2\!\!:\,#3)}

129: \newcommand{\Forall}[3]{(\forall #1\!\!:\,#2\!\!:\,#3)}

130: \newcommand{\Exists}[3]{(\exists #1\!\!:\,#2\!\!:\,#3)}

131: \newcommand{\Union}[3]{(\bigcup #1\!\!:\,#2\!\!:\,#3)}

132: %

133: %

134: %

135: %

136: %

137: %

138: \def\union{\,\bigcup\limits\,}

139: \def\inter{\,\bigcap\limits\,}

140: \newcommand{\true}{\mbox{\it true}}

141: \newcommand{\false}{\mbox{\it false}}

142: \newcommand{\SUM}[3]{ (\sum #1 \!\! : \, #2 \!\!:\, #3) }

143: \newcommand{\IFS}{\mbox{\bf if}}

144: \newcommand{\IF}[1]{ \mbox{\bf if} \, #1 \, \rightarrow \, }

145: \newcommand{\GC}[2]{ #1 \, \rightarrow \, #2 }

146: \newlength{\filength}

147: \settowidth{\filength}{\mbox{\bf f{}i}}

148: \newsavebox{\gcbox}

149: \sbox{\gcbox}{\framebox[\filength]{\rule{0ex}{2ex}}}

150: \newcommand{\BB}[1]{\usebox{\gcbox}\; #1 \, \rightarrow \, }

151: \newcommand{\FI}{\; \mbox{\bf f{}i}}

152: \newcommand{\Skip}{ \mbox{\bf skip} }

153: \newcommand{\DOS}{\mbox{\bf do}}

154: \newcommand{\DO}[1]{\mbox{\bf do} \, #1 \, \rightarrow \,}

155: \newcommand{\OD}{\mbox{\bf od}}

156: \newcommand{\cobegin}{{\bf cobegin}\,}

157: %

158: \newcommand{\coend}{\,{\bf coend}}

159: \newcommand{\Set}[1]{ \hbox{\bf\{} #1 \hbox{\bf\}}}

160: \newcommand{\Bag}[1]{ \{\!| #1  |\!\}}

161: %

162: %

163: %

164: %

165: %

166: %

167: %

168: \newlength{\leftjustindent}

169: \newlength{\@leftjustindent}

170: \setlength{\@leftjustindent}{\leftmargin}

171: \def\leftjust{\let\\\@leftjustcr\let\end\@endleftjust

172:   \addtolength{\@leftjustindent}{\leftjustindent}

173:   \vcenter\bgroup

174:   \halign\bgroup

175:     \hbox to\displaywidth{

176:       \rule{\@leftjustindent}{0ex}$\displaystyle##$\hfill

177:       }\crcr

178: }

179: \def\endleftjust{\crcr\egroup\egroup\endgroup}

180: \def\@endleftjust#1{\crcr\egroup\egroup\@checkend{#1}\endgroup}

181: \def\@leftjustcr{\crcr}

182: 

183: \newcommand{\hoare}[3]{\{{#1}\}\:{#2}\:\{{#3}\}}

184: \renewcommand{\wp}[2]{ {\it wp}({#1},{#2})}

185: \newcommand{\assert}[1]{\!\{#1\}}

186: \newcommand{\atom}[1]{\langle\,{#1}\,\rangle}

187: \newcommand{\lbl}[1]{{#1 \!:\;\,}}

188: \newcommand{\pre}[1]{ {\it pre}({#1})}

189: \newcommand{\post}[1]{ {\it post}({#1}) }

190: \newcommand{\NI}[2]{ {\it NI}({#1},{#2}) }

191: \newcommand{\equi}[3]{ {  {\rm E}_{#1}^{#2}({#3}) }    }

192: \newcommand{\red}[3]{ {  {\rm R}_{#1}^{#2}({#3}) }    }

193: \newcommand{\sparse}{{{\rm SPARSE}}}

194: \newcommand{\tally}{{{\rm TALLY}}}

195: %

196: \newtheorem{theorem}{Theorem}%

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

198: %

199: %

200: %

201: %

202: %

203: %

204: %

205: %

206: %

207: 

208: \newenvironment{proofs}{\noindent{\bf Proof.}\hspace*{1em}}{\qed\bigskip}

209: \newenvironment{proofsketch}{\noindent{\bf Proof Sketch.}\hspace*{1em}}{\qed\bigskip}

210: \newcommand{\sproofof}[1]{\noindent{\bf Proof of {#1}.}\hspace*{1em}}

211: \newcommand{\eproofof}[1]{\hfill \mbox{\qed$_{\mbox{\small {#1}}}$}\quad\medskip}

212: 

213: \newcommand{\qedblob}{\mbox{\rule[-1.5pt]{5pt}{10.5pt}}}

214: \def\literalqed{{\ \nolinebreak\hfill\mbox{\qedblob\quad}}}

215: \def\qedcareful{\literalqed}

216: \def\qed{\literalqed}

217: \def\trueloveqed{{\ \nolinebreak\hfill\mbox{\boldmath

218: \Huge$ \Box$}\nolinebreak\mbox{$\!\!\!\!\!\!

219: {}^{\normalsize\heartsuit}$}}}

220: %

221: %

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

223: \newtheorem{observation}[theorem]{Observation}

224: \newtheorem{fact}[theorem]{Fact}

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

226: \newcommand{\singlespacing}{\let\CS=

227: \@currsize\renewcommand{\baselinestretch}{1}\tiny\CS}

228: \newcommand{\singlespacingplus}{\let\CS=

229: \@currsize\renewcommand{\baselinestretch}{1.25}\tiny\CS}

230: \newcommand{\doublespacing}{\let\CS=

231: \@currsize\renewcommand{\baselinestretch}{1.75}\tiny\CS}

232: \newcommand{\draftspacing}{\let\CS=

233: \@currsize\renewcommand{\baselinestretch}{2.0}\tiny\CS}

234: \newcommand{\foospacing}{\let\CS=

235: \@currsize\renewcommand{\baselinestretch}{1.05}\tiny\CS}

236: \newcommand{\normalspacing}{\singlespacing}

237: \makeatother%

238: 

239: 

240: %

241: \hyphenation{theory theoretical area areas theorem theorems par-allel par-allelize par-allelized threshold Hemaspaan-dra}

242: 

243: %

244: 

245: %

246: %

247: %

248: \def\grantnumber{This work was supported, in part,

249: by NSF grant no.\ DCR-8320274 and Office of Naval

250: Research contract N00014-86-K-0092.}

251: %

252: %

253: %

254: %

255: \mathcode`\0="0030      %

256: \mathcode`\1="0031

257: \mathcode`\2="0032

258: \mathcode`\3="0033

259: \mathcode`\4="0034

260: \mathcode`\5="0035

261: \mathcode`\6="0036

262: \mathcode`\7="0037

263: \mathcode`\8="0038

264: \mathcode`\9="0039

265: %

266: %

267: %

268: %

269: %

270: %

271: 

272: 

273: \newtheorem{definition}[theorem]{Definition}

274: \newtheorem{notation}[theorem]{Notation}

275: \newtheorem{trick}[theorem]{Trick Result}

276: 

277: 

278: %

279: %

280: %

281: %

282: %

283: %

284: 

285: %

286: 

287: %

288: \flushbottom{}

289: %

290: \makeatletter

291: \clubpenalty=\@highpenalty

292: \widowpenalty=\@highpenalty

293: \makeatother

294: 

295: %

296: \let\BLS=\baselinestretch

297: 

298: %

299: %

300: \makeatletter

301: \newcommand{\niceonespacing}{\let\CS=\@currsize\renewcommand{\baselinestretch}{1.1}\tiny\CS}\newcommand{\nicetwospacing}{\let\CS=\@currsize\renewcommand{\baselinestretch}{1.2}\tiny\CS}

302: \newcommand{\nicethreespacing}{\let\CS=\@currsize\renewcommand{\baselinestretch}{1.3}\tiny\CS}

303: \newcommand{\singlespacingplusplus}{\let\CS=\@currsize\renewcommand{\baselinestretch}{1.35}\tiny\CS}

304: \newcommand{\nicefourspacing}{\let\CS=\@currsize\renewcommand{\baselinestretch}{1.4}\tiny\CS}

305: \newcommand{\nicefivespacing}{\let\CS=\@currsize\renewcommand{\baselinestretch}{1.5}\tiny\CS}

306: \newcommand{\nicesixpacing}{\let\CS=\@currsize\renewcommand{\baselinestretch}{1.6}\tiny\CS}

307: \makeatother

308: 

309: %

310: \makeatletter

311: \def\@cite#1#2{[#1\if@tempswa , #2\fi]}

312: \makeatother

313: 

314: %

315: %

316: %

317: \makeatletter

318: \def\@citex[#1]#2{\if@filesw\immediate\write\@auxout{\string\citation{#2}}\fi

319:   \def\@citea{}\@cite{\@for\@citeb:=#2\do

320:     {\@citea\def\@citea{,\linebreak[0]}\@ifundefined

321:        {b@\@citeb}{{\bf ?}\@warning

322:        {Citation `\@citeb' on page \thepage \space undefined}}%

323: \hbox{\csname b@\@citeb\endcsname}}}{#1}}

324: \makeatother

325: 

326: 

327: %

328: \makeatletter

329: \def\ps@thesis{\def\@oddhead{\hfil\rm\thepage\hfil}\def\@oddfoot{}\def\@evenhead{\hfil\rm\thepage\hfil}\def\@evenfoot{}\def\chaptermark##1{}\def\sectionmark##1{}}

330: \makeatother

331: 

332: \makeatletter

333: %

334: %

335: %

336: %

337: \def\foobarpt{\textfont\z@\tenrm 

338:   \scriptfont\z@\ninrm \scriptscriptfont\z@\sevrm

339: \textfont\@ne\tenmi \scriptfont\@ne\ninmi \scriptscriptfont\@ne\sevmi

340: \textfont\tw@\tensy \scriptfont\tw@\ninsy \scriptscriptfont\tw@\sevsy

341: \textfont\thr@@\tenex \scriptfont\thr@@\tenex \scriptscriptfont\thr@@\tenex

342: \def\unboldmath{\everymath{}\everydisplay{}\@nomath\unboldmath

343:           \textfont\@ne\tenmi 

344:           \textfont\tw@\tensy \textfont\lyfam\tenly

345:           \@boldfalse}\@boldfalse

346: \def\boldmath{\@ifundefined{tenmib}{\global\font\tenmib\@mbi\@magscale1\global

347:         \font\tensyb\@mbsy \@magscale1\global\font

348:          \tenlyb\@lasyb\@magscale1\relax\@addfontinfo\@xiipt

349:               {\def\boldmath{\everymath

350:                 {\mit}\everydisplay{\mit}\@prtct\@nomathbold

351:                 \textfont\@ne\tenmib \textfont\tw@\tensyb 

352:                 \textfont\lyfam\tenlyb\@prtct\@boldtrue}}}{}\@xiipt\boldmath}%

353: \def\prm{\fam\z@\tenrm}%

354: \def\pit{\fam\itfam\tenit}\textfont\itfam\tenit \scriptfont\itfam\ninit

355:    \scriptscriptfont\itfam\sevit

356: \def\psl{\fam\slfam\tensl}\textfont\slfam\tensl 

357:      \scriptfont\slfam\tensl \scriptscriptfont\slfam\tensl

358: \def\pbf{\fam\bffam\tenbf}\textfont\bffam\tenbf 

359:    \scriptfont\bffam\ninbf \scriptscriptfont\bffam\ninbf 

360: \def\ptt{\fam\ttfam\tentt}\textfont\ttfam\tentt

361:    \scriptfont\ttfam\nintt \scriptscriptfont\ttfam\nintt 

362: \def\psf{\fam\sffam\tensf}\textfont\sffam\tensf

363:     \scriptfont\sffam\tensf \scriptscriptfont\sffam\tensf

364: \def\psc{\@getfont\psc\scfam\@xiipt{\@mcsc\@magscale1}}%

365: \def\ly{\fam\lyfam\tenly}\textfont\lyfam\tenly 

366:    \scriptfont\lyfam\ninly \scriptscriptfont\lyfam\sevly

367:  \@setstrut \rm}

368: 

369: \makeatother

370: 

371: 

372: 

373: 

374: %

375: 

376: %

377: %

378: %

379: %

380: %

381: %

382: 

383: %

384: \newcommand{\DTIME}[1]{{\mbox{\rm DTIME}}(#1)}

385: \newcommand{\DSPACE}[1]{{\mbox{\rm DSPACE}}(#1)}

386: \newcommand{\NTIME}[1]{{\mbox{\rm NTIME}}(#1)}

387: \newcommand{\NSPACE}[1]{{\mbox{\rm NSPACE}}(#1)}

388: \newcommand{\ATIME}[1]{{\mbox{\rm ATIME}}(#1)}

389: \newcommand{\ASPACE}[1]{{\mbox{\rm ASPACE}}(#1)}

390: 

391: \newcommand{\sharpp}{{\rm \#P}}

392: \newcommand{\sharpsat}{{\rm \#SAT}}

393: \newcommand{\sat}{\mbox{\rm SAT}}

394: \newcommand{\qbf}{\mbox{\rm QBF}}

395: \newcommand{\parityp}{{\rm \oplus P}}

396: \newcommand{\up}{\mbox{\rm UP}}

397: \newcommand{\us}{\mbox{\rm US}}

398: \newcommand{\fewnp}{\mbox{\rm FewNP}}

399: \newcommand{\fewp}{\mbox{\rm FewP}}

400: \newcommand{\coup}{\mbox{\rm coUP}}

401: \newcommand{\e}{\mbox{\rm E}}

402: \renewcommand{\exp}{{\rm EXP}}

403: \newcommand{\NE}{\mbox{\rm NE}}

404: \renewcommand{\ne}{\mbox{\rm NE}}

405: \newcommand{\nexp}{\mbox{\rm NEXP}}

406: \newcommand{\p}{\mbox{\rm P}}

407: \newcommand{\littlep}{{\rm p}}

408: \newcommand{\NP}{\mbox{\rm NP}}

409: \newcommand{\DP}{\mbox{\rm DP}}

410: \newcommand{\np}{\mbox{\rm NP}}

411: \newcommand{\scriptnp}{\mbox{\protect\scriptsize\rm NP}}

412: \newcommand{\tinynp}{\mbox{\protect\tiny\rm NP}}

413: \newcommand{\nt}{\mbox{\rm NT}}

414: \newcommand{\nnt}{\mbox{\rm NNT}}

415: \newcommand{\parityoptp}{{\rm \oplus{}OptP}}

416: \newcommand{\optp}{{\rm OptP}}

417: \newcommand{\diffp}{{\rm D^P}}

418: \newcommand{\pp}{\mbox{\rm PP}}

419: \newcommand{\bpp}{\mbox{\rm BPP}}

420: \newcommand{\zpp}{\mbox{\rm ZPP}}

421: \newcommand{\corp}{\mbox{\rm coRP}}

422: \newcommand{\rp}{\mbox{\rm RP}}

423: \newcommand{\npc}{$\np$-com\-plete}

424: \newcommand{\conp}{\mbox{\rm coNP}}

425: \newcommand{\pspace}{\mbox{\rm PSPACE}}

426: \newcommand{\eespace}{{\rm EESPACE}}

427: \newcommand{\dspace}{{\rm DSPACE}}

428: \newcommand{\psp}{{\pspace}}

429: \newcommand{\pnexp}{{\p^\nexp}}

430: \newcommand{\npnexp}{{\np^\nexp}}

431: \newcommand{\nenp}{{\ne^\np}}

432: \newcommand{\enp}{{\e^\np}}

433: \newcommand{\pnp}{{\p^{\rm NP}}}

434: \newcommand{\npnp}{{\np^{\rm NP}}}

435: %

436: \newcommand{\pnplog}{\mbox{\rm P}^{\scriptnp[\mathcal{O}(\log)]}}

437: \newcommand{\nexpnp}{{\nexp^\np}}

438: \newcommand{\coNP}{{\rm coNP}}

439: \newcommand{\cone}{{\rm CONE}}

440: \newcommand{\sigmatwozero}{{\Sigma_2^0}}

441: \newcommand{\pitwozero}{{\Pi_2^0}}

442: \newcommand{\pithreezero}{{\Pi_3^0}}

443: \newcommand{\sigmathreezero}{{\Sigma_3^0}}

444: \newcommand{\sigmatwo}{{\Sigma_2^p}}

445: \newcommand{\sigmak}{{\Sigma_k^p}}

446: \newcommand{\pitwo}{{\Pi_2^p}}

447: \newcommand{\thetatwo}{{\Theta_2^p}}

448: \newcommand{\deltatwo}{{\Delta_2^p}}

449: \newcommand{\poly}{{\rm poly}}

450: \newcommand{\ph}{{\rm PH}}

451: \newcommand{\few}{{\rm Few}}

452: \newcommand{\fewch}{{\rm FewCH}}

453: \newcommand{\eh}{{\rm EH}}

454: \newcommand{\proof}{\noindent {\bf Proof.}\quad}

455: \def\bull{\vrule height .9ex width .8ex depth -.1ex }

456: %

457: %

458: %

459: %

460: %

461: %

462: \newcommand{\blob}{\mbox{\rule[-1.5pt]{5pt}{10.5pt}}}

463: \newcommand{\lindent}{\qquad}

464: \newcommand{\magicnum}{{ n^{\frac{1-\epsilon}{\epsilon}+\delta}}}

465: \newcommand{\fsup}{{\,f_{super}\,}}

466: \newcommand{\fred}{{\,f_{reduced}\,}}

467: \newcommand{\pne}{{\p^\ne}}

468: \newcommand{\npne}{{\np^\ne}}

469: \newcommand{\nnexarg}{{\nxx^\nexx (x) }}

470: \newcommand{\nnexx}{{\nxx^\nexx  }}

471: \newcommand{\nnex}{{\nxx^\nexx }}

472: \newcommand{\expnp}{{\exp^\np }}

473: \newcommand{\nxx}{{\rm N_{17}}}

474: \newcommand{\nexx}{{\rm NE_{21}}}

475: \newcommand{\seh}{{\rm SEH}}

476: \newcommand{\sexph}{{\rm SEXPH}}

477: \newcommand{\pstar}{{\p_\star}}

478: \newcommand{\nestar}{{\ne_{\,\star}}}

479: %

480: \newcommand{\supersetproper}{  \stackrel{\scriptscriptstyle\superset}{\scriptscriptstyle\not-}}

481: \newcommand{\subsetproper}{  \stackrel{\scriptscriptstyle\subset}{\scriptscriptstyle\not-}}

482: %

483: %

484: %

485: %

486: %

487: %

488: %

489: \newcommand{\superset}{\supset}

490: \newcommand{\superseteq}{\supseteq}

491: 

492: \def\unionfromc{\,\textstyle\bigcup_{\scriptstyle c}\,}

493: \def\unionfromk{\,\textstyle\bigcup_{\scriptstyle k}\,}

494: 

495: 

496: 

497: 

498: 

499: 

500: 

501: 

502: 

503: %

504: %

505: \newcommand{\newlozenge}{\setlength{\fboxsep}{0pt}\setlength{\fboxrule}{.7pt}\framebox[6pt]{\rule{0pt}{9pt}}}

506: 

507: %

508: %

509: 

510: %

511: 

512: 

513: %

514: %

515: %

516: %

517: \def\pair#1{{{\langle\!\!~#1~\!\!\rangle}}}

518: \def\pairs#1{{{\langle\!\!~#1~\!\!\rangle}}}

519: \newcommand{\piso}{\mbox{$\littlep$-iso\-mor\-phic}}

520: \newcommand{\manyonea}{\mbox{$\,\leq_{\rm m}^{{\littlep},\,A}\,$}}

521: \newcommand{\manyone}{\mbox{$\,\leq_{\rm m}^{{\littlep}}$\,}}

522: \newcommand{\paiso}{\mbox{$\littlep^A$-iso\-mor\-phic}}

523: \newcommand{\pisoa}{\paiso}

524: \newcommand{\pisoam}{\mbox{$\littlep^A$-iso\-mor\-phism}}

525: \newcommand{\pisom}{\mbox{$\littlep$-iso\-mor\-phism}}

526: \newcommand{\pselective}{\mbox{$\p$-selec\-tive}}

527: \newcommand{\sigmastar}{\mbox{$\Sigma^\ast$}}

528: \newcommand{\pisnp}{\mbox{$\p=\np$}}

529: \newcommand{\usuba}{\mbox{$U_A$}}

530: \newcommand{\univsuba}{\mbox{$Univ_A$}}

531: \newcommand{\pisnotnp}{\mbox{$\p\neq\np$}}

532: \newcommand{\lb}{\mbox{\{}}

533: \newcommand{\rb}{\mbox{\}}}

534: \newcommand{\pa}{\mbox{$\p^A$}}

535: \newcommand{\calf}{\mbox{$\cal F$}}

536: \newcommand{\calc}{\mbox{$\cal C$}}

537: \newcommand{\dr}{{\tt DodgsonRanking}}

538: \newcommand{\dw}{{\tt DodgsonWinner}}

539: \newcommand{\kw}{{\tt KemenyWinner}}

540: \newcommand{\score}{{\mbox{\it{}Score}}}

541: \newcommand{\dsum}{{\mbox{\it{}DodgsonSum}}}

542: \newcommand{\ds}{{\tt DodgsonScore}}

543: \newcommand{\xthreec}{{\tt ExactCoverByThreeSets}}

544: \newcommand{\satisfiability}{{\tt Satisfiability}}

545: \newcommand{\clique}{{\tt Clique}}

546: \newcommand{\threedm}{{\tt ThreeDimensionalMatching}}

547: \newcommand{\merge}{\mbox{\it{}Merge}}

548: \newcommand{\mergeprime}{\mbox{{\it{}Merge}$'$}}

549: \newcommand{\ncertifiesdefeat}{{\tt CertifiesTieOrDefeat}}

550: \newcommand{\specialdpproblem}{{\tt SpecialProblem}}

551: 

552: \newcommand{\kcolor}{{{\tt \mbox{K-}Colorability}}}

553: \newcommand{\threecolor}{{{\tt 3\mbox{-}Colorability}}}

554: \newcommand{\minthreeuncolor}{{\tt Minimal}\mbox{-}3\mbox{-}{\tt Uncolorability}}

555: \newcommand{\algthreecolor}{{{\tt A\mbox{-}3\mbox{-}Colorability}}}

556: 

557: \newcommand{\ddseqthreecolor}{{{\tt DD\mbox{-}SEQ\mbox{-}3\mbox{-}Colorability}}}

558: \newcommand{\slseqthreecolor}{{{\tt SL\mbox{-}SEQ\mbox{-}3\mbox{-}Colorability}}}

559: 

560: \newcommand{\maxthreesat}{{{\tt MAX\mbox{-}3\mbox{-}SAT}}}

561: 

562: \newcommand{\cfsp}{{{\tt{}CFSP}}}

563: %

564: \def\exactcfsp#1{{\tt{}Exact}\mbox{-}{#1}\mbox{-}{\tt{}CFSP}}

565: 

566: \newcommand{\mindom}{{{\tt Minimum\mbox{-}Dominating\mbox{-}Set}}}

567: \newcommand{\threedomatic}{{{\tt 3\mbox{-}DNP}}}

568: \newcommand{\kdomatic}{{{\tt k\mbox{-}DNP}}}

569: \newcommand{\domatic}{{{\tt DNP}}}

570: \newcommand{\domaticnumber}{{{\tt Domatic} {\tt Number}}}

571: \newcommand{\domaticnumberproblem}{{{\tt Domatic} {\tt Number} {\tt Problem}}}

572: \newcommand{\dnpodd}{{\tt DNP}\mbox{-}{\tt Odd}}

573: \newcommand{\dnpcompare}{{\tt DNP}\mbox{-}{\tt Geq}}

574: \newcommand{\dnpequal}{{\tt DNP}\mbox{-}{\tt Equ}}

575: 

576: \newcommand{\scriptodd}{\mbox{\scriptsize ${\tt odd}$}}

577: \newcommand{\scripteven}{\mbox{\scriptsize ${\tt even}$}}

578: 

579: \newcommand{\conveyorflowshop}{{{\tt Conveyor} {\tt Flow} {\tt Show}}}

580: \newcommand{\conveyor}{{\tt CFSP}}

581: \def\exactdomatic#1{{\tt{}Exact}\mbox{-}{#1}\mbox{-}{\tt{}DNP}}

582: \def\exactconveyor#1{{\tt{}Exact}\mbox{-}{#1}\mbox{-}{\tt{}CFSP}}

583: 

584: 

585: \newcommand{\ddseqint}{{{\tt DD\mbox{-}SEQINT}}}

586: \newcommand{\slseqint}{{{\tt SL\mbox{-}SEQINT}}}

587: 

588: \newcommand{\threesat}{{{\tt 3\mbox{-}SAT}}}

589: 

590: \newcommand{\voter}{\pair}

591: \newcommand{\npa}{\mbox{$\np^A$}}

592: \newcommand{\conpa}{\mbox{$\conp^A$}}

593: \newcommand{\upa}{\mbox{$\up^A$}}

594: \newcommand{\sparses}{\mbox{ sparse $S\,$}}

595: \newcommand{\bigo}{\mbox{$\cal O$}}

596: \newcommand{\condition}{\,\nottoobig{|}\:}

597: \def\land{{\; \wedge \;}}

598: 

599: 

600: %

601: %

602: 

603: %

604: \newcommand{\openNeighbors}{\mbox{$\mbox{\rm{}openNeighbors}_{\mathcal{P}}$}}

605: \newcommand{\openSets}{\mbox{$\mbox{\rm{}openSets}_{\mathcal{P}}$}}

606: \newcommand{\balance}{\mbox{$\mbox{\rm{}balance}_{\mathcal{P}}$}}

607: \newcommand{\surplus}{\mbox{$\mbox{\rm{}surplus}_{\mathcal{P}}$}}

608: \newcommand{\degree}{\mbox{\it deg}}

609: \newcommand{\area}{\mbox{$\mbox{\rm{}area}_{\mathcal{P}}$}}

610: \newcommand{\gap}{\mbox{$\mbox{\rm{}gap}_{\mathcal{P},\mathcal{A}}$}}

611: \newcommand{\sumgap}{\mbox{$\mbox{\rm{}sumgap}_{\mathcal{P},\mathcal{A}}$}}

612: \newcommand{\maxgap}{\mbox{$\mbox{\rm{}maxgap}_{\mathcal{P},\mathcal{A}}$}}

613: \newcommand{\mingap}{\mbox{$\mbox{\rm{}mingap}_{\mathcal{P},\mathcal{A}}$}}

614: \newcommand{\maxdegree}{\mbox{\it max-deg}}

615: \newcommand{\mindegree}{\mbox{\it min-deg}}

616: \newcommand{\mis}{\mbox{\it mis}}

617: \newcommand{\mdg}{\mbox{\it mdg}}

618: \newcommand{\mvc}{\mbox{\it mvc}}

619: \newcommand{\minmdg}{\mbox{\it min-mdg}}

620: \newcommand{\mined}{\mbox{\it min-ed}}

621: %

622: 

623: \newcommand{\parallelnp}{\mbox{$\p_{||}^{\scriptnp}$}}

624: \newcommand{\fooparallelnp}{\mbox{$\p_{||}^{\tinynp}$}}

625: \newcommand{\pnpbf}{\mbox{\rm P}_{\mbox{\protect\scriptsize\rm{}bf}}^{\scriptnp}}

626: \newcommand{\foopnpbf}{\mbox{\rm P}_{\mbox{\protect\tiny\rm{}bf}}^{\tinynp}}

627: 

628: \newcommand{\misequals}{\mbox{\tt MIS$_{\mbox{\footnotesize\tt equal}}$}}

629: \newcommand{\sone}{\mbox{${\cal S}_1$}}

630: \newcommand{\ar}{\mbox{${\cal A}_r$}}

631: \newcommand{\sr}{\mbox{${\cal S}_r$}}

632: \newcommand{\truthtable}{\mbox{$\,\leq_{\rm tt}^{{\littlep}}$\,}}

633: \newcommand{\turing}{\mbox{$\,\leq_{\rm T}^{{\littlep}}$\,}}

634: 

635: \def\bhlevel#1{{{\mbox{\rm{}BH}_{#1}(\np)}}}

636: \newcommand{\bhnp}{{{\mbox{\rm{}BH}(\np)}}}

637: \def\exactcolor#1{{\tt{}Exact}\mbox{-}{#1}\mbox{-}{\tt{}Colorability}}

638: \newcommand{\exactfourcolor}{{\tt{}Exact}\mbox{-}{\tt{}Four}\mbox{-}{\tt{}Colorability}}

639: 

640: \newcommand\seq{\subseteq}

641: \renewcommand\.{\cdot}

642: \newcommand\<{\langle}

643: \renewcommand\>{\rangle}

644: \newcommand\la{\ \leftarrow \ }

645: \newcommand\ra{\ \rightarrow \ }

646: \newcommand\lra{\ \leftrightarrow \ }

647: \newcommand\La{\ \Leftarrow \ }

648: \newcommand\Ra{\ \Rightarrow \ }

649: \newcommand\Lra{\ \Leftrightarrow \ }

650: \newcommand\lola{\ \longleftarrow \ }

651: \newcommand\lora{\ \longrightarrow \ }

652: \newcommand\lolra{\ \longleftrightarrow \ }

653: \newcommand\Lola{\ \Longleftarrow \ }

654: \newcommand\Lora{\, \Longrightarrow \ }

655: \newcommand\Lolra{\ \Longleftrightarrow \ }

656: \newcommand\Lolradef{\ :\Longleftrightarrow \ }

657: 

658: \newcommand{\naturalnumber}{\ensuremath{{  \mathbb{N} }}}

659: \def\nats{\naturalnumber}

660: 

661: \newcommand{\bigoh}{\mathcal{O}}

662: 

663: %

664: 

665: %

666: %

667: %

668: 

669: 

670: %

671: %

672: %

673: %

674: %

675: %

676: %

677: %

678: %

679: %

680: %

681: %

682: %

683: %

684: %

685: %

686: %

687: %

688: %

689: %

690: %

691: %

692: %

693: 

694: 

695: \newenvironment{construction}{\bigbreak\begin{block}}{\end{block}

696:     \bigbreak}

697: 

698: \newenvironment{block}{\begin{list}{\hbox{}}{\leftmargin 1em

699:     \itemindent -1em \topsep 0pt \itemsep 0pt \partopsep 0pt}}{\end{list}}

700: 

701: 

702: %

703: %

704: %

705: %

706: %

707: %

708: %

709: %

710: 

711: \dimen15=0.75em

712: \dimen16=0.75em

713: 

714: \newcommand{\lblocklabel}[1]{\rlap{#1}\hss}

715: 

716: \newenvironment{lblock}{\begin{list}{}{\advance\dimen15 by \dimen16

717:     \leftmargin \dimen15

718:     \itemindent -1em

719:     \topsep 0pt

720:     \labelwidth 0pt

721:     \labelsep \leftmargin

722:     \itemsep 0pt

723:     \let\makelabel\lblocklabel

724:     \partopsep 0pt}}{\end{list}}

725: 

726: 

727: %

728: %

729: 

730: \newenvironment{lconstruction}[2]{\dimen15=#1 \dimen16=#2

731:   \bigbreak\begin{block}}{\end{block}\bigbreak}

732: 

733: 

734: \newcommand{\Comment}[1]{{\sl (\  #1\ )}}

735: 

736: 

737: %

738: %

739: %

740: 

741: %

742: 

743: \newcommand{\finishCycles}{{\mbox{\sc Finish-Cycles}}}

744: \newcommand{\dominate}{{\mbox{\sc Dominate}}}

745: \newcommand{\assign}{{\mbox{\sc Assign}}}

746: \newcommand{\recalculateGaps}{{\mbox{\sc Recalculate-Gaps}}}

747: \newcommand{\finish}{{\mbox{\sc Finish-Graph}}}

748: \newcommand{\handleCriticalVertex}{{\mbox{\sc Handle-Critical-Vertex}}}

749: 

750: 

751: \newenvironment{algorithmus}{\begin{list}

752:    {{\bf Step~\arabic{alg}}.}

753:    {\usecounter{alg}}}{\end{list}}

754: \newcounter{alg}

755: 

756: \newenvironment{algorithmusfall}{\begin{list}

757:    {{\bf Case~\arabic{alg}}:}

758:    {\usecounter{alg}}}{\end{list}}

759: 

760: 

761: 

762: %

763: %

764: %

765: \title{An Exact $2.9416^n$ Algorithm for the Three Domatic Number

766:   Problem\thanks{Work supported in part by the DFG under Grant RO~1202/9-1.}}

767: 

768: \author{

769: Tobias Riege\thanks{Email: ${\tt riege@cs.uni\mbox{-}duesseldorf.de}$.}

770: \quad and \quad

771: J\"{o}rg Rothe\thanks{Email: ${\tt rothe@cs.uni\mbox{-}duesseldorf.de}$.} \\

772: Institut f\"ur Informatik \\

773: Heinrich-Heine-Universit\"at D\"usseldorf \\

774: 40225 D\"usseldorf, Germany

775: }

776: 

777: %

778: \date{June 24, 2005}

779: 

780: 

781: %

782: %

783: %

784: %

785: 

786: 

787: %

788: %

789: %

790: \lefthyphenmin=100

791: 

792: %

793: %

794: %

795: 

796: %

797: %

798: %

799: %

800: %

801: 

802: %

803: %

804: %

805: %

806: %

807: %

808: 

809: %

810: %

811: %

812: %

813: %

814: %

815: %

816: %

817: %

818: %

819: %

820: %

821: 

822: 

823: 

824: %

825: %

826: %

827: %

828: %

829: %

830: %

831: %

832: 

833: %

834: %

835: %

836: %

837: %

838: %

839: %

840: %

841: %

842: \makeatletter

843: \def\@listI{\leftmargin\leftmargini \parsep 4.5pt plus 1pt minus 1pt\topsep

844: 6pt plus 2pt minus 2pt \itemsep  2pt plus 2pt minus 1pt}

845: 

846: \let\@listi\@listI

847: \@listi

848: \makeatother

849: 

850: %

851: 

852: \newcommand{\anvec}[2]{\mbox{$#1_1, \ldots, #1_{#2}$}}

853: \newcommand{\anvecplus}[2]{\mbox{$#1_1, #1_2, \ldots, #1_{#2}$}}

854: 

855: \newcommand{\mujp}{\mbox{$\mu_{j,p}$}}

856: \newcommand{\tjp}{\mbox{$T_{(j,p)}$}}

857: 

858: \newcommand{\naesat}{{{\tt NAE\mbox{-}3\mbox{-}SAT}}}

859: \newcommand{\onethreesat}{{{\tt 1\mbox{-}3\mbox{-}SAT}}}

860: 

861: \newcommand{\Pos}{\mathbb{N^+}}

862: 

863: \newcommand{\eksp}{{\tt Exact}\mbox{-$(k, \sigma, \rho)$-}{\tt Partition}}

864: \newcommand{\ekspargs}[3]{{\tt Exact}\mbox{-$({#1}, {#2}, {#3})$-}{\tt Partition}}

865: \newcommand{\kspset}{\mbox{$(\sigma, \rho)$}}

866: \newcommand{\ksparg}[1]{\mbox{$({#1}, \sigma, \rho)$}}

867: \newcommand{\kspargpartition}[1]{\mbox{$({#1}, \sigma, \rho)$-}{\tt Partition}}

868: \newcommand{\kspargs}[3]{\mbox{$({#1}, {#2}, {#3})$}}

869: \newcommand{\kspargspartition}[3]{\mbox{$({#1}, {#2}, {#3})$-}{\tt Partition}}

870: \newcommand{\kspargszwei}[2]{\mbox{$({#1}, {#2})$}}

871: 

872: 

873: \begin{document}

874: 

875: %

876: %

877: %

878: %

879: %

880: \typeout{WARNING:  BADNESS used to suppress reporting!  Beware!!}

881: \hbadness=3000%

882: \vbadness=10000 %

883: 

884: 

885: 

886: 

887: 

888: 

889: %

890: %

891: %

892: 

893: %

894: %

895: %

896: %

897: %

898: %

899: %

900: %

901: %

902: %

903: \setcounter{page}{1}

904: 

905: 

906: %

907: 

908: \sloppy

909: 

910: %

911: %

912: \setcounter{footnote}{0}

913: %

914: 

915: \maketitle

916: 

917: %

918: %

919: %

920: 

921: %

922: %

923: \begin{abstract}

924: \noindent

925: The three domatic number problem asks whether a given undirected graph

926: can be partitioned into at least three dominating sets, i.e., sets

927: whose closed neighborhood equals the vertex set of the graph.  Since

928: this problem is NP-complete, no polynomial-time algorithm is known for

929: it.  The naive deterministic algorithm for this problem runs in

930: time~$3^n$, up to polynomial factors.  In this paper, we design an

931: exact deterministic algorithm for this problem running in

932: time~$2.9416^n$.  Thus, our algorithm can handle problem instances of

933: larger size than the naive algorithm in the same amount of time.  We

934: also present another deterministic and a randomized algorithm for this

935: problem that both have an even better performance for graphs with

936: small maximum degree.

937: 

938: \vspace*{.5cm}

939: \noindent

940: \begin{tabular}{ll}

941: {\bf Key words:}

942:  & Exact algorithms, domatic number problem 

943: \end{tabular}

944: \end{abstract}

945: 

946: %

947: 

948: %

949: %

950: %

951: %

952: %

953: %

954: \setcounter{page}{1}

955: %

956: %

957: \pagestyle{plain}

958: %

959: \sloppy

960: 

961: %

962: 

963: \setcounter{page}{1}

964: 

965: \section{Introduction}

966: 

967: In this paper, we design a deterministic algorithm for the three

968: domatic number problem, which is one of the standard $\np$-complete

969: problems, see Garey and Johnson~\cite{gar-joh:b:int}.  This problem

970: asks, given an undirected graph~$G$, whether or not the vertex set of

971: $G$ can be partitioned into three dominating sets.  A dominating set

972: is a subset of the vertex set that ``dominates'' the graph in that its

973: closed neighborhood covers the entire graph.  Motivated by the tasks

974: of distributing resources in a computer network and of locating

975: facilities in a communication network, this problem and the related

976: problem of finding a minimum dominating set in a given graph have been

977: thoroughly studied, see, e.g.,

978: \cite{coc-hed:j:towards-theory-of-domination,far:j:domatic-number-strongly-chordal-graphs,bon:j:domatic-number-circular-arc-graphs,kap-sha:j:domatic-number,heg-tel:j:generalized-dominating-sets,fei-hal-kor:c:approximating-domatic-number,rie-rot:j:exact-dnp}.

979: 

980: The exact (i.e., deterministic) algorithm designed in this paper runs

981: in exponential time.  However, its running time is better than that of

982: the naive exact algorithm for this problem.  That is, we improve the

983: trivial $\tilde{\mathcal{O}}(3^n)$ time bound to a time bound

984: of~$\tilde{\mathcal{O}}(2.9416^n)$, where the $\tilde{\mathcal{O}}$

985: notation neglects polynomial factors as is common for exponential-time

986: algorithms.  The point of such an improvement is that a

987: $\tilde{\mathcal{O}}(c^n)$ algorithm, where $c < 3$ is a constant, can

988: deal with larger instances than the trivial $\tilde{\mathcal{O}}(3^n)$

989: algorithm in the same amount of time before the exponential growth

990: rate eventually hits and the running time becomes infeasible.  For

991: example, if $c = \sqrt{3} \approx 1.732$ then we have

992: $\tilde{\bigoh}\left(\sqrt{3}^{\, 2n}\right) = \tilde{\bigoh}(3^n)$,

993: so one can deal with inputs twice as large as before.  Doubling the

994: size of inputs that can be handled by some algorithm can make quite a

995: difference in practice.

996: 

997: Exact exponential-time algorithms with improved running times have

998: been designed for various other important $\np$-complete problems.

999: For example, Dantsin et al.~\cite{dan-etal:j:det-alg-for-ksat} pushed

1000: the trivial $\tilde{\bigoh}(2^n)$ bound for the three satisfiability

1001: problem down to~$\tilde{\bigoh}(1.481^n)$, which was further improved

1002: to $\tilde{\bigoh}(1.473^n)$ by Brueggemann and

1003: Kern~\cite{bru-ker:t:local-search-threesat}.

1004: Schöning~\cite{sch:j:constraint-satisfaction}, Hofmeister et

1005: al.~\cite{hof-sch-sch-wat:c:probabilistic-threesat-alg-further-improved}

1006: and Paturi et al.~\cite{pat-pud-sak-zan:c:exptime-algorithm-for-ksat}

1007: proposed even better randomized algorithms for the satisfiability

1008: problem.  Combining their ideas, the currently best randomized

1009: algorithm for this problem is due to Iwama and

1010: Tamaki~\cite{iwa-tak:t:improved-upper-bound-threesat}, who achieve a

1011: time bound of~$\tilde{\bigoh}(1.324^n)$.

1012: 

1013: The currently best exact time bound of $\tilde{\bigoh}(1.211^n)$ for

1014: the independent set problem is due to

1015: Robson~\cite{rob:j:exact-alg-for-independent-set}.

1016: Eppstein~\cite{epp:c:algs-for-three-sat-threecolor,epp:c:exact-alg-for-graph-coloring}

1017: achieved a $\tilde{\bigoh}(2.415^n)$ time bound for graph coloring and

1018: a $\tilde{\bigoh}(1.3289^n)$ for the special case of graph three

1019: colorability.  Fomin, Kratsch, and

1020: Woeginger~\cite{fom-kra-woe:c:exact-algorithm-for-dominating-set}

1021: improved the trivial $\tilde{\bigoh}(2^n)$ bound for the dominating

1022: set problem to~$\tilde{\bigoh}(1.93782^n)$.  Comprehensive surveys on

1023: this subject have been written by

1024: Woeginger~\cite{woe:b:exact-algorithms-for-np-hard-problems} and

1025: Schöning~\cite{sch:c:algorithmics-in-exponential-time}.

1026: 

1027: 

1028: In designing domatic number algorithms, it might be tempting to exploit known

1029: results (such as Eppstein's $\tilde{\bigoh}(1.3289^n)$ bound) for the

1030: graph three colorability problem, which resembles the three domatic

1031: number problem in that both are partitioning problems.  However, as

1032: Cockayne and Hedetniemi~\cite{coc-hed:j:towards-theory-of-domination}

1033: point out, the theory of domination is dual to the theory of

1034: coloring in the following sense.  Coloring is based on the hereditary

1035: property of independence.  A graph property

1036: is {\em hereditary\/} if whenever some set of vertices has the

1037: property then so does every subset of it.  In contrast, domination is

1038: an {\em expanding\/} property in that every superset of a dominating

1039: set also is a dominating set of the graph.  Further, graph colorability is a

1040: minimum problem, whereas the domatic 

1041: number problem is a maximum problem.  Independence (and thus

1042: colorability) can be seen as a {\em local\/} property, since it

1043: suffices to check the immediate neighborhood of a set of vertices to

1044: determine whether or not it is independent.  In contrast, dominance is

1045: a {\em global\/} property, since in order to check it one has to

1046: consider the relation between the given set of vertices and the entire

1047: graph.  In this sense, determining the domatic number of a graph

1048: intuitively appears to be harder than computing its chromatic number,

1049: notwithstanding that both problems are $\np$-complete.  More to the

1050: point, the algorithms developed for graph coloring seem to be of no

1051: help in designing algorithms for dominating set or domatic number

1052: problems.

1053: 

1054: After introducing some definitions and notation in

1055: Section~\ref{sec:prelims}, we describe and analyze our algorithm in

1056: Section~\ref{sec:algo}; the actual pseudo-code is shifted to the

1057: appendix.  In Section~\ref{sec:bounded-max-degree}, we give another

1058: deterministic and a randomized algorithm, which have an even better

1059: running time for graphs with small maximum degree. Finally, we

1060: summarize and discuss our results in Section~\ref{sec:conclusion}.

1061: 

1062: \section{Preliminaries and Simple Observations}

1063: \label{sec:prelims}

1064: 

1065: We start by introducing some graph-theoretical notation.  We only

1066: consider simple, undirected graphs without loops in this paper.  Let

1067: $G = (V,E)$ be a graph.  Unless stated otherwise, $n$ denotes the

1068: number of vertices in~$G$. The {\em neighborhood of a vertex~$v$ in

1069: $V$\/} is defined by $N(v) = \{ u \in V \condition \{u,v\} \in E \}$,

1070: and the {\em closed neighborhood of~$v$\/} is defined by $N[v] = N(v)

1071: \cup \{ v \}$.  For any subset $S \seq V$ of the vertices of~$G$,

1072: define $N[S] = \bigcup_{v \in S} N[v]$ and $N(S) = N[S] - S$.  The

1073: {\em degree of a vertex $v$ in $G$\/} is the number of vertices

1074: adjacent to~$v$, i.e., $\degree_G(v) = ||N(v)||$.  If the graph $G$ is

1075: clear from the context, we omit the subscript~$G$.  Define the {\em

1076: minimum degree in $G$\/} by $\mindegree(G) = \min_{v \in V}

1077: \degree(v)$, and the {\em maximum degree in $G$\/} by $\maxdegree(G) =

1078: \max_{v \in V} \degree(v)$.  

1079: A path $P_k = u_1 u_2 \cdots u_k$ of length 

1080: $k$ is a sequence of $k$ vertices, where each vertex is adjacent to

1081: its successor, i.e., $\{u_i, u_{i+1} \} \in E$ for $1 \leq i \leq k-1$. 

1082: If, in addition, $\{ u_k, u_1 \} \in E$, then path $P_k$ is said

1083: to be a {\em cycle}, and we write $C_k$ instead of $P_k$.

1084: 

1085: %

1086: %

1087: %

1088: 

1089: \begin{definition}  

1090: Let $G = (V,E)$ be a graph.  A subset $D \seq V$ is a {\em dominating

1091: set of~$G$\/} if and only if $N[D] = V$, i.e., if and only if every

1092: vertex in $G$ either belongs to~$D$ or has some neighbor in~$D$.  The

1093: {\em domination number of~$G$}, denoted~$\gamma(G)$, is the minimum

1094: size of a dominating set of~$G$. The {\em domatic number of~$G$},

1095: denoted~$\delta(G)$, is the maximum number of disjoint dominating sets

1096: of~$G$, i.e., $\delta(G)$ is the maximum $k$ such that $V = V_1 \cup

1097: V_2 \cup \ldots \cup V_k$, where $V_i \cap V_j = \emptyset$ for $1

1098: \leq i < j \leq k$, and each $V_i$ is a dominating set of~$G$.

1099: %

1100: %

1101: %

1102: The {\em dominating set problem\/} asks, given a graph $G$ and a

1103: positive integer~$k$, whether or not $\gamma(G) \leq k$.

1104: %

1105: %

1106: The {\em domatic number problem\/} asks, given a graph $G$ and a

1107: positive integer~$k$, whether or not $\delta(G) \geq k$.

1108: \end{definition}

1109: 

1110: For fixed $k \geq 3$, both the dominating set problem and the domatic

1111: number problem are known to be $\np$-complete, see Garey and

1112: Johnson~\cite{gar-joh:b:int}.  Thus, they are not solvable in

1113: deterministic polynomial time unless $\p = \np$, and all we can hope

1114: for is to design an exponential-time algorithm having a better running

1115: time than the trivial exponential time bound.

1116: %

1117: For exponential-time algorithms, it is common to drop polynomial factors,

1118: as indicated by the $\tilde{\bigoh}$ notation: For functions $f$

1119: and~$g$, we write $f \in \tilde{\bigoh}(g)$ if and only if $f \in

1120: \bigoh(p \cdot g)$ for some polynomial~$p$.

1121: The naive deterministic

1122: algorithm for the dominating set problem runs in time

1123: $\tilde{\bigoh}(2^n)$.  Fomin, Kratsch, and

1124: Woeginger~\cite{fom-kra-woe:c:exact-algorithm-for-dominating-set}

1125: improved this trivial upper bound to~$\tilde{\bigoh}(1.93782^n)$.  For

1126: various restricted graph classes, they achieve even better bounds.

1127: 

1128: The naive deterministic algorithm for the domatic number problem works

1129: as follows: Given a graph~$G$ and an integer~$k$, it sequentially

1130: checks every potential solution (i.e., every possible partition of the

1131: vertex set of $G$ into $k$ sets $D_1, D_2, \ldots, D_k$), and accepts

1132: if and only if a correct solution is found (i.e., if and only if each

1133: $D_i$ is a dominating set).  How many potential solutions are there?

1134: The number of ways of partitioning a set with $n$ elements into $k$

1135: nonempty, disjoint subsets

1136: %

1137: %

1138: %

1139: %

1140: %

1141: %

1142: can be calculated by the Stirling number of the second kind:

1143: $S_2(n,k) = \frac{1}{k!} \sum_{i=0}^{k-1} (-1)^i {k \choose i} (k-i)^n$,

1144: which yields a running time of~$\tilde{\bigoh}(k^n)$.

1145: A better result can be achieved via the dynamic programming across the subsets

1146: technique, which was introduced by Lawler~\cite{law:j:complexity-chromatic} to

1147: compute the chromatic number of a graph by exploiting the fact that every

1148: minimum chromatic partition contains at least one maximum independent set.  By

1149: suitably modifying this technique, one can compute the domatic number of a

1150: graph in time~$\tilde{\bigoh}(3^n)$. This is done by generating all

1151: dominating sets of the graph with increasing cardinality, which takes time

1152: \[

1153: \sum_{k = 0}^n {n \choose k} 2^k = (1 + 2)^n = 3^n.

1154: \]

1155: The difference to Lawler's algorithm lies in the fact that all

1156: dominating sets need to be checked, whereas only maximum independent

1157: sets are relevant to compute the chromatic number.

1158: 

1159: \begin{proposition}

1160: \label{prop:lawler-domatic}

1161: Let $G = (V,E)$ be a graph. Then, the domatic number $\delta(G)$ can

1162: be computed in time~$\tilde{\bigoh}(3^n)$.

1163: \end{proposition}

1164: 

1165: %

1166: One tempting way of designing an improved algorithm for the domatic number

1167: problem might be to exploit the result for the dominating set problem

1168: mentioned above.  However, we observe that no such useful connection between

1169: the two problems exists in general.

1170: %

1171: The first part of Proposition~\ref{prop:mds-dnp} shows that an

1172: arbitrary given minimum dominating set is not necessarily part of a

1173: partition into a maximum number of dominating sets.  The second part

1174: of Proposition~\ref{prop:mds-dnp} shows that, given an arbitrary

1175: partition into a maximum number of dominating sets, it is not

1176: necessarily the case that one set of the partition indeed is a minimum

1177: dominating set.  Thus, for solving the domatic number problem, one

1178: cannot use in any obvious way the exact $\tilde{\bigoh}(1.93782^n)$

1179: algorithm for the dominating set problem by Fomin et

1180: al.~\cite{fom-kra-woe:c:exact-algorithm-for-dominating-set}.

1181: %

1182: %

1183: Proposition~\ref{prop:mds-dnp} is stated for graphs with domatic number~$3$;

1184: it can easily be generalized to graphs with domatic number $k \geq 3$.

1185: The proof of Proposition~\ref{prop:mds-dnp} can be found in the appendix.

1186: 

1187: \begin{proposition}

1188: \label{prop:mds-dnp}

1189: \begin{enumerate}

1190: \item There exists some graph $G$ with $\delta(G) = 3$ such that some minimum

1191:   dominating set $D$ of $G$ is not part of any partition into three dominating

1192:   sets of~$G$.

1193:   

1194: \item There exists some graph $H = (V,E)$ with $\delta(H) = 3$ such

1195:   that for each partition $V = D_1 \cup D_2 \cup D_3$ into three

1196:   dominating sets of $H$ and for each~$i$, $||D_i|| > \gamma(H)$.

1197: %

1198: %

1199: %

1200: \end{enumerate}

1201: \end{proposition}

1202: 

1203: For the three domatic number problem, no algorithm with a running

1204: time better than~$\tilde{\bigoh}(3^n)$ is known.

1205: %

1206: We improve this trivial upper bound to~$\tilde{\bigoh}(2.9416^n)$.

1207: 

1208: We now define some technical notions suitable to measure how

1209: ``useful'' a vertex is to achieve domination of the graph~$G = (V,E)$.

1210: Intuitively, the vertex degree is a good (local) measure, since the

1211: larger the neighborhood of a vertex is, the more vertices are

1212: potentially dominated by the set to which it belongs.  The technical

1213: notions introduced in Definition~\ref{def:notions-for-algorithm} will

1214: be used later on to describe our algorithm.

1215: 

1216: \begin{definition}

1217: \label{def:notions-for-algorithm}

1218:   Let $G = (V,E)$ be a graph with $n$ vertices, and let $\mathcal{P} =

1219:   (D_1, D_2, D_3, R)$ be a partition of $V$ into four sets, $D_1$,

1220:   $D_2$, $D_3$, and~$R$.  The subsets $D_i$ of $V$ will eventually

1221:   yield a partition of $V$ into the three dominating sets (if they

1222:   exist) to be constructed, and the subset $R \seq V$ collects the

1223:   remaining vertices not yet assigned at the current point in the

1224:   computation of the algorithm.  Let $r = || R ||$ be the number of

1225:   these remaining vertices, and let $d = n-r$

1226: %

1227: be the number of vertices already assigned to some set~$D_i$. The {\em

1228:   area of $G$ covered by $\mathcal{P}$} is defined as

1229:   $\area(G) = \sum_{i=1}^3 || N[D_i] ||$.

1230: Note that $\area(G) = 3n$ if and only if $D_1$, $D_2$, and $D_3$ are

1231: dominating sets of~$G$. For a partition~$\mathcal{P}$, we also define

1232: the {\em surplus of graph $G$} as

1233: $\surplus(G) = \area(G) - 3d$.

1234:  

1235: Some of the vertices in $R$ may be assigned to three, not necessarily

1236: disjoint, auxiliary sets~$A_1$, $A_2$, and~$A_3$ arbitrarily.  Let

1237: $\mathcal{A} = (A_1, A_2, A_3)$.  For each vertex $v \in R$ and for

1238: each $i$ with $1 \leq i \leq 3$, define the {\em gap of vertex $v$

1239: with respect to set $D_i$\/} by

1240: \[

1241: \gap(v,i) = \left\{

1242: \begin{array}{ll}

1243:  || N[v] ||

1244:  - || \{ u \in N[v] \condition (\exists w \in N[u]) [ w \in D_i] \} || 

1245:  & \mbox{if $v \notin A_i$} \\

1246:  \bot & \mbox{otherwise,}

1247: \end{array} 

1248: \right.

1249: \]

1250: where $\bot$ is a special symbol that indicates that $\gap(v,i)$ is

1251: undefined for this $v$ and~$i$.  (Our algorithm will make sure

1252: to properly handle the cases of undefined gaps.)

1253: 

1254: Additionally, given $\mathcal{P}$ and~$\mathcal{A}$, define

1255: %

1256: %

1257: for all vertices $v \in R$:

1258: \begin{eqnarray*}

1259: \maxgap(v)

1260:  & = & \max \{ \gap(v,i) \condition 1 \leq i \leq 3 \}, \\

1261: \mingap(v)

1262:  & = & \min \{ \gap(v,i) \condition 1 \leq i \leq 3 \}, \\

1263: \sumgap(v)

1264:  & = & \sum_{i=1}^{3} \gap(v,i).

1265: \end{eqnarray*}

1266: 

1267: Given $G$, $\mathcal{P}$, and~$\mathcal{A}$, define the {\em maximum

1268: gap of $G$\/} and the {\em minimum gap of $G$\/} by taking the maximum

1269: and minimum gaps over all vertices in~$G$ not yet assigned:

1270: \begin{eqnarray*}

1271: \maxgap(G)

1272:  & = & \max \{ \maxgap(v) \condition v \in R \}, \\

1273: \mingap(G)

1274:  & = & \min \{ \mingap(v) \condition v \in R \}.

1275: \end{eqnarray*}

1276: 

1277: Let $\mathcal{P}$ be given. A vertex $u \in V$ is called an

1278: {\em open neighbor of $v \in V$\/} if 

1279: %

1280: $u \in N[v]$ and $u$ has not been assigned

1281: to any set $D_1$, $D_2$, or $D_3$ yet.  A potential dominating set

1282: $D_i$, $1 \leq i \leq 3$, is called an {\em open set of $v \in V$\/}

1283: if its closed neighborhood does not include~$v$, i.e., $v$ is not

1284: dominated by $D_i$. The {\em balance of $v \in V$} is defined as

1285: the difference between the number of open vertices and the number

1286: of open sets.  Formally, define

1287: \begin{eqnarray*}

1288: \openNeighbors(v) & = & \{ u \in N[v] \condition u \in R \}, \\

1289: \openSets(v) & = & \{ i \in \{1,2,3 \} \condition

1290:  v \notin N[D_i] \}, \\

1291: \balance(v) & = & || \openNeighbors(v) || - || \openSets(v) ||.

1292: \end{eqnarray*}

1293: We call a vertex $v \in V$ {\em critical\/} if and only if

1294: $\balance(v) \leq 0$ and $|| \openSets(v) || > 0$.

1295: \end{definition}

1296: 

1297: The proof of the next proposition is straightforward. Once

1298: $\balance(v) = 0$, no two vertices remaining in $N[v] \cap R$ can

1299: be assigned to the same dominating set $D_i$, $1 \leq i \leq 3$,

1300: since $\balance(v)$ would then be negative.

1301: 

1302: \begin{proposition}

1303: \label{prop:critical}

1304: Let $\mathcal{P} = (D_1, D_2, D_3, R)$ be given as in

1305: Definition \ref{def:notions-for-algorithm} , and $v \in V$ be

1306: a critical vertex for this partition. The only way to modify

1307: $\mathcal{P}$ so as to contain three dominating sets is to assign

1308: all vertices $u \in N[v] \cap R$ to distinct dominating sets~$D_i$.

1309: \end{proposition}

1310: 

1311: 

1312: \section{The Algorithm}

1313: \label{sec:algo}

1314: 

1315: Our strategy is to recursively assign the vertices $v \in V$ to obtain

1316: a correct potential solution consisting of a partition into three

1317: dominating sets, $D_1$, $D_2$, and~$D_3$.  Once a previous assignment

1318: of $v$ to some set $D_i$ turns out to be wrong, we remember this by

1319: adding this vertex to~$A_i$.  More precisely, the basic idea

1320: %

1321: is to first pick those vertices with the highest

1322: maximum gap.  While the algorithm is progressing, it dynamically

1323: updates the gaps for every vertex in each step.

1324: %

1325: %

1326: %

1327: %

1328: We now state our main result.

1329: 

1330: \begin{theorem}

1331: \label{the:detalgo}

1332:   The three domatic number problem can be solved by a deterministic algorithm

1333:   running in time~$\tilde{\bigoh}(2.9416^n)$.

1334: \end{theorem}

1335: 

1336: \begin{proofs}

1337: Let $G = (V,E)$ be the given graph.  The algorithm seeks to find a

1338: partition of $V$ into three disjoint dominating sets.

1339: %

1340: Note that every vertex $v \in V$ is contained in one of these sets and is

1341: dominated by the remaining two sets, i.e., it is adjacent to at least one of

1342: their elements. The algorithm is described in pseudo-code in the appendix,

1343: see Figures~\ref{fig:dnp}, \ref{fig:domatic}, \ref{fig:domatic},

1344: \ref{fig:assign}, \ref{fig:recalc}, and~\ref{fig:handle}. Since

1345: $\delta(G) \leq \mindegree(G) + 1$, we may assume that $\mindegree(G)

1346: \geq 2$.

1347: 

1348: The algorithm starts by initializing the potential dominating sets

1349: $D_1, D_2$, and $D_3$ and the auxiliary sets~$A_1$, $A_2$, and~$A_3$,

1350: setting each to the empty set.

1351: The initial partition thus is $\mathcal{P} =

1352: (\emptyset, \emptyset, \emptyset, V)$ and the initial triple of

1353: auxiliary sets is $\mathcal{A} = (\emptyset, \emptyset, \emptyset)$.

1354: 

1355: %

1356: %

1357: %

1358: %

1359: %

1360: %

1361: %

1362: %

1363: %

1364: 

1365: %

1366: %

1367: %

1368: %

1369: %

1370: %

1371: %

1372: %

1373: %

1374: %

1375: %

1376: %

1377: %

1378: %

1379: %

1380: %

1381: %

1382: %

1383: %

1384: %

1385: %

1386: %

1387: %

1388: %

1389: %

1390: 

1391: 

1392: Then, the recursive function $\dominate$ is called for the first time.

1393: It is always invoked with graph $G$, a partition $\mathcal{P} = (D_1,

1394: D_2, D_3, R)$, and a triple $\mathcal{A} = (A_1, A_2, A_3)$ of not

1395: necessarily disjoint auxiliary sets. $\mathcal{P}$ and $\mathcal{A}$

1396: represent a situation in which the vertices in $V - R$ have been

1397: assigned to 

1398: %

1399: $D_1$, $D_2$, and~$D_3$, and $v \in A_i$ means that in some previous

1400: recursive call to function $\dominate$ the vertex $v$ has been

1401: assigned to~$D_i$ without successfully changing $\mathcal{P}$ to

1402: contain three dominating sets.

1403: 

1404: Function $\dominate$ starts by calling $\recalculateGaps$, which

1405: calculates all gaps with respect to $\mathcal{P}$ and~$\mathcal{A}$.

1406: Additionally, $\openNeighbors(v)$, $\openSets(v)$, and $\balance(v)$

1407: are determined for every vertex $v \in V$. 

1408: %

1409: Four trivial cases can occur.

1410: 

1411: \begin{construction}

1412:   \item {\bf Case 1:} The sets $D_1$, $D_2$, and $D_3$ are dominating

1413:   sets of graph $G$. In this case, we are done and may add the

1414:   remaining vertices $v \in R$ to any set~$D_i$, say to~$D_1$.

1415: 

1416:   \item {\bf Case 2:} For some vertex $v \in V$, we have $\balance(v)

1417:         < 0$. That is, there are less vertices in $R \cap N[v]$ than

1418:         dominating sets with $v \notin N[D_i]$. Thus, no matter how the

1419:         vertices in $R \cap N[v]$ are assigned, $\mathcal{P}$ won't

1420:         contain three dominating sets.  We have run into a dead-end and

1421:         return to the previous level of the recursion.

1422: 

1423: %

1424: %

1425: %

1426: %

1427: %

1428: %

1429: 

1430:   \item {\bf Case 3:} There exists a vertex $v \in R$ that is also a

1431:         member of two of the auxiliary sets $A_1$, $A_2$, and

1432:         $A_3$. Hence, vertex $v$ was previously assigned to two

1433:         distinct sets $D_i$ and~$D_j$, $1 \leq i < j \leq 3$, but the

1434:         recursion returned without success. We assign $v$ to the only

1435:         possible set $D_k$ left, with $i \neq k \neq j$.

1436: %

1437: 

1438:   \item {\bf Case 4:} For some vertex $v \in V$, we have $\balance(v)

1439:         = 0$ and $|| \openSets(v) || > 0$.  That is, $v$ is a critical

1440:         vertex, since it is not dominated by all three sets~$D_1$, $D_2$,

1441:         and $D_3$ contained in the current~$\mathcal{P}$, and there are

1442:         as many open neighbors as open sets left for it. Note that this

1443:         is the case for each vertex $v$ with $\degree(v) = 2$ and

1444:         $N[v] \cap R \not= \emptyset$, as $v$ and its two neighbors

1445:         have to be assigned to three different dominating sets. We

1446:         select one of the at most three vertices left in $N[v] \cap

1447:         R$, say~$u$, and call function $\assign(G, \mathcal{P},

1448:         \mathcal{A}, u, i)$ for all $i$ with $u \notin A_i$.

1449: \end{construction}

1450: 

1451: Function $\handleCriticalVertex$ deals with the latter three of these

1452: trivial cases.  After they have been ruled out, one of the remaining

1453: vertices $v \in R$ is selected and assigned to one of the three

1454: sets~$D_i$, under the constraint that a vertex $v \in R$ cannot be

1455: added to $D_i$ if it is already a member of~$A_i$.  This case occurs

1456: whenever the recursion returns

1457: %

1458: because no three dominating sets could be found with this

1459: combination.  The recursion continues by calling $\assign(G,

1460: \mathcal{P}, \mathcal{A}, v, i)$, which adds $v$ to~$D_i$, and then

1461: calls $\dominate(G, \mathcal{P}, \mathcal{A})$.

1462: %

1463: %

1464: If no three dominating sets are found by this choice, we remember this

1465: by adding $v$ to the set~$A_i$. A final call to $\dominate$ is made

1466: without assigning a vertex to one potential dominating set~$D_i$.  If

1467: this call fails, the recursion returns to the previous level.  This

1468: completes the description of the algorithm.  We now argue that it is

1469: correct and estimate its running time.

1470: 

1471: To see that the algorithm works correctly, note that it outputs three

1472: sets~$D_1$, $D_2$, and~$D_3$ only if they each are dominating sets

1473: of~$G$.  It remains to prove that these sets are definitely found in

1474: the recursion tree.  All drop-backs within the recursion occur when,

1475: for the current $\mathcal{P} = (D_1, D_2, D_3, R)$, we have

1476: $\balance(v) < 0$ for some vertex $v \in V$.  Thus,

1477: %

1478: %

1479: $\mathcal{P}$ cannot be modified so as to contain a correct partition

1480: into three dominating sets on this branch of the recursion tree.

1481: Since the algorithm checks every possible partition of $G$ into three

1482: sets, unless it is stopped by such a drop-back, a partition into three

1483: dominating sets will be found, if it exists.  If the algorithm does

1484: not find three dominating sets, it eventually terminates when

1485: returning from the first recursive call of function $\dominate$. It

1486: reports the failure, and thus always yields the correct output.

1487: 

1488: To estimate the running time of the algorithm, an important

1489: observation is that the recalculation of the gaps takes no more than

1490: quadratic time in~$n$, the number of vertices of the graph $G$. Thus,

1491: in terms of the $\tilde{\bigoh}$-notation, the running time of the

1492: algorithm depends solely on the number of recursive calls.  Let $T(m)$

1493: be the number of steps of the algorithm, where $m$ is the number of

1494: potential dominating sets left for all vertices that have not been

1495: selected as yet. Initially, every vertex may be a member of any of the

1496: three dominating sets to be constructed (if they exist), hence $m =

1497: 3n$.

1498: 

1499: There are two scenarios where the algorithm calls function $\dominate$

1500: recursively. If $\handleCriticalVertex$ detects a vertex $v \in V$ as

1501: being critical, it selects a vertex $u \in N[v] \cap R$ and calls

1502: function $\assign$ (and thus $\dominate$) for each $i$ with $u \notin

1503: A_i$. Since every critical vertex $v \in V$ remains critical as long

1504: as $N[v] \cap R \not= \emptyset$, function $\handleCriticalVertex$

1505: will be called until all vertices in $N[v] \cap R$ have been assigned

1506: to any of~$D_1$, $D_2$, and~$D_3$.  Since $|| \openSets(v) || \leq 3$,

1507: at most three vertices in the closed neighborhood of $v$ have not been

1508: assigned when $v$ turns critical.  By Proposition~\ref{prop:critical},

1509: all vertices in $N[v] \cap R$ have to be assigned to different

1510: dominating sets.  If $|| \openNeighbors(v) || = 3$, we have at most

1511: six combinations; if we have two open neighbors for a critical vertex,

1512: there are at most two combinations left; and finally, for one open

1513: neighbor $u \in N[v] \cap R$, there remains only one possible choice

1514: to assign $u$ to one of the sets~$D_1$, $D_2$, and~$D_3$. Thus, in the

1515: worst case, we have $T(m) \leq 6 T(m-6)$, as we will handle three

1516: vertices for which at least two choices for dominating sets are left.

1517: With $m = 3n$, it follows that $T(m) \leq 6^{m/6} = 6^{n/2}$, i.e.,

1518: $T(m) = \tilde{\bigoh}(2.4495^n)$.

1519: 

1520: The only other branching into two different recursive calls happens in

1521: the main body of function $\dominate$, when selecting a vertex $v$

1522: with the currently highest maximum gap with respect to $\mathcal{P}$

1523: and~$\mathcal{A}$.  Two cases might occur.  On the one hand, we might

1524: have considered a correct dominating set $D_i$ for~$v$. If $v$ had not

1525: been looked at so far, i.e., if $v$ is not contained in any set $A_j$,

1526: $1 \leq j \leq 3$, $j \neq i$, we have eliminated all three possible

1527: sets for $v$ to belong to.  Thus, in this case, $T(m) = T(m -3)$. On

1528: the other hand, if the algorithm returns from the recursion and thus

1529: did not make the right choice for~$v$, we have $T(m) = T(m - 1)$,

1530: since $v$ is added to~$A_i$, and function $\dominate$ is called

1531: without assigning any vertex.  Summing up, we have $T(m) \leq T(m-1) +

1532: T(m-3)$.  In the second case, we have already visited vertex $v$ in a

1533: previous stage of the algorithm and unsuccessfully tried to assign it

1534: to some set~$D_j$, with $1 \leq j \leq 3$. There are only two

1535: dominating sets for $v$ left. Either way, if we put~$v$ into the

1536: correct dominating set right away or fail the first time, we have

1537: $T(m) = T(m-2)$. Summing up both cases, we have $T(m) \leq 2

1538: T(m-2)$. Suppose that the first and the second case occur equally

1539: often,

1540: %

1541: i.e., the algorithm considers every vertex twice.  It then follows that

1542: \[

1543: T(m) \leq \frac{1}{2} (T(m-1) + T(m-3)) + \frac{1}{2} (2T(m-2))

1544: \]

1545: with $m = 3n$. Thus, we have $T(m) = \tilde{\bigoh}(3^n)$, and the

1546: trivial time bound cannot be beaten. To improve this running time, we

1547: have to make sure that the recursion tree will not reach its full

1548: depth, i.e., not all vertices are considered by the algorithm or

1549: function $\handleCriticalVertex$ will be called for a sufficiently

1550: large portion of the vertices.

1551: %

1552: %

1553: %

1554: %

1555: %

1556: %

1557: %

1558: %

1559: %

1560: %

1561: %

1562: %

1563: %

1564: %

1565: %

1566: %

1567: %

1568: %

1569: It is clear that the algorithm has found three dominating sets once

1570: $\area(G) = 3n$ (recall the notions from

1571: Definition~\ref{def:notions-for-algorithm}). By selecting the maximum

1572: gap possible for a partition $\mathcal{P}$, we try to reach this goal

1573: as fast as possible. For every vertex $v \in R$ that we assign to one

1574: of the potential dominating sets $D_i$, $1 \leq i \leq 3$, we increase

1575: $\area(G)$ by $\gap(v,i)$, and additionally we add $(\gap(v,i) - 3)$

1576: to $\surplus(G)$.

1577: 

1578: Since the vertices of degree two are critical, they and their

1579: neighbors can be handled in time $\tilde{\bigoh}(2.4495^n)$, as

1580: argued above.  So assume that $\mindegree(G) \geq 3$.  Then, we have

1581: $\maxgap(G) > 3$ at the start of the algorithm. If this condition

1582: remains to hold for at least $3n/4$ steps, we have reached $\area(G) =

1583: 3n$, and the algorithm terminates successfully.  To make use of more

1584: than $3n/4$ vertices, $\maxgap(G)$ has to drop below four at one point

1585: of the computation.  We exploit the fact that up to this point, the

1586: surplus has grown sufficiently large with respect to~$n$.  Decreasing

1587: it will force $\maxgap(G)$ to drop below three, and 

1588: %

1589: this condition can hold only for a certain portion of the remaining

1590: vertices until the algorithm terminates.  To see this, we now analyze

1591: the remaining steps of the algorithm after the given graph $G$ has

1592: reached a certain maximum gap with respect to the current

1593: $\mathcal{P}$ and~$\mathcal{A}$.

1594: %

1595: %

1596: 

1597: If $\maxgap(G) = 0$, the recursion stops immediately. Either we have

1598: already found three disjoint dominating sets (in which case we put the

1599: remaining vertices $v \in R$ into set $D_1$ and halt), or one vertex

1600: has not been dominated by one set $D_i$ in $\mathcal{P}$ yet. Since no

1601: positive gaps exist for the vertices $v \in R$, $\mathcal{P}$ cannot

1602: be modified to a valid partition into three dominating sets.  Function

1603: $\handleCriticalVertex$ returns true immediately after detecting

1604: $\balance(v) < 0$ for some vertex $v \in V$, and function $\dominate$

1605: drops back one recursion level. The question is how many vertices are

1606: left in $R$ when we reach $\maxgap(G) = 0$.

1607: 

1608: \begin{lemma}

1609: \label{lem:mainproof}

1610: Let $G = (V,E)$ be a graph and $\mathcal{P} = (D_1, D_2, D_3, R)$ be

1611: a partition of $V$ as in Definition~\ref{def:notions-for-algorithm}.

1612: Let $r = || R ||$ and $\maxgap(G) = 3$. Then, for at least $r/64$

1613: vertices in~$R$, the algorithm will not recursively call function

1614: $\dominate$.

1615: \end{lemma}

1616: 

1617: \sproofof{Lemma~\ref{lem:mainproof}}

1618: Let $\maxgap(G) = k$ with $k > 0$.  Since $\gap(v,i) \leq k$ for each

1619: $v \in R$ and for each~$i$, $1 \leq i \leq 3$, we have $\sum_{v \in R}

1620: \sumgap(v) \leq 3kr$.  Every vertex $v$ that is selected for a set

1621: $D_i$ with $\gap(v,i) = k$ decreases at least $k$ gaps of the vertices

1622: in $R - \{ v \}$ by one. Otherwise, $\handleCriticalVertex$ would

1623: have found a critical vertex $u \in N[v]$ with $N[u] \cap R = \{ v

1624: \}$.  Then, either $|| \openSets(u) || > 1$ (which implies

1625: $\balance(u) < 0$ and we abort), or $|| \openSets(u) || = 1$, in which

1626: case $v$ is added to the appropriate set $D_i$ without further

1627: branching of function $\dominate$.  Thus, if no critical vertex is

1628: detected, selecting a vertex $v \in R$ for some set $D_i$ decreases at

1629: least $k$ gaps, and since $v$ does not belong to $R$ anymore,

1630: additionally all gaps previously defined for $v$ are now undefined. So

1631: the lowest possible rate at which the gaps are decreased is related to

1632: the maximum gap of~$G$.

1633: 

1634: Now suppose that $\maxgap(G) = 3$ and $\sumgap(v) = 9$ for all

1635: vertices $v \in R$. We always select a vertex $v$ with the highest

1636: summation gap of all vertices $u \in R$ with $\maxgap(u) = 3$.  As

1637: long as there exists a vertex $v \in R$ with $\gap(v,i) = 3$ for

1638: all~$i$, it will be selected by the algorithm.  After calling function

1639: $\recalculateGaps$, the number of gaps equal to three will be

1640: decreased at least by six. If exactly three other gaps of vertices in

1641: $R - \{v\}$ decrease by one in every step, it takes at least $r/4$

1642: vertices until $\sumgap(v) < 9$ for all $v \in R$.  Another $1/4$ of

1643: the $3r/4$ vertices remaining have to be selected until $\sumgap(v) <

1644: 8$. Adding $1/4$ of the $9r/16$ vertices left in~$R$, we have reached

1645: $\maxgap(G) = 2$ with $\sumgap(v) = 6$ for all vertices $v \in R$.

1646: This implies that every defined gap is equal to two. Summing up, we

1647: have selected

1648: \[

1649: \frac{1}{4} \cdot r + \frac{1}{4} \cdot \frac{3}{4} r +

1650:  \frac{1}{4} \cdot \frac{9}{16} r = \frac{37}{64} r

1651: \]

1652: vertices until $\maxgap(G) = 2$, under the constraint that a minimum

1653: number of gaps is reduced in each step, while simultaneously trying to

1654: reduce the maximum summation gap in the fastest possible way.  This

1655: way we reach level $\maxgap(G) = 0$ with as few vertices left in $R$

1656: as possible, which describes the worst case that might happen.

1657: 

1658: Analogously, we can show that $\maxgap(G)$ drops from $2$ to~$1$ after

1659: selecting another $19r/64$ vertices.  And once we have $\maxgap(G) =

1660: 1$, it takes $7r/64$ vertices to get to $\maxgap(G) = 0$.  Now, there

1661: are $r/64$ vertices remaining in~$R$, which do not have to be

1662: processed recursively.

1663: \eproofof{Lemma~\ref{lem:mainproof}}

1664: 

1665: %

1666: %

1667: %

1668: %

1669: %

1670: %

1671: %

1672: %

1673: %

1674: %

1675: %

1676: %

1677: %

1678: %

1679: %

1680: %

1681: %

1682: %

1683: %

1684: %

1685: %

1686: %

1687: %

1688: %

1689: %

1690: %

1691: %

1692: %

1693: %

1694: %

1695: 

1696: Continuing the proof of Theorem~\ref{the:detalgo}, note that

1697: we assumed $\mindegree(G) \geq 3$, so when the gaps are initialized

1698: for graph~$G$, we have $\mingap(v) \geq 4$ for each

1699: vertex $v \in V$.  Thus, more than three vertices

1700: are dominated by the selected set $D_i$ for vertex~$v$.

1701: As long as $\maxgap(G) > 3$ is true, $\surplus(G)$ is

1702: increasing. The only way to lower the surplus is by adding vertices

1703: $v$ to a set $D_i$ with $\gap(v,i) < 3$.  The surplus decreases by one

1704: when $\gap(v,i) = 2$, and it decreases by two when $\gap(v,i)~=~1$.

1705: 

1706: %

1707: %

1708: %

1709: %

1710: %

1711: %

1712: %

1713: Let $S = \surplus(G)$ be the surplus

1714: collected for a partition $\mathcal{P}$ until we reach a point where

1715: $\maxgap(G) = 3$. To make use of the most recursive calls and to even

1716: out the surplus completely, there have to be at least $r = || R ||$

1717: vertices remaining with

1718: \[

1719: 0 \cdot \frac{37r}{64} + 1 \cdot \frac{19r}{64}

1720:  + 2 \cdot \frac{7r}{64} = S,

1721: \]

1722: so $r \geq 64S/33$. A fraction of $1/64$ of these vertices will be

1723: handled by the algorithm without branching into more than one

1724: recursive call, which is at least~$S/33$.

1725: %

1726: %

1727: %

1728: %

1729: %

1730: %

1731: %

1732: %

1733: %

1734: %

1735: %

1736: %

1737: %

1738: %

1739: %

1740: %

1741: %

1742: %

1743: %

1744: %

1745: The question is how big the surplus $S$ might grow and how many vertices

1746: are left in $R$ before $\maxgap(G) = 3$ is reached. The lowest surplus

1747: with as few vertices in $R$ as possible occurs if $\mindegree(G) =

1748: \maxdegree(G) = 3$. Surplus $S$ is increased by one in each step

1749: %

1750: until we arrive at $\maxgap(G) = 3$.  When selecting a vertex $v$ of

1751: degree~$3$ for a set~$D_i$, the gap of its neighbors $u \in N(v)$

1752: and the gaps of the neighbors of every $u$ might be decreased. Summing

1753: up, at most $1 + 3 + 3 \cdot 2 = 10$ vertices can have decreased their

1754: gaps for some~$i$.  After selecting at least $n/10$ vertices for each~$i$,

1755: we have $\mingap(G) = 3$ (in the worst case). From this point on,

1756: we cannot be sure if the next vertex selected for some $D_i$ satisfies

1757: $\gap(v,i) > 3$. But so far we have already collected a surplus of

1758: $S = 3n/10$, and applying this we obtain $64n/110 \leq r \leq 7n/10$.

1759: Thus, for at least $n/110$ vertices we never branch into two different

1760: recursive calls.  Setting $m = 3 (109n/110)$, we obtain a running time

1761: of~$\tilde{\bigoh}(2.9416^n)$.

1762: %

1763: %

1764: %

1765: %

1766: %

1767: %

1768: %

1769: %

1770: %

1771: %

1772: %

1773: %

1774: %

1775: %

1776: %

1777: %

1778: %

1779: %

1780: %

1781: %

1782: %

1783: %

1784: %

1785: %

1786: %

1787: %

1788: %

1789: %

1790: %

1791: %

1792: %

1793: %

1794: %

1795: %

1796: %

1797: %

1798: %

1799: %

1800: %

1801: %

1802: %

1803: %

1804: %

1805: %

1806: %

1807: %

1808: %

1809: %

1810: %

1811: %

1812: %

1813: %

1814: %

1815: %

1816: %

1817: %

1818: %

1819: %

1820: %

1821: %

1822: %

1823: %

1824: %

1825: %

1826: %

1827: %

1828: %

1829: %

1830: %

1831: %

1832: %

1833: %

1834: %

1835: %

1836: %

1837: %

1838: %

1839: %

1840: %

1841: %

1842: %

1843: %

1844: %

1845: %

1846: %

1847: %

1848: %

1849: %

1850: %

1851: %

1852: %

1853: %

1854: %

1855: %

1856: %

1857: %

1858: %

1859: %

1860: %

1861: %

1862: %

1863: %

1864: %

1865: %

1866: %

1867: \end{proofs}

1868: 

1869: %

1870: 

1871: %

1872: 

1873: \section{Graphs with Bounded Maximum Degree}

1874: \label{sec:bounded-max-degree}

1875: 

1876: As seen in the last section, the running time of the algorithm

1877: crucially depends on the degrees of the vertices of~$G$.  If we

1878: restrict ourselves to graphs $G$ with bounded maximum degree (say

1879: $\Delta = \maxdegree(G)$), we can optimize our strategy in finding

1880: three disjoint dominating sets.  In this section, we present a simple

1881: deterministic algorithm, which has a better running time than the

1882: algorithm from Theorem \ref{the:detalgo}, provided that $\Delta$ is low.

1883: By using randomization, we can further improve the running time for

1884: graphs $G$ with low maximum degree.

1885: 

1886: Before stating the two results, note that graphs with maximum degree

1887: two can trivially be partitioned into three dominating sets, if such a

1888: partition exists.  Every component of such a graph is either an

1889: isolated vertex, a path, or a cycle, and each such property can be

1890: recognized in polynomial time.

1891: 

1892: \begin{proposition}

1893: Let $G = (V, E)$ be a given graph with $\maxdegree(G) = 2$. There

1894: exists a partition of the vertices of $G$ into three dominating sets

1895: if and only if every component of $G$ is a cycle of length~$k$ such

1896: that $3$ divides~$k$.

1897: \end{proposition}

1898: 

1899: We use the terms from Definition~\ref{def:notions-for-algorithm} in

1900: Section~\ref{sec:algo} to describe a snapshot within the algorithm.

1901: For any partition $\mathcal{P} = (D_1, D_2, D_3, R)$, some vertices of

1902: $V$ have already been assigned to the potential dominating sets $D_1$,

1903: $D_2$, and~$D_3$, while all the remaining vertices are in~$R$. The

1904: auxiliary sets $\mathcal{A} = (A_1, A_2, A_3)$ will not be needed in

1905: this section.  Only connected graphs are considered, as it is possible

1906: to treat every connected component separately, producing the desired

1907: output within the same time bounds.

1908: 

1909: Table~\ref{tab:results} lists the running times of both the

1910: deterministic and the random algorithm, where the maximum degree of

1911: the input graph is bounded by~$\Delta$, $3 \leq \Delta \leq 8$. Note

1912: that the exact deterministic algorithm from Theorem~\ref{the:detalgo}

1913: in Section~\ref{sec:algo} beats the deterministic algorithm from

1914: Theorem~\ref{the:det-bounded} whenever $\Delta \geq 7$.

1915: 

1916: \begin{table}[!th]

1917: \centering

1918: \begin{tabular}{|c|c|c|c|c|c|c|}

1919: \hline

1920: $\Delta$ & 3 & 4 & 5 & 6 & 7 & 8 \\ \hline

1921: deterministic & $2.2894^n$ & $2.6591^n$ & $2.8252^n$ &

1922:  $2.9058^n$ & $2.9473^n$ & $2.9697^n$ \\ \hline

1923: randomized & $2^n$ & $2.3570^n$ & $2.5820^n$ &

1924:  $2.7262^n$ & $2.8197^n$ & $2.8808^n$ \\ \hline

1925: \end{tabular}

1926: \caption[Summary]{\label{tab:results} Results for

1927:  $\maxdegree(G) = k$, where $3 \leq k \leq 8$}

1928: \end{table}

1929: 

1930: \begin{theorem}

1931: \label{the:det-bounded}

1932: Let $G = (V, E)$ be a graph with $\maxdegree(G) = \Delta$, where

1933: $\Delta \geq 3$.  There exists a deterministic algorithm solving the

1934: three domatic number problem in time

1935: $\tilde{\bigoh}(d^{\frac{n}{\Delta}})$, where

1936: \begin{equation}

1937: \label{equ:d}

1938: d = \sum\limits_{a=0}^{\Delta-2} \left[ {\Delta \choose a}

1939:   \sum\limits_{b=1}^{\Delta-a-1} {\Delta-a \choose b} \right] .

1940: \end{equation}

1941: \end{theorem}

1942: 

1943: \begin{proofs}

1944: The algorithm works as follows. We start with an arbitrary vertex $v

1945: \in V$ and assign it to the first set~$D_1$. In each step, we first

1946: check whether we found a partition $\mathcal{P} = (D_1, D_2, D_3, R)$

1947: into dominating sets $D_1$, $D_2$, and $D_3$. If not, one vertex

1948: $v \in V$ is selected that is not dominated by all three sets $D_1$,

1949: $D_2$, and $D_3$, and additionally has a vertex $u \in N[v]$ in its

1950: closed neighborhood that has already been added to some set~$D_i$, $1

1951: \leq i \leq 3$. It follows that $1 \leq || \openSets(v) || \leq 2$.

1952: 

1953: If $\balance(v) < 0$, we return within the recursion. Otherwise, we

1954: try all combinations to partition the vertices in $N[v] \cap R$,

1955: so that after this step vertex $v$ is dominated by all three

1956: potential dominating sets.  If no such combination leads to a valid

1957: partition, we again return within the recursion.

1958: 

1959: Suppose now that $\balance(v) \geq 0$, $|| \openSets(v) || = 2$,

1960: and $N[v] \cap D_1 \not= \emptyset$. To obtain three disjoint

1961: dominating sets, at least one vertex in $N[v]$ has to be assigned

1962: to~$D_2$, and at least one vertex in $N[v]$ has to be added

1963: to~$D_3$. This limits our choices, especially if the degree of $v$ is

1964: bounded by some constant~$\Delta$.

1965: 

1966: To measure the running time of the algorithm, we consider the worst

1967: case with the most possible combinations that might yield a partition

1968: into three dominating sets.  This occurs when only one vertex $u \in

1969: N[v]$ has already been added to one set, i.e., $|| N[v] \cap (D_1 \cup

1970: D_2 \cup D_3) || = 1$.  If $N[v] \cap D_1 \not = \emptyset$, then any

1971: number between $0$ and $\Delta-2$ of vertices in $N[v] \cap R$ may be

1972: assigned to the same set~$D_1$.  Let this number be~$a$. It follows

1973: that from one to $\Delta-a-1$ vertices remaining in $N[v] \cap R$ are

1974: allowed to be in the next potential dominating set~$D_2$.  This is how

1975: Equation~\ref{equ:d} for $d$ is derived.  After assigning the last

1976: vertices in $N[v] \cap R$ to the dominating set~$D_3$, exactly $\Delta$

1977: vertices have been removed from~$R$.  Thus, we have a worst case

1978: running time of $\tilde{\bigoh}(d^{\frac{n}{\Delta}})$.

1979: Table~\ref{tab:results} lists the running time for graphs with maximum

1980: degree from three to nine.

1981: \end{proofs}

1982: 

1983: In the next theorem, randomization is used to speed up this

1984: procedure. Instead of assigning all vertices in the closed

1985: neighborhood of some vertex $v \in V$ in one step, only one

1986: or two vertices in $N[v] \cap R$ are added to the potential

1987: dominating sets $D_1$, $D_2$, and $D_3$. The goal is to dominate

1988: one vertex by all three sets in one step. We will select the one

1989: or two vertices that are missing for this goal at random.

1990: 

1991: \begin{theorem}

1992: \label{the:rand-bounded}

1993: Let $G = (V, E)$ be a graph with $\maxdegree(G) = \Delta$, where

1994: $\Delta \geq 3$, and let $d$ be defined as in Equation~(\ref{equ:d})

1995: in Theorem~\ref{the:det-bounded}.  For each constant $c > 0$, there

1996: exists a randomized algorithm solving the three domatic number problem

1997: with error probability at most $e^{-c}$ in time

1998: $\tilde{\bigoh}(r^{\frac{n}{2}})$, where

1999: \begin{equation}

2000: \label{equ:r}

2001: r = \frac{d}{3^{\Delta-2}}.

2002: \end{equation} 

2003: \end{theorem}

2004: 

2005: \begin{proofs}

2006: Let graph $G = (V,E)$ be given with $\maxdegree(G) = k$.

2007: As in the deterministic algorithm, we start by adding a random

2008: vertex to the set $D_1$. In every following step, a vertex

2009: $v \in V$ is selected with $0 < || \openSets(v) || < 3$, so it

2010: is $N[v] \cap (D_1 \cup D_2 \cup D_3) \not= \emptyset$.

2011: If $|| \openSets(v) || = 1$, we have $N[v] \cap D_i = \emptyset$ for

2012: one $i$ with $1 \leq i \leq 3$. We randomly choose a vertex

2013: $u \in N[v] \cap R$ and assign it to set $D_i$, in order that

2014: $v$ is dominated by all three sets afterwards. If $|| \openSets(v) ||

2015:  = 2$, we randomly select two vertices $u_1, u_2 \in R$ in the

2016: closed neighborhood of $v$. Another random choice is made when

2017: deciding how to distribute these two vertices among the two

2018: potential dominating sets that have not dominated $v$ up to now.

2019: 

2020: Suppose $G$ indeed has a partition into three dominating

2021: sets. We have to measure the error rate when making our random

2022: choices to estimate the success probability of the algorithm.

2023: In every step, a vertex $v \in V$ is selected with at least

2024: one vertex $u \in N[v]$ in its closed neighborhood that has

2025: already been added to one of the sets $D_1$, $D_2$, or $D_3$.

2026: The highest error occurs when exactly one vertex in $N[v]$ is

2027: not included in $R$, so we restrict our analysis to this case.

2028: To obtain a valid partition into three dominating sets, there

2029: are at most $d$ choices left to partition the vertices

2030: remaining in $N[v] \cap R$. Here, $d$ is the number from

2031: Equation \ref{equ:d}. Once we selected and assigned two

2032: vertices from $N[v] \cap R$ at random, there are

2033: $3^{k-2}$ possibilities left to partition the vertices

2034: in the closed neighborhood of $v$ that are still left in $R$.

2035: Our success rate when selecting the two vertices is

2036: therefore $3^{k-2}/d$.

2037: 

2038: To achieve an error probability of below $e^{-c}$, the

2039: algorithm needs to be executed more than once. The repetition

2040: number of the algorithm equals the reciprocal of

2041: the success rate, which explains Equation \ref{equ:r}.

2042: Since two vertices are processed in every step, the overall

2043: running time is $\tilde{\bigoh}(r^{\frac{n}{2}})$.

2044: \end{proofs}

2045: 

2046: \section{Conclusion}

2047: \label{sec:conclusion}

2048: We have shown that the three domatic number problem can be solved

2049: by a deterministic algorithm in time~$\tilde{\bigoh}(2.9416^n)$.

2050: Furthermore, we presented two algorithms solving the three domatic

2051: number problem for graphs with bounded maximum degree, improving

2052: the above time bound for graphs with small maximum degree. Although

2053: our running times seem to be not too big of an improvement of the

2054: trivial~$\tilde{\bigoh}(3^n)$ bound, they are to our knowledge the

2055: first such algorithms breaking this barrier. For $k > 3$, the decision

2056: problem of whether $\delta(G) \geq k$ can be solved in

2057: time~$\tilde{\bigoh}(3^n)$ by Lawler's dynamic programming

2058: algorithm for the chromatic number, appropriately modified for the domatic

2059: number problem.  Therefore, it would not be reasonable to use our gap

2060: approach of Section~\ref{sec:algo} to decide if $\delta(G) \geq k$

2061: for a graph $G$ and $k > 3$.

2062: 

2063: \medskip

2064: 

2065: \noindent

2066: {\bf Acknowledgement.} We thank Dieter Kratsch for pointing us to

2067: Lawler's algorithm.

2068: %

2069: 

2070: \bibliographystyle{alpha}

2071: 

2072: \bibliography{/home/rothe/BIGBIB/joergbib}

2073: %

2074: %

2075: %

2076: 

2077: 

2078: \clearpage

2079: 

2080: \appendix

2081: 

2082: \section{Proof of Proposition~\ref{prop:mds-dnp}}

2083: 

2084: \begin{proofs}

2085:   Figure~\ref{fig:ex1} shows the graphs $G$ and $H$ whose existence is

2086:   claimed.  In this figure, the numbers $i|j$ within a vertex have the

2087:   following meaning: $i$ indicates which dominating set~$D_i$ this vertex

2088:   belongs to in a fixed partition into three dominating sets, and $j$

2089:   indicates a specific choice of a minimum dominating set $S$ of the graph by

2090:   setting $j = 1$ if and only if this vertex belongs to~$S$.

2091: 

2092: \begin{figure}[!th]

2093: \begin{center}

2094: \input{example1.eepic}

2095: \end{center}

2096: \caption[Counterexamples]{\label{fig:ex1} Graphs $G$ and $H$ for

2097:   Proposition~\ref{prop:mds-dnp}}.

2098: \end{figure}

2099: For the first assertion, look at the graph~$G$ shown on the left-hand side of

2100: Figure~\ref{fig:ex1}.

2101: %

2102: %

2103: %

2104: %

2105: %

2106: %

2107: Note that $\gamma(G) = 2$. In particular, $D = \{u_3, u_5\}$ is a minimum

2108: dominating set of~$G$.  Note further that $\delta(G) = 3$.  In particular, a

2109: partition into three dominating sets of $G$ is given by

2110: $D_1 = \{u_1, u_4, u_7\}$,

2111: $D_2 = \{u_2,u_5\}$, and

2112: $D_3 = \{u_3,u_6\}$.

2113: However, $D$ cannot be part of any partition into three dominating sets, since

2114: the only neighbors of~$u_4$, namely $u_3$ and~$u_5$, belong to~$D$.

2115: 

2116: Note that the minimum dominating set $D_2 = \{u_2,u_5\}$ of $G$ defined above

2117: indeed is part of a partition into three dominating sets.  The second part of

2118: the proposition, however, shows that this is not always the case.  Consider

2119: the graph $H = (V,E)$ shown on the right-hand side of Figure~\ref{fig:ex1}.

2120: %

2121: %

2122: %

2123: %

2124: %

2125: %

2126: %

2127: %

2128: We have $\gamma(H) = 2$ by choosing the minimum dominating set $D = \{ v_1,

2129: v_2 \}$, which is unique in this case.  Again, $\delta(H) = 3$. The only way,

2130: up to isomorphism, to partition the vertex set of $H$ into three dominating

2131: sets is given by

2132: $D_1 = \{v_1, v_7, v_8\}$,

2133: $D_2 = \{v_2, v_6, v_9\}$, and

2134: $D_3 = \{v_3, v_4, v_5\}$. 

2135: Thus, $\min \{||D_1||, ||D_2||, ||D_3||\} > \gamma(H)$ for each partition into

2136: three dominating sets.

2137: \end{proofs}

2138: 

2139: \section{Pseudo-Code of the Algorithm from Theorem~\ref{the:detalgo}}

2140: 

2141: Figures~\ref{fig:dnp}, \ref{fig:domatic}, \ref{fig:assign},

2142: \ref{fig:recalc}, and~\ref{fig:handle} describe the algorithm from

2143: Theorem~\ref{the:detalgo} in pseudo-code.

2144: 

2145: \begin{figure}[!ht]

2146: \begin{construction}

2147: \item {\bf Algorithm for the Three Domatic Number Problem}

2148: \begin{block}

2149:   \item {\bf Input:} Graph $G = (V,E)$ with vertex set $V = \{v_1, v_2, \ldots,

2150:     v_n \}$  and edge set $E$

2151:   \item {\bf Output:} Partition of $V$ into three dominating sets

2152:    $D_1, D_2, D_3 \seq V$ or ``failure''

2153: \end{block}

2154: \begin{block}

2155: %

2156: %

2157:     \item Set each of $D_1$, $D_2$, $D_3$, $A_1$, $A_2$,

2158:       and $A_3$ to the empty set; 

2159:     \item Set $R = V$;

2160:                 \item Set $\mathcal{P} = (D_1, D_2, D_3, R)$;

2161:                 \item Set $\mathcal{A} = (A_1, A_2, A_3)$;

2162:     \item $\dominate(G,\mathcal{P}, \mathcal{A})$;

2163:           \hfill $\slash \slash$  Start recursion

2164:     \item {\bf output} ``failure'' and {\bf halt};

2165: \end{block}

2166: \end{construction}

2167: 

2168: \caption[Algorithm]{\label{fig:dnp} Algorithm for the Three Domatic

2169: Number Problem}

2170: \end{figure}

2171: 

2172: \begin{figure}[!ht]

2173: \begin{construction}

2174: \item {\bf Function} $\dominate(G,\mathcal{P},\mathcal{A})\, \{$ 

2175:       \hfill $\slash \slash$ $\mathcal{P}$ is a partition of graph $G$, 

2176: \item \hfill $\slash \slash$ $\mathcal{A}$ is a triple of auxiliary sets

2177: \begin{block}

2178:   \item $\recalculateGaps(G, \mathcal{P}, \mathcal{A})$;

2179:   \item {\bf if} $($each $D_i$ is a dominating set$)\, \{$

2180:   \begin{block}

2181:     \item $D_1 = D_1 \cup R$;

2182:     \item {\bf output} $D_1, D_2, D_3$;

2183:   \end{block}

2184:   \item $\}$

2185:         \item {\bf if} $(~\mbox{not}~

2186:                \handleCriticalVertex(G, \mathcal{P}, \mathcal{A}))\, \{$

2187:   \begin{block}

2188:                 \item select vertex $v \in R$ with

2189:                 \begin{block}

2190: %

2191:       \item $\maxgap(v) = \maxgap(G)$ and

2192:                         \item $\sumgap(v) = \max \{ \sumgap(u) \condition

2193:                         u \in R \wedge \maxgap(u) = \maxgap(G)\}$;

2194:     \end{block}

2195:                 \item find $i$ with $\gap(v,i) = \maxgap(v)$;

2196:     \item $\assign(G, \mathcal{P}, \mathcal{A}, v, i)$;

2197:                                 \hfill $\slash \slash$ First recursive call

2198:     \item $A_i = A_i \cup \{ v \}$;

2199:               \hfill $\slash \slash$ If recursion fails, put $v$ in 

2200:                                      $A_i$ and try again 

2201:     \item $\dominate(G, \mathcal{P}, \mathcal{A})$;

2202:                         \hfill $\slash \slash$ Second recursive call

2203:   \end{block}

2204:   \item $\}$

2205:   \item {\bf return};

2206: \end{block}

2207: \item $\}$

2208: \end{construction}

2209: 

2210: \caption[Algorithm]{\label{fig:domatic} Recursive function to

2211: dominate graph $G$}

2212: \end{figure}

2213: 

2214: \begin{figure}[!ht]

2215: \begin{construction}

2216: \item {\bf Function} $\assign(G, \mathcal{P}, \mathcal{A}, v, i)\, \{$ 

2217: \begin{block}

2218:   \item $D_i = D_i \cup \{ v \}$;

2219:   \item $R = R - \{ v \}$;

2220:   \item $\dominate(G, \mathcal{P}, \mathcal{A})$;

2221: \end{block}

2222: \item $\}$

2223: \end{construction}

2224: 

2225: \caption{\label{fig:assign} Function to assign vertex $v$ to set $D_i$} 

2226: \end{figure}

2227: 

2228: \begin{figure}[!ht]

2229: \begin{construction}

2230: \item {\bf Function} $\recalculateGaps(G, \mathcal{P}, \mathcal{A})\, \{$ 

2231:       \hfill $\slash \slash$ $\mathcal{P}$ is a partition of graph $G$, 

2232: \item \hfill $\slash \slash$ $\mathcal{A}$ is a triple of auxiliary sets

2233: \begin{block}

2234:   \item {\bf for all} $($vertices $v \in V)\, \{$

2235:         \begin{block}

2236:     \item {\bf if} $($vertex $v \in R)\, \{$

2237:     \begin{block}

2238:       \item {\bf for all} $(i = 1, 2, 3)\, \{$

2239:       \begin{block}

2240:         \item {\bf if} $(v \notin A_i)\, \{$

2241:        $\gap(v,i) = || N[v] || - || \{ u \in N[v] \condition

2242:               (\exists w \in N[u])\, [w \in D_i] \} ||$; $\}$

2243:         \item {\bf else} $\gap(v,i) = \bot$ ;

2244:          \hfill $\slash \slash$ $\bot$ indicates that $\gap(v,i)$ is undefined

2245:       \end{block}

2246:       \item $\}$

2247:       \item $\maxgap(v) = \max_{i \in \{1,2,3\}} \{ \gap(v,i) \}$;

2248:       \item $\mingap(v) = \min_{i \in \{1,2,3\}} \{ \gap(v,i) \}$;

2249:       \item $\sumgap(v) = \sum_{i \in \{1,2,3\}} \gap(v,i)$;

2250:     \end{block}

2251:                 \item $\}$

2252:                 \item $\openNeighbors(v) = \{ u \in  N[v] \condition u \in R \}$;

2253:                 \item $\openSets(v) = \{ i \in \{ 1,2,3 \} \condition

2254:                   v \notin N[D_i] \} $;

2255:     \item $\balance(v) = || \openNeighbors(v) || - || \openSets(v) ||$;

2256:         \end{block}

2257:   \item $\}$

2258:   \item $\maxgap(G) = \max_{v \in R} \{ \maxgap(v) \}$;

2259:   \item $\mingap(G) = \min_{v \in R} \{ \mingap(v) \}$;

2260: \end{block}

2261: \item $\}$

2262: \end{construction}

2263: 

2264: \caption{\label{fig:recalc} Function to recalculate gaps

2265:   after partition has changed} 

2266: \end{figure}

2267: 

2268: \begin{figure}[!ht]

2269: \begin{construction}

2270: \item {\bf Function} boolean 

2271:     $\handleCriticalVertex(G, \mathcal{P}, \mathcal{A})\, \{$

2272: \begin{block}

2273:   \item {\bf for all} $($vertices $v \in V)\, \{$

2274:   \begin{block}

2275:     \item {\bf if} $( \balance(v) < 0 )$ \{ \hfill $\slash \slash$

2276:          impossible to three dominate $v$ with $\mathcal{P}$

2277:                 \begin{block}

2278:       \item {\bf return} true;

2279:     \end{block}

2280:                 \item \} {\bf else if}

2281:                   $(|| \{ i \in \{ 1,2,3 \} \condition v \in A_i \} || == 2)$ \{

2282:         \hfill $\slash \slash$ one choice for $v$ remaining

2283:                 \begin{block}

2284:                   \item select $i$ with $v \notin A_i$;

2285:                   \item $\assign(G, \mathcal{P}, \mathcal{A}, v, i)$;

2286:       \item {\bf return} true;

2287:                 \end{block}

2288:                 \item \} {\bf else if}

2289:                     $(\balance(v) == 0 ~\mbox{and}~ || \openSets(v) || > 0)$ \{

2290:         \hfill $\slash \slash$ $v$ is critical

2291:                 \begin{block}

2292:                   \item select $u \in N[v] \cap R$;

2293:                         \item {\bf for all} $(i$ with $u \notin A_i$ and

2294:                             $v$ not dominated by $D_i)$

2295:                         \begin{block}

2296:                           \item $\assign(G, \mathcal{P}, \mathcal{A}, u, i)$;

2297:                         \end{block}

2298:       \item {\bf return} true;

2299:                 \end{block}

2300:                 \item \}

2301:   \end{block}

2302:   \item $\}$

2303:         \item {\bf return} false;

2304:         \hfill $\slash \slash$ no critical vertices were found

2305: \end{block}

2306: \item $\}$

2307: \end{construction}

2308: 

2309: \caption{\label{fig:handle} Function to handle the critical vertices}

2310: \end{figure}

2311: 

2312: %

2313: %

2314: %

2315: %

2316: %

2317: %

2318: %

2319: %

2320: %

2321: %

2322: %

2323: %

2324: %

2325: %

2326: %

2327: %

2328: %

2329: %

2330: 

2331: \end{document}

2332: 

2333: 

2334: