1: \documentclass[11pt]{article}

2: 

3: \usepackage{eepic,epic}

4: \usepackage{latexsym}

5: \usepackage{amsfonts}

6: \usepackage{amsmath}

7: \usepackage{amssymb}

8: 

9: \bibliographystyle{alpha}

10: 

11: \emergencystretch=2em

12: 

13: %

14: 

15: %

16: \makeatletter%

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

18: \makeatother%

19: 

20: %

21: %

22: %

23: %

24: %

25: %

26: \makeatother%

27: 

28: %

29: %

30: %

31: %

32: 

33: %

34: 

35: \makeatletter%

36: 

37: %

38: %

39: %

40: %

41: \newcount\hour  \newcount\minutes  \hour=\time  \divide\hour by 60

42: \minutes=\hour  \multiply\minutes by -60  \advance\minutes by \time

43: \def\mmmddyyyy{\ifcase\month\or Jan\or Feb\or Mar\or Apr\or May\or Jun\or Jul\or

44:   Aug\or Sep\or Oct\or Nov\or Dec\fi \space\number\day, \number\year}

45: \def\hhmm{\ifnum\hour<10 0\fi\number\hour :%

46:   \ifnum\minutes<10 0\fi\number\minutes}

47: \def\Draft{{\it Draft of \mmmddyyyy}}

48: 

49: %

50: %

51: \topsep 8pt plus2pt minus4pt   %

52: 

53: %

54: %

55: %

56: 

57: %

58: %

59: %

60: 

61: %

62: %

63: %

64: 

65: %

66: %

67: %

68: %

69: 

70: %

71: %

72: %

73: %

74: %

75: %

76: %

77: %

78: %

79: %

80: %

81: %

82: %

83: %

84: %

85: %

86: %

87: %

88: %

89: %

90: %

91: %

92: %

93: %

94: %

95: \def\ps@jtsheadings{%

96: \def\@oddhead{\it\rightmark\hfil\rm\thepage}%

97: \def\@oddfoot{\hfil\Draft}%

98: \if@twoside%

99: \def\@evenhead{\rm\thepage\hfil\it\leftmark}%

100: \def\@evenfoot{\Draft\hfil}%

101: \else

102: \let\@evenhead\@oddhead%

103: \let\@evenfoot\@oddfoot%

104: \fi%

105: }

106: \def\ps@jtsplain{%

107: \def\@oddhead{\hfil\Draft}%

108: \def\@oddfoot{\hfil\rm\thepage\hfil}%

109: \let\@evenfoot\@oddfoot%

110: \if@twoside \def\@evenhead{\Draft\hfil} \else \let\@evenhead\@oddhead \fi

111: }

112: 

113: %

114: %

115: %

116: \def\chaptermark#1{\markboth{\thechapter.\ #1}{\thechapter.\ #1}}%

117: \def\sectionmark#1{\markright{\thesection.\ #1}}

118: \def\subsectionmark#1{\markright{\thesubsection.\ #1}}

119: \def\subsubsectionmark#1{\markright{\thesubsubsection.\ #1}}

120: 

121: %

122: %

123: %

124: %

125: \def\section{\@startsection {section}{1}{\z@}

126:     {3.5ex plus1ex minus.2ex}{2.3ex plus.2ex}{\Large\bf}}

127: \def\subsection{\@startsection{subsection}{2}{\z@}

128:     {3.25ex plus1ex minus.2ex}{1.5ex plus.2ex}{\large\bf}}

129: \def\subsubsection{\@startsection{subsubsection}{3}{\z@}

130:     {3.25ex plus1ex minus.2ex}{1.5ex plus.2ex}{\normalsize\bf}}

131: \def\paragraph{\@startsection{paragraph}{4}{\z@}

132:     {3.25ex plus1ex minus.2ex}{1em}{\normalsize\bf}}

133: \def\subparagraph{\@startsection{subparagraph}{4}{\parindent}

134:     {3.25ex plus1ex minus.2ex}{1em}{\normalsize\bf}}

135: 

136: %

137: %

138: %

139: %

140: %

141: %

142: %

143: %

144: %

145: %

146: %

147: %

148: %

149: %

150: %

151: %

152: %

153: %

154: %

155: %

156: %

157: %

158: %

159: %

160: %

161: %

162: %

163: %

164: %

165: %

166: %

167: %

168: %

169: %

170: %

171: %

172: %

173: %

174: %

175: %

176: %

177: %

178: %

179: %

180: %

181: %

182: %

183: %

184: %

185: %

186: %

187: %

188: %

189: %

190: %

191: %

192: \makeatother%

193: 

194: 

195: %

196: %

197: %

198: %

199: %

200: %

201: %

202: %

203: \makeatletter \@beginparpenalty=10000 \makeatother

204: %

205: %

206: %

207: %

208: %

209: %

210: %

211: %

212: %

213: %

214: %

215: %

216: %

217: %

218: %

219: %

220: %

221: %

222: 

223: 

224: %

225: %

226: 

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

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

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

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

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

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

233: %

234: %

235: %

236: %

237: %

238: %

239: %

240: %

241: %

242: %

243: %

244: %

245: %

246: %

247: %

248: %

249: %

250: %

251: %

252: 

253: 

254: 

255: %

256: %

257: %

258: %

259: %

260: %

261: %

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

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

264: \labelwidth=\leftmargin

265: \advance \labelwidth by -\labelsep

266: \let \makelabel=\desclabel}}

267: \let\enddesc=\endlist

268: 

269: 

270: 

271: 

272: %

273: 

274: \makeatletter %

275: %

276: 

277: %

278: %

279: %

280: %

281: %

282: %

283: %

284: \newlength{\leftjustindent}

285: \newlength{\@leftjustindent}

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

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

288:   \addtolength{\@leftjustindent}{\leftjustindent}

289:   \vcenter\bgroup

290:   \halign\bgroup

291:     \hbox to\displaywidth{

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

293:       }\crcr

294: }

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

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

297: \def\@leftjustcr{\crcr}

298: 

299: %

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

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

302: %

303: %

304: %

305: %

306: %

307: %

308: %

309: %

310: %

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

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

313: \def\qedcareful{\literalqed}

314: \def\qed{\literalqed}

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

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

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

318: %

319: %

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

321: \newtheorem{claim}[theorem]{Claim}

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

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

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

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

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

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

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

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

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

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

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

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

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

335: \newcommand{\normalspacing}{\singlespacing}

336: \makeatother%

337: 

338: 

339: %

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

341: 

342: %

343: 

344: %

345: %

346: %

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

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

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

350: %

351: %

352: %

353: %

354: \mathcode`\0="0030      %

355: \mathcode`\1="0031

356: \mathcode`\2="0032

357: \mathcode`\3="0033

358: \mathcode`\4="0034

359: \mathcode`\5="0035

360: \mathcode`\6="0036

361: \mathcode`\7="0037

362: \mathcode`\8="0038

363: \mathcode`\9="0039

364: %

365: %

366: %

367: %

368: \singlespacingplus

369: %

370: 

371: 

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

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

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

375: 

376: 

377: %

378: %

379: %

380: %

381: %

382: %

383: 

384: %

385: 

386: %

387: \flushbottom{}

388: %

389: \makeatletter

390: \clubpenalty=\@highpenalty

391: \widowpenalty=\@highpenalty

392: \makeatother

393: 

394: %

395: \let\BLS=\baselinestretch

396: 

397: %

398: %

399: \makeatletter

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

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

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

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

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

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

406: \makeatother

407: 

408: %

409: \makeatletter

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

411: \makeatother

412: 

413: %

414: %

415: %

416: \makeatletter

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

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

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

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

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

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

423: \makeatother

424: 

425: 

426: %

427: \makeatletter

428: \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{}}

429: \makeatother

430: 

431: \makeatletter

432: %

433: %

434: %

435: %

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

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

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

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

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

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

442:           \textfont\@ne\tenmi

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

444:           \@boldfalse}\@boldfalse

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

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

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

448:               {\def\boldmath{\everymath

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

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

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

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

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

454:    \scriptscriptfont\itfam\sevit

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

456:      \scriptfont\slfam\tensl \scriptscriptfont\slfam\tensl

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

458:    \scriptfont\bffam\ninbf \scriptscriptfont\bffam\ninbf

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

460:    \scriptfont\ttfam\nintt \scriptscriptfont\ttfam\nintt

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

462:     \scriptfont\sffam\tensf \scriptscriptfont\sffam\tensf

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

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

465:    \scriptfont\lyfam\ninly \scriptscriptfont\lyfam\sevly

466:  \@setstrut \rm}

467: 

468: \makeatother

469: 

470: 

471: 

472: \newcommand{\scriptnp}{\mbox{\scriptsize\rm NP}}

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

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

475: \newcommand{\sedr}{\mbox{${\cal S}^{\mbox{\scriptsize\rm{}ED}}_{r}$}}

476: \newcommand{\smdgr}{\mbox{${\cal S}^{\mbox{\scriptsize\rm{}MDG}}_{r}$}}

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

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

479: \newcommand{\vcgeq}{{\tt VC}_{\mbox{\scriptsize ${\tt geq}$}}}

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

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

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

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

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

485: %

486: \newcommand{\condition}{\,|\:}

487: \newcommand\seq{\subseteq}

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

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

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

491: \newcommand\lora{\ \longrightarrow \ }

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

493: \newcommand{\sedone}{\mbox{${\cal S}^{\mbox{\scriptsize\rm{}ED}}_{1}$}}

494: \newcommand{\vcsedone}{{\tt VC}\mbox{-}{\sedone}}

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

496: \def\nats{\naturalnumber}

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

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

499: \newcommand{\smdgone}{\mbox{${\cal S}^{\mbox{\scriptsize\rm{}MDG}}_{1}$}}

500: \newcommand{\vcsmdgone}{{\tt VC}\mbox{-}{\smdgone}}

501: \newcommand{\vc}{{\tt VC}}

502: \newcommand{\sproof}{\noindent{\bf Proof}\quad}

503: 

504: 

505: %

506: \foospacing

507: %

508: \title{Recognizing When Heuristics Can Approximate Minimum Vertex Covers Is

509:   Complete for Parallel Access to NP\footnote{

510: \protect\singlespacing

511: This work was

512: supported in part by the NSF and the DAAD under grant

513:   NSF-INT-9815095/DAAD-315-PPP-g\"{u}-ab and

514: by the DFG under grant RO~1202/9-1.

515: The first author was supported in part by the NSF under grant NSF-CCR-0311021.

516: The second author was supported in part by a Hei\-sen\-berg Fellowship

517: of the DFG.

518: }}

519: 

520: \author{

521: Edith Hemaspaandra\\

522: Department of Computer Science \\

523: Rochester Institute of Technology \\

524: Rochester, NY 14627, USA\\

525: {\tt eh@cs.rit.edu}

526: \and

527: J\"{o}rg Rothe\\

528: Institut f\"{u}r Informatik\\

529: Heinrich-Heine-Universit\"{a}t\\

530: 40225 D\"{u}sseldorf, Germany \\

531: {\tt rothe@cs.uni-duesseldorf.de}

532: \and

533: Holger Spakowski\\

534: Institut f\"{u}r Informatik\\

535: Heinrich-Heine-Universit\"{a}t\\

536: 40225 D\"{u}sseldorf, Germany \\

537: {\tt spakowsk@cs.uni-duesseldorf.de}

538: }

539: 

540: %

541: \date{January 25, 2005}

542: 

543: 

544: %

545: %

546: %

547: %

548: 

549: 

550: %

551: %

552: %

553: \lefthyphenmin=100

554: 

555: %

556: %

557: %

558: 

559: \setlength{\oddsidemargin}{0.25in}

560: \setlength{\evensidemargin}{\oddsidemargin}

561: \setlength{\textwidth}{6in}

562: \setlength{\textheight}{8in}

563: \setlength{\topmargin}{-0.0in}

564: 

565: %

566: %

567: %

568: %

569: %

570: %

571: 

572: %

573: %

574: %

575: %

576: %

577: %

578: %

579: %

580: %

581: %

582: %

583: %

584: 

585: 

586: 

587: %

588: %

589: %

590: %

591: %

592: %

593: %

594: %

595: 

596: %

597: %

598: %

599: %

600: %

601: %

602: %

603: %

604: %

605: \makeatletter

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

607: 6pt plus 2pt minus 2pt \itemsep  2pt plus 2pt minus 1pt}

608: 

609: \let\@listi\@listI

610: \@listi

611: \makeatother

612: 

613: 

614: \begin{document}

615: 

616: %

617: %

618: %

619: %

620: %

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

622: \hbadness=3000%

623: \vbadness=10000 %

624: 

625: 

626: 

627: 

628: 

629: 

630: %

631: %

632: %

633: %

634: 

635: %

636: \pagestyle{empty}

637: %

638: %

639: %

640: %

641: %

642: %

643: %

644: %

645: \setcounter{page}{1}

646: 

647: 

648: %

649: 

650: \sloppy

651: 

652: %

653: \pagestyle{empty}

654: \setcounter{footnote}{0}

655: %

656: 

657: {\singlespacing

658: 

659: \maketitle

660: 

661: }

662: 

663: %

664: %

665: %

666: 

667: %

668: %

669: 

670: 

671: 

672: \begin{abstract}\noindent

673:   For both the edge deletion heuristic and the

674:   maximum-degree greedy heuristic, we study 

675:   the problem of recognizing those graphs for which

676:   that heuristic can approximate

677:   the size of a minimum vertex cover within a constant factor

678:   of~$r$, where $r$ is a fixed rational number.

679:   Our main results are

680:   that these problems are complete for the class of problems solvable via

681:   parallel access to~$\np$.

682: %

683: %

684: %

685: %

686: %

687:   To achieve these main results, we also show that

688:   the restriction of the vertex cover problem to those graphs for which either

689:   of these heuristics can find an optimal solution remains NP-hard.

690: 

691: 

692: \smallskip\noindent

693: {\bf Key words:} Computational complexity; completeness; minimum vertex cover

694: heuristics; approximation; parallel access to NP.

695: 

696: %

697: %

698: 

699: \end{abstract}

700: 

701: 

702: %

703: 

704: %

705: %

706: \foospacing

707: %

708: %

709: %

710: %

711: %

712: %

713: \pagestyle{plain}

714: %

715: \sloppy

716: 

717: 

718: 

719: 

720: \section{Introduction}

721: 

722: \noindent

723: The minimum vertex cover problem is the problem of finding in a given graph a

724: smallest possible set of vertices that covers at least one vertex of each

725: edge.  The decision version of the minimum vertex cover problem, ${\tt VC}$,

726: is one of the standard NP-complete problems~\cite{gar-joh:b:int}.  To cope

727: with the intractability that appears to be inherent to this problem, various

728: heuristics for finding minimum vertex covers have been proposed.  Two of the

729: most prominent such heuristics are the {\em edge deletion heuristic\/} and the

730: {\em maximum-degree greedy heuristic}, see,

731: e.g.,~\cite{pap-ste:b:optimization,pap:b:complexity}.  These algorithms run in

732: linear time and, depending on the structure of the given input graph, may find

733: a minimum vertex cover, or may provide a good approximation of the optimal

734: solution.

735: 

736: It is common to evaluate heuristics for optimization problems by analyzing

737: their worst-case ratio for approximating the optimal solution.  In this

738: regard, the two heuristics considered behave quite differently: the edge

739: deletion heuristic always approximates the size of a minimum vertex cover

740: within a factor of~$2$ and thus achieves the best approximation ratio known,

741: whereas the maximum-degree greedy heuristic, in the worst case, can have an

742: approximation ratio as bad as logarithmic in the input size.  The latter

743: result follows from the early analysis of the approximation behavior of the

744: greedy algorithm for the minimum set cover problem that was done by

745: Johnson~\cite{joh:j:approximation},

746: Lov{\'a}sz~\cite{lov:j:ratio-of-optimal-covers}, and

747: Chv{\'{a}}tal~\cite{chv:j:greedy-heuristic-for-set-cover} (who studied the

748: weighted version of minimum set cover).  Note that the vertex cover problem is

749: the special case of the set cover problem, restricted so that each element

750: occurs in exactly two sets.  More recently, building on the work of Lund and

751: Yannakakis~\cite{lun-yan:j:approximating-minimization-problems},

752: Feige~\cite{fei:j:approximating-set-cover} showed that, unless NP has slightly

753: superpolynomial-time algorithms, the set cover problem cannot be approximated

754: within $(1 - \epsilon) \ln n$, where $\epsilon > 0$ and $\ln$ denotes the

755: natural logarithm.

756: 

757: In this paper, we study the problem of recognizing those input graphs for

758: which either of the two heuristics can approximate the size of a minimum

759: vertex cover within a constant factor of~$r$, where $r \geq 1$ is a fixed

760: rational number.  Let $\sedr$ and $\smdgr$, respectively, denote this

761: recognition problem for the edge deletion heuristic and for the maximum-degree

762: greedy heuristic.

763: Our main results are: 

764: \begin{description}

765: \item[Theorem~\ref{thm:sedr-thetatwo-complete}] For each rational number $r$

766:   with $1 \leq r < 2$, $\sedr$ is $\parallelnp$-complete.

767:   

768: \item[Theorem~\ref{thm:smdgr-thetatwo-complete}] For each rational number $r

769:   \geq 1$, $\smdgr$ is $\parallelnp$-complete.

770: \end{description}

771: 

772: Here, $\parallelnp$ denotes the class of problems that can be decided in

773: polynomial time by parallel (i.e., truth-table) access to~$\np$.

774: Papadimitriou and Zachos~\cite{pap-zac:c:two-remarks} introduced this class

775: under the name~$\p^{\scriptnp[\mathcal{O}(\log n)]}$, where

776: ``$[\mathcal{O}(\log n)]$'' denotes that at most logarithmically many Turing

777: queries are made to the $\np$ oracle.  Hemaspaandra~\cite{hem:j:sky}

778: proved that $\p^{\scriptnp[\mathcal{O}(\log n)]} = \parallelnp$, and

779: in fact many more characterizations of~$\parallelnp$ are 

780: known~\cite{koe-sch-wag:j:diff,wag:j:bounded}.  

781: Other natural $\parallelnp$-complete problems can be found

782: in the papers by Krentel~\cite{kre:j:optimization},

783: Wagner~\cite{wag:j:min-max}, and Hemaspaandra et

784: al.~\cite{hem-hem-rot:j:dodgson,hem-rot:j:max-independent-set-by-greed}.

785:   

786: The type of recognition problem studied in this paper was investigated for

787: other problems and other heuristics as well.  Bodlaender, Thilikos, and

788: Yamazaki~\cite{bod-thi-yam:j:greedy-for-maximum-independent-sets} defined and

789: studied the analogous problem for the independent set problem and the

790: minimum-degree greedy heuristic, which they denoted by~$\sr$.  They proved

791: that $\sr$ is coNP-hard and belongs to~$\p^{\mbox{\scriptsize $\np$}}$.

792: Closing the gap between these lower and upper bounds, Hemaspaandra and 

793: Rothe~\cite{hem-rot:j:max-independent-set-by-greed} proved that $\sr$ is

794: $\parallelnp$-complete.

795: %

796: %

797: %

798: %

799: As

800: in~\cite{hem-rot:j:max-independent-set-by-greed}, we obtain

801: $\parallelnp$-hardness by reducing from a problem (namely, $\vcgeq$, see

802: Section~\ref{sec:heuristics}) that can be shown to be $\parallelnp$-complete

803: using the techniques of Wagner~\cite{wag:j:min-max}.  Also, we show that the

804: vertex cover problem, restricted to those input graphs for which the

805: heuristics considered can find an optimal solution, remains NP-hard.  We

806: then lift this NP-hardness lower bound to $\parallelnp$-hardness, which proves

807: our main results.  This lifting requires a padding technique such that the

808: given approximation ratio $r$ is precisely met.  In particular, to achieve

809: $\parallelnp$-hardness of~$\smdgr$ for each rational number $r \geq 1$, we

810: modify a construction by Papadimitriou and

811: Steiglitz~\cite{pap-ste:b:optimization} that they use to analyze the

812: worst-case approximation behavior of the maximum-degree greedy heuristic.

813: %

814: %

815: 

816: \section{Two Heuristics for the Vertex Cover Problem}

817: \label{sec:heuristics}

818: 

819: \noindent

820: We use the following notation.

821: Fix the two-letter alphabet $\Sigma = \{0,1\}$.  $\sigmastar$ is the

822: set of all strings over~$\Sigma$. 

823: Let $\pair{\cdot , \cdot} : 

824: \sigmastar \times \sigmastar \rightarrow\, \sigmastar$ be a standard

825: pairing function.

826: %

827: %

828: %

829: For any set~$L$, let $\| L \|$ denote the number of elements of~$L$.

830: %

831: %

832: %

833: 

834: All graphs considered in this paper are undirected nonempty,

835: finite graphs without multiple or reflexive edges.  

836: For any graph~$G$, let $V(G)$ denote the set of vertices of~$G$, and

837: let $E(G)$ denote the set of edges of~$G$.  

838: %

839: %

840: %

841: For any vertex $v \in V(G)$, the {\em degree of $v$\/} (denoted by

842: $\degree_G(v)$) is the number of vertices adjacent to $v$ in~$G$; if $G$ is

843: clear from the context, we omit the subscript and simply write~$\degree(v)$.

844: Let $\maxdegree(G) = \max_{v \in V(G)} \degree(v)$ denote the maximum degree

845: of the vertices of graph~$G$.

846: %

847: %

848: Let $G$ and $H$ be two disjoint graphs.  The {\em disjoint union of $G$ and

849:   $H$\/} is defined to be the graph $U = G \cup H$ with vertex set $V(U) =

850: V(G) \cup V(H)$ and edge set $E(U) = E(G) \cup E(H)$.  The {\em join of $G$

851:   and $H$\/} is defined to be the graph $J = G \bowtie H$ with vertex set $V(J)

852: = V(G) \cup V(H)$ and edge set $E(J) = E(G) \cup E(H) \cup \{\{x,y\}

853: \condition x \in V(G) \ \wedge\ y \in V(H)\}$.

854: 

855: For any graph~$G$, a subset $C \seq V(G)$ is a {\em vertex cover of

856:   $G$\/} if for all edges $\{v, w\} \in E(G)$, $\{v, w\} \cap C \neq

857: \emptyset$.  A vertex cover is said to be a {\em minimum vertex cover of

858:   $G$\/} if it is of minimum size.  For any graph~$G$, let $\mvc(G)$ denote

859: the size of a minimum vertex cover of~$G$.  The vertex cover problem (${\tt

860:   VC}$, for short; see~\cite{gar-joh:b:int}) is defined to be the set of all

861: pairs $\pair{G,k}$ such that $G$ is a graph, $k$ a positive integer, and

862: $\mvc(G) \leq k$.

863: 

864: All hardness and completeness results in this paper are with respect to the

865: polynomial-time many-one reducibility, denoted~$\manyone$.

866: For sets $A$ and~$B$, we say $A \manyone B$ if and only if there exists a

867: polynomial-time computable function $f$ such that for all inputs~$x \in

868: \sigmastar$, $x \in A$ if and only if $f(x) \in B$.

869: 

870: %

871: %

872: %

873: %

874: %

875: %

876: %

877: %

878: %

879: %

880: %

881: %

882: %

883: %

884: %

885: %

886: %

887: %

888: %

889: %

890: %

891: %

892: %

893: %

894: %

895: 

896: We consider the following two heuristics (see,

897: e.g.,~\cite{pap-ste:b:optimization,pap:b:complexity}) for finding a minimum

898: vertex cover of a given graph:

899: \begin{description}

900: \item[Edge Deletion Heuristic (ED):] Given a graph~$G$, the algorithm outputs

901:   a vertex cover $C$ of~$G$.  Initially, $C$ is the empty set.

902:   Nondeterministically choose an edge $\{u, v\} \in E(G)$, add both $u$ and

903:   $v$ to~$C$, and delete $u$, $v$, and all edges incident to $u$ and $v$

904:   from~$G$.  Repeat until there is no edge left in~$G$.

905:   

906: \item[Maximum-Degree Greedy Heuristic (MDG):] Given a graph~$G$, the algorithm

907:   outputs a vertex cover $C$ of~$G$.  Initially, $C$ is the empty set.

908:   Nondeterministically choose a vertex $v \in V(G)$ of maximum degree, add

909:   $v$ to~$C$, and delete $v$ and all edges incident to $v$ from~$G$.  Repeat

910:   until there is no edge left in~$G$.

911: \end{description}

912: 

913: As mentioned in the introduction, these two heuristics have a quite different

914: approximation behavior.  While the worst-case ratio of the MDG algorithm is

915: logarithmic in the input size~\cite{pap:b:complexity,joh:j:approximation}, the

916: ED algorithm always approximates the optimal solution within a factor of~$2$.

917: Thus, despite its extreme simplicity, the edge deletion heuristic achieves the

918: best approximation ratio known for finding minimum vertex

919: covers~\cite{pap:b:complexity}.

920: 

921: The central question raised in this paper is: How hard is it to determine for

922: which graphs $G$ either of these two heuristics can approximate the minimum

923: vertex cover of $G$ within a factor of~$r$, for a given rational number $r

924: \geq 1$?  Let $\mined(G)$ (respectively, $\minmdg(G)$) denote the minimum size

925: of the output set of the ED algorithm (respectively, of the MDG algorithm) on

926: input~$G$, where the minimum is taken over all possible sequences of

927: nondeterministic choices the algorithms can make.  For any fixed rational $r

928: \geq 1$, $\sedr$ (respectively, $\smdgr$) is the class of graphs for which ED

929: (respectively, MDG)

930: can output a vertex cover of size at most $r$ times the size of a minimum

931: vertex cover.  Formally,

932: \begin{eqnarray*}

933: \sedr & = & \{ G \condition \mbox{$G$ is a graph and $\mined(G)

934: \leq r\cdot\mvc(G)$} \} ; \\

935: \smdgr & = & \{ G \condition \mbox{$G$ is a graph and $\minmdg(G)

936: \leq r\cdot\mvc(G)$} \} .

937: \end{eqnarray*}

938: 

939: We will prove that for each fixed rational number $r$ with $1 \leq r < 2$,

940: $\sedr$ is $\parallelnp$-complete, and that for each fixed rational number $r

941: \geq 1$, $\smdgr$ is $\parallelnp$-complete.  To this end, we give reductions

942: from the problem $\vcgeq$, which is defined by

943: \[

944: \vcgeq = \{ \pair{G,H} \condition \mbox{$G$ and $H$ are graphs such that

945:   $\mvc(G) \geq \mvc(H)$}\}.

946: \]

947: It is known that $\vcgeq$ is $\parallelnp$-complete, cf.

948: Wagner~\cite{wag:j:min-max}.  A reduction from any problem in $\parallelnp$ to

949: $\vcgeq$ that in addition has some useful properties (see

950: Lemma~\ref{lem:vcgeq} below) can easily be obtained using the techniques of

951: Wagner~\cite{wag:j:min-max}; see~\cite[Thm.~12]{spa-vog:c:thetatwo} 

952: for an explicit proof of Lemma~\ref{lem:vcgeq}.

953: 

954: \begin{lemma} {\rm{}(cf.~\cite{wag:j:min-max,spa-vog:c:thetatwo})}\quad

955: \label{lem:vcgeq}

956: For any set $X \in \parallelnp$, there exists a polynomial-time computable

957: function $f$ that reduces $X$ to $\vcgeq$ in such a way that for each $x \in

958: \sigmastar$, $f(x) = \pair{G,H}$ is an instance of $\vcgeq$ and

959: \begin{eqnarray*}

960: x \in X     & \Lora & \mvc(G) = \mvc(H) ; \\

961: x \not\in X & \Lora & \mvc(G) < \mvc(H) .

962: \end{eqnarray*}

963: \end{lemma}

964: 

965: \section{The Edge Deletion Heuristic}

966: %

967: %

968: 

969: \noindent

970: Lemma~\ref{lem:vcsedone} below states that the vertex cover problem restricted

971: to graphs in $\sedone$ is NP-hard.  

972: The reduction $g$ from Lemma~\ref{lem:vcsedone}

973: will be used in the proof of the main result of this section,

974: Theorem~\ref{thm:sedr-thetatwo-complete}.  Define the problem

975: \[

976: \vcsedone = \{ \pair{G,k} \condition 

977: \mbox{$G \in \sedone$ and $k \in \nats^+$ and $\mvc(G) \leq k$}\}.

978: \]

979: 

980: \begin{lemma}  

981:   \label{lem:vcsedone}

982:   There is a polynomial-time many-one reduction $g$ from

983:   $\vc$ to $\vcsedone$ transforming any given graph $G$ into

984:   a graph $H\in \sedone$ such that 

985: \begin{equation}

986: \label{eq:vcsedone}

987: \mvc(H) = 2 (\mvc(G) + \| V(G) \|).

988: \end{equation}

989:   Hence, $\vcsedone$ is $\np$-hard.

990: \end{lemma}

991: \sproof

992:   Given any graph~$G$, we construct the graph $H \in \sedone$ 

993: as follows.  For each vertex $v\in V(G)$, create a

994: component $G_v$ that is defined by the vertex set 

995: $V(G_v) = \{ v_1, v_2, v_3, v_4 \}$ and the edge set

996: $E(G_v) = \{ \{ v_1, v_2\} , \{ v_3, v_4\}, \{ v_1, v_3\} \}$.

997: 

998: Define the graph $H$ by joining every pair of components that correspond to

999: adjacent vertices of~$G$:

1000: \begin{eqnarray*}

1001:          V(H) & = & \bigcup_{v\in V(G)} V(G_v); \\

1002:          E(H) & = & \{\{ a_i,b_j\} \condition \mbox{$\{a,b\} \in E(G)$ and 

1003:                       $i,j \in \{ 1, 2, 3, 4 \}$} \}

1004:                     \cup \bigcup_{v\in V(G)} E(G_v).

1005: \end{eqnarray*}  

1006: We now prove Equation (\ref{eq:vcsedone}).

1007: Let $C$ be a minimum vertex cover

1008: of~$G$, i.e., $\mvc(G) = \| C \|$.  Construct a vertex cover $D$ of $H$ as

1009: follows.  For each vertex $v \in C$, add $v_1, v_2, v_3$, and $v_4$ to~$D$;

1010: and for each vertex $w \in V(G) - C$, add $w_1$ and $w_3$ to~$D$.  Hence,

1011: \[

1012: \|D\| = 2 (\| C \| + \| V(G) \|).

1013: \]

1014: Since $mvc(H) \leq \| D \|$, it follows

1015: that $\mvc(H) \leq 2 (\mvc(G) + \| V(G) \|)$.

1016: 

1017: Conversely, let $D$ be a minimum vertex cover of~$H$, i.e., $\mvc(H) = \|D\|$.

1018: Then, it holds that:

1019: \begin{itemize}

1020: \item for each edge $\{ u,v\}\in E(G)$, $V(G_u) \seq D$ or $V(G_v) \seq D$;

1021: \item for each vertex $v\in V(G)$, $\|D \cap V(G_v)\| \geq 2$.

1022: \end{itemize}

1023: Hence,

1024: \begin{eqnarray*}

1025: \| D \| & \geq & 4 \cdot \mvc(G) + 2 (\|V(G)\| - \mvc(G)) \\

1026:         & = & 2 (\mvc(G) + \|V(G)\|).

1027: \end{eqnarray*}

1028: It follows that $ \mvc(H) \geq 2 (\mvc(G) + \| V(G) \|)$, which proves

1029: Equation~(\ref{eq:vcsedone}).

1030: 

1031: It remains to prove that $H \in \sedone$.  Let $C$ be a minimum vertex cover

1032: of~$G$.  The edge deletion algorithm can find a vertex cover of $H$ as

1033: follows.  For every vertex $v\in C$, choose the edges $\{ v_1, v_2\}$ and $\{

1034: v_3, v_4\}$.  For the remaining vertices $w \in V(G) - C$, choose the edge $\{

1035: w_1, w_3\}$.  Thus, $\mined(H) = 2 (\mvc(G) + \| V(G) \|)$.  By

1036: Equation~(\ref{eq:vcsedone}), $\mined(H) = \mvc(H)$, so $H \in \sedone$.~\qed

1037: 

1038: 

1039: \begin{theorem}  

1040: \label{thm:sedr-thetatwo-complete}

1041: For each rational number $r$ with $1 \leq r < 2$, $\sedr$ is

1042: $\parallelnp$-complete.

1043: \end{theorem}

1044: \sproof

1045:   It is easy to see that $\sedr$ is in $\parallelnp$.

1046:   To prove $\parallelnp$-hardness, let $X$ be an arbitrary set

1047:   in~$\parallelnp$, and let $f$ be the reduction from $X$ to $\vcgeq$ stated

1048:   in Lemma~\ref{lem:vcgeq}.  Fix any rational number $r$ with $1 \leq r

1049:   < 2$, and let $\ell$ and $m$ be integers such that $r = \frac{\ell}{m}$.

1050:   Note that $1 \leq m \leq \ell < 2m$.

1051:   

1052:   For any string $x \in \sigmastar$, let $f(x) = \pair{G_1, G_2}$. 

1053:   Since we can add isolated vertices to any graph $G$ without altering

1054:   $\mvc(G)$, we may without

1055:   loss of generality assume that $\|V(G_1)\| = \|V(G_2)\|$.

1056: %

1057: %

1058: %

1059:   Let $g$ be the reduction from

1060:   Lemma~\ref{lem:vcsedone} that

1061:   transforms any given graph $G$ into a graph $H \in \sedone$ such that

1062:   Equation~(\ref{eq:vcsedone}) holds.

1063:   Let $H_1 = g(G_1)$ and $H_2 =

1064:   g(G_2)$.  Thus, both $H_1$ and $H_2$ are in $\sedone$,

1065:   and for $i \in \{1, 2\}$, we have 

1066:   $\mvc(H_i) = 2 (\mvc(G_i) + \| V(G_i) \|)$. 

1067:   

1068:   We will define a graph $\widehat{H}$ and an integer $k \geq 0$ such that:

1069: \begin{eqnarray}

1070: \label{eq:H1H2-mined}

1071: \mined(\widehat{H}) & = & r (m \cdot \mvc(H_2) + 2km); \\

1072: \label{eq:H1H2-mvc}

1073: \mvc(\widehat{H})   & = & m \cdot \mvc(H_1) + 2km.

1074: \end{eqnarray}

1075: 

1076: The reduction mapping any given string $x$ (via the pair $\pair{G_1, G_2}$

1077: obtained according to Lemma~\ref{lem:vcgeq} and via the pair $\pair{H_1, H_2}$

1078: obtained according to Lemma~\ref{lem:vcsedone}) to the graph $\widehat{H}$

1079: such that Equations~(\ref{eq:H1H2-mined}) and~(\ref{eq:H1H2-mvc}) are

1080: satisfied will establish that $X \manyone \sedr$.  In particular, from these

1081: equations, we have that:

1082: 

1083: \begin{itemize}

1084: \item $\mvc(H_2) = \mvc(H_1)$ implies $\mined(\widehat{H}) = r \cdot

1085:   \mvc(\widehat{H})$, and 

1086: \item $\mvc(H_2) > \mvc(H_1)$ implies $\mined(\widehat{H}) > r \cdot

1087:   \mvc(\widehat{H})$.

1088: \end{itemize}

1089: Note that, due to Lemma~\ref{lem:vcgeq}, $\mvc(H_2) \geq \mvc(H_1)$.

1090: 

1091: \begin{figure}[ht]  

1092: \centering

1093: \input{sedr.eepic}

1094: \caption{The graph $\widehat{H}$ constructed from $H_1$ and~$H_2$.}

1095: \label{fig:sedr}  

1096: \end{figure}

1097: 

1098: Look at Figure~\ref{fig:sedr} for the construction of $\widehat{H}$ from $H_1$

1099: and~$H_2$.  The graph $\widehat{H}$ consists of two subgraphs, $L$ and~$R$,

1100: that are joined by the join operation, plus some additional vertices and edges

1101: that are connected to~$R$. Formally, let $H_{1}^{1}, H_{1}^{2}, \ldots,

1102: H_{1}^{m}$ be $m$ pairwise disjoint copies of~$H_1$, and let $H_{2}^{1},

1103: H_{2}^{2}, \ldots, H_{2}^{\ell}$ be $\ell$ pairwise disjoint copies

1104: of~$H_2$.  

1105: Let $k = \ell \| V(H_2) \| + m \|V(H_1)\|$.

1106: Let $I_1$ and $I_2$ be independent sets such that $L$ contains

1107: exactly $k(2m-\ell)$ vertices and $R$ exactly $k\ell$

1108: vertices. (This is possible, because

1109: $k(2m-\ell) - \ell \| V(H_2)\|$ is not negative, since 

1110: $2m - \ell \geq 1$, and $k\ell - m \|V(H_1)\|$ is not negative,

1111: since $\ell \geq 1$.)

1112: Let 

1113: $e_i=\{ a_i, b_i\}$ $(1\le i \le k\ell)$ be additional edges. Every vertex

1114: $a_i$ is adjacent to exactly one vertex in $R$, and each vertex in $R$ is

1115: adjacent to exactly one vertex~$a_i$.  The vertices $a_i$ and $b_i$ are not

1116: adjacent to any other vertices. 

1117: 

1118: \begin{enumerate}

1119: \item

1120: We first determine $\mined(\widehat{H})$.

1121: Let $\widehat{E}$ be a fixed minimum-size output set of the ED algorithm on

1122: input $\widehat{H}$, i.e., $\mined(\widehat{H}) = \| \widehat{E} \|$.

1123: Since $\widehat{E}$ is a vertex cover of $\widehat{H}$, $\widehat{E}$ must

1124: contain $a_i$ or $b_i$ for each $i\in\{ 1,\ldots , k\ell\}$. Since the

1125: ED-algorithm can delete only edges, and $\widehat{E}$ is a minimum-size output

1126: set, it follows that $\widehat{E}$ contains all vertices $a_i$, all vertices

1127: from $R$, and no vertex $b_i$.

1128: 

1129: Let $C_L$ be a minimum-size output set of the ED-algorithm on input $L$.

1130: By construction of $L$, $\|C_L\| = \ell \cdot \mined(H_2)$.  Thus, since

1131: $H_2 \in \sedone$, $\|C_L\|= \ell\cdot\mvc(H_2)$.

1132: 

1133: Define $\widehat{E}' = V(R)\cup C_L\cup \bigcup_{i=1}^{k\ell}\{a_i\}$.

1134: It is easy to see that $\widehat{E}'$ is a minimum-size output set of the ED

1135: algorithm on input $\widehat{H}$.

1136: Hence,

1137: \begin{eqnarray*}

1138:    \mined(\widehat{H}) & = & 2k\ell + \ell\cdot\mvc(H_2)\\

1139:                        & = & r(2km + m\cdot\mvc(H_2)).

1140: \end{eqnarray*}

1141: This proves Equation~(\ref{eq:H1H2-mined}).

1142: 

1143: \item

1144: We now determine $mvc(\widehat{H})$.

1145: Let $\widehat{C}$ be a fixed minimum vertex cover of $\widehat{H}$,

1146: i.e., $\mvc(\widehat{H}) = \|\widehat{C}\|$. Distinguish the following

1147: two cases.

1148: \begin{description}

1149:   \item[Case 1:] $V(R) \seq \widehat{C}$. In this case, $\widehat{C}$ contains all

1150:     vertices from $R$, at least one of $a_i$ or $b_i$ for each $i$, 

1151:     $1\le i \le k\ell$, and a minimum vertex cover of $L$.

1152:     Hence,

1153:     \begin{displaymath}

1154:         \mvc(\widehat{H}) = 2k\ell + \ell\cdot\mvc(H_2).

1155:     \end{displaymath}

1156:   \item[Case 2:] $V(L) \seq \widehat{C}$. In this case, $\widehat{C}$ contains all

1157:     vertices from $L$, each vertex $a_i$, 

1158:     $1\le i \le k\ell$, and a minimum vertex cover of $R$.

1159:     Hence,

1160:     \begin{eqnarray*}

1161:         \mvc(\widehat{H}) & = & k(2m-\ell) + k\ell + m\cdot\mvc(H_1) \\

1162:                           & = & 2km + m\cdot\mvc(H_1).

1163:     \end{eqnarray*}

1164: \end{description}

1165: Since $mvc(H_1) \le \mvc(H_2)$, $m\le \ell$, and $2km\le 2k\ell$, it follows

1166: that

1167: %

1168: %

1169: %

1170: \begin{displaymath}

1171:    mvc(\widehat{H}) = 2km + m\cdot \mvc(H_1).

1172: \end{displaymath}

1173: This proves Equation~(\ref{eq:H1H2-mvc}).

1174: \end{enumerate}

1175: This proves Theorem~\ref{thm:sedr-thetatwo-complete}.~\qed

1176: 

1177: 

1178:    %

1179:    %

1180:    %

1181:    %

1182:    %

1183:    %

1184:    %

1185:    %

1186:    %

1187:    %

1188:    %

1189:    %

1190:    %

1191:    %

1192:    %

1193:    %

1194:    %

1195:    %

1196:    %

1197:    %

1198:    %

1199:    %

1200:    %

1201:    %

1202:    %

1203:    %

1204:    %

1205:    %

1206:    %

1207:    %

1208:    %

1209:    %

1210:    %

1211:    %

1212:    %

1213:    %

1214:    %

1215:    %

1216:    %

1217:    %

1218:    %

1219:    %

1220:    %

1221:    %

1222:    %

1223:    %

1224:    %

1225:    %

1226:    %

1227:    %

1228:    %

1229:    %

1230:    %

1231:    %

1232:    %

1233:    %

1234:    %

1235:    %

1236:    %

1237:    %

1238:    %

1239:    %

1240:    %

1241:    %

1242:    %

1243:    %

1244:    %

1245:    %

1246:    %

1247:    %

1248:    %

1249:    %

1250:    %

1251:    %

1252:    %

1253:    %

1254:    %

1255:    %

1256:    %

1257:    %

1258:    %

1259:    %

1260:    %

1261:    %

1262:    %

1263:    %

1264:    %

1265:    %

1266:    %

1267:    %

1268:    %

1269:    %

1270:    %

1271:    %

1272:    %

1273:    %

1274:    %

1275:    %

1276:    %

1277:    %

1278:    %

1279:    %

1280:    %

1281:    %

1282:    %

1283:    %

1284:    %

1285:    %

1286:    %

1287:    %

1288:    %

1289:    %

1290:    %

1291:    %

1292:    %

1293:    %

1294:    %

1295:    %

1296:    %

1297:    %

1298:    %

1299:    %

1300:    %

1301:    %

1302:    %

1303:    %

1304:    %

1305:    %

1306:    %

1307:    %

1308:    %

1309:    %

1310:    %

1311:    %

1312:    %

1313:    %

1314:    %

1315:    %

1316:    %

1317:    %

1318:    %

1319:    %

1320:    %

1321:    %

1322:    %

1323:    %

1324:    %

1325:    %

1326:    %

1327:    %

1328:    %

1329:    %

1330:    %

1331:    %

1332:    %

1333:    %

1334:    %

1335:    %

1336:    %

1337:    %

1338:    %

1339:    %

1340:    %

1341:    %

1342:    %

1343:    %

1344:    %

1345:    %

1346:    %

1347:    %

1348:    %

1349:    %

1350:    %

1351:    %

1352:    %

1353:    %

1354:    %

1355:    %

1356:    %

1357:    %

1358:    %

1359:    %

1360:    %

1361:    %

1362:    %

1363:    %

1364:    %

1365:    %

1366:    %

1367:    %

1368:    %

1369:    %

1370:    %

1371:    %

1372:    %

1373:    %

1374: 

1375: 

1376: 

1377: \section{The Maximum-Degree Greedy Heuristic}

1378: %

1379: %

1380: 

1381: \noindent

1382: Lemma~\ref{lem:vcsmdgone} below states that the vertex cover problem

1383: restricted to graphs in $\smdgone$ is NP-hard.  The proof of

1384: Lemma~\ref{lem:vcsmdgone} 

1385: is reminiscent of a proof by Bodlaender et

1386: al.~\cite[Thm.~4]{bod-thi-yam:j:greedy-for-maximum-independent-sets}, who show

1387: that the independent set problem restricted to graphs for which the

1388: minimum-degree greedy heuristic can find an optimal solution is NP-hard.

1389: The reduction $g$ from Lemma~\ref{lem:vcsmdgone} will be used in

1390: the proof of the main result of this section,

1391: Theorem~\ref{thm:smdgr-thetatwo-complete}.  Define the problem

1392: \[

1393: \vcsmdgone = \{ \pair{G,k} \condition 

1394: \mbox{$G \in \smdgone$ and $k \in \nats^+$ and $\mvc(G) \leq k$}\}.

1395: \]

1396: 

1397: \begin{lemma}  

1398: \label{lem:vcsmdgone}

1399:   There is a polynomial-time many-one reduction $g$ from

1400:   $\vc$ to $\vcsmdgone$ transforming any given graph $G$ into

1401:   a graph $H\in \smdgone$ such that 

1402: \begin{equation}

1403: \label{eq:vcsmdgone}

1404: \mvc(H) =\mvc(G) + \| E(G) \| (\maxdegree(G) + 1).

1405: \end{equation}

1406:   Hence, $\vcsmdgone$ is $\np$-hard.

1407: \end{lemma}

1408: \sproof

1409:   Given any graph~$G$, we construct the graph $H \in \smdgone$ 

1410: as follows. We replace each edge of $G$ by a gadget that

1411: contains a complete bipartite graph of size $2 (\maxdegree(G) + 1)$.

1412: Formally, $H$ is defined by:

1413: \begin{eqnarray*}

1414:          V(H) & = & V(G) \cup \\

1415:               &   & \bigcup_{e \, = \, \{u,v\} \,\in\, E(G)} 

1416:          \{u_{i}^{e} \condition 1 \leq i \leq \maxdegree(G) + 1\}

1417:          \cup \{v_{i}^{e} \condition 1 \leq i \leq \maxdegree(G) + 1\} ; \\

1418:          E(H) & = & \bigcup_{e \, = \, \{u,v\} \, \in\, E(G)} 

1419:          \left(\{ \{u_{i}^{e}, v_{j}^{e}\} 

1420:          \condition 1 \leq i, j \leq \maxdegree(G) + 1\} 

1421:          \cup \{ \{u, u_{1}^{e}\} \} \cup \{ \{v, v_{1}^{e}\} \}\right).

1422: \end{eqnarray*}  

1423: 

1424: We now prove Equation~(\ref{eq:vcsmdgone}).  Let $C$ be a minimum vertex cover

1425: of~$G$, i.e., $\mvc(G) = \| C \|$.  Note that $\{u, v\} \cap C \neq \emptyset$

1426: for each edge $\{u, v\}$ in~$E(G)$.  Construct a vertex cover $D$ of $H$ as

1427: follows:

1428: \begin{itemize}

1429: \item $D$ contains all vertices from $C$.

1430: \item For every edge $e = \{ u, v \}$ in $E(G)$, add to~$D$:

1431: \begin{itemize}

1432: \item either all vertices $u_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, if

1433:   $u \not\in C$ or if both $u$ and $v$ are in~$C$;

1434: \item or all vertices $v_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, if $v

1435:   \not\in C$.

1436: \end{itemize}

1437: \end{itemize}

1438: It follows that $\mvc(H) \leq \mvc(G) + \| E(G) \| (\maxdegree(G) + 1)$.

1439: 

1440: 

1441: Conversely, let $D$ be a minimum vertex cover of~$H$, i.e., $\mvc(H) = \|D\|$.

1442: Construct a vertex cover $C$ of $G$ as follows.  Initially, set $C = D$.  Let

1443: $e = \{ u, v \}$ be any fixed edge in~$E(G)$.  Suppose that at least one

1444: vertex from $\{ u,v\}$ is in~$D$.  Since $D$ is a vertex cover of~$H$, it

1445: contains at least $\maxdegree(G) + 1$ of the vertices $u_{i}^{e}$

1446: and~$v_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, that correspond to the

1447: edge~$e$.  Remove any $\maxdegree(G) + 1$ such vertices from~$C$.  Suppose now

1448: that neither $u$ nor $v$ is in~$D$.  Since $D$ is a vertex cover of~$H$, it

1449: contains at least $\maxdegree(G) + 2$ of the vertices $u_{i}^{e}$

1450: and~$v_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, that correspond to the

1451: edge~$e$.  Remove any $\maxdegree(G) + 2$ such vertices from~$C$, and add to

1452: $C$ one of $u$ or $v$ instead.  Since the set $C$ thus obtained is a vertex

1453: cover of~$G$, we have $\mvc(H) \geq \mvc(G) + \|E(G)\| (\maxdegree(G) + 1)$,

1454: which proves Equation~(\ref{eq:vcsmdgone}).

1455: 

1456: It remains to prove that $H \in \smdgone$.  Let $C$ be a minimum vertex cover

1457: of~$G$.  The maximum-degree greedy algorithm can find a vertex cover of $H$ as

1458: follows.  For every edge $e = \{ u, v \}$ in~$E(G)$, the MDG algorithm on

1459: input $H$ can choose:

1460: \begin{itemize}

1461: \item either all vertices $u_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, if

1462:   $u \not\in C$ or if both $u$ and $v$ are in~$C$;

1463: \item or all vertices $v_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, if $v

1464:   \not\in C$.

1465: \end{itemize}

1466: Note that the MDG heuristic can always do so, since every vertex in $V(G)$ has

1467: degree at most $\maxdegree(G)$.  Subsequently, all vertices that are not in

1468: $C$ are isolated. Thus, the MDG algorithm can now choose all vertices

1469: from~$C$.  Hence, $\minmdg(H) = \mvc(G) + \| E(G)\| (\maxdegree(G) + 1)$.  By

1470: Equation~(\ref{eq:vcsmdgone}), $\minmdg(H) = \mvc(H)$, so $H \in

1471: \smdgone$.~\qed

1472: 

1473: \medskip

1474: 

1475: Lemma~\ref{lem:mdg-bipartite} below will be used in the proof of

1476: Theorem~\ref{thm:smdgr-thetatwo-complete}.  

1477: The construction of the graph $G$

1478: in this lemma is a modification of a construction given by Papadimitriou and

1479: Steiglitz~\cite[p.~408, Fig.~17-3]{pap-ste:b:optimization}, which shows that

1480: the worst-case approximation ratio of the MDG heuristic can be as bad as

1481: logarithmic in the input size, and so grows unboundedly.  Similar

1482: constructions for achieving the worst-case approximation behavior of the

1483: greedy heuristic solving the more general minimum set cover problem were given

1484: by Johnson~\cite{joh:j:approximation},

1485: Lov{\'a}sz~\cite{lov:j:ratio-of-optimal-covers}, and

1486: Chv{\'{a}}tal~\cite{chv:j:greedy-heuristic-for-set-cover}.

1487: 

1488: \begin{lemma} 

1489: \label{lem:mdg-bipartite}

1490: For all positive integers $n_1$, $n_2$, $\delta$, and $\mu$ satisfying

1491: \begin{equation}

1492: \label{eq:mdg-bipartite}

1493: \mu (\ln \mu - 2 \ln (\delta + 2) - 1) \geq n_1 + n_2 ,

1494: \end{equation}

1495: there exists a bipartite graph $G$ with the following properties:

1496: \begin{enumerate}

1497: \item \label{eq:mdg-bipartite-1}

1498: $V(G) = V \cup \tilde{V}$ such that $V \cap \tilde{V} = \emptyset$

1499:   and both $V$ and $\tilde{V}$ are independent sets, where

1500: \begin{itemize}

1501: \item $V = \{ u_1, u_2, \ldots , u_{n_1}, w_1, w_2, \ldots , w_{\mu}, z_1,

1502:   z_2, \ldots z_{n_2}\}$ and

1503: \item $\tilde{V} = \{ \tilde{u}_1, \tilde{u}_2, \ldots ,

1504:   \tilde{u}_{n_1}, \tilde{w}_1, \tilde{w}_2, \ldots ,

1505:   \tilde{w}_{\mu} \}$.

1506: \end{itemize}

1507:   

1508: \item \label{eq:mdg-bipartite-2}

1509: $\{ \{ u_i, \tilde{u}_i\} \condition 1 \leq i \leq n_1 \} \cup

1510:   \{ \{ w_i, \tilde{w}_i\} \condition 1 \leq i \leq \mu \} \seq E(G)$.

1511:   

1512: \item \label{eq:mdg-bipartite-3}

1513:   Every vertex~$\tilde{u}_i$, where $1 \leq i \leq n_1$, has degree~$1$.

1514:   

1515: \item \label{eq:mdg-bipartite-4}

1516:   For each induced subgraph $S$ of $G$ that can be obtained by deleting

1517:   vertices from~$V$ such that $V \cap V(S) \neq \emptyset$, it holds that

1518:   $\max_{v \in V \cap V(S)} \degree_S(v) > 

1519:    \max_{v \in \tilde{V}} \degree_S(v) + \delta$.

1520: \end{enumerate}

1521: \end{lemma}

1522: \sproof

1523:   Let the constants $n_1$, $n_2$, $\delta$, and $\mu$ be given such that

1524:   Equation~(\ref{eq:mdg-bipartite}) is satisfied.  We describe the

1525:   construction of the graph~$G$.  As stated in the lemma, the vertex set of

1526:   $G$ is given by $V(G) = V \cup \tilde{V}$, where $V$ and~$\tilde{V}$

1527:   are two disjoint independent sets.

1528:   

1529:   Rename the vertices of $V$ by $V = \{ \alpha_1, \alpha_2, \ldots ,

1530:   \alpha_{n_1 + \mu + n_2}\}$.  Let $\tilde{W} = \{ \tilde{w}_1,

1531:   \tilde{w}_2, \ldots , \tilde{w}_{\mu} \}$.  The edge set of $G$ is

1532:   defined as follows:

1533: \begin{itemize}

1534: \item Create the edges $\{ u_i, \tilde{u}_i\}$ for each $i$ with $1 \leq i

1535:   \leq n_1$ and the edges $\{ w_j, \tilde{w}_j\}$ for each $j$ with $1 \leq

1536:   j \leq \mu$.

1537:   

1538: \item Partition $\tilde{W}$ into $\left\lfloor\frac{\mu}{\delta +

1539:       3}\right\rfloor$ disjoint sets $\tilde{W}_{1}^{\delta + 3},

1540:   \tilde{W}_{2}^{\delta + 3}, \ldots ,

1541:   \tilde{W}_{\left\lfloor\frac{\mu}{\delta + 3}\right\rfloor}^{\delta + 3}$

1542:   of size $\delta + 3$ each, possibly leaving out some vertices

1543:   from~$\tilde{V}$ and taking care that no vertex in

1544:   $\tilde{W}_{i}^{\delta + 3}$ already is connected with~$\alpha_i$, $1

1545:   \leq i \leq \left\lfloor\frac{\mu}{\delta + 3} \right\rfloor$.  For each $i$

1546:   with $1 \leq i \leq \left\lfloor\frac{\mu}{\delta + 3} \right\rfloor$,

1547:   connect $\alpha_i$ with each vertex in $\tilde{W}_{i}^{\delta + 3}$ by an

1548:   edge.

1549:   

1550: \item Partition $\tilde{W}$ into $\left\lfloor\frac{\mu}{\delta + 4}

1551:   \right\rfloor$ disjoint sets $\tilde{W}_{1}^{\delta + 4},

1552:   \tilde{W}_{2}^{\delta + 4}, \ldots ,

1553:   \tilde{W}_{\left\lfloor\frac{\mu}{\delta + 4}\right\rfloor}^{\delta + 4}$

1554:   of size $\delta + 4$ each, possibly leaving out some vertices

1555:   from~$\tilde{V}$ and taking care that no vertex in

1556:   $\tilde{W}_{i}^{\delta + 3}$ already is connected

1557:   with~$\alpha_{\left\lfloor\frac{\mu}{\delta + 3}\right\rfloor + i}$, $1 \leq

1558:   i \leq \left\lfloor\frac{\mu}{\delta + 4} \right\rfloor$.  For each $i$ with

1559:   $1 \leq i \leq \left\lfloor\frac{\mu}{\delta + 4} \right\rfloor$, connect

1560:   $\alpha_{ \left\lfloor\frac{\mu}{\delta + 3}\right\rfloor + i}$ with each

1561:   vertex in $\tilde{W}_{i}^{\delta + 4}$ by an edge.

1562:   

1563: \item Partition $\tilde{W}$ into $\left\lfloor\frac{\mu}{\delta + 5}

1564:   \right\rfloor$ disjoint sets $\tilde{W}_{1}^{\delta + 5},

1565:   \tilde{W}_{2}^{\delta + 5}, \ldots ,

1566:   \tilde{W}_{\left\lfloor\frac{\mu}{\delta + 5}\right\rfloor}^{\delta + 5}$

1567:   of size $\delta + 5$ each, possibly leaving out some vertices

1568:   from~$\tilde{V}$ and taking care that no vertex in

1569:   $\tilde{W}_{i}^{\delta + 3}$ already is connected

1570:   with~$\alpha_{\left\lfloor\frac{\mu}{\delta + 3}\right\rfloor +

1571:     \left\lfloor\frac{\mu}{\delta + 4}\right\rfloor + i}$, $1\leq i \leq

1572:   \left\lfloor\frac{\mu}{\delta + 5}\right\rfloor$.  For each $i$ with $1\leq

1573:   i \leq \left\lfloor\frac{\mu}{\delta + 5}\right\rfloor$, connect $\alpha_{

1574:     \left\lfloor\frac{\mu}{\delta + 3}\right\rfloor +

1575:     \left\lfloor\frac{\mu}{\delta + 4}\right\rfloor + i}$ with each vertex in

1576:   $\tilde{W}_{i}^{\delta + 5}$ by an edge.

1577:   

1578: \item Continue in this way until all vertices $\alpha_i$ are connected with

1579:   vertices in~$\tilde{W}$.

1580: \end{itemize}

1581: 

1582: The construction is possible, since Equation~(\ref{eq:mdg-bipartite}) implies

1583: \begin{equation}

1584:        \left\lfloor\frac{\mu}{\delta + 3}\right\rfloor + 

1585:        \left\lfloor\frac{\mu}{\delta + 4}\right\rfloor + \cdots + 

1586:        \left\lfloor\frac{\mu}{\mu - 1}\right\rfloor \geq

1587:        n_1 + \mu + n_2 ,

1588: \end{equation}

1589: and thus there are enough vertices in~$\tilde{W}$.  To see why, note that 

1590: \begin{eqnarray}

1591: \lefteqn{ \left\lfloor\frac{\mu}{\delta + 3}\right\rfloor + 

1592:        \left\lfloor\frac{\mu}{\delta + 4}\right\rfloor + \cdots + 

1593:        \left\lfloor\frac{\mu}{\mu - 1}\right\rfloor } \nonumber \\

1594:  &  =   & \left\lfloor\frac{\mu}{1}\right\rfloor + 

1595:        \left\lfloor\frac{\mu}{2}\right\rfloor + \cdots + 

1596:        \left\lfloor\frac{\mu}{\mu}\right\rfloor 

1597:  - \left( \left\lfloor\frac{\mu}{1}\right\rfloor + 

1598:        \left\lfloor\frac{\mu}{2}\right\rfloor + \cdots + 

1599:        \left\lfloor\frac{\mu}{\delta + 2}\right\rfloor \right) - 1 \nonumber \\

1600: \label{eq:mdg-bipartite-imply-1}

1601:  & \geq & \mu \ln \mu - \mu 

1602: \left( \frac{1}{1} + \frac{1}{2} + \cdots + \frac{1}{\delta + 2}\right) \\

1603: \label{eq:mdg-bipartite-imply-2}

1604:  & \geq & \mu \ln \mu - 2 \mu \ln(\delta + 2) .

1605: \end{eqnarray}

1606: Equations~(\ref{eq:mdg-bipartite-imply-1})

1607: and~(\ref{eq:mdg-bipartite-imply-2}) hold, since $\frac{1}{2} + \frac{1}{3} +

1608: \cdots + \frac{1}{n} \leq \int\limits_{1}^{n} \frac{1}{x} \, dx = \ln n - \ln

1609: 1 = \ln n$ implies for large enough~$n$:

1610: \begin{eqnarray*}

1611: \frac{1}{1} + \frac{1}{2} + \cdots + \frac{1}{n} \leq 2 \ln n 

1612: & \mbox{\quad and \quad} &

1613: \left\lfloor\frac{n}{1}\right\rfloor + 

1614: \left\lfloor\frac{n}{2}\right\rfloor + \cdots + 

1615: \left\lfloor\frac{n}{n}\right\rfloor \geq 1 + n \ln n .

1616: \end{eqnarray*}

1617: 

1618: It is evident from the construction that $G$ has all required properties.  In

1619: particular, to see why Property~\ref{eq:mdg-bipartite-4} holds, let $S$ be any

1620: induced subgraph of $G$ that can be obtained by deleting vertices from~$V$

1621: such that $V \cap V(S) \neq \emptyset$.  Let $y_S = \max_{v \in V \cap V(S)}

1622: \degree_S(v)$.  By construction, $S$ can have only edges of the form $\{u_i,

1623: \tilde{u}_i\}$ or $\{w_j, \tilde{w}_j\}$ or edges that are added during

1624: the stages $\delta + 3, \delta + 4, \ldots , y_S$, where $\delta + i$ denotes

1625: the stage in which $\tilde{W}$ is partitioned into subsets of size $\delta +

1626: i$.  It follows that

1627: \[

1628: \max_{v \in \tilde{V}} \degree_S(v)

1629: \leq 1 + y_S - (\delta + 3) + 1 

1630: = y_S - \delta - 1

1631: < y_S - \delta,

1632: \]

1633: which proves the lemma.~\qed

1634: \begin{theorem}  

1635: \label{thm:smdgr-thetatwo-complete}

1636: For each rational number $r \geq 1$, $\smdgr$ is $\parallelnp$-complete.

1637: \end{theorem}

1638: 

1639: \sproof 

1640:   It is easy to see that $\smdgr$ is in $\parallelnp$.  To prove

1641:   $\parallelnp$-hardness of $\smdgr$, let $X$ be an arbitrary set

1642:   in~$\parallelnp$, and let $f$ be the reduction from $X$ to $\vcgeq$ stated

1643:   in Lemma~\ref{lem:vcgeq}.  For any string $x \in \sigmastar$, let $f(x) =

1644:   \pair{G_1, G_2}$.

1645: %

1646: %

1647: 

1648: 

1649:   It is convenient to consider the special case of $r=1$ and the case of $r>1$

1650:   separately in the proof of Theorem~\ref{thm:smdgr-thetatwo-complete}.  

1651: %

1652: %

1653: %

1654: We start by proving that $\smdgone$ is $\parallelnp$-complete.  

1655: %

1656: %

1657: %

1658: %

1659: We will define a graph

1660: $\widehat{G}$ and an integer $q \geq 0$ such that:

1661: \begin{eqnarray}

1662: \label{eq:minmdg-special}

1663: \minmdg(\widehat{G}) & = & \mvc(G_2) + q ; \\

1664: \label{eq:mvcmdg-special}

1665: \mvc(\widehat{G})    & = & \mvc(G_1) + q .

1666: \end{eqnarray}

1667: 

1668: The reduction mapping any given string $x$ (via the pair $\pair{G_1, G_2}$

1669: obtained according to Lemma~\ref{lem:vcgeq}) 

1670: to the graph $\widehat{G}$ such that

1671: Equations~(\ref{eq:minmdg-special}) and~(\ref{eq:mvcmdg-special}) are

1672: satisfied will establish that $X \manyone \smdgone$.  In particular, from

1673: these equations, we have that:

1674: \begin{itemize}

1675: \item $\mvc(G_2) = \mvc(G_1)$ implies $\minmdg(\widehat{G}) =

1676:   \mvc(\widehat{G})$, and

1677: \item $\mvc(G_2) > \mvc(G_1)$ implies $\minmdg(\widehat{G}) >

1678:   \mvc(\widehat{G})$.

1679: \end{itemize}

1680: Note that, due to Lemma~\ref{lem:vcgeq}, $\mvc(G_2) \geq \mvc(G_1)$.

1681:    

1682: We now describe the construction of~$\widehat{G}$.  

1683: Let $g$ be the reduction from Lemma~\ref{lem:vcsmdgone} and let $H_2 = g(G_2)$.

1684: Thus, $H_2$ is in $\smdgone$ and, by Equation~(\ref{eq:vcsmdgone}),

1685: \begin{equation}

1686: \label{eq:vcsmdgone-2}

1687: \mvc(H_2) = \mvc(G_2) + \| E(G_2) \| (\maxdegree(G_2) + 1).

1688: \end{equation}

1689: Since one can add isolated vertices to any graph $G$

1690: without affecting the values of $\mvc(G)$ or $\minmdg(G)$, we may without loss

1691: of generality assume that

1692: \begin{equation}  

1693: \label{eq:size-of-H2}

1694:         \|V(H_2)\| =  \|V(G_1)\| + \|E(G_2)\| (\maxdegree(G_2) + 1).

1695: \end{equation}

1696: 

1697: \begin{figure}[ht]  

1698: \centering

1699: \input{smdgone.eepic}

1700: \caption{The graph $\widehat{G}$ constructed from $G_1$ and~$H_2$.}

1701: \label{fig:smdgone}  

1702: \end{figure}

1703: 

1704: Look at Figure~\ref{fig:smdgone} for the construction of~$\widehat{G}$ 

1705: from $G_1$ and~$H_2$.  The graph $\widehat{G}$ consists of two subgraphs, 

1706: $L$ and~$R$, that

1707: are joined by the join operation, plus some additional vertices and edges that

1708: are connected to~$L$.  Formally, choose $2j$ new vertices $a_i$ and~$b_i$, $1

1709: \leq i \leq j$, where $j$ is a fixed integer large enough such that the degree

1710: of each vertex in $R$ is larger than the maximum degree of the vertices

1711: in~$L$.  Note that the degree of each vertex in $R$ must remain larger

1712: than the degree of any vertex in $L$ even after some vertices have been

1713: removed from~$R$.

1714: 

1715: Let $B$ be the bipartite matching with the vertex set 

1716: \[

1717: V(B) = \{ a_i \condition 1 \leq i \leq j \} \cup \{ b_i \condition 1 \leq i

1718: \leq j \}

1719: \] 

1720: and the edge set $E(B) = \{\{a_i, b_i\} \condition 1 \leq i \leq j \}$.

1721: Let $R =

1722: G_1$, and let $L$ be the graph with the vertex set $V(L) = \{ a_i \condition 1

1723: \leq i \leq j \} \cup V(H_2)$ and the edge set $E(L) = E(H_2)$.  The graph

1724: $\widehat{G}$ is defined by forming the join $L \bowtie R$, i.e., there are 

1725: edges

1726: connecting each vertex of $L$ with each vertex of~$R$, plus attaching the

1727: vertices~$b_i$, $1 \leq i \leq j$, to $L$ by adding the $j$ edges from~$E(B)$.

1728:     

1729: We first consider $\minmdg(\widehat{G})$.  By our choice of~$j$, each vertex in

1730: $R$ has a degree larger than the degree of any vertex not in~$R$.  Hence, on

1731: input~$\widehat{G}$, the MDG algorithm first deletes all vertices from~$R$.

1732: Subsequently, it can find a minimum vertex cover of~$H_2$, which has size

1733: $\mvc(G_2) + \|E(G_2)\| (\maxdegree(G_2) + 1)$ by

1734: Equation~(\ref{eq:vcsmdgone-2}), and eventually it can choose, say, the

1735: vertices~$a_i$, $1 \leq i \leq j$, to cover the edges of~$B$.  Hence,

1736: \begin{eqnarray*}

1737: \minmdg(\widehat{G}) & = & 

1738:  \|V(G_1)\| + \mvc(G_2) + \| E(G_2)\| (\maxdegree(G_2) + 1) + j \\

1739:  & \stackrel{(\ref{eq:size-of-H2})}{=} & \mvc(G_2) + \| V(H_2)\| + j.

1740: \end{eqnarray*}

1741: We now consider $\mvc(\widehat{G})$.  

1742: Since every vertex cover of $\widehat{G}$ must

1743: contain all vertices of $L$ or all vertices of $R$ to cover the edges

1744: connecting $L$ and~$R$, it follows from Equations~(\ref{eq:vcsmdgone-2})

1745: and~(\ref{eq:size-of-H2}) that:

1746: \begin{eqnarray*}

1747: \mvc(\widehat{G})

1748:  & = & \min\{ \|V(G_1)\| + \mvc(H_2) + j,\ \|V(H_2)\| + j + \mvc(G_1) \}  \\

1749:  & = & \min\{ \mvc(G_2) + \|V(H_2)\| + j,\ \mvc(G_1) + \| V(H_2)\| + j\}.

1750: \end{eqnarray*}

1751: Since $\mvc(G_2) \geq \mvc(G_1)$, it follows that

1752: \[

1753:         \mvc(\widehat{G}) = \mvc(G_1) + \| V(H_2)\| + j.

1754: \]

1755: Hence, setting $q = \| V(H_2)\| + j$, Equations~(\ref{eq:minmdg-special})

1756: and~(\ref{eq:mvcmdg-special}) are satisfied, which completes the proof that

1757: $\smdgone$ is $\parallelnp$-complete.

1758: 

1759: We now turn to the proof that $\smdgr$ is $\parallelnp$-complete for $r>1$.

1760: Fix any rational number $r = \frac{\ell}{m}$, where $\ell$ and $m$ are

1761: integers with $1 \leq m < \ell$.  Without loss of generality, we may assume

1762: that $\gcd(\ell - m, m) = 1$, where $\gcd(a, b)$ denotes the greatest common

1763: divisor of the integers $a$ and~$b$.

1764: Recall that the pair $\pair{G_1, G_2} = f(x)$ of graphs is obtained using the

1765: reduction $f$ from $X$ to $\vcgeq$ according to Lemma~\ref{lem:vcgeq}; hence,

1766: $\mvc(G_2) \geq \mvc(G_1)$.  

1767: 

1768: We will define a graph $\widehat{G}_{r}$ and integers $p, q \geq 0$ such that:

1769: \begin{eqnarray}

1770: \label{eq:G1G2-minmdg}

1771: \minmdg(\widehat{G}_{r}) & = & r (p \cdot \mvc(G_2) + q); \\

1772: \label{eq:G1G2-mvc}

1773: \mvc(\widehat{G}_{r})   & = & p \cdot \mvc(G_1) + q.

1774: \end{eqnarray}

1775: 

1776: The reduction mapping any given string $x$ (via the pair $\pair{G_1, G_2}$

1777: obtained according to Lemma~\ref{lem:vcgeq}) to the graph 

1778: $\widehat{G}_{r}$ such that

1779: Equations~(\ref{eq:G1G2-minmdg}) and~(\ref{eq:G1G2-mvc}) are satisfied will

1780: establish that $X \manyone \smdgr$.  In particular, from these equations, we

1781: have that:

1782: \begin{itemize}

1783: \item $\mvc(G_2) = \mvc(G_1)$ implies $\minmdg(\widehat{G}_{r}) = r \cdot

1784:   \mvc(\widehat{G}_{r})$, and

1785: \item $\mvc(G_2) > \mvc(G_1)$ implies $\minmdg(\widehat{G}_{r}) > r \cdot

1786:   \mvc(\widehat{G}_{r})$.

1787: \end{itemize}

1788: 

1789: We now describe the construction of~$\widehat{G}_{r}$:

1790: \begin{itemize}

1791: \item

1792:   Let $g$ be the reduction from

1793:   Lemma~\ref{lem:vcsmdgone} and 

1794:   let $H_2 = g(G_2)$.

1795:   Thus, $H_2 \in \smdgone$ and Equation~(\ref{eq:vcsmdgone-2}) holds:

1796: \[

1797: \mvc(H_2) = \mvc(G_2) + \| E(G_2) \| (\maxdegree(G_2) + 1).

1798: \]

1799: 

1800: \item Let $G_{1}^{1}, G_{1}^{2}, \ldots, G_{1}^{m}$ be $m$ pairwise disjoint

1801:   copies of~$G_1$, and let $H_{2}^{1}, H_{2}^{2}, \ldots, H_{2}^{\ell}$ be

1802:   $\ell$ pairwise disjoint copies of~$H_2$.

1803:   

1804: \item Let $\tilde{U} = \bigcup_{i = 1}^{\ell} H_{2}^{i}$ be the disjoint union

1805:   of these copies of~$H_2$, and rename the vertices of $\tilde{U}$ by

1806:   $V(\tilde{U}) = \{ \tilde{u}_1, \tilde{u}_2, \ldots ,

1807:   \tilde{u}_{\ell \cdot \|V(H_2)\|} \}$.

1808:   

1809: \item Let $Z = \bigcup_{i = 1}^{m} G_{1}^{i}$ be the disjoint union of

1810:   these copies of~$G_1$, and rename the vertices of $Z$ by $V(Z) = \{ z_1,

1811:   z_2, \ldots , z_{m \cdot \|V(G_1)\|} \}$.

1812:   

1813: \item To apply Lemma~\ref{lem:mdg-bipartite}, choose $n_1 = \ell \cdot

1814:   \|V(H_2)\|$, $n_2 \geq m \cdot \|V(G_1)\|$, and $\delta = \maxdegree(H_2) +

1815:   1$, where the exact value of $n_2$ will be specified below.  Choose the

1816:   constant $\mu$ so as to satisfy Equation~(\ref{eq:mdg-bipartite}):

1817: \[

1818: \mu (\ln \mu - 2 \ln (\delta + 2) - 1) \geq n_2 + n_1 .

1819: \]

1820:     

1821: \item Given the constants $n_1$, $n_2$, $\delta$, and~$\mu$, 

1822:   define $\widehat{G}_{r}$

1823:   to be the bipartite graph $G$ from Lemma~\ref{lem:mdg-bipartite} extended by

1824:   the edges between the $\tilde{u}_i$ vertices that were added above to

1825:   represent the structure of the copies of~$H_2$, and extended by the edges

1826:   between the $z_j$ vertices that were added above to represent the structure

1827:   of the copies of~$G_1$.  

1828:   That is, unlike~$G$, the graph $\widehat{G}_{r}$ is no

1829:   longer a bipartite graph.  

1830:   Formally, the vertex set of $\widehat{G}_{r}$ is given by

1831: \begin{eqnarray*}

1832: V(\widehat{G}_{r}) & = & V(G) = V \cup \tilde{V} , \ \ \mbox{ where} \\

1833: V & = &  \{ u_1, u_2, \ldots , u_{n_1}, w_1, w_2, \ldots , w_{\mu}, z_1,

1834: z_2, \ldots z_{n_2}\} \ \ \mbox{ and} \\

1835: \tilde{V} & = & \{ \tilde{u}_1, \tilde{u}_2,

1836: \ldots , \tilde{u}_{n_1}, \tilde{w}_1, \tilde{w}_2, \ldots ,

1837: \tilde{w}_{\mu} \} ,\end{eqnarray*}

1838: and the edge set of $\widehat{G}_{r}$ is given by

1839: $E(\widehat{G}_{r}) = E(G) \cup E(\tilde{U}) \cup E(Z)$,

1840: where $E(G)$ is constructed as in the proof of Lemma~\ref{lem:mdg-bipartite}.

1841: \end{itemize}

1842: 

1843: This completes the construction of~$\widehat{G}_{r}$.  We now prove

1844: Equations~(\ref{eq:G1G2-minmdg}) and~(\ref{eq:G1G2-mvc}).

1845: 

1846: \begin{enumerate}

1847: \item \label{enum:smdgr-1} We first consider $\minmdg(\widehat{G}_{r})$.  By

1848:   construction, for each vertex $v$ in $\tilde{V}$, we have

1849: \begin{equation}

1850: \label{eq:prop4}

1851: \degree_{\widehat{G}_{r}}(v) \leq 

1852: \degree_{G}(v) + \maxdegree(H_2) < \degree_{G}(v) +

1853: \delta.

1854: \end{equation}

1855: Let $S$ be any induced subgraph of $\widehat{G}_{r}$ 

1856: that can be obtained by deleting

1857: vertices from~$V$ such that $V \cap V(S) \neq \emptyset$.

1858: Property~\ref{eq:mdg-bipartite-4} of Lemma~\ref{lem:mdg-bipartite} and

1859: Equation~(\ref{eq:prop4}) imply that

1860: \[

1861: \max_{v \in V \cap V(S)} \degree_S(v) > \max_{v \in \tilde{V}} \degree_S(v).

1862: \]

1863: Hence, on input~$\widehat{G}_{r}$, 

1864: the MDG algorithm starts by choosing the $n_1 + \mu + n_2$ 

1865: vertices from~$V$, which isolates each vertex $\tilde{w}_i \in

1866: \tilde{V}$ and leaves $\ell$ isolated copies of $H_2$.  Subsequently, since

1867: $H_2 \in \smdgone$, the MDG algorithm can choose a minimum vertex cover in

1868: each of these $\ell$ copies of $H_2$.  By Equation~(\ref{eq:vcsmdgone-2}),

1869: \[

1870: \mvc(H_2) = \mvc(G_2) + \| E(G_2) \| (\maxdegree(G_2) + 1) ,

1871: \]

1872: and hence,

1873: \[

1874: \minmdg(\widehat{G}_{r}) = n_1 + \mu + n_2 + \ell(\mvc(G_2) + \|E(G_2)\|

1875: (\maxdegree(G_2) + 1)) .

1876: \]

1877: 

1878: \item \label{enum:smdgr-2} We now consider $\mvc(\widehat{G}_{r})$. 

1879:    Define the set $C = \tilde{V} \cup D$, 

1880:   where $D$ with $\|D\| = m \cdot \mvc(G_1)$ is a

1881:   minimum vertex cover of~$Z$.  It is obvious from the construction of

1882:   $\widehat{G}_{r}$ that $C$ is a minimum vertex cover 

1883:   of~$\widehat{G}_{r}$.  Hence,

1884: \[

1885: \mvc(\widehat{G}_{r}) = n_1 + \mu + m \cdot \mvc(G_1).

1886: \]

1887: \end{enumerate}

1888:    

1889: To complete the proof, 

1890: we have to choose $n_2 \geq m \cdot \|V(G_1)\|$ such that

1891: Equations~(\ref{eq:G1G2-minmdg}) and~(\ref{eq:G1G2-mvc}) are satisfied for

1892: suitable integers $p$ and~$q$.

1893: Setting $p = m$ and $q = n_1 + \mu$ and requiring

1894: \begin{equation}  \label{mdg::eq::gl3}

1895: n_1 + n_2 + \mu + \ell \cdot \|E(G_2)\| (\maxdegree(G_2) + 1)

1896:  = r(n_1 + \mu )

1897: \end{equation}

1898: or, equivalently,

1899: \begin{equation} \label{mdg::eq::gl4}

1900:         m \cdot n_2 + m \cdot \ell \cdot \|E(G_2)\| (\maxdegree(G_2) + 1))

1901:  = (\ell - m) n_1 + (\ell - m) \mu

1902: \end{equation}

1903: satisfies Equations~(\ref{eq:G1G2-minmdg}) and~(\ref{eq:G1G2-mvc}).  Our

1904: assumption that $\gcd(\ell - m, m) = 1$ implies that

1905: Equation~(\ref{mdg::eq::gl4}) has integer solutions in the variables $n_2$

1906: and~$\mu$.  It is easy

1907: to see that one such solution, say $(n_2, \mu)$, simultaneously (a)~satisfies

1908: Equation~(\ref{eq:mdg-bipartite}), (b)~satisfies that both $n_2$

1909: and~$\mu$ are polynomially bounded in the size of the input of the reduction

1910: being described, and (c)~can be computed 

1911: efficiently~\cite{cla-for:j:efficient-solution-diophantine-equations}.

1912: This completes the proof of the theorem.~\qed

1913: 

1914: 

1915: \bigskip

1916: 

1917: \noindent {\bf Acknowledgments:}  

1918: We thank Dieter Kratsch and Andreas Brandst\"adt for interesting discussions

1919: on graph theory and graph-theoretical notation.

1920: 

1921: 

1922: %

1923: 

1924: \bibliography{/home/inf1/rothe/BIGBIB/joergbib}

1925: %

1926: %

1927: 

1928: %

1929: 

1930: \end{document}

1931: 

1932: \clearpage

1933: 

1934: \appendix

1935: 

1936: \section{Proof of Lemma~\ref{lem:vcsedone}}

1937: \label{app:vcsedone}

1938: 

1939: \sproof

1940:   Given any graph~$G$, we construct a graph $g(G) = H \in \sedone$ such

1941:   that

1942: \begin{equation}

1943: \label{eq:vcsedone}

1944: \mvc(H) = 2 (\mvc(G) + \| V(G) \|).

1945: \end{equation}

1946: Hence, the reduction mapping any pair $\pair{G,k}$ to the pair $\pair{H, 2 (k

1947:   + \| V(G)\|)}$ shows that $\vc \manyone \vcsedone$, so $\vcsedone$ is

1948: $\np$-hard.

1949: 

1950: $H$ is constructed from $G$ as follows.  For each vertex $v\in V(G)$, create a

1951: component $G_v$ that is defined by the vertex set 

1952: $V(G_v) = \{ v_1, v_2, v_3, v_4 \}$ and the edge set

1953: $E(G_v) = \{ \{ v_1, v_2\} , \{ v_3, v_4\}, \{ v_1, v_3\} \}$.

1954: 

1955: Define the graph $H$ by joining every pair of components that correspond to

1956: adjacent vertices of~$G$:

1957: \begin{eqnarray*}

1958:          V(H) & = & \bigcup_{v\in V(G)} V(G_v); \\

1959:          E(H) & = & \{\{ a_i,b_j\} \condition \mbox{$\{a,b\} \in E(G)$ and 

1960:                       $i,j \in \{ 1, 2, 3, 4 \}$} \}

1961:                     \cup \bigcup_{v\in V(G)} E(G_v).

1962: \end{eqnarray*}  

1963: 

1964: We now prove Equation~(\ref{eq:vcsedone}).  Let $C$ be a minimum vertex cover

1965: of~$G$, i.e., $\mvc(G) = \| C \|$.  Construct a vertex cover $D$ of $H$ as

1966: follows.  For each vertex $v \in C$, add $v_1, v_2, v_3$, and $v_4$ to~$D$;

1967: and for each vertex $w \in V(G) - C$, add $w_1$ and $w_3$ to~$D$.  Hence,

1968: $\|D\| = 2 (\| C \| + \| V(G) \|)$. Since $mvc(H) \leq \| D \|$, it follows

1969: that $\mvc(H) \leq 2 (\mvc(G) + \| V(G) \|)$.

1970: 

1971: Conversely, let $D$ be a minimum vertex cover of~$H$, i.e., $\mvc(H) = \|D\|$.

1972: Then, it holds that:

1973: \begin{itemize}

1974: \item for each edge $\{ u,v\}\in E(G)$, $V(G_u) \seq D$ or $V(G_v) \seq D$;

1975: \item for each vertex $v\in V(G)$, $\|D \cap V(G_v)\| \geq 2$.

1976: \end{itemize}

1977: Hence,

1978: \begin{eqnarray*}

1979: \| D \| & \geq & 4 \cdot \mvc(G) + 2 (\|V(G)\| - \mvc(G)) \\

1980:         & = & 2 (\mvc(G) + \|V(G)\|).

1981: \end{eqnarray*}

1982: It follows that $ \mvc(H) \geq 2 (\mvc(G) + \| V(G) \|)$, which proves

1983: Equation~(\ref{eq:vcsedone}).

1984: 

1985: It remains to prove that $H \in \sedone$.  Let $C$ be a minimum vertex cover

1986: of~$G$.  The edge deletion algorithm can find a vertex cover of $H$ as

1987: follows.  For every vertex $v\in C$, choose the edges $\{ v_1, v_2\}$ and $\{

1988: v_3, v_4\}$.  For the remaining vertices $w \in V(G) - C$, choose the edge $\{

1989: w_1, w_3\}$.  Thus, $\mined(H) = 2 (\mvc(G) + \| V(G) \|)$.  By

1990: Equation~(\ref{eq:vcsedone}), $\mined(H) = \mvc(H)$, so $H \in \sedone$.~\qed

1991: 

1992: 

1993: \section{Proof of Lemma~\ref{lem:vcsmdgone}}

1994: \label{app:vcsmdgone}

1995: 

1996: \sproof

1997:   Given any graph~$G$, we construct a graph $g(G) = H \in \smdgone$ such

1998:   that

1999: \begin{equation}

2000: \label{eq:vcsmdgone}

2001: \mvc(H) = \mvc(G) + \| E(G) \| (\maxdegree(G) + 1).

2002: \end{equation}

2003: Hence, the reduction mapping any pair $\pair{G,k}$ to the pair $\pair{H, k + \|

2004:   E(G)\| (\maxdegree(G) + 1)}$ shows that $\vc \manyone \vcsmdgone$, so

2005: $\vcsmdgone$ is $\np$-hard.

2006: 

2007: To construct $H$ from~$G$, we replace each edge of $G$ by a gadget that

2008: contains a complete bipartite graph of size $2 (\maxdegree(G) + 1)$.

2009: Formally, $H$ is defined by:

2010: \begin{eqnarray*}

2011:          V(H) & = & V(G) \cup \bigcup_{e \, = \, \{u,v\} \,\in\, E(G)} 

2012:          \{u_{i}^{e} \condition 1 \leq i \leq \maxdegree(G) + 1\}

2013:          \cup \{v_{i}^{e} \condition 1 \leq i \leq \maxdegree(G) + 1\} ; \\

2014:          E(H) & = & \bigcup_{e \, = \, \{u,v\} \, \in\, E(G)} 

2015:          \left(\{ \{u_{i}^{e}, v_{j}^{e}\} 

2016:          \condition 1 \leq i, j \leq \maxdegree(G) + 1\} 

2017:          \cup \{ \{u, u_{1}^{e}\} \} \cup \{ \{v, v_{1}^{e}\} \}\right).

2018: \end{eqnarray*}  

2019: 

2020: We now prove Equation~(\ref{eq:vcsmdgone}).  Let $C$ be a minimum vertex cover

2021: of~$G$, i.e., $\mvc(G) = \| C \|$.  Note that $\{u, v\} \cap C \neq \emptyset$

2022: for each edge $\{u, v\}$ in~$E(G)$.  Construct a vertex cover $D$ of $H$ as

2023: follows:

2024: \begin{itemize}

2025: \item $D$ contains all vertices from $C$.

2026: \item For every edge $e = \{ u, v \}$ in $E(G)$, add to~$D$:

2027: \begin{itemize}

2028: \item either all vertices $u_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, if

2029:   $u \not\in C$ or if both $u$ and $v$ are in~$C$;

2030: \item or all vertices $v_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, if $v

2031:   \not\in C$.

2032: \end{itemize}

2033: \end{itemize}

2034: It follows that $\mvc(H) \leq \mvc(G) + \| E(G) \| (\maxdegree(G) + 1)$.

2035: 

2036: 

2037: Conversely, let $D$ be a minimum vertex cover of~$H$, i.e., $\mvc(H) = \|D\|$.

2038: Construct a vertex cover $C$ of $G$ as follows.  Initially, set $C = D$.  Let

2039: $e = \{ u, v \}$ be any fixed edge in~$E(G)$.  Suppose that at least one

2040: vertex from $\{ u,v\}$ is in~$D$.  Since $D$ is a vertex cover of~$H$, it

2041: contains at least $\maxdegree(G) + 1$ of the vertices $u_{i}^{e}$

2042: and~$v_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, that correspond to the

2043: edge~$e$.  Remove any $\maxdegree(G) + 1$ such vertices from~$C$.  Suppose now

2044: that neither $u$ nor $v$ is in~$D$.  Since $D$ is a vertex cover of~$H$, it

2045: contains at least $\maxdegree(G) + 2$ of the vertices $u_{i}^{e}$

2046: and~$v_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, that correspond to the

2047: edge~$e$.  Remove any $\maxdegree(G) + 2$ such vertices from~$C$, and add to

2048: $C$ one of $u$ or $v$ instead.  Since the set $C$ thus obtained is a vertex

2049: cover of~$G$, we have $\mvc(H) \geq \mvc(G) + \|E(G)\| (\maxdegree(G) + 1)$,

2050: which proves Equation~(\ref{eq:vcsmdgone}).

2051: 

2052: It remains to prove that $H \in \smdgone$.  Let $C$ be a minimum vertex cover

2053: of~$G$.  The maximum-degree greedy algorithm can find a vertex cover of $H$ as

2054: follows.  For every edge $e = \{ u, v \}$ in~$E(G)$, the MDG algorithm on

2055: input $H$ can choose:

2056: \begin{itemize}

2057: \item either all vertices $u_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, if

2058:   $u \not\in C$ or if both $u$ and $v$ are in~$C$;

2059: \item or all vertices $v_{i}^{e}$, $1 \leq i \leq \maxdegree(G) + 1$, if $v

2060:   \not\in C$.

2061: \end{itemize}

2062: Note that the MDG heuristic can always do so, since every vertex in $V(G)$ has

2063: degree at most $\maxdegree(G)$.  Subsequently, all vertices that are not in

2064: $C$ are isolated. Thus, the MDG algorithm can now choose all vertices

2065: from~$C$.  Hence, $\minmdg(H) = \mvc(G) + \| E(G)\| (\maxdegree(G) + 1)$.  By

2066: Equation~(\ref{eq:vcsmdgone}), $\minmdg(H) = \mvc(H)$, so $H \in

2067: \smdgone$.~\qed

2068: 

2069: 

2070: 

2071: \section{Proof of Lemma~\ref{lem:mdg-bipartite}}

2072: \label{app:mdg-bipartite}

2073: 

2074: \sproof

2075:   Let the constants $n_1$, $n_2$, $\delta$, and $\mu$ be given such that

2076:   Equation~(\ref{eq:mdg-bipartite}) is satisfied.  We describe the

2077:   construction of the graph~$G$.  As stated in the lemma, the vertex set of

2078:   $G$ is given by $V(G) = V \cup \tilde{V}$, where $V$ and~$\tilde{V}$

2079:   are two disjoint independent sets.

2080:   

2081:   Rename the vertices of $V$ by $V = \{ \alpha_1, \alpha_2, \ldots ,

2082:   \alpha_{n_1 + \mu + n_2}\}$.  Let $\tilde{W} = \{ \tilde{w}_1,

2083:   \tilde{w}_2, \ldots , \tilde{w}_{\mu} \}$.  The edge set of $G$ is

2084:   defined as follows:

2085: \begin{itemize}

2086: \item Create the edges $\{ u_i, \tilde{u}_i\}$ for each $i$ with $1 \leq i

2087:   \leq n_1$ and the edges $\{ w_j, \tilde{w}_j\}$ for each $j$ with $1 \leq

2088:   j \leq \mu$.

2089:   

2090: \item Partition $\tilde{W}$ into $\left\lfloor\frac{\mu}{\delta +

2091:       3}\right\rfloor$ disjoint sets $\tilde{W}_{1}^{\delta + 3},

2092:   \tilde{W}_{2}^{\delta + 3}, \ldots ,

2093:   \tilde{W}_{\left\lfloor\frac{\mu}{\delta + 3}\right\rfloor}^{\delta + 3}$

2094:   of size $\delta + 3$ each, possibly leaving out some vertices

2095:   from~$\tilde{V}$ and taking care that no vertex in

2096:   $\tilde{W}_{i}^{\delta + 3}$ already is connected with~$\alpha_i$, $1

2097:   \leq i \leq \left\lfloor\frac{\mu}{\delta + 3} \right\rfloor$.  For each $i$

2098:   with $1 \leq i \leq \left\lfloor\frac{\mu}{\delta + 3} \right\rfloor$,

2099:   connect $\alpha_i$ with each vertex in $\tilde{W}_{i}^{\delta + 3}$ by an

2100:   edge.

2101:   

2102: \item Partition $\tilde{W}$ into $\left\lfloor\frac{\mu}{\delta + 4}

2103:   \right\rfloor$ disjoint sets $\tilde{W}_{1}^{\delta + 4},

2104:   \tilde{W}_{2}^{\delta + 4}, \ldots ,

2105:   \tilde{W}_{\left\lfloor\frac{\mu}{\delta + 4}\right\rfloor}^{\delta + 4}$

2106:   of size $\delta + 4$ each, possibly leaving out some vertices

2107:   from~$\tilde{V}$ and taking care that no vertex in

2108:   $\tilde{W}_{i}^{\delta + 4}$ already is connected

2109:   with~$\alpha_{\left\lfloor\frac{\mu}{\delta + 3}\right\rfloor + i}$, $1 \leq

2110:   i \leq \left\lfloor\frac{\mu}{\delta + 4} \right\rfloor$.  For each $i$ with

2111:   $1 \leq i \leq \left\lfloor\frac{\mu}{\delta + 4} \right\rfloor$, connect

2112:   $\alpha_{ \left\lfloor\frac{\mu}{\delta + 3}\right\rfloor + i}$ with each

2113:   vertex in $\tilde{W}_{i}^{\delta + 4}$ by an edge.

2114:   

2115: \item Partition $\tilde{W}$ into $\left\lfloor\frac{\mu}{\delta + 5}

2116:   \right\rfloor$ disjoint sets $\tilde{W}_{1}^{\delta + 5},

2117:   \tilde{W}_{2}^{\delta + 5}, \ldots ,

2118:   \tilde{W}_{\left\lfloor\frac{\mu}{\delta + 5}\right\rfloor}^{\delta + 5}$

2119:   of size $\delta + 5$ each, possibly leaving out some vertices

2120:   from~$\tilde{V}$ and taking care that no vertex in

2121:   $\tilde{W}_{i}^{\delta + 5}$ already is connected

2122:   with~$\alpha_{\left\lfloor\frac{\mu}{\delta + 3}\right\rfloor +

2123:     \left\lfloor\frac{\mu}{\delta + 4}\right\rfloor + i}$, $1\leq i \leq

2124:   \left\lfloor\frac{\mu}{\delta + 5}\right\rfloor$.  For each $i$ with $1\leq

2125:   i \leq \left\lfloor\frac{\mu}{\delta + 5}\right\rfloor$, connect $\alpha_{

2126:     \left\lfloor\frac{\mu}{\delta + 3}\right\rfloor +

2127:     \left\lfloor\frac{\mu}{\delta + 4}\right\rfloor + i}$ with each vertex in

2128:   $\tilde{W}_{i}^{\delta + 5}$ by an edge.

2129:   

2130: \item Continue in this way until all vertices $\alpha_i$ are connected with

2131:   vertices in~$\tilde{W}$.

2132: \end{itemize}

2133: 

2134: The construction is possible, since Equation~(\ref{eq:mdg-bipartite}) implies

2135: \begin{equation}

2136:        \left\lfloor\frac{\mu}{\delta + 3}\right\rfloor + 

2137:        \left\lfloor\frac{\mu}{\delta + 4}\right\rfloor + \cdots + 

2138:        \left\lfloor\frac{\mu}{\mu - 1}\right\rfloor \geq

2139:        n_1 + \mu + n_2 ,

2140: \end{equation}

2141: and thus there are enough vertices in~$\tilde{W}$.  To see why, note that 

2142: \begin{eqnarray}

2143: \lefteqn{ \left\lfloor\frac{\mu}{\delta + 3}\right\rfloor + 

2144:        \left\lfloor\frac{\mu}{\delta + 4}\right\rfloor + \cdots + 

2145:        \left\lfloor\frac{\mu}{\mu - 1}\right\rfloor } \nonumber \\

2146:  &  =   & \left\lfloor\frac{\mu}{1}\right\rfloor + 

2147:        \left\lfloor\frac{\mu}{2}\right\rfloor + \cdots + 

2148:        \left\lfloor\frac{\mu}{\mu}\right\rfloor 

2149:  - \left( \left\lfloor\frac{\mu}{1}\right\rfloor + 

2150:        \left\lfloor\frac{\mu}{2}\right\rfloor + \cdots + 

2151:        \left\lfloor\frac{\mu}{\delta + 2}\right\rfloor \right) - 1 \nonumber \\

2152: \label{eq:mdg-bipartite-imply-1}

2153:  & \geq & \mu \ln \mu - \mu 

2154: \left( \frac{1}{1} + \frac{1}{2} + \cdots + \frac{1}{\delta + 2}\right) \\

2155: \label{eq:mdg-bipartite-imply-2}

2156:  & \geq & \mu \ln \mu - 2 \mu \ln(\delta + 2) .

2157: \end{eqnarray}

2158: Equations~(\ref{eq:mdg-bipartite-imply-1})

2159: and~(\ref{eq:mdg-bipartite-imply-2}) hold, since $\frac{1}{2} + \frac{1}{3} +

2160: \cdots + \frac{1}{n} \leq \int\limits_{1}^{n} \frac{1}{x} \, dx = \ln n - \ln

2161: 1 = \ln n$ implies for large enough~$n$:

2162: \begin{eqnarray*}

2163: \frac{1}{1} + \frac{1}{2} + \cdots + \frac{1}{n} \leq 2 \ln n 

2164: & \mbox{\quad and \quad} &

2165: \left\lfloor\frac{n}{1}\right\rfloor + 

2166: \left\lfloor\frac{n}{2}\right\rfloor + \cdots + 

2167: \left\lfloor\frac{n}{n}\right\rfloor \geq 1 + n \ln n .

2168: \end{eqnarray*}

2169: 

2170: It is evident from the construction that $G$ has all required properties.  In

2171: particular, to see why Property~\ref{eq:mdg-bipartite-4} holds, let $S$ be any

2172: induced subgraph of $G$ that can be obtained by deleting vertices from~$V$

2173: such that $V \cap V(S) \neq \emptyset$.  Let $y_S = \max_{v \in V \cap V(S)}

2174: \degree_S(v)$.  By construction, $S$ can have only edges of the form $\{u_i,

2175: \tilde{u}_i\}$ or $\{w_j, \tilde{w}_j\}$ or edges that are added during

2176: the stages $\delta + 3, \delta + 4, \ldots , y_S$, where $\delta + i$ denotes

2177: the stage in which $\tilde{W}$ is partitioned into subsets of size $\delta +

2178: i$.  It follows that

2179: \[

2180: \max_{v \in \tilde{V}} \degree_S(v)

2181: \leq 1 + y_S - (\delta + 3) + 1 

2182: = y_S - \delta - 1

2183: < y_S - \delta,

2184: \]

2185: which proves the lemma.~\qed

2186: 

2187: 

2188: 

2189: 

2190: 

2191: %

2192: 

2193: %

2194: %

2195: %

2196: %

2197: 

2198: