1: 

2: 

3: 

4: \magnification=\magstep1

5: \input epsf

6: 

7: 

8: \input amstex

9: \documentstyle{amsppt}

10: 

11: % Comment out EITHER the next 3 lines (PicTeX) ...

12: 

13: %\input pictex

14: %\setcoordinatesystem units <1in,1in>

15: %\linethickness 0.7pt

16: 

17: % ... OR the next 4 lines (Eplain).

18: 

19: %\let\noarrow=0    % comment out if using "\commdiag"

20: \input eplain

21: \input ruled

22: \let\ref=\eplainref  %see Eplain document (v2.6) page 4.

23: 

24: \TagsOnRight

25: \parskip 1ex

26: 

27: \hsize=6.5truein

28: \vsize=9truein

29: 

30: \loadbold

31: \loadeurm

32: 

33: \font\bbbf=cmbx12 scaled\magstep1

34: \font\bbf=cmbx12

35: \font\bbigbf=cmbx12 scaled\magstep1

36: \font\bi=cmbxti10	% bold italic

37: \font\bigbf=cmbx12

38: \font\bigbi=cmbxti10 scaled\magstep1	%big bold italic

39: \font\bigit=cmti12

40: \font\bigrm=cmr12 

41: \font\sc=cmcsc10

42: \font\smallrm=cmr9

43: \font\smallit=cmti9

44: \font\smallbf=cmbx9

45: \font\smallbi=cmbxti10 at 8pt

46: \font\sf=cmss10

47: \font\script=eurm10

48: \font\bigsf=cmss12

49: \font\bigtt=cmtt12

50: \font\smalltt=cmtt8

51: \font\ssf=cmss8 

52: \font\ssmallrm=cmr8

53: \font\smallsl=cmsl8

54: \document

55: 

56: \define\de{\define}

57: \de\AAA{{\tx{${\scr A}$}}}

58: \de\AAa{{\tx{${\Bbb A}$}}}

59: \de\ABBB{{\tx{$A^\BBB$}}}

60: \de\ABN{{\tx{$A^{B,N}$}}}

61: \de\AN{{\tx{$A^N$}}}

62: \de\Ab{{\bf A}}

63: \de\Abi{{\bi A}}

64: \de\Abar{{\tx{$\bar A$}}}

65: \de\Abars{\tx{$\Abar_s$}}

66: \de\Abaru{\tx{$\Abar^u$}}

67: \de\AinKK{\tx{$A \in \KK$}}

68: \de\AinAlgSig{\tx{$A \in \AlgSig$}}

69: \de\AinNStdAlgSig{\tx{$A \in \NStdAlgSig$}}

70: \de\AinStdAlgSig{\tx{$A \in \StdAlgSig$}}

71: \de\Akvec{\tx{$A[\kvec]$}}

72: \de\Akxvec{\tx{$A[\kxvec]$}}

73: \de\Aklxvec{\tx{$A[\kvec,\lxvec]$}}

74: \de\Alg{\tx{\bi Alg}}

75: \de\AlgSig{\Alg\,(\Sig)}

76: \de\Alvec{\tx{$A[\lvec\,]$}}

77: \de\Alxvec{\tx{$A[\lxvec]$}}

78: \de\Aps{\tx{\sf Ap}}

79: \de\As{\tx{$A_s$}}

80: \de\Asbar{\tx{$\Abar_s$}}

81: \de\Asx{\tx{$A_s^*$}}

82: \de\AssLang{\tx{\bi Ass}\Lang}

83: \de\Assumption{\pr{Assumption}\sl}

84: \de\Assumptionn#1{\pr{Assumption #1}\sl}

85: \de\Assumptions{\pr{Assumptions}\sl}

86: \de\Asuu{\tx{$A^\uu_s$}}

87: \de\At{\tx{\bi At}}

88: \de\AtSig{\At(\Sig)}

89: \de\AtSt{\tx{\bi AtSt}}

90: \de\AtStSig{\tx{$\AtSt(\Sig)$}}

91: \de\Atil{{\tx{$\til A$}}}

92: \de\Au{\tx{$A^u$}}

93: \de\Aul{{\tx{$\ul A$}}}

94: \de\Auu{\tx{$A^\uu$}}

95: \de\AuuN{\tx{$A^{\uu,N}$}}

96: \de\Av{\tx{$A^v$}}

97: \de\Aw{\tx{$A^w$}}

98: \de\Ax{{\tx{$A^*$}}}

99: \de\Axs{\tx{$A^*_s$}}

100: \de\Axx{(\Ax)\str}

101: \de\Bbi{\tx{\bi B}}

102: \de\BB{{\tx{${\Bbb B}$}}}

103: \de\BBB{{\tx{${\scr B}$}}}

104: \de\BBbar{\tx{$\ol \BB$}}

105: \de\BBuu{\tx{${\Bbb B}^\uu$}}

106: \de\Bb{\tx{\bf B}}

107: \de\Bool{\tx{\bi Bool}}

108: \de\BoolSig{\tx{$\Bool(\Sig)$}}

109: \de\Bs{\tx{\sf B}}

110: \de\Bss{{\tx{\ssf B}}}

111: \de\Btil{\tx{$\til B$}}

112: \de\Bul{\tx{$\ul B$}}

113: \de\CC{\tx{${\Bbb C}$}}

114: \de\CCn{\tx{$\CC^n$}}

115: \de\CCC{\tx{$\scr C$}}

116: \de\CCCL{\tx{${\CCC}^<$}}

117: \de\CR{\tx{\bi CR}}

118: \de\CRSig{\tx{$\CR(\Sig)$}}

119: \de\CT{{\tx{\bi CT}}}

120: \de\CTSig{\CT(\Sig)}

121: \de\CTsmall{{\tx{\smallbi CT}}}

122: \de\Cbar{\tx{$\ol C$}}

123: \de\Cf{{\it Cf\.}}

124: \de\Chead#1{\bn\cbb{#1}\bn}

125: \de\Comp{\tx{\bi Comp}}

126: \de\CompA{\tx{$\Comp^A$}}

127: \de\CompLengthA{\tx{$\Comp\Length^A$}}

128: \de\CompSeqA#1{\tx{$\Comp\Seq^A(#1)$}}

129: \de\Compu{\tx{\bi Compu}}

130: \de\Con{\tx{\bi Con}}

131: \de\Cond{\tx{\bi Cond}}

132: \de\CondSig{\tx{$\Cond(\Sig)$}}

133: \de\Constrn#1{\pr{Construction \ #1}\rm}

134: \de\Cont{\tx{\bi Cont}}

135: \de\ContA{\tx{$\Cont(A)$}}

136: \de\Conventionn#1{\pr{Convention \ #1}\sl}

137: \de\Cor{\pr{Corollary}\sl}

138: \de\Corn#1{\prn{Corollary #1}\sl}

139: \de\Cors{\pr{Corollaries}\sl}

140: \de\Cs{\tx{\sf C}}

141: \de\Cul{\tx{$\ul C$}}

142: \de\DD{{\tx{${\Bbb D}$}}}

143: \de\DDD{{\tx{$\scr D$}}}

144: \de\Dbar{\tx{$\ol{D}$}}

145: \de\Dbars{\tx{$\ol{D}_s$}}

146: \de\Def{\pr{Definition}\rm}

147: \de\Defn#1{\prn{Definition #1}\rm}

148: \de\Defs{\pr{Definitions}\rm}

149: \de\Defsn#1{\prn{Definitions #1}\rm}

150: \de\Del{\tx{$\Delta$}}

151: \de\Discussion{\pr{Discussion}\rm}

152: \de\Discussionn#1{\pr{Discussion #1}\rm}

153: \de\Ds{\tx{\sf D}}

154: \de\Dss{{\tx{\ssf D}}}

155: \de\Ebar{\tx{$\ol E$}}

156: \de\Eg{{\it E.g.}}

157: \de\ElemInd{\tx{\sf ElemInd}}

158: \de\ElemIndSig{\tx{$\ElemInd(\Sig)$}}

159: \de\EqSort{\tx{\bi EqSort\/}}

160: \de\EqSortSig{\EqSort(\Sig)}

161: \de\Eqb{\tx{\bi Eq}}

162: \de\Eqs{\tx{\sf Eq}}

163: \de\Equiv{\tx{$\pmb{Equiv}$}}

164: \de\Etil{\tx{$\til{E}$}}

165: \de\Example{\pr{Example}}

166: \de\Examplen#1{\prn{Example {#1}}}

167: \de\Examples{\pr{Examples}}

168: \de\Examplesn#1{\prn{Examples {#1}}}

169: \de\Exercise{\pr{Exercise}}

170: \de\Exercisen#1{\prn{Exercise {#1}}}

171: \de\Exercises{\pr{Exercises}}

172: \de\Exercisesn#1{\prn{Exercises {#1}}}

173: \de\F{\tx{\bi F}}

174: \de\FA{\tx{$F^A$}}

175: \de\FAuu{\tx{$F^{A,\uu}$}}

176: \de\FFF{\tx{$\scr F$}}

177: \de\FFt{\tx{\tt F}}

178: \de\FinFuncSig{\tx{$F \in \FuncSig$}}

179: \de\First{\tx{\bi First}}

180: \de\For{\tx{\bi For}}

181: \de\ForA{\tx{$\For(A)$}}

182: \de\ForN{\tx{$\For^N$}}

183: \de\ForNA{\tx{$\ForN(A)$}}

184: \de\ForNSig{\tx{$\ForN(\Sig)$}}

185: \de\ForSig{\tx{$\For(\Sig)$}}

186: \de\ForSigN{\tx{$\For(\SigN)$}}

187: \de\ForSigx{\tx{$\For(\Sigx)$}}

188: \de\Form{\tx{\bi Form}}

189: \de\FormSig{\tx{$\Form(\Sig)$}}

190: \de\Foro{\tx{$\For_0$}}

191: \de\Forolam{\tx{$\For_0^{\lam}$}}

192: \de\Forx{\tx{$\For^\starb$}}

193: \de\ForxA{\tx{$\Forx(A)$}}

194: \de\ForxSig{\tx{$\Forx(\Sig)$}}

195: \de\Fs{{\tx{\smallbi F}}}

196: \de\Ftil{\tx{$\til{F}$}}

197: \de\Func{\tx{\bi Func}}

198: \de\FuncSig{\Func\,(\Sig)}

199: \de\FuncSigp{\tx{$\Func(\Sigp)$}}

200: \de\FuncSigus{\tx{$\FuncSig_\utos$}}

201: \de\GG{\tx{${\Bbb G}$}}

202: \de\GGG{\tx{$\scr G$}}

203: \de\Gtt{\tx{\tt G}}

204: \de\Gam{\tx{$\Gamma$}}

205: \de\Gbar{\tx{$\ol G$}}

206: \de\Gs{{\tx{\sf G}}}

207: \de\Halt{\tx{\bi Halt}}

208: \de\HaltTests{\tx{\sf HaltTest}}

209: \de\HaltA#1{\tx{$\Halt^A\,(#1)$}}

210: \de\Hbar{\tx{$\bar H$}}

211: \de\Hh{\tx{\bi H}}

212: \de\Hs{{\tx{\sf H}}}

213: \de\Hss{{\tx{\ssf H}}}

214: \de\Htt{\tx{\tt H}}

215: \de\Ib{\tx{{\bf I}}}

216: \de\II{{\tx{${\Bbb I}$}}}

217: \de\III{\tx{${\scr I}$}}

218: \de\IIIN{{\tx{$\III^N$}}}

219: \de\IL{\tx{\bi IL}}

220: \de\IO{\tx{\bi IO}}

221: \de\Ibar{{\tx{$\bar I$}}}

222: \de\Ie{{\it I.e.}}

223: \de\Ind{\tx{\sf Ind}}

224: \de\IndSig{\tx{$\Ind(\Sig)$}}

225: \de\Indi{\tx{$\Ind_i$}}

226: \de\IndiSig{\Indi(\Sig)}

227: \de\Init{\tx{\bi Init}}

228: \de\Is{\tx{\sf I}}

229: \de\Iss{{\tx{\ssf I}}}

230: \de\JJJ{\tx{${\scr J}$}}

231: \de\Jbar{{\tx{$\bar J$}}}

232: \de\KK{{\tx{${\Bbb K}$}}}

233: \de\KKbar{\tx{$\ol{\KK}$}}

234: \de\KKN{\tx{${\Bbb K}^N$}}

235: \de\KKuu{\tx{${\Bbb K}^\uu$}}

236: \de\KKx{\tx{${\Bbb K}^*$}}

237: \de\LL{\tx{${\Bbb L}$}}

238: \de\LLL{\tx{${\scr L}$}}

239: \de\Lam{\tx{$\Lambda$}}

240: \de\Lang{\tx{\bi Lang}}

241: \de\LangSig{\tx{$\Lang(\Sig)$}}

242: \de\LangSigx{\tx{$\Lang(\Sigx)$}}

243: \de\Langx{\tx{$\Lang^*$}}

244: \de\LangxSig{\tx{$\Langx(\Sig)$}}

245: \de\Lemma{\pr{Lemma}\sl}

246: \de\Lemman#1{\prn{Lemma #1}\sl}

247: \de\Lemmas{\pr{Lemmas}\sl}

248: \de\Lemmasn#1{\prn{Lemmas #1}\sl}

249: \de\Length{\tx{\bi Length}}

250: \de\Lgths{\tx{\sf Lgth}}

251: \de\Lss{{\tx{\ssf L}}}

252: \de\MMM{\tx{${\scr M}$}}

253: \de\Min{\tx{\bi Min}}

254: \de\MinNStdAlg{\Min\NStdAlg}

255: \de\Mod{\tx{\bi Mod}}

256: \de\NN{{\tx{${\Bbb N}$}}}

257: \de\NNN{{\tx{${\scr N}$}}}

258: \de\NNNB{{\tx{${\NNN^B}$}}}

259: \de\NNNo{{\tx{$\NNN_0$}}}

260: \de\NNu{\tx{${\Bbb N}^\uu$}}

261: \de\NStdAlg{\tx{\bi NStdAlg\/}}

262: \de\NStdAlgSig{\tx{$\NStdAlg\,(\Sig)$}}

263: \de\Nbd{\tx{\bi Nbd}}

264: \de\Nbi{{\tx{\bi N}}}

265: \de\Nbismall{{\tx{\smallbi N}}}

266: \de\Newlengths{\tx{\sf Newlength}}

267: \de\Notation{\pr{Notation}\rm}

268: \de\Notationn#1{\pr{Notation \ #1}\rm}

269: \de\Note{\pr{Note}\rm}

270: \de\Noten#1{\pr{Note #1}\rm}

271: \de\Notes{\pr{Notes}\rm}

272: \de\Notesn#1{\pr{Notes #1}\rm}

273: \de\Ns{\tx{\sf N}}

274: \de\Nss{{\tx{\ssf N}}}

275: \de\Nnu{\tx{$\Nn^\uu$}}

276: \de\Nulls{\tx{\sf Null}}

277: \de\Nullvec{\tx{$\vec{\Nulls}$}}

278: \de\Om{{\tx{$\Omega$}}}

279: \de\Pp{\tx{\bk P}}

280: \de\PA{\tx{$P^A$}}

281: \de\PE{\tx{\bi PE}}

282: \de\PL#1{\tx{$\ProgLang_{#1}$}}

283: \de\PPP{\tx{$\scr P$}}

284: \de\PR{\tx{PR}}

285: \de\PRA{\tx{$\PR(A)$}}

286: \de\PRAutos{\tx{$\PRA_\utos$}}

287: \de\PRSig{\tx{$\PR(\Sig)$}}

288: \de\PRSigb{\tx{$\PRb(\Sigb)$}}

289: \de\PRSigutos{\tx{$\PRSig_\utos$}}

290: \de\PRb{\tx{\bf PR}}

291: \de\PRmin{\tx{$\PR_-$}}

292: \de\PRo{\tx{$\PR_0$}}

293: \de\PRolam{\tx{$\PR_0^{\lam}$}}

294: \de\PRutos{\tx{$\PR_\utos$}}

295: \de\PRx{\PR\str}

296: \de\PRxA{\tx{$\PRx(A)$}}

297: \de\PRxSig{\tx{\PRx(\Sig)}}

298: \de\PRxSigb{\tx{\PRxb(\Sigb)}}

299: \de\PRxb{\PRb\strb}

300: \de\PTE{\tx{\bi PTE}}

301: \de\PTEA{\tx{$\PTE^A$}}

302: \de\PTerm{\tx{\bi PTerm}}

303: \de\PTermSig{\tx{$\PTerm(\Sig)$}}

304: \de\Pbi{\tx{\bi P}}

305: \de\Pf{\n{\bf Proof:\ \,}}

306: \de\Ph{\tx{$\Phi$}}

307: \de\PreStr{\tx{\bi PreStr}}

308: \de\PreStrSig{\PreStr\,(\Sig)}

309: \de\Problem{\pr{Problem}}

310: \de\Problemn#1{\pr{Problem {#1}}}

311: \de\Problems{\pr{Problems}}

312: \de\Problemsn#1{\pr{Problems {#1}}}

313: \de\Propsn#1{\pr{Propositions {#1}}}

314: \de\Proc{\tx{\bi Proc}}

315: \de\ProcN{\tx{$\Proc^N$}}

316: \de\ProcNSig{\ProcN(\Sig)}

317: \de\ProcSig{\Proc(\Sig)}

318: \de\ProcSigN{\Proc(\SigN)}

319: \de\ProcSiguv{\tx{$\Proc(\Sig)_\utov$}}

320: \de\ProcSigx{\tx{$\Proc(\Sigx)$}}

321: \de\Procuv{\tx{$\Proc_\utov$}}

322: \de\ProcuvSig{\tx{$\Procuv(\Sig)$}}

323: \de\Procx{\Proc\str}

324: \de\ProcxSig{\Procx(\Sig)}

325: \de\ProcxSiguv{\tx{$\ProcxSig_\utov$}}

326: \de\Procxuv{\tx{$\Proc^*_\utov$}}

327: \de\Prod{\tx{\bi Prod}}

328: \de\ProdType{\tx{\bi ProdType}}

329: \de\ProdTypeSig{\tx{$\ProdType(\Sig)$}}

330: \de\Prog{\tx{\bi Prog}}

331: \de\ProgLang{\Prog\Lang}

332: \de\ProgTerm{\Prog\Term}

333: \de\ProgTermSig{\tx{$\ProgTerm(\Sig)$}}

334: \de\Prop{\pr{Proposition}\sl}

335: \de\Propn#1{\prn{Proposition #1}\sl}

336: \de\Propp{\tx{\bi Prop}}

337: \de\Props{\pr{Propositions}\sl}

338: \de\Ps{\tx{$\Psi$}}

339: \de\Q{{\bf Q}}

340: \de\QQ{\tx{${\Bbb Q}$}}

341: \de\Qs{\tx{\sf Q}}

342: \de\Question{\pr{Question}\sl}

343: \de\RR{{\tx{${\Bbb R}$}}}

344: \de\RRR{{\tx{$\scr R$}}}

345: \de\RRRB{\tx{$\RRR^B$}}

346: \de\RRRL{\tx{${\RRR}^<$}}

347: \de\RRRN{\tx{$\RRR^N$}}

348: \de\RRn{\tx{$\RR^n$}}

349: \de\RRx{\tx{$\RR\str$}}

350: \de\RRq{\tx{$\RR^q$}}

351: \de\Ref{\tx{\bf Ref:\/}}

352: \de\Rem{\tx{\bi Rem}}

353: \de\RemA{\tx{$\Rem^A$}}

354: \de\RemSeqA#1{\tx{$\Rem\Seq^A(#1)$}}

355: \de\RemSet#1{\tx{\bi RemSet\,$(#1)$}}

356: \de\Remark{\pr{Remark}\rm}

357: \de\Remarkn#1{\prn{Remark #1}\rm}

358: \de\Remarks{\pr{Remarks}\rm}

359: \de\Remarksn#1{\prn{Remarks #1}\rm}

360: \de\Rep{\tx{\bi Rep}}

361: \de\Rest{\tx{\bi Rest}}

362: \de\Rs{\tx{\sf R}}

363: \de\Rss{{\tx{\ssf R}}}

364: \de\SA{\tx{$\bb{S}^A$}}

365: \de\SS{{\tx{${\Bbb S}$}}}

366: \de\SSS{{\tx{${\scr S}$}}}

367: \de\Sarrow{\tx{$S^{\to}$}}

368: \de\Sat{\tx{$S_{\tx{\ssf at}}$}}

369: \de\Sbar{\tx{$\ol \Sbi$}}

370: \de\Sbi{\tx{\bi S}}

371: \de\Seq{\tx{\bi Seq}}

372: \de\Shead#1#2{\bn{\bigbf {#1}\ \ \ {#2}}\sn}

373: \de\Sheads#1#2{\bn{\bigbf {#1}\ \ \ {#2}}}

374: \de\Sig{{\tx{$\varSigma$}}}

375: \de\SigA{\tx{$\Sig(A)$}}

376: \de\SigBBB{\tx{$\Sig^\BBB$}}

377: \de\SigBN{\tx{$\Sig^{B,N}$}}

378: \de\SigN{\tx{$\Sig^N$}}

379: \de\SigT{\tx{$(\Sig,T)$}}

380: \de\Sigb{\tx{$\bs\varSigma$}}

381: \de\Sigbar{\tx{$\ol \Sig$}}

382: \de\Sigo{\tx{$\varSigma_0$}}

383: \de\Sigox{\tx{$\varSigma_0^*$}}

384: \de\Sigone{\tx{$\Sigb_1$}}

385: \de\SigoneInd{\tx{$\Sigb_{\pmb1}$-$\pmb{Ind}\,$}}

386: \de\Sigonex{\tx{$\Sigb_1^*$}}

387: \de\SigonexInd{\tx{$\Sigb_{\pmb1}^{\pmb*}$-$\pmb{Ind}\,$}}

388: \de\SigonexIndl{\tx{$\Sigb_{\pmb1}^{\pmb*}$-$\pmb{Ind_\ell}\,$}}

389: \de\SigonexIndc{\tx{$\Sigb_{\pmb1}^{\pmb*}$-$\pmb{Ind_c}\,$}}

390: \de\SigonexIndi{\tx{$\Sigb_{\pmb1}^{\pmb*}$-$\pmb{Ind_i}\,$}}

391: \de\Sigp{{\tx{$\Sig'$}}}

392: \de\SigpTp{\tx{$(\Sig',T')$}}

393: \de\Sigpp{{\tx{$\Sig''$}}}

394: \de\SigppTpp{\tx{$(\Sig'',T'')$}}

395: \de\Sigstk{\tx{$\Sig^\stkss$}}

396: \de\Siguu{\tx{$\varSigma^\uu$}}

397: \de\SiguuN{\tx{$\varSigma^{\uu,N}$}}

398: \de\Sigx{\tx{$\varSigma^*$}}

399: \de\Snap{\tx{\bi Snap}}

400: \de\SnapA{\tx{$\Snap^A$}}

401: \de\SnapSeqA#1{\tx{$\Snap\Seq^A(#1)$}}

402: \de\Sort{\tx{\bi Sort\/}}

403: \de\SortSig{\Sort(\Sig)}

404: \de\SortSigsmall{\Sortsmall\,(\Sig)}

405: \de\SortSigp{\tx{$\Sort(\Sigp)$}}

406: \de\SortSigx{\tx{$\Sort(\Sigx)$}}

407: \de\Sortsmall{\tx{\smallbi Sort}}

408: \de\Spec{\tx{\sf Spec}}

409: \de\Ss{\tx{\sf S}}

410: \de\Sss{{\tx{\ssf S}}}

411: \de\State{\tx{\bi State}}

412: \de\StateA{\tx{$\State(A)$}}

413: \de\Step{\tx{\bi Step}}

414: \de\Stmt{\tx{\bi Stmt}}

415: \de\StmtSig{\tx{$\Stmt(\Sig)$}}

416: \de\StmtStepA{\tx{$\Stmt\Step^A$}}

417: \de\Std{\tx{\bi Std\/}}

418: \de\StdAlg{\Std\Alg}

419: \de\StdAlgSig{\tx{$\StdAlg\,(\Sig)$}}

420: \de\StdAlgSigx{\tx{$\StdAlg\,(\Sigx)$}}

421: \de\StdMod{\Std\Mod}

422: \de\Str{\tx{\bi Str}}

423: \de\StrSig{\Str\,(\Sig)}

424: \de\SubalgA#1{\tx{${\bi \ Subalg}_A(#1)$}}

425: \de\Tb{\tx{\bf T}}

426: \de\Tbi{\tx{\bi T}}

427: \de\TComp{\tx{\bi TComp}}

428: \de\TE{\tx{\bi TE}}

429: \de\TEA{\tx{$\TE^A$}}

430: \de\TOL{\TagsOnLeft}

431: \de\TOR{\TagsOnRight}

432: \de\TSig{\Tbi(\Sig)}

433: \de\TT{\tx{${\Bbb T}$}}

434: \de\TTT{\tx{$\scr T$}}

435: \de\Tab{\tx{\bi Tab}}

436: \de\TabSig{\Tab(\Sig)}

437: \de\Tbar{\tx{$\ol T$}}

438: \de\Term{\tx{\bi Term}}

439: \de\TermSig{\tx{$\Term(\Sig)$}}

440: \de\TermTup{\Term\Tup}

441: \de\TermTupSig{\tx{$\TermTup(\Sig)$}}

442: \de\Terminology{\pr{Terminology}\rm}

443: \de\Th{\tx{\bi Th}}

444: \de\Thesisn#1{\pr{Thesis \ #1}\sl}

445: \de\Thm{\pr{Theorem}\sl}

446: \de\Thmn#1{\prn{Theorem #1}\sl}

447: \de\Thms{\pr{Theorems}\sl}

448: \de\Tm{\tx{\bi Tm}}

449: \de\TmSig{\tx{$\Tm(\Sig)$}}

450: \de\Tms{\tx{$\Tm_s$}}

451: \de\TmsSig{\tx{$\Tms(\Sig)$}}

452: \de\Ts{\tx{$\Tbi_s$}}

453: \de\TsSig{\tx{$\Ts(\Sig)$}}

454: \de\Tup{\tx{\bi Tup}}

455: \de\Ub{\tx{\bf U}}

456: \de\Ubi{\tx{\bi U}}

457: \de\Univ{\tx{\bi Univ}}

458: \de\Univs{\tx{\sf Univ}}

459: \de\Univuv{\tx{$\Univs_\utov$}}

460: \de\UnivuvA{\tx{$\Univs_\utov^A$}}

461: \de\Univuvx{\tx{$\Univs_\utov^*$}}

462: \de\UnivuvxA{\tx{$\Univs_\utov^{*,A}$}}

463: \de\Univus{\tx{$\Univ_\utos$}}

464: \de\UnivusA{\tx{$\Univ_\utos^A$}}

465: \de\Unspecs{\tx{\sf Unspec}}

466: \de\Updates{\tx{\sf Update}}

467: \de\UpdateD{\tx{$\Updates^D$}}

468: \de\Us{\tx{\sf U}}

469: \de\VVV{\tx{$\scr V$}}

470: \de\Val{\tx{\bi Val}}

471: \de\Var{\tx{\bi Var}}

472: \de\VarSig{\tx{$\Var(\Sig)$}}

473: \de\VarTup{\Var\Tup}

474: \de\VarTupSig{\tx{$\VarTup(\Sig)$}}

475: \de\Vars{\tx{$\Var_s$}}

476: \de\VarsSig{\tx{$\Vars(\Sig)$}}

477: \de\Vbb{\tx{\bf V}}

478: \de\WWW{\tx{${\scr W}$}}

479: \de\Wb{\tx{\bf W}}

480: \de\While{\tx{\bi While}}

481: \de\WhileA{\tx{$\While(A)$}}

482: \de\WhileN{\tx{$\While^\Nbismall$}}

483: \de\WhileNA{\tx{$\WhileN(A)$}}

484: \de\WhileNSig{\tx{$\WhileN(\Sig)$}}

485: \de\WhileSig{\tx{$\While(\Sig)$}}

486: \de\WhileSigN{\tx{$\While(\SigN)$}}

487: \de\WhileSigx{\tx{$\While(\Sigx)$}}

488: \de\Whilebig{\tx{\bigbi While}}

489: \de\WhilebigSig{\tx{$\Whilebig(\Sigb)$}}

490: \de\WhilebigN{\tx{$\Whilebig^\Nbi$}}

491: \de\Whilebigx{\tx{$\Whilebig^\starb$}}

492: \de\Whilex{\tx{$\While^\starb$}}

493: \de\WhilexA{\Whilex(A)}

494: \de\WhilexSig{\Whilex(\Sig)}

495: \de\WhilexSigN{\Whilex(\SigN)}

496: \de\Whilexbig{\tx{$\Whilebig^{\dz{\starb}}$}}

497: \de\XXX{{\tx{${\scr X}$}}}

498: \de\ZZ{\tx{${\Bbb Z}$}}

499: \de\ZZZ{{\tx{${\scr Z}$}}}

500: \de\Zbi{\tx{\bi Z}}

501: \de\Zs{\tx{\sf Z}}

502: \de\Zss{{\tx{\ssf Z}}}

503: \de\Zz{\tx{\sf Z}}

504: \de\abi{\tx{\bi a}}

505: \de\aas{\tx{\sf a}}

506: \de\abar{\tx{$\bar a$}}

507: \de\adt{{\bi adt}}

508: \de\aee{\tx{\bi ae}}

509: \de\ainAu{\tx{$a \in \Au$}}

510: \de\ainAv{\tx{$a \in \Av$}}

511: \de\ainAs{\tx{$a \in \As$}}

512: \de\al{{\tx{$\alpha$}}}

513: \de\alA{\tx{$\alpha^A$}}

514: \de\alb{\tx{$\bs\alpha$}}

515: \de\albar{{\tx{$\ol \alpha$}}}

516: \de\algebras{\tx{\sf algebra}}

517: \de\alhat{{\tx{$\hat{\alpha}$}}}

518: \de\aliass{\tx{\sf alias}}

519: \de\aliasss{{\tx{\ssf alias}}}

520: \de\all{\forall}

521: \de\alvec{\tx{$\vec{\alpha}$}}

522: \de\ands{\tx{\sf and}}

523: \de\ang#1{\tx{$\langle #1 \rangle$}}

524: \de\angg#1{\tx{$\langg #1 \rangg$}}

525: \de\arity#1{\tx{\rm arity$(#1)$}}

526: \de\arr{\arrow <0.1in> [0.2,0.5]}

527: \de\att{{\tx{\tt a}}}

528: \de\atts{{\tx{\smalltt a}}}

529: \de\attx{\att\str}

530: \de\auxs{\tx{\sf aux}}

531: \de\avec{\tx{$\vec{a}$}}

532: \de\ax{\tx{$a^*$}}

533: \de\bA{\tx{$\bb{b}^A$}}

534: \de\bB{\bk{B}}

535: \de\bFor{\tx{\bigbi For}}

536: \de\bForN{\tx{$\bFor^\Nbi$}}

537: \de\bForx{\tx{$\bFor^\starb$}}

538: \de\bProc{\tx{\bigbi Proc}}

539: \de\bb#1{\tx{$\lbb #1 \rbb$}}

540: \de\bbar{[\ns]}

541: \de\bbi{\tx{\bi b}}

542: \de\bbs{\tx{\sf b}}

543: \de\bdot{{\bs\cdot}}

544: \de\be{{\tx{$\beta$}}}

545: \de\beA{\tx{$\beta^A$}}

546: \de\beb{\tx{$\bs\beta$}}

547: \de\begins{\tx{\sf begin}}

548: \de\bigcaplim#1{\underset#1\to\bigcap}

549: \de\bigcon{\bigwedge}

550: \de\bigdis{\bigvee}

551: \de\bigskipn{\bigskip\nin}

552: \de\biu{\tx{\bi u}}

553: \de\bk{\boldkey}

554: \de\bn{\bigskip\nin}

555: \de\bools{\tx{\sf bool}}

556: \de\boolss{{\tx{\ssf bool}}}

557: \de\br{\ |\ }

558: \de\bs{\boldsymbol}

559: \de\btt{{\tx{\tt b}}}

560: \de\btts{{\tx{\smalltt b}}}

561: \de\bttx{\btt\str}

562: \de\bu{$\bullet$}

563: \de\bul{\item{\bu}}

564: \de\bull{\itemm{\bu}}

565: \de\bulll{\itemmm{\bu}}

566: \de\bx{\tx{$b^*$}}

567: \de\capt#1#2{\ce{{\sc #1}. \ \ #2}}

568: \de\card#1{\tx{\bi card\,$(#1)$}}

569: \de\carrierss{\tx{\sf carriers}}

570: \de\cart#1{\tx{\bi cart\, $(#1)$}}

571: \de\cb#1{\ce{\bf #1}}

572: \de\cbb#1{\ce{\bbf #1}}

573: \de\cbbb#1{\ce{\bbigbf #1}}

574: \de\cbi#1{\ce{\bi #1}}

575: \de\cbbi#1{\ce{\bigbi #1}}

576: \de\ccc{{\tx{\bi c}}}

577: \de\ce{\centerline}

578: \de\cf{{\it cf\.}}

579: \de\chead#1{\bn\cb{#1}\mn}

580: \de\chooses{\tx{\sf choose}}

581: \de\cnr#1{\tx{$\ulc #1 \urc$}}

582: \de\comlabs{\tx{\com\bf-\labs}}

583: \de\comlam{\tx{$\com^\lam$}}

584: \de\compb{\tx{\bi comp}}

585: \de\compl#1{\tx{\bi compl\,$(#1)$}}

586: \de\complexs{\tx{\sf complex}}

587: \de\complexss{{\tx{\ssf complex}}}

588: \de\comps{\tx{\sf comp}}

589: \de\con{\land}

590: \de\concon{{\bigcon\nss\nss\nss\bigcon}}

591: \de\conconlim#1{\underset#1\to\concon}

592: \de\conconlims#1#2{\overset{#2}\to{\conconlim{#1}}}

593: \de\cons{\tx{\sf con}}

594: \de\conss{{\tx{\ssf con}}}

595: \de\constantss{\tx{\sf constants}}

596: \de\consts{\tx{\sf const}}

597: \de\constss{{\tx{\ssf const}}}

598: \de\cpxs{\tx{\sf complex}}

599: \de\ccs{\tx{\sf c}}

600: \de\css{{\tx{\ssf c}}}

601: \de\ctt{{\tx{\tt c}}}

602: \de\ctts{{\tx{\smalltt c}}}

603: \de\cttx{\ctt\str}

604: \de\curl#1{\tx{$\{#1\}$}}

605: \de\curly#1{\tx{$\{\,#1\,\}$}}

606: \de\cvals{\tx{\sf cval}}

607: \de\cvec{\tx{$\vec{c}$}}

608: \de\cx{\tx{$c^*$}}

609: \de\da{\downarrow}

610: \de\dash{\item{---}}

611: \de\dashh{\itemm{---}}

612: \de\dashhh{\itemmm{---}}

613: \de\datas{{\tx{\sf data}}}

614: \de\datass{{\tx{\ssf data}}}

615: \de\dcs{\tx{\sf dc}}

616: \de\ddis{\lor\!\!\!\!\!\lor}

617: \de\defaults{\tx{\sf default}}

618: \de\degg#1{\tx{\bi deg$(#1)$}}

619: \de\del{{\tx{$\delta$}}}

620: \de\delb{\tx{$\bs\delta$}}

621: \de\delhat{\hat{\delta}}

622: \de\depth#1{\tx{\bi depth$(#1)$}}

623: \de\dis{\lor}

624: \de\disdis{{\bigdis\nss\nss\nss\bigdis}}

625: \de\disdislim#1{\underset#1\to\disdis}

626: \de\disdislims#1#2{\overset{#2}\to{\disdislim{#1}}}

627: \de\displayquote{\narrower\narrower\smallrm\nin}

628: \de\displaytext{\narrower\narrower\sl\nin}

629: \de\diss{\tx{\sf dis}}

630: \de\divN{\tx{$\divs_{\natss}$}}

631: \de\divNR{\tx{$\divs_\natss^\RRR$}}

632: \de\divs{\tx{\sf div}}

633: \de\dos{\tx{\sf do}}

634: \de\dom#1{\tx{\bi dom$(#1)$}}

635: \de\down{\tx{$^\vee$}}

636: \de\ds{{\tx{\sf d}}}

637: \de\dt{\tx{\bi dt}}

638: \de\dtt{{\tx{\tt d}}}

639: \de\dz{\dsize}

640: \de\ebar{{\tx{$\bar e$}}}

641: \de\eg{{\it e.g.}}

642: \de\ehat{tx{$\hat e$}}

643: \de\el{\tx{$\ell$}}

644: \de\elses{\tx{\sf else}}

645: \de\endpf{\qed\smskip}

646: \de\endpr{\rm\smskip}

647: \de\ends{\tx{\sf end}}

648: \de\enum{\tx{\bi enum}}

649: \de\eps{{\tx{$\epsilon$}}}

650: \de\epsb{{\tx{$\bs\epsilon$}}}

651: \de\eqbool{\tx{$\eqs_{\boolss}$}}

652: \de\eqdata{\tx{$\eqs_\datass$}}

653: \de\eqdf{=_{df}}

654: \de\eqint{\tx{$\eqs_{\intss}$}}

655: \de\eqintvl{\tx{$\eqs_{\intvlss}$}}

656: \de\eqnat{\tx{$\eqs_{\natss}$}}

657: \de\eqnatN{\tx{$\eqs_{\natss}^\NNN$}}

658: \de\eqreal{\tx{$\eqs_{\realss}$}}

659: \de\eqrealR{\tx{$\eqs_{\realss}^\RRR$}}

660: \de\eqs{\tx{\sf eq}}

661: \de\eqss{\tx{$\eqs_s$}}

662: \de\equal{\ &= \ }

663: \de\es{\tx{\sf e}}

664: \de\etal{{\it et al.}}

665: \de\ett{{\tx{\tt e}}}

666: \de\eval{\tx{\bi eval}}

667: \de\evals{\tx{\sf eval}}

668: \de\evec{\tx{$\vec{e}$}}

669: \de\ex{\exists}

670: \de\expands{\tx{\sf expand}}

671: \de\fA{\tx{$f^A$}}

672: \de\falses{\tx{\sf false}}

673: \de\fbi{\tx{\bi f}}

674: \de\fff{\tx{{\sf f}\!{\sf f}}}

675: \de\ffrom{\tx{$\Leftarrow$}}

676: \de\fcheck{\tx{$\check f$}}

677: \de\fhat{\tx{$\hat f$}}

678: \de\first{\tx{\bi first}}

679: \de\fis{\tx{\sf fi}}

680: \de\floor#1{\tx{$\llc #1 \lrc$}}

681: \de\fn{\footnote}

682: \de\fors{\tx{\sf for}}

683: \de\free{\tx{\bi free}}

684: \de\from{\leftarrow}

685: \de\fs{{\tx{\sf f}}}

686: \de\fss{{\tx{\ssf f}}}

687: \de\ftil{\tx{$\til f$}}

688: \de\funcs{\tx{\sf func}}

689: \de\functionss{\tx{\sf functions}}

690: \de\fuu{\tx{$f^\uu$}}

691: \de\fvec{\tx{$\vec{f}$}}

692: \de\gam{{\tx{$\gamma$}}}

693: \de\gamA{\tx{$\gam^A$}}

694: \de\gamb{\tx{$\bs\gamma$}}

695: \de\gamx{\tx{$\gam^*$}}

696: \de\ghat{\tx{$\hat g$}}

697: \de\gtil{\tx{$\til g$}}

698: \de\gn{\tx{\bi gn}}

699: \de\gotos{\tx{\sf goto}}

700: \de\graph{\tx{\bi graph}}

701: \de\gs{{\tx{\sf g}}}

702: \de\gtt{{\tx{\tt g}}}

703: \de\gvec{\tx{$\vec{g}$}}

704: \de\halt{\tx{\bi halt}}

705: \de\hs{{\tx{\sf h}}}

706: \de\htil{\tx{$\til h$}}

707: \de\htt{{\tx{\tt h}}}

708: \de\ibar{{\tx{$\bar{\imath}$}}}

709: \de\id{\tx{\bi id}}

710: \de\ident{\equiv}

711: \de\idi{\tx{$\id_i$}}

712: \de\ie{{\it i.e.}}

713: \de\ifbool{\tx{$\ifs_{\boolss}$}}

714: \de\ifdata{\tx{$\ifs_{\datass}$}}

715: \de\ifff{\ \,\llongtofrom\ }

716: \de\ifffdf{\ \ \tx{$\llongtofrom_{df}$}\ }

717: \de\ifint{\tx{$\ifs_{\intss}$}}

718: \de\ifintvl{\tx{$\ifs_{\intvlss}$}}

719: \de\ifnat{\tx{$\ifs_{\natss}$}}

720: \de\ifnatN{\tx{$\ifs_{\natss}^\NNN$}}

721: \de\ifreal{\tx{$\ifs_{\realss}$}}

722: \de\ifrealR{\tx{$\ifs_\realss^\RRR$}}

723: \de\ifs{\tx{\sf if}}

724: \de\ifss{\tx{$\ifs_s$}}

725: \de\ifstk{\tx{$\ifs_{\stkss}$}}

726: \de\ift#1{&\tx{if \ {#1}}}

727: \de\ii{{\tx{\bi i}}}

728: \de\iii{{\tx{\sf i}}}

729: \de\iinI{{\tx{$i \in I$}}}

730: \de\ims{\tx{\sf im}}

731: \de\imp{\to}

732: \de\impp{\ \,\llongto\ }

733: \de\imps{\tx{\sf imp}}

734: \de\imports{\tx{\sf import}}

735: \de\inbull{\itemitem{\bu}}

736: \de\indash{\itemitem{---}}

737: \de\indentt{\indent\quad}

738: \de\ins{\tx{\sf in}}

739: \de\ints{{\tx{\sf int}}}

740: \de\intss{{\tx{\ssf int}}}

741: \de\intvls{{\tx{\sf intvl}}}

742: \de\intvlss{{\tx{\ssf intvl}}}

743: \de\io{{\tx{$\iota$}}}

744: \de\is{\tx{\sf i}}

745: \de\isom{\cong}

746: \de\itt{{\tx{\tt i}}}

747: %\de\itemm{\itemitem}

748: \de\ivec{\tx{$\vec{\imath}$}}

749: \de\ivecstrut{\vec{\mathstrut i}}

750: \de\jbar{{\tx{$\bar{\jmath}$}}}

751: \de\js{\tx{\sf j}}

752: \de\jvec{\tx{$\vec{\jmath}$}}

753: \de\jvecstrut{\vec{\mathstrut j}}

754: \de\kap{{\tx{$\kappa$}}}

755: \de\kapb{\tx{$\bs\kappa$}}

756: \de\kbar{\tx{$\bar k$}}

757: \de\kinNN{\tx{$k\in\NN$}}

758: \de\ktt{{\tx{\tt k}}}

759: \de\kvec{\tx{$\vec{k}$}}

760: \de\kvecstrut{\vec{\mathstrut k}}

761: \de\kxvec{\tx{$\kvec^*$}}

762: \de\labs{\lamb\tx{\bi abs}}

763: \de\lam{{\tx{$\lambda$}}}

764: \de\lamPR{\lam\PR}

765: \de\lamPRmin{\lam\PRmin}

766: \de\lamPRx{\lam\PRx}

767: \de\lamb{\tx{$\bs\lambda$}}

768: \de\lammuPR{\lam\muu\PR}

769: \de\lammuPRmin{\lam\muu\PRmin}

770: \de\lammuPRx{\lam\muu\PRx}

771: \de\langg{\langle\!|}

772: \de\lb{\linebreak}

773: \de\lbar{\tx{$\ol l$}}

774: \de\lbb{[\![}

775: \de\leasts{\tx{\sf least}}

776: \de\lele#1#2#3{{\tx{$#1 \le #2 \le #3$}}}

777: \de\lel#1#2#3{{\tx{$#1 \le #2 < #3$}}}

778: \de\lgth#1{\tx{\bi lgth$(#1)$}}

779: \de\llc{\llcorner}

780: \de\lle#1#2#3{{\tx{$#1 < #2 \le #3$}}}

781: \de\llongfrom{\tx{$\Longleftarrow$}}

782: \de\llongto{\tx{$\Longrightarrow$}}

783: \de\llongtofrom{\tx{$\Longleftrightarrow$}}

784: \de\loccit{{\it loc\. cit.}}

785: \de\locs{\tx{\sf loc}}

786: \de\longfrom{\longleftarrow}

787: \de\longimp{\longrightarrow}

788: \de\longto{\longrightarrow}

789: \de\longtofrom{\longleftrightarrow}

790: \de\loops{\tx{\sf loop}}

791: \de\lowerbox#1#2{{\lower#1pt\hbox{$#2$}}}

792: \de\lrc{\lrcorner}

793: \de\lss{\tx{\sf less}}

794: \de\lsint{\tx{$\lss_\intss$}}

795: \de\lsintvl{\tx{$\lss_\intvlss$}}

796: \de\lsnat{\tx{$\lss_\natss$}}

797: \de\lsnatN{\tx{$\lss_\natss^\NNN$}}

798: \de\lsreal{\tx{$\lss_\realss$}}

799: \de\lsrealR{\tx{$\lss_\realss^\RRR$}}

800: \de\lvec{\tx{$\vec{l}$}}

801: \de\lxvec{\tx{$\vec{l}^*$}}

802: \de\lxx{\lower.2ex}

803: \de\lxxx{\lower.3ex}

804: \de\lxxxx{\lower.4ex}

805: \de\macros{\tx{\sf macro}}

806: \de\macross{{\tx{\ssf macro}}}

807: \de\magn#1{\magnification=\magstep#1}

808: \de\mapdownl#1{\mapdown\lft{\dz{#1}}}

809: \de\mapdownlr#1#2{\mapdown\lft{\dz{#1}}\rt{\dz{#2}}}

810: \de\mapdownr#1{\mapdown\rt{\dz{#1}}}

811: \de\maprightd#1{\mapright_{\dz{#1}}}

812: \de\maprightdlower#1#2{\mapright_{\lower#1pt\hbox{$\dz{#2}$}}}

813: \de\maprightu#1{\mapright^{\dz{#1}}}

814: \de\maprightud#1#2{\mapright^{\dz{#1}}_{\dz{#2}}}

815: \de\maprighturaise#1#2{\mapright^{\raise#1pt\hbox{$\dz{#2}$}}}

816: \de\maprighturaised#1#2#3{\mapright^{\raise#1pt\hbox{$\dz{#2}$}}_{\dz{#3}}}

817: \de\maprightudlower#1#2#3{\mapright^{\dz{#1}}_{\lower#2pt\hbox{$\dz{#3}$}}}

818: \de\mapupl#1{\mapup\lft{\dz{#1}}}

819: \de\mapuplr#1#2{\mapup\lft{\dz{#1}}\rt{\dz{#2}}}

820: \de\mapupr#1{\mapup\rt{\dz{#1}}}

821: %\de\mapupud#1#2{\mapup^{\dz{#1}}_{\dz{#2}}}

822: \de\mbar{\tx{$\bar m$}}

823: \de\medskipn{\medskip\nin}

824: \de\mg#1{\tx{\bi mg$(#1)$}}

825: \de\mi{\tx{$^-$}}

826: \de\mins{\tx{\sf min}}

827: \de\mn{\medskip\nin}

828: \de\mtt{{\tx{\tt m}}}

829: \de\muCR{\mub\CR}

830: \de\muCRA{\tx{$\muCR(A)$}}

831: \de\muCRSig{\tx{$\muCR(\Sig)$}}

832: \de\muPR{\muu\PR}

833: \de\muPRA{\tx{$\muPR(A)$}}

834: \de\muPRSig{\tx{$\muPR(\Sig)$}}

835: \de\muPRSigb{\tx{$\muPRb(\Sigb)$}}

836: \de\muPRb{\mub\PRb}

837: \de\muPRmin{\muu\PRmin}

838: \de\muPRo{\muu\PRo}

839: \de\muPRolam{\muu\PRolam}

840: \de\muPRx{\muPR\str}

841: \de\muPRxb{\mub\PRxb}

842: \de\muPRxA{\tx{$\muPRx(A)$}}

843: \de\muPRxSig{\tx{$\muPRx(\Sig)$}}

844: \de\muPRxSigutos{\tx{$\muPRxSig_\utos$}}

845: \de\muPRxSigb{\tx{$\muPRxb(\Sigb)$}}

846: \de\muPRxutos{\tx{$\muPRx_\utos$}}

847: \de\mub{\tx{$\bs\mu$}}

848: \de\muu{{\tx{$\mu$}}}

849: \de\mvec{\tx{$\vec{m}$}}

850: \de\mxvec{\tx{$\mvec^*$}}

851: \de\n{\noindent}

852: \de\names{\tx{\sf name}}

853: \de\namess{{\tx{\ssf name}}}

854: \de\nats{\tx{\sf nat}}

855: \de\natss{{\tx{\ssf nat}}}

856: \de\nbar{\tx{$\bar n$}}

857: \de\negs{\tx{\sf neg}}

858: \de\news{\tx{\sf new}}

859: \de\newss{{\tx{\ssf new}}}

860: \de\nil{\tx{$\emptyset$}}

861: \de\nin{\noindent}

862: \de\ninNN{\tx{$n\in\NN$}}

863: \de\nl{\newline}

864: \de\nneg{\neg\!\!\!\!\!\neg}

865: \de\notover{\tx{\bi notover}}

866: \de\notimplies{\tx{$\ \not\nsss\llongto$}}

867: \de\notimplied{\tx{$\ \not\nsss\llongfrom$}}

868: \de\nots{\tx{\sf not}}

869: \de\npn{\NoPageNumbers}

870: \de\ns{\negthinspace}

871: \de\nss{\negthickspace}

872: \de\nsss{\nss\nss}

873: \de\nssss{\nsss\nsss}

874: \de\ntt{{\tx{\tt n}}}

875: \de\nub{\tx{$\bs\nu$}}

876: \de\nx{\tx{$n^*$}}

877: \de\ol{\overline}

878: \de\opcit{{\it op\. cit\.}}

879: \de\ods{\tx{\sf od}}

880: \de\om{{\tx{$\omega$}}}

881: \de\omb{{\tx{$\bs\omega$}}}

882: \de\onto{\twoheadrightarrow}

883: \de\ors{\tx{\sf or}}

884: \de\outs{\tx{\sf out}}

885: \de\ow{&\tx{otherwise}}

886: \de\pa{{\it p.a.}}

887: \de\pairs{\tx{\sf pair}}

888: \de\pe{\tx{\bi pe}}

889: \de\ph{{\tx{$\varphi$}}}

890: \de\pie{\tx{$\pi$}}

891: \de\pmapsto{{\,\overset\ {\cdot} \to\mapsto\,}}

892: \de\pr#1{\mn{\bf #1}. \,}

893: \de\prb#1{\bn{\bbf #1}.\sn}

894: \de\prc#1{\sn{\sc #1}.\ \ }

895: \de\prn#1{\mn{\bf #1}.\,}

896: \de\prims{\tx{\sf prim}}

897: \de\procs{\tx{\sf proc}}

898: \de\procss{{\tx{\ssf proc}}}

899: \de\proj{\tx{\bi proj}}

900: \de\projs{\tx{\sf proj}}

901: \de\ps{{\tx{$\psi$}}}

902: \de\pte{\tx{\bi pte}}

903: \de\pto{{\,\overset\ {\cdot} \to\longto\,}}

904: \de\ptt{{\tx{\tt p}}}

905: \de\pttx{\ptt\str}

906: \de\px{\tx{$p^*$}}

907: \de\qqquad{\qquad \qquad}

908: \de\qqqquad{\qqquad \qqquad}

909: \de\qfors{`\,\fors'}

910: \de\qforxs{`$\,\fors^*$'}

911: \de\qtt{{\tx{\tt q}}}

912: \de\qwhiles{`\,\whiles'}

913: \de\qwhilex{`$\,\whiles^*$'}

914: \de\rR{\tx{$\bk R$}}

915: \de\raisebox#1#2{{\raise#1pt\hbox{$#2$}}}

916: \de\ran#1{\tx{\bi ran$(#1)$}}

917: \de\rand{\ \tx{\rm and}\ }

918: \de\rangg{|\!\rangle}

919: \de\rats{\tx{\sf rat}}

920: \de\ratss{{\tx{\ssf rat}}}

921: \de\rbb{]\!]}

922: \de\reals{\tx{\sf real}}

923: \de\realss{{\tx{\ssf real}}}

924: \de\rec{\tx{\bi rec}}

925: \de\redSig{\,|\,_{\Sig}}

926: \de\reffs{\tx{\sf ref}}

927: \de\refss{{\tx{\ssf ref}}}

928: \de\rel#1{\tx{\rm (rel $#1$)}}

929: \de\rem{\tx{\bi rem}}

930: \de\res{\tx{\sf re}}

931: \de\rest{\restriction}

932: \de\restt{\tx{\bi rest}}

933: \de\rh{\tx{$\rho$}}

934: \de\rtt{{\tx{\tt r}}}

935: \de\rvec{\tx{$\vec{r}$}}

936: \de\rxx{\raise.2ex}

937: \de\rxxx{\raise.3ex}

938: \de\rxxxx{\raise.4ex}

939: \de\sbar{{\tx{$\bar s$}}}

940: \de\scr{\Cal}

941: %\de\scr{\fam\eusbfam}

942: \de\se{\tx{\bi se}}

943: \de\seq{\tx{\bi seq}}

944: \de\seqt{\longmapsto}

945: \de\setss{{\tx{\ssf set}}}

946: \de\shead#1#2{\mn{\bf {#1}\ \ \ {#2}}\sn}

947: \de\sheadb#1#2{\mn{\bf {#1}\ \ \ {#2}}\nl}

948: \de\sheadrun#1#2{\mn{\bf {#1}\ \ \ {#2}.\ \ }}

949: \de\sheads#1#2{\mn{\bf {#1}\ \ \ {#2}}}

950: \de\sig{{\tx{$\sigma$}}}

951: \de\signatures{\tx{\sf signature}}

952: \de\sigo{\tx{$\sig_0$}}

953: \de\sigp{\tx{$\sig'$}}

954: \de\sinSortSig{{\tx{$s \in \SortSig$}}}

955: \de\sinSortSigx{\tx{$s \in \SortSigx$}}

956: \de\skips{\tx{\sf skip}}

957: \de\smskip{\smallskip}

958: \de\smskipn{\smallskip\nin}

959: \de\sn{\smallskip\nin}

960: \de\snap{\tx{\bi snap}}

961: \de\sort#1{\tx{\bi sort\,{$(#1)$}}}

962: \de\sortss{\tx{\sf sorts}}

963: \de\sq{\simeq}

964: \de\sqsseq{\sqsubseteq}

965: \de\srchs{\tx{\sf srch}}

966: \de\ssA{{\ssz A}}

967: \de\ssB{{\ssz B}}

968: \de\ssbe{{\ssz \beta}}

969: \de\sset{\subset}

970: \de\sseq{\subseteq}

971: \de\sshead#1#2{{\bf {#1}\ \ \ {#2}}\sn\!}

972: \de\ssheadrun#1#2{{\bf {#1}\ \ \ {#2}.\ \ }}

973: \de\ssheads#1#2{{\bf {#1}\ \ \ {#2}}}

974: \de\sss{\tx{\sf s}}

975: \de\sz{\ssize}

976: \de\ssz{\sssize}

977: \de\starb{{\tx{$\bk *$}}}

978: \de\state{\tx{\bi state}}

979: \de\stks{{\tx{\sf stk}}}

980: \de\stkss{{\tx{\ssf stk}}}

981: \de\str{\tx{$^*$}}

982: \de\stt{{\tx{\tt s}}}

983: \de\strb{\tx{$^{\bk*}$}}

984: \de\subex{\tx{\bi subex}}

985: \de\subst{\tx{\bi subst}}

986: \de\sucs{\tx{\sf suc}}

987: \de\suu{{\tx{$s^\uu$}}}

988: \de\sx{{\tx{$s^*$}}}

989: \de\tA{\tx{$\bb{t}^A$}}

990: \de\ta{\tx{$\tau$}}

991: \de\tagx{\tag{$*$}}

992: \de\tagxx{\tag{$**$}}

993: \de\tagxxx{\tag{${*}{*}{*}$}}

994: \de\tagxxxx{\tag{${*}{*}{*}{*}$}}

995: \de\tagxxxxx{\tag{${*}{*}{*}{*}{*}$}}

996: \de\tagxxxxxx{\tag{${*}{*}{*}{*}{*}{*}$}}

997: \de\tbar{\tx{$\bar t$}}

998: \de\te{\tx{\bi te}}

999: \de\term{\tx{\bi term}}

1000: \de\that{\tx{$\hat t$}}

1001: \de\thens{\tx{\sf then}}

1002: \de\tif#1{&\tx{if \ {#1}}}

1003: \de\til{\tilde}

1004: \de\tiltil#1{\tilde{\tilde#1}}

1005: \de\tinT{\tx{$t\in\TT$}}

1006: \de\th{\tx{$\theta$}}

1007: \de\tofrom{\leftrightarrow}

1008: \de\tos{\tx{\sf to}}

1009: \de\toto{\rightrightarrows}

1010: \de\trues{\tx{\sf true}}

1011: \de\tstile{\vdash}

1012: \de\tto{\tx{$\Rightarrow$}}

1013: \de\ttofrom{\tx{$\Leftrightarrow$}}

1014: \de\ttstile{\models}

1015: \de\ttt{\tx{{\sf t}\!{\sf t}}}

1016: \de\tup#1#2#3{\tx{$#1_{#2},\dots,#1_{#3}$}}

1017: \de\tuptimes#1#2#3{\tx{$#1_{#2} \times \dots \times #1_{#3}$}}

1018: \de\tvec{\tx{$\vec{t}$}}

1019: \de\tvecstrut{\vec{\mathstrut t}}

1020: \de\twiddle{{\tx{$\sim$}}}

1021: \de\twiddlek{\tx{$\sim_k$}}

1022: \de\tx{\text}

1023: \de\txvec{\tx{$\vec{t^*}$}}

1024: \de\type#1{\tx{\bi type$(#1)$}}

1025: \de\ua{{\tx{$\uparrow$}}}

1026: \de\ub#1{\tx{\underbar{#1}}}

1027: \de\ubi{{\tx{\bi u}}}

1028: \de\ubrivec{\tx{$u | \ivec$}}

1029: \de\uinProdTypeSig{\tx{$u \in \ProdTypeSig$}}

1030: \de\ul{\underline}

1031: \de\ulc{\ulcorner}

1032: \de\up{\tx{$^\wedge$}}

1033: \de\urc{\urcorner}

1034: \de\unspecs{\tx{\sf unspec}}

1035: \de\unspecN{\tx{$\unspecs_\natss$}}

1036: \de\unspecB{\tx{$\unspecs_\boolss$}}

1037: \de\ups{{\tx{$\upsilon$}}}

1038: \de\utos{{\tx{$u \to s$}}}

1039: \de\utov{{\tx{$u \to v$}}}

1040: \de\uu{{\tx{\ssf u}}}

1041: \de\uuu{\tx{{\sf u}\!\i}}

1042: \de\uuus{\tx{$\uuu_s$}}

1043: \de\uvec{\tx{$\vec{u}$}}

1044: \de\ux{{\tx{$u^*$}}}

1045: \de\val{\tx{\bi val}}

1046: \de\vals{\tx{\sf val}}

1047: \de\valss{{\tx{\ssf val}}}

1048: \de\var#1{\tx{\bi var$(#1)$}}

1049: \de\vars{\tx{\sf var}}

1050: \de\vects{\tx{\sf vect}}

1051: \de\vinProdTypeSig{\tx{$v \in \ProdTypeSig$}}

1052: \de\vvec{\tx{$\vec{v}$}}

1053: \de\vvecstrut{\vec{\mathstrut v}}

1054: \de\vx{{\tx{$v^*$}}}

1055: \de\vxvec{\tx{$\vec{v}\,^*$}}

1056: \de\whiles{\tx{\sf while}}

1057: \de\wrt{w.r.t\.}

1058: \de\wvec{\tx{$\vec{w}$}}

1059: \de\wx{{\tx{$w^*$}}}

1060: \de\x{\tx{\bi x}}

1061: \de\xbar{\tx{$\bar x$}}

1062: \de\xib{\tx{$\bs\xi$}}

1063: \de\xii{\tx{$\xi$}}

1064: \de\xinAbaru{\tx{$x \in \Abaru$}}

1065: \de\xinAu{\tx{$x \in \Au$}}

1066: \de\xinAw{\tx{$x \in \Aw$}}

1067: \de\xtt{{\tx{\tt x}}}

1068: \de\xttx{\tx{$\xtt^*$}}

1069: \de\xtts{{\tx{\smalltt x}}}

1070: \de\xttvec{\tx{$\vec{\xtt}$}}

1071: \de\xvec{\tx{$\vec{x}$}}

1072: \de\xx{\tx{$x^*$}}

1073: \de\xxvec{\tx{$\xvec^*$}}

1074: \de\xxx{\tx{${*}{*}{*}$}}

1075: \de\xxxx{\tx{${*}{*}{*}{*}$}}

1076: \de\xxxxx{\tx{${*}{*}{*}{*}{*}$}}

1077: \de\xxxxxx{\tx{${*}{*}{*}{*}{*}{*}$}}

1078: \de\yinAs{\tx{$y \in \As$}}

1079: \de\yinAv{\tx{$y \in \Av$}}

1080: \de\yvec{\tx{$\vec{y}$}}

1081: \de\yx{\tx{$y^*$}}

1082: \de\yxvec{\tx{$\yvec\,^*$}}

1083: \de\ytt{{\tx{\tt y}}}

1084: \de\yttx{\ytt\str}

1085: \de\ztt{{\tx{\tt z}}}

1086: \de\zttx{\ztt\str}

1087: \de\zvec{\tx{$\vec{z}$}}

1088: \de\zx{\tx{$z^*$}}

1089: \de\zxvec{\tx{$\zvec\,^*$}}

1090: 

1091: 

1092: % *** MACROS ***

1093: 

1094: \newcount\mmm

1095: \de\itemn#1{\mmm=#1\par\indentn\mmm \advance\mmm by 1

1096:       \hangindent\mmm \parindent \textindent

1097:  }

1098: 

1099: \newcount\nnn

1100: 

1101: \global\de\indentn#1{

1102: \nnn=#1

1103: \loop \ifnum\nnn>0 

1104:       \indent

1105:       \advance\nnn by -1

1106: \repeat

1107: }

1108: 

1109: \de\itemm{\itemn1}

1110: \de\itemmm{\itemn2}

1111: 

1112: \newcount\ppp

1113: 

1114: \global\de\quadn#1{

1115: \n=#1

1116: \loop \ifnum\ppp>0

1117:         \quad

1118:         \advance\ppp by -1

1119: \repeat

1120: }

1121: 

1122: \de\AEE{\tx{\bi AE}}

1123: \de\AExA{\tx{$\AEE\,_\xtts^A$}}

1124: \de\Abe{\tx{$(A,\be)$}}

1125: \de\Abes{{\tx{$A,\ssbe$}}}

1126: \de\AbsComp{\tx{\bi AbsComp}}

1127: \de\AbsCompA{\tx{$\AbsComp(A)$}}

1128: \de\Ai{\tx{$A_i$}}

1129: \de\Ao{$\tx{\rm A}_0$}

1130: \de\Asprod{\tx{$A_{s_1}\times\dots\times A_{s_n}$}}

1131: \de\Ass{\tx{$A_{s_1},\dots,, A_{s_n}$}}

1132: \de\Asua{\tx{$\As^\ua$}}

1133: \de\AtStx{\tx{$\AtSt\,_\xtts$}}

1134: \de\Aua{\tx{$A^\ua$}}

1135: \de\Avua{\tx{$\Av^\ua$}}

1136: \de\BBBbar{\tx{$\ol \BBB$}}

1137: \de\Bbals{\tx{$\Bb_{\al,s}$}}

1138: \de\Bbalu{\tx{$\Bb_{\al,u}$}}

1139: \de\Bbe{\tx{$(B,\be)$}}

1140: \de\Bbs{\tx{$\Bb_s$}}

1141: \de\Bbu{\tx{$\Bb_u$}}

1142: \de\Bsua{\tx{$B_s^\ua$}}

1143: \de\Bu{\tx{$B^u$}}

1144: \de\Bua{\tx{$B^\ua$}}

1145: \de\Bv{\tx{$B^v$}}

1146: \de\Bvua{\tx{$\Bv^\ua$}}

1147: \de\CS{\tx{\bf CS}}

1148: \de\CSo{\tx{$\CS_0$}}

1149: \de\CSn{\tx{$\CS^n$}}

1150: \de\CSon{\tx{$\CS_0^n$}}

1151: \de\CalLQQe{\tx{$\Cals(\LQQe)$}}

1152: \de\CalQQ{\tx{$\Cals(\QQ)$}}

1153: \de\CalX{\tx{$\Cals(X)$}}

1154: \de\CalXalbar{\tx{$(\CalX, \,\albar)$}}

1155: \de\CalXs{\tx{$\CalX_s$}}

1156: \de\CalXx{\tx{$\CalX^*$}}

1157: \de\Calbar{\tx{$\Cs_\albar$}}

1158: \de\Cals{\tx{$\Cs_\al$}}

1159: \de\CompI{\tx{$\Comp_1$}}

1160: \de\CompIA{\tx{$\Comp_1^A$}}

1161: \de\CompStep{\Comp\Step}

1162: \de\CompStepA{\tx{$\CompStep^A$}}

1163: \de\CompStepxA{\tx{$\CompStep\,_\xtts^A$}}

1164: \de\CompTree{\Comp\Tree}

1165: \de\CompTreeA{\tx{$\CompTree^A$}}

1166: \de\CompTreeASsig{\tx{$\CompTreeA(S,\sig)$}}

1167: \de\CompTreeStage{\tx{\bi CompTreeStage}}

1168: \de\CompTreeStageA{\tx{$\CompTreeStage^A$}}

1169: \de\CompTreeStageB{\tx{$\CompTreeStage^B$}}

1170: \de\CompTreeStagexA{\tx{$\CompTreeStage\,_\xtts^A$}}

1171: \de\CompTreexA{\tx{$\CompTree\,_\xtts^A$}}

1172: \de\CompalA{\tx{$\Comp_\al(A)$}}

1173: \de\ConcRep{\tx{\bi ConcRep}}

1174: \de\ConcRepA{\tx{$\ConcRep(A)$}}

1175: \de\Cua{\tx{$C^\ua$}}

1176: \de\Dal{\tx{$D_\al$}}

1177: \de\Dals{\tx{$D_{\al,s}$}}

1178: \de\Dalu{\tx{$D_\al^u$}}

1179: \de\Dalbar{\tx{$D_\albar$}}

1180: \de\Dalbars{\tx{$D_{\albar,s}$}}

1181: \de\Du{\tx{$D^u$}}

1182: \de\Dx{{\tx{$D^*$}}}

1183: \de\Dxal{{\tx{$D^*_\al$}}}

1184: \de\Dxalbar{\tx{$D^*_\albar$}}

1185: \de\Dxalbars{\tx{$D^*_{\albar,s}$}}

1186: \de\Dxbar{\tx{$\ol{D^*}$}}

1187: \de\FAPCS{\tx{\bi FAPCS}}

1188: \de\Findroots{\tx{\sf Findroots}}

1189: \de\IIIp{\tx{$\III_p$}}

1190: \de\IIItN{\tx{$\III_t^N$}}

1191: \de\Iq{\tx{$I^q$}}

1192: \de\LQQe{\tx{$\tx{L}(\QQ,\es)$}}

1193: \de\LU{\tx{\bf LU}}

1194: \de\LUF{\tx{$\LU_F$}}

1195: \de\LUs{\tx{\smallbf LU}}

1196: \de\LUFs{\tx{$\LUs_F$}}

1197: \de\Lp{\tx{$L^p$}}

1198: \de\MC{\tx{\bf MC}}

1199: \de\MCF{\tx{$\MC_F$}}

1200: \de\MCs{\tx{\smallbf MC}}

1201: \de\MCFs{\tx{$\MCs_F$}}

1202: \de\NNNN{\tx{$\NN^\NN$}}

1203: \de\NNtoAs{\tx{$[\NN\to \!A_s]$}}

1204: \de\NNua{\tx{$\NN^\ua$}}

1205: \de\Nbials{\tx{$\Nbi_{\al,s}$}}

1206: \de\Nbialu{\tx{$\Nbi_{\al,u}$}}

1207: \de\Nbis{\tx{$\Nbi_s$}}

1208: \de\Nbiu{\tx{$\Nbi_u$}}

1209: \de\Omal{\tx{$\Om_\al$}}

1210: \de\Omali{\tx{$\Om_{\al,i}$}}

1211: \de\Omalbar{\tx{$\Om_\albar$}}

1212: \de\Omalbars{\tx{$\Om_{\albar,s}$}}

1213: \de\Omalbarreal{\tx{$\Om_{\albar,\realss}$}}

1214: \de\Omalbaru{\tx{$\Om_\albar^u$}}

1215: \de\Omalbarvua{\tx{$\Omalbarv^\ua$}}

1216: \de\Omalbarv{\tx{$\Om_\albar^v$}}

1217: \de\Omalbarw{\tx{$\Om_\albar^w$}}

1218: \de\Omalreal{\tx{$\Om_{\al,\realss}$}}

1219: \de\Omals{\tx{$\Om_{\al,s}$}}

1220: \de\Omalu{\tx{$\Om_\al^u$}}

1221: \de\Ombe{\tx{$\Om_\be$}}

1222: \de\Ombes{\tx{$\Om_{\be,s}$}}

1223: \de\Ombeu{\tx{$\Om_\be^u$}}

1224: \de\Ombev{\tx{$\Om_\be^v$}}

1225: \de\Ombew{\tx{$\Om_\be^w$}}

1226: \de\Oms{\tx{$\Om_s$}}

1227: \de\Omxs{\tx{$\Om^*_s$}}

1228: \de\PAN{\tx{$P^{A^N}$}}

1229: \de\PAn{\tx{$P^A_n$}}

1230: \de\PAua{\tx{$\PA^\ua$}}

1231: \de\PEabc{\tx{$\PE_\abctts$}}

1232: \de\PEabcA{\tx{$\PE\,_\abctts^A$}}

1233: \de\PPPom{\tx{$\PPP_\om$}}

1234: \de\PPPomp{\tx{$\PPP_\om^+$}}

1235: \de\PPPfin{\tx{$\PPP_\fin$}}

1236: \de\PPPfinp{\tx{$\PPP_\fin^+$}}

1237: \de\PTerms{\tx{$\PTerm_s$}}

1238: \de\PTermx{\tx{$\PTerm_{\,\xtts}$}}

1239: \de\PTermxs{\tx{$\PTerm_{\,\xtts,s}$}}

1240: \de\PfA{\tx{$P_f^A$}}

1241: \de\Pfs{\tx{$P_\fs$}}

1242: \de\Procabc{\tx{$\Proc\,_\abctts$}}

1243: \de\PsqcupA{\tx{$P_\sqcup^A$}}

1244: \de\QA{\tx{$Q^A$}}

1245: \de\QN{\tx{$\QQ^\NN$}}

1246: \de\RRRLN{\tx{$\RRR^{<,N}$}}

1247: \de\RRRi{\tx{$\RRR_1$}}

1248: %\de\RRRinv{\tx{$\RRR_{\invs}$}}

1249: \de\RRRo{\tx{$\RRR_0$}}

1250: \de\RRRp{\tx{$\RRR_p$}}

1251: \de\RRRtN{\tx{$\RRR_t^N$}}

1252: \de\Rat{\tx{\bi Rat\/}}

1253: \de\RestA{\tx{$\Rest\,^A$}}

1254: \de\RestB{\tx{$\Rest\,^B$}}

1255: \de\RestxA{\tx{$\Rest\,_\xtts^A$}}

1256: \de\SE{\tx{\bi SE}}

1257: \de\SExA{\tx{$\SE\,_\xtts^A$}}

1258: \de\Sf{\tx{$S_\fs$}}

1259: \de\SigRRR{\tx{$\Sig(\RRR)$}}

1260: \de\Sigf{\tx{$\Sig_\fs$}}

1261: \de\Sinit{\tx{$S_{init}$}}

1262: \de\StateAcupxua{\tx{$\StateA\cup\{*,\ua\}$}}

1263: \de\StateAua{\tx{$\StateA^\ua$}}

1264: \de\StateB{\tx{$\State(B)$}} 

1265: \de\StatesA{\tx{$\State_s(A)$}}

1266: \de\Statex{\tx{$\State\,_\xtts$}}

1267: \de\StatexA{\tx{$\State\,_\xtts(A)$}}

1268: \de\StatexAua{\tx{$\State\,_\xtts(A)^\ua$}}

1269: \de\Stmtx{\tx{$\Stmt_{\,\xtts}$}}

1270: \de\TExsA{\tx{$\TE\,_{\xtts,s}^A$}}

1271: \de\PTExsA{\tx{$\PTE\,_{\xtts,s}^A$}}

1272: \de\Terms{\tx{$\Term_s$}}

1273: \de\Termx{\tx{$\Term_{\,\xtts}$}}

1274: \de\Termxs{\tx{$\Term_{\,\xtts,s}$}}

1275: \de\Tree{\tx{\bi Tree}}

1276: \de\WhileCC{\tx{\bi WhileCC\/}}

1277: \de\WhileCCSig{\WhileCC(\Sig)}

1278: \de\WhileCCbig{\tx{\bigbi WhileCC}}

1279: \de\WhileCCx{\tx{$\WhileCC^\starb$}}

1280: \de\WhileCCxSig{\WhileCCx(\Sig)}

1281: \de\WhileCCxSigN{\WhileCCx(\SigN)}

1282: \de\WhileCCxbig{\tx{$\WhileCCbig^\starb$}}

1283: \de\XXXo{\tx{$\XXX_0$}}

1284: \de\Xal{\tx{$(X,\al)$}}

1285: \de\Xals{\tx{$X_{\al,s}$}}

1286: \de\Xalu{\tx{$X_\al^u$}}

1287: \de\Xalbar{\tx{$X_\albar$}}

1288: \de\Xalbars{\tx{$X_{\albar,s}$}}

1289: \de\Xnorm{\tx{$(X,\,\normfn)$}}

1290: \de\Xu{\tx{$X^u$}}

1291: \de\Xs{\tx{$X_s$}}

1292: \de\Xsal{\tx{$(X_s,\al)$}}

1293: \de\Xx{{\tx{$X^*$}}}

1294: \de\Xxal{{\tx{$X^*_\al$}}}

1295: \de\Xxalbar{\tx{$X^*_\albar$}}

1296: \de\Xxalbars{\tx{$X^*_{\albar,s}$}}

1297: \de\Xxbar{\tx{$\ol{X^*}$}}

1298: \de\Xua{\tx{$X^\ua$}}

1299: \de\Yua{\tx{$Y^\ua$}}

1300: \de\Zua{\tx{$Z^\ua$}}

1301: \de\abctt{{\tx{$\att,\btt,\ctt$}}}

1302: \de\abctts{{\tx{$\atts,\btts,\ctts$}}}

1303: \de\aexAbe{\tx{$\aee\,_\xtts^\Abes$}}

1304: \de\alalbar{\tx{$(\al,\albar)$}}

1305: \de\albars{{\tx{$\albar_s$}}}

1306: \de\albarreal{{\tx{$\albar_\realss$}}}

1307: \de\albaru{{\tx{$\albar^u$}}}

1308: \de\albarv{{\tx{$\albar^v$}}}

1309: \de\albarx{{\tx{$\ol\alx$}}}

1310: \de\albarxs{{\tx{$\ol \alxs$}}}

1311: \de\alreal{{\tx{$\alpha_\realss$}}}

1312: \de\als{{\tx{$\alpha_s$}}}

1313: \de\alu{{\tx{$\alpha^u$}}}

1314: \de\alx{{\tx{$\alpha^*$}}}

1315: \de\alxs{{\tx{$\alpha_s^*$}}}

1316: \de\bes{{\tx{$\beta_s$}}}

1317: \de\beu{{\tx{$\beta^u$}}}

1318: \de\bev{{\tx{$\beta^v$}}}

1319: \de\bigomp{{}_{\dz\om}^{\dz+}}

1320: \de\bimaprighturaise#1#2{\bimapright^{\raise#1pt\hbox{$\dz{#2}$}}}

1321: \de\cP{\cnr{P}}

1322: \de\cS{\cnr{S}}

1323: \de\choicesttx{\tx{\tt choices\str}}

1324: \de\choosezb{\tx{$\chooses \ \ztt:b$}}

1325: \de\choosezt{\tx{$\chooses \ \ztt:t$}}

1326: \de\comp{\tx{\rm comp}}

1327: \de\complength{\tx{\bi complength}}

1328: \de\complengthxAbe{\tx{$\complength\,_\xtts^\Abes$}}

1329: \de\compseq{\tx{\bi compseq}}

1330: \de\compseqxAbe{\tx{$\compseq\,_{\xtts}^\Abes$}}

1331: \de\compstage{\tx{\bi compstage}}

1332: \de\compstagexA{\tx{$\compstage\,_{\xtts}^A$}}

1333: \de\compstep{\tx{\bi compstep}}

1334: \de\compstepxAbe{\tx{$\compstep\,_{\xtts}^\Abes$}}

1335: \de\comptree{\tx{\bi comptree}}

1336: \de\comptreexA{\tx{$\comptree\,_\xtts^A$}}

1337: \de\delbA{\tx{$\delb_A$}}

1338: \de\delbAu{\tx{$\delb_A^u$}}

1339: \de\delbAv{\tx{$\delb_A^v$}}

1340: \de\delbAw{\tx{$\delb_A^w$}}

1341: \de\delbs{\tx{$\delb^s$}}

1342: \de\delbsA{\tx{$\delb^s_A$}}

1343: \de\delbu{\tx{$\delb^u$}}

1344: \de\delbuA{\tx{$\delb^u_A$}}

1345: \de\domal#1{\tx{$\tx{\bi dom}_\albar(#1)$}}

1346: \de\dsH{\tx{$\ds_\Hss$}}

1347: \de\dss{\tx{$\ds_s$}}

1348: \de\dsu{\tx{$\ds_u$}}

1349: \de\dsx{\tx{$\ds^*$}}

1350: \de\dsxs{\tx{$\ds^*_s$}}

1351: \de\ebarn{\tx{$\ebar[n]$}}

1352: \de\econk{\tx{$e_{\conss[k]}$}}

1353: \de\eqp{\tx{$\eqs_p$}}

1354: \de\enumSig{\tx{$\enum_\Sig$}}

1355: \de\enumC{\tx{$\enums_C$}}

1356: \de\enumX{\tx{$\enums_X$}}

1357: \de\enums{\tx{\sf enum}}

1358: \de\evalG{\tx{$\eval_G$}}

1359: \de\expmins{\tx{\sf expmin}}

1360: \de\fin{{\tx{\rm fin}}}

1361: \de\fmin{\tx{$f^-$}}

1362: \de\foralls{\tx{\sf forall}}

1363: \de\fua{\tx{$f^\ua$}}

1364: \de\gua{\tx{$g^\ua$}}

1365: \de\haltt{\tx{\tt halt}}

1366: \de\hyphenCCb{\tx{\bi -CC}}

1367: \de\invR{\tx{$\invs^\RRR$}}

1368: \de\invs{\tx{\sf inv}}

1369: \de\isints{\tx{\sf isint}}

1370: \de\iI{\tx{$\is_I$}}

1371: \de\iN{\tx{$\is_N$}}

1372: \de\kapn{\tx{$\kap[n]$}}

1373: \de\kbarn{\tx{$\kbar[n]$}}

1374: \de\lp{\tx{$\ell^p$}}

1375: \de\lsp{\tx{$\lss_p$}}

1376: \de\norm#1{\tx{$\|#1\|$}}

1377: \de\normfn{\tx{$\|\cdot\|$}}

1378: \de\peabcAbe{\tx{$\pe\,_\abctts^\Abes$}}

1379: \de\piv{\tx{\rm piv}}

1380: \de\pivo{\tx{$\piv_0$}}

1381: \de\pivom{\tx{$\piv_\om$}}

1382: \de\pteboolsAbe{\tx{$\pte\,_{\boolss,s}^\Abes$}}

1383: \de\ptexsAbe{\tx{$\pte\,_{\xtts,s}^\Abes$}}

1384: \de\qchooses{\tx{`\chooses'}}

1385: \de\restxAbe{\tx{$\restt\,_\xtts^\Abes$}}

1386: \de\scalars{\tx{\sf scalar}}

1387: \de\sexAbe{\tx{$\se\,_\xtts^\Abes$}}

1388: \de\sigG{\tx{$\sig_G$}}

1389: \de\sigs{\tx{$\sig_s$}}

1390: \de\sigzn{\sig\zn}

1391: \de\spaces{\tx{\sf space}}

1392: \de\sprod{\tx{${s_1}\times\dots\times {s_n}$}}

1393: \de\statex{\tx{$\state\,_\xtts$}}

1394: \de\statexA{\tx{$\state\,_\xtts(A)$}}

1395: \de\taun{\tx{$\tau[n]$}}

1396: \de\texboolA{\tx{$\te\,_{\boolss,s}^A$}}

1397: \de\tom{\to_\om}

1398: \de\tomp{\to_\om^+}

1399: \de\totop{\toto^+}

1400: \de\truncs{\tx{\sf truncs}}

1401: \de\vart{\tx{\bi vart}}

1402: \de\vartxbe{\tx{$\vart\,_\xtts^\ssbe$}}

1403: \de\vectors{\tx{\sf vector}}

1404: \de\xo{\tx{$x_0$}}

1405: \de\zn{\{\ztt/n\}}

1406: 

1407: \redefine\Ds{\tx{$D_s$}}

1408: 

1409: 

1410: 

1411: % *** END OF HEADERS, START OF TEXT ***

1412: 

1413: 

1414: 

1415: \cbb{Abstract versus Concrete Computation on Metric Partial Algebras}

1416: \bigskip

1417: \bigskip

1418: \ce{{\bf J.V. Tucker}

1419: }

1420: \sn

1421: \ce{\it Department of Computer Science,}

1422: \ce{\it University of Wales, Swansea  SA2 8PP, Wales}

1423: \ce{\tt J.V.Tucker@swansea.ac.uk}

1424: \mn

1425: \ce{{\bf J.I. Zucker}\footnotemark"*"}

1426: \footnotetext"*\ "{

1427: \n The research of the second author was supported 

1428: by a grant from the Natural Sciences

1429: and Engineering Research Council (Canada)

1430: and by a Visiting Fellowship from the Engineering

1431: and Physical Sciences Research Council (U.K.)

1432: }

1433: \nl

1434: \ce{\it Department of Computing and Software,}

1435: \ce{\it McMaster University, Hamilton, Ont\.  L8S 4L7, Canada}

1436: \ce{\tt zucker@mcmaster.ca}

1437: \bn

1438: \bn

1439: 

1440: \cbb{Abstract}

1441: \mn

1442: {\narrower\smallrm

1443: Data types containing infinite data, such as the real

1444: numbers, functions, bit streams and waveforms, are

1445: modelled by topological many-sorted algebras. In the

1446: theory of computation on topological algebras there is a

1447: considerable gap between so-called abstract and concrete

1448: models of computation. We prove theorems that bridge the

1449: gap in the case of metric algebras with partial operations.

1450: 

1451: With an abstract model of computation on an algebra, the

1452: computations are invariant under isomorphisms and do not

1453: depend on any representation of the algebra. Examples of

1454: such models are the \qwhiles\ programming language and the BCSS model. 

1455: With a concrete model of computation, the computations

1456: depend on the choice of a representation of the algebra

1457: and are not invariant under isomorphisms. Usually, the

1458: representations are made from the set \NN\ of natural numbers,

1459: and computability is reduced to classical computability

1460: on \NN. Examples of such models are computability via

1461: effective metric spaces, effective domain representations, and type

1462: two enumerability.

1463: 

1464: The theory of abstract models is stable:

1465: there are many models of computation, and conditions under

1466: which they are equivalent are largely known. The theory

1467: of concrete models is not yet stable, though it seems to be

1468: converging: several interesting models are known to be

1469: equivalent over special types of topological algebra. We

1470: investigate the problem of comparing the two types of

1471: models and, hence, establishing a unified and stable theory of

1472: computation for topological algebras.

1473: 

1474: First, we show that to compute functions on topological

1475: algebras using an abstract model, it is necessary

1476: that one must use algebras with partial operations and

1477: computable functions that are continuous and multivalued.

1478: This multivaluedness is needed even to compute

1479: single-valued functions, and so {\it abstract models must

1480: be nondeterministic even to compute deterministic

1481: problems\/}. Then we choose the \qwhiles-array programming language

1482: as an abstract model for computing on any data type,

1483: and extend it with a nondeterministic assignment

1484: of ``countable choice". This is the new \WhileCCx\ model.

1485: Finally, we introduce the notion of approximable multivalued

1486: computation on metric algebras.  As a concrete model,

1487: we choose effective metric spaces. 

1488: Among a number of results we prove the following.

1489: 

1490: For any metric algebra $A$ with an effective representation,

1491: any function \WhileCCx\ approximable over $A$ is computable

1492: in the effective representation of the metric algebra

1493: $A$. Conversely, we show that, under certain reasonable

1494: conditions on the effective metric algebra $A$, any

1495: function that is effective is also \WhileCCx\ approximable. We

1496: give an equivalence theorem, and examples 

1497: of algebras where equivalence holds.

1498: 

1499: }

1500: 

1501: \bn

1502: {\bf Keywords:\/}

1503: {\smallrm

1504: data types, abstract models of computation, concrete models of computation,

1505: partial algebra, \qwhiles\ language, countable choice, nondeterminism,

1506: multivalued functions, metric algebras, topological algebras,

1507: approximation by \qwhiles\ programs, effective metric spaces, 

1508: effective Banach spaces}

1509: 

1510: 

1511: \Shead0{Introduction}

1512: The theory of data in computer science is based on many sorted algebras

1513: and homomorphisms.  The theory originates in the 1960s, and has

1514: developed a wealth of theoretical concepts, methods and techniques for

1515: the specification, construction, and verification of software and

1516: hardware systems. It is a significant achievement in computer science

1517: and has exerted a profound influence on programming 

1518: \cite{wirsing,adj78,meseguer-goguen}. 

1519: However, given the

1520: absolutely fundamental nature of its subject matter --- data --- there are

1521: many fascinating and significant open problems.  An important general

1522: problem is:

1523: 

1524: {\displaytext

1525: To develop a comprehensive theory of specification, computation and

1526: reasoning with infinite data.

1527: 

1528: }

1529: By infinite data we mean real numbers, spaces of functions, streams of

1530: bits or reals, waveforms, multidimensional graphics objects, video, and

1531: analogue and digital interfaces. The application areas are obvious:

1532: scientific modelling and simulation, embedded systems, graphics and

1533: multimedia communications. 

1534: 

1535: Data types containing infinite data are modelled by topological 

1536: many-sorted algebras. In this paper we consider computability theory

1537: on topological algebras and investigate the problem

1538: 

1539: {\displaytext

1540: To compare and integrate high-level, representation independent,

1541: abstract models of computation with low-level, representation dependent,

1542: concrete models of computation in topological algebras.

1543: 

1544: }

1545: Computability theory lies at the technical heart of 

1546: theories of both specification

1547: and reasoning about such systems. There are many disparate ways of defining

1548: computable functions on topological algebras and some have (different)

1549: significant mathematical theories. 

1550: In the case of real numbers one can contrast the approaches in books such as

1551: \cite{aberth80,aberth01,pourel-richards,weih:book,bcss}.

1552: 

1553: Generally speaking, the models

1554: of computation for an algebra can be divided into two kinds: the

1555: {\it abstract\/} and {\it concrete\/}. 

1556: 

1557: With an {\it abstract model of computation\/} for an algebra the programs 

1558: do not depend on any representation of

1559: the algebra and are

1560: invariant under isomorphisms.

1561: Abstract models originated in the late 1950s in formalising

1562: flowcharts, and include program schemes and

1563: many general models of recursion.

1564: Examples of such models are the \,\While\ \,

1565: programming language over any algebra

1566: and the 

1567: Blum-Cucker-Shub-Smale model \cite{bss,bcss}

1568: over the rings of real or complex numbers.

1569: The theory of abstract models is stable: there are

1570: many models of computation and the conditions under which they are

1571: equivalent are largely known \cite{tz:book,tz:hb}.

1572: For example, \qwhiles\ programs, flow charts, register machines,

1573: Kleene schemes, etc., are equivalent on {\it any\/} algebra;

1574: the BCSS models are simply instances obtained

1575: by choosing the algebra appropriate to the 

1576: ring or ordered ring.

1577: 

1578: With a {\it concrete model of computation\/} for an algebra the programs 

1579: and computations are not

1580: invariant under isomorphisms, but depend on the choice

1581: of a representation of the algebra. Usually, the representations are

1582: made from the set \NN\ of natural numbers, and computability on an algebra

1583: is reduced to classical computability on \NN. 

1584: Concrete models originated in the 1940s, in formalising the computable

1585: functions on real numbers.

1586: Examples of general models are

1587: computability via 

1588: \bull

1589: effective metric spaces \cite{moscho64},

1590: \bull

1591: computable sequence structures \cite{pourel-richards},

1592: \bull

1593: domain representations 

1594: \cite{stolt-jvt88,stolt-jvt95,edalat95:icomp,edalat97}, and

1595: \bull

1596: type two enumerability \cite{weih:book}. 

1597: 

1598: \n

1599: The theory of concrete models

1600: is not stable though it seems to be converging: several basic models are

1601: known to be equivalent in special cases (see, \eg, \cite{stolt-jvt99:tcs}

1602: where the four general approaches above are shown to be

1603: equivalent). 

1604: 

1605: In the theory of computation on algebras, abstract models are implemented

1606: by concrete models. Thus, the gap between the models is the gap between

1607: high level programming abstractions 

1608: and low level implementations, and can be explored in

1609: terms of the following concepts:

1610: 

1611: 

1612: 

1613: \bul

1614: {\it Soundness of abstract model\/}: 

1615: The functions computable in the abstract model

1616: are also computable in the concrete model.

1617: \bul

1618: {\it Adequacy of abstract model\/}: 

1619: The functions computable in the concrete model

1620: are computable in the abstract model. 

1621: \bul

1622: {\it Completeness of abstract model\/}: 

1623: Functions are computable in the abstract model

1624: if, and only if, they are computable in the concrete model.

1625: 

1626: \n

1627: However, there is a considerable gap between abstract and concrete

1628: models of computation, especially over topological data types. For

1629: example, the popular abstract model in \cite{bcss} is {\it not\/} sound 

1630: for the main

1631: concrete models because of its assumptions about 

1632: the total computability of relations such as equality.

1633: Equality on the real numbers is not everywhere continuous,

1634: but in all the concrete models

1635: computable functions are continuous

1636: (\cf\ Ceitin's Theorem \cite{moscho64}).

1637: The connection between abstract and concrete models of

1638: computation on the real numbers is examined in \cite{tz:top}

1639: where approximation by \qwhiles\ programs

1640: over a {\it particular\/} algebra

1641: was shown to be equivalent to the standard

1642: concrete model of GL computability over the unit interval.

1643: 

1644: First attempts at bridging the gap for all topological algebras

1645: in general have been made in 

1646: \cite{brattka96,brattka:thesis},

1647: using a generalisation of recursion schemes (abstract

1648: computability) and Weihrauch's type two enumerability  (concrete

1649: computability). Here we investigate further the problems in comparing

1650: the two classes of models and in establishing a unified and stable theory

1651: of computation on topological algebras.  We prove new theorems that bridge

1652: the gap in the case of computations on metric algebras with partial

1653: operations. 

1654: 	

1655: By reflecting on a series of examples,

1656: we show that to compute functions on topological algebras, it is

1657: necessary that one must consider 

1658: 

1659: \itemm{($i$)}

1660: algebras with partial operations,

1661: \itemm{($ii$)}

1662: computable functions that are both continuous and multivalued, and

1663: \itemm{($iii$)}

1664: approximations by abstract programs.

1665: 

1666: \n

1667: In particular, {\it multivalued functions are needed, even to compute 

1668: single-valued functions\/}. Thus, to prove an equivalence between abstract and

1669: concrete models we must 

1670: include a nondeterministic construct to define

1671: multivalued functions, and in this way

1672: use nondeterministic abstract  models even to

1673: compute deterministic problems. 

1674: We find that

1675: 

1676: {\displaytext

1677: imperative and other abstract programming

1678: models must be nondeterministic to express even simple programs on 

1679: topological data types.

1680: 

1681: }

1682: We choose the \,\While\ \,programming language as an abstract model for

1683: computing on any data type, and extend it with the {\it nondeterministic

1684: assignment of countable choice\/} having the form:

1685: $$

1686: \xtt\ ::= \ \chooses\ \ztt: b(\ztt, \xtt, \ytt)

1687: $$

1688: where \ztt\ is a natural number variable and $b$ is a Boolean-valued

1689: operation. This new model is called \,\WhileCCx\ \,computability

1690: (`{\it CC\/}' for ``countable choice", 

1691: `\str' for array variables.) \,In particular,

1692: we introduce a notion of {\it approximable multivalued computation\/}, and

1693: formulate and prove the continuity of their semantics. 

1694: We thus have the partial multivalued functions

1695: approximable by a \WhileCCx\ program on $A$.

1696: 

1697: As a concrete model, we choose {\it effective metric spaces\/}; 

1698: this is known to

1699: be equivalent with several other concrete models. 

1700: In computation with effective metric spaces $A$ we pick an enumeration

1701: \al\ of a subspace $X$ of $A$, and construct the subspace

1702: \,\CalX\ \,of \al-computable elements of $A$, enumerated by \albar.

1703: We thus have the partial functions computable on \,\CalX\

1704: \,in the representation \albar.

1705: 

1706: We then prove two theorems that can be summarised

1707: (a little loosely) as follows.

1708: 

1709: \n

1710: {\bi Soundness Theorem\/}: \sl Let $A$ be any  metric partial  algebra with an

1711: effective representation \al. 

1712: Suppose \,\CalX\ \,is a subalgebra of $A$, effective under \albar.

1713: Then  any function $F$ on $A$

1714: that is

1715: \WhileCCx\ \,approximable over $A$ is computable on \,\CalX\ \,in \albar.

1716: \endpr

1717: 

1718: The soundness theorem is technically involved but quite general,

1719: and gives new insight into the semantics of

1720: imperative programs applied to topological data types.

1721: The converse theorem is more restricted in its data types:

1722: 

1723: \n

1724: {\bi Adequacy Theorem\/}: 

1725: \sl Let $A$ be any  metric partial  algebra A with an

1726: effective representation \al.  Suppose the representation \al\ is \,\WhileCCx\

1727: \,computable and dense.  Then  any function \,$F\: A \to A$  

1728: \,that is computable on \,\CalX\ \,in \albar\

1729: and effectively locally uniformly continuous in \al\

1730: is  \,\WhileCCx\ \,approximable over $A$. 

1731: \endpr

1732: 

1733: These are combined into a {\bi Completeness Theorem\/}.

1734: 

1735: The proper statements of these theorems are given as

1736: Theorems A, B and C (in Sections 6, 7 and 8).

1737: Some interesting applications to algebras of real numbers and to

1738: Banach spaces are studied.

1739: 

1740: Here is the structure of the paper. We begin, in Section 1, by

1741: explaining the role of partiality, continuity and multivaluedness

1742: in computation, using simple examples on the real numbers. 

1743: In Section 2 we describe topological

1744: and metric partial algebras and their extensions. In Section 3 we

1745: introduce the \,\WhileCCx\ \,language, give it an algebraic semantics,

1746: and define approximable \,\WhileCCx\ \,computability.

1747: We will see that the \WhileCCx\ \,language has a complex

1748: semantics.  However on total algebras it defines precisely the 

1749: \Whilex\ \,computable functions.

1750: Section 4 is devoted to examples. In Section 5 we prove the

1751: continuity of these \,\WhileCCx\ \,computable multivalued functions. In Section

1752: 6 we introduce our concrete model, effective metric spaces,

1753: and prove a Soundness Theorem (Theorem $A_0$) for the special case

1754: of surjective enumerations

1755: of countable (not necessarliy metric) algebras.

1756: In Section 7

1757: we define the subspace (or subalgebra) 

1758: of elements computable in a metric algebra,

1759: and then prove the more general Soundness Theorem 

1760: (Theorem A) and, in Section 8, the Adequacy Theorem (Theorem B).

1761: These are combined into a Completeness Theorem (Theorem C)

1762: in Section 9.

1763: Concluding remarks are made in Section 10.

1764: 

1765: This work is part of a research programme  --- starting in \cite{tz:book}

1766: and most recently surveyed in \cite{tz:hb} ---

1767: on the theory of computability on algebras, and its application to

1768: specifiability and verifiability in different areas of computer science

1769: and mathematics. Specifically, it has developed from our studies of real

1770: and complex number computation in \cite{tz:sanantonio,tz:top,tz:hb},

1771: stream algebras in \cite{tz:prague,tz:markt}

1772: and metric algebras in \cite{tz:spec}.

1773: 

1774: We thank Vasco Brattka and Kristian Stewart for

1775: invaluable discussions.

1776: 

1777: \bn

1778: 

1779: \itemm{\bbf1\ \ }{\bbf Nondeterminism, many-valuedness, non-extensionality,

1780: \nl

1781: continuity and partiality: \,Some real number examples}

1782: \sn

1783: When one considers the relation between abstract and concrete models,

1784: a number of intriguing problems appear.

1785: We explain them by considering a series of examples.

1786: Then we formulate our strategy for solving these problems.

1787: 

1788: 

1789: Our chosen abstract and concrete models are introduced later (in

1790: Sections 3 and 5, respectively), so we must explain the problems of computing

1791: on the real number data type in rather general terms. First, we sketch

1792: the abstract and concrete forms of the real number data type. The

1793: picture for topological algebras in general will be clear from the

1794: examples.

1795:  

1796: \newpage

1797: 

1798: \itemm{\bf 1.1}

1799: {\bf Abstract versus concrete data types of real numbers;

1800: \,Continuity; 

1801: \nl

1802: Partiality}

1803: \sheadrun{1.1.1}{Abstract and concrete data types of reals}

1804: To compute on the set \RR\  of real numbers with an

1805: abstract model of computation, we have only to select an

1806: algebra $A$ in which \RR\ is a carrier set. Abstract

1807: computability on an algebra $A$ is a computability {\it relative

1808: to\/} $A$: a function is computable over $A$ if it can

1809: be programmed from the operations of $A$ using the programming constructs

1810: of the abstract model.  Clearly, there are infinitely many choices

1811: of operations with which to make an algebra $A$, and hence

1812: there are infinitely many choices of classes of abstractly computable

1813: functions. All the classes of abstractly computable functions  on \RR\

1814: have decent mathematical theories, resembling the theory of

1815: the computable functions on the natural numbers --- thanks to

1816: the general theory of computable functions on many sorted

1817: algebras \cite{tz:hb}.

1818: 

1819:  

1820: In contrast, to compute on \RR\ with a concrete model

1821: of computation, we choose an appropriate concrete

1822: representation $R$, and map

1823: $$

1824: \al\: R\ \to\ \RR

1825: $$

1826: where $R$ is an algebra made from the set \NN\ of natural numbers. 

1827: For example, the map will be based on the fact that the reals

1828: can be built from the rationals, and hence the

1829: naturals, in a variety of equivalent ways (such

1830: as Cauchy sequences, decimal expansions, etc.). The

1831: computability of functions on the reals is

1832: investigated using the theory of computable functions on

1833: \NN, applied to \RR\ via \al.

1834: 

1835:  

1836: To compare this computation theory with abstract models, we

1837: choose an algebra $A$ in which \RR\ is a carrier set

1838: and, in particular, the operations of $A$ are computable with

1839: respect to the representation \al. For example, multiplication by 3 is not

1840: computable in the decimal representation, but the field

1841: operations on \RR\ are computable in the Cauchy sequence

1842: representation.

1843: 

1844: 

1845: We assume that our concrete model is the subspace \,\CS\

1846: \,of Baire space \,\NNNN\

1847: \,consisting of codings of fast Cauchy sequences of rationals,

1848: \ie, sequences \,$(k_n)$ \,of naturals such that

1849: for all $n$ and all $m>n$, \,$|r_{k_m} - r_{k_n}| < 2^{-n}$,

1850: where \,$r_0,r_1,r_2,\dots$

1851: \,is some standard enumeration of the rationals.

1852: The representing function

1853: $$

1854: \al\: \CS\ \to \ \RR

1855: $$

1856: is continuous and onto.

1857: 

1858: 

1859: \sheadrun{1.1.2}

1860: {Continuity}

1861: Computations with real numbers involve infinite data. The

1862: topology of \RR\ defines a process of

1863: approximation for infinite data; the functions on the

1864: data that are continuous in the topology are exactly the

1865: functions that can be approximated to any desired degree

1866: of precision.

1867: 

1868: For abstract models we assume the algebra $A$ that contains

1869: \RR\ is a topological  algebra, \ie, one in which

1870: the basic operations are continuous in its topologies.  We expect

1871: further that all the computable functions will be continuous. 

1872: The class

1873: of functions that can be abstractly computed exactly

1874: can be quite limited! 

1875: With abstract models, 

1876: approximate computations also turn out to be necessary \cite{tz:top}.

1877: 

1878: In the concrete models, moreover, it follows from 

1879: Ceitin's Theorem \cite{moscho64} that if a function

1880: is computable then it is continuous.

1881: 

1882: Thus, in both abstract and concrete approaches, an analysis

1883: of basic concepts shows that 

1884: computability implies continuity.

1885:  

1886: \sheadrun{1.1.3}

1887: {Partiality}

1888: In computing with an abstract model on $A$ we assume 

1889: $A$ has some boolean-valued functions to test data. For

1890: example, in computing on \RR\ we need to

1891: use the functions  

1892: $$

1893: =_R\: \RR^2 \ \to  \ \BB\ \qquad\tx{and}\qquad  <_R\: \RR^2 \ \to\  \BB

1894: $$

1895: where \,$\BB = \curl{\ttt,\fff}$

1896: \,is the set of booleans.

1897: 

1898: Use of these functions presents a problem, since

1899: total continuous boolean-valued functions on the reals 

1900: must be constant. 

1901: This is because the only continuous functions from a connected space

1902: to a discrete space are the constant functions.

1903: Furthermore,

1904: in \cite{tz:top} it was shown that on connected total topological algebras,

1905: the \qwhiles\ and \qwhiles-array computable functions

1906: are precisely the functions explicitly definable by terms over the algebra.

1907: 

1908: To study the full range of real number computations, we

1909: must therefore redefine these tests as {\it partial\/} 

1910: boolean-valued functions.

1911: Computation with partial algebras has interesting effects

1912: on the theory of computable functions,

1913: as indicated in \cite{tz:top}.

1914: 

1915: On the basis of these preliminary remarks on the data type of reals, we 

1916: turn to the examples.

1917: 

1918: \shead{1.2}{Examples of nondeterminism and many-valuedness}

1919: We now look at three examples of computing functions on \RR.

1920: \Examplen{1.2.1: \,Pivot function} \ Define the function

1921: $$ \piv\:\RRn \ \pto\ \curly{1,\dots,n} $$ by $$

1922: \piv(\tup{x}1n) \ = \ \cases \tx{some}\ i: \ x_i\ne0

1923: \tif{such an $i$ exists}\\ \ua \ow \endcases \tag1 $$

1924: Computation of this pivot is a vital step in the Gaussian

1925: elimination algorithm for inverting matrices.

1926: 

1927: Note that (depending on the precise semantics

1928: for the phrase ``some $i$" in (1))

1929: \,\piv\ \,is {\it nondeterministic\/} or (alternatively)

1930: {\it many-valued\/} on 

1931: \,$\dom{\piv} = \RRn\backslash\curl0$.

1932: \ Further:

1933: \sn

1934: ($a$) \,There is no {\it single-valued\/} function

1935: which satisfies the definition (1) and is 

1936: {\it continuous\/} on \,\RRn.

1937: For such a function,

1938: \,being continuous and integer-valued,

1939: would have to be constant 

1940: on its domain \,$\RRn\backslash\curl0$,

1941: \,with constant value (say) $j\in\curly{1,\dots,n}$.

1942: But its value on the $x_j$-axis would have to be different from $j$,

1943: leading to a contradiction.

1944: 

1945: \sn

1946: ($b$) \,However there {\it is\/} a computable (and hence continuous!)

1947: single-valued function

1948: $$

1949: \pivo\: \CSn\ \pto\ \curly{1,\dots,n}

1950: \tag2

1951: $$

1952: with a simple algorithm.

1953: Note however that \,\pivo\ \,is {\it not extensional\/} on 

1954: \,\CSn\  \,(\ie, not well defined on \RRn),

1955: or (equivalently) the map (2) cannot be factored through

1956: \,\RRn:

1957: $$

1958: \commdiag{

1959: \CSn \cr

1960: \mapdown\lft{\dz\al} &\arrow(2,-1)\lft\bdot\rt{\raisebox3{\dz\pivo}}\cr

1961: \RRn &\mapright^\bdot_{\dz\tx{?}}&\ \curly{1,\dots,n}

1962: }

1963: $$

1964: In effect, we can regain continuity (for a single-valued function),

1965: by foregoing extensionality.

1966: 

1967: \sn

1968: ($c$)

1969: \,Alternatively, we can maintain continuity {\it and\/} extensionality

1970: by giving up single-valuedness.

1971: For the many-valued function

1972: $$

1973: \pivom\: \RRn\ \to \ \PPPom(\curl{1,\dots,n})

1974: $$

1975: (where \,$\PPPom(\dots)$

1976: \,denotes the set of countable subsets of \,\dots)

1977: defined by: \,for all $k\in\curl{1,\dots,n}$

1978: $$

1979: k\,\in\,\pivom(\tup{x}1n)\ \ifff\ x_k\ne0,

1980: $$

1981: is {\it extensional\/} and {\it continuous\/},

1982: where a function

1983: $$

1984: f\: A\ \to\ \PPPom(B)

1985: $$

1986: is defined to be continuous iff for all open $Y\sseq B$,

1987: $$

1988: f^{-1}[Y]\ := \ \curly{x\in A\br f(x)\cap Y \ne\nil}

1989: $$

1990: is open in $A$.

1991: (We will consider continuity of many-valued functions

1992: systematically in Section 5.)

1993: 

1994: \Remarksn{1.2.2}

1995: ($i$) The many-valued function \,\pivom\ \,is ``tracked"

1996: (in a sense to be elucidated in Section 6)

1997: by (any implementation of) \,\pivo.

1998: \sn

1999: ($ii$) We could only recover continuity of the \,\piv\ 

2000: \,function by giving up either extensionality

2001: (as in ($b$)) or single-valuedness (as in ($c$)).

2002: \sn

2003: ($ii$) Note however that the complete algorithm for inverting matrices

2004: \,is {\it continuous\/} and {\it deterministic\/}

2005: (hence {\it single-valued\/})

2006: and {\it extensional\/},

2007: even though it contains \,\pivo\ \,as an essential 

2008: component!

2009: \endpr

2010: 

2011: \Examplen{1.2.3: \,``Choose" a rational arbitrarily near a real}

2012: \ Define a function

2013: $$

2014: F\: \RR\times\NN\ \to\ \NN

2015: $$

2016: by

2017: $$

2018: F(x,n) \ =\ \tx{``some"}\ k: \ \ds(x,r_k)\,<\,2^{-n}

2019: \tag3

2020: $$

2021: %where \,$r_k)_{k\in\om}$

2022: where (as before) \,$r_0,r_1,r_2,\dots$

2023: \,is some standard enumeration of the rationals.

2024: Note again (as in Example 1.1): 

2025: \sn

2026: ($a$)

2027: \,There is no {\it single-valued, continuous\/} function $F$

2028: satisfying (3).  This is because such a function, being continuous

2029: with discrete range space, would

2030: have to be constant in the first argument.

2031: \sn

2032: ($b$) \,But there {\it is\/} a single-valued computable (and continuous)

2033: function

2034: $$

2035: F_0\: \CS\times\NN\ \to\ \NN

2036: $$

2037: trivially -- just define

2038: $$

2039: F_0(\xi,n) \ = \ \xi_n.

2040: $$

2041: This is, again, {\it non-extensional\/} on \RR.

2042: 

2043: \sn

2044: ($c$)

2045: \,Further, there is a {\it many-valued, continuous, extensional\/}

2046: function satisfying (1):

2047: $$

2048: F_\om\: \RR\times\NN\ \to\ \PPPom(\NN)

2049: $$

2050: where

2051: $$

2052: F_\om(x,n) \ = \ \curly{k\br \ds(x,r_k)<2^{-n}}.

2053: $$

2054: 

2055: \Examplen{1.2.4: \,Finding the root of a function}

2056: \ This example is adapted from \cite{weih:book}.

2057: Consider the function $f_a$ shown in Figure 1,

2058: where $a$ is a parameter which can assume any real value.

2059: 

2060: \midinsert

2061: \epsfxsize = 4.5in

2062: \ce{\epsfbox{fig1.eps}}

2063: \mn

2064: \ce{\sc Figure 1}

2065: \sn

2066: \endinsert

2067: 

2068: \sn

2069: It is defined by

2070: $$

2071: f_a(x) \ = \

2072: \cases

2073: x+a+2 \ift{$x\le-1$}\\

2074: a-x \ift{$-1\le x\le1$}\\

2075: x+a-2 \ift{$1\le x$}.

2076: \endcases

2077: $$

2078: This function has either 1 or 3 roots, depending 

2079: on the size of $a$.

2080: For $a<-1$, \,$f_a$ has a single (large positive) root;

2081: \,for $a>1$, \,$f_a$ has a single (large negative) root;

2082: \,and for

2083: $-1<a<1$, \,$f_a$ has three roots,

2084: two of which become equal when $a = \pm1$.

2085: 

2086: Let $g$ be the (many-valued) function, such that

2087: $g(a)$ gives all the non-repeated roots of $f_a$.

2088: This is shown in Figure 2.

2089: 

2090: 

2091: \midinsert

2092: \epsfxsize = 2in

2093: \ce{\epsfbox{fig2.eps}}

2094: \mn

2095: \ce{\sc Figure 2}

2096: \sn

2097: \endinsert

2098: 

2099: Again, we have the same situation as in the previous examples:

2100: \sn

2101: ($a$) \,We cannot choose a (single) root of $f_a$ continuously

2102: as a function of $a$.

2103: \sn

2104: ($b$) \,However, one can easily choose and compute 

2105: a root of $f_a$ continuously as a function

2106: of a {\it Cauchy sequence representation\/} of $a$,

2107: \ie, non-extensionally in $a$.

2108: \sn

2109: ($c$) Finally, $g(a)$, as a {\it many-valued\/} function of $a$,

2110: is continuous.

2111: (Note that in order to have continuity,

2112: we must exclude the 

2113: repeated roots of $f_a$, at \,$a=\pm1$.)

2114: 

2115: \Remarkn{1.2.5}

2116: Other examples of a similar nature abound, 

2117: and can be treated similarly;  for example,

2118: the problem of finding, for a given real number $x$,

2119: an integer $n>x$.

2120: %The reader can easily provide the details.

2121: \endpr

2122: 

2123: \newpage

2124: 

2125: \shead{1.3}{Solutions for the abstract model}

2126: In the above three examples we have given:

2127: \itemm{($i$)}

2128: a number of single-valued functions 

2129: \,$f\: \RR^n \to  \RR$ \,that we want to compute;

2130: \itemm{($ii$)}

2131: arguments that they are not continuous;

2132: \itemm{($iii$)}

2133: a prima facie case that they cannot be abstractly computed on the

2134: abstract data type $A$ containing \RR\ because they are not continuous;

2135: \itemm{($iv$)}

2136: a prima facie case that they can be computed in the concrete data

2137: type \CS;

2138: \itemm{($v$)}

2139: arguments that they are selection functions for many-valued

2140: functions on \RR\ that are continuous.

2141: 

2142: \n

2143: At the level of {\it concrete models\/} of computation,

2144: there is not really a problem with the

2145: issues raised by the above examples,

2146: since concrete models work only by 

2147: computations on {\it representations\/}

2148: of the reals

2149: (say by Cauchy sequences),

2150: as described fully in Sections 5 and 7.

2151: 

2152: The real problem arises with 

2153: the construction of {\it abstract models\/}

2154: of computation on the reals

2155: which should model the phenomena illustrated by these examples,

2156: and should, moreover, correspond, in some sense,

2157: to the concrete models.

2158: Thus we have the question:

2159: 

2160: {\displaytext

2161: Can such continuous many-valued functions be computed on the abstract

2162: data type $A$ containing \RR\  using new abstract models of computation? If

2163: they can, are the concrete and abstract models then equivalent?

2164: 

2165: }

2166: 

2167: 

2168: The rest of this paper deals with these issues.

2169: We answer the above question more generally, over many-sorted

2170: partial metric algebras $A$.

2171: 

2172: 

2173: The solution presented in this paper is to extend the \Whilex\ \,programming

2174: language over $A$ \cite{tz:hb} with 

2175: a nondeterministic ``countable choice" programming construct,

2176: so that in the rules of program term formation,

2177: $$

2178: \chooses \ \ztt:b

2179: $$

2180: is a new term of type \,\nats,

2181: \,where \,\ztt\ \,is a variable of type \,\nats

2182: \,and $b$ is a term of type \,\bools.

2183: We will revisit the examples after giving the language definition

2184: in Section 3.

2185:  

2186: 

2187: Alternatively, one could use other abstract models; for example,

2188: one can modify the \,\muPRx\

2189: \,function schemes \cite[\S8.1]{tz:hb}

2190: by replacing the constructive least number

2191: (\muu) operator

2192: $$

2193: f(x) \ \sq \ \mu z\in\NN[g(x,z) = \ttt],

2194: $$

2195: where $g$ is a boolean-valued function,

2196: by a nondeterministic

2197: choice operator:

2198: $$

2199: f(x) \ \sq \ \chooses \ z\in\NN[g(x,z) = \ttt].

2200: $$

2201: 

2202: \mn

2203: Given suitable semantics,

2204: these two approaches turn out to be equivalent.

2205: 

2206: In \cite{brattka:thesis} a more elaborate set of recursive schemes

2207: over many-sorted algebras,

2208: with many-valued operations, was presented.

2209: 

2210: \Shead{2}{Topological partial algebras and continuity}

2211: We define some basic notions concerning 

2212: topological and metric many-sorted partial algebras.

2213: We begin with some basic ideas and examples.

2214: \shead{2.1}{Basic algebraic definitions}

2215: A {\it signature} \Sig\ (for a many-sorted partial algebra)

2216: is a pair consisting of

2217: ($i$) a finite set \SortSig\ of {\it sorts},

2218: and ($ii$) a finite set \FuncSig\ of {\it (basic)

2219: function symbols},

2220: each symbol $F$ having a {\it type}  \ $\tuptimes{s}{1}{m}\to s$,

2221: where $s_1 ,\ldots, s_m, s \in \SortSig$;

2222: in that case we write

2223: \ $F:\ \tuptimes{s}{1}{m}\to s$.

2224: \ (The case $m = 0$ corresponds to {\it constant symbols}.)

2225: 

2226: A 

2227: \Sig-{\it product type\/}

2228: has the form

2229: \ $u = \tuptimes{s}{1}{m}$ \ ($m\ge0$),

2230: where \,\tup{s}{1}{m} \,are \Sig-sorts.

2231: We use the notation

2232: \ $u,v,w,\dots$ \ for \Sig-product types.

2233: 

2234: A partial \Sig-{\it algebra} $A$ has, for each sort $s$ of \Sig,

2235: a non-empty {\it carrier set} \,\As\

2236: \,of sort $s$,

2237: and for each \Sig-function symbol

2238: \ $F:\utos$,

2239: \ a partial function

2240: \ $\FA : \Au \pto \As$,

2241: \,where, for the \Sig-product type \ $u  = \tuptimes{s}{1}{m}$,

2242: \ we write

2243: \ $\Au \ \eqdf \ A_{s_1} \times\dots\times A_{s_m}$.

2244: (The notation \,$f: X\pto Y$ \,refers in general to

2245: a partial function from $X$ to $Y$.)

2246: 

2247: The algebra $A$ is {\it total} if 

2248: \,\FA\ is total for each \Sig-function symbol $F$.

2249: Without such a totality assumption,

2250: $A$ is called {\it partial}.

2251: 

2252: In this paper we deal mainly with partial algebras.

2253: {\it The default assumption is that ``algebra"

2254: refers to partial algebra.}

2255: We will, nevertheless, for the sake of emphasis,

2256: often speak explicitly of ``partial algebras".

2257: 

2258: Given an algebra $A$, we write \,$\Sig(A)$ \,for its signature.

2259: %Also, we sometimes use a function symbol `$F$'

2260: %to stand for its interpretation \FA\

2261: %in $A$.

2262: 

2263: 

2264: \Examplesn{2.1.1}

2265: The following algebras will be used repeatedly as examples 

2266: in this paper.

2267: All but one are total.

2268: \sn

2269: ($a$)

2270: \,The algebra of {\it booleans\/}

2271: has the carrier \ $\BB = \{\ttt,\,\fff\}$

2272: \ of sort \,\bools. 

2273: The signature $\Sig(\BBB)$

2274: and algebra \BBB\ respectively can be displayed as follows:

2275: $$

2276: \boxed{

2277: \matrix \format\l&\quad\l\\

2278: \signatures &\Sig(\BBB)\\

2279: \sortss &\bools\\

2280: \functionss&\trues, \falses:\ \ \to\bools,\\

2281: &\ands, \ors:\bools^2\to\bools\\

2282: &\nots: \bools\to\bools\\

2283: \ends&

2284: \endmatrix

2285: }

2286: \quad

2287: \tx{and}

2288: \quad

2289: \boxed{

2290: \matrix \format\l&\quad\l\\

2291: \algebras &\BBB\\

2292: \carrierss &\BB\\

2293: \functionss&\ttt, \fff:\ \ \to\BB,\\

2294: &\ands^\BBB, \ors^\BBB:\BB^2\to\BB\\

2295: &\nots^\BBB: \BB\to\BB\\

2296: \ends&

2297: \endmatrix

2298: }

2299: $$

2300: Usually the signature can essentially be inferred from the algebra;

2301: indeed we will not define the signature where no confusion will arise.

2302: Further, for notational simplicity, we will not always distinguish between

2303: function names in the signature (\trues, etc.)

2304: and their intended interpretations ($\trues^\BBB = \ttt$, etc.)

2305: \sn

2306: ($b$) \,The algebra \,\NNNo\ \,of naturals

2307: has a carrier \,\NN\ \,of sort \,\nats,

2308: \,together with the zero constant and successor function:

2309: $$

2310: \boxed{

2311: \matrix \format\l&\quad\l\\

2312: \algebras &\NNNo \\

2313: \carrierss &\NN\\

2314: \functionss&0:\ \ \to\NN,\\

2315: &\Ss:\NN\to\NN\\

2316: \ends&

2317: \endmatrix

2318: }

2319: $$

2320: 

2321: ($c$) \,The ring \,\RRRo\ \,of reals

2322: has a carrier \RR\ of sort \reals:

2323: $$

2324: \boxed{

2325: \matrix \format\l&\quad\l\\

2326: \algebras &\RRRo \\

2327: \carrierss &\RR\\

2328: \functionss&0,1:\ \ \to\RR,\\

2329: &+,\times:\RR^2\to\RR,\\

2330: &-:\RR\to\RR\\

2331: \ends&

2332: \endmatrix

2333: }

2334: $$

2335: \endpr

2336: 

2337: 

2338: 

2339: ($d$) The field \,\RRRi\ \,of reals is

2340: formed by adding the multiplicative inverse to the ring \RRRo:

2341: $$

2342: \boxed{

2343: \matrix \format\l&\quad\l\\

2344: \algebras &\RRRi \\

2345: \imports &\RRRo\\

2346: \functionss &\invR:\RR\to\RR\\

2347: \ends&

2348: \endmatrix

2349: }

2350: $$

2351: where \,

2352: $$

2353: \invR(x) \ = \ 

2354: \cases 

2355: 1/x \ift{$x\ne 0$}\\

2356: \ua \ow.

2357: \endcases.

2358: $$

2359: This is an example of a  partial algebra.

2360: More examples of partial algebras wil be given later.

2361: 

2362: Throughout this work we make the following assumption about 

2363: the signatures \Sig.

2364: 

2365: \Assumptionn{2.1.2 \ (Instantiation Assumption)}

2366: {\sl For every sort $s$ of \Sig, there is a closed term of that sort,

2367: called the default term \,\delbs\ \,of that sort.}

2368: \endpr

2369: 

2370: This guarantees the presence of {\it default values} \,\delbsA\ \, 

2371: in a \Sig-algebra $A$ at all sorts $s$, 

2372: and {\it default tuples} \,\delbuA\ \,at all product types $u$.

2373: 

2374: \Defn{2.1.3 \,(Expansions and reducts)}

2375: Let \Sig\ and \Sigp\ be signatures with $\Sig \subset \Sig'$.

2376: \sn

2377: \ ($a$) If $A'$ is a \Sigp-algebra, \,then the \Sig-{\it reduct of} $A'$,

2378: \ $A' \redSig$,

2379: \ is the algebra

2380: of signature \Sig, consisting of the carriers of $A'$

2381: named by the sorts of \Sig\ and equipped with the functions

2382: of $A'$ named by the function symbols of \Sig.

2383: \sn

2384: \ ($b$) If \,$A$ is a \,\Sig-algebra and $A'$ is a \Sigp-algebra,

2385: \,then $A'$ is a {\it \Sigp-expansion} of $A$ iff $A$ is the

2386: \Sig-reduct of $A'$.

2387: \endpr

2388: 

2389: \shead{2.2}{Adding booleans: \,Standard signatures and algebras}

2390: The algebra \BBB\ of booleans (Example 2.1.1($a$))

2391: plays an essential role in computation, as we will see.

2392: This motivates the following definition.

2393: 

2394: \Defn{2.2.1 \ (Standard signature)}

2395: A signature \Sig\ is {\it standard} if

2396: \itemitem{($i$)} 

2397: it contains the signature of booleans, \ie,

2398: \,$\Sig(\BBB) \ \sseq \Sig$, \ and

2399: \itemitem{($ii$)}

2400: The function symbols of \Sig\ include

2401: a {\it conditional\/}

2402: $$

2403: \ifs_s :\bools\times s^2 \to s

2404: $$

2405: for all sorts $s$ of \Sig\ other than \bools.

2406: \endpr

2407: 

2408: Now given a standard signature \Sig,

2409: a sort of \Sig\

2410: is called an {\it equality sort\/}

2411: if \Sig\ includes an

2412: {\it equality  operator}

2413: $$

2414: \eqs_s: s^2 \to \bools.

2415: $$

2416: 

2417: \Defn{2.2.2 \ (Standard algebra)}

2418: Given a standard signature \Sig,

2419: a \Sig-algebra $A$ is a {\it standard} if

2420: \itemm{($i$)}

2421: it is an expansion of \BBB;

2422: \itemm{($ii$)}

2423: the conditional operator on each sort $s$

2424: has its standard interpretation in $A$;

2425: \ie, \,for \,$b\in\BB$ \,and \,$x,y \in \As$,

2426: $$

2427: \ifs^A_s(b,x,y) \ = \

2428: \cases

2429:    x \tif{$b = \ttt$}\\

2430:    y \tif{$b = \fff$};

2431: \endcases

2432: $$

2433: \itemm{($iii$)}

2434: the equality operator

2435: \,$\eqs_s$ \,is interpreted as a {\it partial identity\/} on each

2436: equality sort $s$, \ie, for any two elements of \As,

2437: if they are identical, then

2438: the operator at these arguments returns 

2439: \,\ttt\ \,if it returns anything; and

2440: if they are not identical,

2441: it returns \,\fff\

2442: \,if anything.

2443: More specifically, there are three possible cases.

2444: First, the case

2445: $$

2446: \eqs^A_s(x,y) \ = \ 

2447: \cases

2448:   \ttt \ift{$x=y$}\\

2449:   \fff \ow,

2450: \endcases

2451: $$

2452: \ie, total equality, represents the situation

2453: that equality is ``decidable" or ``computable"

2454: at sort $s$,

2455: for example, when $s=\nats$.

2456: Second, the case

2457: $$

2458: \eqs^A_s(x,y) \ = \ 

2459: \cases

2460:   \ttt \ift{$x=y$}\\

2461:   \ua \ow

2462: \endcases

2463: $$

2464: represents typically the situation that

2465: that equality is ``semidecidable".

2466: An example is given by the initial {\it term algebra\/} 

2467: of an r.e. equational theory.

2468: Third, the case

2469: $$

2470: \eqs^A_s(x,y) \ = \ 

2471: \cases

2472:   \ua \ift{$x=y$}\\

2473:   \fff \ow,

2474: \endcases

2475: $$

2476: represents typically the situation that

2477: that equality is ``co-semidecidable".

2478: Examples are given by the data types

2479: of {\it streams\/} and {\it real numbers\/},

2480: as mentioned in 1.1.3; see Example 2.2.4($c$) below.

2481: \endpr

2482: 

2483: \smskipn

2484: Note that any many-sorted signature \Sig\

2485: can be {\it standardised} to a signature \,\SigBBB\

2486: \,by adjoining the sort \,\bools\ \,together

2487: with the standard boolean operations;

2488: and, correspondingly, any algebra $A$ can be standardised

2489: to an algebra \ABBB\

2490: \,by adjoining the algebra \BBB\

2491: as well as the conditional and equality operators.

2492: 

2493: 

2494: \Examplesn{2.2.4 \,(Standard algebras)}

2495: \sn

2496: ($a$) The simplest standard algebra is the algebra \BBB\ of the booleans

2497: (Example 2.1.1($a$)).

2498: \smskipn

2499: ($b$) A {\it standard algebra of naturals\/} \,\NNN\

2500: \,is formed by standardising the algebra \,\NNNo\ 

2501: \linebreak

2502: (Example 2.1.1($b$)),

2503: with (total) equality and order operations on \NN:

2504: $$

2505: \boxed{

2506: \matrix \format\l&\quad\l\\

2507: \algebras &\NNN \\

2508: \imports &\NNNo, \,\BBB\\

2509: \functionss

2510: &\ifnatN:\BB\times\NN^2\to\NN,\\

2511: &\eqnatN,\,\lsnatN:\NN^2\to\BB\\

2512: \ends&

2513: \endmatrix

2514: }

2515: $$

2516: ($c$) {\it A standard partial algebra \RRR\ on the reals\/} 

2517: is formed similarly by standardising the field \RRRi\ 

2518: (Example 2.1.1($d$)),

2519: %with \reals\ as a partial equality sort,

2520: with partial equality and order operations on \RR:

2521: $$

2522: \boxed{

2523: \matrix \format\l&\quad\l\\

2524: \algebras &\RRR \\

2525: \imports &\RRRi, \,\BBB\\

2526: \functionss

2527: &\ifrealR:\BB\times\RR^2\pto\RR,\\

2528: &\eqrealR,\,\lsrealR:\RR^2\pto\BB\\

2529: \ends&

2530: \endmatrix

2531: }

2532: $$

2533: where 

2534: \TOL

2535: $$

2536: \eqrealR(x,y) \ = \

2537: \cases

2538: \ua \tif{$x=y$}\\

2539: \fff \tif{$x\ne y$}.

2540: \endcases

2541: $$

2542: $$

2543: \lsrealR(x,y) \ = \

2544: \cases

2545: \ttt \tif{$x<y$}\\

2546: \fff \tif{$x>y$}\\

2547: \ua \tif{$x=y$},

2548: \endcases

2549: \tag"and"

2550: $$

2551: \TOR

2552: 

2553: \Discussionn{2.2.5 \,(Semicomputability and co-semicomputability)}

2554: The significance of the partial equality 

2555: and order operations in Example ($c$) above, 

2556: in connection with computability

2557: and continuity, has been touched on in 1.1.3.

2558: The {\it continuity\/} of partial functions will be discussed in \S2.5

2559: (and see in particular Example 2.5.3($b$)).

2560: Regarding {\it computability\/},

2561: these definitions are intended to 

2562: reflect, or capture the intuition of, 

2563: the ``{\it semicomputability\/}" of order and the

2564: ``{\it co-semicomputability\/}" of equality

2565: on (a concrete model of) the reals.

2566: For given two reals $x$ and $y$, 

2567: represented (say) by their infinite decimal expansions,

2568: suppose their decimal digits are being read systematically,

2569: the $n$-th digit of both at step $n$.

2570: Then if $x\ne y$ or $x<y$,

2571: this will become apparent after finitely many steps,

2572: but no finite number of steps can confirm that $x=y$.

2573: \endpr

2574: 

2575: Throughout this paper, we will assume the following,

2576: unless specifically noted to the contrary.

2577: 

2578: \Assumptionn{2.2.6 \,(Standardness Assumption)}

2579: The signature \Sig\ and \Sig-algebra $A$ are standard.

2580: \endpr

2581: 

2582: 

2583: \shead{2.3}{Adding counters: \ N-standard signatures and algebras}

2584: The standard algebra \NNN\ of naturals (Example 2.2.4($b$))

2585: plays, like \BBB, an essential role in computation.

2586: This motivates the following definition.

2587: 

2588: \Defn{2.3.1 \ (N-standard signature)}

2589: A signature \Sig\ is {\it N-standard\/} if

2590: \itemm{($i$)} 

2591: it is standard, \,and

2592: \itemm{($ii$)} 

2593: it contains the standard signature of naturals (Example 2.2.4($b$)), \ie,

2594: \,$\Sig(\NNN) \ \sseq \Sig$.

2595: \endpr

2596: 

2597: \Defn{2.3.2 \ (N-standard algebra)}

2598: Given an N-standard signature \Sig,

2599: a corresponding \Sig-algebra $A$ is {\it N-standard} if 

2600: it is an expansion of \NNN.

2601: \endpr

2602: 

2603: Note that any standard signature \Sig\ can be

2604: N-{\it standardised} to a signature \SigN\ 

2605: by adjoining the sort \nats\ and the operations 0, \Ss,

2606: \eqnat, \lsnat\ and \ifnat.

2607: Correspondingly,

2608: any standard \Sig-algebra $A$ can be N-{\it standardised} 

2609: to an algebra \AN\ 

2610: by adjoining the carrier \NN\

2611: together with the corresponding standard functions.

2612: 

2613: 

2614: \Examplesn{2.3.3 \,(N-standard algebras)}

2615: \itemm{($a$)\,}

2616: The simplest N-standard algebra is the algebra \NNN\ 

2617: (Example 2.2.4($b$)).

2618: \itemm{($b$)\,}

2619: We can N-standardise the standard real algebra

2620: \,\RRR\ (Example 2.2.4($c$))

2621: \,to form the algebra \,\RRRN.

2622: \endpr

2623: 

2624: 

2625: \shead{2.4}{Adding arrays: \ Algebras \,\Ax\ of signature \Sigx}

2626: The significance of arrays for computation

2627: is that they provide

2628: {\it finite but unbounded memory}.

2629: 

2630: Given a standard signature \Sig,

2631: and standard \Sig-algebra $A$, we expand \Sig\ and $A$

2632: in two stages:

2633: 

2634: \itemm{($1^\circ$)}

2635: N-standardise these to form \SigN\ and \AN, as in \S2.3.

2636: 

2637: \itemm{($2^\circ$)}

2638: Define, for each sort $s$ of \Sig, the carrier \Axs\

2639: to be the set of {\it finite sequences\/} or {\it arrays\/} \ax\

2640: over \As, of ``starred sort" \sx.

2641: 

2642: The resulting algebras \Ax\ have signature \Sigx,

2643: which extends \SigN\

2644: by including,

2645: for each sort $s$ of \Sig, the new starred sorts \sx,

2646: and certain new function symbols.

2647: Details are given in \cite[\S2.7]{tz:hb}

2648: and (an equivalent but simpler version) in \cite[\S2.4]{tz:top}.

2649: 

2650: The reason for introducing starred sorts

2651: is the lack of effective coding

2652: of finite sequences within abstract algebras in general.

2653: 

2654: 

2655: 

2656: 

2657: \shead{2.5}{Topological partial algebras}

2658: We now add topologies to our partial algebras,

2659: with the requirement of continuity 

2660: for the basic partial functions.

2661: Background information on topology can be obtained

2662: from any standard text, \eg,

2663: \cite{kelley:book,dugundji:book,engelking:book}.

2664: \Defn{2.5.1}

2665: Given two topological spaces $X$ and $Y$,

2666: a partial function 

2667: \nl

2668: $f:X\pto Y$ 

2669: \,is {\it continuous} if for every open \,$V\sseq Y$,

2670: $$

2671: f^{-1}[V] \ \eqdf \ \curly{x\in X \mid x\in\dom{f} \ \tx{and} \ f(x)\in Y}

2672: $$

2673: is open in $X$.

2674: \endpr

2675: \Defn{2.5.2}

2676: ($a$)

2677: \,A {\it topological partial \Sig-algebra} is a partial \Sig-algebra

2678: with topologies on the carriers

2679: such that each of the basic \Sig-functions 

2680: is continuous.

2681: \smskipn($b$)

2682: \,An (N-){\it standard topological partial algebra} 

2683: is a topological partial algebra which is also an (N-)standard partial algebra,

2684: such that the carriers \BB\ (and \NN) have the discrete topology.  

2685: \endpr

2686: 

2687: \Examplesn{2.5.3}

2688: ($a$) ({\it Discrete algebras.\/})

2689: The standard algebras \,\BBB\ \,and \,\NNN\ \,of booleans and naturals

2690: respectively \,(\S\S2.1, 2.2) 

2691: \,are topological (total) algebras under the discrete topology.

2692: All functions on them are trivially continuous, since the 

2693: carriers are discrete.

2694: \sn

2695: ($b$) ({\it Partial real algebra.\/})

2696: An important standard

2697: topological partial algebra for our purpose is the real algebra \RRR\

2698: (Example 2.2.4($c$)), or its N-standardised version \RRRN\ 

2699: (Example 2.3.3($b$)),

2700: in which 

2701: \RR\ has its usual topology, and \BB\ and \NN\ the discrete topology.

2702: Recall our earlier discussion (1.1.3) of partiality of tests in connection

2703: with continuity, and

2704: note that the partial operations 

2705: \,\eqrealR\ \,and \,\lsrealR\ 

2706: \,are continuous, in the sense of Definition 2.5.1.

2707: \sn

2708: ($c$) ({\it Partial interval algebras.\/})

2709: \ Another useful class of standard topological partial algebras are 

2710: of the form

2711: $$

2712: \boxed{

2713: \matrix \format\l&\quad\l\\

2714: \algebras &\III \\

2715: \imports &\RRR\\

2716: \carrierss & I\\

2717: \functionss&\iI:I\to\RR,\\

2718: &F_1:I^{m_1}\to I,\\

2719: &\quad\dots\\

2720: &F_k:I^{m_k}\to I\\

2721: \ends&

2722: \endmatrix

2723: }

2724: $$

2725: where $I$ is the closed interval \,$[0,1]$

2726: \,(with its usual topology),

2727: \ \iI\ is the embedding of $I$ into \RR,

2728: \,and

2729: \,$F_i: I^{m_i}\to I$

2730: \,are continuous partial functions.

2731: These are called {\it (partial) interval algebras on} $I$.

2732: There are also N-standard versions:

2733: $$

2734: \boxed{

2735: \matrix \format\l&\quad\l\\

2736: \algebras &\IIIN \\

2737: \imports &\RRRN\\

2738: \carrierss & I\\

2739: \functionss&\iI:I\to\RR,\\

2740: &\dots\\

2741: \ends&

2742: \endmatrix

2743: }

2744: $$

2745: \endpr

2746: 

2747: \mn($d$) ({\it N-standard total real algebra\/}.)

2748: \,The algebra \RRRtN\ is

2749: (``$t$" for ``total topological"), defined by

2750: $$

2751: \boxed{

2752: \matrix \format\l&\quad\l\\

2753: \algebras &\RRRtN \\

2754: \imports &\RRRo, \,\NNN,\,\BBB\\

2755: \functionss&\ifrealR:\BB\times\RR^2\to\RR,\\

2756: &\divNR:\RR\times\NN\to\RR,\\

2757: \ends&

2758: \endmatrix

2759: }

2760: $$

2761: Here \RRRo\ is the ring of reals (\S2.1.1($c$)),

2762: \NNN\ is the standard algebra of naturals (2.2.4($b$)),

2763: and \divN\ is division of reals by naturals.

2764: 

2765: Note that \RRRtN\ does not contain (total) boolean-valued 

2766: functions $<$ or $=$ on the reals,

2767: since they are not continuous

2768: (\cf\ the partial functions \eqreal\ and \lsreal\ of \RRR).

2769: It is therefore not an expansion of \RRR.

2770: 

2771: %The reason that \NN, and the function \divN,

2772: %are included in this total algebra

2773: %(unlike the partial algebras \RRRp\ and \IIIp\ of \S3.2)

2774: %is because of its applicability 

2775: %in the theory of approximable computability in Section 9.

2776: 

2777: \Defn{2.5.4 \ (Extensions of topology to \AN\ and \Ax)}

2778: Corresponding to the various algebraic expansions of $A$

2779: detailed in \S\S2.3 and 2.4, there are induced topological expansions.

2780: \sn($a$)

2781: The topological partial N-standard algebra \AN, of signature \SigN,

2782: is constructed from $A$ by giving the new carrier \NN\ the discrete topology.

2783: \smskipn($b$)

2784: The topological partial array algebra \Ax, of signature \Sigx,

2785: is constructed from \AN\ as follows.

2786: Viewing the elements of \Axs\ as (essentially)

2787: arrays of elements of \As\ of finite length,

2788: we can give

2789: \Axs\ the {\it disjoint union} topology

2790: of the sets \,$(\As)^n$ \,of arrays of length $n$,

2791: for all $n\ge0$,

2792: where each set $(\As)^n$ is given the {\it product topology\/}

2793: of the sets \As.

2794: 

2795: The topology on \Ax\ can also be described as follows.

2796: The {\it basic open sets\/} in \Axs\ are of the form

2797: $$

2798: \curly{\ax \in \Axs \mid \Lgths(\ax)> i_n \ \ \tx{and} 

2799: \ \ \ax[i_1]\in U_1,\ \dots,\ \ax[i_n]\in U_n}

2800: $$

2801: for some $n>0$, \,$i_1<\dots<i_n$ \,and

2802: open sets \,$U_1,\dots,U_n\sseq\As$.

2803: 

2804: It is easy to check that \Ax\ is indeed a topological algebra,

2805: \ie, all the new functions of \Ax\ are continuous.

2806: 

2807: \newpage

2808: 

2809: \shead{2.6}{Metric algebra}

2810: A particular type of topological algebra is a {\it metric 

2811: partial algebra}.

2812: This is

2813: a many-sorted standard partial algebra

2814: with an associated metric:

2815: $$

2816: \boxed{

2817: \matrix \format\l&\quad\l\\

2818: \algebras &A\\

2819: \imports &\BBB, \,\RRR\\

2820: \carrierss & \tup{A}1r,\\

2821: \functionss&F^A_1:A^{u_1}\to A_{s_1},\\

2822: &\quad\dots\\

2823: &F^A_k:A^{u_k}\to A_{s_k},\\

2824: &\ds^A_1:A_1^2\to\RR,\\

2825: &\quad\dots\\

2826: &\ds^A_r:A_r^2\to\RR\\

2827: \ends&

2828: \endmatrix

2829: }

2830: $$

2831: where \,\BBB\ \ \,and \,\RRR\

2832: \,are respectively the algebras of booleans and reals 

2833: (Examples 2.1.1($a$), 2.2.4($c$)),

2834: the carriers \,\tup{A}1r\ \,are metric spaces with metrics

2835: \,$\ds^A_1,\dots,\ds^A_r$ \,respectively,

2836: \,\tup{F}1k \,are the \Sig-function symbols

2837: other than \,\tup{\ds}1k,

2838: \,and the (partial) functions

2839: $F^A_i$ are all continuous with respect to these metrics,

2840: where continuity of a partial function is understood as in

2841: Definition 2.5.1.

2842: 

2843: Clearly,

2844: metric algebras can be viewed as special cases of 

2845: {\it topological partial algebras\/}.

2846: 

2847: 

2848: Note that 

2849: the carrier \BB\ (as well as \NN, if present)

2850: has the {\it discrete metric},

2851: defined by

2852: $$

2853: \ds(x,y) \ = \ 

2854: \cases

2855: 0 \tif{$x=y$}\\

2856: 1 \tif{$x\ne y$},

2857: \endcases

2858: $$

2859: which induces the discrete topology.

2860: 

2861: We will often speak of a ``metric algebra $A$",

2862: without stating the metric explicitly.

2863: 

2864: \Examplen{2.6.1}

2865: The partial and total

2866: real algebras \RRR, \RRRN\ and \RRRtN\ (Examples 2.5.3)

2867: can be recast as metric algebras in an obvious way.

2868: \endpr

2869: 

2870: \Remarkn{2.6.2 \ (Extension of metric to \Ax)}

2871: A metric algebra $A$ can be expanded to a metric

2872: algebra \Ax\ of arrays over $A$.

2873: Namely, given a metric \,\dss\ \,on \As,

2874: we define a (bounded) metric \,\dsxs\ \,on \Axs\ as follows:

2875: \,for \,$\ax = (\tup{a}1k), \,\bx=(\tup{b}1l)\in \Axs$:

2876: $$

2877: \dsxs(\ax,\,\bx)\ =

2878: \cases

2879: 1 \ift{$k\ne l$}\\

2880: \min\big(1,\ \max_{i=0}^{k-1}\dss(\ax[i],\bx[i])\big) \ow

2881: \endcases 

2882: $$

2883: This gives the same topology on \Ax\

2884: as that induced by the topology on $A$ (Definition 2.5.4)

2885: \cite{engelking:book}.

2886: 

2887: \Remarkn{2.6.3 \ (Product metric on $A$)}

2888: If $A$ is a \Sig-metric algebra, then

2889: for each \Sig-product sort \,$u=\tuptimes{s}{1}{m}$,

2890: \,we can define a metric  

2891: \,\dsu\ \,on \Au\ by

2892: $$

2893: d_u((x_1,\dots,x_m),(y_1,\dots,y_m)) \ = 

2894: \ \max_{i=1}^m\bigl(d_{s_i}(x_i,y_i)\bigr)

2895: $$

2896: or more generally, by the $\ell_p$ metric

2897: $$

2898: d_u((x_1,\dots,x_m),(y_1,\dots,y_m)) \ = 

2899: \ \bigl(\sum_{i=1}^m(d_{s_i}(x_i,y_i))^p\bigr)^{1/p}

2900: \tag{($1\le p \le\infty$)}

2901: $$

2902: where $p=\infty$ corresponds to the ``max" metric.

2903: This induces the product topology on \Au.

2904: \endpr

2905: 

2906: 

2907: 

2908: \shead{2.7}{W-continuity: \ Another notion of continuity of partial functions}

2909: Recall our definition (2.5.1) of continuity of partial functions:

2910: \,$f:X\pto Y$ 

2911: \,is continuous if for every open \,$V\sseq Y$,

2912: \ $f^{-1}[V]$ \ is open in $X$.

2913: 

2914: This is not the only reasonable definition.

2915: Another definition, used in \cite{weih:book} and 

2916: \cite{brattka96,brattka:thesis}

2917: (henceforth ``W-continuity"),

2918: amounts to saying that $f$

2919: is continuous iff its restriction to its domain

2920: $$

2921: f\rest \dom{f}:\dom{f}\to Y

2922: $$

2923: is continuous (as a total function),

2924: where \,\dom{f} \,has the topology as a subspace of $A$;

2925: or, equivalently, iff for every open \,$V\sseq Y$,

2926: \ $f^{-1}[V]$ \ is open in \,\dom{f}.

2927: 

2928: The following is easily checked:

2929: 

2930: \Propn{2.7.1}

2931: $f$ is continuous \ \ifff\ \ $f$ is W-continuous \,and \,\dom{f} is open.

2932: \endpr

2933: 

2934: \Remarkn{2.7.2}

2935: It is instructive to express these two notions of continuity

2936: in terms of metric spaces. Suppose 

2937: \,$f:X\pto Y$ \,where $X$ and $Y$ are metric spaces.  Then

2938: \itemm{($a$)}

2939: $f$ is continuous iff

2940: $$

2941: \all a\in \dom{f} \,\all\eps>0 \,\ex\del>0 \,\all x\in \Bb(a,\del)

2942: \,\bigl(x\in \dom{f}\con f(x)\in \Bb(f(a),\eps)\bigr).

2943: $$

2944: \itemm{($b$)}

2945: $f$ is W-continuous iff

2946: $$

2947: \all a\in \dom{f} \,\all\eps>0 \,\ex\del>0 \,\all x\in \Bb(a,\del)

2948: \,\bigl(x\in \dom{f}\imp f(x)\in \Bb(f(a),\eps)\bigr).

2949: $$

2950: %which can be rewritten:

2951: %$$

2952: %\all a\in \dom{f} \,\all\eps>0 \,\ex\del>0 \,\all x\in \dom{f}

2953: %\,\bigl(\ds(x,a)<\del \imp \ds(f(x),f(a))<\eps\bigr).

2954: %$$

2955: %\endpr

2956: 

2957: \n

2958: Here \,$\Bb(a,\del)$

2959: \,is the open ball with centre $a$ and radius \del.

2960: \endpr

2961: 

2962: \Examplen{2.7.3}

2963: Consider the partial function \, $f\: \RR\ \pto \ \RR$

2964: \,defined by

2965: $$

2966: f(x) \ = \ 

2967: \cases

2968:   0 \ift{$x$ is an integer}\\

2969:   \ua \ow.

2970: \endcases

2971: $$

2972: Then $f$ is W-continuous, but not continuous.

2973: \endpr

2974: 

2975: 

2976: \newpage

2977: 

2978: \Shead3{`\Whilebig' programming with countable choice}

2979: The programming language $\WhileCC = \WhileCCSig$

2980: \,is an extension of \,\WhileSig\ \cite[\S2.1, 2.13]{tz:hb}

2981: \,with an extra \,\qchooses\ \,rule of term formation.

2982: We give the complete definition 

2983: of its syntax and semantics, using the 

2984: {\it algebraic operational semantics\/} of \cite{tz:hb}.

2985: 

2986: Assume \Sig\ is an N-standard signature,

2987: and $A$ is an N-standard \Sig-algebra.

2988: 

2989: \shead{3.1}{Syntax of \WhileCCSig}

2990: We define four syntactic classes:

2991: {\it variables}, {\it terms}, {\it statements} and

2992: {\it procedures}.

2993: 

2994: \smskipn($a$)

2995: $\Var = \VarSig$

2996: \,is the class of \Sig-{\it program variables\/}, and

2997: for each \Sig-sort $s$, 

2998: \,\Vars\

2999: \,is the class of program variables of sort $s$:

3000: \,$\att^s,\btt^s,\dots,\xtt^s,\ytt^s\dots$.

3001: 

3002: \smskipn($b$)

3003: $\PTerm \,= \PTermSig$

3004: \ is the class of \Sig-{\it program terms\/} \ $t,\dots$,

3005: \ and for each \Sig-sort $s$,

3006: \,\PTerms\

3007: \,is the class of program terms of sort $s$.

3008: These are generated by the rules

3009: $$

3010: t \ ::= \ \xtt^s \mid F(\tup{t}{1}{n}) \mid \chooses \ \ztt^\natss:b

3011: $$

3012: where \,$s,\tup{s}{1}{n}$ \,are \Sig-sorts,

3013: \ $F: \tuptimes{s}{1}{n}\to s$ \ is a \Sig-function symbol,

3014: \,$t_i\in \PTerm_{s_i}$ \,for \,$i=1,\dots,n$ \,($n\ge0$),

3015: and $b$ is  a boolean term, \ie, a term of sort \,\bools.

3016: 

3017: Think of \,\qchooses\ \,as a generalisation of

3018: the {\it constructive least number operator\/}

3019: \,$\leasts\ \ztt:b$

3020: \,which has the value $k$ in case 

3021: \,$b[\ztt/k]$ \,is true

3022: and \,$b[\ztt/i]$ \,is defined and false

3023: for all $i<k$, \,and is undefined in case no such $k$ exists.

3024: 

3025: Here \,`\choosezb' \,selects {\it some\/} value $k$ such that

3026: \,$b[\ztt/k]$

3027: \,is true, if any such $k$ exists (and is undefined otherwise).

3028: Which value is selected depends, in general, on the {\it implementation\/}

3029: of the algebra	$A$.

3030: In our abstract semantics,

3031: we will give the meaning as the set of

3032: {\it all possible $k$'s\/}

3033: (hence ``countable choice").

3034: Any concrete model will select a particular $k$, according to the 

3035: implementation.

3036: 

3037: Note that

3038: the program terms extend the algebraic terms 

3039: (\ie, the terms over the signature \Sig)

3040: by

3041: including in their construction the \,\qchooses\ \,operator,

3042: which is not an operation of \Sig.

3043: An alternative formulation would be to have \,\qchooses\,

3044: \,{\it not\/} as part of the term construction,

3045: but rather as a new atomic program statement: 

3046: \,`$\chooses\ \ztt:b$'.

3047: We prefer the present treatment, as it leads to the construction

3048: of {\it many-valued term semantics\/} (as we will see), 

3049: which is interesting in itself,

3050: and which we would get anyway if we were to extend our

3051: syntax to include many-valued function procedure calls

3052: in our term construction.

3053: 

3054: We write \,$t:s$ \,to indicate that \,$t \in \PTerms$,

3055: \,and for \,$u = \tuptimes{s}{1}{m}$,

3056: \,we write \,$t:u$

3057: \,to indicate

3058: that $t$ is a $u$-{\it tuple} of program terms,

3059: \ie, a tuple of program terms of sorts \ \tup{s}{1}{m}.

3060: 

3061: We also use the notation \,$b,\dots$

3062: \,for boolean terms.

3063: 

3064: \sn($c$)

3065: $\AtSt = \AtStSig$

3066: \ is the class of {\it atomic statements\/} $\Sat,\dots$ \, defined by

3067: $$

3068: \Sat \ ::= \ \skips \br \divs \br \xtt:= t 

3069: $$

3070: where 

3071: \,`\divs' \,stands for ``divergence" (non-teremination),

3072: and 

3073: \,$\xtt:= t$

3074: \,is a

3075: {\it concurrent assignment},

3076: \ where for some product type $u$,

3077: \ $t:u$

3078: \,and \xtt\ is a $u$-tuple of {\it distinct\/} variables.

3079: 

3080: \smskipn($d$)

3081: $\Stmt = \StmtSig$

3082: \,is the class of statements \,$S,\dots$,

3083: \,generated by the rules

3084: $$

3085: S \ ::= \ \Sat \br \ S_1;S_2 \br

3086: \ifs\ b \ \thens\ S_1 \ \elses\ S_2 \ \fis

3087: \br \whiles\ b \ \dos\ S \ \ods

3088: $$

3089: 

3090: 

3091: \smskipn($e$)

3092: $\Proc = \ProcSig$

3093: \ is the class of function procedures

3094: \ $P,Q,\dots$.

3095: \ These have the form

3096: $$

3097: P \ \ident \ \funcs \ \ins \ \att\ \outs \ \btt\ \auxs \ \ctt\

3098: \begins \ S\ \ends

3099: $$

3100: where \att, \btt\ and \ctt\ are lists of {\it input variables},

3101: {\it output variables} and {\it auxiliary} (or

3102: {\it local) variables} respectively,

3103: and $S$ is the {\it body}.

3104: Further, we stipulate:

3105: \bull

3106: \att, \btt\ and \ctt\ \,each consist of distinct variables,

3107: and they are pairwise disjoint,

3108: \bull

3109: all variables occurring in $S$ must be

3110: among \,\att, \btt\ or \ctt,

3111: \bull

3112: the {\it input variables\/} \att\ must not occur on

3113: the lhs of assignments in $S$,

3114: \bull

3115: {\it initialisation condition}: \,$S$ has the form \,$\Sinit;S'$,

3116: \,where \Sinit\ is a concurrent assignment

3117: which initialises all the {\it output\/} and {\it auxiliary variables\/},

3118: \ie, assigns to each variable in \,\btt\ \,and \,\ctt\  

3119: \,the default term (2.1.2)

3120: of the same sort.

3121: 

3122: If \,$\att:u$ \,and \,$\btt:v$, then

3123: $P$ is said to have {\it type} \utov,

3124: \ written \ $P: \utov$.

3125: Its {\it input type} is $u$.

3126: 

3127: 

3128: \shead{3.2}{Algebraic operational semantics of \WhileCC}

3129: We will interpret programs as countably-many-valued

3130: state transformations,

3131: and function procedures as

3132: countably-many-valued functions on $A$.

3133: Our approach

3134: follows the {\it algebraic operational semantics\/}

3135: of \cite[\S\S3.4]{tz:hb}.

3136: First we need some definitions and notation

3137: for many-valued functions.

3138: 

3139: \Notationn{3.2.1}

3140: \itemm{($a$)}

3141: $\PPPom(X)$ \,is the set of all countable subsets of a set $X$,

3142: including the empty set.

3143: \itemm{($b$)}

3144: $\PPPomp(X)$ \,is the set of all countable {\it non-empty\/}

3145: subsets of $X$.

3146: \itemm{($c$)}

3147: We write \,\Yua\ \,for \,$Y\cup\curly{\ua}$,

3148: \,where `\ua' denotes divergence.

3149: \itemm{($d$)}

3150: We write \,$f:X\toto Y$ \,for

3151: \,$f:X\to\PPPom(Y)$.

3152: \itemm{($e$)}

3153: \,We write \,$f:X\totop Y$ \,for

3154: \,$f:X\to\PPPomp(Y)$.

3155: \endpr

3156: 

3157: We will interpret a \WhileCC\ \,procedure

3158: $$

3159: P:\utos

3160: $$

3161: as a countably-many-valued function \PA\ from

3162: \Au\ to \Asua, \,\ie, as a function

3163: $$

3164: \PA:\ \Au\ \to\ \PPPom(\Asua)

3165: $$

3166: or, in the above notation:

3167: $$

3168: \PA: \ \Au\ \totop \ \Asua.

3169: $$

3170: 

3171: \Remarkn{3.2.2 \ (Significance of `\ua')}

3172: Notice that an output of, say, \,$\{2,5,\ua\}$

3173: \,is different from \,$\{2,5\}$,

3174: since the former indicates the possibility of divergence.

3175: So a semantic function will have, for inputs not in its domain,

3176: \,`\ua' \,as a possible output value.

3177: \endpr

3178: 

3179: 

3180: \Defn{3.2.3 \ (States)}

3181: ($a$) \,For each \Sig-algebra $A$, a {\it state} on $A$ is a family

3182: \nl

3183: \ang{\sigs\mid \sinSortSig}

3184: \,of functions

3185: $$

3186: \sigs : \Vars \to \As.

3187: $$

3188: Let \StateA\ be the set of states on $A$,

3189: with elements \ $\sig,\dots$.

3190: \sn

3191: ($b$) \,Let \sig\ be a state over $A$,

3192: \ $\xtt \ident (\tup{\xtt}{1}{n}):u$

3193: \ and

3194: \ $a = (\tup{a}{1}{n}) \in \Au$

3195: \ (for $n\ge1$).

3196: The {\it variant} \,$\sig\{\xtt / a\}$ \,of \sig\

3197: is the state over $A$ formed from \sig\ by replacing its value

3198: at $\xtt_i$ by $a_i$ for $i=1,\dots,n$.

3199: 

3200: 

3201: We give a brief overview of 

3202: {\it algebraic operational semantics\/}.

3203: This was used in \cite{tz:book} for 

3204: deterministic imperative languages with \qwhiles\

3205: and recursion (see \cite{tz:hb} for the case of \WhileSig),

3206: but it can be applied to a wide variety of imperative languages.

3207: It has also been used to analyse compiler correctness \cite{stephenson:thesis}.

3208: It can also be adapted, as we will see, to a nondeterministic language

3209: such as \WhileCCx.

3210: 

3211: Assume ($i$) we have a meaning function for atomic statements

3212: $$

3213: \angg{\Sat} :\ \StateA \ \totop \ \StateAua,

3214: $$

3215: and ($ii$) we have defined a 

3216: pair of functions

3217: $$

3218: \align

3219:   \First&:\ \Stmt \to \AtSt\\

3220:   \RestA&:\ \Stmt \times \StateA \to \Stmt ,

3221: \endalign

3222: $$

3223: where, for a statement $S$ and state \sig,

3224: 

3225: {\displaytext

3226: \ $\First (S)$ \ is an atomic statement

3227: which gives the {\it first} step in the execution of $S$ 

3228: (in any state),

3229: and \ $\RestA (S,\sig)$ \ is a statement

3230: (or, in the present context, a finite set of statements)

3231: which gives the {\it rest} of the execution in state \sig.

3232: 

3233: }

3234: \n

3235: From these we define the

3236: {\it computation step\/} function

3237: $$

3238: \CompStepA : \ \Stmt\times\StateA \totop \ \StateAua

3239: $$

3240: \TOL

3241: $$

3242: \CompStepA (S,\sig) \ = \ \angg{\First(S)}^A\sig.

3243: \tag"by"

3244: $$

3245: \TOR

3246: from which, in turn, we can define 

3247: (for the deterministic language of \cite{tz:hb})

3248: a {\it computation sequence\/} 

3249: or (for the present language)

3250: a {\it computation tree\/}.

3251: The aim is to define a {\it computation tree stage\/} function

3252: $$

3253: \CompTreeStageA: \ \Stmt\times\StateA\times\NN \ \totop \ (\StateAua)^{<\om}

3254: $$

3255: \,where \,$\CompTreeStageA(S,\sig,n)$ \,represents the 

3256: first $n$ stages of \,\CompTreeASsig.

3257: Here 

3258: \,$(\StateAua)^{<\om}$

3259: \,denotes the set of finite sequences 

3260: from 

3261: \,\StateAua,

3262: \,interpreted as finite initial segments of the paths through 

3263: the computation tree.

3264: From this, in turn, are defined the semantics 

3265: of statements and procedures.

3266: 

3267: The intuition behind these semantics

3268: is that 

3269: 

3270: {\displaytext

3271: for any input \xinAu,

3272: \,$\PA(x)$ is the set of all possible outcomes 

3273: (including divergence),

3274: for all possible implementations of the 

3275: \,\qchooses\ \,construct,

3276: including non-constructive implementations!

3277: 

3278: }

3279: \n

3280: For if (for a given input $x$) the only infinite paths through 

3281: the semantic computation tree are non-constructive,

3282: then $\PA(a)$ will still include `\ua'.

3283: 

3284: We now turn to the details of these definitions.

3285: 

3286: 

3287: \mn{\bf ({\bi a}) \,Semantics of program terms.}

3288: \ The meaning of

3289: \,$t\in \PTerm_s$ \,is a function

3290: $$

3291: \tA :\ \StateA \totop \ \Asua.

3292: $$

3293: The definition is by structural induction on $t$:

3294: $$

3295: \align

3296: \bb{\xtt}^A\sig \ =&\ \{\,\sig(\xtt)\,\} \\

3297: \bb{c}^A\sig \ =&\ \{\,c^A\,\} \\

3298: \bb{F(\tup{t}1m)}^A\sig

3299: \ =&\quad\ \,\curly{y\mid 

3300: \ex x_1\in A\cap\bb{t_1}\sig\,\dots\,\ex x_m\in A\cap\bb{t_m}\sig\,:

3301: \,\FA(\tup{x}1m)\da y}\\

3302: &\cup \ \{\,\ua \,\mid \,

3303: \ex x_1\in A\cap\bb{t_1}\sig\,\dots \,\ex x_m\in A\cap\bb{t_m}\sig\,:

3304: \,\FA(\tup{x}1m)\,\ua \,\}\\

3305: %&\cup \ \{\,\ua 

3306: %  \mid \ua \in \bb{t_1}^A\sig\,\dis\,\dots\,\dis\,\ua\in\bb{t_m}^A\sig\,\}\\

3307: &\cup \ \{\,\ua 

3308: \mid \ua \in \bb{t_i}^A\sig\ \,\tx{for some $i$, $1\le i \le m$}\,\}\\

3309: \bb{\ifs(b,t_1, t_2)}^A\sig

3310: \ =&\quad\ \,\curly{y \mid 

3311: \big(\ttt\in\bb{b}^A\sig \,\con\,y\in\bb{t_1}^A\sig\big)

3312: \ \dis

3313: \ \big(\fff\in\bb{b}^A\sig \,\con\,y\in\bb{t_2}^A\sig\big)}\\

3314: &\cup \ \{\,\ua \mid \ua \in \bb{b}^A\sig \,\}\\

3315: \bb{\choosezb}^A\sig

3316: \ =&\quad\ \,\curly{\ninNN  \mid \ttt\in \bA\sigzn}\\

3317: &\cup \ \curly{\ua \mid \all\ninNN

3318: \big(\fff\in\bA\sigzn\ \dis\ \ua\in\bA\sigzn\big)}.

3319: \endalign

3320: $$

3321: Notice that \,$\bb{\choosezb}^A\sig$

3322: \,could include both natural numbers and `\ua',

3323: since for any $n$, \,$\bA\sigzn$ \,could include both \,\ttt\ \,and \,\fff.

3324: 

3325: \mn

3326: {\bf ({\bi b}) \ Semantics of atomic statements}.

3327: \ The meaning of an atomic statement \,$\Sat\in\AtSt$ \,is a function

3328: $$

3329: \angg{\Sat} :\ \StateA \ \totop \ \StateAua

3330: $$

3331: defined by:

3332: $$

3333: \align

3334: \angg{\skips}^A\sig \ =& \ \curl{\sig}\\

3335: \angg{\divs}^A\sig \ =& \ \curly{\ua}\\

3336: \angg{\xtt:=t}^A\sig \ =& 

3337: \quad\ \,\curly{\sig\{\xtt/a\}\mid a\in A\cap\tA\sig}\\

3338: &\cup\ \curly{\ua\mid \ua\in\tA\sig}

3339: \endalign

3340: $$

3341: \mn

3342: {\bf ({\bi c}) \,The \,\First\ \,and\ \,\Rest\ \,operations.}

3343: \ The operation

3344: $$

3345: \First: \ \Stmt \ \to \ \AtSt

3346: $$

3347: is defined exactly as in \cite[\S3.5]{tz:hb},

3348: namely:

3349: $$

3350: \First(S) \ = \

3351: \cases

3352: S \ift{$S$ is atomic}\\

3353: \First(S_1) \ift{$S \ident S_1;S_2$}\\

3354: \skips \ow.

3355: \endcases

3356: $$

3357: The operation

3358: $$

3359: \RestA:\ \Stmt \times \StateA \ \totop \ \Stmt,

3360: $$

3361: %indicating that \,$\RestA(S,\sig)$

3362: %\,is a {\it finite non-empty set\/} of statements,

3363: is defined as follows

3364: (\cf\ \cite[\S3.5]{tz:hb}):

3365: 

3366: \sn

3367: {\it Case 1.}   \ $S$ is atomic. \ Then

3368: $$

3369: \RestA (S,\sig) \ = \ \curly{\skips}.

3370: $$

3371: {\it Case 2.}  \ $S \ \ident \ S_1;S_2$.

3372: \sn\indent

3373: {\it Case 2a.} \ $S_1$ is atomic. Then

3374: $$

3375: \RestA(S,\sig) \ = \ \curly{S_2}.

3376: $$

3377: \indent

3378: {\it Case 2b.} \ $S_1$ is not atomic. Then

3379: $$

3380: \RestA(S,\sig) \ = \ 

3381: \curly{S';S_2 \mid S' \in \RestA(S_1,\sig)}

3382: \ \cup\ \curly{\divs\mid\divs\in\RestA(S_1,\sig)}.

3383: $$

3384: %$$

3385: %\RestA(S,\sig) \ = \ 

3386: %\cases

3387: %\curly{S_2} \tif{$S_1$ is atomic}\\

3388: %\quad\curly{S';S_2 \mid S' \in \RestA(S_1,\sig)} \ow\\

3389: %\cup\ \curly{\divs\mid\divs\in\RestA(S_1,\sig)}.

3390: %\endcases

3391: %$$

3392: {\it Case 3.} \ $S \ \ident \ \ifs \ b \ \thens \ S_1 \ \elses \ S_2 \ \fis$.

3393: \ Then \ $\RestA (S,\sig)$ \ contains all of:

3394: $$

3395: \cases

3396: S_1 \tif{$\ttt\in\bA\sig$},\\

3397: S_2 \tif{$\fff\in\bA\sig$},\\

3398: \divs \tif{$\ua\in\bA\sig$}.

3399: \endcases

3400: $$

3401: Note that more than one condition may hold.

3402: 

3403: \sn

3404: {\it Case 4.} \ $S \ \ident \ \whiles\ b \ \dos \ S_0 \ \ods$.

3405: \ Then \ $\RestA(S,\sig)$ \ contains all of:

3406: $$

3407: \cases

3408: S_0;S \tif{$\ttt\in\bA\sig$},\\

3409: \skips \tif{$\fff\in\bA\sig$},\\

3410: \divs \tif{$\ua\in\bA\sig$}.

3411: \endcases

3412: $$

3413: Note again that more than one condition may hold.

3414: 

3415: \mn

3416: {\bf ({\bi d}) \,Computation step.}

3417: \ From the \First\ function we can define the 

3418: computation step function

3419: $$

3420: \CompStepA : \ \Stmt\times\StateA \totop \ \StateAua

3421: $$

3422: which is like the one-step computation function \,\CompIA\

3423: \,of \cite[\S3.4]{tz:hb}, except for being multi-valued:

3424: $$

3425: \CompStepA (S,\sig) \ = \ \angg{\First(S)}^A\sig.

3426: $$

3427: \mn

3428: {\bf ({\bi e}) \,The computation tree.}

3429: \ The {\it computation sequence\/},

3430: which is basic to the semantics

3431: of \,\While\ \,computations in \cite{tz:hb},

3432: is replaced here by a {\it computation tree\/}

3433: $$

3434: \CompTreeA (S,\sig)

3435: $$

3436: of a statement $S$ at a state \sig.

3437: This is an \om-branching tree,

3438: branching 

3439: %(not according to the outcome of

3440: %boolean tests, like the syntactic computation tree 

3441: %of \cite[\S5.10]{tz:hb}, but)

3442: according to all possible outcomes (\ie, ``output states")

3443: of the one-step computation function \,\CompStepA.

3444: Each node of this tree is labelled by

3445: either a state

3446: or `\ua'.

3447: 

3448: Any actual (``concrete") computation of statement $S$ at state \sig\

3449: corresponds to 

3450: one of the paths through this tree.

3451: The possibilities for any such path are:

3452: 

3453: \itemm{($i$)} it is finite, ending in a leaf containing a state:

3454: the final state of the computation;

3455: 

3456: \itemm{($ii$)} it is finite, ending in a leaf containing `\ua'

3457: (local divergence);

3458: 

3459: \itemm{($iii$)} it is infinite (global divergence).

3460: 

3461: Correspondingly, the function \,\CompA\

3462: \,of \cite[\S3.4]{tz:hb} is replaced by a {\it computation tree stage\/}

3463: function

3464: $$

3465: \CompTreeStageA: \ \Stmt\times\StateA\times\NN \ \totop \ (\StateAua)^{<\om}

3466: $$

3467: \,where \,$\CompTreeStageA(S,\sig,n)$ \,represents the 

3468: first $n$ stages of \,\CompTreeASsig.

3469: %Here 

3470: %\,$(\StateAua)^{<\om}$

3471: %\,denotes the set of finite sequences 

3472: %from 

3473: %\,\StateAua,

3474: %\,interpreted as finite initial segments of the paths through 

3475: %the semantic computation tree.

3476: This is defined (like \CompA) by a simple recursion (``tail recursion")

3477: on $n$:

3478: \sn

3479: {\it Basis:\/}

3480: \,$\CompTreeStageA(S,\sig,0) \ = \ \curl{\sig}$,

3481: \ \ie, \,just the root labelled by \,\sig.

3482: %($n=1$)

3483: %\ $\CompTreeStageA(S,\sig,1)$ \ = \ \curl{\sig}$

3484: %\ is formed by {\it attaching\/} to the root \,\curl{\sig}

3485: %\,leaves\  \,\curl{\sig'} 

3486: %\,for each $\sig'\in \CompStepA(S,\sig)$.

3487: \sn

3488: {\it Induction step:\/}

3489: \,$\CompTreeStageA(S,\sig,n)$

3490: \,is formed by attaching to the root \,\curl{\sig}

3491: \,the following:

3492: \itemm{($i$)\,}

3493: for $S$ atomic:

3494: \,the leaf \,\curl{\sigp},

3495: \,for each \,$\sig' \in \angg{S}^A\sig$

3496: \,(where \,\sigp\ \,may be a state or \,\ua);

3497: \itemm{($ii$)\,}

3498: for $S$ not atomic:

3499: \nl

3500: \,the subtree \,$\CompTreeStageA(S',\sig',n)$,

3501: \,for each \,$\sig' \in \CompStepA(S,\sig)$ \,$(\sig'\ne\ua)$

3502: \,and \,$S'\in \RestA(S,\sig)$, \,as well as

3503: \,the leaf \,\curly{\ua} \,if \,$\tx{`\ua'}\in \CompStepA(S,\sig)$.

3504: 

3505: Then \,\CompTreeASsig\ \,is defined as the ``limit" over $n$ of

3506: \,$\CompTreeStageA(S,\sig,n)$.

3507: 

3508: Note that only the leaves of \,\CompTreeASsig\ 

3509: \,may contain 

3510: `\ua', indicating ``local divergence".

3511: 

3512: \mn

3513: {\bf ({\bi f}) \,Semantics of statements.}

3514: \ From the semantic computation tree we can easily define the 

3515: i/o semantics of statements

3516: $$

3517: \SA :\ \StateA \ \totop \ \StateAua.

3518: $$

3519: Namely, 

3520: 

3521: {\displaytext

3522: \SA\sig\ \,is the set of states and/or `\ua'

3523: \,at all leaves in 

3524: \CompTreeASsig, \,together

3525: with \,`\ua' \,if \,\CompTreeASsig\ \,has an infinite path.

3526: 

3527: }

3528: 

3529: \mn

3530: {\bf ({\bi g}) \,Semantics of procedures.}

3531: \ Finally, if

3532: $$

3533: P \ \ident \ \funcs\ \ins \ \att\ \outs \ \btt\ \auxs \ \ctt

3534: \ \begins \ S\ \ends

3535: \tag1

3536: $$

3537: is a procedure of type \utov,

3538: then its meaning in $A$ is a function

3539: $$

3540: \PA : \ \Au \ \totop \ \Av^\ua

3541: $$

3542: defined as follows (\cf\ \cite[\S3.6]{tz:hb}).

3543: For \xinAu,

3544: $$

3545: %\align

3546: %\PA(x) \ =& \quad \ \ \curly{\sig'(\btt)\mid \sig'\in \SA\sig}\\

3547: %& \ \cup \ \curly{\ua \mid \ua\in\SA\sig}

3548: %\endalign

3549: \PA(x) \ = \ \curly{\sig'(\btt)\mid \sig'\in \SA\sig}

3550: \ \cup \ \curly{\ua \mid \ua\in\SA\sig}

3551: $$

3552: where \sig\ is any state on $A$ such that

3553: \,$\sig[\att] = x$.

3554: 

3555: \Remarkn{3.2.4}

3556: From the initialisation condition (\S3.1($e$))

3557: it follows by a ``functionality lemma" (\cf\ \cite[3.6.1]{tz:hb})

3558: that \,\PA\ \,is well defined.

3559: \endpr

3560: 

3561: \Defn{3.2.5}

3562: A \WhileCC\ \,procedure 

3563: \,$P:\utov$

3564: \,is {\it deterministic\/} on $A$

3565: if for all \xinAu,

3566: \,$\PA(x)$ \,is a singleton.

3567: \endpr

3568: 

3569: \Remarkn{3.2.6 \,(Two concepts of deterministic computation)}

3570: One can distinguish between two notions of deterministic computation:

3571: ($i$) {\it strong deterministic computation\/}, the common concept, 

3572: in which each step of the computation is determinate; \,and

3573: ($ii$) {\it weak deterministic computation\/}, 

3574: in which

3575: the output (or divergence) is 

3576: uniquely determined by (\ie, a unique function of) the input,

3577: but the steps in the computation are {\it not\/} determinate.

3578: A good example of ($ii$) is the Gaussian elimination algorithm (\S1.2.1, \S4.1)

3579: which, although defining a unique function (the inverse of a matrix),

3580: incorporates the (nondeterministic!) pivot function

3581: as a subroutine.

3582: In Definition 3.2.5 and elsewhere in this paper,

3583: we are concerned with the weak sense of deterministic computation.

3584: \endpr

3585: 

3586: \Defn{3.2.7}

3587: ($a$) \,A many-valued function 

3588: \,$F : \ \Au \ \totop \ \Asua$

3589: \,is \WhileCC\ \,computable on $A$

3590: if there is a \WhileCC\ \,procedure $P$

3591: such that \,$F = \PA$.

3592: \sn

3593: ($b$) \,A partial function \,$F:\Au\pto \As$

3594: \,is \,\WhileCC\ \,{\it computable\/} on $A$

3595: if there is a deterministic \,\WhileCC\ \,procedure 

3596: \,$P:\utos$

3597: \,such that for all \xinAu,

3598: \itemm{($i$)}

3599: \ $F(x) \da y \ \impp \ \PA(x) = \{y\}$,

3600: \itemm{($ii$)}

3601: \ $F(x) \ua \ \impp \ \PA(x) = \{\ua\}$,

3602: \endpr

3603: 

3604: 

3605: \Remarkn{3.2.8 \,(Many-valued algebras)}

3606: As we have seen, the semantics for \WhileCC\ procedures

3607: is given by countably many-valued functions.

3608: If we were to start with algebras with many-valued basic operations,

3609: as in \cite{brattka96,brattka:thesis}, the algebraic operational semantics

3610: could handle this just as easily, by adapting the 

3611: clause for the basic \Sig-function $F$ in part ($a$) 

3612: (``Semantics of program terms") 

3613: of the semantic definition above.

3614: \endpr

3615: 

3616: 

3617: \shead{3.3}{The language \WhileCCxSig}

3618: In \cite{tz:top,tz:hb} we worked with

3619: the language \Whilex\ \,rather than \While,

3620: which can be viewed as \While\ augmented by 

3621: auxiliary array and \,\nats\ \,variables \cite[\S3.13]{tz:hb}.

3622: The importance of \Whilex\ 

3623: computability lies in the fact

3624: that it forms the basis

3625: for a generalised Church-Turing Thesis for computability

3626: on abstract many-sorted algebras \cite[\S8]{tz:hb}.

3627: 

3628: Here, similarly,

3629: we will work with the language 

3630: $\WhileCCx = \WhileCCxSig$,

3631: which may be thought of as

3632: \WhileCCSig\ \,augmented by auxiliary array and \,\nats\ \,variables

3633: (or as \WhilexSig\ \,augmented by the \,\qchooses\ \,construct).

3634: More precisely:

3635: 

3636: \Defn{3.3.1 \,(The \WhileCCxSig\ language)}

3637: A \WhileCCxSig\ procedure is a \WhileCC(\Sigx) procedure

3638: in which the input and output variables have sorts in \Sig\ only.

3639: (However the auxiliary variables may have starred sorts or sort \,\nats.)

3640: \endpr

3641: 

3642: \n

3643: Thus a \WhileCCxSig\ procedure defines a countably-many-valued

3644: function on any standard \Sig-algebra.

3645: 

3646: 

3647: 

3648: \shead{3.4}{Some computability issues in the semantics of 

3649: \WhileCCx\ procedures}

3650: Some interesting issues in the semantics of \WhileCCx\ arise 

3651: already in the case of computation over the algebra \NNN\ 

3652: of naturals (Example 2.2.4($b$)).

3653: 

3654: \itemm{{\bf (a)}}

3655: {\bf Elimination of `\chooses' from deterministic \WhileCCx\ programs

3656: over total algebras}

3657: 

3658: \n

3659: The `\chooses' operator can be eliminated from deterministic

3660: \WhileCCx\ \,pro\-ce\-dures 

3661: (\cf\ Definition 3.2.5 and Remark 3.2.6) over total algebras.

3662: 

3663: \Propn{3.4.1}

3664: For any total \Sig-algebra $A$ and \,$f\: \Au\pto\As$,

3665: \sn

3666: \ce{$f$ is \WhileCCx\ computable over $A$

3667: \ $\ifff$

3668: \ $f$ is \Whilex\ computable over $A$.}

3669: \endpr

3670: 

3671: \Pf

3672: ($\tto$)

3673: Let $P$ be a deterministic \WhileCCx\ \,procedure over $A$

3674: which computes $f$.

3675: Since $A$ is total,

3676: evaluation of any boolean term $b$ over $A$

3677: (relative to a state) converges to \ttt\ or \fff\ in $A$.

3678: Further, since $P$ is deterministic,

3679: its output for a given input is independent of the implementation.

3680: Hence every `\chooses' term in $P$ of the form

3681: \,$\chooses \ \ztt: b[\ztt]$

3682: \, can be replaced by a 

3683: \,\qwhiles\ \,loop which 

3684: tests \,$b[0],\,b[1], \,b[2],\ \dots$

3685: \,in turn, \ie,

3686: finds

3687: the {\it least\/} $k$ for which \,$b[k]$

3688: \,is true, if it exists, and diverges otherwise.

3689: \endpf

3690: 

3691: Applying this to the total algebra \NNN, and

3692: recalling that

3693: \Whilex\ computability over \NNN\ 

3694: is equivalent to {\it partial recursiveness\/} 

3695: (\ie, classical computbility) over \NN\ \cite{tz:hb},

3696: we have:

3697: 

3698: \Corn{3.4.2}

3699: For any \,$f\: \NN^m\pto\NN$,

3700: \sn

3701: \ce{$f$ is \WhileCCx\ computable over \NNN\ 

3702: \ $\ifff$

3703: \ $f$ is partial recursive over \NN.}

3704: \endpr

3705: 

3706: \mn

3707: {\bf(b) \,Recursive and non-recursive implementations}

3708: 

3709: \n

3710: The semantics \PA\ of a procedure $P$

3711: is given, for an input $x$,

3712: by {\it all paths\/} of the computation tree 

3713: \,$T=\CompTreeASsig$ \,(where $S$ is the body of $P$)

3714: \,representing {\it all possible computation sequences\/}

3715: for $S$ starting at state \sig, where

3716: \,$\sig[\att] = x$,

3717: \ie, all possible implementations of 

3718: instances of the `\chooses' construct occurring in the execution of $S$

3719: starting at \sig.

3720: This gives rise to interesting computation-theoretic issues

3721: even in the simple case that $A = \NNN$.

3722: In this case we can assume that $T$

3723: is coded as a subset of \NN\

3724: in a standard way.

3725: Now any path of $T$ ending in a leaf is finite,

3726: and therefore (trivially) recursive.

3727: An infinite path or computation sequence

3728: (leading to divergence), however,

3729: may or may not be recursive.

3730: 

3731: \Propn{3.4.3}

3732: There is  a \WhileCCx(\NNN)\ \,procedure $P$

3733: such that its computation tree 

3734: has infinite paths, 

3735: but no recursive infinite paths.

3736: \endpr

3737: 

3738: \Pf

3739: Our construction of $P$ is

3740: based on the construction of a recursive tree

3741: with infinite paths, but no recursive infinite paths 

3742: \cite[V.5.25]{odifreddi:book}.

3743: Let $A$ and $B$ be two disjoint r.e., recursively inseparable sets,

3744: and suppose \,$A = \ran{f}$ \,and \,$B = \ran{g}$ 

3745: \,where $f$ and $g$ are total recursive functions.

3746: The procedure $P$ can be written in pseudo-code as:

3747: $$

3748: \align

3749: &\funcs \ \,\auxs \ \,\ntt,\,\ktt:\nats,\\

3750: &\qqquad \,\choicesttx:\nats\str,\quad

3751:   \curly{\tx{array recording all choices up to present stage \,\ntt}}\\

3752: &\qqquad \,\haltt:\bools\\

3753: &\begins\\

3754: &\qquad\ntt\,:=\,0;\\

3755: &\qquad\choicesttx\,:=\Nulls;\\

3756: &\qquad\haltt\,:=\falses;\\

3757: &\qquad\whiles\ \,\nots\ \,\haltt\ \,\dos\\

3758: &\qqquad \ntt\,:=\,\ntt+1;\\

3759: &\qqquad\choicesttx\,:=\, \Newlengths(\choicesttx, \,\ntt+1);\\

3760: &\qqquad\choicesttx[\ntt]\,:=\, 

3761:   \chooses \ \ztt: (\ztt = 0 \ \,\ors\ \,\ztt = 1);\\

3762: &\qqquad\fors \ \,\ktt \,:= \,0 \ \,\tos\ \,\ntt-1\ \,\dos\\

3763: &\qqquad\qquad

3764: \ifs\ (\choicesttx[\ktt]=0 \ \,\ands\ \,\ktt\in\curly{f(0),\dots,f(\ntt-1)}) 

3765: \ \,\ors\\

3766: &\qqquad\qquad\quad

3767: (\choicesttx[\ktt]=1 \ \,\ands\ \,\ktt\in\curly{g(0),\dots,g(\ntt-1)})\\

3768: &\qqquad\qquad\thens \ \,\haltt\,:=\,\trues\\

3769: &\qqquad\ods\\

3770: &\qquad \ods\\

3771: &\ends.

3772: \endalign

3773: $$

3774: Let

3775: \,$\al_0,\,\al_1,\,\al_2,\,\dots$

3776: \,be the successive values (0 or 1) given by the `\chooses' \,operator

3777: in some given implementation of $P$.

3778: Note that at stage $n$,

3779: $$

3780: \choicesttx[k] \ = \ \al_k \qquad \tx{for} \ \,k=0,\dots,n-1.

3781: $$

3782: Further, the execution diverges if, and only if, 

3783: the set \,$C \eqdf \curly{k\mid \al_k = 1}$

3784: {\it separates\/} $A$ and $B$ (\ie, \,$A\sseq C$ \,and \,$C\cap B = \nil$),

3785: in which case $C$, and hence its characteristic function 

3786: \,$\al = (\al_0, \al_1,\al_2,\dots)$,

3787: are {\it non-recursive\/}.

3788: 

3789: Note finally that for any given 

3790: sequence \al\ of choices,

3791: \al\ is effectively obtainable 

3792: from the corresponding computation sequence or path,

3793: \ie, \al\ is recursive in that path

3794: (with a standard coding of the computation tree).

3795: Hence, since any infinite sequence \al\ is non-recursive,

3796: so is the corresponding infinite path.

3797: \endpf

3798: 

3799: \Remarksn{3.4.4}

3800: (1) \,Clearly, $P$ as defined above

3801: is not semantically equivalent to a \Whilex(\NNN) \,procedure.

3802: This does not contradict Proposition 3.4.1, 

3803: since $P$ is not deterministic.

3804: \sn

3805: (2) \,According to

3806: our semantics above (\S3.2), for $P$ as defined above,

3807: \,$\ua\in\PA()$,

3808: \,\ie, divergence is possible.

3809: However, if we were to restrict all computation sequences

3810: to be recursive, then divergence would not be a possible outcome

3811: for $\PA()$.

3812: The semantics, as we give it (\ie, all possible computation sequences

3813: are included, whether recursive or not) is simpler than this alternative.

3814: In any case, as we will see,

3815: this choice will not affect continuity considerations

3816: (\cf\ Lemmas 5.1.7 and 5.2.1).

3817: \endpr

3818: 

3819: \shead{3.5}{Approximable \,\WhileCCx\ \,computability}

3820: The basic notion of computability that we will be using

3821: in working with metric algebras

3822: is not so much computability,

3823: as rather {\it computable approximability on metric algebras\/}, 

3824: as discussed in \cite[\S9]{tz:top}. We have to adapt the definition

3825: given there to the nondeterministic case

3826: with countable choice.

3827: 

3828: Let $A$ be a metric \Sig-algebra,

3829: $u$ a \Sig-product type and $s$ a \Sig-type.

3830: Let \,$P:\nats\times u \ \to \ s$

3831: \,be a \WhileCCxSigN\ \,procedure.

3832: Put

3833: $$

3834: \PAn \ \eqdf \ \PA(n,\,\cdot\,):\ \Au \ \totop \ \Asua.

3835: $$

3836: Note that that 

3837: for all \xinAu, \,$\PAn(x)\ne \nil$.

3838: 

3839: 

3840: \Defn{3.5.1 \,(\WhileCCx\ \,approximability to a single-valued function)}

3841: \nl

3842: Let \,$F:\Au\pto \As$

3843: \,be a single-valued partial function on $A$.

3844: \sn

3845: ($a$)

3846: \,$F$ is {\it \WhileCCx\ approximable\/} by $P$ on $A$

3847: if for all \,\ninNN\ \,and \,all \xinAu:

3848: $$

3849: x \in \dom{F} \ \impp\ \ua \,\notin \,\PAn(x)\ \sseq\ \Bb(F(x),\,2^{-n}).

3850: \tag1

3851: $$

3852: ($b$) $F$ is {\it strictly \WhileCCx\ approximable\/} by $P$ on $A$

3853: if in addition to (1), 

3854: $$

3855: x \notin \dom{F} \ \impp\ \PAn(x)=\curly{\ua}.

3856: \tag2

3857: $$

3858: 

3859: \Remarkn{3.5.2}

3860: If $F$ is strictly approximable by $P$, then 

3861: (from (1) and (2))

3862: for all \xinAu\ and all $n$:

3863: $$

3864: F(x)\ua \ \ \ifff\ \ \ua \in \PAn(x) \ \ \ifff\ \ \PAn(x) = \curly{\ua}.

3865: $$

3866: \endpr

3867: 

3868: Clearly, \WhileCCx\ \,{\it computability\/} is a special case of

3869: \WhileCCx\ \,{\it approximability\/}.

3870: \endpr

3871: 

3872: \Defn{3.5.3 \ (\WhileCCx\ \,approximability to a many-valued function)}

3873: \nl

3874: Let \,$F:\Au\toto \As$

3875: \,be a countably-many-valued function on $A$.

3876: \sn

3877: \itemm{($a$)}

3878: \,$F$ is {\it \WhileCCx\ approximable\/} by $P$ on $A$

3879: if for all \,\ninNN\ \,and \,all \xinAu:

3880: $$

3881: \aligned

3882: F(x)\ne\nil \ \ \impp\ \ \ua \,\notin \,\PAn(x)\ &\sseq\ 

3883:   \bigcup_{y\in F(x)}\Bb(y,\,2^{-n}) \\

3884: \tx{and} \quad F(x) \ &\sseq\ \bigcup_{y\in\PAn(x)}\Bb(y,\,2^{-n}).

3885: \endaligned

3886: \tag3

3887: $$

3888: Note that 

3889: (assuming $\ua\notin\PAn(x)$)

3890: the r.h.s. of (3) implies 

3891: $$

3892: \dsH(\ol{F(x)},\,\ol{\PAn(x)}) \,\le \,2^{-n},

3893: \tag4

3894: $$

3895: and is implied by 

3896: $$

3897: \dsH(\ol{F(x)},\,\ol{\PAn(x)}) \,< \,2^{-n},

3898: \tag5

3899: $$

3900: where \,$\ol{X}$ \,denotes the closure of $X$,

3901: and \dsH\ is the {\it Hausdorff metric\/} on the set of

3902: closed, bounded non-empty subsets of \As\ \cite[4.5.23]{engelking:book}.

3903: (Actually, the Hausdorff metric applies only to the space of closed 

3904: {\it bounded\/} subsets of a given metric space,

3905: so (4) and (5) should be taken as heuristic statements.)

3906: 

3907: \n

3908: In other words 

3909: (assuming $F(x)\ne0$), for all \xinAu\ and all $n$,

3910: \,{\it each\/} output of $F(x)$ lies within $2^{-n}$ of

3911: some output of $\PAn(x)$,

3912: and vice versa.

3913: 

3914: \itemm{($b$)}

3915: $F$ is {\it strictly \WhileCCx\ approximable\/} by $P$ on $A$

3916: if in addition,

3917: $$

3918: \F(x) = \nil \ \ \impp\ \ \PAn(x) = \curly{\ua}.

3919: $$

3920: \endpr

3921: 

3922: \Remarkn{3.5.4}

3923: (\Cf\ Remark 3.5.2.)

3924: %\nl

3925: \,If $F$ is strictly approximable by $P$, then 

3926: for all \xinAu\ and all $n$:

3927: $$

3928: F(x)=\nil \ \ \ifff\ \ \ua \in \PAn(x) \ \ \ifff\ \ \PAn(x) = \curly{\ua}.

3929: $$

3930: \endpr

3931:  

3932: \bn

3933: 

3934: \itemm{\bbf4\ \ }{\bbf Examples of \WhileCCxbig\ computations and approximating

3935: \nl

3936: computations}

3937: 

3938: %In a partial algebra the search technique of interleaving fails.

3939: %This is because interleaving depends on scheduling 

3940: %{\it terminating\/} computations, and the algebraic operations

3941: %of the algebra cannot be relied upon to terminate 

3942: %(\cf\ \cite[\S5.2]{tz:hb}).

3943: %Of course, the failure of interleaving plays havoc with the theory

3944: %of \While\ computations for partial algebras.

3945: 

3946: \shead{4.1}{Discussion: \,Use of `\chooses' for searching and dovetailing}

3947: Following the examples in Section 1,

3948: the `\chooses' construct was introduced to compute many-valued functions.

3949: Technically, the `\chooses' construct strengthens the power

3950: of the \While\ \,language in performing searches.

3951: In a {\it partial algebra\/}, simple searches 

3952: (\eg, ``find some $x_k$ in an effectively enumerated set 

3953: \,$X = \curly{x_0,x_1,x_2,\dots}$ 

3954: \,satisfying $b(x_k)$")

3955: will obviously fail in general if the search simply

3956: follows the given enumeration of $X$ 

3957: (\ie, testing in turn whether \,$b(x_0)$, $b(x_1)$, $b(x_2)$, \dots 

3958: holds),

3959: since the computation of the boolean predicate $b(x)$ may not terminate

3960: for some $x$.

3961: 

3962: This problem is overcome, at the {\it concrete model\/} level,

3963: by the use of scheduling techniques such as 

3964: {\it interleaving\/} or ``{\it dovetailing\/}":

3965: \,at stage $n$, do $n$ steps in testing whether $b(x_i)$

3966: holds, for $i=0,\dots,n$.

3967: 

3968: An important function of the `\chooses' construct, which 

3969: will recur in our examples, is to simulate such scheduling techniques

3970: at the {\it abstract model\/} level.

3971: This allows searches over any countable subset $X$

3972: of an algebra $A$ that has a computable enumeration

3973: $$

3974: \enumX:\ \NN\ \to\ X,

3975: $$

3976: since we can search $X$ in $A$ by assignments such as

3977: $$

3978: \xtt\ :=\ \enumX(\chooses\ \ztt:b(\enumX(\ztt))).

3979: $$

3980: \endpr

3981: 

3982: \shead{4.2}{Examples}

3983: We now illustrate the use of the \WhileCCx\ \,language

3984: in topological partial algebras

3985: with examples,

3986: which involve computations which are either many-valued,

3987: or approximating, or both.

3988: The examples given in \S1.2 to motivate many-valued abstract computation

3989: are a good place to start.

3990: They can be displayed in the table:

3991: 

3992: \bn

3993: \ruledtable

3994: \| Exact computation | Approximating computation \crthick

3995: Single-valued   \| Gaussian elimination | $e^x$, sin($x$), etc.    \cr

3996: Many-valued \| Approx. points in metric algebra | All simple roots of polynomial

3997: \endruledtable

3998: \sn

3999: 

4000: Examples 4.2.1, 4.2.2 and 4.2.4 below are all based on the 

4001: metric algebra derived from \,\RRRN\ \,(Example 2.3.3($b$)).

4002: 

4003: 

4004: \Examplen{4.2.1 \,(Gaussian elimination)}

4005: This is a single-valued exact computation.

4006: The algorithm can be found in any standard 

4007: text of numerical computation,

4008: \eg, \cite{heath:book}.

4009: It is deterministic, but only in the weak sense

4010: (\cf\ Remark 3.2.6), since it contains, as an

4011: essential component, the computation of the {\it pivot\/} function (\S1.2),

4012: which is many-valued,  and can be formalised simply with the

4013: `\chooses' construct:

4014: $$

4015: \align

4016: &\funcs \ \ \ins\ \ \xtt_1,\dots,\xtt_n\:\reals\\

4017: &\qquad\outs\ \ \itt\:\nats\\

4018: &\qquad\auxs\ \ \ktt\:\nats\\

4019: &\begins\\

4020: &\qquad\itt\,:= 

4021:   \,\chooses \ \ktt\: (\ktt = 1 \ \,\ands\ \,\xtt_1 \ne 0) \ \,\ors\\

4022:   &\qqqquad\quad          (\ktt = 2 \ \,\ands\ \,\xtt_2 \ne 0) \ \ors\\

4023:   &\qqqquad\qqquad \dots\\

4024:   &\qqqquad\quad         (\ktt = n \ \,\ands\ \,\xtt_n \ne 0)\\

4025: &\ends.

4026: \endalign

4027: $$                                                                              

4028: 

4029: 

4030: \Examplen{4.2.2 \,(Approximations to \,${\bk e}^{\bk x}$)}

4031: On the N-standard interval algebra \,\IIIN\ (Example 2.5.3($c$))

4032: we give a \While\ procedure to approximate the function $e^x$ on $I$.

4033: $$

4034: \align

4035: &\funcs \ \ \ins\ \ \ntt\:\nats,\qquad\curly{\tx{degree of approximation}}\\

4036: &\qqquad \xtt\:\intvls\\

4037: &\qquad\outs\ \ \stt\:\reals \qquad\,\curly{\tx{partial sum of power series}}\\

4038: &\qquad\auxs \ \ \ytt\:\reals,\quad\ \,\curly{\tx{current term of series}}\\

4039: &\qqquad \,\ktt\:\nats\qquad\,\curly{\tx{counter}}\\

4040: &\begins\\

4041: &\qquad\ktt\,:=\,0;\\

4042: &\qquad\ytt\,:=\,1;\\

4043: &\qquad\stt\,:=\, 1;\\

4044: &\qquad\whiles\ \,\ktt<2^{\ntt+1}\ \,\dos\\

4045: &\qqquad \ktt\,:=\,\ktt+1;\\

4046: &\qqquad\ytt\,:=\, \ytt\times\iI(\xtt)/\,\iN(\ktt);

4047:     \qquad\curly{\ytt=\xtt^\ktt/\ktt\,!}\\

4048: &\qqquad\stt\,:=\,\stt+\ytt \qqquad\qquad\quad

4049: \curly{\stt = \sum_{i=0}^\ktt \xtt^i/i\,!}\\

4050: &\qquad \ods\\

4051: &\ends

4052: \endalign

4053: $$                                                                              

4054: where \,$\iI:I\to\RR$

4055: \,is the embedding of $I$ in \RR,

4056: which is primitive in \Sig(\IIIN),

4057: and \,$\iN:\NN\to\RR$ \,is the embedding of \NN\ in \RR,

4058: which is easily definable in \While(\RRRN).

4059: 

4060: Denoting the above function procedure by $P$, 

4061: and \IIIN\ by $A$,

4062: we have the semantics

4063: $$

4064: P_n^A\: I \ \to\ \RR

4065: $$

4066: with 

4067: $$

4068: P_n^A(x) \ = \ \sum_{i=0}^{2^{n+1}} {{x^i}\over{i\,!}}

4069: $$

4070: and so for all $x\in I$,

4071: $$

4072: \ds(P_n^A(x),\,e^x)\ < \ 2^{-n},

4073: $$

4074: \ie, $e^x$ is \While\ approximable on \,\IIIN\ by $P$.

4075: 

4076: This computation of $e^x$ is single-valued, but approximating.

4077: 

4078: 

4079: 

4080: 

4081: \Examplen{4.2.3 \,(``Choosing" a member of an enumerated subspace

4082: close to an arbitrary element of a metric algebra)}

4083: Given a metric algebra $A$ with a countable dense

4084: subspace $C$, \ and an enumeration of $C$

4085: $$

4086: \enumC\: \NN\ \onto\ C

4087: $$

4088: in the signature,

4089: we want to compute a function

4090: $$

4091: F\:A\times\NN\ \to\ C

4092: $$

4093: such that

4094: $$

4095: F(a,n) \ =\ \tx{``some" $x\in C$ \,such that \,$\ds(a,x)\,<\,2^{-n}$}.

4096: $$

4097: This is a generalised version of the problem

4098: of approximating reals by rationals (Example 1.2.3).

4099: 

4100: Here is a \WhileCCx\ procedure (in pseudo-code) 

4101: for an exact computation of this function.

4102: (Note that the real-valued function \,$2^{-n}$ \,is \While\ computable on

4103: \,\RRRN, \,and hence on $A$.)

4104: $$

4105: \align

4106: &\funcs \ \ins\ \ \ \att:\spaces,\\

4107: &\qqquad \ntt:\nats\\

4108: &\qquad\outs\ \ \xtt:\spaces\\

4109: &\qquad\auxs\ \ \ktt:\nats\\

4110: &\begins\\

4111: &\qquad\xtt\,:= 

4112:   \,\enumC \big(\chooses \ \ktt: \ds(\att, \enumC(\ktt))<2^{-\ntt}\big)\\

4113: &\ends

4114: \endalign

4115: $$                                                                              

4116: This computation is many-valued, but exact.

4117: 

4118: 

4119: 

4120: \Examplen{4.2.4 \,(Finding simple roots of a polynomial)}

4121: We construct a 

4122: \WhileCC\ procedure to approximate ``some" simple root 

4123: of a polynomial $p(X)$ with real coefficients,

4124: using the method of bisection.

4125: By a {\it simple  root of\/} $p(X)$ we mean a real root 

4126: at which $p(X)$ changes sign.

4127: (See \cite{heath:book}. In practice, a hybrid method is generally used,

4128: involving bisection, Newton's method, etc.)

4129: 

4130: 

4131: Fundamental to the bisection method is the concept 

4132: of 

4133: a {\it bracket\/} for $p(X)$,

4134: which means an

4135: interval \,$[a,b]$

4136: \,such that $p(a)$ and $p(b)$ have opposite signs.

4137: By {\it rational bracket\/},

4138: we mean a bracket with rational endpoints.

4139: 

4140: 

4141: We note the following:

4142: 

4143: \itemm{(1)}

4144: Any bracket for $p$ contains a 

4145: root of $p$

4146: (by the Intermediate Value Theorem),

4147: in fact a simple root of $p$.

4148: 

4149: \itemm{(2)}

4150: Conversely, any simple root of $p$

4151: is contained in

4152: a rational bracket for $p$

4153: of arbitrarily small width.

4154: 

4155: \itemm{(3)}

4156: If $x$ is a simple root of $p$,

4157: then any bracket for $p$ of sufficiently small width

4158: which contains $x$,

4159: contains no other simple root of $p$.

4160: 

4161: \itemm{(4)}

4162: If $[a,b]$ is a bracket for $p$,

4163: then, putting \,$m = (a+b)/2$,

4164: exactly one of the following holds:

4165: \itemmm{$(i)$}

4166: $p(m)= 0$; then $m$ is a root of $p$ (not necessarily simple);

4167: \itemmm{$(ii)$}

4168: $p(m)$ has the same sign as $p(a)$;

4169: then $[m,b]$ is a bracket for $p$;

4170: \itemmm{$(iii)$}

4171: $p(m)$ has the same sign as $p(b)$;

4172: then $[a,m]$ is a bracket for $p$.

4173: 

4174: It follows from the above that

4175: starting with any

4176: rational bracket $J$ for $p$,

4177: we can, by repeated bisection, 

4178: find a nested sequence of rational brackets

4179: $$

4180: J = J_0, \ J_1,\ J_2,\dots

4181: \qquad

4182: \tx{where}

4183: \qquad

4184: \bigcap_{n=0}^\infty J_n = \curl{x}

4185: $$

4186: for some simple root $x$ of $p$.

4187: Then, letting $r_n$ be the left-hand endpoint

4188: of $J_n$, we have a fast Cauchy sequence

4189: \,$\ang{r_n}_n$

4190: \,with limit $x$.

4191: 

4192: One complication with our algorithm

4193: is the occurrence of case ($i$) in (4) above, 

4194: \ie, the case that the midpoint $m$ of the bracket

4195: is itself a root of $p$,

4196: since by the co-semicomputabil\-ity of

4197: equality (Discussion 2.2.5) on \RR\ we can only verify when

4198: $f(m)\ne 0$, not when $f(m)=0$.

4199: We therefore proceed as follows.

4200: By means of the `\chooses' construct,

4201: we search in the middle third (say) of the bracket $[a,b]$

4202: for a ``division point", \ie, a rational point $d$ such that $f(d)\ne0$,

4203: producing 

4204: either $[a,d]$ or $[d,b]$ as a sub-bracket.

4205: 

4206: This new bracket may not halve the width

4207: of $[a,b]$; in the worst case its width

4208: is $2/3$ the width of $[a,b]$.

4209: However a second iteration of this procedure

4210: leads to a bracket of width at most

4211: $(2/3)^2 < 1/2$ the width of $[a,b]$,

4212: and so $2n$ iterations lead to a bracket of width

4213: less than $2^{-n}\,\times$ the width of $[a,b]$.

4214: 

4215: This new bracket may not halve the width

4216: of $[a,b]$; in the worst case its width

4217: is $2/3(b-a)$.

4218: However a second iteration of this procedure

4219: leads to a bracket of width at most

4220: $(2/3)^2 < 1/2$ the width of $[a,b]$,

4221: and so $2n$ iterations lead to a bracket of width

4222: less than $2^{-n}(b-a)$.

4223: 

4224: For convenience,

4225: we will use the following two conservative extensions

4226: to our ``official" programming notation:

4227: 

4228: \itemm{($a$)}

4229: Simultaneously choosing two naturals with a single condition:

4230: $$

4231: \ktt_1,\ktt_2\ := \ \chooses\ \ztt_1, \ztt_2: b[\ztt_1,\ztt_2]

4232: $$

4233: which is easily expressible in \WhileCC\

4234: \,by the use of a primitive recursive pairing function

4235: \pairs\ on \NN\ and its inverses \,$\projs_1, \projs_2$:

4236: $$

4237: \align

4238: \ktt\ &:=\ \chooses\ \ztt: b[\projs_1(\ztt),\projs_2(\ztt)];\\

4239: \ktt_1,\ktt_2\ &:=\ \projs_1(\ktt), \projs_2(\ktt)

4240: \endalign

4241: $$

4242: \itemm{($b$)}

4243: Choosing a rational (of type \reals) satisfying a boolean condition:

4244: $$

4245: %\qtt\ := \ \chooses\ \rtt^\realss: \big(\Rat(\rtt) \ \ands \ b[\rtt]\big)

4246: \qtt\ := \ \chooses\ \rtt^\realss: 

4247: \big(\tx{``\rtt\ is rational"} \ \ands \ b[\rtt]\big)

4248: $$

4249: Let \,$\rats: \NN\to \RR$

4250: \,be a \While-computable enumeration

4251: of the rationals in \RR. 

4252: Then this can be interpreted as:

4253: $$

4254: \qtt := \rats\big(\chooses\ \ktt: b[\rats(\ktt)]\big)

4255: $$

4256: 

4257: \n

4258: Finally, a polynomial $p(X)$  over \RR\

4259: will be represented by an element \px\ of \RRx:

4260: $$

4261: \px\ = \ (\tup{a}0{n-1})\ = \ \sum_{i=0}^{n-1}a_i X^{n-i}

4262: $$

4263: Its evaluation at a point $c$,

4264: denoted by $\px(c)$, 

4265: is easily seen to be

4266: \While(\RRR) computable in \px\ and $c$.

4267: 

4268: 

4269: Now we give a \WhileCCx\  procedure for approximably computing

4270: some simple root of an

4271: input polynomial, in the signature of \RRR.

4272: $$

4273: \align

4274: &\funcs \ \ins\quad \ \ \ntt:\nats,\qquad

4275:   \ \curly{\tx{degree of approximation}}\\

4276: &\qqquad \,\pttx:\reals\str\

4277:   \qquad\curly{\tx{input polynomial, given by list of coefficients}}\\

4278: &\qquad\outs\ \ \ \ \xtt:\reals \qquad\ \curly{\tx{approximation to root}}\\

4279: &\qquad\auxs\ \att,\btt:\reals, \qquad\curly{\tx{endpoints of bracket}}\\

4280: &\qqquad\ \ \,\dtt:\reals,\qquad\curly{\tx{division point of bracket}}\\

4281: &\qqquad \ \ \ \ktt:\nats\qquad\ \curly{\tx{counter}}\\

4282: &\begins\\

4283: &\qquad\ktt\,:=\,0;\\

4284: &\qquad

4285:   \att, \btt\,:=\, \chooses\ \att, \btt: (\tx{``\att\ and \btt\ are rational"} 

4286:   \ \ \ands\ \ \att < \btt < \att+1 \ \ \ands\\

4287: & \qqqquad\qqquad\ \ (\pttx(\att)>0 \ \,\ands\ \,\pttx(\btt)<0)\\

4288: & \qqqquad\qquad\quad

4289:    \ \ors\ \ (\pttx(\att)<0 \ \,\ands\ \,\pttx(\btt)>0));\\

4290: &\qquad\whiles\ \,\ktt<2\ntt\ \,\dos\\

4291: &\qqquad \ktt\,:=\,\ktt+1;\\

4292: &\qqquad \dtt\,:=\, \chooses\ \dtt: (\tx{``\dtt\ is rational"}\ \ \ands

4293:   \ \ (2\att+\btt)/3 < \dtt < (\att+2\btt)/3 \\

4294: &\qqqquad\qqquad \ands\ \ \pttx(\dtt)\ne 0);\\

4295: &\qqquad \ifs\ \ (f(\dtt)>0\ \,\ands\ \,f(\att)>0)

4296:   \ \ \ors \ \ (f(\dtt)<0\ \,\ands\ \,f(\att)<0)\\

4297: &\qqquad\qquad\thens \ \ \att,\btt\,:=\, \dtt,\btt\qquad

4298:     \curly{\tx{new bracket on right part of old}}\\

4299: &\qqquad\qquad\elses \ \ \,\,\att,\btt\,:=\, \att,\dtt\qquad

4300:     \curly{\tx{new bracket on left part of old}}\\

4301: &\qqquad \fis\\

4302: &\qquad \ods;\\

4303: &\qquad \xtt\,:= \,\att \qqqquad\qquad

4304:    \ \ \curly{\tx{$\xtt\,:=\,\btt$ \,would also work here}}\\

4305: &\ends.

4306: \endalign

4307: $$                                                                              

4308: For input natural $n$ and polynomial $p$,

4309: the output is within $2^{-n}$ of some simple root of $p$.

4310: Further, for {\it any\/} simple root $e$ of $p$, there is

4311: {\it some\/} implementation of the `\chooses' operator

4312: which will give an output within $2^{-n}$ of $e$.

4313: Finally, the computation will diverge

4314: if, and only if, $p$ has no simple roots. 

4315: 

4316: This computation is both many-valued and approximating.

4317: 

4318: \newpage

4319: 

4320: \itemm{\bbf5\ \ }{\bbf Countably-many-valued functions; \ 

4321: Continuity of \WhileCCxbig

4322: \nl

4323: computable functions}

4324: 

4325: \mn

4326: In this section we discuss the continuity of

4327: countably-many-valued functions, and then prove that the

4328: countably-many-valued functions

4329: computed by \WhileCCx\ programs

4330: are continuous.

4331: 

4332: \shead{5.1}

4333: {Topology and continuity with countably many values and 

4334: `$\bs \uparrow$'}

4335: Recall Notation 3.2.1.

4336: 

4337: 

4338: \Defn{5.1.1 \ (Totality)}

4339: The function \,$f:X\toto Y$

4340: \,is said to be {\it total\/} if for all \,$x\in X$, \,$f(x)$ 

4341: is a {\it non-empty\/} subset of $Y$,

4342: \ie, if \,$f:X\totop Y$.

4343: \endpr

4344: 

4345: Our semantic functions (in Section 6) will typically be of the form

4346: $$

4347: \Phi: \ \Au \ \totop \ \Avua.

4348: \tag1

4349: $$

4350: 

4351: \Remarkn{5.1.2}

4352: We think of the ``deterministic version" of (1)

4353: as being a total function \,\Ph, \,where

4354: for each $x\in X$, $\Ph(x)$ is a {\it singleton\/},

4355: containing either an element of \Av\ (to indicate convergence)

4356: or `\ua' (to indicate divergence).

4357: %In the present context, this is more appropriate 

4358: %for the representation of a deterministic semantic (partial) 

4359: %function than by specifying \,$\Ph(x) = \nil$ \,for divergence.

4360: (Recall Remark 3.2.2.)

4361: \endpr

4362: 

4363: We must now consider what it means for such a function (1)

4364: to be {\it continuous\/}.

4365: 

4366: \Defn{5.1.3 \ (Continuity)}

4367: Let \,$f\:X\toto Y$, where $X$ and $Y$ are topological spaces.

4368: \sn($a$)

4369: For any \,$V\sseq Y$,

4370: $$

4371: f^{-1}[V] \ \eqdf\ \ \curly{x\in X \br f(x)\cap V \ne \nil},

4372: $$

4373: \ie, \,$x\in f^{-1}[V]$ \,iff 

4374: at least one of the elements of \,$f(x)$ \,lies in $V$.

4375: \sn($b$)

4376: $f$ is {\it continuous\/}

4377: (w.r.t. $X$ and $Y$) iff for all open  \,$V\sseq Y$,

4378: \,$f^{-1}[V]$ \,is open in $X$.

4379: \endpr

4380: 

4381: \Remarksn{5.1.4}

4382: ($a$) For metric spaces $X$ and $Y$, Definition 5.1.3($b$)

4383: becomes: 

4384: \nl

4385: $f\:X\toto Y$ \,is continuous iff

4386: $$

4387: \all a\in X \,\all b\in f(a)\,\all\eps>0 \,\ex\del>0 \,\all x\in \Bb(a,\del)

4388: \,\bigl(f(x)\cap \Bb(b,\eps)\ne\nil\bigr).

4389: $$

4390: ($b$) Definition 5.1.3($b$) reduces to 

4391: the standard definition of continuity for 

4392: total single-valued functions from $X$ to $Y$.

4393: \sn

4394: ($c$) It also reduces to the definition of continuity for partial

4395: single-valued functions (Definition 2.5.1 and Remark 2.7.2($a$)),

4396: as we will see below (Remark 5.1.9).

4397: We must first see how to extend the topology on $Y$ to that on \Yua\

4398: (Definition 5.1.6 below).

4399: \endpr

4400: 

4401: \Defn{5.1.5}

4402: For two functions

4403: \ $f\:  X \toto Y$,

4404: \ $g\:  X \toto Y$,

4405: \ we define 

4406: $$

4407: f\,\sqsseq \,g \ \ \ifffdf\ 

4408: \ \tx{for all} \ x\in X, \ f(x) \,\sseq \,g(x).

4409: $$

4410: \endpr

4411: 

4412: \Defn{5.1.6 \,(Topology on \,\Yua)}

4413: We extend the topology on $Y$ to 

4414: \,\Yua\ ($= Y\cup\curly{\ua}$)

4415: by specifying that the only open set containing \,\curly{\ua}

4416: \,is \,\Yua. (So \,\Yua\ \,is a ``one-point compactification" of $Y$.)

4417: \endpr

4418: 

4419: 

4420: Now, given a function

4421: \,$f : \ X \toto \Yua$,

4422: \,we define functions

4423: \TOL

4424: $$

4425: \align

4426: \fua:& \ X \ \toto \ \Yua\\

4427: \fmin:& \ X \ \toto \ Y

4428: \tag"and"

4429: \endalign

4430: $$

4431: by 

4432: $$

4433: \align

4434: \fua(x) \ &= \ f(x)\cup\curly{\ua}\\

4435: \fmin(x) \ &= \ f(x) \backslash \curly{\ua}.

4436: \endalign

4437: $$

4438: \TOR

4439: In other words, \,\fua\ \,{\it adds\/} \,`\ua' \,to the set \,$f(x)$

4440: \,for each \,$x\in X$

4441: \,and \,\fmin\ \,{\it removes\/} \,`\ua' \,from every such set.

4442: This changes the semantics of $f$ 

4443: (see Remark 3.2.2), but not its {\it continuity properties\/},

4444: as will be seen from the following technical lemma,

4445: which will be used

4446: in the proof of continuity 

4447: of computable functions below (\S5.2).

4448: 

4449: \Lemman{5.1.7}

4450: Let 

4451: \ $f:\ X\ \toto \ Y$ 

4452:  and

4453: \ $g:\ X\ \totop \ \Yua$

4454: \ be any two functions such that 

4455: $$

4456: f\ \sqsseq\ g \ \sqsseq\ \fua,

4457: $$

4458: %for all \,$x\in X$,

4459: %$$

4460: %f(x) \ \sseq \ g(x) \ \sseq \ f(x)\cup\curly{\ua}

4461: %$$

4462: \ie, for all $x\in X$,

4463: \,$g(x) \ne\nil$,

4464: \,and either \,$g(x) = f(x)$ or $g(x) = f(x)\cup \curly{\ua}$.

4465: Then

4466: $$

4467: f\ \ \tx{is continuous} \ \ \ifff\ \ g\ \ \tx{is continuous}.

4468: $$

4469: \endpr

4470: \Pf

4471: ($\tto$)

4472: \ Suppose $f$ is continuous.  

4473: We must show $g$ is continuous.

4474: Let $V$ be an open subset of \Yua.

4475: We must show \,$g^{-1}[V]$ \,is open in $X$.

4476: There are two cases, according as \ua\ is in $V$ or not.

4477: \sn

4478: {\it Case 1}: \,$\ua\notin V$, \ie, $V\sseq Y$.

4479: \ Then $V$ is also open in $Y$ (by definition of the topology on \Yua).

4480: Hence \,$f^{-1}[V]$ \,is open in $X$, and hence

4481: $$

4482: \align

4483: g^{-1}[V] \ &= \ \curly{x\in X \mid g(x)\cap V\ne\nil}\\

4484: &= \ \curly{x\in X\mid f(x)\cap V\ne\nil}\qquad\tx{since $\ua\notin V$}\\

4485: &= \ f^{-1}[V]

4486: \endalign

4487: $$

4488: is open in $X$.

4489: \sn

4490: {\it Case 2}: \,$\ua\in V$.

4491: \ Then $V = \Yua$

4492: (by definition of the topology on \Yua).

4493: Hence 

4494: $$

4495: g^{-1}[V] \ = \ g^{-1}[\Yua] \ = \ X \qquad \tx{(since $g$ is total)},

4496: $$

4497: which is open in $X$.

4498: \sn

4499: ($\ffrom$)

4500: \ Suppose $g$ is continuous.

4501: We must show $f$ is continuous.

4502: Let $V$ be an open subset of $Y$.

4503: We must show \,$f^{-1}[V]$ \,is open in $X$.

4504: Since $V$ is also open in \Yua\ 

4505: (by definition of the topology on \Yua),

4506: $g^{-1}[V]$ \,is open in $X$.  Hence

4507: $$

4508: \align

4509: f^{-1}[V] \ &= \ \curly{x\in X \mid f(x)\cap V\ne\nil}\\

4510: &= \ \curly{x\in X\mid g(x)\cap V\ne\nil}\qquad\tx{since $\ua\notin V$}\\

4511: &= \ g^{-1}[V]

4512: \endalign

4513: $$

4514: is open in $X$.

4515: \endpf

4516: 

4517: 

4518: 

4519: \Corn{5.1.8}

4520: Suppose 

4521: \ $f: X\ \totop \Yua$ \ (\ie, $f$ is total). Then

4522: $$

4523: f\ \ \tx{is continuous} \ \ \ifff\ \ \fmin\ \ \tx{is continuous}

4524: \ \ \ifff\ \ \fua\ \ \tx{is continuous}.

4525: $$

4526: \endpr

4527: 

4528: \Pf

4529: Apply Lemma 5.1.7 twice:

4530: once with \,\fmin\ \,and $f$, 

4531: and once with \,\fmin\ \,and \,\fua.

4532: \endpf

4533: 

4534: 

4535: \Remarkn{5.1.9 \,(Justification of Remark 5.1.4($c$))}

4536: Let \,$f:X\pto Y$ 

4537: \,be a single-valued partial function.

4538: Define 

4539: \sn

4540: ($a$) \,$\fcheck\:X\toto Y$ \,by

4541: $$

4542: \fcheck(x) \ = \

4543: \cases

4544: \curly{f(x)} \ift{$x\in\dom{f}$}\\

4545: \nil \ow

4546: \endcases.

4547: $$

4548: ($b$) \,$\fhat\:X\totop \Yua$ \,by

4549: $$

4550: \fhat(x) \ = \

4551: \cases

4552: \curly{f(x)} \ift{$x\in\dom{f}$}\\

4553: \curly{\ua} \ow.

4554: \endcases

4555: $$

4556: (We can view either \,\fcheck\ \,or \,\fhat\ \,as 

4557: ``representing" $f$ in the present context, \cf\ Remark 5.1.2.)

4558: \,Then

4559: $$

4560: \align

4561: &f\ \ \tx{is continuous \ (according to Def\. 2.5.1)} \\

4562: \ifff\ \ &\fcheck\ \ \tx{is continuous \ (according to Def\. 5.1.3)}\\

4563: \ifff\ \ &\fhat\ \ \tx{is continuous \ (according to Def\. 5.1.3)}

4564: \endalign

4565: $$

4566: The equivalence of the continuity of $f$ and \,\fcheck\ 

4567: \,follows immediately from the definitions.

4568: The equivalence of the continuity of \,\fcheck\ \,and \,\fhat\

4569: \,follows from Lemma 5.1.7.

4570: \endpr

4571: 

4572: \Remarkn{5.1.10 \,(Comparison with W-continuity)}

4573: As in \S2.7, we can consider another notion of continuity 

4574: for functions \,$f:X\toto Y$ \,by modifying Definition 5.1.3($b$);

4575: \,namely,

4576: $f$ is W-continuous

4577: iff for all open  \,$V\sseq Y$,

4578: \,$f^{-1}[V]$ \,is open in \dom{f}.

4579: Note that Lemma 5.1.7, and

4580: the equivalences given in Remark 5.1.9, 

4581: also hold for W-continuity.

4582: \endpr

4583: 

4584: \Lemman{5.1.11}

4585: Given \,$f:X\toto \Yua$, \,extend it to

4586: \,$\ftil:\Xua\toto\Yua$ \,by stipulating that \,$\ftil(\ua) = \ua$.

4587: If $f$ is continuous and total, then \ftil\ is continuous.

4588: \endpr

4589: 

4590: \Pf

4591: Let $V$ be an open subset of \Yua.

4592: We must show \,$\ftil^{-1}[V]$ \,is open in \Xua.

4593: There are two cases:

4594: \sn

4595: {\it Case 1}: \,$\ua\notin V$, \ie, $V\sseq Y$.

4596: Then \,$\ftil^{-1}[V] = f^{-1}[V]$,

4597: \,which is open in $X$, and hence in \Xua.

4598: \sn

4599: {\it Case 2}: \,$\ua\in V$.

4600: \ Then $V = \Yua$

4601: (by definition of the topology on \Yua).

4602: Hence 

4603: $$

4604: \align

4605: \ftil^{-1}[V] \ &= \ \ftil^{-1}[\Yua] \\

4606: &= \ \dom{f}\cup \curly{\ua}\\

4607: &= \ X\cup\curly{\ua} \qquad\tx{(since $f$ is total)}

4608: \endalign

4609: $$

4610: which is open in \Xua.

4611: \endpf

4612: 

4613: \Defn{5.1.12 \,(Composition)}

4614: \nl

4615: ($a$) Suppose \ $f:X\toto Y$ \ and \ $g:Y\toto Z$.

4616: \ We define \ $g\circ f:X\toto Z$ \ by

4617: $$

4618: (g\circ f) (x)\ = \ \bigcup\curly{g(y)\mid y\in f(x)}.

4619: %\ \ \curly{z\in \Zua \br \ex y\in f(x)\cap Y \,(z\in g(y))}

4620: $$

4621: ($b$)

4622: Suppose \ $f:X\toto \Yua$ \ and \ $g:Y\toto\Zua$.

4623: \ We define \ $g\circ f:X\totop \Zua$ \ by

4624: $$

4625: \align

4626: (g\circ f) (x)\ = \ &

4627: \quad\ \,\bigcup\curly{g(y)\mid y\in f(x)\cap Y}\\

4628: %\quad\ \,\curly{z\in \Zua \br \ex y\in f(x)\cap Y \,(z\in g(y))}\\

4629: &\cup\ \curly{\ua \br \ua \in f(x)}

4630: \endalign

4631: $$

4632: \endpr

4633: %Note that $g\circ f$, as defined in ($b$), is the composition

4634: %(in the sense of ($a$)) of \gtil\ (in the notation of Lemma 5.1.11) 

4635: %and $f$.

4636: 

4637: \Propn{5.1.13 \,(Continuity of composition)}

4638: \sn

4639: ($a$) 

4640: If \ $f:X\toto Y$ \ and \ $g:Y\toto Z$ \ are continuous,

4641: then so is \ $g\circ f:X\toto Z$.

4642: \sn

4643: ($b$)

4644: If \ $f:X\totop \Yua$ \ and \ $g:Y\totop \Zua$ \ are continuous,

4645: then so is \ $g\circ f:X\totop \Zua$.

4646: \endpr

4647: 

4648: \Pf

4649: ($a$) Just note that 

4650: for \,$W\sseq Z$,

4651: $$

4652: (g\circ f)^{-1}[W] \ = \ f^{-1}[g^{-1}[W]].

4653: $$

4654: ($b$) 

4655: We give two proofs:

4656: \ ($i$) Note that

4657: $$

4658: (g\circ f)^- \ = \ g^- \circ f^- :\ X \ \toto \ Z

4659: $$ 

4660: and use part ($a$) and Corollary 5.1.8.

4661: \sn

4662: ($ii$)

4663: Note that

4664: for \,$W\sseq \Zua$,

4665: $$

4666: (g\circ f)^{-1}[W] \ = \ f^{-1}[\gtil^{-1}[W]]

4667: $$

4668: (in the notation of Lemma 5.1.11), and apply Lemma 5.1.11.

4669: \endpf

4670: 

4671: \Defn{5.1.14 \,(Union of functions)}

4672: Let \,$f_i:X\toto \Yua$ \,be a family of functions for \iinI.

4673: Suppose for all \,$x\in X$, \,$\bigcup_\iinI f_i(x)$ 

4674: \,is countable.

4675: Then we define 

4676: $$

4677: \bigsqcup_\iinI f_i:\ X \ \toto \ \Yua

4678: $$

4679: by 

4680: $$

4681: \big(\bigsqcup_\iinI f_i\big)(x) \ = \ \bigcup_\iinI f_i(x).

4682: $$

4683: \endpr

4684: 

4685: \Lemman{5.1.15}

4686: If \,$f_i:X\toto \Yua$ \,is continuous for all \iinI,

4687: then so is \,$\bigsqcup_\iinI f_i$.

4688: \endpr

4689: 

4690: \Pf

4691: This follows from the fact that for \,$V\sseq \Yua$,

4692: $$

4693: \big(\bigsqcup_\iinI f_i)^{-1}[V] \ = \ \bigcup_\iinI f_i^{-1}[V].

4694: \tag"$\square$"

4695: $$

4696: 

4697: \Remarkn{5.1.16}

4698: Note that all the results of this subsection (5.1)

4699: hold for {\it arbitrary\/}

4700: multivalued functions  \,$f:X\to \PPP(Y)$,

4701: \,not necessarily countably-many-valued.

4702: \endpr

4703: 

4704: \shead{5.2}{Continuity of \WhileCC\ \,computable functions}

4705: Let $A$ be an N-standard topological \Sig-algebra.

4706: 

4707: In order to prove that \WhileCCx\ \,procedures on $A$ are continuous,

4708: we first state and prove a lemma which says that

4709: such procedures are (almost) equivalent to 

4710: \While\ \,procedures (without `\chooses')

4711: in an extended signature, which includes 

4712: a symbol \,\fs\ \,for an ``oracle function".  Then we apply Lemma 5.1.7.

4713: 

4714: \Lemman{5.2.1 \,(Oracle equivalence lemma)}

4715: Given a \WhileCCSig\ \,statement $S$, and procedure

4716: $$

4717: P \ \ident \ \funcs\ \ins \ \att\ \outs \ \btt\ \auxs \ \ctt

4718: \ \begins \ S\ \ends,

4719: $$

4720: we can effectively construct a \While(\Sigf) statement \,\Sf\ 

4721: \,and procedure

4722: $$

4723: \Pfs \ \ident \ \funcs\ \ins \ \att\ \outs \ \btt\ \auxs \ \ctt

4724: \ \begins \ \Sf\ \ends

4725: $$

4726: \,in a signature \,\Sigf\ \,which extends \,\Sig\

4727: \,by a function symbol \,$\fs: \nats\to\nats$,

4728: \,such that, putting

4729: $$

4730: \PsqcupA \ \eqdf \ \bigsqcup_{f\in\FFF}\PfA,

4731: $$

4732: where \,$\FFF = \NN^\NN$ \,\ is

4733: the set of all functions \,$f\:\NN\to\NN$

4734: \,and \PfA\ is the interpretation of \,\Pfs\ in $A$ formed by interpreting 

4735: \,\fs\ \,as $f$,

4736: \,we have

4737: $$

4738: \PA \ \sqsseq \ \PsqcupA \ \sqsseq \ (\PA)^\ua.

4739: \tag1

4740: $$

4741: \endpr

4742: \n

4743: (Recall Definitions 5.1.14 and 5.1.5,

4744: and the definition of

4745: \,$\PA\: \Au \,\totop\,\Avua$ \,in \S3.2($g$).)

4746: 

4747: \Pf

4748: Intuitively, \,\fs\ \,represents a possible implementation 

4749: of the `\chooses' operator:  \,$\fs(n)$

4750: \,is a possible value for the $n$th call of this operator

4751: in any particular implementation of $P$.

4752: We will then take the union of the interpretations 

4753: over all such possible implementations.

4754: 

4755: In more detail:

4756: the construction of \,\Sf\ \,from $S$ is as follows.

4757: Let \,\ctt\ \,be a new ``counter", \ie,

4758: an auxiliary variable of sort \,\nats\

4759: \,which is not in $S$.

4760: First, it is clear that by ``splitting up" assignments in $S$,

4761: and introducing more auxiliary \,\nats\ \,variables,

4762: we can re-write $S$ in such a way

4763: that every occurrence of the `\chooses' construct

4764: is in the context of an assignment of the form

4765: $$

4766: \ztt' := \chooses \ \ztt:b.

4767: \tag2

4768: $$

4769: where the boolean term $b$ does not contain the `\chooses' construct.

4770: Now replace each assignment of the form (2) by

4771: the pair of assignments

4772: $$

4773: \align

4774: &\ctt := \ctt+1;\\

4775: &\ifs \ \,b\ang{\ztt/\fs(\ctt)} 

4776:   \ \,\thens \ \,\ztt' := \fs(\ctt)\ \,\elses \ \,\divs

4777: \endalign

4778: $$

4779: and initialise the value of \ctt\ (at the beginning of the statement)

4780: to 0.

4781: The result is a \,\Whilex(\Sigf) \,procedure \,\Pfs\

4782: \,with a body \Sf\ \,which, for a given interpretation $f$ of \fs,

4783: ``interprets" successive executions of \,`\chooses' 

4784: by successive values of $f$, when this is possible

4785: (\ie, \,$b\ang{\ztt/f(\ctt)}$ 

4786: \,has \,\ttt\ \,as one of its values), and otherwise, 

4787: causes the execution to diverge.

4788: 

4789: For those $f$ which (for a given input)

4790: always give ``good" values for all the successive executions of \,`\chooses'

4791: assignments (2) in $S$, \,\PfA\ \,will give a possible implementation 

4792: of $P$.

4793: For all other $f$, \,\PfA\ \,will diverge.

4794: Since (for a given input)

4795: each \,\PfA\ \,{\it either\/} simulates one possible implementation 

4796: of successive executions of \,`\chooses' \,in $S$ {\it or\/} diverges,

4797: their ``union" \,\PsqcupA\ \,gives the result of 

4798: {\it all\/} possible implementations

4799: of \,`\chooses', \,plus divergence; hence the conclusion (1).

4800: \endpf

4801: 

4802: \Thmn{5.2.2}

4803: Let

4804: $$

4805: P \ \ident \ \funcs\ \ins \ \att\ \outs \ \btt\ \auxs \ \ctt

4806: \ \begins \ S\ \ends

4807: \tag3

4808: $$

4809: be a \,\WhileCC\ \,procedure,

4810: where \,$\att:u$ and $\btt:v$.

4811: Then the interpretation 

4812: $$

4813: \PA: \ \Au \ \totop \Avua

4814: $$

4815: is continuous.

4816: \endpr

4817: 

4818: \Pf

4819: In the notation of the Oracle Equivalence Lemma (5.2.1):

4820: \,\PfA\ \,is continuous for all \,$f\in\FFF$,

4821: \,by the continuity theorem for \,\While\ \,\cite[\S6.5]{tz:hb}.

4822: Hence \,\PsqcupA\ \,is continuous, by Lemma 5.1.15.

4823: Hence, by (1) and Lemma 5.1.7, so is \,\PA.

4824: \endpf

4825: 

4826: \Remarkn{5.2.3}

4827: In the special case that 

4828: \PA\ is deterministic, \ie, single-valued:

4829: $$

4830: \PA\: \Au \pto \Av,

4831: $$

4832: it follows by Remark 5.1.9 that 

4833: \PA\ is continuous according to our definition (2.5.1)

4834: of continuity for single-valued partial functions.

4835: \endpr

4836: 

4837: \newpage

4838: 

4839: \Corn{5.2.4}

4840: A \WhileCCx\ \,computable function on $A$ is continuous.

4841: \endpr

4842: 

4843: \Pf

4844: Such a function is \WhileCC\ \,computable on \Ax,

4845: hence (by Theorem 5.2.2) continuous on \Ax,

4846: and hence on $A$.

4847: \endpf

4848: 

4849: \shead{5.3}{Continuity of \WhileCCx\ \,approximable functions}

4850: Recall Definiton 3.5.1 and \S2.7.

4851: 

4852: 

4853: \Thmn{5.3.1}

4854: Let $A$ be a metric \Sig-algebra,

4855: and let \,$F\:\Au\pto\Av$.

4856: \sn

4857: ($a$) If $F$ is \WhileCCx\ approximable then $F$ is W-continuous.

4858: \sn

4859: ($b$) If also \,\dom{F} is open in \Au\ \,then $F$ is continuous.

4860: \endpr

4861: 

4862: \Pf

4863: \,Suppose $F$ is approximable on $A$ by the 

4864: \WhileCCx\ 

4865: procedure \,$P\:\nats\times\utov$.

4866: We will show that $F$ is W-continuous, using Remark 2.7.2($b$).

4867: Given \,$a\in \dom{F}$ \,and \,$\eps>0$,

4868: \,choose $N$ such that 

4869: $$

4870: 2^{-N}<{\eps/3}.

4871: \tag1

4872: $$

4873: Then by Definition 3.5.1,

4874: $$

4875: \nil \ \ne \ P^A_N(a)\ \sseq \ \Bb(F(a),\,2^{-N}).

4876: \tag2

4877: $$

4878: Choose $b\in P^A_N(a)$.

4879: By (2), 

4880: $$

4881: \ds(F(a),b)\ < \ 2^{-N}.

4882: \tag3

4883: $$

4884: By Corollary 5.2.3, $P^A_N$ is continuous on $A$, 

4885: and so by Remark 5.1.4($a$),

4886: there exists $\del>0$ such that 

4887: $$

4888: \all x\in \Bb(a,\del), \ P^A_N(x)\cap \Bb(b,\eps/3)\ne\nil.

4889: \tag4

4890: $$

4891: Take any \,$x\in \Bb(a,\del)\cap\dom{F}$.

4892: By Definition 3.5.1 again,

4893: $$

4894: P^A_N(x)\ \sseq \ \Bb(F(x),\,2^{-N})

4895: \tag5

4896: $$

4897: By (4), choose \,$y\in P^A_N(x)\cap\Bb(b,\eps/3)$.

4898: So

4899: $$

4900: \ds(y,b)\ < \ \eps/3

4901: \tag6

4902: $$

4903: and by (5) 

4904: $$

4905: \ds(F(x),y)\ <\ 2^{-N}.

4906: \tag7

4907: $$

4908: Hence

4909: $$

4910: \align

4911: \ds(F(x),F(a)) \ &\le \ \ds(F(x),y)+\ds(y,b)+\ds(b,F(a))\\

4912: &< \ \eps

4913: \endalign

4914: $$

4915: by (7), (6), (3) and (1).

4916: Part ($a$) follows by Remark 2.7.2($b$).

4917: 

4918: Part ($b$) then follows by Proposition 2.7.1.

4919: \endpf

4920: 

4921: \newpage

4922: 

4923: \itemm{\bbf 6\ \ }

4924: {\bbf Concrete computability and the soundness of \WhileCCxbig

4925: \nl

4926: computation on countable algebras}

4927: 

4928: \mn

4929: To compute on a metric algebra $A$ using a concrete model

4930: of computation,

4931: we choose a countable subspace $X$ of $A$ and an enumeration

4932: $$

4933: \al: \ \NN\ \to\ X.

4934: $$

4935: From this we build the space \,\CalX\ \,of \al-computable elements 

4936: of $A$, and enumerate it with 

4937: $$

4938: \albar: \ \NN\ \to\ \CalX.

4939: $$

4940: In this section

4941: we step back from topological algebras and consider

4942: computability on {\it arbitrary countable\/} algebras.

4943: We show that if an algebra $A$ is enumerated and

4944: its basic functions are effective, then

4945: functions that are \WhileCCx\ computable on $A$ are also effective.

4946: This result is a key lemma in the ssoundness theorem for

4947: \WhileCCx\ approximation in the next section.

4948: 

4949: \shead{6.1}{Enumerations and tracking functions for partial functions}

4950: Let 

4951: $$

4952: X \ = \ \ang{\Xs \br \sinSortSig}

4953: $$

4954: be a \SortSig-indexed family of non-empty sets.

4955: 

4956: \Defn{6.1.1}

4957: An {\it enumeration\/} of $X$ is a family 

4958: $$

4959: \al \ = \ \ang{\als\:\Oms\onto \Xs \br \sinSortSig}

4960: $$

4961: of surjective maps \,$\als\:\Oms\onto \Xs$,

4962: \,for some family 

4963: $$

4964: \Om \ = \ \ang{\Om_s \br \sinSortSig}

4965: $$

4966: of sets \,$\Oms\sseq\NN$.

4967: The family $X$ is said to be

4968: {\it enumerated by\/} \al.

4969: We say that

4970: \,$\al\: \Om\onto X$

4971: \,is an {\it enumeration\/} of $X$, 

4972: and call the pair \,\Xal\ \,an {\it enumerated family of sets\/}.

4973: (The notation `$\onto$' denotes surjections, or onto mappings.)

4974: \endpr

4975: 

4976: We also write 

4977: \,$\Oms = \Omals$

4978: \,to make explicit the fact that \,$\Oms = \dom{\als}$.

4979: 

4980: 

4981: \Defn{6.1.2 \ (Tracking and strict tracking functions)}

4982: We use the notation

4983: \nl

4984: \,$X^u = X_{s_1}\times\dots\times X_{s_m}$

4985: \,and

4986: \ $\Omalu = \Om_{\al,s_1}\times\dots\times\Om_{\al,s_m}$,

4987: \,where \,$u=\tuptimes{s}1m$.

4988: \nl

4989: Let \,$F\:\Xu\pto\Xs$ \,and \,$f:\Omalu\pto\Omals$,

4990: \itemm{($a$)}

4991: $f$ is a {\it tracking function with respect to\/} \al,

4992: \,or \,\al-{\it tracking function\/},

4993: \,for $F$,

4994: \,if the following diagram commutes:

4995: $$

4996: \commdiag{

4997: \Xu& &\mapright^{\dz F}_\bdot& &\Xs\cr

4998: \mapup\lft{\dz{\alu}}\rt\bdot& && &\mapup\lft\bdot\rt{\dz{\ \als}}\cr

4999: \NN^m& &\mapright_{\dz f}^\bdot& &\NN

5000: }

5001: $$

5002: \mn

5003: \itemm{}

5004: in the sense that

5005: \,for all \,$k\in \Omalu$

5006: $$

5007: F(\alu(k))\da \ \ \implies

5008: \ f(k)\da \,\con \,f(k)\in \Omals \,\con 

5009: \,F(\alu(k)) = \als(f(k)).

5010: $$

5011: \itemm{($b$)}

5012: $f$ is a {\it strict \al-tracking function\/} for $F$

5013: %if $f$ is an \al-tracking function for $f$ as in ($a$), and also,

5014: if in addition,

5015: for all \,$k\in\Omalu$

5016: $$

5017: f(k)\da \ \ \implies

5018: \ F(\alu(k))\da.

5019: $$

5020: 

5021: Here we use the notation

5022: \ $\alu(k) = (\al_{s_1}(k_1),\dots,$ $\al_{s_m}(k_m))$,

5023: \ where 

5024: %\,$u=\tuptimes{s}1m$ \ and

5025: \ $k = (\tup{k}1m)$.

5026: (We will sometimes drop the type super- and subscripts.)

5027: 

5028: 

5029: 

5030: \Defn{6.1.3 \ (\al-computability)}

5031: ($a$) \,Suppose $A$ is a \SortSig-family,

5032: and \,\Xal\ \,an enumerated subfamily of $A$,

5033: \,\ie, \,$\Xs \sseq A_s$ \,for all \Sig-sorts $s$.

5034: Suppose 

5035: \,$F\:\Au\pto A_s$ \,and \,$f:\NN^m\pto\NN$,

5036: such that 

5037: $$

5038: \align

5039: F\rest \Xu\:\Xu\ &\pto\ \Xs,\\

5040: f\rest \Omalu\:\Omalu\ &\pto\ \Omals,

5041: \endalign

5042: $$

5043: and \,$f\rest \Omalu$ \,is a ({\it strict\/}) \al-{\it tracking function\/}

5044: for \,$F\rest X$.

5045: We then say that $f$ is a ({\it strict\/}) 

5046: \al-{\it tracking function for\/} $F$.

5047: \sn($b$)

5048: \,Suppose now further that $f$ is a {\it computable\/} (\ie, recursive)

5049: partial function.

5050: Then $F$ is said to be ({\it strictly\/}) \al-{\it computable\/}.

5051: \endpr

5052: 

5053: \Remarksn{6.1.4}

5054: ($a$) In the situation of Definition 6.1.3,

5055: we are not concerned with the behaviour of

5056: $F$ off \Xu, or the behaviour of $f$ off \,\Omalu.

5057: \sn

5058: ($b$)

5059: For convenience, we will always assume:

5060: $$

5061: \Om_{\al,\boolss} = \curl{0,1}, \qquad \al_\boolss(0) = \fff, 

5062: \qquad \al_\boolss(1) = \ttt

5063: $$

5064: and also (when \Sig\ is N-standard):

5065: $$

5066: \Om_{\al,\natss} = \NN\ \ \tx{and \,$\al_\natss$ \,is the identity on \NN.}

5067: $$

5068: \endpr

5069: 

5070: 

5071: Assume now that

5072: $A$ is a \Sig-{\it algebra\/}

5073: and \,\Xal\ \,is a \SortSig-family of subsets of $A$,

5074: enumerated by \al.

5075:  

5076: \Defn{6.1.5 \,(Enumerated \Sig-subalgebra)}

5077: \Xal\ is said to be an {\it enumerated \Sig-subalgebra of\/} $A$ if

5078: $X$ is a \Sig-subalgebra of $A$.

5079: \endpr

5080:  

5081: \Defn{6.1.6 \,(\Sig-effective subalgebra)}

5082: Suppose $A$ is a \Sig-algebra and \Xal\ is an enumerated \Sig-subalgebra. 

5083: Then \al\ is said to be

5084: \itemm{($a$)}

5085: \Sig-{\it effective\/} if all the basic \Sig-functions on $A$

5086: are \al-computable; \,and

5087: \itemm{($b$)}

5088: {\it strictly\/} \Sig-{\it effective\/} if all the basic \Sig-functions on $A$

5089: are strictly \al-computable.

5090: \endpr

5091:  

5092: \newpage

5093: 

5094: \shead{6.2}{Soundness Theorem for surjective enumerations}

5095: For the rest of this section we will be considering the special

5096: case of \S6.1 in which the enumerated subalgebra $X$ is $A$ itself,

5097: \ie, we assume the enumeration is {\it onto\/} $A$.

5098: To emphasise this special situation, we will denote the 

5099: enumeration by 

5100: $$

5101: \be\: \Ombe\ \onto\ A,

5102: $$

5103: so that \Abe\ is our {\it enumerated \Sig-algebra\/}.

5104: 

5105: Given such an enumerated algebra \Abe\ and a function

5106: $$

5107: F\:\Au \ \pto \ A_s,

5108: $$

5109: we have two notions of computability for $F$:

5110: \itemm{($i$)} 

5111: {\it abstract\/}, \ie, \WhileCCx\ computability,

5112: as described in Section 3; \,and

5113: \itemm{($ii$)} 

5114: {\it concrete\/}, \ie, \be-computability,

5115: as in Definition 6.1.3, in the special case 

5116: that $X = A$.

5117: 

5118: We will prove a {\it soundness theorem\/} (Theorem \Ao),

5119: for these notions of abstract and concrete computability, 

5120: \ie, \,$(i)\tto(ii)$, \,assuming {\it strict effectiveness\/} of \be.

5121: 

5122: A more general soundness theorem (Theorem A), 

5123: with more general notions of abstract computability

5124: (\WhileCCx\ {\it approximability\/})

5125: and concrete computability

5126: (computability w.r.t\. the {\it computable closure\/}

5127: of an enumeration),

5128: will be proved in Section 7.

5129: 

5130: \Thmn{$\tx{\bf A}_{\bk0}$ \ (Soundness for countable algebras)}

5131: Let \,\Abe\ \,be an enumerated N-standard \Sig-algebra

5132: such that \be\

5133: is strictly \Sig-effective.

5134: If \,$F:\Au\pto A_s$

5135: \,is \,\WhileCCx\ computable on $A$, then $F$ 

5136: is strictly \be-computable on $A$.

5137: \endpr

5138: 

5139: \shead{6.3}{Proof of Soundness Theorem $\tx{\bf A}_{\bk0}$}

5140: Assume, then, that \,\Abe\ \,is an enumerated N-standard \Sig-algebra

5141: and \be\

5142: is strictly \Sig-effective.

5143: 

5144: We will show that each of the semantic functions

5145: listed in \S3.2$(a)$--$(g)$ has a computable 

5146: tracking function.

5147: More precisely, we will work, not with the semantic

5148: functions themselves, but ``localised" functions representing them

5149: (\cf\ \cite[\S4]{tz:hb}).

5150: 

5151: First we will prove a series of results of the form: 

5152: 

5153: \pr{Lemma Scheme 6.3.1}\sl

5154: For each semantic representing function

5155: $$

5156: \Phi: \ \Au \ \totop \ \Avua

5157: $$                                                                        

5158: representing one of the semantic functions

5159: listed in \S3.2$(a)$--$(g)$,

5160: \,there is a 

5161: computable tracking function w.r.t. \be,

5162: \,\ie, a function

5163: $$

5164: \ph:\ \Ombeu\ \pto\ \Ombev

5165: $$

5166: %(where \,$m = \arity{u}$ \,and \,$n=\arity{v}$),

5167: which commutes the diagram

5168: 

5169: $$

5170: \commdiag{

5171: \Au &\ & \bimapright^{\raise1pt\hbox{$\dz\Phi$}} &{}^+& \Avua\cr

5172: \mapup\lft{\dz{\beu}} &\ &&& \mapup\rt{\dz{\,\bev}}\cr

5173: %&&&&\cr

5174: \Ombeu &\ & \mapright_{\dz \ph}^\bdot && \Ombev

5175: }

5176: $$

5177: 

5178: \mn

5179: in the sense that for all \,$k,l\in\Ombeu$:

5180: $$

5181: \align

5182: \ph(k)\da l\ &\implies \ \bev(l)\,\in \,\Ph(\beu(k)),\\

5183: \ph(k)\,\ua \quad \ &\implies \ \ua\,\in \,\Ph(\beu(k)).

5184: \endalign

5185: $$

5186: \endpr

5187: 

5188: \Remarksn{6.3.2}

5189: ($a$) \,Here \,\ph\ \,is a combination ``strict tracking function"

5190: and ``selection function".

5191: We can think of \,\ph\ \,as giving one possible implementation 

5192: of \,\Ph.

5193: (Compare the representative functions for various semantic functions

5194: in \cite[\S4]{tz:hb}.)

5195: \sn

5196: ($b$) \,We are not concerned with the behaviour of \,\ph\ \,on

5197: \,$\NN^m\backslash\Ombeu$

5198: \,(where \,$m = \arity{u}$).

5199: (\Cf\ Remark 6.1.4($a$).)

5200: \endpr

5201: 

5202: Theorem \Ao\ then follows easily (\S6.5) from this lemma scheme.

5203: 

5204: \sn

5205: {\bf Proof of Lemma Scheme 6.3.1:}

5206: \ \ We proceed to prove this lemma scheme

5207: by constructing {\it concrete strict tracking functions for the semantic

5208: functions\/} in \S3.2.

5209: 

5210: Let \xtt\ be a $u$-tuple of variables,

5211: where \,$u = \tuptimes{s}1m$.

5212: Let \,$\PTermx = \PTermx(\Sig)$ \,be the class of all \Sig-terms

5213: with variables among \xtt\ only,

5214: and for all sorts $s$ of \Sig, let \,$\PTermxs = \PTermxs(\Sig)$

5215: \,be the class of such terms of sort $s$.

5216: 

5217: We consider in turn the semantic functions in \S3.2,

5218: or rather versions of these 

5219: {\it localised to\/} \xtt,

5220: \,\ie, defined only in terms of the state values on \xtt\

5221: (\cf\ \cite[\S4]{tz:hb}).

5222: For example, we localise the set 

5223: \,\StateA\

5224: \,of states on $A$ to the set 

5225: $$

5226: \StatexA \ \eqdf \ \Au

5227: $$

5228: of $u$-tuples of elements of $A$,

5229: where a tuple \,\ainAu\ represents a state \sig\

5230: (relative to \xtt) if \,$\sig[\xtt] = a$.

5231: The set \Au\ is, in turn, represented (relative to \be)

5232: by the set \,\Ombeu.

5233: 

5234: We assume an effective coding, or G\"odel numbering,

5235: of the syntax of \Sig.

5236: We use the notation

5237: $$

5238: \cnr{\PTerm_s} \eqdf \ \{\cnr{t} \mid t \in \PTerm_s \},

5239: $$

5240: etc.,  for sets of G\"odel numbers of syntactic expressions.

5241: 

5242: \mn{\bf ({\bi a}) \,Tracking of term evaluation.}

5243: \sn

5244: The function

5245: $$

5246: \PTExsA:\ \PTermxs\times\StatexA \ \totop \ \Asua

5247: $$

5248: defined by

5249: $$

5250: \PTExsA(t,a) \ = \ \tA\sig

5251: $$

5252: for any state \sig\ on $A$ such that \,$\sig[\xtt] = a$,

5253: \,is strictly tracked by a computable function

5254: $$

5255: \ptexsAbe: \ \cnr{\PTermxs}\times \Ombeu \ \pto \ \Ombes

5256: $$

5257: so that the following diagram commutes:

5258: 

5259: $$

5260: \commdiag{

5261: \PTermxs\times\StatexA&\ & \bimapright^{\raise5pt\hbox{$\dz\PTExsA$}}

5262: &{}^+&\Asua\cr

5263: \mapup\lft{\dz{\ang{\enum,\,\beu}}}&\ &&&\mapup\rt{\dz\,\bes}\cr

5264: \cnr{\PTermxs}\times\Ombeu&\ &\mapright_{\dz{\ptexsAbe}}^\bdot&&\Ombes

5265: }

5266: $$

5267: \mn

5268: (where \,\enum\ \,is the inverse of the G\"odel numbering function),

5269: in the sense that 

5270: $$

5271: \aligned

5272: \ptexsAbe(\cnr{t},k)\da l\ &\implies \ \bes(l)\,\in \,\PTExsA(t,\,\beu(k)),\\

5273: \ptexsAbe(\cnr{t},k)\,\ua \quad\ &\implies \ \ua\,\in \,\PTExsA(t,\,\beu(k)).

5274: \endaligned

5275: \tag1

5276: $$

5277: In order to construct such a representing function,

5278: we first define the {\it state variant representing function\/},

5279: \ie, a (primitive) recursive function 

5280: $$

5281: \vartxbe: \ \Ombeu\times\cnr{\Vars}\times\Ombes\ \to \ \Ombes

5282: $$

5283: such that

5284: $$

5285: \beu(\vartxbe(k,\cnr{\ytt},k_0)) \ = \ \beu(e)\curly{\ytt/\bes(k_0)}.

5286: $$

5287: for \,$k\in\Ombeu$, 

5288: \,$\ytt\in\Vars$

5289: \,and \,$k_0\in\Ombes$

5290: \,(\cf\ Definition 3.2.3($b$)).

5291: 

5292: We turn to

5293: the definition of \,$\ptexsAbe(\cnr{t},k)$.

5294: This is by induction on \,\cnr{t},

5295: \,or structural induction on \,$t\in\PTermx$,

5296: over all \Sig-sorts $s$.

5297: The cases are:

5298: 

5299: \bull

5300: $t \ident c$, \ a primitive constant.

5301: Then define

5302: $$

5303: \ptexsAbe(\cnr{t},k) \ = \ k_0

5304: \qquad \tx{where} \qquad \be(k_0) \ = \ c^A.

5305: $$

5306: (Such a $k_0$ exists by the strict \Sig-effectivity of \be).

5307: 

5308: \bull

5309: $t \ident \xtt_i$ \ for some $i=1,\dots,m$, 

5310: \,where \,$\xtt \ident \tup{\xtt}1m$.

5311: Note that \,$k=(\tup{k}1m)\in \Ombeu$.

5312: \ So define 

5313: $$

5314: \ptexsAbe(\cnr{t},k) \ = \ k_i.

5315: $$

5316: 

5317: \bull

5318: $t\ident F(\tup{t}1m)$.

5319: \ Let \,$f$ be a computable strict tracking function for $F$,

5320: which exists by the strict \Sig-effectivity of \be.

5321: Then define

5322: $$

5323: \ptexsAbe(\cnr{t},k) \ \sq 

5324: \ f(\pte\,_{\xtt,s_1}^{A,\ssbe}(\cnr{t_1},k),\,\dots,

5325: \,\pte\,_{\xtt,s_m}^{A,\ssbe}(\cnr{t_m},k))).

5326: $$

5327: From the induction hypothesis applied to \,\tup{t}1m, 

5328: \,the definition of \PTE\ (\S3.2($a$))

5329: and the fact that $f$ {\it strictly\/} tracks $F$,

5330: we can infer (1) for $t$.

5331: 

5332: \bull

5333: $t\ident \ifs(b,t_1,t_2)$. \ Define

5334: $$

5335: \ptexsAbe(t,k)\ \sq\ 

5336: \cases

5337: \ptexsAbe(t_1,k) \tif{$\pte_{\xtt,\boolss}^{A,\ssbe} (b,k) \da 1$}\\

5338: \ptexsAbe(t_2,k) \tif{$\pte_{\xtt,\boolss}^{A,\ssbe} (b,k) \da 0$}\\

5339: \ua           \tif{$\pte_{\xtt,\boolss}^A (b,k) \ \ua$}.\\

5340: \endcases

5341: $$

5342: From the induction hypothesis applied to $b$, $t_0$ and $t_1$,

5343: \,and the definition of \PTE,

5344: \,we can infer (1) for $t$.

5345: 

5346: \bull

5347: $t \ident (\chooses \ \ztt:t_0)$.

5348: \ We define \,$\ptexsAbe(\cnr{t},\,k)$

5349: \,by specifying its computation:

5350: \,find, by dovetailing (recall the discussion in \S4.1!) some $n$ such that

5351: $$

5352: \ptexsAbe(\cnr{t_0}, \,\vartxbe(k,\cnr{\ztt},n)) \ \da \ 1

5353: $$

5354: (remember, \,$\be(1) = \ttt$, by Remark 6.1.4($b$)),

5355: so that \,$\ptexsAbe(\cnr{t},\,k) =$ some such $n$,

5356: \,if it exists, and \,\ua\ \,otherwise.

5357: From the induction hypothesis applied to $t_0$,

5358: \,and the definition of \PTE,

5359: \,we can infer (1) for $t$.

5360: 

5361: \mn{\bf ({\bi b}) \,Tracking of atomic statement evaluation.}

5362: \sn

5363: Let \AtStx\ be the class of atomic statements with variables among \xtt\ only.

5364: The atomic statement evaluation function on $A$ localised to \xtt,

5365: $$

5366: \AExA: \ \AtStx \times \StatexA \ \totop \ \StatexAua,

5367: $$

5368: defined by

5369: $$

5370: \AExA(S,a) \ = \ \angg{S}^A\sig

5371: $$

5372: for any state \sig\ such that \,$\sig[\xtt] = a$,

5373: \,is strictly tracked by a computable function

5374: $$

5375: \aexAbe: \ \cnr{\AtStx}\times \Ombeu \ \pto \ \Ombeu

5376: $$

5377: so that the following diagram commutes:

5378: 

5379: $$

5380: \commdiag{

5381: \AtStx\times\StatexA&\ &\bimaprighturaise5{\AExA}&{}^+ 

5382: &\StatexAua\cr

5383: \mapupl{\ang{\enum,\,\beu}}&\ &&&\mapupr{\,\beu}\cr

5384: \cnr{\AtStx}\times\Ombeu&\ &\maprightd{\aexAbe}^\bdot&&\Ombeu

5385: }

5386: $$

5387: \mn

5388: in the sense that

5389: $$

5390: \aligned

5391: \aexAbe(\cS,k)\da l\ &\implies \ \be(l)\,\in \,\AExA\,(S,\,\be(k)),\\

5392: \aexAbe(\cS,k)\,\ua \quad\ &\implies \ \ua\,\in \,\AExA\,(S,\,\be(k)).

5393: \endaligned

5394: \tag2

5395: $$

5396: The definition of \,$\aexAbe(\cS,k)$ 

5397: \,is given by:

5398: $$

5399: \align

5400: \aexAbe(\cnr{\skips},\,k) \ &\da \ k\\

5401: \aexAbe(\cnr{\divs},\,k) \ &\ \ua\\

5402: \aexAbe(\cnr{\ytt:=t},\,k) \ &\sq 

5403:   \cases

5404:      \vartxbe\,(k,\ytt,l)\tif{$\ptexsAbe(s\cnr{t},k)\da l$}\\

5405:      \ua  \tif{$\ptexsAbe(\cnr{t},k)\ \ua$.}

5406:   \endcases\\

5407: \endalign

5408: $$

5409: Using (1) and the definition of \AExA\ (\S3.2($b$)),

5410: we can infer (2).

5411: 

5412: \mn{\bf ({\bi c}) \,Tracking of \,\First\ \,and \,\Rest\ \,operations.}

5413: \sn

5414: Let \Stmtx\ be the class of statements with variables

5415: among \,\xtt\ \,only. 

5416: Consider the functions \,\First\ \,and \,\RestA\ \,(\S3.2($c$)).

5417: Then \,\First\ \,is

5418: strictly tracked by a computable function

5419: $$

5420: \first \ : \cnr{\Stmt}\ \to\ \cnr{\AtSt}

5421: $$

5422: defined on G\"odel numbers in the obvious way,

5423: so that the following diagram commutes:

5424: 

5425: $$

5426: \commdiag{

5427: \Stmt&\ &\maprightu{\First}&\ &\AtSt\cr

5428: \mapupl{\enum}&&&&\mapupr{\enum}\cr

5429: \cnr{\Stmt}&&\maprightd{\first}&&\cnr{\AtSt}

5430: }

5431: $$

5432: 

5433: \mn

5434: (Note that \,\first, \,unlike most of the other representing functions here,

5435: does not depend on \,\StatexA, \,or, indeed, on $A$ or \,\xtt.)

5436: Next, the localised version of \,\RestA:

5437: $$

5438: \RestxA :

5439: \ \Stmtx\times\StatexA \ \totop \ \Stmtx

5440: $$

5441: defined by

5442: $$

5443: \RestxA(S,a) \ = \ \RestA(S,\sig)

5444: $$

5445: for any state \sig\ such that \,$\sig[\xtt] = a$,

5446: \,is strictly tracked by a computable function

5447: $$

5448: \restxAbe : \ \cnr{\Stmtx}\times\Ombeu\ \pto\ \cnr{\Stmtx}

5449: $$

5450: so that the following diagram commutes:

5451: 

5452: $$

5453: \commdiag{

5454: \Stmtx\times\StatexA&\ &\bimaprighturaise5{\RestxA}&{}^+&\Stmtx\cr

5455: \mapupl{\ang{\enum,\,\beu}}&&&&\mapupr{\enum}\cr

5456: \cnr{\Stmtx}\times\Ombeu&&\maprightud{\bdot}{\restxAbe}&&\cnr{\Stmtx}

5457: }

5458: $$

5459: 

5460: \mn

5461: in the sense that 

5462: $$

5463: \aligned

5464: \restxAbe\,(\cS,k)\da \cnr{S'}\ \implies 

5465: \ &S'\,\in \,\RestA\,(S,\,\be(k)),\\

5466: \restxAbe\,(\cS,k)\,\ua \quad\ \implies \ &\divs\,\in \,\RestA\,(S,\,\be(k))

5467: \endaligned

5468: \tag3

5469: $$

5470: The definition of

5471: \,$\restxAbe(\cS,\,k)$, \,as well as the proof of (3),

5472: are by

5473: induction on \,\cS, \, or structural induction on $S$.

5474: 

5475: \bull

5476: $S$ is atomic. \ Then

5477: $$

5478: \restxAbe (\cS,\,k) \ = \ \cnr{\skips}.

5479: $$

5480: \bull

5481: $S \ \ident \ S_1;S_2$.  \ Then

5482: $$

5483: \restxAbe(\cS,\,k) \ = \ 

5484: \cases

5485: \quad\cnr{S_2} \tif{$S_1$ is atomic}\\

5486: \quad\cnr{\restxAbe(S_1,\,k);S_2} \ow

5487: \endcases

5488: $$

5489: \bull

5490: $S \ \ident \ \ifs \ b \ \thens \ S_1 \ \elses \ S_2 \ \fis$.

5491: \ Then 

5492: $$

5493: \restxAbe (\cS,\,k) \ \sq \

5494: \cases

5495: \cnr{S_1} \tif{$\pteboolsAbe(b,k)\da 1$}\\

5496: \cnr{S_2} \tif{$\pteboolsAbe(b,k)\da 0$}\\

5497: \ua \tif{$\pteboolsAbe(b,k)\ \ua$}.

5498: \endcases

5499: $$

5500: \bull

5501: $S \ \ident \ \whiles\ b \ \dos \ S_0 \ \ods$.

5502: \ Then 

5503: $$

5504: \restxAbe(S,k) \ \sq\

5505: \cases

5506: S_0;S \tif{$\pteboolsAbe(b,k)\da1$},\\

5507: \skips \tif{$\pteboolsAbe(b,k)\da0$},\\

5508: \ua \tif{$\pteboolsAbe(b,k)\ \ua$}.\\

5509: \endcases

5510: $$

5511: 

5512: 

5513: 

5514: \mn{\bf ({\bi d}) \,Tracking of a computation step.}

5515: \sn

5516: The computation step function (\S3.2($d$)) localised to \,\xtt:

5517: $$

5518: \CompStepxA :\  \Stmtx\times\StatexA \ \totop \ \StatexA^\ua

5519: $$

5520: defined by

5521: $$

5522: \CompStepxA(S,a) \ = \ \CompStepA(S,\sig)

5523: $$

5524: for any state \sig\ such that \,$\sig[\xtt] = a$,

5525: \,is represented by the computable function

5526: $$

5527: \compstepxAbe: \ \cnr{\Stmtx}\times\Ombeu \ \pto \ \Ombeu

5528: $$

5529: defined by

5530: $$

5531: \compstepxAbe(\cS,k) \ \sq \ \aexAbe(\first(\cS),\,k).

5532: $$

5533: This makes the following diagram commute:

5534: 

5535: $$

5536: \commdiag{

5537: \Stmtx\times\StatexA&\ &\bimaprighturaise5{\CompStepxA}&{}^+ &\StatexA^\ua\cr

5538: \mapupl{\ang{\enum,\,\beu}}&\ &&&\mapupr{\,\beu}\cr

5539: \cnr{\Stmtx}\times\Ombeu&\ &\maprightud{\bdot}{\compstepxAbe}&&\Ombeu

5540: }

5541: $$

5542: \mn

5543: in the sense that

5544: $$

5545: \aligned

5546: \compstepxAbe\,(\cS,k)\da l\ &\implies 

5547:   \ \be(l)\,\in \,\CompStepxA\,(S,\,\be(k)),\\

5548: \compstepxAbe\,(\cS,k)\,\ua \quad\ &\implies 

5549:   \ \ua\,\in \,\CompStepxA\,(S,\,\be(k).

5550: \endaligned

5551: \tag4

5552: $$

5553: This is proved easily from the definitions and (2).

5554: 

5555: \mn{\bf ({\bi e}) \,Tracking of a computation sequence.}

5556: \sn

5557: Now consider localised versions of the computation tree stage

5558: and computation tree of \S3.2.($e$):

5559: $$

5560: \align

5561: \CompTreeStagexA:\ \Stmtx\times\StatexA\times\NN\ 

5562: &\to\ \PPP((\StatexAua)^{<\om})\\

5563: \CompTreexA:\ \Stmtx\times\StatexA \ &\to \ \PPP((\StatexAua)^{\le\om})

5564: \endalign

5565: $$

5566: We will define a function which {\it selects a path

5567: through the computation tree\/}:

5568: $$

5569: \compseqxAbe: \ \cnr{\Stmtx}\times\Ombeu\times\NN\ \pto 

5570:   \Ombeu\cup \curly{\cnr*}

5571: $$

5572: (where `$*$' is a symbol meaning ``already terminated")

5573: by recursion on $n$:

5574: $$

5575: \align

5576: &\compseqxAbe(\cS,\,k,\,0) \ =\ e \\

5577: &\compseqxAbe(\cS,\,k,\,n+1)\ \sq \\

5578: &\qqquad\cases

5579:   \cnr* \ \ \tx{if $S$ is atomic and $n>0$ and 

5580:       \,$\compseqxAbe(\cS,\,k,\,n)\da$}\\

5581:   \ \ua \ \quad\tx{if $S$ is atomic and $n>0$ and 

5582:      \,$\compseqxAbe(\cS,\,k,\,n)\ \ua$}\\

5583:   \compseqxAbe(\restxAbe(\cS,k),\ \compstepxAbe(\cS,k),\ n)\\

5584:   \qquad \ \tx{otherwise.}

5585: \endcases

5586: \endalign

5587: $$

5588: (This is a ``tail recursion": \ compare definition of

5589: \,\CompIA\ \,in \cite[\S3.4]{tz:hb}.)

5590: 

5591: Writing \,$k_n = \compseqxAbe(\cS,k,n)$,

5592: \,this defines a 

5593: (concrete) {\it computation sequence\/} 

5594: $$

5595: \kbar \ = \ k_0,\ k_1,\ k_2,\ \dots

5596: $$

5597: for $S$ from the 

5598: initial state $k= k_0$.

5599: (Our notation here includes the possibility that 

5600: some of the $k_i$ may be \,\cnr{*} \, or \,\ua.)

5601: As can easily be checked,

5602: there are three possibilities for \,\ebar\

5603: \,(compare the discussion in \S3.2($e$)):

5604: 

5605: \itemm{($i$)}

5606: For some $n$, \,$k_i \in \Ombeu$ \,for all $i\le n$ 

5607: \,and \,$k_i = *$ \,for all $i>n$.

5608: This represents a computation which {\it terminates\/} at stage $n$,

5609: with {\it final state\/} \,$k_n$.

5610: 

5611: \itemm{($ii$)}

5612: For some $n$, \,$k_i \in \Ombeu$ \,for all $i< n$ 

5613: \,and \,$k_i = \ua$ \,for all $i\ge n$.

5614: This represents a {\it non-terminating\/} computation, with

5615: {\it local divergence\/} at stage $n$.

5616: 

5617: \itemm{($iii$)}

5618: For all $i$, \,$k_i\in \Ombeu$.

5619: This represents {\it non-terminating\/} computation, 

5620: with {\it global divergence\/}.

5621: 

5622: We write 

5623: \,$\kbarn = $ the initial segment \,$k_0,k_1,\dots,k_n$,

5624: \,with length \,$\lgth{\kbarn} = n+1$.

5625: We put \,$\lgth{\kbar}=\infty$.

5626: The \,$k_i$ \,are called {\it components\/} of \,\kbar,

5627: \,and of \,\kbarn, \,for all $i\le n$.

5628: 

5629: The computation sequence \,\kbar\ \,then has the following connection 

5630: with the computation tree \,\CompTreexA.

5631: Extend (for now) the definition of \,\be\ \,by

5632: \,$\be(\cnr{*}) = *$,

5633: \,$\be(\ua) = \ua$,

5634: \,and

5635: $$

5636: \align

5637: \be(\kbar) \ &\eqdf\ \be(k_0),\ \be(k_1),\ \be(k_2),\ \dots\\

5638: \be(\kbarn) \ &\eqdf\ \be(k_0),\ \be(k_1),\ \be(k_2),

5639: \ \dots, \ \be(k_n).

5640: \endalign

5641: $$

5642: Let \,$\tau = \CompTreexA(S,\,\be(k))$.

5643: Then

5644: 

5645: \itemm{($i$)}

5646: If the computation sequence \,\kbar\ \,terminates at stage $n$,

5647: then \,\be(\kbarn) \,is a path through \,\ta\

5648: \,from the root to a leaf ($= \be(k_0)$, \,the final state).

5649: 

5650: \itemm{($ii$)}

5651: If for some (smallest) $n$, \,$k_n=\ua$, 

5652: \,then \,\be(\kbarn) \,is a path through \,\ta\

5653: \,from the root to a leaf ($=\ua$, \,local divergence).

5654: 

5655: \itemm{($iii$)}

5656: If for all $n$, \,$k_n\in \Ombeu$,

5657: \,then \,\be(\kbar) is an infinite path through \,\ta\

5658: \,(global divergence).

5659: 

5660: To prove this, we first define

5661: an initial segment of \,\kbar\

5662: \,(including \,\kbar\ \,itself) to be {\it acceptable\/}

5663: if ($i$) no component is equal to `$*$', \,and

5664: \,($ii$) no component, except possibly the last, is equal to \,\ua.

5665: Further, an acceptable initial segment of \,\kbar\ \,is 

5666: {\it maximal (acceptable)\/}

5667: if it has no acceptable extension.

5668: Thus if \,\kbar\ \,is acceptable, it is automatically maximal.

5669: If \,\kbarn\ \,is acceptable,

5670: it is maximal acceptable provided

5671: either \,$k_{n+1}= *$ \,or \,$k_n = \ua$.

5672: We then show:

5673: 

5674: \Lemman{6.3.3}

5675: Given a computation sequence \,$\kbar = k_0,k_1,\dots$

5676: \,for \,\cS\ \,from $k$,

5677: where \,$k_n = \compseqxAbe(\cS,\,k,\,n)$,

5678: \,let \,$\tau = \CompTreexA(S,\,\be(k))$.

5679: Then

5680: with every acceptable initial segment \,\kbarn\ \,of \,\kbar,

5681: \,$\be(\kbarn)$

5682: \,is a path through \,\ta\

5683: \,from the root.

5684: If \,\kbarn\ \,is maximal,

5685: then \,$\be(k_n)$ \,is a leaf.

5686: \endpr

5687: 

5688: \Pf

5689: Put \,$\taun = \CompTreeStagexA(S,\,\be(k_0),\,n)$.

5690: The proof is by induction on $n$,

5691: comparing the inductive definitions of \,$k_n$

5692: \,and \,\taun.

5693: 

5694: \sn

5695: {\it Basis:\/} \ $n=0$.

5696: This is immediate from the definitions:

5697: \,$k_0 = k$, \,and \,$\tau[0] = \curly{\be(k_0)}$.

5698: 

5699: \sn

5700: {\it Induction step:\/}

5701: Assume the induction hypothesis holds for the initial segment

5702: of length $n$ of the computation sequence for \,\cnr{S'} 

5703: \,from $k_1$, where 

5704: $$

5705: \align

5706: S' \ &= \ \restxAbe(\cS,\,\be(k)),\\

5707: e_1 \ &= \ \compseqxAbe(\cS,\,k,\,1)\\

5708:      &\sq\ \compseqxAbe(\restxAbe(\cS,k),\,\compstepxAbe(\cS,k),\,0)\\

5709:      &\sq\ \compstepxAbe(\cS,e)

5710: \endalign

5711: $$

5712: \ie, assume the induction hypothesis for the segment \,\lbar\

5713: \,of length $n$:

5714: $$

5715: l_0,\ l_1,\ l_2,\ \dots\ ,l_n

5716: $$

5717: where \,$l_i = e_{i+1}$ \,($i=1,\dots,n$).

5718: Now apply the inductive definitions for \,$\compseqxAbe (\cS,$ $\,k, \,n+1)$

5719: \,(above) and \,$\CompTreeStagexA(S,\,\be(k),\,n+1)$

5720: \,(\S3.2($e$)), and use (3) and (4).

5721: 

5722: \mn{\bf ({\bi f}) \,Tracking of statement evaluation.}

5723: \sn

5724: First we need a 

5725: constructive computation length function 

5726: $$

5727: \complengthxAbe:\ \cnr{\Stmtx}\times\Ombeu\ \pto\ \NN

5728: $$

5729: by (\cf\ \cite[\S3.4]{tz:hb})

5730: $$

5731: \complengthxAbe(\cS,\,k) \ \sq \ \mu n [\compseqxAbe(\cS,\,k,\,n+1)\da *\,]

5732: $$

5733: \ie, the least $n$ (if it exists) such that for all $i\le n$, 

5734: \,$\compseqxAbe(\cS,\,k,\,i)\da\ \ne *$ \,and\

5735: \,$\compseqxAbe(\cS,\,k,\,n+1)\da*$.

5736: 

5737: Thus \,$\complengthxAbe(\cS,\,k)$ \,is undefined (\ua) 

5738: in the case of local or global divergence

5739: of the computation sequence for \,\cS\ \,from $k$.

5740: 

5741: Now the statement evaluation function (\S3.2($f$)) localised to \,\xtt:

5742: $$

5743: \SExA: \ \Stmtx\times\StatexA\ \ \totop\ \StatexAua

5744: $$

5745: defined by 

5746: $$

5747: \SExA(S,a) \ = \ \SA(\sig)

5748: $$

5749: for any state \sig\ such that \,$\sig[\xtt] = a$,

5750: \,is strictly tracked by the computable function

5751: $$

5752: \sexAbe: \ \cnr{\Stmtx}\times\Ombeu\ \ \pto\ \Ombeu

5753: $$

5754: defined by

5755: $$

5756: \sexAbe(\cS,\,k) \ \sq\ \compseqxAbe(\cS,\,k,\,\complengthxAbe(\cS,\,k)).

5757: $$

5758: This makes the following diagram commute:

5759: 

5760: $$

5761: \commdiag{

5762: \Stmtx\times\StatexA&\ &\bimaprighturaise5{\SExA}&{}^+&\StatexA^\ua\cr

5763: \mapupl{\ang{\enum,\,\beu}}&\ &&&\mapupr{\,\beu}\cr

5764: \cnr{\Stmtx}\times\Ombeu&\ &\maprightud{\bdot}{\sexAbe}&&\Ombeu

5765: }

5766: $$

5767: \mn

5768: in the sense that

5769: $$

5770: \aligned

5771: \sexAbe\,(\cS,k)\da l\ &\implies \ \be(l)\,\in \,\SExA\,(S,\,\be(k)),\\

5772: \sexAbe\,(\cS,k)\,\ua \quad\ &\implies \ \ua\,\in \,\SExA\,(S,\,\be(k)).

5773: \endaligned

5774: \tag5

5775: $$

5776: This result is clear from the definition of

5777: \,\complength\ \,and Lemma 6.3.1.

5778: 

5779: \newpage

5780: 

5781: \mn{\bf ({\bi g}) \,Tracking of procedure evaluation.}

5782: \sn

5783: For a specific triple of lists of variables 

5784: \,$\att:u, \,\btt:v,\,\ctt:w$,

5785: \,let \,\Procabc\ \,be the class of all \WhileCCx\ \,procedures

5786: of type \,\utov, 

5787: \,with declaration

5788: \ `$\ins \ \att\ \outs \ \btt\ \auxs \ \ctt$'.

5789: The procedure evaluation function (\S3.2($g$))

5790: localised to this declaration:

5791: $$

5792: \PEabcA: \ \Procabc\times\Au \totop\ \Avua

5793: $$

5794: defined by 

5795: $$

5796: \PEabcA(P,\,a) \ = \ \PA(a),

5797: $$

5798: is strictly tracked by the computable function

5799: $$

5800: \peabcAbe: \ \cnr{\Procabc}\times\Ombeu\ \pto\ \Ombev

5801: $$

5802: defined by the following algorithm.

5803: Let \,$P\in \Procabc$; \,say

5804: $$

5805: P \ \ident \ \procs\ \ins \ \att\ \outs \ \btt\ \auxs \ \ctt\

5806: \begins \ S\ \ends

5807: $$

5808: and let \,$k_0\in\Ombeu$.

5809: Take any \,$k_1\in\Ombev$

5810: \,and

5811: \,$k_2 \in \Ombew$.

5812: \,(The choice of $k_1$ and $k_2$ is irrelevant,

5813: by Remark 3.2.4.)

5814: Put

5815: \,$k\ident k_0,k_1,k_2$ \,and put

5816: \,$\xtt \ident \att,\btt,\ctt$.

5817: Compute 

5818: \,$\sexAbe(\cS,k)$.  Suppose this converges to 

5819: \,$l\ident \,l_0,l_1,l_2$, \,where 

5820: \,$l_0\in \Ombeu$,

5821: \,$l_1\in \Ombev$

5822: \,and

5823: \,$l_2\in \Ombew$.

5824: Then we define \,$\peabcAbe(\cP,\,k_0) \da l_1$.

5825: The following diagram then commutes:

5826: 

5827: $$

5828: \commdiag{

5829: \Procabc\times\Au&\ &\bimaprighturaise5{\PEabcA}&^+&\Avua\cr

5830: \mapupl{\ang{\enum,\,\beu}}&\ &&&\mapupr{\,\bev}\cr

5831: \cnr{\Procabc}\times\Ombeu&\ &\maprightud{\bdot}{\peabcAbe}&&\Ombev

5832: }

5833: $$

5834: \mn

5835: in the sense that 

5836: $$

5837: \aligned

5838: \peabcAbe\,(\cP,k)\da l\ &\implies \ \be(l)\,\in \,\PEabcA\,(P,\,\be(k)),\\

5839: \peabcAbe\,(\cP,k)\,\ua \quad\ &\implies \ \ua\,\in \,\PEabcA\,(P,\,\be(k)).

5840: \endaligned

5841: \tag6

5842: $$

5843: This is proved from (5)

5844: and the definitions of \,\PE\ \,and \,\pe.

5845: 

5846: This concludes the proof of Lemma Scheme 6.3.1.

5847: \endpf

5848: 

5849: 

5850: \mn

5851: {\bf Proof of Theorem $\tx{\bf A}_{\bk 0}$ \,(conclusion):}

5852: \ \ Suppose 

5853: \,$F:\Au\pto A_s$

5854: \,is \WhileCCx\ computable on $A$.

5855: Then there is a deterministic \WhileCCx\ \,procedure (Definitions 3.2.5/6)

5856: $$

5857: P\: u\ \to \ s

5858: $$

5859: such that for all \ainAu,

5860: $$

5861: \align

5862: F(x) \da y \ &\impp \ \PA(x) = \{y\},\\

5863: F(x) \ \ua \,\ \ \ &\impp \ \PA(x) = \{\ua\}.

5864: \endalign

5865: $$

5866: Hence by ($g$) (above) there is a computable (partial) function 

5867: $$

5868: f\: \Ombeu\ \pto\ \Ombes

5869: $$

5870: which strictly tracks $F$, as required.

5871: \endpf

5872: 

5873: \newpage

5874: 

5875: \Shead7{\bbf Soundness of \WhileCCxbig\ \,approximation}

5876: In this section we address the general situation 

5877: introduced in \S6.2, of a partial metric \Sig-algebra $A$

5878: with an enumerated subalgebra \Xal,

5879: and prove a more general soundness theorem (Theorem A) for 

5880: \WhileCCx\ approximation.

5881: 

5882: 

5883: \shead{7.1}{Enumerated subspace of metric algebra; \,Computational closure}

5884: Let $A$ be an N-standard metric \Sig-algebra,

5885: and  \Xal\ an enumerated \SortSig-family 

5886: \,\ang{\Xsal\mid\sinSortSig}

5887: \,of subsets \,$X_s\sseq A_s$ \,(\sinSortSig).

5888: Each \Xs\ can be viewed as a {\it metric subspace\/}

5889: of the metric space \As.

5890: We call \Xal\ a \SortSig-{\it enumerated (metric) subspace\/} of $A$.

5891: 

5892: %Assume now for convenience that \,$\Omals = \NN$ \,for all sorts $s$.

5893: We define from \Xal\ a family

5894: $$

5895: \CalX \ = \ \ang{\CalXs \br \sinSortSig}

5896: $$

5897: of sets \,\CalXs\ \,of \al-{\it computable elements of\/} \,\As,

5898: \,\ie, limits in \As\ of effectively convergent Cauchy sequences 

5899: (to be defined below)

5900: of elements of \,\Xs,

5901: \,so that

5902: $$

5903: \Xs\ \sseq\ \CalXs\ \sseq \ \As,

5904: $$

5905: with corresponding enumerations

5906: $$

5907: \albars:\ \Omalbars \ \onto \ \CalXs.

5908: $$

5909: Writing \,$\albar = \ang{\albars\mid\sinSortSig}$,

5910: we call the enumerated subspace \,\CalXalbar\

5911: \,the {\it computable closure\/} of \Xal\ in $A$.

5912: 

5913: We will generally be interested in (strictly) \albar-computable

5914: (rather than \al-computable)

5915: functions on $A$ (\cf\ Definition 6.1.3),

5916: as our more general model of {\it concrete computability\/} on $A$.

5917: 

5918: The sets \,$\Omalbars\sseq\NN$

5919: consist of

5920: {\it codes\/} for \,\CalXs\ \,(w.r.t\. \al),

5921: \,\ie, \,pairs of numbers

5922: \,$c = \ang{e,m}$ \,where

5923: \itemm{($i$)}

5924: $e$ is an index for a total recursive function defining a 

5925: sequence \,$\al\circ\curl{e}$ \,in \Xs, \ie, the sequence

5926: $$

5927: \als(\{e\}(0)),\ \als(\{e\}(1)), \ \als(\{e\}(2)), \ \dots\ ,

5928: \tag1

5929: $$

5930: of elements of \Xs,

5931: \itemm{($ii$)}

5932: $m$ is an index for a modulus of convergence for this sequence:

5933: $$

5934: \all k, l\ge\{m\}(n): 

5935: \ \ds_i(\al(\{e\}(k)),\,\al(\{e\}(l))) < 2^{-n} .

5936: \tag2

5937: $$

5938: \sn

5939: For any such code \,$c=\ang{e,m}\in\Omalbars$,

5940: \,$\albars(c)$ \,is defined as the limit in \As\ of the Cauchy sequence (1),

5941: and \,\CalXs \,is the range of \,\albars:

5942: $$

5943: \commdiag{

5944: \Xs&\quad\sseq\quad&\CalXs&\ \ \sseq\ \ &A\cr

5945: \mapup\lft{\dz{\als}}&&\mapup\lft{\dz{\albars}}\cr

5946: \Omals&\   &\Omalbars

5947: }

5948: $$

5949: 

5950: 

5951: \Remarkn{7.1.1}

5952: We may assume, when convenient, 

5953: that the modulus of convergence for a given code is the {\it identity\/},

5954: \ie, replace (2) by the simpler condition

5955: $$

5956: \all k, l\ge n: 

5957: \ \ds_i(\al(\{e\}(k)),\,\al(\{e\}(l))) < 2^{-n} .

5958: $$

5959: or, equivalently,

5960: $$

5961: \all k > n: 

5962: \ \ds_i(\al(\{e\}(k)),\,\al(\{e\}(n))) < 2^{-n} .

5963: \tag3

5964: $$

5965: This is because any code \,$c=\ang{e,m}$ \,satisfying (2)

5966: may be effectively replaced by a code

5967: for the same element of \,\CalXs\ \,satisfying (3),

5968: namely \,$c'=\ang{e',m_1}$,

5969: \,where $m_1$ is a standard code for the identity function on \,\NN,

5970: \,and \,$e'=\comp(e,m)$,

5971: \,where \,$\comp(x,y)$ \,is a primitive recursive function

5972: for ``composition" of (indices of) computable functions, \ie,

5973: \,$\curly{\comp(e,m)}(x) \sq \curly{e}(\curly{m}(x))$.

5974: 

5975: In case of a code \,$c=\ang{e,m_1}$

5976: \,satisfying (3), the sequence (1)

5977: is called a {\it fast (\al-effective) Cauchy sequence\/}.

5978: In such a case

5979: we will often, for simplicity,

5980: refer to $e$ itself as the ``code",

5981: and the argument of \,\albars.

5982: In this way we will shift between ``$c$-codes" and ``$e$-codes" 

5983: as convenient.

5984: \endpr

5985: 

5986: \Remarkn{7.1.2}

5987: In the case \,$s = \nats$,

5988: \,we can simply take \,$\Om_{\albar,\natss} = \Om_{\al,\natss} = \NN$,

5989: and \,$\albar_\natss$ \,and \,$\al_\natss$

5990: \,as the identity mappings on \,\NN.

5991: Similarly, in the case \,$s=\bools$, \,we can take

5992: \,$\Om_{\albar,\boolss} = \Om_{\al,\boolss} = \curly{0,1}$,

5993: with \,$\albar(0) = \al(0) = \fff$

5994: \,and

5995: \,$\albar(1) = \al(1) = \ttt$.

5996: (\Cf\ Remark 6.1.3($b$).)

5997: \endpr

5998: 

5999: \Remarkn{7.1.3 \,(Closure of \alb-computability operation)}

6000: The subspace \,\CalXalbar\ \,is ``computationally closed in $A$",

6001: in the sense that 

6002: the limit of a (fast) \albar-effective Cauchy sequence

6003: of elements of \,\CalX\ \,is again in \CalX,

6004: \,\ie, \,$\Calbar(\CalX) = \CalX$.

6005: %iterating the operation \,$\al\mapsto\albar$

6006: %\,does not add new elements to \CalX.

6007: %In other words, 

6008: ({\it Easy exercise.\/})

6009: \endpr

6010: 

6011: \Remarkn{7.1.4}

6012: We will usually assume that \Omals\ is decidable, 

6013: in fact, that \,$\Omals = \NN$ \,for all sorts $s$, \,which is 

6014: typical in practice, unlike the case for \Omalbar.

6015: (See the following Example.)

6016: \endpr

6017: 

6018: \Examplen{7.1.5 \,(Constructible reals)}

6019: The best known nontrivial example of an enumerated subspace \Xal,

6020: and its extension to a subspace of \al-computable elements,

6021: is the following.  Let $A$ be the metric algebra \RRR\ of reals 

6022: (Example 2.6.1),

6023: with signature \Sig.

6024: Let \,$X_\realss$ \,be the set of rationals \,$\QQ\sset\RR$, 

6025: \,let \,$\Omalreal = \NN$ 

6026: \,and let

6027: $$

6028: \alreal\: \NN\ \to\ \QQ

6029: $$

6030: be a canonical enumeration of \QQ.

6031: Then \,$\CalQQ \eqdf \Cals(X)_\realss\sset \RR$ 

6032: \,is the subspace of {\it recursive\/} or 

6033: {\it constructible reals\/}.

6034: Note that it is a {\it subfield\/} of \RR, and

6035: hence \,\CalX\ \,is a {\it subalgebra\/} of \RRR.

6036: Further, it is easily verified that \albar\ is strictly \SigRRR-effective.

6037: (\Cf\ Definition 6.1.6.)

6038: Note that \,$\Omalreal = \NN$,

6039: \,unlike \,\Omalbarreal, \,which, by contrast, is non-recursive.

6040: (See the previous Remark.)

6041: \endpr

6042: 

6043: 

6044: \Remarkn{7.1.6 \ (Extension of enumeration to \Ax)}

6045: Given an enumeration \al\

6046: of a \Sig-subspace $X$

6047: \,of $A$, we can extend this canonically to an enumeration \alx\

6048: of a \Sigx-subspace \Xx\ \,of \Ax.

6049: ({\it Easy exercise\/}.)

6050: This in turn generates an enumeration \,\albarx\

6051: \,of a \Sigx-subspace \CalXx\

6052: \,of \alx-computable elements of \Ax.

6053: It is easy to see that 

6054: \itemm{($i$)}

6055: if \CalX\ is an \Sig-subalgebra of $A$, then \,\CalXx\ \,is a 

6056: \Sigx-subalgebra of \Ax;

6057: \itemm{($ii$)}

6058: if \albar\ is (strictly) \Sig-effective, then \,\albarx\ \,is 

6059: (strictly) \Sigx-effective.

6060: \sn

6061: We will usually use this extension (of \Xal\  and  \,\CalXalbar)

6062: to \Ax\ implicitly,

6063: \ie, writing `\al' instead of `\alx' etc.

6064: %without displaying the superscript `\str'.

6065: \endpr

6066: 

6067: \shead{7.2}{Soundness Theorem for effective numberings}

6068: We now prove the first main theorem mentioned in the Introduction.

6069: 

6070: \Thmn{A \ (Soundness)}

6071: Let $A$ be an N-standard metric \Sig-algebra,

6072: and \Xal\ an enumerated \SortSig-subspace.

6073: Suppose the enumerated \SortSig-space \CalXalbar\ of

6074: \al-computable elements of $A$ is a \Sig-subalgebra of $A$,

6075: and \albar\ is strictly \Sig-effective.

6076: If \,$F:\Au\pto \As$

6077: \,is \,\WhileCCx-approximable on $A$, then $F$

6078: is \albar-computable on $A$.

6079: \endpr                                                                          

6080: 

6081: \Pf

6082: The proof uses the Soundness Theorem $A_0$

6083: (Section 6), or rather the Lemma Scheme 6.4.1

6084: (specifically, part ($g$) of the proof)

6085: applied to the enumerated subalgebra \,\CalXalbar\ 

6086: \,in place of \Abe.

6087: 

6088: So suppose 

6089: \,$F:\Au\pto \As$

6090: \,is effectively uniformly \WhileCCx\ \,approximable on $A$.

6091: Then there is a \WhileCCxSig\ \,procedure

6092: $$

6093: P:\nats\times u\to s

6094: $$

6095: such that

6096: for all \,\ninNN\ \,and \,all $x\in\dom{F}$:

6097: $$

6098: \ua\,\notin\,\PAn(x)\ \sseq\ \Bb(F(x),\,2^{-n}).

6099: \tag1

6100: $$

6101: (see Definition 3.5.1).

6102: By \S6.4($g$) (applied to \,\CalXalbar\ \,in place of \Abe)

6103: there is a computable function 

6104: $$

6105: f:\ \NN\times\Omalbaru\ \pto\ \Omalbars

6106: $$

6107: which tracks \PA\ strictly,

6108: in the sense that for all \,\ninNN, \,$e\in\Omalbaru$

6109: \,and $e'\in\Omalbars$

6110: \,(and writing $f_n = f(n,\,\cdot\,)$):

6111: $$

6112: \aligned

6113: f_n(e)\da e' \ &\implies\ %e'\in\Omalbars\,\con\,

6114:   \albar(e')\in \PAn(\albar(e)),\\

6115: f_n(e)\ \ua \ \ \ \ &\implies\ \ua\in \PAn(\albar(e)).

6116: \endaligned

6117: \tag2

6118: $$

6119: We will show how to define a partial recursive \albar-tracking function 

6120: $$

6121: g: \ \Omalbaru\ \to \ \Omalbars

6122: $$

6123: for $F$ as follows. 

6124: 

6125: Given any $e\in\Omalbaru$, 

6126: suppose \,$\albar(e)\in\dom{F}$, \,\ie,

6127: $$

6128: F(\albar(e))\da \ \in \As.

6129: \tag3

6130: $$

6131: We must show how to define an \albar-tracking function $g$ for $F$,

6132: \ie, such that

6133: $$

6134: g(e)\in \Omalbars \qquad\tx{and}

6135: \qquad \albar(g(e)) \ = \ F(\albar(e)).

6136: \tag4

6137: $$

6138: By (1), for all $n$

6139: $$

6140: \ua \,\notin \, \PAn(\albar(e)) \ \sseq\ \Bb(F(\albar(e)),\,2^{-n}).

6141: \tag5

6142: $$

6143: Hence by (2), for all $n$

6144: %$$

6145: %f_n(e)\da \,\in\,\Omalbars \qquad\tx{and}\qquad 

6146: %  \albar(f_n(e))\,\in\,\PAn(\albar(e)).

6147: %\tag6

6148: %$$

6149: $$

6150: f_n(e)\da \,\in\,\Omalbars

6151: \tag{6$a$}

6152: $$

6153: and

6154: $$

6155: \albar(f_n(e))\,\in\,\PAn(\albar(e)).

6156: \tag{6$b$}

6157: $$

6158: and so by (6$a$) we may assume (by definition of \Omalbar)

6159: that for all $n$

6160: $$

6161: \tx{$\al\circ\curl{f_n(e)}$ \,is a fast Cauchy sequence, 

6162: with limit \,$\albar(f_n(e))$.}

6163: \tag7

6164: $$

6165: %(Note the notation \curl{e}

6166: %for the partial recursive function with index $e$.)

6167: Also by (6$b$) and (5), 

6168: $$

6169: \ds\bigl(\albar(f_n(e)), \ F(\albar(e))\bigr) \ <\ 2^{-n}.

6170: \tag8

6171: $$

6172: Now let $e'$ be a ``canonical" index for the (partial) function 

6173: $$

6174: \curl{e'}:\ n\ \mapsto \curl{f_n(e)}(n)

6175: \tag9

6176: $$

6177: obtained uniformly effectively in $e$.

6178: So \curl{e'}

6179: is the ``diagonal" function

6180: formed from the sequence of functions with indices

6181: $f_n(e)$.

6182: Consider

6183: the sequence \,$\als\circ \curl{e'}$, \,\ie,

6184: $$

6185: \als(\{e'\}(0)),\ \als(\{e'\}(1)), \ \als(\{e'\}(2)), \ \dots,

6186: \tag{10}

6187: $$

6188: {\bf Claim:}

6189: \,(10) is a Cauchy sequence in \As, with modulus of convergence 

6190: \,$\lam n(n+2)$.

6191: \endpr

6192: 

6193: \n{\bf Proof of claim:}

6194: \ For any $n$ and $k>n$:

6195: $$

6196: \align

6197: &\quad\ \ \ds\big(\al(\curl{e'}(k)),\ \al(\curl{e'}(n)\big) \\

6198: &= \ \ds\big(\al(\curl{f_k(e)}(k),\ \al(\curl{f_n(e)}(n)\big) 

6199:   \qquad\tx{by def\. (9) of $e'$}\\

6200: &\le \ \ds\big(\al(\curl{f_k(e)}(k)),\ \albar(f_k(e))\big)

6201:   \ + \ \ds\big(\albar(f_k(e)),\ \albar(f_n(e))\big)

6202:   \ + \ \ds\big(\albar(f_n(e)),\ \al(\curl{f_n(e)}(n))\big)\\

6203: &=\ \ds_1\ + \ \ds_2\ + \ \ds_3 \qquad\tx{(say)}

6204: \endalign

6205: $$

6206: where

6207: $$

6208: \align

6209: &\ds_1 \ \le \ 2^{-k},\\

6210: &\ds_3 \ \le \ 2^{-n},

6211: \endalign

6212: $$

6213: by (7), and

6214: $$

6215: \align

6216: \ds_2 \ &\le \ \ds\big(\albar(f_k(e)),\ F(\albar(e))\big) 

6217:   \ + \ \ds\big(F(\albar(e)),\ \albar(f_n(e))\big) \\

6218: &< \ 2^{-k} \ + \ 2^{-n}

6219: \endalign

6220: $$

6221: by (8).

6222: Therefore 

6223: $$

6224: \align

6225: \ds\big(\al(\curl{e'}(k)),\ \al(\curl{e'}(n)\big) 

6226:   \ &\le \ \ds_1\ + \ \ds_2\ + \ \ds_3\\

6227: &< \ 2\cdot 2^{-k} \ + \ 2\cdot 2^{-n}\\

6228: &< \ 2^{-n+2}.

6229: \endalign

6230: $$

6231: This proves the claim.

6232: \endpf

6233: Further, by the method of Remark 7.1.1

6234: (composing \curl{e'} with the modulus of convergence),

6235: we can replace the index $e'$ by an $e$-code

6236: $e''$ for a fast Cauchy sequence:

6237: $$

6238: \curl{e''}(n) \ \sq \ \curl{e'}(n+2).

6239: \tag{11}

6240: $$

6241: Then we define 

6242: $$

6243: g(e) \ = \ e''.

6244: \tag{12}

6245: $$

6246: We show that $g$ is an \albar-tracking function for $F$,

6247: \ie, (assuming (3)) we show (4).

6248: %$$

6249: %g(e)\in \Omalbars \qquad\tx{and}

6250: %\qquad \albar(g(e)) \ = \ F(\albar(e)).

6251: %\tag{12}

6252: %$$

6253: Since \,$\al\circ \curl{e''}$ \,is a 

6254: fast Cauchy sequence, with the same limit in $A$

6255: (if it exists) as \,$\al\circ \curl{e'}$ \,(by its definition (11)), 

6256: \,to prove (4) it is enough to show (by (12)) that 

6257: $$

6258: \al(\curl{e'}(n)) \ \to \ F(\albar(e)) \qquad \tx{as}\qquad n \ \to\ \infty.

6259: \tag{13}

6260: $$

6261: This follows since

6262: $$

6263: \aligned

6264: \ds\big(\al(\curl{e'}(n),\ F(\albar(e) )\big)

6265: \ &= \ \ds\big(\al(\curl{f_n(e)}(n)),\ F(\albar(e))\big)

6266: \qqquad\tx{by def\. (9) of $e'$}\\

6267: \ &\le\ \ds\big(\al(\curl{f_n(e)}(n)), \ \albar(f_n(e))\big)

6268: \ + \ \ds\big(\albar(f_n(e)), \ F(\albar(e))\big)\\

6269: \ &<\ 2^{-n}\ \ + \ \ 2^{-n} \qquad\qqqquad\tx{by (7) and (8)}\\

6270: \ &=\  2^{-n+1}

6271: \endaligned

6272: $$

6273: proving (13).

6274: \endpf

6275: 

6276: \Remarkn{7.2.1}

6277: A deterministic version of Theorem A

6278: (\ie, without `\chooses')

6279: was proved in \cite{stewart:thesis}.

6280: 

6281: 

6282: \newpage

6283: 

6284: \itemm{\bbf 8 \ }{\bbf Interpretation of concrete in abstract model:

6285: \,Adequacy of 

6286: \nl

6287: \WhileCCxbig\ \,approximation}

6288: 

6289: \shead{8.1}{Adequacy Theorem}

6290: In this section we will prove Theorem B, a converse to the result of

6291: the previous section.

6292: Assume that $A$ is an N-standard metric \Sig-algebra,

6293: and \Xal\ an enumerated \Sig-subspace,

6294: with \al-computable closure \,\CalXalbar.

6295: 

6296: Note that we are not assuming in this section that 

6297: \,\CalX\ \,is a subalgebra of $A$,

6298: \,or even that

6299: \,\albar\ \,is \Sig-effective.

6300: 

6301: Before stating the theorem,

6302: we need a definition.

6303: 

6304: \Defn{8.1.1 \,(\al-effective local uniform continuity)}

6305: A partial function

6306:  \,$F:\Au\pto\As$

6307: \,is {\it effectively locally uniformly continuous\/\ } 

6308: (with respect to \al) \,if

6309: there is a recursive sequence 

6310: $$

6311: (k_0,l_0),\ (k_1,l_1),\ (k_2,l_2),\ \dots

6312: $$

6313: of pairs of naturals such that

6314: $$

6315: \dom{F} \ \sseq\ \bigcup_{i=0}^\infty\Bbu\big(\al(k_i),\,2^{-l_i}\big) 

6316: $$

6317: and there is a total recursive function

6318: \ $\LUF: \NN^2\to\NN$

6319: \ (a modulus of local uniform continuity for $F$) such that

6320: for all $i$, all \,$x,y\in\Bbu\big(\al(k_i),\,2^{-l_i}\big)\cap\dom{F}$,

6321: \,and all $n$:

6322: $$

6323: \dsu(x,y)\,< \,2^{-\LUFs(i,n)} \ \implies\ \ \dss(F(x),F(y))\,<\,2^{-n}.

6324: $$

6325: Here \,$\Bbu(a,\del)$ \,is the open ball in \Au\ with centre 

6326: $a$ and radius \del.

6327: (Recall the definition (2.6.3) of the product metric \dsu\ on \Au.).

6328: \endpr

6329: 

6330: \Examplen{8.1.2}

6331: This phenomenon typically occurs

6332: in the situation where $A$ is a countable union of 

6333: neighbourhoods with compact closure;

6334: for example, in the algebra \,\RRRp\ \,of reals,

6335: \,\RR\ \,is the union of the neighbourhoods 

6336: \,$(-k,\,k)$ \,for $k=1,2,\dots$.

6337: Then a continuous function $F$ on $A$ will be uniformly continuous

6338: on each of these neighbourhoods.

6339: \endpr

6340: 

6341: We are now ready for the theorem.

6342: 

6343: \Thmn{B \ (Adequacy)}

6344: Let $A$ be an N-standard metric \Sig-algebra,

6345: \Xal\ an enumerated \SortSig-subspace,

6346: and \,\CalXalbar\ the \SortSig-subspace

6347: \,of \al-computable elements of $A$.  Suppose 

6348: that for all \Sig-sorts $s$:

6349: \sn

6350: ($i$) \ \Xs\ is dense in \As, \,and

6351: \sn

6352: ($ii$) \ $\als:\NN\to\As$ \,is \WhileCCx-computable on $A$.

6353: \sn

6354: Let \,$F:\Au\pto \As$ \,be a function on $A$ 

6355: such that

6356: \sn

6357: ($iii$) \ $F$

6358: is effectively locally uniformly continuous w.r.t\. \al, \,and

6359: \sn

6360: ($iv$) \ \dom{F} is open.

6361: \sn

6362: If $F$ is strictly \albar-computable on $A$,

6363: then $F$ is \WhileCCx\ \,approximable on $A$.

6364: \endpr

6365: 

6366: Note the extra condition in Theorem B

6367: (apart from assumptions ($i$)--($iv$)),

6368: that $F$ be {\it strictly\/} \albar-computable.

6369: 

6370: \Remarkn{8.1.3}

6371: From the proof of the theorem, it will be apparent that 

6372: only sorts $s$ in the domain of $F$ have to satisfy condition ($i$),

6373: and only sorts $s$ in the domain or range of $F$ have to

6374: satisfy condition ($ii$).

6375: \endpr

6376: 

6377: The proof uses the following notation.

6378: 

6379: \Notationn{8.1.4}

6380: For any \,$k\in\NN$, \,let \,\econk\ \,be a canonical index for 

6381: the constant function \,on \,\NN\

6382: \,with constant value $k$,

6383: \ie, for all $n\in\NN$, 

6384: $$

6385: \curl{\econk}(n) \ = \ k.

6386: $$

6387: Note that \,$\econk\in\Omalbar$ \,and

6388: $$

6389: \albar(\econk) \ = \ \al(k).

6390: $$

6391: \endpr

6392: 

6393: \shead{8.2}{Proof of Theorem B: \,Overview}

6394: As an aid to the reader,

6395: we first give an informal overview 

6396: of the proof of Theorem B.  

6397: (See Figure 3.)

6398: 

6399: \midinsert

6400: \epsfxsize = 4in

6401: \ce{\epsfbox{fig3.eps}}

6402: \mn

6403: \ce{\sc Figure 3}

6404: \sn

6405: \endinsert

6406: 

6407: Given the assumptions \,$(i)\to(iv)$

6408: \,of Theorem B,

6409: suppose 

6410: \,$F:\Au\,\pto\,\As$ \,is strictly \albar-computable by 

6411: \,$f:\Omalbaru\pto\Omalbars$.

6412: (In Figure 3, we represent $f$ as mapping 

6413: \,\Omalu\ \,to \,\Omals,

6414: \,rather than mapping

6415: \,\Omalbaru\ \,to \,\Omalbars,

6416: \,as a useful approximation,

6417: as we will see.)

6418: We must describe a \WhileCCx\ \,procedure

6419: which approximates $F$ on $A$.

6420: 

6421: Let \xinAu.

6422: Suppose \,$F(x)\da y$.

6423: By the {\it density\/} of \,$X=\ran{\alu}$ in \Au,

6424: and by the {\it openness\/} of \dom{F},

6425: for each $n$

6426: we can find

6427: (using the \,`\chooses' \,operator,

6428: as well as the \WhileCCx\ \,computability of \al)

6429: an element \,$k_n$ of \Omalu\ 

6430: \,such that 

6431: \,$x_n \eqdf \alu(k_n) \in\dom{F}$,

6432: \,and also 

6433: \,$\ds(x_n,x) < 2^{-n}$.

6434: 

6435: Now compute 

6436: an element \,$l_n$ of \Omalu\ 

6437: which is a close approximation to

6438: $f(k_n)$, or rather to $f(e_{\conss[k_n]})$.

6439: More precisely, let

6440: \,$e'_n \eqdf f(e_{\conss[k_n]})$, 

6441: \,and let 

6442: \,$l_n \,\eqdf \,\curl{e'_n}(n)$.

6443: Then 

6444: \,$\ds(\al(l_n), \albar(e'_n)) < 2^{-n}$.

6445: Put \,$y_n = \al(l_n)$.

6446: 

6447: We must now check that the mapping 

6448: \,$(x,n)\mapsto y_n$

6449: \,defined above is \WhileCCx\

6450: \,computable, and approximates $F$.

6451: By {\it effective local uniform continuity\/} of $F$,

6452: since 

6453: \,$(x_n)_n$

6454: \,is a fast Cauchy sequence with limit $x$,

6455: \,$(y_n)_n$

6456: \,is a Cauchy sequence with computable modulus of convergence and limit $y$.

6457: Note also that \WhileCCx\ \,computability of $y_n$

6458: (as a function of $x$ and $n$)

6459: uses the \WhileCCx\ \,{\it computability\/} of \al.

6460: Hence we can define a \WhileCCx\

6461: \,procedure 

6462: \,$P:\nats\times u \,\to\,s$

6463: \,with \,$P^A(n,x)$

6464: \,equal to

6465: the set of all such $y_n$, obtainable

6466: in this way from all possible implementations of the

6467: \,`\chooses' \,operator.

6468: Hence $F$ is computably approximable by $P$.

6469: 

6470: We turn to a precise proof of the theorem.

6471: 

6472: 

6473: \shead{8.3}{Proof of Theorem B}

6474: First we show, from assumption ($iii$), 

6475: that $F$ has a \WhileCCx\ \,modulus of continuity,

6476: \ie, a function

6477: $$

6478: \MCF:\ \Au\times\NN\ \pto\ \NN

6479: $$

6480: such that \,$\dom{F}\sseq\dom{\MCF}$, \,and

6481: for all \,$x,y\in\dom{F}$ \,and for all $n$,

6482: $$

6483: %\ds(x,y)\,<\,2^{-\MCFs(\dz{x,n})}\ \implies \ \ds(F(x),F(y))\,<\,2^{-n}.

6484: \ds(x,y)\,<\,2^{-\MCFs(\dz{x,n})}\ \implies \ \ds(F(x),F(y))\,<\,2^{-n}.

6485: \tag1

6486: $$

6487: A \WhileCCx\ \,algorithm for this

6488: is easily constructed as follows

6489: (using the notation of Definition 8.1.1).

6490: With input \xinAu\ \,and $n$:

6491: \,first find $i$ such that 

6492: $$

6493: x\in\Bb(\al(k_i),\ 2^{-l_i}).

6494: \tag2

6495: $$

6496: (If $x\notin\dom{F}$, 

6497: there may be no such $i$, and the algorithm 

6498: for $\MCF(x,n)$ would then diverge, which is fine, from our viewpoint.)

6499: \,Note that 

6500: the sequences ($k_i$) and ($l_i$)

6501: are computable, and also

6502: (by assumption ($ii$))

6503: \,\al\ \,is \WhileCCx\  computable.

6504: We also use 

6505: the primitive operations 

6506: \,\ds\ \,and \,`$<$' (partial!) on \,\RR,

6507: \,as well as the \,`\chooses' \,construct,

6508: in ``finding" a suitable $i$.

6509: 

6510: Next (by (2)) find a natural number $d_0$ such that 

6511: $$

6512: \ds(x,\,\al(k_i)) \,+ \,2^{-d_0} \ < \ 2^{-l_i}.

6513: \tag3

6514: $$

6515: Here again we use

6516: the \WhileCCx\ \,computability of \al,

6517: and

6518: the primitive operations

6519: \,\ds, \,`$+$' \,and \,`$<$

6520: on \,\RR, \,

6521: as well as the \,`\chooses' \,construct,

6522: to find a suitable $d_0$.

6523: 

6524: From (2) and (3),

6525: $$

6526: \Bb(x,\ 2^{-d_0}) \ \sseq\ \Bb(\al(k_i),\ 2^{-l_i}).

6527: $$

6528: So define

6529: $$

6530: \MCF(x,n) \ := \ \max(d_0,\,\LUF(i,n))

6531: $$

6532: which is \,\WhileCCx\ \,computable, by the above remarks.

6533: 

6534: Now we will describe (in pseudo-\WhileCCx\ \,code)

6535: an algorithm for a \WhileCCx-computable function 

6536: $$

6537: G:\ \NN\times\Au\ \totop\ \Asua

6538: $$

6539: which approximates $F$, in the sense that

6540: for all $n$ and all $x\in\dom{F}$,

6541: %\,$G_n(x)\sseq\As$, \,and 

6542: $$

6543: %\tx{for all} \ \ y\,\in \,G_n(x), \quad \ds(F(x),\,y)\ <\ 2^{-n}.

6544: G_n(x)\ \sseq\ \Bb(F(x),\,2^{-n})\ \sseq\ A_s.

6545: \tag4

6546: $$

6547: With input \,$n,\,x$:

6548: \sn

6549: ($1^\circ$)

6550: \ Compute 

6551: $$

6552: M\ :=\ \MCF(x, \,n+1).

6553: \tag5

6554: $$

6555: \ ($2^\circ$)

6556: We want to find some $k$ such that both

6557: $$

6558: \ds(\al(k),\ x) \ < \ 2^{-M}

6559: \tag6

6560: $$

6561: and

6562: $$

6563: \albar(\econk) \ = \ \al(k) \,\in\,\dom{F}.

6564: \tag7

6565: $$

6566: Assume $x\in\dom{F}$.

6567: By the density assumption ($i$)

6568: and \ openness assumption ($iv$),

6569: such a $k$ exists.

6570: Further, by assumption, $F$ has a 

6571: computable strict \albar-tracking function $f$.

6572: Then (7) is equivalent to

6573: $$

6574: f(\econk)\da.

6575: \tag8

6576: $$

6577: So using the \,`\chooses' \,construct again,

6578: search for some $k$ satisfying both (6) and (8).

6579: (Note that in practice 

6580: this `\chooses' operation would be implemented by dovetailing ---

6581: recall the discussion in \S4.1.)

6582: \sn

6583: ($3^\circ$)

6584: Compute \,$f(\econk)\da e'$.

6585: By (7), \,$e'\in\Omal$ \,and

6586: $$

6587: F(\al(k)) \ = \ F(\albar(\econk)) \ = \ \albar(f(\econk)) \ = \ \albar(e').

6588: $$

6589: Hence by (1), (5) and (6),

6590: $$

6591: \ds\big(F(x),\ \albar(e')\big)\ 

6592:   = \ \ds\big(F(x),\ F(\al(k))\big) \ < \ 2^{-n-1}.

6593: \tag9

6594: $$

6595: ($4^\circ$)

6596: \ Finally compute 

6597: $$

6598: y\ :=\ \al\big(\{e'\}(n+1)\big)

6599: \tag{10}

6600: $$

6601: This is possible by assumption ($ii$) again.

6602: Then, since \,$\al\circ\{e'\}$ \,is a fast Cauchy sequence,

6603: $$

6604: \ds\big(y,\ \albar(e')\big) \ = 

6605: \ \ds\big(\al(\{e'\}(n+1)),\ \albar(e')\big) \ \le\ 2^{-n-1}.

6606: \tag{11}

6607: $$

6608: Hence by (11) and (9),

6609: $$

6610: \align

6611: \ds\big(y,\,F(x)\big) \ &\le

6612: \ \ds\big(y,\,\albar(e')\big)\,+\,\ds\big(\albar(e'),\,F(x)\big)\\

6613: &<\ 2^{-n-1}\,+\,2^{-n-1}\\

6614: &=\ 2^{-n}.

6615: \endalign

6616: $$

6617: Define \,$G_n(x)$ \,to be the set of all possible $y$ computed as in (10),

6618: by all possible implementations of the \,`\chooses' \,construct

6619: as used in the above algorithm.

6620: %(see Discussion 3.2.7).

6621: Then $G$ satisfies (4), and is \WhileCCx\ \,computable, 

6622: by the above discussion.

6623: \endpf

6624: 

6625: 

6626: \Shead9{Completeness of \WhileCCxbig\ \,approximation}

6627: Under certain assumptions,

6628: we can combine Theorems A and B into a single equivalence,

6629: Theorem C below.

6630: We will then look at several examples of metric algebras where our

6631: abstract and concrete models are equivalent according to this Theorem.

6632: 

6633: \shead{9.1}{Effective openness}

6634: Note first the following problem:

6635: Theorem A concludes with \albar-computability of $F$,

6636: whereas Theorem B assumes {\it strong\/} \albar-computability.

6637: To deal with this, we must make an assumption 

6638: of ``effective openness" of \dom{F}.

6639: This is handled by strengthening the ``effective local uniform continuity"

6640: assumption, as follows.

6641: 

6642: Assume, as before, that $A$ is an N-standard metric

6643: \Sig-algebra,

6644: \Xal\ \,is an enumerated \Sig-subspace of $A$,

6645: and \,\CalXalbar\ \, is its computable closure in $A$.

6646: 

6647: \Defn{9.1.1 \ (\al-effective openness)}

6648: A subset $U$ of \Au\ ($u$ a \Sig-product type)

6649: is {\it effectively open\/} (with respect to \al)

6650: \,if there is a recursive sequence

6651: $$

6652: (k_0,l_0),\ (k_1,l_1),\ (k_2,l_2),\ \dots

6653: $$

6654: of pairs of naturals such that

6655: $$

6656: U \ =\ \bigcup_{i=0}^\infty\Bbu\big(\al(k_i),\,2^{-l_i}\big).

6657: $$

6658: \endpr

6659: 

6660: \Defn{9.1.2 \ (Strong \al-effective local uniform continuity)}

6661: A partial function

6662:  \,$F:\Au\pto\As$

6663: \,is {\it strongly effectively locally uniformly continuous\/} 

6664: (with respect to \al) 

6665: \,if there is a recursive sequence

6666: $$

6667: (k_0,l_0),\ (k_1,l_1),\ (k_2,l_2),\ \dots

6668: $$

6669: of pairs of naturals such that

6670: $$

6671: \dom{F} \ =\ \bigcup_{i=0}^\infty\Bbu\big(\al(k_i),\,2^{-l_i}\big)

6672: \tag1

6673: $$

6674: and

6675: there is a total recursive function

6676: \ $\LUF: \NN^2\to\NN$

6677: \ (a modulus of local uniform continuity for $F$) such that

6678: for all $i$, all \,$x,y\in\Bbu\big(\al(k_i),\,2^{-l_i}\big)$,

6679: \,and all $n$:

6680: $$

6681: \ds(x,y)\,< \,2^{-\LUFs(i,n)} \ \implies\ \ \ds(F(x),F(y))\,<\,2^{-n}.

6682: $$

6683: \endpr

6684: 

6685: \Remarkn{9.1.3}

6686: The only difference between effective local uniform continuity 

6687: (Definition 8.1.1) and the ``strong" version above

6688: is the equality in equation (1).

6689: \endpr

6690: 

6691: Let

6692: \,$F:\Au\pto \As$

6693: \,be a function on $A$.  Then clearly:

6694: 

6695: \Lemman{9.1.4}

6696: Strong \al-effective local uniform continuity of $F$ 

6697: implies \al-effective openness of \dom{F}.

6698: \endpr

6699: 

6700: \Lemman{9.1.5}

6701: Suppose \dom{F} is \al-effectively open,

6702: and \albar\ is strictly \Sig-effective.

6703: Then

6704: $$

6705: \tx{$F$ is \,\albar-computable}

6706: \ \ \ifff

6707: \ \ \tx{$F$ is strictly \,\albar-computable.}

6708: $$

6709: \endpr

6710: 

6711: \Pf

6712: ($\tto$)

6713: Note first that the assumptions imply that 

6714: $$

6715: \domal{F} \ \eqdf\ (\albar)^{-1}(\dom{F})

6716: \ = \ \curly{e\in \Omalbaru\mid \albar(e)\in \dom{F}}

6717: $$

6718: is an r.e. (recursively or computably enumerable) 

6719: subset of \NN, since for all $e\in\NN$

6720: $$

6721: e\in \domal{F} \ \ifff\ 

6722: \ex i \,\big[\ds(\albar(e), \al(k_i)) \,< \,2^{-l_i}\big]

6723: $$

6724: (in the notation of Definition 9.1.2)

6725: which is an r.e\. 

6726: condition, by strict \albar-computability of \,\ds\ \,and \,\lsreal\

6727: \,(implied by strict \Sig-effectiveness of \albar).

6728: Hence, if $f$ is a computable \albar-tracking function for $F$,

6729: it can be replaced by a strict \albar-tracking function $f'$,

6730: defined by

6731: $$

6732: f'(e) \ \simeq\ 

6733: \cases

6734: f(e) \ift{$e\in\domal{F}$}\\

6735: \ua \ow

6736: \endcases

6737: $$

6738: which is easily seen to be computable.

6739: \endpf

6740: \Lemman{9.1.6}

6741: Suppose \dom{F} is \al-effectively open,

6742: and the mappings \,$\als:\NN\to\As$ \,are \WhileCCx\ computable.

6743: Then

6744: $$

6745: \tx{$F$ is \WhileCCx-approximable}

6746: \ \ \ifff

6747: \ \ \tx{$F$ is strictly \WhileCCx-approximable.}

6748: $$

6749: \endpr

6750: \n

6751: (Recall Definition 3.5.1.)

6752: \,The proof is an easy exercise.

6753: 

6754: 

6755: \shead{9.2}{Completeness}

6756: We are ready to state the completeness theorem 

6757: for \WhileCCx\ approximability relative to

6758: \albar-computability.

6759: 

6760: \newpage

6761: 

6762: \Thmn{C \,(Completeness)}

6763: Let $A$ be an N-standard metric \Sig-algebra,

6764: and \Xal\ an enumerated \SortSig-subspace.

6765: Suppose the enumerated \SortSig-space \CalXalbar\ of

6766: \al-computable elements of $A$ is a \Sig-subalgebra of $A$.

6767: Assume also that for all \Sig-sorts $s$,

6768: \sn

6769: ($i$) \ \albar\ \,is strictly \Sig-effective,

6770: \sn

6771: ($ii$) \ \Xs\  \,is dense in \As, \,and

6772: \sn

6773: ($iii$) \ $\als:\NN\to\As$ \,is \WhileCCx-computable on $A$.

6774: \sn                               

6775: Let \,$F:\Au\pto \As$

6776: \,be a function on $A$,

6777: such that

6778: \sn

6779: ($iv$) \ $F$

6780: is strongly effectively locally uniformly continuous w.r.t\. \al.

6781: \sn

6782: Then

6783: $$

6784: \tx{$F$ is (strictly) \WhileCCx\ approximable on $A$}

6785: \ \ \ifff

6786: \ \ \tx{$F$ is (strictly) \,\albar-computable on $A$.}

6787: $$

6788: \endpr

6789: \n

6790: Note that the word ``strictly" in the equivalence

6791: may be omitted or inserted in either side at will.

6792: 

6793: \Pf

6794: From Theorems A and B, together with Lemmas 9.1.4, 9.1.5 and 9.1.6.

6795: \endpr

6796: 

6797: \sheads{9.3}{Examples of the application of the Completeness Theorem}

6798: \mn

6799: {\bf ({\bi a}) \,Canonical enumerations\/}

6800: \sn

6801: The purpose of this example is 

6802: to make plausible condition ($iii$) of Theorem C

6803: (and, of course, condition ($ii$) of Theorem B in Section 8),

6804: \ie, the assumption of \WhileCCx\ computability

6805: of the enumeration \al, by describing a commonly occurring

6806: situation which implies it.

6807: 

6808: Suppose \Xal\ is an enumerated \Sig-subalgebra of $A$.

6809: 

6810: \Defn{9.3.1}

6811: The enumeration 

6812: \,$\al\:\NN\onto X$

6813: \,is {\it effectively determined by a system of generators\/}

6814: \,$G = \ang{g^s_0,g^s_1,g^s_2,\dots}_\sinSortSig$ 

6815: \,if, and only if,

6816: \sn($i$) \,$G$ generates $X$ as a \Sig-subalgebra of $A$; 

6817: \sn($ii$) \,\al\ \,is defined as the composition of the maps

6818: $$

6819: \commdiag{

6820: \NN\,\maprighturaise3{\enumSig} \,\TermSig\,\maprighturaise3{\evalG}\,X

6821: }

6822: $$

6823: where \,\enumSig\ \,is the inverse of the G\"odel numbering

6824: of \,\TermSig, \,and \,\evalG\

6825: \,is the term evaluation induced by $G$,

6826: %$$

6827: %\xtt_i^s \ \mapsto \ g_i^s,

6828: %$$

6829: \ie,

6830: $$

6831: \evalG(t) \ = \ \bb{t}\sigG,

6832: $$

6833: where \sigG\ is the state defined by

6834: $$

6835: \sigG(\xtt_i^s) \ = \ g_i^s

6836: $$

6837: for some standard enumeration

6838: \,$\xtt_0^s, \,\xtt_1^s, \,\xtt_2^s, \dots$

6839: \,of the \Sig-variables of sort $s$; \,and

6840: \sn($iii$)

6841: if, for any \Sig-sort $s$,

6842: the sequence 

6843: \,\ang{g^s_0,g^s_1,g^s_2,\dots}

6844: \,is finite, then each $g^s_i$ is a \Sig-constant,

6845: \,whereas if this sequence is infinite,

6846: then the map

6847: \,$i \mapsto g_i^s$

6848: \,is a \Sig-function.

6849: \endpr

6850: 

6851: An enumeration constructed in this way is called {\it canonical\/}

6852: w.r.t\. $G$.

6853: 

6854: \Remarkn{9.3.2 \,(Totality of \evalG)}

6855: We assume here that \,\evalG\

6856: \,(and hence \al) is total.

6857: This is achieved by assuming that either

6858: \itemm{($i$)}

6859: $A$ is total, \,or

6860: \itemm{($ii$)}

6861: \TermSig\ \,is replaced by some decidable subset 

6862: \,$\Term'(\Sig)$

6863: \,on which \,\evalG\ \,is total

6864: (for example, omitting all terms involving division by 0).

6865: \sn

6866: Either one of these assumptions holds in each of the following examples;

6867: for example, ($i$) holds in example ($b$) below, and ($ii$) in example ($c$),

6868: resulting in the same ``canonical" enumeration \al\ of \QQ\

6869: in both cases (even though the algebras are different).

6870: \endpr

6871: 

6872: \Propn{9.3.3}

6873: If \,\al\ \,is effectively determined 

6874: by a system of generators, then the canonical

6875: enumerations \,\als\ \,are \Whilex\ computable for all \Sig-sorts $s$.

6876: \endpr

6877: 

6878: \Pf

6879: This follows from the \Whilex\ computability 

6880: of term evaluation \cite[Cor\. 4.7]{tz:hb}.

6881: \endpf

6882: 

6883: The significance of the above definition

6884: and proposition is this:

6885: it is quite common for an enumeration to be effectively determined

6886: by a system of generators;

6887: and in such a situation,

6888: condition ($ii$) in Theorem B, and ($iii$) in Theorem C,

6889: will be (more than) satisfied.

6890: This will be the case in the following examples.

6891: \endpr

6892: 

6893: \mn

6894: {\bf ({\bi b}) \,Partial real algebra\/}

6895: \sn

6896: Recall the example (7.1.5) of the enumeration \al\

6897: of \QQ\ as a subspace of the N-standardised metric 

6898: algebra \,\RRRN\ \,of reals

6899: (Examples 2.5.3($b$) and 2.6.1)

6900: and the corresponding enumeration \albar\

6901: of the set \,\CalQQ\ \,of {\it recursive reals\/}.

6902: Note that \al\ is canonical,

6903: being effectively determined by the generators \curl{0,1},

6904: and is hence \Whilex\ computable over \RRR.

6905: Further, \QQ\ is dense in \RR,

6906: \,\CalQQ\ \,is a subfield of \RR,

6907: and \albar\ is strictly \SigRRR-effective.

6908: We then have,

6909: as a corollary to Theorem C:

6910: 

6911: %Now let 

6912: %$$

6913: %F\:\RR^n\ \pto\ \RR

6914: %$$

6915: %be any partial function on \RR\ which is strongly effectively

6916: %locally uniformly continuous.

6917: %(Examples of such a function are all the common functions

6918: %of elementary calculus, such as \ $1/x$, $\log(x)$ and $\tan(x)$.)

6919: 

6920: \Corn{9.3.4}

6921: Suppose 

6922: \,$F\:\RR^n\ \pto\ \RR$

6923: \,is strongly effectively

6924: locally uniformly continuous.

6925: Then

6926: %\sn

6927: %\ce{$F$ is (strictly) \WhileCCx-approximable on \RRRN\ \ \ifff\ 

6928: %\ $F$ is (strictly) \albar-computable on \RR.}

6929: $$

6930: \multline

6931: \tx{$F$ is (strictly) \WhileCCx-approximable on \RRRN}\\

6932: \llongtofrom

6933: \ \ \tx{$F$ is (strictly) \albar-computable on \RR.}

6934: \endmultline

6935: $$

6936: \endpr

6937: 

6938: Examples of functions satisfying the assumption

6939: (and also the equivalence)

6940: are all the common (partial) functions

6941: of elementary calculus, such as \ $1/x$, \,$\log x$ \,and \,$\tan x$.

6942: 

6943: \newpage

6944: 

6945: \mn

6946: {\bf ({\bi c}) \,Banach spaces with countable bases\/}

6947: \sn

6948: Let $X$ be a Banach space over \RR\ with a countable basis 

6949: \,$e_0,e_1,e_2,\dots $,

6950: which means that any element $x\in X$ can be represented uniquely

6951: as an infinite sum

6952: $$

6953: x \ = \ \sum_{i=0}^\infty r_i e_i

6954: $$

6955: with coefficients \,$r_i\in\RR$

6956: (where the infinite sum is understood as denoting convergence

6957: of the partial sums in the norm of $X$).

6958: (Background on Banach space theory can be found in

6959: any of the standard texts, \eg, \cite{royden,taylor-lay}.)

6960: To program with $X$, we construct a many-sorted algebra \XXX\ of the form

6961: \mn

6962: $$

6963: \boxed{

6964: \matrix \format\l&\quad\l\\

6965: \algebras &\XXX \\

6966: \imports &\RRRN\\

6967: \carrierss &X\\

6968: \functionss

6969: &0\:\ \ \to X,\\

6970: &+\:X^2 \to X,\\

6971: &-\:X \to X,\\

6972: &\odot\:\RR\times X\to X,\\

6973: &\normfn\:X\to \RR,\\

6974: &\es\:\NN\to X,\\

6975: &\ifs_X\:\BB\times X^2\to X\\

6976: \ends&

6977: \endmatrix

6978: }

6979: $$                                                                              

6980: \mn

6981: where \,$\odot$ \,is scalar multiplication,

6982: \,\normfn\ \,is the norm function and

6983: and \,\es\ \,is the enumeration of the basis:

6984: \,$\es(i) = e_i$.

6985: Note that the algebras \BBB\ and \NNN\ are 

6986: implicitly imported, as parts of \RRRN,

6987: so that there are four carriers:

6988: \,$X$, \RR, \BB\ \,and \,\NN,

6989: \,of sorts 

6990: \,\vectors, \,\scalars, \,\bools\ \,and \,\nats\

6991: \,respectively.

6992: 

6993: 

6994: Let \,$\Sig = \Sig(\XXX)$.  

6995: Let \Sigo\ be \Sig\ without the norm function \,\normfn,

6996: \,and let \XXXo\ be the reduct of \XXX\ to \Sigo.

6997: Then \Sigo\ is the signature of an N-standardised vector space

6998: over \RR, with explicit countable basis.

6999: 

7000: This can be turned into a metric algebra

7001: in the standard way,

7002: by defining a distance function on $X$

7003: in terms of the norm:

7004: $$

7005: \ds(x,y) \ \eqdf\ \norm{x-y}.

7006: $$

7007: 

7008: Let \,$\LQQe\sset X$ \,be the set of all finite 

7009: linear combinations of basis elements from \es\ with 

7010: coefficients in \QQ.

7011: The following are easily shown:

7012: \bull

7013: \LQQe\ \,is countable; in fact it has a canonical enumeration

7014: $$

7015: \al\: \NN\ \onto\ \LQQe

7016: $$

7017: w.r.t\. the generators \,\es,

7018: \,which (by ($a$) above)

7019: is \Whilex\ computable;

7020: \bull

7021: \LQQe\ is dense in $X$;

7022: \bull

7023: \LQQe, with scalar field \QQ\

7024: (together with carriers \NN\ and \BB)

7025: is a \Sigo-subalgebra of \XXXo.

7026: 

7027: \sn

7028: Now let \,(\CalLQQe,\ \albar) \,be the enumerated subspace of 

7029: \al-computable vectors.  Then we can see that

7030: 

7031: \bull

7032: \CalLQQe, with scalar field \,\CalQQ\

7033: (together with carriers \NN\ and \BB)

7034: is also a \Sigo-subalgebra of \XXXo; 

7035: \,and moreover,

7036: 

7037: \bull

7038: \albar\ \,is strictly \Sigo-effective.

7039: 

7040: \sn

7041: However \,\CalLQQe\ \,is {\it not\/} necessarily a {\it normed\/} 

7042: subspace of \XXX, since it may not be

7043: closed under \,\normfn,

7044: \,\ie, \,\norm{x} may not be in \,\CalQQ\ \,for all $x\in \CalLQQe$;

7045: for example, if \XXX\ is the space \,\lp\ \,or

7046: \,$\Lp[0,1]$

7047: \,where $p$ is a nonrecursive real (see Examples 9.3.8 below).

7048: We must therefore make an explicit assumption

7049: for the Banach space \,\Xnorm\ \,with respect to 

7050: both the {\it closure\/} of \,\CalLQQe\ \,under \,\normfn,

7051: \,and the \,\albar-{\it computability\/} of \,\normfn.

7052: 

7053: \Assumptionn{9.3.5 \,(\albar-computable norm assumption for \,\Xnorm)}

7054: \nl

7055: For all $x\in \CalLQQe$, \,$\norm{x}\in\CalQQ$.

7056: Furthermore, the norm function \,\normfn\

7057: \,is strictly \,\albar-computable.

7058: \endpr

7059: 

7060: \n

7061: As we will see, many common examples of Banach spaces satisfy 

7062: this assumption.

7063: 

7064: %\Remarkn{9.3.6}

7065: Note that ssumption 9.3.5 is equivalent to the 

7066: following (apparently weaker) assumption,

7067: which is often easier to prove:

7068: \endpr

7069: 

7070: \Assumptionn{9.3.6 \,(\alalbar-computable norm assumption for \,\Xnorm)}

7071: For all $x\in \LQQe$, \,$\norm{x}\in\CalQQ$.

7072: Further, \,\normfn\ \,has a computable \alalbar-tracking function,

7073: \ie, a computable function \,$f\:\NN\to\NN$ 

7074: \,such that the following diagram commutes:

7075: 

7076: \bn

7077: $$

7078: \commdiag{

7079: \LQQe& &\maprighturaise3{\normfn}& &\CalQQ\cr

7080: \mapupl{\al}& && &\mapupr{\ \albar}\cr

7081: \NN& &\maprightd{f}& &\Omalbar

7082: }

7083: $$

7084: \endpr

7085: 

7086: \mn

7087: Suppose now that \,\Xnorm\ \,satisfies the \albar-computable norm assumption.

7088: Then the \Sigo-subalgebra \,\CalLQQe\ \,of \XXXo\

7089: can be expanded to a \Sig-subalgebra of \XXX\ 

7090: (which we will also write as \,\CalLQQe), enumerated by \albar,

7091: which is strictly \Sig-effective.

7092: 

7093: Now let \,$F\:X\to \RR$ \,be a (total) {\it linear functional on\/} $X$.

7094: $F$ is said to be {\it bounded\/} if for some real $M$,

7095: $$

7096: |F(x)| \ \le \ M\norm{x} \ \ \ \tx{for all}  \ \,x\in X.

7097: \tag1

7098: $$

7099: Write \norm{F} for the least $M$ for which (1) holds.

7100: Then if $F$ is bounded,

7101: $$

7102: |F(x)-F(y)| \ \le\ \norm{F}\cdot\norm{x-y} \ \ \ \tx{for all}\  \,x,y\in X,

7103: $$

7104: and so $F$ is uniformly continuous, in fact it is clearly 

7105: {\it effectively locally uniformly continuous\/},

7106: and strongly so (since it is total).

7107: We may therefore apply Theorem C to $F$.

7108: 

7109: \Corn{9.3.7 \,(Completeness for computation on Banach spaces)}

7110: Let $X$ be a Banach space over \RR\ with countable basis, and

7111: let \,\CalLQQe\ \,be the enumerated subspace of \al-computable vectors,

7112: where \al\ is a canonical enumeration of the subspace \,\LQQe.

7113: Suppose \Xnorm\ satisfies the \,\alalbar-computable norm assumption.

7114: Then for any bounded linear functional $F$ on $X$,

7115: $$

7116: \tx{$F$ is (strictly) \WhileCCx\ approximable on \XXX}

7117: \ \ifff

7118: \ \tx{$F$ is (strictly) \,\albar-computable on $X$,}

7119: $$

7120: where \XXX\ is the N-standard algebra formed from $X$ as above.

7121: \endpr

7122: 

7123: Finally we give examples of Bananch spaces which satisfy this

7124: \albar-computable norm assumption.

7125: 

7126: \Examplesn{9.3.8 \,(Banach spaces with computable norms)}

7127: \itemm{($i$)}

7128: For $1\le p<\infty$, we have the space \,\lp\ 

7129: \,of all sequences $x = \ang{x_n}_{n=0}^\infty$

7130: \,of reals such that \,$\sum_{n=0}^\infty |x_n|^p < \infty$,

7131: with norm defined by

7132: $$

7133: \norm{x}_p \ = \ \big(\sum_{n=0}^\infty |x_n|^p\big)^{1/p},

7134: $$

7135: and a countable basis given by \,$e_i = \ang{e_{i,n}}_{n=0}^\infty$, 

7136: \,where 

7137: $$

7138: e_{i,n} \ = \

7139: \cases

7140: 1 \ift{$i=n$},\\

7141: 0 \ow.

7142: \endcases

7143: $$

7144: It is not hard to see that 

7145: $$

7146: \tx{\sl if $p$ is a recursive real, then 

7147: \,\lp\ \,satisfies the computable norm assumption,}

7148: $$

7149: and hence Corollary 9.3.7 can be applied to it.

7150: 

7151: \itemm{($ii$)}

7152: For $1\le p<\infty$, we have the space 

7153: \,$\Lp[0,1]$

7154: \,of all Lebesgue measurable functions $f$ on 

7155: the unit interval $[0,1]$ 

7156: such that \,$\int_0^1 |f|^p<\infty$,

7157: \,with norm defined by

7158: $$

7159: \norm{f}_p \ = \ \big(\int_0^1 |f|^p\big)^{1/p},

7160: $$

7161: and a countable basis given by (\eg)

7162: some standard enumeration of all 

7163: step functions on $[0,1]$ with rational values

7164: and (finitely many) rational points of discontinuity,

7165: or of all

7166: polynomial functions on $[0,1]$ with rational coefficients.

7167: Again, it is not hard to see that 

7168: $$

7169: \tx{\sl if $p$ is a recursive real, then \,$\Lp[0,1]$

7170: \,satisfies the computable norm assumption,}

7171: $$

7172: and hence Corollary 9.3.7 can be applied to it.

7173: 

7174: \itemm{($iii$)}

7175: The space \,$C[0,1]$ 

7176: \,of all continuous functions $f$ on $[0,1]$,

7177: with norm defined by

7178: $$

7179: \norm{f}_{\tx{sup}} \ = \ \sup_{t\in I}|f(t)|

7180: $$

7181: and a countable basis given by a standard enumeration

7182: of all 

7183: zig-zag functions on $[0,1]$ with (finitely many) turning points

7184: with rational coordinates,

7185: or of all

7186: polynomial functions on $[0,1]$

7187: with rational coefficients.

7188: Again, we see that 

7189: $$

7190: \tx{\sl $C[0,1]$ \,satisfies the computable norm assumption.}

7191: $$

7192: 

7193: 

7194: \Shead{10}{Conclusion}

7195: We have compared two theories of computable functions on topological

7196: algebras, one based on an abstract, high level model of programming and

7197: another based on a concrete, low-level implementation model. Our

7198: examples and results here, combined with our earlier results 

7199: \cite{tz:top,tz:hb}

7200: and those of Brattka \cite{brattka96,brattka:thesis}, 

7201: show that the following are surprisingly necessary features

7202: of a comprehensive theory of computation on topological algebras:

7203: 

7204: \item{1.}

7205: The algebras have partial operations.

7206: \item{2.}

7207: Functions are both continuous and multivalued. 

7208: \item{3.}

7209: Classical algorithms in analysis require nondeterministic

7210: constructs for their proper expression

7211: in programming languages.

7212: \item{4.}

7213: Indeed, multivalued subfunctions are needed to compute even single-valued

7214: functions, and abstract models must be nondeterministic even to compute

7215: deterministic problems.

7216: \item{5.}

7217: Abstract models and effective approximations by abstract models

7218: are generally sound for concrete models. 

7219: \item{6.}

7220: Abstract models even with approximation or limit operators are

7221: adequate to capture concrete models only in special circumstances.

7222: \item{7.}

7223: Nevertheless there are interesting examples where equivalence holds.

7224: \item{8.}

7225: The classical computable functions of analysis can be characterised

7226: by abstract models of computation.

7227: 

7228: Specifically, we examined abstract computation by the basic

7229: imperative model of \qwhiles-array programs. Many

7230: algorithms in practical computation are presented in

7231: pseudo-code based on the \qwhiles\ language.  To meet the

7232: requirement of feature 2 above we added the simplest form of

7233: countable choice to the assignments of the language, and we

7234: defined the \WhileCCx\ approximable computations. We proved

7235: a Soundness Theorem (Theorem A) and an Adequacy Theorem

7236: (Theorem B), and combined these into a Completeness Theorem (Theorem C), 

7237: in the case of metric algebras with partial

7238: operations. We considered algebras 

7239: of real numbers and Banach spaces where equivalence

7240: theorems hold.

7241: 

7242: There are, of course, interesting technical questions to

7243: answer in working out the details of the computability

7244: theory for the \WhileCCx\ model (\cf\ the theory

7245: for single-valued functions on total algebras in

7246: \cite{tz:hb}). There are several other important

7247: abstract models of computation that may be extended

7248: with nondeterminsitic constructs in order to establish

7249: equivalence with concrete models. The abstract model of

7250: schemes in \cite{brattka:thesis} is quite general in a number of

7251: ways. The topological properties of many valued functions

7252: are also in need of investigation.

7253: 

7254: However, returning to the general problem posed in the

7255: Introduction, the features 1--8 above suggest that new

7256: research directions are needed to develop a comprehensive

7257: theory of specification, computation and reasoning

7258: with infinite data. What are the appropriate programming

7259: constructs for working with topological computations? What

7260: specification techniques are appropriate for continuous

7261: systems? What logics are needed to support verification

7262: of programs that approximate functions?  Our work on

7263: computation suggests that some advanced semantic features

7264: are necessary. It suggests that the nondeterminism that

7265: played an important role in programming methodologies of

7266: the late 1970s  (\eg, \cite{dijkstra:book} seems to be needed in

7267: the proper development of topological programming. There

7268: are plenty of algorithms in scientific modelling, numerical

7269: analysis and graphics to investigate, using such new theoretical tools.

7270: 

7271: 

7272: \bigskip

7273: \bigskip

7274: \goodbreak

7275: 

7276: \cbb{References}\bigskip

7277: 

7278: \bibliographystyle{alpha}

7279: \bibliography{abbrev,bib}

7280: 

7281: \enddocument

7282: