1: \documentclass[10pt]{amsart}

2: \usepackage{amssymb}

3: \usepackage{amscd}

4: \usepackage{verbatim}

5: %\usepackage{curves}

6: \usepackage{epsfig}

7: %\addtolength{\textheight}{80pt}

8: %\addtolength{\textwidth}{80pt}

9: %\addtolength{\hoffset}{-40pt}

10: %\addtolength{\voffset}{-40pt}

11: 

12: 

13: \begin{document}

14: 

15: \newcommand\nin{\notin}

16: \newcommand\identity{\operatorname{id}}

17: \newcommand\id{\operatorname{id}}

18: \newcommand\Id{\operatorname{Id}}

19: %\newcommand\Real{\mathbb{R}}

20: \newcommand\pos{\Real^+}

21: \newcommand\Rnp{\Real\setminus\{0\}}

22: \newcommand\nzero{\setminus\{0\}}

23: \newcommand\Cx{\mathbb{C}}

24: \newcommand\Cxp{\Cx^+}

25: \newcommand\Cxm{\Cx^-}

26: \newcommand\Nat{\mathbb{N}}

27: \newcommand\halfNat{{\frac{1}{2}}\mathbb{N}}

28: \newcommand\intgr{\mathbb{Z}}

29: \newcommand\im{\operatorname{Im}}

30: \newcommand\re{\operatorname{Re}}

31: \newcommand\sign{\operatorname{sign}}

32: \newcommand\codim{\operatorname{codim}}

33: \newcommand\End{\operatorname{End}}

34: \newcommand\Ker{\operatorname{Ker}}

35: \newcommand\Hom{\operatorname{Hom}}

36: \newcommand\ideal{{\mathcal I}}

37: \newcommand\Span{\operatorname{span}}

38: \newcommand\Range{\operatorname{Ran}}

39: \newcommand\graph{\operatorname{graph}}

40: \newcommand\slim{\operatornamewithlimits{s-lim}}

41: \newcommand\diag{\operatorname{diag}}

42: \newcommand\Rn{\Real^n}

43: \newcommand\Rm{\Real^m}

44: \newcommand\RN{\Real^N}

45: \newcommand\RtN{\Real^{2N}}

46: \newcommand\RM{\Real^M}

47: \newcommand\sphere{\mathbb{S}}

48: \newcommand\Sn{\sphere^{n-1}}

49: \newcommand\Sm{\sphere^{m-1}}

50: \newcommand\Snp{\sphere^n_+}

51: \newcommand\Smp{\sphere^m_+}

52: \newcommand\SN{\sphere^{N-1}}

53: \newcommand\SNp{\sphere^N_+}

54: \newcommand\circlep{\sphere^1_+}

55: \newcommand\Phom{P_{h}}

56: \newcommand\Shom{S_{h}}

57: \newcommand\distance{\operatorname{dist}}

58: \newcommand\cl{\operatorname{cl}}

59: \newcommand\interior{\operatorname{int}}

60: \newcommand\Fa{\operatorname{Fa}}

61: \newcommand\ff{\operatorname{ff}}

62: \newcommand\mf{\operatorname{mf}}

63: \newcommand\cf{\operatorname{cf}}

64: \newcommand\scf{\operatorname{sf}}

65: \newcommand\lf{\operatorname{lb}}

66: \newcommand\rf{\operatorname{rb}}

67: \newcommand\indfam{{\mathcal K}}

68: \newcommand\calX{{\mathcal X}}

69: \newcommand\calK{{\mathcal K}}

70: \newcommand\calF{{\mathcal F}}

71: \newcommand\calO{{\mathcal O}}

72: \newcommand\calC{{\mathcal C}}

73: \newcommand\calCL{{\mathcal C}_{\text L}}

74: \newcommand\calCR{{\mathcal C}_{\text R}}

75: %\newcommand\Cinf{{\mathcal C}^{\infty}}

76: \newcommand\Cinf{\CI}

77: \newcommand\dist{{\mathcal C}^{-\infty}}

78: \newcommand\dCinf{\dot\Cinf}

79: \newcommand\ddist{\dot\dist}

80: \newcommand\Cj{{\mathcal C}^j}

81: \newcommand\Linf{L^{\infty}}

82: \newcommand\bcon{{\mathcal A}}

83: \newcommand\bconc{{\mathcal A}_{\text{phg}}}

84: \newcommand\Sch{{\mathcal S}}

85: \newcommand\temp{\Sch^{\prime}}

86: \newcommand\Diff{\operatorname{Diff}}

87: \newcommand\Diffb{\operatorname{Diff}_{\text{b}}}

88: \newcommand\Diffc{\operatorname{Diff}_{\text{c}}}

89: \newcommand\Diffsc{\operatorname{Diff}_{\text{sc}}}

90: \newcommand\DiffI{\operatorname{Diff}_{\text{I}}}

91: \newcommand\DiffIq{\operatorname{Diff}_{\text{I},q}}

92: \newcommand\supp{\operatorname{supp}}

93: \newcommand\ssupp{\operatorname{sing\ supp}}

94: \newcommand\csupp{\operatorname{cone\ supp}}

95: \newcommand\esupp{\operatorname{ess\ supp}}

96: \newcommand\Fr{{\mathcal F}}

97: \newcommand\Frinv{\Fr^{-1}}

98: \newcommand\bop{{\mathcal B}}

99: \newcommand\spec{\operatorname{spec}}

100: \newcommand\pspec{\spec_{pp}}

101: \newcommand\cspec{\spec_{c}}

102: \newcommand\FIO{{\mathcal I}}

103: \newcommand\SP{\operatorname{SP}}

104: \newcommand\Symc{S_c}

105: \newcommand\Symca{S_c^{\alpha}}

106: \newcommand\Symczero{S_c^{0,...,0}}

107: \newcommand\sci{{}^{\text{sc}}}

108: \newcommand\sct{\sci T^*}

109: \newcommand\scct{\sci\bar{T}^*}

110: \newcommand\Csc{C_{\text{sc}}}

111: \newcommand\SNpscd{(\SNp)^2_{\text{sc}}}

112: \newcommand\scdiag{\Delta_{\text{sc}}}

113: \newcommand\projscl{\pi^L_{\text{sc}}}

114: \newcommand\projscr{\pi^R_{\text{sc}}}

115: \newcommand\scHL{\sci H^{2,0}_{|\zeta|^2-\lambda^2}}

116: \newcommand\scHrg{\sci H^{2,0}_{\sqrt{g}}}

117: \newcommand\Hsc{H_{\text{sc}}}

118: \newcommand\WF{\operatorname{WF}}

119: \newcommand\WFp{\operatorname{WF^{\prime}}}

120: \newcommand\WFsc{\operatorname{WF}_{\text{sc}}}

121: \newcommand\WFscp{\operatorname{WF_{sc}^{\prime}}}

122: \newcommand\WFC{\operatorname{WF}_C}

123: \newcommand\WFCi{\operatorname{WF}_{C_i}}

124: \newcommand\elliptic{\operatorname{ell}}

125: \newcommand\Psop{\operatorname{\Psi}}

126: \newcommand\Psiscrs{\operatorname{\Psi_{sc}^{-2,\infty}}}

127: \newcommand\Psiscr{\operatorname{\Psi_{sc}^{-2,0}}}

128: \newcommand\Psiscrm{\operatorname{\Psi_{sc}^{0,2}}}

129: \newcommand\PsiscHam{\operatorname{\Psi_{sc}^{2,0}}}

130: \newcommand\Psisci{\operatorname{\Psi_{sc}^{*,*}}}

131: \newcommand\Psiscid{\operatorname{\Psi_{sc}^{0,0}}}

132: \newcommand\Psiscis{\operatorname{\Psi_{sc}^{0,\infty}}}

133: \newcommand\Psiscsi{\operatorname{\Psi_{sc}^{-\infty,0}}}

134: \newcommand\Psiscs{\operatorname{\Psi_{sc}^{-\infty,\infty}}}

135: \newcommand\Psiscalg{\operatorname{\Psi_{sc}^{\infty,-\infty}}}

136: \newcommand\nullHam{{\mathcal N}}

137: \newcommand\charD{\Sigma_{\Delta-\lambda^2}}

138: \newcommand\charLap{\Sigma_{\Delta-\lambda}}

139: \newcommand\Snl{\Sn_{\lambda}}

140: \newcommand\SNl{\SN_{\lambda}}

141: \newcommand\gammat{\tilde\gamma}

142: \newcommand\gammasc{\gamma}

143: \newcommand\taut{\tilde\tau}

144: \begin{comment}

145: \newcommand\Nout{N^+_{\lambda}}

146: \newcommand\Nin{N^-_{\lambda}}

147: \newcommand\Nio{N^{\pm}_{\lambda}}

148: \newcommand\El{E_{\lambda}}

149: \newcommand\Elt{\tilde E_{\lambda}}

150: \newcommand\Eil{E^i_{\lambda}}

151: \newcommand\Ejl{E^j_{\lambda}}

152: \newcommand\Eajl{E^{\alpha_j}_{\lambda}}

153: \newcommand\Eilt{\tilde E^i_{\lambda}}

154: \newcommand\Np{N^+}

155: \newcommand\Nm{N^-}

156: \newcommand\Npm{N^{\pm}}

157: \newcommand\Fin{F^-(\lambda)}

158: \newcommand\Fini{F^-_i(\lambda)}

159: \newcommand\Fout{F^+(\lambda)}

160: \newcommand\Fouti{F^+_i(\lambda)}

161: \newcommand\Foutj{F^+_j(\lambda)}

162: \newcommand\Rout{R^+_{\lambda}}

163: \newcommand\Routl{R^+_{\lambda^2}}

164: \newcommand\Routsgnl{R^{\sign\lambda}_{\lambda^2}}

165: \newcommand\Rin{R^-_{\lambda}}

166: \newcommand\Rinl{R^-_{\lambda^2}}

167: \newcommand\Rinsgnl{R^{-\sign\lambda}_{\lambda^2}}

168: \newcommand\Rio{R^{\pm}_{\lambda}}

169: \newcommand\Riol{R^{\pm}_{\lambda^2}}

170: \newcommand\Roi{R^{\mp}_{\lambda}}

171: \newcommand\Roil{R^{\mp}_{\lambda^2}}

172: \newcommand\Riob{R^{\pm}}

173: \newcommand\Roib{R^{\mp}}

174: \newcommand\Tio{T^{\pm}}

175: \newcommand\Tiob{T^{\pm}_{\ff}}

176: \newcommand\Toi{T^{\mp}}

177: \newcommand\Toib{T^{\mp}_{\ff}}

178: \newcommand\TIiob{T_I^{\pm}}

179: \newcommand\Rinb{R^-}

180: \newcommand\Rinbsgnl{R^{-\sign\lambda}}

181: \newcommand\Tin{T^-}

182: \newcommand\Tinb{T^-_{\ff}}

183: \newcommand\TIinb{T^-_I}

184: \newcommand\Routb{R^+}

185: \newcommand\Routbsgnl{R^{\sign\lambda}}

186: \newcommand\Tout{T^+}

187: \newcommand\Toutb{T^+_{\ff}}

188: \newcommand\TIoutb{T^+_I}

189: \newcommand\Rlkf{(|\xib|^2-(\lambda-i0)^2)^{-1}}

190: \newcommand\Rlk{\rho_0(\lambda)}

191: \newcommand\Rmlk{\rho_0(-\lambda)}

192: \newcommand\Rpmlk{\rho_0(\pm\lambda)}

193: \newcommand\Rlka{\rho_1(\lambda)}

194: \newcommand\Rlkb{\rho_2(\lambda)}

195: \newcommand\Rilk{\rho_i(\lambda)}

196: \newcommand\reduced{\natural}

197: \newcommand\Rlf{R_0(\lambda)}

198: \newcommand\Rla{R_1(\lambda)}

199: \newcommand\Rlb{R_2(\lambda)}

200: \newcommand\Ril{R_i(\lambda)}

201: \newcommand\Rlj{R_j(\lambda)}

202: \newcommand\Rlft{R_0(\lambda)}

203: \newcommand\Rflambda{R_0^{\reduced}(\sigma)}

204: \newcommand\RV{R^{\reduced}_V}

205: \newcommand\Rfsigma{R_0^{\reduced}(\sigma)}

206: \newcommand\Rfsigmah{R_0^{\reduced}(\sigma^{1/2})}

207: \newcommand\Rfzero{R_0^{\reduced}(0)}

208: \newcommand\RlV{R^{\reduced}_V(\sigma)}

209: \newcommand\RlVi{R^{\reduced}_{V_i}(\sigma)}

210: \newcommand\RlVt{R_V(\lambda)}

211: \newcommand\RlVtL{{R}_V^L(\lambda)}

212: \newcommand\RlVtR{{R}_V^R(\lambda)}

213: \newcommand\RlVit{{R}_{V_i}(\lambda)}

214: \newcommand\RlVta{{R}_V^{(1)}(\lambda)}

215: \newcommand\RlVtk{{R}_V^{(k)}(\lambda)}

216: \newcommand\RlVatV{{R}_{V_{\alpha}}(\lambda)V_{\alpha}}

217: \newcommand\RlVatVa{{R}_{V_{\alpha_1}}(\lambda)V_{\alpha_1}}

218: \newcommand\RlVatVb{{R}_{V_{\alpha_2}}(\lambda)V_{\alpha_2}}

219: \newcommand\RlVatVk{{R}_{V_{\alpha_k}}(\lambda)V_{\alpha_k}}

220: \newcommand\RlVatVkk{{R}_{V_{\alpha_{k+1}}}(\lambda)V_{\alpha_{k+1}}}

221: \newcommand\RlVaptV{{R}_{V_{\alpha'}}(\lambda)V_{\alpha'}}

222: \newcommand\RlVapptV{{R}_{V_{\alpha''}}(\lambda)V_{\alpha''}}

223: \newcommand\RlVajtV{{R}_{V_{\alpha_j}}(\lambda)V_{\alpha_j}}

224: \newcommand\RlVaktV{{R}_{V_{\alpha_k}}(\lambda)V_{\alpha_k}}

225: \newcommand\RlVakktV{{R}_{V_{\alpha_{k+1}}}(\lambda)V_{\alpha_{k+1}}}

226: \newcommand\Tl{T(\lambda)}

227: \newcommand\Tlt{\tilde\Tl}

228: \newcommand\Tltp{\tilde T'(\lambda)}

229: \newcommand\Tltpp{\tilde T''(\lambda)}

230: \newcommand\Tli{T_i(\lambda)}

231: \newcommand\Tlit{\tilde\Tli}

232: \newcommand\Tlip{T_i'(\lambda)}

233: \newcommand\Tlipp{T_i''(\lambda)}

234: \newcommand\Tlj{T_j(\lambda)}

235: \newcommand\Tla{T_{\alpha}(\lambda)}

236: \newcommand\Tlaa{T_{\alpha_1}(\lambda)}

237: \newcommand\Tlab{T_{\alpha_2}(\lambda)}

238: \newcommand\Tlak{T_{\alpha_k}(\lambda)}

239: \newcommand\Tlakt{\tilde\Tlak}

240: \newcommand\Tlaj{T_{\alpha_j}(\lambda)}

241: \newcommand\Tlajj{T_{\alpha_{j+1}}(\lambda)}

242: \newcommand\Tlajp{T_{\alpha_j}'(\lambda)}

243: \newcommand\Tlajpt{\tilde\Tlajp}

244: \newcommand\Tlajt{\tilde\Tlaj}

245: \newcommand\Tlakk{T_{\alpha_{k+1}}(\lambda)}

246: \newcommand\Tlakkp{T_{\alpha_{k+1}}'(\lambda)}

247: \newcommand\Tlap{T_{\alpha'}(\lambda)}

248: \newcommand\Tlapt{\tilde\Tlap}

249: \newcommand\Tlapp{T_{\alpha''}(\lambda)}

250: \newcommand\Tkl{T^{(k)}(\lambda)}

251: \newcommand\Tcl{T^{\flat}(\lambda)}

252: \newcommand\Fl{F(\lambda)}

253: \newcommand\BlVt{\tilde B_V(\lambda)}

254: \newcommand\KBlVt{K_{\BlVt}}

255: \newcommand\BlVaat{B_{V_{\alpha_1}}(\lambda)}

256: \newcommand\BV{B_V}

257: \newcommand\Bone{B_1}

258: \newcommand\Btwo{B_2}

259: \newcommand\Bthree{B_3}

260: \newcommand\Banyj{B_j}

261: \end{comment}

262: \newcommand\PlV{P_V(\lambda)}

263: \newcommand\PlVc{P_V^{\flat}(\lambda)}

264: \newcommand\Pl{P_0(\lambda)}

265: \newcommand\SVl{S_V(\lambda)}

266: \newcommand\Sjr{S_j^{\reduced}}

267: \newcommand\Rkp{{\mathcal R}^k_+}

268: \newcommand\Rkm{{\mathcal R}^k_-}

269: \newcommand\Rkpm{{\mathcal R}^k_{\pm}}

270: \newcommand\Phys{{\mathcal P}}

271: \newcommand\Pc{\overline{\mathcal P}}

272: \newcommand\pip{\pi^{\perp}}

273: \newcommand\xit{\tilde\xi}

274: \newcommand\zetat{\tilde\zeta}

275: \newcommand\etat{\tilde\eta}

276: \newcommand\sigmat{\tilde\sigma}

277: \newcommand\sigmahat{\hat\sigma}

278: \newcommand\thetat{\tilde\theta}

279: \newcommand\psit{\tilde\psi}

280: \newcommand\phit{\tilde\phi}

281: \newcommand\chit{\tilde\chi}

282: \newcommand\rhot{\tilde\rho}

283: \newcommand\xib{\bar\xi}

284: \newcommand\zetab{\bar\zeta}

285: \newcommand\thetab{\bar\theta}

286: \newcommand\etab{\bar\eta}

287: \newcommand\iotal{\iota_{\lambda}}

288: \newcommand\rhoat{\rhot_{\alpha_1}}

289: \newcommand\Lambdat{\tilde\Lambda}

290: \newcommand\poles{\Lambda'}

291: \newcommand\rpoles{\Lambda_p}

292: \newcommand\thresholds{\Lambda}

293: \newcommand\Vt{\tilde V}

294: \newcommand\half{{\frac{1}{2}}}

295: \newcommand\sigmah{\sigma^{1/2}}

296: \newcommand\bX{\partial X}

297: \newcommand\Deltabt{\tilde\Delta_0}

298: \newcommand\strip{\Omega_T}

299: \newcommand\Kf{K^{\flat}}

300: \newcommand\Gs{G^{\sharp}}

301: \newcommand\Gt{\tilde G}

302: \newcommand\Gb{\bar G}

303: \newcommand\Osc{\sci\Omega}

304: \newcommand\Osch{\sci\Omega^{\half}}

305: \newcommand\Oscmh{\sci\Omega^{-\half}}

306: \newcommand\Isc{I_{sc}}

307: \newcommand\Qzl{Q^0_{-\lambda}}

308: \newcommand\Lie{{\mathcal L}}

309: \newcommand\bl{{\text b}}

310: \newcommand\scl{{\text{sc}}}

311: \newcommand\sccl{{\text{scc}}}

312: \newcommand\Scl{{\text{3sc}}}

313: %\newcommand\Scl{{\text{Sc}}}

314: %\newcommand\ScLl{{\text{3sc,L}}}

315: \newcommand\ScLl{{\text{Sc,L}}}

316: %\newcommand\ScRl{{\text{3sc,R}}}

317: \newcommand\ScRl{{\text{Sc,R}}}

318: %\newcommand\Sccl{{\text{3scc}}}

319: \newcommand\Sccl{{\text{Scc}}}

320: \newcommand\sfl{{\operatorname{s\Phi}}}

321: \newcommand\sfi{{}^\sfl}

322: \newcommand\sus{{\text{sus}}}

323: \newcommand\Osfh{\sfi\Omega^{\half}}

324: \newcommand\Isf{I_\sfl}

325: \newcommand\XXb{X^2_\bl}

326: \newcommand\XXsc{X^2_\scl}

327: \newcommand\XXSc{X^2_\Scl}

328: \newcommand\XXScL{X^2_\ScLl}

329: \newcommand\XXScR{X^2_\ScRl}

330: \newcommand\MMsc{M^2_\scl}

331: \newcommand\Deltab{\Delta_\bl}

332: \newcommand\Deltasc{\Delta_\scl}

333: \newcommand\DeltaSc{\Delta_\Scl}

334: \newcommand\DeltaScL{\Delta_\ScLl}

335: \newcommand\DeltaScR{\Delta_\ScRl}

336: \newcommand\prs{\sigma}

337: \newcommand\Nsc{N_\scl}

338: \newcommand\Nscp{N_{\scl,p}}

339: \newcommand\Nff{N_{\ff}}

340: \newcommand\Nffz{N_{\ff,0}}

341: \newcommand\Nffzp{N_{\ff,0,p}}

342: \newcommand\Nffl{N_{\ff,l}}

343: \newcommand\Nffml{N_{\ff,-l}}

344: \newcommand\Nmf{N_{\mf}}

345: \newcommand\Nmfz{N_{\mf,0}}

346: \newcommand\Nmfl{N_{\mf,l}}

347: \newcommand\Nmfml{N_{\mf,-l}}

348: \newcommand\ffb{\operatorname{bf}}

349: \newcommand\Ffb{\operatorname{bf'}}

350: \newcommand\ffsc{\operatorname{sf}}

351: \newcommand\ffSc{\operatorname{sf_C}}

352: \newcommand\Ffsc{\operatorname{sf'}}

353: \newcommand\rff{\rho_{\ff}}

354: \newcommand\rmf{\rho_{\mf}}

355: \newcommand\rffb{\rho_{\ffb}}

356: \newcommand\rffsc{\rho_{\ffsc}}

357: \newcommand\rFfsc{\rho_{\Ffsc}}

358: \newcommand\rffSc{\rho_{\ffSc}}

359: \newcommand\rinf{\rho_{\infty}}

360: \newcommand\CL{C_L}

361: \newcommand\CR{C_R}

362: \newcommand\betab{\beta_\bl}

363: \newcommand\betasc{\beta_\scl}

364: \newcommand\betaSc{\beta_\Scl}

365: \newcommand\BetaSc{\bar\beta_\Scl}

366: \newcommand\betaScL{\beta_\ScLl}

367: \newcommand\betaScR{\beta_\ScRl}

368: \newcommand\sfT{{}^\sfl T^*}

369: \newcommand\sfN{{}^\sfl N^*}

370: \newcommand\ScT{{}^\Scl T^*}

371: \newcommand\SccT{{}^\Scl \bar T^*}

372: \newcommand\ScS{{}^\Scl S^*}

373: \newcommand\scS{{}^\scl S^*}

374: \newcommand\Tb{{}^\bl T}

375: \newcommand\Tsc{{}^\scl T}

376: \newcommand\Tsf{{}^\sfl T}

377: \newcommand\TSc{{}^\Scl T}

378: \newcommand\CSc{C_\Scl}

379: \newcommand\Lambdasc{{}^\scl\Lambda}

380: \newcommand\XXXb{X^3_\bl}

381: \newcommand\XXXsc{X^3_\scl}

382: \newcommand\XXXSc{X^3_\Scl}

383: \newcommand\XXXScO{X^3_{\Scl,O}}

384: \newcommand\XXXScF{X^3_{\Scl,F}}

385: \newcommand\XXXScS{X^3_{\Scl,S}}

386: \newcommand\XXXScC{X^3_{\Scl,C}}

387: \newcommand\KDsc{\operatorname{KD^{\half}_\scl}}

388: \newcommand\SDsc{\operatorname{SD^{\half}_\scl}}

389: \newcommand\SDsf{\operatorname{SD^{\half}_\sfl}}

390: \newcommand\KDSc{\operatorname{KD^{\half}_\Scl}}

391: \newcommand\KDScEF{\operatorname{KD^{E,F}_\Scl}}

392: \newcommand\Oh{\operatorname{\Omega^{\half}}}

393: \newcommand\WFSc{\WF_\Scl}

394: \newcommand\WFScmf{\WF_{\Scl,\mf}}

395: \newcommand\WFScff{\WF_{\Scl,\ff}}

396: \newcommand\WFScs{\WF_{\Scl,\prs}}

397: \newcommand\WFScp{\WF'_\Scl}

398: \newcommand\WFScmfp{\WF'_{\Scl,\mf}}

399: \newcommand\WFScffp{\WF'_{\Scl,\ff}}

400: \newcommand\WFScsp{\WF'_{\Scl,\prs}}

401: \newcommand\Diffscc{\Diff_\sccl}

402: \newcommand\DiffSc{\Diff_\Scl}

403: \newcommand\Diffsf{\Diff_\sfl}

404: \newcommand\DiffScc{\Diff_\Sccl}

405: \newcommand\DiffscI{\Diff_{\scl,\text{I}}}

406: \newcommand\VscI{\Vf_{\scl,\text{I}}}

407: \newcommand\DiffsV{\operatorname{Diff}_{\sus(V)}}

408: \newcommand\DiffsVsc{\operatorname{Diff}_{\sus(V),\scl}}

409: \newcommand\DiffsVCsc{\operatorname{Diff}_{\sus(V)-C,\scl}}   

410: \newcommand\Psisc{\Psop_\scl}

411: \newcommand\Psiscc{\Psop_\sccl}

412: \newcommand\PsiSc{\Psop_\Scl}

413: \newcommand\PsiScc{\Psop_\Sccl}

414: \newcommand\PsiSccml{\Psop^{m,l}_\Sccl}

415: \newcommand\PsiScxx{\Psop^{*,*}_\Scl}

416: \newcommand\PsiScml{\Psop^{m,l}_\Scl}

417: \newcommand\PsiScmz{\Psop^{m,0}_\Scl}

418: \newcommand\PsiScmmz{\Psop^{-m,0}_\Scl}

419: \newcommand\PsiSckz{\Psop^{k,0}_\Scl}

420: \newcommand\PsiScmmml{\Psop^{-m,-l}_\Scl}

421: \newcommand\Psiscmkk{\Psop^{-k,k}_\scl}

422: \newcommand\Psiscmmmkk{\Psop^{-m-k,k}_\scl}

423: \newcommand\Psiscmoo{\Psop^{-1,1}_\scl}

424: \newcommand\Psiscmz{\Psop^{m,0}_\scl}

425: \newcommand\Psiscmmz{\Psop^{-m,0}_\scl}

426: \newcommand\PsiSckmkl{\Psop^{km,kl}_\Scl}

427: \newcommand\PsiScmplp{\Psop^{m',l'}_\Scl}

428: \newcommand\PsiScmmpllp{\Psop^{m+m',l+l'}_\Scl}

429: \newcommand\Psiscml{\Psop^{m,l}_\scl}

430: \newcommand\PsiScid{\Psop^{0,0}_\Scl}

431: \newcommand\PsiSczo{\Psop^{0,1}_\Scl}

432: \newcommand\PsiScmii{\Psop^{-\infty,\infty}_\Scl}

433: \newcommand\PsiScmiz{\Psop^{-\infty,0}_\Scl}

434: \newcommand\PsiScmoo{\Psop^{-1,1}_\Scl}

435: \newcommand\PsisCid{\Psop^{0,0}_{\scl-C}}

436: \newcommand\PsisC{\Psop_{\scl-C}}

437: \newcommand\Psiinf{\Psop_{\infty}}

438: \newcommand\Psiinfid{\Psop_{\infty}^0}

439: \newcommand\PsiFinf{\Psop_{\infty-\Fr}}

440: \newcommand\PsisVscml{\Psop^{m,l}_{\sus(V),\scl}}

441: \newcommand\PsisVsc{\Psop_{\sus(V),\scl}}

442: \newcommand\PsisVpsc{\Psop_{\sus(V_p),\scl}}

443: \newcommand\PsisVCSc{\Psop_{\sus(V)-C,\scl}}

444: \newcommand\SFinf{S_{\infty-\Fr}}

445: \newcommand\YsVC{Y^2_{\sus(V)-C,\scl}}

446: \newcommand\ffYsc{\ffsc_{\sus(V)}}

447: \newcommand\SXC{S(X;C)}

448: \newcommand\Ios{I_{\text{os}}}

449: \newcommand\pbL{\pi^2_{\bl,{\text L}}}

450: \newcommand\pbR{\pi^2_{\bl,{\text R}}}

451: \newcommand\pscL{\pi^2_{\scl,{\text L}}}

452: \newcommand\pscR{\pi^2_{\scl,{\text R}}}

453: \newcommand\PbO{\pi^3_{\bl,{\text O}}}

454: \newcommand\PscO{\pi^3_{\scl,{\text O}}}

455: \newcommand\PScO{\pi^3_{\Scl,{\text O}}}

456: \newcommand\PScF{\pi^3_{\Scl,{\text F}}}

457: \newcommand\PScC{\pi^3_{\Scl,{\text C}}}

458: \newcommand\PScS{\pi^3_{\Scl,{\text S}}}

459: \newcommand\pScL{\pi^2_{\Scl,{\text L}}}

460: \newcommand\pScR{\pi^2_{\Scl,{\text R}}}

461: \newcommand\CLF{\CL^F}

462: \newcommand\CLO{\CL^O}

463: \newcommand\CLS{\CL^S}

464: \newcommand\CLC{\CL^C}

465: \newcommand\DeltaYb{\Delta_{\bl,Y}}

466: \newcommand\DeltaYsc{\Delta_{\sus-\scl}}

467: \newcommand\Vf{{\mathcal V}}

468: \newcommand\Vb{{\mathcal V}_{\bl}}

469: \newcommand\Vsc{{\mathcal V}_{\scl}}

470: \newcommand\VSc{{\mathcal V}_{\Scl}}

471: \newcommand\Vsf{{\mathcal V}_{\sfl}}

472: \newcommand\VfI{\Vf_{\text{I}}}

473: \newcommand\VfIq{\Vf_{\text{I},q}}

474: \newcommand\scH{{}^\scl H}

475: \newcommand\scHg{\scH_g}

476: \newcommand\xh{\hat x}

477: \newcommand\xb{\bar x}

478: \newcommand\Yh{\hat Y}

479: \newcommand\Yb{\bar Y}

480: \newcommand\hb{\bar h}

481: \newcommand\xih{\hat\xi}

482: \newcommand\etah{\hat\eta}

483: \newcommand\muh{\hat\mu}

484: \newcommand\mut{\tilde\mu}

485: \newcommand\mub{\bar\mu}

486: \newcommand\mubh{\widehat{\bar\mu}}

487: \newcommand\yb{\bar y}

488: \newcommand\phib{\bar \phi}

489: \newcommand\ub{\bar u}

490: \newcommand\Qb{\bar Q}

491: \newcommand\Wbp{{\bar W}^\perp}

492: \newcommand\Wp{W^\perp}

493: \newcommand\Kt{\tilde K}

494: \newcommand\Wt{\tilde W}

495: \newcommand\Ut{\tilde U}

496: \newcommand\xt{\tilde x}

497: \newcommand\yt{\tilde y}

498: \newcommand\ft{\tilde f}

499: \newcommand\fs{f^{\sharp}}

500: \newcommand\at{\tilde a}

501: \newcommand\htil{\tilde h}

502: \newcommand\gt{\tilde g}

503: \newcommand\Ht{\tilde H}

504: \newcommand\Mt{\tilde M}

505: \newcommand\St{\tilde S}

506: \newcommand\Pt{\tilde P}

507: \newcommand\Rt{\tilde R}

508: \newcommand\qt{\tilde q}

509: \newcommand\Qt{\tilde Q}

510: \newcommand\Xb{\bar X}

511: \newcommand\lambdat{\tilde\lambda}

512: \newcommand\epst{\tilde\epsilon}

513: \newcommand\At{\tilde A}

514: \newcommand\Ah{\hat A}

515: \newcommand\Bh{\hat B}

516: \newcommand\Gh{\hat G}

517: \newcommand\Hh{\hat H}

518: \newcommand\Qh{\hat Q}

519: \newcommand\Ph{\hat P}

520: \newcommand\Nh{\hat N}

521: \newcommand\Sh{\hat S}

522: \newcommand\Gcal{{\mathcal G}}

523: \newcommand\GcalC{{\mathcal G}_C}

524: \newcommand\Jcal{{\mathcal J}}

525: \newcommand\JcalC{{\mathcal J}_C}

526: \newcommand\Miff{\ \text{iff}\ }

527: \newcommand\Mif{\ \text{if}\ }

528: \newcommand\Mand{\ \text{and}\ }

529: \newcommand\Mor{\ \text{or}\ }

530: \newcommand\Mst{\ \text{s.t.}\ }

531: 

532: \newcommand\pa{\partial}

533: \newcommand\sub{\operatorname{sub}}

534: \newcommand\fc{\operatorname{fc}}

535: \def\Vsphi{{}^{s\Phi}\Nu}

536: \def\tx{\tilde x}

537: \def\sphihalfdens{{}^{s\Phi}\Omega^{1/2}}

538: \def\mf{\operatorname{mf}}

539: \def\Diff{\operatorname{Diff}}

540: \renewcommand\sc{\operatorname{sc}}

541: \newcommand\Diag{\operatorname{Diag}}

542: \setcounter{secnumdepth}{3}

543: \newtheorem{lemma}{Lemma}[section]

544: \newtheorem{prop}[lemma]{Proposition}

545: \newtheorem{thm}[lemma]{Theorem}

546: \newtheorem{cor}[lemma]{Corollary}

547: \newtheorem{result}[lemma]{Result}

548: \newtheorem*{thm*}{Theorem}

549: \numberwithin{equation}{section}

550: \theoremstyle{remark}

551: \newtheorem{rem}[lemma]{Remark}

552: \theoremstyle{definition}

553: \newtheorem{Def}[lemma]{Definition}

554: \def\signature#1#2{\par\noindent#1\dotfill\null\\*

555: {\raggedleft #2\par}}

556: 

557: \renewcommand\div{\operatorname{div}}

558: \newcommand\Cal{\mathcal}

559: \newcommand\Lap{\varDelta}

560: \newcommand\del{\partial}

561: \renewcommand\a{\alpha}

562: %\predefine\b{\barunder}

563: \renewcommand\b{\beta}

564: \newcommand\ga{\gamma}

565: %\predefine\d{\dotunder}

566: \renewcommand\d{\delta}

567: \renewcommand\th{\theta}

568: \newcommand\e{\epsilon}

569: \newcommand\s{\sigma}

570: %\predefine\o{\orsted}

571: \renewcommand\o{\omega}

572: \newcommand\Sig{\Sigma}

573: \newcommand\CI{{\Cal C}^{\infty}}

574: \newcommand\RR{{\Bbb R}}

575: \newcommand\ZZ{{\Bbb Z}}

576: \newcommand\CC{{\Bbb C}}

577: \newcommand\xa{x_{\alpha}}

578: %\newcommand\xb{x_{\beta}}

579: \newcommand\ab{\alpha \beta}

580: \newcommand\Ma{M_{\alpha}}

581: \newcommand\Va{V_{\alpha}}

582: \newcommand\Mb{M_{\beta}}

583: \newcommand\Hab{H_{\alpha\beta}}

584: \newcommand\bT{{}^bT}

585: \newcommand\bcT{{}^bT^*}

586: \newcommand\bw{\bigwedge}

587: \newcommand\bbw[1]{{}^b\bw{}^{#1}}

588: \newcommand\Xe{X_{\epsilon}}

589: \newcommand\ba{\beta \alpha}

590: \newcommand\ha{\frac12}

591: \newcommand\ra{\rho_{\alpha}}

592: \newcommand\be{{}^b\eta}

593: \newcommand\gab{\ga_{\ab}}

594: \newcommand\Spin{\text{Spin\,}}

595: \newcommand\ind{\text{ind\,}}

596: \newcommand\Sdet{\text{Sdet\,}}

597: \newcommand\se{s_{\e}}

598: \newcommand\gep{g_{\e}}

599: \newcommand\bg{\beta \gamma}

600: \newcommand\ag{\alpha \gamma}

601: 

602: %AH

603: \newcommand\etag{\eth_{\Cal G}}

604: %\newcommand\Id{\operatorname{Id}}

605: \newcommand\tr{\operatorname{tr}}

606: \newcommand\proj{\operatorname{proj}}

607: \newcommand\gi{\operatorname{generalized inverse of}}

608: \newcommand\olS{\overline S}

609: 

610: %RBM

611: \newcommand\bH[1]{{}^b\kern-1pt H^{#1}}

612: \newcommand\bL{{}^b\kern-1pt L}

613: \newcommand\rest{\restriction}

614: \newcommand\nul{\operatorname{null}}

615: \newcommand\clos{\operatorname{clos}}

616: \newcommand\ins[1]{\overset\circ{#1}}

617: \newif\ifwantold

618: \newcommand\old[1]{\ifwantold #1\else\relax\fi}

619: \newif\ifhide

620: \let\pa\del

621: \newcommand\sola{{\frak o}}

622: \newcommand\SO{\operatorname{SO}}

623: \newcommand\Xo{X_{\epsilon_0}}

624: %\newcommand\Vb{{\Cal V}_b}

625: \newcommand\Hba{H_{\beta\alpha}}

626: \newcommand\APS{\operatorname{APS}}

627: %\newcommand\sign{\operatorname{sign}}

628: \newcommand\AS{\operatorname{AS}}

629: \newcommand\Ch{\operatorname{Ch}}

630: \newcommand\Refsect[1]{\S\ref{#1}}

631: \newcommand\Mwhere{\text{ where }}

632: \newcommand\Mnear{\text{ near }}

633: \newcommand\lang{\big\langle}

634: \newcommand\rang{\big\rangle}

635: \newcommand\lb{\operatorname{lb}}

636: \newcommand\rb{\operatorname{rb}}

637: \newcommand\bF{\operatorname{bf}}

638: \newcommand\Ls{L^{\#}}

639: \def\ilabel#1{\label{#1}}

640: 

641: \def\thebibliography#1{\section*{References}\list

642:  {[\arabic{enumi}]}{\settowidth\labelwidth{[#1]}\leftmargin\labelwidth

643:  \advance\leftmargin\labelsep

644:  \usecounter{enumi}}

645:  \def\newblock{\hskip .11em plus .33em minus .07em}

646:  \sloppy\clubpenalty4000\widowpenalty4000

647:  \sfcode`\.=1000\relax}

648: \let\endthebibliography=\endlist

649: 

650: 

651: \def\onto{\to}

652: \def\hz{\hat z}

653: \def\hzz{\hat z_0}

654: \def\hzp{\hat z'}

655: \def\pa{\partial}

656: \def\({\left(}

657: \def\){\right)}

658: \def\hV{\hat V_i *}

659: \def\ol{\overline}

660: \def\zetahat{\hat \zeta'}

661: \def\modzeta{|\zeta'|}

662: \def\zhat{\hat z}

663: \def\Sch{Schr\"odinger }

664: \def\bRN{\overline {{\Bbb R}^N}}

665: \def\scTX{{}^{sc} T^*_{\pa X} X}

666: \def\scRN{{}^{sc} T^*_{S^{N-1}} \overline {{\Bbb R}^N}}

667: \def\tscRN{{}^{sc} T^* \overline {{\Bbb R}^N}}

668: \def\scpsdo#1#2{{}^{sc}\Psi^{#1, #2}_{cl}(\overline {{\Bbb R}^N})}

669: \def\scbs#1#2{\sigma_{#2}}

670: \def\scwf{{}^{sc} \text{WF}}

671: \def\legsym#1{{}^{sc} \sigma^{#1} }

672: \def\half{{\frac1{2}}}

673: \def\modS{\text{ mod } {\Cal S} }

674: \def\olV{ {\overline{V}}}

675: \def\hypp{ \hat y''}

676: \def\zetapy{ {\zeta''}^{\perp}}

677: \def\Id{\operatorname{Id}}

678: \def\bgamma{\overline \gamma}

679: \def\bL{\overline L}

680: \def\Gs{{G^\sharp}}

681: \def\Gsl{G^\sharp(\lambda)}

682: \def\Gt{\tilde G}

683: \def\Gh{\hat G}

684: \def\exL{ \overline L_{k, \alpha}}

685: \def\Id{\operatorname{Id}}

686: \def\RN{{\Bbb R}^N}

687: \def\lam{\lambda}

688: \def\Rml{R_-(\lambda)}

689: \def\uka{u_{l, \alpha}}

690: \def\eka{e_{l, \alpha}}

691: \def\tphi{\tilde \phi}

692: \def\utl{u_{\theta, \lambda}}

693: \def\e{\epsilon}

694: \def\vka{v_{l, \alpha}}

695: \def\WF{\operatorname{WF}}

696: \def\CR{\operatorname{CR}}

697: \def\wzp{\widehat{z'}}

698: \def\wzpp{\widehat{z''}}

699: \def\cl{\operatorname{cl}}

700: \def\interior{\operatorname{int}}

701: \def\uol{u_{\omega, \lambda}}

702: \def\dnu#1{\frac{\partial {#1}}{\partial \nu}}

703: \def\ac{\operatorname{ac}}

704: \def\td{\tilde}

705: \def\ot{\leftarrow}

706: \def\cdotinfty{\dot C^\infty}

707: \def\sm{\sigma_{\operatorname{max}}}

708: \def\lzr{\langle z \rangle}

709: \def\ep{\epsilon}

710: \def\Op{\operatorname{Op}}

711: \def\Rea{\operatorname{Re}}

712: \def\Imag{\operatorname{Im}}

713: \def\phg{\operatorname{phg}}

714: \def\Nu{{\Cal V}}

715: \def\ang#1{\langle {#1} \rangle}

716: \def\psido{$\Psi$.d.o. }

717: \def\psidos{$\Psi$.d.o.s }

718: \def\resA{(A-\lam)^{-1}}

719: \def\cl{\operatorname{cl}}

720: \def\slim{\operatorname{s-lim}}

721: \def\ord{\operatorname{ord}}

722: \def\ad{\operatorname{ad}}

723: \def\Psizcl{\Psi^{\ZZ}_{\cl}}

724: \def\sc{\operatorname{sc}}

725: \def\X2sc{X^2_{\sc}}

726: \def\ilg{\operatorname{ilg}}

727: \def\inc{\operatorname{in}}

728: \def\out{\operatorname{out}}

729: \def\ang#1{\big\langle #1 \big\rangle}

730: \def\GISwf{ {}^{\operatorname{GIS}} \WF}

731: \def\op{\operatorname{op}}

732: \def\Rn{\RR^n}

733: \def\sigb{\sigma_b}

734: \long\def\skip#1{}

735: \def\CIdot{\dot \CI}

736: \def\symrel#1#2#3{\sigma^{#1}_{\mathcal{#2}, #3}}

737: \def\sc{\operatorname{sc}}

738: \def\sf{\operatorname{s\Phi}}

739: \def\scOh{{}^{\sc}{\Omega^{\half}}}

740: \def\sfOh{{}^{\sf}{\Omega^{\half}}}

741: \def\schd{\left|\frac{dx dy}{x^{n+1}}\right|^\half}

742: \def\xt{\tilde x}

743: \def\jsc{j_{\sc}}

744: \newcommand\symbint{\sigma_{\operatorname{int}}}

745: \newcommand\symbb{\sigma_{\pa}}

746: \def\Lt{\tilde L}

747: \def\pit{\tilde \pi}

748: \def\ybar{\overline{y}}

749: \def\sfOh{{}^{\sf} \Omega^\half}

750: \def\Sm{S^{[m]}}

751: \def\Rp{R_+ (\lam)}

752: \def\Rm{R_- (\lam)}

753: \def\Rpm{R_\pm (\lam)}

754: \def\Rpln#1{R_+^{(#1)}(\lam)}

755: \def\XXbt{\tilde{\XXb}}

756: \def\Lh{\hat L}

757: \def\Lhp{\hat L^+}

758: \def\Lpl{L_+(\lam)}

759: \def\Lplh{\hat L_+(\lam)}

760: \def\Lsl{\Ls(\lam)}

761: \def\bff{\operatorname{bf}}

762: \def\kid{K_{\Id}}

763: \def\Rt{\tilde R}

764: 

765: 

766: \title[The resolvent for Laplace-type operators] 

767: {The resolvent for Laplace-type operators on asymptotically conic spaces}

768: \author{Andrew Hassell}

769: \address{Centre for Mathematics and its Applications, Australian National

770:   University, Canberra ACT 0200 Australia}

771: \email{hassell@maths.anu.edu.au}

772: \author[Andr\'as Vasy]{Andr\'as Vasy}

773: \address{Department of Mathematics, University of California, Berkeley,

774:   CA 94720, U.S.A.}

775: \email{andras@math.berkeley.edu}

776: 

777: %\centerline{\today}

778: 

779: \begin{abstract}Let $X$ be a compact manifold with boundary, and $g$ a

780: scattering metric on $X$, which may be either of short range or `gravitational'

781: long range type. Thus, $g$ gives $X$ the geometric structure of a complete

782: manifold with an asymptotically conic end. 

783: Let $H$ be an operator of the form $H =

784: \Lap + P$, where $\Lap$ is the Laplacian with respect to $g$ and

785: $P$ is a self-adjoint first order scattering differential operator with

786: coefficients vanishing at $\pa X$ and satisfying a `gravitational'

787: condition. We define a symbol calculus for Legendre distributions on

788: manifolds with codimension two corners and use it to give a

789: direct construction of the resolvent kernel of $H$, $R(\sigma + i0)$, for

790: $\sigma$ on the positive real axis. 

791: In this approach, we do not use the limiting

792: absorption principle at any stage; instead we construct a parametrix  

793: which solves the resolvent equation up to a compact error term and then use

794: Fredholm theory to remove the error term. 

795: \end{abstract}

796: \maketitle

797: 

798: \section{Introduction} Scattering metrics are a class of Riemannian 

799: metrics which describe manifolds which

800: are geometrically complete, and asymptotically conic at infinity. 

801: We consider manifolds which have 

802: one end which is diffeomorphic to $Y \times [1, \infty)_r$, where

803: $Y$ is a closed manifold, and is metrically asymptotic to $dr^2 + r^2 h$,

804: where $h$ is a Riemannian metric on $Y$, as $r \to \infty$. The precise

805: definition is given in Definition~\ref{def-sc-met} below. Examples include

806: the standard metric and the Schwartzschild metric on Euclidean space. 

807: 

808: In this paper we give a direct construction of the

809: outgoing resolvent kernel, $R(\sigma + i0) = (H - (\sigma + i0))^{-1}$, for

810: $\sigma$ on the real axis, where $H$ is a perturbation of the Laplacian with

811: respect to a scattering metric. The incoming resolvent kernel, $R(\sigma -

812: i0)$, may be obtained by taking the formal adjoint kernel.  

813: 

814: 

815: The strategy of the proof is to compactify the space to a compact manifold

816: $X$ and use the scattering calculus of Melrose, as well as the calculus of

817: Legendre distributions of Melrose-Zworski, extended by us in \cite{HV2}. The

818: oscillatory behaviour of the resolvent kernel is analyzed in terms of the

819: `scattering wavefront set' at the boundary. Using propagation of singularity

820: theorems for the scattering wavefront set leads to an ansatz for the

821: structure of the resolvent kernel as a sum of a pseudodifferential term and

822: Legendre distributions of various types. The calculus of Legendre

823: distributions allows us to construct a rather precise parametrix for the

824: resolvent in this class, with an error term $E$ which is compact. Using the

825: parametrix, we show that one can make a finite rank correction to the

826: parametrix which makes $\Id + E$ invertible, and thus can correct the

827: parametrix to the exact resolvent. 

828: 

829: As compared to the method of \cite{HV2},

830: where the authors previously constructed the resolvent, the construction is

831: direct in two senses. First, we write down rather explicitly a parametrix

832: for $R(\sigma + i0)$ and then solve away the error using Fredholm

833: theory. In \cite{HV2}, by contrast, the resolvent was constructed via the

834: spectral measure, which itself was constructed from the Poisson

835: operator. Second, we make no use of the limiting absorption principle; that

836: is, we work directly at the real axis in the spectral variable rather than

837: taking a limit $R(\sigma + i\ep)$ as $\ep \to 0$. We then prove a posteriori

838: that the operator constructed is equal to this limit. 

839: 

840: \  

841: 

842: Let us briefly describe the main result here. We consider an operator $H$ of the

843: form $H = \Lap + P$ acting on half densities, where $P$ is, in the first

844: place, a short range 

845: perturbation, that is, a first order self-adjoint differential

846: operator with coefficients vanishing to

847: second order at infinity. (Later, we show that there is a simple extension

848: to metrics and perturbations of `long range gravitational type', which

849: includes the Newtonian or Coulomb potential and

850: metrics of interest in general relativity.) We remark that

851: the Riemannian half-density $|dg|^{1/2}$ trivializes the half-density

852: bundle, and operators on functions can be regarded as operators on

853: half-densities via this trivialization.

854: Given $\lam > 0$, we solve for a kernel $\Rt(\lam)$ on $X^2$ which satisfies

855: \begin{equation}

856: (H - \lam^2) \Rt(\lam) = K_{\Id},

857: \ilabel{eq-Rt}\end{equation}

858: where $K_{\Id}$ is the kernel of the identity operator on half

859: densities. More precisely, we 

860: consider this equation on $\XXb$, which is the space $X^2$ with the corner

861: blown up. This allows us to use the scattering wavefront set at the `front

862: face' (the face resulting from blowing up the corner) 

863: to analyze singularities, which is an absolutely crucial part of the

864: strategy. The kernel $\Rt(\lam)$ is also required to satisfy a wavefront set

865: condition at the front face, which is the analogue of the outgoing

866: Sommerfeld radiation condition. 

867: 

868: We cannot find $\Rt(\lam)$ exactly in one step, so first we look for an

869: approximation $G(\lam)$ of it.   

870: The general strategy is to find $G(\lam)$ which solves away the singularities

871: of the right hand side, $K_{\Id}$, of \eqref{eq-Rt}. Singularities should be

872: understood both in the sense of interior singularities and oscillations, or

873: growth, at the boundary, as measured by the scattering wavefront set. 

874: 

875: The first step is to find a pseudodifferential approximation which solves

876: away the interior singularities of $K_{\Id}$, which is a conormal

877: distribution supported on the diagonal. This can be done and removes

878: singularities except at the boundary of the diagonal, where $H - \lam^2$ is

879: not elliptic (in the sense of the boundary wavefront set). In fact, the

880: singularities which remain lie on a Legendrian submanifold $N^* \diag_\bl$

881: at the boundary of $\diag_\bl$ (see \eqref{Ndiag}). Singularities of

882: $G(\lam)$ can be expected to propagate in a Legendre submanifold $\Lpl$ which

883: is the bicharacteristic flowout from the intersection of $N^* \diag_\bl$

884: and the characteristic variety of $H - 

885: \lam^2$. (The geometry here is precisely that of the fundamental solution of

886: the wave operator in $\RR^{n+1}$, which is captured by the intersecting

887: Lagrangian calculus of Melrose-Uhlmann \cite{Rbm-Uhl:Intersecting}.) This

888: Legendre has conic singularities at another Legendrian, $\Lsl$, 

889: which is `outgoing'. Thus, in view of the calculus of Legendre distributions

890: of Melrose-Zworski and the authors, the simplest one could hope for is that

891: the resolvent on the real axis is the sum of a pseudodifferential term, an

892: intersecting Legendre 

893: distribution associated to $(N^* \diag_\bl, \Lpl)$ and a Legendre conic pair

894: associated to $(\Lpl, \Lsl)$. This is the case:

895: 

896: \begin{thm}\label{main-result} Let $H$ be a short range perturbation of a

897: short range scattering metric on $X$. Then, for $\lam > 0$, the outgoing

898: resolvent $R(\lam^2 + i0)$ lies in the class \eqref{seek}, that is, it is the

899: sum of a scattering pseudodifferential operator of order $-2$, an

900: intersecting Legendre distribution of order $-1/2$ associated to $(N^*

901: \diag_\bl, \Lpl)$ and a Legendrian conic pair associated to $(\Lpl, \Lsl)$

902: of orders $-1/2$ at $\Lpl$, $(n-2)/2$ at $\Lsl$ and $(n-1)/2$ at the left

903: and right boundaries. 

904: 

905: If $H$ is of long range gravitational type, then the same result holds

906: except that the Legendre conic pair is multiplied by a complex power of the

907: left and right boundary defining functions. 

908: \end{thm} 

909: 

910: This theorem was already proved in our previous work \cite{HV2}, so it is the

911: method that is of principal interest here. By comparison with \cite{HV2},

912: the proof is conceptually much shorter; it does not use any results from

913: \cite{RBMSpec} or \cite{RBMZw}, though it makes substantial use of machinery

914: from 

915: \cite{RBMZw}. But the main point we wish to emphasize is that the proof

916: works directly on the spectrum and 

917: nowhere uses the limiting absorption principle, a method of attack that we think

918: will be useful elsewhere in scattering theory. It seems that things which

919: are easy to prove with this method are difficult with the limiting absorption

920: principle, and vice versa. For example, it is immediate from our results

921: that if $f$ is compactly supported in the interior of $X$, then $u = R(\lam^2 +

922: i0)f$ is such that $x^{-(n-1)/2} e^{-i\lam/x} u \in \CI(X)$, while it is not

923: so easy to see that the resolvent is a bounded operator from $x^l L^2$ to

924: $x^{-l} L^2$ for any $l > 1/2$. Using the limiting absorption principle, it

925: is the second statement that is much easier to derive (following

926: \cite{GIS} for example). Thus, we hope that

927: this type of approach will complement other standard methods in scattering

928: theory. 

929: 

930: \

931: 

932: 

933: In the next section, we describe the machinery required, including the

934: scattering calculus on manifolds with boundary, the scattering-fibred

935: calculus on manifolds with codimension two corners, and Legendre

936: distributions in these contexts. The b-double space, which is a blown up

937: version of the double space $X \times X$ which carries the resolvent kernel,

938: is also described. The discussion here is rather concise, but there are more

939: leisurely treatments in \cite{RBMZw} and \cite{HV2}. 

940: 

941: The third section gives a symbol calculus for Legendre distributions on

942: manifolds with codimension two corners. This is a straightforward

943: generalization from the codimension one case. 

944: 

945: The fourth section is the heart of the paper, where we construct the

946: parametrix $G(\lam)$ for the resolvent kernel. Propagation of singularity

947: theorems show that the simplest space of functions in which one could hope

948: to find the resolvent kernel is given by \eqref{seek}. We can in fact

949: construct a parametrix for the resolvent in this class. In the fifth section

950: this is extended to long range metrics and perturbations. 

951: 

952: In the final section we show that one can modify the parametrix so that the

953: error term $E(\lam)$ is such that $\Id + E(\lam)$ is invertible. This is done by

954: showing that the range of $H - \lam^2$ on the sum of $\CIdot(X)$ and $G(\lam)

955: \CIdot(X)$ is dense on suitable weighted Sobolev spaces. Thus the parametrix

956: may be corrected to an exact solution of \eqref{eq-Rt}. Such a result also

957: shows the absence of positive eigenvalues for $H$. Finally, we show that the

958: kernel so constructed has an analytic continuation to the upper half plane

959: and agrees with the resolvent there. 

960: 

961: {\it Notation and conventions. } On a compact manifold with boundary, $X$, we

962: use $\dCinf(X)$ to denote the class of smooth functions, all of whose

963: derivatives 

964: vanish at the boundary, with the usual Fr\'echet topology, and $\dist(X)$ to

965: denote its 

966: topological dual. On the radial compactification of $\RR^n$ these correspond

967: to the space of Schwartz functions and tempered distributions, respectively. The

968: Laplacian $\Delta$ is taken to be positive. The space $L^2(X)$ is taken with

969: respect to the Riemannian density induced by the scattering metric $g$. This

970: density has the form $a \, dx dy/x^{n+1}$ near the boundary, where $a$ is

971: smooth.  

972:  

973: 

974: {\it Acknowledgements. } We wish to thank Richard Melrose and Rafe Mazzeo for

975: suggesting the problem, and for many very

976: helpful conversations. A.\ H.\ is grateful to the Australian Research

977: Council for financial support. A.\ V.\ thanks the NSF for partial support,

978: NSF grant \#DMS-99-70607. 

979: 

980: 

981: \section{Preliminaries}

982: 

983: \subsection{Scattering calculus}

984: 

985: \begin{comment}

986: In this paper we give a symbol calculus for the Legendre distributions

987: introduced in \cite{HV2} and use it to construct the resolvent kernel of

988: self-adjoint operators $H$ of the form $H = \Lap + P$, where $\Lap$ is the

989: Laplacian with respect to a scattering metric and $P$ is a first order

990: `scattering differential operator' (defined in the next subsection)

991: with coefficients vanishing at $\pa X$. A more sophisticated symbol calculus

992: for Legendre distributions on manifolds with corners of arbitary codimension

993: will be given in \cite{HV4}. 

994: \end{comment}

995: 

996: Let $X$ be a manifold with boundary $\pa X = Y$. Near the boundary we will

997: write local coordinates in the form $(x,y)$ where $x$ is a boundary defining

998: function and $y$ are coordinates on $Y$ extended to a collar neighbourhood

999: of $\pa X$. 

1000: 

1001: We begin by giving the definition of a scattering metric.  

1002: The precise requirements for the metric

1003: (and many other things besides) are easiest to formulate in terms of a

1004: compactification of the space. Taking the function $x = r^{-1}$ as a

1005: boundary defining function and adding a copy of $Y$ at $x=0$ yields a

1006: compact manifold, $X$, with boundary $\pa X = Y$. Then the definition of

1007: scattering metric is given in terms of $X$ in Definition~\ref{def-sc-met}

1008: below. Regularity statements 

1009: for the metric coefficients are in terms of the 

1010: $\CI$ structure on $X$; this is a strong requirement, being equivalent to

1011: the existence of a complete asymptotic expansion, together

1012: with all derivatives, in inverse powers of $r$ as $r \to \infty$. The

1013: benefit of such a strong requirement is that we get complete asymptotic

1014: expansions for the resolvent kernel, and correspondingly, mapping properties

1015: of the resolvent on spaces of functions with complete asymptotic

1016: expansions. 

1017: 

1018: \begin{Def}\ilabel{def-sc-met} 

1019: A (short range) scattering metric on $X$ is a Riemannian metric

1020: $g$ in the interior of $X$ which takes the form

1021: \begin{equation}

1022: g = \frac{dx^2}{x^4} + \frac{h'}{x^2},

1023: \label{sc-metric-1}\end{equation}

1024: where $h'$ is a smooth symmetric 2-cotensor on $X$ which restricts to the

1025: boundary to be a metric $h$ on $Y$ \cite{RBMSpec}. A long range scattering

1026: metric is a 

1027: metric in the interior of $X$ which takes the form

1028: \begin{equation}

1029: g = a_{00} \frac{dx^2}{x^4} + \frac{h'}{x^2},

1030: \label{sc-metric-2}\end{equation}

1031: where $a_{00}$ is smooth on $X$, $a_{00} = 1 + O(x)$, and $h'$

1032: is as above \cite{Vasy:Geometric}. If $a_{00} = 1 - cx + O(x^2)$ for some

1033: constant $c$ we call $g$ a gravitational long range scattering metric. 

1034: \end{Def}

1035: 

1036: {\bf Examples.} Flat Euclidean space has a metric which in polar coordinates

1037: takes the form

1038: $$

1039: dr^2 + r^2 d\omega^2,

1040: $$

1041: where $d\omega^2$ is the standard metric on $S^{n-1}$. Compactifying Euclidean

1042: space as above, we obtain a ball with $x = r^{-1}$ as boundary defining

1043: function, and then the flat metric becomes

1044: $$

1045: \frac{dx^2}{x^4} + \frac{d\omega^2}{x^2},

1046: $$

1047: which is a short range scattering metric. 

1048: 

1049: The Schwartzschild metric on $\RR^n$ takes the form near infinity

1050: $$

1051: \left( 1 - \frac{2m}{r} \right) dr^2 + r^2 d\omega^2,

1052: $$

1053: which under the same transformation leads to a gravitational long range

1054: scattering metric 

1055: $$

1056: (1 - 2mx) \frac{dx^2}{x^4} + \frac{d\omega^2}{x^2}.

1057: $$

1058: The constant $m = c/2$ is interpreted as the mass in general relativity. 

1059: 

1060: \

1061: 

1062: The natural Lie Algebra corresponding to the class of scattering metrics

1063: on $X$ is the scattering Lie Algebra 

1064: $$

1065: \Vsc(X) = \{ V \mid V = x W, \ \text{where } W \text{ is a } \CI \text{ vector

1066:   field on } X \text{ tangent to } \pa X \}.

1067: $$

1068: Clearly this Lie Algebra can be localized to any open set. In the interior

1069: of $X$, it consists of all smooth vector fields, while near the boundary it

1070: is equal to the $\CI(X)$-span of the vector fields $x^2 \pa_x$ and $x

1071: \pa_{y_i}$. Hence it is the space of smooth sections of a vector bundle,

1072: denoted $\Tsc X$, the scattering tangent bundle. Any scattering metric

1073: turns out to be a smooth fibre metric on $\Tsc X$. The dual bundle, denoted

1074: $\sct X$, is called the scattering cotangent bundle; near the boundary,

1075: smooth sections are generated over $\CI(X)$ by $dx/x^2$ and $dy_i/x$. A

1076: general point in $\sct _p X$ can be thought of as the value of a

1077: differential $d(f/x)$ at $p$, where $f \in \CI(X)$, and in terms of local

1078: coordinates $(x,y)$ near $\pa X$ can be written 

1079: $\tau dx/x^2 + \mu_i dy_i/x$,

1080: yielding local coordinates $(x,y,\tau,\mu)$ on $\sct X$ near $\pa X$. 

1081: 

1082: The scattering

1083: differential operators of order $k$, denoted $\Diffsc ^k(X)$, are those given

1084: by sums of products of at most $k$ scattering vector fields. There are two

1085: symbol maps defined for $P \in \Diffsc^k(X)$. The first is the `usual' symbol map,

1086: denoted here $\symbint^k(P)$, which maps to $S^k(\sct X)/S^{k-1}(\sct X)$,

1087: where $S^k(\sct X)$ denotes the classical symbols of order $k$ on

1088: $\sct X$.

1089: The second is the boundary symbol, $\symbb(P) \in S^k(\sct_{\pa X} X)$, 

1090: which is the {\it full} symbol of $P$ restricted to $x=0$. This is well

1091: defined since the Lie Algebra $\Vsc(X)$ has the property $[\Vsc(X), \Vsc(X)]

1092: \subset x \Vsc(X)$, so commutators of scattering vector fields vanish to an

1093: additional order at the boundary. Dividing the interior symbol

1094: $\symbint^k(P)$ by $|\xi|^k_g$, where $|\cdot|_g$ is the metric on $\sct X$

1095: determined by the scattering metric, we get a function on the sphere bundle

1096: of $\sct X$. This may be combined with the boundary symbol into a joint symbol,

1097: $\jsc^k(P)$, a function on $\Csc(X)$ which is the topological space obtained

1098: by gluing together the sphere bundle of $\sct X$ with the the fibrewise

1099: radial compactification of $\sct_{\pa X} X$ along their common boundary. 

1100: 

1101: \begin{comment}

1102: We will often be interested in differential operators acting on

1103: half-densities on $X$. Vector fields act on densities by Lie

1104: derivation. The action of a vector field $V$ on a nonvanishing half density

1105: $\alpha$ is determined by the condition

1106: $$

1107: V(\alpha^2) = 2 \alpha V(\alpha),

1108: $$

1109: and on an arbitrary half density is determined by 

1110: $$

1111: V(f\alpha) = f V(\alpha) + V(f) \alpha.

1112: $$

1113: If a differential operator $P$ is written as a sum of products of vector

1114: fields, 

1115: $$

1116: P = \sum_i \sum_{j=1}^{k(i)} V^{(i)}_j + f,

1117: $$

1118: then the action of $P$ on a half-density is defined by 

1119: $$

1120: P(\alpha) = \sum_i \sum_{j=1}^{k(i)} V^{(i)}_j(\alpha) + f \alpha;

1121: $$

1122: this is well defined since $\mathcal{L}_V \mathcal{L}_W -  \mathcal{L}_W  

1123: \mathcal{L}_V =  \mathcal{L}_{[V,W]}$. 

1124: \end{comment}

1125: 

1126: The scattering pseudodifferential operators are defined in terms of the

1127: behaviour of their Schwartz kernels on the scattering double space 

1128: $\XXsc$, a blown up version of the

1129: double space $X^2$. This is defined by

1130: \begin{equation}\begin{gathered}

1131: \XXb = [X^2; (\pa X)^2] \qquad \text{and} \\

1132: \XXsc = [\XXb; \pa \diag_\bl ],

1133: \end{gathered}\end{equation}

1134: and $\diag_\bl$ is the lift of the diagonal to $\XXb$. The lift of

1135: $\diag_\bl$ to $\XXsc$ is denoted $\diag_\scl$. The blowup notation $[; ]$

1136: is that of Melrose: see \cite{RBMCalcCon} or \cite{tapsit}. The boundary

1137:   hypersurfaces are 

1138: labelled lb, rb, bf and sf; see figure~\ref{fig:XXb}. The scattering

1139: pseudodifferential operators of order $k$, acting on half densities,

1140: $\Psisc^k(X)$, are those given by $\KDsc$-valued distribution kernels which

1141: are classical conormal at $\diag_\scl$, of order $k$, uniformly to the

1142: boundary, and rapidly vanishing at lb, rb, bf. (Here $\KDsc$ is the pullback

1143: of the bundle $\pi_l^* \scOh(X) \otimes \pi_r^* \scOh(X)$ over $X^2$ to $\XXb$.)

1144: The space $\Psisc ^{k,l}(X)$ is defined 

1145: to be $x^l \Psisc ^{k}(X)$. 

1146: 

1147: \begin{figure}\centering

1148: \epsfig{file=Scatt2.eps,width=12cm,height=4cm}

1149: \caption{The b-double space and the scattering scattering space}

1150: \label{fig:XXb}    

1151: \end{figure}

1152: 

1153: The joint symbol map $\jsc^k$ extends from $\Diffsc^k(X)$ to $\Psisc^k(X)$

1154: multiplicatively,

1155: $$

1156: \jsc^k(A) \cdot \jsc^m(B) = \jsc^{k+m}(AB),

1157: $$

1158: and such that there is an exact sequence

1159: $$

1160: \begin{CD}

1161: 0 @>>> \Psisc^{m-1,1}(X) @>>> \Psisc^{m}(X) @>{\jsc^{m}}>> 

1162: \CI(\Csc(X)) @>>> 0.

1163: \end{CD}

1164: $$

1165: A scattering pseudodifferential operator $A \in \Psisc^k(X)$ is said to be

1166: elliptic at a point $q \in \Csc(X)$ if $\jsc^k(A)(q) \neq 0$. It is said to 

1167: have 

1168: elliptic interior symbol (boundary symbol) if $\jsc^k(A)$ does not vanish at

1169: fibre-infinity (spatial infinity), and is said to be totally elliptic if

1170: $\jsc^k(A)$ vanishes nowhere. The characteristic variety of $A$,

1171: $\Sigma(A)$, is the zero set of $\jsc^k(A)$. 

1172: 

1173: The scattering wavefront set of a tempered distribution $u \in \dist(X)$

1174: (the dual space of $\CIdot(X)$, the space of smooth functions on $X$

1175: vanishing with all derivatives at the boundary) is the closed subset of

1176: $\Csc(X)$ whose complement is

1177: \begin{equation}

1178: \scwf(u)^\complement = \{ q \in \Csc(X) \mid \ \exists A \in \Psisc^0(X)

1179: \text{ elliptic at } q \text{ such that } Au \in \CIdot(X) \}.

1180: \end{equation}

1181: \begin{comment} 

1182: There is also a graded version, that is, a wavefront set $\scwf^{m,l}(u)$

1183: with respect to the weighted Sobolev space $H^{m,l}(X) = x^l H^m(X)$ which

1184: is the space of functions which are mapped to $L^2(X)$ (with respect to the

1185: measure induced by the scattering metric $g$) by $\Psisc^k(X)$ (or

1186: equivalently $\Diffsc^k(X)$ if $k$ is a natural number). This the subset of

1187: $\Csc(X)$ whose complement is

1188: \begin{equation}

1189: \scwf^{m,l}(u)^\complement = \{ q \in \Csc(X) \mid \ \exists A \in \Psisc^0(X)

1190: \text{ elliptic at } q \text{ such that } Au \in H^{m,l}(X) \}.

1191: \end{equation}\end{comment}

1192: The interior part of the wavefront set (at fibre-infinity) is a familiar

1193: object, the standard wavefront set introduced by H\"ormander (except that

1194: each ray of the standard wavefront set is thought of here as a point in the

1195: cosphere bundle). 

1196: In this paper we are mostly interested in the part of the scattering

1197: wavefront set at spatial infinity. In fact, the operators $H$ we shall study

1198: will have elliptic interior symbol, uniformly to the boundary, 

1199: so in view of the next theorem, solutions

1200: of $(H - \sigma) u = 0$ must have wavefront set contained in the

1201: part of $\Csc(X)$ at spatial infinity, which we denote $K$ 

1202: (that is, $K = \sct_{\pa X} X$).

1203: 

1204: There is a natural contact structure on $K$ induced by the symplectic form

1205: $\omega$ on $T^* X$. Writing $\omega$ in terms of the rescaled cotangent

1206: variables $\tau, \mu$ and contracting with the vector $x^2 \pa_x$ yields the

1207: 1-form

1208: $$

1209: \chi = \iota_{x^2 \pa_x} \omega = d\tau + \mu \cdot dy,

1210: $$

1211: which is nondegenerate, and therefore a contact form. A change of boundary

1212: defining function $x' = a x$ changes $\chi$ by a factor $a^{-1}$, so the

1213: contact {\it structure} is totally well-defined. Given a Hamiltonian $h$ on

1214: $K$, the Hamiltonian vector field on $K$ determined by the contact form is

1215: $$

1216: \frac{\pa h}{\pa \mu_i} \frac{\pa}{\pa y_i} + \Big(

1217: -\frac{\pa h}{\pa y_i} + \mu_i \frac{\pa h}{\pa \tau} \Big) \frac{\pa }{\pa

1218:   \mu_i} + \Big( h - \mu_i \frac{\pa h}{\pa \mu_i} \Big) \frac{\pa}{\pa

1219:   \tau}.

1220: $$

1221: This is the same as $x^{-1} V_h$ restricted to $x=0$, where $V_h$ is the

1222: Hamilton vector field on $\sct X$ induced by $h$. 

1223: Integral curves of this vector field are called bicharacteristics of $h$ (or

1224: of $A$, if $h$ is the boundary symbol of $A$). 

1225: 

1226: Under a coordinate change, the variables $\tau$ and $\mu$ change according

1227: to

1228: $$

1229: \tau' = a \tau, \quad \mu' = a \mu \frac{ \pa y}{\pa y'} - \tau \frac{\pa

1230:   a}{\pa y'} \qquad x' = a x.

1231: $$

1232: Since $a > 0$, this shows that the subset 

1233: \begin{equation}

1234: K_- = \{ (y, \tau, \mu) \in K \mid \tau \leq 0 \}

1235: \ilabel{Kminus}

1236: \end{equation} 

1237: is invariantly defined. This is important in the definition of the outgoing

1238: resolvent, see \eqref{outgoing}. 

1239: 

1240: The boundary part of the scattering wavefront set behaves very much as the

1241: interior wavefront set part behaves, and in particular we have a propagation

1242: of singularities result for operators of real principal type:

1243: 

1244: \begin{thm}[Melrose] Suppose $A \in \Psisc^{k}(X)$ has elliptic interior

1245: symbol, and 

1246: real boundary symbol. Then for

1247: $u \in \Cinf(X^o) \cap \dist(X)$,

1248: we have 

1249: \begin{equation}

1250: \WFsc(Au) \subset \WFsc(u), 

1251: \ilabel{prop-1}

1252: \end{equation}

1253: \begin{equation}

1254: \WFsc(u) \setminus \WFsc(Au) \subset \Sigma(A),  

1255: \ilabel{prop-2}\end{equation} 

1256: and 

1257: \begin{equation}\begin{gathered}

1258: \WFsc(u) \setminus \WFsc(Au) 

1259: \text{ is a union of } \\ \text{ maximally

1260: extended bicharacteristics of } A \text{ inside } 

1261: \Sigma(A) \setminus \WFsc(Au).

1262: \end{gathered}\ilabel{prop-3}\end{equation}

1263: \end{thm}

1264: 

1265: Thus, if $Au = 0$, then $\scwf(u) \subset K$ and consists of a union of

1266: maximally extended bicharacteristics of $A$ inside $\Sigma(A)$. 

1267: 

1268: As well as a boundary principal symbol defined on $K$, scattering

1269: pseudodifferential operators also have a boundary subprincipal symbol. This

1270: is the $O(x)$ term of the full symbol at the boundary when the operator is

1271: written in Weyl form. It is important to keep in mind that it depends on a 

1272: choice of product structure at the boundary; it does not enjoy quite the

1273: same invariance properties as does the standard (interior) subprincipal

1274: symbol.  A practical formula to use for differential 

1275: operators with symbol in left-reduced form, ie, such that 

1276: $$

1277: \sigma \left( a(x,y) (x^2 D_x)^j (x D_y)^\alpha \right) = a(x,y) \tau^j

1278: \mu^\alpha, \quad D = -i \pa, \quad \alpha =

1279: (\alpha_1, \dots, \alpha_{n-1}),

1280: $$

1281: is that for $\sigma(P) = p(y, \tau, \mu) + x q(y, \tau, \mu) + O(x^2)$,

1282: the boundary subprincipal symbol of $P$ is given by

1283: \begin{equation}

1284: \sigma_{\sub}(P) = q + \frac{i}{2} \left( \frac{\pa^2 p}{\pa y_i \pa \mu_i}

1285: - (n-1) \frac{ \pa p}{\pa \tau} + \mu_i \frac{ \pa^2 p}{\pa \mu_i \pa \tau}

1286: \right).

1287: \ilabel{subpr-symb}\end{equation}

1288: 

1289: \begin{lemma}\label{lem:sub-pr} 

1290: Let $g$ be a short range scattering metric, let $x$ be a

1291: boundary defining function with respect to which 

1292: $g = dx^2/x^4 + h'/x^2$, and let $H$ be a short range perturbation of the

1293: Laplacian with respect to $g$. Then in local coordinates $(x,y)$, the 

1294: sub-principal symbol of $H$ vanishes at $\mu = 0$. 

1295: \end{lemma}

1296: 

1297: \begin{proof} The operator $H$ may be written

1298: $$

1299: H = (x^2 D_x)^2 + (n-1) i x^3 D_x + x^2 \Lap_\pa + a_{ij}(x,y) x^3 D_{y_i}

1300: D_{y_j} + Q, \quad Q \in x^2 \Diffsc(X). 

1301: $$

1302: Thus the left-reduced symbol as above is

1303: $$

1304: \sigma(H) = \tau^2 + h_{ij}\mu_i \mu_j + x\left( (n-1) i \tau + a_{ij} \mu_i

1305: \mu_j \right) + O(x^2).

1306: $$

1307: Hence the sub-principal symbol is 

1308: \begin{equation}

1309: \sigma_{\sub}(H) = i \frac{\pa h_{ij}}{\pa y_i} \mu_j + a_{ij} \mu_i \mu_j,

1310: \ilabel{mu=0}\end{equation}

1311: which vanishes when $\mu = 0$. 

1312: \end{proof}

1313: 

1314: We now define the gravitational condition for the perturbation $P$. 

1315: 

1316: \begin{Def} A first order scattering differential operator $P$ on $X$ is

1317: said to be 

1318: short range if it lies in $x^2 \Diffsc^1(X)$, and long range if it lies in $x

1319: \Diffsc^1(X)$. Let $g$ be a scattering metric and $x$ a boundary defining

1320: function with respect to which $g$ takes the form \eqref{sc-metric-1} or

1321: \eqref{sc-metric-2}. $P$ is said to be 

1322: of long range gravitational type with respect to $g$ if it has the form

1323: $$

1324: P = x \left( \sum_i a_i (x \pa_{y_i}) + b x^2 \pa_x + c \right),

1325: $$

1326: near $x=0$, where $a_i$, $b$ and $c$ are in $\CI(X)$, and for some

1327: constants $b_0$ and $c_0$, $b = b_0 + O(x)$ and $c = c_0  + O(x)$. 

1328: \end{Def}

1329: 

1330: The point of the short range condition is that then the subprincipal symbol of

1331: both $H = \Lap +  P$ vanishes at the radial sets $\mu = 0, \tau = \pm \lam$

1332: of $H - \lam^2$, whilst in the long range gravitational case, the subprincipal

1333: symbol is constant. In the general long range case, the subprincipal symbol

1334: is an arbitrary function on the radial set, which causes some inconvenience

1335: (but not insuperable difficulties) in constructing the parametrix for

1336: $(H-\lam^2 - i0)^{-1}$. 

1337: 

1338: \subsection{Legendre distributions} An important special case that occurs

1339: often is that $\scwf(u)$ is a Legendre submanifold, or union of Legendre

1340: submanifolds, of $K$; moreover, in many cases, $u$ is a Legendre

1341: distribution, which means that it has a WKB-type expansion, the product of a

1342: oscillatory and smooth term, as discussed below, which makes it

1343: particularly amenable to analysis. 

1344: 

1345: We let $\dim X = n$, so that $\dim K = 2n-1$. 

1346: A Legendre submanifold of $K$ is a submanifold $G$ of dimension $n-1$ such that $\chi

1347: \restriction G = 0$. Such submanifolds have several nice properties. One is

1348: that if a Hamiltonian, $h$, is constant on $G$ then its Hamilton vector

1349: field is tangent to $G$. Another is that Legendre submanifolds may be

1350: generated in the following way: If $F$ is a submanifold of dimension $n-2$,

1351: such that $\chi$ vanishes on $F$, and if the Hamilton vector field of $h$ is

1352: nowhere tangent to $F$, then the union of bicharacteristics of $h$ passing

1353: through $F$ is

1354: (locally) a Legendre submanifold. 

1355: 

1356: Let $G$ be a Legendre submanifold, and let 

1357: $q \in G$. A local (nondegenerate) parametrization of $G$ near $q$ is a

1358: function $\phi(y,v)$ defined in a neighbourhood of $y_0 \in Y$ and $v_0 \in

1359: \RR^k$, such that $d_v \phi = 0$ at $q' = (y_0, v_0)$, $q = (y,

1360: d_{(x,y)}(\phi/x))$ at $q'$, % in local coordinates $(y, \tau, \mu)$ on $K$, 

1361: $\phi$ satisfies the nondegeneracy hypothesis

1362: $$

1363: d\Big( \frac{ \pa \phi}{\pa v_i} \Big) \text{ are linearly independent at }

1364: C = \{ (y,v) \mid d_v \phi = 0 \}, 1 \leq i \leq k,

1365: $$

1366: and near $q$,

1367: \begin{equation}

1368: G = \{ \big(y, d_{(x,y)} \left( \frac{\phi}{x} \right) \big) \mid (y,v) \in C \}.

1369: \ilabel{correspondence}\end{equation}

1370: A Legendre distribution of order $m$ associated to $G$ is a half-density of

1371: the form $u = (u_0 + \sum_{j=1}^N u_j) \nu$, where $\nu$ is a smooth section

1372: of the scattering half density bundle, $u_0 \in \CIdot(X)$, and

1373: $u_j$ is supported in a coordinate patch $(x,y)$ near the boundary, with an

1374: expression 

1375: $$

1376: u_j = x^{m+n/4-k/2} \int_{\RR^k} e^{i\phi_j(y,v)/x} a_j(x,y,v) \, dv,

1377: $$

1378: where $\phi_j$ locally parametrizes $G$ and $a_j \in \CI(X \times \RR^k)$,

1379: with compact support in $v$. Melrose and Zworski showed that $u_j$ can be

1380: written with respect to any local parametrization, up to an error in

1381: $\CIdot(X)$. The set of such half-densities is denoted $I^m(X, G;

1382: \scOh)$. The scattering wavefront set of $u \in I^m(X, G; \scOh)$ is

1383: contained in $G$.  

1384: 

1385: An intersecting Legendre distribution is associated to a pair of Legendre

1386: submanifolds, $\Lt = (L_0, L_1)$, where $L_1$ is a manifold with boundary such

1387: that $L_0$ and $L_1$ intersect cleanly at $\pa L_1$. A local parametrization

1388: of $(L_0, L_1)$ near $q \in L_0 \cap L_1$ is a function $\phi(y,v,s)$

1389: defined in a neighbourhood of $q' = (y_0,v_0,0)$ in $Y \times \RR^k \times [0,

1390: \infty)$ such that $d_v \phi = 0$ at $q'$, $q = (y, d_{(x,y)}(\phi/x))(q')$,

1391: %in local coordinates $(y, \tau, \mu)$ on $K$, 

1392: $\phi$ satisfies the nondegeneracy hypothesis

1393: $$

1394: ds, \ d\phi, \ \text{ and }

1395: d\Big( \frac{ \pa \phi}{\pa v_i} \Big) \text{ are linearly independent at }

1396: q', \ 1 \leq i \leq k,

1397: $$

1398: and near $q$,

1399: $$ \begin{gathered}

1400: L_0 = \{ \big(y, d_{(x,y)} \left( \frac{\phi}{x} \right) \big) \mid s=0, d_v

1401: \phi = 0 \}, \\ 

1402: L_1 = \{ \big(y, d_{(x,y)} \left( \frac{\phi}{x} \right) \big) \mid s\geq 0,

1403: d_s \phi = 0, \ d_v \phi = 0 \}. 

1404: \end{gathered}

1405: $$

1406: A Legendre distribution of order $m$ associated to $\Lt$ is a 

1407: half-density of

1408: the form $u = u_0 + (\sum_{j=1}^N u_j)\nu$, where $\nu$ is a smooth

1409: scattering half-density, $u_0 \in I^m_c(X, L_1; \scOh) +

1410: I^{m+1/2}(X, L_0; \scOh)$ (the subscript $c$ indicates that the microlocal

1411: support does not meet the boundary of $L_1$), and

1412: $u_j$ is supported in a coordinate patch $(x,y)$ near the boundary, with an

1413: expression 

1414: $$

1415: u_j = x^{m+n/4-(k+1)/2} \int_0^\infty ds \int e^{i\phi_j(y,v,s)/x} a_j(x,y,v,s) 

1416: \, dv  ,

1417: $$

1418: where $\phi_j$ locally parametrizes $(L_0, L_1)$ and 

1419: $a_j \in \CI(X \times \RR^k \times [0, \infty))$,

1420: with compact support in $v$ and $s$. Again, $u_j$ can be

1421: written with respect to any local parametrization, up to an error in

1422: $\CIdot(X)$. The set of such half-densities is denoted 

1423: $I^m(X, \Lt; \scOh)$. The scattering wavefront set of $u \in I^m(X, \Lt;

1424: \scOh)$ is contained in $L_0 

1425: \cup L_1$.

1426: 

1427: \

1428: 

1429: A Legendre distribution associated to a conic Legendrian pair is associated

1430: to a pair of Legendre submanifolds $\Gt = (G, \Gs)$ where $\Gs$ is a

1431: projectable Legendrian (that is, the projection from $\sct X$ to $Y$ is a

1432: diffeomorphism restricted to $\Gs$) 

1433: and $G$ is an open Legendrian submanifold such that

1434: $\overline{G} \setminus G$ is contained in $\Gs$ and $\overline{G}$ has at

1435: most a conic singularity at $\Gs$. We further assume that $\tau \neq 0$ on

1436: $\Gs$, so that we may change coordinates to a new boundary defining function

1437: such that $\Gs$ is parametrized by the phase function $1$. In these

1438: coordinates, the

1439: condition that $\overline{G}$ has a conic singularity at $\Gs$ means that 

1440: $\overline{G}$ lifts to a smooth

1441: submanifold with boundary, $\Gh$, on the blown-up space

1442: \begin{equation}

1443: [\sct X; \ \{ x = 0, \mu = 0 \}],

1444: \ilabel{Ghat}\end{equation}

1445: intersecting the front face of \eqref{Ghat} transversally. In local

1446: coordinates $(x,y,\tau,\mu)$, coordinates near the front face are 

1447: \begin{equation}

1448: x/|\mu|, \ y, \ \tau, \ |\mu| \text{ and } \hat \mu,

1449: \ilabel{hat-coords}\end{equation}

1450: and we require that $\Gh$ is given by the vanishing of

1451: $n$ smooth functions of these variables with linearly independent

1452: differentials, and that $d|\mu| \neq 0$ at $\pa 

1453: \Gh$. 

1454: 

1455: 

1456: A local parametrization

1457: of $\Gt$ near $q \in \overline{G} \cap \Gs$ is a function $\phi(y,v,s) =

1458: 1 + s \psi(y,v,s)$

1459: defined in a neighbourhood of $q' = (y_0,v_0,0)$ in $Y \times \RR^k \times [0,

1460: \infty)$ such that $\phi_0$ parametrizes $\Gs$ near $q$, 

1461: $d_v \phi = 0$ at $q'$, $q = (y, d_{(x,y)}(\phi/x))(q')$, $\phi$

1462: satisfies the nondegeneracy hypothesis

1463: $$

1464: ds, \ d\psi, \ \text{ and }

1465: d\Big( \frac{ \pa \psi}{\pa v_i} \Big) \text{ are linearly independent at }

1466: q', \ 1 \leq i \leq k,

1467: $$

1468: and near $q$,

1469: $$ 

1470: \Gh = \{ (0, y, -\phi, s d_y \psi, \widehat{d_y \psi}) 

1471: \mid d_v \phi = 0, \ d_s \psi = 0 \},

1472: $$

1473: in the coordinates \eqref{hat-coords}. 

1474: A Legendre distribution of order $(m,p)$ associated to $(G, \Gs)$ is a 

1475: half-density of

1476: the form $u = u_0 + (\sum_{i=1}^N u_i)\nu$, where $\nu$ is as above, 

1477: $u_0 \in I^m_c(X, G; \scOh) +

1478: I^{p}(X, \Gs; \scOh)$ (the subscript $c$ indicates that the microlocal

1479: support does not meet $\Gs$), and

1480: $u_j$ is supported in a coordinate patch $(x,y)$ near the boundary, with an

1481: expression 

1482: $$

1483: u_j =  \int_0^\infty ds \int e^{i\phi_j(y,v,s)/x} a_j(y,v,x/s,s) \Big

1484: ( \frac{x}{s} \Big)^{m+n/4-(k+1)/2} s^{p + n/4 - 1} 

1485: \, dv  ,

1486: $$

1487: where $\phi_j$ locally parametrizes $(G, \Gs)$ and 

1488: $a_j \in \CI(X \times \RR^k \times [0, \infty) \times [0, \infty))$,

1489: with compact support in $v$, $x/s$ and $s$. Here $u_j$ can be

1490: written with respect to any local parametrization, up to an error in

1491: $I^p(X, \Gs; \scOh)$. The set of such half-densities is denoted 

1492: $I^{m,p}(X, \Gt; \scOh)$. 

1493: The wavefront set of $u \in I^m(X, G; \scOh)$ is contained in $G

1494: \cup \Gs$.

1495: 

1496: \subsection{Codimension 2 corners} In this subsection we briefly review the

1497: extension of the theory of Legendre distributions to manifolds with

1498: codimension 2 corners and fibred boundaries given in \cite{HV2}. 

1499: 

1500: Let $M$ be a compact manifold with codimension 2 corners. The boundary

1501: hypersurfaces will be labelled $\mf, H_1, \dots, H_d$, where the $H_i$ are

1502: endowed with fibrations $\pi_i : H_i \to Z_i$ to certain closed manifolds

1503: $Z_i$ and mf (the `main face') is given the trivial fibration

1504: $\operatorname{id} : \mf \to \mf$. The collection of fibrations is denoted

1505: $\Phi$. It is assumed that $H_i \cap H_j = \emptyset$ if $i \neq j$.

1506: It is also assumed that the fibres of $\pi_i$ intersect $H_i \cap \mf$

1507: transversally  and therefore induce a fibration from $H_i \cap \mf \to

1508: Z_i$. Further, it is assumed that a total boundary defining function $x$ is

1509: given, which is distinguished up to multiplication by positive functions

1510: which are constant on the fibres of $\pa M$. 

1511: 

1512: Near $H \cap \mf$, where $H = H_i$ for some $i$, there are coordinates $x_1,

1513: x_2, y_1, y_2$ such that $x_1$ is a boundary defining function for $H$,

1514: $x_2$ is a boundary defining function for mf, $x_1x_2 = x$, and the

1515: fibration on $H$ takes the form

1516: $$

1517: (y_1, x_2, y_2) \mapsto y_1.

1518: $$

1519: 

1520: Associated with this structure is a Lie Algebra of vector fields

1521: $$

1522: \Vsf(M) = \{ V \mid V \in \CI, V \text{ is tangent to } \Phi \text{ at } \pa M,

1523: V(x) = O(x^2) \}.

1524: $$

1525: This is the space of smooth sections of a vector bundle, denoted $\Tsf M$. 

1526: The dual space is denoted $\sfT M$. A point in $\sfT _p M$ may be thought of

1527: as a differential $d(f/x)$ at $p$, where $f$ is a smooth function on $M$

1528: constant on the fibres at $\pa M$. A basis for $\sfT _p M$, for $p \in M$

1529: near $\mf \cap H$, is given by 

1530: $dx/x^2$, $dx_1/x$, $dy_1/x$, $dy_2/x^2$. Writing $q \in \sfT M$ as

1531: $$

1532: q = \tau \frac{dx}{x^2} + \tau_1 \frac{dx_1}{x} + \mu_1 \cdot

1533: \frac{dy_1}{x} + \mu_2 \cdot \frac{dy_2}{x_2} 

1534: $$

1535: gives coordinates 

1536: \begin{equation}

1537: (x_1, \ x_2, \ y_1, \ y_2, \ \tau, \ \tau_1, \ \mu_1, \ \mu_2)

1538: \ilabel{sfT-coords}\end{equation} 

1539: on $\sfT M$ near $\mf \cap H$. 

1540: 

1541: The differential operators of order at most $k$ generated over $\CI(M)$ by

1542: $\Vsf(M)$ are denoted $\Diffsf ^k(M)$. Near the interior of mf, the Lie

1543: Algebra $\Vsf(M)$ localizes to the scattering Lie Algebra $\Vsc(\Mt)$, where

1544: $\Mt$ denotes the noncompact manifold with boundary $M \setminus \cup_i

1545: H_i$. Consequently, we have a boundary symbol $\symbb(P)$, $P \in \Diffsf

1546: ^k(M)$ taking values in $S^k(\sct _{\pa \Mt} \Mt)$ over the interior of

1547: mf. In fact the symbol extends to an element of $S^k (\sfT _{\mf} M)$

1548: continuous up to the boundary of mf.

1549: 

1550: For each fibre $F$ of $H$, there is a subbundle of $\Tsf M$ consisting of

1551: all vector fields vanishing at $F$. The annihilator subbundle of $\sfT M$ is

1552: denoted $\sct (H; F)$ since it is isomorphic to the cotangent space of the

1553: fibre. The quotient bundle, $\sfT M / \sct (H; F)$ is denoted $\sfN Z_i$

1554: since it is the pullback of a bundle over $Z_i$. The fibration $\pi_i$

1555: induces a fibration

1556: \begin{equation}\begin{gathered}

1557: \pit_i : \sfT _{\mf \cap H} M \to \sfN Z_i \\

1558: (y_1, y_2, \tau, \tau_1, \mu_1, \mu_2) \mapsto (\tau, y_1, \mu_1).

1559: \end{gathered}\ilabel{pit}

1560: \end{equation}

1561: 

1562: We next describe three types of contact structures associated with the

1563: structure of 

1564: $M$. Since $\Vsf (M)$ is locally the scattering structure near the interior

1565: of mf, there is an induced contact structure on $\sfT M$ over the interior

1566: of mf. In local coordinates, the contact form looks like

1567: $$

1568: \chi = \iota_{x^2 \pa_x} (\omega) = d\tau + \tau_1 dx_1 + \mu_1 dy_1 + x_1

1569: \mu_2 dy_2.

1570: $$

1571: We see from this that at $x_1 = 0$, $\chi$ is degenerate. However,

1572: restricted to $\mf \cap 

1573: H$, $\chi$ is the lift of a form $\chi_{Z_i}$ on $\sfN (Z_i)$,

1574: namely $d\tau + \mu_1 dy_1$, which is nondegenerate on $\sfN (Z_i)$. This

1575: determines 

1576: our second type of contact structure (one for each $i$). 

1577: The third type of contact structure is that on $\sct

1578: _{\pa F} (H_i; F)$ induced by $\Vsf (M)$ for each fibre $F$ of $H_i$, since

1579: it restricts to the scattering vector fields on each fibre. In local

1580: coodinates, this looks like $d\tau_1 + \mu_2 dy_2$. 

1581: 

1582: Using these three contact structures we define Legendre submanifolds and

1583: Legendre distributions.

1584: 

1585: \begin{Def}\label{M-Leg-submfld}

1586: A  Legendre submanifold $G$ of $\sfT M$ is a Legendre

1587: submanifold of $\sfT_{\mf}M$ which is transversal to $\sfT_{H_i \cap

1588: \mf}M$ for each $H_i$,

1589: for which the map \eqref{pit} induces a fibration from $\pa G$ to $G_1$,

1590: where $G_1$ is a Legendre

1591: submanifold of $\sfN Z_i$, whose fibers are Legendre submanifolds

1592: of $\sct_{\partial F} F$.

1593: \end{Def}

1594: 

1595: A projectable Legendrian (one such that the projection from $\sfT _{\mf} M

1596: \to \mf$ is a diffeomorphism when restricted to $G$) is always of the form 

1597: $$

1598: \text{ graph } \big( d \big(\frac{\phi}{x} \big) \big) = \{ \big(\ybar,

1599: d\big(\frac{\phi(\ybar)}{x}\big)\big)  \mid \ybar \in \mf \}

1600: $$

1601: for some smooth function $\phi$ constant on the fibres of $\pa M$. We then

1602: say that $\phi$ parametrizes $G$. In

1603: general, let $G$ be a Legendre submanifold of $\sfT M$, and let $q \in

1604: G$. If $q$ lies above the interior of mf, then a local parametrization of

1605: $G$ near $q$ is as described in the previous subsection, so consider

1606: $q \in \pa G$ lying in $\sfT _{\ybar_0} M$, where $\ybar_0 \in \mf \cap H$. 

1607: A local (nondegenerate) parametrization of $G$ near $q$ is a

1608: function $\phi(x_1, y_1, y_2, v, w)$ of the form 

1609: \begin{equation}\label{eq:fib-ph-9}

1610: \phi(x_1, y_1, y_2,v, w)=\phi_1(y_1,v)+x_1\phi_2(x_1,y_1,y_2,v, w),

1611: \end{equation}

1612: defined in a neighbourhood of $q' = (\ybar_0,v_0,w_0) \in \mf \times

1613: \RR^{k_1 + k_2}$, such that $d_{v,w} \phi = 0$ at $q'$, 

1614: $$

1615: q = \big(\ybar_0, d\big(\frac{\phi(\ybar_0)}{x}\big)\big) \text{ at } q'

1616: $$ 

1617: in local coordinates \eqref{sfT-coords}, $\phi$

1618: satisfies the nondegeneracy hypothesis at $q'$

1619: \begin{equation}\label{eq:corner-20}

1620: d_{(y_1,v)}\frac{\pa \phi_1}{\partial {v_j}}, j=1,\ldots, k

1621: \text{ and } d_{(y_2,w)} \frac{\pa \phi_2}

1622: {\partial {w_{j'}}},\ j'=1,\ldots,k' \text{ linearly independent,}

1623: \end{equation}

1624: and near $q$,

1625: \begin{equation}

1626: G = \{ \big(\ybar, d\big(\frac{\phi(\ybar)}{x}\big)\big) \mid 

1627: d_v \phi = d_w \psi = 0 \}.

1628: \end{equation}

1629: A Legendre distribution of order $(m; r_1, \dots r_d)$ associated to $G$ is

1630: a half-density 

1631: such that for any $\upsilon_i \in \CI(M)$ whose support does not intersect

1632: $H_k$, for $k \neq i$, $\upsilon_i u$ is of

1633: the form $u = u_0 + (\sum_{j=1}^N u_j + \sum_{j=1}^M u_j')\nu$, where 

1634: $\nu$ is a smooth section of the half-density bundle $\sfOh$ induced by

1635: $\sfT M$, 

1636: $u_0 \in \CIdot(X; \sfOh)$, and $u_j$, $u'_j$ have expressions 

1637: \begin{equation}\begin{gathered}\label{eq:dist-1}

1638: u_j(x_1, x_2,y_1, y_2)

1639: = \int e^{i\phi_j(x_1, y_1, y_2,v,w)/x} a_j(x_1, x_2, y_1, y_2,v,w)\\

1640: x_2^{m-(k+k')/2+N/4}x_1^{r_i-k/2+N/4-f_i/2}\,dv\,dw 

1641: \end{gathered}\end{equation}

1642: with $N=\dim M$, $a_j\in\Cinf_c([0,\epsilon)\times

1643: U\times\RR^{k+k'})$, $U$ open in $\mf$, $f_i$ the dimension of the fibres

1644: of $H_i$ and 

1645: $\phi_j$ a phase function parametrizing a Legendrian $G$ on

1646: $U$, and 

1647: \begin{equation}

1648: u'_j(x_1, y_1,z)

1649: = \int e^{i\psi_j(y_1,w)/x} a_j(x,y_1,z,w)

1650: x^{r_i-k/2+N/4-f_i/2}\,dw 

1651: \ilabel{eq:dist-2}

1652: \end{equation}

1653: with $N=\dim M$, $a_j\in\Cinf_c([0,\epsilon)\times

1654: U\times\RR^{k})$, $U$ open in $H$, $f_i$ as above,

1655: $\psi_j$ a phase function parametrizing the Legendrian $G_1$. 

1656: 

1657: \begin{Def}\label{M-leg-submfld-conicpts}

1658: A  Legendre pair with conic points, $(G,\Gs)$,

1659: in $\sfT M$ consists of two  Legendre submanifolds $G$ and $\Gs$

1660: of $\sfT M$ which form an intersecting pair with conic points in $\sct_{\td M} \td M$ 

1661: such that for each $H_i$ the

1662: fibrations of $G$ and $\Gs$ induced by \eqref{pit} have the

1663: same Legendre submanifold $G_1$ of $\sfN Z_i$ as base and for

1664: which the fibres are intersecting pairs of 

1665: Legendre submanifolds with conic points

1666: of $\sct_{\partial F} F$. 

1667: \end{Def}

1668: 

1669: The Legendrian $\Gs$ is required to be projectable, so it parametrized by a

1670: phase function $\phi(\ybar)$ which is constant on the fibres of $\pa

1671: M$. Thus, $x' = x/\phi$ is another admissible total boundary defining

1672: function. With respect to $x'$, $\Gs$ is

1673: parametrized by the function $1$. Thus, without loss of generality we may

1674: assume that coordinates have been chosen so that $\Gs$ is parametrized by

1675: $1$. This simplifies the coordinate form of the blowup

1676: \eqref{Ghat}. Coordinates near $\pa \Gh$ then are 

1677: \begin{equation}

1678: \ybar = (x_1, y_1, y_2), \ \tau,\ \tau_1/|\mu_2|, \ \mu_1/|\mu_2|, \ 

1679: \muh_2 = \mu_2/|\mu_2|, \text{ and } |\mu_2|, 

1680: \ilabel{Ghat-coords}\end{equation}

1681: the last of which is a boundary defining function for $\Gh$ (see

1682: \cite{HV2}). 

1683: 

1684: As a consequence of definition~\ref{M-leg-submfld-conicpts}, $\Gh$ is a

1685: compact manifold with corners in 

1686: \begin{equation}

1687: [\sfT_{\mf} X; \ \{ x = 0, \mu_1 = \mu_2 = \tau_1 = 0 \}],

1688: \ilabel{Ghat-2}\end{equation}

1689: with one boundary hypersurface

1690: at the intersection of $\Gh$ and $\sfT _{\mf \cap H_i} M$ for each $i$ 

1691: (which are

1692: mutually nonintersecting, since the $H_i$ are mutually nonintersecting), and

1693: one at the intersection of $\Gh$ and the front face of the blowup in

1694: \eqref{Ghat-2}. If $q$ lies in the interior of $\Gh$  then the situation is as

1695: for Legendrians in the scattering setup. If $q$ is on the boundary of $\Gh$,

1696: but does not lie over $H_i$ for some $i$, then the situation is as for

1697: Legendrian conic pairs as in the previous subsection. If $q$ is on the

1698: boundary of $\Gh$ but not in $\Gs$ then the situation is as above. Thus the

1699: only situation left to describe is if $q$ is in the corner of $\Gh$, lying

1700: above $\ybar_0 \in \mf \cap H_i$ say. 

1701: 

1702: A local parametrization of $(G, \Gs)$ near $q$ (in coordinates as chosen

1703: above) is a function 

1704: $$\phi(x_1, y_1, y_2, s, w) = 1 + sx_1 \psi(x_1, y_1, y_2, s, w),

1705: $$

1706: with $\psi$ defined in a neighbourhood of $q' = (\ybar_0, 0, w_0) \subset M

1707: \times [0, \infty) \times \RR^k$, such that 

1708: $d_w \psi = 0$ at $q'$, 

1709: $$

1710: d_{y_2} \psi \ \text{ and } d_{(y_2,w)} \left( \frac{\pa \psi}{\pa

1711:   w_i} \right) \text{ are linearly independent at } q',

1712: $$

1713: and such that near $q \in \Gh$, 

1714: $$

1715: \Gh = \{ (\ybar, -\phi, \frac{d_{x_1}(x_1 \psi)}{|d_{y_2} \psi|},

1716: \frac{d_{y_1}(x_1 \psi)}{|d_{y_2} \psi|}, \widehat{ d_{y_2} \psi },

1717: {|d_{y_2} \psi|} \},

1718: $$

1719: in the coordinates \eqref{Ghat-coords}. 

1720: 

1721: A Legendre distribution of order $(m,p; r_1, \dots r_d)$ associated to $(G,

1722: \Gs)$ is a half-density 

1723: such that for any $\upsilon_i \in \CI(M)$ whose support does not intersect

1724: $H_k$, for $k \neq i$, $\upsilon_i u$ is of

1725: the form $u = u_0 + (\sum_{j=1}^N u_j + \sum_{j=1}^M u_j')\nu$, where 

1726: $u_0 \in \CIdot(X; \sfOh)$, and $u_j$, $u'_j$ have expressions 

1727: \begin{equation}\begin{gathered}\label{eq:dist-1p}

1728: u_j(x_1, x_2,y_1, y_2)

1729: = \int_0^\infty ds \int dw \, e^{i\phi_j(x_1, y_1, y_2,v,w)/x} 

1730: a_j(x_1, s, x_2/s, y_1, y_2,w)\\

1731: \left( \frac{x_2}{s}\right)^{m-(k'+1)/2+N/4} s^{p-1+N/4}

1732: x_1^{r_i-k/2+N/4-f_i/2} 

1733: \end{gathered}\end{equation}

1734: with $N=\dim M$, $a_j\in\Cinf_c([0,\epsilon)\times

1735: U\times\RR^{k+k'})$, $U$ open in $\mf$, $f_i$ the 

1736: dimension of the fibres of $H_i$ and

1737: $\phi_j$ a phase function parametrizing a Legendrian $G$ on

1738: $U$, and where $u'_j$ is as in \eqref{eq:dist-2}. 

1739: 

1740: 

1741: \subsection{The b-double space} Here we analyze the b-double space $\XXb$,

1742: where $X$ is a compact manifold with boundary, from the perspective of

1743: manifolds with corners with fibred boundaries. The manifold with corners

1744: $\XXb$ has three boundary hypersurfaces: lb and rb, which are the lifts of

1745: the left and right boundaries $\pa X \times X$, $X \times \pa X$ of $X^2$ to

1746: $\XXb$, and bf, coming from the blowup of $(\pa X)^2$ (see figure

1747: \ref{fig:XXb}). Thus, lb 

1748: and rb have natural projections to $\pa X$. The fibres of lb and rb meet bf

1749: transversally, so we may identify bf 

1750: as the `main face' mf of $\XXb$. Given coordinates $(x,y)$ or

1751: $z$ on $X$, we denote the lift to $\XXb$ via the left, resp. right

1752: projection by $(x',y')$ or $z'$, resp. $(x'',y'')$ or $z''$. 

1753: We may take the distinguished total boundary defining function to

1754: be $x'$, for $\sigma = x'/x'' < C$ and $x''$, for $\sigma >

1755: C^{-1}$. These are compatible since their ratio is constant on fibres on the

1756: overlap region $C^{-1} < \sigma < C$ (this is trivially true since the

1757: fibres of bf are points). 

1758: 

1759: These data

1760: give $\XXb$ the structure of a manifold with corners with fibred boundary as

1761: defined above. The $\sf$-vector fields then are the same as the sum of the scattering

1762: Lie Algebra $\Vb(X)$ lifted to $\XXb$ from the left and right factors. 

1763: 

1764: On $\XXb$, and $\sfT \XXb$, it is most convenient to use coordinates lifted

1765: from $X$ and $\sct X$. Near lb, 

1766: but away from bf, we use coordinates $(x',y',z'')$ and coordinates $(\tau',

1767: \mu', \zeta'')$ on $\sfT \XXb$ where we write a covector $q \in \sfT \XXb$

1768: as

1769: $$

1770: q = \tau' \frac{dx'}{{x'}^2} + \mu' \cdot \frac{dy'}{x'} + \zeta'' \cdot

1771: dz''.

1772: $$

1773: Similarly near rb, but away from bf, we use coordinates $(z',x'',y'';

1774: \zeta',\tau'',\mu'')$. Near $\lb \cap \operatorname{bf}$, we use $(x'',

1775: \sigma, y', y'')$ 

1776: with corresponding coordinates $(\tau, \kappa, \mu', \mu'')$, by

1777: writing $q \in \sfT \XXb$ as

1778: $$

1779: q = \tau \frac{dx'}{{x'}^2} + \kappa \frac{ d\sigma}{x'} + \mu'

1780: \cdot \frac{dy'}{x'} + \mu'' \cdot \frac{dy''}{x''}.

1781: $$

1782: However, we may also use scattering cotangent coordinates $(\tau', \tau'', \mu',

1783: \mu'')$ lifted from $\sct X$, where we write

1784: $$

1785: q = \tau' \frac{dx'}{{x'}^2} + \tau'' \frac{dx''}{{x''}^2} + \mu'

1786: \cdot \frac{dy'}{x'} + \mu'' \cdot \frac{dy''}{x''}.

1787: $$

1788: This gives

1789: \begin{equation}

1790: \tau' = \tau + \sigma \kappa \quad \tau''

1791:   = -\kappa.

1792: \ilabel{tau}\end{equation}

1793: The coordinates $(x'', \sigma, y', y'')$ hold good near bf as long as we

1794: stay away from rb, when we need to switch to $(x', \sigma^{-1}, y', y'')$. 

1795: The cotangent coordinates $(\tau', \tau'', \mu', \mu'')$ are good coordinates

1796: globally near bf; notice that the roles of $(\tau, \tau_1, \mu_1, \mu_2)$

1797: are played 

1798: by $(\tau', \tau'', \mu', \mu'')$ near lb and $(\tau'', \tau', \mu'', \mu')$

1799: near rb. 

1800: 

1801: The operator $H$ can act on half-densities on $\XXb$ by acting either on the

1802: left or the right factor of $X$; these operators are denoted $H_l$ and $H_r$

1803: respectively. For $H = \Lap + P$, where $P \in x \Diffsc^1(X)$, the Hamilton

1804: vector field induced by $H_l$ and

1805: the contact structure on $\sfT_{\bff} \XXbt$, with

1806: respect to $x'$, takes the form

1807: \begin{equation}

1808: V_l = 2 \tau' \sigma \frac{\pa}{\pa \sigma} + 

1809: 2\tau' \mu' \frac{\pa}{\pa \mu'} - h' \frac{\pa}{\pa \tau'} + \left

1810: ( \frac{\pa h'}{\pa \mu'} \frac{\pa}{\pa y'} - \frac{\pa

1811: h'}{\pa y'} \frac{\pa}{\pa \mu'}  \right) \quad h' = h(y', \mu').

1812: \ilabel{Vl}\end{equation}

1813: Similarly, the Hamilton vector field induced by $H_r$ and

1814: the contact structure on $\sfT_{\bff} \XXbt$, with

1815: respect to $x''$, takes the form

1816: \begin{equation}

1817: V_r = -2 \tau'' \sigma \frac{\pa}{\pa \sigma} + 

1818: 2\tau'' \mu'' \frac{\pa}{\pa \mu''} - h'' \frac{\pa}{\pa

1819:   \tau''} + \left 

1820: ( \frac{\pa h''}{\pa \mu''} \frac{\pa}{\pa y''} - \frac{\pa h''}{\pa y''} 

1821: \frac{\pa}{\pa \mu''}  \right) \quad h'' = h(y'', \mu'').

1822: \ilabel{Vr}\end{equation}

1823: Notice that $V_l$ and $V_r$ commute. 

1824: 

1825: 

1826: 

1827: 

1828: 

1829: \section{Symbol calculus for Legendre distributions}

1830: \subsection{Manifolds with boundary}

1831: Let $X$ be a manifold with boundary of dimension $N$, and let

1832: $$

1833: u = x^q (2\pi)^{-k/2-n/4}

1834: \left( \int e^{i\phi(y,v)/x} a(x,y,v) \, dv \right) \schd \in I^m(X, G; \scOh)

1835: $$

1836: be a Legendre distribution of order $m$. Let $C = \{ (y,v) \mid d_v \phi = 0

1837: \}$ and let $\lambda$ be a set of functions in $(y,v)$-space such that

1838: $(\lambda, d_v \phi)$ form local coordinates near $C$. We temporarily define 

1839: the symbol relative to the

1840: coordinate system $\mathcal{Z} = (x,y)$ and the parametrization $\phi$ to be

1841: the half density on $G$ given by

1842: \begin{equation}

1843: \symrel{m}{Z}{\phi}(u) = \big( a(0,y,v) \restriction C \big) \left| \frac

1844: { \pa (d_v \phi, \lam)}{\pa (y,v)} \right| ^{-\half} |d\lam|^\half.

1845: \ilabel{symrel}\end{equation}

1846: Here we have used the correspondence \eqref{correspondence} between 

1847: $C$ and $G$. 

1848: 

1849: 

1850: If we change coordinate system, the symbol changes by

1851: \begin{equation}

1852: \sigma^{m}_{\mathcal{Z}', \phi}(u) = \symrel{m}{Z}{\phi}(u)

1853: a^{n/4-m}

1854: e^{-i\rho(\mathcal{Z}', \mathcal{Z})}, \quad a = \frac{x'}{x}

1855: \ilabel{change}\end{equation}

1856: where

1857: \begin{equation}

1858: \rho(\mathcal{Z}', \mathcal{Z}) = \Big\{ a \mu_i \frac{\pa y_i}{\pa x'} 

1859:  - \tau \frac{\pa a}{\pa x'} \Big\}

1860: \restriction x = 0.

1861: \ilabel{E-factor}\end{equation}

1862: If the parametrization is changed, then by \cite{FIO1}, the symbol changes by

1863: \begin{equation}

1864: \symrel{m}{Z}{\psi}(u) = \symrel{m}{Z}{\phi}(u) e^{i\pi(\sign d^2_{vv}\psi -

1865:   \sign d^2_{v'v'}\phi)/4};

1866: \ilabel{M-factor}\end{equation}

1867: the exponential is a locally constant function. We use these

1868: transformation factors to define two line bundles, the $E$-bundle over $\sct

1869: X$ which is

1870: defined by the transition functions \eqref{E-factor}, and the Maslov bundle

1871: over $G$ which is defined by the transition functions

1872: \eqref{M-factor}. (These bundles will be described in much more detail in

1873: \cite{HV4}.) Defining the bundle $\Sm (G) = |N^* \pa X|^{m-n/4} \otimes E

1874: \otimes M(G)$ over $G$, we obtain an invariant symbol map from \eqref{symrel}

1875: $$

1876: \sigma^m : I^m(X, G; \scOh) \to \CI(G; \Omega^\half \otimes \Sm (G)).

1877: $$

1878: 

1879: 

1880: The elements of the symbol calculus for Legendre distributions on

1881: manifolds with boundary have been given by Melrose and Zworski \cite{RBMZw}:

1882: 

1883: 

1884: \begin{prop} The symbol map induces an exact sequence

1885: $$

1886: 0 \to I^{m+1}(X, G; \scOh) \to I^m(X, G; \scOh) \to

1887: \CI(G, \Omega^\half \otimes S^{[m]}(G)) \to 0.

1888: $$

1889: If $P \in \Psi^k(X; \scOh)$ and $u \in I^m(X, G; \scOh)$, 

1890: then $Pu \in I^m(X, G; \scOh)$ and 

1891: $$

1892: \sigma^m(Pu) = \Big( \sigma(P) \rest G \Big) \sigma^m(u).

1893: $$

1894: Thus, if the symbol of $P$ vanishes on $G$, then $Pu \in I^{m+1}(X, G;

1895: \scOh)$. The symbol of order $m+1$ of $Pu$ in this case is

1896: \begin{equation}

1897: \Big( -i \mathcal{L}_{H_p} -i \big(\half + m - \frac{N}{4} \big) \frac{ \pa

1898:   p}{\pa \tau} + p_{\sub} \Big) \sigma^m(u) \otimes |dx|,

1899: \ilabel{transport}\end{equation}

1900: where $H_p$ is the Hamilton vector field of $p$, the principal symbol of

1901: $P$, and $p_{\sub}$ is the subprincipal symbol of $P$. 

1902: \end{prop}

1903: 

1904: The symbol calculus for intersecting Legendre distributions is easily

1905: deduced from Melrose and Uhlmann's calculus of intersecting Lagrangian

1906: distributions. The symbol takes values in a bundle over $L_0 \cup L_1$. Let

1907: $\rho_1$ be a boundary defining function for $\pa L_1$ as a submanifold of

1908: $L_0$, and $\rho_0$ be a boundary defining function for $\pa L_1$ as a

1909: submanifold of $L_1$. To

1910: define the symbol, note that the symbol on $L_0$ is defined by continuity from

1911: distributions microsupported away from $L_1$, and takes values in

1912: \begin{equation}

1913: \rho_1^{-1} \CI(\Omega^{1/2}(L_0)\otimes S^{[m+1/2]}(L_0)) = 

1914: \rho_1^{-1/2} \CI(\Omega_b^{1/2}(L_0 \setminus \pa L_1)

1915: \otimes S^{[m+1/2]}(L_0)),

1916: \ilabel{int-Leg-dist}\end{equation}

1917: while the symbol on

1918: $L_1$ defined by continuity from

1919: distributions microsupported away from $\pa L_1$ takes values in

1920: $$

1921: \CI(\Omega^{1/2}(L_1) \otimes S^{[m]}(L_1)) =

1922: \rho_0^{1/2}\CI(\Omega_b^{1/2}(L_1) \otimes S^{[m]}(L_1)).

1923: $$

1924: Melrose and Uhlmann showed

1925: that the Maslov factors were canonically isomorphic on $L_0 \cap L_1$, so

1926: $S^{[m]}(L_0)$ is naturally isomorphic to $S^{[m]}(L_1)$ over $L_0 \cap

1927: L_1$. Canonical restriction of the

1928: half-density factors to $L_0 \cap L_1$ gives terms in $\CI(\Omega^\half(L_0

1929: \cap L_1) \otimes S^{[m]}(L_1) \otimes |N^*_{L_0}\pa L_1|^{-1/2} \otimes

1930: |N^* \pa X|^{1/2}$ and $\CI(\Omega^\half(L_0

1931: \cap L_1) \otimes S^{[m]}(L_1) \otimes |N^*_{L_1}\pa L_1|^{1/2}$

1932: respectively. In 

1933: fact $|N^*_{L_0}\pa L_1| \otimes |N^*_{L_1}\pa L_1| \otimes |N^* \pa

1934: X|^{-1}$ is canonically trivial; an

1935: explicit trivialization

1936: is given by 

1937: \begin{equation}

1938: (d\rho_0, d\rho_1, x^{-1}) \mapsto x^{-1} \omega(V_{\rho_0},

1939: V_{\rho_1}) \restriction L_0 \cap L_1,

1940: \ilabel{triv}\end{equation}

1941: where $V_{\rho_i}$ are the Hamilton vector fields of the

1942: functions $\rho_i$ extended into $\sct X$, and $\omega$ is the standard

1943: symplectic form. Thus the two bundles are naturally isomorphic over the

1944: intersection. We define the bundle $S^{[m]}(\Lt)$ to be that bundle such

1945: that smooth sections of $\Omega^{1/2}_b(\Lt) \otimes S^{[m]}(\Lt)$ are precisely

1946: those 

1947: pairs $(a,b)$ of sections of $\rho_1^{-1} \CI(\Omega^{1/2}(L_0)\otimes

1948: S^{[m+1/2]}(L_0))$ and $\rho_0^{1/2}\CI(\Omega^{1/2}_b(L_1) \otimes 

1949: S^{[m]}(L_1))$ such that 

1950: \begin{equation}

1951: \rho_1^{1/2} b = e^{i\pi/4} (2\pi)^{1/4} \rho_0^{-1/2} a \text{ at }L_0 \cap L_1

1952: \ilabel{compat}\end{equation}

1953: under the

1954: identification \eqref{triv}. The symbol maps of order $m$ on $L_1$ and

1955: $m+1/2$ on $L_0$ then extend in a natural way to a symbol map of order $m$

1956: on $\Lt$ taking values in $\Omega_b^{1/2}(\Lt) \otimes S^{[m]}(\Lt)$. 

1957: 

1958: \begin{prop} The symbol map on $\Lt$ yields an exact sequence

1959: \begin{equation}

1960: %\begin{CD}

1961: 0 \to I^{m+1}(X, \Lt; \scOh) \to I^m(X, \Lt; \scOh) \to

1962: \CI(\Lt, \Omega^\half_b \otimes S^{[m]}) \to 0. 

1963: %\end{CD}

1964: \ilabel{ex-int-1}\end{equation}

1965: Moreover, if we consider just the symbol map to $L_1$, there is an exact

1966: sequence

1967: \begin{multline}

1968: 0 \to I^{m+1}(X, \Lt; \scOh) + I^{m+\half}(X, L_0; \scOh) \to I^m(X, \Lt;

1969: \scOh)  \\  

1970: \to \CI(L_1, \Omega^\half \otimes S^{[m]}) \to 0.

1971: \ilabel{ex-int-2}\end{multline}

1972: If $P \in \Psi^k(X; \scOh)$ and $u \in I^m(X, \Lt; \scOh)$, 

1973: then $Pu \in I^m(X, \Lt; \scOh)$ and 

1974: $$

1975: \sigma^m(Pu) = \Big( \sigma(P) \rest \Lt \Big) \sigma^m(u).

1976: $$

1977: Thus, if the symbol of $P$ vanishes on $L_1$, then $Pu$ is an element of

1978: $I^{m+1}(X, \Lt;

1979: \scOh) + I^{m}(X, L_0; \scOh)$. The symbol of order $m+1$ of $Pu$ on $L_1$

1980: in this case is given by \eqref{transport}. 

1981: \end{prop}

1982: 

1983: For a conic pair of Legendre submanifolds $\Gt = (G, \Gs)$, with $\Gh$ the

1984: desingularized submanifold obtained by blowing up $\Gs$, the symbol is

1985: defined by continuity from the regular part of $G$. The symbol calculus then

1986: takes the form

1987: 

1988: \begin{prop} Let $s$ be a boundary defining function for $\Gh$. Then 

1989: there is an exact sequence

1990: $$

1991: 0 \to I^{m+1,p}(X, \Gt; \scOh) \to I^m(X, \Gt; \scOh) \to

1992: s^{m-p}\CI(\Gh, \Omega^\half_b \otimes S^{[m]}(\Gh)) \to 0.

1993: $$

1994: If $P \in \Psi^k(X; \scOh)$, and $u \in I^{m,p}(X, \Gt; \scOh)$, 

1995: then $Pu \in I^{m,p}(X, \Gt; \scOh)$ and 

1996: $$

1997: \sigma^m(Pu) = \Big( \sigma(P) \rest \Gh \Big) \sigma^m(u).

1998: $$

1999: If the symbol of $P$ vanishes on $G$, then $Pu \in I^{m+1,p}(X, \Gt;

2000: \scOh)$. The symbol of order $m+1$ of $Pu$ in this case is given by

2001: \eqref{transport}. 

2002: \end{prop}

2003: 

2004: 

2005: 

2006: \subsection{Codimension two corners} When we have codimension two corners,

2007: then essentially the same results hold by continuity from the main face. 

2008: The symbol is defined as a half-density on $G$ by continuity from the

2009: interior of mf, where the scattering situation applies. 

2010: We must restrict to

2011: differential operators, however, since pseudodifferential operators have not

2012: been defined in this context. 

2013: 

2014: Let $M$ be a manifold with codimension 2 corners with fibred boundaries,

2015: let $N = \dim M$, and

2016: let $G$ be a Legendre distribution. Let $\rho_i$ be a boundary defining 

2017: function for $H_i$. The Maslov bundle $M$ and the E-bundle are defined via

2018: the scattering structure over the interior of $G$ and extend to smooth

2019: bundles over the whole of $G$ (that is, they are smooth up to each boundary

2020: of $G$ at $\sfT _{H_i \cap \mf} M$). Let $S^{[m]}(G) = M(G) \otimes E

2021: \otimes |N^* \mf|^{m-N/4} \otimes |N^* H_1|^{m-N/4} \otimes \dots \otimes

2022: |N^* H_d|^{m-N/4}$. Finally let ${\bf r}$ stand for $(r_1, \dots, r_d)$,

2023: and let $\rho^{\bf r} = \prod_i \rho_i^{r_i}$. 

2024: 

2025: \begin{prop} There is an exact sequence

2026: $$

2027: 0 \to I^{m+1,{\bf r}}(M, G; \sfOh) \to I^{m,{\bf r}}(M, G; \sfOh) \to 

2028: \rho^{m-{\bf r}} \CI(G, \Omega^\half_b \otimes S^{[m]}(G)) \to 0.

2029: $$

2030: If $P \in \Diff(M; \sfOh)$ and $u \in I^{m,{\bf r}}(M, G; \sfOh)$, 

2031: then $Pu \in I^{m,{\bf r}}(M, G; \sfOh)$ and 

2032: $$

2033: \sigma^m(Pu) = \Big( \sigma(P) \rest G \Big) \sigma^m(u).

2034: $$

2035: Thus, if the symbol of $P$ vanishes on $G$, then $Pu \in I^{m+1,{\bf r}}(M, G;

2036: \sfOh)$. The symbol of order $m+1$ of $Pu$ in this case is given by

2037: \eqref{transport}. 

2038: \end{prop}

2039: 

2040: For a conic pair of Legendre submanifolds $\Gt = (G, \Gs)$, with $\Gh$ the

2041: desingularized submanifold obtained by blowing up $\Gs$, the symbol calculus

2042: takes the form

2043: 

2044: \begin{prop}\label{ex-conic-2} Let $s$ be a boundary defining function for

2045: $\Gh$ at $\Gh \cap \Gs$. Then  

2046: there is an exact sequence

2047: \begin{multline}

2048: 0 \to I^{m+1,p; {\bf r}}(M, \Gt; \sfOh) \to I^{m,p; {\bf r}}(M, \Gt; \sfOh) \\

2049: \to \rho^{m-{\bf r}} s^{m-p} \CI(\Gh, \Omega^\half_b \otimes S^{[m]}(\Gh))

2050: \to 0. 

2051: \end{multline}

2052: If $P \in \Diff(M; \sfOh)$, and $u \in I^{m,p;{\bf r}}(M, \Gt; \sfOh)$, 

2053: then $Pu \in I^{m,p;{\bf r}}(M, \Gt; \sfOh)$ and 

2054: $$

2055: \sigma^m(Pu) = \Big( \sigma(P) \rest \Gh \Big) \sigma^m(u).

2056: $$

2057: If the symbol of $P$ vanishes on $G$, then $Pu \in I^{m+1,p;{\bf r}}(M, \Gt;

2058: \sfOh)$. The symbol of order $m+1$ of $Pu$ in this case is given by

2059: \eqref{transport}. 

2060: \end{prop}

2061: 

2062: The proofs of these propositions are omitted, since they are easily deduced

2063: from the codimension one case.  

2064: 

2065: 

2066: 

2067: 

2068: 

2069: \section{Parametrix construction}

2070: In this section, we consider self-adjoint operators $H$ of the form $\Lap +

2071: P$, where $\Lap$ is the positive Laplacian with respect to a short-range

2072: metric on a compact manifold with boundary, $X$, and $P \in x^2

2073: \Diffsc^1(X)$ is a short-range perturbation of $\Lap$. In the following

2074: section, we consider metrics and perturbations of long range gravitational

2075: type. Let $R(\sigma)$ denote the resolvent $(H - \sigma)^{-1}$ of $H$. 

2076: 

2077: In this section, we 

2078: directly construct a parametrix $G(\lam)$ for \eqref{eq-Rt} whose error term

2079: $E(\lam) = (H - \lam^2) 

2080: G(\lam) - \Id$ is compact. Using Fredholm theory and a unique continuation

2081: theorem we solve away the error, giving us a Schwartz kernel $\Rt(\lam)$.

2082: We then show that $\Rt(\sqrt{\sigma})$ has an analytic continuation (as a

2083: distribution on $X^2$) 

2084: to the upper half $\sigma$ plane which

2085: agrees with the resolvent $R(\sigma)$ there. This proves that $\Rt(\lam)$ and

2086: $R(\lam^2 + i0)$ coincide on the real axis. 

2087: 

2088: The distribution $\Rt(\lam)$ has the defining property that

2089: \begin{equation}

2090: (H - \lam^2) \Rt(\lam) = \Id,

2091: \ilabel{defining}\end{equation}

2092: as an operator on $\CIdot(X; \scOh)$, and that

2093: \begin{equation}

2094: \scwf_{\XXbt}(\Rt(\lam)) \subset K_- \text{ as defined in } \eqref{Kminus}.

2095: \ilabel{outgoing}

2096: \end{equation}

2097: Here $\XXbt = \XXb \setminus \{ \lb \cup \rb \}$ is regarded as an open

2098: manifold with boundary, so that we can talk about the scattering wavefront

2099: set over the interior of bf. Equation \eqref{outgoing} is the microlocal

2100: version of the outgoing Sommerfeld radiation condition. (For example, if

2101: $\lam > 0$, 

2102: $e^{i\lam/x}$ has wavefront set in $K_-$, while $e^{-i\lam/x}$ does not.)

2103: 

2104: Equation \eqref{defining} means that the kernel of $\Rt(\lam)$, which we also

2105: denote by $\Rt(\lam)$ by an abuse of notation, satisfies

2106: \begin{equation}

2107: (H_l - \lam^2) \Rt(\lam) =K_{\Id}, 

2108: \ilabel{diag}\end{equation}

2109: where $K_{\Id}$ is the kernel of the identity operator, i.e.\ it is

2110: a delta distribution on the diagonal, and $H_l$ is the operator

2111: $H$ acting on the left factor of $X$ in $X\times X$.

2112: 

2113: There are four main steps in the construction. First we find an

2114: approximation to $\Rt(\lam)$ in the scattering calculus, $G_1(\lam)

2115: \in \Psisc^{-2}(X)$, which removes the singularity on the diagonal in

2116: \eqref{diag}. This leaves an error which, when viewed on the b-double space

2117: $\XXb$, is singular at the boundary of the diagonal $\pa \diag_\bl$. In

2118: fact, it is Legendrian at a Legendre submanifold lying over $\pa \diag_\bl$

2119: which we denote $N^* \diag_\bl$ (see \eqref{Ndiag}). We solve this error

2120: away locally 

2121: near $\pa \diag_\bl$ using an intersecting Legendrian construction which is

2122: due (in the Lagrangian setting) to Melrose and Uhlmann

2123: \cite{Rbm-Uhl:Intersecting}; the 

2124: singularities inside $N^* \diag_\bl \cap \Sigma(H_l - \lam^2)$ propagate in a

2125: Legendre submanifold $\Lpl$. This Legendre submanifold intersects both lb

2126: and rb, and an `outgoing' Legendre submanifold $\Lsl$; $(\Lpl, \Lsl)$ form

2127: an conic pair of Legendre submanifolds and we can find a conic Legendre pair

2128: which solves away the error up to an error term which is Legendrian only at

2129: $\Lsl$, ie we can solve away the errors at $\Lpl$ completely. Finally, this

2130: outgoing error is solved away, using a very standard argument in scattering

2131: theory, at lb and bf, leaving an error $E(\lam)$ which is compact on

2132: weighted $L^2$ spaces $x^l L^2(X)$ for all $l > 1/2$. 

2133: 

2134: Thus, we seek $G(\lam)$ (and $\Rt(\lam)$) in the class 

2135: \begin{equation}

2136: \Psisc^{2,0}(X) + I^{-\frac1{2}}(N^* \diag_\bl, \Lpl; \scOh) +

2137: I^{-\frac{1}{2}, 

2138:   \frac{n-2}{2}; \frac{n-1}{2}, \frac{n-1}{2}}(\Lpl, \Ls(\lam); \scOh)

2139: \ilabel{seek}\end{equation}

2140: where the second term is an intersecting Legendrian distribution and the

2141: third is a Legendre conic pair with orders $-1/2$ at $\Lpl$, $(n-2)/2$ at

2142: $\Lsl$ and $(n-1)/2$ at lb and rb. In this class of distributions there is a

2143: unique solution $\Rt(\lam)$ to \eqref{defining} and \eqref{outgoing}. 

2144: 

2145: To avoid cumbersome notation, $Q$ will denote a generic correction to

2146: the para\discretionary{-}{metrix}{metrix}

2147: constructed so far, and $E$ will denote a generic error. The values of these

2148: symbols is allowed to change from line to line. 

2149: 

2150: 

2151: \subsection{Pseudodifferential approximation}\label{sec:pseudodiff-approx}

2152: The first step in constructing

2153: $G(\lam)$ is a very 

2154: standard argument. We seek $G_1(\lam) \in \Psisc^{-2}(X)$ such that

2155: $$

2156: (H - \lam^2) G_1(\lam) = \Id + E_1(\lam), \quad E_1(\lam) \in

2157: \Psisc^{-\infty}(X).

2158: $$

2159: This will mean that the error term $E_1(\lam)$ has a smooth kernel

2160: (times the standard half-density) on $\XXsc$, so that we have solved away

2161: completely 

2162: the singularity along the diagonal. 

2163: 

2164: The standard elliptic argument applies here 

2165: since the interior symbol of $H - \lam^2$

2166: is elliptic. Thus, we first choose any $Q \in \Psisc^{-2}(X)$ whose

2167: interior symbol is $|\cdot|_g^{-2} = (\sigma^2(H - \lam^2))^{-1}$. Then

2168: $$

2169: (H - \lam^2) Q = \Id + E, \quad E \in \Psisc^{-1}(X).

2170: $$

2171: Multiplying $Q$ by a finite Neumann series $(\Id + E + \dots + E^{k-1})$ thus

2172: gives an error $E^k \in \Psisc^{-k}(X)$. Taking an 

2173: asymptotic limit gives us a $G_1(\lam) \in \Psisc^{-2}(X)$ with the

2174: desired error term. 

2175: 

2176: 

2177: \subsection{Intersecting Legendrian construction} 

2178: \label{sec:intersecting-Leg-constr}

2179: 

2180: In the next step of the construction we move to $\XXb$, and view the error

2181: $E_1(\lam)$ from the first step of the construction on $\XXb$ rather than

2182: $\XXsc$. On $\XXb$ it has a smooth kernel except at $\pa \diag_\bl$ where it

2183: has a conic singularity. That is, at $\pa \diag_\bl = \{ x' = 0,

2184: \sigma \equiv x'/x'' = 1, y' = y'' \}$,  the kernel is a smooth 

2185: (and compactly supported) function of $x'$, $S = (\sigma-1)/x'$, $Y =

2186: (y'-y'')/x'$ and $y'$; this is easy to see since these are smooth

2187: coordinates on sf $\subset \XXsc$. Using the Fourier transform, we write

2188: $$

2189: E_1(\lam) = \left( \int_{\RR^n} e^{i((y'-y'')\cdot \eta + (\sigma - 1)

2190:   t)/x'} 

2191: a(x',y',\eta, t) \, d\eta \, dt \right) \nu 

2192: $$

2193: where $a$ is smooth in all variables, and in addition Schwartz in $(\eta,

2194: t)$. The phase function $(y'-y'')\cdot \eta + (\sigma - 1)t$ parametrizes

2195: the Legendrian

2196: \begin{equation}

2197: N^* \diag_\bl = \{ y' = y'', \sigma = 1, \mu' = -\mu'', \tau' = -\tau'' \}.

2198: \ilabel{Ndiag}\end{equation}

2199: Therefore, $E_1(\lam)$ is a Legendre distribution of order $0$

2200: associated to $N^* \diag_\bl$. (To be pedantic, $E_1(\lam)$

2201: does not fall strictly in the class of Legendre distributions as defined by

2202: Melrose and Zworski since its microsupport is not compact; from

2203: \eqref{Ndiag} we see that the microsupport is a vector

2204: bundle over $\pa \diag_\bl$. It is instead an `extended Legendre distribution'

2205: as defined in \cite{Hassell:Plane}. However this is of no significance since

2206: the symbol 

2207: is rapidly decreasing in each fibre of the vector bundle, hence all

2208: constructions we wish to perform here are valid in this context.)

2209: 

2210: Observe that $\symbb(H_l - \lam^2) = {\tau'}^2 + h(y',\mu') - \lam^2$ vanishes

2211: on a codimension one submanifold of $N^* \diag_\bl$, and does so

2212: simply. Consider the vector field $V_l$ which is given by \eqref{Vl}. Since

2213: ${\tau'}^2 

2214: + h = \lam^2 \neq 0$ on $\Sigma(H_l - \lam^2)$, at least one of the coefficients

2215: of $\pa_\sigma$ and $\pa_{\tau'}$ in \eqref{Vl} is nonzero, so $V_l$ is

2216: transverse to $N^* 

2217: \diag_\bl$ at the intersection with $\Sigma(H_l - \lam^2)$. We define

2218: $L^\circ(\lam)$ to be the flowout Legendrian from $N^* \diag_\bl \cap

2219: \Sigma(H_l - \lam^2)$ 

2220: with respect to $V_l$, and $L^\circ_\pm(\lam)$ to be the flowout in the

2221: positive, resp. negative direction with respect to $V_l$. Thus, at least

2222: locally near $N^* \diag_\bl$, $L^\circ_\pm (\lam)$ are smooth manifolds with

2223: boundary. Notice that by \eqref{tau}, $N^* \diag_\bl$ is contained in $\tau

2224: = 0$ and $V_l(\tau) < 0$ at $N^* \diag_\bl$. Thus, at least locally near

2225: $N^* \diag_\bl$, $L^\circ_+(\lam)$ is contained in $K_- = \{ \tau \leq 0 \}$.

2226: The global properties of $L^\circ_\pm (\lam)$ are studied in the next

2227: section; in this section we only work microlocally near $N^* \diag_\bl$. 

2228: 

2229: 

2230: The first step in solving away the error $E_1(\lam)$ from the previous step is

2231: to find an intersecting Legendrian $Q \in 

2232: I^{-1/2} (\XXb, (N^*\diag_\bl, \Lpl); \sfOh)$ such that 

2233: \begin{equation}

2234: (H_l - \lam^2) Q - E_1(\lam) \in I^{1}(\XXb, N^* \diag_\bl; \sfOh) +

2235: I^{\frac{3}{2}} (\XXb, (N^* \diag_\bl, \Lpl), \sfOh),

2236: \ilabel{err-1}\end{equation}

2237: microlocally near $N^* \diag_\bl$. To do this we choose $Q$ with symbol on

2238: $N^* \diag_\bl$ equal to $\symbb^0(H_l - \lam^2)^{-1}

2239: \sigma^0(E_+^{(1)}(\lam))$. This is an admissible symbol on $N^* \diag_\bl$ by

2240: \eqref{int-Leg-dist}, and \eqref{ex-int-1}, 

2241: since $\symbb^0(H_l - \lam^2)$ is a boundary defining

2242: function for $\Lpl$ on $N^* \diag_\bl$. It determines the value of the

2243: symbol on $\Lpl$ at the boundary by \eqref{compat}. We extend this symbol by

2244: requiring that the transport equation, \eqref{transport}, be satisfied. 

2245: This equation is a first order

2246: linear ODE with smooth coefficients, so there is a unique solution in a

2247: neighbourhood of $N^* \diag_\bl$. Then the symbol of order $-1/2$ of 

2248: $(H_l - \lam^2) R_0 - E_+^{(1)}(\lam)$ vanishes, and there is an additional

2249: order of vanishing on $\Lpl$ since the transport equation is satisfied.

2250: Thus by \eqref{ex-int-2} the error term is as in \eqref{err-1}.

2251: 

2252: We now show inductively that we can solve away an error $E_k$ which is in

2253: $I^{k}(N^* \diag_\bl) + I^{k+1/2} (N^*\diag_\bl, \Lpl)$ with a term

2254: $Q_k \in I^{k+1/2}(N^*\diag_\bl, \Lpl)$, 

2255: up to an error which is in $I^{k+1}(N^* \diag_\bl) + I^{k+3/2}

2256: (N^*\diag_\bl, \Lpl)$. The argument is the same as above: we take the symbol

2257: of order $k$ on $N^*\diag_\bl$ equal to $\symbb^0(H_l - \lam^2)^{-1}

2258: \sigma^k(E_k)$, and the symbol on $\Lpl$ to solve away the symbol of

2259: order $k+1/2$ of $E_k$ when the transport operator is applied to it. 

2260: Taking an asymptotic sum of $Q$ and the $Q_k$'s gives us an error term

2261: which is microlocally trivial near $N^* \diag_\bl$. 

2262: By cutting off away from $\pa \diag_\bl$, we obtain an error 

2263: $$

2264: E_2(\lam) \in I^{1/2}_c(\Lpl),

2265: $$

2266: where the subscript $c$ indicates that the microlocal support is compact and

2267: disjoint from the intersection with $N^* \diag_\bl$. 

2268: 

2269: 

2270: \subsection{Structure of $L(\lam)$}

2271: In this section we analyze the global structure of $L^\circ(\lam)$. This was

2272: defined as the flowout from $N^* \diag_\bl \cap \Sigma(H_l - \lam^2)$ by the

2273: vector field $V_l$. In fact, it is quite easy to see that $N^* \diag_\bl

2274: \cap \Sigma(H_l - \lam^2) = N^* \diag_\bl \cap \Sigma(H_r - \lam^2)$. Moreover,

2275: neither $V_l$ nor $V_r$ is tangent to $N^* \diag_\bl$ at any point contained

2276: in $\Sigma(H_l - \lam^2)$, but the difference $V_l - V_r$ is tangent to $N^*

2277: \diag_\bl$. Since $V_l$ and $V_r$ commute, this shows that the flowout with

2278: respect to $V_l$ is the same as the flowout with respect to $V_r$. We will

2279: soon see that the symbols of our parametrix on $\Lpl$, defined so as to

2280: satisfy the left transport equation, also satisfy the right transport

2281: equation. 

2282: 

2283: 

2284: It is convenient to write down $L^\circ(\lam)$ explicitly. Indeed, the

2285: computation of Melrose and Zworski can be applied with a minor change

2286: (that takes care of the behavior in $\sigma$) to deduce that

2287: \begin{equation}\begin{split}\label{eq:sp-1c}

2288: L^\circ(\lam) 

2289: =&\{(\theta,y',y'',\tau',\tau'',\mu',\mu''):\ \exists(y,\muh)\in S^*\bX,

2290: \ s,s'\in(0,\pi),\ \text{s.t.}\\

2291: &\quad \sigma = \tan \theta = \frac{\sin s'}{\sin s},

2292: \ \tau'=\lambda\cos s',

2293: \ \tau''=-\lambda\cos s,\\

2294: \quad(y',\mu')&=\lambda\sin s'\exp(s'H_{\half h})(y,\muh),

2295: (y'',\mu'')=-\lambda\sin s \exp(s H_{\half h})(y,\muh)\}\\

2296: &\cup T_+(\lam) \cup T_-(\lam), \quad T_\pm (\lam) = 

2297: \{(\sigma,y,y,\pm \lambda,\mp \lambda,0,0):\ \sigma\in(0,\infty),\ y\in\bX\}

2298: \end{split}\end{equation}

2299: The sets $T_\pm (\lam)$ are, 

2300: for fixed $y$, integral curves of both vector fields, and they appear

2301: separately only because we used the parameterization of Melrose-Zworski.

2302: The smooth structure near $T_\pm (\lam)$ follows from the

2303: flowout description, but is not apparent in this parameterization;

2304: we discuss it below while describing the closure of $L^\circ(\lam)$.

2305: 

2306: The closure $L(\lam)$ of $L^\circ(\lam)$ is $\sfT\XXb$ is clear from the

2307: above description; it is

2308: \begin{equation}

2309: \cl L =L(\lam)\cup F_{\lambda}\cup F_{-\lambda}

2310: \end{equation}

2311: where

2312: \begin{equation}\begin{split}\label{eq:sp-1}

2313: L(\lam) 

2314: =&\{(\theta,y',y'',\tau',\tau'',\mu',\mu''):\ \exists(y,\muh)\in S^*\bX,

2315: \ s,s'\in[0,\pi],\\

2316: &\quad (\sin s)^2+(\sin s')^2>0,\ \text{s.t.}\\

2317: &\quad \sigma = \tan \theta = \frac{\sin s'}{\sin s},

2318: \ \tau'=\lambda\cos s',

2319: \ \tau''=-\lambda\cos s,\\

2320: \quad(y',\mu')&=\lambda\sin s'\exp(s'H_{\half h})(y,\muh),

2321: (y'',\mu'')=-\lambda\sin s \exp(s H_{\half h})(y,\muh)\}\\

2322: & \cup T_+(\lam) \cup T_-(\lam), \quad \text{and} \\

2323: F_{\lam} &= \{(\sigma,y',y'',-\lam,-\lam,0,0) \mid 

2324: \exists \text{ geodesic of length } \pi \text{ connecting } y', \, y'' \}.

2325: \end{split}\end{equation}

2326: 

2327: 

2328: Note that the requirement $(\sin s)^2+(\sin s')^2>0$ just

2329: means that $s$ and $s'$ cannot take values in $\{0,\pi\}$ at the same time.

2330: The set $L(\lam) \setminus L^o(\lam)$ comprises those points where one of $s$,

2331: $s'$ takes values in $\{ 0, \pi \}$ while the other lies in $(0, \pi)$. The

2332: sets $T_\pm (\lam)$ in \eqref{eq:sp-1} comprise the limit points where $s$ and

2333: $s'$ converge either both to $0$ or both to $\pi$, whilst $F_{\pm \lam}$

2334: comprise the limit points as $s \to 0$ and $s' \to \pi$ or vice versa. 

2335: 

2336: 

2337: 

2338: 

2339: The smooth structure near $T_\pm (\lam)$ becomes apparent 

2340: if we note that near $\tau'=\lambda$,

2341: $\tau''=-\lambda$, $\sigma\in[0,C)$ where $C>1$, $L(\lam)$ is

2342: given by

2343: \begin{equation}\begin{split}\label{eq:sp-1-a}

2344: \{(\sigma&,y',y'',\tau',\tau'',\mu',\mu''):\ \exists(y,\mu)\in T^*\bX,

2345: \ |\mu|<C^{-1},\ \sigma\in[0,C)\ \text{s.t.}\\

2346: &\tau'=\lambda(1-|\sigma\mu|^2)^{1/2},

2347: \ \tau''=-\lambda(1-|\mu|^2)^{1/2},\\

2348: &(y',\mu')=\lambda\exp(f(\sigma\mu)V_h)(y,\sigma\mu),

2349: \ (y'',\mu'')=-\lambda\exp(f(\mu)V_h)(y,\mu)\}

2350: \end{split}\end{equation}

2351: where

2352: $f(\mu)={|\mu|}^{-1}\arcsin(|\mu|)$

2353: is smooth and $f(0)=1$. Thus, the differential of the map

2354: \begin{equation}

2355: (y,\mu)\mapsto-\lambda\exp(f(\mu)V_h)(y,\mu)=(y'',\mu'')

2356: \end{equation}

2357: is invertible near $\mu=0$, so it gives a diffeomorphism near $|\mu|=0$.

2358: Hence, $\sigma$ and $(y'',\mu'')$ give coordinates on $L(\lam)$ in this

2359: region, so $L(\lam)$ is smooth here. Away from $T_+(\lam)$, coordinates on

2360: $L(\lam)$ are $\sigma$, $y''$, $\muh''$ and $s$. 

2361: 

2362: In the coodinates $(y, \hat \mu, s, s')$, the vector field $V_l$ is given by

2363: $\sin s' \pa_{s'}$ and $V_r$ is given by $\sin s \pa_s$. The intersection of

2364: $L(\lam)$ and $N^* \diag_\bl$ is given by $\{ s = s' \}$. Thus

2365: $\Lpl$ is given by $\{ s \leq s' \}$. On $\Lpl$, $\tau = \tau' + \sigma

2366: \tau''$ by \eqref{tau}, so 

2367: $$

2368: \tau = \lam \frac{\sin (s-s')}{\sin s} \leq 0 \text{ on } \Lpl.

2369: $$

2370: Thus, any distribution in $I^m(N^* \diag_\bl, \Lpl)$ satisfies condition

2371: \eqref{outgoing}. 

2372: 

2373: We also define

2374: \begin{equation}

2375: \Ls(\lam)=\{(\theta,y',y'',-\lam,-\lam,0,0):\ y',y''\in\bX,\ \theta

2376: \in[0,\pi/2]\},

2377: \label{eq:L-sharp}\end{equation}

2378: so $\Ls(\lam)$ is a Legendrian submanifold of $\sct_{\ffb}\XXbt$, and

2379: \begin{equation}

2380: \cl L\cap\Ls(\pm \lam)=F_{\pm\lambda}.

2381: \end{equation}

2382: 

2383: \begin{prop}\ilabel{conicpair} The pair 

2384: \begin{equation}

2385: \Lt (\lam) = (L(\lam), \Ls(\lam) \cup \Ls(-\lam))

2386: \end{equation}

2387: is a pair of intersecting Legendre manifolds with conic points.

2388: \end{prop}

2389: 

2390: \begin{proof} 

2391: We must show that when the set $\{ t q \mid t > 0, q \in \Lsl \}$ is blown

2392: up inside $\sfT \XXb$, the closure of $L(\lam)$ is a smooth manifold with

2393: corners which meets the front face of the blowup transversally. 

2394: Let us restrict attention to a neighbourhood of $\Lsl$; the case of

2395: $\Ls(-\lam)$ is similar. 

2396: Consider the vector field $V_l + V_r$. By \eqref{Vl} and \eqref{Vr}, in

2397: $\Sigma(H_l - \lam^2) \cap \Sigma(H_r - \lam^2)$ this is given by

2398: $$

2399: 2(\tau' - \tau'') \sigma \pa_\sigma + 2\tau' \mu' \cdot \pa_{\mu'} + 

2400: 2\tau'' \mu'' \cdot \pa_{\mu''} + \pa_{\mu'} h' \cdot \pa_{y'} - 

2401: \pa_{y'} h' \cdot \pa_{\mu'} + \pa_{\mu''} h'' \cdot \pa_{y''} - 

2402: \pa_{y''} h'' \cdot \pa_{\mu''}

2403: $$

2404: This is

2405: equal to $-2\lam$ times the b-normal vector field $\mu' \cdot \pa_{\mu'} +

2406: \mu'' \cdot \pa_{\mu''}$ plus a sum of vector fields which have the form 

2407: $\rho V$, where $\rho$ vanishes at $\Lsl$ and $V$ is tangent to lb and

2408: $\Lsl$ (all

2409: considerations taking place inside $\Sigma(H_l - \lam^2) \cap \Sigma(H_r

2410: - \lam^2)$). Thus, under blowup of $\{ t q \mid t > 0, q \in \Lsl \}$, $V_l +

2411: V_r$ lifts to a vector field of the form

2412: \begin{equation}

2413: V_l + V_r = -2\lam s \pa_s + s W,

2414: \ilabel{VlVr}\end{equation}

2415: where $W$ is smooth and tangent to the boundary of $\hat L(\lam)$, 

2416: and so dividing by $s$

2417: yields a nonvanishing normal vector field plus a smooth tangent

2418: vector field. As above, such a vector field has a continuation

2419: across the boundary to the double of $\hat L(\lam)$ (across the front face) as

2420: a smooth 

2421: nonvanishing vector field. This holds true smoothly up to the corner with

2422: lb, so $\hat L(\lam)$ is a smooth manifold with corners. 

2423: \end{proof}

2424: 

2425: 

2426: \subsection{Smoothness of symbols}\label{smoothness} In the next stage of

2427: the construction, we 

2428: solve away the error $E_2(\lam)$ which is microsupported in the interior

2429: of $\Lpl$. This involves solving the transport equation globally on

2430: $\Lpl$. In view of Proposition~\ref{conicpair}, we can expect the

2431: construction to involve Legendrian conic pairs with respect to $(\Lpl,

2432: \Lsl)$. In order for the symbol to be quantizable to such a conic pair, we

2433: need to show regularity of the symbol on $\Lplh$, so that it lies in the

2434: symbol space of the exact sequence from Proposition~\ref{ex-conic-2}. 

2435: That is, the symbol of order $j-1/2$ on $\Lpl$ should lie in

2436: \begin{equation}

2437: \rho_{\lb}^{n/2-j} \rho_{\rb}^{n/2-j} \rho_{\#}^{(n-1)/2-j} \CI(\Lplh; 

2438: \Omega_b^{1/2} \otimes S^{[j-1/2]}(\Lplh)).

2439: \ilabel{regularity}

2440: \end{equation}

2441: (We will ignore the symbol bundle $S^{[j-1/2]}$ in the rest of this

2442: section.)

2443:  

2444: 

2445: To do this, we observe that the symbol on $\Lpl$ automatically satisfies

2446: the transport equation for the right Hamilton vector field. To see this,

2447: let $G_2(\lam)$ be the approximation to $\Rt(\lam)$ constructed so far, with

2448: $$

2449: (H_l - \lam^2) G_2(\lam) - K_{\Id} = E_2(\lam),

2450: $$

2451: and

2452: $$

2453: \scwf(E_2(\lam)) \subset \Lpl \cap \{ \tau < -c \}

2454: $$

2455: for some $c > 0$. Consider applying $H_r - \lam^2$ to $E_2(\lam)$. Since $H_l$

2456: and $H_r$ commute, and $H_l K_{\Id} = H_r K_{\Id}$, we have

2457: $$\begin{aligned}

2458: (H_l - \lam^2) \left( (H_r - \lam^2) G_2(\lam) - \Id \right) 

2459: &= (H_r - \lam^2) \left ( (H_l - \lam^2) G_2(\lam) - \Id \right) \\ 

2460: &= (H_r - \lam^2) E_2(\lam).

2461: \end{aligned}$$

2462: We claim that $\scwf( (H_r - \lam^2) G_2(\lam) - K_{\Id})$ is contained in $

2463: \{ \tau < -c \}$. For if there is a point where $\tau \geq -c$, then by

2464: \eqref{prop-3}, the maximal bicharacteristic ray in 

2465: \begin{multline}

2466: \Sigma(H_l - \lam^2) \setminus \scwf\Big((H_l - \lam^2)\big((H_r - \lam^2) G_2(\lam)

2467: - K_{\Id}\big)\Big) \\ = \Sigma(H_l - \lam^2) \setminus \scwf((H_r - \lam^2)

2468: E_2(\lam))

2469: \end{multline}

2470: lies in $\scwf( (H_r - \lam^2) G_2(\lam) - K_{\Id})$. Such rays always propagate

2471: into $\tau > 0$. But

2472: $$

2473: \scwf(G_2(\lam)) \subset \{ \tau \leq 0 \}, \quad

2474: \scwf(K_{\Id}) \subset \{ \tau \leq 0 \}, 

2475: $$

2476: so this is impossible. Consquently, $(H_r - \lam^2) G_2(\lam) - K_{\Id}$ has

2477: no scattering wavefront set

2478: for $\{ \tau \geq -c \}$, and so the symbols of $G_2(\lam)$ must obey the

2479: right transport equations in this region. By cutting off the symbols closer

2480: and closer 

2481: to the boundary of $\Lpl$, we see that the right transport equations must be

2482: satisfied everywhere on $\Lpl$. 

2483: 

2484: Let us examine the form of these transport equations at the boundary of

2485: $\Lpl$. Near lb, we have coordinates $(y'', \mu'', \sigma)$ near $T_+(\lam)$

2486: and $(y'', \muh'', \sigma, s)$ away from $T_+(\lam)$, which are valid

2487: coordinates for $\sigma < 2$, say. The situation near rb is similar so the

2488: argument will be omitted. 

2489: 

2490: In either set of coordinates, the left vector field, restricted to $\Lpl$, takes

2491: the form 

2492: $$

2493: V_l = 2\tau' \sigma \pa_\sigma .

2494: $$

2495: Also, by Lemma~\ref{lem:sub-pr}, the subprincipal symbol of $H_l - \lam^2$,

2496: which is equal to the subprincipal symbol of $H - \lam^2$ in the

2497: singly-primed coordinates, vanishes where $\mu'$ vanishes, and $\mu' = 0$ at

2498: lb on $\Lpl$. Therefore, by \eqref{transport}, the transport equation for the

2499: symbol of order $-1/2$ takes the

2500: form

2501: \begin{equation} 

2502: \Big( -i\big({\mathcal L}_{V_l} 

2503: - n \tau' \big) + 

2504: \sigma f \Big) a_0 |\frac{d\sigma}{\sigma} dy'' d\mu''|^\half = 0, \ f \in

2505: \CI(\Lpl),

2506: \ilabel{left-eqn-0}

2507: \end{equation}

2508: near $T_+(\lam)$, or 

2509: \begin{equation} 

2510: \Big( -i\big({\mathcal L}_{V_l} 

2511: - n \tau' \big) + 

2512: \sigma f \Big) a_0 |\frac{d\sigma}{\sigma} \frac{ds}{s} dy'' d\muh''|^\half

2513: = 0, \ f \in \CI(\Lpl),

2514: \ilabel{left-eqn}

2515: \end{equation}

2516: away from $T_+(\lam)$,

2517: which gives an equation for $a_0$ of the form

2518: \begin{equation}

2519: -i\tau' \big( \pa_\sigma  + f \big) (\sigma^{-n/2}a_0) = 0,

2520: \quad f \in \CI(\Lpl).

2521: \ilabel{eq-Vl}\end{equation}

2522: This shows that $\sigma^{-n/2} a_0$ is smooth across $\sigma = 0$. 

2523: 

2524: To show regularity near $\Lsl$, we use the fact that the symbol satisfies

2525: both the right and left transport equation. We take the sum of the transport

2526: equations that obtain when we use the total boundary defining function $x'$

2527: for $H_l$, and $x''$ for $H_r$. The right transport operator with respect to

2528: $x''$ takes the form

2529: $$

2530: -i\big({\mathcal L}_{V_r} - n \tau'' \big) + p_{\sub} 

2531: $$

2532: However, by \eqref{change} the symbol written in terms of $x''$ is equal to

2533: $(x''/x')^{-1/2-n/2}$ times the symbol written in terms

2534: of $x'$. Since we are writing the symbol in terms of $x'$, we must include a

2535: factor $\sigma^{-1/2-n/2}$ to be consistent with \eqref{left-eqn}. 

2536: This gives an equation for $a_0$ of the form

2537: \begin{equation} 

2538: \Big( -i\big({\mathcal L}_{V_r} 

2539: - n \tau'' \big) + 

2540: p_{\sub}(y'',\mu'',\tau'') \Big)  

2541: \Big( \sigma^{-1/2-n/2} a_0 |\frac{d\sigma}{\sigma} \frac{ds}{s} dy''

2542: d\muh''|^\half \Big) = 0.

2543: \ilabel{right-eqn}

2544: \end{equation}

2545: In view of the term $-2\tau'' \sigma \pa_\sigma$ in

2546: the formula \eqref{Vr} for $V_r$, and since $p_{\sub}$ vanishes at $s=0$ since

2547: $\mu'' = 0$ there, we get an equation for $a_0$ 

2548: \begin{equation}

2549: \big( V_r + \tau'' + s f' \big) a_0 = 0 \quad f' \in \CI(\Lplh).

2550: \ilabel{eq-Vr}

2551: \end{equation}

2552: Combining with the left transport equation gives an equation which, using

2553: \eqref{VlVr} and the fact that $\tau' = \tau'' = -\lam$ at $s=0$ takes the form

2554: $$

2555: 2\lam \big( s \pa_s - \frac{n-1}{2} + sW + s\tilde f \big) a_0 = 0,

2556: $$

2557: where $W$ is tangent to the boundary of $\Lplh$ and $\tilde f$ is smooth on

2558: $\Lplh$. This may be written

2559: $$

2560: \big( \pa_s + W + \tilde f \big) \big( s^{-(n-1)/2} a_0 \big) = 0,

2561: $$

2562: This together with \eqref{eq-Vl} shows that $a_0 \in \sigma^{n/2} s^{(n-1)/2}

2563: \CI(\Lplh)$. 

2564: 

2565: \

2566: 

2567: It follows that there is a Legendre distribution $I^{-\frac1{2},

2568:   \frac{n-2}{2}; \frac{n-1}{2}, \frac{n-1}{2}}(\Lpl,

2569: \Lplh)$ which has the correct symbol of order $-1/2$ at $\Lpl$. Thus it

2570: solves the equation

2571: \begin{equation}

2572: (H_l - \lam^2) Q - E(\lam) \in I^{\frac{3}{2}, \frac{n}{2};

2573:  \frac{n+3}{2}, \frac{n-1}{2}}(\XXb, \Lpl, \Lsl; \sfOh),

2574: \end{equation}

2575: where $(n+3)/2$ is the order of

2576: vanishing at lb and $(n-1)/2$ is the order of vanishing at rb. The order

2577: of improvement at lb is two since not only is the Legendrian $G_1$ of

2578: Definition~\ref{M-Leg-submfld} at lb characteristic for $H_l - \lam^2$, but

2579: the transport 

2580: operator for symbols of order $(n-1)/2$ vanishes, so we automatically get two

2581: orders of improvement there. At rb however we can expect no improvement. 

2582: As shown above, $Q$ will automatically satisfy the equation

2583: $$

2584: (H_r - \lam^2) Q - E_2(\lam) \in I^{\frac{3}{2}, \frac{n}{2};

2585:  \frac{n-1}{2}, \frac{n+3}{2}}(\XXb, \Lpl, \Lsl; \sfOh).

2586: $$

2587: Let us assume by induction that we have a kernel which solves the left equation

2588: above up to an error in 

2589: \begin{equation}

2590: I^{k+\frac{1}{2}, \frac{n}{2};

2591:  \frac{n+3}{2}, \frac{n-1}{2}}(\XXb, \Lpl, \Lsl; \sfOh)

2592: \ilabel{error-l-k}

2593: \end{equation}

2594: and hence the right equation up to an

2595: error in 

2596: \begin{equation}

2597: I^{k+\frac{1}{2}, \frac{n}{2};

2598:  \frac{n-1}{2}, \frac{n+3}{2}}(\XXb, \Lpl, \Lsl; \sfOh)

2599: \end{equation}

2600: We wish to improve this by one order

2601: at $\Lpl$. To do this, we choose $Q_k \in I^{k-\frac1{2}, \frac{n-2}{2}; 

2602: \frac{n-1}{2}, \frac{n-1}{2}}(\Lpl, \Lplh)$ to have the symbol of order

2603: $k-1/2$ on $\Lpl$ 

2604: which solves the left transport equation (and

2605: therefore the right transport equation) on $\Lpl$. We need to investigate 

2606: the regularity of this symbol to see if it extends to a Legendrian conic

2607: pair. The argument is very analogous to the one above, but now we have error

2608: terms on the right hand side. In the first region, after removing the

2609: half-density factor, we get an equation of the form

2610: \begin{equation}

2611: -2i\tau' \big( \sigma \pa_\sigma + (-\frac{n}{2} + k) + \sigma f \big) a_k =

2612: b_k, 

2613: \quad f \in \CI(\Lpl).

2614: \ilabel{eq-Vl-k}\end{equation}

2615: The term $b_k$ comes from the error to be solved away. Since the error term

2616: is of order $k+1/2$ at $\Lpl$ and order $(n+3)/2$ at lb, 

2617: $b_k \in \sigma^{n/2-k+1}

2618: \CI(\Lpl)$. This shows that $a_k \in \sigma^{n/2-k}

2619: \CI(\Lpl)$, as desired. Similarly, in the second region, near the corner

2620: $\lb \cap \Lsl$, by combining the vector fields $V_l + V_r$ we get an

2621: equation of the form 

2622: \begin{equation}

2623: -2i\tau'' \Big( - s \pa_s + s W +( - \frac{n-1}{2} + k) + s \sigma f

2624: \Big) a_k = b_k, \quad f \in \CI(\Lpl)

2625: \ilabel{eq-Vr-k}\end{equation}

2626: with $b_k$ again the error to be solved away. To calculate its order of

2627: vanishing at $s=0$, consider the transport

2628: equation for symbols of order $(n-2)/2$ at $\Lsl$. Noting that the

2629: subprincipal symbols vanish identically on $\Lsl$, the left transport

2630: operator is

2631: $$ 

2632: -i \big( \mathcal{L}_{V_l} -  \tau' \big) 

2633: $$

2634: whilst the right transport operator with respect to $x''$ is 

2635: $$ 

2636: -i \big( \mathcal{L}_{V_r} -  \tau'' \big) 

2637: $$

2638: To write this with respect to $x'$ we must conjugate by $\sigma$ (by

2639: \eqref{change}). In view of the term $-2 \tau'' \sigma \pa_\sigma$, this

2640: changes the operator to

2641: $$ 

2642: -i \big( \mathcal{L}_{V_r} +  \tau'' \big) .

2643: $$

2644: The sum of these two operators vanishes on $\Lsl$ so actually the right hand

2645: side in  

2646: \eqref{eq-Vr-k} comes from a term in $I^{k+\frac{1}{2}, \frac{n+2}{2};

2647:  \frac{n+1}{2}, \frac{n-1}{2}}(\XXb, \Lpl, \Lsl; \sfOh)$. From

2648: Proposition~\ref{ex-conic-2} we

2649: see that $b_k \in s^{(n-1)/2-k+1} \sigma^{n/2-k} \CI(\Lplh)$, 

2650: one power in $s$ better than might be expected. 

2651: This shows that $a_k \in s^{(n-1)/2-k} \sigma^{n/2-k}

2652: \CI(\Lplh)$, as desired. Therefore, one can find a Legendre conic pair with

2653: symbol of order $k-1/2$ on $\Lpl$ equal to $a_k$ which solves away the error

2654: term of order $k+1/2$ at $\Lpl$. 

2655: This completes the inductive step. By asymptotically

2656: summing these correction terms, we end up with an approximation $G_3(\lam)$ to the

2657: resolvent kernel with an error $E_3(\lam)$ in $I^{\frac{n-2}{2};

2658: \frac{n+3}{2}, \frac{n-1}{2}}(\XXb, \Lsl; \sfOh)$. That is, we have solved

2659: away the scattering wavefront set 

2660: of the error term at $\Lpl$ completely. 

2661: 

2662: \subsection{Solving away outgoing error}\label{outgoing-error}

2663: In the last step of the construction of the parametrix, we solve away the

2664: error to infinite order at bf and lb. We begin by considering the expansion

2665: at rb. By construction, the parametrix $G_3(\lam)$ constructed so far has an

2666: expansion at rb

2667: $$

2668: G_3(\lam) \sim e^{i\lam/x''} {x''}^{(n-1)/2} \sum_{j \geq 0} {x''}^j

2669: g_j(z',y'') \cdot \nu' \cdot \nu'',

2670: $$

2671: where $g_j(z',y'') \cdot \nu' \in I^{-n/4-j, n/4 - 1/2-j}(G_{y''}(\lam),

2672: \Gs(\lam))(X; \scOh)$,

2673: $G_{y''}(\lam)$ is the fibre Legendrian of Definition~\ref{M-Leg-submfld} and

2674: $y''$ is regarded as a smooth parameter. The factors $\nu'$, $\nu''$ are the

2675: Riemannian half-density factors on $X$ lifted to $\XXb$ via the left and

2676: right projections, respectively. We will ignore the half-density factors

2677: from here on; since $\Lap(a \cdot \nu' \cdot \nu') = \Lap(a) \cdot \nu'

2678: \cdot \nu'$, this only has the effect of changing $H = \Lap + P$ to $\Lap +

2679: P'$ for some $P'$ with the same properties as $P$. 

2680: 

2681: The error term after applying $H - \lam^2$ to $G_3(\lam)$ has the form

2682: \begin{equation}

2683: E_3(\lam) \sim e^{i\lam/x''} {x''}^{(n-1)/2} \sum_{j \geq 0} {x''}^j

2684: e_j(z',y''),

2685: \ilabel{error3}\end{equation}

2686: where $e_j \in {x'}^{(n+1)/2-j} e^{i\lam/x'} \CI(X \times Y)$. Again we regard

2687: $y''$ as a parameter. Thus we have

2688: \begin{equation}

2689: (H - \lam^2) g_j = e_j  

2690: \ilabel{gj}\end{equation}

2691: 

2692: Consider the problem of solving away errors of the

2693: form $e_j$, to infinite order at bf (of course we cannot solve the errors

2694: away exactly without begin able 

2695: to solve $(H - \lam^2) u = f$ exactly, which we cannot do until we have

2696: constructed the resolvent kernel!). 

2697: If we apply $H - \lam^2$ to a series of the form 

2698: \begin{equation}

2699: e^{i\lam/x} x^{(n-1)/2 - k}\sum_{j \geq 0} x^j b_j, \quad b_j \in \CI(X),

2700: \ilabel{series1}\end{equation}

2701: we get a series of the form

2702: \begin{equation} 

2703: e^{i\lam/x} x^{(n+1)/2 - k} \sum_{j \geq 0} x^j c_j, \quad c_j \in \CI(X),

2704: \quad c_0 = 2i\lam k b_0.

2705: \ilabel{series2}\end{equation}

2706: Thus, we can add to $g_j$ a series of the form \eqref{series1} to solve away

2707: the powers greater than $(n+1)/2$, but the power $(n+1)/2$ presents a

2708: problem (without introducing logarithmic terms), because of the vanishing of

2709: $k$ in \eqref{series2}, unless the 

2710: coefficient of the $x^{(n+1)/2}$ term happens to be zero. We need the

2711: following results.  

2712: 

2713:  

2714: 

2715: \begin{lemma}\label{lemma:one-order-improvement}

2716: If $g \in  e^{i\lam/x} x^{(n-1)/2 - k} \CI(X)$, $k = 1, 2, \dots$, satisfies 

2717: $$

2718: (H - \lam^2) g = x^{(n+1)/2} e^{i\lam/x} \CI,

2719: $$

2720: then 

2721: $$

2722: (H - \lam^2) g = x^{(n+3)/2} e^{i\lam/x} \CI.

2723: $$

2724: \end{lemma}

2725: 

2726: \begin{proof} It follows inductively using \eqref{series2} that the

2727: coefficient of order $(n+1)/2 - l$ vanishes, for $l = k, k-1,

2728: \dots$. Thus, actually $g \in e^{i\lam/x} x^{(n-1)/2} \CI(X)$. Then

2729: \eqref{series2} 

2730: shows that the next coefficient also vanishes. 

2731: \end{proof}

2732: 

2733: \begin{cor} The same result holds if the condition $g \in  e^{i\lam/x}

2734: x^{(n-1)/2 - k} 

2735: \CI(X)$ is replaced by $g \in I^{p, (n-1)/2 - n/4 - k}(K, \Gs)$ for any

2736: $p$ and any Legendre conic pair $(K, \Gs)$.

2737: \end{cor}

2738: 

2739: \begin{proof} Apply the above argument to the symbol at $\Gs$.

2740: \end{proof}

2741: 

2742: Thus, for each $j$, we can modify $g_j$ by a series of the form

2743: \eqref{series1} until the error term is of the form ${x'}^{(n+1)/2}

2744: e^{i\lam/x'} \CI(X \times Y)$. Then applying the Corollary to $g_j$,

2745: we find that unsolvable term of order ${x'}^{(n+1)/2}$ vanishes. 

2746: Therefore, we can solve away the $e_j$ to infinite order

2747: at bf. Thus, we may assume that our error in $E_3(\lam)$ vanishes to infinite

2748: order at the corner bf $\cap \rb$.  

2749: 

2750: 

2751: Next, we solve the error away at $\Ls$. This involves solving the transport

2752: equation 

2753: \begin{equation}

2754: i\lam \Big( \sigma \pa_{\sigma} + (\frac{1}{2} + j)

2755: \Big) a_j  = b_j.

2756: \end{equation}

2757: The equation for $a_0$ then is

2758: $$

2759: \Big( \sigma \pa_{\sigma} + \frac{1}{2} \Big) a_0 = b_0,

2760: $$

2761: and $b_0$ is rapidly decreasing at rb and is in $\sigma^{3/2} \CI(\Lsl)$ at

2762: lb. There is a unique solution which is rapidly decreasing at rb and in

2763: $\sigma^{-1/2} \CI$ at lb. We can thus find a correction term which reduces

2764: the error to $I^{(n+2)/2; (n+3)/2, (n-1)/2}(\Lsl)$, 

2765: with infinite order vanishing at bf $\cap \rb$. 

2766: Inductively, assume that we have reduced the error to $I^{n/2 + k; (n+3)/2,

2767: (n-1)/2}(\Lsl)$, with infinite order vanishing at bf $\cap \rb$. 

2768: The transport equation for $a_k$ is then

2769: $$

2770: \Big( \sigma \pa_{\sigma} + \frac{1}{2} + k \Big) a_k = b_k,

2771: $$

2772: where inductively, $b_k$ is rapidly decreasing at rb and is in $\sigma^{1/2 - k}

2773: \CI(\Lsl)$ at lb. There is a unique solution rapidly decreasing at

2774: rb and in $\sigma^{-1/2 + k} \CI(\Lsl)$ at lb. A Legendre distribution in 

2775: $I^{n/2 + k - 1, (n-1)/2, (n-1)/2}(\Lsl)$ with $a_j$ as symbol then reduces

2776: the error to $I^{n/2 + k + 1; (n+3)/2,

2777: (n-1)/2}(\Lsl)$, with infinite order vanishing at bf $\cap \rb$, so this

2778: completes the inductive step. Taking an asymptotic sum of such correction

2779: terms yields a parametrix $G_4(\lam) = G(\lam)$ leaving an error which is the

2780: sum of a term supported away from rb 

2781: of the form 

2782: $$

2783: e^{i\lam/x'} {x'}^{(n+3)/2} a(y', z'')

2784: $$

2785: with $a$ smooth and rapidly decreasing at bf, plus a term supported away

2786: from lb of the form

2787: $$

2788: e^{i\lam/x''} {x''}^{(n-1)/2} b(y'', z')

2789: $$

2790: at rb, with $b$ rapidly decreasing at bf. The error at lb can be solved away

2791: using \eqref{series1}-\eqref{series2}, leaving an error term $E_4(\lam)$

2792: which can be expressed on the

2793: blown-down space $X^2$ as 

2794: $$

2795: E_+(\lam) = e^{i\lam/x''} {x''}^{(n-1)/2} b(z', z'')

2796: $$

2797: with $b$ smooth on $X^2$ and rapidly decreasing at $x' = 0$. Such an error

2798: term is compact on the weighted $L^2$ space $x^l L^2(X)$ for any $l > 1/2$

2799: (where $L^2$ is taken with respect to the metric density). 

2800: This completes the construction of the parametrix. 

2801: 

2802: 

2803: 

2804: \section{Long range case}

2805: The case of long range metrics and long range perturbations, 

2806: $P\in x\Diffsc^{1}(X)$, requires only

2807: minor modifications in the parametrix construction until the last step

2808: (removing the outgoing error). In particular,

2809: there is no change in the construction of the pseudodifferential

2810: approximation.

2811: In the intersecting Legendrian construction, as well as solving the

2812: transport equations on $L^\circ(\lambda)$, the only difference is in the

2813: structure of the subprincipal symbol, which no longer obeys

2814: Lemma~\ref{lem:sub-pr}. Thus, the arguments in section~\ref{smoothness} and section~\ref{outgoing-error} have

2815: to be modified. Let $q$ denote the boundary subprincipal symbol of

2816: $H$. Notice that in the gravitational long range case, $q$ is a constant,

2817: but in the general long range case, $q$ is an arbitrary smooth function of $y$

2818: which is a quadratic on each fibre of $K$ over $Y$. 

2819: Let $q_l$ and $q_r$ denote the lift of $q$ to $\XXb$ via the left,

2820: respectively right, projection. 

2821: 

2822: Let us now discuss the necessary modifications to sections~\ref{smoothness}

2823: and \ref{outgoing-error}. Equation \eqref{left-eqn-0} becomes

2824: \begin{equation} 

2825: -i\big({\mathcal L}_{V_l} 

2826: - n \tau' +iq_l\big)  a_0 |\frac{d\sigma}{\sigma} dy'' d\mu''|^\half = 0,

2827: \end{equation}

2828: near $T_+(\lam)$. Thus,

2829: \eqref{eq-Vl} is replaced by

2830: \begin{equation}

2831: -i\tau' \big( \pa_\sigma  + f \big) (\sigma^{-n/2

2832: -i\frac{q_l}{2\lambda}}a_0) = 0,

2833: \quad f \in \CI(\Lpl).

2834: \end{equation}

2835: Thus, now we conclude that in this region

2836: $a_0$ is of the form

2837: $$

2838: \sigma^{\frac{n}{2}+i\frac{q_l}{2\lambda}}\CI(\Lpl).

2839: $$

2840: 

2841: Next, in the second region, at the corner $\lf\cap L^\sharp(\lam)$,

2842: the right transport equation \eqref{eq-Vr} becomes

2843: \begin{equation}

2844: \big( V_r + \tau'' +iq_r+ s f' \big) a_0 = 0

2845: \quad f' \in \CI(\Lplh).

2846: \end{equation}

2847: Adding this to the left transport equation yields

2848: $$

2849: 2\lam \big( s \pa_s - \frac{n-1}{2}-i\frac{q_l}{2\lam}-i\frac{q_r}{2\lam}

2850:  + sW + s\tilde f \big) a_0 = 0,

2851: $$

2852: which now gives that $a_0$ is of the form

2853: $$

2854: \sigma^{\frac{n}{2}+i\frac{q_l}{2\lambda}}

2855: s^{\frac{n-1}{2}+i\frac{q_l + q_r}

2856: {2\lambda}}\CI(\Lplh).

2857: $$

2858: 

2859: Combining this with the other similar results at $\rf$ and the interior

2860: of $L^\sharp(\lam)$, we deduce that

2861: $$

2862: a_0\in\rho_{\lb}^{n/2+i\frac{q_l}{2\lambda}}

2863: \rho_{\rb}^{n/2+i\frac{q_r}{2\lambda}}

2864: \rho_{\#}^{(n-1)/2+i\frac{q_l + q_r}{2\lambda}}

2865: \CI(\Lplh;

2866: \Omega_b^{1/2} \otimes S^{[-1/2]}).

2867: $$ 

2868: 

2869: In the general long range case, 

2870: the dependence of $q$ on $y$, and its appearance 

2871: in the exponent of the boundary defining functions

2872: $\rho_{\lb}$, etc.,

2873: means that differential operators from the left

2874: factor, acting on a Legendre function with principal symbol $a_0$,

2875: introduce logarithmic terms. For example, in a neighborhood

2876: of $\lb$ in $\Lpl$ the error term $b_k$

2877: in \eqref{eq-Vl-k} for $k=1$ will

2878: take the form

2879: $$

2880: b_1=\sigma^{\frac{n}{2}+i\frac{q_l}{2\lambda}}

2881: ((\log\sigma)^2 c_2+\log\sigma\, c_1+c_0),\quad c_j\in\Cinf(\Lpl),\ j=0,1,2.

2882: $$

2883: Then the transport equation for $a_1$ takes the form

2884: \begin{equation} 

2885: \Big( -i\big({\mathcal L}_{V_l} 

2886: +(- n+2) \tau' +iq_l\big) +

2887: \sigma f \Big) a_1 |\frac{d\sigma}{\sigma} dy'' d\mu''|^\half = 0, \ f \in

2888: \CI(\Lpl),

2889: \end{equation}

2890: \begin{equation}

2891: -i\tau' \big( \sigma \pa_\sigma  -\frac{n}{2}+1 -iq_l/(2\lambda)

2892: + \sigma f \big) a_1 = b_1,

2893: \quad f \in \CI(\Lpl).

2894: \end{equation}

2895: Hence, near $\lb$ but away from $L^\sharp(\lam)$,

2896: $a_1$ will take the form

2897: $$

2898: a_1=\sigma^{\frac{n}{2}-1+i\frac{q_l}{2\lambda}}

2899: ((\log\sigma)^2 c'_2+\log\sigma\, c'_1+c'_0),

2900: \quad c'_j\in\Cinf(\Lpl).

2901: $$

2902: 

2903: A similar discussion works at the other boundary faces of $\Lh^+(\lam)$,

2904: with up to quadratic factors in each of $\log \rho_{\lb}$,

2905: $\log \rho_{\rb}$, $\log_{\rho_\sharp}$, and can be repeated (with

2906: progressively higher powers of logarithms) for all $a_k$'s.

2907: 

2908: Since the most important long-range case is the gravitational case

2909: where the subprincipal symbol is constant, 

2910: and since it makes the discussion more transparent,

2911: in what follows we make the assumption that

2912: $$

2913: \Lap \text{ and } P \ \text{are of long range gravitational type},

2914: $$

2915: which implies that $q$ is constant. Let

2916: $$

2917: \alpha = \frac{q}{2\lam}.

2918: $$

2919: The point is that in this case the powers of $\rho_{\lb}$, etc., above

2920: are constant, thus no logarithmic factors arise when we apply $H-\lam^2$

2921: to such Legendre functions.

2922: Then

2923: $$

2924: a_k\in\rho_{\lb}^{n/2+i\alpha}

2925: \rho_{\rb}^{n/2+i\alpha-k}

2926: \rho_{\#}^{(n-1)/2+2i\alpha-k}

2927: \CI(\Lplh;

2928: \Omega_b^{1/2} \otimes S^{[-1/2]}),

2929: $$

2930: and asymptotic summation gives an outgoing error

2931: $$

2932: E_+(\lam) \in e^{i\lam/x'}e^{i\lam/x''}{x''}^{(n-1)/2+i\alpha}

2933: {x'}^{(n+1)/2+i\alpha} \CI(\XXb;

2934: \scOh).

2935: $$

2936: Since $\alpha$ is a constant, \eqref{series1}-\eqref{series2} are

2937: still true ($k$ need

2938: not be an integer; it suffices that it is a constant), except that

2939: now

2940: $$

2941: c_0=(2i\lam k+2\lam\alpha)b_0 .

2942: $$

2943: Since now we are taking $k=l+i\alpha$,

2944: where $l$ is an integer, we can solve away the series, provided that

2945: the coefficient of the $l=0$ term vanishes, which is assured just as in

2946: Lemma~\ref{lemma:one-order-improvement}. The rest of the argument

2947: requires only similar modifications as compared to the short-range case,

2948: so we conclude, as there, that we can modify the parametrix to obtain

2949: an error term of the form

2950: $$

2951: E_4(\lam)=e^{i\lambda/x''}(x'')^{(n-1)/2+i\alpha}b(z',z'')

2952: $$

2953: with $b$ smooth on $X^2$ and rapidly decreasing at $x'=0$.

2954: 

2955: \begin{comment}

2956: The case of general long-range $P\in\Diffsc^{1,1}(X)$ requires only

2957: simple, but rather tedious, modifications in this argument by forcing

2958: one to follow the logarithmic factors, and do iterative arguments

2959: in decreasing powers of the logarithms as well.

2960: \end{comment}

2961: 

2962: \begin{comment}

2963: To describe the structure in general, we introduce the space of polyhomogeneous

2964: conormal functions (half-densities)

2965: on $\Lh^+(\lam)$ corresponding to the index set

2966: $$

2967: \calK_k=\{(m,p):\ m,p\in\Nat,\ p\leq 2\min\{m,k\}\}

2968: $$

2969: on every boundary hypersurface of $\Lh^+(\lam)$. Then

2970: $c\in\bconc^{\calK_k}(\Lh^+(\lam))$ means

2971: that at $\lb$, $c$ takes the form

2972: $$

2973: \sum_{m=0}^\infty \sum_{p=0}^{2\min\{m,k\}}\rho_{\lb}^m(\log\rho_{\lb})^p

2974: c_{m,p},

2975: $$

2976: where $c_{m,p}$ is $\Cinf$ away from the corner of $\hat \Lpl$, and

2977: has a similar expansion there.

2978: Then an argument similar to the one given above gives

2979: $$

2980: a_k\in\rho_{\lb}^{n/2-k+i\frac{\sigma_{1,1}(P_l)}{2\lambda}}

2981: \rho_{\rb}^{n/2-k+i\frac{\sigma_{1,1}(P_r)}{2\lambda}}

2982: \rho_{\#}^{(n-1)/2-k+i\frac{\sigma_{1,1}(P_l)+\sigma_{1,1}(P_r)}{2\lambda}}

2983: \bconc^{\calK_k}(\Lplh;

2984: \Omega_b^{1/2} \otimes S^{[-1/2]}).

2985: $$ 

2986: ****It seems that this is not optimal at the corners, since we have

2987: $(\rho_{\lb}\rho_{\#})^{i\sigma_{1,1}(P_l)/(2\lambda)}$, so the

2988: differentiations in $y'$ only give rise to $(\log \rho_{\lb}+\log\rho_{\#})^2$

2989: for $k=1$, i.e.\ product terms such as $(\log\rho_{\lb})^2(\log\rho_{\#})^2$

2990: do not arise.****

2991: 

2992: Asymptotically summing the correction terms with principal symbols

2993: $a_k$ now yields an error of the form

2994: $$

2995: E_+(\lam)\in

2996: e^{i\lam/x'}e^{i\lam/x''}{x''}^{\frac{n-1}{2}+i\frac{\sigma_{1,1}(P_r)}{2\lambda}}

2997: {x'}^{\frac{n+1}{2}+i\frac{\sigma_{1,1}(P_l)}{2\lambda}} \bconc^\calK(\XXb;

2998: \scOh),

2999: $$

3000: where the our index family assigns the index set

3001: $$

3002: \calK=\{(m,p):\ m,p\in\Nat,\ p\leq 2m\}

3003: $$

3004: to every boundary hypersurface. (Thus, the maximum power of $\log$ is

3005: finite for each $a_k$ above, but the asymptotic summation forces us to allow

3006: arbirarily high powers of $\log$.)

3007: \end{comment}

3008: 

3009: 

3010: \section{Resolvent from parametrix}

3011: 

3012: In the previous two sections, we constructed a parametrix $G(\lam)$ for $\Rt(\lam)$ 

3013: which satisfies

3014: $$

3015: (H_l - \lam^2) G(\lam) = K_{\Id} + E(\lam),

3016: $$

3017: where $E(\lam)$ has a kernel which is of the form $e^{i\lam/x''} {x''}^{(n-1)/2} 

3018: {x'}^\infty \CI(\XXb; \scOh)$. Thus, it is a Hilbert-Schmidt kernel on

3019: $x^l L^2(X)$ for every $l > 1/2$, and in particular is compact. In fact,

3020: we also see directly from its form that $E(\lam):x^l L^2(X)\to\dCinf(X)$

3021: for $l>1/2$. Crude

3022: estimates (such as Schur's Lemma)

3023: show that $G(\lam)$ acts as a bounded operator from $x^l L^2$ to $x^{-l} L^2$

3024: for large enough $l>1/2$;

3025: more refined estimates, which we do not need here,

3026: show that in fact this is true for any $l >

3027: 1/2$. Thus, the equation above becomes an operator equation

3028: $$

3029: (H - \lam^2) G(\lam) = \Id + E(\lam)

3030: $$

3031: from $x^l L^2$ to $x^{-l} L^2$. 

3032: 

3033: \subsection{Finite rank perturbation}

3034: To correct $G(\lam)$ to the actual $\Rt(\lam)$,

3035: we must solve away the error term $E(\lam)$. Thus, we would like $\Id + E(\lam)$

3036: to be invertible. However, this is certainly not

3037: necessarily the case as things stand; if for example we modified $G(\lam)$ 

3038: by subtracting from it the rank one operator $G(\lam)(\phi) \langle \phi,

3039: \cdot \rangle$, for some $\phi \in \CIdot(X; \scOh)$, then the modified $G(\lam)$

3040: would be microlocally indistinguishable from the old one, but would

3041: annihilate $\phi$, so the modified $\Id + E(\lam)$ would not be invertible.  

3042: 

3043: Since $\Id + E(\lam)$ is compact, it has a null space and cokernel of the same

3044: finite dimension $N$. 

3045: To make $\Id + E(\lam)$ invertible, we try to correct $G(\lam)$ by adding to it a

3046: finite rank term 

3047: \begin{equation}

3048: \sum_{i=1}^N \phi_i \langle x^{2l} \psi_i, \cdot \rangle .

3049: \ilabel{finiterank}\end{equation}

3050: Here $\langle, \ \rangle$ denotes $L^2$ pairing, $\psi_i$ should lie in

3051: $L^2$,

3052: and the factor of $x^{2l}$ is included to ensure that it acts

3053: on $x^l L^2$. We require that $\phi_i$ are in $x^{-l} L^2$ so that

3054: \eqref{finiterank} is bounded from $x^{l} L^2 \to x^{-l} L^2$. We wish to

3055: choose $\phi_i$ and $\psi_i$ so that 

3056: $$

3057: \Id + E(\lam) + \sum_{i=1}^N ((H - \lam^2) \phi_i) \langle x^{2l} \psi_i, \cdot

3058: \rangle 

3059: $$

3060: is invertible. This is possible if we can choose $x^{l} \psi_i$ to span the

3061: null space of $\Id + E(\lam)$ and $(H - \lam ^2)\phi_i$ to span a subspace

3062: supplementary to 

3063: the range. Note that if $(\Id+E(\lam))u=0$ and $u\in x^l L^2$, then

3064: $u=-E(\lam)u$, so the mapping properties of $E(\lam)$ imply that

3065: $u\in\dCinf(X)$. Thus, we automatically have

3066: $\psi_i\in\dCinf(X)$ above. To proceed, we need the following lemma. 

3067: 

3068: \begin{lemma}\label{density}

3069: Let $l>1/2$. Then the image of $H - \lam^2$ on

3070: the sum of $\dCinf(X)$ and the range of $G(\lam)$ applied to $\dCinf(X)$

3071: is dense in $x^l L^2$.  

3072: \end{lemma}

3073: 

3074: \begin{rem}

3075: Note that $(H-\lam^2)G(\lam)g=(\Id+E(\lam))g\in\dCinf(X)$ if $g\in\dCinf(X)$,

3076: and for $u\in\dCinf(X)$, $(H-\lam^2)u\in\dCinf(X)$, so the image of

3077: $H-\lam^2$ on the space in the statement of the lemma is a subspace

3078: of $\dCinf(X)$.

3079: \end{rem}

3080: 

3081: \begin{proof} To proceed, we give the proof for short range $H$; the proof

3082: for long range $H$ requires only minor modifications. 

3083: 

3084: Let $\mathcal{M}$ be the subspace of $x^l L^2$ given by the 

3085: image of $H - \lam^2$ on

3086: the sum of $\dCinf(X)$ and the range of $G(\lam)$ applied to $\dCinf(X)$. 

3087: If $\mathcal{M}$ is not dense, then there is a function $f \in x^l L^2$

3088: orthogonal to 

3089: $\mathcal{M}$. Since for

3090: $u\in\dCinf(X)$ implies

3091: $(H-\lam^2)u\in\mathcal{M}$, $f$ satisfies

3092: \begin{equation}\begin{gathered}

3093: \lang x^{-l} f, x^{-l} (H - \lam^2) u \rang = 0 \quad \forall \ u \in

3094: \CIdot(X) \\

3095: \Rightarrow \lang (H - \lam^2 ) x^{-2l} f, u \rang = 0 \quad \forall \ u \in

3096: \CIdot(X) \\

3097: \Rightarrow (H - \lam^2) h = 0, \quad h = x^{-2l} f. 

3098: \end{gathered}\end{equation}

3099: where we used that $H$ is symmetric on $\dCinf(X)$.

3100: On the other hand, $G(\lam)$ maps $\dCinf(X) \to x^{(n-1)/2}

3101: e^{i\lam/x} \CI(X)$, and for any

3102: $g\in\dCinf(X)$, $(H-\lam^2)G(\lam)g=(\Id+E(\lam))g\in\dCinf(X)$, hence

3103: $(\Id+E(\lam))g\in\mathcal{M}$. In addition, $E(\lam)^*$, with kernel

3104: $E(\lam)^*(z',z'')

3105: =\overline{E(\lam,z'',z')}$, maps $\dist(X)\to x^{(n-1)/2}e^{-i\lambda/x}

3106: \CI(X)$,

3107: so we have

3108: \begin{equation}\begin{gathered}

3109: \lang x^{-l} f, x^{-l} (H - \lam^2) G(\lam) g \rang =

3110: \lang x^{-l} f, x^{-l} (\Id + E(\lam)) g \rang = 0 \quad \forall \ g \in

3111: \CIdot(X) \\

3112: \Rightarrow \lang (\Id + E(\lam)^* ) x^{-2l} f, g \rang = 0 \quad \forall \ g \in

3113: \CIdot(X) \\

3114: \Rightarrow (\Id + E(\lam)^*) h = 0. 

3115: \end{gathered}\end{equation}

3116: If $h = - E(\lam)^* h$, then $h$ has the form $x^{(n-1)/2} e^{-i\lam/x}

3117: \Cinf(X)$, ie it is incoming. A standard argument then implies that $h

3118: \equiv 0$. Let $h = x^{(n-1)/2} e^{-i\lam/x} h_0(y) + \tilde h$, where $\tilde

3119: h \in x^{(n+1)/2} e^{-i\lam/x} \CI(X)$. Green's formula yields

3120: \begin{equation}\begin{gathered}

3121: 0 = \int_X \overline h (H - \lam^2) h  - h (H - \lam^2) \overline h \\

3122: = 2i\lam \lim_{\ep \to 0} \int\limits_{\{ x = \ep \}} \overline h x^2 \pa_x h  -

3123: h x^2 

3124: \pa_x \overline h 

3125: = 2i\lam \int\limits_Y |h_0(y)|^2,

3126: \end{gathered}\end{equation}

3127: so $h_0 \equiv 0$. It then follows iteratively from \eqref{series1} and 

3128: \eqref{series2} that the expansion of $h$

3129: at the boundary of $X$ vanishes identically, that is, that $h \in

3130: \CIdot(X)$. Finally a unique continuation theorem, see

3131: e.g.\ \cite[Chapter~XVII]{Hor}, shows $h=0$ identically. 

3132: This means

3133: that $\mathcal{M}$ is indeed dense in $x^{l} L^2$.

3134: \end{proof}

3135: 

3136: Thus, we can choose the $\phi_i \in x^{(n-1)/2}

3137: e^{i\lam/x} \CI(X)$ above so that $(H - \lam^2) \phi_i\in\dCinf(X)$

3138: span a supplementary subspace of range $\Id + E(\lam)$.

3139: The modified parametrix then satisfies 

3140: $$

3141: (H  -\lam^2) G_5(\lam) = \Id + E_5(\lam),

3142: $$

3143: where $E_5(\lam)$ has the same form as $E(\lam)$ but in addition is 

3144: invertible on $x^l L^2$ for all $l > 1/2$. 

3145: 

3146: \subsection{Resolvent}

3147:  The inverse $\Id + S(\lam)$ of $\Id + E_5(\lam)$ is Hilbert-Schmidt on $x^l

3148:  L^2$ since this is true of $E_5(\lam)$. Moreover, since 

3149: $$

3150: S(\lam) = -E_5(\lam) + E_5(\lam)^2 + E_5(\lam) S(\lam) E_5(\lam),

3151: $$

3152: it is easy to see that also $S(\lam) \in e^{i\lam/x''} {x''}^{(n-1)/2} 

3153: {x'}^\infty \CI(\XXb; \scOh)$. Our solution for the kernel $\Rt(\lam)$ is

3154: then

3155: $$

3156: \Rt(\lam) = G_5(\lam) ( \Id + S(\lam)).

3157: $$

3158: It is not hard to show that $G_5(\lam) S(\lam)$ has the form

3159: $${x'}^{(n-1)/2}

3160: {x''}^{(n-1)/2} e^{i\lam/x'} e^{i\lam/x''} \CI(\XXb; \scOh),$$

3161: so $\Rt(\lam)$

3162: has the desired microlocal regularity \eqref{seek}. 

3163: 

3164: {\it Remark. } Lemma~\ref{density} directly shows the absence of positive

3165: eigenvalues of $H$. Suppose that $(H - \lam^2) u = 0$ and $u \in x^s H^2(X)$

3166: for some $s > -1/2$. This would certainly be the case of an eigenfunction

3167: since $H$ has elliptic interior symbol, so $u$ would lie in $H^k(X)$ for

3168: every $k$. We need to show that

3169: for all functions $g\in x^{(n-1)/2}e^{i\lam/x}

3170: \Cinf(X)$ the equation

3171: \begin{equation}

3172: \lang (H - \lam^2) u, g \rang = \lang u, (H - \lam^2) g \rang = 0

3173: \ilabel{ef-symmetry}

3174: \end{equation}

3175: holds. Indeed, this implies

3176: that $u$ is $L^2$-orthogonal to the image of $H - \lam^2$

3177: acting on

3178: $x^{(n-1)/2}e^{i\lam/x}

3179: \Cinf(X)$,

3180: or equivalently that $x^{2l} u \in x^l L^2$ is orthogonal in

3181: $x^l L^2$ to the image of $H - \lam^2$ acting on $x^{(n-1)/2}e^{i\lam/x}

3182: \Cinf(X)$ in $x^l L^2$. Then

3183: Lemma~\ref{density} shows that $u \equiv 0$.

3184: 

3185: To deduce \eqref{ef-symmetry} for $g\in x^{(n-1)/2}e^{i\lam/x}

3186: \Cinf(X)$,

3187: let $\phi\in\Cinf(\RR)$, identically $1$ on $[1,\infty)$, $0$ near the

3188: origin.

3189: Then

3190: \begin{equation}

3191: \begin{split}

3192: \lang (H-\lam^2)u, g\rang&=\lim_{t\to 0}\lang (H-\lam^2)u, \phi(x/t)g\rang

3193: =\lim_{t\to 0} \lang u,(H-\lam^2)\phi(x/t)g\rang\\

3194: &=\lim_{t\to 0} \lang u,\phi(x/t)(H-\lam^2)g\rang

3195: +\lim_{t\to 0} \lang u, [H,\phi(x/t)]g\rang\\

3196: &=\lang u,(H-\lam^2)g\rang+\lim_{t\to 0} \lang u, [H,\phi(x/t)]g\rang.

3197: \end{split}\end{equation}

3198: Note that $[H,\phi(x/t)]$ is uniformly bounded (i.e.\ with bounds

3199: independent of $t$)

3200: as a map $x^l H^1\to x^{l+1}L^2$, and in fact $[H,\phi(x/t)]\to 0$ strongly

3201: as $t\to 0$. Applying this with $l=-1/2-\ep$, $\ep>0$ sufficiently small,

3202: we see that the last term goes to $0$ as $t\to 0$, proving

3203: \eqref{ef-symmetry}.

3204: 

3205: 

3206: \subsection{Analytic continuation}

3207: It is not hard to show that the kernel $\Rt(\lam)$ constructed

3208: above continues analytically (as a distribution on $\XXsc$)

3209: into $\Imag \lam \geq 0, \Rea \lam > 0$. We

3210: complete the proof of Theorem~\ref{main-result} by showing that this 

3211: analytic continuation coincides with the outgoing resolvent, $R(\lam^2)$,

3212: for $\Imag \lam > 0$. 

3213: 

3214: Our parametrix is defined as an asymptotic sum of

3215: symbols, which is really a sum with cutoff functions inserted

3216: (see \cite[Proposition~18.1.3]{Hor} for an explicit construction). The

3217: cutoffs

3218: depend on $C^k$ norms of a finite number of symbols and ensure that the

3219: sum

3220: converges in $C^k$ for all $k$. If the symbols are

3221: holomorphic in $\lam$

3222: then the $C^k$ norms may be taken uniform on compact subsets of

3223: $\lam$. Since holomorphy is preserved under uniform limits, we need only

3224: check that the phase and symbols analytically continue in some explicit

3225: parametrization of the Legendrians. 

3226: 

3227: It is standard that the pseudodifferential approximation $G_1(\lam)$

3228: analytically continues. Blowing down sf, we solve away

3229: the error as an intersecting Legendrian,

3230: see Section~\ref{sec:intersecting-Leg-constr}. 

3231: \begin{comment}As apparent from

3232: the explicit parameterization of $L^\circ(\lam)$, \eqref{eq:sp-1c},

3233: $L^\circ(\lam')$ is the image of $L^\circ(\lam)$ under dilation

3234: by $\lam'/\lam$ in the fibers of $\sct_{\ffb}\XXb$, hence there exist

3235: parameterizations with very simple (certainly holomorphic) behavior in

3236: $\lam$.

3237: \end{comment} 

3238: Let $\phi$ be a local parametrization of the Legendrian $L(1)$ near $L(1)

3239: \cap N^*\diag_\bl$. Then it is easily checked that the phase function

3240: $$

3241: k \phi + s(\lam - k)

3242: $$

3243: locally parametrizes $(N^*\diag_\bl, L^+(\lam))$. Since the variable $s$

3244: takes

3245: nonnegative values, the function $e^{i(k\phi + s(\lam - k))/x}$ 

3246: continues to $\Imag \lam > 0$. 

3247: 

3248: 

3249: Away from sf, the value of $\tau$ is strictly negative on the

3250: Legendrian, and so the phase is of the form

3251: $$

3252: e^{i\lam \phi/x}

3253: $$

3254: where $\phi$ is positive on the Legendrian, independent of $\lam$.

3255: By restricting the support of

3256: the symbol sufficiently, therefore, we may assume that $\phi$ is positive

3257: everywhere on the support of the integral. Thus this also analytically 

3258: continues to the upper half plane with uniform bounds.

3259: 

3260: The symbols are defined by iteratively solving transport equations

3261: of the form

3262: $$

3263: \Big( -i \mathcal{L}_{H_p} -i \big(\half + m - \frac{N}{4} \big) \frac{

3264: \pa

3265:   p}{\pa \tau} + p_{\sub} \Big) a_j=b_j,

3266: $$

3267: where $b_0=0$.

3268: These equations

3269: are solved along $L^\circ(\lam)$, i.e.\ if we consider amplitudes

3270: in an explicit parameterization of the Legendrian, then along

3271: the critical submanifold $C_\phi=\{(0,y,u):\ d_u\phi=0\}$, where

3272: $\lambda\phi/x$ is the phase function as above. Note that $C_\phi$

3273: is independent of $\lam$, and it is identified with $L^\circ(\lam)$

3274: via the map $C_\phi\ni(0,y,u)\mapsto(0,y,d_{(x,y)}(\lam\phi/x)|_{(0,y,u)})$.

3275: Along $C_\phi$ the transport equation

3276: becomes an ODE whose coefficients depend on $\lam$ polynomially, since the

3277: only $\lam$ dependence of the coefficients arises from this identification

3278: map,

3279: and $H_p$, $\frac{ \pa  p}{\pa \tau}$, $p_{\sub}$

3280: are polynomial in the fiber variables.

3281: Thus, the solution $a_j$ of the transport equation,

3282: as a function on $D\times C_\phi$, $D$ a neighborhood of the positive

3283: real axis in $\CC=\CC_\lam$, is

3284: holomorphic (in $\lam$), provided $b_j$ is (here we

3285: identify $C_\phi$ with $L^\circ(\lam)$). Note that

3286: the $b_j$'s arise because solving the transport equations only

3287: guarantees that the `error term' $E_3(\lam)$, arising from the application

3288: of $H-\lam^2$ to $R_3(\lam)$, is one order lower than expected, so for each

3289: $\lam$,

3290: $b_j|_{C_\phi}$

3291: depends on $a_i$, $i<j$ near $C_\phi$, and not just on $a_i|_{C_\phi}$.

3292: (In fact, $b_j|_{C_\phi}$ depends on a finite number of terms of the

3293: Taylor series of $a_i$, $i<j$, at $C_\phi$.)

3294: To ensure that the

3295: $b_j$ are holomorphic in $\lam$, we define the $a_i$ {\em near} $D\times

3296: C_\phi$,

3297: rather than {\em at} $D\times C_\phi$, e.g. by introducing a local product

3298: decomposition $C_\phi\times U$, $U\subset\RR^k$, of the parameter space

3299: near $C_\phi$, and pull-back the $a_i$, first defined on $D\times C_\phi$,

3300: by the projection. Then, having constructed $a_i$, $i<j$, $b_j$ will

3301: be holomorphic in $\lam$ near, hence at, $D\times C_\phi$, so $a_j$ is also

3302: holomorphic at $D\times C_\phi$, hence it extends to be holomorphic near

3303: $D\times C_\phi$. If we express the amplitudes $a_j$ with respect to

3304: a different parameterization of $L^\circ(\lam)$, which

3305: is still of the form $\lambda\phit/x$, then the new amplitudes $\at_j$

3306: will

3307: still be holomorphic functions of $\lam$, so holomorphy is preserved

3308: at the overlap of parameterizations of different parts of $L^\circ(\lam)$.

3309: This completes the inductive argument.

3310: 

3311: 

3312: Therefore, our parametrix constructed above may be assumed holomorphic in

3313: some set $B(\ep, \lam_0) \cap \{ \Imag \lam \geq 0 \}$, for some $\lam_0 > 0$.

3314: It

3315: is

3316: easy to see that for non-real $\lam$, the parametrix is in the small

3317: calculus, since the positivity of $\phi$ implies that the exponent

3318: of $e^{i\lam\phi/x}$ has negative real part, and is therefore rapidly

3319: decreasing at bf, lb and rb. 

3320: The finite rank correction may be taken independent of $\lam$ if we

3321: chose $\ep>0$ sufficiently small. Then, 

3322: we have

3323: $$

3324: (\Lap  -\lam^2) G_5(\lam) = \Id + E_5(\lam), \quad \Imag \lam \geq 0, \ \Rea \lam > 0,

3325: $$

3326: where all terms are holomorphic in some small open set as above, 

3327: $E_5(\lam)$ is invertible on $x^l

3328: L^2$ for all $l > 1/2$, and off the real axis, $G_5(\lam)$ and $E_5(\lam)$ are in

3329: the small calculus. Define $\Id + S(\lam)$ to be the inverse of $\Id +

3330: E_5(\lam)$ on $x^l L^2$ for some fixed $l$. By the symbolic functional

3331: calculus \cite{HV1}, for $\Imag \lam > 0$,

3332: $S$ is a family of scattering pseudodifferential operators which is

3333: clearly

3334: holomorphic. Then $\Rt(\lam) \equiv G_5(\lam)(\Id + S(\lam))$

3335: satisfies $(H - \lam^2)\Rt(\lam) = \Id$ on $x^l L^2$. But by

3336: self-adjointness,

3337: and the symbolic functional calculus, 

3338: for $\Imag \lam > 0$, $(H - \lam^2)$ has a pseudodifferential inverse on

3339: $L^2$. Since $R(\lam)$ is a bounded operator on $L^2$ for $\Imag \lam > 0$ it must

3340: be the 

3341: inverse. Therefore we have shown the inverse on the real axis constructed

3342: above continues as a Schwartz kernel to the upper half plane and agrees

3343: with the resolvent there. This completes the proof of

3344: Theorem~\ref{main-result}. 

3345: 

3346: \begin{rem}

3347: The only place where we used that $\Imag \lam\geq 0$ is to make our

3348: parametrix act on, and its error compact on, weighted Sobolev

3349: spaces. Namely, in the last step

3350: of the construction, i.e.\ when we add a finite rank perturbation to

3351: remove the error $E(\lam)$, we need $E(\lam)$ to be a compact

3352: operator on $x^l L^2$ for $l>1/2$. However, the kernel of $E(\lam)$ is of

3353: the form

3354: $e^{i\lam/x''} {x''}^{(n-1)/2} 

3355: {x'}^\infty \CI(\XXb; \scOh)$, and for $\Imag \lam<0$ the real part of the

3356: exponent is positive, so the kernel of $E(\lam)$ is not even a tempered

3357: distribution on $\XXb$. In particular, it does not even map $\dCinf(X)$

3358: to $\dist(X)$. The same statement holds for $G(\lam)$ as well.

3359: \end{rem}

3360: 

3361: 

3362: 

3363: 

3364: 

3365: \bibliographystyle{plain}

3366: \bibliography{sm}

3367: \end{document}

3368: 

3369: 

3370: 

3371: 

3372: 

3373: 

3374: 

3375: 

3376: 

3377: 

3378: 

3379: 

3380: 

3381: 

3382: 

3383: 

3384: 

3385: 

3386: