1: \documentclass[12pt]{article}

2: 

3: \pagestyle{headings}

4: 

5: \usepackage{multicol}

6: 

7: \usepackage{amssymb}

8: \usepackage[leqno]{amsmath}

9: \usepackage{amsthm}

10: 

11: \usepackage{graphicx,color}

12: 

13: \usepackage{eso-pic}

14: \definecolor{lightgray}{gray}{.85}

15: \usepackage{subfigure}

16: %\usepackage{showlabels}

17: 

18: \usepackage{xr-hyper}

19: \usepackage[bookmarks,pdfnewwindow]{hyperref}

20: 

21: \hypersetup{

22: colorlinks=true,

23: linkcolor=black,

24: citecolor=black,

25: pdfauthor={Victor Ivrii},

26: pdftitle={Sharp Spectral Asymptotics for 2-dimensional Schr\"odinger operator with a strong but degenerating magnetic field. II},

27: pdfsubject={Sharp Spectral Asymptotics},

28: pdfkeywords={Microlocal Analysis, Magnetic Schr\"odinger Operator},

29: baseurl={http://www.math.toronto.edu/ivrii/Research/preprints/},

30: }

31: 

32: \externaldocument[IRO6-]{IRO6}[http://www.math.toronto.edu/ivrii/Research/preprints/IRO6.pdf]

33: 

34: %%%%%%%%%

35: %%	romans

36: 

37: %\newcommand\A{{\rm A}}

38: %\newcommand\B{{\rm B}}

39: %\newcommand\C{{\rm C}}

40: %\newcommand\D{{\rm D}}

41: %\newcommand\E{{\rm E}}

42: %\newcommand\F{{\rm F}}

43: %\newcommand\G{{\rm G}}

44: %\newcommand\H{{\rm H}}

45: %\newcommand\I{{\rm I}}

46: %\newcommand\J{{\rm J}}

47: %\newcommand\K{{\rm K}}

48: %\newcommand\L{{\rm L}}

49: %\newcommand\M{{\rm M}}

50: %\newcommand\N{{\rm N}}

51: %\newcommand\O{{\rm O}}

52: %\newcommand\P{{\rm P}}

53: %\newcommand\Q{{\rm Q}}

54: %\newcommand\R{{\rm R}}

55: %\newcommand\S{{\rm S}}

56: %\newcommand\TT{{\rm T}}

57: %\newcommand\U{{\rm U}}

58: %\newcommand\V{{\rm V}}

59: \newcommand\W{{\rm W}}

60: \newcommand\MW{{\rm {MW}}}

61: %\newcommand\X{{\rm X}}

62: %\newcommand\Y{{\rm Y}}

63: %\newcommand\Z{{\rm Z}}

64: \newcommand\Def{{\overset {\rm {def}}{\ =\ }}}

65: \renewcommand\And{{\rm {and\;\;}}}

66: \newcommand\as{{\rm {as}\;\;}}

67: \newcommand\at{{\rm {at}\;}}

68: \newcommand\out{{\rm {out}}}

69: \newcommand\inn{{\rm {inn}}}

70: \newcommand\per{{\rm {per}}}

71: 

72: \newcommand\const{{\rm {const}}}

73: \newcommand\dist{{\rm {dist}}}

74: \newcommand\eff{{\rm {eff}}}

75: %\newcommand\either{{\rm {either}}}

76: %\newcommand\ess{{\rm {ess}}}

77: %\newcommand\even{{\rm {even}}}

78: %\newcommand\sc{{\rm {sc}}}

79: \newcommand\for{{\rm {for}\;}}

80: %\newcommand\h{{\rm h}}

81: \newcommand\In{{\rm {in\;}}}

82: %\newcommand\If{{\rm {if\;}}}

83: %\newcommand\inter{{\rm {int}}}

84: %\newcommand\loc{{\rm {loc}}}

85: \newcommand\new{{\rm {new}}}

86: %\newcommand\odd{{\rm {odd}}}

87: %\newcommand\old{{\rm {old}}}

88: \newcommand\Or{{\rm {or}}}

89: %\newcommand\on{{\rm {on\;}}}

90: %\newcommand\TF{{\rm {TF}}}

91: %\newcommand\SS{{\rm {SS}}}

92: %\newcommand\HS{{\rm {HS}}}

93: %\newcommand\sc{{\rm {sc}}}

94: %\newcommand\app{{\rm {app}}}

95: %\newcommand\el{{\rm {el}}}

96: \newcommand\corr{{\rm {corr}}}

97: %\newcommand\provided{{\rm {provided\;}}}

98: %\newcommand\reg{{\rm {reg}}}

99: \newcommand\st{{\rm {st}}}

100: %\newcommand\with{{\rm {with}\;}}

101: \newcommand\w{{\rm w}}

102: %\newcommand\x{{\rm x}}

103: %\newcommand\x{{\rm x}}

104: %\newcommand\z{{\rm z}}

105: %\newcommand\OF{{\rm {OF}}}

106: %\newcommand\WF{{\rm {WF}}}

107: %\newcommand\Weyl{{\underline {\rm {Weyl}}}}

108: 

109: %%	barred

110: %%%%%%%%%

111: 

112: \newcommand\balpha{{\bar \alpha}}

113: \newcommand\bsigma{{\bar {\mathstrut \sigma}}}

114: \newcommand\bl{{\bar {\mathstrut l}}}

115: \newcommand\bx{{\bar {\mathstrut x}}}

116: \newcommand\by{{\bar {\mathstrut y}}}

117: %\newcommand\bt{{\bar {\mathstrut t}}}

118: %\newcommand\btau{\bar {\mathstrut \tau }}

119: \newcommand\bz{{\bar {\mathstrut z}}}

120: %\newcommand\bgamma{\bar {\mathstrut \gamma }}

121: \newcommand\bareta{{\bar {\mathstrut \eta }}}

122: \newcommand\bxi{\bar {\mathstrut \xi}}

123: %\newcommand\btau{\bar {\mathstrut \tau }}

124: %\newcommand\bep{\bar {\mathstrut \epsilon }}

125: %\newcommand\bgamma{\bar {\mathstrut \gamma }}

126: %\newcommand\brho{\bar {\mathstrut \rho }}

127: 

128: %%	boldsymbols

129: %%%%%%%%%

130: 

131: %\newcommand\boldalpha{{\boldsymbol \alpha }}

132: %\newcommand\boldbeta {{\boldsymbol \nu }}

133: %\newcommand\boldrho{{\boldsymbol \rho }}

134: %\newcommand\boldpi{{\boldsymbol \pi }}

135: %\newcommand\boldsigma{{\boldsymbol \sigma }}

136: %\newcommand\boldgamma{{\boldsymbol \gamma }}

137: %\newcommand\boldPhi{{\bold\Phi }}

138: %\newcommand\boldPsi{{\bold\Psi }}

139: \newcommand\blangle{{\boldsymbol \langle }}

140: \newcommand\brangle{{\boldsymbol \rangle }}

141: \newcommand\bv{{\boldsymbol | }}

142: 

143: %%	operatornames

144: %%%%%%%%%

145: 

146: \newcommand\Ad{\operatorname{Ad}}

147: %\newcommand\Char{\operatorname{Char}}

148: \newcommand\codim{\operatorname{codim}}

149: %\newcommand\deg{\operatorname{deg}}

150: \newcommand\diag{\operatorname{diag}}

151: %\newcommand\diam{\operatorname{diam}}

152: %\newcommand\dim{\operatorname{dim}}

153: %\newcommand\graph{\operatorname{graph}}

154: %

155: \newcommand\Hess{\operatorname{Hess}}

156: %\newcommand\Hom{\operatorname{Hom}}

157: \renewcommand\Im{\operatorname{Im}}

158: %\newcommand\ind{\operatorname{ind}}

159: %\newcommand\indM{\operatorname{ind_\M}}

160: \newcommand\Ker{\operatorname{Ker}}

161: \newcommand\Ran{\operatorname{Ran}}

162: %\newcommand\mes{\operatorname{mes}}

163: %\newcommand\mod{\operatorname{mod}}

164: %\newcommand\Op{\operatorname{Op}}

165: %\newcommand\Opw{\operatorname{Op^\w}}

166: %\newcommand\Ran{\operatorname{Ran}}

167: \newcommand\rank{\operatorname{rank}}

168: \renewcommand\Re{\operatorname{Re}}

169: %\newcommand\sgn{\operatorname{sgn}}

170: \newcommand\sign{\operatorname{sign}}

171: \newcommand\Spec{\operatorname{Spec}}

172: %\newcommand\Specess{\operatorname{Spec_\ess}}

173: \newcommand\supp{\operatorname{supp}}

174: %\newcommand\symb{\operatorname{symb}}

175: \newcommand\tr{\operatorname{tr}}

176: \newcommand\Tr{\operatorname{Tr}}

177: %\newcommand\vrai{\operatorname{vrai}}

178: \newcommand\Res{\operatorname{Res}}

179: \newcommand\ResR{\operatorname{Res_\bR}}

180: %\newcommand\Vol{\operatorname{vol}}

181: %\newcommand\Voll_#1{\operatorname{vol_#1}}

182: %\newcommand\Trr_#1{\operatorname{Tr_#1}}

183: %\newcommand\t_#1{\operatorname{tr_#1}}

184: %\newcommand\TTrr^#1{\operatorname{Tr^#1}}

185: %\newcommand\t^#1{\operatorname{tr^#1}}

186: %\newcommand\Hes_#1{\operatorname{Hess_{#1}}}

187: %\newcommand\Hessigma {\operatorname{Hess ^\sigma }}

188: 

189: %%	special symbols

190: %%%%%%%%%

191: 

192: %\newcommand\bi{{\bar \imath}}

193: %\newcommand\bj{{\bar\jmath}}

194: %\newcommand\hi{{\hat\imath}}

195: %\newcommand\hj{{\hat\jmath}}

196: %\newcommand\ti{{\tilde \imath}}

197: %\newcommand\tj{{\tilde \jmath}}

198: 

199: %%	mathbbb symbols

200: %%%%%%%%%

201: 

202: \newcommand\bR{{\mathbb R}}

203: \newcommand\bC{{\mathbb C}}

204: \newcommand\bH{{\mathbb H}}

205: \newcommand\bK{{\mathbb K}}

206: %\newcommand\bN{{\mathbb N}}

207: %\newcommand\bR{{\Bbb R}}

208: \newcommand\bS{{\mathbb S}}

209: \newcommand\bT{{\mathbb T}}

210: \newcommand\bZ{{\mathbb Z}}

211: %

212: 

213: %%	mathbbb symbols

214: %%%%%%%%%

215: \newcommand\cA{{\mathcal A}}

216: \newcommand\cB{{\mathcal B}}

217: \newcommand\cC{{\mathcal C}}

218: \newcommand\cD{{\mathcal D}}

219: \newcommand\cE{{\mathcal E}}

220: \newcommand\cF{{\mathcal F}}

221: \newcommand\cG{{\mathcal G}}

222: \newcommand\cH{{\mathcal H}}

223: \newcommand\cI{{\mathcal I}}

224: \newcommand\cJ{{\mathcal J}}

225: \newcommand\cK{{\mathcal K}}

226: \newcommand\cL{{\mathcal L}}

227: \newcommand\cM{{\mathcal M}}

228: \newcommand\cN{{\mathcal N}}

229: \newcommand\cO{{\mathcal O}}

230: \newcommand\cP{{\mathcal P}}

231: \newcommand\cQ{{\mathcal Q}}

232: \newcommand\cR{{\mathcal R}}

233: \newcommand\cS{{\mathcal S}}

234: \newcommand\cT{{\mathcal T}}

235: \newcommand\cU{{\mathcal U}}

236: \newcommand\cV{{\mathcal V}}

237: \newcommand\cW{{\mathcal W}}

238: \newcommand\cX{{\mathcal X}}

239: \newcommand\cY{{\mathcal Y}}

240: \newcommand\cZ{{\mathcal Z}}

241: 

242: \renewcommand{\thefootnote}{{\arabic{footnote})}}

243: 

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

245: 

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

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

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

249: 

250: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

251: 

252: \theoremstyle{definition}

253: 

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

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

256: \newtheorem{condition}[theorem]{Condition}

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

258: 

259: \setcounter{section}{-1}

260: \newcommand{\sect}[1]{\setcounter{equation}{0}\section{#1}}

261: 

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

263: 

264: \newenvironment{claim}[1][(\theequation)]{\refstepcounter{equation}\smallskip

265: \begin{trivlist}

266: \item[{\hskip\labelsep#1}]}{\smallskip\end{trivlist}}

267: 

268: \newcounter{note}

269: 

270: \newenvironment{note}[1]{\refstepcounter{note}{\color{red}$^\thenote$}\reversemarginpar

271: \marginpar{ \hskip-40pt\colorbox{yellow}{\begin{minipage}{70pt} {\color{red}\thenote\ #1}\end{minipage}}}}

272: 

273: \newenvironment{highlight}[1]{\colorbox{yellow}{#1}}

274: 

275: \newenvironment{claim*}[1]{\smallskip

276: \begin{trivlist}

277: \item[{\hskip\labelsep#1}]}{\smallskip\end{trivlist}}

278: 

279: \newenvironment{pdeq}[1][]{\refstepcounter{equation}}{}

280: 

281: \setlength{\textwidth}{163truemm}

282: 

283: %\overfullrule=0pt

284: 

285: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

286: %%% BEGIN %

287: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

288: 	

289: \begin{document}

290: 

291: \title{%

292: Sharp Spectral Asymptotics for 2-dimensional Schr\"odinger operator with a strong

293: but degenerating magnetic field. II}

294: 

295: \author{%

296: Victor Ivrii

297: \footnote{Work was partially supported by NSERC grant OGP0138277.}

298: }

299: \maketitle

300: 

301: {\abstract

302: I consider the same operator as in part I \cite{IRO6} assuming however that

303: $\mu \ge Ch^{-1}$ and $V$ is replaced by $(2l+1)\mu h F+W$ with $l\in \bZ^+$. Under some non-degeneracy conditions I recover remainder estimates up to $O \bigl( \mu^{-{\frac 1 \nu}}h^{-1} +1\bigr)$ but now case $\mu \ge Ch^{-\nu}$ is no more forbidden and the principal part is of magnitude $\mu h^{-1}$. 

304: 

305: \endabstract}

306: 

307: %\AddToShipoutPicture*{

308: %\AtTextCenter{

309: %\makebox(0,0)[c]{\resizebox{\textwidth}{!}{

310: %\rotatebox{55}{\textsf{\textbf{\color{lightgray} Draft}}}}}

311: %}

312: %}

313: 

314: \setcounter{section}{5}

315: \sect{Modified $V$. I. $\mu \le \epsilon \lowercase {h}^{-\nu}$}

316: \subsection{Introduction}

317: 

318: This paper is a continuation of \cite{IRO6} which is considered as Part I. I consider spectral asymptotics of the magnetic Schr\"odinger operator

319: \begin{equation}

320: A= {\frac 1 2}\Bigl(\sum_{j,k}P_jg^{jk}(x)P_k -V\Bigr),\qquad P_j=D_j-\mu V_j,\quad V= (2l+1)\mu h F+W

321: \label{6-1}

322: \end{equation}

323: where $g^{jk}$, $V_j$, $W$ are smooth real-valued functions of $x\in \bR^2$, $l\in \bZ^+$ (i.e. $l=0,1,\dots$) and

324: $(g^{jk})$ is positive-definite matrix, $0<h\ll 1$ is a Planck parameter and

325: $\mu \gg1$ is a coupling parameter. I assume that $A$ is a self-adjoint operator and all the conditions are satisfied in the ball $B(0,1)$,

326: $F=F_{12}g^{-{\frac 1 2}}$, $F_{12}=\partial_{x_1}V_2-\partial_{x_2}V_1$,

327: $g=\det (g^{jk})^{-1}$. 

328: 

329: Further, exactly as in \cite{IRO6}, I assume that

330: \begin{equation}

331: F \asymp |x_1|^{\nu -1},\qquad \nu\in \bZ^+,\ \nu \ge 2

332: \label{6-2}

333: \end{equation}

334: and thus with no loss of the generality I can assume that

335: \begin{equation}

336: V_1=0,\qquad V_2\asymp x_1^\nu.

337: \label{6-3}

338: \end{equation}

339: Furthermore,  I assume that \underline{either}

340: \begin{pdeq}\label{6-4}\end{pdeq}

341: \begin{align}

342: &\pm W\ge \epsilon_0, \qquad &&\text{as\ } l\ge 0\tag*{$(\theequation)_\pm$}

343: \label{6-4-pm}\\

344: \intertext{and as $l=0$ only sign ``$+$'' is interesting \underline{or}}

345: &|\partial_{x_2} W/f |\ge \epsilon_0, \qquad f \Def F x_1^{1-\nu}.

346: \label{6-5}

347: \end{align}

348: Also as in \cite{IRO6}, I am interested in the asymptotics of

349: $\int e(x,x,0)\psi (x)\,dx$ where $e(x,y,\tau)$ is the Schwartz kernel of the spectral projector $E(\tau)$ of $A$ and $\psi\in C_0^\infty (B(0,{\frac 1 2}))$

350: is a cut-off function and I expect the main part of it to be

351: $\int \cE^\MW (x,0)\psi (x)\,dx$ where $\cE^\MW$ is defined by (0.8)\,\footnote{\label{foot-1} References by default are to \cite{IRO6}.}

352: which is of magnitude $\mu h^{-1}$. \emph{I am assuming without mention that $\psi$ is supported in the small but fixed vicinity of $\{x_1=0\}$.}

353: 

354: In the sharp contrast to the analysis of Part I the case $\mu \ge C h^{-\nu}$ is not ``forbidden'' anymore as well as zone

355: $\cZ''=\{|x_1|\ge {\bar\gamma}_1\Def C(\mu h)^{-1/(\nu -1)}\}$.

356: On the contrary, as $\mu h\ge C$ this zone becomes the main contributor to the principal part of asymptotics which now is of magnitude $\mu h^{-1}$ instead of

357: $h^{-2} (\mu h)^{-1/(\nu -1)}$ as it was in \cite{IRO6}.

358: Actually I will time to time slightly change the definition of ${\bar\gamma}_1$, replacing it by

359: ${\bar\gamma}_1=\epsilon(\mu h)^{-1/(\nu -1)}$ and back and changing respectively definition of zones.

360: 

361: Section 6 is devoted to the case  of $\mu \le \epsilon h^{-\nu}$. Analysis in zone  $\cZ'\Def \{|x_1|\le 2{\bar\gamma}_1\}$ remains basically the same and the main attention is paid here to the formally forbidden zone $\cZ''$. The main results here are theorems \ref{thm-6-10},  \ref{thm-6-11} and \ref{thm-6-17}.

362: 

363: As $\mu \ge \epsilon h^{-\nu}$ this separation to zones is no more reasonable and will be modified. In section 7 I analyze the case of 

364: $\epsilon h^{-\nu}\le \mu \le  Ch^{-\nu}$.  The main results here are theorems \ref{thm-7-3} and  \ref{thm-7-4}.

365: 

366: Further, in section 8 analyze the case of  $\mu \ge  Ch^{-\nu}$. The main results here are theorems \ref{thm-8-9},  \ref{thm-8-10},  \ref{thm-8-11} and  \ref{thm-8-12}.

367: 

368: Finally, appendix A is devoted to asymptotics of some one-dimensional Schr\"odiunger operators associated with (\ref{6-1}).

369: 

370: 

371: \subsection{Simple Rescaling}

372: 

373: As in \cite{IRO6} the simple rescaling arguments help us to get the easy but not sharp results. 

374: 

375: \medskip\noindent 

376: {\bf \thesubsection.1 }

377: In this and the next subsubsection I assume that $\mu \le Ch^{-\nu}$.Rescaling arguments in the zone

378: $\cZ'$ work exactly in the same manner as in \cite{IRO6} leading to the asymptotics of

379: $\int e(x,x,0)\psi'(x)\,dx$ with the principal part

380: $\int \cE^\MW (x,0)\psi' (x)\,dx$ and the remainder estimate $O(h^{-1})$ where $\psi'(x)$ and $\psi''(x)$ are cut-off functions supported in zones $\cZ'$ and $\cZ''$ (defined as above) respectively; one can take

381: $\psi'(x)=\psi (x) \psi_0(x_1/{\bar\gamma}_1)$, $\psi''=\psi-\psi'$ where $\psi_0\in C_0^\infty $ is supported in $(-1,1)$ and equals 1 in $[-{\frac 1 2}, {\frac 1 2}]$.

382: 

383: However the contribution of the previously forbidden zone $\cZ''$ to the remainder estimate is

384: \begin{equation*}

385: O\Bigl(\int_{\{{\bar\gamma}_1\le \gamma \le 1\}} \gamma^{-2}\, d\gamma\Bigr)= O\bigl({\bar\gamma}_1^{-1}\bigr)

386: \end{equation*}

387: which is $O\bigl(h^{-1}\bigr)$ due to assumption $\mu \le Ch^{-\nu}$ and the contribution of $\cZ''$ to the principal part is

388: \begin{equation}

389: \int \cE^\MW (x)\psi ''(x)\,dx=

390: {\frac 1 {4\pi}}\mu h^{-1}l_\pm \int\psi '' |F|\sqrt g\, dx,\qquad

391: l_\pm \Def l+{\frac 1 2}(-1\pm 1)

392: \label{6-6}

393: \end{equation}

394: under condition \ref{6-4-pm}.

395: 

396: Under condition (\ref{6-5}) the above arguments remain true for the contribution of the subzone $\cZ''\cap \{|W|\ge C\gamma\}$; for the contribution of the zone

397: $\cZ''\cap \{|W|\le C\gamma\}$ one needs to take in account correction term\footnote{\label{foot-2}See section 6 of \cite{Ivr1}.}

398: $\sum_m\kappa_m \mu_\eff h_\eff ^{1+2m}$ for the case $\mu_\eff h_\eff\ge 1$, $h_\eff\le 1$ where in the rescaling and division arguments

399: $\mu_\eff=\mu \gamma^{\nu-{\frac 1 2}}$,

400: $h_\eff =h\gamma ^{-\frac 3 2}$ and the number of balls is $O(1)$ for each $\gamma$. Then the total contribution of this correction terms is

401: $O\bigl(\mu h \bigr)$ as $\nu \ge 3$ and $O\bigl(\mu h |\log h|\bigr )$ as $\nu=2$.

402: 

403: \medskip\noindent

404: {\bf \thesubsection.2 }

405: Replacing $\psi$ by $x_1\psi$ in the above arguments one gains factor $\gamma$ in each integrand; then the total contribution of the zone $\cZ'$ to the remainder estimate becomes

406: \begin{equation*}

407: O\Bigl(\int \mu^{-1}h^{-1}\gamma^{1-\nu}\times \gamma \times \gamma^{-2}\,d\gamma \Bigr) = O\bigl( \mu^{-1/\nu}h^{-1}\bigr)

408: \end{equation*}

409: which is exactly what I want. On the other hand, the contribution of zone $\cZ''$ to the remainder estimate becomes

410: $O(\gamma^{-1}\,d\gamma)=O(|\log h|)$ which is what we want as 

411: $\mu \le C(h|\log h|)^{-\nu}$ only. To fix it under condition \ref{6-4-pm} one can notice that zone $\cZ''$ is  the spectral gap and therefore the contribution of the individual ball to the remainder estimate is

412: $O(\gamma h_\eff^s)$ with $h_\eff= h/\gamma$ rather than $O(1)$ and therefore the total contribution of zone $\cZ''$ to the remainder estimate is $O(1)$.

413: 

414: As before, under condition (\ref{6-5}) these arguments are applicable in the subzone $\cZ''\cap \{|W|\ge C\gamma\}$ with $h_\eff= h/(\gamma |\W|^{1/2})$

415: as long as $h_\eff\le 1$. This leads to $O(1)$ estimate of the contribution of the subzone $\cZ'' \cap\{ |W|\ge C\gamma, |W|^{1/2}\gamma \ge h\}$ to the remainder. One can see easily that the integral of $\gamma^{-1}$ taken over subzones $\cZ'' \cap\{ |W|\ge C\gamma, |W|^{1/2}\gamma \le h\}$

416: and $\cZ'' \cap\{ |W|\le C\gamma\}$ is $O(1)$ as well. Thus rescaling arguments

417: provide remainder estimate $O\bigl(\mu^{-{1/\nu}}+1\bigr)$ if $\psi$ contains an extra factor $x_1$ and under condition (\ref{6-5}) correction terms are taken into account.

418: 

419: Therefore

420: 

421: \begin{claim}\label{6-7}

422: As $\mu \le Ch^{-\nu}$ in what follows one can assume without any loss of the generality that $\psi (x)= \psi_1(x_1)\psi_2(x_2)$.

423: \end{claim}

424: 

425: \medskip\noindent

426: {\bf \thesubsection.3 }

427: As $\mu \ge Ch^\nu$ arguments of subsubsection \thesubsection.1 work as $\{|x_1|\ge C h\}$ providing $O(h^{-1})$ contribution of this zone to the remainder estimate while the contribution of zone $\{|x_1|\le C h\}$ will be $O(\mu h^{\nu -1})$. The main part of the asymptotics will be the same as above.

428: 

429: Moreover, arguments of subsubsection \thesubsection.1 work as $\{|x_1|\ge C h\}$ providing $O(1)$ contribution of this zone to the remainder estimate as $\psi$ is replaced by $x_1\psi$

430: while the contribution of zone $\{|x_1|\le C h\}$ will be $O(\mu h^\nu )$.

431: 

432: In the next section I will improve these latter results.

433: 

434: \subsection{Estimates. I}

435: In section 2 and subsections 4.1--4.4 of \cite{IRO6} various properties of operator $A$ were proven in the outer and inner zones

436: $\cZ_\out =\{{\bar \gamma}\le |x_1|\le 2{\bar\gamma}_1\}$ and

437: $\cZ_\inn =\{|x_1|\le 2{\bar\gamma}\}$ with

438: ${\bar\gamma}\Def C \mu^{-1/\nu}$ as long as ${\bar\gamma}\le {\bar\gamma}_1$ i.e. $\mu \le \epsilon h^{-\nu}$.

439: These properties were proven first in section 2 under assumption

440: \begin{align}

441: &C\le \mu \le \epsilon (h|\log h|)^{-\nu}\label{6-8}\\

442: \intertext{using standard microlocal analysis with logarithmic uncertainty principle and then in subsections 4.1--4.4 under assumption}

443: & \epsilon (h|\log h|)^{-\nu}\le \mu \le \epsilon h^{-\nu}\label{6-9}

444: \end{align}

445: applying microlocal analysis for $h$-pseudo-differential operators with respect to $x_2$ with operator-valued symbols -- operators in the auxiliary space $\bH = L^2(\bR_{x_1})$; I remind that in te case (\ref{6-9}) localization was done with respect to $\xi_2$ rather $x_1$.

446: 

447: Therefore in both cases (\ref{6-8}), (\ref{6-9}) in the redefined outer zone

448: \begin{equation}

449: \cZ_\out=\bigl\{{\bar\gamma}\le |x_1|\le {\bar\gamma}'_1=\epsilon {\bar\gamma}_1\bigr\}

450: \label{6-10}

451: \end{equation}

452: (with the small constant $\epsilon$) all these arguments remain true leading us eventually to the following statements:

453: 

454: \begin{proposition}\label{prop-6-1} Let conditions $(\ref{6-2})$ and $(\ref{6-4})_+$ be fulfilled. Let $\psi=\psi(x_2)$ be supported in

455: $B(0,{\frac 1 2})$ and let $\varphi=\varphi (\xi_2)$

456: be supported in the strip

457: \begin{equation}

458: \cY_\gamma=

459: \bigl\{\mu \gamma ^\nu \le |\xi_2|\le 2\mu \gamma^\nu\bigr\}

460: \label{6-11}

461: \end{equation}

462: with $C_1{\bar\gamma}\le \gamma \le \epsilon_1 {\bar\gamma}_1$

463: Then

464: 

465: \smallskip

466: \noindent

467: (i) As $\mu \le \epsilon h^{ -\nu}$ estimates

468: \begin{equation}

469: | F_{t\to h^{-1}\tau} \chi_T(t)\Gamma (Qe) |\le Ch^s

470: \label{6-12}

471: \end{equation}

472: and

473: \begin{equation}

474: \cR' = |\Gamma \bigl(Q e \bigr) - h^{-1}\int _{-\infty}^0 \Bigl( F_{t\to h^{-1}\tau} {\bar\chi}_T(t)\Gamma (Qe)\Bigr)\,d\tau |\le

475: C\mu^{-1} \gamma^{1-\nu}h^{-1}

476: \label{6-13}

477: \end{equation}

478: hold with $Qe=\varphi (hD_2) \bigl(e\psi\bigr)$, $e=e(x,y,\tau)$,

479: $|\tau|\le \epsilon$, $T\in [T_0,T_1]$, $T_0=Ch|\log h|$, $T_1=\epsilon \mu^{-1}\gamma^{-\nu}$;

480: 

481: \smallskip

482: \noindent

483: (ii) Moreover, under condition $(\ref{6-8})$ statement (i) holds with $Q=\psi_1\psi$, $\psi_1=\psi_1(x_1)$ supported in $\cZ_\gamma =

484: \{\gamma\le |x_1|\le 2\gamma\}$.

485: \end{proposition}

486: 

487: \begin{corollary} \label{cor-6-2} Let conditions $(\ref{6-2})$ and $(\ref{6-4})_+$ be fulfilled. Let $\psi=\psi(x_2)$

488: be supported in $B(0,{\frac 1 2})$ and $\varphi=\varphi(\xi_2)$

489: be supported in the outer zone defined in the terms of $\xi_2$

490: \begin{equation}

491: \cY_\out=

492: \bigl\{C_0\le |\xi_2|\le \epsilon \bigl(\mu h^\nu\bigr)^{-1/(\nu -1)}\bigr\}.

493: \label{6-14}

494: \end{equation}

495: Then

496: 

497: \smallskip

498: \noindent

499: (i) As $\mu \le \epsilon h^{ -\nu}$ estimate

500: \begin{equation}

501: \cR' \le C\mu^{-1/\nu} h^{-1}

502: \label{6-15}

503: \end{equation}

504: holds.

505: 

506: \smallskip

507: \noindent

508: (ii) Moreover, under condition $(\ref{6-8})$ statement (i) holds with $Q=\psi_1\psi$, $\psi_1=\psi_1(x_1)$ supported in $\cZ_\out$.

509: \end{corollary}

510: 

511: On the other hand, under condition  $(\ref{6-4})_-$ the whole zone $\cZ'=\cZ_\inn \cup \cZ_\out $ will be forbidden leading us to the following statement not having analogues in \cite{IRO6}:

512: 

513: \begin{proposition}\label{prop-6-3} Let conditions $(\ref{6-2})$ and $(\ref{6-4})_-$ be fulfilled. Let $\psi=\psi(x)$, $\psi_1=\psi_1(x_1)$ be supported in

514: $B(0,{\frac 1 2})$ and $\cZ'$ respectively and let

515: $\varphi =\varphi (\xi_2)$ be supported in the zone

516: \begin{equation}

517: \cY'=\{|\xi_2|\le \epsilon \bigl(\mu h^\nu\bigr)^{-1/(\nu -1)}\}.

518: \label{6-16}

519: \end{equation}

520: Then

521: 

522: \smallskip

523: \noindent

524: (i) $|Q e|\le Ch^s$ with $Qe=\varphi (hD_2) \bigl(e\psi\bigr)$, $e=e(x,y,\tau)$,

525: $|\tau|\le \epsilon$ as $\mu \le \epsilon h^{-\nu}$;

526: 

527: \smallskip

528: \noindent

529: (ii) Moreover, under condition $(\ref{6-8})$ statement (i) holds with $Q=\psi_1\psi$, $\psi_1=\psi_1(x_1)$ supported in $\cZ'$.

530: \end{proposition}

531: 

532: Therefore as $\mu \le \epsilon h^{ -\nu}$ and condition $(\ref{6-4})_+$ is fulfilled one needs to discuss the contribution of the inner zone

533: $\cZ_\inn =\{|x_1|\le {\bar\gamma} \}$ or equivalently

534: $\cY_\inn = \{|\xi_2|\le C_0\}$\,\footnote{\label{foot-3} These two definitions are essentially equivalent under condition (\ref{6-8}) but in the case (\ref{6-9}) one needs always use definition in the frames of $\xi_2$.} to the remainder estimate.

535: Furthermore one needs to consider the contribution of the previously forbidden zone

536: $\cZ''=\{|x_1|\ge {\bar\gamma}'_1\}$ or equivalently

537: $\cY''=\{|\xi_2|\ge \epsilon\bigl(\mu h^\nu\bigr)^{-1/(\nu -1)}\}$\,$^{\ref{foot-3}}$

538: to the remainder estimate.

539: 

540: The inner zone is analyzed exactly as in section 2 and subsections 4.1--4.4 of \cite{IRO6} leading us eventually to

541: 

542: \begin{proposition}\label{prop-6-4-pm} Let conditions $(\ref{6-2})$ and $(\ref{6-4})_+$ be fulfilled. Let $\psi=\psi(x_2)$ and $\psi_1=\psi_1(x_1)$ be supported in

543: $B(0,{\frac 1 2})$ and $\cZ_\inn$ respectively and let

544: $\varphi =\varphi (\xi_2)$ be supported in $\cY_\inn= \{ |\xi_2|\le C_0\}$. Then

545: all the results of section 2 and subsections 4.1--4.4 of {\rm\cite{IRO6}} remain true; in particular

546: 

547: \smallskip

548: \noindent

549: (i) As $\mu \le Ch^{\delta -\nu}$

550: \begin{equation}

551: \cR'' \Def |\Gamma \bigl(Q e \bigr) - h^{-1}\sum _j \int _{-\infty}^0 \Bigl( F_{t\to h^{-1}\tau} {\bar\chi}_{T_j}(t)\Gamma (Q_je)\Bigr)\,d\tau |\le C\mu^{-1/\nu} h^{-1}

552: \label{6-17}

553: \end{equation}

554: with $Qe=\varphi (hD_2) \bigl(e\psi\bigr)$, $e=e(x,y,\tau)$, $Q=\sum_j Q_j$ and

555: $|\tau|\le \epsilon$ where partition $Q_j$ and $Ch|\log h|\le T_j$ are defined following formula $(3.28)$ in {\rm \cite{IRO6}};

556: 

557: \smallskip

558: \noindent

559: (ii) Moreover, under nondegeneracy condition

560: \begin{pdeq}\label{6-18}\end{pdeq}

561: \begin{equation}

562: \sum_{1\le k\le m}|\partial_{x_2}^k \bigl({\frac W f}\bigr)|\ge \epsilon_0.

563: \label{6-18-m}

564: \tag*{$(\theequation)_m$}

565: \end{equation}

566: $\cR''$ does not exceed $C\mu^{-1/\nu} h^{-1}$ as $\mu \le \epsilon h^{-\nu}$;

567: 

568: \smallskip

569: \noindent

570: (iii) On the other hand, in the general case $\cR''$ does not exceed

571: $C\mu^{-1/\nu} h^{-1}+Ch^{-\delta}$ as $\mu \le \epsilon h^{-\nu}$;

572: 

573: \smallskip

574: \noindent

575: (iv) Furthermore, under condition $(\ref{6-8})$ all statements (i)--(iii) hold with $Q=\psi_1\psi$.

576: \end{proposition}

577: 

578: \begin{remark}\label{rem-6-5} In frames of proposition \ref{prop-6-4-pm} estimate

579: (\ref{6-12}) holds for $Q=Q_m$ and $T\in [T_m, T'_m]$ with $T'_m$ defined by (\ref{IRO6-2-98}) (it was denoted by $T_1$ then).

580: \end{remark}

581: 

582: \subsection{Estimates. II}

583: 

584: To investigate zone $\cZ''$ I will apply the theory of operators with operator-valued symbols. However, as $\mu \le \epsilon (h|\log h|)^{-\nu}$ one can apply a usual microlocal analysis with logarithmic uncertainty principle.

585: 

586: So, let us consider $A$ as $h$-pseudo-differential operator $\cA (x_2, h D_2)$ with operator-valued symbol $\cA(x_2,\xi_2)$. However, before doing this one can assume without any loss of the generality that $g^{11}=1$, $g^{12}=0$ and therefore

587: \begin{multline}

588: \cA(x_2,\xi_2)=

589: {\frac 1 2}\biggl(h^2D_1^2 + \sigma^2 (x) \bigl(\xi_2 -\mu V_2(x)\bigr)^2 - (2l+1)\mu h F -W(x)\biggr), \\

590: V_2=\phi (x){\frac 1 \nu}x_1^\nu

591: \label{6-19}

592: \end{multline}

593: with $\phi (x)=1$ as $x_1=0$; then $f=\sigma\phi$.

594: 

595: Further, for given $x_2$ by change of variable $x_1$ one can transform $\cA$ unitarily to the similar operator with $\phi =1$ and with

596: \begin{equation}

597: \sigma =1 \qquad \text{as\;} x_1=0;

598: \label{6-20}

599: \end{equation}

600: but this new operator is multiplied from the left and the right by $\alpha (x)$.

601: So operator $\cA(x_2,\xi_2)$ is unitary equivalent to

602: \begin{multline}

603: \cA' (x_2,\xi_2)= \\

604: {\frac 1 2}\alpha (x)

605: \biggl(h^2D_1^2 + \sigma^2 (x) \Bigl(\xi_2 -\mu {\frac 1 \nu}x_1^\nu \Bigr)^2 -(2l+1)\mu h \sigma (x)x_1^{\nu -1}-W_0(x)\biggr)

606: \alpha (x).

607: \label{6-21}

608: \end{multline}

609: Note that $W_0=W/f$ as $x_1=0$ and thus \emph{conditions \ \ref{6-4-pm}, $(\ref{6-5})$ and \ref{6-18-m} are reformulated in terms of $W_0$\/} obviously.

610: 

611: Proposition \ref{prop-A-3}(ii) of Appendix A implies that under condition \ref{6-4-pm} zone

612: $\cY''\setminus \cY''_0=\{\epsilon (\mu h^\nu)^{-1/(\nu -1)} \le |\xi_2|\le

613: 2C(\mu h^\nu)^{-1/(\nu -1)}\}$ is microhyperbolic with respect to $\xi_2$ and thus one can extend $\cY'$ to zone

614: ${\bar\cY}' \Def \{ |\xi_2|\le 2C(\mu h^\nu)^{-1/(\nu -1)}\}$ resulting in the following statement:

615: 

616: \begin{proposition}\label{prop-6-6} Let conditions $(\ref{6-2})$ and\ref{6-4-pm} be fulfilled. Then estimate $\cR'\le C$ holds as $\cR'$ is defined by $(\ref{6-13})$ with $Qe=\varphi (hD_2)(e\psi)$, $\varphi $ supported in the zone $\cY''\setminus \cY''_0$, $T\in [T_0,T_1]$, $T_0=Ch|\log h|$,

617: $T_1=\epsilon (\mu h^\nu )^{-1/(\nu-1)}$, $\mu\le \epsilon h^{-\nu}$.

618: \end{proposition}

619: 

620: Furthermore,  proposition \ref{prop-A-3}(i) implies that under condition \ref{6-4-pm} zone

621: $\cY''_0=\{|\xi_2|\ge C(\mu h)^{-1/(\nu-1)}\}$ is forbidden on energy levels $|\tau|\le \epsilon$ as long as $\mu \le \epsilon h^{-\nu}$ is forbidden; namely

622: \begin{equation}

623: | F_{t\to h^{-1}\tau }{\bar\chi}_T(t) (Qu)(x,y,t) |\le CT h^s

624: \qquad \forall \tau:|\tau|\le\epsilon

625: \label{6-22}

626: \end{equation}

627: as $Q\psi =\varphi (hD_2)(u\psi)$ with $\varphi $ supported in the zone $\cY''_0$ and therefore its contribution to the remainder $\cR'$ defined by (\ref{6-13}) is negligible as well:

628: 

629: \begin{proposition}\label{prop-6-7} Let conditions $(\ref{6-2})$ and \ref{6-4-pm} be fulfilled. Then estimate

630: $\cR'\le Ch^s$ holds as $\cR'$ is defined by $(\ref{6-13})$ with

631: $Qe=\varphi (hD_2)(e\psi)$, $\varphi $ supported in the zone $\cY''_0$,

632: $T\ge T_0=Ch|\log h|$, $\mu\le \epsilon h^{-\nu}$.

633: \end{proposition}

634: 

635: The analysis of all zones under condition (\ref{6-5}) will be done in subsection 6.7.

636: 

637: \subsection{Calculations. I}

638: 

639: In this subsection I will change partition: instead of $\cZ'$ and $\cZ''$ I will consider ${\bar\cY}'$ and $\cY''_0$ obtained if I redefine

640: ${\bar\gamma}_1= C(\mu h)^{-1/(\nu -1)}$; respectively change definitions and notations of zones $\cY_\out$, $\cZ_\out$, $\cZ'$, $\cZ''$.

641: 

642: After estimates were derived in two previous subsections under assumption

643: $Ch^{-1}\le \mu \le \epsilon h^{-\nu}$ and condition \ref{6-4-pm} calculations in zone ${\bar\cY}'$ are done exactly as in section 3 and subsection 4.4 of \cite{IRO6}.

644: 

645: On the other hand, calculations in zone $\cY''_0$ as $\mu \le \epsilon h^{-\nu}$ are rather obvious under assumptions

646: $Ch^{-1}\le \mu \le \epsilon h^{-\nu}$ and \ref{6-4-pm}. Therefore I arrive to the intermediate estimate

647: \begin{equation}

648: |\int \Bigl( \bigl(\varphi (hD_2)e\bigr)(x,x,0)\ - (2\pi h)^{-1}

649: \int {\rm e}(x_1,x_1;x_2,\xi_2,0)\varphi(\xi_2)\,d\xi_2\Bigr)

650: \psi _2(x_2)\, dx|\le R

651: \label{6-23}

652: \end{equation}

653: where $R$ is an estimate already derived in the corresponding conditions (also see below) and $\varphi\in C_0^\infty (-\epsilon',\epsilon')$ with sufficiently small constant $\epsilon'$.

654: 

655: Then the same estimate holds with $\psi(x_2)$ replaced by $\psi (x)$ such that

656: $\psi (x)=\psi _2(x_2)$ as $|x_1|\le C_1\epsilon'$ because this transition leads to a negligible error. I take $\psi$ also satisfying $\psi(x)=0$ as $|x_1|\ge 2C_1\epsilon'$. Then in the latter estimate I can replace $\varphi $ by 1. Really, then the error would be

657: \begin{equation}

658: |\int \Bigl( \bigl((1-\varphi (hD_2))e\bigr)(x,x,0)\ - (2\pi h)^{-1}

659: \int {\rm e}(x_1,x_1;x_2,\xi_2,0)(1-\varphi(\xi_2))\,d\xi_2\Bigr)

660: \psi (x)\, dx|

661: \label{6-24}

662: \end{equation}

663: and replacing $\psi$ by $\psi'$ equal to $\psi$ as $|x_1|\ge 2C_2^{-1}\epsilon'$

664: and equal to 0 as $|x_1|\le C_2^{-1}\epsilon'$ leads to a negligible error. However, to expression (\ref{6-24}) modified this way one can apply the theory of operators with non-degenerating magnetic field and then to estimate expression (\ref{6-24}) by $C$.

665: 

666: Thus I derived (\ref{6-23}) with $\varphi$ replaced by 1 and $\psi_2(x_2)$ replaced by some ``special'' function $\psi(x)$. Then due to rescaling arguments like in subsubsection 6.2.2 the same estimate holds for a general function $\psi(x)$ supported in $\{|x_1|\le 2C_1\epsilon'\}$. Thus I arrive to

667: 

668: \begin{proposition}\label{prop-6-8} Let conditions $(\ref{6-2})$ and $(\ref{6-4})_+$ be fulfilled. Then 

669: 

670: \smallskip

671: \noindent

672: (i) As \underline{either} $\mu \le h^{\delta-\nu}$ \underline{or} condition \ref{6-18-m} is fulfilled and $\mu \le \epsilon h^{-\nu}$ the following estimate holds

673: \begin{equation}

674: \cR_I\Def |\int \Bigl( e(x,x,0)\ - (2\pi h)^{-1}

675: \int {\rm e}(x_1,x_1;x_2,\xi_2,0)\,d\xi_2\Bigr)

676: \psi (x)\, dx|\le C\mu^{-{\frac 1 \nu }}h^{-1}

677: \label{6-25}

678: \end{equation}

679: where here and below ${\rm e}(x_1,y_1; x_2,\xi_2,\tau)$ is the Schwartz kernel of the spectral projector of operator $\cA(x_2,\xi_2)$ defined by $(\ref{6-19})$ and $\delta >0$ is an arbitrarily small exponent;

680: 

681: \smallskip

682: \noindent

683: (ii) In the general case with $\mu \le \epsilon h^{-\nu}$ estimate

684: \begin{equation}

685: \cR_I \le C\mu^{-{\frac 1 \nu }}h^{-1} +Ch^{-\delta}

686: \label{6-26}

687: \end{equation}

688: holds.

689: \end{proposition}

690: 

691: I remind that in both statements of proposition \ref{prop-6-8} the principal part of asymptotics has magnitude $\asymp \mu h^{-1}$ (as $\mu \ge h^{-1}$).

692: 

693: On the other hand, under condition

694: $(\ref{6-4})_-$ zone $\cY'$ becomes forbidden and thus I arrive to

695: 

696: \begin{proposition}\label{prop-6-9} Let conditions $(\ref{6-2})$ and $(\ref{6-4})_-$ be fulfilled and $l\ge 1$. Then for

697: $Ch^{-1}\le \mu \le \epsilon h^{-\nu}$ estimate $\cR_I\le C$ holds while the principal part of asymptotics has magnitude $\asymp \mu h^{-1}$.

698: \end{proposition}

699: 

700: \subsection{Calculations. II}

701: 

702: Transition to the auxiliary operator $\cA_0$ without increasing error estimates could be done easily in zone $\cY_\out$ exactly as it was done in the proof of propositions \ref{IRO6-prop-3-3} and \ref{IRO6-prop-3-4} while arguments of \ref{IRO6-prop-3-8} etc work in zone $\cY_\inn$. 

703: 

704: On the other hand, this transition in zone $\cY''_0$ is obvious under condition \ref{6-4-pm}, and I arrive to two theorems below as $\mu \le h^{-\nu}|\log h|^{-K}$ and function $\psi$ is ``special'' in the sense of the previous subsection. Then the same arguments as there extend theorem to general $\psi$.

705: 

706: Furthermore, under condition \ref{6-4-pm} the case

707: $h^{-\nu}|\log h|^{-K}\le \mu \le \epsilon h^{-\nu}$ is analyzed exactly as in section 4 of Part I leading to the extension of these theorems to

708: $\mu \le \epsilon h^{-\nu}$:

709: 

710: \begin{theorem}\label{thm-6-10} Let conditions $(\ref{6-2})$ and $(\ref{6-4})_+$ be fulfilled. Then

711: 

712: \smallskip

713: \noindent

714: (i) As \underline{either} $\mu \le h^{\delta-\nu}$ \underline{or} condition \ref{6-18-m} is fulfilled and $\mu \le \epsilon h^{-\nu}$ 

715: \begin{align}

716: &\cR^*\Def |\int \Bigl( e(x,x,0)- {\tilde\cE}^\MW (x,0)\Bigr)\psi (x)\,dx - \int \cE^\MW_\corr (x_2,0) \psi(0,x_2)\,dx_2|

717: \label{6-27}\\

718: \intertext{does not exceed $C\mu^{-1/\nu}h^{-1}$ where}

719: &\cE^\MW_\corr (x,\tau)\Def (2\pi h)^{-1}

720: \int {\rm e}_0(x_1, x_1;x_2, \xi_2, \tau, \hbar) \,dx_1 d\xi_2-

721: \int {\tilde \cE}^\MW_0 (x,\tau)\,dx_1,

722: \label{6-28}

723: \end{align}

724: $\cE^\MW$ is Magnetic Weyl approximation\footnote{\label{foot-4} See e.g. $(\ref{IRO6-0-8})$.} and here and below ${\rm e}_0(x_1,y_1; x_2,\xi_2,\tau)$ is the Schwartz kernel of the spectral projector of operator $\cA_0(x_2,\xi_2)$ defined by $(\ref{6-19})$ and with $\alpha, \phi, \sigma, W$ restricted to $\{x_1=0\}$ and $\cE^\MW_0$ is Magnetic Weyl approximation for this operator.

725: 

726: \smallskip

727: \noindent

728: (ii) In the general case with $\mu \le \epsilon h^{-\nu}$ estimate

729: $\cR^*\le C\mu^{-1/\nu}+Ch^{-\delta}$ holds.

730: \end{theorem}

731: 

732: \begin{theorem}\label{thm-6-11} Let conditions $(\ref{6-2})$ and $(\ref{6-4})_-$ be fulfilled and $l\ge 1$. Then as

733: $Ch^{-1}\le \mu \le \epsilon h^{-\nu}$ estimate $\cR^*\le C$ holds while the principal part of asymptotics has magnitude $\asymp \mu h^{-1}$.

734: \end{theorem}

735: 

736: \begin{remark}\label{rem-6-12} 

737: Obviously the same approximate expressions  (\ref{IRO6-3-52}), \ref{IRO6-3-52-*}, \ref{IRO6-3-52-**} hold for the part of $\cE^\MW_\corr$ ``associated'' with $\cY_\inn$;

738: \end{remark}

739: 

740: \subsection{Estimates under condition $(\ref{6-5})$}

741: 

742: I start from the remainder estimate in zone ${\bar\cY}'$ which is trivial:

743: 

744: \begin{proposition}\label{prop-6-13} Let conditions $(\ref{6-2})$, $(\ref{6-20})$ and $(\ref{6-5})$ be fulfilled. Then

745: 

746: \smallskip

747: \noindent

748: (i) Estimate $(\ref{6-13})$ holds with $Qe=\varphi (hD_2)(e\psi)$, $\varphi $ supported in the strip $\cY_\gamma$ with the same restrictions to $\gamma$ and

749: the same $T_0$, $T_1$ as in proposition \ref{6-1}(i);

750: 

751: \smallskip

752: \noindent

753: (ii) Furthermore, the same estimate holds as $\varphi $ is supported in zone $\cY_\inn$ and $\gamma={\bar\gamma}_0=\mu^{-1/\nu}$;

754: 

755: \smallskip

756: \noindent

757: (iii) Therefore $\cR'$ defined by $(\ref{6-13})$ does not exceed  $C\mu^{-1/\nu}h^{-1}$ as $\varphi$ is supported in zone ${\bar\cY}'$ and $T=T_0$.

758: \end{proposition}

759: 

760: Now let us analyze zone $\cY''_0$ under condition $(\ref{6-5})$:

761: 

762: \begin{proposition}\label{prop-6-14} Let conditions $(\ref{6-2})$, $(\ref{6-20})$ and $(\ref{6-5})$ be fulfilled. Then estimate

763: $\cR'\le C$ holds as $\cR'$ is defined by $(\ref{6-13})$ with

764: $Qe=\varphi (hD_2)(e\psi)$, $\varphi $ supported in the zone $\cY''_0$.

765: \end{proposition}

766: 

767: \begin{proof} (i) Let us note first that estimate

768: \begin{equation}

769: | F_{t\to h^{-1}\tau }\bigl({\bar\chi}_{T_1}(t)- {\bar\chi}_{\bar T}(t)\bigr) (Qu)(x,y,t) |\le Ch^s

770: \qquad \forall \tau:|\tau|\le\epsilon

771: \label{6-29}

772: \end{equation}

773: holds with $T_1= \epsilon \mu^{-1} \gamma^{-\nu}$, ${\bar T}=Ch|\log h|$

774: as $Qu=\varphi (hD_2)(u\psi)$, $\varphi $ supported in the strip

775: $\cY_{(\gamma)}=\{\mu \gamma^\nu \le |\xi_2|\le 2 \gamma^\nu\}$ with

776: $\gamma \ge C{\bar\gamma}_1$.

777: 

778: Really, us consider a partial trace $\Gamma' (Qu)$ (with respect to $x_1$). Due to proposition \ref{prop-A-3} the propagation speed with with respect to $x_2$ does not exceed $C|\xi_2|^{-1}\asymp C(\mu \gamma^\nu)^{-1}$ and the propagation speed with respect to $\xi_2$ does not exceed $C$\,\footnote{\label{foot-5} Under some assumptions this would be equivalent to the estimate of the the average propagation speed with respect to $x_1$ of $Qu$ by

779: $C\gamma (\mu \gamma^\nu)^{-1}$; further one can estimate

780: average propagation speed with respect to $x_2$ of $Qu$ by

781: $C(\mu \gamma^\nu)^{-1}$

782: as well.}; moreover, under condition (\ref{6-5}) this propagation speed with respect to $\xi_2$  is greater than $\epsilon$.

783: 

784: On the other hand, an obvious estimate

785: \begin{equation}

786: | F_{t\to h^{-1}\tau } {\bar\chi}_{T_0}(t) \Gamma(Qu)(t) |\le

787: C\mu\gamma^\nu h^{-1}\times T_0= C\mu \gamma^\nu |\log h|

788: \label{6-30}

789: \end{equation}

790: holds where the first factor is $\mu_\eff h_\eff^{-1}\gamma^{-1}$; furthermore, due to (\ref{6-29}) this estimate holds for the left-hand expression with $T_0$ replaced by $T_1$.

791: 

792: Therefore the contribution of the strip $\cY_\gamma$ to the remainder estimate does not exceed

793: \begin{equation}

794: C\mu \gamma^\nu |\log h| \times T_1^{-1}= C|\log h|

795: \label{6-31}

796: \end{equation}

797: and therefore the total contribution of $\cY''_0$ to the remainder estimate does not exceed

798: $C|\log h| \int \gamma^{-1}\,d\gamma \asymp C|\log h|^2$.

799: 

800: This estimate is as good as I need for $\mu \le Ch^{-\nu}|\log h|^{-2\nu}$. However for $Ch^{-\nu}|\log h|^{-2\nu}\le \mu \le \epsilon h^{-\nu}$ I would like to improve it getting rid of two logarithmic factors.

801: 

802: \medskip

803: \noindent

804: (ii) Getting rid off one of them is easy: rescaling $t \mapsto t/T$, $(x_j-y_j)\mapsto (x_j-y_j)/T$, $\mu \mapsto \mu T$, $h\mapsto h/T$

805: estimates for Schr\"odinger operator with strong non-degenerate magnetic field

806: \cite{Ivr1}, section 6 (with arbitrary parameters $\mu$ and $h$ such that 

807: $\mu h \ge C$) I arrive to two following inequalities

808: \begin{align}

809: & |F_{t\to h^{-1}\tau}\chi_T(t) \Gamma (Qu) |\le C\mu \bigl({\frac h T} \bigr)^s

810: \label{6-32}\\

811: & |F_{t\to h^{-1}\tau}{\bar \chi}_T(t) \Gamma (Qu) |\le C\mu \label{6-33}

812: \end{align}

813: as $h\le T\le 1$, $|\tau|\le \epsilon$ under condition

814: $|W|+|\nabla W|\ge \epsilon_0$. Then using our standard scaling

815: $x_1\mapsto x_1/\gamma$, $x_2\mapsto (x_2-y_2)/\gamma$,

816: $\mu \mapsto \mu_\eff =\mu \gamma^\nu$, $h\mapsto h_\eff=h/\gamma$ and 

817: $T\mapsto T/\gamma$ I arrive to estimate (\ref{6-30}) without logarithmic factor

818: \begin{equation}

819: | F_{t\to h^{-1}\tau } {\bar\chi}_T(t) \Gamma(Qu)(t) |\le

820: C\mu \gamma^\nu

821: \tag*{$(\ref*{6-30})^*$}\label{6-30-*}

822: \end{equation}

823: as $|\tau|\le\epsilon$, $T/\gamma \le \epsilon \mu \gamma^\nu \iff T\le T'_1=\epsilon \mu \gamma^{\nu +1} $. Further,  and  to

824: $(\ref{6-29})$ this estimate holds as $h\le T\le T_1=\epsilon \mu \gamma^\nu $ provided $T'_1\ge Ch$ i.e. $\gamma \ge {\bar \gamma}_1$.

825: 

826: Then the contribution of the strip $\cY_\gamma$ to the remainder $\cR'$ is $C$ and therefore the total estimate is $C|\log h|$.

827: 

828: \medskip

829: \noindent

830: (iii) To get rid off the second logarithmic factor I need to further increase $T_1$ in the previous arguments and for this purpose I need for each $\gamma$ to make $x_2$-partition of $\cY_\gamma$ of the size

831: \begin{equation}

832: \ell = \epsilon |V(0,x_2)| +{\bar \ell},\qquad {\bar\ell}\ge C\gamma.

833: \label{6-34}

834: \end{equation}

835: 

836: Consider first elements $\cU_{\gamma,\ell}$ with $\ell\ge C{\bar\ell}$. For every such element on levels $\tau$ with $|\tau|\le \epsilon \ell$ after rescaling

837: \begin{equation}

838: x_2\mapsto x_2 \ell^{-1},\quad h\mapsto h' =h \ell ^{-\frac 3 2},\quad

839: t\mapsto t \ell^{-1},\quad

840: \mu\mapsto \mu'=\mu \ell^{\frac 1 2}

841: \label{6-35}

842: \end{equation}

843: I am in the elliptic situation.

844: 

845: Therefore contribution of each such element to the remainder estimate does not exceed $C\mu '(h')^s$ and therefore the total  contribution of such elements is negligible as ${\bar\ell}= h^\delta$.

846: 

847: So I need to consider only elements $\cU'_\gamma=\cU_{\gamma,\ell}$ with

848: $\ell \asymp {\bar\ell}=h^\delta$. For such elements after rescaling (\ref{6-35}) I can apply estimate \ref{6-30-*}; then scaling back I get the same estimate \ref{6-30-*} again but with $Q=\psi' (x_2)\varphi (hD_2)$

849: supported in $\cU'_\gamma$, $|\tau|\le \epsilon \ell$ and

850: $Ch|\log h|\ell^{-1}\le T\le T_1=\epsilon \mu \gamma^{\nu+1}$\,\footnote{\label{foot-6} It is consistent with the fact that support of $\psi'$ is of the length $\ell$ but now ${\bar T}=Ch|\log h|/\ell$. }. Furthermore, applying (\ref{6-29}) I can increase $T_1$ to 

851: $\epsilon \mu \gamma^\nu$.

852: 

853: So far I gained nothing: the estimate I  proved alone would bring me the same final remainder estimate $C|\log h|$ as before but now I can further increase $T_1$  and thus reduce the remainder estimate.

854: 

855: Namely, let us consider propagation in the time direction in which $|\xi_2|$ increases. If only propagation with respect to $\xi_2$ was considered, until time $\epsilon_3 \mu$ it would be confined to zone

856: \begin{equation*}

857: \bigl\{\epsilon_0 \le |\xi_2|\bigl(\mu \gamma^\nu + |t|\bigr)^{-1}\le C\bigr\} \subset

858: \bigl\{{\frac 1 2}\mu \gamma^\nu \le |\xi_2|\le \epsilon _1\mu\bigr\}

859: \end{equation*}

860: and thus to

861: $\{|x_1|\le \epsilon_3\}$. 

862: 

863: However let us note that the propagation speed with respect to $x_2$ does not exceed $C\ell/|\xi_2|$ as $\ell\ge C|V|+{\bar\ell}$. Therefore one can prove easily that propagation, which started in the zone

864: $\{|x_2|\le {\frac 1 2}, |V|\le h^\delta\}$ as I have assumed,

865: until time $T^*_1=\mu \gamma^\nu h^{-\delta_1}$ is confined to a bit larger zone

866: $\bigl\{|x_2|\le {\frac 3 4}, |V|\le h^{\delta/2}\bigr\}$ of the same type.

867: 

868: Therefore estimate \ref{6-30-*} holds with $Ch ^{1-2\delta} \le T\le T^*_1$. Then due to the Tauberian approach contribution of each partition element $\cU'_\gamma$ to the remainder estimate does not exceed $

869: C\mu \gamma^\nu T^{* -1}_1= Ch^{\delta_1}$ and the contribution of the whole strip $\cY_\gamma$ does not exceed $Ch^{\delta_1}$ as well and of the whole zone $\cY''_0$ does not exceed $Ch^{\delta_2}$.

870: 

871: Clearly, at some moment I increased slightly $T_0$ but after summation over partition was done I can (using negligibility of the trace on

872: $[Ch|\log h|, h^{1-\delta}]$ time interval on energy levels

873: $|\tau|\le \epsilon$) return to original ${\bar T}$.

874: \end{proof}

875: 

876: \subsection{Calculations under condition $(\ref{6-5})$}

877: 

878: Calculations in zone ${\bar\cY}'$ are exactly as in \cite{IRO6}. However one should be more careful with calculations in zone $\cY''_0$.

879: 

880: Let me remind that according to subsection 6.2 \cite{Ivr1} in the nondegenerate case with $\mu h\ge C$ the operator in question is reduced to one-dimensional $\mu^{-1}h$-pdo  $B(x_2,\mu^{-1}hD_2, h^2)$\,\footnote {\label{foot-7} Where $x_2$ is not our original $x_2$.} with the ``main symbol'' $B(x_2,\xi_2, 0)= W \circ \Psi$ and therefore the contribution of the partition element to the final answer will be given as in subsection 6.6 by magnetic Weyl expression

881: $\int \cE^\MW (x,0)\psi (x)\, dx$ plus correction terms

882: $ \mu h ^{1+2m} \int \varkappa_{l,m} (x)\psi (x) \, dx$, $m=0,1,\dots$.

883: 

884: After rescaling $\mu\mapsto \mu \gamma^\nu$, $h\mapsto h/\gamma$, $dx\mapsto \gamma^{-2}dx$ these terms are transformed into

885: \begin{equation}

886: \mu h^{1+2m} \int \varkappa_{l,m} (x,\gamma )\psi (x) \gamma^{\nu -2m -3} \,dx

887: \label{6-36}

888: \end{equation}

889: integrated over zone $\{{\bar \gamma}_1\le \gamma \le \epsilon\}$.

890: 

891: One can see easily that if there was an extra factor $\gamma$ one would be able to rewrite this expression (\ref{6-36}) modulo $O(1)$ into the similar expression with integration over $\{\gamma \le \epsilon\}$ as

892: $2m+2 < \nu$\,\footnote{\label{foot-8} thus resulting in exactly expression $\kappa_{l,m} \mu h^{1+2m}$ as in non-degenerate case.}

893: or to simply skip it as $2m+2 > \nu$ or to get a term which is

894: $O\bigl(\mu h^\nu |\log h|\bigr)$ as $2m+2=\nu$. To gain this extra factor one needs to consider the difference of expressions $\int e(x,x,0)\psi (x)\,dx$ for two operators with $g^{jk}(x)$, $f(x)$, $V(x)$ coinciding as $x_1=0$. As this second operator it is natural to pick up the simplest one i.e.

895: \begin{equation}

896: A_0= {\frac 1 2}\biggl(h^2D_1^2 + \bigl(hD_2-\mu x_1^\nu/\nu\bigr)^2 - (2l+1)\mu h x_1^{\nu-1} - W(x_2)\biggr).

897: \label{6-37}

898: \end{equation}

899: Therefore I arrive to

900: 

901: \begin{proposition}\label{prop-6-15} Under condition $(\ref{6-5})$ estimate

902: \begin{multline}

903: |\int \biggl( e(x,x, 0) - e_0(x,x,0) -\cE^\MW (x,0) + \cE^\MW_0 (x,0)\biggr)

904: \psi (x)\,dx - \\

905: \sum \kappa'_{l,m} \mu h^{1+2m}| \le C\mu^{-{1/\nu}}h^{-1}

906: \label{6-38}

907: \end{multline}

908: holds as $\mu \le h^{-\nu}|\log h|^{-K}$ where $e_0$ and $\cE^\MW_0$ are defined

909: for operator $A_0$.

910: \end{proposition}

911: 

912: \begin{claim}\label{6-39}

913: Now in what follows I can consider operator $A_0$ instead of $A$.

914: \end{claim}

915: 

916: Then I can apply the standard method of successive approximations with unperturbed operator $\cA (y_2,hD_2)$ and plug the results of successive approximations into expression

917: \begin{equation}

918: h^{-1}\int_{-\epsilon}^0 \biggl( F_{t\to h^{-1}\tau} {\bar\chi}_T(t) \Gamma (Qu)\biggr)\,d\tau

919: \label{6-40}

920: \end{equation}

921: which calculates exactly contribution of the ``problematic'' eigenvalue $\lambda_l$ of the corresponding one-dimensional Schr\"odinger operator; I remind that $T={\bar T}=Ch|\log h|$.

922: 

923: Thus while the main part of asymptotics is estimated by

924: $C\mu h^{-2}\gamma ^\nu T = C\mu h^{-1}\gamma^\nu |\log h|$, each next term seemingly acquires factor

925: \begin{equation}

926: Ch^{-1}\bigl(\mu h \gamma^{\nu-1}\bigr)^{1/2}T^2 \asymp

927: Ch \bigl(\mu h \gamma^{\nu-1}\bigr)^{1/2}|\log h|^2;

928: \label{6-41}

929: \end{equation}

930: since the propagation speed with respect to $x_2$ is estimated by

931: $C_0(\mu h\gamma^{\nu -1})^{1/2}$ such factor could be larger than 1.

932: 

933: In fact however, $C_0(\mu h\gamma^{\nu -1})^{1/2}$ is the estimate for the instant propagation speed only. Using instead  the mentioned reduction to a one-dimensional $\mu^{-1}h$-pdo one can find that the propagation speed with respect to $x_2$ is estimated by $C_0\mu ^{-1}$ if magnetic field is non-degenerate and then in the canonical coordinates for time $T={\bar T}$ the shift of $(x'_2,\xi'_2)$ will be estimated by $C_0\bigl(\mu^{-}h |\log h|\bigr)^{1/2}$ which is the smallest distance allowed by the logarithmic uncertainty principle\footnote{\label{foot-9} Since $\mu^{-1}h$-Fourier Integral Operators are involved later one needs the same distance in each $(x,\xi)$ direction.} and this would persist if one returns back to the original 

934: $(x_2, \mu^{-1}\xi_2)$; so one would be able to estimate $(x_2-y_2)$ on the time interval in question by

935: $C_0\bigl(\mu^{-1}h |\log h|\bigr)^{1/2}$.

936: 

937: In the degenerate case described here one must replace $\mu$, $h$ by

938: $\mu \gamma^\nu$, $h/\gamma$ respectively and then multiply by $\gamma$ thus producing final estimate for $|x_2-y_2|$

939: \begin{equation}

940: \varrho\Def C\bigl(\mu ^{-1}h \gamma ^{1-\nu} |\log h|\bigr)^{1/2} \asymp

941: Ch {\bar \gamma}_1^{{\frac 1 2}(\nu -1)}\gamma ^{-{\frac 1 2}(\nu -1)}

942: |\log h|^{\frac 1 2}

943: \label{6-42}

944: \end{equation}

945: and therefore each next term acquires factor $\varrho |\log h|$. Then $m$-th term of the final answer is estimated by

946: \begin{equation}

947: C\mu h^{-1}\varrho^{m-1} |\log h|^K \asymp C\mu h^{m-2}

948: \gamma^{\nu - {\frac 1 2}(\nu -1)(m-1)}

949: {\bar\gamma}_1^{{\frac 1 2}(\nu-1)(m-1)}|\log h|^K.

950: \label{6-43}

951: \end{equation}

952: After integration over $\gamma^{-1}\,d\gamma$ with

953: ${\bar\gamma}_1\le \gamma \le \epsilon$ expression (\ref{6-43}) results in

954: $C\mu h^{m-2}{\bar\gamma}_1^\nu|\log h|^K$ as

955: $\nu - {\frac 1 2}(\nu -1)(m-1)\le 0$ or in

956: $C(\mu ^{-1}h)^{(m-3)/2}|\log h|^K$ otherwise.

957: One can check easily that in either case the answer is $O(|\log h|^K)$ as

958: $m\ge 3$ and only terms with $m=1,2$ should be considered more carefully under condition (\ref{6-8}).

959: 

960: \smallskip

961: On the other hand, the main term appears as (\ref{6-40}) with $u$ replaced by ${\bar u}$ and modulo negligible one can rewrite it with any $T\ge {\bar T}$, in particular with $T=\infty$ which leads to

962: \begin{equation}

963: (2\pi h)^{-1} \int

964: {\rm e}_0(x_1,x_1,0; x_2,\xi_2)\psi (x_1)\varphi (\xi_2) \,dx_2\,d\xi_2

965: \label{6-44}

966: \end{equation}

967: where I remind that ${\rm e}_0(x_1,y_1,0; x_2,\xi_2)$ is the Schwartz kernel of the spectral projector of one-dimensional Schr\"odinger operator $\cA_0(x_2,\xi_2)$.

968: 

969: \smallskip

970: Let us consider terms with $m=2$ i.e. expression (\ref{6-40}) with $u$ replaced by ${\bar u}_1$; similarly to analysis of (i) one can estimate contribution of $O\bigl((x_2-y_2)^2\bigr)$ terms in the perturbation

971: $\cA (x_2,hD_2) - \cA (y_2,hD_2)$ by $C|\log h|^K$. Therefore one should consider only

972: $\cA (x_2,hD_2) - \cA (y_2,hD_2)= (x_2-y_2)B_1 (y_2)$ in which case

973: ${\bar u}_1$ is defined by (3.23) without the last term since $B_1$ commutes with $(x_2-y_2)$:

974: \begin{equation}

975: u \mapsto {\bar u}_1=-ih\sum_{\varsigma=\pm }\varsigma {\bar G}^\varsigma B_1

976: {\bar G}^\varsigma [{\bar A}, x_2-y_2] {\bar G}^\varsigma \delta(t)\delta (x_2-y_2)\delta(x_1-y_1).

977: \label{6-45}

978: \end{equation}

979: 

980: One needs to multiply this by $h^{-1}\psi$, integrate with respect to $\tau$ and apply $\Gamma$ to it. Obviously since

981: for odd $\nu$ operators ${\bar G}^\varsigma$ and $[{\bar A}, x_2-y_2]$ are even and odd respectively as $x_1\mapsto -x_1$, $\xi_2\mapsto -\xi_2$ the answer would be 0 if $\psi$ is even with respect to $x_1$.

982: 

983: To cover the case of even $\nu$ and general $\psi$ let us note that $B_1$ commutes with ${\bar G}^\varsigma$ considered as operators in the auxiliary space $L^2(\bR^1_{x_1})$. Then if ${\bar G}^\varsigma$ commuted with $\psi$, taking trace and integrating with respect to $\tau$ would result in

984: \begin{equation*}

985: \const \cdot \partial_{\xi_2} B_1 \sum_{\varsigma=\pm }

986: \varsigma \Tr \bigl( {\bar G}^\varsigma \psi\bigr)

987: \end{equation*}

988: which after integration over $\xi_2$ results in $0$.

989: 

990: However ${\bar G}^\varsigma$ does not commute with $\psi$, so instead of 0 one gets

991: \begin{equation*}

992: \const \cdot B_1 \sum_{\varsigma=\pm }\Tr

993: \varsigma\Bigl({\bar G}^\varsigma

994: \bigl(\partial_{\xi_2} {\bar G}^\varsigma\bigr) \Bigl({\bar G}^\varsigma

995: [{\bar A},\psi]\Bigr)

996: \end{equation*}

997: and to this expression one can apply the same type of transformations and calculations as in the proof of proposition \ref{prop-6-15} resulting in

998: the expressin $\sum_m \kappa_{l,m} \mu h^{1+2m}$ where coefficients $\kappa_{l,m}$ are changed as needed.

999: 

1000: Therefore combining with the results for zone ${\bar\cY}'$ I arrive to

1001: 

1002: \begin{proposition}\label{prop-6-16} For a model operator

1003: \begin{multline}

1004: |\int \biggl(e_0(x,x,0) - (2\pi h)^{-1}\int

1005: {\rm e}_0(x_1,x_1,0; x_2,\xi_2)\,d\xi_2\biggr)\psi (x)\,dx -\\

1006: \sum \kappa_m \mu h^{1+2m}|\le C\mu^{-1/\nu}h^{-1}

1007: \label{6-46}

1008: \end{multline}

1009: as $\mu \le Ch^{-\nu}|\log h|^{-K}$.

1010: \end{proposition}

1011: 

1012: Further, combining this with proposition \ref{prop-6-14} \ I get as

1013: $\mu \le h^{-\nu}|\log h|^{-K}$ estimate (\ref{6-47}):

1014: 

1015: \begin{theorem}\label{thm-6-17} Under condition $(\ref{6-5})$ estimate

1016: \begin{multline}

1017: |\int \biggl( e(x,x, 0) -

1018: (2\pi h)^{-1} \int {\rm e}_0(x_1,x_1,0;x_2,\xi_2)\,d\xi_2 \\-\cE^\MW (x,0) + \cE^\MW_0 (x,0)\biggr)

1019: \psi (x)\,dx -

1020: \sum \kappa_{l,m} \mu h^{1+2m} | \le C\mu^{-{1/\nu}}h^{-1}

1021: \label{6-47}

1022: \end{multline}

1023: holds as $\mu \le \epsilon h^{-\nu}$.

1024: \end{theorem}

1025: 

1026: \begin{proof} To finish the proof of this theorem one needs to cover the case

1027: $h^{-\nu}|\log h|^{-K}\le \mu \le \epsilon h^{-\nu}$, getting rid of the term $|\log h|^K$ in the error estimates.

1028: 

1029: The first problematic error comes from the correction terms in proposition \ref{prop-6-15}, namely from the terms of the type

1030: $\mu h^{1+2m}\int \varkappa_{l,m}(x_2)\gamma^{\nu -2m-3+k}\,dx$ with $k\ge 1$,

1031: $\nu -2m-3+k=-1$ and this error term is $O(1)$ unless $k=1$, $\nu=2m+1$ in which case it it is $\kappa'_l \mu h^\nu |\log h|$. This is possible only for odd $\nu$ in which case operator $\cA_0$ is even with respect to $x_1\mapsto -x_1$,

1032: $\xi_2\mapsto -\xi_2$ but perturbation contains exactly one factor $x_1$ and therefore it is odd and after integration with respect to $x_1$, $\xi_2$ this correction term results in 0 if $\psi$ is even with respect to $x_1$.

1033: 

1034: Further, one needs to consider terms corresponding to $m=3$ in the successive approximations leading to proposition \ref{prop-6-16} and there one can replace

1035: $\cA _0(x_2,\xi_2)-\cA _0(y_2,\xi_2)$ by $B_1 (x_2-y_2)$, and also terms corresponding to $m=2$ in the same successive approximations and there one can replace $\cA _0(x_2,\xi_2)-\cA _0(y_2,\xi_2)$ by $B_2 (x_2-y_2)^2$.

1036: 

1037: To calculate the contribution of such terms one can apply the same approach as in the proof of proposition \ref{prop-6-15} and the contribution of $\gamma$-admissible partition element with respect to $x_1$ will be

1038: \begin{equation*}

1039: \sum _m \mu h^{1+2m} \int \varkappa_{l,m,k}(x_2) \psi (x) \gamma^{\nu-2m-3+k}\,dx

1040: \end{equation*}

1041: with $k\ge 0$; however since this expression should be $O(|\log h|^K)$ all the terms but those with $\nu\le 2m+1$, $k\ge 1$ should vanish; further, the total contribution of all remaining terms save those with $\nu = 2m+1$ and $k=1$ is $O(1)$, which leaves us with no ``bad'' terms for even $\nu$ and with one ``bad'' term $\kappa'_l \mu h^\nu \log h$ for odd $\nu$, $m=(\nu-1)/2$. However, parity considerations with respect to $x_1$ show that this term should vanish as well.

1042: \end{proof}

1043: 

1044: \begin{remark}\label{rem-6-18-m}(i)

1045: All the coefficients $\varkappa_{l,*}$ and $\kappa_{l,*}$  vanish for $l=0$.

1046: 

1047: \smallskip

1048: \noindent

1049: (ii) Obviously the same approximate expressions  (\ref{IRO6-3-52}), \ref{IRO6-3-52-*}, \ref{IRO6-3-52-**} as in \cite{IRO6} hold for part $\cE^\MW_\corr$ ``associated'' with $\cY_\inn$;

1050: \end{remark}

1051: 

1052: \sect{Modified $V$. II. $\epsilon_0 \lowercase {h}^{-\nu}\le \mu \le C_0\lowercase { h}^{-\nu}$}

1053: 

1054: 

1055: Now I will consider the intermediate case

1056: \begin{equation}

1057: \epsilon_0 h^{-\nu}\le \mu \le C_0 h^{-\nu}

1058: \label{7-1}

1059: \end{equation}

1060: with arbitrarily small constant $\epsilon_0$ and arbitrarily large constant $C_0$; this case which described the largest possible values in \cite{IRO6}

1061: now is no more than transition to the next section.

1062: 

1063: \subsection{Estimates}

1064: 

1065: Let us denote by

1066: $\lambda_n(\xi_2)$ eigenvalues of operator

1067: \begin{equation}

1068: {\bf a}^0= {\frac 1 2}\biggl(D_1^2 + \bigl(\xi_2 - x_1^\nu/\nu\bigr)^2 - (2l+1)x_1^{\nu-1}\biggr);

1069: \label{7-2}

1070: \end{equation}

1071: then $\Lambda_n(x_2,\xi_2)=\lambda_n (\xi_2) - {\frac 1 2} W(x_2)$ are eigenvalues of

1072: ${\bf a}= {\bf a}^0-W(x_2)$.

1073: 

1074: My main nondegeneracy assumption will be

1075: \begin{equation}

1076: |\Lambda_n|+ (|\xi_2|+1) |\partial_{\xi_2}\Lambda_n| +|\partial_{x_2}\Lambda_n|\ge \epsilon_0\qquad \forall n,\xi_2,

1077: \label{7-3}

1078: \end{equation}

1079: may be coupled with \ref{6-4-pm}. This condition (\ref{7-3}) follows from

1080: (\ref{6-5}); further, it follows from \ref{6-4-pm} for $|\xi_2|\ge C$.

1081: On the other hand, since $\lambda_n\to 0$ and

1082: $\xi_2\partial_{\xi_2}\lambda_n\to 0$ as $|\xi_2|\to \infty$, condition (\ref{7-3}) implies that $|W|+|\partial_{x_2}W|\ge \epsilon_0$ and therefore

1083: locally one of conditions \ref{6-4-pm}, (\ref{6-5}) must be fulfilled. 

1084: 

1085: Obviously, under conditions (\ref{7-1}),(\ref{7-3}) for each $\xi_2$ number of eigenvalues of one-dimensional operator

1086: \begin{equation}

1087: \cA_0= {\frac 1 2}\biggl(h^2D_1^2 + \bigl(\xi_2 - \mu x_1^\nu/\nu\bigr)^2 - (2l+1)x_1^{\nu-1}-W\biggr)

1088: \label{7-4}

1089: \end{equation}

1090: below level $c_0$ does not exceed $C$.

1091: 

1092: Further, note that condition (\ref{7-3}) for eigenvalues of $\cA_0$ is equivalent to the same condition for eigenvalues of ${\bf a}$. Then I easily arrive to

1093: 

1094: \begin{proposition}\label{prop-7-1} Under conditions $(\ref{7-1}),(\ref{7-3})$

1095: contribution to the remainder estimate of the zone $\{|\xi_2|\le C\}$ is $O(1)$.

1096: \end{proposition}

1097: 

1098: Furthermore, analysis in the zone $\cY''_0$ under condition (\ref{7-1}) does not differ from the analysis as $\mu\le \epsilon h^{-\nu}$. Namely

1099: 

1100: \begin{claim}\label{7-5}

1101: Under conditions (\ref{7-1}) and \ref{6-4-pm} operator $\cA_0$ and thus operator $\cA$ is elliptic in the zone $\cY''_0\Def \{|\xi_2|\ge C\}$ and the contribution of $\cY''_0$ to the remainder estimate is negligible.

1102: \end{claim}

1103: 

1104: \begin{claim}\label{7-6}

1105: Similarly, under conditions (\ref{7-1}) and (\ref{6-5}) operator $\cA$ is microhyperbolic in the zone $\cY''_0\Def \{|\xi_2|\ge C\}$ and the contribution of $\cY''_0$ to the remainder estimate is $O(1)$.

1106: \end{claim}

1107: 

1108:  Therefore

1109: 

1110: \begin{proposition}\label{prop-7-2} Let conditions $(\ref{7-1}),(\ref{7-3})$

1111: and one of conditions \ref{6-4-pm}, $(\ref{6-5})$ be fulfilled. Then the remainder estimate is $O(1)$ where the principal part is defined by $(\ref{6-40})$.\end{proposition}

1112: 

1113: \subsection{Calculations} 

1114: Calculations in this case also do not differ from those in section 6 leading to

1115: the following statements

1116: 

1117: \begin{theorem}\label{thm-7-3}

1118: Let conditions $(\ref{7-1})$, $(\ref{7-3})$ and \ref{6-4-pm} be fulfilled. Then $\cR_I$ defined by $(\ref{6-25})$ and $\cR^*$ defined by $(\ref{6-27})$ do not exceed $C$.

1119: \end{theorem}

1120: 

1121: \begin{theorem}\label{thm-7-4}

1122: Let conditions $(\ref{7-1})$ and $(\ref{6-5})$ be fulfilled. Then left-hand expressions of $(\ref{6-38})$, $(\ref{6-46})$ and $(\ref{6-47})$ do not exceed $C$.

1123: \end{theorem}

1124: 

1125: 

1126: \sect{Modified $V$. III. $ \mu \ge C_0\lowercase { h}^{-\nu}$}

1127: 

1128: Now I consider the previously forbidden case

1129: \begin{equation}

1130: \mu \ge C_0 h^{-\nu}

1131: \label{8-1}

1132: \end{equation}

1133: with sufficiently large constant $C_0$. In this case all zones should be redefined. Also the difference between $l=0$ and $l\ge 1$ becomes crucial.

1134: 

1135: 

1136: \subsection{Estimates. I}

1137: 

1138: As $|\xi_2|\asymp \mu \gamma^\nu$, $\gamma \ge C_1(\mu^{-1}h )^{1/(\nu+1)}$ let us consider first eigenvalues $\Lambda_n(x_2,\xi_2)$ of operator $\cA (x_2,\xi_2)$.

1139: Then proposition \ref{prop-A-3} implies instantly that

1140: 

1141: \begin{claim}\label{8-2} As $n\ne l$ and $|\xi_2|\asymp \mu \gamma^\nu$, 

1142: $\gamma \ge C_1(\mu^{-1}h )^{1/(\nu+1)}$

1143: \begin{equation*}

1144: \Lambda_n (x_2,\xi_2)  \asymp (n-l) \mu h \gamma ^{\nu-1}

1145: \end{equation*}and signs of the left and right-hand expressions coincide and \begin{equation}\label{8-3}

1146: \Lambda_l(x_2,\xi_2)=\omega_l h^2\gamma^{-2}-{\frac 1 2}W(x_2)+

1147: O\Bigl(h^2\gamma^{-1} + h^2 (\mu^{-1}h)^2\gamma^{-4-2\nu}\Bigr), \quad

1148: \omega_l > 0\ \text{as\;}\ l\ge1.

1149: \end{equation}

1150: \end{claim}\vskip-20pt

1151: Therefore

1152: 

1153: \begin{claim}\label{8-4} As $l\ge 1$ zone

1154: $\cY'''\Def \bigl\{  C_0(\mu h^\nu)^{1/(\nu+1)}\le |\xi_2|\le 

1155:  \epsilon \mu h^\nu \bigr\}$ 

1156: is elliptic and its contribution to the remainder estimate is $O(h^s)$.

1157: \end{claim}

1158: On the other hand,

1159: 

1160: \begin{claim}\label{8-5}

1161: Under condition \ref{6-4-pm} zone $\cY''\Def \bigl\{|\xi_2|\ge C\mu h^\nu\bigr\}$ is elliptic as well and its contribution to the remainder estimate is  $O(h^s)$ as well for $l\ge 0$.

1162: \end{claim}

1163: 

1164: Therefore as $l\ge 1$ and condition \ref{6-4-pm} is fulfilled, one needs to analyze only two remaining zones

1165: $\cX_1=\bigl\{\epsilon \rho_1\le |\xi_2|\le C\rho_1 \bigr\}$, 

1166: $\rho_1=\mu h^\nu$ and

1167: $\cX_0=\bigl\{|\xi_2|\le C_0\rho_0 \bigr\}$, $\rho_0=(\mu h^\nu)^{1/(\nu+1)}$.

1168: 

1169: In the zone $\cX_1$ propagation speed with respect to $x_2$ is in average $\asymp \rho^{-1}$ (with $\rho=\rho_1$)  due to proposition \ref{prop-A-3} again and the propagation speed with respect to $\xi_2$ is in average $O(1)$ and therefore one can take

1170: \begin{equation}

1171: T_0=Ch|\log h|,\qquad T_1=\epsilon_1 \rho_1

1172: \label{8-6}

1173: \end{equation}

1174: and for $T\in [T_0,T_1]$

1175: propagation on the energy levels $\tau\in [-\epsilon,\epsilon]$ which started in $B(0,{\frac 1 2})$ does not leave $B(0,1)$ but the shift with respect to $x_2$

1176: is $\asymp \rho^{-1}T$ and it satisfies logarithmic uncertainty principle and thus the spectral trace is negligible.

1177: 

1178: \begin{remark}\label{rem-8-1}

1179: One should be more careful as $\mu \ge h^{-M}$ with arbitrarily large $M$ and use $\log \mu$ instead of $|\log h|$.

1180: \end{remark}

1181: 

1182: Therefore

1183: \begin{equation}

1184: |F_{t\to h^{-1}\tau}{\bar \chi}_T(t) (Qu)|

1185: \label{8-7}

1186: \end{equation}

1187: does not exceed $Ch^{-1}\rho T_0= C \rho |\log h|$ where $Q$ is a partition element corresponding to $\cX_1$, $|\tau|\le \epsilon$. Therefore due to Tauberian arguments  the contribution of this zone to the remainder is

1188: $O(h^{-1}T_0 /T_1)=O(|\log h|)$. One can get rid off this superficial logarithmic factor both in the estimate of (\ref{8-7}) and in the remainder estimate; standard details I leave to the reader. So,

1189: 

1190: \begin{proposition}\label{prop-8-2} Let $l\ge 1$ and conditions \ref{6-4-pm} and $(\ref{8-1})$ be fulfilled. Then as $Q$ is supported in the zone $\cX_1$ expression $(\ref{8-7})$ does not exceed $C\rho_1$ and the contribution of $\cX_1$

1191: to the remainder estimate is $O(1)$.

1192: \end{proposition}

1193: 

1194: Therefore I am left with the zone

1195: $\cX_0=\bigl\{|\xi_2|\le C_0(\mu h^\nu)^{1/(\nu+1)} \bigr\}$. Let us fix $x_2$. I don't know if eigenvalue $\lambda_n(\xi_2)$ of ${\bf a}^0(\xi_2)$ vanishes in $\cX_0$ (may be even with some of its derivatives)\footnote{\label{foot-10} It clearly happens for even $\nu$ and $n<l$.}

1196: but I know that if it happens then $n\le c_1$; moreover due to the analyticity of $\lambda _n(\xi_2)$ it can happen only in no more then $C_1$ points and due to proposition \ref{prop-A-3}  and the analyticity of $\lambda _n(\xi_2)$  

1197: \begin{equation}

1198: \lambda _n (\eta)\sim \alpha (\eta-{\bar\eta})^r

1199: \label{8-8}

1200: \end{equation}

1201: for some $\alpha \ne 0$ and $r=1,2,\dots$   near each such point ${\bar\eta}$, $\alpha$ and $r$ depend on ${\bar\eta}={\bar\eta}_{n,k}$ $k=1,\dots, K$ (depending on $\nu,l$ as well).  Further,   two eigenvalues do not vanish simultaneously.

1202: 

1203: But then condition \ref{6-4-pm} will provide non-degeneracy. Really, in our assumptions an ellipticity is broken only in the strips of the type

1204: \begin{equation}

1205: \cY=\bigl\{ |\xi_2 -{\bar\eta}\rho_0|\asymp C\Delta\bigr\},\qquad\Delta = \rho_0^{1-2/r},

1206: \label{8-9}

1207: \end{equation}

1208: and the average propagation speed with respect to $x_2$ is of magnitude

1209: $\rho_0^{-1}|\xi_2 -{\bar\eta}|^{r-1}\asymp \rho_0^{(2-r)/r}$ there

1210: and therefore one can take

1211: \begin{equation}

1212: T_1=\epsilon \rho_0^{-(2-r)/r},\qquad

1213: T_0 = Ch|\log h| \rho_0^{-(2-r)/r} \Delta^{-1}\asymp h|\log h|,\qquad \Delta = \rho_0^{1-2/r}.

1214: \label{8-10}

1215: \end{equation}

1216: Therefore for $Q$ supported in the strip $\cY$ expression (\ref{8-7}) does not exceed $Ch^{-1}\Delta \times T_0 = C|\log h| \rho_0^{-(2-r)/r}$

1217: and contribution of $\cY$ to the remainder estimate does not exceed this expression multiplied by $T_1^{-1}$ i.e. $Ch|\log h|$. Furthermore, using standard methods one can easily get rid off the superficial logarithmic factor both in the estimate of (\ref{8-7}) and the remainder estimate:

1218: 

1219: \begin{proposition}\label{prop-8-3} Let $l\ge 1$ and conditions \ref{6-4-pm} and $(\ref{8-1})$ be fulfilled. Then as $Q$ is supported in the strip $\cY$ described by $(\ref{8-9})$, expression $(\ref{8-7})$ does not exceed $C\rho_0^{-(2-r)/r}$ and the contribution of $\cY$ to the remainder estimate is $O(1)$. 

1220: \end{proposition}

1221: 

1222: Therefore I arrive to

1223: 

1224: \begin{proposition}\label{prop-8-4}

1225: Let $l\ge 1$ and conditions \ref{6-4-pm} and $(\ref{8-1})$ be fulfilled. Then the remainder estimate is $O(1)$ while the principal part is given by $(\ref{6-40})$ for different strips with any $T\in [T_0,T_1]$ defined by $(\ref{8-10})$ for strip $\cY$ under conditions $(\ref{8-8})-(\ref{8-9})$ and by $(\ref{8-6})$ for strip $\cX_1$.

1226: \end{proposition}

1227: 

1228: I would like to note that

1229: 

1230: \begin{proposition}\label{prop-8-5} Let $l\ge 1$ and conditions $(\ref{6-4})_-$ and $(\ref{8-1})$ be fulfilled. Then 

1231: 

1232: \smallskip

1233: \noindent

1234: (i) Zone $\cX_1$ is elliptic and its contribution to the remainder estimate is $O(h^s)$;

1235: 

1236: \smallskip

1237: \noindent

1238: (ii) Furthermore if also condition

1239: \begin{equation}

1240: \lambda_n(\eta)\ne 0 \qquad \forall n,\eta

1241: \label{8-11}

1242: \end{equation}

1243: is fulfilled\,\footnote{\label{foot-11} However I cannot check condition (\ref{8-11}).} then the remainder estimate is $O(h^s)$.

1244: \end{proposition}

1245: 

1246: \subsection{Estimates. II}

1247: 

1248: Let us consider the special case $l=0$; I remind that  then only eigenvalue $\lambda_0(\eta)$ should be considered and that condition $(\ref{6-4})_-$ leads then to the asymptotics with the principal part 0 and remainder estimate $O(h^s)$ and therefore is excluded from the further consideration.

1249: 

1250: Further, as $\nu$ is odd $\lambda_0=0$ identically, condition $(\ref{6-4})_+$ provides ellipticity everywhere. Thus I arrive to

1251: 

1252: \begin{proposition}\label{prop-8-6}

1253: Let $l=0$, $\nu$ be odd and conditions $(\ref{6-4})_+$ and $(\ref{8-1})$ be fulfilled. Then the remainder estimate is $O(h^s)$ while the principal part is given by $(\ref{6-40})$.

1254: \end{proposition}

1255: 

1256: On the other hand, as $l=0$, $\nu$ is even and condition $(\ref{6-4})_+$ holds due to proposition \ref{prop-A-7} ellipticity is violated only in the strip

1257: \begin{equation}

1258: \cY=\{ \epsilon_1 \Delta \le |\xi_2 - \eta \rho_0 | \le C\Delta\}, \qquad \eta \asymp |\log \rho_0|^{\nu/(\nu+1)},\;

1259: \Delta =\rho_0 |\log \rho_0|^{-1/(\nu+1)}

1260: \label{8-12}

1261: \end{equation}

1262: where as before $\rho_0 =(\mu h^\nu)^{1/(\nu+1)}$. In this strip propagation speed with respect to $x_2$ is $\asymp \Delta^{-1}$ and again

1263: \begin{equation}

1264: T_0= Ch|\log h|,\qquad T_1= \epsilon \Delta

1265: \label{8-13}

1266: \end{equation}

1267: and expression (\ref{8-7}) does not exceed

1268: $Ch^{-1}\Delta T_0 = C\Delta |\log h|$ and the remainder estimate is

1269: $O(|\log h|)$. Further, by the standard arguments one can get rid off the superficial logarithmic factors. Thus

1270: 

1271: \begin{proposition}\label{prop-8-7}

1272: Let $l=0$, $\nu$ be even and conditions $(\ref{6-4})_+$ and $(\ref{8-1})$ be fulfilled. Then the remainder estimate is $O(1)$ while the principal part is given by $(\ref{6-40})$ with $T_0,T_1$ defined by $(\ref{8-13})$.

1273: \end{proposition}

1274: 

1275: \subsection{Estimates. III}

1276: 

1277: Now I want to derive estimates under condition \ref{6-4-pm} replaced by (\ref{6-5}). Without condition \ref{6-4-pm} some zones cease to be elliptic and should be reexamined:

1278: 

1279: \begin{claim}\label{8-14}

1280: As $l\ge 1$ these zones are

1281: $\bigl\{|\xi_2|\ge C \mu h^\nu\bigr\}$ and also\end{claim}\vskip-20pt

1282: \begin{claim}\label{8-15}

1283: As $l\ge 1$ these zones are ``inner parts'' of the strips  described by (\ref{8-9}), namely,

1284: $\cY=\bigl\{ |\xi_2 - {\bar\eta }\rho_0 | \le \epsilon_1\Delta\bigr\}$.

1285: \end{claim}\vskip-20pt

1286: \begin{claim}\label{8-16} As $l=0$, $\nu $ even this zone is

1287: $\bigl\{|\xi_2|\ge C\rho_0 |\log \rho_0|^{\nu/(\nu+1)}\bigr\}$;

1288: \end{claim}\vskip-20pt

1289: \begin{claim}\label{8-17} As $l=0$, $\nu$ odd this zone is 

1290: $\bigl\{|\xi_2|\le \epsilon \mu\bigr\}$.

1291: \end{claim}

1292: 

1293: Since condition (\ref{6-5}) provides $T_0=Ch|\log h|$ anyway contribution of (\ref{8-9})-type strips to the remainder estimate will be $O(1)$ again. The standard partition-rescaling arguments in all other zones bring contribution of all other zones to $O(\log \mu)$; however additional arguments of the proof of proposition \ref{prop-6-14} allow us to reduce it to $O(1)$. Therefore

1294: 

1295: \begin{proposition}\label{prop-8-8} Let conditions $(\ref{8-1})$ and $(\ref{6-5})$ be fulfilled. Then

1296: the remainder estimate is $O(1)$ while the principal part of the asymptotics is given by $(\ref{6-40})$ for different zones with any $T\in [T_0,T_1]$, $T_0=Ch|\log h|$ and $T_1$ defined as in propositions \ref{prop-8-2}--\ref{prop-8-7}.

1297: \end{proposition}

1298: 

1299: \subsection{Calculations. I}

1300: In this subsection I give the principal parts of asymptotics already derived under condition \ref{6-4-pm} in more explicit form.

1301: 

1302: First of all, consider method of successive approximations fixing $x_2=y_2$. Then while contribution of the strip of the width

1303: $\Delta $ in $\xi_2$ to the principal part is of magnitude $\Delta h^{-1}$, each next term of successive approximations acquires factor

1304: $|\partial_{\xi_2}\Lambda _n| T \times T/h\asymp

1305: (\partial_{\xi_2}\Lambda _n) h|\log h|^2$ with $T=T_0$ where $\Lambda_n$ is an eigenvalue of $\cA$.

1306: Further one needs to consider only strips where ellipticity fails and then $\Delta \asymp |\partial_{\xi_2}\Lambda_n|^{-1}$.

1307: 

1308: So, the first, the second and the the third terms do not exceed

1309: \begin{pdeq}\label{8-18}\end{pdeq}

1310: \begin{equation}

1311: Ch^{-1}|\partial_{\xi_2}\Lambda _n|^{-1},\quad

1312: C |\log h|^2, \quad

1313: C h |\partial_{\xi_2}\Lambda_n |\cdot |\log h|^4

1314: \tag*{$(\theequation)_{1-3}$}\label{8-18-*}

1315: \end{equation}

1316: respectively.

1317: 

1318: Actually the second term in the successive approximations is $O(1)$. Really, considering the second term which corresponds to the linear part $(x_2-y_2)\partial_{y_2}\cA(y_2,hD_2)$ of the perturbation one can rewrite it 

1319: as the result of direct calculations in the form including 

1320: $\partial_{x_2}\partial_{\xi_2}\Lambda _n =0$; on the other hand considering the second term corresponding to the rest $(x_2-y)^2\cB (x_2,y_2,hD_2)$ of the perturbation one can estimate it easily by $O(h^\delta)$.

1321: 

1322: Now I can rewrite the principal part of the asymptotics as

1323: \begin{equation}

1324: (2\pi h)^{-1}\int {\rm e}(x_1,x_1,0;x_2,\xi_2)\psi (x)\,d\xi_2 dx

1325: \label{8-19}

1326: \end{equation}

1327: with error not exceeding already achieved remainder estimate which is either $O(1)$ or $O(h^\infty)$ (where remainder estimate $O(h^\infty)$ corresponds to the elliptic case and no successive approximations are needed at all).

1328: 

1329: Let us consider the contribution of the strips where ellipticity is broken to the error; I remind it does not exceed the minimum of all three expressions in \ref{8-18-*}. Then $(\ref{8-18})_3$ is obviously $O(1)$ in all cases with the singular exception of the strip  (\ref{8-9}) with $r=1$, 

1330: $\rho h \ge |\log h|^{-K}$. However in this case $(\ref{8-18})_1$ is $O(1)$

1331: unless $|\log h|^{-K}\le \rho h \le |\log h|^K$ and one can still handle this case getting rid off the superficial logarithmic factors in $(\ref{8-18})_{1,3}$  by the standard arguments. Thus I arrive to

1332: 

1333: \begin{theorem}\label{thm-8-9} Let conditions \ref{6-4-pm} and $(\ref{8-6})$ be fulfilled. Then

1334: 

1335: \smallskip

1336: \noindent

1337: (i) Asymptotics with the principal part given by $(\ref{8-19})$ holds with the remainder estimate $O(1)$;

1338: 

1339: \smallskip

1340: \noindent

1341: (ii) Furthermore, as $l=0$, $\nu$ is odd this asymptotics holds with the remainder estimate $O(h^\infty)$.

1342: \end{theorem}

1343: 

1344: Furthermore, fixing $W$ at $x_1=0$ and $\alpha=1 $ and thus replacing $\cA$ by $\cA^0$ to the pilot model operator, I can apply the method of successive approximation again; then each next term  gets an extra factor

1345: $C\gamma T_0 h^{-1} |\log h|$ with $\gamma=(\mu ^{-1}|\xi_2|)^{1/\nu}$ and only strips where ellipticity breaks should be counted. Also one can see easily that

1346: 

1347: \begin{claim}\label{8-20}

1348: The error does not exceed the second term

1349: $Ch^{-2}T_0 \Delta \gamma$\,\footnote{\label{foot-12} I skip superficial logarithmic factors one can easily get rid off by the standard arguments.}.

1350: Furthermore, for odd $\nu$ and perturbation, which is odd with respect to $x_1$, the second term is 0 and therefore the error does not exceed the sum of the second term with a perturbation $O(x_1^2)$ and the third term with a perturbation $O(x_1)$ i.e. $Ch^{-3}T_0^2 \Delta \gamma^2$\,$^{\ref{foot-12}}$.

1351: \end{claim}

1352: 

1353: Thus, I just list the different cases:

1354: 

1355: \begin{claim}\label{8-21}

1356: As $l\ge 1$ and condition $(\ref{6-4})_+$ is fulfilled the main contribution to the error is provided by the zone $\cX_1$ with $\xi_2 \asymp \mu h^\nu$ and

1357: $\gamma \asymp h$ and of the width

1358: $\Delta \asymp \mu h^\nu$; so the error is $O\bigl(\mu h^\nu\bigr)$. The contributions of (\ref{8-9})-type strips  are much smaller;

1359: \end{claim}

1360: \vskip-10pt

1361: \begin{claim} \label{8-22} As $l\ge 1$ and condition $(\ref{6-4})_-$ is fulfilled the main contribution to the error is provided by (\ref{8-9})-type strips with the largest possible $r$; then $\xi_2 =O\bigl( (\mu h^\nu)^{1/(\nu+1)}\bigr)$,

1362: $\gamma \asymp (\mu ^{-1}h)^{1/(\nu+1)}$ and

1363: $\Delta \asymp (\mu h^\nu)^{(r-2)/r(\nu+1)}$; so the error is

1364: $O\bigl( (\mu h^\nu)^{-\delta}\bigr)$ with $\delta = 2/r(\nu+1)$ anyway;

1365: \end{claim}

1366: \vskip-10pt

1367: \begin{claim} \label{8-23} As $l= 0$, $\nu$ is even and condition $(\ref{6-4})_+$ is fulfilled the main contribution to the error is provided by  $\cX_1$ with

1368: $\xi_2 \asymp (\mu h^\nu)^{1/(\nu+1)}|\log (\mu h^\nu)|^{\nu/(\nu+1)}$,

1369: $\gamma \asymp (\mu ^{-1}h)^{1/(\nu +1)} |\log (\mu h^\nu)|^{1/(\nu+1)}$ and of the width

1370: $\Delta \asymp (\mu h^\nu )^{1/(\nu+1)}|\log (\mu h^\nu)|^{-1/(\nu+1)}$;

1371: so the error is $O(1)$ anyway;

1372: \end{claim}

1373: \vskip-10pt

1374: \begin{claim} \label{8-24} As $l= 0$, $\nu$ is odd and condition $(\ref{6-4})_+$ is fulfilled the error is just $O(h^\infty)$.

1375: \end{claim}

1376: 

1377: Thus I arrive to asymptotics with the principal part

1378: \begin{equation}\label{8-25}

1379: (2\pi h)^{-1}\int {\rm e}_0(x_1,x_1,0;x_2,\xi_2)\psi (x)\,d\xi_2 dx

1380: \end{equation}

1381: and remainder estimates described in Theorem \ref{thm-8-10} below:

1382: 

1383: \begin{theorem}\label{thm-8-10} Let condition $(\ref{8-1})$ be fulfilled. Then

1384: 

1385: \smallskip

1386: \noindent

1387: (i) As $l\ge 1$ and condition $(\ref{6-4})_+$ is fulfilled asymptotics with the principal part given by $(\ref{8-25})$ holds with the remainder estimate 

1388: $O(\mu h^\nu)$;

1389: 

1390: \smallskip

1391: \noindent

1392: (ii) As \underline{either} $l\ge 1$ and condition $(\ref{6-4})_-$ is fulfilled \underline{or} $l=0$, $\nu$ is even and condition $(\ref{6-4})_+$ is fulfilled asymptotics with the principal part given by $(\ref{8-25})$ holds with the remainder estimate $O(1)$;

1393: 

1394: \smallskip

1395: \noindent

1396: (iii) Furthermore, as $l=0$, $\nu$ is odd and condition $(\ref{6-4})_+$ is fulfilled the same asymptotics holds with the remainder estimate $O(h^\infty)$.

1397: \end{theorem}

1398: 

1399: \subsection{Calculations. II}

1400: 

1401: In this subsection I give in more explicit form the principal parts of asymptotics already derived under condition (\ref{6-5}). Basically I need

1402: to reconsider only the external formerly elliptic zones described by 

1403: (\ref{8-14})--(\ref{8-17}). The analysis in the first of them is not different from the analysis under condition \ref{6-4-pm}; analysis in the second one repeats the proof of theorem \ref{thm-6-17}; analysis in two latter is rather obvious. Thus I arrive to two following theorems:

1404: 

1405: \begin{theorem}\label{thm-8-11} Let conditions $(\ref{6-5})$ and $(\ref{8-6})$ be fulfilled. Then asymptotics with the principal part $(\ref{8-19})$

1406: holds with the remainder estimate $O(1)$.

1407: \end{theorem}

1408: 

1409: \begin{theorem}\label{thm-8-12} Let conditions $(\ref{6-5})$ and $(\ref{8-6})$ be fulfilled. Then

1410: 

1411: \smallskip

1412: \noindent

1413: (i) As $l\ge 1$ estimate

1414: \begin{multline}

1415: \cR^{**}\Def|\int \biggl( e(x,x, 0) -

1416: (2\pi h)^{-1} \int {\rm e}_0(x_1,x_1,0;x_2,\xi_2)\,d\xi_2 \\-\cE^\MW (x,0) + \cE^\MW_0 (x,0)\biggr)

1417: \psi (x)\,dx -

1418: \sum \kappa_{l,m} \mu h^{1+2m} | \le C\mu h^\nu

1419: \label{8-26}

1420: \end{multline}

1421: holds;

1422: 

1423: \smallskip

1424: \noindent

1425: (ii) As $l=0$ estimate

1426: \begin{multline}

1427: |\int \biggl( e(x,x, 0) -

1428: (2\pi h)^{-1} \int {\rm e}_0(x_1,x_1,0;x_2,\xi_2)\,d\xi_2 \\-\cE^\MW (x,0) + \cE^\MW_0 (x,0)\biggr)

1429: \psi (x)\,dx  | \le C

1430: \label{8-27}

1431: \end{multline}

1432: holds.

1433: \end{theorem}

1434: 

1435: \appendix\sect{Appendix: Eigenvalues of 1D operators}

1436: 

1437: \subsection{General observations}

1438: 

1439: In this Appendix $\lambda_n(\eta)$ ($n=0,1,\dots$) denote eigenvalues of one-dimensional pilot-model Schr\"odinger operators with $\mu=h=1$

1440: \begin{align}

1441: &{\bf a}^0(\eta) = D ^2 +(\eta - x^\nu/\nu)^2 - (2l+1) x^{\nu-1}

1442: \label{A-1}\\

1443: \intertext{or more general operator}

1444: &{\bf a} (\eta) =(1+\alpha_1 x+ \beta_1^2x^2) D ^2 +

1445: (1+\alpha_2 x+ \beta_2^2x^2)(\eta - x^\nu/\nu)^2 -\label{A-2} \\

1446: &\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad(2l+1) (1+\alpha_3 x)x^{\nu-1}

1447: \notag

1448: \end{align}

1449: with $\nu=2,3,\dots$ and $\beta_j> \alpha_j^2/2$.

1450: 

1451: One can prove easily the following   statement:

1452: 

1453: \begin{proposition}\label{prop-A-1} Let $l\in \bR$. Then

1454: 

1455: \smallskip

1456: \noindent

1457: (i) As $|\eta |\le C_0$ the spacing between two consecutive eigenvalues $\lambda_n$ and $\lambda_{n+1}$ with $n\le c_0$ is

1458: $\asymp 1$;

1459: 

1460: \smallskip

1461: \noindent

1462: (ii) For  operator $(\ref{A-1})$ with odd $\nu$ \ $\lambda_n(-\eta)=\lambda_n(\eta)$;

1463: 

1464: \smallskip

1465: \noindent

1466: (iii) For even $\nu$ and $\eta \le 0$ \

1467: $\lambda_n (\eta)\ge (1-\epsilon)\eta^2 - C_1$ \; $\forall n=0,1,\dots$.

1468: \end{proposition}

1469: 

1470: However, the case of even $\nu$ and $\eta\to -\infty$ is rather exceptional:

1471: 

1472: \begin{proposition}\label{prop-A-2} As

1473: $\eta \ge C_0$ (and thus also as $\eta \le -C_0$ and $\nu$ is odd)

1474: 

1475: \smallskip

1476: \noindent

1477: (i) The spacing between eigenvalues with $n\le c_0$ is

1478: $\asymp (1+|\eta|)^{(\nu-1)/\nu}$;

1479: 

1480: \smallskip

1481: \noindent

1482: (ii) As $n< l$ \ ($l< n \le c_0$)

1483: $\lambda_n(\eta)$ is less than (greater than respectively)

1484: \newline $\epsilon (n-l) (1+|\eta|)^{(\nu-1)/\nu}$\,\footnote{\label{foot-?}

1485: Thus leaving the special case $n=l\in \bZ^+$ for the further analysis.}.

1486: \end{proposition}

1487: 

1488: \begin{proof} Proof follows from the proof of proposition \ref{prop-A-3} below.

1489: 

1490: \end{proof}

1491: 

1492: \subsection{Asymptotic behavior of $\lambda_l(\eta)$ as

1493: $\eta\to \infty$ as $l\ge1$}

1494: 

1495: In this subsection I prove

1496: 

1497: \begin{proposition}\label{prop-A-3} (i) For operator $(\ref{A-1})$ with $l\ge 1$ as $\eta \to +\infty$

1498: (and thus also as $\eta \to -\infty$ and $\nu$ is odd)

1499: \begin{equation}

1500: \lambda_l(\eta)= \kappa \eta^{-2/\nu}+ O\bigl( \eta^{-(\nu+3)/\nu}\bigr)

1501: \label{A-3}

1502: \end{equation}

1503: with $\kappa > 0$;

1504: 

1505: \smallskip

1506: \noindent

1507: (ii) For operator $(\ref{A-2})$ with $l\ge 1$ as $\eta \to +\infty$

1508: (and thus as $\eta \to -\infty$ and $\nu$ is odd)

1509: \begin{equation}

1510: \partial_{\alpha_j} \lambda_l(\eta)\bigr|_{\alpha = \beta =0} =

1511: \kappa_j \eta + O(\eta^{-1/\nu})

1512: \label{A-4}

1513: \end{equation}

1514: with $\kappa_1=\kappa_2=-\kappa_3/2$, $\alpha=(\alpha_1,\alpha_2,\alpha_3)$,

1515: $\beta=(\beta_1,\beta_2,\beta_3)$ and furthermore

1516: \begin{equation}

1517: \sum_{1\le j\le 3}\partial_{\alpha_j} \lambda_l(\eta)\bigr|_{\alpha = \beta =0} =

1518: \kappa_4 \eta ^{1/\nu}\lambda_l + O(\eta^{-2/\nu}).

1519: \label{A-5}

1520: \end{equation}

1521: \end{proposition}

1522: 

1523: \begin{proof} (i) Let us plug $\eta =\gamma^\nu/\nu$ with $\gamma\gg 1$ where in the case even $\nu$ this is the only scenario and in the case of odd $\nu$ analysis of scenario $\xi_2=-\gamma^\nu/\nu$ is done by the symmetry. Then after shift

1524: $x\mapsto x+\gamma$ operator ${\bf a}^0(\eta)$ is transformed into operator

1525: \begin{multline*}

1526: D^2 + x^2\Bigl( \gamma^{\nu-1}+ {\frac 1 2}(\nu-1)x \gamma^{\nu-2}+

1527: {\frac 1 6}(\nu-1)(\nu-2)x^2\gamma^{\nu-3}+ \dots \Bigr)^2\\

1528: - (2l+1)\Bigl(\gamma^{\nu-1} + (\nu-1)x \gamma^{\nu-2}+

1529: {\frac 1 2}(\nu-1)(\nu-2) x^2\gamma^{\nu -3} +\dots\Bigr)\

1530: \end{multline*}

1531: and after rescaling $x\mapsto x\gamma^{(1-\nu)/2}$ this operator is transformed into $\gamma^{\nu-1}{\bf b}_\varepsilon$ where

1532: \begin{multline*}

1533: {\bf b}_\varepsilon= D^2 + x^2\Bigl( 1+

1534: {\frac 1 2}(\nu-1)x \varepsilon +

1535: {\frac 1 6}(\nu-1)(\nu-2)x^2\varepsilon^2 + \dots \Bigr)^2\\

1536: - (2l+1)\Bigl(1 + (\nu-1)x \varepsilon+

1537: {\frac 1 2}(\nu-1)(\nu-2) x^2\varepsilon^2 +\dots \Bigr)

1538: \end{multline*}

1539: with $\varepsilon=\gamma^{-(\nu+1)/2}$. Then

1540: \begin{multline*}

1541: {\bf b}_\varepsilon= \underbrace{D^2 + x^2 -(2l+1)}_{{\bf h}_0}

1542: +\varepsilon \underbrace{(\nu-1) \Bigl( x^3 -(2l+1)x\Bigr)}_{{\bf h}_1} +\\

1543: \varepsilon^2\underbrace{(\nu-1)\Bigl( ({\frac 7 {12}}\nu-{\frac {11}{12}}) x^4 -{\frac 1 2}(2l+1)(\nu-2)x^2\Bigr)}_{{\bf h}_2} +O\bigl(\varepsilon^3\bigr)

1544: \end{multline*}

1545: and let us denote by $\Lambda_\varepsilon$ and $U_\varepsilon$ its eigenvalue close to $0$ and the corresponding eigenfunction. Then

1546: \begin{equation}

1547: \Lambda _\varepsilon = \omega_1\varepsilon +\omega_2 \varepsilon^2 +\dots\qquad\text{and}\qquad

1548: U_\varepsilon = u _0 + u_1\varepsilon + u_2\varepsilon^2 \dots

1549: \label{A-6}

1550: \end{equation}

1551: where obviously $u_0=\upsilon_l$ is a Hermite function, $\omega_1=\omega_3=\dots=0$

1552: and

1553: \begin{equation}

1554: {\bf h}_0 u_1 + {\bf h}_1 u_0=0\qquad

1555: {\bf h}_0 u_2 +{\bf h}_1u +{\bf h}_2u_0= \omega_2 u_0.

1556: \label{A-7}

1557: \end{equation}

1558: Then

1559: \begin{equation}

1560: \omega_2 = \langle {\bf h}_1u +{\bf h}_2u_0, u_0\rangle=

1561: -\langle u , {\bf h}_0u\rangle +\langle {\bf h}_2u_0, u_0\rangle.

1562: \label{A-8}

1563: \end{equation}

1564: It is known that $(x-iD)\upsilon_k=(2k+2)^{1/2}\upsilon_{k+1}$,

1565: $(x+iD)\upsilon_k=(2k)^{1/2}\upsilon_{k-1}$ and therefore

1566: \begin{align*}

1567: &x \upsilon_l = &&{\frac 1 2}\Bigl(

1568: (2l+2)^{1/2}\upsilon_{l+1} + (2l)^{1/2}\upsilon_{l-1}\Bigr),\\

1569: &x^2 \upsilon_l = &&{\frac 1 4}\Bigl(

1570: (2l+2)^{1/2}(2l+4)^{1/2}\upsilon_{l+2} + (4l+2)\upsilon_l +

1571: (2l)^{1/2}(2l-2)^{1/2}\upsilon_{l-2}\Bigr),\\

1572: &\bigl(x ^2-2l-1)\upsilon_l = &&{\frac 1 4}\Bigl(

1573: (2l+2)^{1/2}(2l+4)^{1/2}\upsilon_{l+2} -

1574: 2(2l+1) \upsilon_l+

1575: (2l)^{1/2}(2l-2)^{1/2}\upsilon_{l-2}\Bigr),\\

1576: &x\bigl(x ^2-2l-1)\upsilon_l = &&{\frac 1 8}\Bigl(

1577: (2l+2)^{1/2}(2l+4)^{1/2}(2l+6)^{1/2}\upsilon_{l+3} -

1578: (2l+2)^{1/2}(2l-2)\upsilon_{l+1}-\\

1579: &&&\hphantom{ {\frac 1 2}\Bigl( }(2l)^{1/2}(2l+4)\upsilon_{l-1}+

1580: (2l)^{1/2}(2l-2)^{1/2}(2l-4)^{1/2}\upsilon_{l-3}\Bigr),

1581: \end{align*}

1582: which imply

1583: \begin{multline*}

1584: \hskip-\labelsep\langle {\bf h}_0 u,u\rangle = \\{\frac 1 {64}}(\nu-1)^2\Bigl( {\frac 1 6}(2l+2)(2l+4)(2l+6) + {\frac 1 2} (2l+2)(2l-2)^2 -{\frac 1 2} (2l)(2l+4)^2 -{\frac 1 6} (2l) (2l-2)(2l-4)\Bigr)=\\

1585: {\frac 1 {16}} (\nu-1)^2 \Bigl(-2l^2-2l+3\Bigr).

1586: \end{multline*}

1587: On the other hand

1588: \begin{multline*}

1589: \langle {\bf h}_2u_0,u_0\rangle = (\nu-1)\Bigl(

1590: ({\frac 7 {12}}\nu-{\frac {11}{12}}) \|x^2u_0\|^2 -

1591: {\frac 1 2}(\nu-2)(2l+1)\|xu_0\|^2\Bigr)= \\

1592: (\nu-1)({\frac 7 {12}}\nu-{\frac {11}{12}}){\cdot}{\frac 1 {16}}\Bigl(

1593: (2l+2)(2l+4)+(4l+2)^2+(2l)(2l-2)\Bigr) -\\

1594: {\frac 1 4}(\nu-1)(\nu-2)\cdot (2l+1)^2=\\

1595: (\nu-1)(7\nu-11){\cdot}{\frac 1 {16}}\Bigl(2l^2+2l+1\Bigr) -

1596: {\frac 1 4}(\nu-1)(\nu-2)(2l+1)^2

1597: \end{multline*}

1598: and

1599: \begin{multline*}

1600: \omega_2= {\frac 1 {16}} (\nu-1)\biggl( (7\nu-11)\bigl(2l^2+2l+1\bigr) -

1601: 4(\nu-2)\bigl(4l^2+4l+1\bigr)- (\nu-1) \bigl(-2l^2-2l+3\bigr)\biggr)=\\

1602: {\frac 1 2} (\nu-1)l (l+1),

1603: \end{multline*}

1604: Therefore $\Lambda _\varepsilon = \omega_2 \varepsilon^2+O(\varepsilon^4)$ as $\varepsilon \to 0$ (because $\omega_3=0$ as well) which implies statement (i) with $\kappa =\omega_2\nu^{-2/\nu}$.

1605: 

1606: \smallskip

1607: \noindent

1608: (ii) After obvious transformations

1609: \begin{equation*}

1610: \partial_{\alpha_j} \lambda_l(\eta)\bigr|_{\alpha = \beta =0} =

1611: \gamma^{\nu-1} \langle {\bf k}_j U_\varepsilon, U_\varepsilon\rangle

1612: \end{equation*}

1613: with

1614: \begin{align*}

1615: &{\bf k}_1 = (\gamma + \varepsilon x )D^2,\\

1616: &{\bf k}_2= x^2\Bigl( 1+

1617: {\frac 1 2}(\nu-1)x \varepsilon +

1618: {\frac 1 6}(\nu-1)(\nu-2)x^2\varepsilon^2 + \dots \Bigr)^2\\

1619: &{\bf k}_3=- (2l+1)\Bigl(1 + (\nu-1)x \varepsilon+

1620: {\frac 1 2}(\nu-1)(\nu-2) x^2\varepsilon^2 +\dots \Bigr)

1621: \end{align*}

1622: and therefore

1623: \begin{equation*}

1624: \langle {\bf k}_j U_\varepsilon, U_\varepsilon\rangle = \gamma

1625: \langle {\bf k}'_j u_0,u_0\rangle +O\bigl(\varepsilon^2\gamma\bigr)

1626: \end{equation*}

1627: with ${\bf k}'_1=D^2$, ${\bf k}'_2=x^2$, ${\bf k}'_3=-(2l+1)$

1628: which implies (\ref{A-4}). 

1629: 

1630: Known equalities

1631: $\langle x^2\upsilon_l,\upsilon_l\rangle =

1632: \langle D^2\upsilon_l,\upsilon_l\rangle = (2l+1)/2$ imply that

1633: $\kappa_1=\kappa_2=-\kappa_3/2$. Further,

1634: $\sum_{1\le j\le 3} \langle {\bf k}_jU_\varepsilon, U_\varepsilon\rangle

1635: = \gamma \lambda_l +O(\varepsilon^2)$ which implies (\ref{A-5}).

1636: \end{proof}

1637: 

1638: \subsection{More general operators}

1639: 

1640: Now I consider operator

1641: \begin{equation}

1642: \cA (y,\eta)\Def \beta\Bigl(\alpha h^2D^2 \alpha + \alpha ^{-2}(\eta -\mu x^\nu/\nu)^2 -(2l+1)\mu h x^{\nu-1}\Bigr)\beta

1643: \label{A-9}

1644: \end{equation}

1645: with

1646: \begin{equation}

1647: \alpha =\alpha (x,y),\quad \beta=\beta(x,y) ,\quad \alpha (0,y)=1,\quad

1648: c_0^{-1}\le \beta \le c_0.

1649: \label{A-10}

1650: \end{equation}

1651: Let $\lambda_n$ be eigenvalues of $\cA$. Changing 

1652: $x\mapsto \gamma (\mu^{-1}h)^{1/(\nu+1)}x$ and

1653: $\eta \mapsto (\mu h^\nu)^{1/(\nu+1)}$ respectively  I arrive to operator (\ref{A-9})  again  with

1654: $\mu = h=1$ and $\alpha$, $\beta$ replaced by

1655: $\alpha ((\mu^{-1}h)^{1/(\nu+1)}x, y)$, $\beta ((\mu^{-1}h)^{1/(\nu+1)}x, y)$ and with a factor  $(\mu h^\nu)^{2/(\nu+1)}$.

1656: 

1657: \begin{proposition}\label{prop-A-4} Let conditions $(\ref{A-9}),(\ref{A-10})$ be fulfilled. Then

1658: 

1659: \smallskip

1660: \noindent

1661: (i) $\lambda_n(\eta)\ge C_0(\mu h^\nu)^{2/(\nu+1)}$ as $n\ge C$;

1662: 

1663: \smallskip

1664: \noindent

1665: (ii) As $|\eta |\le C_0(\mu h^\nu)^{2/(\nu+1)}$ the spacing between consecutive eigenvalues with $n\le c_0$ is $\asymp (\mu h^\nu)^{2/(\nu+1)}$ and

1666: \begin{equation}

1667: |\partial_y^p \partial_\eta ^q\lambda_n(y,\eta)|\le C_{pq} (\mu^{-1}h)^{p/(\nu+1)}

1668: (\mu h^\nu)^{(2-q)/(\nu+1)};

1669: \label{A-11}

1670: \end{equation}

1671: 

1672: \smallskip

1673: \noindent

1674: (iii) For even $\nu$ and $\eta \le 0$ \quad

1675: $\lambda_n(y,\eta)\ge (1-\epsilon) \eta^2 -C_1$, $n=0,1,\dots$

1676: \end{proposition}

1677: 

1678: \begin{proposition}\label{prop-A-5}

1679: As $\eta \ge C_0(\mu h^\nu)^{1/(\nu+1)}$ (and thus also as

1680: $\eta \le -C_0(\mu h^\nu)^{1/(\nu+1)}$ and  $\nu$ is odd)

1681: 

1682: \smallskip

1683: \noindent

1684: (i) The spacing between eigenvalues with $n\le c_0$ is

1685: $\asymp |\eta|^{(\nu-1)/\nu} (\mu h^\nu)^{1/\nu}$;

1686: 

1687: \smallskip

1688: \noindent

1689: (ii) As $n<l$ \ ($l<n \le c_0$) $\lambda_n(y,\eta)$ is less than (greater than respectively) \newline

1690: $\epsilon (n-l)\bigl((\mu h^\nu)^{2/(\nu+1)} + |\eta|^{(\nu-1)/\nu} (\mu h^\nu)^{1/\nu}\bigr)$ 

1691: and these eigenvalues satisfy

1692: \begin{equation}

1693: |\partial_y^p \partial_\eta ^q\lambda_n(y,\eta)|\le C_{pq}(\mu^{-1}h)^{p/(\nu+1)}

1694: |\eta |^{-q}|\lambda_n(y,\eta)|;

1695: \label{A-12}

1696: \end{equation}

1697: 

1698: \smallskip

1699: \noindent

1700: (iii) As $\eta \ge C_0(\mu h^\nu)$ (and thus as

1701: $\eta \le -C_0(\mu h^\nu)$ and  $\nu$ is odd) 

1702: $|\lambda_l(y,\eta)|\le \epsilon_0$.

1703: \end{proposition}

1704: 

1705: An extra analysis is needed for our purposes as $n=l$ and

1706: \begin{equation}

1707: \mu h^\nu \ge C_1

1708: \label{A-13}

1709: \end{equation}

1710: with arbitrarily large $C_1$.

1711: 

1712: \begin{proposition}\label{prop-A-6}

1713: Let condition $(\ref{A-13})$ be fulfilled and $l\ge 1$. Then

1714: as $\eta \ge C_0(\mu h^\nu)^{1/(\nu+1)}$ 

1715: \begin{equation}

1716: \lambda_l (y,\eta)\asymp (\mu h^\nu /\eta)^{2/\nu}\qquad\text{and}\qquad

1717: \eta \partial_\eta \lambda_l (y,\eta)\asymp (\mu h^\nu /\eta)^{2/\nu}.

1718: \label{A-14}

1719: \end{equation}

1720: \end{proposition}

1721: 

1722: \subsection{Case of $\lambda_l$ as $l=0$}

1723: 

1724: Here cases of  odd and even $\nu$ differ drastically. Note first that

1725: \begin{equation}

1726: {\bf a}^0 (\eta)= \bigl(i D + \xi_2 - x^\nu/\nu\bigr)

1727: \bigl(-i D + \xi_2 - x^\nu/\nu\bigr)

1728: \label{A-15}

1729: \end{equation}

1730: and as $\nu$ is odd operator ${\bf a}^0(\eta)$ has the bottom eigenvalue $\lambda_0(\eta)$ with eigenfunction defined from

1731: $\bigl(-\partial + \xi_2 - x^\nu/\nu\bigr)v=0$ i.e. $v=\exp (\xi_2 x - x^{\nu+1}/\nu(\nu+1))$ and therefore $\lambda_0(\eta)$ is identically $0$.

1732: 

1733: Similarly, as $\beta=1$ operator $\cA$ defined by (\ref{A-9}) is equal modulo $O(h^2)$ to operator

1734: \begin{multline}

1735: \cB(y,\eta)\Def h^2 \alpha ^2 D +\alpha^{-2} (\eta -\mu x^\nu/\nu)^2- \mu h x^{\nu-1}=\\

1736: \bigl(ihD \alpha +\alpha^{-1} (\eta -\mu x^\nu/\nu)\bigr)

1737: \bigl(-\alpha ihD +\alpha^{-1} (\eta -\mu x^\nu/\nu)\bigr)

1738: \label{A-16}

1739: \end{multline}

1740: and I arrive to the statement (i) of

1741: 

1742: \begin{proposition}\label{prop-A-7} (i) For odd $\nu$ the bottom eigenvalue of $\cB(y,\eta)$ is $0$;

1743: 

1744: \smallskip

1745: \noindent

1746: (ii) For even $\nu$ the bottom eigenvalue of $\cB(y,\eta)$ is

1747: $(\mu h^\nu)^{2/(\nu+1)} \Lambda \bigl(y,\eta (\mu h^\nu)^{-1/(\nu+1)}\bigr)$

1748: where

1749: \begin{align}

1750: &C^{-1}\exp (-C\eta^{(\nu+1)/\nu}) \le \Lambda (y,\eta) \le

1751: C \exp (-\epsilon \eta^{(\nu+1)/\nu}),\label{A-17}\\

1752: &\epsilon \eta ^{1/\nu}\le -\partial_\eta \bigl(\log \Lambda (y,\eta)\bigr) \le C \eta^{1/\nu}.

1753: \label{A-18}

1754: \end{align}

1755: \end{proposition}

1756: 

1757: \begin{proof} I need to consider the case of even $\nu$ only. The same representation (\ref{A-15}) shows that $\lambda_0 (y,\eta)>0$. However, since this eigenfunction is fast decaying outside of the potential well, one can do

1758: the same shift and rescaling as before and using arguments of \cite{HeMa} to prove that

1759: $\Lambda_0 (y,\eta) \sim k\exp (-k_2 \eta^{(\nu+1)/\nu})$. Also one can prove easily

1760: that $\partial_\eta \Lambda_0 (y,\eta)\sim -k_3 \eta ^{1/\nu} k\exp (-k_2 \eta^{(\nu+1)/\nu})$ as $\eta \ge C$ with $k_3=kk_2(1+\nu)/\nu$. Estimates 

1761: (\ref{A-17}), (\ref{A-18}) follow from this.

1762: \end{proof}

1763: 

1764: \input IRO7.bbl

1765: \end{document}

1766: 

1767: