1: \documentclass[12pt]{amsart}

2: \usepackage{amsmath,amssymb,amsthm,epsfig}

3: \textheight=22cm

4: \textwidth=14cm

5: \voffset=-1cm

6: \hoffset=-1cm

7: \def\torus{\mathbb T}

8: \def\real{\mathbb R}

9: \def\complex{\mathbb C}

10: \def\integer{\mathbb Z}

11: \def\const{\text{const}}

12: \def\rk{\qopname\relax o{rk}}

13: \def\Im{\qopname\relax o{Im}}

14: \def\Iso{\qopname\relax o{Iso}}

15: \newtheorem{theorem}{Theorem}

16: \newtheorem{lemma}{Lemma}

17: \theoremstyle{remark}

18: \newtheorem{ex}{Example}

19: \newtheorem{case}{Case}

20: \newtheorem{conjecture}{Conjecture}

21: \newtheorem{remark}{Remark}

22: \author{I.\,Dynnikov}

23: \author{ S.\,Novikov}

24: \address{Dept.\ of Mech.\ and Math., Moscow State University,

25: Moscow 119992 GSP-2, Russia}

26: \email{dynnikov@mech.math.msu.su}

27: \address{Landau Institute for theoretical physics, Kosygina str.~2,

28: Moscow 119334, Russia; IPST, University of Maryland, College Park,

29: MD 20742, USA}

30: \email{novikov@glue.umd.edu}

31: \title{Topology of quasiperiodic functions on the plane}

32: \thanks{The work of I.\,Dynnikov was supported in part by

33: Russian Foundation for Basic Research (grant no.~02-01-00659);

34: the work of S.\,Novikov was supported in part by

35: the Council of the Russian Academy of Science (grant ``Mathematical

36: methods of nonlinear dynamics'')}

37: 

38: 

39: 

40: \begin{document}

41: 

42: \maketitle

43: 

44: 

45: \begin{abstract}This article describes a topological theory of

46: quasiperiodic functions on the plane. The development of

47: this theory was started (in different

48: terminology) by the Moscow topology group in early 1980s.

49: It was motivated by the needs of solid state physics, as a partial

50: (nongeneric) case of Hamiltonian foliations of  Fermi

51: surfaces with multivalued Hamiltonian function~\cite{N}. The

52: unexpected discoveries of their topological properties that were made in

53: 1980s \cite{Z,N1} and 1990s \cite{D,D',D1} have finally led to

54: nontrivial physical conclusions \cite{NM,NM1} along the lines of

55: the so-called geometric strong magnetic field limit

56: \cite{LAK}. A very fruitful new point of view comes from

57: the reformulation of that problem

58: in terms of quasiperiodic functions and an extension to higher dimensions

59: made in 1999 \cite{N2}. One may say that,

60: for single crystal normal metals put in a magnetic field,

61: the semiclassical trajectories of electrons in the space of

62: quasimomenta are exactly the level lines of the quasiperiodic function with

63: three quasiperiods that is the dispersion relation restricted to a

64: plane orthogonal to the magnetic field. General studies of the

65: topological properties of levels of quasiperiodic functions on the

66: plane with any number of quasiperiods were started in 1999 when

67: certain ideas were formulated for the case of four quasiperiods

68: \cite{N2}. The last section of this work contains a complete

69: proof of these results based on the technique developed in \cite{D2,D3}.

70: Some new physical applications of the general problem were found

71: recently \cite{M}.

72: \end{abstract}

73: 

74: \section{Quasiperiodic functions}

75: 

76: Let $\torus^n=\real^n/\integer^n$ denote the

77: $n$-dimensional torus,

78: $\nu\!:\real^n\rightarrow\torus^n=\real^n/\integer^n$ the standard

79: projection.

80: 

81: We say that a real smooth function $\varphi(y)=\varphi(y^1,\dots,y^k)$

82: on the $k$-plane $\real^k$ is \emph{quasiperiodic} with $n$

83: quasiperiods (frequencies) if it can be represented in the form

84: $\varphi(y)=f(x(y))$:

85: \begin{equation}

86: \varphi=f\circ\nu\circ \iota,

87: \end{equation}

88: where $\iota:\real^k\rightarrow\real^n$ is an affine imbedding:

89: $$x^s=a^s_ry^r+x^s_0,$$

90: $f=f(x):\torus^n\rightarrow\real$ is some

91: smooth function, and $n\geqslant 2$ is the minimal possible integer for

92: which such a function $f$ and an affine imbedding $\iota$ exist. Here

93: $s=1,\dots,n$ and $r=1,\dots,k$. In the theory of quasicrystals, people call

94: the space $\real^k$ (where $k=2,3$) the

95: \emph{physical space} and the space $\real^n$

96: the \emph{superspace}. Every $n$-periodic function $f(x)$ generates a

97: family of \emph{descendants}, which are obtained by varying

98: the initial vector

99: $x_0=(x^1_0,\dots,x^n_0)$ in the superspace $\real^n$.

100: Any two descendants $\varphi_1(y)$, $\varphi_2(y)$

101: of the same function $f$ are said to be

102: \emph{related}. They have the same frequencies

103: and obtain the following property: for any $\varepsilon>0$,

104: there is a shift $y\mapsto y+a$ in the physical space

105: such that the shifted function

106: $\varphi_2(y+a)$ is $\varepsilon$-close to $\varphi_1(y)$:

107: $$|\varphi_2(y+a)-\varphi_1(y)|<\varepsilon\qquad\forall y\in\real^k.$$

108: 

109:  Any

110: linear function $\lambda:\real^k\rightarrow\real$ of the form

111: $\lambda(y)=\ell(x(y))$ or $\lambda=\ell\circ \iota$, where the

112: linear function $\ell:\real^n\rightarrow\real$ belongs to the dual

113: (or ``reciprocal'') lattice $(\integer^n)^*$ (\emph{i.e.}, we have

114: $\ell(\integer^n)\subset\integer$) is called a \emph{frequency} of

115: $\varphi$. The set of all frequencies form a free abelian group

116: with $n$ natural  generators

117: $\lambda_1=\ell^1\circ\iota,\dots,\lambda_1=\ell^n\circ\iota$ where

118: the functions

119: $\ell^s(x)=x^s$, $s=1,\ldots,n$, are dual to

120: the basic periods. We call this group

121: the \emph{group of frequencies}. It is a free

122: abelian subgroup $\Gamma^*$ of the dual vector space $\real^*$,

123: and it is the

124: same for the whole family of related quasiperiodic functions (descendants of the

125: same $n$-periodic function $f$).

126: 

127: Analytically, any $n$-periodic function can be presented in the form

128: of a trigonometric series

129: 

130: $$f(x)=\sum_{\ell\in(\integer^n)^*} c_\ell\exp\big(2\pi

131: i\ell(x)\big)$$

132: 

133: Therefore, any  quasiperiodic function can be presented in a

134: similar form:

135: $$\varphi(y)=\sum_{\lambda\in\Gamma^*}b_{\lambda}\exp\big(2\pi

136: i\lambda(y)\big)= \sum_mb_m\exp\Big(2\pi i

137: \sum_{s=1}^nm_s\lambda_s(y)\Big),$$

138: where $m=(m_1,\dots,m_n)

139: \in\integer^n$. By definition, the set of basic frequencies

140: $\lambda_s$ generates the space $\real^k$ over the field $\real$.

141: 

142: For the space $\real^k$ endowed with a Euclidean metric there is a natural

143: identification $\real^k\cong(\real^k)^*$, so the subgroup of

144: frequencies can be treated as a subgroup $\Gamma\cong\Gamma^*\subset

145: \real^k$ in the physical space $\real^k$.

146: 

147: There is an affine symmetry semigroup associated with each family

148: of related quasiperiodic functions. By definition, this semigroup

149: $\widetilde{G}$ consists of all affine transformations

150: $$g:\real^k\rightarrow \real^k$$

151: of the physical space

152: such that

153: $$g(\Gamma^*)\subset\Gamma^*,$$

154: where $\Gamma^*$ is treated as

155: a subset of the group of translations: $\Gamma^*\subset \real^k$.

156: For the Euclidean space $\real^k$ we define the \emph{symmetry group}

157: $G\subset\widetilde{G}$ consisting only of isometries $g$ such

158: that $g(\Gamma^*)=\Gamma^*$.

159: 

160: This group satisfies the general definition of a

161: quasicrystallographic group introduced by S.\,Novikov and

162: A.\,Veselov in 1980s in order to answer the question: what is the

163: symmetry of quasicrystals (see~\cite{Pi})? According to that

164: definition the intersection $G\bigcap\real^k\subset\Iso(\real^k)$

165: of a quasicrystallographic group

166: with the subgroup of translations should be a finitely generated free

167: abelian group. In our case it is exactly the group $\Gamma^*$. The

168: definition allows  the ``rotational'' quotient group

169: $G/(G\bigcap \real^k)\subset O_k$ to be infinite.

170: S.\,Piunikhin studied these

171: groups for $k=2,3$ in a series of works (see \cite{Pi}).

172: 

173: \begin{ex} Consider the two-dimensional case, $k=2$. Let $\theta$ be a

174: unimodular complex number, $|\theta |=1$, $\theta=\exp(i\psi)$,

175: satisfying the equation

176: $$P(\theta)=\theta^n+a_1\theta^{n-1}+\ldots+a_{n-1}\theta+1=0,$$

177: where all coefficients are  integer-valued, $a_s\in

178: \integer$, and we have $a_s=a_{n-s}$. The complex numbers

179: (or real two-vectors)

180: $\lambda_1=1, \lambda_2=\theta,\dots, \lambda_n=\theta^{n-1}$

181: generate a group of frequencies $\Gamma^*\subset

182: \complex=\real^2$  with nontrivial rotational symmetry $g\rightarrow

183: g\exp(i\psi)$. It is easy to find such a polynomial $P$ with

184: a root at $\theta=\exp(i\psi)$, where the

185: ratio $\psi/2\pi$ is irrational. There are very complicated

186: quasicrystallographic groups for $k=3$ (see \cite{Pi}).

187: 

188: \end{ex}

189: 

190: \section{Quasiperiodic functions in analysis, geometry, and

191: dynamical systems motivated by natural sciences}

192: 

193: \subsection{Quasiperiodic functions on the real line}

194: 

195: Consider the case $k=1$. In the XIX century,

196: one-dimensional quasiperiodic functions

197: with $n$ quasiperiods appeared in the theory of

198: completely integrable Hamiltonian systems of the classical

199: mechanics with $n$ degrees of freedom. According to so-called

200: Liouville's theorem, the integrability follows from the existence of

201: $n$ smooth independent pairwise commuting integrals of motion. If their

202: common level sets are compact, then the time dependence of the space

203: coordinates along a trajectory can be described by

204: quasiperiodic functions $x^r(t)$ with (at most) $n$ quasiperiods.

205: So, all studies of perturbations of

206: completely integrable systems should start with quasiperiodic

207: unperturbed background. A lot of fundamental work has been done in

208: this area (see \cite{DKN}).

209: 

210: \subsection{Quasiperiodic functions in the theory of nonlinear PDE}

211: Completely integrable PDE systems of the theory of solitons give rise

212: to quasiperiodic functions with $k>1$. There are very famous

213: ($1+1$) PDE systems such as KdV ($u_t=6uu_x+u_{xxx}$) or sine-Gordon

214: ($u_{tt}=u_{xx}+\sin \{u(x,t)\}$), which are completely integrable by the

215: so-called \emph{inverse scattering transform method for rapidly

216: decreasing initial values}. A countable number of continuous

217: families of exact

218: smooth real ``finite-gap'' solutions of these equations were discovered

219: in 1970s (see~\cite{DKN}). These solutions are quasiperiodic

220: functions in $x,t$, and depend on many parameters $a$, $a'$:

221: $$u(x,t)=F(xU+Vt+U_0;a)$$

222: for KdV, and

223: $$\exp\big(iu(x,t)\big)=F'(U'x+V't+U'_0;a')$$ for sine-Gordon. Here

224: $u(x,t)$ is real in both cases, $F,F'$ are $n$-periodic smooth

225: functions in $n$ variables (\emph{i.e.}, smooth functions on the real

226: $n$-torus). They can be expressed through special functions, namely,

227: theta-functions of a hyperelliptic Riemann surface

228: of genus~$n$. $U$, $U'$, $V$, $V'$ are the $n$-vectors of

229: periods of some Abelian differentials of the second kind

230: (see \cite{N1}). Let us

231: mention that, for the sine-Gordon system, the function $u=1/i\,\log

232: F'$ is generically a multivalued function on the ``real''

233: $n$-torus imbedded in the complex $2n$-dimensional Jacoby torus

234: associated with a complex hyperelliptic Riemann surface. Here we

235: have $k=2$. For famous completely integrable $(2+1)$ systems

236: (like KP, and others) one comes to quasiperiodic solutions of the form

237: $u(x,y,t)$, which are quasiperiodic functions

238: in $k=3$ physical variables. When studying the

239: dependence of the solution on so-called higher times one may arrive at any

240: value of $k$.

241: 

242: \subsection{Quasiperiodic functions and quasicrystals}

243: 

244: Completely different examples come from solid

245: state physics. In 1980s a new type of 2D and 3D media was

246: discovered. People named them ``quasicrystals''. The optical

247: analysis of the location of atoms gave an evidence for the group

248: of frequencies being incompatible with an ordinary crystal structure.

249: For example,

250: for $k=2$, the observed group of frequencies $\Gamma^*$ might be generated

251: by the 5th roots of unity:

252: $$\lambda_r=\eta^r\in\Gamma^*,\quad r=0,1,2,3,\quad\eta^5=1,\quad P(\eta)=0,$$

253: where

254: $$P(z)=z^4+z^3+z^2+z+1.$$

255: 

256: Recall that our extension of the idea of symmetry allows the

257: rotational symmetry to be even infinite.

258: 

259: There are two mathematical models of quasicrystals.

260: Let us think of atoms in

261: the physical space $\real^k$ as being located in a discrete set of

262: points $x_A$ such that

263: there exists a couple of positive ``radii'' $\rho_1$, $\rho_2$ with the properties:

264: \begin{itemize}

265: \item[a.]

266: We have $|x_A-x_{A'}|\geq\rho_2$ for all pairs $A,A'$ with $A\neq

267: A'$;

268: \item[b.]

269: For every point $x\in \real^k$, there exists a point $x_{A}$

270: such that $|x_A-x|\leqslant \rho_1$.

271: \end{itemize}

272: We call this set of points

273: \emph{quasiperiodic} if the distribution $\sum_A\delta(x-x_A)$ can be

274: decomposed into a Fourier series with finitely generated free

275: abelian group of frequencies $\Gamma^*$.

276: 

277: In another model, our

278: physical space $\real^k$ is endowed with

279: a ``quasiperiodic tiling''. This means the following:

280: \begin{itemize}

281: \item[a.] The space is

282: covered by countably many polytopes $P_B$,

283: $\real^k=\bigcup_B P_B$, where $P_B\bigcap P_{B'}$ is a face for any pair $B,B'$.

284: \item[b.]

285: Up to shift, there is only a finite number of different polytopes

286: $P_1,\dots,P_q$ among them.

287: \item[c.]

288: Let us associate some constant $c_q$ with every polytope $P_q$

289: and consider a function that is equal to $c_j$ everywhere

290: in the interior of any $P_B$ obtained from $P_j$ by a shift. We obtain a

291: piecewise constant function $c(x)$ in $x\in \real^k$ defined

292: (at a full measure set)

293: by our tiling and the choice of the constants $c_j$.

294: The tiling is said to be \emph{quasiperiodic} if, for every

295: choice of constants $c_j$, the function $c(x)$ is quasiperiodic,

296: \emph{i.e.}, can be presented in the form of a

297: trigonometric series with finitely generated free abelian group $\Gamma^*$ of

298: frequencies.

299: \end{itemize}

300: 

301: 

302: There is a famous tiling of

303: the plane $\real^2$ by rhombi of two types: one with

304: angles $\pi/5$ and $4\pi/5$,

305: and the other with angles $2\pi/5$ and $3\pi/5$.

306: It is called the Penrose tiling.

307: This tiling is quasiperiodic, which was discovered a few years later

308: after Penrose's original work

309: (see the history and details of the subject in~\cite{Pi}). An

310: interesting idea of ``local rules'' was developed by

311: physicists and mathematicians in order to explain the growth of

312: quasicrystals in terms of tilings. In this model, the atoms are

313: located at the vertices of the tiles.

314: 

315: Both models can be obtained from the following construction. Let a

316: ``superlattice'' $\Gamma$ of full rank be given

317: in the superspace $\real^n$, and the

318: superspace be presented as the direct sum

319: $\real^n=\real^k\bigoplus\real^{n-k}$,

320: where $\real^k$ is the physical subspace.

321: Let

322: $p:\real^n\rightarrow \real^k$ and $q:\real^n\rightarrow

323: \real^{n-k}$ be the natural projections.

324: Fix a

325: finite $(n-k)$-polyhedron $D\subset

326: \real^{n-k}$ and consider the ``tubular $D$-neighborhood''

327: $D_q=q^{-1}(D)\subset \real^n$ of the physical subspace

328: $\real^k\subset\real^n$.

329: Assume that the boundary of the polyhedron $D$ is disjoint from

330: $q(\Gamma)$, or equivalently, $\partial D_q\cap\Gamma=\varnothing$.

331: Then the set of points

332: $$p(\Gamma\cap D_q)\subset\real^k$$

333: is quasiperiodic in the sense of the definition given above.

334: 

335: By taking a certain polytope

336: decomposition of the space $\real^n$ associated with the lattice

337: $\Gamma$ and the polyhedron $D$, one obtains a quasiperiodic

338: tiling of $\real^k$

339: whose tiles are the intersections of $\real^k$ with

340: the $n$-cells of the decomposition (see survey

341: article \cite{Pi}).

342: 

343: Very interesting examples of nontrivial

344: symmetry groups come from the superspace $\real^4$ endowed with

345: the Minkovski metric and

346: a lattice $\Gamma\cong\integer^4$ such that the

347: physical subspace $\real^2$, which is spacelike

348: (\emph{i.e.}\ Euclidean), is

349: invariant under some lattice-preserving mapping from the group

350: $O(3,1)$.

351: 

352: The superspaces $\real^{l,m}$, where $l+m=n$, might also appear in

353: interesting cases.

354: 

355: \subsection{Quasiperiodic functions in the theory of conductivity}

356: 

357: Here we describe the situation that is the main motivation for

358: our topological and dynamical theory.

359: 

360: For every single

361: crystal normal metal, we have a lattice $\Gamma$ in the physical

362: space $\real^3$. However, our geometrical constructions will live in

363: a completely different space, namely the $3$-torus of

364: quasimomenta $\torus^3$, which

365: is the quotient space of the \emph{dual} $3$-space $(\real^3)^*\cong\real^3$

366: by the dual

367: (reciprocal) lattice $\Gamma^*\cong\integer^3$. The ``Bloch'' states of quantum

368: electrons are parameterized by pairs $(m,p)$,

369: where $p$ is a point in the space of

370: quasimomenta, $p\in \torus^3=\real^3/\Gamma^*$, and $m$ is a

371: natural number, which is the index of

372: a branch of the dispersion relation

373: $f(p)=\epsilon_m(p):\torus^3\rightarrow \real$. In what follows

374: we will always deal with just one branch only, so we

375: drop the index $m$ in the notation. We assume that $f(p)=\epsilon(p)$ is a Morse

376: function on the $3$-torus or, in other words,

377: a three-periodic Morse function on the

378: covering Euclidean space $\real^3$. At zero temperature all

379: electrons occupy the ``Dirac sea'' $\epsilon(p)\leqslant \epsilon_F$

380: where the ``Fermi energy'' $\epsilon_F$ depends on the number of free

381: electrons in the metal. We assume that $\epsilon_F$ is a regular value

382: for the Morse function $f=\epsilon(p)$. At low temperatures we

383: are dealing only with

384: ``excited'' electrons nearby the

385: Fermi level $\epsilon(p)=\epsilon_F$.

386: 

387: The Fermi level looks geometrically as a

388: two-dimensional surface $M_F\subset \torus^3$

389: in the space of quasimomenta. This

390: surface is nonsingular and homologous to zero in the $3$-torus. Let

391: us assume that it is connected.

392: 

393: The \emph{topological rank} $r(M_F)$

394: of the Fermi surface is defined as the rank of the image

395: of the first homology group of $M_F$ under the mapping

396: $i_*:H_1(M_F,\integer)\rightarrow H_1(\torus^3,\integer)\cong\integer^3$

397: induced by the inclusion $i:M_F\hookrightarrow\torus^3$.

398: Since $i_*(H_1(M_F))$ is a sublattice in $\integer^3$,

399: we always have $r\in\{0,1,2,3\}$.

400: 

401: For

402: example, the topological rank of the Fermi surface of

403: lithium is equal to zero (in this case, the Fermi

404: surface looks like a topological $2$-sphere), whereas

405: it is equal to three for

406: copper, gold, platinum, and some other noble metals. For gold, for

407: example, the genus of the Fermi surface is equal to four.

408: 

409: The problem that we will consider is most difficult when the topological

410: rank of the Fermi surface is maximal possible, \emph{i.e.}, equal to three.

411: One can easily show that the genus of the Fermi surface must be

412: greater than or equal to the topological rank.

413: 

414: Interesting dynamical phenomena occur in the presence of a

415: magnetic field. In the semiclassical approximation, an electron,

416: which is considered as a point in the space of quasimomenta, moves

417: along constant energy lines in the plane $\real^2_{B,p_0}$

418: orthogonal to the magnetic field $B$ and passing through the

419: initial position $p_0$ of the quasimomentum.

420: 

421: One may say that this is a

422: Hamiltonian system on the $3$-torus of quasimomenta with Poisson

423: bracket $$\{p_j,p_l\}=\frac ec\,B_{jl}=\frac ec\,\varepsilon_{jlq}B^q$$ and

424: Hamiltonian $f=\epsilon(p)$:

425: $$\frac{dp_j}{dt}=\{p_j,\epsilon(p)\},$$

426: so

427: the motion preserves the energy and a linear Casimir of the Poisson

428: bracket. The level sets of this Casimir are planes orthogonal to the magnetic

429: field. The trajectories can be treated as  leaves of the

430: Hamiltonian foliation on the Fermi surface given by the

431: equation $\omega=0$ where $\omega$ is the following closed $1$-form:

432: $\omega=\sum_jB^jdp_j|_{M_F}$.

433: 

434: According

435: to the ``strong magnetic field limit'' principle worked out by

436: I.\,Lifshitz, M.\,Azbel, M.\,Kaganov, and V.\,Peschanski in early 1960s,

437: all essential properties of the electrical conductivity in the presence of

438: a reasonably strong uniform magnetic field $B$ should follow from the

439: structure of the

440: dynamical system on the Fermi surface described above (see

441: \cite{LAK,Ab,Ki,Zi}). For ordinary normal metals (like gold, for

442: example) one may use this approximation for magnetic fields strong

443: enough in the human sense (like

444: $1\,\mathrm{Tesla}<|B|<10^3\,\mathrm{Tesla}$ for low

445: temperatures; recall: $1\,\mathrm{Tesla}=

446: 10^4\,\mathrm{Gauss}$). If the magnetic field is too strong, then

447: the semiclassical approximation will not be valid.

448: If the magnetic field is too weak, then the electron quasimomentum drift will

449: be too slow, and the distance that the quasimomentum passes for the

450: characteristic time of the electron free motion will become

451: insufficient to affect the observable conductivity.

452: 

453: However, in 1960s the study of the just mentioned dynamical system

454: was only started. Some conceptual mistake

455: was then made in \cite{LP} and further investigation

456: was stopped, and resumed only many years later in works

457: \cite{N,Z,N1,D,D1,NM,NM1,D2,D3,DM,MN,MN1,MN2}.

458: 

459: What is crucial for us here is following:

460: 

461: \medskip

462: \centerline{\parbox{0.8\textwidth}{%

463: the electron trajectories coincide with connected components of

464: the level curves $\epsilon(p)=\epsilon_F$ of the function

465: $\epsilon$ restricted to the planes orthogonal

466: to the magnetic field $B$; in other words, they are connected

467: components of the level curves of functions that form a family of

468: related quasiperiodic functions with three quasiperiods. }}

469: \medskip

470: 

471: In work \cite{N2} an extension of these studies to

472: a larger number of quasiperiods was started. In particular, some

473: new  ideas and results   were formulated for the case $n=4$. The

474: present  work contains the first complete proof of those (properly corrected)

475: statements. The proof is based on the topological technique developed in

476: \cite{D2,D3}.

477: 

478: Let a constant Poisson bracket $B_{jk}$ of rank two be given

479: on the $n$-torus. Then every Hamiltonian

480: $f(p)=\epsilon(p):\torus^n\rightarrow \real$ defines a Hamiltonian

481: system whose trajectories are exactly the level lines

482: $\epsilon(p)=\const$ of the restriction

483: of the function $\epsilon$ to the planes $\real^2_{B,a}$

484: defined as follows.

485: There exist exactly $n-2$ independent linear

486: Casimirs $K_1,\dots,K_{n-2}$, $K_j(p)=K_j^lp_l$, such that

487: $\{p_s,K_j\}=0=K_j^lB_{sl}$. We put

488: $$\real^2_{B,a}=\{K_1=a_1,\dots,K_{n-2}=a_{n-2}\},$$

489: where $a=(a_1,\dots,a_{n-2})$.

490: So our trajectories

491: are exactly the levels of quasiperiodic

492: functions on the two-planes $\real^2_{B,a}$, which form

493: the family of descendants of the $n$-periodic function

494: $\epsilon(p)$. They depend on the constants $a_1,\dots,a_{n-2}$.

495: 

496: Topological study of this problem is the central part of this

497: article.

498: 

499: 

500: Modern experimental technology allows to construct surfaces with

501: a variety of prescribed small fluctuations. In particular, it is

502: possible to make  a quasiperiodic construction with any number of

503: quasiperiods. It presents us a two-dimensional weak quasiperiodic

504: electric potential $V(x,y)$. In a strong magnetic

505: field $B$ electrons move along the surface. After averaging we

506: obtain a slow motion along the level curves $V(x,y)=\const$. These

507: studies, experimental and theoretical, were done originally for

508: periodic potentials with $n\leqslant 2$ periods only (see

509: \cite{B}), but it was pointed out in work \cite{M} that

510: quasiperiodic potentials can also appear here; new predictions

511: were made for quasiperiodic cases with three and four quasiperiods

512: based on the topological results obtained in a series of works

513: of the present authors (\cite{N2,D3,MN2}).

514: 

515: \section{Topology and dynamics of quasiperiodic functions on the

516: plane: the case of three quasiperiods. The electrical conductivity in

517: metals}

518: 

519: We address the following general question. How may the level lines

520: $\varphi=\const$ of a quasiperiodic function $\varphi$ on the plane

521: with $n$

522: quasiperiods look like? In a generic situation, such a level line

523: is a one-dimensional submanifold of $\real^2$, i.e.\ a union of

524: curves. We will call these curves ``trajectories'' because in our studies they

525: have been appearing as semiclassical electron trajectories

526: on the Fermi surface in the presence of a magnetic field since early 1980s when

527: this problem was posed in work \cite{N} as a problem of

528: topology and dynamical systems. It corresponds to the case of three

529: quasiperiods only. Some of those curves may be closed in $\real^2$

530: (compact) and others nonclosed in $\real^2$ (open). Let us ask the

531: following questions.

532: 

533: \begin{description}

534: \item[Question 1] Is the size of the compact trajectories uniformly bounded (for a

535: fixed level of $\varphi$)?

536: \item[Question 2] Do the open trajectories have some nice asymptotic behavior?

537: \end{description}

538: 

539: The first results were

540: obtained in work \cite{Z}. It became clear in the second half

541: of 1980s  that the proper form of Question~2 is the following:

542: does any open trajectory has a ``strong asymptotic direction'' in

543: $\real^2$, \emph{i.e.}, lie in a strip of uniformly bounded width

544: and passes through the strip

545: ``from $-\infty$ to $+\infty$''? This specification of the problem

546: was made in article \cite{N1}. In work \cite{D} the

547: results of \cite{Z} were improved accordingly to the new formulation of

548: the problem. The important breakthrough was made in work

549: \cite{D1}, but for a long time there was no applications.

550: Physical applications were found later in works~\cite{NM,NM1}.

551: 

552: In the physically important  case $n=3$, the positive answer to

553: our Question~1 follows easily from a quite elementary argument.

554: For $n>3$ it is more difficult, and it will be discussed later.

555: Question~2 is highly nontrivial already for $n=3$.

556: As mentioned above, the asymptotic behavior of open electron

557: trajectories, \emph{i.e.}, of open connected components

558: of a level line of a quasiperiodic

559: function with three quasiperiods, was studied in

560: \cite{Z,D,D1}. It became finally

561: clear after work \cite{D1} that for the family of

562: related quasiperiodic function corresponding to

563: a ``typical'' direction of the

564: magnetic field (which is regarded as the direction of a

565: plane $\real^2\subset \real^3$),

566: either their level lines do not have open components at all or

567: the open components all have a strong asymptotic direction.

568:  The

569: latter means that each open curve has a parametrization

570: $\gamma(t)$ such that the following holds for some nonzero two-vector

571: $(x_1,y_1)$:

572: \begin{equation}\label{strong}

573: \gamma(t)=(x_0,y_0)+t\cdot(x_1,y_1)+O(1).\end{equation}

574: Below we will explain precisely

575: what  `a typical direction' means,

576: and provide references to the papers containing proofs of

577: the corresponding results. In the first work \cite{Z}

578: completed in the note

579: \cite{D}, this type of result was obtained for the special case of

580: small perturbations of a magnetic field having ``rational''

581: direction. We shall return to this special case in the next section

582: where we discuss the

583: quasiperiodic functions with four quasiperiods.

584: 

585: Applications of this studies to explaining the electrical

586: conductivity in a strong magnetic field are presented in

587: works \cite{NM,NM1}. They are based on the results of the Lifshitz

588: school of 1960s. Physicists calculated the contribution of

589: individual trajectories of simple types to the conductivity

590: tensor. These calculations have become a part of textbooks (see

591: \cite{LAK,Ab}). In the case when all trajectories are

592: compact the conductivity

593: components orthogonal to the magnetic field $B$ decrease as $|B|^{-1}$ or

594: $|B|^{-2}$ when $|B|$ grows while the direction of $B$

595: remains fixed. Some special examples of open trajectories lying in

596: finitely wide strips were found at the same time and their

597: contribution to the conductivity was

598: calculated. As pointed out in \cite{NM,NM1},

599: one can easily extend the just mentioned

600: calculation to the case of general trajectories of the same type.

601: The projection to the plane orthogonal to $B$

602: of the part of the conductivity tensor contributed by such a trajectory

603: has two eigenvalues one of which is zero and the other nonzero.

604: Since the contribution of closed trajectories tends to zero

605: when $|B|$ grows, the observable conductivity tensor

606: for a strong enough $B$ will depend on open trajectories only.

607: 

608: \smallskip

609: \centerline{\parbox{0.8\textwidth}

610: {However, the observable physical

611: conductivity tensor is formed by the contributions of all electron

612: trajectories as the sum of them. What conclusion about this tensor

613: can be made from the qualitative dynamical properties of

614: that system, which was defined on the quantum level?}}

615: \smallskip

616: 

617: In order to obtain a nontrivial new physical result, one needs

618: more than the theorems explicitly formulated in~\cite{D1}.

619: But luckily an additional crucial property also holds

620: for our dynamical system,

621: and this can be extracted from the proofs of the main theorems

622: of works \cite{Z,D1}. The property, which we call

623: \emph{topological resonance}, implies the following for the behavior

624: of trajectories.

625: 

626: \smallskip

627: \centerline{\parbox{0.8\textwidth}{For a ``typical'' family $\{\varphi_a\}$

628: of related quasiperiodic functions $\varphi_a(p)=\epsilon(p)|_{\real^2_{B,a}}$,

629: the strong asymptotical direction $\eta_B$ of noncompact trajectories is the

630: same for all trajectories.

631: Moreover, there exist an integral two-plane $\mu\subset

632: \real^3$ (\emph{i.e.}, a plane generated by two reciprocal lattice vectors,

633: $\mu\cap\integer^3\cong\integer^2$)

634: such that $\eta_B$ has the direction of the intersection of $\mu$

635: with the plane orthogonal to the magnetic field:

636: $$\eta_B\in\mu\cap \real^2_B.$$

637: This integer plane $\mu$ is

638: locally rigid, \emph{i.e.}, it remains unchanged under small

639: variations of the direction of the magnetic field.}}

640: \smallskip

641: 

642: The topological resonance property of our dynamical system

643: makes possible serious applications. It was missed in the classical works

644: of physicists, and a conceptual mistake was made in \cite{LP},

645: where calculations led to a result contradicting to this

646: property. This mistake was revealed and corrected only in works

647: \cite{NM,NM1,MN}.

648: 

649: For a strong enough $B$, the direction $\eta_B$ is a

650: zero eigenvector of the projection of the conductivity tensor to the

651: plane orthogonal to the magnetic field. The

652: integral plane $\Pi\subset \real^3$ is

653: directly observable by measuring the zero eigenvector $\eta_B$ for two

654: or more magnetic fields $B$ close to each other.

655: 

656: We refer the reader to recent article \cite{MN} for a more detailed

657: physical discussions.

658: 

659: Let us describe the picture topologically. Consider all our objects

660: in the universal covering space $\real^3$ with the reciprocal lattice

661: $\Gamma^*=\integer^3\subset \real^3$ and the three-periodic Fermi surface

662: $$\widehat{M_F}=\nu^{-1}(M_F)\subset \real^3$$

663: covering the compact one

664: $$M_F\subset

665: \torus^3.$$

666: (Recall that $\nu$ stays for the standard projection $\real^3\rightarrow\torus^3$.)

667: The electron trajectories in the covering space are

668: connected components of the intersections of the

669: three-periodic Fermi surface with planes orthogonal to $B$. Let

670: $M_0(B)$ be the closure of the union of all compact trajectories,

671: and let $L(B)$ be the closure of its complement in the Fermi surface:

672: $$L(B)=\overline{M_F\setminus M_0(B)}.$$

673: 

674: Let $L_l(B)$ be the connected components of $L(B)$.

675: In the typical case, $L(B)$ is a compact two-manifold with boundary

676: $$\partial

677: L(B)=\bigcup_{l,s} \beta_{ls},$$

678: where

679: $$\partial L_l(B)=\bigcup_s\beta_{sl}.$$

680: All boundary  curves $\beta_{sl}$ are

681: saddle connection cycles. In the typical case, we may assume that

682: every cycle $\beta_{sl}$ joins a saddle critical point to itself, since

683: all the other cases have

684: positive codimension in the appropriate functional space.

685: (In particular, we assume that there

686: is no rational linear dependence between the components of $B$,

687: and that the Hamiltonian foliation defined by $\omega=\sum

688: B^idp_i|_{M_F}=0$ has only Morse singularities and

689: does not have saddle connections between different saddles.)

690: 

691: The part $M_0$ of the Fermi surface

692: can be presented as the union of ``cylinders'' $Z_q$,

693: $M_0=\bigcup_q Z_q$,

694: whose interior consists of regular compact trajectories and

695: ``bases'' are either saddle connection cycles or

696: isolated points (centers). There are finitely many such

697: cylinders, and they are obviously compact.

698: This immediately implies a positive

699: answer to Question~1 posed in the beginning of this section:

700: 

701: \begin{center}the size of all compact trajectories is uniformly bounded.

702: \end{center}

703: 

704: 

705: We call the

706: pieces $L_l$ of the Fermi surface the \emph{carriers of open

707: trajectories}. By construction, every open trajectory

708: (in $\torus^3$) is contained in one of the carriers,

709: and, in the generic case, is everywhere dense in it.

710: Let $D^2_{ls}\subset \real^2_{B,a}$ be planar two-discs

711: orthogonal to the magnetic field such that $\partial

712: D^2_{ls}=\beta_{ls}$. We define the ``closure'' $N_l$ of every carrier

713: $L_l$ as follows: $$N_l=M_l\cup\Big(\bigcup_{s} D^2_{ls}\Big).$$

714: By construction we also have

715: $$N_l\bigcap N_{l'}=\varnothing$$

716: for $l\neq l'$.

717: 

718: We call

719: our system {\it stable topologically completely integrable} if

720: the genus of each surface $N_l $ is equal to one, and this

721: picture is stable under arbitrary small enough perturbations of the magnetic field.

722: 

723: We call the system {\it chaotic} if the genus of some $N_l$ is greater

724: than one. According to the main theorem of \cite{D1}, the latter

725: situation is always topologically unstable.

726: 

727: According to \cite{NM,NM1,MN}, what is important for physical

728: applications, is the following \emph{topological resonance}

729: property of our system. In the stable topologically completely

730: integrable case, all the closures $N_l$ of the carriers of open

731: trajectories have the same up to sign nonzero homology class:

732: $$[N_l]=\pm \mu\in H_2(\torus^3,\integer),\quad\mu\neq 0,$$

733: which is an indivisible element of the group

734: $H_2(\torus^3,\integer)\cong\integer^3$. The number of the tori

735: $N_l\subset \torus^3$ is even because the sum of their homology classes is

736: equal to the class of Fermi surface, which is zero.

737: (Note that every homologically nontrivial connected closed nonselfintersecting

738: two-manifold $M\subset \torus^3$ always represent an indivisible

739: homology class. Any two such submanifolds with empty

740: intersection represent the same homology class up to

741: sign.) The class $\mu\in H_2(\torus^3,\integer)\cong\integer^3$

742: is presented by three relatively prime integers $\mu(B)=(m_1,m_2,m_3)$.

743: 

744: The integral vector $\mu(B)$ remains

745: unchanged under small perturbations of $B$. Therefore, there

746: is an open set on the sphere $S^2$ with the same $\mu(B)$.

747: This set as well as the integral vector $\mu$ is

748: an observable characteristics of our system,

749: and it can be found experimentally by measuring the

750: conductivity tensor in the presence of strong

751: enough magnetic fields having generic directions.

752: The stable topologically integrable

753: case occurs for all directions $B/|B|\in S^2$

754: of the magnetic field from an everywhere dense open subset of

755: $S^2$. It was proved

756: in \cite{D2,D3} (by two different methods)

757: that this picture may be not valid

758: for directions $B/|B|\in S^2$ of the magnetic field from

759: a nonempty subset whose codimension is at least one.

760: 

761: In terms of quasiperiodic functions, we can say that ``non-typical''

762: functions $\varphi_a(p)=\epsilon(p)|_{\real^2_{B,a}}$ with three quasiperiods,

763: \emph{i.e.}, such that open connected

764: components of their level sets don't have a strong asymptotic direction,

765: all lie in a subset that has codimension one (in some

766: natural sense). Examples of level lines with chaotic behavior in

767: the case $n=3$ were constructed in \cite{D2}. We call such level lines

768: \emph{strongly chaotic trajectories}. Interesting attempts were

769: made in order to find physical properties of the conductivity in these

770: cases. For some special examples it was done in work \cite{DM}

771: but in general the stochastic properties of these trajectories are

772: unknown. A.\,Maltsev formulated the following conjecture.

773: 

774: \begin{conjecture}

775: The contribution of strongly chaotic trajectories to the

776: conductivity tensor tends to zero when $|B|$ grows (remaining in

777: a reasonable range), which includes the conductivity in

778: the direction of the  magnetic field itself.

779: \end{conjecture}

780: 

781: Previously, S.\,Tsarev constructed a ``weakly chaotic'' example

782: (unpublished, see work \cite{D2}). In his case, there is a rational dependence

783: between the components of $B$, and there is just one carrier of

784: open trajectories in our sense, which coincides with the Fermi surface.

785: However, the closure of any trajectory in $\torus^3$ is not the whole

786: surface, but just a half of it, which is homeomorphic

787: to a $2$-torus with two holes. The holes are not homologically

788: trivial in $\torus^3$, so they are not regarded as closed in $\real^3$.

789: In Tsarev's example, the nonclosed level lines still have an

790: asymptotic direction in a weaker sense,

791: $$\gamma(t)=(x_0,y_0)+t\cdot(x_1,y_1)+o(t),$$

792: but the projection of any trajectory to a straight line perpendicular to $(x_1,y_1)$

793: is unbounded.

794: 

795: It is interesting to look at the behavior of trajectories in the

796: special (nongeneric) case of the Fermi surface

797: $$\epsilon(p)=\cos(p_1)+\cos(p_2)+\cos(p_3)=0.$$

798: Examples of this type were investigated numerically and

799: analytically in works \cite{D2,RdL}. There are chaotic trajectories for the set of

800: magnetic fields whose Hausdorf dimension is presumably equal to some

801: $\alpha$ with

802: $$1<\alpha<2.$$

803: There are many (in fact, infinitely many)

804: different stable topologically completely

805: integrable zones on the sphere $S^2$ having different integral

806: characteristics $\mu(B)\in\integer^3$. We

807: call this type of examples \emph{generic symmetric

808: levels}, see below.

809: 

810: We conjecture the following.

811: 

812: \begin{conjecture}

813: (i) For a generic connected smooth two-manifold $M_F\subset

814: \torus^3$ homologous to zero, the set of chaotic directions of

815: the magnetic field has Hausdorf dimension less than one in $S^2$.

816: 

817: (ii) For a generic $1$-parametric smooth family $M_{F,t}\subset \torus^3$

818: of such Fermi surfaces this set has Hausdorf

819: dimension less than two.

820: \end{conjecture}

821: 

822: A detailed investigation of this problem containing the proofs of

823: all topological statements needed for physical

824: applications found in \cite{NM,NM1} is performed in

825: \cite{D2,D3}. Special attention is paid there to one-parametric families of

826: Fermi surfaces that are levels of the same Morse function

827: $f:\torus^3\rightarrow \real$:

828: $$M_c=\{f(p)=c\}.$$

829: It is proved that, for any $B$ from a stability zone,

830: open trajectories live on the levels $M_c$ from a connected interval $c_1(B)\leqslant

831: c\leqslant c_2(B)$ of the real line. For a $B$ away from the stability

832: zones, the strongly chaotic

833: behavior might appear only on a single level $c(B)\in\real$.

834: As a corollary we obtain the following result:

835: 

836: \smallskip

837: \centerline{\parbox{0.8\textwidth}{For a function

838: f=$\epsilon(p)$ with symmetry $\epsilon(p+p_0)=-\epsilon(p)$,

839: where $p_0\in\torus^3$ is some shift,

840: strongly chaotic trajectories cannot appear on the levels $M_c$ with

841: $c\neq0$ because otherwise they must appear on $M_{-c}$, too,

842: for the same $B$, which is impossible.

843: In such a case we call the level $c=0$ a \emph{generic symmetric level}.}}

844: \smallskip

845: 

846: According to

847: our conjecture, the Hausdorf dimension of the set of $B/|B|\in S^2$ for which

848: the strongly chaotic behavior occurs on such a level is equal to some $\alpha<2$.

849: 

850: The surface $\sum_{j=1}^{j=3} \cos(p_j)=0$ gives an example of a generic

851: symmetric level with $p_0=(\pi,\pi,\pi)$.

852: 

853: Some more details about chaotic

854: trajectories and stability zones for the case of three

855: quasiperiods will be given below. They will be needed for

856: proving our main

857: result about quasiperiodic functions with four quasiperiods (see the next

858: section).

859: 

860: \section{The stable topological complete integrability for $n=4$

861: quasiperiods}

862: 

863: Let us consider now the case of $n=4$ (or more) quasiperiods.

864: For every direction $\Pi$ of two-planes in $\real^n$ (\emph{i.e.}, a two-dimensional

865: vector subspace $\Pi\subset\real^n$), the

866: original $n$-periodic Morse function

867: $f:\torus^n\rightarrow \real$

868: defines a family of descendants $\{\varphi_a(y)\}$ on the family

869: affine two-planes $\real^2_{\Pi,a}\subset

870: \real^n$ having direction $\Pi$.

871: 

872: We call the level $\{f=c\}$

873: of the function $f$ \emph{topologically completely

874: integrable} (\emph{TCI}) for the direction $\Pi$ if, for each

875: $\varphi_a$ from the family, all regular connected components of the level

876: line $\varphi_a(y)=c$ are either compact or have a strong asymptotic

877: direction. We call this level \emph{stable

878: TCI} if this property remains unchanged under small

879: perturbations of the function $f$ and the direction $\Pi$,

880: which is a point in the Grassmanian manifold $G_{n,2}$.

881: 

882: We say that

883: the Stable TCI level satisfies the {\it topological

884: resonance} condition (for given $\Pi$) if there exists an integral hyperplane

885: $\mu\subset \real^n$, $\mu\cap\integer^n\cong\integer^{n-1}$,

886: such that all open regular trajectories have

887: the same asymptotical direction $\eta_\Pi$ that coincides with

888: the direction of the straight line $\mu\cap\Pi\cong\real$.

889: Since $\mu$ is integral, it must remain unchanged under small perturbations of

890: anything.

891: 

892: Let us point out that even the ``trivial case'' $n=2$ is

893: meaningful (as a subject of the elementary Morse theory on the

894: $2$-torus): for a generic double-periodic function on the plane

895: there exists a level $f=c$ with a connected component presenting a nontrivial

896: indivisible

897: homology class $\mu\in H_1(\torus^2,\integer )$.

898: All other components of every level are

899: either homologically trivial or homologous to $\pm\mu$. For

900: Morse functions with exactly four critical points and critical values

901: $c_0<c_1<c_2<c_3$, all levels $f=c$, with $c_1<c<c_2$, have exactly two

902: connected components, whose homology classes are $\pm\mu$. All other nonsingular

903: levels are either compact or empty.

904: 

905: \begin{description}

906: \item[Question]

907: Consider famous real nonsingular quasiperiodic

908: finite-gap solutions of the KdV equation

909: $$u(x,t)=2\partial_x^2\log\Theta (xU+tV+U_0) + c_{\Gamma}$$ with

910: an arbitrary number of quasiperiods (or gaps). Are the levels

911: $u(x,t)=c$ always stable topologically completely integrable or

912: they can be chaotic? How to find their strong asymptotic

913: direction and their integer-valued characteristic $\mu$?

914: \end{description}

915: 

916: P.\,Grinevich pointed out to us that, for real smooth finite-gap

917: solutions of the KdV equation,

918: the $n$-periodic function $f=2\partial_U^2 \log

919: \Theta(\eta_1,\dots,\eta_n)+c_{\Gamma}$ on the $\eta$-space

920: is always a Morse

921: function on the real $n$-torus with $2^n$ critical points, simply

922: because it can be reduced to the form

923: $f=\sum_{j=1}^{j=n}\alpha_j\sin x_j$ by a diffeomorphism of the

924: torus isotopic to the identity. There is a canonical  lattice in the

925: $\eta$-space generated by the so-called $a$-cycles which are the

926: real finite gaps of the 1D Schr\"odinger operator on a hyperelliptic

927: ``spectral'' Riemann surface $\Gamma$. The real constants

928: $\alpha_j$ depend on the spectrum. As a conclusion, we get the

929: following:

930: 

931: \smallskip

932: \centerline{\parbox{0.8\textwidth}{For

933: generic real nonsingular two-gap solutions of the KdV equation, there

934: exist a critical value $c_{\mathrm{cr}}$ such that all constant speed

935: levels $u(x,t)=c$,

936: $$c_1=c_{\Gamma}-c_{\mathrm{cr}}<c<c_{\Gamma}+c_{\mathrm{cr}}=c_2$$

937: are periodic

938: perturbations  of a family of straight lines with integral

939: direction $m_1:m_2$ on the plane

940: with lattice. This direction is locally rigid, but globally depends

941: on the constants $\alpha_j$. All other levels are either compact

942: or empty. We call it the \emph{topological speed} of the solution.}}

943: \smallskip

944: 

945: The computational studies of this problem for finite-gap solutions

946: are now being investigated numerically.

947: 

948: We concentrate now on the case $n=4$ quasiperiods. Work

949: \cite{N2} presents an idea of the proof that, for every generic Morse

950: function $f$ and a noncritical generic level $f=c$, there exists an open

951: everywhere dense set of two-plane directions

952: $\Pi\in G_{4,2}$ for which the level $f=c$

953: is stable TCI. The proof of this theorem

954: requires the use of some extension of the results of work

955: \cite{D3}. Here we make some corrections to the statement

956: of~\cite{N2} and provide a complete proof.

957: 

958: We believe that a generic level is stable TCI for all directions $\Pi$

959: from a subset $S\subset G_{4,2}$ whose measure is full in $G_{4,2}$. However,

960: we don't have an idea how to prove this conjecture. For $n>4$

961: nothing like that is expected.

962: 

963: Now we start a detailed investigation of the case $n=4$. Even

964: Question~1 of the previous section presents a difficulty here.

965: It is possible that a

966: single level set of a quasiperiodic function with four

967: quasiperiods contains a family of compact components without an

968: upper bound for their size. An example can be constructed easily.

969: 

970: However, we are going to show that there is an open everywhere

971: dense open set of

972: quasiperiodic functions $\varphi$ with four quasiperiods such that

973: the level lines $\varphi=\const$ have the same qualitative

974: behavior as those in the typical case of three quasiperiods.

975: The precise formulation is as follows.

976: 

977: \begin{theorem}\label{th1}

978: There exists an open everywhere dense

979: subset $S\subset C^\infty(\torus^4)$ of $4$-periodic

980: functions $f$ and an open everywhere dense subset

981: $X_f\subset G_{4,2}$ depending on $f$ such that

982: any level $M^3_c=\{f=c\}$ of $f$ is stable TCI (or does not

983: contain open trajectories at all) for any $\Pi\in X_f$.

984: 

985: Moreover, for any regular open trajectory, the remainder

986: term $O(1)$ in (\ref{strong}) as well as

987: the diameter of any compact trajectory are bounded from above by a constant $C$

988: not depending on the affine plane $\real^2_{\Pi,a}$ containing the trajectory,

989: provided that the level $c$ and the direction $\Pi\in X_f$ are fixed.

990: \end{theorem}

991: 

992: Let us make a remark about notation and terminology.

993: Once we switched to the case of four quasiperiods,

994: our problem is no longer relevant to the discussed above physical model

995: of conductivity in normal metals in the presence of a magnetic field.

996: So we change the notation for the coordinates in $\real^n$ from $p_l$, which

997: we used for quasimomentum, to more customary, $x_l$, $l=0,1,2,3$,

998: and don't longer think

999: of the lattice $\integer^4\subset\real^4$ as the one dual to

1000: some physical lattice. We think of the ``magnetic field''

1001: $B$ as a linear mapping from $\real^4$ to $\real^2$ (or from $\real^3$

1002: to $\real$ in the $n=3$ case) such that $\Pi=\ker(B)$.

1003: Thus, by $\real^2_{\Pi,a}$ we mean the two-plane $B^{-1}(a)$,

1004: where $a\in\real^2$. In the case $n=3$ we may also think of $B$ as a vector

1005: perpendicular to the plane $\Pi$.

1006: However, we keep calling connected components of

1007: the intersections of $M_c^3$ with the two-planes $\real^2_{\Pi,a}$

1008: trajectories, just for briefness.

1009: 

1010: We start by recalling results of~\cite{D2,D3} for the

1011: three-dimensional case in the form needed to prove our theorem. Let

1012: $B:\real^3\rightarrow\real$ be a linear function of irrationality

1013: degree three, \emph{i.e.}, of the form $B(x)=B_1x_1+B_2x_2+B_3x_3$,

1014: where $B_1$, $B_2$, $B_3$ are reals linearly independent over $\integer$, and

1015: let $f:\torus^3\rightarrow\real$ be a generic smooth function. By

1016: `generic' we mean that $f$ does not satisfy certain conditions

1017: that have codimension $\geqslant1$. However, we shall pay

1018: attention to codimension one singularities as, in order to deal

1019: with the four-dimensional case, we are going to consider

1020: one-parametric families of three-dimensional pictures.

1021: 

1022: We abuse notation by using the same letter $f$ for the lift of $f$

1023: to the covering $3$-space $\real^3$. We use notation $M^2_c$

1024: for the level set $f^{-1}(c)$ in

1025: $\torus^3$ and $\widehat{M^2_c}$ for its cover in $\real^3$. By

1026: $\gamma_{a,c}$ we denote the whole intersection of $\widehat{M^2_c}$

1027: with the plane $\real^2_{\Pi,a}=B^{-1}(a)$. So, the trajectories that we are studying are

1028: regular connected components of $\gamma_{a,c}$ or their projections to $\torus^3$.

1029: 

1030: First of all, consider closed trajectories on $M^2_c$. Notice

1031: that, since we assumed $B$ to be of maximal irrationality degree,

1032: a trajectory in $\real^3$ is closed if and only if so is its image

1033: in $\torus^3$. Without the assumption on $B$ this may be not true,

1034: since a closed trajectory in $\torus^3$ may then be

1035: non-homologous to zero, in which case its cover in $\real^3$

1036: consists of infinite ``periodic'' trajectories treated as ``open''

1037: in the physical applications.

1038: 

1039: For a generic $f$, compact trajectories on every $M^2_c$ form

1040: finitely many cylinders whose bases are either saddle connections

1041: or extrema of the restriction $B|_{M^2_c}$. Obviously, the length

1042: of compact trajectories is bounded from above by some constant.

1043: 

1044: Let $U$ be the set of $c$ such that $\gamma_{a,c}$ has unbounded

1045: connected components for some $a$. In other words, $c\in U$ if and

1046: only if $M^2_c$ contains open trajectories that are not saddle

1047: connections. The following picture, which was sketched in

1048: the previous section, can be extracted from work

1049: \cite{D3}:

1050: 

1051: \smallskip

1052: \noindent\hskip 0.1\textwidth{\parbox{0.8\textwidth}{The

1053: set $U$ is either a closed interval, $U=[c_-,c_+]$, or

1054: just one point, $U=\{c_0\}$.

1055: 

1056: If $U=[c_-,c_+]$ is a nontrivial interval, then for any $c\in U$,

1057: there is a (unique) family of two-tori $\torus^2_{c,1},\ldots,

1058: \torus^2_{c,2k}$ (with $k$ depending on $c$)

1059: imbedded in $\torus^3$ such that

1060: \begin{enumerate}

1061: \item Each $\torus^2_{c,i}$ consists of the closure of some

1062: open trajectory on $M^2_c$ and a few (may be zero) planar

1063: disks perpendicular to $B$;

1064: \item Every open trajectory is contained by whole in one of

1065: the tori $\torus^2_{c,i}$;

1066: \item All the tori $\torus^2_{c,i}$ define the same up to sign

1067: nonzero homology class $\mu$ in $H_2(\torus^3,\integer)$;

1068: \item For all but finitely many $c$, the tori $\torus^2_{c,i}$

1069: are pairwise disjoint, and, in this case, a sufficiently

1070: small variation of $c$ causes small deformation of the tori.

1071: For the exceptional $c$'s they can be

1072: made disjoint by a small perturbation. At such a $c$, a couple of tori

1073: is born or killed.

1074: \end{enumerate}

1075: All the picture is stable in this case, which means that after a

1076: small enough perturbation of $f$, the interval $U$ and the family

1077: of tori $\torus^2_{c,i}$ are perturbed slightly. In particular,

1078: the homology class $\mu$ fixed.}}\hfill\\

1079: 

1080: \begin{remark}

1081: In the setting of papers~\cite{D2,D3},

1082: the function $f$ was assumed to be fixed

1083: and the point of concern was the dependence of the behavior of

1084: our dynamical system on the magnetic field $B$ and on the

1085: level of the function $f$. The stability

1086: of the whole picture under small perturbations of the function $f$

1087: was not discussed. However, the arguments of those works can be

1088: easily modified in order to prove such stability. Indeed,

1089: one of the key observations in~\cite{D2,D3} is that, locally,

1090: the qualitative behavior of the trajectories (including the existence

1091: of a strong asymptotic direction) depends only

1092: on finitely many parameters, which are certain critical values of

1093: the ``height'' function $B(x)$ restricted

1094: to the surface $\{f=\const\}$. (For instance,

1095: the existence of strongly chaotic examples

1096: was proved in~\cite{D2} by specifying the combinatorial

1097: structure of the surface and particular values of the

1098: parameters.) It is easy to see that those parameters

1099: behave nicely under small perturbations of $f$,

1100: so extending the arguments of~\cite{D2,D3}

1101: to this, more general type of perturbations requires

1102: almost no additional work.

1103: \end{remark}

1104: 

1105: Let us describe the three-dimensional picture in more details. For

1106: a generic level surface $M^2\subset\torus^3$, the structure of

1107: trajectories on $M^2$ is as follows. Compact trajectories form a

1108: few open cylinders whose ``ends'' approach either an extremum

1109: point of the function $f|_{M^2}$ or a saddle connection cycle, see

1110: Fig.~\ref{cylinders}.

1111: \begin{figure}[ht]

1112: \centerline{\epsfig{file=cc1.eps,height=100pt}\hskip1cm

1113: \epsfig{file=cc2.eps,height=100pt}\hskip1cm\epsfig{file=cc3.eps,height=100pt}}

1114: \caption{Cylinders of compact

1115: trajectories}\label{cylinders}

1116: \end{figure}

1117: The rest of the surface (if not empty) consists of an

1118: even number of two-tori with or without holes, and each hole is a

1119: saddle connection cycle. Each hole can be glued up by a planar

1120: disk perpendicular to the vector $B$. We denote the obtained

1121: surface by $N$. The preimage $\widehat N\subset\real^3$ of $N$

1122: under the projection $\nu:\real^3\rightarrow\torus^3$ is a family of

1123: finitely deformed periodic ``wrapped''  planes in $\real^3$, see

1124: Fig.~\ref{wrapped}.

1125: \begin{figure}[ht]

1126: \centerline{\epsfig{file=wp.eps,height=100pt}}

1127: \caption{A wrapped plane}\label{wrapped}

1128: \end{figure}

1129: 

1130: What happens to $\widehat N$ when the surface $M^2$ changes?

1131: Small deformations of the surface $M^2$ cause just small

1132: deformations of the tori and their covering planes. Suppose we

1133: have a generic $1$-parametric family of surfaces $M^2(t)$. This

1134: means that we consider a generic $1$-parametric family of

1135: functions $f_t:\torus^3\rightarrow\real$, and for each $t$, the

1136: surface $M^2(t)$ is defined by the equation $f_t(x)=\const$.

1137: 

1138: When the parameter $t$ varies, the connected components of $N$ are

1139: just deformed while they stay apart from each other. However,

1140: eventually two tori can collide and disappear or, on the contrary,

1141: a pair of tori can be born. This occurs when $M^2(t)$ traverses a

1142: subset which has codimension one in a natural sense. It is not

1143: important here whether $M^2(t)$ is the family of level surface of

1144: a single function or an arbitrary generic one-parametric family of

1145: surfaces.

1146: 

1147: The generic tori collision was described in~\cite{D3}.

1148: It was assumed in \cite{D3} that the family of

1149: surfaces $M^2(t)$ is the family of level surfaces of the same

1150: function, $M^2(t)= \{x\in\torus^3\;|\;f(x)=t\}$. However, the

1151: argument is exactly the same for an arbitrary generic family of

1152: surfaces.

1153: 

1154: The following two types of tori collision are possible

1155: in the generic case.

1156: \begin{enumerate}

1157: \item

1158: A cylinder of closed trajectories with bases attached to two

1159: different components of $N$ collapses. The corresponding

1160: codimension-one condition has the following form: two different

1161: saddles get joined by a saddle connection.

1162: This causes an ``interaction'' of pairs of open

1163: trajectories lying on the collided tori, which turns

1164: them into infinitely many closed trajectories,

1165: see Fig.~\ref{degenerate-cylinder}.

1166: \begin{figure}[ht]

1167: \centerline{\epsfig{file=ccc1.eps,height=200pt}

1168: \hskip0.5cm\raisebox{100pt}{$\rightarrow$}\hskip0.5cm

1169: \epsfig{file=ccc2.eps,height=200pt}

1170: \hskip0.5cm\raisebox{100pt}{$\rightarrow$}\hskip0.5cm

1171: \epsfig{file=ccc3.eps,height=200pt}}

1172: \caption{Collapse of a cylinder}

1173: \label{degenerate-cylinder}

1174: \end{figure}

1175: \item

1176: A Morse-type surgery occurs on $M^2$ that results in a one-handle added

1177: to the surface. The behavior of trajectories is shown in Fig.~\ref{morse}.

1178: \begin{figure}[ht]

1179: \centerline{\epsfig{file=ms1.eps,height=200pt}

1180: \hskip0.5cm\raisebox{100pt}{$\rightarrow$}\hskip0.5cm

1181: \epsfig{file=ms2.eps,height=200pt}

1182: \hskip0.5cm\raisebox{100pt}{$\rightarrow$}\hskip0.5cm

1183: \epsfig{file=ms3.eps,height=200pt}}

1184: \caption{A Morse surgery destroying open trajectories}\label{morse}

1185: \end{figure}

1186: \end{enumerate}

1187: 

1188: It is important to note here that whenever $N$ consists of just

1189: two tori and $M^2(t)$ passes a singularity of one of the two types

1190: mentioned above, then all open trajectories get destroyed, so that

1191: $N$ is empty right after the critical event. However, the

1192: following is not proven to be impossible in a generic

1193: one-parametric family of surfaces: bearing and canceling of a

1194: pair of tori occurs alternatingly at moments $t_1,t_2,t_3,\dots$

1195: so that the sequence $(t_n)$ converges to some $t_*$ and an

1196: ergodic regime occurs on $M^2(t_*)$. The latter means that there

1197: is an open trajectory on $M^2(t_*)$ whose closure has genus more

1198: than one (actually, it should then be equal to three). The

1199: existence of such ergodic regimes was proven in~\cite{D2,D3}, and it

1200: was shown only that such regimes satisfy a codimension one

1201: condition. However, we will not need to deal with ergodic regimes

1202: in order to prove our result.

1203: 

1204: Now we turn to the case when $M^2(t)=M^2_t$ is the family of level

1205: surfaces of a generic function on $\torus^3$.

1206: 

1207: Consider the restriction of the function $f$ to the plane $\real^2_{\Pi,a}$

1208: for some $a$. Let $V\subset\real^2_{\Pi,a}$ be the union of all compact

1209: components of $\gamma_{a,c}$ over all $c$, and $W\subset\real^2_{\Pi,a}$ the

1210: union of all unbounded components of $\gamma_{a,c}$. We have

1211: $\real^2_{\Pi,a}=V\cup W$, $V\cap W=\varnothing$, $V$ is open. Notice:

1212: connected components of $V$ are not necessarily bounded. Let

1213: $V_1,V_2,\ldots$ be the connected components of $V$. It is easy to

1214: see that $f$ is constant on $\partial V_i$ for any $i$.

1215: 

1216: For $x\in\real^2_{\Pi,a}$ we put

1217: $$\overline f(x)=\left\{\begin{array}{ll}f(x)&\text{if }x\in W,\\

1218: f(\partial V_i)&\text{if }x\in V_i.\end{array}\right.$$

1219: By doing so for all $a$, we obtain a new function $\overline f:

1220: \real^3\rightarrow\real$. We use the following notation:

1221: $$N_c=\{x\in\torus^3\;;\;\overline f(x)=c\}.$$

1222: The function $f$ and its level sets $N_c$

1223: have the following properties.

1224: 

1225: \begin{lemma}

1226: The function $\overline f$ is a well defined continuous function

1227: on $\torus^3=\real^3/\integer^3$.

1228: 

1229: If $U=\{c_0\}$, then $\overline f\equiv c_0$.

1230: 

1231: If $U=[c_-,c_+]$, where $c_-<c_+$, then for all but finitely many

1232: $c\in(c_-,c_+)$, we have

1233: $$N_c=\bigcup\limits_i\torus^2_{c,i}.$$

1234: 

1235: For all $c\in[c_-,c_+]$ a small regular neighborhood of

1236: $N_c$ is homeomorphic to the union of a few copies of

1237: $\torus^2\times[0,1]$.

1238: \end{lemma}

1239: 

1240: \begin{proof}

1241: In the case $U=\{c_0\}$ our claim is trivial.

1242: 

1243: Assume that $U=[c_-,c_+]$ with $c_-<c_+$.

1244: By construction, $\torus^2_{c,i}\cap M_c$ consists of open trajectories,

1245: thus, we have $\overline f(x)\equiv c$ on $\torus^2_{c,i}\cap M_c$.

1246: The whole torus $\torus^2_{c,i}$ is obtained from

1247: $\torus^2_{c,i}\cap M_c$ by attaching disks each of which

1248: lies in the plane $\real^2_{\Pi,a}$ for some $a$. The boundary of such a disk is

1249: a part of a singular unbounded component of the level set

1250: of $f|_{\real^2_{\Pi,a}}$. By construction, we have $\overline f\equiv c$

1251: in such a disk. Therefore, we always have

1252: $\torus^2_{c,i}\subset N_c$.

1253: 

1254: Let us look at what happens with the tori $\torus^2_{c,i}$ when

1255: $c$ varies. For all but finitely many $c$ the tori

1256: $\torus^2_{c,i},\torus^2_{c,j}$ are disjoint if $i\ne j$.

1257: Moreover, for any $c\ne c'$ and any $i,j$, the tori $\torus^2_{c,i}$

1258: $\torus^2_{c',j}$ are always disjoint.

1259: 

1260: When $c$ varies, tori $\torus^2_{c,i}$ are continuously deformed

1261: except at a few values of $c$, where one of the following happens:

1262: 1) two tori collide and then disappear; 2) two tori are

1263: newly born. The latter event is opposite to the first one.

1264: 

1265: Let us describe torus collision in more detail.

1266: At the moment of the collision we have a closed domain $W$

1267: in $\torus^3$ that has the form of the manifold

1268: $\torus^2\times[0,1]$ in which some intervals

1269: $x\times[0,1]$ are collapsed to a point. There may be

1270: just one such points $x$ or a closed disk $D^2\subset\torus^2$

1271: of such points. The first case corresponds to an index one or index two

1272: Morse critical point of $f$, whereas the latter

1273: corresponds to a degenerate cylinder of closed trajectories.

1274: 

1275: The interior of the domain $W$ is filled by compact trajectories

1276: and, by construction, the function $\overline f$ is constant

1277: inside $W$. Thus, $W$ is a connected component of some $N_c$,

1278: since $W$ is squeezed between the two collided tori. We call such

1279: a $W$ \emph{pseudotorus}.

1280: 

1281: So, we have the following picture. The decomposition of $\torus^3$

1282: into the union of (connected components of) $N_c$ over all $c$

1283: is nothing else but a trivial fibration over $S^1$ with fibre $\torus^2$,

1284: with a few fibres replaced by pseudotori.

1285: 

1286: Schematically this is shown in Fig.~\ref{pseudotori}.

1287: \begin{figure}[ht]

1288: \centerline{\epsfig{file=pseudot.eps,height=150pt}}

1289: \caption{A family of tori with a few replaced by pseudotori}

1290: \label{pseudotori}

1291: \end{figure}

1292: 

1293: \end{proof}

1294: 

1295: Now we return to the four-dimensional case. Let $\Pi\in G_{4,2}$

1296: be a two-plane defined by a linear mapping $B:\real^4\rightarrow\real^2$.

1297: By $\real^2_{\Pi,a,b}$ we denote the affine plane $B^{-1}(a,b)\subset\real^4$,

1298: and by $M^4_{\leqslant c}$ (respectively, $M^4_{\geqslant c}$)

1299: the subset of $\torus^4$ defined by the inequality $f(x)\leqslant c$

1300: (respectively, $f(x)\geqslant c$).

1301: 

1302: Let $N\subset\torus^4$ be a submanifold (or, more generally, a subset).

1303: We say that $N$ is \emph{essentially below} (respectively, \emph{essentially

1304: above}) $M^3_c$ if for any $a,b\in\real$,

1305: the intersection $\widehat N\cap\real^2_{\Pi,a,b}$

1306: is disjoint from all unbounded components of $\widehat M^4_{\geqslant c}

1307: \cap\real^2_{\Pi,a,b}$ (respectively, of $\widehat M^4_{\leqslant c}\cap\real^2_{\Pi,a,b}$).

1308: Thus, the property of $N$ to be essentially below $M^3_c$ depends on $\Pi$.

1309: 

1310: The following two facts are proved by analogy with the 3D case.

1311: 

1312: \begin{lemma}

1313: If $N$ is essentially below or essentially

1314: above $M^3_c$ for a given $\Pi$ then this remains

1315: true after a small perturbation of $\Pi$, $f$, and $c$.

1316: \end{lemma}

1317: 

1318: \begin{lemma}

1319: If there exists a homologically nontrivial $3$-torus $N$

1320: which is essentially above or essentially below $M^3_c$,

1321: then the assertion of Theorem~\ref{th1} is true

1322: for these specific $f$, $\Pi$, and $c$.

1323: \end{lemma}

1324: 

1325: Thus, in order to prove Theorem~\ref{th1} it suffices to

1326: show that for everywhere dense set of pairs $(f,\Pi)$,

1327: and for each $c$,

1328: there exists a homologically nontrivial $3$-torus $N\subset\torus^4$

1329: which is essentially below or essentially above $M^3_c$.

1330: 

1331: Let $B=(\ell_1,\ell_2)$ be a couple of

1332: linear functions on $\real^4$ such that

1333: \begin{enumerate}

1334: \item

1335: the function $\ell_1$ is rational, i.e., $\ell_1\in(\integer^4)^*$;

1336: \item

1337: the restriction of $\ell_2$ to the integral three-plane $\ell_1=0$

1338: has irrationality degree three.

1339: \end{enumerate}

1340: Obviously, the set of $2$-planes $\Pi=\ker B$ defined by

1341: $\ell_1,\ell_2$ of this form is everywhere dense in $G_{4,2}$.

1342: 

1343: Without loss of generality, we assume that $\ell_1(x)=x_0$,

1344: $\ell_2(x)=H_1x_1+H_2x_2+H_3x_3$, where

1345: $x=(x_0,x_1,x_2,x_3)\in\real^4$, $\qopname\relax

1346: o{rank}_\integer\langle H_1,H_2,H_3\rangle=3$. We consider the

1347: $4$-torus $\torus^4$ as a one-parametric family of three-tori

1348: $\torus^3_t=\{x_0=t\}$. For any $t\in[0,1]$, we deal with the

1349: restrictions $f_t$ and $\ell_{2,t}$ of respectively $f$ and

1350: $\ell_2$ to $\torus^3_t$ as in the three-dimensional case. We

1351: introduce $U_t=[c_{t-},c_{t+}]$, $\overline{f_t}$, $N_{t,c}$ as

1352: before.

1353: 

1354: Let us consider the dependence of the interval $U_t$ on $t$.

1355: The endpoints $c_{t\pm}$ of the interval $U_t$ are continuous functions

1356: of $t$. Moreover, in the regions where $c_{t+}>c_{t-}$

1357: these functions are piecewise smooth. This follows from

1358: the fact that locally, near a generic $t$, they are defined

1359: by a condition of the form: two saddles on $M^2_{t,c_{\pm}}$

1360: are connected by a separatrix. Here we call such intervals \emph{stability zones}.

1361: To every stability zone there corresponds an integral vector

1362: $\mu\in H_1(\torus^3,\integer)=\integer^3$, which we call

1363: the \emph{label} of the zone.

1364: 

1365: Figure~\ref{cpm} shows how the functions $c_{t\pm}$ may look like

1366: in the generic case.

1367: It is possible that at some $t$ we have $c_{t+}=c_{t-}$,

1368: see Fig.~\ref{cpm}a). This may

1369: occur at the boundary of a stability zone or at $t$ such that

1370: the open trajectories in $M^3_t$ are chaotic.

1371: \begin{figure}[p]

1372: \noindent a)

1373: 

1374: \centerline{\begin{picture}(425,160)

1375: \put(0,30){\epsfig{file=cpma.eps,width=400pt}}

1376: \put(0,10){\vector(1,0){425}} \put(422,0){$t$}

1377: \put(150,80){$c_{t-}$} \put(150,130){$c_{t+}$}

1378: \put(402,71){$c_0(t)$}

1379: \end{picture}}

1380: 

1381: 

1382: \vskip1cm

1383: \noindent b)

1384: 

1385: \centerline{\begin{picture}(425,160)

1386: \put(0,30){\epsfig{file=cpmb.eps,width=400pt}}

1387: \put(0,10){\vector(1,0){425}} \put(422,0){$t$}

1388: \put(150,76){$c_{t-}$} \put(150,135){$c_{t+}$}

1389: \multiput(0,152)(10,0){41}{\line(1,0)5}

1390: \put(408,149){$(c_-)_{\mathrm{max}}$}

1391: \multiput(0,99)(10,0){41}{\line(1,0)5}

1392: \put(408,96){$(c_+)_{\mathrm{min}}$}

1393: \end{picture}}

1394: 

1395: \vskip1cm

1396: \noindent c)

1397: 

1398: \centerline{\begin{picture}(425,160)

1399: \put(0,30){\epsfig{file=cpmc.eps,width=400pt}}

1400: \put(0,10){\vector(1,0){425}} \put(422,0){$t$}

1401: \put(150,62){$c_{t-}$} \put(150,145){$c_{t+}$}

1402: \multiput(0,106)(10,0){41}{\line(1,0)5}

1403: \put(408,89){$(c_-)_{\mathrm{max}}$}

1404: \multiput(0,92)(10,0){41}{\line(1,0)5}

1405: \put(408,103){$(c_+)_{\mathrm{min}}$}

1406: \end{picture}}

1407: \caption{Functions $c_{t\pm}$ in the generic case}\label{cpm}

1408: \end{figure}

1409: For all such $t$ we have $c_{t+}=c_{t-}=c_0(t)$, where $c_0(t)$ is

1410: a piecewise smooth function of $t$. It is defined locally by a

1411: condition of the form: the sum of ``heights'' of certain saddles

1412: equals to zero, see~\cite{D2,D3}.

1413: 

1414: It is most likely that any ``chaotic" $t$ must be an accumulating

1415: point of an infinite sequence of stability zones. In other words,

1416: it cannot happen that the equality $c_{t+}=c_{t-}$ holds

1417: everywhere in a nontrivial interval $(t_1,t_2)$. However, this

1418: does not follow directly from the previous works~\cite{D2,D3}, and

1419: will not be used here.

1420: 

1421: \begin{lemma}\label{=/=}

1422: The equality

1423: \begin{equation}\label{c+=c-}

1424: \min_tc_{t+}=\max_tc_{t-}

1425: \end{equation}

1426: does not hold for a generic $f$.

1427: \end{lemma}

1428: 

1429: \begin{proof}

1430: We should consider the following three cases.

1431: 

1432: Case 1. For all $t$ we have $c_{t+}>c_{t-}$. Then there exists

1433: a smooth periodic function $g(t)$ such that

1434: $c_{t+}>g(t)>c_{t-}$. Condition~(\ref{c+=c-})

1435: will not hold if we disturb the function $f$ in the following way:

1436: $f(t,x_1,x_2,x_3)\mapsto f_\varepsilon(t,x_1,x_2,x_3)=f(t,x_1,x_2,x_3)+

1437: \varepsilon g(t)$, where $|\varepsilon|$ is sufficiently small.

1438: So,~(\ref{c+=c-}) impose a codimension one condition on $f$ in this case.

1439: 

1440: Case 2. For two different $t=t_1,t_2$ we have $c_{t+}=c_{t-}$.

1441: Then condition~(\ref{c+=c-}) will not hold after an arbitrary

1442: perturbation $f\mapsto f+\varepsilon g(t)$, where $g(t)$

1443: is an arbitrary function with $g(t_1)\ne g(t_2)$.

1444: 

1445: Case 3. There is exactly one $t=t_0$ such that $c_{t+}=c_{t-}=c_0$ holds,

1446: and we have $c_{t+}>c>c_{t-}$ for all $t\ne t_0$.

1447: Let us take $t$ close to $t_0$. The interval $[c_{t-},c_{t+}]$

1448: is then small, which means that, when $c$ varies from

1449: $(c_{t-}-\delta)$ to $(c_{t+}+\delta)$ with $\delta>0$,

1450: a pair of tories $N_{t,c}$ is born at $c=c_{t-}$ and then almost

1451: immediately destroyed at $c=c_{t+}$. Therefore,

1452: there are two cylinders of closed trajectories on $M_{t,c_0}$

1453: of very small height.

1454: 

1455: Now let us fix $c=c_0$ and vary $t$. When $t$ approaches $t_0$

1456: from the left, say, we will have two closed trajectory cylinders

1457: that get degenerate at the moment $t=t_0$. When $t$ passes $t_0$,

1458: two cylinders must appear again. The main point here is that

1459: those, new, cylinders appear from \emph{the same} pair of degenerate

1460: cylinders. Indeed, the pair of degenerate cylinders that we obtain

1461: when $t$ approaches $t_0$ from the left cuts $M_{c_0,t_0}$

1462: into two tori. Since the irrationality degree of $\ell_2$

1463: is equal to three, one can show that there may no other degenerate cylinder on

1464: $M_{t_0,c_0}$.

1465: 

1466: Thus, we have the following picture. When $t$ goes from $t_0-\delta$

1467: to $t_0+\delta$, two closed trajectory cylinders degenerate

1468: and then regenerate again. Let $h_1(t)$, $h_2(t)$ be there heights.

1469: So, we not only have $h_1(t_0)=h_2(t_0)=0$, but also

1470: $h_1'(t_0)=h_2'(t_0)=0$, which impose a codimension two

1471: condition on the function $f$.

1472: \end{proof}

1473: 

1474: By construction we have

1475: 

1476: \begin{lemma}\label{<>}

1477: For any $t$ and $c>c_{t+}$ (respectively, $c<c_{t-}$),

1478: the torus $\torus^3_t$ is essentially

1479: below (respectively, essentially above) $M^3_c$.

1480: \end{lemma}

1481: 

1482: 

1483: According to Lemma~\ref{=/=} only the following two cases are possible:

1484: 1) $\min_tc_{t+}<\max_tc_{t-}$; 2) $\min_tc_{t+}>\max_tc_{t-}$.

1485: In Case~1, for any $c$ we have either $c>\min_tc_{t+}$

1486: or $c<\max_tc_{t-}$, and by Lemma~\ref{<>} we are done.

1487: 

1488: So, it remains to consider Case~2,

1489: $\min_tc_{t+}>\max_tc_{t_-}$. This inequality means,

1490: in particular, that the intervals $U_t$

1491: have a nontrivial intersection $U=\cap_tU_t=[c_-,c_+]$, and we

1492: have just one stability zone, which covers the whole circle $S^1$.

1493: This is illustrated in Fig.~\ref{cpm}c).

1494: 

1495: We define a function $\overline f:\torus^4\rightarrow\real$

1496: as in the three-dimensional case by considering

1497: intersections $M_c^3\cap\real^2_{\Pi,a,b}$, and introduce notation

1498: $$N_c=\{x\in\torus^4\;;\;\overline f(x)=c\}.$$

1499: By construction, $\overline{f_t}$

1500: coincides with the restriction of $\overline f$ to $\torus^3_t$, and we

1501: have

1502: $$N_c=\bigcup\limits_tN_{t,c}.$$

1503: 

1504: 

1505: For almost

1506: any $c$, $t$, the intersection $N_c$ with $\torus^3_t$ is either empty

1507: or consists

1508: of $2$-tori, and all those tori have the same up to sign homology class

1509: $\alpha\in H_2(\torus^3,\integer)$.

1510: In this case, the whole torus $\torus^4$ has the structure

1511: of a trivial $\torus^2$-bundle over $\torus^2$ with a $1$-parametric

1512: family of fibres replaced by pseudotori.

1513: \begin{figure}[ht]

1514: \centerline{\begin{picture}(400,320)

1515: \put(0,20){\epsfig{file=t2.eps,height=300pt}}

1516: \put(0,10){\vector(1,0){310}} \put(305,0){$t$} \put(400,160){\hbox

1517: to0pt{\hss Pseudotori}} \put(340,170){\vector(-1,1){55}}

1518: \put(340,160){\vector(-2,-3){37}}

1519: \end{picture}}

1520: \caption{Level lines of $\overline f$ on $\torus^2$}\label{t2}

1521: \end{figure}

1522: The function

1523: $\overline f$ is constant over each fibre, so it can be considered

1524: as a function on $\torus^2$.

1525: 

1526: Figure~\ref{t2} illustrates the structure of level lines of $\overline f$

1527: viewed as a function on $\torus^2$. The preimage of a generic point

1528: is a $2$-torus imbedded in $\torus^4$, and all those $2$-tori are

1529: ``parallel". Whenever the function $\overline f$ has an extremum on

1530: the line $t=\mathrm{const}$, the preimage of the critical point

1531: is a pseudotorus.

1532: 

1533: Including a pseudotorus into a $1$-parametric family of $2$-tori

1534: does not change the topological type of the union of the tori. Indeed,

1535: as we mentioned above, a small regular neighborhood of a pseudotorus

1536: in $\torus^3$ is homeomorphic to $\torus^2\times(0,1)$. In other words,

1537: attaching collars $\torus^2\times(0,1)$ to a pseudotorus again

1538: gives $\torus^2\times(0,1)$.

1539: 

1540: We conclude the following from this.

1541: 

1542: \begin{lemma}

1543: For almost all $c$ each connected component

1544: of $N_c$ will be homeomorphic to $\torus^3$. The exceptions are those

1545: $c$ that are critical values of the function $\overline f$ on $\torus^2$.

1546: \end{lemma}

1547: 

1548: For a generic $f$ we obtain a generic Morse function $\overline f$

1549: on $\torus^2$. For such a function, there must be an interval $[c_1,c_2]$

1550: such that, whenever we have $c\in[c_1,c_2]$,

1551: the level line $\overline f=c$ contains a closed curve non-homologous to

1552: zero in $\torus^2$. The preimage $N_c$ of this level line in $\torus^4$

1553: is a $3$-torus non-homologous to zero.

1554: Thus, we get the following.

1555: 

1556: \begin{lemma}

1557: There exist $c_1$, $c_2$ such that $c_1<c_2$ and both $N_{c_1}$

1558: and $N_{c_2}$ contain a connected component homeomorphic to $\torus^3$

1559: and non-homologous to zero.

1560: \end{lemma}

1561: 

1562: It remains to notice that whenever $c>c_1$ the hypersurface $N_{c_1}$

1563: is essentially below $M^3_c$, and whenever $c<c_2$ the hypersurface

1564: $N_{c_2}$ is essentially above $M^3_c$. Thus, for all $c$

1565: we have a non-homologous to zero $3$-torus which is either essentially

1566: above or essentially below $M^3_c$, and we are done in Case~2.

1567: 

1568: 

1569: 

1570: \begin{thebibliography}{99}

1571: \bibitem{N}

1572: S.~P.~Novikov. Hamiltonian formalism and a multivalued analog

1573: of Morse theory. (Russian) {\it Uspekhi Mat.\ Nauk}~{\bf37} (1982),

1574: no.~5, 3--49; translation in {\it Russian Math.\

1575: Surveys}~{\bf37}~(1982);

1576: {\tt http://genesis.mi.ras.ru/$\sim$snovikov/74.zip}.

1577: \bibitem{Z}

1578: A.\,V.\,Zorich. Novikov's problem on semiclassical motion of electron

1579: in homogeneous magnetic field close to rational. (Russian)

1580: {\it Uspekhi Mat.\ Nauk}~{\bf39}~(1984), no.\ 5, 235--236; translation

1581: in {\it Russian Math.\ Surveys}~{\bf39} (1984).

1582: \bibitem{N1}

1583: S.\,P.\,Novikov. Quasiperiodic structures in topology.

1584: {\it Topological methods in modern mathematics\/} (Stony Brook, NY, 1991),

1585: 223--233, Publish or Perish, Houston, TX, 1993.

1586: \bibitem{D}

1587: I.\,A.\,Dynnikov.

1588: A proof of S.\,P.\,Novikov's conjecture for the case of small

1589: perturbations of rational magnetic fields. (Russian)

1590: {\it Uspekhi Mat.\ Nauk\/} {\bf47} (1992), no. 3(285), 161--162;

1591: translation in {\it Russian Math. Surveys\/} {\bf47} (1992), no. 3,

1592: 172--173.

1593: \bibitem{D'}

1594: I.\,A.\,Dynnikov.

1595: S.\,P.\,Novikov's problem on the semiclassical motion of an electron.

1596: (Russian) {\it Uspekhi Mat. Nauk\/} {\bf48} (1993),

1597: no. 2(290), 179--180; translation in {\it Russian Math. Surveys\/}

1598: {\bf48} (1993), no. 2, 173--174.

1599: \bibitem{D1}

1600: I.\,A.\,Dynnikov.

1601: A proof of the conjecture of S.\,P.\,Novikov on the semiclassical

1602: motion of an electron. (Russian) {\it Mat.\ Zametki\/} {\bf53}

1603: (1993), no.~5, 57--68; translation in {\it Math Notes\/}

1604: {\bf53} (1993), no.~5--6, 495--501.

1605: \bibitem{NM}

1606: S.\,P.\,Novikov and A.\,Ya.\,Maltsev.

1607: Topological quantum characteristics observed in the investigation

1608: of the conductivity in normal metals. (Russian)

1609: {\it Pis'ma Zh.\ Eksp.\ Teor.\ Fiz.} {\bf63} (1996), no.~10, 809--813;

1610: translation in

1611: {\it JETP Letters\/} {\bf63} (1996), no.~10, 855--860.

1612: \bibitem{NM1}

1613: S.\,P.\,Novikov and A.\,Ya.\,Maltsev.

1614: Topological phenomena in normal metals. (Russian)

1615: {\it Uspekhi Phys.\ Nauk\/} {\bf41} (1998), no.~3, 231--239;

1616: {\tt arXiv:cond-mat/9709007}.

1617: \bibitem{LAK}

1618: I.\,Lifshitz, M.\,Ya.\,Azbel, M.\,I.\,Kaganov.

1619: {\it Elektronnaya teoriya metallov}. (Russian) Nauka, Moscow,

1620: 1971; translation: {\it Electron theory of metals}, Consultants

1621: Bureau, New York,  1973.

1622: \bibitem{N2}

1623: S.\,P.\,Novikov.

1624: Levels of quasiperiodic functions on a plane, and Hamiltonian systems.

1625: (Russian) {\it Uspekhi Mat.\ Nauk\/} {\bf54} (1999),

1626: no.~5(329), 147--148; translation in {\it Russian Math.\ Surveys\/}

1627: {\bf54} (1999), no.~5, 1031--1032;

1628: {\tt arXiv:math-ph/9909032}.

1629: \bibitem{M}

1630: A.~Ya.Maltsev.

1631: Quasiperiodic functions theory and the

1632: superlattice potentials for a two-dimensional electron gas.

1633: {\it Journal of Mathematical Physics\/} {\bf45}, no.~3 (March 2004),

1634: 1128--1149;

1635: {\tt arXiv:cond-mat/0302014}.

1636: \bibitem{DKN}

1637: V.~I.~Arnold, S.~P.~Novikov (editors), {\it Encyclopedia

1638: Mathematical Sciences, Dynamical Systems, vol 4: Completely

1639: Integrable Systems and Symplectic Geometry}. Springer Verlag,

1640: second edition (revised), 2001:\\

1641:  1.\ V.\,Arnold, A.\,Givental;

1642: \\

1643:  2.\ B.\,Dubrovin, I.\,Krichever, S.\,Novikov.

1644: \bibitem{Pi}

1645: T.\,Q.\,T.\,Le, S.\,Piunikhin, and V.\,Sadov.

1646: The geometry of quasicrystals. (Russian) {\it Uspekhi Mat.\

1647: Nauk\/} {\bf48} (1993), no.~1(289), 41--102; translation in {\it

1648: Russian Math.\ Surveys\/} {\bf48} (1993), no.~1, 37--100.

1649: \bibitem{LP}

1650: I.\,M.\,Lifshitz, V.\,G.\,Peschanski.

1651: Halvanomagnetic characteristics of metals with open Fermi surfaces,~II.

1652: (Russian) {\it Zh.\ Eksp.\ Teor.\ Fiz.} {\bf 38} (1960), 188--193;

1653: translation in {\it JETP.}.

1654: \bibitem{Ab}

1655: A.~A.~Abrikosov.

1656: {\it Osnovy torii metallov}. (Russian) Nauka, Moscow, 1987;

1657: translation: {\it Fundamentals of theory of metals},

1658: North-Holland, Amsterdam, 1988.

1659: \bibitem{Ki}

1660: C.\,Kittel, {\it Quantum theory of solids}. John

1661: Wiley \& Sons Inc., New York, London, 1963.

1662: \bibitem{Zi}

1663: J.\,M.\,Ziman. {\it Principles of the theory of solids}.

1664: Cambridge Univ.\ Press, 1972.

1665: \bibitem{MN}

1666: A.\,Ya.\,Maltsev and S.\,P.\,Novikov.

1667: Dynamical Systems, Topology, and Conductivity in Normal Metals.

1668: {\it Journal of Statistical Physics\/} {\bf115} no.~1, April 2004, 31-46;

1669: {\tt arXiv:cond-mat/0312708}.

1670: \bibitem{MN1}

1671: A.\,Ya.\,Maltsev and S.\,P.\,Novikov.

1672: Quasiperiodic functions and dynamical systems in quantum solid state

1673: physics. Dedicated to the 50th anniversary of IMPA. {\it Bull.\ Braz.\

1674: Math.\ Soc. (N.S.)\/} {\bf 34} (2003), no.~1, 171--210;

1675: {\tt arXiv:math-ph/0301033}.

1676: \bibitem{MN2}

1677: A.\,Ya.\,Maltsev, S.\,P.\,Novikov.

1678: Topology, Quasiperiodic functions and the transport phenomena.

1679: {\it Topology in condensed matters}, M.\,I.\,Monastyrsky (ed.),

1680: Sprinter Verlag, 2004, to appear;

1681: {\tt arXiv:cond-mat/0312710}.

1682: \bibitem{D2}

1683: I.\,A.\,Dynnikov.

1684: Semiclassical motion of the electron. A proof of the Novikov conjecture

1685: in general position and counterexamples.

1686: {\it Solitons, geometry, and topology\/{\rm:} on the crossroad},

1687: 45--73, Amer.\ Math.\ Soc.\ Transl.\ Ser.~2, 179,

1688: {\it Amer.\ Math.\ Soc.}, {\it Providence}, {\it RI}, 1997

1689: \bibitem{D3}

1690: I.\,A.\,Dynnikov.

1691: The geometry of stability regiones in Novikov's problem on

1692: the semiclassical motion of the electron. (Russian)

1693: {\it Uspekhi Mat.\ Nauk}, {\bf 54} (1999), no.~1, 21--60.

1694: \bibitem{RdL}

1695: R.\,De Leo. Numerical Analysis of the Novikov Problem

1696: of a Normal Metal in a Strong Magnetic

1697: Field. {\it SIAM Journal of Applied Dynamical Systems\/}

1698: {\bf 2}:4 (2003), 517--545.

1699: \bibitem{DM}

1700: I.\,A.\,Dynnikov and A.\,Ya.\,Maltsev.

1701: Topological characteristics of electron spectra in monocrystals.

1702: (Russian) {\it Zh.\ Eksp.\ Teor.\ Fiz.} {\bf112}:1

1703: (1997), 371--378; translation in {\it JETP\/} {\bf85}:1 (1997),

1704: 205--208.

1705: \bibitem{B}

1706: C.\,W.\,J.\,Beenaker.  Guiding-center-drift resonance in a

1707: periodically modulated two-dimensional electron gas.

1708: {\it Phys.\ Rev.\ Lett.} {\bf62} (1989), no.~17, 2020--2023.

1709: 

1710: 

1711: \end{thebibliography}

1712: 

1713: \end{document}

1714: