1: 

2: \documentclass[12pt]{article}

3: \usepackage{latexsym,amssymb,psfig}

4: 

5: %\documentstyle[12pt]{article}

6: \textheight 9in

7: \textwidth 6in

8: \font\of=msbm10 scaled 1200

9: \def\R{\mbox{\of R}}

10: \def\C{\mbox{\of C}}

11: \def\Z{\mbox{\of Z}}

12: \def\N{\mbox{\of N}}

13: \renewcommand\arraystretch{1.3}

14: \hoffset =-1cm

15: \voffset =-2cm

16: 

17: \newtheorem{definition}{Definition}

18: \newtheorem{remark}{Remark}

19: \newtheorem{theorem}{Theorem}%[section]

20: \newtheorem{lemma}{Lemma}

21: \newtheorem{proposition}{Proposition}

22: \newtheorem{corollary}{Corollary}

23: 

24: \title{ Complete hyperelliptic integrals of the first kind and their

25: non-oscillation }

26: 

27: \author{Lubomir Gavrilov \\

28: \normalsize \it Laboratoire Emile Picard , CNRS UMR 5580,  Universit\'e

29: Paul Sabatier\\

30: \normalsize \it 118, route de Narbonne, 31062 Toulouse Cedex, France \\

31: \\ Iliya D. Iliev\\

32: \normalsize \it Institute of Mathematics, Bulgarian Academy of Sciences\\

33: \normalsize \it P.O. Box

34: 373, 1090 Sofia, Bulgaria } \date{}

35: 

36: \begin{document}

37: \maketitle

38: \begin{abstract}

39: Let $P(x)$ be a real polynomial of degree $2g+1$, $H=y^2+P(x)$ and

40: $\delta(h)$ be an oval contained in the level set $\{H=h\}$.

41: We study complete Abelian integrals of the form

42: $$I(h)=\int_{\delta(h)}

43: \frac{(\alpha_0+\alpha_1 x+\ldots + \alpha_{g-1}x^{g-1})dx}{y},

44: \;\;h\in \Sigma,$$

45: where $\alpha_i$ are real and $\Sigma\subset \R$ is a maximal open interval

46: on which a continuous family of ovals $\{\delta(h)\}$ exists. We show that

47: the $g$-dimensional real vector space of these integrals is not Chebyshev

48: in general: for any $g>1$, there are hyperelliptic Hamiltonians $H$ and

49: continuous families of ovals $\delta(h)\subset\{H=h\}$, $h\in\Sigma$,

50: such that the Abelian integral $I(h)$ can

51: have at least $[\frac32g]-1$ zeros in $\Sigma$. Our main result is Theorem

52: \ref{main} in which we show that when $g=2$, exceptional families of ovals

53: $\{\delta(h)\}$ exist, such that the corresponding vector space is still

54: Chebyshev.

55: \end{abstract}

56: 

57: \section{Introduction}

58: 

59: 

60: 

61: 

62: 

63: 

64: Take real polynomials  $H,f,g\in \R[x,y]$ and let $\delta (h) \subset

65: \{(x,y)\in \R^2: H(x,y)=h\}$, $h\in\Sigma$,

66: be a continuous family of ovals.

67: For  sufficiently small $\varepsilon$ and generic $H, f, g,$

68: the limit cycles of the perturbed plane Hamiltonian system

69: $$

70: dH + \varepsilon (f dx + g dy) = 0

71: $$

72: which tend to certain ovals from the continuous

73: family when $\varepsilon\to 0$, are in one-to-one

74: correspondence with the zeros of the complete Abelian integral

75: $$

76: I(h) = \int_{\delta (h)} f(x,y) dx + g(x,y) dy, \quad h\in\Sigma.

77: $$

78: For this reason the problem of finding the zeros of $I(h)$ in terms of the

79: degrees of $H,f,g$

80: was called  by Arnold \cite[p. 313]{Arnold1} the

81: ``{\it weakened 16th Hilbert problem}"  (compare to Hilbert \cite{hil}, see

82: also Arnold \cite{Arnold2,Arnold1+,ten}).

83: Note that the level sets $ \{H=h\} $ will contain in general several

84: continuous families of ovals which need be considered separately.

85: 

86: It follows from the Varchenko-Khovanskii theorem that the number of the

87: zeros of $I(h)$ is bounded

88: by a constant $N(n,d)<\infty$, uniformly in all $H,f,g$, such that

89: $deg\,f\leq d$,  $deg\,g \leq d$,  $deg\,H\leq n$.

90: 

91: In the so called ``hyperelliptic case" ($H=y^2+P(x)$) it was proved by

92: Novikov and Yakovenko\cite{nov-yak} that there exists an algorithm producing

93: a function $C(n,d)$ such that $\infty > C(n,d)> N(n,d)$,

94: which is given by a tower function (an iterated exponent) of

95: height at least five. Their

96: proof is based on the analytic properties of a suitable Picard-Fuchs system

97: satisfied by Abelian

98: integrals (including the magnitude of the coefficients of the system and some

99: restrictions on its monodromy group).

100: 

101: The progress in solving the weakened 16th Hilbert problem (finding the

102: {\it exact} number of the zeros of

103: $I(h)$) concerned so far the ``elliptic case" only (the complex algebraic

104: curve $\{H=h\}$

105: is of genus at most one). It was proved for instance

106:  that in several cases,

107: the vector space ${\cal A}_{H,d}$

108: of Abelian integrals of degree $d$ polynomials along the ovals of $H$,

109: obeys the so called {\it Chebyshev property} (the number

110: of the zeros of each integral is smaller than the dimension of the vector

111: space ${\cal A}_{H,d}$), see \cite{Petrov86,gav98,gav99,gas-li}.

112: In this relation Arnold asked in \cite[the 7th problem]{ten} whether the

113: $g$-dimensional vector

114: space of Abelian integrals

115: \begin{equation}

116: \label{question}

117: I(h)=\int_{\delta(h)}

118: \frac{(\alpha_0+\alpha_1 x+\ldots + \alpha_{g-1}x^{g-1})dx}{y},

119: \;\;h\in \Sigma

120: \end{equation}

121:  where

122: $

123: \delta (h)\subset \{(x,y)\in \R^2:\; y^2+P(x)=h\},\;\;g=[\frac12(deg\, P-1)]>1,

124: $

125: is Chebyshev.

126: 

127: In an attempt to solve this problem Givental \cite{giv} obtained a

128: non-oscillation theorem for Lagrangian

129: planes of the Picard-Fuchs system satisfied by the Abelian integrals

130: $\int_{\delta(h)}x^idx/y$. He

131: used the fact that every Picard-Fuchs system has a Hamiltonian (or Poisson)

132: structure. He failed,

133: however, to produce bounds for the zeros of the Abelian integrals.

134: 

135: In the present paper we try to explore the algebro-geometrical properties

136: of the Abelian integrals,

137: without using the Picard-Fuchs system. Our main result is Theorem

138: \ref{main} in which we show that

139: when $g=2$ and $deg\,P=5$, exceptional families of ovals

140: $\{\delta(h)\}$, $ h\in \Sigma$, exist such that every Abelian integral of

141: the form

142: $$

143: I(h)=\int_{\delta(h)}

144: \frac{(\alpha_0+\alpha_1 x)dx}{y},\quad \alpha _0^2+\alpha _1^2\neq 0

145: $$

146: has at most one simple zero on the  interval $\Sigma$. At a first sight

147: this seems to be an easy

148: observation (equivalent to the claim that $(\int_{\delta(h)}

149: xdx/y)/(\int_{\delta(h)}dx/y)$ is

150: monotonous on $\Sigma$). This is, however, the first result of such a kind

151: for non-elliptic curves

152: $\{H=h\}$. The proof uses the Riemann bilinear relations on differentials

153: of the first kind together

154: with  the fact that a Jacobian variety with its polarization cannot be a

155: direct product of

156: principally polarized Abelian varieties. Our arguments can be adapted to

157: other situations

158: (for instance when the degree of $P(x)$ is $6$), but we shall not do

159: this here.

160: 

161: We give also a negative answer (Proposition \ref{nocheb}) to the initial

162: question posed by Arnold:

163: it turns out that there exist Abelian integrals of the form

164: (\ref{question}) with exactly

165: $[\frac32g]-1$ zeros in a neighborhood of the origin.

166: 

167: 

168: 

169: \section{The cyclicity at a center}

170: In this section we prove that in general, the integral (\ref{question})

171: does not belong to a Chebyshev space. Consider first the particular case

172: when $n=5$. Let $H= y^2+ P(x)$ where $P(x)= x^2(1+a_1 x +a_2 x^2+a_3 x^3)$,

173: $a_1,a_2, a_3 \in \R$, $a_3\neq 0$, and denote by

174: $\delta (t) \subset \{y^2+P(x)=t\}$  a continuous family of cycles vanishing

175: at the origin as $t$ tends to zero.

176: 

177: 

178: \begin{proposition}

179: \label{cyclicity}

180: The Abelian integral

181: $$

182: I(t) = \int_{\delta (t)} \frac{(\alpha_0+\alpha_1 x) dx}{y}=

183: \alpha_0 I_0(t)+\alpha_1 I_1(t)\not\equiv 0

184: $$

185: can have a zero at the origin of multiplicity at most two. Moreover, for

186: fixed $\alpha_1, a_2, a_3$, $\alpha_1 a_3\neq 0$, there exist a sufficiently

187: small

188: $\varepsilon >0$  and $\alpha_0$, $a_1$ in a small neighborhood of zero,

189: such that $I(t)$ has exactly two simple zeros in the interval

190: $(0,\varepsilon )$.

191: \end{proposition}

192: The above shows that the real vector space generated by the functions

193: $I_0(t)$, $I_1(t)$ is not Chebyshev. Recall that a real vector space $V$ of

194: functions defined on some interval $\Sigma $ is said to be Chebyshev,

195: provided that each $f\in V$ has at most $\dim V - 1$ zeros (counted with

196: multiplicity) on $\Sigma $, and Chebyshev with accuracy $m$, if

197: each $f\in V$ has at most $\dim V+m - 1$ zeros there.

198: 

199: 

200: \vspace{2ex}

201: \noindent

202: {\bf Proof of Proposition \ref{cyclicity}.}

203: For a small $x$, denote $X=x(1+a_1 x +a_2 x^2+a_3 x^3)^{1/2}$.

204: Then an easy calculation yields the inverse transformation

205: $$\textstyle

206: x=\varphi(X)\sim X-\frac12a_1X^2+(\frac58a_1^2-\frac12a_2)X^3

207: -(a_1^3-\frac32a_1a_2+\frac12a_3)X^4+\ldots \quad \mbox{\rm as}\quad

208: X\to 0.$$

209: Therefore, for small positive $t$,

210: $$\begin{array}{l}{\displaystyle

211: I(t)=\int_{y^2+X^2=t}\frac{[\alpha_0+\alpha_1\varphi(X)]

212: \varphi'(X)}{y}dX}\\[5mm]

213: {\displaystyle

214: = \int_{y^2+X^2=t} \frac{\alpha_0[1+O(X^2)] - \alpha_1[\frac32a_1X^2

215: +\frac52(\frac{21}{8}a_1^3-\frac72a_1a_2+a_3)X^4 +O(X^6)]}{y}dX}\\[5mm]

216: =2\pi\alpha_0[1+O(t)]-2\pi\alpha_1[\frac34a_1t+

217: \frac{15}{16}(\frac{21}{8}a_1^3- \frac72a_1a_2+a_3)t^2 +O(t^3)].

218: \end{array}$$

219: When $\alpha_0=a_1=0$, the function $I(t)$ has a double zero at the

220: origin since $\alpha_1 a_3\neq 0$.

221: Taking $\alpha_1=1$, $|\alpha_0|<\!\!<|a_1|<\!\!<|a_3|$ and

222: $\alpha_0 a_1>0>a_1a_3$, one finishes the proof of the proposition. $\Box$

223: 

224: Clearly, the above approach could be applied to hyperelliptic Hamiltonians

225: of any degree that have a center.

226: 

227: Denote by ${\cal H}_n$ the set of Hamiltonians

228: $H=y^2+x^2+a_1 x^3+\ldots +a_{n-2}x^n$, where $a_k\in \R$, $a_{n-2}\neq 0$.

229: Denote by $g=[\frac12(n-1)]\geq 2$ the genus of the hyperelliptic curve.

230: Given $H\in {\cal H}_n$, let $\delta(t)$, $t\in (0, T_H)$, be an oval from

231: the continuous family of ovals surrounding the center at the origin.

232: Consider the vector space ${\cal A}_H$ formed by the Abelian integrals

233: 

234: 

235: \begin{equation}

236: \label{g}

237: I(t) = \int_{\delta (t)} \frac{G(x) dx}{y},\quad t\in(0,T_H),

238: \quad G\in \R[x],\quad deg\, G<g.

239: \end{equation}

240: Let us point out that $dim\,{\cal H}_H=g$ in general, but

241: $dim\,{\cal H}_H=[\frac12(g+1)]$ provided that $H$ contains no odd degree

242: monomials. In the proposition below, we give a partial answer to the question

243: raised by V. Arnold'd in \cite{ten}.

244: 

245: 

246: 

247: \begin{proposition}

248: \label{nocheb}

249: There exist $H\in{\cal H}_n$ and a polynomial $G$ such that the

250: corresponding integral $(\ref{g})$ has exactly $[\frac32g]-1$ small positive

251: simple zeroes. Thus, such a space ${\cal A}_H$ could be Chebyshev with

252: accuracy at least $[\frac12g]$ in $(0,T_H)$.

253: \end{proposition}

254: 

255: 

256: 

257: Given an integer $k\geq 0$ and $H\in{\cal H}_n$, denote

258: $$I_k(t)=\int_{\delta(t)}\frac{x^kdx}{y}, \quad 0<t<T_H.$$

259: The proof of the above proposition is based on the following lemma.

260: 

261: 

262: 

263: \begin{lemma}

264: \label{int} The following asymptotic expansions hold for

265: small positive $t$:

266: 

267: \vspace{1ex}

268: \noindent

269: {\em(i)}$\quad I_{2k}(t)= t^k[c_k+O(t)],\;\;\mbox{\it where} \;\;

270: c_k=2\pi\frac{(2k-1)!!}{(2k)!!},$

271: 

272: \vspace{1ex}

273: \noindent

274: {\em(ii)}$\quad\displaystyle I_{2k+1}(t)=-{\textstyle\frac12}

275: \sum\limits_{j=1}^g a_{2j-1}(2k+2j+1)t^{k+j}[c_{k+j}+O(t)],$

276: 

277: \vspace{1ex}

278: \noindent

279: therefore the coefficients in the last expansion belong to the ideal

280: generated by the odd-numbered parameters $\{a_1, a_3,\ldots, a_{2g-1}\}$.

281: \end{lemma}

282: 

283: 

284: 

285: \noindent

286: {\bf Proof.} Take as above $X=x(1+a_1 x +\ldots+ a_{n-2} x^{n-2})^{1/2}$,

287: then the inverse transformation $x=\varphi(X,a_1,a_2,\ldots,a_{n-2})$

288: for small $X$ reads $ x=X+A_1X^2+A_2X^3 +A_3X^4+\ldots$,

289: where the coefficients $A_j=A_j(a_1,a_2,\ldots,a_{n-2})$ are polynomials.

290: It is easy to see that if one attaches a weight $k$ to the coefficient $a_k$,

291: then $A_j$ is a polynomial of weight $j$. Indeed, directly from the definition

292: of $\varphi$ it follows that

293: $$\varphi(\omega X,a_1,a_2,\ldots,a_{n-2})=\omega\varphi(X, \omega a_1,

294: \omega^2 a_2,\ldots, \omega^{n-2}a_{n-2}).$$

295: Applying this identity to

296: $$\varphi(X, a_1, \ldots, a_{n-2})=X(1+\sum_{k=1}^\infty A_k(a_1,\ldots,

297: a_{n-2})X^k),$$

298: we get $A_k(\omega a_1, \ldots, \omega^{n-2} a_{n-2})=\omega^k

299: A_k(a_1, \ldots, a_{n-2})$.

300: To calculate the coefficient at $a_k$ in $A_k$, we take $a_1=\ldots=a_{k-1}=0$,

301: and easily obtain that it equals $-\frac12$. Therefore, one deduces that

302: $A_k=-\frac12a_k+A_k^*$ where $A_k^*$ belongs to the polynomial ideal

303: generated by $a_1,\ldots, a_{k-1}$.

304: 

305: In case (i), one immediately obtains that for small positive $t$,

306: $$I_{2k}(t)=\frac{1}{2k+1}\int_{y^2+X^2=t}\frac{d\varphi^{2k+1}(X)}{y}=

307: \int_{y^2+X^2=t}\frac{(X^{2k}+\ldots)dX}{y}=c_kt^k+O(t^{k+1}).

308: $$

309: To calculate the integral in case (ii), we write $\varphi^{2k+2}(X)$

310: in the form

311: $$\varphi^{2k+2}=X^{2k+2}\left(1+\sum\limits_{j=1}^\infty

312: (-{\textstyle\frac12}a_j+A_j^*)X^j\right)^{2k+2}$$

313: $$=X^{2k+2}\left(1+\sum\limits_{j=1}^\infty

314: [(2k+2)(-{\textstyle\frac12}a_j+A_j^*)X^j+O(X^{j+1})]\right)$$

315: $$=X^{2k+2}\left(1-\sum\limits_{j=1}^{n-2} (k+1)a_jX^j[1+O(X)]\right)$$

316: $$=-(k+1)\sum\limits_{j=1}^g a_{2j-1}X^{2k+2j+1}[1+O(X^2)]+ \Phi(X)$$

317: where $\Phi(X)$ is a series in $X$ containing only even order powers.

318: Therefore,

319: $$I_{2k+1}(t)=\frac{1}{2k+2}\int_{y^2+X^2=t}\frac{d\varphi^{2k+2}(X)}{y}$$

320: $$=-{\textstyle\frac12}\sum\limits_{j=1}^g a_{2j-1}(2k+2j+1)t^{k+j}

321: [c_{k+j}+O(t)].\;\;\Box$$

322: 

323: \vspace{2ex}

324: \noindent

325: {\bf Proof of Proposition \ref{nocheb}.} Denote $m=[\frac12g]$ and take

326: in (\ref{g})

327: $$G(x)=\sum_{k=0}^{m-1}\gamma_{2k}x^{2k}-x^{2m-1}.$$

328: Then by Lemma \ref{int},

329: $$I(t)=\sum_{k=0}^{m-1}\gamma_{2k}t^k[c_k+O(t)]+{\textstyle\frac12}

330: \sum_{j=1}^g a_{2j-1}(2m+2j-1)t^{m+j-1}[c_{m+j-1}+O(t)].$$

331: Now, freeze the even-numbered coefficients in $H$ and then choose $\gamma_{2j}$

332: and $a_{2j-1}$ so that $\gamma_{2j}\gamma_{2j+2}<0$,

333: $a_{2j-1}a_{2j+1}<0$, $\gamma_{2m-2}a_1<0$ and

334: $$|\gamma_0|<\!\!<|\gamma_2|<\!\!<\ldots<\!\!<|\gamma_{2m-2}|

335: <\!\!<|a_1|<\!\!<\ldots <\!\!<|a_{2g-1}|<\!\!<1.$$

336: Such a choice guarantees that $I(t)$ will have exactly $m+g-1=[\frac32g]-1$

337: zeroes in a certain small interval $(0,\varepsilon)$. $\Box$

338: 

339: This result raises the problem to describe the hyperelliptic Hamiltonians

340: $H$ and the continuous families of ovals $\{\delta(h)\}$ for which the

341: space of integrals (\ref{question}) obeys the Chebyshev property.

342: In the rest of the paper, we concentrate our efforts on the simplest case

343: when $deg\,P=5$. We begin with some preliminaries.

344: 

345: \section{The normal form, the bifurcation diagram and the Dynkin diagram

346: for $n=5$}

347: We consider the hyperelliptic Hamiltonian $H=\frac12y^2+P(x)$

348: where $P\in \R[x]$ and $deg\,P=5$. Assume that a certain level set

349: $\{H=h\}$ contains an oval (compact closed smooth curve without critical

350: points). As the Poincar\'e index of the respective Hamiltonian vector

351: field $X_H$ is zero, then $X_H$ has at least two different real critical

352: points. Without any loss of generality, we can place all the four critical

353: points at $(0,0)$, $(\mu, 0)$, $(\lambda,0)$ and $(1,0)$, where either

354: $0\leq\mu\leq\lambda\leq 1$ (the real case)

355: or $\bar\mu=\lambda\in\C\setminus\R$ (the complex case).

356: Finally, we can assume that the coefficient at $x^5$ is positive.

357: Then after rescaling of the $y$ variable, we come to the normal form

358: \begin{equation}

359: H(x,y)=\frac12y^2-\frac{\lambda\mu}{2}x^2+

360: \frac{\lambda+\mu+\lambda\mu}{3}x^3-\frac{1+\lambda+\mu}{4}x^4+\frac15x^5

361: \equiv \frac12y^2+P(x)

362: \label{normal}

363: \end{equation}

364: and the Hamiltonian flow $X_H$ takes the form

365: \begin{equation}

366: \label{system}

367: \begin{array}{l}

368: \dot{x}=y,\\

369: \dot{y}=-x(x-\mu)(x-\lambda)(x-1).

370: \end{array}

371: \end{equation}

372: Until the end of the paper, we will use this normal form of $H$ only.

373: 

374: Clearly, the origin is a hyperbolic saddle if $\mu\neq 0$, and $(1,0)$ is

375: a nondegenerate center provided that $\lambda\neq 1$. In addition,

376: when $0<\mu<\lambda<1$ in the real case, $(\mu, 0)$ is a center

377: and $(\lambda, 0)$ is a saddle. Denote by $h_0, h_\mu, h_\lambda$ and $h_1$

378: the corresponding critical levels of $H$. One has

379: $$\begin{array}{l}

380: h_0=0,\quad h_1=-\frac{1}{60}(3-5\lambda-5\mu+10\lambda\mu),\\[2mm]

381: h_\lambda=-\frac{\lambda^3}{60}(3\lambda^2-5\lambda\mu-5\lambda+10\mu),\\[2mm]

382: h_\mu=-\frac{\mu^3}{60}(3\mu^2-5\lambda\mu-5\mu+10\lambda). \end{array}$$

383: If $0<\mu<\lambda<1$, we have $h_0>h_\mu$, $h_\lambda>h_\mu$, $h_\lambda>h_1$.

384: The bifurcation diagram $D$ in the real case consists of the boundary

385: of the triangle $T=\{0\leq\mu\leq\lambda\leq 1\}$ and the curve $\gamma:\;

386: h_\lambda=h_0$ having an equation

387: $$\mu=\frac{3\lambda^2-5\lambda}{5(\lambda-2)},$$

388: see Fig. \ref{bdrc}.

389: 

390: \begin{figure}[htbp]

391: \begin{center}

392:   \psfig{file=save1.eps}

393: \end{center}

394: \caption{ Bifurcation diagram in the $(\lambda,\mu)$-plane in the real case}

395: \label{bdrc}

396: \end{figure}

397: 

398: There are two open components in $T\setminus D$ giving the

399: two generic cases. Namely, for parameters $\lambda, \mu$ above the

400: curve $\gamma$, $X_H$ has three period annuli, and below this curve $X_H$

401: has just two period annuli. There are also five codimension-one and four

402: codimension-two degenerate (nongeneric) cases corresponding

403: to the parameters on the bifurcation diagram $D$. In the complex case,

404: there is a unique period annulus around $(1,0)$ which terminates

405: at the saddle-loop through $(0,0)$ and the phase portraits for all

406: $\lambda\in\C\setminus\R$ are topologically equivalent. We will denote

407: the period annuli (continuous families of ovals) around $(1,0)$, $(\mu,0)$

408: and the eight loop by ${\cal O}_1$, ${\cal O}_\mu$ and ${\cal O}_e$,

409: respectively.

410: 

411: Below we describe the Dynkin diagram of the real polynomial $H(x,y)$

412: when the critical levels are distinct. Recall that it is a graph

413: defined in the same way as the Dynkin diagram (or D-diagram)

414: of a germ of an analytical function with an isolated singularity, see

415: \cite{Arnold,dim}. The vertices of the diagram are in one-to-one

416: correspondence with the vanishing cycles of the polynomial. Two

417: vertices are connected by an edge if the intersection number of the

418: cycles is not zero. To an edge one associates also a label equal

419: to the intersection number, but it will be omitted here because is of no

420: importance for us. The vanishing cycles ``vanish" along suitable paths,

421: and hence the Dynkin diagram depends also on the family of paths. The

422: precise definition in our situation is as follows.

423: 

424: Consider first the real case $0<\mu<\lambda<1$. There are five possible

425: distributions of the critical values as follows:

426: $$\begin{array}{lll}

427: h_1 < h_\mu < h_\lambda < h_0, &

428: h_1 < h_\mu < h_0 < h_\lambda, &

429: h_\mu < h_1 < h_0 < h_\lambda, \\

430: h_\mu < h_0 < h_1 < h_\lambda, &

431: h_\mu < h_1 < h_\lambda < h_0. &

432: \end{array}$$

433: For definiteness, assume that $h_1<h_\mu < h_\lambda < h_0$.

434: Denote ${\cal D} = \C \backslash [h_\lambda , \infty)$ and let

435: $$

436: l_0,l_1,l_\lambda,l_\mu:\; [0,1] \rightarrow \{h\in \C: Im\,h \geq 0\}

437: $$

438: be continuous paths connecting some fixed regular value

439: $\tilde{h} \in {\cal D}$ to

440: $h_0$, $h_1$, $h_\lambda$, $h_\mu$ respectively, as shown on

441: Fig. \ref{realdynk}.

442: 

443: % \begin{figure}

444: % \setlength{\unitlength}{.03\textwidth}

445: %

446: % \begin{picture}(30,15)(0,0)

447: % \put(2,8){

448: % \put(5,0){\vector(1,0){15}}

449: % \put(12,-7){\vector(0,1){15}}

450: % \qbezier[200](9,-6,)(14.96,0)(9,6)

451: % \put(2,0){\qbezier[200](15,-6,)(9.04,0)(15,6)}

452: % \put(20.5,-.3){$Re \lambda$}

453: % \put(11,8.5){$Im \lambda$} }

454: % \end{picture}

455: % \caption{The hyperbola  $\Gamma:\; h_\lambda=h_\mu$ in the complex case}

456: % \label{fhyp}

457: % \end{figure}

458: 

459: The Dynkin diagram of $H$ with respect to the above paths

460: is a graph with four vertices corresponding to the

461: four families of cycles vanishing along the paths

462: $l_0$, $l_1$, $l_\lambda$, $l_\mu$, which we denote by

463: $\delta_0(h)$, $\delta_1(h)$, $\delta_\mu (h)$, $\delta_\lambda (h)$

464: respectively. Two vertices are connected by an

465: edge if the intersection number of the cycles for $h=\tilde{h}$

466: is not zero (in which case it is $\pm 1$ and may be supposed equal

467: to one). The computation of the Dynkin diagram of a real polynomial

468: with real critical points only is well known, see for instance

469: \cite{cam,Arnold}. It yields the diagram shown on Fig. \ref{realdynk}.

470: We note that the computation depends only on the distribution of the

471: four critical points $0<\mu<\lambda<1$ and not on the distribution of

472: the respective critical values (hence, the Dynkin diagram is the same

473: for all possible arrangements of the critical values

474: with corresponding paths $l_0$, $l_1$, $l_\lambda$, $l_\mu$).

475: 

476: Consider now the complex case when $\lambda = \bar{\mu}\not\in\R $.

477: The method we use is a deformation of $H$ to a polynomial having real

478: critical points. Denote ${\cal D}=\C\backslash [h_0,\infty)$ and let

479: $$

480: l_0, l_1, l_\lambda, l_\mu:\; [0,1] \rightarrow {\cal D}

481: $$

482: be continuous paths connecting some fixed regular value

483: $\tilde{h} \in {\cal D}$ to $h_0$, $h_1$, $h_\lambda$, $h_\mu$

484: respectively, as shown on Fig. \ref{dynkin1}. The Dynkin diagram

485: related to these paths is defined as above. Consider in

486: the complex plane the hyperbola $\Gamma$ defined by the equation

487: \begin{equation}

488: \label{hyperbola}

489: \Gamma:\;\;

490: h_\lambda=h_\mu\quad \Leftrightarrow\quad (Im\,\lambda)^2=

491: 5Re\,\lambda(Re\,\lambda-1)

492: \end{equation}

493: and drawn on Fig. \ref{imconj}.

494: 

495: 

496: \begin{figure}

497: \setlength{\unitlength}{.03\textwidth}

498: \begin{picture}(30,15)(0,0)

499: %critical values

500: \put(0,5){

501:  \begin{picture}(15,15)(0,0)

502:  \put(0,0){\circle*{.3}} \put(0,0.5){$h_1$}

503:  \put(3,0){\circle*{.3}}  \put(3,0.5){$h_\mu$}

504:  \put(6,0){\circle*{.3}} \put(6,0.5){$h_\lambda$}

505:  \put(9,0){\circle*{.3}} \put(9,0.5){$h_0$}

506:  \put(6,0){\thicklines{\vector(1,0){6}}}

507:  %\put(9,0){\thicklines{\vector(1,0){3}}}

508: 

509:  \qbezier[20](0,0)(0,5)(6,6)

510:  \qbezier[20](3,0)(3,2)(6,6)

511:  \qbezier[20](6,0)(5,3)(6,6)

512:  \qbezier[20](9,0)(7.5,2.5)(6,6)

513:  \put(6,6){\circle*{.3}} \put(6.5,6){$\tilde{h}$}

514:  \end{picture}}

515: %Dynkin diagram

516: \put(20,5){

517:  \begin{picture}(15,15)(0,0)

518:  \put(0,0){\line(1,0){3}} \put(0,0){\circle*{.3}}

519:  \put(3,0){\line(1,0){3}} \put(3,0){\circle*{.3}}

520:  \put(6,0){\line(1,0){3}} \put(6,0){\circle*{.3}}

521:  \put(9,0){\circle*{.3}}

522: 

523:  \put(0,.5){$\delta _1$}

524:  \put(2.5,.5){$\delta _\lambda$}

525:  \put(5.5,.5){$\delta _\mu$}

526:  \put(8.5,.5){$\delta _0$}

527: 

528:  \end{picture}

529:               }

530: \put(7,0){ (i)}

531: \put(22,0){ (ii)}

532: \end{picture}

533: \caption{

534: (i) The paths $l_0$, $l_1$, $l_\lambda$, $l_\mu$ in the real case

535: $0<\mu<\lambda<1$ when $h_1<h_\mu<h_\lambda<h_0$;

536: (ii) Dynkin diagram in the real case $0<\mu <\lambda <1$ }

537: \label{realdynk}

538: \end{figure}

539: 

540: 

541: Suppose for definiteness that $\lambda$ lies inside its left branch.

542: First, by continuous deformation of $\lambda = \bar{\mu }$ such that

543: in the course of the deformation $\lambda $ remains in the left branch

544: of (\ref{hyperbola}), we may achieve that $\lambda =\mu<0<1$.

545: This, combined with $h_\lambda =h_\mu <0$ already implies that

546: \begin{itemize}

547: \item

548: the intersection number of $\delta _1(\tilde{h})$ and $\delta _0(\tilde{h})$

549: is equal to $\pm 1$;

550: \item

551: the intersection number of $\delta _\lambda (\tilde{h})$,

552: $\delta _\mu  (\tilde{h})$ with $\delta _1 (\tilde{h})$ is zero;

553: \item

554: the intersection number of $\delta _\lambda (\tilde{h})$ and

555: $\delta _\mu  (\tilde{h})$ is equal to $\pm 1$;

556: \item

557: The intersection number of either $\delta _\lambda (\tilde{h})$ or

558: $\delta _\mu(\tilde{h})$ with $\delta _0(\tilde{h})$ is equal

559: to $\pm 1$.

560: \end{itemize}

561: Finally, we may use the real structure of the complex algebraic  curve

562: $$\Gamma _h = \{(x,y)\in \C^2 :\; H(x,y) =h\}$$

563: for $h\in \R$.

564: This structure is defined by an antiholomorphic involution $i$ on the

565: curve which is the complex conjugation. It induces on its hand an

566: involution $i_*$ of $H_1(\Gamma _h, \Z)$, $(i_* )^2=id$.

567: It is easy to see that for $h<0$ we have

568: $$

569: i_* \delta _\lambda(h) = \delta _\mu(h) ,\quad

570: i_* \delta _0(h) = - \delta _0(h) .

571: $$

572: This combined with the identity

573: $$

574: \overline{\langle \delta (h) , \gamma (h)\rangle} =

575: - \langle i_*\delta (h) , i_*\gamma (h)\rangle

576: $$

577: shows that

578: $$\langle \delta_\lambda  (h) , \delta _0 (h) \rangle=

579: \langle \delta_\mu   (h) , \delta _0 (h)\rangle.

580: $$

581: This implies the Dynkin diagram as shown on Fig. \ref{dynkin1}(ii)).

582: The remaining cases shown there are studied in a similar way.

583: 

584: \begin{figure}

585: \setlength{\unitlength}{.03\textwidth}

586: 

587: \begin{picture}(30,15)(0,0)

588: %critical values

589: \put(0,5){

590:    \begin{picture}(15,15)(0,0)

591: \put(0,0){\circle*{.3}} \put(0,0.5){$h_1$}

592: \put(3,0){\circle*{.3}}  \put(3,0.5){$h_0$}

593: \put(1.5,1.5){\circle*{.3}} \put(1.5,2){$h_\lambda$}

594: \put(1.5,-1.5){\circle*{.3}} \put(2,-1.5){$h_\mu$}

595: \put(3,0){\thicklines{\vector(1,0){9}}}

596: 

597: \qbezier[20](0,0)(0,5)(6,6)

598: \qbezier[20](3,0)(3,2)(6,6)

599: \qbezier[20](1.5,1.5)(4,5)(6,6)

600: \qbezier[25](1.5,-1.5)(3,3.5)(6,6)

601: \put(6,6){\circle*{.3}} \put(6.5,6){$\tilde{h}$}

602:    \end{picture}}

603: %Dynkin diagram

604: \put(20,5){

605:    \begin{picture}(15,15)(0,0)

606: \put(0,0){\line(1,0){3}} \put(0,0){\circle*{.3}}

607: \put(3,0){\line(1,1){2.1213}} \put(3,0){\circle*{.3}}

608: \put(3,0){\line(1,-1){2.1213}}

609: \put(5.1213,2.1213){\circle*{.3}}

610: \put(5.1213,-2.1213){\circle*{.3}}

611: 

612: \put(2.5,0.5){$\delta _0$}

613: \put(0,0.5){$\delta _1$}

614: \put(5.6213,2.1213){$\delta _\lambda$}

615: \put(5.6213,-2.1213){$\delta _\mu$}

616: 

617: \put(5.1213,-2.1213){\line(0,1){4.2426}}

618:    \end{picture}

619:               }

620: \put(7,0){ (i)}

621: \put(22,0){ (ii)}

622: 

623: \end{picture}

624: \begin{picture}(30,15)(0,0)

625: %Dynkin diagram

626: \put(3,7){

627:  \begin{picture}(15,15)(0,0)

628:  \put(0,0){\line(1,0){3}} \put(0,0){\circle*{.3}}

629:  \put(3,0){\line(1,1){2.1213}} \put(3,0){\circle*{.3}}

630:  \put(3,0){\line(1,-1){2.1213}}

631:  \put(5.1213,2.1213){\circle*{.3}}

632:  \put(5.1213,-2.1213){\circle*{.3}}

633: 

634:  \put(2.5,0.5){$\delta _1$}

635:  \put(0,0.5){$\delta _0$}

636:  \put(5.6213,2.1213){$\delta _\lambda$}

637:  \put(5.6213,-2.1213){$\delta _\mu$}

638: 

639:  \put(5.1213,-2.1213){\line(0,1){4.2426}}

640:     \end{picture}

641:           }

642: %Dynkin diagram

643: \put(20,5){

644:  \begin{picture}(15,15)(0,0)

645:  \put(0,0){\line(1,0){4}} \put(0,0){\circle*{.3}}

646:  \put(4,0){\line(0,1){4}} \put(4,0){\circle*{.3}}

647:  \put(0,4){\line(1,0){4}}

648:  \put(0,0){\line(0,1){4}}

649:  \put(4,4){\circle*{.3}}

650:  \put(0,4){\circle*{.3}}

651: 

652:  \put(-.5,- 0.8){$\delta _0$}

653:  \put(4.3,-.8){$\delta _\lambda $}

654:  \put(-.5,4.5){$\delta _\mu$}

655:  \put(4.3,4.5){$\delta _1$}

656: 

657:  \put(0,0){\line(1,1){4}}

658:  \end{picture}

659:               }

660: \put(7,0){ (iii)}

661: \put(22,0){ (iv)}

662: 

663: \end{picture}

664: \caption{(i) The paths $l_0$, $l_1$, $l_\lambda$, $l_\mu$ in the case when

665: $\lambda \not \in \R$ and $h_\lambda \neq h_\mu $;

666: (ii) Dynkin diagram in the case when $\lambda \not \in \R$ lies inside the

667: left branch of the hyperbola $\Gamma$;

668: (iii) Dynkin diagram in the case when $\lambda \not \in \R$ lies inside the

669: right branch of the hyperbola $\Gamma$;

670: (iv) Dynkin diagram in the case when $\lambda \not \in \R$ lies between the

671: two branches of the hyperbola $\Gamma$.}

672: \label{dynkin1}

673: \end{figure}

674: 

675: We finish this section by introducing a property of the continuous

676: families of ovals which is dependent on the structure of the related

677: Dynkin diagram. Suppose that the real polynomial $H= y^2+ P(x)$ has

678: distinct critical values. Let $\delta (h) \subset \{H=h\}$ be a

679: continuous family of ovals defined on a maximal open interval

680: $\Sigma =(h_c,h_s)$, where for $h=h_c$ the oval degenerates to a point

681: $\delta (h_c)$ which is a center and for $h=h_s$ the oval becomes a

682: homoclinic loop of the Hamiltonian system $dH=0$. The family

683: $\{ \delta (h)\}$ represents a continuous family of cycles vanishing

684: at the center $\delta (h_c)$.

685: To formulate our main result we shall need the following

686: \begin{definition}

687: \label{excd}

688: {\em It is said that $\{ \delta (h)\}$ is {\em an exceptional family of ovals},

689: provided that for every polynomial one-form $\omega $ the Abelian integral

690: $$

691: I(h)= \int_{\delta (h)} \omega , \quad h\in \Sigma

692: $$

693: allows an analytic continuation in the sector $S_\varepsilon (h_s)= \{h\in \C:

694: Arg (h-h_s) \in (-\varepsilon, 2\pi + \varepsilon )\}$ for some

695: strictly  positive

696: $\varepsilon$. The corresponding family of vanishing cycles represented by

697: $\delta (h)$ is called {\em an exceptional family of vanishing cycles.}}

698: \end{definition}

699: The above definition has in fact a geometric nature.

700: Indeed, $I(h)$ has an analytic continuation in

701: the sector  $S_\varepsilon (h_s)$ if and only if $\delta (h)$ has an

702: appropriate intersection number with each of the other families of

703: vanishing cycles. For instance, if $\lambda$, $\mu$ are real and

704: $h_1 < h_\mu < h_\lambda < h_0$, then the family of cycles $\delta_1(h)$ is

705: exceptional, because it has zero intersection number with $\delta _0(h)$.

706: On the contrary, the family $\delta _\mu (h)$ is not exceptional, because it

707: has a non-zero intersection number with  $\delta _0(h)$.

708: The remaining possible distributions of the critical

709: values, as well as the case $\lambda, \mu \not \in \R$, are studied in the

710: same way. We summarize all this in the following

711: \begin{proposition}

712: \label{exceptional}

713: Suppose that the Hamiltonian $H(x,y)$ is taken in a normal form

714: $(\ref{normal})$ and has four distinct critical values. Then:

715: 

716: \vspace{1ex}

717: \noindent

718: $(i)$

719: In the real case $(0 <\mu< \lambda <1)$, the continuous family of ovals

720: ${\cal O}_1$ surrounding the  center

721: $(1,0)$ is exceptional, and the continuous family of ovals ${\cal O}_\mu$

722: surrounding the center $(\mu,0)$ is not exceptional.

723: 

724: \vspace{1ex}

725: \noindent

726: $(ii)$

727: In the complex case $(\lambda = \bar{\mu }\not \in \R)$,

728: the continuous family of  ovals ${\cal O}_1$

729: is exceptional if and only if $\lambda$ lies inside the left

730: branch of the hyperbola $\Gamma$ $($see

731: $(\ref{hyperbola})$ and {\em Fig. \ref{imconj})}.

732: \end{proposition}

733: 

734: \section{The period integral in a complex domain}

735: Assume that the level curve $\{H=h\}$ contains an oval $\delta(h)$ and

736: consider the integral (oriented along with the vector field (\ref{system}))

737: $$I_0(h)=\oint_{\delta(h)}\frac{dx}{y}$$

738: which is the derivative of the area inside $\delta(h)$ and hence

739: determines the period of the periodic orbit lying on $\delta(h)$.

740: Denote by $\alpha(h)$ and by $\beta(h)$ the minimal and the maximal

741: values of $x$ for $(x,y)\in\delta(h)$. These are among the real solutions

742: of the equation $P(x)=h$. Then one can express $I_0(h)$ in the

743: form

744: $$I_0(h)=\sqrt{2}\int_{\alpha(h)}^{\beta(h)}\frac{dx}{\sqrt{h-P(x)}}.$$

745: Consider first the case $0<\mu<\lambda<1$ and assume that $\delta(h)$ belongs

746: to either of the continuous families of

747: ovals ${\cal O}_\mu$, ${\cal O}_1$ surrounding centers $(\mu,0)$ and

748: $(1,0)$, respectively. The corresponding families of cycles vanish

749: respectively at $h=h_\mu$ and $h=h_1$. Denote this value in both cases

750: by $h_c$ and assume that the real ovals are defined in  $\Sigma=(h_c, h_s)$.

751: Finally, denote by $x_c$ the $x$-coordinate of the corresponding center.

752: We keep the same notation in the complex case and in the

753: cases when either $\mu=0$ or $\mu=\lambda$  or $\lambda=1$.

754: Then $x_c$ is the $x$-coordinate of the unique nondegenerate center

755: (as long as $(\lambda,\mu)\not\in\{(1,0), (1,1)\}$).

756: 

757: The integral $I_0(h)$ has an analytic continuation for $h<h_s$. We wish to

758: prove that $I_0(h)\neq 0$ in  $(-\infty, h_c)$. The precise statement

759: is as follows.

760: 

761: \begin{proposition}

762: \label{nozero}

763: Assume that the parameters $\lambda,\mu$ in $(\ref{normal})$ are either

764: real or lie inside the hyperbola $\Gamma$. Then the integral $I_0(h)$

765: related to any period annulus around a nondegenerate center of

766: $(\ref{system})$ has an analytic continuation in $(-\infty,h_s)$

767: and takes positive values there.

768: \end{proposition}

769: {\bf Proof.}

770: The claim that $I_0(h)$ has an analytic continuation in the interval

771: $(-\infty,h_s)$ is obvious in the complex case ($\lambda=

772: \bar{\mu}\not\in\R$). In the real case this follows from the Dynkin

773: diagram shown on Fig. \ref{realdynk}. For instance, if

774: $h_1 < h_\mu < h_\lambda < h_0$ as on Fig. \ref{realdynk}, and if

775: $h_s=h_\lambda$, $h_c= h_1$, then the claim is obvious. If $h_s=h_\lambda$,

776: $h_c= h_\mu$, then the claim is true again because the intersection number

777: of $\delta_\mu$ and $\delta_1$ is equal to zero. The remaining four

778: configurations of critical values and period annuli are examined in the

779: same way (with the help of Fig. \ref{realdynk}). Note, however, that the

780: integral $I_0(h)$ corresponding to the period annulus ${\cal O}_e$

781: (in this case $\Sigma =(h_\lambda, 0)$) has no analytic continuation on

782: the real axis outside $\Sigma$. This is so because the singularities in

783: both $h=h_0$ and $h=h_\lambda$ are unavoidable. The same conclusion holds

784: for some of the degenerate cases as well.

785: 

786: 

787: Next we shall prove the positivity of the period integral $I_0(h)$

788: on $(-\infty,h_s)$. We first choose a proper formula to present $I_0(h)$

789: when $h<h_c$. Take $z=x+iy\in\C$, $h<h_s$ and consider the polynomial

790: $h-P(z)$. By Taylor's formula,

791: $$\begin{array}{rl}

792: h-P(z)=&\!\!\!h-P(x)+\frac12y^2P''(x)-\frac{1}{24}y^4P''''(x)

793: -iy[P'(x)-\frac16y^2P'''(x)+\frac15y^4]\\[2mm]

794: \equiv&\!\!\!h-Q(x,y)-iyR(x,y)\end{array}$$

795: Given $h<h_s$, consider the equation  $h-P(z)=0$ and denote by

796: $\zeta_k(h)$, $k=1,2$, the two branches of the algebraic function defined by

797: $h-P(\zeta(h))\equiv 0$ which satisfy $\zeta(h_c)=(x_c,0)$.

798: These functions are unique since $P'(z)\neq 0$ for $z\neq 0,\mu,\lambda,1$.

799: For $h_c<h<h_s$, one has $\zeta_1(h)=(\alpha(h),0)=\alpha(h)$ and

800: $\zeta_2(h)=(\beta(h),0)=\beta(h)$ where $\alpha, \beta$ are as above.

801: For $h<h_c$, one has $\zeta_1(h)=\bar\zeta_2(h)\in\C$.

802: For a similarity in notation, we put $\zeta_1(h)=\alpha(h)$ and

803: $\zeta_2(h)=\beta(h)=\bar\alpha(h)$ where $Im\,\alpha<0$.

804: Denote by $C_h$ the curve in the complex plane connecting the

805: points $\alpha(h)$, $\beta(h)$ along $R(x,y)=0$, if $h<h_c$,

806: and along the real line $y=0$, if $h_c<h<h_s$. Then the integral

807: $I_0(h)$ is expressed as

808: \begin{equation}

809: \label{cmplx}

810: I_0(h)=\sqrt{2}\int_{C_h}\frac{dz}{\sqrt{h-Q(x,y)}},\quad h<h_s.

811: \end{equation}

812: Let $h<h_c$. We will establish below that $y=Im\,z$ can be used as a local

813: coordinate on $C_h$. Thus, $dz=[x'(y)+i]dy$ where $x(-y)=x(y)=x$. Denote for

814: short $y_h=Im\,\beta(h)$ and write $\int_{C_h}=\int_{-y_h}^0+\int_0^{y_h}$.

815: Replacing $y$ with $-y$ in the first integral, we obtain

816: \begin{equation}

817: \label{y}

818: I_0(h)=2\sqrt{2}\int_0^{y_h}\frac{dy}{\sqrt{Q(x(y),y)-h}},\quad h<h_c.

819: \end{equation}

820: 

821: Below we study the curve ${\cal R}:\;R(x,y)=0$ and establish that on this

822: curve the real-valued function $Q(x,y)-h$ is positive for all $h<h_c$.

823: 

824: Recall that $\cal R$ has an equation

825: $\frac15y^4-\frac16y^2P'''(x)+P'(x)=0$. We will study how this algebraic

826: curve changes when varying the parameters $\lambda,\mu$. It turns out that

827: a local analysis can be used to determine the global behavior of $\cal R$.

828: To begin with, we observe that the Poincar\'e index of any sufficiently big

829: circle $x^2+y^2=r^2$ subject to the vector field $dR(x,y)=0$ is $-3$.

830: The critical points of $R$ are determined from $R_x=R_y=0$, namely

831: 

832: \vspace{1ex}

833: (i) $y=0$, $P''(x)=0$;

834: 

835: \vspace{1ex}

836: (ii) $y^2=\frac{5}{12}P'''(x)$, $P''(x)-\frac{5}{72}P'''(x)P''''(x)=0$.

837: 

838: \vspace{1ex}

839: \noindent

840: It is easy to verify that the Hessian $R_{xx}R_{yy}-R_{xy}^2$ is negative

841: at any (real) nondegenerate critical point $(\xi,\eta)$. Indeed, it equals

842: $-\frac13(P'''(\xi))^2$ and

843: $-\frac{5}{108}P'''(\xi)(P''''(\xi))^2-\frac49(P'''(\xi))^2$, respectively

844: in cases (i) and (ii). Hence, $R$ has in general three nondegenerate saddles

845: and when two of the saddles collide, the resulting degenerate

846: critical point has only hyperbolic sectors around. In particular,

847: (ii) has no solutions for $0<\mu<\lambda<1$, since all the

848: three critical points are then given by (i). Also, in this case the curve

849: $\cal R$ is free of critical points because the value of $R$ at the saddle

850: $(\xi,0)$ is $P'(\xi)\neq 0$. In general, one can prove that $\cal R$

851: has no critical points outside the real line. Indeed, taking a critical point

852: $z=\xi+i\eta$ with $\eta\neq 0$, one has

853: $P'(z)=R(\xi,\eta)+\eta(R_y(\xi,\eta)+iR_x(\xi,\eta))=R(\xi,\eta)\neq 0$

854: unless $z=\lambda$ or $z=\mu$ (which may happen in the complex case).

855: However, by (ii), $(Re\,\lambda,\pm Im\,\lambda)$ is a critical point

856: if and only if $\lambda$ lies on the hyperbola $\Gamma$,

857: which is not our case. We have proved that the only critical points

858: $\cal R$  may have are $(0,0)$, $(\lambda,0)$ and $(1,0)$

859: provided that $\mu=0$, $\lambda=\mu$, $\lambda=1$, respectively.

860: 

861: As a consequence, the branches of the level curve $\cal R$ do not intersect

862: outside the real line and all they do escape to infinity (in the directions

863: of $x$ and $y$ altogether, as shown by the asymptotics at infinity). This

864: is because any closed compact curve necessarily surrounds a critical point

865: having an index $+1$ and $R(x,y)=const$ has no such points.

866: 

867: Assume first that $0<\mu<\lambda<1$. Then the four branches of $\cal R$ have

868: no common points at all. This implies that the branches through

869: $(0,0)$ and $(\mu,0)$ go to $-\infty$ in the $x$-direction while the

870: branches through $(\lambda,0)$ and $(1,0)$ go to $+\infty$, see Fig.

871: \ref{curve}. When running any of the branches, the $y$ variable changes

872: in a monotone way. (Otherwise, there would exist a horizontal line $y=const$

873: having at least 6 intersections with $\cal R$ which is impossible.)

874: Finally, denote by $\xi$ any of $0,1,\lambda,\mu$. The local equation

875: of $\cal R$ near the point $(\xi,0)$ is

876: $y^2=[6P''(\xi)/P'''(\xi)](x-\xi)+O((x-\xi)^2)$ where the coefficient

877: is negative for $\xi=0$, positive for $\xi=1$ and changes sign if

878: $\xi=\lambda,\mu$. This latter is because $P'''(\lambda)$ and

879: $P'''(\mu)$ change sign in $T$.

880: On the $(\lambda,\mu)$ curve where $P'''$ vanishes, the local equation

881: of $\cal R$

882: becomes $y^4\sim-5P''(\xi)(x-\xi)$ for $\xi=\lambda,\mu$.

883: Summing up, we obtain the form of the curve $\cal R$ as given in

884: Fig. \ref{curve}.

885: The arrows indicate the direction of the Hamiltonian vector field $dR=0$.

886: 

887: \begin{figure}[htbp]

888: \begin{center}

889:   \psfig{file=save2.eps}

890: \end{center}

891: \caption{The curve $R(x,y)=0$ for the  case $0<\mu<\lambda<1$.}

892: \label{curve}

893: \end{figure}

894: 

895: It is easy to see that $q(y)=Q(x(y),y)-h$ does not change sign when

896: $y\in(0,y_h)$. As $q(y_h)=0$, it suffices to show that $q'(y)\neq 0$.

897: We first note that $\dot{y}=-R_x<0$ on the branch through $(h_c,0)$

898: we consider. Then, by Cauchy-Riemann equations,

899: $$Q_x=(yR)_y,\quad Q_y=-(yR)_x,$$

900: and we obtain on $R=0$ that

901: $$q'=Q_y+x'Q_x=-\frac{y}{R_x}(R_x^2+R_y^2)<0.$$

902: This implies that $Q(x(y),y)-h>0$ for $y\in(0,y_h)$

903: and hence by (\ref{y}), $I_0(h)>0$ for $h<h_s$.

904: 

905: The same analysis is applicable when $\mu=0$ or $\mu=\lambda$ or $\lambda=1$,

906: the unique difference is that on Fig. \ref{curve} the corresponding

907: branches touch each other at $y=0$.

908: 

909: Let us turn now to the complex case. Apart of the real case, the branches

910: through $(0,0)$ and $(1,0)$ (let us denote them by $R_0$ and $R_1$) may

911: bifurcate significantly. This is because they are not necessarily monotone

912: with respect to $y$ and may escape to either $-\infty$ or $+\infty$, in the

913: $x$-direction. The complex plane is divided by the hyperbola $\Gamma$

914: into three parts. In the outer domain

915: $(Im\,\lambda)^2>5Re\,\lambda(Re\,\lambda-1)$, $R_0$ and $R_1$ behave as

916: in Fig. \ref{curve},

917: but are not necessarily monotone in $y$. Inside the left branch of the

918: hyperbola, $R_0$ and $R_1$ look like the branches through $\lambda$ and $1$

919: on Fig. \ref{curve}, respectively. And inside the right branch, $R_0$ and

920: $R_1$ behave as the branches through $0$ and $\mu$ on Fig. \ref{curve},

921: respectively. All this is easily verified by a simple deformation argument,

922: with a starting point on the appropriate side of triangle $T$.

923: 

924: To avoid complications, we regard only the parameters $\lambda$,

925: $\mu$ in the complex plane which belong to the interior of $\Gamma$

926: (which is the case to be considered in our main theorem). Taking

927: $\lambda=a+ib$, then $b^2<5a(a-1)$.

928: We are going to prove that $R_1$ is monotone with respect to $y$

929: in this domain. If so, one obtains just as in the real case that

930: $q(y)=Q(x(y),y)-h>0$ and the proof is complete.

931: 

932: Consider first the interior of the left branch of $\Gamma$. Then $R_1$ lies

933: in the half-plane $x\geq 1$. To establish monotonicity, we rewrite $R(x,y)$

934: as a polynomial in $t=x-1$. One obtains

935: \begin{equation}

936: \label{descartes}

937: R=t^4+(3-2a)t^3+(3-4a+a^2+b^2-2y^2)t^2+(1-2a+a^2+b^2+(2a-3)y^2)t+R(1,y).

938: \end{equation}

939: Assume that $a\leq\frac12$. As the first two coefficients of the polynomial

940: are positive, then according to Descartes rule of signs, (\ref{descartes})

941: can have three positive roots only if the coefficient at $t^2$ is negative

942: and the coefficient at $t$ is positive. However, these two conditions

943: contradict each other. Therefore each line $y=const$ can intersect $R_1$

944: in at most two points (counting multiplicity) and hence in a single point

945: because the number of intersections should be an odd number.

946: 

947: Now take $\lambda$ in the interior of the right branch of $\Gamma$ and denote

948: by $(\xi,\eta)$ the right-most point of $R_1$ in the half-plane $y\geq 0$.

949: Thus, $\xi\geq 1$ and $\eta\geq 0$. As above, one obtains with

950: $t=\xi-x \geq 0$ that

951: \begin{equation}

952: \label{descart1}

953: {\textstyle

954: R=t^4-\frac16P''''(\xi)t^3+(\frac12P'''(\xi)-2y^2)t^2-(P''(\xi)

955: -\frac16P''''(\xi)y^2)t+\frac15(y^2-\frac{5}{12}P'''(\xi))^2.}

956: \end{equation}

957: As $\eta=0$ if $P'''(\xi)\leq 0$ and $\eta^2=\frac{5}{12}P'''(\xi)$

958: otherwise, the coefficient at $t^2$ is negative for $y>\eta$.

959: Then by Descartes rule,

960: (\ref{descart1}) has only two positive roots and therefore the part of

961: $R_1$ above the line $y=\eta$ is monotone. It remains to consider the

962: part below it (if $\eta>0)$. In this domain (\ref{descart1}) can

963: have two complex roots and Descartes rule does not yield the required result.

964: We can proceed as follows. Assume that the part of $R_1$ in the upper

965: half-plane lying below $y=\eta$

966: is not monotone. Denote by $(\xi_1,\eta_1)$ the first local maximum on $R_1$

967: (starting from $(1,0)$), thus $1<\xi_1<\xi$ and $0<\eta_1<\eta$. Consider

968: also the branch $R_+$ of $\cal R$ which is placed in the upper half-plane

969: to the right of the line $x=\xi$ (and above the line $y=\eta_1$).

970: An easy calculation yields the equations

971: $y=\pm\sqrt{5\pm2\sqrt5}(x-\frac{1+2a}{4})$

972: for the tangential lines of $\cal R$ at infinity. As their common point

973: is different from $(\xi_1,\eta_1)$, there would always exist a tangential

974: line $l$ to $R_+$ at a finite point, which goes through $(\xi_1,\eta_1)$.

975: However, then $l$ would intersect $R_+$ in at least two points and $R_1$

976: in at least three points, which is impossible because $\cal R$ is a quartic

977: curve. The monotonicity of $R_1$ is established. $\Box$

978: 

979: \vspace{2ex}

980: \noindent

981: \begin{proposition}

982: \label{nozero1}

983: Assume that $\delta(h)$ is an exceptional family of ovals according to

984: Definition $\ref{excd}$. Then the period integral

985: $I_0(h)= \int_{\delta(h)} dx/y$

986: has an analytic continuation in the complex domain ${\cal D}= \C \setminus

987: [h_s,\infty)$ and does not vanish there. \end{proposition}

988: {\bf Proof.}

989: The proof that $I_0(h)$ has an analytic continuation in $\cal D$

990: follows immediately from the definition of an exceptional family of ovals.

991: To count the zeros of the analytic function $I_0(h)$ in $\cal D$,

992: we apply the argument principle to the complex domain

993: ${\cal D}_r$  obtained from $\cal D$ by removing the discs

994: $\{h\in \C: |h-h_{s}|< r \}$, $\{h\in \C: |h| > 1/r \}$, where $ r>0$

995: is a small enough constant.  For this purpose we

996: evaluate the increase of the argument of $I_{0}(h)$ when $h$ runs

997: the boundary of ${\cal D}_r$ in a positive direction.

998: 

999: We consider first the real case ($0< \mu < \lambda < 1$).

1000: The Picard-Lefschetz formula implies that in a suitable complex

1001: neighborhood of $h_s$ it holds

1002: $$I_{0}(h) = \varphi(h) \log(h-h_{s}) + \psi(h), \quad

1003: \varphi(h)= \frac{1}{2\pi} \int_{\delta_{s}(h)} \frac{dx}{y}

1004: $$

1005: where $\varphi,\psi$ are locally holomorphic and, as it is easily

1006: checked, $\varphi(h_s)\neq 0$.

1007: 

1008: 

1009: The behavior of $I_{0}(h)$ for $|h|$ sufficiently big follows from

1010: the Picard-Lefschetz formula too, but the argument is more delicate.

1011: Let $l$ be a simple closed path which makes one turn in a positive

1012: direction around all the critical values of $H$. It induces a

1013: monodromy map

1014: $$

1015: l_* : H_1(\Gamma_{\tilde{h}},\Z) \rightarrow H_1(\Gamma_{\tilde{h}},\Z),

1016: \quad \Gamma_{\tilde{h}}= \{(x,y)\in \C:\, H(x,y)=\tilde{h} \} \; .

1017: $$

1018: It is shown for instance in \cite{gav97} that $l_{*}$ coincides

1019: with the operator of classical monodromy of the singularity

1020: $\frac{1}{2}y^2 + \frac{1}{5}x^5 $ (which is the highest

1021: weighted homogeneous part of $H$). The characteristic polynomial $P(z)$

1022: of $l_*$ is now easily computed to be equal to

1023: $P(z)= (z^5+1)/(z+1)$ (see \cite{bri}). This shows that in a neighborhood

1024: of $h=\infty$ on the projective sphere the integral

1025: $I_0(h)$ is a meromorphic function of $h^{1/10}$. The substitution

1026: $$

1027: x \rightarrow x h^{1/5},\quad y \rightarrow y h^{1/2},\quad h \rightarrow

1028: \infty

1029: $$

1030: defines an isotopy of the regular fibers of $H$ and the regular fibers

1031: of its weighted homogeneous part

1032: $\frac{1}{2}y^2 + \frac{1}{5}x^5 $ \cite{gav97}.

1033: From this fact we deduce that

1034: $$I_{0}(h)=c_{\infty}h^{1/5-1/2} (1+O(h^{-1/10})).$$

1035: For a further use we note that

1036: $c_{\infty} \neq 0$. Indeed, the one-form $dx/y$ is not cohomologous to zero

1037: on the regular fibers of the polynomial $\frac{1}{2}y^2 + \frac{1}{5}x^5 $

1038: (this follows from \cite[Theorem 1.1]{gav98a}).

1039: On the other hand, the form of the characteristic polynomial of $l_{*}$ shows

1040: that it is irreducible over the field of rational numbers. Therefore the span

1041: of $l_*^k \delta$, $k=0, 1, 2, 3$ generates the whole

1042: homology group of the fiber. This implies that if $c_\infty=0$, then the

1043: restriction of  $dx/y$ on the regular fibers of

1044: $\frac{1}{2}y^2 + \frac{1}{5}x^5$  defines the zero

1045: cohomology class which is a contradiction.

1046: 

1047: 

1048: The above consideration shows that the decrease of the argument of $I_{0}(h)$

1049: along $\{h\in \C: |h| = 1/r \}$ is close to $3 \pi/5$ and the increase of the

1050: argument of $I_{0}(h)$  along $\{h\in \C: |h-h_{s}| = r \}$ is close to zero.

1051: 

1052: 

1053: We claim further that the imaginary part of $I_{0}(h)$ along

1054: $(r+h_s,1/r)$ does not vanish. Namely, denote the two determinations of

1055: $I_{0}(h)$ on $(r+h_s,1/r)$ by $I_{0}^{\pm}(h)$. Similarly, denote by

1056: $\delta^{\pm}(h)$ the two determinations of the continuous family

1057: $\delta(h)$ on $(r+h_s,1/r)$.

1058: Since $I_{0}$ is a real analytic function  along $(-\infty, h_{s})$, we have

1059: $$I_{0}^{+}(h)= \overline{I_{0}^{-}(h)},\quad

1060: \delta^{+}(h) =\overline{\delta^{-}(h)}, \quad h>h_{s}

1061: $$

1062: and hence $2\,\sqrt{-1} \, Im\, I_{0}^{+}(h) =

1063: I_{0}^{+}(h) - I_{0}^{-}(h)$. On the other hand $\delta(h)$ is an exceptional

1064: family of ovals. Therefore the Picard-Lefschetz formula implies that for

1065: $h> h_s$,

1066: $$

1067: \delta^{+}(h) - \delta^{-}(h) = \gamma(h),\qquad

1068: I_{0}^{+}(h) - I_{0}^{-}(h) = \int_{\gamma(h)} \frac{dx}{y}

1069: $$

1070: where $\gamma(h)$ is a continuous family of cycles vanishing at

1071: $h=h_{s}$, with intersection number $\langle\gamma,\delta^-\rangle=1$.

1072: 

1073: Finally, we note that after substituting $y$ by $\sqrt{-1} y$,

1074: the family of cycles $\gamma(h)$ corresponds to a family

1075: of ovals surrounding a center. Moreover, in the real case the full

1076: bifurcation diagram (consisting of the boundary of $T$ and the parts of

1077: the curves $h_\lambda=h_0$, $h_\mu=h_1$, $h_0=h_1$ inside) images onto

1078: itself. For later use we also mention that in the complex case, the

1079: interior of the hyperbola $\Gamma$ images onto itself. All this holds

1080: because the corresponding Hamiltonians when taken in a normal form

1081: (\ref{normal}) are related by the formula

1082: $$H(x, \sqrt{-1}y,\lambda,\mu)=h_1-H(1-x,y,1-\mu,1-\lambda).$$

1083: Hence, it follows from Proposition \ref{nozero}  that

1084: $\int_{\gamma(h)} dx/y$ does not vanish along $(h_{s},\infty)$. Therefore

1085: if  $h\in (r+h_s,1/r)$, then $Im\, I_{0}^{\pm}(h) \neq 0$ and hence

1086: the increase of the argument of $Im\, I_{0}^{\pm}(h)$

1087: along $(r+h_{s},1/r)$ is at most $\pi$.

1088: 

1089: Summing up the above information we conclude that the increase of the

1090: argument of $I_{0}(h)$ along the boundary of ${\cal D}_r$ is strictly

1091: less than $2\pi$. By the argument principle, $I_{0}(h)$ does not vanish

1092: in ${\cal D}_r$, and hence in $\cal D$.

1093: 

1094: The complex case when $\lambda = \bar{\mu }\not \in \R$ is studied in the

1095: same way. $\Box$

1096: 

1097: \section{The Chebyshev property}

1098: Suppose that the real polynomial $H= y^2+ P(x)$ has distinct critical values.

1099: Let $\delta (h) \subset \{H=h\}$ be a continuous family of ovals defined on

1100: a maximal open interval

1101: $\Sigma =(h_c,h_s)$, where for $h=h_c$ the oval degenerates to a point

1102: $\delta (h_c)$ which is a

1103: center and for $h=h_s$ the oval becomes a homoclinic loop of the

1104: Hamiltonian system $dH=0$. Denote ${\cal D}=\C\setminus[h_s,\infty)$.

1105: The following theorem is our main result.

1106: 

1107: 

1108: \begin{theorem}

1109: \label{main}

1110: Let $\delta (h)$, $h\in \Sigma $ be an exceptional family of ovals

1111: of $(\ref{normal})$. Then the real vector space

1112: ${\cal A}=\{\alpha_0 I_0 + \alpha_1 I_1: \alpha_0, \alpha_1 \in \R \}$

1113: of Abelian integrals

1114: $$I_0(h) =  \int_{\delta (h)} \frac{ dx}{y}, \qquad I_1(h) =

1115: \int_{\delta (h)} \frac{x dx}{y}, \qquad h\in \Sigma $$

1116: is Chebyshev in $\cal D$ and hence in $\Sigma$.

1117: \end{theorem}

1118: {\bf Proof.}

1119: Let $\delta(h)$ where $h \in \Sigma = (h_{c},h_{s})$ be an exceptional

1120: family of ovals. According to Proposition \ref{nozero1}, it suffices to

1121: show that the analytic function $F(h)= \alpha_0 + I_{1}(h)/I_{0}(h)$ has

1122: at most one zero in the domain $\cal D$. To prove this claim we apply the

1123: argument principle to $F(h)$ in the complex domain ${\cal D}_r$ obtained

1124: from $\cal D$ by removing the discs

1125: $\{h\in \C: |h-h_{s}|< r \}$, $\{h\in \C: |h| > 1/r \}$,

1126: where $ r>0$ is a small enough constant.

1127: 

1128:  As in the proof of Proposition \ref{nozero1}, we may check that:

1129: \begin{itemize}

1130:      \item  if $h\sim \infty$, then $I_{1}(h)$ is meromorphic

1131:       with respect to $h^{-1/10}$ and

1132:       $$I_{1}(h) = c_{\infty} h^{2/5-1/2}(1+O(h^{-1/10}))

1133:       \;\;\mbox{\rm where}\;\; c_{\infty} \neq 0;$$

1134: 

1135:       \item if $h \sim h_{s}$, then

1136:       $$I_{1}(h) = \varphi_1(h) \log(h-h_{s}) + \psi_1(h),

1137:  \quad \varphi_1(h)= \frac{1}{2\pi} \int_{\delta_{s}(h)} \frac{x\,dx}{y},$$

1138:   where $\varphi_1$ and $\psi_1$ are locally holomorphic. Moreover,

1139:   it is easily verified that

1140:   $\lim\limits_{h \rightarrow h_{s}} I_{1}(h)/I_{0}(h)=

1141:   \varphi_1(h_s)/\varphi(h_s) = x_{s} $

1142:   where $(x_{s},0)$ is the saddle point corresponding to the

1143:   critical value $h_{s}$.

1144: \end{itemize}

1145: 

1146: \noindent

1147: Therefore the increase of the argument of $F(h)$ along

1148: $\{h\in \C: |h| = 1/r \}$

1149: is close to $2 \pi/5$ and the increase of the argument of $F(h)$

1150: along $\{h\in \C: |h| = r \}$ is close to zero or strictly negative.

1151: 

1152: We claim further that

1153: \begin{itemize}

1154:         \item  the imaginary part of $F(h)$ along $(r+h_{s},1/r)$

1155:         does not vanish.

1156: \end{itemize}

1157: 

1158: 

1159: \noindent

1160: If this claim were proved then summing up the above information we would get

1161: that the argument of $F(h)$ increases along the boundary of ${\cal D}_r$

1162: by at most $2\pi + 2 \pi/5$. Thus the

1163: function $F(h)$ can have at most one zero in  ${\cal D}_r$  and hence

1164: in $\C \backslash [h_{s}, \infty)$.

1165: 

1166: It remains to prove the above claim. Recall that, since $\delta (h)$ is an

1167: exceptional family of ovals, then the two determinations of $F(h)$

1168: over $(h_s,\infty)$, namely $F^\pm(h)$, are analytic

1169: along $(h_s,\infty)$. The function $F(h)$ is real analytic along

1170: $(-\infty,h_s)$ and hence

1171: $$

1172: 2\,\sqrt{-1}\,Im\, F(h) = F^+(h) - F^-(h) \;\; \forall h\in (h_s,\infty).

1173: $$

1174: Denote by $\delta ^\pm(h)$ the two determinations of

1175: $\delta (h)$, $h\in {\cal D}$ on $(h_s,\infty)$, and let

1176: $$\Gamma_{h}~=~\{(x,y)\in \C^2: H(x,y) =h\}.$$

1177: It was shown in the proof of Proposition~\ref{nozero1} that

1178: $$

1179: \int_{\delta^{\pm} (h)} \frac{ dx}{y} \neq 0\quad\forall h \in (h_s,\infty).

1180: $$

1181: Then we obtain for each $h\in (h_s,\infty)$

1182: $$

1183:    2\,\sqrt{-1}\,Im\, F(h)  =  \frac{\int_{\delta^+(h)} \frac{x dx}{y}}

1184:    {\int_{\delta^{+} (h)} \frac{dx}{y}}

1185:                   - \frac{\int_{\delta^{-} (h)} \frac{x dx}{y}}

1186:    {\int_{\delta^{-} (h)} \frac{ dx}{y}}

1187:  =  \frac{ det \left( \begin{array}{cc}

1188:                 \int_{\delta^{+} (h)} \frac{x dx}{y} & \int_{\delta^{+}

1189: (h)} \frac{ dx}{y}  \\[1mm]

1190:                 \int_{\delta^{-} (h)} \frac{x dx}{y} & \int_{\delta^{-}

1191: (h)} \frac{ dx}{y} \end{array} \right) }{

1192:         |\int_{\delta^{\pm} (h)} \frac{ dx}{y}|^{2}}.

1193: $$

1194: Therefore we have to prove that the analytic function

1195: $$

1196: \Delta (h) =  det \left( \begin{array}{cc}

1197:                 \int_{\delta^{+} (h)} \frac{x dx}{y} & \int_{\delta^{+}

1198: (h)} \frac{ dx}{y}  \\[1mm]

1199:                 \int_{\delta^{-} (h)} \frac{x dx}{y} & \int_{\delta^{-}

1200: (h)} \frac{ dx}{y} \end{array} \right),\quad  h \in (h_s,\infty)

1201: $$

1202: does not vanish.

1203: 

1204: Suppose first that $h=h_0 \in (h_s,\infty)$ is a critical value of $H$.

1205: The affine curve $\Gamma_{h_0}$ is singular of arithmetic genus two. Let

1206: $\bar{\Gamma} _{h_0}$ be the corresponding completed and normalized Riemann

1207: surface. As the geometric genus of $\bar{\Gamma} _{h_0}$ is one and the intersection

1208: number $\langle \delta^{+},\delta^- \rangle$ equals $+1$, then

1209: $\delta^{+}(h_0)$ and $\delta^-(h_0)$ form a canonical homology basis of

1210: $\bar{\Gamma} _{h_0}$. We have $P(x)=(x-x_0)^2(x-x_1)(x-x_2)(x-x_3)$

1211: where $x_i\neq x_j$ for $i\neq j$. Therefore, $(x-x_0)dx/y$ induces a

1212: holomorphic one-form on $\bar{\Gamma} _{h_0}$ and $dx/y$ induces a

1213: meromorphic one-form of the third kind on $\bar{\Gamma} _{h_0}$ with simple

1214: poles at $P^\pm$, where $P^\pm$ are the two distinct pre-images

1215: of the singular point $(x_0,0)\in \Gamma _{h_0}$ under the canonical

1216: projection $ \bar{\Gamma} _{h_0}\rightarrow \Gamma_{h_0}$. The reciprocity

1217: law for differentials of the first and third kind \cite{gh} applied to

1218: $(x-x_0)dx/y$, $dx/y$ on the elliptic curve $\bar{\Gamma} _{h_0}$ gives

1219: \begin{eqnarray*}

1220: \Delta (h) & = & 2\pi \sqrt{-1}\left[

1221: Res|_{P^+}\left(\frac{dx}{y} \int_{P^0}^{P} \frac{(x-x_0)dx}{y}\right)

1222: + Res|_{P^-}\left(\frac{dx}{y} \int_{P^0}^{P} \frac{(x-x_0)dx}{y}\right)

1223: \right]\\

1224: &=& \frac{\pm 2 \pi \sqrt{-1}}{\sqrt{(x_0-x_1)(x_0-x_2)(x_0-x_3)}}

1225: \int_{P^-}^{P^+}\frac{dx}{\sqrt{(x-x_1)(x-x_2)(x-x_3)}

1226: }

1227: \end{eqnarray*}

1228: where the path of integration is on the Riemann surface

1229: $ \bar{\Gamma} _{h_0}$ of the affine elliptic curve

1230: $$\{y^2= (x-x_1)(x-x_2)(x-x_3) \}$$

1231: and $P^\pm = (x_0,\pm \sqrt{(x_0-x_1)(x_0-x_2)(x_0-x_3)}) $. The function

1232: $$

1233: P \rightarrow \int^P \frac{dx}{\sqrt{(x-x_1)(x-x_2)(x-x_3)}}

1234: $$

1235: is an uniformizing variable on the elliptic curve $ \bar{\Gamma} _{h_0}$

1236: which shows that

1237: $$

1238: \int_{P^-}^{P^+}\frac{dx}{\sqrt{(x-x_1)(x-x_2)(x-x_3)}} = 0

1239: \;\;\Rightarrow\;\; P^+=P^- .

1240: $$

1241: As $P^+ \neq P^-$, then $\Delta (h_0) \neq 0$.

1242: 

1243: 

1244: Suppose now that $h\in (h_s,\infty)$ is a regular value of $H$. The

1245: intersection number $\langle \delta^{+},\delta^- \rangle$ equals $+1$ and

1246: it is easy to check that we may choose a canonical homology basis

1247: $\delta_{1}(h),\ldots,\delta_{4}(h)$ of the lattice

1248: $H_{1}(\bar{\Gamma}_{h},\Z)$ such that $\delta_{1}(h) = \delta^{+}$,

1249: $\delta_{3}(h) = \delta^{-}$. Further, let $\omega_{1}, \omega_{2}$ be a

1250: normalized basis of holomorphic differentials on $\bar{\Gamma}_{h}$. The

1251: corresponding period lattice $\Pi$ reads

1252: $$

1253: \Pi = \left(

1254: \begin{array}{cccc}

1255:         \int_{\delta_{1} (h)} \omega_{1} & \int_{\delta_{2} (h)} \omega_{1} &

1256:         \int_{\delta_{3} (h)} \omega_{1} & \int_{\delta_{4} (h)} \omega_{1}\\

1257: \int_{\delta_{1} (h)} \omega_{2} & \int_{\delta_{2} (h)} \omega_{2} &

1258:         \int_{\delta_{3} (h)} \omega_{2} & \int_{\delta_{4} (h)} \omega_{2}

1259: \end{array}

1260: \right) =

1261: \left(

1262: \begin{array}{cccc}

1263:         1 & 0 &

1264:         a & b  \\

1265: 0 & 1 &

1266:         c & d

1267: \end{array}

1268: \right)

1269: $$

1270: where

1271: $$

1272: \left(

1273: \begin{array}{c}

1274:         \omega _1 \\

1275:         \omega _2

1276: \end{array}

1277: \right) =

1278: \left(

1279: \begin{array}{cc}

1280:    \int_{\delta_{1} (h)} \frac{xdx}{y} & \int_{\delta_{2} (h)}\frac{xdx}{y}\\

1281:    \int_{\delta_{1} (h)} \frac{dx}{y} & \int_{\delta_{2} (h)}\frac{dx}{y}

1282: \end{array}

1283: \right)^{-1}

1284: \left(

1285: \begin{array}{c}

1286:         \frac{xdx}{y} \\

1287:       \frac{ dx}{y}

1288: \end{array}

1289: \right) .

1290: $$

1291: We get that

1292: $$

1293: \Delta (h) = \det \left(

1294: \begin{array}{cc}

1295:     \int_{\delta_{1} (h)} \frac{xdx}{y} & \int_{\delta_{2} (h)}\frac{xdx}{y}\\

1296:     \int_{\delta_{1} (h)} \frac{dx}{y} & \int_{\delta_{2} (h)}\frac{dx}{y}

1297: \end{array}

1298: \right) \det

1299: \left(

1300: \begin{array}{cc}

1301:         1 & a \\

1302:      0 & c

1303: \end{array}

1304: \right)

1305: $$

1306: and hence

1307: $$

1308: \Delta (h)=0 \;\;\Leftrightarrow \;\; c=0.

1309: $$

1310: Let $\Lambda $ be the lattice generated by the columns of $\Pi $.

1311: The Riemann bilinear relations \cite{gh} on $\Pi $ imply

1312: $$

1313: b=c.

1314: $$

1315: Therefore if $\Delta (h)=0$, then the Jacobian variety

1316: $J(\bar{\Gamma} _{h}) = \C^2/\Lambda $ is

1317: a direct product of the two elliptic curves

1318: \begin{equation}

1319: \label{curves}

1320: \C / \{\Z \oplus a \Z  \},\quad \C / \{\Z \oplus d \Z  \}.

1321: \end{equation}

1322: It is well known that this is impossible (e.g. the Remark on page 49 in

1323: \cite{arbarello}) which completes the proof of Theorem \ref{main}.

1324: 

1325: For convenience of the reader, we present a proof of the last claim.

1326: Consider the Riemann theta function

1327: $$

1328: \theta(z)=  \sum_{m\in \Z^{2}} e^{\pi \sqrt{-1} \langle m,Bm\rangle + 2 \pi

1329: \sqrt{-1}\langle m,z\rangle}

1330: $$

1331: and the corresponding theta divisor

1332: $\Theta = \{z \in  J(\bar{\Gamma}_{h}): \;\theta (z)=0 \}$ where

1333: $$

1334: B= \pmatrix{ a & b  \cr c & d},\quad z= \pmatrix{z_{1}\cr z_{2}},\quad

1335:  \langle m,z\rangle = m_{1}z_{1}+ m_{2}z_{2}.

1336: $$

1337: By the Riemann theorem, the divisor $\Theta$ is isomorphic to the curve

1338: $\bar{\Gamma} _{h}$ and in particular is irreducible. If $\Delta(h)=0$,

1339: then $b=c=0$ and $\theta(z)$ factorizes into a product of two elliptic

1340: theta functions

1341: $$

1342: \theta(z)=

1343: \sum_{m_{1}\in \Z} e^{\pi \sqrt{-1} a m_{1}^{2} + 2 \pi

1344: \sqrt{-1} m_{1} z_{1}}

1345: \sum_{m_{2}\in \Z} e^{\pi \sqrt{-1} d m_{2}^{2} + 2 \pi

1346: \sqrt{-1} m_{2} z_{2}}  .

1347: $$

1348: Therefore the Riemann theta divisor $\Theta$ is a product of the two

1349: elliptic curves (\ref{curves}) which is a contradiction.  $\Box$

1350: 

1351: \section{Concluding remarks}

1352: We finish this paper with some general hypotheses concerning the zeros of

1353: the function $\alpha_0I_0+\alpha_1I_1$. In the previous section we

1354: demonstrated that this linear combination belongs to a Chebyshev space,

1355: provided that the continuous family of ovals $\{\delta(h)\}$ is exceptional.

1356: In the non-exceptional case a big part of our proof still holds,

1357: provided that $\lambda$ and $\mu$ are real. This fact combined

1358: with the exact local result obtained in Proposition \ref{cyclicity}

1359: suggests the following more general statement.

1360: 

1361: \vspace{2ex}

1362: \noindent

1363: {\bf Conjecture 1.}

1364: Assume that the polynomial $P(x)$ has distinct real critical values.

1365:  Then the corresponding

1366: vector space of Abelian integrals $\alpha_0 I_0+\alpha_1I_1$

1367: is Chebyshev with accuracy at most one in ${\cal D}$.

1368: 

1369: %It turns out that, at least locally in a neighborhood of a center, these two

1370: %bounds do not coincide. The bifurcation curves in the $\lambda,\mu$-space

1371: %which separate domains in which the centers have different cyclicity turn

1372: %out to be algebraic, and hence globally defined. This suggests immediately

1373: %a global conjecture which we formulate below.

1374: 

1375: \vspace{2ex}

1376: \noindent

1377: One may also ask whether the upper bound (two) for the number of

1378: the zeros near the center always holds in the whole interval $\Sigma$ and

1379: next, whether the bounds for the real zeros when $h\in\Sigma$ and for the

1380: complex zeros when $h$ belongs to a certain complex extension of $\Sigma$

1381: do coincide. It turns out that the answers of all these questions are

1382: negative. In particular, we establish below that when $P(x)$ has

1383: complex critical values, then there are three zeros in $\Sigma$

1384: for appropriate $\lambda\in\C$.

1385: 

1386: 

1387: One can try to predict the possible number of real zeros by studying

1388: the behavior of the ratio $F(h)=I_1(h)/I_0(h)$ near the ends of

1389: the interval $\Sigma$. For reader's convenience, we first recall some

1390: formulas.

1391: 

1392: Let us take a continuous function $f(x)$ and consider the integral

1393: $$I(h)=\int_{\delta(h)}\frac{f(x)dx}{y},\quad h\in\Sigma$$

1394: where $\Sigma=(h_c,h_s)$ for the families ${\cal O}_1$, ${\cal O}_\mu$

1395: and $\Sigma=(h_s,h_0)$ for the family ${\cal O}_e$.

1396: Using Picard-Lefschetz formula or direct calculations of integrals,

1397: one can obtain the following asymptotical expansions in a neighborhood

1398: of the critical levels $h_s$ corresponding to a saddle or a cuspidal point

1399: $(x_s,0)$:

1400: 

1401: \vspace{1ex}

1402: \noindent

1403: (i) For the saddle-loop and eight-loop cases,

1404: $$I(h)=-c_sf(x_s)\log|h-h_s|+\varphi(h)$$

1405: where $c_s$ is a specific positive constant independent on $f$ and

1406: $\varphi(h)$ is a continuous function in a neighborhood of $h_s$.

1407: We are not going to write up the exact expression of $c_s$ since do

1408: not need it here.

1409: 

1410: \vspace{1ex}

1411: \noindent

1412: (ii) For the heteroclinic loop cases,

1413: $$I(h)=-(c_0f(0)+c_s f(x_s))\log|h-h_s|+\varphi(h)$$

1414: where $c_0$, $c_s$ and $\varphi(h)$ are as above.

1415: 

1416: \vspace{1ex}

1417: \noindent

1418: (iii) For the cuspidal loop cases,

1419: $$I(h)=c_sf(x_s)|h-h_s|^{-1/6}+\varphi(h),$$

1420: where $c_s$ and $\varphi(h)$ are as above.

1421: 

1422: \vspace{1ex}

1423: \noindent

1424: (In fact, these formulas hold in a much more general context.)

1425: Applying the above expansions to $F(h)$, we get $F(h_s)= x_s/(1+c_0/c_s)$

1426: for the heteroclinic loop case and $F(h_s)=x_s$ for all the remaining cases.

1427: Moreover, what concerns the saddle-loops, it is evident that

1428: $F(h)-x_s\neq 0$ in $\Sigma$ and the sign of the expression depends on

1429: the position of the saddle with respect to the family of ovals.

1430: Consider now the case of ${\cal O}_e$ where the situation is more

1431: complicated. For $h=h_\lambda$, we obtain from (\ref{normal}) that

1432: $$\textstyle y^2=\frac25(x-\lambda)^2p(x),\quad p(x)=(x-x_1)(x-x_2)(x_3-x)$$

1433: where $x_1<0<x_2\leq\lambda\leq x_3$ and $x_2$, $x_3$ are the left-most and

1434: the right-most points of the eight-loop (or the cuspidal loop, as a

1435: limit case). By (i)--(iii), the function $\varphi(h)=I_1(h)-\lambda I_0(h)$

1436: is continuous in a neighborhood of $h_\lambda$, which yields immediately that

1437: $$\varphi(h_\lambda)=

1438: \sqrt{10}\left(\int_\lambda^{x_3}-\int_{x_2}^\lambda\right)\frac{dx}

1439: {\sqrt{p(x)}}.$$

1440: Clearly, $\varphi(h_\lambda)\to-\infty$ when $x_2\to x_1$,

1441: $\varphi(h_\lambda)\to\pm \sqrt{10}\int_{x_2}^{x_3}dx/\sqrt{p(x)}$

1442: when $\lambda\to x_2$ and $\lambda\to x_3$, respectively.

1443: As $x_2=x_1$ is equivalent to $h_\lambda=0$,

1444: $\lambda=x_2$ is equivalent to $\lambda=\mu$ and $\lambda=x_3$

1445: is equivalent to $\lambda=1$, this means that the curve

1446: $$\gamma_s=\{(\lambda,\mu)\in T:\;I_1(h_\lambda)-\lambda I_0(h_\lambda)=0\}$$

1447: connects the points $(0,0)$ and $(1,1)$ and lies in the part of $T$ above

1448: $\gamma$. Surprisingly, although expressed in terms of incomplete elliptic

1449: integrals, it turns out that $\gamma_s$ is a part of an algebraic curve of

1450: degree 6. This can be verified by transforming the integrals to a standard

1451: form. From the explicit algebraic equation thus obtained it is easily seen

1452: that $\gamma_s$ is a simple and connected curve (we omit the details).

1453: The curve $\gamma_s$ is an element of the bifurcation diagram because for

1454: $h$ close to $h_\lambda$, the function $F(h)-\lambda$ will change the sign

1455: when crossing $\gamma_s$ in the parameter space.

1456: 

1457: Among other, the above analysis implies that for $\lambda=\mu$, the function

1458: $\alpha_0I_0+\alpha_1I_1=I_0(\alpha_0+\alpha_1 F)$ can have three zeros

1459: in $(h_1,h_\lambda)\cup(h_\lambda,h_0)$. Hence, the same fact remains true

1460: for ${\cal O}_1$ and the respective $\Sigma=(h_1,h_0)$ in the complex case,

1461: at least for parameters $\lambda=a+ib$ with $a\in(0,1)$ and $|b|$ small.

1462: Based on the behavior of $F$ near the other endpoint which corresponds

1463: to a center (see below), we conjecture that the domain in the parameter space

1464: $\lambda\in\C$ where the equation $F(h)=const$ can have three real zeros,

1465: is surrounded by a curve on which they join into a triple zero.

1466: 

1467: Looking at the bifurcation diagram again, let us denote by $\gamma_c$

1468: the curve

1469: $$\textstyle\gamma_c=\{(\lambda,\mu)\in T:\;

1470: \lambda=\frac{3\mu^2-2\mu}{2\mu-1}\}

1471: \quad\mbox{\rm (in the real case),}$$

1472: $$\gamma_c=\{\lambda\in \C:\; |\lambda-2|=1\}

1473: \quad\mbox{\rm (in the complex case).}$$

1474: Further, let us denote by $\gamma^*$ the (hypothetical) curve

1475: $$\gamma^*=\{\lambda\in \C:\;

1476: \exists h^*\in\Sigma\;\; \mbox{\rm with}\;\; F'(h^*)=F''(h^*)=0\}.$$

1477: We believe that $\gamma^*$ is a simple closed curve in the complex plane

1478: going through $(0,0)$ and $(1,0)$. On $\gamma^*$, a bifurcation

1479: of a triple real zero into a simple one occurs.

1480: 

1481: Next, in the real case, denote by $\Omega_\mu$ the part of $T$ placed

1482: between the curves $\gamma_c$ and $\gamma$, by $\Omega_e$ the part of

1483: $T$ placed between the curve $\gamma_s$ and the line $\lambda=\mu$

1484: and finally, in the complex case, denote by $\Omega_1$

1485: the unit disk inside $\gamma_c$ and by $\Omega_1^*$

1486: the interior of $\gamma^*$ (see figures \ref{imconj} and  \ref{reconj}).

1487: %\begin{figure}[htbp]

1488: %\begin{center}

1489: %  \psfig{file=save3.eps}

1490: %\end{center}

1491: %\caption{ Bifurcation diagram}

1492: %\label{reconj}

1493: %\end{figure}

1494: 

1495:         \begin{figure}

1496: \setlength{\unitlength}{.04\textwidth}

1497: 

1498: \begin{picture}(25,15)(5,0)

1499: \put(2,8){

1500:  \put(5,0){\vector(1,0){20}}

1501: \put(12,-7){\vector(0,1){15}}

1502: \qbezier[200](9,-6,)(14.96,0)(9,6)

1503: \put(4,0){\qbezier[200](15,-6,)(9.04,0)(15,6)}

1504: \put(12,0){

1505: \qbezier[50](0,0)(0,6)(2,6)

1506: \qbezier[50](0,0)(0,-6)(2,-6)

1507: \qbezier[50](2,6)(4,6)(4,0)

1508: \qbezier[50](2,-6)(4,-6)(4,0)

1509: }

1510: \put(16,0){

1511: \qbezier[100](0,0)(0,2)(2,2)

1512: \qbezier[100](0,0)(0,-2)(2,-2)

1513: \qbezier[100](2,2)(4,2)(4,0)

1514: \qbezier[100](2,-2)(4,-2)(4,0)}

1515: \put(25.5,-.3){$Re \lambda$}

1516: \put(13.5,.3){$\Omega_1^*$}

1517: \put(17,.3){$\Omega_1$}

1518: \put(11,-.8){$0$}

1519: \put(16.5,-.8){$1$}

1520: \put(20.5,-.8){$3$}

1521: \put(15,-6){$\gamma^*$}

1522: \put(19.5,-2){$\gamma_c$}

1523: \put(19.7,6){$\Gamma$}

1524: \put(8,6){$\Gamma$}

1525: \put(11,8.5){$Im \lambda$} }

1526: \end{picture}

1527: \caption{Partition of the parameter space in the complex case 

1528: $\lambda = \bar{\mu} \in \C$.}

1529: \label{imconj}

1530: \end{figure}

1531: \begin{figure}

1532: \setlength{\unitlength}{.04\textwidth}

1533: 

1534: \begin{picture}(25,25)(0,0)

1535: \put(0,0){\vector(1,0){22}}

1536: \put(0,0){\vector(0,1){20}}

1537: \put(20,0){\line(0,1){20}}

1538: 

1539: \qbezier[100](0,0)(10,4.5)(20,6)

1540: \qbezier[100](0,0)(9,6)(20,10)

1541: \qbezier[100](0,0)(15,10)(20,20)

1542: \put(0,0){\line(1,1){20}}

1543: \put(0,-.6){$0$}

1544: \put(20,-.6){$1$}

1545: \put(22,.6){$\lambda$}

1546: \put(0.6,19.6){$\mu$}

1547: \put(10,3.2){$\gamma_c$}

1548: \put(16,7){$\Omega_\mu$}

1549: \put(14,8.1){$\gamma$}

1550: \put(17,14.6){$\gamma_s$}

1551: \put(10,9.1){$\Omega_e$}

1552: 

1553: \end{picture}

1554: \caption{Partition of the parameter space in the real case $\lambda,\mu \in \R$.

1555: }

1556: \label{reconj}

1557: \end{figure}

1558: \begin{proposition}

1559: \label{twocycles}

1560: The vector spaces ${\cal A}_\mu$, ${\cal A}_e$, ${\cal A}_1$ of Abelian

1561: integrals $\alpha_0I_0+\alpha_1I_1$ corresponding to the continuous

1562: families of ovals ${\cal O}_\mu$ for $(\lambda,\mu)\in\Omega_\mu$,

1563: ${\cal O}_e$ for $(\lambda,\mu)\in\Omega_e$, and

1564: ${\cal O}_1$ for $\lambda\in\Omega_1\cup\Omega_1^*$, are not Chebyshev

1565: in the respective intervals $\Sigma$.

1566: \end{proposition}

1567: {\bf Proof.} The assertion follows from local analysis.

1568: Consider first ${\cal O}_\mu$. Then $\Sigma=(h_\mu, 0)$ and

1569: $F(h_\mu)=\mu>0$, $F(0)=0$. By the proof of Proposition \ref{cyclicity},

1570: we obtain that $F'(h_\mu)>0$ is equivalent to $a_1<0$. This coefficient

1571: is easily calculated from the normal form (\ref{normal}) and the

1572: result is

1573: $$a_1=\frac{P'''(\mu)}{3P''(\mu)}=

1574: \frac{6\mu^2-4\lambda\mu-4\mu+2\lambda}{3\mu(\lambda-\mu)(1-\mu)}$$

1575: which proves the claim. The proof in the case of ${\cal O}_1$

1576: is similar. The assertion for ${\cal O}_e$ follows from the

1577: definition of $\gamma_s$. $\Box$

1578: 

1579: \vspace{1ex}

1580: Making use of the whole information about the local behavior of $F$ near

1581: the endpoints of $\Sigma$, one can formulate the following hypothesis.

1582: 

1583: \vspace{2ex}

1584: \noindent

1585: {\bf Conjecture 2.} (i).

1586: The vector spaces ${\cal A}_\mu$, ${\cal A}_e$ are Chebyshev with accuracy

1587: one in $\Sigma$ for the cases considered in Proposition \ref{twocycles},

1588: and they are Chebyshev in $\Sigma$ for all the remaining cases.

1589: 

1590: (ii). The vector space ${\cal A}_1$ is Chebyshev with accuracy one in

1591: $\Sigma$ for $\lambda\in\Omega_1$, it is Chebyshev with accuracy two in

1592: $\Sigma$ for $\lambda\in \Omega_1^*$ and it is Chebyshev in $\Sigma$ for

1593: the remaining cases.

1594: 

1595: (iii). Given $\alpha_0$ and $\alpha_1$, the total number of zeros of

1596: the function $\alpha_0I_0+\alpha_1I_1$ relative to all the period annuli

1597: is three and this bound is attained in $\Omega_e\cup\Omega_1^*$.

1598: 

1599: \vspace{2ex}

1600: \noindent

1601: It is plausible that Conjecture 2 is a consequence of Conjecture 1, except in

1602: the case when the polynomial $P(x)$ has complex critical points.

1603: 

1604: The next problem in difficulty should be to study the space of Abelian

1605: integrals of the first kind related to genus three hyperelliptic curves. We

1606: hope that this problem can be settled by combining the methods of the present

1607: paper with deformation arguments as in \cite{Gav4}. Properly understood,

1608: this might lead to the solution of the problem for arbitrary genus $g$.

1609: 

1610: \vspace{2ex}

1611: \noindent

1612: {\bf Acknowledgments.} This research is partially supported by the

1613: Ministry of Science and Education of Bulgaria under Grant no. 1003/2000.

1614: 

1615: \begin{thebibliography}{99}

1616: \bibitem{cam}

1617: %

1618: N. A'Campo,

1619: Le groupe de monodromie du d\'eploiement des singularit\'es isol\'ees de

1620: courbes planes. I. {\it Math. Ann.} {\bf 213} (1975), 1--32.

1621: %

1622: \bibitem{arbarello} E. Arbarello, M. Cornalba, P.A. Griffiths, J. Harris,

1623: {\it Geometry of Algebraic Curves}, vol. 1, Springer, New York, 1985.

1624: %

1625: \bibitem{Arnold} V.I. Arnold, S.M. Gusein-Zade, A.N. Varchenko,

1626: {\it Singularities of Differentiable Maps}, vols. 1 and 2,

1627: Monographs in Mathematics, Birkh\"auser, Boston, 1985 and 1988.

1628: %

1629: \bibitem{Arnold0} V.I. Arnold, Yu.S. Il'yashenko, Ordinary Differential

1630: Equations, in: {\it Dynamical Systems} I, Encyclopaedia of Math. Sci.,

1631: vol. 1, Springer, Berlin, 1988.

1632: %

1633: \bibitem{Arnold1} V.I. Arnold, {\it Geometrical Methods in the Theory of

1634: Ordinary Differential Equations}, Springer, New York, 1988.

1635: %

1636: \bibitem{Arnold2} V.I. Arnold, Sur quelques probl\`emes de la

1637: th\'eorie des syst\`emes dynamiques, {\it Topological Methods in Nonlinear

1638: Analysis}, {\bf  4} (1994), 209--225.

1639: %

1640: \bibitem{Arnold1+} V.I. Arnold, Some unsolved problems in the theory of

1641: differential equations and mathematical physics,

1642: {\it Russian Math. Surveys} {\bf 44} (1989), no. 4, 157--171.

1643: %

1644: \bibitem{ten} V.I. Arnold, Ten problems, in: {\it Theory of

1645: singularities and its applications}, {\it Adv. Soviet Math.} {\bf 1},

1646: pp. 1--8, Amer. Math. Soc., Providence, 1990.

1647: %

1648: \bibitem{bri} E. Brieskorn, Die Monodromie der isolierten Singularit\"{a}ten

1649: von Hyperfl\"{a}chen, {\it  Manuscripta Math.} {\bf 2} (1970), 103--161.

1650: %

1651: \bibitem{dim} A. Dimca, {\it Singularities and Topology of Hypersurfaces},

1652: Springer, Berlin, 1992.

1653: %

1654: \bibitem{gas-li} A. Gasull, W. Li, J. Llibre, Zh. Zhang, Chebyshev property

1655: of complete elliptic integrals and its application to Abelian integrals,

1656: {\it Pacific J. Math.}  {\bf 202} (2002), no. 2, 341--362.

1657: %

1658: \bibitem{gav97} L. Gavrilov, Isochronicity of plane polynomial

1659: Hamiltonian systems, {\it Nonlinearity} {\bf 10} (1997), no. 2, 433--448.

1660: %

1661: \bibitem{gav98a} L. Gavrilov, Petrov modules and zeros of Abelian

1662: integrals, {\it Bull. Sci. Math.} {\bf 122} (1998), no. 8, 571--584.

1663: %

1664: \bibitem{gav99} L. Gavrilov, Abelian integrals related to Morse polynomials

1665: and perturbations of plane Hamiltonian vector fields, {\it Ann. Inst. Fourier}

1666: {\bf 49} (1999), no. 2, 611--652.

1667: %

1668: \bibitem{gav98} L. Gavrilov, Nonoscillation of elliptic integrals related

1669: to cubic polynomials with symmetry of order three, {\it Bull. London Math.

1670: Soc.} {\bf 30} (1998), no. 2,  267--273.

1671: %

1672: \bibitem{Gav4} L. Gavrilov, The infinitesimal 16th Hilbert problem in the

1673: quadratic case, {\it Invent. Math.} {\bf 143} (2001), 449--497.

1674: %

1675: \bibitem{giv} A.B. Givental, Sturm's theorem for hyperelliptic integrals,

1676: {\it Leningrad Math. J.} {\bf 1} (1990), 1157--1163.

1677: %

1678: \bibitem{gh} P.A. Griffiths, J. Harris, {\it Principles of algebraic

1679: geometry}, Pure and Appl. Math., John Wiley and Sons, New York, 1978.

1680: %

1681: \bibitem{hil} D. Hilbert, Mathematische probleme,

1682: {\it Gesammelte Abhandlungen} III, Springer-Verlag, Berlin (1935),

1683: pp. 403--479.

1684: %

1685: \bibitem{nov-yak} D. Novikov and S. Yakovenko, Tangential Hilbert problem

1686: for perturbations of hyperelliptic Hamiltonian systems, {\it Electron. Res.

1687: Announc. Amer. Math. Soc.} {\bf 5} (1999), 55--65 (electronic).

1688: %

1689: % \bibitem{zade} S.M. Husein-Zade, Dynkin diagrams of singularities of

1690: % functions of two variables, {\it Functional Anal. Appl.} {\bf 8}  (1974),

1691: % 10--13.  % 295-300.

1692: %

1693: \bibitem{Petrov86} G.S. Petrov, Number of zeros of complete elliptic integrals,

1694: {\it Funct.Anal.Appl.} {\bf 18} (1984), 72--73; {\bf 20} (1986), 37--40;

1695: {\bf 21 } (1987), 87--88; {\bf 22} (1988), 37--40;

1696: {\bf 23} (1989), 88--89; {\bf 24} (1990), 45--50.

1697: %

1698: \bibitem{pont} L.S. Pontryagin, On dynamic systems close to Hamiltonian

1699: systems, {\it Zh. Eksp. Teor. Fiz.} {\bf 4} (1934), 234--238 (Russian).

1700: %

1701: \bibitem{Roussarie1} R. Roussarie, {\it Bifurcation of Planar Vector Fields

1702: and Hilbert's sixteenth Problem}, Progress in Mathematics, vol. 164,

1703: Birkh{\"a}user, Basel, 1998.

1704: %

1705: \end{thebibliography}

1706: 

1707: \end{document}

1708: %=======================================================================

1709: 

1710: 

1711: 

1712: 

1713: 

1714: 

1715: 

1716: 

1717: 

1718: 

1719: 

1720: 

1721: 

1722: 

1723: 

1724: 

1725: 

1726: 

1727: 

1728: 

1729: 

1730: 

1731: 

1732: 

1733: 

1734: 

1735: 

1736: 

1737: 

1738: 

1739: 

1740: 

1741: 

1742: 

1743: 

1744: 

1745: 

1746: 

1747: 

1748: 

1749: 

1750: 

1751: 

1752: