1: 

2: \documentclass{amsart}

3: 

4: \usepackage{epsfig,amssymb}

5: 

6: \newcommand\ns[1]{\mbox{\small $#1$}}

7: \newcommand\sm[1]{\mbox{\small #1}}

8: 

9: 

10: \newtheorem{theorem}{Theorem}

11: \newtheorem{lem}{Lemma}

12: \newtheorem{cor}{Corollary}

13: 

14: \begin{document}

15: 

16: \title{\bf 

17: On the polynomial moment problem.

18: }

19: \author {F. Pakovich}

20: \address{Department of Mathematics, Weizmann Institute of Science,

21: Rehovot 76100, Israel}

22: \email{pakovich@wisdom.weizmann.ac.il}

23: 

24: \subjclass{Primary 30E99; Secondary 34C99}

25: 

26: \date{\today}

27: 

28: \keywords{moment condition; Poincare center-focus problem; polynomial Abel

29: equation; polynomials; moments; compositions; functional equations}

30: 

31: \maketitle

32: 

33: \def\d{{\rm d}}

34: \def\I{{\rm I}}

35: 

36: \def\C{{\mathbb C}}

37: \def\N{{\mathbb N}}

38: \def\P{{\mathbb P}}

39: \def\D{{\mathbb D}}

40: \def\Z{{\mathbb Z}}

41: \def\phi{{\varphi }}

42: \def\d{{\rm d\,}}

43: \def\deg{{\rm deg\,}}

44: \def\Det{{\rm Det}}\def\dim{{\rm dim\,}}

45: \def\Ker{{\rm Ker\,}}

46: \def\Gal{{\rm Gal\,}}

47: \def\St{{\rm St\,}}

48: \def\Sym{{\rm Sym\,}}

49: \def\Mon{{\rm Mon\,}}

50: 

51: 

52: 

53: \vskip 0.2cm

54: 

55: 

56: \section{Introduction.}

57: 

58: In this paper we treat the following "polynomial moment problem":

59: {\it for a complex polynomial $P(z)$ and distinct $a,b \in \C$

60: such that $P(a)=P(b)$ to describe polynomials $q(z)$

61: such that

62: $$\int^b_a P^i(z)q(z)\d z=0 \eqno(*)$$

63: for all integer non-negative $i.$

64: }

65: 

66: The polynomial moment problem was

67: proposed in the series of papers of M. Briskin, J.-P. Francoise and

68: Y. Yomdin \cite{bfy1}-\cite{bfy5} as an infinitesimal version of

69: the center problem for the polynomial Abel equation in the complex domain

70: in the frame

71: of a programme concerning the classical Poincare center-focus

72: problem for the polynomial vector field on the plane.

73: It was suggested that the following "composition condition" 

74: imposed on $P(z)$ and $Q(z)=\int q(z)\d z$ is necessary

75: and sufficient for the pair $P(z),$ $q(z)$ to satisfy (*):

76: {\it there exist polynomials $\tilde P(z),

77: \tilde Q(z), W(z)$ such that} $$P(z)=\tilde P(W(z)), \ \ \

78: Q(z)=\tilde Q(W(z)), \ \ \ {\it and} \ \ \ W(a)=W(b).\eqno(**)$$

79: 

80: It is easy to see that the composition condition is sufficient:

81: since after the change of variable $z\rightarrow W(z)$ 

82: the way of integration becomes closed, the sufficientness

83: follows from the Cauchy theorem.

84: The necessity of the composition condition in the case when $a,b$ are not

85: critical points of $P(z)$

86: was proved by C. Christopher

87: in \cite{c} (see also the paper of N. Roytvarf \cite{ro}

88: for a similar result) and in

89: some other special cases by M. Briskin, J.-P. Francoise and

90: Y. Yomdin in the papers cited above.

91: Nevertheless, in general the composition conjecture fails to be true

92: as it was shown by the author in \cite{pa}. 

93: 

94: In this paper we give a solution of the polynomial moment problem

95: in the case when $P(z)$ is indecomposable that is when $P(z)$

96: can not be represented as a composition

97: $P(z)=P_1(P_2(z))$ with non-linear polynomials $P_1(z),$ 

98: $P_2(z).$ We show that in this case the

99: composition conjecture is true without any restrictions on points $a,b.$

100: 

101: 

102: \pagebreak

103: 

104: 

105: \noindent{\bf Theorem 1.} {\it 

106: Let $P(z),q(z)$ be complex polynomials and let $a,b$ be distinct complex

107: numbers such that $P(a)=P(b)$ and

108: $$\int^b_a

109: P^i(z)q(z)\d z=0$$

110: for $i\geq 0.$ Suppose that $P(z)$ is indecomposable. Then there exists a

111: polynomial $\tilde Q(z)$ such that

112: $Q(z)=\int q(z)\d z=\tilde Q(P(z))$.}

113: 

114: \vskip 0.2cm

115: 

116: We also examine the following condition which is stronger than (*):

117: $$\int^b_a

118: P^i(z)Q^{j}(z)Q^{\prime}(z)\d z=0$$ for $i\geq 0,$

119: $j\geq 0.$

120: If $\gamma$ is a curve which is the image of the segment $[a,b]$

121: in $\C^2$ under the map $z\rightarrow (P(z),Q(z))$

122: then this condition is equivalent

123: to the condition that $\int_{\gamma}\omega =0$ for all global

124: holomorphic 1-forms $\omega$ in $\C^2$ ("the moment condition").

125: For an oriented simple closed curve $\delta$ of class $C^2$ in $\C^2$ 

126: the moment condition is necessary

127: and sufficient to be a boundary of a bounded analytic variety

128: $\Sigma$ in $\C^2;$ it is a special case of

129: the result of R. Harwey and B. Lawson \cite{hl}. 

130: The case when $\delta$ is an image

131: of $S^1$ under the map $z \rightarrow (f(z), g(z)),$ where $f(z),g(z)$

132: are functions analytic in an annulus containing $S^1$ was investigated

133: earlier by J. Wermer \cite{w1}:

134: in this case the moment condition is equivalent to the condition that

135: there exists a finite Riemann surface $\Sigma$ with border $S^1$

136: such that $f(z),$ $g(z)$ have an 

137: analytic extension to $\Sigma.$

138: 

139: Unlike to condition (*) the more restrictive moment

140: condition imposed on polynomials $P(z),$ $Q(z)$ 

141: turns out to be equivalent to composition condition (**). 

142: We show that actually even weaker 

143: condition is needed.

144: 

145: \vskip 0.2cm

146: 

147: \noindent{\bf Theorem 2.} {\it 

148: Let $P(z),Q(z)$ be complex polynomials and let $a,b$ be distinct complex

149: numbers such that $P(a)=P(b)$ and 

150: $$\int^b_a P^i(z)Q^j(z)Q^{\prime}(z)\d z=0$$

151: for $0\leq i\leq \infty,$ $0\leq j \leq d_a+d_b-2,$

152: where $d_a$ (resp.

153: $d_b$) is the multiplicity of the point $a$ (resp. $b$) with respect to

154: $P(z).$ Then there exist polynomials $\tilde P(z),$

155: $\tilde Q(z),$ $W(z)$ such that $P(z)=\tilde P(W(z)),$

156: $Q(z)=\tilde Q(W(z)),$ and $W(a)=W(b).$}

157: 

158: \vskip 0.2cm

159: 

160: Note that if $a,b$ are not critical points of $P(z)$

161: that is if $d_a=d_b=1$ then conditions of the theorem reduce to 

162: condition (*) and therefore theorem 2

163: includes as a particular case the result of C. Christopher.

164: 

165: 

166: 

167: \section{Proofs.}

168: 

169: 

170: \subsection{Lemmata about branches of $Q(P^{-1}(z))$.}

171: 

172: \vskip 0.2cm

173: 

174: \noindent Let $P(z)$ and $Q(z)$ be rational functions and

175: let $U\subset \C$ be a domain in which there exists a single-valued branch

176: $p^{-1}(z)$ of the algebraic function $P^{-1}(z).$

177: Denote by $Q(P^{-1}(z))$

178: the complete algebraic function obtained by the analytic

179: continuation of the functional element $\{U,Q(p^{-1}(z))\}.$

180: Since the monodromy group $G(P^{-1})$ of the algebraic function

181: $P^{-1}(z)$ is transitive this definition does not depend

182: of the choice of $p^{-1}(z).$ Denote by $d (Q(P^{-1}(z)))$ the degree of

183: the algebraic function $Q(P^{-1}(z))$ that is the number of its

184: branches.

185: 

186: 

187: \begin{lem}

188: 

189: Let $P(z),Q(z)$ be rational functions.

190: Then $$d (Q(P^{-1}(z)))=\deg P(z)/[\C(z):\C(P,Q)].$$

191: 

192: \end{lem}

193: 

194: \noindent{\it Proof.} Since any algebraic relation over $\C$ between

195: $Q(p^{-1}(z))$ and $z$ supplies

196: an algebraic relation between $Q(z)$ and $P(z)$ and vice versa

197: we see that $d (Q(P^{-1}(z)))=[\C(P,Q):\C(P)].$ As

198: $[\C(P,Q):\C(P)]=[\C(z):\C(P)]/[\C(z):\C(P,Q)]$ the lemma follows

199: now from the observation that $[\C(z):\C(P)]=\deg P(z).$

200: 

201: \vskip 0.2cm

202: 

203: Recall that by L\"{u}roth theorem 

204: each field $k$ such that $\C \subset k \subset \C(z)$ and

205: $k\neq \C$ is of the form $k=\C(R),$ $R\in \C(z)\setminus \C.$ Therefore,

206: the field $\C(P,Q)$ is a proper subfield of $\C(z)$ if and only if

207: $P(z)=\tilde P(W(z)),$ $Q(z)=\tilde Q(W(z))$

208: for some rational functions $\tilde P(z),$ $\tilde Q(z),$

209: $W(z)$ with $\deg W(z) > 1;$ in this case we say that $P(z)$ and

210: $Q(z)$ have a common right divisor in the composition algebra.

211: The lemma 1 implies the following explicit criterion which

212: essentially due to Ritt \cite{ri} (cf. also \cite{c}, \cite{ro}).

213: 

214: \vskip 0.2cm

215: 

216: \noindent{\bf Corollary 1.} {\it Let $P(z),Q(z)$ be rational functions.

217: Then $P(z)$ and $Q(z)$ have a common right divisor in

218: the composition algebra if and only if

219: $$Q(p^{-1}(z))=Q(\tilde p^{-1}(z))\eqno(1)$$

220: for two different branches $p^{-1}(z),$ $\tilde p^{-1}(z)$

221: of $P^{-1}(z).$}

222: 

223: \vskip 0.2cm

224: 

225: \noindent{\it Proof.} Indeed, by lemma 1 the field $\C(P,Q)$ 

226: is a proper subfield of $\C(z)$ if and only if $d (Q(P^{-1}(z)))<\deg

227: P(z).$ On the other hand, the last inequality is clearly equivalent to

228: condition (1).

229: 

230: \begin{lem}

231: 

232: Let $P(z),$ $Q(z)$ be rational functions, $\deg P(z)=n.$ 

233: Suppose that there exist

234: $a_i\in \C,$ $1\leq i \leq n,$ not all equal between themselves such that

235: $$\sum_{i=1}^na_iQ(p^{-1}_i(z))=0. \eqno(2)$$

236: If, in addition, the group $G(P^{-1})$ is doubly transitive 

237: then $Q(z)=\tilde Q(P(z))$ for some rational function $\tilde Q(z).$ 

238: 

239: \end{lem}

240: 

241: 

242: \noindent{\it Proof.}

243: Let $G\subset S_n$ be a permutation

244: group and let $\rho_G: G\rightarrow GL(\C^n)$ be

245: the permutation representation of $G$  

246: that is $\rho_G(g),$ $g\in G$ is the linear map which sends 

247: a vector $\vec a=(a_1,a_2, ... , a_n)$ to the vector

248: $\vec{a_g}=(a_{g(1)},a_{g(2)}, ... , a_{g(n)}).$

249: It is well known

250: (see e.g. \cite{wi}, Th. 29.9) that $G$ is doubly transitive if and 

251: only if $\rho_G$ is

252: the sum of the identical representation and an absolutely irreducible

253: representation. Clearly, the one-dimensional $\rho_G$-invariant subspace

254: $E\subset \C^n$

255: corresponding to the identity representation is generated by the vector

256: $(1,1, ... ,1).$ Therefore, since the Hermitian inner product

257: $(\vec{a},\vec{b})=a_1 \bar b_1+ a_2 \bar b_2 + ... + a_n \bar

258: b_n$ is invariant with respect to $\rho_G,$ the group $G$ is doubly

259: transitive if and only if the subspace $E$ and its orthogonal complement

260: $E^{\perp}$ are the only $\rho_G$-invariant subspaces of $\C^n.$

261: 

262: Suppose that (2) holds. In this case  

263: also $$\sum_{i=1}^na_iQ(p^{-1}_{\sigma(i)}(z))=0 \eqno(3)$$ for all

264: $\sigma\in G(P^{-1})$ by the analytic continuation. To prove the lemma 

265: it is enough to show that

266: $Q(p^{-1}_{i}(z))=Q(p^{-1}_{j}(z))$ for all $i,j,$ $1\leq

267: i,j \leq n;$ then by lemma 1

268: $[\C(z):\C(P,Q)]= \deg P(z)=[\C(z):\C(P)]$ and therefore

269: $Q(z)=\tilde Q(P(z))$ for some rational function

270: $\tilde Q(z)$. Assume the converse i.e. that there exists 

271: $z_0\in U$ such that not all $Q(p^{-1}_{i}(z_0)),$

272: $1\leq i \leq n,$ are equal between themselves. 

273: Without loss of generality we can suppose that all $Q(p^{-1}_{i}(z_0)),$

274: $1\leq i \leq n,$ are finite.

275: Consider the subspace

276: $V\subset \C^n$ generated by the vectors $\vec v_{\sigma},$

277: $\sigma\in G(P^{-1}),$ where $\vec v_{\sigma}=(Q(p^{-1}_{\sigma(1)}(z_0)),

278: Q(p^{-1}_{\sigma(2)}(z_0)), ...

279: , Q(p^{-1}_{\sigma(n)}(z_0)).$ Clearly, $V$ is $\rho_{G(P^{-1})}$-invariant

280: and $V\neq E.$

281: Moreover, it follows from (3) that $V$ is

282: contained

283: in the orthogonal complement $A^{\perp}$ of the subspace $A\subset \C^n$

284: generated by the vector $(\bar a_1, \bar a_2, ... , \bar a_n).$ 

285: Since $A\neq E$ we see that $V$ is a

286: proper $\rho_G$-invariant subspace of $\C^n$ distinct from $E$ and

287: $E^{\perp}$ that contradicts the assumption that

288: the group $G(P^{-1})$ is doubly transitive.

289: 

290: \subsection{Lemma about preimages of domains.}

291: For a polynomial $P(z)$ denote by $c(P)$ the set of finite critical

292: values of $P(z)$. 

293: 

294: \begin{lem}

295: 

296: Let $P(z)$ be a polynomial and let $U\subset \C\P^1$ be an unbounded simply 

297: connected domain such that $c(P)\cap U=\emptyset.$ 

298: Then $P^{-1}\{U\}$ is conformally

299: equivalent to the unit disk and $P^{-1}\{\partial U\}$ is connected.

300: 

301: \end{lem}

302: 

303: 

304: \noindent{\it Proof.} Indeed, by the Riemann theorem $U$

305: is conformally equivalent to the unit disk $\D$ whenever $\partial

306: U$

307: contains more than one point. It follows from $c(P)\cap U=\emptyset$

308: that $\partial U$ contains a unique point

309: if and only if $P(z)$ has a unique finite critical value $c$ and

310: $\partial U=c;$ in this case there exist linear functions

311: $\sigma_1,$ $\sigma_2$ such that $\sigma_1(P(\sigma_2(z)))=z^n,$ 

312: $n\in \N$ and the lemma is obvious.

313: Therefore, we can suppose that $U\cong \D.$ Since $c(P)\cap U=\emptyset$

314: the restriction of the map

315: $P(z)\,:\, \C\P^1\rightarrow \C\P^1$ on $

316: P^{-1}\{U\}\setminus P^{-1}\{\infty\}$ is a covering map.

317: As $U \setminus\infty $

318: is conformally equivalent to the punctured

319: unit disc $\D^*$ it follows from covering spaces

320: theory that $ P^{-1}\{U\}\setminus

321: P^{-1}\{\infty\}$ is a disjoint union of domains $\cup U_i$

322: conformally equivalent to $\D^*$ such that all induced maps

323: $f_i \,:\, \D^* \rightarrow \D^*$ are of the form

324: $z\rightarrow z^{l_i},$ $l_i\in \N.$

325: But, as $P^{-1}\{\infty\}=\{\infty\},$ there may be only one such a

326: domain. Therefore, the preimage $P^{-1}\{U\}$ is conformally

327: equivalent to the unit disk. In particular, since

328: $P^{-1}\{\partial U\}=

329: \partial P^{-1}\{U\}$ we see that $P^{-1}\{\partial U\}$ is

330: connected.

331: 

332: \subsection{Proof of theorem 2: the case of a regular value.}

333: 

334: In this section we investigate the case

335: when $t_0=P(a)=P(b)$ is not a

336: critical value of the polynomial $P(z).$

337: For a simple closed curve $M\subset \C$ denote by $D_M^+$ (resp. by

338: $D_M^-$) the domain that is interior (resp. exterior) with respect to $M.$

339: 

340: Let $L\subset \C$ be a simple closed curve

341: such that $t_0\in L$ and $c(P)\subset D_L^+.$ Denote by $\vec{L}$ the

342: same curve considered as an embedded into the complex plane oriented

343: graph. By definition, the graph $\vec{L}$ has one vertex $t_0$ and one

344: counter-clockwise oriented edge

345: $l.$ Let $\vec{\Omega}=P^{-1}\{\vec{L}\}$ be an oriented graph which is

346: the preimage of the graph $\vec{L}$ under the mapping $P(z): \C

347: \rightarrow \C,$ i.e. vertices of $\vec\Omega$ are preimages of $t_0$ and

348: oriented edges of $\vec\Omega$ are preimages of $l.$ 

349: As $L\cap c(P)=\emptyset$ the graph $\vec\Omega$ has $n=\deg P(z)$ vertices

350: and $n$ edges. Furthermore, by lemma 3

351: the graph $\vec\Omega=P^{-1}\{\partial D_L^-\}$ is connected.

352: Therefore, as a point set in $\C$ the graph $\vec\Omega$ is

353: a simple closed curve. Let $l_j,$ $1\leq j \leq n,$ be oriented

354: edges of $\vec\Omega$ and let $a_j$ (resp. $b_j$) be the

355: starting (resp. ending) point of

356: $l_j.$ We will suppose that edges of $\vec\Omega$

357: are numerated by such a way that $a_1=a$ and that under a moving

358: around the domain $P^{-1}\{D_L^-\}$ along its boundary $\vec\Omega$

359: the edge $l_{i},$ $1\leq i \leq n-1,$ is followed by the edge $l_{i+1}$

360: (see fig. 1).

361: 

362: Let $U\subset \C$ be a simply connected domain such that

363: $U\cap c(P)=\emptyset$ and $L\setminus \{t_0\}\subset U.$ By monodromy

364: theorem, in such a domain there exist $n$ single-valued branches of

365: $P^{-1}(t).$ Denote by $p^{-1}_j(t),$ $1 \leq j \leq n,$

366: the single-valued branch of $P^{-1}(t)$ defined in $U$ by the

367: condition $p_j^{-1}\{l\setminus t_0\}=l_j\setminus\{a_j,b_{j}\};$

368: such a numeration of branches of $P^{-1}(t)$

369: means that the analytic continuation of the functional

370: element $\{U,p_j^{-1}(t)\},$ $1 \leq j \leq n-1,$ along $L$

371: is the functional element $\{U,p_{j+1}^{-1}(t)\}.$

372: Let $l_k,$ $k<n,$ be the edge of $\Omega$ such that $b_k=b$ and

373: let $\Gamma=\{l_1,l_2, ... ,l_k\}$ be the oriented path in the graph

374: $\Omega$ joining the vertices $a_1=a$ to $b_k=b.$

375: For $t\in U$ set

376: $\phi(t)=\sum_{j=1}^k

377: Q(p^{-1}_j(t)).$ Clearly, $\phi(t)$ is analytic on $U$ and extends to a 

378: continuous on $U\cup t_0$ function

379: since $\sum_{j=1}^k Q(a_j)-\sum_{j=1}^k Q(b_j)=Q(a)-Q(b)=0.$

380: 

381: \begin{figure}[t]

382: \centering

383: \input{fig1.pstex_t}

384: \caption{}

385: \end{figure}

386: 

387: Consider an analytic on $\C\P^1\setminus L$ function

388: $$I(\lambda)=\oint_L\frac {\phi(t)}{t-\lambda}\d t=

389: \int_{\Gamma}\frac{Q(z)P^{\prime}(z)\d z}{P(z)-\lambda}.$$

390: More precisely, the integral above defines two analytic functions:

391: one of them $I^+(\lambda)$ is analytic in $D_L^+$ and the other

392: one $I^-(\lambda)$ is analytic in $D_L^-.$

393: Furthermore, calculating the Taylor expansion at infinity 

394: we see that condition (*) reduces

395: to the condition that $I^-(\lambda)\equiv 0$ in $D_L^-.$

396: Therefore, by a well-known result about integrals of the

397: Cauchy type (see e.g. \cite{mu}, p. 63) the function $\phi(t)$

398: is the boundary value on $L$ of the analytic in $D_L^+$ function 

399: $I^+(\lambda)$. It follows from the uniqueness theorem for boundary values

400: of analytic functions that the functional element 

401: $\{U,\phi(t)\}$ can be analytically continued along any curve 

402: $M\subset D_L^+.$

403: As $c(P)\subset D_L^+$ this fact implies that 

404: $\{U,\phi(t)\}$ can be analytically continued along any curve 

405: $M\subset \C.$

406: Therefore, by the monodromy theorem, the element $\{U,\phi(t)\}$ 

407: extends to a single-valued analytic function in the whole complex plane.

408: In particular,

409: the analytic continuation of $\{U,\phi(t)\}$ along any closed

410: curve coincides with 

411: $\{U,\phi(t)\}.$ On the other hand, by construction the analytic

412: continuation of $\{U,\phi(t)\}$ along the curve $L$ is

413: $\{U, \phi_L(t)\},$ where $\phi_L(t)=\sum_{j=2}^{k+1}

414: Q(p_{j}^{-1}(t)).$ It follows from $\phi(t)=\phi_L(t)$ 

415: that $Q(p_{1}^{-1}(t))=Q(p_{k+1}^{-1}(t))$

416: and by corollary 1 we conclude that $P(z)$ and $Q(z)$ have a

417: common right divisor in the composition algebra.

418: 

419: As the field $\C(P,Q)$ is a proper subfield of $\C(z)$ and $P(z),$ 

420: $Q(z)$ are polynomials it is easy to prove that $\C(P,Q)=\C(W)$ for some

421: polynomial $W(z),$ $\deg W(z) > 1.$ It means that 

422: $P(z)=\tilde P(W(z)),$ $Q(z)=\tilde Q(W(z))$ for 

423: some polynomials $\tilde P(z),$ $\tilde Q(z)$ such that

424: $\tilde P(z)$ and $\tilde Q(z)$ have no a

425: common right divisor in the composition algebra.

426: Let us show that $W(a)=W(b).$

427: Since $t_0$ is not a critical value of the polynomial $P(z)=\tilde

428: P(W(z))$ the chain rule implies that $t_0$ is not a critical value of 

429: the polynomial $\tilde P(z).$

430: Therefore, if $W(a)\neq W(b)$ then 

431: after the change of variable $z\rightarrow W(z)$

432: in the same way as above we find that $\tilde P(z)=\bar

433: P(U(z))$, $\tilde Q(z)=\bar

434: Q(U(z))$ for some polynomials $\bar P(z), \bar Q(z), U(z)$ with

435: $\deg U(z) >1$ that contradicts the fact that

436: $\tilde P(z),$ $\tilde Q(z)$ have no a

437: common right divisor in the composition algebra.

438: This completes the proof in the case when $z_0$ is not a critical value

439: of $P(z).$

440: 

441: \begin{figure}[t]

442: \centering

443: \input{fig2.pstex_t}

444: \caption{}

445: \end{figure}

446: 

447: \subsection{Proof of theorem 2: the case of a critical value.}

448: Assume now that $t_0=P(a)=P(b)$ is a critical value of

449: $P(z).$ In this case let $L$ be a

450: simple closed curve such that $t_0\in L$ and $c(P)\setminus

451: t_0\subset D_L^+.$ Consider again a graph 

452: $\vec\Omega=P^{-1}\{\vec{L}\}$. 

453: Since $P^{-1}\{D_L^-\}$ is still 

454: conformally equivalent to the unit disk by lemma 3, we see

455: that the graph $\vec\Omega$ topologically is the boundary of a disc

456: although it is not a simple closed curve any more.

457: Let $l_j,$ $1\leq j \leq n,$ be

458: oriented edges of $\vec\Omega$ and let $a_j$ (resp. $b_j$)

459: be the starting (resp. the ending)

460: point of $l_j.$

461: Let us fix again such a numeration of edges of $\vec\Omega$ that $a_1=a$

462: and that under a moving around the domain $P^{-1}\{D_L^-\}$ along its

463: boundary $\vec\Omega$

464: the edge $l_{i},$ $1\leq i \leq n-1,$ is followed by the edge $l_{i+1}.$

465: As above denote by $U$ a domain in $\C$  

466: such that $U\cap c(P)=\emptyset,$ $L\setminus \{t_0\}\subset U$

467: and let $p^{-1}_j(t),$ $1 \leq j \leq n,$ be the single-valued

468: branch of $P^{-1}(t)$ defined in $U$ by the

469: condition $p_j^{-1}\{l\setminus t_0\}=l_j\setminus\{a_j,b_{j}\}.$

470: If $k<n$ is a number such that $b_k=b$ then

471: for the same reason as above the function

472: $\phi(t)=\sum_{j=1}^k Q(p^{-1}_j(t))$ 

473: extends to an analytic in $U \cup D_L^+$ function but this fact does 

474: not imply now that $\phi(t)$ extends to an analytic in the whole

475: complex plane function since $D_L^+$ does not contain

476: $t_0\in c(P).$ Nevertheless,

477: if $V$ is a simply connected domain such that $U\subset V$ and

478: $t_0\notin V$ then $\phi(t)$ still extends to a single-valued analytic

479: function in $V.$ 

480: In particular, 

481: the analytic continuation of 

482: $\{U,\phi(t)\}$ along any simple closed curve $M$ such that 

483: $t_0\subset D_M^-$ coincides with

484: $\{U,\phi(t)\}.$

485: 

486: Let $t_1\in U$ be a point and let $M_1$ (resp. $M_2$) be a 

487: simple closed curve

488: such that $t_1\in M_1,$ $M_1\cap c(P)=\emptyset$

489: and $D^+_{M_1}\cap c(P)=t_0$ (resp. $t_1\in M_2,$ 

490: $M_2\cap c(P)=\emptyset$ and

491: $D^+_{M_2}\cap c(P)=c(P)\setminus t_0$). Define a permutation

492: $\rho_1\in S_n$ (resp. $\rho_2\in S_n$)

493: by the condition that the functional element

494: $\{U,p^{-1}_{\rho_1(j)}(t)\}$

495: (resp. $\{U,p^{-1}_{\rho_2(j)}(t)\}$) is the result

496: of the analytic continuation of the

497: functional element $\{U,p^{-1}_{j}(t)\},$ $1 \leq j \leq n,$ from $t_1$ 

498: along the curve $M_1$ (resp. $M_2$). 

499: Having in mind the identification of the set of elements

500: $\{U,p^{-1}_{j}(t)\},$

501: $1 \leq j \leq n,$ with the set of oriented edges of the graph

502: $\vec\Omega$ the permutations $\rho_1,\rho_2$ 

503: can be described as follows:

504: $\rho_1$ cyclically permutes the edges of $\vec\Omega$ 

505: around the vertices from which they go while cycles

506: $(j_1,j_2,...,j_k)$ of $\rho_2$ correspond to

507: simple cycles $(l_{j_1}, l_{j_2},..., l_{j_k})$

508: of the graph $\vec\Omega$ and 

509: $\rho_1\rho_2=(12...n)$ (see fig. 2).

510: 

511: To unload notation denote temporarily the element 

512: $\{U,Q(p^{-1}_{i}(t))\},$ $1\leq i \leq n,$ by $s_i.$

513: Since $t_0\subset D_{M_2}^-$ we have:

514: $$

515: 0=\sum_{j=1}^{k} s_{\rho_2(j)}-\sum_{j=1}^{k} s_{j}=s_{\rho_2(k)}+\sum_{j=1}^{k-1}\left[

516: s_{\rho_2(j)}-

517: s_{j+1}\right]-s_{1}. \eqno(4)$$

518: Using $\rho_1\rho_2=(12...n)$ we can rewrite (4) as

519: $$s_{\rho^{-1}_1(k+1)}-s_{1}+\sum_{j=1}^{k-1}\left[

520: s_{\rho_2(j)}-s_{\rho_1\rho_2(j)}\right]=0.$$

521: Therefore, by the analytic continuation

522: $$s_{\rho_1^{f-1}(k+1)}-s_{\rho_1^f(1)}+\sum_{j=1}^{k-1}

523: \left[s_{\rho_1^{f}\rho_2(j)}-s_{\rho_1^{f+1}\rho_2(j)}

524: \right]=0 \eqno(5)$$  for $f\geq 0.$

525: Summing equalities (5) from $f=1$ to $f=o(\rho_1),$

526: where $o(\rho_1)$ is the order of the permutation $\rho_1,$

527: changing the order of summing,

528: and observing that 

529: $$\sum_{f=0}^{o(\rho_1)-1}\left[

530: s_{\rho_1^{f}\rho_2(j)}-

531: s_{\rho_1^{f+1}\rho_2(j)}\right]=

532: s_{\rho_2(j)}-s_{\rho_1^{o(\rho_1)}\rho_2(j)}=0$$ 

533: we obtain:

534: $$\sum_{f=0}^{o(\rho_1)-1}s_{\rho_1^f(k+1)}=

535: \sum_{f=0}^{o(\rho_1)-1}s_{\rho_1^f(1)}.\eqno(6)$$

536: If $a,b$ are not critical points

537: of $P(z)$ then

538: $p^{-1}_{\rho_1(1)}(t)=p^{-1}_{1}(t),$ 

539: $p^{-1}_{\rho_1(k+1)}(t)=p^{-1}_{k+1}(t)$

540: and (6) reduces to the equality

541: $Q(p^{-1}_{k+1}(t))=Q(p^{-1}_{1}(t)).$

542: 

543: Suppose now that at least one of the points $a,b$ is a critical 

544: point of $P(z).$

545: Observe that (6) is hold for any polynomial $Q(z)$ such that 

546: $q(z)=Q^{\prime}(z)$ satisfies

547: (*).Therefore, 

548: substituting in (6) $Q^j(z),$ $2\leq j \leq d_a+d_b-1,$ instead of $Q(z)$ we 

549: conclude that $$\sum_{s=0}^{o(\rho_1)-1}Q^j(p^{-1}_{\rho_1^s(k+1)}(t))=

550: \sum_{s=0}^{o(\rho_1)-1}Q^j(p^{-1}_{\rho_1^s(1)}(t))\eqno(7)$$

551: for all $j,$ $1\leq j \leq \d_b+d_b-1.$ Consider

552: a Vandermonde determinant $D=\parallel d_{j,i} \parallel,$

553: where $d_{j,i}=Q^j(p^{-1}_i(t)),$ 

554: $0\leq j \leq d_a+d_b-1$ and $i$ ranges the set of different indices

555: from the cycles of $\rho_1$ containing $1$ and $k+1.$

556: Since (7) implies that $D=0$ we conclude again that

557: $Q(p^{-1}_i(t))=Q(p^{-1}_j(t))$ for some $i\neq j,$ $1\leq i,j \leq n.$

558: Therefore, $P(z)$ and $Q(z)$ have a 

559: common right divisor in the composition algebra 

560: and we can finish the proof by the same argument as in  

561: section 2.3 taking into account that the multiplicity 

562: of a point $c\in \C$ with respect to $P(z)=\tilde P(W(z))$ is greater or

563: equal then the multiplicity of the point $W(c)$ with respect to 

564: $\tilde P(z).$

565: 

566: \pagebreak

567: 

568: \subsection{Proof of theorem 1.}

569: Suppose at first that $n=\deg P(z)$ is a prime number.

570: In this case the degree of the algebraic function 

571: $Q(P^{-1}(t))$ equals either $n$ or $1$

572: since $d(Q(P^{-1}(t)))$ divides $\deg P(z).$

573: If $d(Q(p^{-1}(t)))= n$ then 

574: Puiseux expansions at infinity

575: $$Q(p^{-1}_i(t))=\sum_{k\geq 

576: k_0}a_k\varepsilon^{ik}t^{\frac{k}{n}},\eqno(8)$$

577: $1\leq i \leq n,$ $a_k\in \C,$ $\varepsilon=exp(2\pi i/n),$

578: contain a coefficient $a_k\neq 0$ such that $k$ is not 

579: a multiple of $n.$

580: Substituting (8) in (6)

581: we conclude that $\varepsilon^k$ satisfies an algebraic 

582: polynomial with integer coefficients distinct from

583: the $n$-th cyclotomic polynomial $\Phi_n(z)=1+z+ ...

584: +z^{n-1}$.

585: Since $\varepsilon^k$ is a primitive $n$-th root of unity

586: it is a contradiction. Therefore, $d(Q(p^{-1}(t)))= 1$

587: and $Q(z)=\tilde Q(P(z))$ for some polynomial $\tilde Q(z)$.

588: 

589: Suppose now that $n$ is composite.

590: Since $P(z)$ is indecomposable

591: the group $G(P^{-1})$ is primitive by the Ritt theorem \cite{ri}.

592: By the Schur theorem (see e.g. \cite{wi}, Th. 25.3)

593: a primitive permutation group of composite degree $n$ 

594: which contains an $n$-cycle is doubly transitive. Therefore,

595: by lemma 2 equality (6) implies that

596: $Q(z)=\tilde Q(P(z))$ for some polynomial $\tilde Q(z)$. 

597: 

598: \vskip 0.2cm

599: 

600: 

601: 

602: \noindent{\bf Acknowledgments.} I am grateful to 

603: Y. Yomdin for drawing my attention to the polynomial moment

604: problem and for stimulating discussions.

605: 

606: 

607: 

608: \bibliographystyle{amsplain}

609: \begin{thebibliography}{10}

610: 

611: 

612: 

613: \bibitem {bfy1} M. Briskin, J.-P. Francoise, Y. Yomdin,

614: \textit{Une approche au probleme du centre-foyer de Poincare,}

615: C. R. Acad. Sci., Paris, Ser. I, Math. 326, No.11, 1295-1298 (1998).

616: 

617: 

618: \bibitem {bfy2} M. Briskin, J.-P.

619: Francoise, Y. Yomdin,

620: \textit{Center conditions, compositions of polynomials and moments on

621: algebraic curve}, Ergodic Theory Dyn. Syst. \textbf{19}, no 5,

622: 1201-1220 (1999).

623: 

624: 

625: \bibitem {bfy3} M. Briskin, J.-P. Francoise, Y. Yomdin,

626: \textit{Center condition II: Parametric and model center problems}, Isr.

627: J. Math. \textbf{118}, 61-82 (2000).

628: 

629: 

630: \bibitem {bfy4} M. Briskin, J.-P. Francoise, Y. Yomdin,

631: \textit{Center condition III: Parametric and model center problems}, Isr.

632: J. Math. \textbf{118}, 83-108 (2000).

633: 

634: \bibitem {bfy5} M. Briskin, J.-P. Francoise, Y. Yomdin,

635: \textit{Generalized moments, center-focus conditions and compositions of

636: polynomials,} in "Operator theory, system theory and related topics", 

637: Oper. Theory Adv. Appl.,  \textbf{123}, 161--185 (2001). 

638: 

639: \bibitem {c} C. Christopher, \textit{Abel equations: composition

640: conjectures and

641: the model problem}, Bull. Lond. Math. Soc. \textbf{32}, No.3,

642: 332-338 (2000).

643: 

644: \bibitem {hl} F. Harvey, H. Lawson, \textit{

645: On boundaries of complex analytic varieties. I}, 

646: Ann. Math., II. Ser. \textbf{102}, 223-290 (1975).

647: 

648: \bibitem {pa} F. Pakovich, \textit{A counterexample to the composition

649: conjecture}, Proc. Am. Math. Soc. (to appear).

650: 

651: \bibitem {ri} J. Ritt, \textit{Prime and composite polynomials,} Trans.

652: Amer. Math. Soc.  \textbf{23} (1922), no. 1, 51--66 (1922)

653: 

654: \bibitem {ro} N. Roytvarf, \textit{Generalized moments, composition of

655: polynomials and Bernstein classes}, in "Entire functions in modern

656: analysis. B.Ya. Levin memorial volume", Isr. Math. Conf. Proc.

657: \textbf{15}, 339-355 (2001).

658: 

659: \bibitem {mu} N. Muskhelishvili, \textit{Singular integral

660: equations,} Groningen/Holland: P. Noordhoff, (1953).

661: 

662: \bibitem {w1} J. Wermer, \textit{

663: The hull of a curve in $C^n$,} Ann. Math., II. Ser. 68, 550-561 (1958).

664: 

665: 

666: \bibitem {wi} H. Wielandt, \textit{

667: Finite permutation groups,}

668: New York and London: Academic Press, (1964). 

669: 

670: 

671: 

672: 

673: 

674: 

675: 

676: 

677: 

678: 

679: \end{thebibliography}

680: 

681: 

682: 

683: 

684: 

685: 

686: 

687: 

688: 

689: 

690: 

691: 

692: 

693: 

694: 

695: 

696: 

697: 

698: \end{document}

699: 

700: 

701: 

702: 

703: 

704: 

705: