1: \documentclass{article} 

2: \usepackage{amssymb}

3: \usepackage{amsmath}

4: \begin{document} 

5: \input epsf.sty

6: 

7: 

8: 

9: 

10:  

11: \title{On polynomials orthogonal to all powers of a given

12:  polynomial on a segment}  

13: \author{F. Pakovich}

14: \date{} 

15: 

16: 

17: \maketitle

18: 

19: 

20: %\tableofcontents

21: %\pagebreak

22: 

23: 

24: \def\be{\begin{equation}}

25: \def\ee{\end{equation}}

26: \def\bs{$\square$ \vskip 0.2cm}

27: \def\d{{\rm d}} 

28: \def\D{{\rm D}} 

29: \def\I{{\rm I}} 

30: \def\C{{\mathbb C}} 

31: \def\N{{\mathbb N}} 

32: \def\P{{\mathbb P}}

33: \def\Z{{\mathbb Z}}

34: \def\R{{\mathbb R}} 

35: \def\ord{{\rm ord}}

36: 

37: \def\e{\eqref}

38: \def\phi{{\varphi}}

39: \def\v{{\varepsilon}} 

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

41: \def\Det{{\rm Det}}

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

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

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

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

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

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

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

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

50: \def\GCD{{\rm GCD }}

51: \def\LCM{{\rm LCM }}

52: \def\mod{{\rm mod\ }}

53: 

54: \def\bp{\begin{proposition}}

55: \def\ep{\end{proposition}}

56: \def\bt{\begin{theorem}}

57: \def\et{\end{theorem}}

58: \def\be{\begin{equation}}

59: \def\l{\label}

60: \def\ee{\end{equation}}

61: \def\bl{\begin{lemma}}

62: \def\el{\end{lemma}}

63: \def\bc{\begin{corollary}}

64: \def\ec{\end{corollary}}

65: \def\pr{\noindent{\it Proof. }}

66: \def\note{\noindent{\bf Note. }}

67: \def\bd{\begin{definition}}

68: \def\ed{\end{definition}}

69: 

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

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

72: \newtheorem{definition}{Definition}[section]

73: \newtheorem{corollary}{Corollary}[section]

74: \newtheorem{proposition}{Proposition}[section]

75: 

76: 

77: 

78: \section{Introduction}

79: 

80: In this paper we investigate the following

81: ``polynomial moment problem":

82: {\it for a complex polynomial $P(z)$ and distinct complex numbers $a,b$

83: to describe polynomials $q(z)$

84: such that}

85: \be \l{1}

86: \int^b_a P^i(z)q(z)\d z=0

87: \ee

88: {\it for all integer non-negative $i$.}

89: This problem attracted 

90: attention recently in the

91: series of papers \cite{bby}-\cite{by}, \cite{y1}, where (1) arose

92: as an infinitesimal version of the

93: center problem for the Abel differential equation with polynomial

94: coefficients in the complex domain.

95: 

96: The example of $P(z)=z$ shows that $q(z)=0$ can be the only

97: polynomial solution to

98: \eqref{1} since the Stone-Weierstrass

99: theorem implies that the only continuous complex-valued function

100: which is orthogonal to all powers of $z$ is zero. On the other

101: hand, if $P(a)=P(b),$ then non-trivial polynomial solutions to (1) always

102: exist. Indeed, it is enough to set

103: $q(z)=R^{\prime}(P(z))P^{\prime}(z),$ where $R(z)$ is any

104: complex polynomial.  Then for any $i\geq 0$ $$\int^b_a 

105: P^i(z)q(z)\d z=\int^{P(b)}_{P(a)}y^i R^{\prime}(y)\d y=0.$$

106: 

107: 

108: More generally, the following ``composition condition" imposed

109: on $P(z)$ and $Q(z)=\int q(z)\d z$ is sufficient for polynomials

110: $P(z),q(z)$ to satisfy (1): {\it there exist polynomials $\tilde

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

112: \be \l{2}

113: P(z)=\tilde P(W(z)), \

114: \ Q(z)=\tilde Q(W(z)), \ {\it and} \ W(a)=W(b).

115: \ee

116: The

117: sufficiency of condition (2) follows from $W(a)=W(b)$ after the

118: change of variable $z \rightarrow W(z)$. It was suggested in the

119: papers \cite{bfy1}-\cite{bfy5} (``the composition conjecture") that,

120: under the additional assumption $P(a)=P(b),$

121: condition (1) is actually equivalent to   

122: condition (2). This conjecture is true in several special cases.

123: In particular, when $a,b$ are not critical points of $P(z)$   

124: (\cite{c}), when $P(z)$ is indecomposable (\cite{pa2}),

125: and in some other cases (see \cite{pry}, \cite{ro}, and the papers

126: cited above).

127: 

128: Nevertheless, in general the composition conjecture fails to be 

129: true; a class of counterexamples to the composition conjecture 

130: was constructed in \cite{pa1}. These counterexamples exploit

131: polynomials

132: which admit 

133: ``double decompositions'' of the form

134: $P(z)=A(B(z))=C(D(z)),$ where $A(z),$ $B(z),$ $C(z),$ $D(z)$

135: are non-linear polynomials. If $P(z)$ is such a polynomial and,

136: in addition, the equalities $B(a)=B(b),$ $D(a)=D(b)$ hold, then

137: for any polynomial

138: $Q(z),$ which can be represented as $Q(z)=E(B(z))+F(D(z))$ for some

139: polynomials $E(z),F(z),$ condition (1) is satisfied with

140: $q(z)=Q^{\prime}(z)$ by linearity.

141: On the other hand, it can be shown (see \cite{pa1}) that if $\deg B(z)$ and 

142: $\deg D(z)$ are coprime then condition (2) is not satisfied already for

143: $Q(z)=B(z)+D(z)$. The simplest explicit counterexample

144: to the composition conjecture has

145: the following form:

146: $$ P(z)=T_6(z), \ \ \ \ q(z)=T_2^{\prime}(z)+T_3^{\prime}(z),\ \ \ \

147: a=-\sqrt{3}/2. \ \ \ \ b=\sqrt{3}/2,$$

148: where $T_n(z)$ denotes $n$-th Chebyshev polynomial.

149: %Note that, by the second Ritt theorem,

150: %double decompositions with $\deg A(z)=\deg D(z),$

151: %$\deg B(z)=\deg C(z)$ and $\deg B(z),\deg D(z)$

152: %coprime

153: %are equivalent either to decompositions with $$A(z)=z^nR^m(z),\ \ \ 

154: %B(z)=z^m, \ \ \ C(z)=z^m,\ \ \ D(z)= z^nR(z^m)$$ for a polynomial $R(z)$

155: %and ${\rm GCD}(n,m)=1$ or to decompositions with $$A(z)=T_m(z), \ \ \ B(z)=

156: %T_n(z),\ \ \ 

157: %C(z)= T_n(z), \ \ \ D(z)=T_m(z)$$

158: %for Chebyshev polynomials $T_n(z),$ $T_m(z)$ and ${\rm GCD}(n,m)=1$ (see

159: %\cite{ri},

160: %\cite{sch}).

161: 

162: The counterexamples above

163: suggest to transform the composition conjecture 

164: as follows (\cite{pa4}): {\it non-zero polynomials $P(z),$ $q(z)$ satisfy 

165: condition (1) 

166: if and only if $\int q(z)\d z$ can be represented as a {\it sum} of

167: polynomials $Q_j$ such that

168: \be \l{cc}

169: P(z)=\tilde P_j(W_j(z)), \ \ \

170: Q_j(z)=\tilde Q_j(W_j(z)), \ \ \ {\it and} \ \ \ W_j(a)=W_j(b)

171: \ee

172: for some $\tilde P_j(z), \tilde Q_j(z), W_j(z)\in \C[z]$.}

173: Note that we do not make any additional assumptions on the 

174: values of $P(z)$ at the points $a,b$ any more. 

175: In particular, the conjecture

176: implies that non-zero polynomials orthogonal to

177: all powers of a given polynomial $P(z)$ on $[a,b]$ exist if and only

178: if $P(a)=P(b).$ 

179: For $P(z)=T_n(z)$

180: conjecture \eqref{cc} was verified in \cite{pa3}.

181: 

182: This paper is organized as follows. We start from the description  of

183: necessary and sufficient

184: conditions for polynomials $P(z),q(z)$ to satisfy \eqref{1} in terms

185: of linear relations between

186: branches of the algebraic function $Q(P^{-1}(z)),$ where $P^{-1}(z)$ denotes

187: the algebraic function which is inverse to $P(z).$ More precisely,

188: let $P^{-1}_{i}(z),$ $1\leq i \leq n,$ be single-valued branches of $P^{-1}(z)$

189: in a simply-connected domain $U\subset \C$

190: containing no critical values of $P(z).$

191: We show that there exists a system of

192: equations 

193: \be \l{su}

194: \sum_{i=1}^nf_{s,i}Q(P^{-1}_{i}(z))=0, \ \ \ \ \ \ \ \ 1\leq s \leq \deg P(z),

195: \ee

196: where $f_{s,i}\in\{0,-1,1\},$

197: such that \eqref{1} holds if and only if \eqref{su} holds.

198: This system depends on $P(z),a,b$ and can be calculated explicitly

199: via a special graph $\lambda_P,$ called the ``cactus'' of $P(z),$

200: which contains

201: all the information about the monodromy of $P(z).$

202: 

203: 

204: The criterion

205: \eqref{su}

206: has a number of applications.

207: For example, it allows us

208: to reduce an infinite set of

209: equations \eqref{1} to a finite set of equations $v_k=0,$ ${0\leq k \leq M,}$

210: where $v_k$ are initial coefficients of

211: the Puiseux expansions of the combinations of branches in \eqref{su} and

212: $M$ depends only on degrees of $P(z)$ and $Q(z)$.

213: Furthermore, using the equivalence of \eqref{1} and \eqref{su},

214: we provide a variety of different conditions on a

215: collection $P(z),a,b$ 

216: under which conditions \eqref{1} and 

217: \eqref{2} are equivalent; 

218: such conditions are of interest because

219: of applications to the 

220: Abel equation (see \cite{bby}, \cite{bry}, \cite{by}). 

221: Essentially the finding of such conditions reduces to the finding conditions

222: under which system \eqref{su} implies that

223: \be \l{eb}

224: Q(P^{-1}_{i_1}(z))=Q(P^{-1}_{i_2}(z))

225: \ee 

226: for some $i_1\neq i_2.$

227: In its turn these conditions can be naturally given in terms

228: of combinatorics of the graph $\lambda_P.$

229: 

230: 

231: While an explicit form of system \eqref{su} depends on

232: the collection $P(z),a,b,$

233: there exists a necessary condition for \e{1} to be satisfied the form

234: of which is invariant with respect to $P(z),a,b.$

235: Namely, it is known (\cite{pa2}, \cite{pry}) that \e{1}

236: implies the equality  

237: \be \l{e1}

238: \frac{1}{d_{a}}\sum_{s=1}^{d_{a}}

239: Q(P_{a_s}^{-1}(z))=\frac{1}{d_{b}}\sum_{s=1}^{d_{b}}

240: Q(P_{b_s}^{-1}(z)),

241: \ee if $P(a)=P(b)$, or

242: the system 

243: \be \l{e2}

244: \frac{1}{d_{a}}\sum_{s=1}^{d_{a}}

245: Q(P_{a_s}^{-1}(z))=0, \ \ \ \ \ \ \ \ \ \ \frac{1}{d_{b}}\sum_{s=1}^{d_{b}}

246: Q(P_{b_s}^{-1}(z))=0,

247: \ee

248: if $P(a)\neq P(b),$ where

249: $P_{a_1}^{-1}(z),$

250: $P_{a_2}^{-1}(z), ... , P_{a_{f_a}}^{-1}(z)$

251: (resp. $P_{b_1}^{-1}(z),$ $P_{b_2}^{-1}(z),$ ... , \linebreak

252: $P_{b_{f_b}}^{-1}(z)$) denote

253: the branches

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

255: which map points close

256: to $P(a)$ (resp. $P(b)$)

257: to points close to $a$ (resp. $b$).

258: 

259: 

260: In the fourth section of this paper we investigate equation \eqref{e1} and 

261: system \eqref{e2}. In particular, we establish a specific geometric

262: property of the monodromy groups of polynomials 

263: from which we deduce that if \e{e1} or \e{e2} is satisfied

264: for $P(z),Q(z)\in \C[z],$ $\deg P(z)=n, \deg Q(z)=m,$

265: then for coefficients of the Puiseux expansions near infinity

266: \be \l{ps2}

267: Q(P^{-1}_j(z))=\sum_{k=-m}^{\infty}

268: u_k\varepsilon_n^{jk}z^{-\frac{k}{n}}

269: \ee

270: the equality $u_k=0$ holds whenever $\GCD(k,n)=1.$

271: This fact agrees with conjecture \eqref{cc} and, in particular,

272: implies that for $P(z),q(z)$ satisfying \eqref{1} the numbers

273: $n$ and $m$ can not be coprime.

274: As an application of our analysis of \eqref{e1},\eqref{e2} 

275: we show in the fifth section that conditions \eqref{1} and \eqref{2}

276: are equivalent if at least one from points $a,b$ is not a critical

277: point of $P(z)$ or if $\deg P(z)=p^r$ for a prime number $p.$

278: 

279: Finally,

280: on the base of obtained results,

281: in the sixth section we show that for any

282: collection $P(z),a,b$ with $\deg P(z)<10$

283: conditions \eqref{1}

284: and \eqref{2} are equivalent except the case when 

285: $P(z),a,b$ is linearly equivalent to the collection

286: $T_6(z),- \sqrt{3}/2,\sqrt{3}/2.$ 

287: Since for $P(z)=T_n(z)$ all solutions to \eqref{1} were obtained

288: in \cite{pa3}, this provides

289: the complete solution of the polynomial moment problem for

290: $P(z),a,b$ with $\deg P(z)<10.$

291: 

292: 

293: \section{Criterion for a polynomial to be orthogonal to all

294: powers of a given polynomial} 

295: 

296: \subsection{Cauchy type integrals of algebraic functions} 

297: A quite general approach to the polynomial moment problem

298: was proposed in the paper \cite{pry} concerning Cauchy type integrals of

299: algebraic functions

300: \be \l{ci}

301: I(t)=I(\gamma,g,t)=\frac{1} {2\pi i}\int_{\gamma} \frac{g(z)dz}

302: {z-t}\,.

303: \ee

304: In this subsection we briefly recall it (see \cite{pry} for

305: details) and outline in this context the approach of this paper.

306: 

307: 

308: First of all notice that condition \eqref{1} is equivalent

309: to the condition

310: \be \l{1+}

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

312: \ee

313: for $i\geq 0,$ where $Q(z)=\int q(z) \d z$ is normalized

314: by the condition ${Q(a)=Q(b)=0}$

315: ($Q(a)$ always equals $Q(b)$

316: by \e{1} taken for $i=0$). 

317: Furthermore, vice versa, condition \e{1+}

318: with $Q(a)=Q(b)=0$ implies that \e{1} holds with $q(z)=Q^{\prime}(z).$

319: (Actually, it was the condition

320: \eqref{1+} that appeared initially in the papers

321: on differential equations cited above).

322: 

323: Indeed, condition \e{1+} is equivalent to the condition that

324: the function 

325: $$

326: H(t)= \int_a^b \frac{Q(z)P^{\prime}(z)dz}{P(z)-t} \ \ \ \ \

327: $$ vanishes identically near infinity, since near infinity 

328: $$H(t)=-\sum_{i=0}^{\infty}m_it^{-(i+1)}, \ \ \ \  {\rm where}

329: \ \ \ \  m_i=

330: \int^b_a P^i(z)Q(z)P^{\prime}(z)\d z.$$

331: On the other hand,

332: we have:  

333: \begin{multline} \l{mul}

334: \frac{dH(t)}{dt}=\int_a^b 

335: \frac{Q(z)P^{\prime}(z)dz}{(P(z)-t)^2}=

336: -\int_a^b Q(z)d\left(\frac{1}{P(z)-t}\right)=\\

337: =\frac{Q(a)}{P(a)-t}-\frac{Q(b)}{P(b)-t}+\tilde H(t),

338: \end{multline}

339: where

340: $$\tilde H(t)= \int_a^b \frac{q(z)dz}{P(z)-t}.$$ Since 

341: near infinity 

342: $$\tilde H(t)=-\sum_{i=0}^{\infty}\tilde m_it^{-(i+1)}, \ \ \ \  {\rm where}

343: \ \ \ \  \tilde m_i=

344: \int^b_a P^i(z)q(z)\d z,$$

345: it follows from \e{mul} that conditions \e{1} and \e{1+} are equivalent

346: whenever \linebreak

347: ${Q(a)=Q(b)=0.}$

348: 

349: Furthermore, performing the change of variable

350: $z\rightarrow P(z),$

351: %the vanishing of $H(t)$

352: %near infinity becomes equivalent to

353: %the vanishing near infinity of Cauchy type integral

354: %\e{ci}, where $\gamma=P([a,b])$ and

355: %$g(z)$ is an algebraic function obtained by the analytic continuation

356: %of a germ of the algebraic function

357: %$g(z)=Q(P^{-1}(z))$ along $\gamma.$

358: we see that $H(t)$ coincides with integral \e{ci} where $\gamma=P([a,b])$ and

359: $g(z)$ is an algebraic function obtained by the analytic continuation

360: of a germ of the algebraic function

361: $g(z)=Q(P^{-1}(z))$ along $\gamma.$

362: Integral representation \eqref{ci}

363: defines a collection of univalent regular functions

364: $I_i(t);$ each $I_i(t)$ is defined in a domain $D_i$ of the complement

365: of $\gamma$ in $\C\P^1$. Denote by $I_{\infty}(t)$

366: the function defined in

367: the domain $D_{\infty}$ containing infinity. Then the vanishing of $H(t)$

368: near infinity becomes equivalent to the equality $I_{\infty}(t)\equiv 0.$

369: 

370: 

371: More generally,

372: consider integral \e{ci}, where $\gamma$ is a curve in the complex

373: plane $\C$ and $g(z)$ is any 

374: ``piecewise-algebraic'' function on $\gamma$. More precisely, we assume that 

375: after removing from $\gamma$ a finite

376: set of points $\Sigma_{\gamma},$

377: the set $\gamma\setminus \Sigma_{\gamma}$ is an union

378: of topological segments $\cup \gamma_s$ such that 

379: for each $\gamma_s$ there exists a  

380: domain $V_s\supset \gamma_s$ and an analytic in $V_s$ 

381: algebraic function $g_s(z)$ such that $g(z)$

382: on $\gamma_s$ coincides with

383: $g_s(z).$ Furthermore, we 

384: assume that at the points of $\Sigma_{\gamma}$, complete analytic

385: continuations $\hat g_s(z)$ of $g_s(z)$ can ramify

386: but do not have 

387: poles. 

388: %Integral representation \eqref{ci}

389: %defines a collection of univalent regular functions

390: %$I_i(t);$ each $I_i(t)$ is defined in a domain $D_i$ of the complement

391: %of $\gamma$ in $\C\P^1$. Denote by $I_{\infty}(t)$

392: %the function defined in

393: %the domain $D_{\infty}$ containing infinity.

394: Below we sketch conditions for $I_{\infty}(t)$

395: to be a rational function; if these conditions are satisfied, then in

396: order to verify the equality 

397: $I_{\infty}(t)\equiv 0$

398: it is enough to examine possible poles.

399: 

400: 

401: 

402: Denote by $\Sigma_g$ the set 

403: of all singularities of $\hat g_s(z)$

404: in $\C\P^1.$

405: Show that any element $(I_i(t),U_i)$ can be analytically continued

406: along any curve

407: $S=S_{t_1,t_2}$ connecting points $t_1,t_2 \in \C\P^1$

408: and avoiding points from the sets $\Sigma_g$

409: and $\Sigma_{\gamma}.$ 

410: First of all notice that if $t_2\in \partial U_i$ then 

411: an analytical extension of $(I_i(t),U_i)$

412: to a domain containing $t_2$ 

413: is given simply by the integral

414: $I(\tilde\gamma,g,t),$ where $\tilde \gamma$ is a small

415: deformation of the $\gamma$

416: such that $t_2\in \tilde U_i.$

417: Furthermore, if $S=S_{t_1,t_2}$ is a simple curve connecting points

418: $t_1\in U_i,$ $t_2\in U_j,$ where

419: $U_i,$ $U_j$ are domains with a common segment of the boundary

420: $\gamma_{s}$ and $(g_s,V_s)$ is

421: the corresponding algebraic function, 

422: then the well-known 

423: boundary property of Cauchy type integrals (see e.g. \cite{mus}) implies that

424: $$(I_i(t),U_i\cap V_s)=(I_j(t),U_j\cap V_s)+(g_s,V_s).$$

425: Therefore, the analytic continuation of 

426: $(I_i(t),U_i)$ along $S$ can be defined via

427: the analytic continuation of the right side of this formula.

428: 

429: Finally,

430: for arbitrary domains $U_i,$ $U_j$ and

431: a curve $S=S_{t_1,t_2}$ connecting points $t_1\in U_i$ and $t_2\in U_j,$ 

432: the analytic continuation $(I_i(t),U_i)_S$ of 

433: the element $(I_i(t),U_i)$ along $S$ can be defined inductively as follows.

434: Let $S\cap \gamma =\{c_1,c_2, ... ,c_r\},$ 

435: $c_s\in V_s,$ $1\leq s \leq r,$

436: and 

437: let $(g_1,V_1), (g_2,V_2), ... , (g_r,V_r)$ be 

438: the corresponding algebraic functions.

439: Define a germ $g_{\gamma,S}$ of an algebraic function near the point 

440: $t_2$ by the formula:

441: $$g_{\gamma,S}=\sum_{i=1}^r (g_i,V_i)_{S_{c_i}},  $$

442: where $(g_i,V_i)_{S_{c_i}}$, $1\leq i \leq r,$ denotes the analytic 

443: continuation

444: of the element $(g_i,V_i)$ (taken with the sign corresponding

445: to the orientation of the crossing of $S$ and $\gamma$)

446: along a part of $S$ from $c_i$ to $t_2.$ 

447: Then by induction we have:

448: \be \l{skl}

449: (I_i(t),U_i)_{S}=(I_j(t),U_j)+g_{\gamma,S}.

450: \ee 

451: In particular, a complete analytic continuation $\hat I_i(t)$ of the element

452: $(I_i(t),U_i)$ is

453: a multi-valued analytic function with a finite set of

454: singularities $\Sigma_{\hat I_i} \subset \Sigma_g \cup \Sigma_{\gamma}.$

455: 

456: From formula \e{skl} one deduces the following criterion

457: (\cite{pry}):

458: {\it $\hat I_i(t)$ is a rational function 

459: if and only if the equality

460: \be \l{us}

461: g_{\gamma,S}=0

462: \ee

463: holds for any curve

464: $S=S_{t_1,t_2}$ as above with $t_1=t_2\in U_i.$}

465: Indeed, the necessity of 

466: \e{us} is obvious. To establish the sufficiency

467: observe that \e{us} implies, in particular, that $\hat I_i(t)$ has

468: no ramification in its singularities. 

469: Since

470: a Cauchy type integral near the end of the line of integration $u$

471: (and, similarly, near points of $\Sigma_{\gamma}$) is

472: a sum of a part which has a logarithmic branching at $u$

473: with a part that bounded at

474: $u$ (see e.g. \cite{mus}) and

475: $\hat g_s(z)$ do not have 

476: poles at $\Sigma_{\gamma},$

477: this fact implies that the singularities of $\hat I_i(t)$

478: located on $\gamma$ are removable.

479: On the other hand, if a singularity

480: $t_0$ of $\hat I_i(t)$ is contained in $\C\P^1\setminus \gamma,$ then

481: formula \e{skl} implies that

482: $t_0\in \Sigma_g.$ Since $\hat I_i(t)$ has no ramification

483: at $t_0$ it follows that in this case $t_0$ 

484: is a pole the worst.

485: 

486: Although the method above in principle is constructive

487: its practical application is rather

488: difficult since the calculation of sums $g_{\gamma,S}$ 

489: is complicated.

490: In this paper we propose a modification

491: of the method above

492: designed

493: specially for the polynomial moment problem.

494: This modification

495: permits to avoid any analytic continuations  

496: and allows us to obtain a necessary and sufficient conditions

497: for equality \e{1} to be satisfied in 

498: a closed and convenient form.

499: The idea is

500: to choose a very special way of integration $\Gamma$ connecting points

501: $a,b$ (we can use any of them since integrals \e{1}

502: do not depend on $\Gamma$). It turns out that 

503: $\Gamma$ can by chosen  

504: so that $\C\P^1\setminus P(\Gamma)$ consists of

505: a {\it unique} domain. Then condition $I_{\infty}(t)\equiv 0$

506: simply reduces to the condition that

507: the corresponding algebraic functions $g_s(z)$ vanish on $P(\Gamma).$

508: Furthermore, we choose $\Gamma$ as a subset of a special tree $\lambda_P,$

509: called the cactus of $P(z),$ which contains

510: all the information about the monodromy $P(z).$ 

511: This allows us to relate properties of the collection

512: $P(z),a,b$ which are connected to the polynomial moment 

513: problem to combinatorial properties of the pair consisting of

514: the tree $\lambda_P$ and the   

515: path 

516: connecting points $a,b$ on $\lambda_P.$ 

517: 

518: 

519: \subsection{Cacti} \l{cacti}

520: To visualize the monodromy group of a polynomial $P(z)$

521: it is convenient to consider a graphical object $\lambda_P$ called the

522: {\it cactus} of $P(z)$ (see e. g. \cite{cac1}, \cite{cac2}). 

523: 

524: Let $c_1,c_2, ... ,c_k$ be all

525: finite critical values of $P(z)$ and let $c$ be a not critical value. Draw 

526: a star $S$ joining $c$ with $c_1,c_2, ... ,c_k$ by non intersecting arcs

527: $\gamma_1, \gamma_2, ... ,\gamma_k.$ We will suppose that $c_1,c_2, ...

528: ,c_k$ are numerated in such a way that in a counterclockwise rotation

529: around $c$ the arc $\gamma_s$, $1 \leq s \leq k-1,$ is followed by the arc

530: $\gamma_{s+1}.$ By definition, the cactus $\lambda_P$ is the preimage of $S$

531: under the map $P(z)\,:\, \C \rightarrow \C.$ More precisely, we consider

532: $\lambda_P$ as a $k+1$-colored graph embedded into the Riemann sphere:

533: vertices of $\lambda_P$ colored by the $s$-th color, where $1\leq s \leq 

534: k,$ are preimages of the point $c_s,$ vertices colored by the $k+1$-th 

535: color (to 

536: be definite we will suppose that it is the white color) are 

537: preimages

538: of the point $c,$ and edges of $\lambda_P$ are preimages of the arcs

539: $\gamma_s,$ $1 \leq s \leq k.$ It is not difficult to show that the graph

540: $\lambda_P$ is connected and has no cycles. Therefore,

541: $\lambda_P$ is a plane tree.  

542: 

543: The valency of a non-white vertex $z$ of $\lambda_P$

544: coincides with the multiplicity of $z$ with respect to $P(z)$ while

545: all white vertices of $\lambda_P$ are of the same valency

546: $n=\deg P(z).$ The set of all edges of $\lambda_P$ adjacent to a

547: white vertex $w$ is called by a {\it star} of $\lambda_P$ centered at $w.$

548: Clearly, $\lambda_P$ has $nk$ edges and $n$ stars. The set of stars

549: of $\lambda_P$ is naturally identified with the set of branches of $P^{-1}(z)$

550: as follows.

551: Let $U$ be a

552: simply connected domain containing no critical values of $P(z)$ such 

553: that $S\setminus\{c_1,c_2,

554: ... ,c_k\}\subset U.$ By the monodromy theorem in $U$ there exist

555: $n$ single valued branches

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

557: Any such a branch $P_i^{-1}(z),$ $1\leq i \leq n,$

558: maps $S\setminus\{c_1,c_2, ... ,c_k\}$ into a star of

559: $\lambda_P$ and we will label the corresponding star by the symbol $S_i$

560: (see Fig. 1).

561: 

562: 

563: \begin{figure}

564: \medskip

565: \epsfxsize=12truecm

566: \centerline{\epsffile{1.eps}}

567: \smallskip

568: \centerline{Figure 1}

569: \medskip

570: \end{figure}

571: 

572: The cactus $\lambda_P$ permits to reconstruct the monodromy group

573: $G_P$ of $P(z).$ Indeed, $G_P$ is generated by  

574: the permutations $g_{s}\in S_n,$ $1\leq s \leq k,$ where 

575: $g_{s}$ is defined by the condition

576: that the analytic continuation of

577: the element $(P_i^{-1}(z),U),$ $1\leq i \leq n,$ along a counterclockwise

578: oriented loop $l_s$ around $c_s$

579: is the element $(P_{g_s(i)}^{-1}(z),U).$ Having in mind the identification of 

580: the set of stars of $\lambda_P$ with the set of branches of $P^{-1}(z),$

581: the permutation $g_s,$ $1\leq s \leq k,$ can be 

582: identified with the permutation $\hat g_s,$

583: $1\leq s \leq k,$

584: acting

585: on the set of starts of $\lambda_P$ in the following way: $\hat g_s$ sends the

586: star $S_i,$ $1\leq i \leq n,$ to the ``next'' one in a counterclockwise

587: direction around its vertex of color $s.$ For example,

588: for the cactus shown on Fig. 1 we have: $g_1=(1)(2)(37)(4)(5)(6)(8),$

589: $g_2=(1)(2)(3)(47)(56)(8),$ $g_3=(1238)(4)(57)(6).$

590: 

591: Note that since $P(z)$ is a polynomial, the permutation

592: $g_{\infty}=g_1g_2 ... g_k$ is a cycle of

593: length $n.$ Usually, we will choose the numeration of $S_i,$ $1\leq i \leq

594: n,$ in such a way that this cycle coincides with the cycle $(12...n).$

595: 

596: 

597: \subsection{Criterion}\l{criter} In this subsection

598: we give explicit necessary and sufficient conditions for $P(z),$ $q(z)\in

599: \C[z]$ and $a,b\in \C,$ $a\neq b,$ to satisfy \e{1}, \e{1+}.

600: For this propose we choose the way of integration $\Gamma_{a,b}$ connecting

601: $a,b$ so

602: that $\Gamma_{a,b}$ would be a subset of $\lambda_P.$ 

603: 

604: 

605: More precisely, for any $P(z)\in \C[z]$ and $a,b\in \C$ let us define

606: an {\it extended} cactus

607: $\tilde \lambda_P=\tilde \lambda_P(c_1,c_2,\, ... \,,c_{\tilde k})$

608: as follows. Let $c_1,c_2, ... ,c_{\tilde k}$ be all finite critical values

609: of $P(z)$ complemented by $P(a)$ or $P(b)$ (or by both of them) if $P(a)$

610: or $P(b)$ is not a critical value of $P(z).$ Consider an extended star

611: $\tilde S$ connecting $c$ with $c_1,c_2, ... ,c_{\tilde k}$ and set

612: $\tilde \lambda_P=P^{-1}\{\tilde S\}$ (we suppose that $c$ is chosen

613: distinct from $P(a),P(b)$). Clearly, $\tilde \lambda_P$ considered as a

614: $\tilde k+1$ colored graph is still connected and has no cycles.

615: Furthermore,

616: by construction the points $a,b$ are vertices of $\tilde \lambda_P.$ Since

617: $\tilde \lambda_P$ is connected there exists an oriented path $\Gamma_{a,b}

618: \subset \tilde \lambda_P$

619: with the starting point $a$ and the ending point $b.$ Moreover, since

620: $\tilde \lambda_P$ has no cycles there exists exactly one such a path. We

621: choose $\Gamma_{a,b}$ as a new way of integration.

622: 

623: Let $U$ be a domain as above and let $Q(z)=\int q(z) \d z$ be normalized

624: by the condition ${Q(a)=Q(b)=0}.$ 

625: For each

626: $s,$ $1\leq s \leq \tilde k,$ define a linear combination

627: $\phi_s(z)$ of branches $Q(P^{-1}_{i}(z)),$ $1\leq i \leq n,$ in $U$

628: as follows. Set 

629: \be \l{piz}

630: \phi_s(z)=\sum_{i=1}^nf_{s,i}Q(P^{-1}_{i}(z)), 

631: \ee

632: where $f_{s,i}\neq 0$ if and only if the path

633: $\Gamma_{a,b}$ passes through a vertex $v$ of the star

634: $S_i$

635: colored by the

636: $s$-th color

637: (we do not take into account the stars $S_i$ for which $\Gamma_{a,b}\cap S_i$

638: contains only the point $v$). Furthermore, if under a moving along $\Gamma_{a,b}$ the vertex $v$

639: is followed by the center of $S_i$ then $f_{s,i}=-1$ otherwise

640: $f_{s,i}=1$.  As an example consider the cactus shown on Fig. 1. Then 

641: for the path

642: $\Gamma_{a,b}$ pictured by the fat 

643: line we 

644: have:  $$\phi_1(z)=

645: -Q(P^{-1}_{2}(z))+Q(P^{-1}_{3}(z))- Q(P^{-1}_{7}(z)),$$ $$\phi_2(z)=

646: Q(P^{-1}_{7}(z))-Q(P^{-1}_{4}(z)),$$ $$\phi_3(z)=

647: Q(P^{-1}_{2}(z))-Q(P^{-1}_{3}(z))+ Q(P^{-1}_{4}(z)).$$

648: \vskip 0.1cm

649: \bt \l{t1}

650: Let $P(z),q(z)\in \C[z]$, $a,b\in \C,$ $a\neq b,$

651: and let $\tilde \lambda_P(c_1,c_2,\, ... \,,c_{\tilde k})$ be an extended

652: cactus corresponding to the collection $P(z),a,b.$

653: Then 

654: \eqref{1} holds if and only if 

655: the equality $\phi_s(z)\equiv 0$ holds

656: in $U$ for any $s,$ $1\leq s \leq \tilde k.$

657: \et 

658: 

659: \pr Indeed,

660: condition \eqref{1+} is equivalent to the condition that the function

661: $$

662: H(t)= \int_{\Gamma_{a,b}} \frac{Q(z)P^{\prime}(z)dz}{P(z)-t} \ \ \ \ \

663: $$ vanishes identically near infinity.

664: On the other hand, using

665: the change of variable $z\rightarrow P(z),$ we can express the function

666: $H(t)$ as a sum of Cauchy type integrals of algebraic functions as follows:

667: \be \l{f}

668: H(t)=\sum_{s=1}^{\tilde k}\int_{\gamma_s}\frac{\phi_s(z)}{z-t}\, \d

669: z.

670: \ee

671: Since this formula implies that $H(t)$ is analytic in a domain

672: $V=\C\P^1\setminus S$ we see that the vanishing of $H(t)$

673: near infinity is equivalent to the

674: condition that $H(t)\equiv 0$ in $V.$

675: 

676: Let $z_0$ be an interior point of $\gamma_s,$ $1\leq s \leq \tilde k.$ 

677: Then by the well-known boundary property of

678: Cauchy type integrals (see e.g. \cite{mus}) we have:

679: 

680: $$

681: \lim_{t \to z_0}\!\!\!\!\,^+H(t)-\lim_{t \to z_0}

682: \!\!\!\!\,^-H(t)=\phi_s(t_0),

683: $$

684: where the limits are taken respectively for $t$ 

685: tending to $z_0$ from the ``left'' and 

686: from the ``right'' parts of $V$ with respect to $\gamma_s.$ 

687: If $H(t)\equiv 0$ in $V,$ then 

688: $$

689: \lim_{t \to z_0}\!\!\!\!\,^+H(t)=\lim_{t \to z_0}

690: \!\!\!\!\,^-H(t)=0,

691: $$

692: and, therefore, $\phi_s(z_0)= 0.$

693: Since this equality holds for any

694: interior point $z_0$ of any arc $\gamma_s,$ $1\leq s \leq \tilde k,$ 

695: we conclude that  

696: $\phi_s(z)\equiv 0,$ $1\leq s \leq \tilde k,$ in $U.$ On the other hand,

697: if $\phi_s(z)\equiv 0,$ $1\leq s \leq \tilde k,$ in $U,$ then

698: it follows directly from formula \e{f} that

699: $H(t)\equiv 0$ in $V.$

700: 

701: 

702: \vskip 0.2cm

703: Note that some of equations $\phi_s(z)\equiv 0,$ $1\leq s \leq \tilde k,$

704: could be trivial. 

705: This happens exactly for the values $s$ such that

706: the path $\Gamma_{a,b}$ does not pass through vertices

707: colored by the $s$-th color.

708: Note also that

709: equations \e{piz}

710: are linearly dependent. Indeed, 

711: for each $i$ such that there exists an index $s,$ 

712: $1\leq s

713: \leq \tilde k,$ with $f_{s,i}\neq 0$ there exist exactly two such

714: indices $s_1,s_2$ and $c_{s_1,i}=-c_{s_2,i}.$ Therefore, the equality

715: $$\sum_{s=1}^{\tilde k}\phi_s(t)=0$$

716: holds in $U$.

717: 

718: 

719: \subsection{Checking the criterion}

720: Let $P(z),Q(z)\in \C[z],$ $\deg P(z)=n,$ $\deg Q(z)=m.$

721: Let $U$ be a simply connected domain containing no critical values of $P(z)$

722: and let $P_i^{-1}(z),$ $1\leq i \leq n,$ be branches of $P^{-1}(z)$

723: in $U.$

724: In this subsection we provide a simple estimation 

725: for the order of a zero in $U$ of a function of the form

726: $$\psi(z)=\sum_{i=1}^nf_iQ(P^{-1}_i(z)), \ \ \ \ \ \ \ \ \ \ \ \ f_i\in \C,$$

727: via the degrees of $P(z)$ and $Q(z).$ This reduces 

728: the verification of the criterion

729: to the calculation of a finite set of initial coefficients of

730: Puiseux expansions of functions \e{piz}

731: and, as a corollary, provides a practical method for checking an infinite

732: set of equation \e{1} in a finite number of steps.

733: 

734: 

735: \bl\l{ur} If $\psi(z)\neq 0$ then $\psi(z)$

736: satisfies an equation

737: \be \l{urr}

738: y^N(z)+a_1(z)y^{N-1}(z)+\ ... \ + a_N(z)=0,

739: \ee

740: where $a_j(z)\in \C[z],$

741: $a_N(z)\neq 0,$ and $N\leq n!.$

742: Furthermore, 

743: $\deg a_j(z)\leq \left(\frac{m}{n}\right)^{j},$ $1\leq j \leq N.$

744: \el

745: 

746: \pr Indeed, if $\psi(z)\neq 0$ then, since $\psi(z)$ is a sum of algebraic

747: functions, $\psi(z)$ itself is an algebraic function and

748: therefore satisfies an algebraic equation \e{urr}

749: with $a_i(z)\in \C(z),$ $1\leq j \leq N.$ Furthermore, we can

750: suppose that this equation

751: is irreducible. Then $a_N(z)\neq 0$

752: and the number $N$ 

753: coincides with the number of different analytic continuations

754: $\psi_j(z)$ of

755: $\psi(z)$ along closed curves. Clearly,

756: $N$ can be bounded by the number $N_1$ of different elements of

757: the monodromy group of $P(z).$ In its turn, $N_1$ is bounded by the

758: number of elements of the full symmetric group $S_n.$ Hence, $N\leq n!.$

759: 

760: Furthermore, since $P(z),Q(z)$ are polynomials, the rational functions

761: $a_j(z),$ $1\leq j \leq N,$ as the symmetric functions

762: of $\psi_j(z),$ $1\leq j \leq N,$ have no poles in $\C$ and therefore

763: are polynomials. Finally, since near infinity branches $P^{-1}_i(z),$

764: $1\leq i \leq n,$ 

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

766: are represented 

767: by the Puiseux series 

768: \be 

769: P^{-1}_i(z)=\sum_{k=-1}^{\infty}v_k\varepsilon_n^{ik}z^{-\frac{k}{n}},

770: \ \ \ \ \ \ \ \ v_k\in \C, \ \ \ \ \ \ \ \ \varepsilon_n=exp(2\pi i/n),

771: \ee

772: the first non-zero exponent of the Puiseux series 

773: at infinity for the functions

774: $\psi_j(z),$ $1\leq j \leq N,$ 

775: is less or equal than $m/n.$ It follows that

776: $\deg a_j(z)\leq \left(\frac{m}{n}\right)^{j},$ $1\leq j \leq N.$

777: 

778: \bc Let $z_0 \in U.$ To verify that $\psi(z) \equiv 0$ it is enough

779: to check that the first $\left(\frac{m}{n}\right)^{n!}+1$ coefficients of

780: the series

781: $\psi(z)=\sum_{k=0}^{\infty} v_k (z-z_0)^k$ vanish.

782: 

783: \ec

784: 

785: \pr Indeed, suppose that $\ord_{z_0}\psi(z) > \left(\frac{m}{n}\right)^{n!}$

786: but $\psi(z)\neq 0.$ Then,

787: by lemma \ref{ur}, $\psi(z)$ satisfies \e{urr}, where 

788: $\deg a_j(z)\leq \left(\frac{m}{n}\right)^{j}\leq \left(\frac{m}{n}

789: \right)^{n!},$ $1\leq j \leq N,$ and $a_N\neq 0.$

790: It follows that

791: $$\ord_{z_0}\{\psi^N(z)\} >

792: \ord_{z_0}\{a_{i_1}(z)\psi^{N-i_1}(z)\} >\ ... \ >\ord_{z_0}\{a_{i_k}(z)

793: \psi^{N-i_k}(z)\},$$ where $0 < i_1 < i_2 <\ ... \ <i_k=N$ are all indices

794: for which $a_i(z) \neq 0.$

795: Therefore,

796: $$\ord_{z_0}\{\psi^N(z)+

797: a_1(z)\psi^{N-1}(z) +\ ... \ +a_{N}(z)\}

798: =\ord_{z_0}\{a_{N}(z)\}<\infty $$ in contradiction with equality

799: \e{urr}.

800: 

801: 

802: 

803: \section{Application to a description of definite polynomials}

804: In this section, as a first

805: application of theorem \ref{t1}, we provide

806: a number of conditions

807: on a collection $P(z),a,b,$ where $P(z)\in \C[z],$ $a,b\in \C,$

808: $a\neq b,$ under which conditions \e{1} and \e{2}

809: are equivalent; such collections are called {\it definite} and

810: are of interest because of applications to the Abel equation (see \cite{bby},

811: \cite{bry}, \cite{by}).

812: 

813: \subsection{A combinatorial condition for a change of variable}

814: 

815: The simplest form of the equality 

816: $\phi_s(z)=0$ 

817: is equality \e{eb}.

818: Furthermore, \e{eb} has a clear compositional meaning.

819: 

820: 

821: \bl \l{lcomp} The equality \eqref{eb} holds if and only if

822: \be \l{comp}

823: P(z)=\tilde P(W(z)), \ \ \ \ \ \ \ \ Q(z)=\tilde Q(W(z))

824: \ee

825: for

826: some polynomials $\tilde P(z),$ $\tilde Q(z),$ $W(z)$ with $\deg W(z)>1.$

827: \el

828: 

829: The proof of this lemma easily follows from the L\"{u}roth theorem

830: (see e.g. \cite{pa2}, \cite{ro}).

831: If condition \e{comp} is satisfied we say that

832: {\it polynomials $P(z),$ $Q(z)$ have a (non-trivial)

833: common right divisor in the composition algebra of polynomials.}

834: 

835: Below we give a convenient combinatorial

836: condition on a collection $P(z),a,b$ which implies that for any 

837: $q(z)$ satisfying \e{1} polynomials $P(z),$ $Q(z)=\int q(z)\d z$ have a

838: common right divisor in the composition algebra of polynomials.

839: The use of this condition permits, after the change of variable

840: $z\rightarrow W(z),$

841: to reduce the solution of the polynomial

842: moment problem for a polynomial $P(z)$ to that for a polynomial

843: of lesser degree $\tilde P(z).$

844: 

845: Let $\tilde \lambda_P$ be a $\tilde k+1$ colored extended cactus

846: corresponding to a collection $P(z),a,b$ and let

847: $\Gamma_{a,b}\subset \lambda_P$

848: be the path connecting points $a,b$ on $\tilde \lambda_P.$ 

849: For each $s,$ $1\leq s \leq \tilde k,$

850: define the weight $w(s)$ of the $s$-th color

851: on $\Gamma_{a,b}$ as a number of vertices $v\in \Gamma_{a,b}$

852: colored by the $s$-th color with the convention that vertices $a,b$ are

853: counted with the coefficient $1/2.$ For example, for $\Gamma_{a,b}$

854: shown on Fig. 1 we have $w(1)=w(3)=3/2,$ $w(2)=1.$ 

855: 

856: \bt \l{rt}  Let $P(z),q(z)\in \C[z]$, $q(z)\neq 0,$

857: $a,b\in \C,$ $a\neq b$ satisfy \e{1}. Suppose that there exists

858: $s,$ $1\leq s \leq \tilde k,$ such that $w(s)=1$ on $\Gamma_{a,b}.$

859: Then $P(z),Q(z)$ have a common right divisor in the composition algebra.

860: \et

861: \pr Indeed, the construction of $\Gamma_{a,b}$ implies that if

862: $w(s)=1,$ then $f_{s,i}\neq 0$

863: exactly for two values $i_1,i_2,$ $1\leq i_1,i_2 \leq n.$ Moreover, for these values we have

864: $c_{s,{i_1}}=-c_{s,{i_2}}$ and

865: hence the equality $\phi_s(z)=0$

866: reduces to \e{eb}. Therefore,

867: $P(z)$ and $Q(z)$ have a common right divisor in the composition algebra

868: by lemma \ref{lcomp}.

869: 

870: \subsection{Reduction}

871: 

872: 

873: 

874: Although condition \eqref{comp} in general is weaker than condition \eqref{2} 

875: it turns out that in order to prove

876: that for any collection

877: $P(z),a,b,$ $a\neq b,$

878: satisfying some condition $\mathcal R$

879: conditions \eqref{1} and \eqref{2} are equivalent

880: it is often enough to show 

881: that for any such a collection

882: condition \e{1} implies condition \e{comp}.

883: Say that a condition $\mathcal R$

884: is {\it compositionally stable} if for any

885: collection $P(z),a,b,$ $a\neq b,$ satisfying $\mathcal R$

886: such that $P(z)=\tilde P(W(z))$ for

887: some $\tilde P(z), W(z)\in \C[z],$ ${\deg W(z) >1,}$ $W(a)\neq W(b),$   

888: the collection $\tilde P(z),W(a),$ $W(b)$ also

889: satisfies $\mathcal R.$ For instance, the following condition is

890: compositional stable: at least one point from $a,b$ is

891: not a critical point of $P(z).$ An other example of a

892: compositional stable condition

893: is the following one: $\deg P(z)=p^r,$ where $p$ is a prime.

894: 

895: 

896: 

897: \vskip 0.2cm

898: 

899: \bl \l{rl}

900: Let $\mathcal R$ be

901: a compositionally stable condition.

902: Suppose that for any collection $P(z),a,b,$ $a\neq b,$

903: satisfying $\mathcal R,$

904: condition \e{1} implies condition \e{comp}.

905: Then for any collection $P(z),a,b,$ $a\neq b,$

906: satisfying $\mathcal R$ conditions \e{1} and \e{2} are equivalent.

907: \el

908: 

909: \vskip 0.2cm

910: 

911: \pr

912: Let $P(z),a,b$ be a collection satisfying $\mathcal R.$

913: Suppose that \e{1} holds for some $q(z)\in \C[z].$ Then it follows from

914: \e{comp} that $\C(P,Q)$ is a proper

915: subfield of $\C(z).$ Therefore, by the L\"{u}roth theorem

916: \be\l{lu}

917: \C(P,Q)=\C(W_1)

918: \ee for some rational function $W_1(z),$ $\deg W_1(z) > 1,$ 

919: and without loss of generality we can assume that

920: $W_1(z)$ is a polynomial. To prove the lemma it is enough to show

921: that the equality $W_1(a)=W_1(b)$ holds. 

922: 

923: Observe that equality \e{lu} is equivalent to the statement

924: that 

925: \be \l{comp+}

926: P(z)=P_1(W_1(z)), \ \ \ \ \ \ \ \ Q(z)=Q_1(W_1(z))

927: \ee

928: for

929: some polynomials $P_1(z),$ $Q_1(z)$ such that

930: $P_1(z)$ and $Q_1(z)$ have no a

931: common right divisor in the composition algebra.

932: On the other hand, 

933: performing the change of variable $z\rightarrow W_1(z)$

934: we see that condition \e{1} is satisfied also for

935: $P_1(z), Q_1^{\prime}(z), W_1(a),W_1(b).$

936: Therefore, if $W_1(a)\neq W_1(b),$ then,

937: taking into account that $\mathcal{R}$ is compositionally stable,

938: we find that $$P_1(z)=P_2

939: (W_2(z)), \ \ \ \ \ \ \ Q_1(z)=

940: Q_2(W_2(z))$$ for some $P_2(z), Q_2(z), W_2(z)\in \C[z]$ with

941: $\deg W_2(z) >1$ that contradicts the fact that

942: $P_1(z),$ $Q_1(z)$ have no a

943: common right divisor in the composition algebra.

944: 

945: 

946: \vskip 0.2cm

947: 

948: 

949: 

950: 

951: 

952: \vskip 0.2cm

953: 

954: 

955: 

956: 

957: 

958: \subsection{Description of some classes of definite

959: polynomials}

960: 

961: As a first application of theorem 

962: \ref{rt} and lemma \ref{rl} we give a simple proof

963: of the following assertion conjectured in \cite{pry}.

964: 

965: \bc \l{c1} Let $P(z),q(z)\in \C[z]$, $q(z)\neq 0,$

966: $a,b\in \C,$ $a\neq b.$ Suppose that

967: $P(a)=P(b)=c_1$ and that all the

968: points of the preimage

969: $P^{-1}(c_1)$ except possibly $a,b$ are not critical points

970: of $P(z).$ Then conditions \e{1} and \e{2} are equivalent. 

971: \ec

972: 

973: \pr

974: Since the chain rule implies that

975: the condition of the corollary is compositionally

976: stable it is enough to show that 

977: $P(z),$ $Q(z)$ have a

978: common right divisor in the composition algebra.

979: To establish it observe that $\Gamma_{a,b}$ can not pass through

980: vertices of $\lambda_P$ of the valency $1$ distinct from $a,b.$

981: Therefore, the

982: condition of the corollary implies 

983: that $w(1)=1.$ It follows now from theorem \ref{rt} that

984: $P(z),$ $Q(z)$ have a

985: common right divisor in the composition algebra.

986: 

987: \vskip 0.2cm A slight modification of the idea used in the proof

988: of corollary \e{c1}

989: leads to the following

990: statement.

991: 

992: 

993: \bc Let $P(z),q(z)\in \C[z]$, $q(z)\neq 0,$

994: $a,b\in \C,$ $a\neq b.$ Suppose that

995: $P(a)=P(b)=c_1$ and that for any critical

996: value $c$ of $P(z)$ except possibly $c_1$ the preimage $P^{-1}(c)$

997: contains only one critical point. Then conditions \e{1} and \e{2} are 

998: equivalent. \ec

999: 

1000: \pr

1001: Again, it follows from the chain rule that 

1002: the condition of the corollary is compositionally stable.

1003: Furthermore, observe that

1004: the path $\Gamma_{a,b}$ contains at least one vertex $v$ 

1005: of a color $s\neq 1.$  

1006: Since $\Gamma_{a,b}$ can not pass through vertices of the

1007: valency 

1008: $1$ distinct from $a,b,$ it follows from the condition

1009: of the corollary that 

1010: the equality

1011: $w(s)=1$ holds and, therefore, by theorem \ref{rt},

1012: $P(z),$ $Q(z)$ have a

1013: common right divisor in the composition algebra.

1014: 

1015: \vskip 0.2cm

1016: 

1017: Finally, we give a new proof of an

1018: assertion from the paper \cite{pry}

1019: which provides some geometric condition for a collection $P(z),a,b$

1020: to be definite. It turns out that

1021: this assertion actually also can be regarded as a particular

1022: case of theorem \ref{rt}.

1023: For a curve $M$ 

1024: denote by $V_{M,\infty}$

1025: the domain from the collection of domains $\C\P^1\setminus M$ which contains infinity. 

1026: For an oriented curve $L$ and points 

1027: $d_1,d_2\in L$ 

1028: denote by $L_{d_1,d_2}$ the part of

1029: $L$ between $d_1$ and $d_2.$ 

1030: 

1031: \bc Let $P(z),q(z)\in \C[z]$, $q(z)\neq 0,$ 

1032: $a,b\in \C,$ $a\neq b.$ Suppose that

1033: $P(a)=P(b)=c_1$ and that

1034: there exists a curve $L$ 

1035: connecting points $a,b$

1036: such that $c_1$

1037: is a simple point of $P(L)$ and $c_1\in \partial V_{P(L),\infty}.$

1038: Then conditions \e{1} and \e{2} are 

1039: equivalent.

1040: \ec 

1041: 

1042: 

1043: \pr

1044: We will keep the notation introduced in subsection \ref{cacti}

1045: and \ref{criter}.

1046: Let $a^+$ (resp. $b^-$) be a point on $L$ near the point $a$

1047: (resp. $b$) and

1048: let $U$ be a

1049: simply connected domain containing no critical values of $P(z)$ such 

1050: that the sets $S\setminus\{c_1,c_2,... ,c_{\tilde k}\},$

1051: $P(L_{a,a^+})\setminus c_1,$

1052: and $P(L_{b^-,b})\setminus c_1$ are subsets of

1053: $U.$ Recall that there is a natural correspondence between branches

1054: $P_i^{-1}(z),$ $1\leq i \leq n,$ of

1055: $P^{-1}(z)$ in $U$ and stars of the cactus $\lambda_P:$

1056: branch $P_i^{-1}(z)$ maps $U$ on a domain $U_i$ containing $S_i.$

1057: Denote by $U_{j_1}$ (resp. $U_{j_2}$) the domain containing the point $a^+$

1058: (resp. $b^-$). Then by construction the result of the analytic continuation

1059: of the element $(P^{-1}_{j_1}(z),U)$ along the curve

1060: $P(L_{a^+,b^-})$ is the element $(P^{-1}_{j_2}(z),U).$

1061: \begin{figure}

1062: \medskip

1063: \epsfxsize=4.5truecm

1064: \centerline{\epsffile{gg.eps}}

1065: \smallskip

1066: \centerline{Figure 2}

1067: \medskip

1068: \end{figure}

1069: Let $c_0$ be an interior point of $U$ close to $c_1.$

1070: Consider a small deformation $M$ of the curve $P(L_{a,b})$ obtained

1071: as follows: change the part of $P(L_{a,b})$ connecting $c_1$

1072: and $P(a^+)$

1073: to an arc $\gamma^+\subset U$ connecting $c_0$ with $P(a^+)$

1074: and, similarly, change the part of $P(L_{a,b})$ connecting

1075: $P(b^-)$ and $c_1$ 

1076: to an arc $\gamma^-\subset U$ connecting $P(b^-)$ and $c_0$ (see Fig. 2).

1077: 

1078: Let now $l_M=l_{i_1}^{j_1}l_{i_2}^{j_2}\, ...\, l_{i_r}^{j_r}$ 

1079: be the image of the curve $M$ in 

1080: $\pi_1(X, c_0),$ where $X=\C\P^1\setminus \{c_1,c_2,c_3, \, ... \,

1081: c_{k}\}.$ 

1082: Since the result of the analytic continuation

1083: of the element $(P^{-1}_{j_1}(z),U)$ along the curve

1084: $M$ is still the element $(P^{-1}_{j_2}(z),U),$

1085: the final element of the chain of stars 

1086: $$\Omega=<S_{j_1}\,,\;S_{g_{i_1}^{j_1}(j)}\,,\;

1087: S_{g_{i_1}^{j_1}g_{i_2}^{j_2}(j)}

1088: \,,\;

1089: S_{g_{i_1}^{j_1}g_{i_2}^{j_2}g_{i_3}^{j_3}(j)}\,,\;...\;,\;

1090: S_{g_{i_1}^{j_1}g_{i_2}^{j_2}\,...\; g_{i_r}^{j_r}(j)}>$$

1091: is the star $S_{j_2}.$

1092: In particular, the path $\Gamma_{a,b}$ is contained in

1093: $\Omega.$ 

1094: Since $c_1\in V_{M,\infty},$  

1095: the loop $l_1$ does not appear

1096: among the loops $l_{i_1},l_{i_2},\,...\,l_{i_r}.$

1097: Therefore, the common vertex of any two successive stars in the chain

1098: $\Omega$ is not contained in the set $P^{-1}(c_1).$

1099: In particular, among of vertices of $\Gamma_{a,b}$ there are no preimages

1100: of $c_1$ distinct from $a,b$ and hence

1101: $w(1)=1$ on $\Gamma_{a,b}.$

1102: 

1103: To finish the proof notice that the condition of the

1104: corollary is compositionally stable. Indeed, if 

1105: $L$ is a curve connecting points $a,b$

1106: such that $c_1=P(a)=P(b)$

1107: is a simple point of $P(L)$ and $c_1\in \partial V_{P(L),\infty},$

1108: then $W(L)$ is a curve connecting points $W(a),W(b)$

1109: such that $c_1=\tilde P(W(a))=\tilde P(W(b))$

1110: is a simple point of $\tilde P(W(L))$ and $c_1\in \partial V_{\tilde

1111: P(W(L)),\infty}.$

1112: 

1113: 

1114: \vskip 0.2cm

1115: 

1116: 

1117: 

1118: 

1119: 

1120: \section{On functional equations

1121: \e{e1} and \e{e2}}

1122: 

1123: \subsection{Derivation of \e{e1} and \e{e2}}

1124: Let $U$ be a simply connected domain containing no critical values of $P(z)$

1125: such that $S\setminus\{c_1,c_2, ... ,c_{\tilde k}\}\subset U.$

1126: Denote by 

1127: $P_{a_1}^{-1}(z),$

1128: $P_{a_2}^{-1}(z), ... , P_{a_{f_a}}^{-1}(z)$

1129: (resp. $P_{b_1}^{-1}(z),$ $P_{b_2}^{-1}(z),$ ... , 

1130: $P_{b_{f_b}}^{-1}(z)$)

1131: the branches

1132: of $P^{-1}(z)$ in $U$ 

1133: which map points close

1134: to $P(a)$ (resp. $P(b)$)

1135: to points close to $a$ (resp. $b$). 

1136: In particular, $f_a$ (resp. $f_b$)

1137: equals the multiplicity of the point $a$ (resp. $b$) 

1138: with respect to $P(z).$ It was shown in \cite{pa2} for $P(a)=P(b)$

1139: and in \cite{pry} in general case that condition \e{1} implies

1140: equality \e{e1} or system \e{e2}, where

1141: as above $Q(z)=\int q(z) \d z$ is normalized

1142: by the condition ${Q(a)=Q(b)=0}.$

1143: 

1144: For the sake of self-containedness of this paper below

1145: we provide a short derivation of \e{e1}, \e{e2} from

1146: theorem \ref{t1}.

1147: 

1148: 

1149: \bp \l{brc} Suppose that condition \eqref{1} holds. Then, if $P(a)=P(b),$ 

1150: equation \e{e1}

1151: holds in $U$.

1152: Furthermore, if $P(a)\neq P(b),$ then system \e{e2} holds

1153: in $U.$

1154: \ep

1155: 

1156: \pr Suppose first that $P(a)= P(b)=c_1.$ Examine the relation 

1157: $$\phi_1(z)=\sum_{i=1}^nf_{1,i}Q(P^{-1}_{i}(z))=0.$$ 

1158: Let $i,$ $1\leq i \leq n,$ be an index such that $f_{1,i}\neq 0$

1159: and let $x$ be a vertex of the star $S_i$ such that $P(x)=c_1.$

1160: Observe that if $x\neq a,$ $x\neq b,$ then there exists an index

1161: $\tilde i$ such 

1162: that $x$ is also a vertex of the star $S_{\tilde i}$ and 

1163: $f_{1,\tilde i}=-f_{1,i}.$ Furthermore, we have $\tilde i=g_1^j(i)$

1164: for some natural number $j.$ Therefore, $\phi_1(z)$ has the form

1165: $$\phi_1(z)=-Q(P^{-1}_{i_a}(z))+ $$

1166: $$

1167: Q(P^{-1}_{i_1}(z))-Q(P^{-1}_{g_1^{j_1}(i_1)}(z))+

1168: \, ... \, +

1169: Q(P^{-1}_{i_r}(z))-Q(P^{-1}_{g_1^{j_r}(i_r)}(z))$$

1170: $$+Q(P^{-1}_{i_b}(z))=0,$$

1171: where $i_a$ (resp. $i_b$) is the index such that 

1172: $a\subset S_{i_a}$

1173: (resp. $b\subset S_{i_b}$), $i_1,i_2,\, ... \, i_r$

1174: are some other indices and $j_1,j_2,\, ... \, j_r$

1175: are natural numbers.

1176: 

1177: Let $n_1$ be the order of the element $g_1$ in the group $G_P.$

1178: For each $s,$ \linebreak ${0\leq s \leq n_1-1,}$ the equality

1179: $$-Q(P^{-1}_{g^s_1(i_a)}(z))+ $$

1180: $$

1181: Q(P^{-1}_{g^s_1(i_1)}(z))-Q(P^{-1}_{g_1^{j_1+s}(i_1)}(z))+

1182: \, ... \, +

1183: Q(P^{-1}_{g^s_1(i_r)}(z))-Q(P^{-1}_{g_1^{j_r+s}(i_r)}(z))$$

1184: $$+Q(P^{-1}_{g^s_1(i_b)}(z))=0$$ holds by the analytic continuation

1185: of the equality 

1186: $\phi_1(z)=0.$ Summing these equalities and taking into account that

1187: for any $i,$ $1\leq i \leq n,$ and any natural number $j$ we have:

1188: $$\sum_{s=0}^{n_1-1} 

1189: Q(P^{-1}_{g^s_1(i)}(z))=\sum_{s=0}^{n_1-1} Q(P^{-1}_{g_1^{j+s}(i)}(z)),$$

1190: we obtain equality \e{e1}.

1191: 

1192: In the case when $P(a)\neq P(b)$ the proof is similar: if $P(a)=c_1,$

1193: $P(b)=c_2,$ then one must examine relations $\phi_1(z)=0$ and $\phi_2(z)=0.$

1194: 

1195: \vskip 0.2cm

1196: Note that if points $a,b$ are not critical

1197: points of $P(z),$ then \eqref{e1} reduces to \eqref{eb}

1198: while \eqref{e2} leads to the equality $q(z)\equiv 0.$

1199: In view of lemmas \ref{lcomp} and \ref{rl} this implies

1200: immediately the following result from \cite{c}

1201: (see also \cite{pa2},\cite{pry}).

1202: 

1203: \bc \l{cris}

1204: Let $P(z),q(z)\in \C[z]$, $q(z)\neq 0,$

1205: $a,b\in \C,$ $a\neq b.$ Suppose that

1206: $a,$ $b$ are not critical points  of $P(z).$

1207: Then conditions \e{1} and \e{2} are 

1208: equivalent. \ec

1209: 

1210: 

1211: \subsection{Relations between branches of $Q(P^{-1}(z))$}

1212: In this subsection we examine how linear relations between branches of \linebreak $Q(P^{-1}(z))$ over $\C$ reflect on the structure of coefficients of Puiseux

1213: expansions of $Q(P^{-1}(z))$ near

1214: infinity.

1215: 

1216: Let $P(z)$ be a non-constant polynomial of degree $n$

1217: and 

1218: let $z_0\in \C$ be a non-critical

1219: value of $P(z).$

1220: If

1221: $\vert z_0\vert $ is sufficiently large then in a neighborhood $U_{z_0}$ 

1222: of $z_0$ each

1223: branch of $P^{-1}(z)$ can be 

1224: represented

1225: by a Puiseux series centered at infinity. More precisely,

1226: if $P_0^{-1}(z)$ is a fixed branch of $P^{-1}(z)$ near $z_0$ then

1227: in $U_{z_0}$ we have: $$P_0^{-1}(z)=

1228: \sum_{k=-1}^{\infty}v_kz^{-\frac{k}{n}}, \ \ \ \ \ \ \ \ v_k\in \C, \ \ \ 

1229: \ \ \ \ \ \varepsilon_n=exp(2\pi i/n),$$

1230: where $z^{\frac{1}{n}}$ is a branch of

1231: the algebraic function which is inverse to $z^n$ in $U_{z_0}.$  

1232: If $l$ is a loop around infinity then 

1233: the result of the analytic continuation of the branch 

1234: $P^{-1}_0(z)$ along $l^j,$ $0\leq j \leq n-1,$ is represented

1235: by the series 

1236: \be \l{cn}

1237: P^{-1}_j(z)=\sum_{k=-1}^{\infty}v_k\varepsilon_n^{jk}z^{-\frac{k}{n}}.

1238: \ee

1239: The numeration of branches of $P^{-1}(z)$ near $z_0$ defined by equation

1240: \eqref{cn} 

1241: is called canonical. Clearly, such a numeration depends on

1242: the choice of $P^{-1}_0(z).$ Nevertheless, any canonical numeration 

1243: induces

1244: the same cyclic ordering of branches of $P^{-1}(z)$ in $U_{z_0}.$

1245: This cyclic ordering also will be called canonical.

1246: For any non-zero polynomial $Q(z),$ $\deg Q(z)=m,$  

1247: the composition $Q(P^{-1}_j(z)),$  $0\leq j \leq n-1,$

1248: is represented near $z_0$ by the series \e{ps2}

1249: obtained by the substitution of series \eqref{cn} in $Q(z).$

1250: 

1251: Let $U$ be a simply-connected domain containing no critical values of

1252: $P(z)$ such that some linear combination of branches

1253: of $Q(P^{-1}(z))$ over $\C$ identically vanishes in $U$.

1254: Considering in case of necessity a bigger domain we can 

1255: suppose without loss of generality that $\infty \in \partial U.$

1256: Then series \e{cn} converge in a domain

1257: $V\subset U.$

1258: Furthermore, we can assume that

1259: the numeration of branches of $P^{-1}(z)$ in $U$ 

1260: is induced by a canonical numeration

1261: of branches of $P^{-1}(z)$ in $V$.

1262: If equality

1263: \be \l{x}

1264: \sum_{i=0}^{n-1}f_jQ(P^{-1}_j(z))=0, \ \ \ \ \ \ f_j\in \C, 

1265: \ee

1266: holds in $U,$ then substituting in \e{x} expansions \e{ps2}

1267: we see that \e{x} reduces to the system

1268: $$\sum_{j=0}^{n-1}f_ju_k\varepsilon^{kj}_n=0, \ \ \ \ \ \ \ k\geq -m.$$

1269: Introducing the notation $F(z)=\sum_{j=0}^{n-1}f_jz^j$

1270: and summing up we get:

1271: 

1272: \bl \l{la} The equality \e{x}

1273: holds in $U$ if and only if 

1274: for any $k\geq -m$ 

1275: either

1276: $u_k=0$

1277: or $F(\varepsilon^k_n)=0.$

1278: \el

1279: 

1280: In particular, since all $u_k$ can not

1281: vanish

1282: and $\deg F(z)<n,$ the following statement is true.

1283: 

1284: \bc \l{la+}

1285: If equality \e{x} holds in $U$

1286: %for non-constant polynomials $P(z),Q(z),$

1287: then $F(\varepsilon^r_n)=0$ for at least one $r,$ ${0\leq r \leq n-1.}$ On the other hand, for at least one $r,$ $0\leq r \leq n-1,$ the equality $u_k=0$ holds whenever $k\equiv r \ \mod n.$

1288: \ec

1289: 

1290: \subsection{Lemma about monodromy groups of polynomials}

1291: In order to relate \e{e1}, \e{e2}

1292: with coefficients of Puiseux

1293: expansions of $Q(P^{-1}(z))$ near

1294: infinity

1295: we are going to examine

1296: which roots of unity can be roots of the corresponding polynomial

1297: \be \l{e11}

1298: r(z)=\frac{1}{d_{a}}\sum_{s=1}^{d_{a}}

1299: z^{a_s}-\frac{1}{d_{b}}\sum_{s=1}^{d_{b}}

1300: z^{b_s},

1301: \ee or common roots of the corresponding pair of polynomials

1302: \be \l{e22}

1303: r_1(z)=\frac{1}{d_{a}}\sum_{s=1}^{d_{a}}

1304: z^{a_s}, \ \ \ \ \ \ \ \ \ \ r_2(z)=\frac{1}{d_{b}}\sum_{s=1}^{d_{b}}

1305: z^{b_s}.

1306: \ee

1307: For this propose we establish now a 

1308: geometric property of monodromy groups of polynomials 

1309: which concerns the mutual arrangement

1310: of indices $a_1,$

1311: $a_2, ... , a_{f_a}$

1312: and $b_1,$ $b_2,$ ... , 

1313: $b_{f_b}$ under assumption that the numeration of branches is canonical.

1314: 

1315: Let $P(z)\in \C[z],$ $\deg P(z)=n,$ $a,b\in \C,$ $a\neq b.$

1316: Let $U$

1317: be a simply-connected domain containing no critical values of $P(z)$ 

1318: such that $P(a),P(b), \infty \in \partial U.$ 

1319: Fix a canonical numeration of branches of $P^{-1}(z)$ in $U$

1320: and let $P_{u_1}^{-1}(z),$

1321: $P_{u_2}^{-1}(z), ... ,$ $P_{u_{f_a}}^{-1}(z)$

1322: (resp. $P_{v_1}^{-1}(z),$ $P_{v_2}^{-1}(z), ... ,

1323: P_{v_{f_b}}^{-1}(z)$) be

1324: the branches of $P^{-1}(z)$ in $U$ which map points close

1325: to $P(a)$ (resp. $P(b)$)

1326: to points close to the point $a$ (resp. $b$) numbered

1327: by means of this numeration.

1328: The lemma below describes the

1329: mutual position on the unit circle of the sets

1330: $V(a)=\{ \v_n^{a_1},\v_n^{a_2}, ... , \v^{a_{f_a}}_n \}$ and

1331: $V(b)=\{\v_n^{b_1},\v_n^{b_2}, ... ,\v_n^{b_{f_b}}\},$

1332: where $\varepsilon_n=exp(2\pi i/n).$

1333: Let us introduce the following

1334: definitions.

1335: Say that two sets of points $X,Y$ on the unit circle $S_1$ are

1336: {\it disjointed} if there exist $s_1, s_2 \in S_1$

1337: such that all points from $X$ are on

1338: the one of two connected components of $S_1\setminus \{s_1,

1339: s_2\}$

1340: while all points from $Y$ are on

1341: the other one. Say that $X,Y$ are

1342: {\it almost disjointed} if $X\cap Y$ consists of a single point $s_1$

1343: and there exists a point $s_2\in S_1$ such that

1344: all points from $X\setminus s_1$ are on

1345: the one of two connected components of $S_1\setminus \{s_1,

1346: s_2\}$ while all points from $Y\setminus s_1$ are on

1347: the other one.

1348: 

1349: \vskip 0.2cm

1350:  

1351:  

1352: \noindent{\bf Monodromy Lemma.} {\it 

1353: The sets $V(a)$ and $V(b)$ are almost disjointed. Furthermore,

1354: if $P(a)= P(b)$ then $V(a)$ and $V(b)$ are disjointed.

1355: }

1356:  

1357:  

1358: \vskip 0.2cm

1359:  

1360: \noindent{\it Proof.}

1361: Consider first the case when $P(a)=P(b).$

1362: \begin{figure}

1363: \medskip

1364: \epsfxsize=9truecm

1365: \centerline{\epsffile{G1.eps}}

1366: \smallskip

1367: \centerline{Figure 3}

1368: \medskip

1369: \end{figure} 

1370: Let $M\subset U$ be a simple curve connecting points $P(a)=P(b)$

1371: and $\infty.$

1372: Consider the preimage

1373: $P^{-1}\{M\}$ of $M$ under the map $P(z)\,:\,\C\P^1\rightarrow

1374: \C\P^1.$

1375: It is convenient to consider $P^{-1}\{M\}$ as a bicolored graph

1376: $\Omega$ embedded into the Riemann sphere:

1377: the black vertices of $\Omega$ are preimages of

1378: $P(a)=P(b),$ the

1379: unique white vertex is the preimage of $\infty,$ and

1380: the edges of $\Omega$ are preimages of $M$ (see Fig. 3). 

1381: Since the multiplicity of the vertex $\infty$ equals $n$ and $\Omega$

1382: has $n$ edges, $\Omega$ is connected.

1383: The edges of $\Omega$ are identified with branches of $P^{-1}(z)$ in $U$

1384: as follows:

1385: to a branch $P^{-1}_k(z),$ $1\leq k \leq n,$  

1386: corresponds the edge $e_k$

1387: such that $P^{-1}_k(z)$ maps $M\setminus \{P(a),\infty\}$ into $e_k.$

1388: For any vertex $v$ of $\Omega$ the orientation of $\C\P^1$

1389: induces a natural cyclic ordering on edges of $\Omega$ adjacent to

1390: $v.$ In particular, taking $v=\infty,$ we obtain a 

1391: cyclic ordering on edges of $\Omega.$

1392: Clearly, this cyclic ordering coincides with that induced by

1393: the canonical cyclic ordering of branches of $P^{-1}(z)$ in $U.$

1394: Let $E_a=\{e_{a_1},e_{a_2},\, ... \,, e_{a_{f_a}}\}$

1395: (resp. $E_b=\{e_{b_1},e_{b_2},\, ... \,, e_{b_{f_b}}\}$) be the union

1396: of edges of

1397: $\Omega$ which are adjacent to the vertex $a$ (resp. $b$).

1398: Let $D$

1399: be the domain from the collection of domains

1400: $\C\P^{1}\setminus E_a$ which contains point $b$

1401: and let $e_s,e_t\in E_a$ be the edges which bound $D.$

1402: Clearly, all the edges from $E_a$ are contained in

1403: $\C\P^1\setminus D.$

1404: Therefore, the lemma is equivalent to the following statement: 

1405: the domain $D$ contains $e_h\setminus {\infty}$

1406: for all $e_h\in E_b.$ But the last statement is a corollary

1407: of the Jordan theorem since an edge $e_h\in E_b$ can intersect $e_s$ or   

1408: $e_t$ only at infinity.

1409: 

1410: 

1411: In the case when $P(a)\neq P(b)$ the proof is modified as follows.

1412: Divide the boundary of $U$ into three

1413: parts $M_1,M_2,M_3,$ where $M_1$ connects the point $\infty$ with the 

1414: point 

1415: $P(a),$ $M_2$ connects the point $\infty$ with the point 

1416: $P(b),$ and $M_3$ connects the point $P(a)$ with the point 

1417: $P(b).$ 

1418: Consider now $P^{-1}\{\partial U\}$ as a graph

1419: $\Omega$ embedded into the Riemann sphere. The vertices of $\Omega$ 

1420: are divided into three groups: the first one consists of vertices  

1421: that are preimages of $\infty,$ the second one consists of vertices  

1422: that are preimages of $P(b),$ and the third one consists of vertices  

1423: that are preimages of $P(a).$ Similarly, the edges of $\Omega$ 

1424: also are divided into three groups: the first one consists of edges

1425: that are preimages of $M_1,$ the second one consists of edges 

1426: that are preimages of $M_2,$ and the third one consists of edges 

1427: that are preimages of $M_3.$ Finally, the faces of $\Omega$ are divided 

1428: into two groups: the first one consists of faces

1429: that are preimages of $U$ and the second one consists of faces

1430: that are preimages of $\C\P^1\setminus \overline{U}$

1431: (see Fig. 4). 

1432: 

1433: 

1434: \begin{figure}

1435: \medskip

1436: \epsfxsize=11truecm

1437: \centerline{\epsffile{G2.eps}}

1438: \smallskip

1439: \centerline{Figure 4}

1440: \medskip

1441: \end{figure} 

1442:  

1443: 

1444: The faces from the first group are identified with branches of $P^{-1}(z)$ 

1445: in $U$

1446: as follows:

1447: %if $N_3$ is a numeration of branches of $P^{-1}(z)$ in $V$

1448: to a branch $P^{-1}_k(z),$ $1\leq k \leq n,$  

1449: corresponds the face $f_k$

1450: such that $P^{-1}_k(z)$ maps bijectively $U$ on $f_k\setminus \partial f_k.$

1451: The edges from the corresponding groups which bound $f_k$ will be denoted 

1452: by

1453: $e^1_k, e^2_k, e^3_k$ correspondingly.  

1454: Note that in a counterclockwise direction around infinity the edge 

1455: $e^1_k,$ $1\leq k \leq n,$ is

1456: followed by the edge $e^2_k.$

1457: The canonical cyclic ordering of

1458: branches of $P^{-1}(z)$ in $U$ induces a cyclic ordering of faces of the

1459: first group of $\Omega$ and this  

1460: ordering coincides with that induced by the

1461: orientation of $\C\P^1.$

1462: Let $E_a^1=\{e^1_{a_1},e^1_{a_2},\, ... \,, e^1_{a_{f_a}}\}$

1463: (resp. $E_b^2=\{e^2_{b_1},e^2_{b_2},\, ... \,, e^2_{b_{f_b}}\}$)

1464: be the union of edges from the first (resp. 

1465: the second) group

1466: $\Omega$ which are adjacent to the vertex $a$ (resp. $b$).

1467: Let $D$

1468: be the domain from the collection of domains

1469: $\C\P^{1}\setminus E_a^1$ which contains point $b.$

1470: %and let $e_s^1,e_t^1\in E_a$ be the edges which bound $D.$

1471: Once again the Jordan theorem implies that

1472: all the edges from $E_a^1$ are contained in

1473: $\C\P^1\setminus D$ while

1474: $D$ contains $e_h^2\setminus {\infty}$

1475: for all $e_h^2\in E_b^2.$ Taking into account that

1476: for any $k,$ $1\leq k \leq n,$

1477: the edge $e^1_k$ is 

1478: followed by $e^2_k$ this fact implies

1479: that $V(a)$ and $V(b)$ are almost disjointed.

1480: Note that, in contrast to the case when 

1481: $P(a)=P(b),$

1482: now the sets $V(a)$ and $V(b)$ can have a non-empty intersection

1483: consisting of a single element.

1484: 

1485: \subsection{On coefficients of Puiseux expansion of $Q(P^{-1}(z))$}

1486: 

1487: In this subsection we deduce from the monodromy lemma the following

1488: important

1489: property of

1490: the Puiseux expansions \e{ps2}

1491: for pairs $P(z),Q(z)$

1492: satisfying \e{e1}, \e{e2}.

1493: 

1494: 

1495: \bt \l{pu}

1496: Let $P(z),Q(z)\in \C[z]$, $\deg P(z)=n,$ 

1497: $a,b\in \C,$ $a\neq b.$

1498: Suppose that \e{e1} or \e{e2} holds.

1499: Then $u_k=0$ for any $k$ such that $\GCD(k,n)=1.$ 

1500: 

1501: \et

1502: 

1503: \pr 

1504: Suppose first that $P(a)=P(b).$ 

1505: Then lemma \ref{la} implies that 

1506: $u_k = 0$ whenever the number $\varepsilon_n^{k}$ is not

1507: a root of the polynomial \e{e11}.

1508: Let us show that if $\GCD(k,n)=1$ 

1509: then

1510: the equality $r(\varepsilon_n^{k})=0$ is impossible.

1511: Indeed, if $(k,n)=1,$

1512: then $\varepsilon_n^{k}$

1513: is a primitive $n$-th root of unity.

1514: Since

1515: the $n$-th cyclotomic polynomial $\Phi_n(z)$ is irreducible over $\Z$,

1516: the equality $r(\varepsilon_n^{k})=0$ 

1517: implies that $\Phi_n(z)$ divides $r(z)$ in the ring $\Z[z].$

1518: Therefore, the primitive $n$-th root of unity

1519: $\varepsilon_n=exp(2\pi i/n)$ also is

1520: a root of $r(z)$ and hence the equality   

1521: $$\sum_{s=1}^{d_a}\varepsilon_n^{a_s}/d_a =

1522: \sum_{s=1}^{d_b}\varepsilon_n^{b_s}/d_b$$ holds.

1523: The last equality is equivalent to the statement

1524: that the mass centers of the sets $V(a)$ and $V(b)$ 

1525: coincide. But this contradicts to the monodromy lemma. Indeed, 

1526: the mass center of a system of points in $\C$

1527: is inside of the convex envelope of this system and therefore

1528: the mass centers of disjointed sets must be distinct.

1529: 

1530: If $P(a)\neq P(b)$ then, similarly, the inequality $u_k\neq 0$ for

1531: $\GCD(k,n)=1$ implies that

1532: $$\sum_{s=1}^{d_a}\varepsilon_n^{a_s}/d_a =0, \ \ \ \ \ 

1533: \sum_{s=1}^{d_b}\varepsilon_n^{b_s}/d_b=0.$$

1534: But this again contradicts the monodromy lemma. Indeed, the fact that  

1535: the sets $V(a)$ and $V(b)$ are almost disjoint implies that at least 

1536: one

1537: from these sets is contained in an open half plane bounded by a line

1538: passing through the origin and therefore the mass center of this set

1539: is distinct from zero.

1540: 

1541: 

1542: 

1543: \bc

1544: Let $P(z),q(z)\in \C[z]$, $q(z)\neq 0,$ $\deg P(z)=n,$ 

1545: $\deg Q(z)=m,$ 

1546: $a,b\in \C,$ $a\neq b.$ Suppose that \e{1} holds.

1547: Then $\GCD(m,n)>1.$

1548: \ec

1549: 

1550: \pr Since in expansions \e{cn}

1551: the coefficient $v_{-1}$ is distinct from zero,

1552: the coefficient $u_{-m}=v_{-1}^m$ in expansions \e{ps2}

1553: is distinct from zero.

1554: Since \e{1} implies \e{e1} or \e{e2}

1555: by proposition \ref{brc} it follows from

1556: theorem \ref{pu} that $\GCD(m,n)>1.$

1557: 

1558: \vskip 0.2cm

1559: Notice that theorem \ref{pu} agrees with conjecture

1560: \e{cc}. Indeed, if

1561: \be \l{a}

1562: Q(z)=\tilde Q_1(W_1(z))+\tilde Q_1(W_1(z))+ \ ... \ + \tilde Q_r(W_r(z)),

1563: \ee

1564: where $W_1(z),W_2(z),...,W_r(z)$ are (non-trivial) right divisors of

1565: $P(z)$ in the composition algebra, 

1566: $$P(z)=\tilde P_1(W_1(z))=\tilde P_2(W_2(z))=\ ... \ =

1567: \tilde P_r(W_r(z)),$$

1568: %for non-linear

1569: %$W_1(z),W_2(z),...,W_r(z)$$W_1(z),W_2(z),...,W_r(z)$

1570: then

1571: the expansion \e{ps2} has the form

1572: $$Q(P^{-1}(z))=\tilde Q_1(\tilde P_1^{-1}(z))+\tilde Q_2(\tilde

1573: P_2^{-1}(z))+\

1574: ... \ +\tilde Q_r(\tilde P_r^{-1}(z)).$$

1575: Since $\deg \tilde P_j(z) < n,$ $1\leq j \leq r,$ it follows

1576: easily that

1577: $u_k=0$ for any $k$ such that $\GCD(k,n)=1.$

1578: Conjecturally, vice versa, equalities

1579: $u_k=0$ for all $k$ with $\GCD(k,n)=1$ imply

1580: that $Q(z)$ has form \e{a} at least under some additional

1581: assumptions. We plan to discuss this topic in another paper.

1582: 

1583: 

1584: 

1585: 

1586: 

1587: \section{Further description of definite polynomials}

1588: 

1589: 

1590: \subsection{Case when $a$ or $b$ is not a critical

1591: point of $P(z)$}

1592: As a first application of the Puiseux expansions technique 

1593: we provide in this section the following generalization of

1594: corollary \ref{cris}.

1595: 

1596: \bt \l{alo} Let $P(z),q(z)\in \C[z]$, $q(z)\neq 0,$ 

1597: $a,b\in \C,$ $a\neq b.$ Suppose that at least one from 

1598: points $a$ and $b$ is not a critical point of the polynomial $P(z).$

1599: Then

1600: conditions \e{1} and \e{2} are equivalent.

1601: \et 

1602: \pr Since the condition of the theorem is compositionally

1603: stable it follows from lemmas \ref{rl}, \ref{lcomp}

1604: that we only must show that equality \e{eb} holds.

1605: To be definite suppose that the point $a$ is not a critical point of 

1606: $P(z).$

1607: By proposition \ref{brc} either the system

1608: \be \l{pk}

1609: Q(p_{a_1}^{-1}(z))=0, \ \ \ \ \ \ \ \  \sum_{s=1}^{d_{b}}

1610: Q(p_{b_s}^{-1}(z))=0

1611: \ee

1612: or the equality

1613: \be \l{pk2}

1614: Q(p_{a_1}^{-1}(z))=\frac{1}{d_{b}}\sum_{s=1}^{d_{b}}

1615: Q(p_{b_s}^{-1}(z))

1616: \ee

1617: holds. Nevertheless, since the first equation of system \e{pk}

1618: leads to the equality ${q(z)\equiv 0,}$ we must only consider

1619: equation \e{pk2}.

1620: 

1621: Applying lemma \ref{la} we see 

1622: that for any $k$ such that $u_k\neq 0$

1623: the equality  

1624: $$d_b(\varepsilon_n^k)^{a_1}=\sum_{s=1}^{d_{b}} (\varepsilon_n^k)^{b_s} $$ 

1625: holds. 

1626: The triangle inequality implies that

1627: this is possible only if 

1628: $$(\varepsilon_n^k)^{a_1} =(\varepsilon_n^k)^{b_1}=(\varepsilon_n^k)^{b_2}= 

1629: ... =(\varepsilon_n^k)^{b_{d_b}}.$$ 

1630: Therefore,

1631: $$Q(p_{a_1}^{-1}(z))=

1632: Q(p_{b_1}^{-1}(z))=Q(p_{b_2}^{-1}(z))= ... =

1633: Q(p_{b_{d_s}}^{-1}(z))

1634: .$$

1635: 

1636: 

1637: 

1638: 

1639: \subsection{Case when

1640: $\deg P(z)=p^n$}

1641: In this subsection we deduce from theorem \ref{pu} the solution

1642: of the polynomial moment problem in the case

1643: when $\deg P(z)=p^r$ for $p$ prime.

1644: 

1645: \bt \l{prime} Let $P(z),q(z)\in \C[z],$ $q(z)\neq 0,$ 

1646: $a,b\in \C,$ $a\neq b.$

1647: Suppose that $\deg P(z)=p^r,$

1648: where $p$ is a prime number. Then conditions \e{1} and \e{2} are equivalent.

1649: \et

1650: 

1651: \pr Again, since the condition of the theorem is compositionally

1652: stable, it is enough to show that \e{eb} holds. Consider

1653: expansion \e{ps2}. By theorem \ref{pu} the equality

1654: $u_k=0$ holds for any $k$ with $\GCD(k,p^r)=1.$

1655: Show that this fact implies the equality

1656: $$

1657: Q(P^{-1}_j(z))=Q(P^{-1}_{j+p^{r-1}}(z))

1658: $$

1659: for any $j,$ $0\leq j\leq n-1.$

1660: Indeed, we have:

1661: $$ 

1662: Q(P^{-1}_j(z))-Q(P^{-1}_{j+p^{r-1}}(z))

1663: =\sum_{k=-m}^{\infty}

1664: w_kz^{-\frac{k}{n}},

1665: $$

1666: where

1667: $$

1668: w_k=u_k(\varepsilon_{p^r}^{jk}-\varepsilon_{p^r}^{(j+p^{r-1})k}).

1669: $$

1670: If $\GCD(k,p^r)=1$ then $u_k=0$ and hence $w_k=0.$

1671: Otherwise, $k=p\tilde k$ for some $\tilde k\in \Z.$ 

1672: Therefore,

1673: $$\varepsilon_{p^r}^{(j+p^{r-1})k}=

1674: \varepsilon_{p^r}^{jk}\varepsilon_{p^r}^{p^{r}\tilde k}=

1675: \varepsilon_{p^r}^{jk}$$ and hence again

1676: $w_k=0.$

1677: 

1678: 

1679: \subsection{Case when $P(z)$ is indecomposable}

1680: %In this subsection we give, using the obtained results,

1681: Theorems \ref{t1}, \ref{prime} allow us to give 

1682: a short proof of the theorem proved in \cite{pa2}, \cite{pa4}

1683: which describes solutions to \e{1} in case when $P(z)$

1684: is indecomposable that is

1685: can not be represented as a composition

1686: $P(z)=P_1(P_2(z))$ with non-linear polynomials $P_1(z),$

1687: $P_2(z).$

1688: \bt \l{nepr}

1689: Let $P(z),q(z)\in \C[z],$ $q(z)\neq 0,$ 

1690: $a,b\in \C,$ $a\neq b.$

1691: Suppose that $P(z)$ is indecomposable.

1692: Then conditions \e{1} and \e{2} are equivalent. In more details,

1693: $Q(z)$ is a polynomial in $P(z)$ 

1694: and $P(a)=P(b).$

1695: \et

1696: 

1697: \pr 

1698: Once again we only must prove that \e{eb} holds.

1699: Suppose the contrary that is that

1700: all $Q(P^{-1}_i(z)),$ $1\leq i \leq n,$ where $n=\deg P(z)$ are

1701: different; then the monodromy group $G$ of the algebraic

1702: function $Q(P^{-1}(z))$ obtained by the complete analytic continuation

1703: of $Q(P^{-1}_i(z)),$ $1\leq i \leq n,$ 

1704: coincides with that of $P^{-1}(z)$.

1705: Since $P(z)$ is indecomposable, $G$ is primitive

1706: by the Ritt theorem \cite{ri}.

1707: Since for the case when $n=\deg P(z)$ is a prime number the statement follows

1708: from theorem \ref{prime} we can suppose that $n$ is a composite number.

1709: By the Schur theorem (see e.g. \cite{wi}, Th. 25.3)

1710: a primitive permutation group of composite degree $n$

1711: which contains an $n$-cycle is doubly transitive.

1712: 

1713: Recall now the following fact:

1714: roots

1715: $\alpha_i,$ $1\leq i \leq n,$ of an

1716: irreducible algebraic equation over a field $k$ of characteristic zero

1717: with doubly transitive Galois group can not satisfy any

1718: relation $$\sum_{i=1}^{n} c_i\alpha_i=0,\ \ \ \ \ \ c_i\in k,$$ except

1719: the case when $c_1=c_2= ... =c_n$ 

1720: (see \cite{gi}, Proposition 4, or, in the context of algebraic functions, \cite{pa2}, lemma 2).

1721: Since the monodromy group of an algebraic function coincides with

1722: the Galois group of the equation over $\C(z)$

1723: which defines this function, 

1724: it follows that if all $Q(P^{-1}_i(z)),$ $1\leq i \leq n,$ are

1725: different, then equality \e{x} is possible only when

1726: \be \l{ko}

1727: f_1=f_2= ... =f_n.

1728: \ee 

1729: On the other hand, for any non-trivial equation $\phi_s(z)=0$

1730: appeared in theorem \ref{t1}

1731: the equality \e{ko} is impossible by construction.

1732: This contradiction completes the proof. 

1733: 

1734: 

1735: \section{Solution of the polynomial moment problem

1736: for polynomials of degree less than 

1737: \protect

1738: $10$}

1739: 

1740: In this section

1741: %using obtained results combined with some combinatorial

1742: %reasoning involving cacti

1743: we provide a complete solution of 

1744: the polynomial moment problem

1745: for polynomials of degree less than 

1746: $10.$

1747: 

1748: 

1749: For an extended cactus $\tilde \lambda_P$ and a path $\Gamma_{a,b}$ define

1750: {\it the 

1751: skeleton}

1752: $\hat\Gamma_{a,b}$ of $\Gamma_{a,b}$

1753: as follows.

1754: Draw the path

1755: $\Gamma_{a,b}$ separately from the graph $\tilde\lambda_P$ 

1756: and erase all its white vertices.

1757: Number the edges of the obtained graph $\hat\Gamma_{a,b}$ so that

1758: the number of an edge $e_k$ coincides with

1759: the number of the star $S_k$ of $\tilde\lambda_P$

1760: for which $e_k\subset S_k.$   

1761: The number of edges of

1762: $\hat\Gamma_{a,b}$ is called the length

1763: $l(\hat\Gamma_{a,b})$ of $\hat\Gamma_{a,b}.$

1764: For example, the skeleton 

1765: $\hat\Gamma_{a,b}$ of the path $\Gamma_{a,b}$ from Fig. 1

1766: is shown on Fig. 5; here $l(\hat\Gamma_{a,b})=4.$

1767: 

1768: \medskip

1769: \epsfxsize=8truecm

1770: \centerline{\epsffile{2.eps}}

1771: \smallskip

1772: \centerline{Figure 5}

1773: \medskip

1774: 

1775: 

1776: \bt 

1777: Let $P(z),q(z)\in \C[z]$, $q(z)\neq 0,$ 

1778: $a,b\in \C,$ $a\neq b,$ satisfy \e{1}.

1779: Suppose that $\deg P(z)<10.$ Then

1780: either condition \e{2} holds  or

1781: %$\deg P(z)=6$ and

1782: there exist linear

1783: functions $L_1(z),L_2(z)$ such that

1784: $$L_2(P(L_1(z)))=T_6(z), \ \ \ \ L_1^{-1}(a)=-\sqrt{3}/2,

1785: \ \ \ \ L_1^{-1}(b)=\sqrt{3}/2,$$

1786: and $$Q(L_1(z))=A(T_3(z))+B(T_2(z))$$

1787: for some $A(z),B(z)\in \C[z].$

1788: \et

1789: 

1790:  

1791: 

1792: \pr First of all observe that any natural number $n<10$ distinct

1793: from 6

1794: is either a prime number or a degree of a prime number.

1795: Therefore, it follows from theorem \ref{prime}

1796: that

1797: it suffices to consider the case

1798: when $\deg P(z) = 6.$ Furthermore, in view of theorem \ref{alo}

1799: we can suppose that the points $a,b$ are critical points of $P(z).$

1800: Finally notice that in order to prove that condition \e{2}

1801: holds for $P(z),$ $q(z)$ satisfying \e{1} with $\deg P(z)=6$

1802: it is enough to establish equality \e{comp}.

1803: Indeed, if $W(a)\neq W(b)$ in \e{comp} then performing the change of

1804: variable $z\rightarrow W(z)$ we see

1805: that \e{1} holds

1806: for $\tilde P(z), \tilde Q(z), W(a), W(b).$ 

1807: If $\deg W(z)$ equals 3 or 2,

1808: then it follows from theorem 

1809: \ref{nepr} that ${\tilde Q(z)=R(\tilde P(z))}$ for some $R(z)\in \C[z]$ and 

1810: $\tilde P(W(a))=\tilde P(W(b)).$ Therefore, \e{2} holds

1811: with $W(z)=P(z), \tilde Q(z)=R(z).$ On the other hand, if 

1812: $\deg W(z)=6$ in \e{comp} then necessary   

1813: $W(a)= W(b)$ since otherwise $\tilde Q(z)$ would be orthogonal to

1814: all powers of $z$ on the segment $W(a),W(b).$ 

1815: In particular, in view of lemma \ref{lcomp}, we see that

1816: in order to prove that conditions \e{1} and \e{2} are equivalent

1817: it is enough to establish \e{eb}.

1818: 

1819: 

1820: 

1821: Since $\deg P(z)= 6,$ clearly $l(\hat\Gamma_{a,b})\leq 6.$ Moreover,

1822: since the points $a,b$ are critical points of $P(z),$

1823: the valency of the corresponding vertices of $\tilde \lambda_P$ is

1824: at least 2, and, therefore, actually $l(\hat\Gamma_{a,b})\leq 4.$

1825: Consider all possible cases. First of all observe that

1826: the equality $l(\hat\Gamma_{a,b})=1$ is

1827: impossible. Indeed,

1828: in this case theorem \ref{t1} implies that

1829: $Q(P^{-1}_i(z))=0,$ where $i$ is 

1830: the number of the unique edge of $\hat\Gamma_{a,b},$ 

1831: and therefore $q(z)\equiv 0.$

1832: Furthermore, if $l(\hat\Gamma_{a,b})=2$ 

1833: then,

1834: since adjacent vertices of $\hat\Gamma_{a,b}$ have different 

1835: colors,

1836: $\hat\Gamma_{a,b}$ can be of

1837: one from the following two forms shown on Fig. 6.

1838: \vskip 0.2cm

1839: \medskip \epsfxsize=12truecm \centerline{\epsffile{3.eps}} \smallskip

1840: \centerline{Figure 6} \medskip \noindent

1841: In both cases for the middle

1842: vertex $y$ we have $w(y)=1.$

1843: Therefore by theorem \ref{rt} equality \e{comp} holds and

1844: hence conditions \e{1} and \e{2} are equivalent.

1845: Observe, however, that the first configuration shown on Fig. 6 is actually

1846: not realizable since \e{2} implies that $P(a)=P(b).$

1847: 

1848: Consider now the case when $l(\hat\Gamma_{a,b})=3.$ It is not

1849: difficult to see that in this case either again $w(y)=1$ for some color

1850: $y$ or $\hat\Gamma_{a,b}$ has the form shown on Fig. 7.

1851: 

1852: \medskip

1853: \epsfxsize=8truecm

1854: \centerline{\epsffile{4.eps}}

1855: \smallskip

1856: \centerline{Figure 7}

1857: \medskip

1858: \noindent

1859: Let us examine the last case.

1860: Since for the skeleton shown on Fig. 7 

1861: we have $P(a)\neq P(b),$ it follows

1862: from proposition \ref{brc} that

1863: system \ref{e2} holds.

1864: Furthermore, the equality $\deg P(z)= 6$ implies that for at least one

1865: point $s,$ $s\in \{a,b\},$

1866: the following two conditions are satisfied: the multiplicity of the 

1867: vertex $s$ of the graph $\tilde \lambda_P$

1868: equals 2 and the connectivity component of $\tilde \lambda_P\setminus s$

1869: which does not contain $\Gamma_{a,b}$ consists of a unique star.

1870: To be definite suppose that $s=a.$

1871: Then, in notation of \ref{cacti},

1872: the first condition implies that \be \l{gra}

1873: \sum_{s=1}^{d_{a}}

1874: Q(P_{a_s}^{-1}(z))=Q(P^{-1}_{i_1}(z))+Q(P^{-1}_{g_x(i_1)}(z))=0 \ee and 

1875: the second one that

1876: $g_y(g_x(i_1))=g_x(i_1).$ Therefore, the analytic continuation of

1877: \eqref{gra} along the loop $l_y$ leads to the equality \be \l{gra2}

1878: Q(P^{-1}_{i_2}(z))+Q(P^{-1}_{g_x(i_1)}(z))=0. \ee Now equalities

1879: \eqref{gra},\eqref{gra2} imply that 

1880: $Q(P^{-1}_{i_1}(z))=Q(P^{-1}_{i_2}(z))$

1881: %Since this equality implies \e{zae}

1882: and we conclude as above that the configuration

1883: shown on Fig. 7 is not realizable.

1884: 

1885: Consider finally the case when $l(\hat\Gamma_{a,b})=4.$

1886: Since $\hat\Gamma_{a,b}$ has 5 vertices, 

1887: either $w(y)=1$ for some color

1888: $y$ or $\hat\Gamma_{a,b}$ is two-colored. In the last case

1889: $\hat\Gamma_{a,b}$ has the form shown on Fig. 8

1890: \vskip 0.2cm

1891: \medskip

1892: \epsfxsize=8truecm

1893: \centerline{\epsffile{z.eps}}

1894: \smallskip

1895: \centerline{Figure 8}

1896: \medskip

1897: \noindent

1898: and the corresponding cactus $\tilde \lambda_P$ is a 6-chain

1899: (the cactus with 6 stars of the

1900: maximal diameter).

1901: Furthermore, since $\deg P(z)=6,$ it follows from

1902: the Riemann-Hurwitz formula that

1903: $$\sum_{z\in \C\P^1}({\rm mult}_zP-1)=10.$$ Since

1904: ${\rm mult}_{\infty}P-1=5$ and

1905: the combinatorics of $\tilde \lambda_P$ imply that 

1906: $$

1907: \sum_{P(z)=c_x}({\rm mult}_zP-1)=3, \ \ \ \ \ \ \ \

1908: \sum_{P(z)=c_y}({\rm mult}_zP-1)=2,$$

1909: we conclude that $P(z)$ has only two finite critical values $c_x,c_y.$

1910: 

1911: It follows from the Riemann existence theorem (see e.g. \cite{cac2})

1912: that a complex polynomial with given

1913: critical values 

1914: %$c_x,c_y$ 

1915: is defined by its cactus up to a 

1916: linear change of variable.

1917: On the other hand, it is easy to see using the

1918: formula $T_n(\cos \phi)=\cos n\phi$ that 

1919: $T_n(z)$ has only two critical values $-1,1$ and that

1920: all critical points of $T_n(z)$ are simple,

1921: Therefore, the corresponding cactus is a chain.

1922: In particular, for $P(z)=T_6(z)$ the corresponding cactus realized

1923: as the preimage

1924: of the segment $[-1,1]$ (considered as a star connecting

1925: $0$ with points $1$ and $-1$) has the form shown on Fig. 9

1926: (white vertices are omitted).

1927: 

1928: %\begin{figure}

1929: \vskip 0.5cm

1930: \medskip

1931: \epsfxsize=9.5truecm

1932: \centerline{\epsffile{zz.eps}}

1933: %\smallskip

1934: \centerline{Figure 9}

1935: \medskip

1936: \noindent

1937: %\end{figure}

1938: Therefore, if we choose linear functions

1939: $L_1(z), L_2(z)$ such that:

1940: $$L_1^{-1}(a)=-\sqrt{3}/2, \ \ \ L_1^{-1}(b)=\sqrt{3}/2,

1941: \ \ \ 

1942: %critical values of the polynomial

1943: %$L_2(P(L_1(z)))$ are $-1,1,$ and

1944: %$L_2(P(-\sqrt{3}/2))=L_2(P(\sqrt{3}/2))=-1,$

1945: L_2(c_x)=-1,\ \ \ L_2(c_y)=1,$$

1946: the polynomial $L_2(P(L_1(z)))$ will be equal $T_6(z).$

1947: 

1948: Finally, the last assertion of the theorem follows from

1949: the main result of the paper \cite{pa3} where 

1950: all solutions to \eqref{1} for $P(z)=T_n(z)$

1951: were described.

1952: 

1953: 

1954: 

1955: \bibliographystyle{amsplain}

1956: \begin{thebibliography}{10}

1957: 

1958: 

1959: \bibitem {bby}  M. Blinov, M. Briskin, Y. Yomdin, {\it Local center conditions

1960: for a polynomial Abel equation and cyclicity

1961: of its zero solution}, Journal d'Anal. Math, to appear.

1962:   

1963: \bibitem {bfy1} M. Briskin, J.-P. Francoise, Y. Yomdin,

1964: \textit{Une approche au probleme du centre-foyer de Poincare,}

1965: C. R. Acad. Sci., Paris, Ser. I, Math. 326, No.11, 1295-1298 (1998).

1966: 

1967: 

1968: \bibitem {bfy2} M. Briskin, J.-P.

1969: Francoise, Y. Yomdin,

1970: \textit{Center conditions, compositions of polynomials and moments on

1971: algebraic curve}, Ergodic Theory Dyn. Syst. \textbf{19}, no 5,

1972: 1201-1220 (1999).

1973: 

1974: 

1975: \bibitem {bfy3} M. Briskin, J.-P. Francoise, Y. Yomdin,

1976: \textit{Center condition II: Parametric and model center problems}, Isr.

1977: J. Math. \textbf{118}, 61-82 (2000).

1978: 

1979: 

1980: \bibitem {bfy4} M. Briskin, J.-P. Francoise, Y. Yomdin,

1981: \textit{Center condition III: Parametric and model center problems}, Isr.

1982: J. Math. \textbf{118}, 83-108 (2000).

1983: 

1984: \bibitem {bfy5} M. Briskin, J.-P. Francoise, Y. Yomdin,

1985: \textit{Generalized moments, center-focus conditions and compositions of

1986: polynomials,} in ``Operator theory, system theory and related topics",   

1987: Oper. Theory Adv. Appl.,  \textbf{123}, 161--185 (2001).

1988: 

1989:  

1990: \bibitem {bry} M. Briskin, N. Roytvarf, Y. Yomdin,

1991: {\it Center conditions at infinity for Abel differential equation},

1992: preprint, 2003.

1993: 

1994: 

1995: \bibitem {by} M. Briskin, Y. Yomdin,

1996: {\it Tangential Hilbert problem for Abel equation},

1997: Mosc. Math. J., to appear.

1998: 

1999: \bibitem {c} C. Christopher, \textit{Abel equations: composition

2000: conjectures and

2001: the model problem}, Bull. Lond. Math. Soc. \textbf{32}, No.3,

2002: 332-338 (2000).

2003: 

2004: 

2005: \bibitem {cac1} M. El Marraki, N. Hanusse, J. Zipperer, A. Zvonkin,

2006: \textit{Cacti, braids and complex polynomials}, Sim. Lothar. Combin.

2007: \textbf{37}, 

2008: Art. B37b, 36 pp, 1996.

2009:   

2010:  

2011: \bibitem {gi} K. Girstmair, {\it Linear dependence of zeros 

2012: of

2013: polynomials and construction of primitive elements}, Manuscripta 

2014: Math. 39 (1982), no. 1, 81--97. 

2015: 

2016: \bibitem{cac2} S. Lando, A. Zvonkin, \textit{Graphs on surfaces and their 

2017: applications}, Encyclopaedia of Mathematical 

2018: Sciences, \textbf{141}, Low-Dimensional Topology, II,

2019: Springer-Verlag, Berlin, 2004. 

2020: 

2021: 

2022: \bibitem {mus} N.I. Muskhelishvili, {\it Singular integral equations.

2023: Boundary problems of function theory and their application to mathematical

2024: physics,} Dover Publications, Inc., New York, 1992. 447 pp.

2025: 

2026: \bibitem {pa1} F. Pakovich, {\it A counterexample to the ``Composition 

2027: Conjecture''},

2028: Proc. AMS, 130, no. 12 (2002), 3747-3749.

2029:  

2030: \bibitem {pa2} F. Pakovich, {\it On the polynomial moment problem}, Math. 

2031: Research Letters

2032: {\bf 10}, (2003), 401-410.

2033:  

2034: \bibitem {pa3} F. Pakovich, {\it On polynomials orthogonal to all powers

2035: of a Chebyshev polynomial on a segment}, Isr. J. Math, to appear.

2036:  

2037: \bibitem {pa4} F. Pakovich, {\it Polynomial moment problem},

2038: Addendum           

2039: to the paper {\it Center Problem for Abel Equation, Compositions of 

2040: Functions,

2041: and Moment Conditions} by Y. Yomdin, Mosc. Math. J.,

2042: \textbf{3} (2003), no. 3, 1167-1195.

2043: 

2044: \bibitem {pry} F. Pakovich, N. Roytvarf and Y. Yomdin. {\it Cauchy type 

2045: integrals

2046: of Algebraic functions,} Isr. J. Math, to appear.

2047: 

2048: 

2049: \bibitem {ri} J. Ritt, \textit{Prime and composite polynomials,} Trans.

2050: Amer. Math. Soc.  \textbf{23}, no. 1, 51--66 (1922).

2051: 

2052: \bibitem {ro} N. Roytvarf, \textit{Generalized moments, composition of

2053: polynomials and Bernstein classes}, in "Entire functions in modern

2054: analysis. B.Ya. Levin memorial volume", Isr. Math. Conf. Proc.

2055: \textbf{15}, 339-355 (2001).

2056: 

2057: 

2058: \bibitem {sch} A. Schinzel, \textit{Polynomials with special regard to

2059: reducibility}, Encyclopedia of Mathematics and Its Applications

2060: \textbf{77}, Cambridge University Press, 2000.

2061: 

2062: \bibitem {wi} H. Wielandt, \textit{Finite permutation groups,}

2063: New York and London: Academic Press, 1964.

2064: 

2065: \bibitem {y1}  Y. Yomdin, \textit{Center Problem for Abel Equation,

2066: Compositions of Functions, and Moment Conditions},

2067: Mosc. Math. J., \textbf{3} (2003), no. 3, 1167-1195.

2068: 

2069: \end{thebibliography}

2070: 

2071: 

2072: 

2073: 

2074: 

2075: 

2076: \end{document}

2077: