1: \documentclass[12pt]{article}
2: \usepackage{epsfig}
3: \def\const{{\mathrm{const}}}
4: \def\id{{\mathrm{id}}}
5: \def\R{\mathbf{R}}
6: \def\cross{\times}
7: \def\nu{\circ}
8: \def\C{{\mathbf{C}}}
9: \def\bC{{\mathbf{\overline C}}}
10: \def\mod{\mathrm{mod}\;}
11: \def\Ai{\mathrm{Ai}}
12: \title{Nevanlinna functions with real zeros}
13: \author{A. Eremenko\thanks{Supported by NSF grants
14: DMS-0100512 and DMS-0244547.}\,
15: and S. Merenkov\thanks{
16: Supported by NSF grants DMS-0400636 and DMS-0244547.}
17: }
18: \date{January 2005}
19: \begin{document}
20: \maketitle
21:
22: \section{Introduction}
23:
24: Consider a differential equation
25: \begin{equation}
26: \label{DE}
27: w''+Pw=0,
28: \end{equation}
29: where $P$ is a polynomial of the independent variable. Every solution
30: $w$ of this equation is an entire function.
31: We are interested in solutions $w$ whose all roots are real.
32: If (\ref{DE}) has two linearly independent solutions with
33: this property then $\deg P=0$, see \cite{G2,HR}.
34: Here we study equations (\ref{DE}) that have
35: at least one solution with all roots real.
36:
37: The question of describing equations (\ref{DE}) with this property
38: was proposed by S. Hellerstein
39: and J. Rossi in \cite[Probl. 2.71]{H}. According to \cite{G1},
40: up to trivial changes of variables,
41: only the following four examples were known until recently.
42: \begin{itemize}
43: \item
44: $\deg P=0.$ If $k$ is real, all solutions of $w''+k^2w=0$
45: are trigonometric functions.
46: \item
47: $\deg P=1$. The Airy equation $w''-zw=0$,
48: has a solution $\Ai(z)$ whose
49: roots lie on the negative ray.
50: \item
51: $\deg P$ is even and $w=p\exp q$,
52: where $p$ and $q$ are polynomials, and all
53: roots of $p$ are real. In this case, the set
54: of roots of $w$ is evidently
55: finite.
56: For example, if $P(z)=1-z^2+2n,$ where $n$ is a positive integer,
57: then
58: the equation (\ref{DE}) has solutions $w=H_n(z)\exp(-z^2/2),$
59: where $H_n$ are Hermite's polynomials whose roots are all real.
60:
61: \item
62: $P(z)=az^4+bz^2-c$. Gundersen \cite{G1} proved that
63: for every $a>0$ and $b\geq 0$ one can find
64: an infinite set of real numbers $c$, such that some solution
65: of $(\ref{DE})$
66: has infinitely many roots,
67: almost all of them real. When $b=0$ this result was earlier obtained
68: by Titchmarsh \cite[p. 172]{T}.
69: \end{itemize}
70:
71: Here and in what follows ``almost all'' means ``all except finitely many''.
72: Recently Kwang C. Shin \cite{shin} proved a similar result for
73: a degree $3$ polynomial:
74: \begin{itemize}
75:
76: \item
77: For every real $a$ and $b\leq 0$ there exist an infinite set
78: of positive numbers $c$ such that the equation
79: $w''+(z^3+az^2+bz-c)w=0$ has a solution with infinitely many roots,
80: almost all of them real.
81: \end{itemize}
82:
83: On the other hand, Gundersen \cite{G2} proved the following theorems:
84: \vspace{.1in}
85:
86: \noindent
87: {\bf Theorem A} {\em If $d=\deg P\equiv 2\; (\mod\, 4),$
88: and $w$ is a solution of $(\ref{DE})$, with almost all roots real,
89: then $w$
90: has only
91: finitely many roots.}
92: \vspace{.1in}
93:
94: \noindent
95: {\bf Theorem B} {\em If $(\ref{DE})$ possesses a solution $w$
96: with infinitely many real zeros,
97: then $P$ is a real polynomial, and $w$
98: is proportional to a real function.}
99: \vspace{.1in}
100:
101: We also mention a result of Rossi and Wang \cite{RW} that
102: if (\ref{DE}) has a solution with infinitely many roots, all of
103: them real, then at least half of all roots of $P$ are non-real.
104: In view of Theorem B we restrict from now on to the case
105: of real polynomials $P$ in (\ref{DE}). Our results are:
106: \vspace{.1in}
107:
108: \noindent
109: {\bf Theorem 1} {\em For every $d$,
110: there exist $w$ satisfying $(\ref{DE})$ with $\deg P=d$ and such that
111: all roots of $w$ are real.
112: For every positive $d$ divisible by $4$, there exist $w$ with infinitely
113: many roots, all of them real, as well as $w$ with any prescribed finite
114: number of roots, all of them real.}
115: \vspace{.1in}
116:
117: \noindent
118: {\bf Theorem 2} {\em Let $w$ be a solution of
119: the equation $(\ref{DE})$
120: whose all roots are real, and $d=\deg P$. Then:
121: \newline
122: (a) For $d\equiv 0\,(\mod 4)$ the set of roots of
123: $w$ is either finite or unbounded from above and below (as a subset of the real axis).
124: \newline
125: (b) For odd $d$ the roots of $w$ lie on a ray, and there
126: are infinitely many of them.}
127: \vspace{.1in}
128:
129: Theorem 2 can be generalized to the case that almost all roots
130: of $w$ are real.
131:
132: In comparison with the existence results of Titchmarsh, Gundersen and Shin
133: mentioned above, our Theorem 1 gives more precise information on the
134: zeros $w$: they are all real. On the other hand we
135: can tell less about the polynomial $P$.
136:
137: Our proofs are based on a geometric
138: characterization of meromorphic functions of
139: the form
140: $f=w_1/w_2$, where $w_1$ and $w_2$ are linearly independent solutions of
141: (\ref{DE}), due to F. and R. Nevanlinna \cite{Fne,RNe1}, which will be
142: explained in the next section.
143:
144: \section{A class of meromorphic functions}
145:
146: We associate with (\ref{DE}) another differential equation
147: \begin{equation}
148: \label{SDE}
149: \frac{f'''}{f'}-\frac{3}{2}\left(\frac{f''}{f'}\right)^2=2P.
150: \end{equation}
151: The expression in the left hand side of (\ref{SDE}) is called the
152: {\em Schwarzian derivative} of $f$. The following well-known fact
153: is proved by simple formal computation.
154: \vspace{.1in}
155:
156: \noindent
157: {\bf Proposition 1} {\em The relation $f=w_1/w_2$ gives a
158: bijective correspondence
159: between solutions $f$ of $(\ref{SDE})$ and classes of proportionality of
160: pairs $(w_1,w_2)$ of
161: linearly independent solutions of $(\ref{DE})$.}
162: \vspace{.1in}
163:
164: Thus when $P$ is a polynomial, all solutions of (\ref{SDE}) are
165: meromorphic in the complex plane, and they are all obtained from each other
166: by post-composition with a fractional-linear transformation.
167: We call solutions of equations (\ref{SDE}) with polynomial right hand side
168: {\em Nevanlinna functions}.
169: Equation (\ref{SDE}) has a real solution if and only if $P$ is real.
170:
171: It is easy to see that meromorphic functions $f$ satisfying (\ref{SDE}) are
172: local homeomorphisms. In other words, $f'(z)\neq 0$ and all poles are
173: simple.
174:
175: F. and R. Nevanlinna gave a topological characterization
176: of all meromorphic functions $f$
177: which may occur as solutions of (\ref{SDE}). To formulate their result
178: we recall several definitions.
179:
180: A {\em surface} is a connected Hausdorff topological manifold of
181: dimension $2$ with countable base.
182:
183: A continuous map $\pi:X\to Y$ of surfaces is called
184: {\em topologically holomorphic} if it is open and discrete.
185: According to a theorem of Stoilov \cite{S} this is equivalent to
186: the following property. For every $x\in X$, there is a positive integer $k$ and
187: complex local coordinates $z$ and $w$ in neighborhoods of $x$ and $\pi(x)$,
188: such that $z(x)=0$ and the map $\pi$
189: has the form $w=z^k$ in these coordinates. The integer $k=k(x)$ is called
190: the
191: {\em local degree} of $\pi$ at the point $x$. So $\pi$ is
192: a local homeomorphism if and only if $n=1$ for every $x\in X$.
193:
194: A pair $(X,\pi)$ where $X$ is a surface and $\pi:X\to\bC$ a topologically
195: holomorphic map
196: is called {\em a surface spread over the sphere}
197: (\"Uberlagerungsfl\"ache in German).
198: Two such pairs $(X_1,\pi_1)$
199: and $(X_2,\pi_2)$ are considered
200: equivalent if there is a homeomorphism $h:X_1\to X_2$ such that
201: $\pi_2=\pi_1\circ h$.
202: So, strictly speaking, a surface spread over the sphere
203: is an equivalence class of such pairs.
204:
205: If $f:D(R)\to\bC$ is a meromorphic function in some
206: disc $D(R)=\{ z:|z|<R\},\; R\leq\infty$ then $(D(R),f)$ defines a
207: surface spread over the sphere.
208: We will call the equivalence class of $(D(R),f)$ the
209: {\em surface
210: associated with $f$}. It is the same as the Riemann surface of $f^{-1}$,
211: as it is defined on \cite[p. 288]{Ah}, completed with algebraic branch
212: points as in \cite[p. 300]{Ah}.
213:
214: In the opposite direction, suppose that $(X,\pi)$ is a surface spread over
215: the sphere. Then there exists a unique conformal structure on $X$ which makes
216: $\pi$ holomorphic. If $X$ is open and simply connected, the Uniformization
217: Theorem says that there exists a conformal homeomorphism
218: $\phi:D(R)\to X$, where $R=1$ or $\infty$.
219: This $\phi$ is defined up to a conformal automorphism of $D(R)$.
220: The function $f=\pi\circ\phi$ is meromorphic in $D(R)$, and $(X,\pi)$
221: is (a representative of) the surface associated with $f$.
222:
223: If $R=\infty$ we say that $(X,\pi)$ is of {\em parabolic type}.
224:
225: We consider surfaces spread over the sphere $(X,\pi)$ where
226: $X$ is open and simply connected\footnote{That is homeomorphic to the plane.},
227: $\pi$ a local homeomorphism, and subject to additional conditions below.
228:
229: Suppose that for some finite set $A\subset\bC$ the
230: restriction
231: \begin{equation}
232: \label{S}
233: \pi:X\backslash\pi^{-1}(A)\to\bC\backslash A\quad\mbox{is a covering map}.
234: \end{equation}
235: Fix an open topological disc
236: $D\subset \bC$ containing exactly one point $a$ of the set $A$.
237: If $V$ is a connected component
238: of $\pi^{-1}(D\backslash \{ a\})$ then
239: the restriction
240: \begin{equation}
241: \label{V}
242: \pi:V\to D\backslash \{ a\}
243: \end{equation}
244: is a covering of a ring, and its degree $k$
245: does not depend on the choice of the disc $D$.
246: The following cases are possible.
247: \newline
248: a) $k=\infty$. Then (\ref{V}) is a universal covering, $V$ is simply
249: connected and its boundary consists of a single simple curve in $X$
250: tending to ``infinity'' in both directions. In this case we say that $V$
251: defines a {\em logarithmic singularity} over $a$.
252: Notice that the number of logarithmic singularities over $a$
253: is independent of the choice of $D$.
254: \newline
255: b) If $k<\infty$ and there exists a point $x\in X$ such that
256: ${\widetilde{V}}=V\cup\{ x\}$ is an open topological disc, then
257: $\pi:{\widetilde{V}}\to D$ is a ramified covering and
258: has local degree $k$ at $x$. As we assume that $\pi$ is a local
259: homeomorphism, only $k=1$ is possible, so
260: $\pi:{\widetilde{V}}\to D$ is a homeomorphism.
261: \newline
262: c) If $k<\infty$ but there is no point $x\in X$ such that $V\cup\{ x\}$
263: is an open disc, then we can add to $X$ such an ``ideal point'' and define
264: the topology on ${\widetilde{X}}=X\cup\{ x\}$ so that it remains a surface.
265: Evidently ${\widetilde{X}}$ is a sphere, and our local homeomorphism
266: extends to a topologically holomorphic map between spheres whose
267: local degree equals one everywhere except possibly one point.
268: It easily follows that the local degree equals one everywhere,
269: the extended map is a homeomorphism. This implies that our original
270: map $\pi$ is an embedding.
271:
272: So in any case the degree of the map (\ref{V})
273: can be only $1$ or $\infty$.
274: \vspace{.1in}
275:
276: \noindent
277: {\bf Definition}
278: We say that $(X,\pi)$ is an {\em N-surface} if $X$ is open and simply
279: connected,
280: $\pi$ is a local homeomorphism, condition (\ref{S}) is
281: satisfied, and there
282: are only finitely many logarithmic singularities.
283: \vspace{.1in}
284:
285: A simple topological argument shows that aside from the case when
286: $\pi$ is an embedding, the number of logarithmic singularities is at least two.
287: All cases with two logarithmic singularities can be reduced by
288: a fractional-linear transformation of $\bC$ to the case
289: $\exp:\C\to\bC$.
290:
291: The name N-surface is chosen in honor of F. and R. Nevanlinna.
292: The complete official name of this object would be ``An open simply
293: connected surface spread over the sphere without algebraic branch points
294: and with finitely many logarithmic singularities''.
295: \vspace{.1in}
296:
297: \noindent
298: {\bf Theorem C} (i) {\em Every N-surface is of parabolic type, that is its
299: associated functions are meromorphic in the plane $\C$.}
300: \newline
301: (ii) {\em If an N-surface $(X,\pi)$ has $n\geq 2$ logarithmic singularities then the
302: associated meromorphic functions $f=\pi\circ\phi$ satisfy a differential
303: equation $(\ref{SDE})$ in which $\deg P=n-2.$}
304: \newline
305: (iii) {\em For every polynomial $P$, every solution $f$ of $(\ref{SDE})$ is
306: a meromorphic function in the plane whose associated surface
307: is an N-surface with $n=\deg P+2$ logarithmic singularities.}
308: \vspace{.1in}
309:
310: Two meromorphic functions $f_1$ and $f_2$ are called
311: equivalent if $f_1(z)=f_2(az+b)$ with $a\neq 0$.
312: Theorem C establishes a
313: bijective correspondence between equivalence
314: classes of Nevanlinna functions and N-surfaces.
315:
316: In this paper we use only statements (i) and (ii) of Theorem C.
317: The connection between N-surfaces and differential equations
318: was apparently discovered by F. Nevanlinna who proved (iii) in
319: \cite{Fne}.
320: Statements (i) and (ii) were proved for the first time by
321: R. Nevanlinna in \cite{RNe1}.
322: Then Ahlfors
323: \cite{Ah1} gave an alternative proof based on completely different ideas.
324: A modern version of this second proof uses quasiconformal mappings
325: \cite{GO}. This modern proof is reproduced in \cite{DV}.
326: All these authors were primarily interested in the theory of
327: meromorphic functions, and used differential equations as a tool.
328: Apparently, the only application of Theorem C to differential equations
329: is due to Sibuya \cite{Sib} who deduced from it the existence of
330: equations (\ref{DE}) with prescribed Stokes multipliers.
331:
332: In view of Theorem C, to obtain our results, we only need to
333: single out those N-surfaces that are associated with
334: real meromorphic functions with real zeros.
335:
336: \section{Speiser graphs}
337:
338: We recall a classical tool for explicit construction and visualization
339: of N-surfaces.
340: It actually applies to all surfaces spread over the
341: sphere that satisfy (\ref{S}).
342: First we
343: suppose that a surface spread over the sphere $(X,\pi)$ satisfying (\ref{S}) is given, and $A=\{ a_1,\ldots,a_q\}$
344: is the set in (\ref{S}). We call elements of $A$ {\em base points}.
345: Consider a {\em base curve} that is an oriented Jordan
346: curve $\Gamma$ passing
347: through $a_1,\ldots,a_q$. Choosing a base curve defines a cyclic order
348: on $A$, and we assume that the enumeration is consistent with this
349: cyclic order and interpret the subscripts as remainders modulo $q$.
350:
351: The base curve $\Gamma$ divides the Riemann sphere $\bC$ into two
352: regions which we denote $D^\cross$ and $D^\circ$, so that
353: when $\Gamma$ is traced according to its orientation, the region $D^\cross$ is
354: on the left. We choose points
355: $\times\in D^\cross$ and $\circ\in D^\circ$, and connect these
356: two points by $q$ disjoint simple arcs $L_j$ so that each $L_j$ intersects
357: $\Gamma$ at exactly one point, and this point belongs to the arc
358: $(a_j,a_{j+1})\subset\Gamma$.
359: We obtain an {\em embedded graph} $L\subset\bC$ having two
360: vertices $\cross$ and $\circ$ and $q$ edges $L_j$.
361: This embedded graph defines a cell decomposition of the sphere,
362: whose $2$-cells (faces) are components of the complement of $L$,
363: $1$-cells (edges) are the open arcs $L_j$ and $0$-cells (vertices)
364: are the points $\cross$ and $\circ$.
365: Each face contains exactly one base point.
366:
367: The preimage of this cell decomposition under $\pi$ is a cell
368: decomposition of $X$, because as we saw in the previous section,
369: each component of the preimage of a cell is a cell of the same dimension.
370: The $1$-skeleton $S=\pi^{-1}(L)\subset X$
371: is a connected
372: properly embedded graph in $X$.
373: As $S$ completely defines the cell decomposition, we will permit ourselves to
374: follow the tradition and speak of this graph instead of the cell decomposition,
375: and use such expressions as
376: ``faces of $S$'' meaning the faces of the cell decomposition.
377:
378: We label vertices of $S$ by $\cross$ and $\circ$, according to their images
379: under $\pi$, and similarly label the faces by the base points $a_j$.
380: Our labeled graph $S$ (or more precisely, the labeled cell decomposition)
381: has the following properties:
382: \vspace{.1in}
383:
384: \noindent
385: 1. Every edge connects a $\cross$-vertex
386: to a $\circ$ vertex.
387: \vspace{.1in}
388:
389: \noindent
390: 2. Every vertex belongs to the boundaries
391: of exactly $q$ faces having all $q$ different labels.
392: \vspace{.1in}
393:
394: \noindent
395: 3. The face labels have cyclic order
396: $a_1,\ldots,a_q$ anticlockwise around each $\cross$-vertex, and the opposite
397: cyclic order around each $\circ$-vertex.
398: \vspace{.1in}
399:
400: The labeled graph $S$ is called the {\em Speiser
401: graph} or the {\em line complex} of the surface spread over the sphere
402: $(X,\pi)$.
403:
404: A face of $S$ is called {\em bounded} if its boundary consists of finitely
405: many edges and vertices. It follows from property 1 that the numbers
406: of edges and vertices on the boundary of a bounded
407: face are equal and even.
408: If $a$ is a base point, all solutions of the equation $\pi(x)=a$
409: belong to the bounded faces labeled by $a$,
410: and each such face contains exactly
411: one solution.
412: If $k$ is the local degree of $\pi$ at this point $x$ then the face
413: is a $2k$-gon, that is bounded by $2k$ edges and $2k$ vertices.
414:
415: Suppose now that $X$ is a surface and a labeled cell decomposition of $X$
416: with $1$-skeleton $S$
417: is given such that
418: 1, 2 and 3 hold.
419: If we choose a set $A\subset\bC$ of $q$ points and a
420: curve $\Gamma\subset\bC$ passing through the
421: points of the set $A$,
422: and define
423: $L$ as above, then there exists a
424: topologically holomorphic map $\pi$ such that
425: $S=\pi^{-1}(L)$. This map $\pi$
426: is unique up to pre-composition with a homeomorphism
427: of $X$.
428: A verification of this statement is contained in
429: \cite{GO}. We recall the construction.
430:
431: The labels of faces define labels of edges: an edge is labeled by $j$
432: if it belongs to the common boundary of faces with labels $a_j$ and $a_{j+1}$.
433: This defines a map of the $1$-skeleton to the $1$-skeleton $L$
434: of the cell decomposition of the sphere: each edge of $S$ labeled by $j$
435: is mapped onto $L_j$ homeomorphically, and such that the orientation
436: is consistent with the vertex
437: labeling. It is easy to see that this map is a covering $S\to L$.
438: The boundary of each face covers a topological circle formed by two
439: adjacent
440: edges and two vertices of $L$. Such map extends to a ramified covering
441: between faces, ramified only over the base points (unramified for
442: $2$-gonal or unbounded faces).
443:
444: For a given cyclically ordered set $A$ and a surface $X$,
445: the correspondence between topologically holomorphic maps $\pi$
446: and Speiser graphs
447: is not canonical: it depends on the choice of the base curve $\Gamma$.
448: (It is the isotopy class of $\Gamma$ with fixed set $A$ that matters).
449:
450: It is easy to single out those Speiser graphs that correspond to N-surfaces:
451: the ambient surface $X$ is open and simply connected, and two
452: additional properties hold:
453: \vspace{.1in}
454:
455: \noindent
456: 4. Each face has either two or infinitely many boundary edges.
457: \vspace{.1in}
458:
459: \noindent
460: 5. The set of unbounded faces is finite.
461: \vspace{.1in}
462:
463: Property 4 corresponds to the assumption that $\pi$ is a local homeomorphism,
464: and property 5 follows from the fact that unbounded faces correspond to
465: ``logarithmic singularities'' that is to the components $V$ in (\ref{V})
466: where the covering has infinite degree.
467:
468: So we have
469: \vspace{.1in}
470:
471: \noindent
472: {\bf Proposition 2} {\em Let $(X,\pi)$ be an N-surface, and $a$ a basis point.
473: Then each solution of the equation $\pi(x)=a$ is contained in a face
474: which is a $2$-gon, and is labeled by $a$. Each such face contains exactly
475: one solution of this equation.}
476: \vspace{.1in}
477:
478: For each Speiser graph $S$ corresponding to an N-surface, we
479: define a new graph $T(S)$ with the same vertices:
480: two vertices are connected by a single edge in $T$ if they
481: are connected by at least one edge in $S$.
482: Thus $T(S)$ is obtained from $S$ by dropping multiple edges.
483: Property 4 of $S$ implies that $T$ is a tree. Faces of $T(S)$ are
484: exactly the unbounded faces of $S$.
485: We assume that vertices and unbounded faces inherit their labels from
486: $S$.
487:
488: The tree $T(S)$ is properly embedded in $X$. Each vertex
489: has at least two and at most $q$ adjacent edges in $T(S)$
490: and the cyclic order of face labels around the $\cross$-vertices
491: of $T(S)$ is the same as in $S$.
492:
493: Suppose that $S$ has more than two unbounded faces. Then
494: the tree $T(S)$ has $n$ maximal (by inclusion) infinite ``branches'' having
495: all vertices of degree $2$, of the
496: form
497: $$-\circ-\cross-\circ-\cross-\circ-\ldots\quad\mbox{or}\quad
498: -\cross-\circ-\cross-\circ-\cross-\ldots$$
499: where $n$ is the number of logarithmic singularities of $(X,\pi)$.
500: Such branches are called {\em logarithmic ends}.
501: A tree $T(S)$ is the union of its logarithmic ends and a finite
502: subtree.
503: %Let us fix the anticlockwise cyclic order on logarithmic ends or,
504: %which is equivalent, on the faces of $T$. Then we obtain a cyclic sequence
505: %$b_1,\ldots,b_n$ of labels of the faces of $T$.
506: %(This order is in general different from the cyclic order of the set
507: %$A=\{ a_1,\ldots,a_q\}$, and $n$ is in general different from $q$,
508: %but each $b_j\in A$).
509:
510: \section{Symmetric Speiser graphs}
511:
512: A {\em symmetric} surface spread over the sphere is defined as
513: a triple $(X,\pi,s)$, where
514: $s:X\to X$ is a homeomorphism such that $s\circ s=\id$
515: and $\pi\circ s(x)=\overline{\pi(x)}$,
516: and the bar denotes complex conjugation. It is clear that such $s$ is
517: an anticonformal homeomorphism of $X$. If $(X,\pi)$ is
518: of parabolic type, and $\phi:\C\to X$ a conformal homeomorphism
519: then $\phi^{-1}\circ s\circ\phi$ is an anticonformal involution of the
520: complex plane. Each such involution is conjugate to $z\mapsto\overline{z}$
521: by a conformal automorphism of $\C$. So for a symmetric surface
522: spread over the sphere there exists a uniformizing map $\phi$ with
523: the property $\phi(\overline{z})=s\circ\phi(z),\, z\in\C$.
524: The set of fixed points of $s$ is called the {\em axis} (of symmetry);
525: it is the image of the real line under $\phi$.
526: %Two symmetric surfaces spread over the sphere $(X_1,\pi_1,s_1)$ and
527: %$(X_2,\pi_2,s_2)$ are called equivalent if there exists a homeomorphism
528: %$h:X_1\to X_2$ such that $\pi_2=\pi_1\circ h$ and $s_2=h\circ s_1\circ h^{-1}$.
529:
530: If $f$ is a real function meromorphic in $\C$ then its associated
531: surface has a natural involution which makes it a symmetric
532: surface spread over the sphere. In the opposite direction,
533: to a symmetric surface spread over the sphere of parabolic type,
534: a real meromorphic function is associated.
535:
536: It is clear that the set $A=\{ a_1,\ldots,a_q\}$ of basis points
537: of a symmetric N-surface is invariant under complex conjugation.
538: Suppose for a moment that at most two of the points $a_1,\ldots,a_q$ are real.
539: Then there exists a base curve $\Gamma$ passing through $a_1,\ldots,a_q$
540: which is symmetric with respect to complex conjugation.
541: Choosing the $\cross$ and $\circ$ points on the real axis we
542: can perform the construction of the Speiser graph symmetrically.
543: The resulting graph $S$ and the tree $T(S)$ will be {\em symmetric} in
544: the following sense. The involution $s$ will send each vertex to
545: a vertex with the same label, each edge to an edge and each face to a face
546: with complex conjugate label.
547:
548: In the general case, that more than two basis points are allowed
549: on the real line, one has to modify a little the definition of the Speiser
550: graph. Let $a$ be a real base point of a symmetric N-surface $(X,\pi,s)$.
551: Consider an open (round) disc $D\subset\bC$ centered at $a$ and not
552: containing other base points. Let $V$ be a component of $\pi^{-1}(D)$
553: which defines a logarithmic singularity (see Section 2).
554: Then one of the following: either $V$ is invariant under $s$, or $s(V)=V'$
555: where $V'$ is another component of $\pi^{-1}(D)$, disjoint from $V$. In
556: the latter case, $V$ is disjoint from the symmetry axis.
557: In the former case, we will call the logarithmic singularity {\em real}.
558:
559: We claim that there are at most two real logarithmic singularities.
560: Indeed, for a real logarithmic singularity, the intersection of
561: $V$ with the symmetry axis consists of a ``ray'', and there cannot be more
562: than two disjoint ``rays'' on the symmetry axis.
563:
564: The symmetry axis divides $X$ into two ``halfplanes'', and each non-real
565: logarithmic singularity (more precisely, its defining region $V$)
566: belongs to one of these ``halfplanes''. Thus
567: the non-real logarithmic singularities are split into two classes, say
568: $C_+$ and $C_-$
569: according to the ``halfplane'' they belong, and the regions $V$ and $V'$
570: always belong to different classes.
571:
572: The real logarithmic singularities lie over at most two
573: basis points.
574:
575: Let $a$ be a real basis point
576: such that there are no
577: real logarithmic singularity over $a$. Consider a homeomorphism $\eta_+$ of
578: the Riemann sphere which is identical outside $D$ and sends the point $a$
579: to the point $a+i\epsilon$, where $\epsilon>0$ is so small that $a+i\epsilon\in
580: D$. Let $\eta_-(z)=\overline{\eta_+(\overline{z})}$.
581: We deform our map $\pi$ in the following way:
582: $$\pi^*(x)=\left\{\begin{array}{ll}\eta_+\circ\pi(x),\; x\in V, &V\in C_+,\\
583: \eta_-\circ\pi(x),\; x\in V, &V\in C_-,\\
584: \pi(x) &\mbox{otherwise.}
585: \end{array}\right.
586: $$
587: Evidently, the new N-surface is symmetric.
588: Let $E$ be the set projections of real logarithmic singularities,
589: ${\mathrm{card}\; E}\leq 2$.
590: Performing the deformation described above for all real base points
591: except two of them, $a'$ and $a''$, such that $E\subset\{ a',a''\}$ we
592: obtain a new symmetric $N$-surface which has the property
593: that only two basis points are real. So a symmetric base curve
594: can be chosen and a symmetric Speiser graph constructed.
595: The Speiser graph of this deformed surface
596: does not depend on $\epsilon$
597: as soon as $\epsilon$ is positive and small enough, and we call
598: it a {\em symmetric Speiser graph of} $(X,\pi)$. The number of
599: basis points of $(X,\pi^*)$ is larger than that of
600: $(X,\pi)$; to preserve properties 2 and 3, we can use two
601: different
602: labels, say $a^+$ and $a^-$ for a basis point $a$ as above.
603: The faces of $S$ over $a+i\epsilon$ are labeled by $a^+$, those
604: over $a-i\epsilon$
605: by $a^-$.
606:
607: Symmetric Speiser graphs have all the properties 1-5 listed above, and
608: in addition, they are preserved by an orientation-reversing involution
609: of the ambient surface.
610: This includes the action of $s$ on labels
611: if we consider
612: $a^+$ and $a^-$ as complex conjugate.
613:
614:
615:
616: Given a symmetric Speiser graph
617: one can construct
618: a symmetric surface spread over the sphere corresponding to this Speiser
619: graph. First we replace all labels $a^+$ by $a+i\epsilon$ and $a^-$ by
620: $a-i\epsilon$, then choose a symmetric basis curve passing through the
621: new basis points, points $\cross$ and $\circ$ on the real axis
622: and a symmetric graph $L$, and perform all
623: construction preserving symmetry.
624: Then we apply the inverse of the deformation described above to place the basis points
625: in their original position.
626:
627: We will need two simple properties of symmetric Speiser graphs $S$ and
628: trees $T(S)$.
629: \vspace{.1in}
630:
631: \noindent
632: A. If a logarithmic end intersects the axis then it is contained in
633: the axis.
634: \vspace{.1in}
635:
636: \noindent
637: B. Every edge either belongs to the axis or is disjoint from it.
638: \vspace{.1in}
639:
640: Consider a symmetric N-surface satisfying the following
641: \vspace{.1in}
642:
643: \noindent
644: {\bf Assumptions}
645: {\em The number of logarithmic ends is at least $3$,
646: zero is a basis point, there is at most one
647: real logarithmic singularity not lying over zero, and the symmetric
648: Speiser graph does not have labels $0^+$ or $0^-$. }
649: \vspace{.1in}
650:
651: Comments. The assumption that there are at least three logarithmic ends
652: excludes only the trivial cases when $P=\const$.
653: The assumption that $0$ is a basis point does not restrict generality
654: because an extra basis point can be always added.
655: The third assumption excludes the cases when the number of real zeros
656: is finite (if there are two non-zero real logarithmic singularities,
657: then $f$ has non-zero limits along the real axis as $x\to+\infty$
658: and $x\to-\infty$,
659: so the set of real zeros is finite). If this third assumption is
660: satisfied, we can always construct the symmetric
661: Speiser graph in such a way that it does not contain
662: labels $0^+$ and $0^-$.
663: \vspace{.1in}
664:
665: \noindent
666: {\bf Proposition 3}.
667: {\em Let $S\subset\C$ be a symmetric Speiser graph corresponding to
668: a symmetric N-surface, $f$ an associated real Nevanlinna function,
669: and the above Assumptions are satisfied.
670: Then all zeros of $f$ are real if and only if $S$ has the following property:
671: each vertex belongs either to the axis of symmetry, or to the boundary
672: of an unbounded face labeled by $0$.}
673: \vspace{.1in}
674:
675: {\em Proof}. Suppose that all zeros of $f$ are real.
676: This means that all $\pi$-preimages of $0$ lie on the axis of symmetry.
677: By Proposition 2, these preimages are in bijective
678: correspondence with $2$-gonal faces $F$ labeled by zero.
679: We have for such faces $F\cap s(F)\neq\emptyset$,
680: and thus by the symmetry of the graph,
681: $F=s(F)$. It follows that both vertices on the boundary of
682: $F$ belong to the axis.
683:
684: Every vertex belongs to the boundary of some face
685: labeled by $0$. If this face is bounded we conclude from the above
686: that the vertex lies
687: on the symmetry axis. This proves necessity of the condition of Proposition 3.
688:
689:
690: Now we suppose that each vertex belongs either to the symmetry axis or to
691: the boundary of an unbounded face labeled by zero. As every vertex
692: belongs to the boundary of {\em only one} face labeled by $0$, we conclude
693: that every $2$-gonal face labeled by zero has one boundary vertex
694: on the axis, and thus
695: its other boundary vertex also belongs to the axis.
696: We conclude that this face is symmetric, and thus the $\pi$-preimage of $0$ contained in this face belongs to the axis.
697: \vspace{.1in}
698:
699: {\em Proof of Theorem 2}. The cases $P=0$ and $d=0$ are trivial,
700: so we assume
701: that $d\geq 1$. If there are two real logarithmic singularities
702: over non-zero points then $d$ is even and $f$ has finitely
703: many zeros, so there is nothing to prove.
704: Thus we suppose from now on that the Assumptions stated above
705: are satisfied.
706:
707: If the number $n=\deg P+2$ of logarithmic ends is even,
708: then either none or two of them belong to the axis of symmetry.
709:
710: If none of the logarithmic ends belongs to the axis then
711: $f$ has finitely many zeros.
712:
713: Now suppose that there are two logarithmic ends on the axis
714: of symmetry.
715: Consider one of them.
716: Let $a$ and $b$ be the labels of the two unbounded
717: faces adjacent to it.
718: Then $a\neq b$ and $a=\overline{b}$, so neither
719: $a$ nor $b$ can be $0$.
720: This implies that all vertices on this logarithmic end belong to the
721: boundaries of $2$-gonal faces whose labels are $0$, and
722: thus we obtain an infinite sequence of real zeros.
723: As there are two logarithmic ends on the axis of symmetry,
724: the sequence of zeros is unbounded from above and below.
725: This proves (a).
726:
727: That for $d=4k, k\geq 1$ both cases actually occur
728: is demonstrated by Figures 1 and 2.
729:
730: If $n$ is odd then exactly one logarithmic
731: end belongs to the axis of symmetry.
732: The other end of the symmetry axis is contained in an infinite face
733: and bisects it (by symmetry).
734: It follows that $\pi$ has a limit on this other
735: end of the symmetry axis.
736: If the limit is non-zero, (b) immediately follows.
737: If the limit is zero, we notice that this end of the axis belongs to
738: a neighborhood $V$ of a real logarithmic singularity over $0$,
739: and again statement (b) follows.
740: \vspace{.1in}
741:
742: {\em Proof of Theorem 1}. It is enough to display a Speiser graph for each case.
743: For simplicity we only show in figures 1-5 the trees $T(S)$ on the left of each picture
744: and the basis curves with basis points
745: on the right. It can be easily seen that each of our trees has unique
746: extension to a symmetric Speiser graph $S$ that has the property
747: described in Proposition 3.
748: \vspace{.1in}
749:
750: {\em Remarks}. Suppose that all basis points of a symmetric N-surface
751: are real. Then we can construct
752: another kind of Speiser graph which we call {\em almost symmetric},
753: without using the perturbation procedure described above. Namely, take the
754: real axis as the basis curve, and choose $\cross$ and $\circ$ at the
755: points $\pm i$. The corresponding Speiser graph is preserved by the
756: involution,
757: except that the vertex labels are now interchanged. It is easy to see
758: that in such almost symmetric graph there can be no vertices on the axis,
759: and {\em exactly one edge} of $T(S)$ intersects the axis. We conclude
760: that
761:
762: {\em A real Nevanlinna function with at least three logarithmic
763: singularities and only real asymptotic values
764: can have at most one
765: real zero}.
766: \newpage
767:
768: \begin{center}
769: \input{deg0a.tex}
770: \vspace{.1in}
771:
772: Figure 1. $d\equiv 0\,(\mod 4)$, the sequence of zeros infinite in both
773: directions.
774: \end{center}
775: \newpage
776: \begin{center}
777: \input{deg4a.tex}
778: \vspace{.1in}
779:
780: Figure 2. $d\equiv 0\,(\mod 4)$, the sequence of zeros is finite.
781: \end{center}
782: \newpage
783: \begin{center}
784: \input{deg1a.tex}
785: \vspace{.1in}
786:
787: Figure 3. $d\equiv 1\,(\mod 4)$.
788: \end{center}
789: \newpage
790: \begin{center}
791: \input{deg2b.tex}
792: \vspace{.1in}
793:
794: Figure 4. $d\equiv 2\,(\mod 4).$
795: \end{center}
796: \newpage
797: \begin{center}
798: \input{deg3a.tex}
799: \vspace{.1in}
800:
801: Figure 5. $d\equiv 3\,(\mod 4).$
802: \end{center}
803:
804:
805: \newpage
806: \begin{thebibliography}{11}
807: \bibitem{Ah} L. Ahlfors, Complex Analysis, McGraw-Hill, NY, 1979.
808: \bibitem{Ah1} L. Ahlfors, \"Uber eine in der neueren
809: Wertverteilungstheorie betrachtete Klasse transcendenter Funktionen, Acta math.,
810: (1932) 375-406.
811: \bibitem{H} D.A. Brannan and W.K. Hayman,
812: Research problems in complex analysis,
813: Bull. London Math. Soc., 21 (1989) 1-35.
814: \bibitem{DV} D. Drasin and A. Weitsman, Meromorphic functions with large sums of deficiencies, Adv. Math., 15 (1975) 93--126.
815: \bibitem{GO} A.A. Gol'dberg and I.V. Ostrovskii, Distribution of values of meromorphic functions, Moscow, Nauka, 1970 (Russian)
816: \bibitem{G1} G. Gundersen, Solutions of $f''+P(z)f=0$ that have almost all
817: real zeros, Ann. Acad. Sci. Fenn. Math. 26 (2001) 483--488.
818: \bibitem{G2} G. Gundersen, On the real zeros of solutions of $f''+A(z)f=0$,
819: where $A(z)$ is entire, Ann. Acad. Sci. Fenn. Math. 11 (1986) 275-294.
820: \bibitem{HR} S. Hellerstein and J. Rossi, Zeros of meromorphic solutions
821: of second order linear differential equations, Math. Z., 192 (1986)
822: 603--612.
823: \bibitem{Fne} F. Nevanlinna, \"Uber eine Klasse meromorpher
824: Funktionen,
825: 7 Congress Math. Scand., Oslo, 1929.
826: \bibitem{RNe1} R. Nevanlinna, \"Uber Riemannsche Fl\"achen mit endlich vielen
827: Windungspunkten, Acta math., 58 (1932), 292--373.
828: \bibitem{Ne2} R. Nevanlinna, Analytic functions, Springer, NY, 1953.
829: \bibitem{RW} J. Rossi and S. Wang, The radial oscillation of solutions to ODE's
830: in the complex domain, Proc. Edinburgh Math. Soc., (2) 39 (1996) 473--483.
831: \bibitem{shin} K. Shin, A new class of examples related to a question of
832: Hellerstein and Rossi, Preprint, 2002.
833: \bibitem{Sib} Y. Sibuya, Global theory of a second order linear ordinary differential equation with a polynomial coefficient, North-Holland, Amsterdam, 1975.
834: \bibitem{S} S. Stoilov, Le\c{c}ons sur les principes topologiques de la th\'eorie
835: des fonctions analytiques, Paris, Gauthier-Villars, 1956.
836: \bibitem{T} E. Titchmarsh, Eigenfunction expansions associated with second-order
837: differential equations, vol. 1, second edition, Oxford, Oxford UP, 1962.
838: \end{thebibliography}
839:
840: {\em
841:
842: Purdue University
843:
844: West Lafayette IN
845:
846: eremenko@math.purdue.edu
847: \vspace{.2in}
848:
849: University of Michigan
850:
851: Ann Arbor MI
852:
853: merenkov@umich.edu}
854:
855: \end{document}
856:
857:
858: To each tract of a meromorphic function corresponds
859: a point $a\in\bC$ which is called an asymptotic value. In the case of
860: Nevanlinna functions, asymptotic values coincide with deficient values
861: and $\delta(a,f)=2n(a)/(d+2),$ where $n(a)$ is the number of tracts with
862: the asymptotic value $a$. Nevanlinna functions were originally used
863: to solve the Inverse Problem of the value distribution theory
864: (see, for example \cite{Nev1,Nev2,GO}. Asymptotic values of
865: a Nevanlinna function can be also interpreted as the Stokes' multipliers
866: of the differential equation (\ref{DE}). This interpretation was
867: used by Sibuya \cite{Sibuya} to prove the existence of
868: differential equations (\ref{DE}) with prescribed Stokes
869: multipliers.
870:
871: There is a natural cyclic ordering of asymptotic values,
872: $a_0,a_1,\ldots,a_{d+1}$, where the subscripts are the residues modulo
873: $n$.
874: Sibuya showed that exactly those sequences are possible for Nevanlinna
875: functions which
876: satisfy $a_j\neq a_{j+1},\; j=0,\ldots,d+1$.
877:
878: This leads us to the question, which sequences of asymptotic values
879: may occur
880: for Nevanlinna functions with real zeros. To state the
881: answer, we introduce the following normalization:
882:
883: (i) In case b) of Theorem 2, suppose that zeros of $w$ belong to the
884: negative ray. This can be always achieved by a change of the
885: independent variable $z\mapsto\pm z+c,\;c\in \R$.
886:
887: (ii) Suppose that $a_0$ is the asymptotic value corresponding to the
888: smallest argument of a tract.
889:
890: Then we have
891: \vspace{.1in}
892:
893: \noindent
894: {\bf Theorem 3} {\em A sequence $(a_0,\ldots,a_{d+1}$ is a
895: normalized sequence of asymptotic values of a Nevanlinna function
896: with real zeros if and only if it satisfies the following conditions:
897:
898: a) If $d\equiv 0 (\mod\, 4)$:
899: $$a_j=\overline{a_{d+1-j}},\quad 0\leq j\leq d/2,$$
900: $$a_{2k+1}=0,\quad k=0,\dots,d/2-1,$$
901: $$?$$
902:
903: b) If $d\equiv 1 (\mod 4)$:
904: $$\quad a_j=\overline{a_{d-j}},\quad 1\leq j\leq (d-1)/2,$$
905: $$a_{2k+1}=0,\quad k=0,\ldots,(d-1)/2,$$
906:
907: c) If $d\equiv 2 (\mod 4)$;
908: %$$ it ishe following sequences of asymptotic values are
909: possible
910:
911:
912:
913:
914: \section{Construction of solutions}
915:
916: Our construction is based on the work of R. Nevanlinna (\circite[Ch.]{N} and
917: references there). First we recall, how some simply connected
918: Riemann surfaces,
919: spread over $\bC$ are described by labeled Speiser graphs
920: (Line Complexes, Streckenkomplexen).
921:
922: A {\em Speiser graph} is a bipartite
923: connected graph, whose vertices all have the same degree $q\geq 2$,
924: properly embedded in the plane. In what follows we
925: identify a Speiser graph with its image
926: $\Gamma\subset\C$. It is a custom to mark the vertices with $\circross$ and $\o$,
927: so that the edges connect vertices of different kind. The components
928: of $\C\backslash\Gamma$ are called {\em faces}, each of them is a
929: simply connected region, bounded
930: by an even number of edges, or by infinitely many edges, making one chain.
931:
932: Suppose we have a cyclic sequence $a_1,\ldots,a_q$ of distinct points
933: in $\bC$,
934: so that the subscript is a residue modulo $q$.
935: Then we can label the faces of a Speiser graph $\Gamma$ with points $a_j$,
936: so that the following rule is satisfied:
937:
938: \noindent{\em when going anticlockwise around a vertex,
939: the labels are
940: encountered in their cyclic order,
941: if the vertex is an $\circross$ and in the opposite cyclic order, if the vertex is
942: an $\o$.}
943:
944: Labeling the faces induces a labeling of edges by residues modulo $q$
945: in the following way: if an edge belongs to the common boundary
946: of two faces labeled by $a_j$ and $a_{j+1}$, we label this edge with $j$.
947:
948: A Speiser graph, whose faces and vertices are
949: labeled this way, will be called
950: a {\em labeled Speiser graph}. Now we choose an base curve
951: $\gamma$ which passes through $a_1,\ldots,a_q$ according
952: to their cyclic order. Such $\gamma$ is usually called a
953: {\em base curve}. We denote by $\gamma_j$ the arc of $\gamma$
954: between $a_j$ and $a_{j+1}$.
955:
956: Given a labeled Speiser graph, and a base curve, one constructs a
957: simply connected Riemann surface $S$, spread over $\bC$ in the following
958: way. The base curve divides the sphere $\bC$ into two complementary
959: regions. We call $D$ that one, whose oriented boundary coincides
960: with $\gamma$, and $\Delta$ the other one. Let $X$ and $O$ be the sets
961: of $\circross-$ and $\o-$ vertices of $\Gamma$. We consider the
962: disjoint union of
963: products $S':=(D\times X)\circup (\Delta\times O)$, equipped with the
964: natural projection $\pi:S'\to\bC,\; \pi(z,v)=z$.
965:
966: For every edge $e$ of the Speiser graph we add to $S'$ an open arc
967: by the following rule. Suppose that $e$ has label $j$, and connects
968: two vertices $v'\in X$ and $v''\in O$. Then we add to $S'$ the open common
969: boundary of $D\times v$ and $\Delta\times v'$, which
970: projects to $\gamma_j$.
971:
972: For every face with finitely many sides of $\Gamma$, we add a point
973: to $S'$ by the following rule. Let $v_1,\ldots,v_{2m}$ be all vertices on the
974: boundary of this face, $v_1\in X$, and $a$ be the label of this face.
975: Then we add the common boundary point of
976: $D\times{v_1},\ldots,\Delta\times{v_{2m}}$, which projects to $a$.
977:
978: After all these additions we obtain an open simply connected surface
979: $S$, equipped with a topologically holomorphic map $\pi:S\to\bC$.
980: There is a unique conformal structure on $S$, which makes $\pi$ holomorphic.
981: Thus an open simply connected surface is defined. By the uniformization
982: theorem, there exists a conformal isomorphism $\phi$, so that
983: $\pi\circ\phi$ is a meromorphic function, either in the plane,
984: or in the unit disc.
985:
986: We need a very special case of this construction, when $\pi$ is unramified
987: (so that the only finite-sided faces have $2$ sides),
988: and the number $n$ of infinite-sided faces is finite. For this case
989: we have the following theorem of R. Nevanlinna
990:
991: \noindent
992: {\bf Theorem C}. {\em Let $S$ be an open simply connected Riemann surface spread
993: over $\bC$, with no critical points and $n\in [2,\infty)$ logarithmic branch
994: points. Then $S$ is of parabolic type, and meromorphic function
995: $f=\pi\circ\phi$ satisfies a Schwarz differential equation
996: $$\frac{f'''}{f'}-\frac{3}{2}\left(\frac{f''}{f'}\right)^2=2P,$$
997: where $P$ is a polynomial of degree $d=n-2$.}
998: \vspace{.1in}
999:
1000: Now we state some additional conditions, which will guarantee that
1001: the function $f$ given by Theorem C has only real zeros.
1002:
1003: We choose $\gamma=\{ z:|z-1|=1\}$, oriented anticlockwise,
1004: so that $D$ is the disc of radius $1$ centered at $1$, and $\Delta$ is
1005: its complement. Then we choose $q$ points on $\gamma$, ordered
1006: anticlockwise, so that $a_q=0$, and $a_j=\overline{a_{q-j}},$ for
1007: $j=1,\ldots,q-1.$ Here the bar means complex conjugation. It is also
1008: convenient to introduce conjugate index $\overline{j}=q-j$.
1009:
1010: Speiser graphs in our construction will satisfy two additional assumptions:
1011: \vspace{.1in}
1012:
1013: \noindent
1014: (i) the labeled imbedded graph $\Gamma$ is symmetric with respect
1015: to the real axis. This means that the reflection with respect to
1016: the real axis sends vertices to vertices of the same type $X$ or $O$,
1017: edges to edges with conjugate labels, and faces to faces with complex
1018: conjugate labels.
1019: \vspace{.1in}
1020:
1021: \noindent
1022: (ii)
1023: All $2$-gonal faces, labeled with $a_q=0$ are invariant under the reflection.
1024: \vspace{.1in}
1025:
1026: Now suppose that our labeled Speiser graph and the base curve satisfy
1027: all these conditions. Let $f$ be the corresponding meromorphic function,
1028: normalized so that $f(0)=1$ and $f'(0)=1$. First we show that $f$ is a
1029: real function.
1030:
1031: Our surface $S$ has an anti-conformal involution $I$, defined in the
1032: following way: $I(z,v)=(\overline{z},\overline{v}).$ Because of
1033: the normalization $\phi$ splits the complex conjugation and $I$.
1034: Let $L$ be the line
1035: fixed by this involution. Then $\phi(\R)=L$. By (ii) $\pi^{-1}(0)\subset L$.
1036: So all zeros of $f$ are real.
1037:
1038:
1039: Now we consider the differential equation (\ref{DE}). Our meromorphic
1040: function $f$ factors as $w_1/w_0$, where $w_i$ are two linearly independent
1041: solutions. So $w_1$ has only real zeros.
1042:
1043: It remains to construct the labeled graphs satisfying all stated properties.
1044:
1045:
1046:
1047: \begin{thebibliography}{1}
1048: \bibitem{GO} A. A. Gol'dberg and I. V. Ostrovskii,
1049: Distribution of values of meromorphic functions,
1050: Moscow, Nauka, 1970 (Russian).
1051: \bibitem{G1} G. Gundersen, Solutions of $f''+P(z)f=0$ that have almost all
1052: real zeros, Preprint.
1053: \bibitem{G2} G. Gundersen, On the real zeros of solutions of $f''+A(z)f=0$,
1054: where $A(z)$ is entire, Ann. Acad. Sci. Fenn., Math. 11 (1986) 275-294.
1055: \bibitem{H} D.A. Brannan and W.K. Hayman,
1056: Research problems in complex analysis,
1057: Bull. London Math. Soc., 21 (1989) 1-35.
1058: \bibitem{HR} S. Hellerstein and J. Rossi, Zeros of meromorphic solutions
1059: of second order linear differential equations, Math. Z., 192 (1986)
1060: 603--612.
1061: \bibitem{Nev1} R. Nevanlinna, Analytic Functions, Springer
1062: \bibitem{Nev2} R. Nevanlinna,
1063: \bibitem{Sib} Y. Sibuya,
1064: \bibitem{T} E.C. Titchmarsh, Eigenfunction expansions,
1065: associated with second-order differential equations, Part I, Second Edition,
1066: Oxford University Press, 1962.
1067: \end{thebibliography}
1068:
1069: {\em Purdue University
1070:
1071: eremenko@math.purdue.edu
1072:
1073: University of Michigan
1074:
1075: merenkov@umich.edu}
1076: \end{document}
1077:
1078: