math0211021/zoll.tex
1: %&latex
2: \documentclass[12pt]{article}
3: \usepackage{latexsym}
4: \usepackage{amssymb}
5: \usepackage[OT2,OT1]{fontenc}
6: \usepackage{graphicx}
7: \DeclareGraphicsExtensions{.eps, .jpg}
8: \def\cyr{%
9: \renewcommand\rmdefault{wncyr}%
10: \renewcommand\sfdefault{wncyss}%
11: \renewcommand\encodingdefault{OT2}%
12: \normalfont
13: \selectfont}
14: \DeclareTextFontCommand{\textcyr}{\cyr}
15: \def\p#1{\frac{\partial}{\partial #1}}
16: \def\half{\frac{1}{2}}
17: \newcommand{\Dye}{\mbox{\cyr   D}}
18: \newcommand{\Ge}{\mbox{\cyr  \em G}}
19: \newcommand{\C}{\mathbb{C}}
20: \newcommand{\CC}{\mathbb{C}}
21: \newcommand{\HH}{\mathbb{H}}
22: \newcommand{\CP}{\mathbb C\mathbb P}
23: \newcommand{\RP}{\mathbb R\mathbb P}
24: \newcommand{\PP}{\mathbb P}
25: \newcommand{\RR}{\mathbb R}
26: \newcommand{\NN}{\mathbb{N}}
27: \newcommand{\Z}{\mathbb{Z}}
28: \newcommand{\ZZ}{\mathbb{Z}}
29: \renewcommand{\d}{\mathrm{d}}
30: \newcommand{\notD}{{\rm D\hskip-0.65em / }}
31: \newcommand{\D}{\mathrm{D}}
32: \newcommand{\G}{\mathcal{N}}
33: \renewcommand{\O}{\mathcal{O}}
34: \renewcommand{\P}{\mathbb{P}}
35: \newcommand{\SL}{\mathrm{SL}}
36: \newcommand{\zz}{{\mathfrak z}}
37: \newcommand{\DD}{{\rm I\hskip-0.22em D}}
38: \newcommand{\1}{{\rm 1\hskip-0.25em I}}
39: %\newcommand{\be}{\begin{equation}\label}
40: %\newcommand{\ee}{\end{equation}}
41: %\newcommand{\bea}{\begin{eqnarray}\label}
42: %\newcommand{\eea}{\end{eqnarray}}
43: %\newcommand{\proof}{ \noindent {\bf Proof:}}
44: 
45: 
46: \newtheorem{thm}{Theorem}[section]
47: \newtheorem{defn}[thm]{Definition}
48: \newtheorem{prop}[thm]{Proposition}
49: \newtheorem{cor}[thm]{Corollary}
50: \newtheorem{lem}[thm]{Lemma}
51: \newtheorem{conjecture}[thm]{Conjecture}
52: \newenvironment{remark}{\medskip \noindent {\bf Remark.}}{\hfill $\diamondsuit$\\}
53: \newenvironment{proof}{\medskip \noindent {\bf Proof.}}{\hfill $\blacksquare$\\}
54: \topmargin0pt
55: \headheight0pt
56: \headsep0pt
57: \oddsidemargin0pt
58: \textheight23cm
59: \textwidth16cm
60: \begin{document}
61: \title{Zoll Manifolds and Complex Surfaces}
62: \author{Claude LeBrun\thanks{Supported 
63: in part by  NSF grant DMS-0072591.}     ~and  
64: L.J. Mason}
65: 
66: \maketitle 
67: 
68: 
69: \abstract{
70: We classify compact surfaces with 
71:  torsion-free 
72: affine connections
73: for which every geodesic is a simple closed curve.  In the process, 
74: we obtain  completely  new proofs of all the 
75:  major results \cite{beszoll}  concerning the Riemannian case. 
76:   In contrast to previous work, our 
77: approach is 
78: twistor-theoretic, and    depends  fundamentally on the 
79: fact that, up to biholomorphism, there is only one 
80:  complex structure on $\CP_2$.} 
81: 
82: \section{Introduction}
83: 
84: A  {\em Zoll metric} on a smooth manifold $M$  is a  Riemannian
85: metric $g$   whose geodesics are all simple closed curves of equal length. 
86: This terminology \cite{guillzoll} celebrates Otto Zoll's (now century-old)
87:  discovery \cite{zoll}
88: that $S^2$  admits many such metrics besides the obvious metrics of 
89:  constant curvature \cite{beszoll}. Indeed, in terms of  cylindrical coordinates
90: $(z,\theta)\in [-1,1]\times [0, 2\pi ]$, 
91: \begin{equation}
92: \label{ansatz}
93: g= \frac{\left[1+f(z)\right]^2}{1-z^2}dz^2 + (1-z^2)d\theta^2
94: \end{equation}
95: defines a Zoll metric on $S^2$
96:  for any smooth odd function 
97: $$
98: f: [-1 , 1] \to (-1, 1) , ~~ f(-z) = -f(z) 
99: $$
100: which vanishes at the end-points of the interval. A formal perturbation argument of 
101: Funk \cite{funk} later indicated that, modulo isometries and rescalings, 
102:  the {general}  Zoll metric
103: on $S^2$ depends on  one {odd} function 
104: $f: S^2 \to \RR$. This formal calculation was later turned
105: into a theorem by Guillemin \cite{guillzoll}, whose proof
106: depends, e.g.\ on an implicit function theorem 
107: of Nash-Moser type. Because the function $f$ is required to 
108: satisfy $f(-\vec{x})=-f(\vec{x})$, 
109: however,  these constructions never  give rise to  non-standard Zoll
110: metrics on $\RR\PP^2$. Indeed, the so-called Blaschke conjecture,  
111: proved by 
112: Leon  Green \cite{grezoll}, asserts that, up to isometries
113: and rescaling, the only  Zoll metric on $\RR\PP^2$
114: is the standard one. For an outstanding survey of these results, as well as 
115:  an exploration of their higher-dimensional Riemannian generalizations,
116: see \cite{beszoll}.
117: 
118: The aims of the present article are twofold. First of all, instead of
119: limiting ourselves to the study of  Riemannian metrics, we will more generally 
120: consider symmetric affine connections $\nabla$, and 
121: ask how many such connections on a given manifold $M$ have the property
122: that all of their geodesics are simple closed curves. In order
123: to make this  a sensible problem, however, one must first observe 
124: that for any   $1$-form $\beta$ on $M$, 
125: the  symmetric affine connection $\hat{\nabla}$ defined by 
126: $$\hat{\nabla}_{\bf u} {\bf v} = \nabla_{\bf u}{\bf v} + \beta ({\bf u}) {\bf v} + \beta ({\bf v}) {\bf u}$$ 
127: has exactly the same {\em unparameterized} geodesics
128: as the  connection $\nabla$; two connections
129: related in this manner are said to be {\em projectively equivalent},
130: and  obviously one should therefore only try to classify 
131: such connections  modulo projective equivalence.  
132: 
133: 
134: Even in this rather general setting, our methods will allow us to 
135:  obtain results  very much like  
136: to the classical Riemannian results alluded to above. Indeed, in 
137: \S \ref{prelim}, we begin by showing that  the only
138: compact surfaces which admit Zoll projective  connections are $S^2$
139: and $\RR\PP^2$. In \S \ref{rip}, we then go on to show 
140: that,   modulo  diffeomorphisms, there is only one such
141:  projective class of  connections on $\RR\PP^2$.
142: Finally, in \S \ref{zoe}, we prove that 
143: there is a non-trivial moduli space of such 
144: projective classes on $S^2$, locally  parameterized by the  space of 
145: vector fields on $\RP^2$. 
146: 
147: But  even in the Riemannian case,   we seem to 
148: have something fundamentally 
149: new to contribute to the subject, as our proofs rest on foundations 
150:  completely
151:  different   
152:  from those used by  of our predecessors. 
153: Blaschke's unsuccessful approach to the problem of classifying
154: Zoll metrics on $\RP^2$ amounted to a direct attempt to identify the 
155: space of all geodesics with the standard dual projective plane
156: $\RP^{2*}$, the points of which which are by definition the   real projective lines 
157: $\RP^1$ in $\RP^2$.
158: The essence of  our method is to instead use {\em complex}, rather than real,
159: projective geometry to solve the problem. 
160: Indeed, we will 
161:  construct a  complex $2$-manifold
162: from any given   Zoll structure, modeled on the  dual complex projective plane $\CC\PP_2^*$.
163: The punch line of the proof is then that, up to biholomorphism, there is 
164: \cite{bpv,yau} 
165: only one complex structure on $\CP_2$. 
166: Our proof of the generalized Blaschke conjecture 
167: then proceeds by recognizing the points of $\RP^2$ as the set of those 
168: complex projective lines $\CP_1$ in this $\CP_2$
169: which are invariant under the action of a certain anti-holomorphic involution. 
170: By contrast, 
171: the flexibility of Zoll structure on $S^2$ arises because 
172:  the points in this case are instead represented 
173: by holomorphic disks with boundary on a totally real 
174: embedding of $\RP^2$ in $\CP_2$;  deformations of  this
175: embedding then correspond to deformations of
176: the Zoll structure. In this way, we are not only able to 
177: construct the general small deformation of the standard 
178: Zoll structure without recourse to Nash-Moser, 
179:  but, more importantly, we  are also
180: able to glean a significant amount of information 
181: regarding arbitrary Zoll structures, even when they are quite
182: far from the model case. 
183: 
184: Finally, by way of  an appendix, 
185: this article ends where it began,  with a discussion of the 
186: axisymmetric  case.  After all, since we have chosen to 
187: generalize Zoll's problem by focusing on projective structures, 
188: it is only fitting that we should also generalize
189: Zoll's  construction by  writing down all the 
190: axisymmetric Zoll projective structures on $S^2$ in closed form. 
191: In the process, we are able to show how the conceptual framework 
192: used in \S \ref{zoe}  can be implemented in concrete,
193: calculational terms. We  hope  that our discussion of this special case will not only
194: help clarify 
195:  our general approach, but also make it seem all the more compelling. 
196: 
197: \pagebreak 
198:  
199: \section{Zoll  Projective Structures} \label{prelim}
200: 
201: 
202: 
203: We  begin  by recalling the notion \cite{schouten} of projective equivalence of 
204: affine connections.
205: \begin{defn}
206: Two  torsion-free affine connections $\nabla$ and $\hat{\nabla}$ on a
207:  manifold $M$ are said to be {\em projectively equivalent}
208: if they have the same  geodesics, considered as {\em unparameterized} curves. 
209: \end{defn}
210: This condition may be re-expressed as
211: the requirement that 
212: $$
213: \hat{\nabla}_{\bf v}{\bf v}\propto {\bf v} ~~~ \Longleftrightarrow ~~~ {\nabla}_{\bf v}{\bf v}\propto {\bf v}.
214: $$
215: We therefore have  \cite{schouten}
216: \begin{prop}
217: Two $C^k$ symmetric affine connections $\nabla$ and $\hat{\nabla}$ 
218: are projectively equivalent iff
219: $$\hat{\nabla}_{\bf u} {\bf v} = \nabla_{\bf u}{\bf v} + \beta ({\bf u}) {\bf v} + \beta ({\bf v}) {\bf u}$$ 
220: for some $C^k$  $1$-form $\beta$. 
221: \end{prop}
222: Here a connection is said to be of differentiability class $C^k$ 
223: with respect to a fixed $C^{k+2}$ structure if 
224: the covariant derivative of any $C^{k+1}$ vector field is 
225: a $C^k$ tensor field; this is equivalent to requiring that 
226: the Christoffel
227: symbols
228: $$
229: \Gamma^j_{k\ell} = \left\langle dx^j , \nabla_{\p {x^k}}\p {x^\ell}\right\rangle
230: $$
231: are all $C^k$ functions in any admissible local coordinate system. 
232: We also note, in passing, that the symmetric (or torsion-free) condition 
233: employed here can been
234: imposed without any loss of generality; given  an {\em arbitrary} affine connection, one
235: can construct a unique torsion-free connection with precisely the same {\em parameterized}
236: geodesics by replacing the Christoffel symbols with their symmetrizations:
237:  $$
238: \Gamma^j_{k\ell} \rightsquigarrow \hat{\Gamma}^j_{k\ell}
239: =\frac{1}{2}\left(\Gamma^j_{k\ell} + \Gamma^j_{\ell k}\right)~.
240: $$
241: \begin{defn}
242: A $C^k$ {\em projective structure} on a smooth manifold is  the 
243: projective equivalence class $[\nabla]$ of some torsion-free $C^k$ affine
244: connection $\nabla$. 
245: \end{defn}
246: By definition, 
247: a  projective structure $[\nabla]$ on $M$  defines a certain family of geodesics;
248: these are to be thought of as abstract immersed curves in $M$,  without preferred parameterizations. Conversely, a projective structure is completely
249: specified once its geodesics are known.
250: 
251: In this paper, we will be interested in projective structures for
252: which every  geodesic is a simple closed curve. 
253: 
254: \begin{defn}
255: Let $\nabla$ be a $C^1$  torsion-free affine connection on a 
256:    smooth manifold $M$. We will say that the projective equivalence class
257: $[\nabla ]$ of 
258: $\nabla$   is a {\em  Zoll projective structure}
259:  if 
260:  the image ${\mathfrak C}$ of any maximal geodesic of $\nabla$
261: is an embedded circle $S^1\subset M$.
262: \end{defn}
263: 
264: %However, it will temporarily be technically convenient 
265: %to focus on those Zoll manifolds satisfying an additional assumption:
266: 
267: %
268: %\begin{defn}
269: %A Zoll projective structure $[\nabla ]$ on $M$ will be called
270: %{\em docile} if 
271: %$ \langle w_1(TM) , [{\mathfrak C}] \rangle \in \ZZ_2$
272: %is the same for all of its geodesic circles ${\mathfrak C}$. 
273: %\end{defn}
274: 
275: %This hypothesis may be restated as stipulating that
276: %a geodesic circle ${\mathfrak C}$  can have an orientable neighborhood
277: % $U\subset M$ only if  {\em every} geodesic circle has such a neighborhood. 
278: %Of course, this is  automatically true whenever  $M$
279: %is orientable. 
280: 
281: If $c: (a,b) \looparrowright M$ is any immersed curve, its derivative 
282: $dc/dt$ is non-zero at every point, so that $[dc/dt ]$ is a well-defined
283: element of the {\em projectivized tangent bundle}
284: $$
285: \PP TM = (TM -0_M)/\RR^\times ;
286: $$
287: thus $t\mapsto [dc/dt]$ defines a  curve $\tilde{c} : \RR \to \PP TM$, called
288: the 
289:  {\em canonical lift}  of $c$. 
290: Given a  $C^k$ Zoll projective structure $[\nabla ]$ on $M$, the canonical lifts 
291: of its geodesics 
292:  give us a $C^k$ 
293:  foliation $\mathcal F$ of  $\PP TM$ by circles. Let  $N$ denote the
294: leaf space of this foliation. 
295: 
296: \begin{defn}
297: Let $(M,[\nabla ])$ be an $n$-manifold with $C^k$ Zoll projective structure. 
298: We will say that $[\nabla ]$ is {\em tame} if the corresponding 
299: foliation $\mathcal F$ of $\PP TM$
300: by lifted geodesics is 
301:  {\em locally trivial},  in the sense that   each  
302: leaf has a neighborhood which is $C^k$ diffeomorphic to 
303: $\RR^{2n-2} \times S^1$ in such a manner that every leaf 
304:  corresponds to a  circle of the form  $\{ pt \} \times S^1$. 
305: \end{defn}
306: 
307: 
308: These local trivializations give $N$ the structure
309: of a $C^k$  $(2n-2)$-manifold in a canonical manner, making the quotient map 
310: $\nu : \PP TM \to N$ into a $C^k$  submersion.  We will call the surface
311: $N$ the space of (undirected) geodesics of the tame  Zoll  projective structure
312: $[\nabla ]$. The situation is encapsulated by a diagram 
313: \setlength{\unitlength}{1ex}
\begin{center}\begin{picture}(20,17)(0,3)
\put(10,17){\makebox(0,0){$\PP TM$}}
\put(2,5){\makebox(0,0){$M$}}\put(18,5){\makebox(0,0){$N$}}
\put(15,12){\makebox(0,0){$\nu$}}
\put(5,12){\makebox(0,0){$\mu$}}
\put(11,15.5){\vector(2,-3){6}}
\put(9,15.5){\vector(-2,-3){6}}
\end{picture}\end{center}
314: which we shall refer to as the (real) {\em  double fibration} of $[\nabla ]$. 
315: Here $\mu : \PP TM \to M$ of course denotes the bundle projection. 
316: Notice that, by construction, the tangent spaces of the fibers of 
317: $\mu$ and $\nu$ are everywhere linearly independent:
318: $$(\ker \mu_*)\cap (\ker \nu_* ) =0.
319: $$
320: Moreover, the restriction of  $\nu$  to any fiber  of $\mu$
321: gives us an embedding  $\RR\PP^{n-1}\hookrightarrow N$.
322: 
323: Fortunately, as we will show in Theorem \ref{bingo} below,
324: this  desirable picture is   applies 
325: to every compact Zoll surface. 
326: A key step in this direction is the following: 
327: 
328: \begin{prop} \label{meow} 
329: Any   Zoll projective structure $[\nabla ]$ 
330: on a  compact  {\em orientable} surface $M^2$  is tame. 
331: \end{prop}
332: \begin{proof}
333: Because $M$ is assumed to be a compact surface, 
334:  $\PP TM$ is a compact $3$-manifold, and 
335: the Zoll projective structure $[\nabla ]$ 
336: gives us a foliation $\mathcal F$ of  $\PP TM$ by circles. 
337: However, a theorem of
338: Epstein \cite{epstein} asserts that any
339: foliation of a compact $3$-manifold by circles is a 
340: Seifert fibration. Thus any leaf of  $\mathcal F$ has 
341: a basis of neighborhoods modelled on 
342: $$(\CC \times S^1)/\ZZ_m,$$
343: where the $\ZZ_m$ action on $\CC \times S^1\subset \CC^2$
344: is generated by $(z_1, z_2)\mapsto (e^{2\pi i\ell / m}z_1, e^{2\pi i /m}z_2)$,
345: for some integer $\ell$. All we therefore need to show is that
346: no leaf is non-trivially covered by nearby leaves.
347: 
348: %There are two cases to consider. First, suppose that 
349: %
350: %${\mathfrak C}$ is a closed geodesic with $\langle w_1 (TM) , {\mathfrak C}\rangle =0$.
351: %Then 
352: 
353: Now, because we have assumed that $M$ is orientable, any geodesic circle 
354: ${\mathfrak C}$ has a tubular neighborhood diffeomorphic to the cylinder
355: $S^1 \times \RR$.  Moreover, by Epstein's result, 
356: the lift of  ${\mathfrak C}$ to $\PP TM$  has a standard neighborhood
357: whose projection to $M$  is contained in the given cylindrical neighborhood. 
358: Thus, 
359: any geodesic circle ${\mathfrak C}'$  with initial point
360: and tangent 
361: sufficiently close to those of ${\mathfrak C}$ will remain within our 
362: cylindrical neighborhood, and indeed will do so in such a manner that  the projection ${\mathfrak C}'\to {\mathfrak C}$
363: induced by $S^1 \times \RR\to S^1$ has non-zero derivative everywhere, and so will be 
364:  a covering map. 
365: However,  our tubular neighborhood 
366:  $S^1 \times \RR$ can be identified with  
367:   $\RR^2 -0$ in such a manner that 
368: ${\mathfrak C}$ becomes the unit circle, and the degree of the covering 
369: becomes the winding number of ${\mathfrak C}'$ around the origin.
370: But since ${\mathfrak C}'$ has been transformed into an embedded curve
371: in the plane, the Jordan curve theorem tells us that 
372: its winding number around the origin has absolute value 
373: $\leq 1$. Thus  the covering map in question must have degree $1$.
374: The associated foliation $\mathcal F$ of  $\PP TM$  is therefore  trivial in a neighborhood of 
375:  the lift of ${\mathfrak C}$. 
376: %
377: %If instead $\langle w_1 (TM) , {\mathfrak C}\rangle \neq 0$, then ${\mathfrak C}$  has a neighborhood
378: %diffeomorphic to the M\"obius band, and our nearby geodesic ${\mathfrak C}'$ then becomes
379: %an embedded circle in this band. But  the docile condition
380: % now stipulates that $\langle w_1 (TM) , {\mathfrak C}'\rangle \neq 0$,
381: %too, so the projection ${\mathfrak C}'\to {\mathfrak C}$ must  have odd degree. Double
382: %covering the M\"obius band by the cylinder $S^1\times \RR$
383: %thus results in double coverings  of both 
384: %${\mathfrak C}$ and ${\mathfrak C}'$ by circles,
385: %which we will respectively denote by  $\tilde{{\mathfrak C}}$ and $\tilde{{\mathfrak C}}'$. But the degree of ${\mathfrak C}'\to {\mathfrak C}$ now 
386: %equals the degree of $\tilde{{\mathfrak C}}\to \tilde{{\mathfrak C}}'$, and the latter
387: % must be equal to one, by the same application of the Jordan curve theorem used before. 
388: %Thus ${\mathfrak C}'\to {\mathfrak C}$ is a diffeomorphism, and the associated foliation $\mathcal F$ of  $\PP TM$  
389: %is  therefore  locally  trivial, as claimed. 
390: \end{proof}
391: 
392: 
393: 
394: Next, we wish to determine precisely which compact surfaces
395: admit  Zoll projective structures. Our solution to this problem begins with 
396:  the following   simple observation:
397: 
398: \begin{lem}\label{cover}
399: Let $[\nabla]$ be a tame Zoll projective structure on an $n$-manifold
400: $M$. Let $\varpi : \tilde{M}\to M$ be the universal cover of $M$.
401: Then $[\varpi^*\nabla ]$ is a tame Zoll projective structure on
402: $\tilde{M}$. 
403: \end{lem}
404: 
405: \begin{proof}
406: If $(M,[\nabla])$ is a tame Zoll manifold, all the 
407: lifted geodesics are freely homotopic embedded circles
408: in $\PP TM$; this is true because $\PP TM$
409: is connected, and  is the union 
410: of `trivializing' open sets for the foliation $\mathcal F$, in which all the  circular leaves are 
411:  freely homotopic. 
412: Hence all the 
413: geodesic circles in $M$ are freely homotopic. Moreover, 
414: by considering the geodesic circles through a given point $p\in M$,
415: one obtains a base-point homotopy between 
416: any geodesic circle ${\mathfrak C}\subset M$ and its reverse-parameterized
417: version $\overline{{\mathfrak C}}$. Hence ${\mathfrak C}$ either represents  an   element of order 
418: $1$ or $2$ in $\pi_1 (M, p)$. Thus either ${\mathfrak C}$ or a 2-fold cover $\hat{{\mathfrak C}}\to {\mathfrak C}$
419: lifts to the universal cover $\tilde{M}$ as  an embedded circle,
420: and this circle is geodesic with respect to the pull-back
421: connection $\varpi^*\nabla$. 
422: %Since all the geodesic circles of $(M, [\nabla ])$ are 
423: %freely homotopic, it follows that either (a) all of them
424: %lift to $\tilde{M}$ as embedded circles, or else (b) all of them 
425: %have $2$-fold covers which lift to embedded circles in 
426: %$\tilde{M}$. 
427: Acting on each such lift by the action of
428: $\pi_1(M)$, we thus see that every geodesic of $(\tilde{M}, [\varpi^* \nabla ])$
429: is an embedded circle., and  
430: $[\varpi^*\nabla ]$ is therefore a Zoll projective structure on 
431: $\tilde{M}$. 
432: 
433: 
434: It remains to show that $[\varpi^*\nabla ]$ is tame. To see this, 
435: first observe that $\mathcal F$ of $\PP TM$  
436: pulls back to the foliation $\hat{\mathcal F}$ of $\PP T\tilde{M}$ given 
437: by lifted geodesics of $[\varpi^*\nabla ]$. Moreover, the induced map
438: $\hat{\varpi}: \PP T\tilde{M}\to \PP TM$
439: is a covering map. If $U\subset \PP TM$ is any connected  open set, 
440: and if $\hat{U}\subset  \PP T\tilde{M}$ is any connected
441: component of $\hat{\varpi}^{-1}(U)$, then $\hat{\varpi}|_{\hat{U}}: \hat{U}\to U$
442: is also a covering map. But if $U$ is  a trivializing neighborhood for  $\mathcal F$,  then the finite cover $\hat{U}$ of $U\approx S^1\times \RR^{2n-2}$
443: will therefore provide a local trivialization of $\hat{\mathcal F}$.
444:  Since $\PP T\tilde{M}$ is covered by 
445: such neighborhoods, this shows that $(\tilde{M}, [\varpi^*\nabla ])$
446: is tame, as claimed. 
447: \end{proof}
448: 
449: This leads to  constraints on the topology of $M$. 
450: 
451: \begin{lem}\label{top}
452: Suppose that the $n$-manifold  $M$ 
453: admits a tame Zoll projective structure $[\nabla]$.
454: Then $M$ is compact, and has finite fundamental group.
455: Moreover, every two points $x$ and $x'$ of $M$ are joined by a 
456: geodesic of $\nabla$. 
457: \end{lem}
458: 
459: 
460: \begin{proof}
461: Choose an arbitrary point  $x\in M$. 
462: In $\PP TM$, consider the union 
463: $$\hat{X}=\nu^{-1}\left(\nu \left[\mu^{-1}(x)\right] \right)$$ of the lifts of geodesics through 
464: $x$. Then $\hat{X}$ is a compact
465: differentiable   $n$-manifold. 
466: But since  $\mu^{-1}(x)\subset \hat{X}$
467: is an $\RR\PP^{n-1}$ whose normal bundle is the universal
468:  line bundle, $\hat{X}$ may be    blown down
469: along $\mu^{-1}(x)$
470: to produce a new compact differentiable  $n$-manifold\footnote{We remark in passing that 
471: it is not difficult to show that ${X}$ is always diffeomorphic to $\RR\PP^n$.}  
472:  ${X}$.
473: Moreover, 
474: $\mu$ induces a differentiable map 
475: ${\wp}:{X}\to M$. Indeed, 
476: if $\check{x}\in X$ denotes the point 
477: obtained by blowing down $\mu^{-1}(x)$, 
478:  then, in a neighborhood of $\check{x}$, 
479: $\wp$ is modeled on the exponential map of $\nabla$
480: near $0\in T_xM$. In particular, $\check{x}$
481: is a regular point of  $\wp$.  But, because $[\nabla ]$ is Zoll,  
482: a  geodesic circle can pass through $x$ 
483: {\em only  once},  
484: so it follows that ${\wp}^{-1}(x)=\{ \check{x}\}$. Thus $x$ is a regular value of 
485: the proper map $\wp$ with $\# {\wp}^{-1}(x)=1$.  This shows that   
486: the $\bmod$-$2$ degree
487: of the proper map $\wp$ is $1\in \ZZ_2$. In particular, 
488:  ${\wp}$ is onto, and 
489:  $M= {\wp}({X})$ is therefore compact. The very definition
490: of the surjective map $\wp$ now tells us that 
491: any point $x'$ of $M$ is joined to $x$ by 
492: some geodesic of $\nabla$.
493: 
494: Since the universal cover $\tilde{M}$ also admits a tame Zoll projective 
495: structure by Lemma \ref{cover}, the above argument now also shows that
496: $\tilde{M}$ is  compact. Hence the universal covering map $\varpi : \tilde{M}\to M$
497: is  finite-to-one, and $\pi_1(M)$ is therefore finite, as claimed. 
498: \end{proof}
499: 
500: Applying this  to the two-dimensional case, we  obtain the following:
501: 
502: \begin{prop} \label{class} 
503: A compact surface $M^2$ admits a  Zoll projective structure iff 
504: $M$ is diffeomorphic to either $S^2$ or $\RR \PP^2$. 
505: \end{prop}
506: \begin{proof}
507: By pulling the projective structure back to a double cover $\tilde{M}$ of
508: $M$ if necessary, we obtain a  Zoll
509: projective structure  on a compact orientable surface $\tilde{M}$, 
510: and this pulled-back structure is then 
511:   tame by 
512: Proposition \ref{meow}. This  forces $\tilde{M}$, and hence $M$, to have 
513:  finite fundamental group 
514: by Lemma \ref{top}. The classification of compact surfaces then tells us that
515: $M$  must be diffeomorphic to 
516: either $S^2$ or $\RR \PP^2$. Conversely, the Levi-Civita connection $\triangledown$
517: of  the standard, homogeneous metric determines a Zoll projective structure 
518: $[\triangledown ]$ on either of these spaces. 
519: \end{proof}
520: 
521: The following information thus becomes pertinent to our discussion: 
522: 
523: \begin{lem}\label{order}
524: If $M=S^2$, $|\pi_1(\PP TM)|=4$. If $M=\RR\PP^2$, 
525: $|\pi_1(\PP TM)|=8$.
526: \end{lem}
527: \begin{proof}
528: The unit bundle of $S^2$ may be identified with  $SO(3)$
529: by thinking of the first column of an orthogonal matrix as a point of
530: $S^2\subset \RR^3$, and the second  column as a unit tangent vector
531: at that point. 
532: Thus  $\PP TS^2$ may  be identified with $SO(3)/\ZZ_2$, where the 
533: $\ZZ_2$ action is generated by left multiplication by
534: $$\left[
535: \begin{array}{ccc}
536: 1&0&0\\0&-1&0\\0&0&-1
537: \end{array}\right] ~.
538: $$
539: Lifting to the universal cover $Sp(1)=S^3\subset \HH^{\times}$ of $SO(3)$, we thus have
540: $\PP TS^2=Sp(1)/\ZZ_4$, where the $\ZZ_4$ is generated by $i$. 
541: Hence $\pi_1(\PP TS^2)\cong \ZZ_4$ has order $4$, as claimed. 
542: 
543: The antipodal map on $S^2$ acts on the unit tangent bundle 
544: via
545: $$
546: \left[
547: \begin{array}{ccc}
548: -1&0&0\\0&-1&0\\0&0&1
549: \end{array}\right] \in SO(3) ~,
550: $$
551: and this lifts to $Sp(1)$ as $\pm k$. 
552: Thus $\PP T\RR\PP^2= Sp(1)/ \{ \pm 1 , \pm i , \pm j , \pm k\}$, and hence 
553: $\pi_1(\PP T\RR\PP^2)\cong \{ \pm 1 , \pm i , \pm j , \pm k\}$
554:  has order $8$, 
555: as claimed. 
556: \end{proof}
557: 
558: In particular, $\pi_1 (\PP TM^2 )$ must be finite.  Hence: 
559: 
560: \begin{prop}
561: Let $(M,[\nabla ])$ be a  compact surface with tame Zoll projective structure. Then its space $N$ 
562: of unoriented geodesics is diffeomorphic to $\RP^2$.
563: \end{prop}
564: \begin{proof}
565: The group homomorphism  
566: $$\nu_{\natural}: \pi_1 (\PP TM)\to \pi_1 (N)$$
567: induced by the fibration $\nu$ is surjective, since  
568:  each fiber of $\nu$ is path connected. But Proposition \ref{class} and 
569: Lemma \ref{order} together 
570: tell us  that 
571: $\PP TM$ has finite fundamental group. Hence $\pi_1(N)$ is finite, 
572: and the classification of $2$-manifolds therefore tells us that $N$ must be 
573: diffeomorphic to either 
574: $S^2$ or $\RR\PP^2$. 
575: But we also know that  $N$ is not simply connected, since it has a non-trivial cover
576: $\tilde{N}$, given by the space of  {\em directed} geodesics of $[\nabla]$. 
577: This shows that $N\approx \RP^2$, as claimed. 
578: \end{proof} 
579: 
580: Next, we would like to understand the  topological structure of  the $S^1$-bundle 
581: $$\nu : \PP TM \to N .$$
582: Our method will  simultaneously allow us to analyze the conjugate points of the 
583: projective structure $[\nabla ]$. Let us thus begin by recalling the notion of a Jacobi field.  
584: 
585: 
586: If $\nabla$ is a connection on a manifold $M$, and if 
587: $c: (a,b) \to M$ is an affinely  parameterized 
588: geodesic of $\nabla$, then a 
589:  {\em Jacobi field} 
590: along   $c$
591: is by definition a vector field 
592: ${\bf y}\in \Gamma (c^*TM)$ 
593: along $c$ 
594: which satisfies the linear differential equation
595: $$\nabla_{\bf v}\nabla_{\bf v}{\bf y}= R_{{\bf v}{\bf y}}{\bf v},$$
596: where $R$ denotes the  curvature tensor of $\nabla$, and where 
597:  the standard tangent vector
598: $${\bf v}=\frac{dc}{dt}$$
599: of our parameterized geodesic 
600: satisfies the {\em auto-parallel condition} 
601: \begin{equation}\nabla_{\bf v} {\bf v} =0.\label{parl}\end{equation}
602: It is not difficult to see that ${\bf y}$ is a Jacobi field iff 
603: it is  locally 
604: the joining vector field for a  $1$-parameter family of geodesics of $\nabla$.
605: More precisely,  for any $[a',b']\subset (a,b)$, there is an  $\varepsilon > 0$ and 
606: a differentiable map \begin{eqnarray*}
607: \hat{c}: [a',b']\times (-\varepsilon, \varepsilon )&\to& M\\
608: (t,u) &\mapsto& \hat{c}(t,u)
609: \end{eqnarray*}
610: with $\hat{c}(t,0)=c(t)$, such that, setting 
611: $$\tilde{{\bf v}}=  \frac{\partial \hat{c}}{\partial t}, ~~ \tilde{{\bf y}} = \frac{\partial \hat{c}}{\partial u},$$
612: one has 
613: $$\nabla_{\tilde{{\bf v}}}\tilde{{\bf v}}=0$$
614: and $$\tilde{{\bf y}}|_{u=0}= {\bf y}.$$
615: 
616: 
617: The notion of a Jacobi field is not actually projectively invariant, but there is a 
618: closely related concept which {\em is}.
619: 
620: \begin{defn} Let $[\nabla ]$ be a $C^1$ projective connection on $M$, and 
621: let ${\mathfrak C}\looparrowright M$  be any geodesic of $[\nabla ]$.
622: Then a section ${\mathfrak Y} $ of the normal bundle $TM/T{\mathfrak C}$ of ${\mathfrak C}$ will be called 
623: a {\em Jacobi class} on ${\mathfrak C}$ iff,
624: near any given point $p\in {\mathfrak C}$,  
625: $${\mathfrak Y} \equiv {\bf y} \bmod T{\mathfrak C}$$
626: for some locally defined Jacobi field ${\bf y}$.
627: \end{defn}
628: In other words, ${\mathfrak Y} $ is a Jacobi class iff it locally joins infinitesimally
629: separated {\em unparameterized} geodesics. Thought of this
630: way, it thus becomes immediately apparent that 
631: the notion of Jacobi class is projectively invariant. 
632: 
633: \begin{defn} Let $[\nabla ]$ be a $C^1$ projective connection on $M$, and 
634: let ${\mathfrak C}\looparrowright M$  be any geodesic of $[\nabla ]$.
635: We will say that two points $p,q\in {\mathfrak C}$ are {\em conjugate}
636: along ${\mathfrak C}$ iff there is a Jacobi class  ${\mathfrak Y} $ on ${\mathfrak C}$ 
637: with ${\mathfrak Y} (p)={\mathfrak Y} (q)=0$. 
638:  \end{defn}
639: Very roughly, conjugate points are thus the places where two 
640: infinitesimally separated geodesics of $[\nabla]$ meet. 
641: 
642: 
643: 
644: 
645: Let us now make all  of this more explicit in the special case of $\dim M=2$. If 
646: ${\mathfrak C}\looparrowright M$ is a  geodesic of an affine connection $\nabla$ on 
647: a surface $M$, the normal bundle $TM/T{\mathfrak C}$
648: is  a real line bundle $E\to {\mathfrak C}$. Since $T{\mathfrak C}\subset TM$ is
649: parallel, $\nabla$ defines a connection ${\mathfrak D}$ on $E$. 
650: Let us take an affine parameterization
651: $c: (a,b)\to {\mathfrak C}$, so that ${\bf v}=dc/dt$
652: satisfies (\ref{parl}). Let us then trivialize $c^*E\to (a,b)$  by means of $[{\bf e}]$,
653: where ${\bf e}\not\propto {\bf v}$ is a generic  parallel section of $c^*TM$, and 
654: where the brackets $[\cdot]$ indicate the equivalence class $\bmod  ~ T{\mathfrak C}$. 
655: Defining $\kappa : (a,b)\to \RR$ by 
656: $$\kappa =  r({\bf v},{\bf v}),$$
657: where $r_{ab}={R^c}_{acb}$
658: is the Ricci tensor of $\nabla$, we then have 
659: $$
660: R_{{\bf v}{\bf e}}{\bf v}\equiv -\kappa {\bf e} \bmod {\bf v}, 
661: $$
662: so that $y(t) {\bf e}\equiv {\bf y}\bmod T{\mathfrak C}$ for some Jacobi field 
663: ${\bf y}$ iff $y : (a,b)\to \RR$ satisfies the second order linear differential equation 
664: \begin{equation}\label{jack}
665: \frac{d^2y}{dt^2}+ \kappa y=0.\end{equation}
666: More abstractly, (\ref{jack}) becomes
667: \begin{equation}\label{lack}
668: {\mathfrak D}_{\bf v}{\mathfrak D}_{\bf v}{\mathfrak Y} + r({\bf v},{\bf v}) {\mathfrak Y} =0\end{equation}
669: in terms of the connection ${\mathfrak D}$ induced on the normal bundle 
670: $E$, and this in turn generalizes to becomes 
671: \begin{equation}\label{quack}
672: {\mathfrak D}_{\bf v}{\mathfrak D}_{\bf v}{\mathfrak Y} -{\mathfrak D}_{\nabla_{\bf v}{\bf v}}{\mathfrak Y} + r({\bf v},{\bf v}) {\mathfrak Y} =0\end{equation}
673: if we drop the auto-parallel condition (\ref{parl}) on our tangent field ${\bf v}$.
674: Let us remark  that if 
675:  $\nabla$ is replaced by the projectively equivalent connection $\hat{\nabla}$
676: defined by 
677: $$\hat{\nabla}_{\bf u}{\bf v} = \nabla_{\bf u}{\bf v}+ \beta ({\bf u}) {\bf v} + \beta ({\bf v}) {\bf u} ,$$
678: one then has
679: \begin{eqnarray*}
680: \hat{\mathfrak D}_{\bf v}\hat{\mathfrak D}_{\bf v} {\mathfrak Y} &=& 
681: {\mathfrak D}_{\bf v} {\mathfrak D}_{\bf v} {\mathfrak Y}
682: + 2 \beta ({\bf v}) {\mathfrak D}_{\bf v} {\mathfrak Y}
683: + \left[{\bf v} \beta ({\bf v})+ \beta ({\bf v})^2 \right] {\mathfrak Y}  ,
684: \\
685: \hat{\mathfrak D}_{\hat{\nabla}_{\bf v}{\bf v}}{\mathfrak Y} &=& 
686: {\mathfrak D}_{{\nabla}_{\bf v}{\bf v}}{\mathfrak Y} + 2 \beta ({\bf v}) 
687: {\mathfrak D}_{\bf v}{\mathfrak Y} 
688: + \left[\beta (\nabla_{\bf v}{\bf v})+ 2 \beta ({\bf v})^2\right] {\mathfrak Y} , \\
689: \hat{r}({\bf v},{\bf v}) &=& r ({\bf v},{\bf v}) + (n-1) 
690: \left[ \beta (\nabla_{\bf v}{\bf v}) - {\bf v} \beta ({\bf v}) + \beta({\bf v})^2\right] , 
691: \end{eqnarray*}
692: so that blind, brute-force calculation does indeed show  that 
693: (\ref{quack}) is   projectively invariant   in dimension $n=2$, as   previously 
694: deduced by pure thought.
695: 
696: 
697: Now the vector space of solutions of (\ref{jack}) is  two dimensional, corresponding to  choices of 
698: $y$ and $y'$ at an arbitrary base-point of the interval $(a,b)$. Let
699: $\{ y_1, y_2\}$ be an arbitrary basis for this solution space, and 
700: consider the Wronskian 
701: $$W(t)= \left|\begin{array}{cc}
702: y_1(t)&y_1'(t)\\y_2(t) & y_2'(t)
703: \end{array}
704: \right|=y_1y_2'-y_2y_1'.
705: $$
706: The differential equation (\ref{jack})
707: then tells us that 
708: \begin{eqnarray*}
709: \frac{dW}{dt}&=&y_1'y_2'+ y_1y_2''-y_2'y_1'-y_2y_1''\\
710: &=&y_1(-\kappa y_2)- y_2(-\kappa y_1) =0,
711: \end{eqnarray*}
712: so that $W(t)$ is constant. Moreover,  this
713: constant must be non-zero, since $y_1$ and $y_2$ have linearly 
714: independent initial values at the base-point. 
715:  The map 
716:  \begin{eqnarray*} 
717:  \phi : (a,b) & \longrightarrow &~~~~\RP^1\\ 
718: t~~~&\mapsto&  [y_1(t):y_2(t)]\end{eqnarray*} 
719: is therefore well defined for all $t$, since $y_1$ and $y_2$ cannot
720:  simultaneous vanish. Moreover, $\phi$ is an {\em immersion}, since
721: $$\frac{d}{dt}\left( \frac{y_1}{y_2}\right) = \frac{W}{y_2^2}
722: ~~~ \mbox{  and  } ~~~\frac{d}{dt}\left( \frac{y_2}{y_1}\right) = -\frac{W}{y_1^2}
723: $$
724: are never zero.  Geometrically, $\phi$ may be interpreted as sending $x\in (a,b)$
725: to the set of Jacobi classes ${\mathfrak Y}$ with ${\mathfrak Y} (x)=0$, since a Jacobi class 
726: $$y(t) = \lambda_2 y_1(t) -\lambda_1 y_2 (t)\not\equiv 0$$
727: vanishes at $x$ iff $[ \lambda_1 : \lambda_2] =\phi (x) := [y_1(x) : y_2(x)]$. 
728: In particular, two points are conjugate along $c\left[(a , b )\right]$ 
729: iff they have the same image under $\phi$. 
730: 
731: 
732: 
733: For a tame $C^k$ Zoll projective structure 
734: $[\nabla ]$ on a surface $M^2$, there are two linearly independent  
735:  Jacobi
736: classes defined along the entirety of any  closed geodesic ${\mathfrak C}$; indeed, if $y\in N$
737: represents ${\mathfrak C}$ in the space of geodesics, $T_yN$ is naturally
738: in one-to-one correspondence with the space of Jacobi classes
739: along ${\mathfrak C}$ via $\mu_*\circ (\nu_*)^{-1}$. 
740: The  above construction thus gives
741: us a $C^{k+1}$ covering map $\phi : {\mathfrak C}\to \RP^1$
742: for every geometrically closed geodesic ${\mathfrak C}$, and this map is uniquely defined
743: modulo the action of $SL(2, \RR )$ on $\RP^1$. 
744: The order  of the covering $\phi : {\mathfrak C} \to RP^1$ will be called the 
745: {\em conjugacy number}
746: of the geodesic, since it exactly counts how many points of ${\mathfrak C}$ are  
747: conjugate to $x\in {\mathfrak C}$, of course including  $x$ itself. We will 
748: now see that this number actually has a rather deeper meaning. 
749: 
750: 
751: \begin{prop}\label{blimey} 
752: Let $[\nabla ]$ be a tame $C^k$ Zoll projective connection, $1\leq k \leq \infty$,
753:  on a compact $2$-manifold $M$, and 
754: consider the $C^{k-1}$ map
755:  \begin{eqnarray*} 
756: \varphi :\PP TM &\longrightarrow& ~~~~\PP TN\\
757: z~~~~&\mapsto& \nu_* \left( \ker \mu_{*z}\right) ,
758:  \end{eqnarray*} 
759: where $\mu_*$ and  $\nu_*$ denote the derivatives of $\mu$ and $\nu$,
760: respectively. Then $\varphi$ is 
761: a  covering map. Moreover, the order of the covering $\varphi$ exactly equals 
762:  the conjugacy number of 
763: any closed geodesic ${\mathfrak C}\subset M$. In particular,
764: all the geodesics of  $[\nabla ]$ have the same conjugacy number. 
765: \end{prop}
766: \begin{proof}
767: Let us first notice that  we have a commutative diagram 
768: \setlength{\unitlength}{1ex}
\begin{center}\begin{picture}(36,19)(0,3)
\put(10,17){\makebox(0,0){$\PP TM$}}
769: \put(18,19){\makebox(0,0){$\varphi$}}
\put(18,5){\makebox(0,0){$N$}}
770: \put(26,17){\makebox(0,0){$\PP TN$}}
\put(15,12){\makebox(0,0){$\nu$}}
771: \put(21,12){\makebox(0,0){$\pi$}}
\put(11,15.5){\vector(2,-3){6}}
772: \put(25,15.5){\vector(-2,-3){6}}
773: \put(14,17){\vector(1,0){8}}
\end{picture}\end{center}
774: where $\pi$ denotes the relevant canonical projection. Moreover, 
775: since $N$ is by definition the leaf space of the 
776:   foliation $\mathcal F$, we also know that $\varphi$ maps  
777: each leaf of  $\mathcal F$ to a different
778:  fiber of $\pi$. 
779: 
780: 
781: Now 
782: the tangent space of $N$ at any point can be canonically
783: identified with the space of Jacobi classes on the corresponding
784: geodesic in $M$. With this identification,   $\varphi$ then sends a point
785: of a  geodesic ${\mathfrak C}$ (identified, by lifting, with a leaf of $\mathcal F$) to the 
786:  set of Jacobi classes which vanish at that point. In other words, 
787: on each leaf of $\mathcal F$, thought of as a geodesic ${\mathfrak C}\subset M$
788: of $[\nabla ]$, $\varphi$ precisely coincides with the map $\phi$ described above. 
789: This shows that $\varphi$ immerses each leaf in $\PP TN$ as a 
790: fiber of $\pi$.  Since $\nu$ is a submersion, it follows, for $k\geq 2$,
791: that $\varphi_*$ is injective, and hence that $\varphi$ is a 
792: local diffeomorphism; for $k=1$, one instead may observe that 
793:  $\varphi$ must be injective on some
794:  neighborhood of any point, and so must be 
795: a local homeomorphism by the  open mapping theorem.
796: But since  $\PP TM$ is compact,   this implies that  $\varphi$ is  a  covering map. 
797: Moreover, the order of this covering is precisely the number
798: of points on a leaf of $\mathcal F$ which are sent to the same point of a fiber 
799: $\pi$. This shows that the order of covering $\varphi$ is precisely the
800: conjugacy number of any geodesic of $[\nabla]$. 
801: \end{proof}
802: 
803: 
804: If $X$ is any manifold,  let us use ${\mathbb S}TX$ to denote the 
805: sphere bundle $(TX- 0_X)/\RR^+$.  In other words, 
806: ${\mathbb S}TX$ may be thought of as the set of unit tangent vectors
807:  for an arbitrary Riemannian metric on $X$. 
808: 
809: \begin{thm} \label{wooster}
810: If $[\nabla ]$ is any $C^k$ Zoll projective structure, $k \geq 1$, on $M\approx S^2$,
811: its conjugacy number is two, and 
812: there is a $C^{k-1}$ diffeomorphism  $\PP TM \approx {\mathbb S}TN$ such that 
813: $\nu$ becomes the canonical projection ${\mathbb S}TN \to N$.
814: Moreover, the real line bundle $\ker \mu_*$ over $\PP TM$
815: is trivial.
816: \end{thm}
817: 
818: 
819: \begin{proof} Let us first recall that Proposition \ref{meow} tells us 
820: that $[\nabla ]$ is tame. But now, 
821: with a nod to  Lemma \ref{order}, we see that 
822: the covering map 
823:  $\varphi : \PP TM \to \PP TN$
824:  has order 
825: $$\frac{|\pi_1(\PP TN )| }{ |\pi_1 (\PP TM )|}=
826: \frac{|\pi_1(\PP T\RR\PP^2 )| }{|\pi_1 (\PP TS^2 )|} =\frac{8}{4} =2,$$
827: and the conjugacy number is therefore $2$, by Proposition \ref{blimey}. 
828: 
829: Now notice that the real line bundle $\ker \mu_*$ over $\PP TS^2$
830: is trivial. Indeed, after the choice of a metric and orientation, 
831: $\PP TS^2$ can be identified with the $SO(2)$ bundle of 
832: oriented orthonormal frames divided by $\langle -{\bf 1}\rangle \subset SO(2)$, 
833: and carries an induced ${\mathfrak so}(2)$ action which trivializes 
834: $\ker \mu_*$.. 
835: 
836: Imitating our  construction of $\varphi$, we now  obtain a diagram 
837: \setlength{\unitlength}{1ex}
\begin{center}\begin{picture}(36,19)(0,3)
\put(10,17){\makebox(0,0){$\PP TM$}}
838: \put(18,19){\makebox(0,0){$\hat{\varphi}$}}
\put(18,5){\makebox(0,0){$N$}}
839: \put(26,17){\makebox(0,0){${\mathbb S}TN$}}
\put(15,12){\makebox(0,0){$\nu$}}
840: \put(21,12){\makebox(0,0){$\wp$}}
\put(11,15.5){\vector(2,-3){6}}
841: \put(25,15.5){\vector(-2,-3){6}}
842: \put(14,17){\vector(1,0){8}}
\end{picture}\end{center}
843: by defining $\hat{\varphi}(z)= \RR^+\nu_*({\bf v}_z)$; here $\wp: {\mathbb S}TN \to N$
844: of course denotes the canonical projection. 
845: Now   $\hat{\varphi}$ is a  covering map, since it  lifts
846: $\varphi$. 
847: But 
848: $$\frac{|\pi_1(S TN )| }{ |\pi_1 (\PP TM )|}=
849: \frac{|\pi_1(S T\RR\PP^2 )| }{|\pi_1 (\PP TS^2 )|} =\frac{4}{4} =1,$$
850: so it now follows that $\hat{\varphi}$ is a homeomorphism if $k=1$, and 
851: a diffeomorphism if $k\geq 2$. 
852: \end{proof}
853: 
854: This finally allows us to definitively  dispense with the tame condition.
855: 
856: \begin{thm} \label{bingo}
857: Any $C^1$ Zoll projective structure  on a compact surface
858: $M^2$ is tame.  
859: \end{thm}
860: \begin{proof} 
861: Proposition \ref{meow} covers   the orientable case, so we  may assume 
862: henceforth that 
863:  $M$ is non-orientable.  Proposition \ref{class} then tells us that 
864: $M$ is diffeomorphic to $\RP^2$, so we have 
865: $M= \tilde{M}/\langle a\rangle$, where 
866: $\tilde{M}\approx S^2$, and where $a :  \tilde{M} \to  \tilde{M}$
867: corresponds to the  antipodal map on $S^2$. 
868: Any Zoll projective structure  on $M$ then pulls
869: back to a tame Zoll projective structure on $\tilde{M}$
870: each of whose geodesics is  sent to some geodesic by $a$.
871: If $\tilde{N}\approx \RP^2$ is the space of 
872: unoriented geodesics of $\tilde{M}$, then $a$ thus induces
873: a diffeomorphism $\hat{a}: \tilde{N}\to \tilde{N}$. We claim  
874: that $\hat{a}$ is in fact the identity. 
875: 
876: Suppose not. Then $\hat{a}$ generates a non-trivial $\ZZ_2$ action.
877: But any  action by a finite group of diffeomorphisms is isometric with respect to some 
878:  Riemannian metric, and so has fixed-point set
879: consisting of a disjoint union of closed submanifolds. In our case, 
880: the fixed-point set would be  a finite  union  of disjoint circles and points. 
881: Moreover, the quotient $\tilde{N}/\ZZ_2$ would have Euler characteristic 
882: $$\chi (\tilde{N}/\ZZ_2) = \frac{\chi (\tilde{N})+m}{2} =  \frac{1+m}{2} ,$$
883: where $m$ is the number of isolated fixed points. Since the Euler
884: characteristic is an integer, this shows that $\hat{a}$
885: has at least one  isolated fixed point. At such an isolated fixed point, the
886: derivative of $\hat{a}$ must be $-1$, as this is   the unique  order-$2$ element of 
887: $O(2)$  with trivial  $+1$-eigenspace. 
888: 
889: Back in $\tilde{M}$, this fixed point would correspond to a geodesic circle 
890: $\mathfrak C$ with $a ({\mathfrak C})={\mathfrak C}$
891: along which $a_*$ induced the action 
892: ${\mathfrak Y} \mapsto -{\mathfrak Y}$ on the vector space of Jacobi classes. 
893: In particular, the zero locus of a Jacobi class ${\mathfrak Y}\not\equiv 0$
894: would  necessarily be sent to itself by $a_*$. But since $\tilde{M}$ has
895: conjugacy number $2$ by Theorem \ref{wooster}, and since $a$ has no fixed points, 
896: this means that $a$ acts 
897: on $\mathfrak C$ by sending each point to the  unique other point
898: to which it is conjugate. Now trivialize the normal bundle $E= T\tilde{M}/T{\mathfrak C}$,
899: so that we can talk about whether a non-zero element of $L$ is `positive'
900: or `negative'. Then, since any Jacobi class ${\mathfrak Y}\not\equiv 0$
901: meets the zero section of $E$ transversely in exactly $2$ points, 
902: the subsets of $\mathfrak C$ given by ${\mathfrak Y} > 0$ and
903: ${\mathfrak Y} < 0$ are necessarily intervals, and are necessarily 
904: interchanged by the fixed-point-free map $a$. But since 
905: ${\mathfrak Y} \mapsto -{\mathfrak Y}$
906: under $a_*$, this shows that $a_*$ acts on the normal bundle $E$
907: in an {\em orientation-preserving} manner. Moreover, $a_*$ is also
908: orientation-preserving on $T{\mathfrak C}$, since $a : {\mathfrak C}
909: \to {\mathfrak C}$ has no fixed point. 
910:  Hence $a$ acts on $\tilde{M}$ in an orientation-preserving 
911: manner --- contradicting the fact that, by construction,  $a$ is an orientation-reversing map!
912: 
913: This contradiction shows that $\hat{a}$ must be the identity on $\tilde{N}$. Hence 
914: $a_*$ induces an action on 
915: $\PP T\tilde{M}$ which sends each leaf to itself, and  holonomy around any leaf  
916: in $\PP TM$ is therefore trivial. Hence the given Zoll projective structure on
917: $M\approx \RP^2$ is tame, as claimed. 
918:  \end{proof}
919: 
920: In particular, it now makes sense to talk about the 
921: conjugacy number of any Zoll projective structure 
922: on $\RP^2$. 
923: 
924: \begin{thm} \label{jeeves} 
925: If $[\nabla ]$ is any $C^k$  Zoll projective structure, $k\geq 1$, 
926:  on $M\approx \RR\PP^2$,
927: its conjugacy number is $1$. Moreover, 
928: there is a $C^{k-1}$ diffeomorphism $\PP TM \approx \PP TN$ such that 
929: $\nu$ becomes the canonical projection $\PP TN \to N$, and such that 
930:  $\ker \mu_*\to\PP TM$
931: becomes  the `tautological' real line bundle
932:  $L \to \PP TN$, whose  frame bundle bundle is the principal $\RR^\times$-bundle 
933: $(TN - 0_N )\to \PP TN$. 
934: \end{thm}
935: 
936: \begin{proof}
937: Since $[\nabla ]$ is tame by Theorem \ref{bingo}, we are free to  
938: consider the covering map $\varphi : \PP TM \to \PP TN$
939: of Proposition \ref{blimey}. By construction,  the tautological
940: line bundle $L\to \PP TN$ then satisfies $\varphi^*L=\ker \mu_*$.
941: Since 
942: the order of this covering is $$\frac{|\pi_1(\PP TN )| }{ |\pi_1 (\PP TM )|}=
943: \frac{|\pi_1(\PP T\RR\PP^2 )| }{|\pi_1 (\PP T\RR\PP^2 )|} =1,$$
944: we conclude that  $\varphi$ is a homeomorphism, and  
945:  the conjugacy number is therefore $1$ by Proposition \ref{blimey}. 
946: Moreover,  the same argument also shows that $\varphi$ is actually a  diffeomorphism
947: if $k\geq 2$.
948:  \end{proof}
949: 
950: \begin{cor} \label{lcapl}
951: For  any  Zoll projective structure  $[\nabla ]$ on $M\approx \RR\PP^2$,
952: any two distinct points are joined by a unique geodesic circle $\mathfrak C$. 
953: \end{cor} 
954: \begin{proof}
955: As in the proof of Lemma \ref{top}, 
956: let 
957: $$\hat{X}=\nu^{-1}\left(\nu \left[\mu^{-1}(x)\right] \right)$$ be the 
958: union of the lifts of geodesics through 
959: $x$. Then $\hat{X}$ is a compact
960: differentiable   surface and may be    blown down
961: along $\mu^{-1}(x)$
962: to produce a new smooth compact  surface $X$.   
963: Since $\hat{X}$ is a circle bundle over the circle $\ell_x =  \nu \left[\mu^{-1}(x)\right]$,
964: and since a neighborhood of  $\mu^{-1}(x)$ is a M\"obius band $B$, 
965: it follows that 
966:  $X$  contains a  M\"obius band $B'=\hat{X}-B$, 
967: and hence  is not orientable. 
968: 
969:  On the other hand, Theorem \ref{jeeves}
970: tells us that each geodesic in $M$ has conjugacy number $1$, and
971: hence  no point $x'\neq x$ is conjugate to $x$ along any geodesic. 
972: Hence the canonical projection $\hat{X}\to M$ is an  immersion 
973: away from $\mu^{-1}(x)$, and the induced map $\wp :X\to M$ is therefore
974: an immersion everywhere.  Since $X$ is compact, $\wp$  is therefore
975:  a covering map. Since $X$ is not simply connected and 
976: $\pi_1 (M) = \ZZ_2$, it follows that $\wp$ is a one-to-one and onto. 
977: But, by the very definition of $\wp$, this means that there is one 
978: and only one geodesic between $x$ and any other point $x'\neq x$
979: in $M$. 
980: \end{proof}
981: 
982: 
983: 
984: \begin{cor} \label{lowercase}
985: Let $(M^2, [\nabla ])$ be a compact surface with Zoll projective structure.
986: Let ${\mathfrak C}\subset M$ be any geodesic circle.  
987: Then the following conditions are equivalent:
988: \begin{itemize}
989: \item $\langle w_1 (M) , [{\mathfrak C}]\rangle = 1 \in \ZZ_2$;
990: \item the conjugacy number of ${\mathfrak C}$ is odd;
991: \item $M$ is not orientable;
992: \item $M$ is diffeomorphic to $\RP^2$.
993: \end{itemize}
994: \end{cor}
995: 
996: 
997: \begin{proof}
998: At points where a Jacobi class ${\mathfrak Y}\not\equiv 0$ 
999: vanishes along ${\mathfrak C}$, the covariant derivative ${\mathfrak D}_{\bf v}{\mathfrak Y}$
1000: must be nonzero, since ${\mathfrak Y}$ satisfies (\ref{jack}). Thus the 
1001: mod-$2$ reduction of the conjugacy
1002: number of ${\mathfrak C}$ calculates $\langle w_1 (E) , [{\mathfrak C}]\rangle$, where 
1003: $E=TM/T{\mathfrak C}$ is the normal bundle, and this of course coincides with 
1004:  $\langle w_1 (M) , [{\mathfrak C}]\rangle := \langle w_1 (TM) , [{\mathfrak C}]\rangle$, 
1005: since $T{\mathfrak C}$ is trivial. But Theorems \ref{jeeves} and \ref{wooster} 
1006: tell us that the only possible values of the conjugacy number are $1$ and $2$, and 
1007: that the value of the conjugacy number determines whether $M$ is diffeomorphic
1008: to $\RP^2$ or $S^2$. 
1009: \end{proof}
1010: 
1011: The same argument also yields the following: 
1012: 
1013: \begin{cor}\label{uppercase}
1014: Let $(M^2, [\nabla ])$ be a compact surface with Zoll projective structure.
1015: Let ${\mathfrak C}\subset M$ be a geodesic circle.  
1016: Then the following conditions are equivalent:
1017: \begin{itemize}
1018: \item $\langle w_1 (M) , [{\mathfrak C}]\rangle = 0\in \ZZ_2$;
1019: \item the conjugacy number of ${\mathfrak C}$ is even;
1020: \item $M$ is orientable;
1021: \item $M$ is diffeomorphic to $S^2$.
1022: \end{itemize}
1023: \end{cor}
1024: 
1025: Let us now take a moment  to  compare our definitions with those 
1026: previously used by others in the Riemannian context \cite{beszoll,guillzoll}. 
1027: 
1028: \begin{prop}\label{others}
1029: Let $(M^2,g)$  be a compact surface with $C^k$ Riemannian metric, $2\leq k\leq\infty$. Let
1030: $\triangledown$ be the Levi-Civita connection of $g$. Then $[\triangledown]$ is a 
1031: $C^{k-1}$  Zoll projective
1032: structure on $M$ iff the geodesics of $g$ are all simple  
1033: closed curves of equal length. 
1034: \end{prop} 
1035: \begin{proof}
1036: If $[\triangledown]$ is  a Zoll projective structure,  Theorem \ref{bingo} then
1037: tells us it is tame,
1038: and its geodesic circles are therefore freely homotopic to one
1039: another {\em through  geodesic circles}. 
1040: But the affinely parameterized closed geodesics of $g$ are
1041: precisely those differentiable maps $c: S^1 \to M$ which are critical points of 
1042: the energy functional
1043: $$E( c ) = \int_{S^1} g(  c^\prime (t) ,  c^\prime (t) ) dt ~;$$
1044:  thus the energy is necessarily constant for any $1$-parameter
1045: family of closed geodesics. This shows that  the 
1046: geodesic circles of $g$ must all have equal 
1047: energy,  and hence equal 
1048: length. 
1049: \end{proof}
1050: 
1051: 
1052: 
1053: 
1054: We conclude this section with an aside which  plays no r\^ole whatsoever 
1055: in what follows, but which, in light of Proposition \ref{others},
1056:  has a certain intrinsic interest. 
1057: Given a  Zoll projective structure $[\nabla ]$ on a compact 
1058: surface $M$, it is natural to  ask whether there is a connection $\nabla$ 
1059: representing $[\nabla ]$ such that every affinely parameterized geodesic
1060:  is {\em
1061: periodic}. 
1062: The answer is affirmative.
1063: 
1064: \begin{prop}
1065: If $[\nabla ]$ is any Zoll projective structure on a compact surface $M^2$,
1066: then there is a symmetric affine connection $\nabla\in [\nabla ]$ 
1067: for which each affinely parameterized geodesic extends as a 
1068: periodic function $c: \RR \to M$. 
1069: \end{prop}
1070: 
1071: \begin{proof} If
1072: $M=S^2$, let 
1073: $\omega$ be an arbitrary area form on $M$, and let $\nabla$ be \cite{schouten} the 
1074: unique connection in the equivalence class such that $\nabla \omega =0$.
1075: If $c: [a,b]\to M$ is an affine parameterization of a geodesic of $\nabla$,
1076: with $c(b)=c(a)$ and $c^{\prime}(b)=\lambda c^{\prime}(a)$, then
1077:  any   parallel vector field ${\bf e}$ along $c$
1078:  must satisfy ${\bf e}(b)=  \lambda^{-1} {\bf e}(a)\bmod c^{\prime}$. 
1079: Now the Zoll  condition guarantees the 
1080: existence of  a two-parameter family of  solutions of  (\ref{lack}) which  satisfy the
1081: "periodicity" condition 
1082: $${\mathfrak Y}|_{c (b)} = {\mathfrak Y} |_{c(a)}, 
1083: ~~ {\mathfrak D}{\mathfrak Y} |_{c (b)}  = {\mathfrak D} {\mathfrak Y}  |_{c(a)} .$$
1084:  Every solution of (\ref{jack})  
1085: must therefore 
1086: satisfy $$y(b)=\lambda y(a), ~~~ y^\prime (b) = \lambda^2 y^\prime (a).$$ 
1087: Hence the Wronskian $W= y_1y_2^\prime - y_2y_1^\prime$ of two linearly
1088: independent solutions of (\ref{jack}) must  satisfy 
1089: $W(b)=\lambda^3 W (a)$. But $W$ is constant! Thus $\lambda = 1$, and 
1090: the given geodesic is therefore periodic. But this argument applies to {\em any}
1091: geodesic on $M$. 
1092: Hence every geodesic of 
1093: the chosen connection $\nabla$  is   periodic, and the claim follows if $M=S^2$. 
1094: 
1095: The case of $\RR\PP^2$ now follows easily; one simply  takes the area form 
1096: $\omega$ on $S^2$  to be anti-invariant under the antipodal map  
1097: $a : S^2\to S^2$, and then notices that the corresponding connection 
1098: $\nabla$ then descends to 
1099: $\RR\PP^2$. 
1100: \end{proof}
1101: 
1102: 
1103: 
1104: %\pagebreak 
1105: 
1106: 
1107: \section{The Blaschke  Conjecture Revisited}
1108: \label{rip}
1109: 
1110: If $[\nabla ]$ is a Zoll projective structure on a compact surface $M$, 
1111: we saw in \S \ref{prelim}  that its space of unoriented geodesics $N$
1112: is diffeomorphic to $\RP^2$.  Now notice that $N$ also comes
1113: equipped with a family 
1114: $$\ell_x = \nu [ \mu^{-1} (x) ]$$
1115: of embedded circles $\ell_x\subset N$, $x\in M$. (For any given $x\in M$, 
1116: this is to say that 
1117:  $\ell_x$  consists  precisely of the geodesics  passing through $x$.)
1118: If we were simply given $N$ and this family of curves, we could then 
1119: completely reconstruct the given projective structure on $M$. 
1120: Indeed, $M$ could be redefined as the parameter space or `moduli space' 
1121: of these curves $\ell_x$, and the geodesics ${\mathfrak C}_y \subset M$  would then become the
1122: set of curves $\ell_x$ passing through some given point $y\in N$. 
1123: The utility of this point of view might seem to be rather questionable, however,
1124: as there is no obvious  geometric structure one might impose on $N$ 
1125: in order to keep track of which embedded circles $\ell \subset N$ are to be the 
1126: elements of the family $\{ \ell_x \}_{x\in M}$.  However, our main observation,
1127: extrapolated from  a twistor correspondence due to 
1128: Hitchin \cite{hitproj} and the first author \cite{lebthes}, 
1129: is that one {\em can} naturally keep track of these curves by `complexifying' the
1130: picture, and embedding $N$ in a complex $2$-manifold ${\mathcal N}$. 
1131: 
1132: 
1133: 
1134: Let us suppose we are given a $C^2$ Zoll projective structure $[\nabla ]$
1135: on $M=\RP^2$.
1136:  Consider the $\CP_1$-bundle 
1137: $$\PP T_\CC M = \left( \CC \otimes TM - 0_M
1138: \right)
1139: /\CC^\times,$$
1140: and observe that the circle bundle
1141: $$\PP T M = \left(  TM - 0_M
1142: \right)
1143: /\RR^\times $$
1144: is a hypersurface in the $4$-manifold $\PP T_\CC M$. 
1145: For brevity, we introduce the notation
1146: $${\mathcal Z}= \PP T_\CC M, ~~ Z = \PP TM.$$
1147: Because each fiber of $\PP T_\CC M$ has a canonical 
1148: complex structure $J^\parallel$, the normal bundle of $\PP T M\subset \PP T_\CC M$
1149: is just $J^\parallel (\ker \mu_*)$, where $\mu : \PP T M \to M$
1150: is the bundle projection. Now recall that  
1151: our Zoll projective structure gives us a foliation 
1152: $\mathcal F$ of $\PP T M$ by circles,  and the leaves of 
1153: $\mathcal F$ are precisely the fibers of a $C^2$ submersion 
1154: $\nu : \PP T M \to N\approx \RP^2$. Moreover, 
1155: Theorem \ref{jeeves} tells us that there is a $C^1$ 
1156: diffeomorphism $\varphi : \PP T M \to \PP T N$ 
1157: such that the real line bundle $\ker \mu_*$ 
1158: becomes the pull-back $\varphi^* L$ of the 
1159: tautological line bundle $L\to  \PP T N$.
1160: The latter line bundle is by definition a sub-bundle of 
1161: $\pi^* TN$, where $\pi : \PP TN \to N$ is the canonical 
1162: projection; namely, for any non-zero vector ${\bf v}\in T_yN$, 
1163: the fiber over $[{\bf v} ] \in \PP TN$ is 
1164: $L_{[{\bf v}]}= \mbox{span} ({\bf v}) \subset T_yN$. In particular, 
1165: there is a tautological $C^1$ `blowing down' map 
1166: $\psi : L\to TN$ which is a diffeomorphism 
1167: away from  the zero section $\PP TN$ 
1168: of $L$, but collapses this zero section to the zero section
1169: $N$ of $TN$ via $\pi : \PP TN \to N$. 
1170: On the other hand,  the tubular neighborhood theorem tells
1171: us that $Z=\PP T M$ has a neighborhood $\hat{\mathcal V}$ in ${\mathcal Z}=\PP T_\CC M$
1172: which is $C^\infty$ diffeomorphic to the total space of 
1173: $J^{\parallel}\ker \mu_*$, in such a manner that the derivative along 
1174: $Z$ is the identity. Letting $\mathcal V$ denote the total space of 
1175: $TN=T\RP^2$, we then have a $C^1$ map $\tilde{\psi} : \hat{\mathcal V}\to {\mathcal V}$
1176: which corresponds to $\psi$ via our $C^1$ diffeomorphism $J^{\parallel}\ker \mu_*\to L$. 
1177: We may now define a 
1178:  new $C^1$ compact $4$-manifold 
1179: $${\mathcal N} = {\mathcal U} \cup_{\tilde{\psi}} {\mathcal V}$$
1180: by gluing together ${\mathcal U} := {\mathcal Z}-Z$
1181: and ${\mathcal V}=TN$ via $\tilde{\psi}$. 
1182:  By construction, 
1183: we also  have a $C^1$ 
1184: `blowing down' map 
1185: $$\Psi : {\mathcal Z} \to {\mathcal N},$$
1186: given by the identity on ${\mathcal U}$ and by $\tilde{\psi}$ on $\hat{\mathcal V}$. 
1187: 
1188: %Even though we have only given ${\mathcal N}$ a $C^1$ structure, 
1189: %it nonetheless admits  charts in which $\Psi$ appears to be more regular:
1190: 
1191: %Working lemma
1192: 
1193: %\begin{lem}
1194: %Let $[\nabla ]$ be a $C^2$ Zoll projective connection on $M=\RP^2$,
1195: %and let $\Psi : {\mathcal Z}\to {\mathcal N}$ be constructed as above. 
1196: %Then every point of $N$  is contained in the domain $U$ of some  
1197: %$C^1$ coordinate chart $\varsigma: U\to \RR^4$ such that 
1198: %$\varsigma \circ \Psi$ is  $C^2$.
1199: %\end{lem}
1200: 
1201: %Note that this Lemma does {\em not} make ${\mathcal N}$ into a
1202: %$C^2$ manifold, because the special charts we have constructed
1203: %need not be related to each other by $C^2$ transition functions. 
1204: 
1205: If we suppose that $[\nabla ]$ is
1206: $C^k$ for $k > 2$, the above construction allows us to  impose a  
1207: $C^{k-1}$ structure on ${\mathcal N}$  
1208: in such a manner that $\Psi$ becomes a $C^{k-1}$ map. 
1209: While this will actually turn out to be  technically useful, 
1210: the reader should be warned, however, that such a 
1211: $C^{k-1}$ structure is in no sense be
1212: {\em natural} or {\em canonical},  because 
1213: it depends on the $(k-1)$-jet of 
1214:  our identification of the tubular neighborhood $\hat{\mathcal V}$ 
1215: with $L\to \PP TN$, and such a choice 
1216:   is  uniquely specified  by the geometry only when $k=2$;
1217: for this reason, we will refer to such  a choice  as a {\em provisional}  $C^{k-1}$ structure. 
1218: Fortunately, however, this apparent shortcoming will 
1219: soon  be remedied. 
1220: Indeed, the thrust of our argument is that  that $[\nabla ]$ induces  a 
1221: certain complex structure $J$ on $\mathcal N$, 
1222: and so endows 
1223: $\mathcal N$  with a canonical $C^{\infty}$  structure.
1224: In order to see this, we will proceed by first constructing a  certain 
1225:  involutive complex distribution ${\bf D}$ on $\PP T_\CC M$,
1226: and then analyzing its image under $\Psi$. 
1227: 
1228: 
1229: Since ${\mathcal Z}= \PP T_\CC M$, we have a bundle projection,
1230: which we will denote by 
1231:  $\hat{\mu}: {\mathcal Z}\to M$. The sub-bundle ${\bf V} = \ker \hat{\mu}_*\subset T{\mathcal Z}$
1232: will be called the {\em vertical sub-bundle}.  Now
1233: choose a connection 
1234:  $\nabla$ representing
1235: the given projective structure $[\nabla ]$,
1236: and let ${\bf H}\subset T{\mathcal Z}$ be the {\em horizontal sub-bundle},
1237: corresponding to parallel transport with respect to $\nabla$,
1238: so that we have a 
1239:  direct-sum decomposition
1240: $$
1241: T{\mathcal Z} = {\bf V}\oplus {\bf H}.
1242: $$
1243: Complexifying these bundles, we thus have
1244: $$T_\CC {\mathcal Z} = {\bf V}_\CC\oplus {\bf H}_\CC ,$$
1245: where $T_\CC  {\mathcal Z} = \CC \otimes T{\mathcal Z}$, etc. 
1246: Notice that the derivative of the projection also gives us a 
1247:  canonical isomorphism
1248: $$\hat{\mu}_*: {\bf H}_\CC \stackrel{\cong}{\longrightarrow} \hat{\mu}^*T_\CC M.$$
1249: 
1250: 
1251: Using this picture, we will now  define two 
1252: line sub-bundles 
1253: $${\bf L}_j \subset T_\CC {\mathcal Z} = \CC \otimes T{\mathcal Z}, ~~
1254: j=1,2.$$ To this end, let us  
1255: first  recall that each fiber of ${\mathcal Z}\to M$ 
1256: is a $\CP_1$, so that we  have a fiber-wise complex structure tensor
1257: $$J^\parallel : {\bf V}\to {\bf V}, ~~ (J^\parallel)^2= - {\bf 1},$$
1258: and we  define ${\bf L}_1\subset {\bf V}_\CC$ to be the $(-i)$-eigenspace of 
1259: $J^\parallel$:
1260: $${\bf L}_1 = {\bf V}^{0,1}_{J^\parallel}.$$
1261: On the other hand, each element of ${\mathcal Z} = \PP T_\CC M$
1262: may be identified with a $1$-dimensional complex-linear subspace 
1263: of $T_\CC M$, and 
1264:  this picture gives us  
1265: a tautological line sub-bundle  ${\bf L}_2$ of ${\bf H}_\CC\cong \hat{\mu}^* T_\CC M$:
1266: \begin{equation}\label{dumbo}
1267: \left. {\bf L}_2\right|_{[{\bf w}]} = (\hat{\mu}_{*[{\bf w}]})^{-1}( \mbox{span } {\bf w}).
1268: \end{equation}
1269: Set \begin{equation}
1270: \label{recipe}
1271: {\bf D}= {\bf L}_1\oplus {\bf L}_2 \subset T_\CC {\mathcal Z} .
1272: \end{equation}
1273: Then ${\bf D}$ is a $C^2$ distribution of complex $2$-planes on ${\mathcal Z}$.
1274: We will now see that ${\bf D}$ is involutive, in the sense that 
1275: $$[ C^1 ({\bf D}) , C^1 ({\bf D})] \subset C^{0} ({\bf D}).$$
1276: Moreover,  ${\bf D}$ will turn out to be  unchanged if
1277: we replace $\nabla$ with a projectively equivalent 
1278: connection $\hat{\nabla}$. 
1279: 
1280: Indeed, let  $(x^1, x^2): {\Omega} \to \RR^2$ be a local coordinate system 
1281: on ${\Omega}\subset M$, and let
1282: $$
1283: \Gamma^j_{k\ell }= \left\langle ~dx^j ~,~ \nabla_{\frac{\partial}{\partial x^k}}{\frac{\partial}{\partial x^\ell}}~\right\rangle
1284: $$
1285: be the corresponding Christoffel symbols of the connection 
1286: $\nabla$. We can then introduce local coordinates
1287: $(x^1, x^2, \zeta ) : \hat{\mu}^{-1} ({\Omega}) \to \RR^2 \times \CC$
1288: on $\hat{\mu}^{-1} ({\Omega})\subset {\mathcal Z}$ by 
1289: $$
1290: \left[
1291: \left. 
1292: \left(
1293: \frac{\partial}{\partial x^1} + \zeta \frac{\partial}{\partial x^2}
1294: \right)\right|_{(x^1 , x^2)}
1295: \right]   \longleftrightarrow (x^1,x^2,\zeta ) .$$ 
1296: Then, in these coordinates, 
1297: ${\bf L}_1$ is spanned by $\partial/\partial \bar{\zeta}$, whereas 
1298: ${\bf L}_2$ is spanned  by 
1299: $$
1300: \Xi_0 = \frac{\partial}{\partial x^1}+ \zeta \frac{\partial}{\partial x^2}+ 
1301: Q(x,\zeta , \zeta ) \frac{\partial}{\partial \zeta} + Q (x, \zeta , \overline{\zeta} )
1302:  \frac{\partial}{\partial \overline{\zeta}}, 
1303: $$
1304: where 
1305: $$Q(x, u, v) = 
1306:  - \Gamma^2_{11} - \Gamma^2_{12} (u+v) -\Gamma^2_{22}uv 
1307: + \Gamma^1_{11} v+ \Gamma^1_{12}v(u+v)
1308: +  \Gamma^1_{22}uv^2$$
1309: encodes  the Christoffel symbols $\Gamma^j_{k\ell}$ of our
1310: chart, which are of course functions of $x=(x^1,x^2)$. 
1311: In particular, $\bf D$ is spanned by $\partial /\partial \overline{\zeta}$ and 
1312: \begin{equation}
1313: \Xi = \frac{\partial}{\partial x^1}+ \zeta \frac{\partial}{\partial x^2}+ 
1314: P(x,\zeta ) \frac{\partial}{\partial \xi},\label{zuppa}
1315: \end{equation}
1316: where $\zeta = \xi + i \eta$ and where
1317: \begin{equation}
1318: P(x,\zeta)   = Q(x,\zeta, \zeta) = - \Gamma^2_{11}
1319: +\left[\Gamma^1_{11}-2 \Gamma^2_{12}\right]\zeta
1320: + 
1321: \left[ 2\Gamma^1_{12}-\Gamma^2_{22}\right]\zeta^2
1322: + \Gamma^1_{22}\zeta^3
1323: \label{pesce}
1324: \end{equation}
1325: is evidently of the same differentiability class as $\nabla$. 
1326: But 
1327: $$\left[ ~\frac{\partial}{\partial \overline{\zeta}} ~,~\Xi~
1328: \right ]
1329: =
1330: \left[ ~ \frac{\partial}{\partial \overline{\zeta}} ~ ,~ \frac{\partial}{\partial x^1}+ \zeta \frac{\partial}{\partial x^2}+ 
1331: P(x,\zeta ) \frac{\partial}{\partial \xi} ~ \right ]=0 ~,$$ 
1332: because 
1333: $$\frac{\partial}{\partial \overline{\zeta}}\zeta= 0, \hspace{1cm}
1334:  \frac{\partial}{\partial \overline{\zeta}}P(x,\zeta ) =0.$$
1335: It therefore follows that 
1336: ${\bf D}= \mbox{span } \{ \Xi , \partial /\partial \bar{\zeta} \}$
1337: is involutive, as claimed. 
1338: 
1339: 
1340: Notice that  the replacement 
1341: $$\Gamma^i_{jk} \rightsquigarrow \Gamma^i_{jk} + \delta^i_j\beta_k + \beta_j \delta^i_k$$
1342: leaves  $P(x,\zeta)$ unaltered. Thus replacing $\nabla$ with 
1343: a projectively equivalent  connection $\hat{\nabla}$ leaves  $\Xi$ unchanged, and 
1344:  ${\bf D}=\mbox{span} \{ \Xi , \partial/\partial \overline{\zeta}\}$ 
1345: is   therefore 
1346: projectively invariant.  
1347: 
1348: 
1349: The distribution ${\bf D}$ does not quite define a complex structure on ${\mathcal Z}$, 
1350: because  certain real tangent vectors  are elements of 
1351: ${\bf D}$. Indeed, notice that, because ${\bf D}$ is the direct sum of ${\bf L}_1\subset {\bf V}_\CC$ and 
1352: ${\bf L}_2 \subset {\bf H}_\CC$, and because the projections $T_\CC M \to  {\bf V}_\CC$
1353: and $T_\CC M \to  {\bf H}_\CC$ commute with complex conjugation, any  real
1354: element of ${\bf D}$ must have real components in ${\bf L}_1$ and ${\bf L}_2$. 
1355: But since ${\bf L}_1$ contains no non-zero real element, we therefore have
1356: $$
1357: {\bf D}\cap \overline{{\bf D}} = 
1358: ({\bf L}_1\cap \overline{\bf L}_1)+ ({\bf L}_2\cap \overline{\bf L}_2) = 
1359: {\bf L}_2\cap \overline{\bf L}_2.$$
1360: On the other hand, equation (\ref{dumbo}) tells us that 
1361: ${\bf L}_2$ contains a non-zero real element precisely at the
1362: hypersurface $Z=\PP TM$ in ${\mathcal Z}=\PP T_\CC M$: 
1363: \begin{equation}
1364: \dim ({\bf D}_z\cap \overline{{\bf D}}_z ) = 
1365: \left\{ \begin{array}{cl}
1366: 0, & z\not\in Z \\
1367: 1, & z\in Z .
1368: \end{array}
1369: \right.
1370: \label{acs}
1371: \end{equation}
1372: Indeed,  ${\bf L}_2|_{Z}$ is simply 
1373: the complexification 
1374: $\CC \otimes \ker \nu_*$ of the  tangent space 
1375: of the foliation $\mathcal F$ of $\PP TM$ by lifted geodesics. 
1376: This  observation gives a somewhat more
1377: geometric explanation for the previously noted projective 
1378: invariance of ${\bf D}$. Indeed,  in equation 
1379: (\ref{pesce})
1380:  we   carefully chose our  complex vector field  $\Xi$  so that  at the locus $Z$, 
1381: given by $\eta =0$, 
1382: $\Xi$  is real and tangent to $\mathcal F$, with coefficients that 
1383: are holomorphic in $\zeta=\xi + i \eta$, and so determined by the behavior of $\Xi$ along
1384: $\eta =0$. 
1385: 
1386: 
1387: \begin{prop}\label{key} Let $[\nabla]$ be a Zoll projective structure which is 
1388: represented by a $C^3$ connection $\nabla$ on $M\approx \RP^2$. Then 
1389: there is a unique integrable almost-complex structure $J$ on $\mathcal N$ such that 
1390: $$\Psi_* [{\bf D}] \subset T^{0,1}({\mathcal N}, J).$$
1391: The unique $C^\infty$  structure on ${\cal N}$
1392: associated with its maximal  atlas of  $J$-compatible complex charts
1393: is compatible with the previously-constructed $C^1$ structure on 
1394: $\mathcal N$, so that  $\Psi : {\mathcal Z} \to {\mathcal N}$  remains a $C^1$ map relative to 
1395: this smooth structure; moreover, 
1396: $\Psi$ actually becomes  $C^3$ on the open dense set ${\mathcal Z}-Z$.
1397: Moreover, if $[\nabla ]$ is  represented by a 
1398: $C^{k,\alpha}$ connection $\nabla$ on $M$,  $3\leq k\leq \infty$,
1399:   $0< \alpha < 1$, 
1400: and if ${\cal N}$ is again given the natural $C^\infty$  structure
1401: associated with  $J$,
1402: then  $\Psi : {\mathcal Z} \to {\mathcal N}$ is actually a $C^{k+1,\alpha}$ map
1403: on  ${\mathcal Z}-Z$. 
1404: \end{prop}
1405: 
1406: \begin{remark}
1407: With the same hypotheses, 
1408: we will later also show (remark, page \pageref{tricky}) that
1409: $\Psi$ is actually   $C^{k+1,\alpha}$ on {\em all} of $\mathcal Z$.
1410: \end{remark}
1411: 
1412: \begin{proof}
1413: We begin by defining $J$ point-wise. 
1414: On the open set ${\mathcal N} -N = \Psi ({\mathcal Z}-Z)$, we may do this by 
1415: first observing that 
1416: $$
1417: T_\CC ({\mathcal N} -N) = \Psi_*{\bf D} \oplus \overline{\Psi_*{\bf D}} 
1418: $$
1419: by (\ref{acs}) and the fact that $\Psi|_{{\mathcal Z}-Z}$ is a diffeomorphism;
1420: on ${\mathcal N}-N$, we now set 
1421: $$J=
1422: \left[
1423: \begin{array}{cc}
1424: -i& 0\\
1425: 0& +i
1426: \end{array}\right]
1427: $$
1428: with respect to this direct sum decomposition. 
1429: On the other hand, since ${\mathcal V}\subset {\mathcal N}$
1430: is, by definition,  a copy of the total space of $TN\to N$, 
1431: we have 
1432:  a canonical identification 
1433: $$T{\mathcal N}|_N = TN\oplus TN,$$ where the first factor is tangent to
1434: $N$, and where the second factor is transverse to it; 
1435: and along $N\subset {\mathcal N}$ we  can therefore  set 
1436: $$J=
1437: \left[
1438: \begin{array}{cc}
1439: 0& -1\\
1440: 1& 0
1441: \end{array}\right]
1442: $$
1443: with respect to this second direct sum decomposition. 
1444: This defines the almost complex structure $J$ at all points of $\mathcal N$.
1445: 
1446: While it is not yet even yet clear that this $J$ is  
1447: continuous, 
1448:  it is at least  easy to see that $\Psi_*{\bf D} \subset T^{0,1}({\mathcal N}, J)$. Indeed, 
1449: by construction, $\Psi_*{\bf D} = T^{0,1}({\mathcal N}, J)$ away from $N$. 
1450:  On the other hand, $\Psi_*{\bf D} = \Psi_*{\bf V}^{0,1}$ along $Z$, and since 
1451: we used $J^\parallel$ to pick out the normal factor of $T{\mathcal Z}|_{Z}= TZ \oplus L$
1452: before blowing down, 
1453:  $\Psi_*\circ J^\parallel= J \circ \Psi_*$ on ${\bf V}|_{Z}$,  and it
1454: follows that $\Psi_*{\bf D} \subset  T^{0,1}({\mathcal N}, J)$ along $Z$, too. 
1455: Moreover, $J$ is certainly the {\em only} almost-complex structure 
1456: with this property, since, for any $y\in N$,   $$T_y{\mathcal N}= 
1457: \Psi_* {\bf V}_x\oplus \Psi_*{\bf V}_{x'}$$
1458: whenever $x\neq x'$ are 
1459:  distinct  points  of the geodesic ${\mathfrak C}_y\subset M$ represented by 
1460: $y$. 
1461: 
1462: Now since $[\nabla]$ has been assumed to be $C^3$, we can 
1463: can give  ${\mathcal N}$  a `provisional' $C^2$ structure, compatible with 
1464: its fixed $C^1$ structure,  relative to which 
1465:   $\Psi$ 
1466: becomes a $C^2$ map. We now claim that $J$ is  actually 
1467: Lipschitz continuous in the associated charts on ${\mathcal N}$. 
1468: Of course, this is is only a non-trivial statement near a point $y\in N$, since 
1469: the restriction of 
1470: $J$ to ${\mathcal N}-N$ corresponds, via $\Psi$, to a 
1471:  $C^3$ almost-complex structure on ${\mathcal Z}-Z$.
1472: 
1473: Now  let us recall that  we have written down an 
1474: explicit local framing $(\Xi, \partial/\partial \overline{\zeta})$ of  
1475: ${\bf D}$ such that $[\Xi, \partial/\partial \overline{\zeta}]=0$, and such that 
1476: $\Xi$ is real along
1477: $Z=\PP TM$, and spans the tangent space of the foliation $\mathcal F$
1478: there. 
1479: Giving an arbitrary  leaf $\hat{{\mathfrak C}}_y$ a parameter $t$ such that 
1480: $\Xi = d/dt$ along the leaf, then, for any $C^2$ function $f$ on ${\mathcal N}$
1481:  we  have
1482: \begin{eqnarray*}
1483: \frac{d}{dt}\left[ \Psi_* (\frac{\partial}{\partial \overline{\zeta}}) f\right]&=& 
1484: \frac{d}{dt}\frac{\partial}{\partial \overline{\zeta}} \Psi^* f\\
1485: &=&\Xi\frac{\partial}{\partial \overline{\zeta}}\Psi^* f\\
1486: &=& \frac{\partial}{\partial \overline{\zeta}}\Xi\Psi^* f\\
1487: &=&  \frac{\partial}{\partial \overline{\zeta}}\left[ \Psi_*(\Xi)f\right],
1488: \end{eqnarray*}
1489: so that, setting $\zeta = \xi + i\eta$,  
1490: $$\frac{d}{dt}\left[ \Psi_* (\frac{\partial}{\partial \overline{\zeta}})\right] 
1491: =  \frac{\partial}{\partial \overline{\zeta}}\left[ \Psi_*(\Xi)\right] =
1492:   \frac{i}{2} \frac{\partial}{\partial \eta}
1493: \left[ \Psi_*(\Xi)\right]
1494: $$
1495: at $y\in N$, 
1496: since $\Psi_*(\Xi)\equiv 0$ along $Z$, where $\eta=0$.
1497: Here the 
1498:  right-hand side should   be  interpreted as the
1499:   invariant  derivative 
1500: {\em at a zero} of a section of a  vector bundle on 
1501:  $\Sigma_x := \Psi [\hat{\mu}^{-1} (x)]\cong \CP_1$. 
1502: On the other hand, 
1503: $$\Psi_* \left(\frac{\partial}{\partial \overline{\zeta}}\right)\in T_y^{0,1}({\mathcal N},J)$$
1504: for all $t$, by our previous discussion, so it follows that 
1505: $$\left.
1506: \frac{\partial}{\partial \eta}
1507: \left[ \Psi_*(\Xi)\right]\right|_{\eta =0} \in T_y^{0,1}({\mathcal N},J),$$
1508: too. Along $\Sigma_x$, we therefore have, 
1509: near an arbitrary point $y\in N$,    two 
1510: continuous sections of $T^{1,0}$
1511: given by ${\bf e}_1=\Psi_*(\partial/\partial \overline{\zeta})$ and 
1512: $$
1513: {\bf e}_2 = \left\{
1514: \begin{array}{cc}
1515: \left[ \Psi_*(\Xi)\right]/\eta& \eta \neq 0\\
1516: \frac{\partial}{\partial \eta}
1517: \left[ \Psi_*(\Xi)\right]&\eta =0. 
1518: \end{array}
1519: \right.
1520: $$
1521: These sections are linearly independent at every point, and so 
1522: span $T^{1,0}_y$, because $\det (\Psi_*)$ only vanishes to first
1523: order at $Z$. Moreover, since $\Psi$ appears to be  $C^2$ in our coordinates, 
1524:  these sections are both appear to be continuously differentiable  in our chart,
1525: with derivatives that may be expressed in any coordinate
1526: system in terms of partial derivatives of $\Psi$ of 
1527: order $\leq 2$. Hence $J$ is also differentiable,  and in particular is Lipschitz, along
1528: $\Sigma_x$, with Lipschitz constant controlled by the 
1529: partial derivatives of $\Psi$ of order $\leq 2$.
1530: Since the family $\{ \Sigma_x\}$ sweeps out all the 
1531: radial lines  in our tubular neighborhood  $TN$  of $N\subset {\mathcal N}$,
1532: if follows that the tensor field $J$ on ${\mathcal N}$ is  Lipschitz. 
1533: 
1534: Now recall that Rademacher's theorem asserts that all  the distributional 
1535: first partial derivatives of a Lipschitz function are 
1536:  locally  bounded measurable functions. 
1537: The Nijenhuis tensor 
1538: $$\tau ({\bf v},{\bf w}) = [{\bf v}, {\bf w}] - [J{\bf v}, J{\bf w}] +J[{\bf v}, J{\bf w}] +J[J{\bf v}, {\bf w}]$$
1539: of our almost-complex structure $J$ is therefore well-defined in the distributional sense, 
1540: and has components that are
1541: locally  bounded and  measurable. 
1542: On the other hand, since ${\bf D}$ is involutive and 
1543:  $\Psi |_{{\mathcal Z}-Z}$ is a diffeomorphism, 
1544: $\tau \equiv 0$ on a set of full measure, and therefore 
1545: vanishes in the distributional sense. But Hill and Taylor  \cite{hiltay}
1546:  have
1547: recently shown that the Newlander-Nirenberg theorem 
1548: holds for Lipschitz almost complex structures for which  $\tau =0$
1549: in just this  distributional sense. Thus every point of $\mathcal N$
1550: has a neighborhood on which we can find a pair $(z^1, z^2)$ of differentiable 
1551: complex-valued functions with $dz^k\in \Lambda^{1,0}({\mathcal N}, J)$
1552: and $dz^1\wedge dz^2 \neq 0$. Taking these to be the complex
1553: coordinate systems gives ${\mathcal N}$ the structure of a 
1554: compact complex surface. In particular, this gives ${\mathcal N}$ a
1555: specific real-analytic structure, and hence  a specific $C^\infty$ structure. 
1556: 
1557: Finally,  we  address  
1558:  the smoothness of  $\Psi : {\mathcal Z}\to {\mathcal N}$. 
1559: Suppose that $\nabla$ is of differentiability class $C^{k,\alpha}$, and 
1560: suppose that $f$ is a holomorphic function on some open subset
1561: of ${\mathcal N}$; we then consider the function  $\Psi^*f$
1562: on ${\mathcal Z}=\PP T_\CC M$. Now,  by \cite{hiltay},
1563: $f$ is a $C^1$ function with respect to our (original, unchanged) $C^1$ structure on 
1564: ${\mathcal N}$, 
1565: and $\Psi^* f$ is therefore a $C^1$ function, since $\Psi$ was
1566: $C^1$ by construction.  Moreover, since $\Psi_* {\bf D} \subset T^{0,1}{\mathcal N}$, 
1567: $\Psi^* f =0$ solves the Cauchy-Riemann equations 
1568: $\bar{\partial}_{\bf D} (\Psi^* f)=0$  with respect to the $C^{k,\alpha}$ almost-complex
1569: structure which ${\bf D}$ determines on  ${\mathcal Z}-Z$.
1570: But since $\bar{\partial}_{\bf D}+ \bar{\partial}_{\bf D}^*$, defined with 
1571: respect to an arbitrary $C^{k, \alpha}$ Hermitian metric on ${\mathcal Z}-Z$,
1572:  is a
1573: first-order elliptic system with $C^{k, \alpha}$ coefficients, 
1574: elliptic regularity  \cite{morrey} tells us 
1575: that $\Psi^* f$ is $C^{k+1,\alpha}$ on ${\mathcal Z}-Z$. 
1576: Applying these observation when $f$ is any local complex coordinate 
1577: $z^j$ on ${\mathcal N}$ then shows that $\Psi$
1578: belongs to  the claimed differentiability class. 
1579: \end{proof}
1580: 
1581: 
1582: \begin{remark} \label{rocky}
1583: The above proof uses a powerful recent analytic theorem  in order to obtain 
1584: the result without too much hard work. Most readers will find it reassuring, however, that 
1585: older technology may instead be used to prove a workable version of the proposition 
1586: at  the price of a half-dozen derivatives and  a certain amount of careful calculation.
1587: Moreover, this approach has the added benefit of providing some immediate
1588: added information concerning the regularity of
1589: $\Psi$ along $Z\subset {\mathcal Z}$.  
1590: In particular, 
1591: those primarily interested in the $C^\infty$ case  might
1592:  well prefer the following  elementary argument. 
1593: 
1594: 
1595: Suppose that $\nabla$ is a $C^k$ connection, where $k=2\ell +2$. 
1596: Choose $C^{2\ell+2}$ local real 
1597: coordinates $(\check{y}^1,\check{y}^2)$ on $U\subset N$, and pull 
1598: them back to $Z=\PP TM$ so as to obtain
1599:  $C^{2\ell+2}$ functions $y^\jmath = \nu^*\check{y}^\jmath$
1600:  on $\nu^{-1}U\subset Z$. By construction, these solve   the
1601: equation $\Xi y^\jmath=0$.  We now extend the $y^\jmath$ as 
1602: $C^{\ell+2}$ complex-valued  functions 
1603: ${\mathfrak z}^\jmath$ defined on an open set in 
1604: ${\mathcal Z}$ 
1605:  by requiring  that $\partial y^\jmath/\partial
1606: \bar{\zeta}$ vanish to order $\ell-1$ along $Z$.
1607: This  completely specifies the $\ell$-jet of the 
1608: function, and we must have 
1609: $$
1610: {\mathfrak z}^\jmath(x^1,x^2,\xi,\eta) =\sum_{r=0}^{\ell} \frac{i^r}{r!}
1611: \eta^r \left.
1612:  \frac{\partial^r y^\jmath}{\partial \xi^r}\right|_{(x^1,x^2,\xi)}  ~~+ O(\eta^{\ell +1}). 
1613: $$
1614: Indeed, this recipe does indeed give us  
1615: \begin{eqnarray*}
1616:  \frac{\partial {\mathfrak z}^\jmath}{\partial \overline{\zeta}}
1617: &=&\frac{1}{2}
1618:  \left( \frac{\partial}{\partial \xi}
1619: + i\frac{\partial}{\partial \eta}\right)  \left( \sum_{r=0}^{\ell} \frac{i^r}{r!} \eta^r \frac{\partial^r y^\jmath}{\partial \xi^r}
1620:  ~~+ O(\eta^{\ell +1}) \right)
1621: \\&=& 
1622: \frac{1}{2}
1623: \sum_{r=0}^{\ell} \frac{i^r}{r!} \eta^r\frac{\partial^{r+1} y^\jmath}{\partial \xi^{r+1}}
1624: -\frac{1}{2}
1625: \sum_{r=1}^{\ell} \frac{i^{r-1}}{(r-1)!} \eta^{r-1}\frac{\partial^r y^\jmath}{\partial \xi^r} ~~+ O(\eta^{\ell})
1626: \\&=& 
1627: \frac{1}{2}
1628: \sum_{r=0}^{\ell} \frac{i^r}{r!} \eta^r\frac{\partial^{r+1} y^\jmath}{\partial \xi^{r+1}}
1629: -\frac{1}{2}
1630: \sum_{r=0}^{\ell-1} \frac{i^{r}}{r!} \eta^{r}\frac{\partial^{r+1} y^\jmath}{\partial \xi^{r+1}} ~~+ O(\eta^{\ell})
1631: \\&=& 
1632: \frac{1}{2}
1633: \frac{i^{\ell}}{\ell!} \eta^{\ell}\frac{\partial^{\ell+1} y^\jmath}{\partial \xi^{\ell+1}} ~~+ O(\eta^{\ell})
1634: \\&=&   O(\eta^{\ell}) , 
1635: \end{eqnarray*}
1636: and  since the cancellation is a  term-by-term matter, 
1637: uniqueness of the $\ell$-jet follows. But since our condition on the 
1638: $\ell$-jet is obviously independent of the choice of 
1639: coordinates 
1640: $(x^1,x^2)$ on $M$, global 
1641: existence now follows by patching together any such
1642: local choices via a partition of unity.
1643: 
1644: The uniqueness argument also has another useful consequence. Notice that there 
1645: certainly are $C^2$ coordinates $(\tilde{\mathfrak z}^1, \tilde{\mathfrak z}^2)$ 
1646: for our provisional $C^2$ structure on
1647: ${\mathcal N}$ whose restrictions to $N$ are the $\check{y}^\jmath$, and 
1648: which are satisfy $\overline{\partial}_J \tilde{\mathfrak z}^\jmath =0$
1649: to $0^{th}$ order along $N$, since the restriction of $J$ to 
1650: $T{\mathcal N}|_N$ is $C^{k-1}$. 
1651: But pulling these back to ${\mathcal Z}$ would gives us $C^2$ functions 
1652: killed by  $\partial/\partial\overline{\zeta}$ to $0^{th}$ order along $Z$, and the
1653: $\ell=1$ version of the above calculation therefore gives 
1654: $$
1655: \Psi^*\tilde{\mathfrak z}^\jmath =  {\mathfrak z}^\jmath + O(\eta^2).
1656: $$
1657: It follows that $({\mathfrak z}^1, {\mathfrak z}^2)$ is
1658: actually a $C^1$ complex-valued coordinate system on 
1659: ${\mathcal N}$. Our strategy will now be to analyze the 
1660: the almost-complex structure $J$ 
1661: by thinking of $(x^1,x^2,\xi,\eta)\mapsto ({\mathfrak z}^1, {\mathfrak z}^2)$
1662: as a representation of 
1663: $\Psi$ in special coordinates
1664: 
1665:  
1666: To this end, we next
1667:  observe that, since  $[\Xi , \frac{\partial}{\partial \overline{\zeta}} ]=0$, the 
1668: $C^{\ell+1}$ function $\Xi {\mathfrak z}^\jmath$ satisfies 
1669: $$
1670: \frac{\partial^m }{\partial \overline{\zeta}^m} \Xi {\mathfrak z}^\jmath = \Xi \frac{\partial^m }{\partial \overline{\zeta}^m}{\mathfrak z}^\jmath =
1671: \Xi O(\eta^{\ell-m+1}) = O(\eta^{\ell-m+1}),
1672: $$
1673: so that  
1674: $$\left.
1675: \left(\frac{\partial}{\partial {\xi}} +i\frac{\partial}{\partial {\eta}}
1676: \right)^m (\Xi {\mathfrak z}^\jmath)\right|_{\eta=0} \equiv 0,
1677: $$
1678: for $m=0, \ldots, \ell$.
1679: But since $\Xi {\mathfrak z}^\jmath \equiv 0$ along $\eta =0$, this tells us that 
1680: $$\left.
1681: \frac{\partial^m }{\partial {\eta}^m} (\Xi {\mathfrak z}^\jmath)\right|_{\eta=0} \equiv 0
1682: $$
1683: for $m = 0 , \ldots , \ell$, and hence  that
1684: $$
1685: \Xi {\mathfrak z}^\jmath = O (\eta^{\ell+1}).
1686: $$
1687: 
1688: Now we 
1689: have already shown, by an 
1690: elementary argument, that the almost-complex structure $J$ is characterized,
1691: in a point-wise manner, 
1692: by the fact that $\Psi_*\partial/\partial \overline{\zeta}$ and $\Psi_* \Xi$ are 
1693: always elements  of $T^{0,1}({\mathcal N},J)$. 
1694: Since $\mbox{span} \{ \partial/\partial {\mathfrak z}^1 , \partial/\partial {\mathfrak z}^2\}$ contains 
1695: the image of $\partial/\partial \overline{\zeta}$ and (trivially) $\Xi$ along 
1696: the locus $N$ given by $\Im  m {\mathfrak z}^\jmath =0$, we must therefore have 
1697: $$T^{0,1}({\mathcal N},J)|_{\Im m {\mathfrak z}^\jmath =0}= \mbox{span}\left\{ \frac{\partial}{\partial \overline{{\mathfrak z}}^\jmath} 
1698: \right\}_{\jmath=1,2} ~~,$$
1699: and 
1700: $$
1701: T^{*1,0}({\mathcal N},J)|_{\Im m {\mathfrak z}^\jmath =0}= \mbox{span}\left\{ d{\mathfrak z}^\jmath\right\}_{\jmath=1,2}~~.$$
1702: Elsewhere, 
1703: $$T^{*1,0}({\mathcal N},J)= 
1704: \mbox{span}\left\{ d{\mathfrak z}^\jmath -\sum_\imath a^\jmath_\imath d\overline{{\mathfrak z}}^\imath
1705: \right\}_{\jmath=1,2}$$
1706: and
1707: $$T^{0,1}({\mathcal N},J)= \mbox{span} \left\{ \frac{\partial}{\partial \overline{{\mathfrak z}}^\jmath}
1708: + \sum_\imath a^\imath_\jmath \frac{\partial}{\partial {{\mathfrak z}}^\imath}
1709: \right\}_{\jmath=1,2} ~~,$$
1710: where the $a^\jmath_\imath$ are to be found by solving the equation 
1711: $$
1712: \left[
1713: \begin{array}{cc}
1714: a_1^1&a^1_2
1715: \\a^2_1&a^2_2
1716: \end{array}
1717: \right] 
1718: \left[
1719: \begin{array}{cc}
1720: \Xi \overline{{\mathfrak z}}^1& \frac{\partial \overline{{\mathfrak z}}^1}{\partial \overline{\zeta}}
1721: \\
1722: \Xi \overline{{\mathfrak z}}^2& \frac{\partial \overline{{\mathfrak z}}^2}{\partial \overline{\zeta}}
1723: \end{array}
1724: \right]
1725: =
1726: \left[
1727: \begin{array}{cc}
1728: \Xi {{\mathfrak z}}^1& \frac{\partial {{\mathfrak z}}^1}{\partial \overline{\zeta}}
1729: \\
1730: \Xi {{\mathfrak z}}^2& \frac{\partial {{\mathfrak z}}^2}{\partial \overline{\zeta}}
1731: \end{array}
1732: \right].
1733: $$
1734: But 
1735: $$\frac{\partial}{\partial \overline{\zeta}} \overline{{\mathfrak z}}^\jmath = \frac{\partial y^\jmath}{\partial \xi}  + O (\eta),$$
1736: and 
1737: $$
1738: \Xi \overline{{\mathfrak z}}^\jmath = \Xi \left ( -i\eta \frac{\partial y^\jmath}{\partial \overline{\xi}} + O(\eta^2)\right)
1739: = -i \eta [\Xi ,  \frac{\partial }{\partial \xi} ] y^\jmath
1740: = i\eta  \frac{\partial y^\jmath}{\partial x^2} + i\eta P^\prime (\xi) \frac{\partial y^\jmath}{\partial {\xi}} + O (\eta^2),
1741: $$
1742: so that 
1743: $$
1744:  \left|
1745: \begin{array}{cc}
1746: \Xi \overline{{\mathfrak z}}^1& \frac{\partial \overline{{\mathfrak z}}^1}{\partial \overline{\zeta}}
1747: \\
1748: \Xi \overline{{\mathfrak z}}^2& \frac{\partial \overline{{\mathfrak z}}^2}{\partial \overline{\zeta}}
1749: \end{array}
1750: \right|
1751: = i\eta 
1752:  \left|
1753: \begin{array}{cc}
1754: \frac{\partial y^1}{\partial x^2} + P^\prime (\xi) \frac{\partial y^1}{\partial {\xi}} 
1755: & \frac{\partial {y}^1}{\partial \xi}
1756: \\
1757: \frac{\partial y^2}{\partial x^2} + P^\prime (\xi) \frac{\partial y^2}{\partial {\xi}} 
1758: & \frac{\partial {y}^2}{\partial \xi}
1759: \end{array}
1760: \right| + O (\eta^2) = i\eta  \frac{\partial(y^1,y^2)}{\partial(x^2,\xi)}
1761:   + O(\eta^2).
1762: $$
1763: But ${\partial(y^1,y^2)}/{\partial(x^2,\xi)}\neq 0$ everywhere, since $\Xi$ is always 
1764: linearly independent from $\partial/\partial x^2$ and $\partial/\partial\xi$. Thus 
1765: \begin{eqnarray*}
1766: \left[
1767: \begin{array}{cc}
1768: a_1^1&a^1_2
1769: \\a^2_1&a^2_2
1770: \end{array}
1771: \right] &=& 
1772: \left[
1773: \begin{array}{cc}
1774: \Xi {{\mathfrak z}}^1& \frac{\partial {{\mathfrak z}}^1}{\partial \overline{\zeta}}
1775: \\
1776: \Xi {{\mathfrak z}}^2& \frac{\partial {{\mathfrak z}}^2}{\partial \overline{\zeta}}
1777: \end{array}
1778: \right]
1779: \left[
1780: \begin{array}{cc}
1781: \Xi \overline{{\mathfrak z}}^1& \frac{\partial \overline{{\mathfrak z}}^1}{\partial \overline{\zeta}}
1782: \\
1783: \Xi \overline{{\mathfrak z}}^2& \frac{\partial \overline{{\mathfrak z}}^2}{\partial \overline{\zeta}}
1784: \end{array}
1785: \right]^{-1}
1786: \\ &=&
1787: \left[
1788: \begin{array}{cc}
1789: O(\eta^{\ell+1})& O(\eta^{\ell})\\
1790: O(\eta^{\ell+1})&O(\eta^{\ell})
1791: \end{array}
1792: \right]
1793: \frac{1}{i\eta} \left(
1794:  \frac{\partial (x^2,\xi)}{\partial (y^1,y^2)} + O (\eta)
1795: \right) 
1796: \left[
1797: \begin{array}{cc}
1798:  \frac{\partial \overline{{\mathfrak z}}^2}{\partial \overline{\zeta}}& -\frac{\partial \overline{{\mathfrak z}}^1}{\partial \overline{\zeta}}
1799: \\
1800: -\Xi \overline{{\mathfrak z}}^2&\Xi \overline{{\mathfrak z}}^1
1801: \end{array}
1802: \right]
1803: \\ &=&
1804: \left(
1805:  \frac{\partial (\xi,x^2)}{\partial (y^1,y^2)} + O (\eta)
1806: \right) 
1807: \left[
1808: \begin{array}{cc}
1809: O(\eta^{\ell})& O(\eta^{\ell-1})\\
1810: O(\eta^{\ell})& O(\eta^{\ell-1})
1811: \end{array}
1812: \right]
1813: \left[
1814: \begin{array}{cc}
1815: O(\eta^0)& O(\eta^0)
1816: \\
1817: O(\eta)& O(\eta)
1818: \end{array}
1819: \right]
1820: \\
1821: &=& O(\eta^{\ell-1}) 
1822: \end{eqnarray*}
1823: More precisely, for $(x^1,x^2,\xi, \eta)$ in any fixed compact set, there is a constant 
1824: $C$ such that 
1825: $$|a^\jmath_\imath | < C~ |~\eta ~|^{\ell -1}.$$
1826: For the corresponding set in ${\mathcal N}$, this becomes
1827: the statement that 
1828: $$|a^\jmath_\imath | < C_1  ~|~\Im m ~\vec{\mathfrak{z}} ~|^{\ell -1}.$$
1829: But since $\Psi$ is a proper map, it only takes a finite 
1830: number of closed coordinate balls to cover the inverse image of 
1831: any compact set in $\mathcal N$, and hence we have 
1832: $$|a^\jmath_\imath | < C_2 ~|~\Im m ~\vec{\mathfrak{z}} ~|^{\ell -1}$$
1833: as long as $\vec{\mathfrak z}=({\mathfrak z}^1, {\mathfrak z}^2)$ 
1834: is constrained to lie in any fixed compact set. 
1835: 
1836: 
1837: Since $(x^1,x^2,\xi,\eta)\mapsto ({\mathfrak z}^1,{\mathfrak z}^2)$ is 
1838: a $C^{\ell+1}$ diffeomorphism away from $\eta =0$, the 
1839: $a^\jmath_\imath$ are $C^{\ell+1}$ functions of the 
1840: $({\mathfrak z}^1,{\mathfrak z}^2)$ away from $\Im m ~{\mathfrak z}^1 = \Im m ~{\mathfrak z}^2
1841: =0$, and on the other hand we have seen that they vanish to
1842: order $\ell -2$ along this bad locus. Thus the 
1843: $a^\jmath_\imath$ are $C^{\ell-2}$ functions of the 
1844: ${\mathfrak z}^\jmath$, and the  complex structure
1845: $J$ on ${\mathcal N}$  is $C^{\ell-2}$
1846: in these coordinates. If $\ell -2 \geq 1$, the Nijenhuis tensor 
1847: therefore vanishes identically by continuity, since  it is already known to
1848: vanish on an open dense set. 
1849: If $\ell-2\geq 4$, or in other words if $[\nabla ]$  is at least $C^{14}$,
1850: we may therefore apply the original Newlander-Nirenberg theorem
1851: \cite{newnir} to get $C^{\ell-2}$ functions
1852: $(z^1,z^2)$ of $({\mathfrak z}^1, {\mathfrak z}^2)$ which are holomorphic
1853: with respect to $J$. The Malgrange refinement \cite{malgrange} of Newlander-Nirenberg
1854: may similarly be applied if $\ell-2\geq 2$, 
1855: or in other words if $[\nabla ]$ is at least $C^{10}$. 
1856: %These coordinates are $C^1$ with respect to our fixed differentiable
1857: %structure on $\mathcal N$, and are therefore differentiable with respect to each 
1858: %other; but the transition functions between them are 
1859: %holomorphic, so they make $\mathcal N$ into a complex manifold, as
1860: %promised. 
1861: The rest of the proof  then proceeds as before. Notice, however, 
1862: that this second argument also  directly verifies  that  $\Psi : {\mathcal Z}\to {\mathcal N}$
1863: is at least $C^{[k/2]-3}$ along $Z\subset {\mathcal Z}$.
1864: \label{road}
1865: \end{remark}
1866: 
1867: 
1868: Having constructed our compact complex surface ${\mathcal N}$,
1869: we will now try to unmask its identity. To this end, 
1870: recall that we originally assembled ${\mathcal N}$ from two open sets, 
1871: ${\mathcal U}={\mathcal Z}-Z$ and ${\mathcal V} \approx T\RP^2$. 
1872: However,  ${\mathcal U}$
1873: may be identified with the space of all almost-complex structures\footnote{Indeed, 
1874: the fact that $\bf D$ is a complex structure on
1875: $\mathcal U$  thus naturally arises in the context of the 
1876: O'Brian-Rawnsley 
1877: generalization \cite{obiwon} of the    Atiyah-Hitchin-Singer 
1878: approach \cite{AHS} to  twistor theory.} on 
1879: $M$, since an almost-complex structure is completely characterized 
1880: by its $(0,1)$-tangent space, and in dimension $2$ this may be taken
1881: to be any $1$-dimensional subspace of $T_\CC M$ which is not spanned
1882: by a real vector. 
1883:  Thus ${\mathcal U}\to M$
1884: may be identified with the space  of pairs $([h], \circlearrowleft)$, 
1885: where $h$ is a Riemannian metric on some tangent space $T_xM$, 
1886: $[h]$ is its conformal class, and $\circlearrowleft$ denotes a choice of
1887: orientation of $T_xM$. Since  the space of Riemannian metrics
1888: is a convex cone,  ${\mathcal U}$ therefore canonically 
1889:  deform retracts to the set of point-wise orientations
1890: on  $M$, once we choose a
1891: single `background' Riemannian metric $h_0$ on $M$.
1892: But the $2$-fold cover $\tilde{M}$ of $M$ by its set of local 
1893: orientations is evidently just $S^2$, since $M=\RP^2$
1894: by assumption. This shows that ${\mathcal U}$ is homotopy equivalent to
1895: $S^2$. 
1896: 
1897: With this observation in hand, we are now in a position to 
1898: list   some identifying traits of our complex surface $({\mathcal N}, J)$. 
1899:  
1900:  \begin{prop} \label{complexn} 
1901: Let $[\nabla ]$ be a Zoll projective structure on $M=\RP^2$,
1902: and let $N\approx \RP^2$ denote the corresponding space of
1903: unoriented geodesics. Then there is a compact complex surface
1904: $\mathcal N$ and an embedding $N\hookrightarrow {\mathcal N}$
1905: such that 
1906: \begin{itemize}
1907: \item $\pi_1 ({\mathcal N}) =0$; 
1908: \item there is an anti-holomorphic involution $\sigma : {\mathcal N}
1909: \to {\mathcal N}$ with fixed-point set N; 
1910: \item for all $x \in M$, there is a  
1911: complex curve $\Sigma_x \subset 
1912: {\mathcal N}$, 
1913: $\Sigma_x \cong \CP_1$, such that $$\ell_x = \Sigma_x\cap  N;$$
1914: \item the $\Sigma_x$ all represent the same element of $\pi_2 ({\mathcal N})$; and
1915: \item if $x$ and $x'$ are distinct points of $M$, then 
1916: $\Sigma_x$ and $\Sigma_{x'}$ are transverse, and meet in exactly one point. 
1917: \end{itemize}
1918: \end{prop}
1919: 
1920: \begin{proof}
1921: By construction, 
1922: ${\mathcal N}= {\mathcal U}\cup {\mathcal V}$, where ${\mathcal U}={\mathcal Z} - Z$
1923: and ${\mathcal V} =TN \approx T\RP^2$. But we have just seen that 
1924:  ${\mathcal U}$ deform retracts to $S^2$. Moreover, ${\mathcal V}$ deform 
1925: retracts to $N\approx \RP^2$,  and the inclusion map 
1926: $\jmath : {\mathcal U}\cap {\mathcal V} \hookrightarrow  {\mathcal V}$
1927: is homotopic to the bundle projection $\wp : (TN-0_{N}) \to N$. 
1928: Because ${\mathcal U}$ is simply connected and ${\mathcal U}\cap {\mathcal V}$ is connected, 
1929: the Seifert-van Kampen theorem 
1930: tells us that 
1931: $$\pi_1 ({\mathcal N} ) = \frac{\pi_1 ({\mathcal V})}{ \jmath_{\natural} [\pi_1 ({\mathcal U} \cap {\mathcal V})]} = 
1932: \frac{\pi_1 (N ) }{\wp_{\natural}
1933:  [\pi_1 (TN-0_N)]}.$$
1934: But $\wp_{\natural} : \pi_1 (TN-0_N) \to \pi_1 (N)$ is surjective, since the 
1935: fibers of $\wp$ are path connected. Hence ${\mathcal N}$
1936: is simply connected. 
1937: 
1938: Complex conjugation $\PP T_\CC M \to \PP T_\CC M$
1939: sends the distribution ${\bf D}$ to its conjugate $\overline{{\bf D}}$. The 
1940: induced involution $\sigma : {\mathcal N}\to {\mathcal N}$ is therefore 
1941: anti-holomorphic, and obviously has fixed point set precisely consisting of $N$. 
1942: 
1943:  For each $x\in M$, set $\Sigma_x = \wp (\PP T_{x\CC} M)$. Then 
1944: $\Sigma_x$ is an embedded genus $0$ complex curve in ${\mathcal N}$. 
1945: Since  the fibers of $\PP T_\CC M$ are all homotopic, 
1946: so are their images in $\mathcal N$. Moreover, since the fibers of 
1947: $\PP T_\CC M$ are all disjoint, we must have $\Sigma_x \cap \Sigma_{x'} \subset N$.
1948: But, by construction, $\Sigma_x \cap N = \ell_x$, and so 
1949: $$\Sigma_x \cap \Sigma_{x'} = (\Sigma_x \cap N)  \cap (\Sigma_{x'} \cap N) = \ell_x
1950: \cap \ell_{x'}, $$
1951: and if $x\neq x'$ this 
1952:  consists of precisely one point $y$, representing  the unique geodesic joining 
1953: $x$ to $x'$; cf.  Corollary \ref{lcapl}. Now $\Sigma_x$ and $\Sigma_{x'}$ are both 
1954: $\sigma$-invariant, so $T_y\Sigma_x \cap T_y \Sigma_{x'}$ is invariant under
1955: the complex anti-linear involution 
1956: $\sigma_*$ of $T_y{\mathcal N}$, which we may identify with 
1957: complex conjugation  on $\CC\otimes T_yN$. 
1958: But since $T_y\ell_x\cap T_y\ell_{x'}=0$, 
1959: its complexification $T_y\Sigma_x\cap T_y\Sigma_{x'}$
1960: is also zero, and $\Sigma_x$ and $\Sigma_{x'}$ therefore intersect
1961: transversely, at the unique point $y$, exactly as claimed. 
1962: \end{proof}
1963: 
1964: 
1965: 
1966: We now come to the key step in our proof, which is to 
1967: observe that $\mathcal N$ must be biholomorphic to $\CP_2$. 
1968: It is a deep and remarkable  fact \cite{yau} that, 
1969: up to biholomorphism, 
1970:  $\CP_2$ is the only simply connected complex surface of Euler characteristic $3$,
1971: and it might therefore be tempting to now  invoke this powerful result, much as we will 
1972: later do in  \S \ref{zoe} below.  However, we will 
1973:  actually need to know a great deal about the  biholomorphism 
1974: $F: {\mathcal N}\to \CP_2$, and for this reason it is 
1975: in every sense more satisfactory   to instead make use of 
1976: the following low-tech lemma, based on the classical ideas of Castelnuovo,
1977: Enriques and 
1978: Kodaira;  cf. 
1979:   \cite[Proposition V.4.3]{bpv}.
1980: As a courtesy to the reader, as well as  to emphasize the 
1981: elementary nature   of  the result, we  include a short, complete proof.
1982: 
1983: \begin{lem}\label{castle}
1984: Let ${\mathcal S}$ be a simply connected compact complex surface, 
1985: equipped with 
1986: a fixed homology class
1987: ${\bf a} \in  H_2 ({\mathcal S}, \Z)$ such that  ${\bf a}\cdot {\bf a} =1$. 
1988: For every 
1989: $p\in {\mathcal S}$,  suppose that there exists a non-singular, embedded
1990:  complex curve $\Sigma\subset {\mathcal S}$ of  genus $0$
1991: passing through  $p$, with homology class $[\Sigma]={\bf a}$. Then
1992: ${\mathcal S}$ is biholomorphic to $\CP_2$,  in such  a manner  that 
1993: all of the given curves  become projective lines. 
1994: \end{lem} 
1995: 
1996: \begin{proof}
1997: Since  the 
1998: Fr\"olicher spectral sequence
1999: of any complex surface degenerates at the $E_1$ level \cite[Theorem IV.2.7]{bpv}, 
2000: we have 
2001: $$H^1({\mathcal S},\CC) \cong H^1({\mathcal S},{\mathcal O}) \oplus H^0({\mathcal S},\Omega^1),$$
2002: so  the assumption that $\pi_1 ({\mathcal S})=0$ 
2003: immediately implies that $H^1({\mathcal S}, {\mathcal O}) =0$. 
2004: But  the divisor line bundle  ${\mathcal O}(\Sigma)$ 
2005: of  any of the curves $\Sigma\subset {\mathcal S}$ fits into an exact sequence 
2006: \begin{equation}
2007: \label{rest}
2008: 0\to {\mathcal O} \stackrel{{f}\cdot}{\to}  {\mathcal O}(\Sigma) \to n_{\Sigma} \to 0
2009: \end{equation}
2010: of sheaves on ${\mathcal S}$, where $n_{\Sigma}$ 
2011: is the normal sheaf of $\Sigma$, extended 
2012: to ${\mathcal S}$ by $0$,
2013: and where ${f}\cdot$ denotes multiplication by a holomorphic section 
2014: ${f}$ of ${\mathcal O}(\Sigma)$ which vanishes only at $\Sigma$, 
2015: with $d{f}\neq 0$ along $\Sigma$. 
2016:  Now the normal bundle of $\Sigma$ has degree 
2017: ${\bf a}\cdot {\bf a} =1$,  and thus  $n_{\Sigma}$ can 
2018: be identified with the unique degree-$1$ holomorphic line bundle ${\mathcal O}(1)$ on $\CP_1$. 
2019: Since $H^1({\mathcal S}, {\mathcal O}) =0$, the 
2020: long exact sequence in cohomology induced by (\ref{rest}) 
2021: therefore gives us the short exact sequence 
2022: \begin{equation}\label{shorty}
2023: 0\to \CC \stackrel{{f}\cdot}{\to} 
2024:  \Gamma ({\mathcal S}, {\mathcal O}(\Sigma)){\to} \Gamma (\CP_1 , {\mathcal O}(1))\to 0. 
2025: \end{equation} 
2026: In particular,  $H^0( {\mathcal S}, {\mathcal O}(\Sigma)) \cong \CC^3$; moreover,  
2027:  there is a holomorphic section of ${\mathcal O}(\Sigma)$ which is 
2028: non-zero at any given point of ${\mathcal S}$. The  associated  map 
2029: $$F: {\mathcal S} \to {\mathbb P}[H^0( {\mathcal S}, {\mathcal O}(\Sigma))^*]\cong \CP_2,$$
2030: is  thus everywhere defined. Also notice that 
2031:   $F(\Sigma)$ is a 
2032: projective line  ${\mathcal P}  \subset \CP_2$,  and that  
2033: the derivative of $F$ is  of maximal rank at any point $p$ of 
2034: $\Sigma$, since (\ref{shorty}) 
2035:  allows us to produce two  sections of ${\mathcal O}(\Sigma)$, $f$ and another one, 
2036:  which vanish at $p$, but have linearly 
2037: independent
2038: derivatives there.
2039: 
2040: Since 
2041: $H^1({\mathcal S}, \O)=0$, 
2042: the exact sequence 
2043: $$\cdots \to H^1({\mathcal S}, {\mathcal O}) \to H^1 ({\mathcal S}, {\mathcal O}^* ) \stackrel{c_1}{\to} H^2({\mathcal S}, \Z ) \to \cdots  $$
2044:  tells us that holomorphic line bundles on ${\mathcal S}$ are classified by their 
2045: first Chern classes.
2046: But if 
2047: $\Sigma$ and $\Sigma'$ are two complex curves in the homology class ${\bf a}$,
2048: their  divisor line bundles ${\mathcal O}(\Sigma)$ and ${\mathcal O}(\Sigma')$ both have 
2049:  Chern class equal to the Poincar\'e dual of ${\bf a}$.  
2050: Thus ${\mathcal O}(\Sigma)\cong {\mathcal O}(\Sigma')$, and 
2051: $\Gamma ({\mathcal S}, {\mathcal O}(\Sigma))=  \Gamma ({\mathcal S}, {\mathcal O}(\Sigma'))$. The 
2052:  holomorphic map 
2053: $F: {\mathcal S}\to \CP_2$ determined by $\Sigma$ therefore also maps $\Sigma'$ biholomorphically
2054: to a projective line ${\mathcal P}'$, and the derivative of $F$ has maximal rank at every
2055: point of $\Sigma'$. 
2056: Since, by hypothesis,  we may find such a  curve through 
2057: any point, $F$ is a local biholomorphism. But  since ${\mathcal S}$ is compact,
2058: $F$ is therefore a covering map; and 
2059: since $\CP_2$ is simply connected, we conclude  that 
2060:  $F$ is a biholomorphism. \end{proof} 
2061: 
2062: 
2063: \begin{thm} \label{rigid}
2064: Let $(M,[\nabla])$ be a compact 2-manifold with  Zoll projective structure 
2065: of odd conjugacy number. Assume that 
2066:  $\nabla$ is  of differentiability class $C^{k,\alpha}$,  for some $k\geq 3$,
2067: and some 
2068: $\alpha \in (0,1)$.
2069: Then there is a $C^{k+2, \alpha}$  diffeomorphism 
2070: $\Phi : M\stackrel{\approx}{\longrightarrow} \RP^2$ such that   
2071: $[\nabla]=  [ \Phi^* \triangledown]$,
2072: where $\triangledown$ is 
2073:   the Levi-Civita connection 
2074: $\triangledown$ of the standard, constant curvature 
2075: Riemannian metric 
2076: $g$ on $\RP^2$. 
2077: \end{thm}
2078: \begin{proof}
2079: By  Proposition \ref{complexn}, the entire complex surface ${\mathcal N}$ is swept out 
2080: by the genus zero curves $\Sigma_x$, $x\in M$, and the 
2081: homology class 
2082: $[\Sigma_x ]\in H_2({\mathcal N}, \ZZ)$
2083: is independent of $x$. Moreover, this homology class has self-intersection
2084: $$[\Sigma_x ]\cdot [\Sigma_x ] = [\Sigma_x ]\cdot [\Sigma_{x'} ] = 
2085: 1 ,$$
2086: since $\Sigma_x$ and $\Sigma_{x'}$ intersect transversely  in one point
2087: whenever $x\neq x'$. 
2088:  Lemma \ref{castle} therefore tells us that 
2089: there is a biholomorphism $F: {\mathcal N}\to \CP_2$ which sends each 
2090: of the 
2091:  complex curves $\Sigma_x$  to a corresponding projective line $\CP_1\subset \CP_2$.
2092: 
2093: Now the anti-holomorphic involution $\sigma : {\mathcal N} \to {\mathcal N}$
2094: induces an anti-holomorphic involution $\tilde{\sigma}= F\circ\sigma\circ F^{-1}:
2095: \CP_2\to \CP_2$. By taking the 
2096:  Jacobian determinant of this map, we then obtain 
2097:  to an anti-holomorphic involution $\tilde{\sigma}^* : K \to K$ 
2098: of the canonical line bundle 
2099: $K=\Lambda ^{2,0}$ of $\CP_2$.  But $K$ has a unique holomorphic 
2100: cube-root $K^{1/3}$, the frame bundle of which is the universal cover of 
2101: the 
2102: frame bundle of $K$; 
2103: and covering space theory now  tells us that 
2104: $\tilde{\sigma}^*$ has three possible anti-holomorphic lifts $\varrho : K^{1/3} \to 
2105: K^{1/3}$, differing by  multiplicative factors of a cube-root of unity. Choose
2106: any such lift, and observe that $\varrho^2$ is the identity on any fiber over 
2107: the fixed-point locus $F (N)$ of $\tilde{\sigma}$; since $F(N)$ is totally real
2108: and 
2109: of maximal dimension, 
2110: the principle of analytic 
2111: continuation therefore implies that 
2112: the holomorphic map 
2113: $\varrho^2$ must therefore be the identity. The anti-linear map
2114: $$\varrho^* : \Gamma (\CP_2 , {\mathcal O}(K^{-1/3}))\to  \Gamma (\CP_2 , {\mathcal O}(K^{-1/3}))
2115: $$
2116: therefore satisfies $(\varrho^*)^2={\bf 1}$. It is therefore diagonalizable over $\RR$,
2117: with eigenvalues $\pm 1$, and, because it is anti-linear, it  can be put in the form 
2118: $$(z_1, z_2 , z_3 ) \mapsto (\bar{z}_1, \bar{z}_2 , \bar{z}_3 )$$
2119: by choosing a suitable basis for 
2120: $\Gamma (\CP_2 , {\mathcal O}(K^{-1/3}))\cong \Gamma (\CP_2 , {\mathcal O}(1))\cong \CC^3$. 
2121: But $[z_1 : z_2 : z_3 ]$  gives us a set of homogeneous coordinates on 
2122: $\CP_2$, so we have succeeded in identifying  $\sigma : {\mathcal N} \to {\mathcal N}$ 
2123: with the
2124: standard 
2125: complex conjugation on $\CP_2$. In the process, we have thereby 
2126: identified $N$ with $\RP^2\subset \CP_2$, and each complex curves
2127: $\Sigma_x$ with a complex projective line $\CP_1$ which is invariant
2128: under complex conjugation. 
2129: 
2130: 
2131: Now let  $ \CP_2^*=\PP (\CC^{3*})$ denote  the dual projective
2132: plane of $\CP_2= \PP (\CC^{3})$, and consider the map 
2133: \begin{eqnarray*}
2134: \Phi_0 : M &\to&  \CP_2^*\\
2135: x &\mapsto& F(\Sigma_x)^\perp, 
2136: \end{eqnarray*}
2137: where $\perp$ denotes the usual correspondence between lines in 
2138: $\CP^2$ and points in $\CP^{2*}$. 
2139: We claim that $\Phi_0$ is of differentiability class $C^{k+2,\alpha}$.
2140: Indeed, let ${\mathcal C}\subset {\mathcal U}$ be a (non-compact) holomorphic
2141: curve which is transverse to the fibers of $\hat{\mu}$, obtained by setting
2142: some local complex coordinate ${\mathfrak z}^1$ equal to zero. Since the almost
2143:  complex structure on $\mathcal U$ is of class $C^{k,\alpha}$, elliptic regularity 
2144: tells us that the 
2145: local complex coordinates $({\mathfrak z}^1,{\mathfrak z}^2)$ are of class $C^{k+1,\alpha}$,
2146: and   $\mathcal C$ is therefore   representable as the image of a 
2147: $C^{k+1,\alpha}$ map  from  an open set in $\CC$ to $\mathcal U$.
2148: But the projection from  $\mathcal C$ to $M$ is a local diffeomorphism,
2149: and so $\mathcal C$ may locally be thought of as the graph of a $C^{k+1,\alpha}$
2150: local section $\varsigma$ of ${\mathcal U}\to M$. But such a section 
2151: is precisely a local almost-complex structure on $M$ of 
2152: differentiablity class $C^{k+1,\alpha}$. Since the map 
2153: $F\circ \Psi \circ \varsigma$ is holomorphic with respect to this $C^{k+1,\alpha}$
2154: almost-complex structure, and it is  therefore of class $C^{k+2,\alpha}$
2155: by elliptic regularity. But on the domain of this function, 
2156: $F(\Sigma_x)^\perp$ is the unique line joining $F(\Psi ({\varsigma}(x))$
2157: to its complex conjugate, and so can be expressed in homogeneous
2158: coordinates as $$\Phi_0 (x) = F(\Psi ({\varsigma}(x))\times \overline{F(\Psi ({\varsigma}(x))},$$
2159: where $\times : \CC^3 \times \CC^3 \to \CC^{3*}$ is the 
2160: vector cross-product. Since $M$ is covered by the domains of 
2161: such local almost-complex structures $\varsigma$, this shows that 
2162:  $\Phi_0$ is $C^{k+2,\alpha}$
2163: on all of $M$. 
2164: 
2165: 
2166: Now notice that 
2167: $\Phi_0$
2168: is also an immersion, because $\Psi$ is a diffeomorphism on 
2169: ${\mathcal U}$, and the section of the normal bundle 
2170: of $\Sigma_x\subset {\mathcal N}$ corresponding to a non-zero
2171: element of $T_xM$ is therefore never identically zero. 
2172: Moreover,  because each $F(\Sigma_x)$ is invariant under complex conjugation, 
2173:   $\Phi_0 (M)$ actually lies
2174: in the real dual projective plane $\RP^{2*}\subset \CP_2^*$. 
2175: Thus, $\Phi_0$ actually gives us a $C^{k+2,\alpha}$  immersion  
2176: $$
2177: \Phi : M \to   \RP^{2*}$$
2178: which can be described as 
2179: $$
2180: x \mapsto  F(\ell_x)^\perp .
2181: $$
2182: But since $M$ is a compact $2$-manifold, this immersion must be a covering map, 
2183: and since $\pi_1(M)\cong \pi_1 (\RP^{2*})=\ZZ_2$, it follows that 
2184: $\Phi$ is a diffeomorphism. Moreover,$\Phi$ sends the geodesic ${\mathfrak C}_y$
2185: to the set of projective lines through the point $F(y)\in \RP^2$, or
2186: in other words to the projective line $F(y)^\perp$ in $\RP^{2*}$.
2187: This shows that $\Phi_* \nabla$ has the same geodesics as 
2188: the Levi-Civita connection $\triangledown$ of the standard  metric
2189: $g$ on $\RP^{2*}$, so that $\Phi^*\triangledown$ is projectively equivalent
2190: to $\nabla$. Identifying $\RP^2$ with $\RP^{2*}$ via  any isometry
2191:    now proves the claim. 
2192:  \end{proof}
2193: 
2194: \begin{remark}\label{tricky}
2195: Much the same trick used to check the regularity of $\Phi_0$
2196: also allows one to show that  $\Psi : {\mathcal Z}\to {\mathcal N}$
2197: is actually $C^{k+1,\alpha}$ along $Z$.
2198: Indeed, let $\varsigma_0$, $\varsigma_1$ and $\varsigma_\infty$ be three smooth  sections 
2199: of ${\mathcal U}\to M$ over a coordinate domain $U\subset M$ whose values are
2200: all distinct at each point. In terms of our local coordinates $(x^1,x^2,\zeta)$, 
2201: these correspond to three complex-valued functions $\zeta_\ell (x) = \zeta (\varsigma_\ell)$,
2202: $\ell=0,1,\infty$,
2203: whose values are all distinct, and never real. Set
2204: $$\tilde{\zeta} (x,\zeta)= \frac{[\zeta - \zeta_0(x)][\zeta_\infty(x)-\zeta_1(x)]}{[\zeta_1(x) -\zeta_0(x)][\zeta_\infty(x)
2205:  -\zeta]},$$ 
2206: so that $\tilde{\zeta}(x,\zeta_\ell(x))=\ell$  for each $x=(x^1,x^2)$ and 
2207: $\ell=0,1,\infty$. Choose  an inhomogeneous coordinate system on $\CP_2$
2208: such  that $z^1(F(\Psi(\varsigma_\ell(0,0)))$, $\ell=0,1,\infty$, are 
2209: all finite and distinct, 
2210: and, for $x$ in a neighborhood of $0$,  
2211: set
2212: $$ (z^1_\ell (x), z^2_\ell(x)) =F \circ \Psi (\varsigma_\ell (x^1,x^2)), ~~ \ell = 0,1,\infty. $$
2213: Then, in these coordinates, $F\circ\Psi$ must explicitly be given by 
2214: $$(x,\zeta) \mapsto \left( 
2215: \frac{\lambda z^1_0(x)+\tilde{\zeta}(x,\zeta)z^1_\infty(x)}{\lambda(x)+\tilde{\zeta}(x,\zeta)},
2216: \frac{\lambda z^2_0(x)+\tilde{\zeta}(x,\zeta)z^2_\infty(x)}{\lambda(x)+\tilde{\zeta}(x,\zeta)}
2217: \right),$$
2218: where
2219: $$
2220: \lambda (x^1,x^2)= \frac{z^1_\infty(x)-z^1_1(x)}{z^1_1(x)-z^1_0(x)},
2221: $$
2222: since each $\CP_1$ fiber of ${\mathcal Z}\to M$ is sent to holomorphically to a projective line 
2223: in $\CP_2$ by $F\circ \Psi$. If $\nabla$ is $C^{k,\alpha}$, 
2224: this shows, albeit quite indirectly, that $\Psi$ is 
2225: $C^{k+1,\alpha}$ on all of  ${\mathcal Z}$, and not just on 
2226: ${\mathcal U}= {\mathcal Z}-Z$. Needless to say, however, a direct analytic proof of this
2227: fact, perhaps along the lines of \cite{eastgrah}, would be highly desirable. 
2228: \end{remark}
2229: 
2230: If we start with a Zoll metric $h$ on $M=\RP^2$, rather than just a 
2231: Zoll projective structure, the complex surface $\mathcal N$ comes
2232: equipped with  a \label{conehead} 
2233: certain additional  complex curve
2234: ${\mathcal Q} \subset {\mathcal N}$. Indeed, let  us consider the locus 
2235: $${\mathcal C} = \{ [v] \in \PP T_\CC M ~|~ h(v,v) = 0\}, $$
2236: where $h$ has been extended from $TM$ to $T_\CC M$ as a 
2237: {\em complex bilinear} form, and set
2238: $${\mathcal Q} = \Psi [{\mathcal C}].$$
2239:  In any inhomogeneous coordinate $\zeta$
2240: on the fiber $T_{x\CC}M$, $h(v,v)$ becomes a quadratic polynomial of
2241: degree  $2$, and the corresponding locus in $\PP T_{x\CC}M$
2242: thus consists of  two points, perhaps counted with multiplicity. However, since $h$ is real,
2243: ${\mathcal C}$ is invariant under complex conjugation, so a root of multiplicity 
2244: two would have to lie in  the real slice $\PP T_x M$; but the latter is impossible, 
2245: since $h$ is a positive-definite inner product on $T_xM$. Thus
2246: ${\mathcal C}$ intersects each fiber of $\PP T_\CC M$ in precisely 
2247: two points, neither of which is  in $\PP TM$. 
2248:  Indeed, if we choose to think of ${\mathcal U} = {\mathcal Z}-Z$
2249: as the bundle of all point-wise almost-complex structures on $M$, 
2250: $\mathcal C$ is  consists precisely of those almost-complex structures
2251:  which are orthogonal transformations of $T_xM$ with respect to $h$; and there
2252: are exactly two of these for each $x$, corresponding to the two possible 
2253: orientations of $T_xM$. 
2254: 
2255: 
2256: Now ${\mathcal C}$
2257: is horizontal with respect to the Levi-Civita connection $\triangledown$,
2258: since parallel transport preserves $h$. This not only implies that 
2259: ${\mathcal C}$ meets each fiber of $\PP T_\CC M$ transversely, but also, 
2260: more importantly, that  
2261: there is a non-zero element  $\Xi_0$
2262: of ${\bf D}$ which is   tangent to ${\mathcal C}$ at each point.
2263: Thus  ${\mathcal C}$
2264: is a complex curve in $\PP T_\CC M - \PP TM$, and its diffeomorphic image 
2265: ${\mathcal Q} = \Psi [{\mathcal C}]$ is a complex submanifold of  $\mathcal N$. 
2266: Since $\mathcal C$ is invariant under complex conjugation, the
2267: corresponding curve ${\mathcal Q}\subset {\mathcal N}$ is therefore
2268: invariant under the action of $\sigma : {\mathcal N}\to {\mathcal N}$. 
2269: Moreover, since $\mathcal C$ meets each fiber of $\PP T_\CC M$
2270: transversely, in two points $\not\in \PP TM$, it follows that
2271: $\mathcal Q$ meets $\Sigma_x$ transversely, in two points,
2272: for any $x\in M$. 
2273: 
2274: Also notice  that the bundle projection $\mu : 
2275: \PP T_\CC M\to M$ induces a $2$-to-$1$ covering map
2276:  $\varpi: {\mathcal C}\to M\approx \RP^2$,
2277: so ${\mathcal C}$ is therefore 
2278: compact --- and indeed, must be diffeomorphic to  $S^2$.
2279: Moreover, this covering map  $\varpi$ is a {\em conformal map} from the 
2280: Riemann surface ${\mathcal C}$ to the Riemannian manifold
2281: $(M,h)$, since 
2282: $$\varpi_* [T^{0,1}_{[v]}{\mathcal C}]=\mbox{span} (v)\subset T_\CC M , $$
2283: and $h(v,v)=0$.  With this observation in hand, we may now  prove the following:
2284: 
2285: \begin{thm} \label{rumple}
2286: Let $(M,h)$ be a 
2287:  Riemannian $2$-manifold whose geodesics are
2288: all embedded circles of length $\pi$. 
2289: If $M$ is not simply connected, there is a   diffeomorphism 
2290: $\Phi : M\stackrel{\approx}{\longrightarrow} \RP^2$ such that   
2291: $h=  \Phi^* g$,
2292: where $g$ is the standard curvature $1$ 
2293: Riemannian metric 
2294:  on $\RP^2$. \end{thm}
2295: \begin{proof}
2296: With these hypotheses, 
2297: the Hopf-Rinow theorem tells us  that   $M$ is  necessarily compact, since,
2298: for any $x\in M$, 
2299: the  closed disk of radius $\pi/2$ in  
2300: $T_xM$ will surject  
2301: onto $M$ under the exponential map.  Proposition \ref{others},
2302: therefore tells us that $[\triangledown ]$ is a Zoll projective structure
2303: on the compact surface $M$. Now assume henceforth that $M$ is not simply connected, 
2304: We then know that $M\approx \RP^2$ by Proposition \ref{class}, 
2305: and that $[\triangledown ]$    has conjugacy number $1$ by 
2306: Theorem \ref{jeeves}. 
2307: 
2308: Now the proof of Theorem \ref{rigid} tells us that 
2309: that there is a biholomorphism $F: {\mathcal N}\to \CP_2$ such that 
2310: the  $F(\Sigma_x)$ is a projective lines $\CP_1\subset \CP_2$
2311: for each $x\in M$, and such that $F\circ \sigma \circ F^{-1}$ is the 
2312: complex conjugation map 
2313: $$
2314: [ z^1 : z^2 : z^3 ]  \mapsto  [\bar{z}^1: \bar{z}^2: \bar{z}^3 ] .
2315: $$
2316: Thus, $F( {\mathcal Q})$ is a non-singular compact complex curve in 
2317: $\CP_2$ which is invariant under complex conjugation, and which 
2318: meets certain projective lines transversely, in two points. Hence 
2319:  $F( {\mathcal Q})$ is a non-singular conic,  and  
2320: so is the zero locus of a quadratic polynomial 
2321: $$0= q(z)= \sum_{j,k=1}^3q_{jk}z^jz^k.$$
2322: But since  $F( {\mathcal Q})$ is invariant under complex conjugation,
2323: it is also the zero locus of $\overline{q(\overline{z})}$, so that both
2324: $$\sum_{j,k=1}^3(\Re e ~q_{jk})z^jz^k
2325: ~~~\mbox{ and }~~~
2326: \sum_{j,k=1}^3(\Im m ~q_{jk}) z^jz^k$$
2327: vanish along $F( {\mathcal Q})$; and at least one of these
2328: quadratic forms is non-trivial, since $q\not\equiv 0$. 
2329: Thus $F( {\mathcal Q})$ is the zero locus of a real 
2330: quadratic form, represented by a real symmetric $3\times 3$ matrix 
2331: $A= [a_{jk}]$. But any such $A$  is similar, over $GL(3, \RR  )$, to
2332: a diagonal matrix whose  entries are all in $\{  1, 0, - 1\}$.  
2333: On the other hand, 
2334: since $F ({\mathcal Q}) \cap \RP^2= \emptyset$,
2335:   the quadratic form represented by  $A$  must be 
2336:  definite. Thus, by a suitable real change of 
2337: coordinates,  we may arrange for
2338: our map $F: {\mathcal N}\to \CP_2$ to send
2339: $\mathcal Q$ to the standard conic ${\mathcal Q}_0$ given by 
2340: $$ (z^1)^2+ (z^2)^2+ (z^3)^2 =0$$
2341: without sacrificing any of the previously used properties of $F$.
2342: 
2343: 
2344: On the other hand, we can repeat the entire construction for the 
2345: standard metric $g$ on $\RP^2$. The map $\Phi: M\to \RP^2$
2346: constructed in Theorem \ref{rigid} is then characterized by 
2347: $$\Phi (x) = \tilde{x} \Longleftrightarrow F(\Sigma_x)= \tilde{F}(\tilde{\Sigma}_{\tilde{x}})$$
2348: where untilded letters pertain to $(M,h)$ and tilded ones pertain to 
2349: $(\RP^2, g)$. But since we have arranged for both ${\mathcal C}$ and 
2350: $\tilde{\mathcal C}$ to map biholomorphically to ${\mathcal Q}_0\subset \CP_2$,
2351: it follows that 
2352: \begin{eqnarray*}
2353:   F\left[\Psi [\varpi^{-1}(x)]\right]&=& F(\Sigma_x)\cap {\mathcal Q}_0\\
2354: \tilde{F}\left[\tilde{\Psi} [\tilde{\varpi}^{-1}(\tilde{x})]\right]
2355: &=& \tilde{F}(\tilde{\Sigma}_{\tilde{x}})\cap {\mathcal Q}_0.
2356: \end{eqnarray*}
2357: The holomorphic map $$\hat{\Phi} = \left(\left. (\tilde{F}\circ \tilde{\Psi})\right|_{\tilde{\mathcal C}}
2358: \right)^{-1}\circ (F\circ \Psi ) :
2359: {\mathcal C}\to \tilde{\mathcal C}$$
2360: therefore makes the diagram 
2361: \setlength{\unitlength}{1ex}
\begin{center}\begin{picture}(20,17)(0,3)
\put(2,17){\makebox(0,0){$\mathcal C$}}
2362: \put(18,17){\makebox(0,0){$\tilde{\mathcal C}$}}
\put(2,5){\makebox(0,0){$M$}}
2363: \put(18,5){\makebox(0,0){$\RP^2$}}
\put(0,12){\makebox(0,0){$\varpi$}}
2364: \put(20,12){\makebox(0,0){$\tilde{\varpi}$}}
2365: \put(10,6.5){\makebox(0,0){$\Phi$}}
2366: \put(10,19){\makebox(0,0){$\hat{\Phi}$}}
\put(18,15.5){\vector(0,-1){9}}
2367: \put(2,15.5){\vector(0,-1){9}}
2368: \put(3.5,17){\vector(1,0){13}}
2369: \put(3.5,5){\vector(1,0){12}}
\end{picture}\end{center}
2370: commute, and, since $\varpi$ and $\tilde{\varpi}$ are
2371: both conformal maps, it follows that $\Phi$ is also conformal.
2372: In other words, 
2373:  $\Phi^*g = e^{2u}h$ for some smooth function $u : M\to \RR$.
2374: But the Levi-Civita connection $\tilde{\nabla}$ of $\Phi^*g$
2375: is then related to the Levi-Civita connection $\nabla$ 
2376: of $h$ by
2377: $$\tilde{\nabla}_{\bf v}{\bf w}- \nabla_{\bf v}{\bf w} = du ({\bf v}) {\bf w} + du ({\bf w}) {\bf v} + 
2378: h({\bf v},{\bf w}) \mbox{ grad}_h u .$$
2379: However, the proof of Theorem \ref{rigid} tells us that $\tilde{\nabla}$ and
2380: $\nabla$ are also projectively equivalent; that is,
2381: $$\tilde{\nabla}_{\bf v}{\bf w}- \nabla_{\bf v}{\bf w} = \beta ({\bf v}) {\bf w} + \beta ({\bf w}) {\bf v} $$
2382: for some $1$-form $\beta$. 
2383: Thus 
2384: $$\beta ({\bf v}) {\bf w} + \beta ({\bf w}) {\bf v} = 
2385: du ({\bf v}) {\bf w} + du ({\bf w}) {\bf v} + 
2386: h({\bf v},{\bf w}) \mbox{ grad}_h u$$
2387: for all vectors ${\bf v}$ and ${\bf w}$. 
2388: But if, for example,   we take ${\bf v}$ and ${\bf w}$ to be 
2389: orthonormal, with 
2390: $\beta ({\bf w}) =0$, we  then have
2391: $ \beta ({\bf v}) ~ {\bf w}=
2392: du ({\bf v}) ~{\bf w} + du ({\bf w})~ {\bf v}$, so that 
2393: $du ({\bf v}) = \beta ({\bf v})$,  $du ({\bf w})=0=\beta ({\bf w})$; thus 
2394: $du$ and $\beta$ must have the same components in 
2395: the basis $({\bf v}, {\bf w})$, and 
2396: hence  $\beta = du$. But if  instead we take ${\bf w}={\bf v}\neq 0$,
2397: we  instead obtain
2398: $$
2399: 2 ~du ({\bf v}) ~{\bf v} + |{\bf v}|^2\mbox{ grad}_h u = 2~\beta ({\bf v}) ~{\bf v}, 
2400: $$
2401: and the substitution $\beta = du$ then tells us that $\mbox{grad}_h u=0$.
2402: Hence $u$ is constant. But, by hypothesis, 
2403:  $h$ is  normalized  so that its
2404: geodesic circles all have the same length as those of $g$. The constant
2405: $e^{2u}$ must therefore equal $1$, and $\Phi$ is therefore an isometry 
2406: between $(M,h)$ and $(\RP^2, g)$. 
2407: \end{proof} 
2408: 
2409: This is essentially equivalent \cite{beszoll} to the classical Blaschke conjecture
2410: first proved  by Leon Green \cite{grezoll} in the early 1960s. 
2411: 
2412: \begin{cor}[Blaschke Conjecture] 
2413: Let $(M,h)$ be a compact  
2414:  Riemannian \linebreak 
2415: $2$-manifold for  which  the cut locus of each 
2416: point $x\in M$ is a  one-point set $\{ x'\} \subset  M$. Then 
2417: there is a diffeomorphism 
2418: $\Phi : M\stackrel{\approx}{\longrightarrow} S^2$ such that   
2419: $h= c \Phi^* g$,
2420: where  $g$ is the standard curvature $1$ 
2421: Riemannian metric 
2422:  on $S^2$, and $c$ is some positive constant. \end{cor}
2423: 
2424: \begin{proof}
2425: On a compact Riemannian manifold, any minimizing geodesic 
2426: segment necessarily has finite length, so every arc-length-parameterized geodesic 
2427: emanating from $x$ must arrive at the cut locus $\{ x'\}$, and must first do so 
2428: precisely at  time  $\mbox{dist}(x,x')$. But since $x'$ represents the first
2429: conjugate point on each geodesic leaving $x$, we see, by following these
2430: geodesics backwards, that $x$ is an element of the cut locus of $x'$, and our
2431: hypothesis therefore implies that the cut locus of $x'$ is {\em exactly} 
2432: $\{ x\}$. Thus $x\mapsto x'$ is an involution $\imath : M\to M$. Moreover, every geodesic
2433: of $M$ is a simple closed curve, and $\imath$ maps every such
2434: geodesic circle to itself, by a rotation of $180^\circ$. In particular,  
2435:  $\imath$ is an isometry, and  is therefore smooth. Moreover, $\mbox{dist}(x,\imath (x))$
2436: is independent of $x$ along any particular geodesic, and thus 
2437: is constant on $M$. Thus 
2438:  the geodesics of the 
2439: the quotient Riemannian metric on 
2440: $M/\langle \imath \rangle$ are all 
2441: simple closed  curves of equal length.
2442: After a suitable rescaling, Theorem \ref{rumple} therefore tells us that 
2443: the non-simply-connected Zoll manifold 
2444: $M/\langle \imath \rangle$ becomes isometric to the standard $\RP^2$, and
2445: hence that $M$ becomes isometric to the standard $S^2$. 
2446: \end{proof}
2447: 
2448: \begin{remark}
2449: Since the Christoffel symbols of the Levi-Civita connection of $h$ are
2450: expressed  in terms of the first derivatives of $h$, Theorem \ref{rigid} 
2451:  constructs an isometry of class $C^{k+2,\alpha}$
2452: when we assume that $h$ is itself of class $C^{k+1,\alpha}$, $3\leq k$,
2453: $0< \alpha < 1$.   Thus the regularity of the map $\Phi$
2454: in Theorem \ref{rumple} is actually  
2455: optimal, as one  certainly has every right to expect. 
2456: 
2457: It is  more important, however,  to inquire as to
2458:  the minimal level of differentiability  
2459: needed for our proof of Theorem \ref{rumple}.  
2460: If we  assume that $h$ is of class $C^4$, then the proof goes through,
2461: although the  constructed map $\Phi$ would appear only  
2462: to be $C^4$. Nonetheless, 
2463:  $\Phi^*g$ is still $C^3$, and its Gauss curvature
2464: is therefore the pull-back of the Gauss curvature of $g$. This
2465: shows any $C^4$  Zoll metric 
2466: $h$ on $\RP^2$ must have constant curvature. 
2467: However,    Green's proof  \cite{grezoll} actually 
2468: draws the same conclusion even if  $h$ is merely assumed to be $C^3$. 
2469: It would thus be extremley gratifying if  there were 
2470:  some way of improving the present arguments so as to 
2471: make them  work when, for example,  $[\nabla]$ is  merely asssumed to be 
2472: of class $C^2$!
2473: \end{remark}
2474: 
2475: \pagebreak 
2476: 
2477: 
2478: 
2479: \section{Zoll Structures on the $2$-Sphere}
2480: \label{zoe}
2481: 
2482: In light of our success in understanding Zoll structures of odd conjugacy number, 
2483: it now seems reasonable to ask what our  techniques can tell us about 
2484: the even case.  Let us 
2485: therefore suppose that we are given a $C^3$ Zoll projective structure 
2486: $[\nabla ]$ of even conjugacy number on a compact $2$-manifold $M$. By 
2487: Corollary \ref{uppercase},  $M$ is then diffeomorphic to $S^2$.
2488: Let us fix some orientation of $M$, and observe that 
2489: $${\mathcal U}={\mathcal Z}-Z=
2490: \PP T_\CC M - \PP TM$$
2491:  can once again be identified with the space of 
2492: all point-wise almost-complex structures on $M$. 
2493: Thus 
2494: $${\mathcal U} = {\mathcal U}_+ \cup {\mathcal U}_- , $$
2495: where 
2496: ${\mathcal U}_+$ (respectively, ${\mathcal U}_-$) consists
2497: of those almost-complex structures which are compatible 
2498: (respectively, incompatible)
2499: with the given orientation 
2500: of $M$. These are both connected sets; indeed, either can 
2501: be identified with the space of all point-wise conformal structures
2502: on $M$.  Let us now consider the compact $4$-manifold-with-boundary
2503: $${\mathcal Z}_+:={\mathcal U}_+\cup Z,$$
2504: with   $\partial {\mathcal Z}_+ =Z$.
2505: We can identify 
2506: ${\mathcal Z}_+$  with the non-zero, semi-positive
2507: elements of $\odot^2T^*M$, modulo rescaling.  
2508: Relative to some chosen `background' metric 
2509: $h_0$ on $M\approx S^2$, we can then identify  
2510: ${\mathcal Z}_+\to M$ as the  unit disk bundle in  the traceless,
2511: symmetric bilinear forms $\odot^2_0T^*M$. 
2512: From a topological view-point, this 
2513: allows us to think of   ${\mathcal Z}_+$ as
2514:  the unique oriented $2$-disk bundle of 
2515: Euler class $4$ over $S^2$. 
2516: 
2517: Let us now give the normal bundle $J^\parallel \ker\mu_*$
2518: of $Z=\partial {\mathcal Z}_+$ the 
2519: `inward pointing' orientation, and then give $\ker \mu_*$
2520: the corresponding orientation. Having made such a choice, 
2521: Theorem \ref{wooster} then tells us that $\nu : Z\to N$ can be 
2522: canonically identified with the circle bundle ${\mathbb S}TN\to N$,
2523: in such a way that $J^\parallel \ker\mu_*$ is canonically
2524: identified with the pull-back of the (trivial) tautological 
2525: line bundle over ${\mathbb S}TN$, meaning the 
2526:  sub-bundle  
2527: $L\subset \pi^* TN$, where $\pi :{\mathbb S} TN \to N$ is the canonical 
2528: projection, whose fiber at $[{\bf v} ] \in \PP TN$
2529: is $\mbox{span} ({\bf v})$. Now, with respect to  the canonical `outward pointing' orientation 
2530: of $L\to {\mathbb S}TN$, let 
2531: $L^+$ be the $[0,\infty )$-bundle consisting of vectors
2532: which are not inward pointing. By the tubular neighborhood
2533: theorem, $Z= \partial {\mathcal Z}_+$ has a neighborhood $\hat{\mathcal V}$ in 
2534: ${\mathcal Z}_+$ which can be identified with $L^+$
2535: via a $C^1$ diffeomorphism whose derivative along the zero section 
2536: of $L$ is given by our previous identification of $J^\parallel \ker\mu_*$
2537: and $L$. But we have an obvious  $C^1$ `blowing down' map 
2538: $\psi : L^+\to TN$, and, letting $ {\mathcal V}$ denote the 
2539: total space of $TN$, 
2540:  this now corresponds to
2541: a $C^1$ map $\tilde{\psi} : \hat{\mathcal V}\to {\mathcal V}$
2542: which is a diffeomorphism on the complement of $Z$. 
2543: We may now define a 
2544:  differentiable  
2545:  $4$-manifold  
2546: $${\mathcal N} = {\mathcal U}_+ \cup_{\tilde{\psi}} {\mathcal V}$$
2547: by gluing together ${\mathcal U}_+$
2548: and ${\mathcal V}=TN$ via $\tilde{\psi}$. 
2549:  By construction, 
2550: we   have a surjective $C^1$ 
2551: `blowing down' map 
2552: $$\Psi : {\mathcal Z}_+ \to {\mathcal N},$$
2553: given by the identity on ${\mathcal U}_+$ and by $\tilde{\psi}$ on $\hat{\mathcal V}$,
2554: so in particular we know that ${\mathcal N}$ is compact. 
2555: Moreover, if $[\nabla ]$ is $C^k$, we can once again 
2556: impose a `provisional' $C^{k-1}$ structure on 
2557: $\mathcal N$ so that $\Psi$ will become a $C^{k-1}$ map. 
2558: 
2559: 
2560: Now $\mathcal Z$ still carries an involutive complex distribution 
2561: $\bf D$, and the proof of Proposition \ref{key}, supplemented by the
2562: remark on pp. \pageref{rocky}---\pageref{road},  then proves the following:  
2563: 
2564: 
2565: 
2566: \begin{prop}\label{cle} Let $[\nabla]$ be a Zoll projective structure which is 
2567: represented by a $C^3$ connection $\nabla$ on $M\approx S^2$. Then 
2568: there is a unique integrable almost-complex structure $J$ on $\mathcal N$ such that 
2569: $$\Psi_* [{\bf D}] \subset T^{0,1}({\mathcal N}, J).$$
2570: The unique $C^\infty$  structure on ${\cal N}$
2571: associated with its maximal  atlas of  $J$-compatible complex charts
2572: is compatible with the previously-constructed $C^1$ structure on 
2573: $\mathcal N$, so that  $\Psi : {\mathcal Z} \to {\mathcal N}$  remains a $C^1$ map relative to 
2574: this smooth structure.
2575: Moreover, if  $\nabla$ is of class $C^{2k+6}$, then $\Psi$ is $C^k$. 
2576: \end{prop}
2577: 
2578: 
2579: In order to unmask the identity of the complex surface $({\mathcal N}, J)$, 
2580: we will now call in the heavy artillery, in the form of the 
2581:  following fundamental result, which  is   due to Yau \cite{yau}.
2582: We include the synopsis of a complete proof,
2583: both as a courtesy to the reader, and for our    own enjoyment.
2584: 
2585: \begin{lem}[Yau] \label{wao}
2586: Let ${\mathcal S}$ be a simply connected compact complex surface
2587: with $b_2({\mathcal S})=1$. 
2588:  Then ${\mathcal S}$ is biholomorphic to
2589: $\CP_2$.
2590: \end{lem}
2591: \begin{proof}
2592: Any   compact, oriented,   simply connected
2593: $4$-manifold ${\mathcal S}$ has   Euler characteristic 
2594: $\chi ({\mathcal S})  = 2+ b_2({\mathcal S})$, so that 
2595: $\chi ({\mathcal S})  = 3$ if $b_2({\mathcal S})=1$.
2596: On the other hand,  if $b_2({\mathcal S})=1$, the
2597:  signature $\tau ({\mathcal S})$ is evidently 
2598:  $\pm 1$, where the  $\pm$  sign indicates whether   the intersection form of 
2599: ${\mathcal S}$ is positive or  negative definite. 
2600: But our $\mathcal S$ is assumed to admit a  complex structure, so its first Chern class
2601:  has self-intersection 
2602: $$
2603: c_1^2 ({\mathcal S}) = 2\chi ({\mathcal S}) + 3\tau ({\mathcal S}) = 6\pm 3 > 0,
2604: $$
2605:  and the intersection form 
2606: $H^2({\mathcal S}, {\ZZ}) \times H^2({\mathcal S},
2607: {\ZZ}) 
2608: \to \ZZ$ therefore 
2609: cannot be negative definite. Thus  $\tau ({\mathcal S})=1$, and 
2610: $c_1^2 ({\mathcal S})= 6+ 3 = 9$. 
2611: Since this same calculation also shows that there is a holomorphic line bundle
2612: of positive self-intersection, Grauert's criterion implies \cite{bpv} that 
2613: ${\mathcal S}$ is projective algebraic.  But since  $H^2({\mathcal S}, \ZZ)\subset 
2614: H^2({\mathcal S}, \RR)\cong \RR$, 
2615:  and 
2616: $c_1 ({\mathcal S})\neq 0$,  this can only happen if 
2617:  $c_1  ({\mathcal S}) = \pm [\omega ]$ for some K\"ahler form $\omega$. 
2618: 
2619: 
2620: Now if we had
2621:  $c_1  ({\mathcal S}) = - [\omega ]$, the Aubin/Yau  theorem \cite{aubin,yau} 
2622: would tell us that 
2623: $\mathcal S$ admitted a K\"ahler-Einstein metric  of negative Ricci curvature. 
2624: However, one has the Gauss-Bonnet-like formula
2625: $$
2626:  \chi - 3\tau = \frac{1}{8\pi^2} \int_{\mathcal S} \left[
2627: 3 |W_-|^2 - \frac{|\stackrel{\circ}{r}|^2}{2} 
2628: \right]d\mu$$
2629: for any K\"ahler metric on any compact complex surface, 
2630: where $\stackrel{\circ}{r}$ is the trace-free  Ricci-curvature, and 
2631: where the anti-self-dual Weyl  curvature 
2632: $W_-$ is  the only piece of the 
2633: curvature tensor not determined by the Ricci tensor. 
2634: For our manifold, $\chi = 3\tau$, whereas $\stackrel{\circ}{r}$ vanishes
2635: for any Einstein metric, so we would conclude that $W_-\equiv 0$.
2636:  Our K\"ahler-Einstein manifold would therefore necessarily 
2637: have negative sectional curvature,
2638: and so would have  contractible universal cover.  
2639: But   ${\mathcal S}$ has been assumed to be  compact and simply connected,  so this is a
2640: contradiction. 
2641: 
2642: We must therefore have  $c_1  ({\mathcal S}) =  [\omega ]$
2643: for some K\"ahler metric. Set $L= K^{-1/3}$, where $K=\Lambda^{2,0}$
2644: is once again the canonical bundle,  so that $L$ is  the unique positive line bundle on 
2645: ${\mathcal S}$ with $c_1^2 (L) =1$.  By the Kodaira vanishing theorem,
2646: $H^p ({\mathcal S}, {\mathcal O}( L))=0$ for  $p > 0$, and the Hirzebruch-Riemann-Roch
2647: theorem therefore tells us that 
2648: $$h^0 ({\mathcal S}, {\mathcal O}( L)) =\left\langle \left(1+\frac{c_1}{2}+\frac{c_1^2+ 
2649: c_2}{12}\right) \exp (\frac{c_1}{3}) , [{\mathcal S}]\right\rangle 
2650: = \frac{11}{36}c_1^2 +
2651: \frac{1}{12} c_2 = 3.$$
2652: Moreover, if $\Sigma \subset {\mathcal S}$ is the curve cut out by 
2653: the vanishing of any non-trivial  holomorphic section of $L$, then,
2654: because $L$ is positive on every curve  and satisfies $L\cdot L =1$, 
2655:  $\Sigma$ can have only 
2656: one irreducible component, and the zero of the section can only 
2657: have multiplicity $1$ at a generic point of $\Sigma$.  
2658: %Since the ideal sheaf   of $\Sigma$ therefore
2659: %coincides with ${\mathcal O}(L^*)$, we have an exact sequence
2660: %$$0\to {\mathcal O}\to {\mathcal O}(L) \to \O_\Sigma (L)\to 0,$$
2661: %and this tells us that there is a $2$-dimensional space of 
2662: %sections of $L|_\Sigma$ which arise as restrictions of
2663: %holomorphic sections of $L$ on $\mathcal S$. 
2664: If
2665: $\hat{\Sigma}$ is the normalization of $\Sigma$, the pull-backs of these
2666:  sections  therefore give us a $2$-dimensional 
2667: space of sections of the degree-$1$ line bundle  $L|_{\hat{\Sigma}}$.
2668: But since $\hat{\Sigma}$ is connected, Abel's theorem tells us that 
2669: this gives us
2670:   a biholomorphism $\hat{\Sigma}\to \CP_1$; and   since 
2671: this map  is induced  by  pull-backs of  sections
2672: from ${\mathcal S}$,  
2673: $\hat{\Sigma}\to {\mathcal S}$ is an embedding, so that 
2674:  $\Sigma= \hat{\Sigma}$ is  a non-singular embedded  curve. 
2675: Moreover, there is no point of $\Sigma$ at which every section 
2676: of $L$ vanishes. This shows that  the linear system   $|L|$
2677: has empty base locus, and 
2678: %,
2679: %since we can produce a section which is non-zero at any point of 
2680: %any chosen $\Sigma$; 
2681: %in other words,  we have shown that the base locus of the system 
2682: %$L$ is empty. 
2683: the sections of $L$ therefore give us a 
2684: well-defined holomorphic map
2685: $$
2686: F : {\mathcal S}  \to \PP [H^0({\mathcal S}, {\mathcal O}( L))^*]\cong \CP_2 .
2687: $$
2688: But since the inverse image of any $\CP_1\subset \CP_2$ is a smooth complex 
2689: curve $\Sigma$ which is carried biholomorphically onto its image, this
2690: map is a degree-$1$ holomorphic submersion, and  is therefore a biholomorphism.  
2691: \end{proof}
2692: 
2693: 
2694: Let us next recall that a differentiable 
2695: $n$-dimensional submanifold $X$ of a complex $n$-manifold 
2696: $(Y^{2n}, J)$ is said to be {\em totally real} if 
2697: $T_pX\cap J(T_pX) =0$ at each $p\in X$. 
2698: When $n=2$, which is the case of interest to us here, 
2699: this is equivalent to the statement that $T_pX$ is never
2700: a $1$-dimensional complex  subspace of $(T_pY, J) \cong \CC^2$. 
2701: This  is  of course an open condition on $T_pX$; indeed, 
2702: for $n=2$, this simply amounts to the observation that 
2703: since $Gr_{1}(\CC^2)=\CP_1$ is a closed submanifold of 
2704: $Gr_2(\RR^4) \cong (S^2\times S^2 ) /\ZZ_2$. 
2705: To that  any submanifold which is $C^1$ close to a totally 
2706: real submanifold will itself  be totally real.  
2707: 
2708: It  will also be convenient to introduce some terminology specifically tailored 
2709: to discussions of 
2710: differentiable embeddings of $\RP^2$ into $\CP^2$. 
2711: 
2712: \begin{defn}
2713: A differentiable  embedding  $\jmath: \RP^2\hookrightarrow \CP^2$ will be said to be 
2714: {\em weakly unknotted} if there exists a diffeomorphism $\mbox{\cyr   f}: \CP_2 \to \CP_2$ such that $\jmath = \mbox{\cyr   f}\circ j$, where $j: \RP^2\hookrightarrow \CP^2$ 
2715: is the standard embedding $[x:y:z]\mapsto [x:y:z]$.
2716: \end{defn}
2717: 
2718: \begin{remark}
2719: By composing  with   complex conjugation $\CP_2\to \CP_2$ 
2720:  if necessary, we may always arrange for
2721:  $\mbox{\cyr   f}$ to  induce the identity on homology.  But since  two self-homeomorphisms of a simply connected 
2722: compact $4$-manifold are $C^0$-isotopic iff they induce the same
2723: maps on homology \cite{frequin}, our diffeomorphism
2724:   $\mbox{\cyr  f}$ would then be in the identity component of 
2725: of the homeomorphism group of $\CP_2$. Thus any weakly unknotted embedding 
2726: of $\RP^2$ in $\CP^2$, as defined above,  may  be moved through locally flat
2727: {\sl topological}  embeddings 
2728: so as to ``unknot'' it into the standard $\RP^2$. {\em A priori}, however, 
2729: it might still be impossible to carry out this unknotting process by
2730: a path of {\em smooth} embeddings.
2731: \end{remark}
2732: 
2733: \begin{thm} Let $[\nabla]$ be a $C^3$ Zoll projective structure on
2734: an oriented surface 
2735: $M\approx S^2$. Then,
2736: up to  a projective linear transformation, the projective
2737: structure 
2738:  $[\nabla]$ uniquely determines a differentiable, totally real,
2739: weakly unknotted
2740:  embedding
2741: of the space of geodesics $N\approx \RP^2$ into $\CP_2$. 
2742: If $[\nabla]$ is $C^\infty$, so is the embedding. 
2743: Moreover, the image of each of the circles $\ell_x\subset N$, $x\in M$,
2744:  bounds a holomorphic embedding  of the 
2745: disk $D^2\hookrightarrow \CP_2$, and the interiors
2746: of these disks foliate the  complement $\CP_2-N$.
2747: \end{thm}
2748: 
2749: 
2750: 
2751: \begin{proof} By construction, the smooth $4$-manifold ${\mathcal N}$  can be
2752:  obtained by gluing  the unit disk bundle in $T\RP^2$
2753: to the  Euler-class-$4$ $D^2$ bundle over $S^2$ via an orientation-reversing diffeomorphism 
2754: of their common boundary, which is the Lens space $X=S^3/\ZZ_4$. 
2755: However, the diffeomorphism type of the pair $({\mathcal N},N)$ 
2756: only depends on the isotopy class of the diffeomorphism $X\to X$.
2757: But the group of orientation-preserving diffeomorphisms of $X$ is connected
2758: \cite{diffm3}, so it follows that the diffeotype of  the pair 
2759: $({\mathcal N},N)$  is independent of which  Zoll projective structure 
2760: $[\triangledown ]$ on $S^2$ we use. However, the standard structure 
2761: $[\nabla ]$ gives us the pair 
2762: $(\CP_2 , \RP^2 )$.  Thus 
2763: there is a diffeomorphism $\mbox{\cyr \em f}: \CP_2 \to {\mathcal N}$
2764: with $\mbox{\cyr \em  f}(\RP^2 )=N$.
2765: 
2766: 
2767: In particular, this argument 
2768: says that ${\mathcal N}$ is diffeomorphic to $\CP_2$. 
2769:  Lemma \ref{wao} therefore
2770: tells us that there is a {\em biholomorphism} $F: {\mathcal N}\to  \CP_2$, 
2771: and this $F$ is unique modulo composition with elements of    $PSL (3, \CC )$. 
2772: The promised embedding $N\hookrightarrow \CP_2$ is then given by 
2773: $F|_N$, 
2774: whereas the promised disks are the images of the the fibers of 
2775: ${\mathcal Z}_+\to M$ under $F\circ \Psi$. 
2776: Moreover,  since the diffeomorphism $\mbox{\cyr  f}= F\circ \mbox{\cyr \em  f}: \CP_2 \to \CP_2$  
2777: sends 
2778: $\RP_2$ to  $F(N)$, our embedding $F|_N$
2779: is weakly unknotted,  and we are done. 
2780: \end{proof}
2781: 
2782: 
2783: 
2784: 
2785: Now, in order to invert the above onstruction, let us  instead suppose that we 
2786: are given a totally real submanifold $N\approx \RP^2$ of $\CP_2$,
2787: and attempt to construct a suitable family of holomorphic disks $D\hookrightarrow \CP_2$
2788: with boundary $\partial D = S^1 \hookrightarrow N$;  these circles
2789: in $N$ will then eventually become the curves $\ell_x$ corresponding to a
2790: Zoll projective structure on $S^2$. Our method of  accomplishing this
2791: will be to invoke  the inverse function theorem, and so 
2792: will apply only when the given 
2793:  embedding $N\hookrightarrow \CP_2$
2794: is $C^1$ close to the standard  embedding 
2795: $\RP^2 \hookrightarrow \CP_2$. Thus, relative to a choice of tubular neighborhood,
2796: we will 
2797:  henceforth assume that $N$  is represented by 
2798: by a section of the normal bundle of $\RP^2$. This allows us a
2799: further technical simplification, since,  if such a section
2800: has sufficiently small $C^1$-norm. Also notice that the
2801: normal bundle of $\RP^2\subset \CP_2$ can be canonically identified,
2802: via the complex structure, with $T\RP^2$, and so may also be identified 
2803: with $T^*\RP^2$ by means of the standard Riemannian metric.
2804: Thus the freedom of choosing the submanifold $N\subset \CP^2$ 
2805: can be conveniently parameterized by the space of $1$-forms on 
2806: on $\RP^2$ of sufficiently small $C^1$-norm. 
2807: 
2808: For the standard projective structure on $S^2$, the 
2809: disks in question are obtained by considering those
2810: complex projective lines $\CP_1\subset \CP_2$ which 
2811: are complexifications of some real projective line 
2812: $\RP^1\subset \RP^2$, and then choosing one of
2813: the hemispheres into which such a  $\CP_1$
2814: is divided by the corresponding $\RP^1$. 
2815: In order to understand these disks more explicitly, let us begin with the 
2816:  standard homogeneous coordinates
2817: $[z_1 : z_2 : z_3 ]$ on $\CP_2$, with the usual convention that  
2818: $\RP^2$ is represented by $z_1, z_2 , z_3$ real, 
2819: and consider the affine chart  $(\zz_1 , \zz_2 )$ on $\CP_2$ defined by 
2820: $$
2821: \zz_1  =  \frac{z_1-iz_2}{z_1+iz_2} ~, \hspace{0.5in}
2822: \zz_2  =  \frac{z_3}{z_1+iz_2}~.
2823: $$
2824: This chart realizes  $\RP^2-[0:0:1]$ as the M\"obius band $B\subset \CC^2$ 
2825: given by 
2826: $$
2827: \zz_1\overline{\zz}_1  =  1  ~, \hspace{0.5in}
2828:  \zz_1\overline{\zz}_2 = \zz_2 .  
2829: $$
2830: Note that  we may also parameterize $B$ by 
2831: \begin{eqnarray*}
2832: \zz_1 & = & e^{i\theta} \\
2833: \zz_2 & = & t e^{i\theta /2} ,
2834: \end{eqnarray*}
2835: where  the real coordinates  $(\theta , t)$ are  best  thought of  as really 
2836: taking values in the {\em abstract} M\"obius band $\RR^2/\ZZ$
2837: corresponding to    the $\ZZ$-action  generated by 
2838: $$ (\theta , t ) \mapsto (\theta + 2\pi , -t ).$$
2839: 
2840: Now the projective line $z_3=0$ in $\CP_2$ corresponds, in this picture, to 
2841: the complex affine line $\zz_2=0$; and one hemisphere of this $\CP_1$
2842: is  the disk $|\zz_1| \leq 0$ in this affine complex 
2843: line, the  boundary of which is the circle 
2844: $\theta \mapsto (e^{i\theta}, 0)$  in $B$. 
2845: How many other ways can one  holomorphically  the disk $D\subset \CC$ 
2846: in $\CC^2$ in such a manner that its boundary $\partial D = S^1$ both lies on
2847: $B$, and is homotopic in $B$ to $\theta \mapsto (e^{i\theta}, 0)$? Projecting
2848: any such disk to the $\zz_1$ axis would give a degree-$1$ holomorphic map
2849: $D\to \CC$ with boundary map a degree-$1$ map $S^1\to S^1$, and any
2850: such map is of course given by a M\"obius transformation
2851: \begin{equation}
2852: \label{moby}
2853: \zeta \mapsto \frac{a\zeta  + b}{\overline{a}+ \overline{b}\zeta} ~, |a|^2-|b|^2 =1. 
2854: \end{equation}
2855: Thus, after composition with a M\"obius transfomation, 
2856: any such disk is the graph $\zz_2 = F (\zz_1)$ of a holomorphic function $F$ 
2857: on the unit disk $|\zz_1| \leq 1$. 
2858: However, the requirement that $F(\partial D)$ lie in $B$ says that 
2859: $$
2860:  F(e^{i\theta}) = e^{i\theta} \overline{F(e^{i\theta})}.
2861: $$ 
2862: If $F$ has power series expansion
2863: $$F(\zz_1 ) = \sum_{\ell =0}^\infty a_\ell \zz_1^\ell , $$
2864: our boundary condition becomes  
2865: $$
2866: \sum_{\ell=0}^\infty a_\ell e^{i\ell\theta} = \sum_{\ell=-\infty}^1 \overline{a}_{-\ell+1} e^{i\ell\theta}.
2867: $$
2868: Hence every such disk is   the graph of an affine linear function
2869: $$\zz_2 = a + \bar{a}\zz_1$$
2870: restricted to the unit disk $|\zz_1|\leq 1$, where $a=a_0$. 
2871: Each of these disks exactly represents one hemisphere of the 
2872: projective line $\CP_1\subset \CP_2$ given by 
2873: $$z_3= (2~\Re e ~a) ~z_1 + (-2~\Im m ~a) ~z_2,$$ 
2874:  and the boundaries of these disks are  thus 
2875: precisely  the real projective lines $\RP^1\subset \RP^2$
2876: which do not pass through the point $[0:0:1]$ which was excluded by our
2877: choice of coordinates. By considering all possible permutations of the 
2878: homogeneous coordinates $z_1,z_2,z_3$, one obtains the entire
2879: family of disks corresponding to the points of $S^2$ equipped with
2880: its standard projective structure. 
2881: 
2882: 
2883: We now consider the problem of constructing an analogous family of 
2884: disks with boundaries on a submanifold $N\subset \CP_2$ which is $C^1$ 
2885: near to $\RP^2\subset \CP_2$. To do this, it is enough to completely 
2886: analyze the corresponding problem arising when 
2887:  intersection of the M\"obius Band $B$ and a large ball
2888: is replaced with  a section of its normal bundle, since $N$ is covered
2889: by a finite number of pieces of this form. 
2890: 
2891: To this end, we will begin by considering maps of the circle $S^1$
2892: to the abstract M\"obius band $\RR^2/\ZZ$
2893: with winding number $1$. For reasons of technical transparency,
2894: we will consider maps of Sobolev class $L^2_k$, where $k\geq 1$.
2895: Let us recall  that the Cauchy-Schwarz inequality 
2896: immediately implies the Sobolev embedding theorem in this case, since
2897: any smooth, real valued  function $f$ on the line satisfies
2898: \begin{equation}
2899: \label{kosher}
2900: | f (a) - f(b) | \leq \left( \int_a^b\left|\frac{ df}{dx}\right|^2 dx\right)^{1/2} |a-b|^{1/2},
2901: \end{equation}
2902: whence $L^2_k (S^1)\subset C^{k-1,\frac{1}{2}}(S^1)$.
2903: In particular,  maps from the circle of class  $L^2_k$ are continuous,
2904: and   it  thus makes sense to talk about  winding numbers of such  maps. 
2905: Moreover, this shows that point-wise multiplication of functions
2906: gives us  a continuous bilinear map 
2907: $L^2_k (S^1)\times L^2_k (S^1)\to  L^2_k(S^1)$. Also note
2908: that  the composition of 
2909: of any $C^k$ function with an $L^2_k$ function is again an  $L^2_k$
2910: function. 
2911: 
2912: 
2913:  We will freely identify  $L^2_k(S^1)$ with the real Hilbert space of 
2914: real-valued $L^2_k$ functions of  $\theta \in [0,2\pi ]$ 
2915: with 
2916: $u(\theta +2\pi ) = u(\theta  )$, and we will also need to consider the
2917: real Hilbert space $\tilde{L}^2_k(S^1)$ of $L^2_k$ sections of the M\"obius band, 
2918: which we may think of as functions 
2919:  of $\theta \in [0,2\pi ]$ 
2920: with $u(\theta +2\pi ) = -u(\theta  )$. 
2921: %Either of these may of course be thought of as a
2922: %subspace of the complex Hilbert space $\tilde{L}^2_k(S^1,\CC)$
2923: %of complex-valued $L^2_k$ functions on the circle. 
2924: Since any continuous section of the M\"obius band must have a zero, 
2925: (\ref{kosher}) tells us that 
2926: any $u\in \tilde{L}^2_k$, $k\geq 1$,  satisfies 
2927: $$\sup |u| \leq \sqrt{\pi}\left( \int_0^{2\pi}\left|\frac{du}{d\theta}\right|^2d\theta \right)^{1/2}\leq \sqrt{\pi} \|u\|_{L^2_k} ,$$
2928: so the elements $u$ of the ball of radius $R/\sqrt{\pi}$ in $\tilde{L}^2_k$ may be thought of as
2929: defining a section $\theta \mapsto (\theta, u(\theta ))$ 
2930: of    the {\em finite} M\"obius strip 
2931: $$
2932: B^R = \left( \RR \times [-R ,  R ]\right)/\ZZ ,
2933: $$
2934: where the $\ZZ$ action is again generated by $(\theta , t ) \mapsto (\theta + 2\pi , -t )$. 
2935: We will use $C^k (B^R)$ to denote the real Banach space of $C^k$ real-valued
2936: functions on this strip, and 
2937: $$\tilde{C}^k(B^R)= \{ h: \RR \times [-R ,  R ]\stackrel{C^k}{\to} \RR ~|~ h(\theta +2\pi , -t) = 
2938: -h(\theta , t)
2939:  \} $$ to denote the real  Banach space
2940: of $C^k$ sections of the non-trivial real line bundle on $B^R$, the Banach-space norms 
2941: being of course the suprema of the absolute values of all partial derivatives of 
2942: order $\leq k$. 
2943: 
2944: Any pair $(h_1,h_2) \in C^{k+1}(B^R) \times \tilde{C}^{k+1}(B^R)$ defines 
2945: an embedding  $B^R\hookrightarrow \CC^2$  by 
2946: $$(\theta , t ) \mapsto \left(e^{h_1(\theta , t )+i\theta}, [t+ ih_2(\theta , t )]e^{i\theta /2}\right),$$
2947: and any $C^{k+1}$ submanifold $N\subset \CP_2$ which is sufficiently close to 
2948: the standard 
2949: $\RP^2 \subset \CP_2$ can be written as a finite union of images of 
2950: such embeddings of finite strips via suitable systems of inhomogeneous coordinates. 
2951: The general $L^2_k$ embedding of $S^1$
2952: inside this strip with winding number $1$ can then be written as
2953: $$\theta \mapsto  
2954:  \left(e^{h_1(\theta + u_1(\theta) , u_2(\theta) )+i[\theta + u_1(\theta)]},
2955:  [u_2(\theta)+ ih_2(\theta + u_1(\theta) , u_2 (\theta )  )]e^{i(\theta + u_1(\theta)) /2}\right)$$
2956: for $u_1\in L^2_k(S^1)$ and $u_2\in \tilde{L}^2_k (S^1)^R$, where 
2957:  $\tilde{L}^2_k(S^1)^R$ denotes the open ball of radius $R/\sqrt{\pi}$ 
2958: centered at the origin in $\tilde{L}^2_k(S^1)$.
2959: This motivates us to consider the  maps of Banach manifolds   
2960: $${\mathcal F}_1,  {\mathcal F}_2: 
2961: L^2_k (S^1) \times \tilde{L}^2_k(S^1)^R \times 
2962: C^{k+\ell}(B^R) \times \tilde{C}^{k+\ell}(B^R) \longrightarrow L^2_k (S^1, \CC) \times L^2_k (S^1, \CC), $$
2963: given by 
2964: $$
2965: [{\mathcal F}_1 (u_1 , u_2 , h_1 , h_2) ] (\theta ) = 
2966: \exp \left[ h_1\Big(\theta + u_1(\theta) , u_2(\theta) \Big)+i\Big( \theta + u_1(\theta)\Big)\right]
2967: $$
2968: and 
2969: $$
2970: [{\mathcal F}_2 (u_1 , u_2 , h_1 , h_2) ] (\theta ) = 
2971: \left[ 
2972: u_2(\theta)+ ih_2\Big(\theta + u_1(\theta) , u_2 (\theta )  \Big)\right]  \exp \left( i\frac{\theta + u_1(\theta)}{2} \right).
2973: $$
2974: These maps are both  $C^\ell$; in particular, for $\ell \geq 1$  they have bounded continuous
2975: derivatives  given by 
2976: $$
2977: (\dot{u}_1 , \dot{u}_2 , \dot{h}_1 , \dot{h}_2) ] 
2978: \stackrel{{\mathcal F}_{1*}}{\longmapsto}
2979: \Big[ \dot{h}_1(\theta +u_1, u_2) 
2980: +\Big( i+ \frac{\partial h_1}{\partial \theta}\Big) \dot{u}_1
2981:  + \frac{\partial h_1}{\partial t}\dot{u}_2  \Big] e^{h_1+i(\theta + u_1)}
2982: $$
2983: and 
2984: $$
2985: (\dot{u}_1 , \dot{u}_2 , \dot{h}_1 , \dot{h}_2) ] 
2986: \stackrel{{\mathcal F}_{2*}}{\longmapsto} 
2987: \left[ \Big( \frac{iu_2-h_2}{2}+i \frac{\partial h_2}{\partial \theta}\Big)\dot{u}_1+ 
2988: \Big( 1+ i\frac{\partial h_2}{\partial t}\Big)\dot{u}_2   + i\dot{h}_2(\theta +u_1, u_2) \right]  e^{i(\theta + u_1) /2},
2989: $$
2990: where $h_1$,  $h_2$,  and their first  partial derivatives  with respect to $\theta$ and $t$
2991: are understood to be evaluated at 
2992: $(\theta + u_1(\theta) , u_2 (\theta )  )$, and thus are  functions of class $L^2_k$ 
2993: which depend continuously on $(u_1,u_2,h_1,h_2)$. 
2994: In particular, notice 
2995:  that the derivatives of these maps at the origin  are respectively given by 
2996: $$
2997: [{\mathcal F}_{1*{\bf 0}} (\dot{u}_1 , \dot{u}_2 , \dot{h}_1 , \dot{h}_2) ] (\theta ) = 
2998: \Big[ \dot{h}_1(\theta , 0)  + i\dot{u}_1 (\theta) \Big] e^{i\theta} 
2999: $$
3000: and 
3001: $$
3002: [{\mathcal F}_{2*{\bf 0}} (\dot{u}_1 , \dot{u}_2 , \dot{h}_1 , \dot{h}_2) ] (\theta ) = 
3003: \Big[ \dot{u}_2 (\theta)  + i\dot{h}_2(\theta , 0) \Big]  e^{i\theta/2}~.
3004: $$
3005: 
3006: 
3007: Next, we introduce the orthogonal projection  
3008: $$\Pi: L^2 (S^1, \CC) \to L^2{\downarrow}$$ 
3009: to the closed linear subspace 
3010: $$L^2{\downarrow}=
3011: \left\{ \sum_{\ell <0} a_\ell e^{i\ell\theta }~\Big|~a_\ell\in \CC , ~\sum_{\ell <0} |a_\ell |^2< \infty
3012: \right\} \subset  L^2 (S^1, \CC)$$
3013:  of  {\em negative frequency}
3014: functions 
3015: given by 
3016: $$
3017: \Pi(\sum_{\ell=-\infty}^\infty a_\ell e^{i\ell\theta } ) = \sum_{\ell=-\infty}^{-1} a_\ell e^{i\ell\theta } .$$
3018: This a bounded linear operator, and indeed has operator norm $1$ .  
3019: Notice that the kernel of $\Pi$ precisely consists of
3020: those  $L^2$ function on the circle which arise as the  boundary values
3021: of holomorphic functions on the disk. 
3022: Set 
3023: $$
3024: L^2_{k}{\downarrow}= \left\{ \sum_{\ell <0} a_\ell e^{i\ell\theta }
3025: ~\Big|~a_\ell\in \CC , ~\sum_{\ell <0} \ell^{2k}|a_\ell |^2< \infty
3026: \right\}
3027: =  L^2_k (S^1, \CC) \cap L^2{\downarrow}. 
3028: $$
3029: and  notice that 
3030: $$\Pi: L^2_k (S^1, \CC)\to L^2_{k}{\downarrow}$$
3031: is also bounded, and  indeed again has operator norm $1$. 
3032: 
3033: Similarly, let us define 
3034: $$\mbox{\cyr p}: L^2_k (S^1, \CC)\to \CC$$
3035: by 
3036: $$
3037: \mbox{\cyr p} (\sum_{\ell=-\infty}^\infty a_\ell e^{i\ell\theta } ) = 
3038: a_0 .$$
3039: 
3040: 
3041: 
3042: \begin{remark}
3043: The linear map  $\Pi$ is closely related to the Hilbert transform on the 
3044: circle, and can be explicitly be realized \cite{taylor} as the singular integral operator  
3045: $$
3046: [\Pi (u)](\theta)  = u(\theta)  - \frac{e^{-i\theta}}{2\pi}  ~ p.v. \int_0^{2\pi} \frac{u(\phi) d\phi
3047:  }{e^{i(\phi -\theta ) }-1}. 
3048: $$
3049: This can be used \cite{hilltai} to show that $\Pi$ is also bounded with
3050: respect to in $C^{k,\alpha}$ norms. However,  
3051: we have chosen, in the spirit of \cite{bisdisk}, to emphasize
3052: Sobolev norms here, as this has the advantage of keeping
3053: the technical details to a minimum. 
3054: \end{remark}
3055: 
3056: Now, for $k, \ell \geq 1$,  consider the $C^\ell$ map 
3057: $$
3058: L^2_k (S^1) \times \tilde{L}^2_k(S^1)^R \times 
3059: C^{k+\ell}(B^R) \times \tilde{C}^{k+\ell}(B^R)  \stackrel{{\mathcal F}}{\longrightarrow}  
3060: L^2_{k}{\downarrow} \times L^2_{k}{\downarrow}  \times   C^{k+\ell}(B^R) \times 
3061: \tilde{C}^{k+\ell}(B^R)
3062: \times \CC \times   \CC \times  \RR 
3063: $$
3064: of real Banach manifolds
3065: defined by 
3066: $$
3067: {\mathcal F}=( \Pi \circ {\mathcal F}_1) \times ( \Pi \circ {\mathcal F}_2) \times  
3068: \mbox{\cyr L} \times \tilde{\mbox{\cyr L}}
3069: \times (\mbox{\cyr p} \circ {\mathcal F}_1) \times ( \mbox{\cyr p} \circ {\mathcal F}_2)
3070: \times 
3071: \mbox{\cyr sh}  , 
3072: $$
3073: where
3074: $$\mbox{\cyr L} : L^2_k (S^1) \times \tilde{L}^2_k(S^1)^R \times 
3075: C^{k+\ell}(B^R) \times \tilde{C}^{k+\ell}(B^R) \longrightarrow C^{k+\ell}(B^R)$$
3076: and 
3077: $$\tilde{\mbox{\cyr L}} : L^2_k (S^1) \times \tilde{L}^2_k(S^1)^R \times 
3078: C^{k+\ell}(B^R) \times \tilde{C}^{k+\ell}(B^R) \longrightarrow \tilde{C}^{k+\ell}(B^R)$$
3079: are the factor projections, while 
3080: $$\mbox{\cyr sh} : L^2_k (S^1) \times \tilde{L}^2_k(S^1)^R \times 
3081: C^{k+\ell}(B^R) \times \tilde{C}^{k+\ell}(B^R) \longrightarrow \RR$$
3082: is  given by 
3083: $$
3084: \mbox{\cyr sh} (u_1,u_2,h_1,h_2)= 
3085: \frac{1}{2\pi}\int_0^{2\pi} u_1(\theta ) 
3086: ~ d\theta .
3087: $$
3088: Since $\Pi$, $\mbox{\cyr L}$, $\tilde{\mbox{\cyr L}}$, $\mbox{\cyr p}$ and
3089: $\mbox{\cyr sh}$ are all bounded linear operators, 
3090: this map is $C^1$, with  derivative  given by 
3091: $$
3092: {\cal F}_* = (\Pi\circ {\mathcal F}_{1*}) \times (\Pi\circ {\mathcal F}_{2*})
3093: \times \mbox{\cyr L}\times \tilde{\mbox{\cyr L}}\times
3094: (\mbox{\cyr p}\circ {\mathcal F}_{1*}) \times (\mbox{\cyr p}\circ {\mathcal F}_{2*})
3095: \times 
3096:  \mbox{\cyr sh} .$$
3097: In particular, for any
3098: $$
3099: \dot{u}_1= b_0+ \sum_{\ell=1}^\infty b_\ell\cos (\ell\theta ) + c_\ell \sin (\ell \theta ) 
3100: $$
3101: in $L^2_k(S^1)$,
3102: and any 
3103: $$
3104: \dot{u}_2= \sum_{\ell=0}^\infty \
3105: \tilde{b}_\ell\cos \left[(\ell + {\textstyle \frac{1}{2}}) \theta \right] + \tilde{c}_\ell \sin 
3106: \left[(\ell +  {\textstyle \frac{1}{2}})  \theta \right] 
3107: $$
3108: in $\tilde{L}^2_k(S^1)$, 
3109: we see that 
3110:  the derivative of $\mathcal F$ at the origin is explicitly by 
3111: $$ {\cal F}_{*{\bf 0}}\left[
3112: \begin{array}{c}
3113: \dot{u}_1\\ 
3114: \dot{u}_2\\ 
3115: \dot{h}_1\\ 
3116: \dot{h}_2
3117: \end{array}\right]=  
3118: \left[\begin{array}{c}
3119:  \Pi \left(\dot{h}_1 (\theta, 0)e^{i\theta}\right)+  
3120: \sum_{\ell =2}^\infty{ \frac{-c_{\ell}+ib_{\ell}}{2}}e^{-i(\ell -1)\theta} \\   
3121:  \Pi \left(i\dot{h}_2 (\theta, 0)e^{i\theta /2}\right) +  \sum_{\ell =1}^\infty{ \frac{\tilde{b}_{\ell}+
3122: i\tilde{c}_{\ell}}{2}}e^{-i\ell\theta}
3123: \\ \dot{h}_1\\ \dot{h}_2\\   
3124: \mbox{\cyr p} \left(\dot{h}_1 (\theta, 0)e^{i\theta}\right) + 
3125: \frac{-c_{1}+ib_{1}}{2}\\  
3126: \mbox{\cyr p} \left(i\dot{h}_2 (\theta, 0)e^{i\theta /2}\right) +
3127: {\frac{\tilde{b}_0+i\tilde{c}_0}{2}} \\ 
3128: b_0 \end{array}\right] ~.
3129: $$
3130: Since ${\cal F}_{*{\bf 0}}$
3131: manifestly has bounded inverse, the Banach-space inverse function theorem \cite{schwartz}
3132: tells us that there is an open  neighborhood $\mathfrak U$ 
3133: of ${\bf 0}\in  L^2_k (S^1) \times \tilde{L}^2_k(S^1)^R \times 
3134: C^{k+\ell}(B^R) \times \tilde{C}^{k+\ell}(B^R)$ and an open  neighborhood $\mathfrak V$
3135: of ${\bf 0}\in L^2_{k}{\downarrow} \times L^2_{k}{\downarrow}  \times   C^{k+\ell}(B^R) \times \tilde{C}^{k+\ell}(B^R)
3136: \times \CC \times \CC \times \RR$ such that 
3137: $${\mathcal F}|_{\mathfrak U} : {\mathfrak U} {\longrightarrow}
3138:   {\mathfrak V} $$ is a 
3139: diffeomorphism. For any $h_1, h_2$ of sufficiently small $C^{k+\ell}$ norm, we 
3140: therefore  obtain a $5$-parameter family of holomorphic disks
3141: $D\to \CP_2$ with boundaries on the graph of $(h_1, h_2)$ by considering 
3142: the unique disks with boundary values specified by 
3143: $({\mathcal F}|_{\mathfrak U})^{-1}[  {\mathfrak V}\cap (\{(0,0,h_1,h_2)\} \times 
3144: \CC \times \CC \times \RR )
3145: ]$. On the other hand, not all of these disks correspond to   geometrically
3146: distinct unparameterized disks, since any parameterized  disk gives rise to 
3147: a $3$-parameter family 
3148: of other parameterized disks by composition with M\"obius transformations
3149: of the form (\ref{moby}). However, we can easily kill this ``gauge freedom'' 
3150: by   instead 
3151: considering the $2$-parameter family of disks whose boundary values are 
3152:  by the circles 
3153: $$({\mathcal F}|_{\mathfrak U})^{-1} (0,0,h_1,h_2,-w^2, w, 0) 
3154: , ~~w\in \CC , |w| < \varepsilon .$$
3155: The other disks in our original $5$-parameter family can then all be obtained by 
3156: composing the disks in this $2$-parameter family with M\"obius transformations. 
3157: Notice, however, that we have now carefully constructed our disks so that their centers 
3158: are on the complex curve
3159: $${\mathfrak z}_1 + {\mathfrak z}_2^2=0$$
3160: in $\CC^2$, and that our parameter $w$ exactly sweeps out a neighborhood of 
3161: the origin in this curve. However, this curve is just an affine chart on the conic ${\mathcal 
3162: Q}\subset \CP_2$ given by 
3163: $$
3164: z_1^2+z_2^2 + z_3^2 =0 .
3165: $$
3166: Now the subgroup $SO (3) \subset PSL (3, \CC )$ preserves both
3167: ${\mathcal Q}$ and $\RP^2\subset \CP_2$, and acts transitively 
3168: on both ${\mathcal Q}$ and the set of real projective lines $\RP^1\subset \RP^2$. 
3169: Thus, by 
3170: considering only affine charts $({\mathfrak z}_1 , {\mathfrak z}_2)$
3171: related to our original choice by the action of $SO (3)$, we can construct 
3172: a  collection  of families of disks so that their centers  run through 
3173:  a finite open cover of ${\mathcal Q}\approx S^2$, in  a uniform
3174: manner depending on the submanifold $N \subset \CP_2$, thought of as 
3175: the graph of a  section of  the normal bundle of 
3176: $\RP^2\subset \CP_2$  of sufficiently small $C^{k+\ell}$ norm, corresponding to 
3177: $(h_1,h_2)$ in local coordinates. 
3178: Since ${\mathcal F}|_{\mathfrak U}$ is a  diffeomorphism,
3179: we can also arrange that these disks coincide up to M\"obius transformations on overlaps
3180: by at worst restricting to a smaller open set of $N$'s in the $C^{k+\ell}$ 
3181: topology. 
3182:  This yields  the 
3183: following result: 
3184: 
3185: \begin{prop}
3186: \label{smoothdisks}
3187:  If  $N\subset \CP_2$ is the image of any embedding
3188: $\RP^2\hookrightarrow  \CP_2$ which is sufficiently close to  the standard one in the 
3189: $C^{2k-1}$ topology, then $N$ contains a unique family 
3190: of embedded oriented circles $\ell_x \subset N$, $x\in S^2$, each of  which 
3191: bounds an  embedded holomorphic 
3192: disk $D^2\subset \CP_2$, and each of which is 
3193:  $L^2_k$ close (and hence $C^{k-1}$
3194: close) to the image of an  oriented real  projective line $\RP^1\hookrightarrow \RP^2$.
3195: Moreover, if $k \geq 2$, the corresponding family of holomorphic disks   
3196: can be realized  by a fiber-wise holomorphic,  
3197: $C^{k-1}$ map from the unit disk bundle in the ${\mathcal O}(4)$ 
3198: complex line bundle over $S^2= \CP_1$. 
3199: these disks are all embedded, and their interiors foliate $\CP^2 - N$. 
3200: \end{prop}
3201: \begin{proof}
3202:  Locally, our family of disks has
3203: been found by using ${\mathcal F}^{-1}$ to construct
3204:  a $C^{k-1}$ map from an open set  
3205: $W\subset \CC$ to the space of $L^2_k$ maps from the circle to $N$ which bound maps 
3206: of the $2$-disk. But, provided that $k \geq 2$, 
3207:  the inclusion $L^2_k\hookrightarrow C^{k-1}$ is a bounded linear map, and
3208: the maximum principle tells us that we therefore have a 
3209: $C^{k-1}$ map from $W$  into 
3210: the $C^{k-1}$ maps of the disk to $\CP_2$. But any such  map is 
3211: given by a $C^{k-1}$ map $W\times D^2\to \CP_2$. 
3212: Since we have also arranged for the centers of our disks to 
3213: land on the conic $\mathcal Q$, our various local families 
3214: of disks are related by M\"obius transformations which fix the origin,
3215: and so are elements of $U(1)$; moreover, these transformations
3216: are $C^{k-1}$ functions of our parameters, and so determine 
3217: a $C^{k-1}$ disk bundle over ${\mathcal Q}\approx S^2$. 
3218: 
3219: 
3220:  
3221: Now our family of disks is a $C^{k-1}$ map {\cyr f} from this
3222: disk bundle to $\CP_2$, and sends the zero section 
3223: to ${\mathcal Q}$. In our $({\mathfrak z}^1, {\mathfrak z}^2)$
3224:  coordinates, each our disks is  $C^{k-1}$ close
3225: to a disk in a complex line ${\mathfrak z}^2=a +\bar{a}{\mathfrak z}^1$.
3226: By possibly shrinking our neighborhood of  $N$'s, we can thus arrange  that 
3227: each is  embedded, and transverse to ${\mathcal Q}$. 
3228: Similarly, we can arrange for the derivative of {\cyr f}  
3229: to be non-zero everywhere, since  locally the 
3230: map is $C^1$ close to our model example. 
3231: Moreover,
3232: each of our $N$'s can be obtained from $\RP^2$ by applying 
3233: a self-diffeomorphism of $\CP_2$ which is $C^{k-1}$ close to the 
3234: identity, and the push-forward of the 
3235: local functions $|{\mathfrak z}^1|^2$ by these diffeomorphisms
3236: will result in functions which are sub-harmonic on each disk of the 
3237: family, and  the maximum principle therefore  shows each
3238: of the disks will meet $N$ only along its boundary. 
3239: Thus {\cyr f} gives us a proper local diffeomorphism, and hence a covering map, from 
3240: the interior of the disk bundle to $\CP_2 - N$; but $\CP_2 - N$
3241: is simply connected, so {\cyr f} is a diffeomorphism on the interior
3242: of our disk bundle. 
3243: In particular, the zero section of of our disk bundle, which  is 
3244: sent to $\mathcal Q$, has self-intersection ${\mathcal Q}\cdot {\mathcal Q}=2^2=4$,
3245: so our disk bundle has first Chern class $4$, and so must be $C^{k-1}$ 
3246: isomorphic to the unit disk bundle in ${\mathcal O}(4)$. \end{proof}
3247: 
3248: Thus, we  have constructed  a family of  curves $\ell_x \subset N$,
3249: $x\in S^2$, which bound holomorphic disks. We now wish  to  consider the 
3250:  of curves ${\mathfrak C}_y\subset S^2$, 
3251: $y\in N$, obtained by considering the set of all $\ell_x$'s passing through $y$,  
3252: and we would  like to assert that these must be  the geodesics 
3253: of a unique Zoll projective connection $[\nabla ]$ on $M=S^2$.
3254: Our proof of this assertion will hinge on  
3255: 
3256: \begin{lem} \label{rival}
3257:  Let $M$ be a smooth connected 
3258: $2$-manifold,  $\varpi: {\mathcal X}\to M$  a smooth 
3259: $\CP_1$-bundle, and   $\rho : {\mathcal X}\to {\mathcal X}$  an involution,
3260: commuting with the projection $\varpi$, whose  fixed-point  set 
3261: ${\mathcal X}_\rho$ is an $S^1$-bundle over $M$ which disconnects
3262: ${\mathcal X}$ into two closed $2$-disk bundles ${\mathcal X}_\pm$
3263: with common boundary ${\mathcal X}_\rho$. 
3264: Suppose that $\Dye \subset T_\CC {\mathcal X}$ is
3265: a    distribution of complex $2$-planes on ${\mathcal X}$
3266: such that 
3267: \begin{itemize}
3268: \item 
3269: $\rho^* \Dye = \overline{\Dye}$; 
3270: \item   the restriction of $\Dye$ to ${\mathcal X}_+$
3271: is $C^{k}$, $k \geq 1$, 
3272: and  involutive; 
3273: \item $\Dye \cap \ker \varpi_*$  is the $(0,1)$ tangent space of the $\CP_1$ fibers
3274: of $\varpi$;  and 
3275: \item  the restriction of $\Dye$ to a fiber of ${\mathcal X}$  has 
3276: $c_1= -3$ with respect to the complex orientation.
3277: \end{itemize}
3278: Then there is a unique $C^{k-1}$
3279: projective structure $[\nabla ]$ on $M$ such that $\Dye$ is 
3280: obtained from the  
3281: associated involutive distribution $\bf D$ on $\PP T_\CC M$
3282: given by the recipe (\ref{recipe}), pulled back by a uniquely
3283: determined  $C^{k}$ diffeomorphism $\phi: 
3284: {\mathcal X}\to \PP T_\CC M$ which makes the diagrams 
3285: \setlength{\unitlength}{1ex}
\begin{center}\begin{picture}(80,17)(0,3)
\put(10,17){\makebox(0,0){${\mathcal X}$}}
3286: \put(18,19){\makebox(0,0){$\phi$}}
\put(18,5){\makebox(0,0){$M$}}
3287: \put(26,17){\makebox(0,0){$\PP T_\CC M$}}
\put(11,15.5){\vector(2,-3){6}}
3288: \put(25,15.5){\vector(-2,-3){6}}
3289: \put(12,17){\vector(1,0){10}}
3290: \put(36,10){{and}}
\put(52,17){\makebox(0,0){$\mathcal X$}}
3291: \put(68,17){\makebox(0,0){$\PP T_\CC M$}}
\put(52,5){\makebox(0,0){$\mathcal X$}}
3292: \put(68,5){\makebox(0,0){$\PP T_\CC M$}}
\put(50,12){\makebox(0,0){$\rho$}}
3293: \put(70,12){\makebox(0,0){$c$}}
3294: \put(60,6.5){\makebox(0,0){$\phi$}}
3295: \put(60,19){\makebox(0,0){${\phi}$}}
\put(68,15.5){\vector(0,-1){9}}
3296: \put(52,15.5){\vector(0,-1){9}}
3297: \put(53.5,17){\vector(1,0){11}}
3298: \put(53.5,5){\vector(1,0){11}}
\end{picture}\end{center}
3299: commute, where $c: \PP T_\CC M \to \PP T_\CC M$ denotes 
3300: the usual complex conjugation map. 
3301: \end{lem}
3302: \begin{proof}
3303: Let us begin by noticing that, since   $\Dye= \rho^*\overline{\Dye}$ is continuous on the
3304: closed sets ${\mathcal X}_+$ and 
3305: ${\mathcal X}_-$, it is continuous on all of $\mathcal X$. 
3306: Also notice that the  $\phi$ makes the above diagrams commute.
3307: 
3308: 
3309: Now let $L_1$ be the $(0,1)$ tangent space of the fibers. By hypothesis, 
3310: $L_1\subset \Dye$, so that  $L_2= \Dye /L_1$ is a well
3311: defined complex line bundle. Also notice that, since $\Dye \cap \ker \varpi_* = L_1$,
3312:  the fibers of 
3313: $L_2$ are carried injectively into $T_\CC M$ by $\varpi_*$. 
3314: We may therefore define a  continuous map 
3315: $\phi :  {\mathcal X} \to \PP T_\CC M$ by  $z\mapsto  \varpi_* (L_{2}|_z) = 
3316:  \varpi_* (D_z)$. Now let $\zeta$ be a smooth, fiber-wise holomorphic
3317: coordinate on $\mathcal X$, and notice that the corresponding 
3318: vertical vector field $\partial / \partial \overline{\zeta}$
3319: is a smooth section of $D$.  Next, near any point of the interior of ${\mathcal X}_+$,
3320: let $\mathfrak w$ be any other local section of $\Dye$ which is linearly independent
3321: from  $\partial / \partial \overline{\zeta}$, and then notice that the involutivity hypothesis
3322: $[C^1 (\Dye) , C^1 (\Dye) ]\subset
3323: C^0 (\Dye)$
3324: tells us that 
3325: $$\frac{\partial}{\partial \overline{\zeta}} \left( \varpi_* ({\mathfrak w}) \right)
3326: =  \varpi_* ({\mathcal L}_{\frac{\partial}{\partial \overline{\zeta}}}{\mathfrak w})
3327: =  \varpi_* \left(\left[  \frac{\partial}{\partial \overline{\zeta}} ,{\mathfrak w} \right]\right) 
3328: \equiv 0 \bmod  \varpi_* ({\mathfrak w}),$$ 
3329: so that $\phi$ is fiber-wise holomorphic on the interior ${{\mathcal X}}_+$. 
3330: But since $\phi = c\circ \phi \circ \rho$, it then follows that $\phi$ is also 
3331: fiber-wise holomorphic on the interior ${{\mathcal X}}_-$.
3332: But since $\phi$ is also continuous across ${\mathcal X}_\rho = {\mathcal X}_+
3333: \cap {\mathcal X}_-$, this implies that $\phi$ is actually fiber-wise holomorphic on
3334: all of ${\mathcal X}$.  
3335: 
3336: 
3337: 
3338: Now the restriction of $L_2$ to $\varpi^{-1}(x)$
3339: is the  pull-back, via $\phi$, of the tautological 
3340: $\O(-1)$ line bundle over $\PP (\CC\otimes T_{x} M)\cong \CP_1$.
3341: Since $L_1$ is the $(0,1)$ tangent space of $\varpi^{-1}(x)$, and  $\varpi^{-1}(x)
3342: \cong \CP_1$, 
3343:  $c_1(L_1)=-2$ on any fiber of $\varpi$.  On the other hand, 
3344: $c_1(D)=-3$ on $\varpi^{-1}(x)$, by hypothesis. Adjunction therefore
3345: tells us that $c_1(L) = -1$ on any fiber. However,  $c_1({\mathcal O}(-1))=-1$ on 
3346:  $\CP_1$, and we have just observed that 
3347: the $\phi^*c_1( {\mathcal O}(-1)) = c_1(L_2)$. This shows that the fiber-wise 
3348: degree of $\phi$ is $(-1)/(-1)=+1$. But since $\phi$ is also fiber-wise holomorphic,
3349: it follows that $\phi$ maps each fiber of $\mathcal X$   biholomorphically
3350: to the corresponding  fiber of $\PP T_\CC M$. 
3351: This in turn implies that    $\phi$ is $C^k$ on all of ${\mathcal X}$,
3352: since it sends any three pointwise-distinct local $C^k$ sections of ${\mathcal X}_+$
3353: to  three pointwise-distinct local $C^k$ sections of $\PP T_\CC M$, and 
3354: $\phi$  is then algebraically determined by its value along these
3355: sections. 
3356: 
3357: 
3358: 
3359: 
3360: Let us now try to analyze the   distribution of complex $2$-planes
3361: $\phi_* D$ on ${\mathcal Z}= \PP T_\CC M$. To this end, let us begin
3362: by choosing an arbitrary $C^{k-1}$ torsion-free affine connection 
3363: $\nabla_0$ on $M$, and then considering the  corresponding $C^{k-1}$ integrable 
3364: distribution of complex $2$-planes ${\bf D}_0$ on $\mathcal Z$
3365: given by  (\ref{recipe}). By construction, $\phi_* D$ and
3366: ${\bf D}_0$ both intersect the vertical in the $(0,1)$ tangent spaces of 
3367: the fibers. Moreover, letting ${\bf V}^{0,1}$ denote the $(0,1)$ vertical tangent bundle of 
3368: $\PP T_\CC M$,  ${\bf D}_0/ {\bf V}^{0,1} = (\phi_* D )/ {\bf V}^{0,1} = {\mathcal O}(-1)$, where
3369: ${\mathcal O}(-1)$ of course denotes the tautological line bundle. Thus there is 
3370: a unique  continuous  section $\gamma$ 
3371:  of ${\bf V}^{1,0}\otimes {\mathcal O}(1)$ 
3372: such that ${\mathfrak w} \in {\bf D}_0$ 
3373: iff 
3374: ${\mathfrak w} + \gamma (\pi_*{\mathfrak w} ) \in \phi_* \Dye$;
3375: here we  have used the notation ${\bf V}^{1,0}=\overline{{\bf V}^{0,1}}$
3376: and ${\bf V}^{1,0}\otimes {\mathcal O}(1) = {\mathcal H}om ( {\mathcal O}(-1) , {\bf V}^{1,0} )$.
3377: Moreover, the regularity of $\Dye$ guarantees that $\gamma$ is 
3378: $C^{k-1}$ away from the real slice $\PP TM \subset \PP T_\CC M$. 
3379: Now, let ${\mathfrak w}$ be a $C^{ k-1}$ local section of ${\bf D}_0$ 
3380: for which $\pi_* {\mathfrak w}$
3381: is a fiber-wise holomorphic section of ${\mathcal O}(-1)$;
3382: such a section may always be constructed by 
3383: multiplying a generic section by  a  suitable complex-valued function. 
3384: Set $f \partial/\partial \zeta = \gamma (\pi_*{\mathfrak w} )$. 
3385: Then, away from the real slice,  the involutivity 
3386: of $\phi_* \Dye$ and ${\bf D}_0$ then tells us  that 
3387: $$
3388: \left[  \frac{\partial}{\partial \overline{\zeta}} ,{\mathfrak w} \right] \equiv  0 \bmod \frac{\partial}{\partial \overline{\zeta}}
3389: $$
3390: and
3391: $$
3392: \left[  \frac{\partial}{\partial \overline{\zeta}} ,{\mathfrak w} + f  \frac{\partial}{\partial {\zeta}}\right]
3393:  \equiv  0 \bmod \frac{\partial}{\partial \overline{\zeta}} ~, 
3394: $$
3395: so that
3396: $$
3397:  \frac{\partial f }{\partial \overline{\zeta}} \frac{\partial}{\partial {\zeta}}  \equiv  0 \bmod \frac{\partial}{\partial \overline{\zeta}} ,
3398: $$
3399: and hence 
3400: $
3401: \partial f /\partial \overline{\zeta} = 0
3402: $.
3403: This shows that $\gamma$ is fiber-wise holomorphic
3404: away from the real slice. But $\gamma$ is also continuous
3405: across the real slice. It follows that $\gamma$  is fiber-wise holomorphic
3406: on all of $\PP T_\CC M$. 
3407: 
3408: Now any holomorphic section of $(T^{1,0}\CP_1)\otimes {\mathcal O}(1)\cong {\mathcal O}(3)$ 
3409: arises from a unique trace-free element of $\CC^2 \otimes \odot^2 (\CC^2)^*$. 
3410: Thus $\gamma$
3411: is uniquely expressible as a   trace-free symmetric tensor field
3412: $$
3413: \mbox{\cyr  g} \in T_\CC M \otimes \odot^2 T^*_\CC M .
3414: $$
3415: Since $\gamma$ is $C^{k-1}$ away from the real slice, 
3416: it follows that $\mbox{\cyr  g}$ must be $C^{k-1}$. 
3417: Moreover, because $\phi_* \Dye$ and ${\bf D}_0$ are both 
3418: sent to their complex conjugates by $c$, so is $\gamma$, and 
3419: $\mbox{\cyr  g}$ is therefore  real-valued. Setting
3420: $\nabla = \nabla_0 +  \mbox{\cyr  g}$ now gives us  a
3421: $C^{k-1}$ symmetric affine connection on $M$ such that 
3422: $\phi_* \Dye$ coincides with the distribution $\bf D$ defined
3423: by
3424: (\ref{recipe}). Since this last requirement certainly also  determines  $\nabla$ up to
3425: projective equivalence, we are therefore done. 
3426: \end{proof}
3427: 
3428: This allows us to finally show that our constructed families of holomorphic 
3429: disks actually give us Zoll projective structures.
3430: 
3431: \begin{thm}
3432: Let $N$ be any embedding of $\RP^2$ into $\CP_2$ which is 
3433: $C^{2k+5}$ close to the standard one. Let $\{\ell_x~|~ x\in S^2\}$ be the 
3434: constructed family of circles which bound holomorphic disks.
3435: For each $y\in N$, set 
3436: $${\mathfrak C}_y =\{ x\in S^2 ~|~ y\in \ell_x \}.$$
3437: Then there is a unique $C^k$ Zoll projective 
3438: structure $[\nabla ]$ on $S^2$ for which every ${\mathfrak C}_y$
3439: is a geodesic.  
3440: \end{thm}
3441: \begin{proof}
3442: Let ${\mathcal X}_+$ be the unit disk bundle in ${\mathcal O}(4)$, and let 
3443: $\mathcal X$ be its double, obtained by identifying two copies
3444: of ${\mathcal X}_+$
3445: along their boundaries. Let ${\mathcal X}_-$ be the second copy 
3446: of  ${\mathcal X}_+$, and let $\rho : {\mathcal X}\to {\mathcal X}$
3447: be the smooth map which interchanges  ${\mathcal X}_+$ and  ${\mathcal X}_-$.
3448: Notice that one may think of   ${\mathcal X}\to S^2$ as the fourth Hirzebruch surface,
3449: and that, while  ${\mathcal X}$ is itself diffeomorphic to $S^2\times S^2$, the 
3450: `real slice' ${\mathcal X}_\rho \to S^2$ is the circle bundle of Euler class $4$. 
3451: 
3452: Next, we consider the constructed family of holomorphic disks $\mbox{\cyr f} : {\mathcal X}_+
3453: \to \CP_2$ with boundary on $N$. Let $\mbox{\cyr f}_*^{1,0}: T_\CC {\mathcal X}_+
3454: \to \mbox{\cyr f}^*T^{1,0} \CP_2$ be the $(1,0)$ component of its derivative. 
3455: Since  $\det \mbox{\cyr f}_*^{1,0}$ is $C^{k+1}$ close to the corresponding,
3456: non-zero   expression  arising in the  model case of the linear embedding 
3457: $\RP^2\hookrightarrow \CP_2$,  it is also non-zero for every embedding
3458: in an appropriate neighborhood with respect to the topology in question.
3459: Thus we may arrange for $\Dye = \ker \mbox{\cyr f}_*^{1,0}$ to be a $C^{k+1}$ 
3460: distribution of complex $2$-planes on ${\mathcal X}_+$ for each   of the 
3461: embeddings in question. Moreover, $\Dye$ is involutive on the 
3462: interior of ${\mathcal X}_+$, since $\mbox{\cyr f}$ is a diffeomorphism 
3463: there, and sends $\Dye$  to the involutive distribution  $T^{0,1}\CP_2$.
3464: 
3465: Along ${\mathcal X}_\rho = \partial {\mathcal X}_+$, note that $\Dye$ is spanned by 
3466: $\partial/\partial \overline{\zeta}$ and the  distribution of real lines
3467: tangent to the fibers of 
3468: $$\mbox{\cyr f}|_{\partial {\mathcal X}_+}:  {\mathcal X}_\rho\to N.$$
3469: We may therefore extend $\Dye$  
3470: to  ${\mathcal X}_-$ by declaring it  equal to $\rho^*\overline{\Dye}$ on this set. 
3471: The resulting distribution is  $C^0$ close to the one corresponding to 
3472: the model case, and so has $c_1 (\Dye ) =-3$ on every fiber of 
3473: $\mathcal X$. Thus the hypotheses of Lemma \ref{rival}
3474: are all fulfilled, and we therefore obtain  a unique $C^k$ projective structure $[\nabla ]$ on 
3475: $M=S^2$ for which $\Dye$ corresponds to $\bf D$ via $\phi$. But
3476: $\phi$ sends ${\mathcal X}_\rho$ diffeomorphically to $\PP TM$, and 
3477: the fibers of $\mbox{\cyr f}|_{\partial {\mathcal X}_+}$ are thereby sent to 
3478: a foliation $\mathcal F$ of $\PP TM$ by circles which is horizontal 
3479: with respect to $[\nabla ]$, and must coincide with  the foliation by lifted
3480: $[\nabla ]$-geodesics.   The projective structure $[\nabla ]$ is 
3481: therefore Zoll, and so tame by Theorem \ref{bingo}. The space of geodesics
3482: $\tilde{N}$ of $[\nabla]$ is then a compact manifold diffeomorphic to $\RP^2$,
3483: and comes equipped with a tautological submersion to $N$; this 
3484: map is necessarily a covering map, and hence is a diffeomorphism by 
3485: comparison of fundamental groups. In particular, the
3486: ${\mathfrak C}_y$ are precisely the geodesics of the constructed  
3487: projective structure.\end{proof}
3488: 
3489: 
3490: 
3491: 
3492: We now address 
3493: the issue of determining when  a given  projective
3494: structure can be represented by the Levi-Civita connection of 
3495: a Riemannian metric. 
3496: 
3497: Suppose that $g$ is a Zoll metric on $M \approx S^2$. 
3498: Then, in analogy with the construction on page
3499: \pageref{conehead}, we obtain a preferred holomorphic 
3500: curve ${\mathcal C}\subset {\mathcal Z}_+$ of genus zero
3501: and self-intersection $4$ 
3502: by considering $T^{0,1}M$ for the unique complex structure 
3503: compatible with $g$ and the fixed orientation of $M$. 
3504: The image ${\mathcal Q}= \Psi [{\mathcal C}]$ of this
3505: Riemann surface is then an embedded, non-singular rational curve 
3506: of self-intersection $4$ in 
3507: in 
3508: ${\mathcal N}\cong \CP_2$, and so must be a non-singular   conic\footnote{Note
3509:  that one thus does not  need to invoke  
3510:  Yau's deep contribution to Theorem \ref{wao} 
3511:  in this  Zoll metric case, insofar as the 
3512: existence of a rational curve of positive self-intersection
3513:  in   ${\mathcal N}$ forces this compact  complex surface
3514:  to be  rational,   
3515: for strictly classical reasons \cite[Proposition V.4.3]{bpv}.}. 
3516: After a projective linear
3517: transformation, we may thus identify ${\mathcal Q}$ with 
3518: the smooth 
3519: conic  given by 
3520: $$z_1^2+z_2^2+z_3^2=0 .$$
3521: Henceforth, we will impose this choice as a matter of convention. 
3522: 
3523: 
3524: Now observe that 
3525: the Riemann surface $\mathcal C$ is one of the two connected components of 
3526: the locus $g ({\bf v},{\bf v}) =0$ in $\PP T_\CC M$. The complement of
3527: this locus is doubly covered by
3528: $$UT_\CC M = \{ {\bf v} \in T_\CC M ~|~ g( {\bf v},{\bf v}) = 1 \} ~, $$
3529:  which we will think of as a fiber-wise complexification of
3530: the unit tangent bundle of $(M,g)$. However, $UT_\CC M$
3531: may be canonically identified, using $g$, with
3532:  $$UT^*_\CC M = \{ \eta \in T^*_\CC M ~|~ g^{-1}( \eta , \eta ) = 1 \} ~, $$
3533: and we may thus equip $UT_\CC M$ with a complex-valued $2$-form 
3534: $\Upsilon$ obtained by restricting $d\Theta$ to $UT^*_\CC M$, where
3535: $\Theta = \sum_{j=1}^2 y_j dx^j$ is the tautological complex-valued $1$-form 
3536: on $T^*_\CC M$. Moreover, it is not hard  to see that  ${\bf D}= \ker \Upsilon$
3537: on $UT_\CC M$, since, taking geodesic normal coordinates 
3538: around an arbitrary point, we have
3539: $$g = (dx^1)^2 + (dx^2)^2 + O (|x|^2),$$
3540: and hence 
3541: $$\Upsilon = d \left(
3542: \frac{dx^1 + \zeta dx^2}{\sqrt{1+\zeta^2+ O (|x|^2)}}
3543: \right) =
3544: \frac{d\zeta \wedge (\zeta dx^1 - dx^2)}{(1+\zeta^2)^{3/2}}+ O (x^1,x^2)~,$$
3545: and $\ker \Upsilon$ is therefore spanned by 
3546: $\partial /\partial \overline{\zeta}$ and 
3547: $\Xi = \partial /\partial {x^1} + \zeta ~\partial /\partial {x^2}+ O (x^1,x^2)$.  
3548:  Away from the real slice, $\Upsilon$ is therefore a closed 
3549: form of type $(2,0)$ with respect to $\bf D$, and hence is holomorphic;  and,
3550: by the last calculation,  
3551:  $\Upsilon \otimes \Upsilon$ descends to 
3552: $\PP T_\CC M - {\mathcal C}$ so as to have a pole of order $3$
3553: along $\mathcal C$. On the other hand, the restriction of $\Upsilon$ to the unit circle
3554: bundle of $M$ is real-valued, and descends to the space of oriented geodesics by  symplectic reduction \cite{beszoll,weinstein}, and
3555: $\Upsilon$ thus gives rise to a continuous $2$-form on the double cover of 
3556: $\CP_2 -{\mathcal Q}$ which is holomorphic on the the complement of $N$, and hence 
3557: holomorphic everywhere. Thus $\Upsilon \otimes \Upsilon$ is a well-defined
3558: meromorphic section of $K^2$ on $\CP_2$ with  polar locus $3{\mathcal Q}$,
3559: and it follows that 
3560: $$
3561: \Upsilon = \lambda \frac{z_1 ~dz_2\wedge dz_3 + z_2 ~dz_3 \wedge dz_1 + z_3~ dz_1\wedge d_2}{
3562: (z_1^2 + z_2^2 + z_3^2)^{3/2}
3563: }
3564: $$
3565: for some constant $\lambda \in \CC$. However, we also know that 
3566: the restriction of $\Upsilon$ to the double cover $\tilde{N}\approx S^2$ of 
3567: $N$ is real. Since $\tilde{N}$ is homotopic to the double cover $S^2$
3568: of the standard $\RP^2\subset \CP_2$, and since $\Upsilon$ is 
3569: closed, 
3570: we have
3571: $$
3572: \int_{\tilde{N}} \Upsilon = \int_{S^2} \lambda \frac{z_1 ~dz_2\wedge 
3573: dz_3 + z_2 ~dz_3 \wedge dz_1 + z_3~ dz_1\wedge d_2}{
3574: (z_1^2 + z_2^2 + z_3^2)^{3/2}} = 4\pi \lambda ~,
3575: $$
3576: and it follows that $\lambda$ must be real. 
3577: Since $\Upsilon$ is real along $\tilde{N}$, we thus conclude  that 
3578: the Riemannian condition implies that 
3579:  $N$  is Lagrangian with respect to the 
3580: sign-ambiguous symplectic structure
3581: \begin{equation}
3582: \label{sympform}
3583: \omega = \pm \Im m \left(
3584: \frac{z_1 ~dz_2\wedge dz_3 + z_2 ~dz_3 \wedge dz_1 + z_3~ dz_1\wedge d_2}{
3585: (z_1^2 + z_2^2 + z_3^2)^{3/2}
3586: }
3587: \right) 
3588: \end{equation}
3589: on  $\CP_2 -{\mathcal Q}$. 
3590: 
3591: Conversely, suppose the  surface $N$ corresponding to a given projective 
3592: structure $[\nabla ]$ on $M=S^2$ does avoid the conic $\mathcal Q$ 
3593: and is Lagrangian with respect to the sign-ambiguous symplectic structure
3594: $\omega$ on $\CP_2 - {\mathcal Q}$. Assume, moreover, that $\mathcal Q$ 
3595: generates $H_2 ( \CP_2 - N, \ZZ )\cong \ZZ$, as certainly happens whenever $N$
3596: is sufficiently close to the standard $\RP^2 \subset \CP_2$. Then, since the 
3597: family of holomorphic disks associated with $[\nabla ]$ generate
3598: $H_2 ( \CP_2 , N ; \ZZ )$, their Poincar\'e duals generate $H^2 (  \CP_2 - N, \ZZ )$,
3599: and each holomorphic disk therefore has intersection number $+1$ with 
3600: ${\mathcal Q}$, and hence geometrically intersects $\mathcal Q$
3601: transversely in a unique point. This gives us a diffeomorphism 
3602: ${\mathcal Q} \to M$, and hence fixes a conformal structure on 
3603: $M\approx S^2$. Since we also have an orientation, the structure group 
3604: of the circle bundle ${\mathbb S}TM = (TM - 0_M) /\RR^+$ is thereby reduced to $SO (2)$;
3605: let $\partial /\partial \theta$ be the vertical vector field on ${\mathbb S}TM$
3606: which generates the corresponding $SO (2)$ action. 
3607: Now make a particular choice of $\Upsilon$ by choosing the real constant $\lambda \neq 0$, 
3608: and pull it back 
3609:  to the double cover ${\mathbb S}TM$ of $\PP TM$;
3610:  this is a real-valued $2$ form on ${\mathbb S}TM$, and 
3611: we then have a  map ${\mathbb S}TM \hookrightarrow  T^*M$
3612: given by  $\Upsilon ( \xi, \cdot) $, the image of which is the set of 
3613: co-vectors for a unique Riemannian metric $g$ on $M$ in the given
3614: conformal class. For this metric, the foliation $\mathcal F$ is given by symplectic reduction,
3615: and the geodesics of our projective structure are then exactly those of $g$.   
3616: Thus: 
3617: 
3618: 
3619: \begin{thm}
3620: Let $N \hookrightarrow \CP_2$ be a totally real  embedding of $\RP^2$ 
3621: which corresponds to a projective structure $[\nabla ]$ on $M\approx S^2$. 
3622: Then there is a Riemannian metric $g$ on $M$ whose Levi-Civita
3623: connection $\triangledown$ belongs to the projective class 
3624: $[\nabla ]$ iff, after a $PSL(3,\CC )$ transformation
3625: of $\CP_2$, the surface $N$ avoids the conic $\mathcal Q$, and 
3626: is Lagrangian with respect to the signed symplectic structure 
3627: $\omega$ on $\CP_2 - {\mathcal Q}$. Moreover, such a 
3628: Lagrangian embedding ccompletely determines the metric $g$ up to 
3629: an overall multiplicative constant. \end{thm}
3630: 
3631: \pagebreak
3632: 
3633: \section{Concluding Remarks}
3634: 
3635: 
3636: A number of important technical issues remain to be 
3637: resoved in connection to our treatment of Zoll structures
3638: on $S^2$. For example, while we have shown that one can 
3639: associate a totally real  embedding of $\RP^2$ in 
3640: $\CP_2$ with each Zoll projective connection on $S^2$, 
3641: and that such embedded surfaces can conversely  be 
3642: used to determine a projective connection on $S^2$, one loses
3643: a ridiculous number of derivatives in following the 
3644: story full circle to ones starting point. Ideally, one might hope that 
3645:  $C^{k,\alpha}$ projective structures on $S^2$ should  exactly correspond to 
3646: $C^{k+1,\alpha}$ surfaces  $N\subset \CP_2$. However, we are at present quite
3647: far from being able to make such an assertion in either direction. 
3648: 
3649: What is worse, we do not at present know  that our family of disks either exists or is 
3650: unique when $N$ is very from the the standard $\RP^2$. Nonetheless, optimism
3651: might well be appropriate in the present instance. Indeed, let us throw
3652: caution to the wind and hazard the following: 
3653: 
3654: \begin{conjecture}
3655: The moduli space of Zoll metrics on $S^2$ is connected. Moreover, 
3656: once we mark our Zoll structures by choosing an orthonormal frame
3657: at some base-point, 
3658: the  
3659: moduli space of marked Zoll structures is in natural $1$-$1$ correspondence with the set of
3660: totally real Lagrangian embeddings of $\RP^2 \hookrightarrow (\CP_2 -{\mathcal C}, \omega )$
3661: which are homotopic to the standard embedding.
3662: \end{conjecture}
3663: 
3664: In fact, it does not seem not hard to show that the set of $N\subset \CP_2$ 
3665: carrying suitable families of embedded holomorphic disks
3666: is open, but there are a numerous 
3667:  technical difficulties  involved in trying to show that it is closed, since
3668: sequences of embedded disks may  have singular limits,
3669: and one tends, in the limit,
3670:  to lose regularity  of the  dependence of families on
3671: parameters . Moreover,  
3672:  one would need to know
3673: that the relevant set of   Lagrangian $\RP^2$'s
3674: in $(\CP_2 -{\mathcal C}, \omega )$ is actually connected
3675: for this program to ultimately succeed. Fortunately, however, 
3676: the latter is similar to problems already solved by 
3677: Eliashberg \cite{elih,elipol} and his co-workers, so there is ample reason to 
3678: hope for  such a  program to be viable. 
3679: 
3680: One might also want to hazard an analogous conjecture about Zoll projective 
3681: structures. However, this would seem to be a considerably more difficult
3682: problem, as there is as yet no good mechanism for trying to show that 
3683: two weakly unknotted embeddings of $\RP^2$ in $\CP_2$ are
3684: actually isotopic. On the other hand, Gromov's $h$-principle \cite{gromh,elih} at least 
3685: provides a rather complete reduction of  questions concerning isotopy
3686: through totally real submanifolds to questions of isotopy in the usual, 
3687: elementary sense. 
3688: 
3689: 
3690: It seems improbable that the methods we have developed here
3691: will shed  much light on higher-dimensional Zoll manifolds, at least in the 
3692: near term. However, 
3693: our techniques certainly have obvious extensions which could be brought to bear on   Zoll-like 
3694:  Lorentzian $3$-manifolds \cite{mogul},  special classes of split-signature 
3695: Einstein manifolds  \cite{massplit} and certain problems in  Yang-Mills fields 
3696: \cite{maswood}.
3697:  We 
3698: look forward to watching the further development of the present circle of 
3699: ideas in connection with these problems. 
3700:  
3701: 
3702: 
3703: \pagebreak 
3704: 
3705: \appendix
3706:   
3707: 
3708: 
3709: \section{Appendix: The axisymmetric cases}
3710: The main result of this appendix is a formula for the general
3711: axisymmetric Zoll projective structure close to that for the round
3712: sphere.  We first give the formulae for the connection, and show that
3713: it gives rise to a Zoll projective structure.  We go on to show how
3714: these examples are calculated from the twistor correspondence.
3715: Although this latter step is not, strictly speaking, required in the
3716: logical structure, this was how these examples were obtained and it is
3717: difficult to see how the formulae would be obtained otherwise, or
3718: indeed aspects of the proof of the Zoll property.  Furthermore, it
3719: provides a family of detailed worked examples of the construction
3720: given in the body of the paper.
3721: 
3722: \subsection{Axisymmetric Zoll examples}
3723: In order to introduce the formulae, we first recall Zoll's original
3724: family of axisymmetric metrics expressed here in spherical polar coordinates  
3725: $$
3726: g=(F -1)^2d\phi ^2 +\sin^2\phi ~
3727: d\theta ^2\, .
3728: $$
3729: This metric is
3730: the same as that given by (\ref{ansatz}),  after the
3731: coordinate transformation $z=\cos\phi$ and the substitution
3732: $F(\phi)=-f(\cos\phi)$.  
3733: 
3734: We will express the general Zoll projective sructure in terms of the
3735: difference between a compatible affine connection and the metric
3736: connection of the above metric.  
3737: Consider the orthonormal frame 
3738: $$
3739: ({\bf e}_1,{\bf e}_2)=(\frac{1}{F
3740: -1}\p \phi ,\frac{1}{\sin\phi }\p \theta )
3741: $$
3742: and dual co-frame
3743: $(\theta^1,\theta^2)=(({F -1})d\phi, \sin\phi d\theta)$.
3744: In this frame, it is straightforward to calculate that the connection
3745: 1-form is 
3746: $$
3747: \omega=\frac{\cot\phi}{F -1}\theta^2\, .
3748: $$ 
3749: The associated Levi-Civita connection, $\nabla^g$, gives the most
3750: general axisymmetric Zoll projective structure that is compatible with
3751: a metric (at least close to the round metric).
3752: 
3753: In general, a compatible torsion-free affine connection for a
3754: projective structure can be given by a connection $\nabla$ such that,
3755: with $$
3756: \gamma_{ij}^k=\left\langle\theta^k,(\nabla_i-\nabla^g_i){\bf e}_j\right\rangle \, ,
3757: $$ $\gamma_{ij}^k$ is symmetric on the $ij$ indices (so that $\nabla$
3758: is torsion-free) and trace free; this last condition corresponds to
3759: fixing the connection in the projective equivalence class by requiring
3760: that it preserve the metric volume form.  
3761: 
3762: The general axisymmetric Zoll projective structure close to the round
3763: sphere is obtained with the choice 
3764: $$
3765: \gamma_{11}^i=0\, , \quad
3766: \gamma_{22}^1=-\frac{h^2\cot\phi}{F-1}
3767: \, , \quad
3768: \gamma_{21}^1= \frac1{3(F-1)} \left(\frac{\partial h}{\partial \phi} -\frac{2h}{\sin\phi\cos\phi}
3769: \right)
3770: $$
3771: where all the other components of $\gamma_{ij}^k$ are determined by
3772: the trace and symmetry conditions and $h=h(\phi)$ is a smooth function
3773: of $\phi$ vanishing in some small neighborhood of $0$ and $\pi$ and
3774: odd under $\phi\rightarrow \pi-\phi$.
3775: 
3776: 
3777: 
3778: This information can be encapsulated in the geodesic spray.  
3779: This is the vector field on the projective tangent bundle $PTS^2$ that at
3780: $(v,x)\in PT_xS^2$ is the horizontal lift of the vector
3781: $v$ at $x$.  We parametrize the fibre of $PTS^2$ by $\zeta$
3782: corresponding to the vector ${\bf e}_1+\zeta {\bf e}_2$.  Then the geodesic spray
3783: from the projective structure above is given by
3784: \begin{equation}\label{zollspray}
3785: \Xi
3786: =\p \phi  + \left(\frac{F-1}{\sin\phi}\right)\zeta \p \theta 
3787: -\zeta\left((1+\zeta^2(1+h^2))\cot\phi   -
3788: \zeta \left(\frac{\partial h}{\partial \phi } -
3789: \frac{2h}{\sin\phi \cos\phi }\right) \right) \p \zeta 
3790: \end{equation}
3791: on $PTS^2$ defines a Zoll projective structure if the smooth functions
3792: $F(\phi)$ and $h=h(\phi)$ are respectively odd and even under $\phi
3793: \rightarrow\pi-\phi$.  For regularity at $\phi=\pi/2$, we further
3794: require that $h$ should vanish (and hence to second order) at $\phi
3795: =\pi/2$.  For regularity at $\phi=0,\pi$, we assume that $F$ and $h$
3796: vanish in some small neighborhood of these values. This is actually stronger
3797: than necessary,  but makes the proof of the Zoll property more
3798: straightforward; the minimal requirement would be to just stipulate  that they be smooth
3799: functions of $\cos\phi$ that vanish at $\phi=0$ and $\pi$.  
3800: 
3801: 
3802: The metric case occurs when $h=0$, and in this case
3803: there is the preferred overall scaling factor that gives the arc-length
3804: parameterisation; this arises on dividing by $(F
3805: -1)\sqrt{(1+\zeta^2)}$.  To see that the above gives a multiple of the
3806: geodesic spray in this case, coordinatize the tangent bundle by
3807: $(\mu_1,\mu_2)\rightarrow \mu_1 {\bf e}_1+\mu_2 {\bf e}_2$.  Then the horizontal
3808: lift of ${\bf e}_1$ is just ${\bf e}_1$ since $\omega({\bf e}_1)=0$ and the horizontal
3809: lift of ${\bf e}_2$ is ${\bf e}_2-\omega({\bf e}_2) (\mu_1\p {\mu_2} -
3810: \mu_2\p {\mu_1})$.  Thus, using the affine coordinate
3811: $\zeta=\mu_2/\mu_1$ on the projective tangent bundle, the geodesic
3812: spray will be
3813: $$
3814: {\bf e}_1+\zeta \left({\bf e}_2 -(1+\zeta^2)\frac{\cot\theta}{F
3815: -1}\p \zeta \right)
3816: $$
3817: and this can be seen to be proportional to the formula given above
3818: when $h=0$ as required.  If we wish to normalize the horizontal part
3819: to have unit length, then we must divide by $\sqrt{(1+\zeta^2)}$ and
3820: this will give the overall factor required to give proper length
3821: parameterisation.
3822: 
3823: 
3824: We first give a direct proof of the Zoll property, and then in the
3825: subsequent sections we show how the formula arises from the twistor
3826: construction.  (The direct proof of the Zoll property below in fact
3827: will use equations arising in the twistor derivation below, but it is
3828: easily checked that these follow directly from the form of the
3829: geodesic spray above.  It is difficult, however, to see how they might
3830: have been anticipated without the twistor construction.)
3831: 
3832: \begin{thm}
3833: Equation(\ref{zollspray}) defines a Zoll projective structure for all
3834: smooth odd functions $F$ and even functions $h$ with $h(\pi/2)=0$
3835: and both $h$ and $F$ vanishing in some neighborhood of
3836: $\phi=0$.  
3837: \end{thm}
3838: 
3839: 
3840: The proof is divided into two parts.  We first 
3841: analyze the flow of the projection of the geodesic spray under
3842: $q:PTS^2\rightarrow \RP^1\times[0,\pi]$,
3843: $q(\zeta,\phi,\theta)=(\zeta,\phi)$, to the space
3844: of orbits of $\partial/\partial\theta$ in $PTS^2$.  We show first that the orbits
3845: of the projected flow are circles, and secondly that the lifts of these
3846: to orbits of the full geodesic spray are also circles in $PTS^2$.
3847: 
3848: 
3849: \smallskip
3850: 
3851: \noindent
3852: 1)
3853: We first study the integral curves of
3854: $q_*\Xi$ for $(\zeta,\phi )\in \RP^1\times[0,\pi/2]$.
3855: Introduce the angular coordinate $\psi\in
3856: [0,\pi)$ on $\RP^1$ by $\zeta=\tan\psi$ so that $\psi$ is a
3857: smooth coordinate near $\zeta=\infty$.  Then, the flow
3858: becomes
3859: \begin{eqnarray}
3860: \dot \phi& =&\sin\phi \cos\psi\, ,\nonumber \\ 
3861: \dot\psi &=&
3862: -\sin\psi\left((1+h^2\sin^2\psi)\cos\phi -\cos\psi \sin\psi \left(\frac{\partial h}{ \partial \phi}
3863: \sin\phi  -\frac{2h}{\cos\phi} \right)\right)\label{flow}
3864: \end{eqnarray}
3865: where $\dot\phi=d\phi/d t$ for the time parameter $t$ along the flow
3866: defined by
3867: $$
3868: \frac d{dt}=\sin\phi\cos\psi q_*\Xi\, .
3869: $$
3870: These additional factors yield a smooth flow by inspection noting in
3871: particular that our requirement that $h(\pi/2)=0$ implies that
3872: $h/\cos\phi$ is smooth.  Note also that this flow is invariant under
3873: the reflection in $\phi=\pi/2$: $(\psi,\phi,t)\rightarrow
3874: (\psi,\pi-\phi,-t)$. 
3875: 
3876: [The perceptive reader might have noticed that the direction of the flow
3877: changes sign across the identification of $\psi=\pi$ with $\psi=0$.
3878: Although the flow defines a smooth distribution in the projective tangent
3879: bundle away from the fixed points, to obtain a flow with continuous
3880: direction, we would need to work on the double cover obtained by
3881: factoring the tangent bundle by the positive scalings.  This will not
3882: be a problem in the following as $\psi=0$ or $\pi$ is a flow line.]
3883: 
3884: 
3885: \begin{lem}
3886: The flow of $q_*\Xi$ has fixed points at $(\psi,\phi)=(0,0), (0,\pi)$
3887: and $(\pi/2,\pi/2)$.  The integral
3888: curves of $q_*\Xi$ are smoothly embedded curves in $(\zeta,\phi )\in
3889: \RP^1\times[0,\pi]$ on which, for $\phi\in[0,\pi/2]$ (resp.\
3890: $\phi\in[\pi/2,\pi]$) the coordinate $\phi$ decreases (resp.\
3891: increases) from $\pi/2$ to a unique minimum (resp.\ maximum) value and
3892: then increases (resp.\ decreases) again to $\pi/2$.  The extrema occur
3893: when $\psi=\pi/2$.
3894: \end{lem} 
3895: 
3896: 
3897: The fixed points are where both the right hand sides vanish, so that,
3898: from $\dot\phi=0$ we obtain either $\phi=0,\pi$ or $\psi=\pi/2$.
3899: At $\phi=0,\pi$, $h=0$ and so we find $\dot\psi=\mp\sin\psi$, i.e., a
3900: fixed point at $\psi=0(=\zeta)$.  At $\psi=\pi/2$, we find that
3901: $\dot\psi=0$ iff $\cos\phi=0$, i.e., $\phi=\pi/2$.
3902: 
3903: It is clear from the first of equations (\ref{flow}) that for
3904: $\phi\in(0,\pi)$, $\dot\phi$ only vanishes when $\psi=\pi/2$.  The
3905: second derivative at $\psi=\pi/2$ can be calculated to give
3906: $$
3907: \frac{\partial^2\phi}{\partial\psi^2}= (1+h^2)\cot\phi
3908: $$
3909: and it can be seen that this second derivative $\partial^2\phi/\partial\psi^2$ is positive
3910: for $\phi \in(0,\pi/2)$ and so this
3911: must be a minimum.  Similarly on $\phi\in(\pi/2,\pi)$, $\phi$ can only
3912: be a maximum at a stationary point.  Thus, on an integral curve in
3913: $\phi\in(0,\pi/2)$,
3914: $\phi $ will descend to a unique minimum value, at which
3915: $\zeta=\infty$ and then increase again.  
3916: $\Box$
3917: 
3918: \smallskip
3919: 
3920: The key issue now is as to whether we can make these integral curves  
3921: join up into a circle.  Firstly  note that $\psi=0$ and $\phi=0,\pi$
3922: are all flow lines, and these are the only flow lines limiting onto
3923: the fixed points $(0,0)$ and $(0,\pi)$ as we have assumed that $h=0$
3924: in a neighborhood of $\phi=0$ and of $\pi$, and this means that the flow
3925: lines in those neighborhoods are precisely those of the flat case,
3926: and these are precisely the level curves of $\sin\phi\sin\psi$.
3927:  
3928: 
3929: Let us suppose that a curve starts at some 
3930: value of $\psi\in (0,\pi/2)$.  Then $\phi$ will descend to a minimum
3931: and either (a) increase up to $\pi/2$ again, or (b) the minimum will be
3932: $\phi=0$.  In case (a), the reflection of the orbit under the
3933: involution $(\psi,\phi,t)\rightarrow (\psi,\pi-\phi,-t)$ will be an
3934: orbit in $\phi\in (\pi/2,\pi)$ and this will join up to make a
3935: circular orbit.  Case (b) will be the case $\psi=0$ since the orbit
3936: must intersect $\phi=0$ at $\psi=0$, since the complement of that
3937: point in $\phi=0$ is a regular orbit on its own, but the only orbit in
3938: a neighborhood of $\phi=0$ that intersects this fixed point is
3939: $\psi=0$ (or $\phi=0$).  
3940: 
3941: Thus, all the orbits of the flow are circles, except the above
3942: mentioned fixed points and special orbits that limit onto the fixed
3943: points; this gives the flow diagram
3944: \ref{flowdiagram}. 
3945: \begin{figure}[ht]
3946: \caption{The flow diagram for the projected flow}\label{flowdiagram}
3947: \begin{center}
3948: \includegraphics[width=10cm,height=10cm,angle=0]{flow.eps}
3949: \end{center}
3950: \end{figure}
3951: 
3952: \medskip
3953: 
3954: 2)  We now wish to show that these orbits in the $(\psi,\phi)$ plane
3955:     only lift to give closed $S^1$ orbits in the full projective tangent
3956:     bundle of the sphere.  In the above coordinates, the equation for
3957:     $\theta$ will become
3958: \begin{equation}\label{zolllift}
3959: \dot\theta=(F -1)\sin\psi\, .
3960: \end{equation}
3961: In order for the geodesics to be circles, we need to prove that the
3962: integral of the right hand side around an integral curve of $q_*\Xi$
3963: is 0 modulo $2\pi$ for each integral curve.  The first and second
3964: terms in the right hand side of equation (\ref{zolllift}) are
3965: respectively odd and even under $\theta\rightarrow \pi-\theta$.  Since
3966: the integral curves of $q_*\Xi$ are even, the first part will
3967: automatically 
3968: integrate to zero.  We need to show, then, that the second part
3969: will in fact integrate to $0$ modulo $2\pi$ on all integral curves.
3970: 
3971: To integrate $\dot\theta=-\sin\psi$, from equations
3972: (\ref{liftspray}) and (\ref{beta}) in the lifting part of the twistor
3973: construction, we note that, with $1/a=(h-i)|\cos\phi|$
3974: $$
3975: \omega= \frac12 \arg {\frac{(1-\zeta/\bar a)}{1-\zeta/a}}
3976: $$
3977: satisfies
3978: $$
3979: \dot\omega=-\sin\psi\, .
3980: $$
3981: (We leave it to the assiduous reader to show that equations
3982: (\ref{liftspray}) and (\ref{beta}) follow independently of the twistor
3983: construction.)  Thus, $\theta=\omega$ is the solution to the even part
3984: of the $\theta$ flow.  However, $\omega$ is the argument of a single
3985: valued complex function on the $(\psi,\phi)$--plane, and so, when we
3986: do a complete circuit around an integral curve of $q_*\Xi$ returning
3987: to our original point, the argument must return to zero modulo
3988: $2\pi$. $\Box$
3989: 
3990: 
3991: \subsection{The twistor construction in the axisymmetric case}
3992: In \S\ref{homcase} we study the structure of the action of
3993: axisymmetry on the twistor space and the correspondence for the round
3994: metric.  In \S\ref{defms} we give the axisymmetric deformations of
3995: the real slice.  The subsequent subsection \S\ref{constrdisc} is
3996: devoted to constructing the holomorphic disks, and then finally in
3997: \S\ref{constrproj} the associated projective struture is constructed.
3998: 
3999: \subsubsection{The round sphere}\label{homcase}
4000: We consider the action of the standard rotation on $\RR^3$, its
4001: complexified action on $\C^3$ and induced action on $\CP^2$.  With
4002: coordinates $(z,\tilde z, z_0)$, the $S^1$ action is generated by the real
4003: part of the holomorphic vector field
4004: $$
4005: \p \theta =i(z\p z-\tilde z\p {\tilde z})\, ,
4006: $$ 
4007: where $\RR^3$ is taken to be $\tilde z=\bar z$ and $z_0=\bar z_0$.  If
4008: we remove the the fixed points $(1,0,0)$, $(0,1,0)$ and $(0,0,1)$, the
4009: generic orbits form the pencil of conics $(1-w)z_0^2=w z \tilde z$
4010: that are tangent to the line $z=0$ at $(0,1,0)$ and also to the line
4011: $\tilde z =0$ at $(1,0,0)$.  The degenerate orbits consist of the
4012: double line $z_0=0$ at $w=0$ and the pair of lines $z=0$ and $\tilde
4013: z=0$ at $w=1$.  They determine a fibration of
4014: $\CP^2-\{(1,0,0),(0,1,0)\}$ over $\CP^1$ with affine coordinate $w$,
4015: and, away from the exceptional fibres at $w=0,1$, we can coordinatise
4016: $\CP^2$ with $(w,\xi)=(z_0^2/(z_0^2+z\tilde z), z/z_0)$.
4017: % (which generate rationally all the usual affine coordinates in the sense
4018: % that, for example, $\tilde \xi:=\tilde z/z_0= (-1+1/w) /\xi$ etc.). 
4019: In these coordinates $\p \theta =i\xi\p \xi$.
4020: 
4021: The real slice, $\RP^2$, is given by $w\in
4022: [0,1]$ and $|\xi|^2=-1 + 1/w $. Note that the orbit $z_0=0$ intersects
4023: $\RP^2$ in a real line, whereas the orbit $\{z=0\} \cup \{\tilde z=0\}$
4024: intersects $\RP^2$ in a single point.  All the other real orbits are
4025: contractible circles in $\RP^2$.
4026: 
4027: 
4028: Introduce spherical polar coordinates $(\phi ,\theta )$ on $S^2$ so that
4029: the symmetry is $\p \theta $.  We can coordinatize the fibres of the
4030: projective tangent bundle by $\zeta$ so that $\zeta$ corresponds
4031: to the vector $\p \phi  + \frac{\zeta}{ \sin \phi } \p \theta $.
4032: (These coordinates will then only break down at the fixed points.)
4033: The lines in $\CP^2$ corresponding to points of $S^2$ are $2
4034: z_0=\tan\phi  (e^{i\theta }z+e^{-i\theta } \tilde z)$.  In terms of
4035: $\zeta$, and the coordinates $(w,\xi)$ on $\CP^2$, the holomorphic
4036: disks  are the images of the upper-half plane in $\zeta$ under
4037: \begin{equation}\label{rounddiscs}
4038: w=\frac{\zeta^2\sin^2\phi 
4039: }{1+\zeta^2}
4040: \, , \qquad \xi=e^{i\theta }
4041: \frac{\zeta \cos\phi  +i}{\zeta\sin\phi }
4042: \, ,
4043: \end{equation}
4044: and when $\zeta$ is real the image lies in $\RP^2$.\footnote{A
4045: global and invariant formulation can be obtained in index notation by
4046: letting $z_i$, $i=1,\cdots 3$ be homogeneous coordinates on $\CP^2$,
4047: and $x^i$ coordinates on $\RR^3$, then the open disk in $\CP^2$
4048: corresponding to $x^i$ on $S^2$ is given by the condition that $iz_i\bar
4049: z_j\varepsilon^{ijk}$ be a positive multiple of $x^k$.}
4050: 
4051: It is worth noting for later use that, on these disks, $\zeta=
4052: \sqrt{w/(\sin^2\phi -w)}$ defines the square root in the
4053: upper-half plane. 
4054: 
4055: 
4056: 
4057: In these coordinates, the geodesic spray takes the form:
4058: $$
4059: \Xi=\p \phi  +\frac{\zeta}{\sin\phi }\p \theta 
4060: -\cot\phi 
4061: (1+\zeta^2)\zeta\p \zeta\, .
4062: $$
4063: 
4064: It should also be noted that the conserved quantity associated to the
4065: axial symmetry $\p \theta $ and metric $g=d\phi ^2+\sin^2\phi  d\theta ^2$
4066: is
4067: $$
4068: \frac{g(\p \theta ,\p \phi  +\frac{\zeta}{\sin\phi }\p \theta )}
4069: {\sqrt{g(\p \phi  +\frac{\zeta}{\sin\phi }\p \theta ),\p \phi 
4070: +\frac{\zeta}{\sin\phi }\p \theta )}} =\sqrt w\, .
4071: $$
4072: This formula can also be derived intrinsically on $\CP^2$; namely, $\sqrt w$
4073: is the Hamiltonian for $\p \phi $ using the symplectic form
4074: associated to the conic ${\mathcal Q}$  defined in
4075: equation 
4076: (\ref{sympform}).

4077: \subsubsection{Deformation of the real slice}\label{defms}
4078: We will represent a circle invariant deformed embedding of $\RP^2$
4079: into $\CP^2$ as the set given by $$ w=\gamma(\phi )\, , \mbox{ and }
4080: |\xi|^2= e^{g(\phi )} \left|\frac{1-\gamma(\phi )}{\gamma(\phi )}\right| $$ for
4081: $\phi\in [0,\pi/2]$.  Here $g$ is a smooth real function with compact
4082: support in $(0,\pi/2)$ and $\gamma:[0,\pi/2]\rightarrow\C$ is a smooth
4083: embedded curve from $w=0$ to $w=1$ such that $\gamma(\phi )=\sin^2\phi
4084: $ on the complement of some compact subset of $(0,\pi/2]$. 
4085: 
4086: In the homogeneous case, $\gamma(\phi)=\sin^2\phi$, and $g=0$.  The
4087: compact support of the deviation from the homogeneous case will
4088: guarantee smoothness of this deformation near the degenerate fibre
4089: $z_0=0$.  In particular, the embedding of $\RP^2$ into $\CP^2$ near
4090: the fixed line $z_0=0$, is the same as the canonical embedding, and so
4091: the holomorphic disks  near those at $z_0=0$ will be those above in
4092: equation (\ref{rounddiscs}) and so we will not need to concern
4093: ourselves with singular behaviour there.
4094: 
4095: 
4096: 
4097: 
4098: These assumptions amount to the assumption that our $S^1$--invariant
4099: Zoll projective structure on $S^2$ will have two fixed points
4100: corresponding to $\phi =0,\pi$ in a neighborhood of each of which
4101: the projective structure will be that of the round sphere, and exactly 
4102: one of the $S^1$ orbits will be a geodesic with $\phi =\pi/2$.  
4103: 
4104: In the metric case we will have that $\gamma(\phi )=\sin^2\phi $
4105: since the square of the conserved quantity is determined by the
4106: geometry of the action on $\CP^2$ relative to its fixed symplectic
4107: structure.  It will necessarily be equal to $w$, and will be real on
4108: the real slice.  The nontrivial information in this case is contained
4109: only in the function $g(\phi )$.
4110: 
4111: For later convenience, we extend $\gamma$ and $g$ to $\phi \in
4112: [0,\pi]$ by $\gamma(\phi )=\gamma(\pi-\phi )$ and $g(\phi )=g(\pi-\phi
4113: )$. The data of the location of the deformation of $\RP^2$ could be
4114: represented more economically by expressing the curve $\gamma$ as a
4115: graph of the imaginary part over the real interval $[0,\pi/2]$.
4116: However the formulation above will allow us to make a convenient
4117: choice of the coordinate $\phi$ later.
4118: 
4119: 
4120: \subsubsection{Construction of the holomorphic disks}\label{constrdisc}
4121: The problem of finding the deformed disks  with boundary on the
4122: deformed real slice decomposes into two parts: firstly that of
4123: finding the projection of the disk to the $w$--Riemann sphere with
4124: boundary on the projection of the real slice, the curve $\gamma$, and
4125: secondly, the problem of lifting the disk to $\CP^2$.
4126: 
4127: \medskip
4128: 
4129: \noindent
4130: 1) The projected disks  must have their boundary on some subinterval of
4131: the curve $\gamma$.  This subinterval must include the end at
4132: $\phi=0$: this end corresponds to the line $z_0=0$ and each boundary
4133: of a disk must be homologous to this line, but because these are all
4134: generators of the homlogy of $\RP^2$, they must intersect each other
4135: at least once.
4136: 
4137: Thus, the first task is to find, for each $\phi \in [0,\pi/2]$, a map
4138: $\zeta \rightarrow w(\zeta,\phi )$ from the upper half plane into the
4139: $w$-Riemann sphere such that the boundary of the disk is mapped to the
4140: image of the interval $[0,\phi ]$ under $\gamma$.
4141: 
4142: To analyze this, first consider the conformal map $$ w\rightarrow
4143: v(w,\phi )=\sqrt{\frac{w}{\gamma(\phi )-w}}\, , $$ where we fix the
4144: branch of the square root by requiring that, near $w=0$,
4145: $v\sqrt{\gamma(\phi )}$ lies in the upper-half-plane (there is no
4146: obstruction to choosing $\sqrt{\gamma(\phi )}$ as $\phi $ varies so
4147: that it is positive for small $\phi $).  In the $v$-Riemann sphere,
4148: the image of $\gamma([0,\phi ])$ is a continuously differentiable
4149: embedded circle tangent to $\sqrt{\gamma(\phi )}\times$ the real axis
4150: at the origin and passing through the point $v=\infty$.  It will be
4151: smooth except possibly at $0$ and $\infty$.  Thus the branch defined
4152: above is well defined and determines a region $V_\phi $ in the
4153: $v$-plane as the image of the complement of $\gamma([0,\phi ])$.
4154: 
4155: By the Riemann mapping theorem there will exist a conformal map from
4156: the upper-half-plane in $\zeta$ to $V_\phi $ and hence to the
4157: complement of $\gamma([0,\phi ])$ in the $w$-Riemann sphere.  It will
4158: be smooth with non-vanishing derivative up to and including the
4159: boundary on the $v$-Riemann sphere except possibly at $0$ and $\infty$
4160: where it is nevertheless guaranteed to be continuous \cite[p. 340]{tayapp}.
4161:   It is worth emphasizing that while Proposition \ref{smoothdisks}
guarantees that the disks will be smoothly embedded in $\CP^2$, but
4162: they will be tangent to the fibres of the projection along the orbits
4163: of the complexified axisymmetry at $w=0$ and $\gamma(\phi)$.  Hence,
4164: the projection of the disks to the $w$-Riemann sphere will be smooth
4165: up to $\gamma([0,\phi ])$ except at the points $0$ and $\gamma(\phi )$
4166: which will be ramification points of order 2.  Using a Mobius
4167: transformation of the upper-half plane to itself, this map
4168: $w(\zeta,\phi )$ can be chosen so that $$ w(\zeta,\phi )=\zeta^2
4169: \sin^2\phi + O(\zeta^3)\, , $$ at $\zeta=0$ and $w(\zeta,\phi
4170: )=\gamma(\phi ) - k(\phi )\gamma'(\phi ) \zeta^{-2} + O(\zeta^{-3})$
4171: at $\zeta=\infty$ for some real $k(\phi )>0$.
4172: 
4173: For later use we define the function $s(\zeta,\phi )$
4174: for $\zeta\in\RR$, $\phi \in [0,\pi]$ by the condition that
4175: $$
4176: \gamma(s(\zeta,\phi )) = w(\zeta,\phi )\, .
4177: $$ 
4178: In the following we extend both $w(\zeta,\phi )$ and $s(\zeta,\phi
4179: )$ to $\phi \in[0,\pi]$ so that they are even functions under $\phi
4180: \rightarrow \pi-\phi $.
4181: 
4182: 
4183: \medskip
4184: 
4185: \noindent
4186: 2) We now wish to find the lift of these conformal mappings to disks
4187: in $\CP^2$ with boundary on the deformed real slice.  To do this we
4188: need to obtain $\xi(\zeta,\phi ,\theta )$ holomorphic on the
4189: upper-half-plane in $\zeta$ such that, for $\zeta\in\RR$, 
4190: $$
4191: |\xi(\zeta,\phi ,\theta )
4192: |^2 = e^{g(s(\zeta,\phi)
4193: )}\left|\frac{1-\gamma(s(\zeta,\phi ))} {\gamma(s(\zeta,\phi) )}\right|\, .  
4194: $$
4195: By symmetry we must have $\xi(\zeta,\phi ,\theta )=e^{i\theta
4196: }\xi(\zeta,\phi ,0)$.
4197: 
4198: The orbits of the complexified axisymmetry corresponding to $w\neq
4199: 0,1$ are regular orbits.  Thus for $w(\zeta,\phi)\neq 0,1$, the lift
4200: $\xi(\zeta,\phi ,\theta )$ cannot meet $\xi=0$ or $\infty$ since
4201: $\xi=0$ is part of the orbit $w=1$ and $\xi=\infty$ is the orbit
4202: $w=0$.  However, as $w\rightarrow 0$ we must have, by the above
4203: condition on the real slice, $|\xi|^2\rightarrow
4204: |(1-w)/w|\rightarrow\infty$.  Furthermore, if $\phi =\pi/2$, $w=1$ is
4205: a real point on the boundary of the conformal mapping and must
4206: therefore lift to the real point $\xi=\tilde\xi=0$.  Conversely, at
4207: $w=1$, but $\phi \neq \pi/2$, the point $w=1$ is not a real point on
4208: the disk and so we cannot have both $\xi=0$ and $\tilde\xi=0$.  Hence
4209: either we will have $\xi=0$ and $\tilde \xi\neq 0$, or $\xi\neq 0$ and
4210: $\tilde \xi= 0$.  We can therefore assume that, by continuity from the
4211: round sphere case, $\xi\neq 0$ for $\phi \in [0,\pi/2)$, and
4212: $\tilde\xi\neq 0$ for $\phi \in (\pi/2,\pi]$.
4213: 
4214: By taking logs, the problem of lifting the conformal maps to disks in
4215: $\CP^2$, can be reduced to an abelian problem.  However, we cannot
4216: proceed completely naively as we will still have $\xi\rightarrow
4217: \infty$ as $w\rightarrow 0$, although we can guarantee that either
4218: $xi$ or $\tilde\xi$ will be non-vanishing. We work first on $\phi \in
4219: (0,\pi/2)$ so that $\xi\neq 0$, and divide that problem into a part
4220: that is regular on taking logs, and one that can be handled
4221: explicitly.  Set $$
4222: \xi(\zeta,\phi ,\theta )=e^{i\theta  +
4223: G(\zeta,\phi )}\Gamma(\zeta,\phi )
4224: $$
4225: then we wish to find $G(\zeta,\phi )$ that is
4226: holomorphic for $\Im m\zeta >0$ such that for $\zeta$ real
4227: $$
4228: \Re eG(\zeta ,\phi )=g(s(\zeta,\phi ))
4229: $$
4230: and similarly we wish to find $\Gamma(\zeta,\phi )$ holomorphic on the
4231: upper half plane  in $\zeta$, such that for $\zeta$ real
4232: $$
4233: |\Gamma(\zeta,\phi )|^2= 
4234: \left|\frac{1-\gamma(s(\zeta,\phi )}{\gamma(s(\zeta,\phi ))}\right|\, .
4235: $$
4236: The first problem is solved in a standard way by a contour integral
4237: along the real axis
4238: $$
4239: G(\zeta,\phi )=\frac{1}{2\pi i}
4240:  \oint \frac{\Re eG(\mu,\phi )}{\mu -\zeta}
4241: d \mu - \frac{1}{2\pi i}
4242:  P.V. \int \frac{\Re eG(\mu,\phi )}{\mu }d \mu 
4243: $$
4244: where the purpose of the last term is to remove the ambiguity
4245: associated with the addition of a constant (in $\zeta$ but perhaps
4246: with $\phi $--dependence) to the
4247: imaginary part of $G$.  This choice ensures $\Im mG(0,\phi )=0$.
4248: 
4249: The problem for $\Gamma$ cannot be solved so simply in the above way.
4250: First we define the complex function $a(\phi )$ in the upper half
4251: plane by the condition $w(a,\phi )=1$, i.e., the image in the
4252: $\zeta$ plane of $w=1$.  Then the function
4253: $$
4254: \Gamma(\zeta,\phi )=i\sqrt{\frac{(1-\zeta/\bar a)}{(1-\zeta
4255: /a)}\frac{(1-w)}{w}} 
4256: $$
4257: makes sense for $\zeta$ in the upper half plane  since the function whose root is
4258: taken does not vanish on the upper half plane .  We choose the branch for the square
4259: root that tends towards $i/\zeta\sin\phi $ as $\phi $ and
4260: $\zeta$ tend to zero.  Then $\Gamma$ as defined is non-vanishing,
4261: holomorphic in the upper half plane  and has the required modulus when
4262: $\zeta\in\RR$ as then $|(1-\zeta/\bar a)/(1-\zeta/a)|=1$.
4263: 
4264: 
4265: For $\phi \in [\pi/2,\pi]$ we work with $\tilde\xi$ as that will be
4266: non-zero on this interval.  However, 
4267: $$
4268: |\tilde \xi(\zeta,\phi ,\theta )|^2 =\frac{|1-w|^2}{|w^2\xi^2|}=
4269:  e^{-g(s(\zeta,\phi )}\frac{|1-\gamma(s,\zeta,\phi )|}
4270: {|\gamma(s,\zeta,\phi )|}\, . 
4271: $$
4272: and so the solution will be
4273: $$
4274: \tilde\xi=e^{-i\theta -G(\zeta,\pi-\phi )}\Gamma(\zeta,\pi-\phi )\, ,
4275: $$
4276: where the $\Gamma$ and $G$ are the functions obtained above.
4277: 
4278: \subsubsection{Construction of the projective structure}\label{constrproj}
4279: To reconstruct the corresponding projective connection on $S^2$, we
4280: wish to construct the vector field determining the geodesic spray on
4281: the correspondence space, $PTS^2$.  We use coordinates $(\phi ,\theta )$
4282: on $S^2$, and $\zeta\in\RR$ on the fibres of $PTS^2$.  
4283: We construct the geodesic spray $\Xi$ in two steps:
4284: 
4285: \medskip
4286: 
4287: \noindent
4288: 1) Under the projection $q:(\zeta, \phi ,\theta )\rightarrow
4289: (\zeta, \phi )$, $\Xi$ projects to $q_*
4290: \Xi=\p \phi -p(\zeta,\phi)\p \zeta$ for some $p(\zeta,\phi)$.  The
4291: function $w$ is constant along the geodesic spray so that
4292: $q_*\Xi w=0$ which gives $p=\partial_ \phi w/\partial_ \zeta w$.
4293: 
4294: When $\zeta\in\RR$, $w=\gamma(s(\zeta,\phi )$, so 
4295: $$
4296: p(\phi ,\zeta)=\frac{\gamma' \partial s/\partial \phi  }{\gamma' \partial s/\partial \zeta }=
4297: \frac{\partial s/\partial \phi  }{ \partial s /\partial\zeta }
4298: $$
4299: is real.  Thus  $p$ can be extended meromorphically
4300: over the $\zeta$ Riemann sphere by defining it in the lower-half
4301: plane to be 
4302: the complex conjugate of the pullback under $\zeta\rightarrow
4303: \bar\zeta$.  The fact that it is real for $\zeta\in\RR$ ensures
4304: continuity and hence holomorphy there.  It does, however, have simple
4305: poles at $\zeta=0,\infty$ as $\p \zeta w$ has simple zeroes
4306: there. However, the chosen form at $\zeta=0$ implies that in fact, $p$
4307: vanishes at $\zeta=0$.  Thus, since $p\p \zeta$ is globally holomorphic
4308: except a simple
4309: pole at $\zeta=\infty$ (as a vector field on the Riemann sphere), zero at $\zeta=0$ and real for
4310: $\zeta$ real, we can write 
4311: $$
4312: p=\zeta(\Gamma_2\zeta^2+\Gamma_1\zeta +\cot\phi )
4313: $$
4314: where $\Gamma_1$ and $\Gamma_2$ are real functions of $\phi $ and the
4315: $\cot\phi $ follows from the expansion at $\zeta=0$.
4316: 
4317: Note here that since $w$ and $s$ are even functions under
4318: $\phi \rightarrow \pi-\phi $, $\Gamma_2$, $\Gamma_1$ and
4319: $\cot\phi $ are odd as they involve the $\phi $ derivatives of $s$.
4320: 
4321: We will need the fact later that $\Gamma_1$ and $\Gamma_2$ can be
4322: expressed in terms of $a(\phi )$ and its first derivative by using
4323: the condition
4324: \begin{equation}\label{aeq}
4325: q_*\Xi (\zeta -a)|_{\zeta=a}=0
4326: \end{equation}
4327: which follows from the fact that $\zeta=a$ corresponds to $w=1$
4328: which is a holomorphic curve in $\CP^2$.  This yields the equation
4329: $$
4330: \p \phi  a+ a(\Gamma_2 a^2 +\Gamma_1 a +\cot\phi )=0
4331: $$
4332: and this together with its complex conjugate yields
4333: \begin{equation}\label{connection}
4334: \Gamma_2=\frac{\sin\phi }{\bar a -a}\p \phi \left(\frac{a-\bar
4335: a}{|a|^2\sin\phi }\right) \, , 
4336: \quad \Gamma_1=\frac{\sin\phi }{\bar a-a}\left(\bar
4337: a\p \phi \left(\frac {1}{a\sin\phi }\right) - c.c.\right)\, .
4338: \end{equation}
4339: The number of free functions here is two: either the pair $\Gamma_1$
4340: and $\Gamma_2$ or, equivalently, the real and imaginary parts of $a$.
4341: This is to be compared to the one free function we have in the data of
4342: the curve $\gamma(\phi )$ in the reduced twistor space and the second
4343: free function we have in choosing the coordinate $\phi $, which, up to
4344: now, has been arbitrary (at least away from $\phi =0,\pi/2$).  We will
4345: fix this coordinate freedom subsequently.
4346: 
4347: \medskip
4348: 
4349: \noindent
4350: 2) The next step is to lift $q_*\Xi$ to the vector field $\Xi$ 
4351:    on the full correspondence space $PTS^2$ that annihilates also $\xi$
4352:    or equivalently $\tilde \xi$.   We will have
4353: $$
4354: \Xi= q_*\Xi - \frac{(q_*\Xi \xi)}{\partial_ \theta  \xi}\p \theta  
4355: =q_*\Xi - \frac{(q_*\Xi \xi)}{i\xi}\p \theta  
4356: =q_*\Xi +i(q_*\Xi \log\xi)\p \theta  
4357: \, .
4358: $$
4359: In order to proceed further, note that the coefficient of $\p \theta $ is
4360: $iq_*\Xi\log\xi$, and this is
4361:    (a) holomorphic over upper-half-plane in $\zeta$, and 
4362: (b) is real for $\zeta\in\RR$
4363: since the imaginary part of
4364: $$
4365: i q_*\Xi \log \xi|_{\zeta=\bar \zeta }= i\left(\p \phi  -
4366: \frac{\partial s/\partial \phi }{\partial s/\partial\zeta 
4367: }\frac{\partial}{\partial\zeta}\right)\log \xi
4368: $$
4369: is just $q_*\Xi \log |\xi|$ but $\log|\xi|=\Re e\log \Gamma + \Re
4370: G$ is a function of $\zeta$ and $\phi$ only through $s$, and such
4371: functions of $s$ alone are annihilated by
4372: $q_*\Xi$ by construction.  Thus, the imaginary part of the right
4373: hand side of the above equation vanishes for $\zeta\in\RR$.  Hence,
4374: we can extend it meromorphically over the $\zeta$--Riemann sphere by
4375: setting it to be the complex conjugate of the pullback under
4376: $\zeta\rightarrow \bar\zeta$ for $\Im\zeta<0$ and noting that
4377: reality at $\zeta\in\RR$ implies continuity and hence holomorphy
4378: across the real axis.
4379: 
4380: The function $iq_*\Xi \log\xi$ divides into two parts:
4381: $$
4382: iq_*\Xi \log\xi= iq_*\Xi G(\zeta,\phi )+ iq_*\Xi
4383: \log\Gamma \, ,
4384: $$
4385: and since $w$ is constant along $q_*\Xi$, the second part reduces
4386: to
4387: $$
4388: iq_*\Xi \log\Gamma = \frac{i}{2}q_*\Xi  \log \frac
4389: {1-\zeta/\bar a}{1-\zeta/a} \, .
4390: $$
4391: They are both holomorphic on the full $\zeta$ sphere, except with
4392: poles at $\zeta=\infty$ since $q_*\Xi$ has one there.  However,
4393: they will also have a simple zero at $\zeta=0$ since the imaginary
4394: parts of $G$ and the above expression for $iq_*\Xi\log \Gamma$
4395: vanish there by construction.  (The possible apparent poles in
4396: $iq_*\Xi\log\Gamma$ are removable as a consequence of equation
4397: \ref{aeq}.) Therefore 
4398: \begin{equation}\label{liftspray}
4399: iq_*\Xi G=\frac{F(\phi)}{\sin\phi} \zeta \, , \quad\mbox{ and }\quad
4400: iq_*\Xi\log\Gamma=\beta(\phi )\zeta
4401: \end{equation}
4402: for some real functions
4403: $F$ and $\beta$ and the
4404: geodesic spray is 
4405: $$
4406: \Xi=\p \phi  +\left(\frac{F}{\sin\phi} +\beta\right) \zeta\p \theta  -
4407: \zeta(\Gamma_2\zeta^2+\Gamma_1\zeta +\cot\phi )\p \zeta\, .
4408: $$
4409: Using the above and equations (\ref{aeq}) and (\ref{connection}) we
4410: calculate directly that
4411: $$
4412: \beta=-\Gamma_2 \Im ma\, .
4413: $$
4414: 
4415: When $\phi \in[\pi/2,\pi]$ we should note first that $G$ and $\Gamma$
4416: are even functions under $\phi \rightarrow\pi-\phi $.  Hence,
4417: $F$ and $\beta$ are, as defined, odd functions.  However, there
4418: is a further sign change on using $\tilde\xi$ instead of $\xi$ for
4419: $\beta$ which yields an {\em even} contribution for $\beta$ and odd
4420: for $F$ and $p$, i.e., for $\phi \in[\pi/2,\pi]$ 
4421: $$
4422: \Xi=\p \phi  +\left(-\frac{F(\pi-\phi )}{\sin\phi}
4423: + \beta(\pi-\phi )\right) \zeta\p \theta  
4424: +p(\pi-\phi , \zeta) \p \zeta\, .
4425: $$
4426: 
4427: 
4428: \medskip
4429: 
4430: 
4431: We now fix the choice of the coordinate $\phi $ which up to now has
4432: been arbitrary except near $\phi =0$
4433: and $\pi/2$.  We do this by imposing 
4434: $$
4435: \Im m\frac 1a=  - |\cos\phi |\, 
4436: $$
4437: (note that $a$ must always be in the upper half plane , and must be even
4438: under $\phi \rightarrow \pi-\phi $).  This gives
4439: \begin{equation}\label{beta}
4440: \beta=-1/\sin\phi 
4441: \end{equation}
4442: Introduce the function $h(\phi)$ by 
4443: $$
4444: \Re e\frac 1a=h|\cos\phi |
4445: $$ and this leads to the formulae
4446: $$
4447: \Gamma_1=-\p \phi  h
4448: +\frac{2h}{\sin\phi \cos\phi } \, , \quad
4449: \Gamma_2= \cot\phi  \left(1+h^2\right)\, .
4450: $$
4451: This leads to our final formula for the geodesic spray 
4452: \begin{equation}\label{zollspray2}
4453: \Xi
4454: =\p \phi  + \frac{F-1}{\sin\phi}\zeta \p \theta 
4455: -\left((1+ \zeta^2+\zeta^2h^2)\cot\phi   - 
4456: \zeta \left(\frac{\partial h}{\partial \phi}   -
4457: \frac{2h}{\sin\phi \cos\phi } \right)\right)\zeta \p \zeta 
4458: \end{equation}
4459: where $F$ must be
4460: odd under $\phi \rightarrow\pi-\phi $ and $h$ must be even.  For
4461: regularity, $h$
4462: should vanish to second order at $\phi =\pi/2$.  From the assumption
4463: that the twistor data was zero in some small neighborhood of the fixed line
4464: $z_0=0$, we also deduce that the functions $h$ and $F$ should
4465: vanish in some small neighborhood of $\phi =0, \pi$.
4466: This is the formula that leads to the expressions given at the beginning of
4467: this appendix.
4468: 
4469: 
4470: \vfill
4471: 
4472: \noindent 
4473: {\sc 
4474: Department of Mathematics, SUNY, Stony Brook, NY 11794-3651 USA\\
4475: The Mathematical Institute, 24-29 St Giles,
4476: Oxford OX1 3LB,  England}
4477: 
4478: 
4479: \bigskip
4480: 
4481: \bigskip
4482: 
4483: \noindent 
4484: {\bf Acknowledgments.}
4485: The first author would like to thank Denny Hill, 
4486: Dusa McDuff, and Dennis Sullivan for helpful conversations, as well as  
4487: Bob Gompf and Yasha Eliashberg for some helpful e-mail. 
4488: The second author  would like to thank Mike Eastwood and Rafe Mazzeo for useful
4489: discussions, and MSRI for its hospitality during the early stages of the writing of  this paper.
4490: 
4491: \pagebreak
4492: \begin{thebibliography}{10}
4493: 
4494: \bibitem{AHS}
4495: {\sc M.~F. Atiyah, N.~J. Hitchin, and I.~M. Singer}, {\em Self-duality in
4496:   four-dimensional {R}iemannian geometry}, Proc. Roy. Soc. London Ser. A, 362
4497:   (1978), pp.~425--461.
4498: 
4499: \bibitem{aubin}
4500: {\sc T.~Aubin}, {\em Equations du type {M}onge-{A}mp\`{e}re sur les
4501:   vari\'{e}t\'{e}s {K\"a}hl{\'e}riennes compactes}, C. R. Acad. Sci. Paris,
4502:   283A (1976), pp.~119--121.
4503: 
4504: \bibitem{bpv}
4505: {\sc W.~Barth, C.~Peters, and A.~V. de~Ven}, {\em Compact Complex Surfaces},
4506:   Springer-Verlag, 1984.
4507: 
4508: \bibitem{beszoll}
4509: {\sc A.~L. Besse}, {\em Manifolds {A}ll of {W}hose {G}eodesics {A}re {C}losed},
4510:   Springer-Verlag, Berlin, 1978.
4511: \newblock With appendices by D. B. A. Epstein, J.-P. Bourguignon, L.
4512:   B\'erard-Bergery, M. Berger and J. L. Kazdan.
4513: 
4514: \bibitem{bisdisk}
4515: {\sc E.~Bishop}, {\em Differentiable manifolds in complex {E}uclidean space},
4516:   Duke Math. J., 32 (1965), pp.~1--21.
4517: 
4518: \bibitem{diffm3}
4519: {\sc F.~Bonahon}, {\em Diff\'eotopies des espaces lenticulaires}, Topology, 22
4520:   (1983), pp.~305--314.
4521: 
4522: \bibitem{eastgrah}
4523: {\sc M.~Eastwood and C.~R. Graham}, {\em The involutive structure on the
4524:   blow-up of {${\bf R}^n$} in {${\bf C}^n$}}, Comm. Anal. Geom., 7 (1999),
4525:   pp.~609--622.
4526: 
4527: \bibitem{elih}
4528: {\sc Y.~Eliashberg and N.~Mishachev}, {\em Introduction to the
4529:   {$h$}-principle}, vol.~48 of Graduate Studies in Mathematics, American
4530:   Mathematical Society, Providence, RI, 2002.
4531: 
4532: \bibitem{elipol}
4533: {\sc Y.~Eliashberg and L.~Polterovich}, {\em Unknottedness of {L}agrangian
4534:   surfaces in symplectic {$4$}-manifolds}, Internat. Math. Res. Notices,
4535:   (1993), pp.~295--301.
4536: 
4537: \bibitem{epstein}
4538: {\sc D.~B.~A. Epstein}, {\em Periodic flows on three-manifolds}, Ann. of Math.
4539:   (2), 95 (1972), pp.~66--82.
4540: 
4541: \bibitem{frequin}
4542: {\sc M.~H. Freedman and F.~Quinn}, {\em Topology of 4-manifolds}, Princeton
4543:   University Press, Princeton, NJ, 1990.
4544: 
4545: \bibitem{funk}
4546: {\sc P.~Funk}, {\em {\"U}ber {F}l\"achern mit lauter geschlossenen
4547:   geod\"atischen {L}inien}, Math. Ann., 74 (1913), pp.~278--300.
4548: 
4549: \bibitem{grezoll}
4550: {\sc L.~W. Green}, {\em Auf {W}iedersehensfl\"achen}, Ann. of Math. (2), 78
4551:   (1963), pp.~289--299.
4552: 
4553: \bibitem{gromh}
4554: {\sc M.~Gromov}, {\em Partial differential relations}, vol.~9 of Ergebnisse der
4555:   Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related
4556:   Areas (3)], Springer-Verlag, Berlin, 1986.
4557: 
4558: \bibitem{guillzoll}
4559: {\sc V.~Guillemin}, {\em The {R}adon transform on {Z}oll surfaces}, Advances in
4560:   Math., 22 (1976), pp.~85--119.
4561: 
4562: \bibitem{mogul}
4563: \leavevmode\vrule height 2pt depth -1.6pt width 23pt, {\em Cosmology in {$(2 +
4564:   1)$}-dimensions, cyclic models, and deformations of {$M\sb {2,1}$}}, vol.~121
4565:   of Annals of Mathematics Studies, Princeton University Press, Princeton, NJ,
4566:   1989.
4567: 
4568: \bibitem{hilltai}
4569: {\sc C.~D. Hill and G.~Taiani}, {\em Families of analytic discs in {${\bf
4570:   {C}}^{n}$} with boundaries on a prescribed {C}{R} submanifold}, Ann. Scuola
4571:   Norm. Sup. Pisa Cl. Sci. (4), 5 (1978), pp.~327--380.
4572: 
4573: \bibitem{hiltay}
4574: {\sc C.~D. Hill and M.~E. Taylor}, {\em Integrability of rough almost complex
4575:   structures}, J. Geom. Anal.
4576: \newblock to appear.
4577: 
4578: \bibitem{hitproj}
4579: {\sc N.~J. Hitchin}, {\em Complex manifolds and {E}instein's equations}, in
4580:   Twistor {G}eometry and {N}onlinear {S}ystems (Primorsko, 1980), Springer,
4581:   1982, pp.~73--99.
4582: 
4583: \bibitem{lebthes}
4584: {\sc C.~R. LeBrun}, {\em Spaces of {C}omplex {G}eodesics and {R}elated
4585:   {S}tructures}.
4586: \newblock D. Phil. thesis, Oxford University, 1980.
4587: 
4588: \bibitem{malgrange}
4589: {\sc B.~Malgrange}, {\em Sur l'int\'egrabilit\'e des structures
4590:   presque-complexes}, in Symposia Mathematica, Vol. II (INDAM, Rome, 1968),
4591:   Academic Press, London, 1969, pp.~289--296.
4592: 
4593: \bibitem{massplit}
4594: {\sc L.~J. Mason}, {\em Global solutions of the self-duality equations in split
4595:   signature}, in Further {A}dvances in {T}wistor {T}heory. {V}ol. {I}{I}, L.~J.
4596:   Mason, L.~P. Hughston, and P.~Z. Kobak, eds., Longman Scientific \&
4597:   Technical, Harlow, 1995.
4598: 
4599: \bibitem{maswood}
4600: {\sc L.~J. Mason and N.~M.~J. Woodhouse}, {\em Integrability, {S}elf-{D}uality,
4601:   and {T}wistor {T}heory}, The Clarendon Press Oxford University Press, New
4602:   York, 1996.
4603: \newblock Oxford Science Publications.
4604: 
4605: \bibitem{morrey}
4606: {\sc C.~B. Morrey, Jr.}, {\em Multiple integrals in the calculus of
4607:   variations}, Springer-Verlag New York, Inc., New York, 1966.
4608: 
4609: \bibitem{newnir}
4610: {\sc A.~Newlander and L.~Nirenberg}, {\em Complex analytic coordinates in
4611:   almost complex manifolds}, Ann. of Math. (2), 65 (1957), pp.~391--404.
4612: 
4613: \bibitem{obiwon}
4614: {\sc N.~R. O'Brian and J.~H. Rawnsley}, {\em Twistor spaces}, Ann. Global Anal.
4615:   Geom., 3 (1985), pp.~29--58.
4616: 
4617: \bibitem{schouten}
4618: {\sc J.~A. Schouten}, {\em Ricci-{C}alculus. {A}n introduction to tensor
4619:   analysis and its geometrical applications}, Springer-Verlag, Berlin, 1954.
4620: \newblock 2d. ed.
4621: 
4622: \bibitem{schwartz}
4623: {\sc J.~T. Schwartz}, {\em Nonlinear functional analysis}, Gordon and Breach
4624:   Science Publishers, New York, 1969.
4625: \newblock Notes by H. Fattorini, R. Nirenberg and H. Porta, with an additional
4626:   chapter by Hermann Karcher, Notes on Mathematics and its Applications.
4627: 
4628: \bibitem{taylor}
4629: {\sc M.~Taylor}, {\em Pseudo differential operators}, Springer-Verlag, Berlin,
4630:   1974.
4631: \newblock Lecture Notes in Mathematics, Vol. 416.
4632: 
4633: \bibitem{tayapp}
4634: {\sc M.~E. Taylor}, {\em Partial differential equations. {I}}, vol.~115 of
4635:   Applied Mathematical Sciences, Springer-Verlag, New York, 1996.
4636: \newblock Basic theory.
4637: 
4638: \bibitem{weinstein}
4639: {\sc A.~Weinstein}, {\em On the volume of manifolds all of whose geodesics are
4640:   closed}, J. Differential Geometry, 9 (1974), pp.~513--517.
4641: 
4642: \bibitem{yau}
4643: {\sc S.-T. Yau}, {\em {C}alabi's conjecture and some new results in algebraic
4644:   geometry}, Proc. Nat. Acad. USA, 74 (1977), pp.~1789--1799.
4645: 
4646: \bibitem{zoll}
4647: {\sc O.~Zoll}, {\em {\"U}ber {F}l\"achen mit {S}charen geschlossener
4648:   geod\"atischer {L}inien}, Math. Ann., 57 (1903), pp.~108--133.
4649: 
4650: \end{thebibliography}
4651: 
4652: 
4653: \end{document}
4654: