1:
2: \documentstyle[12pt,psfig] {article}
3: %\input{psfig}
4: \textwidth 5.8in
5: \oddsidemargin 0.6in
6: \topmargin -.4in
7: \topskip 0pt
8: \headheight 12pt
9: \footskip 18pt
10: \footheight 12pt
11: \textheight 8.75in
12: \baselineskip 24pt
13:
14: \newtheorem{theo}{Theorem}
15: \newtheorem{lem}{Lemma}
16: \newtheorem{slm}{SubLemma}
17: \newtheorem{defi}{Definition}
18: \newtheorem{conj}{Conjecture}
19: \newtheorem{prop}{Proposition}
20: \newtheorem{prob}{Open problem}
21: \newtheorem{ex}{Example}
22: \newtheorem{q}{Question}
23: \newtheorem{cor}{Corollary}
24: \newtheorem{rem}{Remark}
25:
26: \begin {document}
27: \bibliographystyle{plain}
28: \title{ Weak Limits of Riemannian Metrics in Surfaces with integral Curvature Bound}
29:
30: \author{ Xiuxiong Chen\\
31: Department of Mathematics \\
32: Stanford University \\
33: Palo Alto, CA 94305 }
34: \date{ Sep. 19, 1996}
35: \maketitle
36: \section{ Introduction}
37: {\bf 1.1 Introduction to the problem.} We study the limit of a sequence
38: of Riemannian metrics on a surface under some suitable conditions.
39: Let $\Omega $ be any open domain and let
40: ${\cal{G}}(\Omega)$ be the set of all smooth Riemannian metrics on
41: $\;\Omega.\;$
42: Any two metrics $g_1,\;g_2 $ in ${\cal{G}} (\Omega)$
43: are called pointwise conformally equivalent if they are related under
44: multiplication by a smooth, positive function on $\Omega.\;$ This relation
45: is denoted by $g_1\propto g_2.\;$ Define the curvature energy function and area
46: function for a metric $g$ in $\Omega$ as follows:
47: \[ K(g,\Omega) = \int_{\Omega} K_{g} ^2 \; d\;g,\qquad A(g,\Omega) =
48: \int_{\Omega} d\;g ,\]
49: where $K_{g}$ is the scalar curvature of $g,$ and $d\,g$ is the area (volume) element.
50: For a given metric
51: $g_{0}$ in $\Omega$, define a function space
52: $\overline {{\cal{S}} (g_{0},C_1, C_2,\Omega)}\;$
53: to be the completion of the following set under any reasonable topology:
54: \[ {\cal{S}} (g_{0},C_1, C_2, \Omega) = \{ g \in {\cal{G}}(\Omega) |
55: g\propto g_0,\; A(g,\Omega) = C_1,\;K(g,\Omega) \leq C_2\},
56: \]
57: here $ C_1 $ and $ C_2$ are generic constants.\\
58:
59: We are mainly concerned with the two following questions: (a) Given a
60: sequence of metrics $\;\{g_k ,\;k \in \bf N \} $ in $ {\cal{S}}
61: (g_{0},C_1, C_2, \Omega),\; $ what is the set of its cluster points? (b) When can
62: one conclude that there must exist at least one limit and, if so,
63: what are its geometric properties? We have constructed an example of a
64: sequence of metrics with no subsequence that converges in the elementary
65: sense. Therefore, one must devise a geometrically reasonable topology in
66: the function space of metrics; in particular, the area functional
67: is continuous and the energy functional is lower-semi-continuous. \\
68:
69: Our main result is Theorem A at section \ref{th:theorem A}.
70: It could be summarized
71: as the following: As is shown in Figure \ref{fg:closed},
72: there is a subsequence of $\{g_n\}$
73: which locally weakly converges in $H^{2,2}$ (functions up to second derivative
74: are in $L^2$) to a Riemannian metric $f_0.\;$
75: However, this weak convergence is not on all of the surface $\Omega,$
76: but on $\Omega $ with a number of points $\{p_i\}$ deleted. There is a positive
77: amount of energy and area concentrations at each point $p_i.\;$
78: At each point $p_i,$ we use a rescaling argument to construct
79: a sequence of Riemannian metrics in $S^2$ with a small disk deleted
80: (the size of the disk approaches zero when the sequence takes a limit).
81: This renormalized sequence of metrics then (have a subsequence) converges to a
82: metric $f_i$ in $S^2$ with a finite number of points deleted. We then call
83: this metric a ``bubble metric.''
84: Iterating this process at each
85: new bubble point of $f_i$, and so on. The final ``limit'' of
86: the subsequence (passing to the diagonal subsequence) is a disjoint
87: union of these bubble metrics, which are defined
88: in different surfaces.
89: % $\Omega$ and a finite number of punctured 2-spheres, organized in a tree structure
90: %where $\Omega$ is the root vertex.
91: Each metric in the ``limit'' has a special
92: property that if it vanishes at one point in its domain,
93: it then vanishes everywhere in its domain.
94: While a bubble metric might be a metric in 2-sphere with constant curvature,
95: generically it is a metric defined on a punctured sphere and it
96: has a singular angle at each punctured point.\\
97:
98:
99: \begin{figure}
100: \centerline{\psfig{figure=new1.eps}}
101: \caption{Bubbles on Bubbles}
102: \label{fg:closed}
103: \end{figure}
104:
105:
106: \noindent {\bf 1.2 Extremal K\"ahler metrics.} The proposed problem
107: is motivated from the study of the existence problem of extremal K\"ahler
108: metric in a K\"ahler manifold $M.\;$ An extremal K\"ahler
109: metric is a critical point for the energy functional:
110: \[ E(g) = \int_M \;K_g^2 \;d\,g \]
111: on the space of K\"ahler metrics in a fixed K\"ahler class. The Euler
112: equation for the critical metric is (assume $\partial M =\emptyset$):
113: \[ {K_g}_{,\alpha \beta} = 0,\qquad \forall\; 1\leq \alpha,\;\beta \leq n, \]
114: where $n$ is the complex dimension of the manifold. \\
115:
116: This problem first appeared in \cite{calabi82}.
117: E. Calabi proposed to use the heat flow
118: method to solve this problem when the manifold
119: admits no holomorphic vector field. The heat flow he suggested is:
120: \[ {{\partial g_{\alpha \overline{\beta}}}\over{\partial t}} =
121: {K_g}_{,\alpha \overline{\beta}},
122: \qquad\forall\; 1\leq \alpha,\;\beta \leq n. \]
123: This flow indeed decreases the energy function $E$ along its trajectory.
124: To obtain the long term existence and the convergence as $ t\rightarrow \infty,\;$
125: one needs to understand the following question:
126: what is the weak topology of the set of metrics in a
127: fixed K\"ahler class with bounded energy?
128: An initial approach to this question would be
129: to study it in the case of complex dimension one, reducing the problem to
130: the one just described. Observe that any Riemannian metric on a surface is
131: also a K\"ahler metric; any two metrics $g_1,\; g_2$ are in the same K\"ahler
132: class if and only if $ g_1 \propto g_2\;$ and $ \;\int_\Omega d\;g_1 =\int_\Omega d\;g_2$ if $ \partial \Omega = \emptyset. $\\
133:
134: \noindent {\bf 1.3 Uniformization theorem and Dirichlet Problem.}
135: The selection of $L^2$ norm (rather than any $L^p$ norm with $p > 1$)
136: of the scalar curvature as energy function is not essential
137: as far as the weak topology of the function space is concerned.
138: It is significant, however, if we
139: consider the corresponding variational problem. The Euler equation
140: of the energy functional is:
141: \begin{equation} \triangle_g \;K_g + K_g^2 = C \; {\rm (generic\; constant).}
142: \label{eq:real:euler}
143: \end{equation}
144:
145: This Equation is called the extremal equation. Any metric satisfies
146: equation (~\ref{eq:real:euler}) is called an extremal metric, even if it
147: is only a stationary point of the energy functional. \\
148:
149: Let $ \Omega $ be any domain with smooth boundary; let $g_0$ be a smooth
150: metric in $\Omega$ which could be extended smoothly to a slight larger domain.
151: We want to ask if there always exists a metric $g,$ in a pointwise
152: conformal class of $g_0,$ which solves
153: equation (~\ref{eq:real:euler}) and satisfies the Dirichlet boundary condition:
154: \begin{equation}
155: g|_{\partial \Omega} = g_0|_{\partial \Omega},
156: \qquad {\partial g \over \partial n}|_{\partial \Omega}
157: ={\partial g_0 \over \partial n}|_{\partial \Omega}. \label{eq:direchlt}
158: \end{equation}
159: \noindent {\bf Conjecture 1. } {\it There always exists a solution to equation (1) with
160: Dirichlet boundary condition (\ref{eq:direchlt}), while solution metric
161: is pointwise conformal to the initial metric $g_0$.}\\
162:
163: The Euler equation (~\ref{eq:real:euler}) has an equivalent complex version:
164: \begin{equation}
165: {\partial \over \partial {\overline{z}}} {K_g}_{,zz} = 0,\label{eq:complex:euler}
166: \qquad {K_g}_{,zz} = {{\partial^2 K_g }\over{\partial z^2}} - 2\cdot
167: {{\partial K_g }\over{\partial z }} \cdot {{\partial \varphi}\over{\partial z}},
168: \end{equation}
169: where $g = e^{2\varphi} |d\,z|^2 $ locally.\\
170:
171: The Euler equation has two important special cases:
172: the first special case is the following
173: \begin{equation}
174: {K_g}_{,zz} = 0,
175: \label{eq:hciu}
176: \end{equation}
177: while the second special case is the following
178: \begin{equation} {K_g} \equiv C,\qquad {\rm or}\; - \triangle \varphi = C \cdot e^{2 \varphi}.
179: \label{eq:constant}
180: \end{equation}
181:
182: Any metric solves the equation (~\ref{eq:hciu})
183: has a special property that the Hessian of its
184: curvature is proportional to the metric tensor. Therefore, we may denote these
185: metrics as HCMU metrics (``Hessian of Curvature of Metric is Umbilical'').
186: If the Conjecture 1
187: were established, it would be desirable to understand the obstructions for
188: the existence of any HCMU metric and obstructions to the existence of
189: any constant curvature metric in a domain with appropriate Dirichlet boundary
190: condition~(\ref{eq:direchlt}). \\
191:
192: In the special case when $\partial \Omega = \emptyset$, any extremal metric also
193: has a constant curvature. Recall that the classical uniformization theorem
194: in a surface with no boundary asserts that any Riemannian metric is pointwise
195: conformal to a metric with constant curvature. Therefore, the Conjecture 1
196: (if proved), would generalize the classical uniformization
197: theorem in a surface with no boundary to any domain with smooth boundaries. \\
198:
199: Consider another special case where the boundaries are a set of isolated points.
200: To replace the Dirichlet boundary conditions, one requires all of the metrics
201: have a prescribed conical angle at each boundary point.
202: Such a surface is called ``a surface with conical singularities''
203: (see \cite{MTrojanov:conical91} for definition).\\
204:
205: \noindent {\bf Open Problem 1.} {\it Is any Riemannian metric on a singular surface
206: pointwise conformal to an extremal metric with the same
207: angle at each singular point.}\\
208:
209: In this special case, there have been plenty of attempts
210: ( mostly by analysts)
211: to generalize the classical uniformization theorem to surfaces with
212: conical singularities. Most work has concentrated on finding a
213: constant curvature metric in a pointwise conformal class.
214: However, we believe our approach may be more fruitful,
215: since not all surfaces with conical singularities support
216: a constant scalar curvature metric. Our program involves two related
217: but independent problems. The first problem is to use direct variational
218: method to give a positive answer to the above problem. For this purpose,
219: we need to study the weak compactness of the function space of Riemannian metrics with
220: finite energy and area (which is the subject of this study).
221: The second problem is to study the obstructions of
222: existence of any HCMU metric
223: and constant curvature metric in such surfaces. This second problem
224: is discussed in \cite{chen943}, where we give a necessary condition
225: for these surfaces to admit any HCMU
226: metric with non-constant curvature.\\
227:
228: \noindent {\bf 1.4 Bubbling phenomenon.} An important feature
229: of Theorem A is ``bubbling phenomenon.''
230: The bubbling phenomenon was first observed by
231: Sacks-Uhlenbeck \cite{Uhlenbeck} in 1979,
232: when they studied the existence theorem for harmonic maps
233: between two spheres. Since then, it has been studied and recognized in a wide
234: variety of geometric differential equations (see \cite{ParkerWolfson} for
235: further references). The solution spaces to these
236: equations are non-compact in any reasonable topology. The key
237: observation was that the non-compactness is
238: associated with the concentration of the energy density at isolated
239: points and that, by using the conformal invariance
240: of the equations, one could renormalized the solutions around
241: these points to obtain other solutions. This re-normalization process
242: is commonly referred to as ``bubbling.''\\
243:
244: Our ``bubbling'' procedure appears to be
245: very similar to the re-normalization process employed first
246: by Sacks-Uhlenbeck in 1979. However, there are some significant differences.
247: First, the function space is not a solution space
248: of any elliptic equation. Second, in most geometric
249: problems where the bubbling phenomenon occurs,
250: the energy function involves only the first derivatives
251: of the ``function'' in the solution space. However, the energy
252: functional here involves the second derivatives. To make the matter worse,
253: it involves only the Laplacian of the conformal parameter function, which
254: exerts a very weak control on the size of the metric.
255: These differences dictate a new approach
256: other than the standard one to solve the problem.
257: For instance, in most of these problems where bubbling
258: phenomenon occurs, one usually obtains a weak convergence result without too much
259: difficulty. The hard part is to show that the bubble points are
260: isolated. However, we have to do it exactly in the opposite order here.
261: The definition of a ``bubble point'' then becomes rather tricky, because
262: there is no weak convergent subsequence to work with. To overcome
263: this difficulty, we introduce
264: the notion of ``pseudo bubble point,'' where a subsequence of
265: metrics has a positive amount of energy and area concentration.
266: Unfortunately, the set of pseudo bubble points could be a
267: dense set in the domain.\\
268:
269: \noindent {\bf 1.5 Thick-thin Decomposition.}
270: In a higher dimensional
271: compact manifold, the Cheeger-Gromov theorem \cite{cheeger86} states
272: that any sequence of metrics in a compact manifold has a convergent
273: subsequence, provided that the sectional curvature is uniformly bounded,
274: the volume is bounded from below, and the diameter is bounded from above.
275: Similar results to \cite{cheeger86} were obtained
276: in \cite{Gromove81},\cite{Green88} and \cite{Peter87}
277: as well.
278: The following corollary of the theorem A could be regarded as a 2 dimensional version
279: of the Cheeger-Gromov thick-thin decomposition theorem,
280: under a weaker integral condition on
281: the curvature tensors.
282: \\
283:
284: \noindent {\bf Corollary B.} {\it For any locally weakly convergent sequence
285: of surfaces $ \{(\Omega,g_n),\;n \in {\bf N}\}\;$
286: where $ g_n \in {\cal{S}}(g_{0}, C_1, C_2, \Omega), $ and for any number $\epsilon > 0,$
287: there exist two integers
288: $ N_{thick} \;{\rm and }\; N_{thin} $ which depend only on $\epsilon$ and the total
289: energy $\sqrt{C_1 \cdot C_2} $ of this sequence (independent of $n$).
290: There exists a decomposition of
291: $(\Omega,g_n)$ into $ N_{thick}$ of thick components
292: $ \{(\Omega_{\alpha},g_n|_{\Omega_{\alpha}})\} $ (indexed by $ I_{thick}$) and $N_{thin}$ of
293: thin components $\{(\Omega_{\beta},g_n|_{\Omega_{\beta}})\}$ (indexed by $ I_{thin}$),
294: such that (see Figure~\ref{fg: thin-thick} on p.~\pageref{fg: thin-thick}): 1) $ \Omega = \displaystyle {\bigcup_{\alpha \in I_{thick}}} \Omega_{\alpha}
295: \displaystyle {\bigcup_{ \beta \in I_{thin}}} \Omega_{\beta}. \;$ 2)
296: For any fixed $\alpha \in I_{thick},$ except one, $ (\Omega_{\alpha}, g_n|_{\Omega_{\alpha}}) $
297: locally weakly converges to a metric in $S^2$ with a finite number of
298: small disks deleted; the other thick component locally weakly converges to a metric
299: in $\Omega$ with a finite number of disks deleted. Moreover, the size of each
300: deleted disk could be made as small as needed. 3) Each thick component is
301: self-connected; however, they are mutually disconnected if all of
302: the thin components are removed from the surface.
303: 4) Each thin component is topologically $ S^1 \times (a,b)$ and the
304: length of any concentric circle $S^1 \times \{x\} ( x \in (a,b)) $ is strictly
305: less than $\epsilon.$ }\\
306:
307: \noindent {\bf Remark 1.} {\it The terms ``thin'' and ``thick'' used here,
308: are slightly
309: different from what are originally used in \cite{cheeger86}.
310: For instance, the metrics in
311: a thin part in above corollary do not necessary have a
312: lower bound on the scalar curvature. }\\
313:
314: We initially hoped that both numbers $N_{thin}$ and $N_{thick}$ would be independent
315: of $\epsilon.\;$ However, we have constructed a sequence of rotationally
316: symmetric metrics in ${\cal{S}}(g_0,C_1, C_2, S^2)$ such that this sequence yields
317: as many thick components as needed when $\epsilon \rightarrow 0,$ without
318: incurring a blowing up of the energy functional. We first constructs a sequence of
319: metrics in a sequence of disks where the boundary of each disk is a smooth closed geodesic;
320: the length of the boundary geodesic tends to 0, while both the energy functional
321: and area are kept uniformly bounded from above
322: (see example 2 in p. \pageref{example} for details). We then construct a sequence of
323: metrics in a sequence of cylinder where both boundary circles are geodesics;
324: the length of the boundary geodesics tends to zero, while the energy functional
325: and area functional could make to be arbitrarily small. Using these metrics as building
326: block, we could construct a sequence of metrics with
327: bounded energy and area as in Figure \ref{fg:mid-bubbles}, where the limit
328: of metrics splits into as many parts as desired. Henceforth, Corollary {\bf B}
329: in its present form is the best one we could expect.\\
330:
331: \begin{figure}
332: \centerline{\psfig{figure=new2.eps}}
333: \caption{Rotationally symmetric Bubbles}
334: \label{fg:mid-bubbles}
335: \end{figure}
336:
337:
338: Motivated by the work of \cite{cheeger86}, C. Barvard and P. Pansu \cite{BavardP88}
339: studied the divergence problem of a sequence of metrics in any surface with
340: pointwise curvature bounded, allowing the conformal structure to be varied.
341: They have constructed some examples which show that the compactness fails
342: if the conformal structure is not bounded. As a matter of fact, the weak
343: compactness still fails even
344: if the conformal structure is fixed. Following the work of
345: C. Barvard and P. Pansu \cite{BavardP88}, M. Trojanov
346: \cite{MTrojanov:concentration91} first considered a sequence
347: of Riemannian metrics in a surface with a $L^p (\;\forall \;p > 1)$
348: norm of curvature (with respect to a fixed background metric) uniformly
349: bounded from above. He then showed such a sequence of metrics
350: either has a convergent subsequence or has at least one singular point.
351: However, the ``bubbles on bubbles'' phenomenon is not observed
352: in \cite{MTrojanov:concentration91}\\
353:
354: \noindent {\bf 1.6 Analytical approach.} In a local coordinate
355: system, any Riemannian metric $g$ could be expressed in
356: terms of its conformal parameter function $\varphi:$
357: \[ g = e^{2\varphi} (d\,x^2 + d\,y^2).\]
358: Therefore, $g$ can then be regarded as a solution
359: of scalar curvature equation:
360: \begin{equation} -\triangle\;\varphi = K \cdot e^{2\varphi}.
361: \label{eq:anaEQ} \end{equation}
362:
363: H. Brezis and F. Merle \cite{BrezisM91}
364: had studied the weak compactness
365: of the solution space of this equation. They consider a sequence
366: of pointwise conformal metrics in an open disk. It is assumed that
367: the $L^p (\forall \;p>1) $ norm of curvature is
368: uniformly bounded from above and the curvature function
369: is non-negative. They \cite{BrezisM91}
370: observed only the first level of bubbles, but not bubbles
371: on bubbles phenomenon. \\
372:
373: In both problems, difficulties arise because the right side of
374: equation (\ref{eq:anaEQ}) is only in $L^1.\;$ Interested readers are encouraged
375: to compare the main theorems of \cite{BrezisM91} with the
376: Theorem~\ref{theo:lo:weak} and~\ref{theo:lo:bubble}
377: in Section 3, where the problem is discussed from an analytic perspective.
378: There are some striking similarities which underscore the connections
379: between theses two problems. However, there are also some fundamental differences
380: between these two problems. It is assumed in \cite{BrezisM91}
381: that either the scalar curvature function
382: is non-negative, or the area element is in $L^{p'}
383: ({1 \over p} + { 1 \over p'} = 1).\;$ The compactness fails in our problem precisely
384: because that the scalar curvature function changes sign and the area element
385: is only in $L^1.\;$ \\
386:
387: \noindent {\bf 1.7 Organization}. In Section 2, we introduce the corresponding
388: local weak compactness
389: problem and conclude a local version of weak convergence theorem. Also in this
390: section, we analyze the sequence of metrics near a bubble point via blowing
391: up and conclude a theorem of bubbles on bubbles. This Section is the central
392: piece of this work. In Section 3, we essentially restate the weak compactness
393: theorem in a geometric context. In Section 4, we outline a bubbling procedure
394: and obtain a theorem of bubbles on bubbles.
395:
396:
397: \section { Local problem from an analytic viewpoint}
398:
399: \subsection{Introduction}
400: In this section, we consider the problem of weak compactness of a sequence of metrics
401: in a local coordinate disk. In one coordinate chart $(D,z)$, any
402: metric $g$ is defined as:
403: \begin{equation} g = e^{2 \varphi} (d x^2 + d y^2), \label{eq:lo:metric}
404: \end{equation}
405: and the curvature function is:
406: \begin{equation} K = - { {\triangle \varphi}\over {e^{2 \varphi}}}.
407: \label{eq:lo:curv}
408: \end{equation}
409:
410: A metric $g$ is said to have a finite area $C_1$ and a finite energy $C_2$
411: if and only if the following conditions are met:
412: \begin{equation}
413: \left\{ \begin{array}{ccc} \int\limits_{D}
414: e^{2 \varphi} dx dy & \leq & C_1, \\
415:
416: \int\limits_{D} {{(\triangle \varphi) ^2}
417: \over {e^{ 2 \varphi}}} d\, x d\,y & \leq & C_2.
418: \end{array} \right.
419: \label{eq:lo:bound}
420: \end{equation}
421:
422: A sequence of metrics $\{g_n\}$ where $g_n = e^{ 2\varphi_n} (d\,x^2 + d\,y^2)$
423: is said to have finite area $C_1$ and energy $C_2$ if and only if each $\varphi_n$
424: satisfies the inequality~(\ref{eq:lo:bound}). From this point on, in this
425: Section, we will use either $\{\varphi_n\}$ or $\{g_n\}$ to denote
426: a sequence of metrics with finite area $C_1$ and energy $C_2,$ unless
427: otherwise specified. \\
428:
429: The questions raised in Section 1.1 are : (1)
430: for a sequence of metrics $\{\varphi_n \}$
431: satisfies the inequalities (\ref{eq:lo:bound}),
432: does this sequence of functions have a uniform bound in $L^{\infty}(D)$?
433: (2) what is the weak limit of $\{\varphi_n\}$ under some reasonable topology?
434: \\
435:
436: {\small
437: \begin{rem} H. Brezis and F. Merle\cite{BrezisM91} considered a sequence of metrics
438: $\{\varphi_n\}$ satisfies the following equation:
439: \[ - \triangle \varphi_n = K_n \cdot e^{2 \varphi_n}\]
440: in an open disk $D$, where $ K_n \geq 0 $ and $K_n \in L^{p}(D),\;e^{2 \varphi_n}
441: \in L^{p'}(D) $ where $ {1\over p} + {1 \over p'} = 1).\;$ They proved that
442: one of the following three alternatives holds true (mutually exclusive):
443: \begin{enumerate}
444: \item Vanishing case: $ \varphi_n \rightarrow - \infty $ uniformly in any compact
445: subset of $D.\;$
446: \item Convergence: there exists a function $ \varphi \in H^{2,2}(D) $ such
447: that $ \varphi_n \rightharpoonup \varphi $ weakly in $H^{2,2}_{loc}(D). $
448: \item There exists a finite number of bubble points $\{p_1,p_2,\cdots,
449: p_m\}$ such that as a measure,
450: \[ K_n \cdot e^{2\varphi_n} \rightharpoonup \sum_i \alpha_i \cdot \delta_{p_i}. \]
451: \end{enumerate}
452:
453: They conjectured that $ \alpha_i = 4 \pi \cdot m_i$ for some integer $m_i.\;$
454: This conjecture was proved by Y. Y. Li and I. Shafrir\cite{LiSh94}.
455: However, it remains open whether
456: $m_i$ actually equals $1$.
457:
458: Our problem differs from the problem consider by H. Brezis and F. Merle
459: significantly. We quote their results here for comparison. The non-compactness
460: occurs in our case is precisely because the curvature
461: changes sign in a small neighborhood and the area element is not in
462: $L^{p'}$ for any $p'> 1.\;$
463: \end{rem}
464: }
465:
466: For any sub-domain $\Omega$ in $D$, re-label the
467: energy and area for a conformal parameter functions as:
468: \[
469: A_c(\varphi, \Omega) = \int_{\Omega} e^{2 \varphi} d\,x d\,y, \qquad
470: K_c(\varphi, \Omega) = \int_{\Omega} {{(\triangle \varphi) ^2} \over {e^{ 2 \varphi}}}
471: d\,x d\,y.
472: \]
473:
474: A ``$0$'' metric should have ``$0$'' area and energy. Since a ``$0$''
475: metric has a conformal parameters function $-\infty$, we define:
476: $A_c(-\infty,\Omega) = K_c(-\infty,\Omega)= 0.\;$\\
477:
478: For the convenience of notations, we add ``$-\infty$'' into
479: $H^{2,2}(\Omega)$. The resulted space is denoted by
480: ${\hat{H}}^{2,2}(\Omega).\;$
481: A sequence of functions $\{\varphi_n\} \in H^{2,2}(\Omega) $
482: weak converges to a function $\varphi_0$ in
483: ${\hat{H}}^{2,2}_{loc}(\Omega)$ if one of the
484: following two alternatives holds true (mutually exclusive):
485: \begin{enumerate}
486: \item (Vanishing case): If $\varphi_0 \equiv -\infty, $
487: then $\varphi_n \rightarrow -\infty$
488: uniformly in any compact sub-domain of $\Omega.\; $
489: \item (Non-vanishing case): If $\varphi_0 \in H^{2,2}(\Omega), $ then
490: $ \varphi_n \rightharpoonup \varphi_0\;{\rm weakly\; in} \; H^{2,2}_{loc}(\Omega).$
491: \end{enumerate}
492:
493: \begin{defi} A point $p$ is said to be a bubble point of $\{\varphi_i\}$ if
494: for any $r > 0$,
495: \begin{equation}
496: \displaystyle {\underline{\lim}}_{n \rightarrow \infty}
497: \int\limits_{D_{r}(p)} {{(\triangle \varphi_n) ^2} \over {e^{ 2 \varphi_n}}}
498: dx\,dy \geq \alpha>0,\qquad
499: \displaystyle {\underline{\lim}}_{n \rightarrow \infty} \int\limits_{D_{r}(p)}
500: e^{2 \varphi_n} d\,x d\,y \geq \beta>0. \label{eq:bubble}
501: \end{equation}
502: where $ D_{r}(p)$ denotes a
503: coordinate disk centered at $p$ with radius $r.\;$
504: The largest possible numbers $\alpha $ and $\beta$ are the concentration weights
505: of the energy function
506: and area function at this point $p.\;$
507: \end{defi}
508:
509: Clearly, if $p$ is a bubble point of $\{\varphi_n\}$, then $p$ is
510: a bubble point of any subsequence of $\{\varphi_n\}.$ \\
511:
512: \noindent {\bf Example 1.} {\it Let
513: $g_n = {{n^2}\over{(1+ n^2\cdot |z + n^{-0.33} |^2)^2}} |d\,z|^2$ be a
514: sequence of metrics in $S^2$ with a constant curvature of 1. This sequence
515: of metrics then converges to $0$ at every point (including the point $z = 0$)
516: on $S^2.\;$ However, the concentrations of energy and area at $z=0$ are $4\pi, 4\pi$.
517: The metrics could be renormalized as: $\tilde{g}_n(z) =g_n((z-n^{-0.33})/n).\; $
518: This new sequence $\tilde{g}_n$
519: weakly converges to a metric in $S^2$ with constant curvature. }\\
520:
521:
522: The main theorems in this Section are:
523:
524: \begin{theo} Let $\{\varphi_n,\; n\in {\bf N}\}$ be a sequence of metrics
525: in $H^{2,2}(D)$ with a finite area $C_1$ and energy $C_2$. There exists a
526: subsequence $\{\varphi_{n_j},j \in {\bf N}\}$ of $\{\varphi_n\},$
527: a finite number of bubble points
528: $\{p_1,p_2,\cdots,p_m\}(0\leq m \leq \sqrt{\frac{C_1 \cdot C_2}{4\pi^2}})$
529: with respect to $\{ \varphi_{n_j}, j \in {\bf N}\}, $ and a metric $\varphi_0
530: \in {\hat{H}}^{2,2}_{loc}(D \setminus \{p_1,p_2,\cdots,p_m\})$ such that:
531:
532: \[ \varphi_{n_j} \rightharpoonup \varphi_0\;{\rm in}\; {\hat{H}}^{2,2}_{loc}
533: (D \setminus \{p_1,p_2,\cdots,p_m\}).\]
534:
535: If the energy and area concentrations in each bubble point $p_i$ are
536: $A_{p_i}$ and $K_{p_i}$ for any $i \in [1,m],$ then:
537: \begin{eqnarray}
538: \displaystyle{\lim_{j\rightarrow \infty}} A_c(\varphi_{n_j},D) &
539: = & A_c(\varphi_0,D\setminus\{p_1,p_2,\cdots,p_m\} ) +
540: \sum_{i=1}^{m} A_{p_i} \label{eq:lo:weak:area} \\
541: \displaystyle{\lim_{j\rightarrow \infty}} K_c(\varphi_{n_j},D)
542: & \geq & K_c(\varphi_0,D\setminus\{p_1,p_2,\cdots,p_m\} ) +
543: \sum_{i=1}^{m} K_{p_i}. \label{eq:lo:weak:energy}\end{eqnarray}
544: \label{theo:lo:weak}
545: \end{theo}
546:
547: \begin{rem}
548: The equality in formula~\ref{eq:lo:weak:energy} holds if
549: $\{\varphi_n\}$ minimizes the energy function.
550: \end{rem}
551:
552: \begin{theo}
553: For any metric $\varphi$ with a finite area $C_1$ and energy $C_2$
554: in $D\setminus \{0\}$,
555: define $\phi(r) = {1 \over 2 \pi} \int_{0}^{2\pi}
556: \varphi(r\cos \theta, r\sin \theta)\, d\theta.\; $
557: The following three statements hold true:
558: \begin{enumerate}
559: \item $\displaystyle{\lim_{r \rightarrow 0}} (\varphi(r\cos \theta, r\sin \theta) + \ln r) = -\infty. $
560: \item $\displaystyle {\lim_{r\rightarrow 0}} \;\phi'_r(r)\cdot r $
561: exists and is finite.
562: \item There exists a constant $\beta \in (0,1) $ and two constants
563: $C_3, C_4$ such that:
564: \[ {1\over \beta} (\phi(r) + \ln r) + C_3 \leq \varphi(r\cos \theta, r\sin\theta)
565: + \ln r \leq \beta (\phi(r) + \ln r) + C_4.\]
566: \end{enumerate}
567: \label{theo:lo:prop}
568: \end{theo}
569:
570: \begin{theo} (Bubbles on bubbles).
571: Let $\{\varphi_n\}$ be a sequence of metrics in $D$ with finite area
572: $C_1$ and finite energy $C_2$. Suppose that $ p=0$ is
573: the only bubble point in $D$ with area concentration $A_p$
574: and energy concentration $K_p.\;$ Suppose there exists a
575: metric $\varphi_0 \in {\hat{H}}^{2,2}(D\setminus\{p\})$ such that
576: $\varphi_n \rightharpoonup \varphi_0$ in
577: $ {\hat{H}}^{2,2}_{loc}(D\setminus\{p\}).\;$
578: A sequence of numbers $\{\epsilon_n \searrow 0\}$ can
579: be chosen to re-normalize the sequence of metrics as:
580: $ \phi_n(x,y) = \varphi_{n}(\epsilon_{n} \cdot x,\epsilon_{n} \cdot y)
581: + \ln \epsilon_{n} (\forall n \in {\bf N}).\;$
582: There exists a subsequence $\{\varphi_{n_j},j \in {\bf N}\}$
583: of $\{\varphi_n\},$ a
584: finite number of bubble points $\{q_1,q_2,\cdots,q_m\}
585: ( 0 \leq m\leq \sqrt{\frac{A_p\cdot K_p}{4\pi^2}})$
586: with respect to the subsequence of metrics $\{ \phi_{n_j}\},\;$
587: a metric $\phi_0\in {\hat{H}}^{2,2}( S^{2}\setminus
588: \{\infty, q_1,q_2,\cdots,q_m\}) $ such that:
589: \[ \phi_{n_j} \rightharpoonup \phi_0 \;{\rm in}\; {\hat{H}}^{2,2}_{loc}
590: ( S^{2}\setminus \{\infty, q_1,q_2,\cdots,q_m\}).\]
591:
592: If the energy and area concentrations of $\{\phi_n\}$
593: at each point $q_i$ are $K_{q_i} $ and $A_{q_i},$ then:
594: \begin{eqnarray}
595: A_p & \geq & A_c(\phi_0,S^2 \setminus\{q_1,q_2,\cdots, q_m\})
596: + \sum_{i=1}^{m} A_{q_i}\label{eq:blow:area}\\
597: K_p & \geq & K_c(\phi_0,S^2 \setminus\{q_1,q_2,\cdots, q_m\})
598: + \sum_{i=1}^{m} K_{q_i}. \label{eq:blow:energy}
599: \end{eqnarray}
600: \label{theo:lo:bubble}
601:
602: If $\phi_0 \equiv -\infty$( vanishing case), then $m \geq 2 $
603: and $p\, ( z = 0) $ is a bubble point of $\{\phi_{n_j}, j \in {\bf N}\}. \;$
604:
605: \end{theo}
606:
607: \begin{rem}
608: The difference of the left side and right side of the inequality~\ref{eq:blow:area}
609: represents the amount of area lost during the blowing up procedure. If this
610: amount is zero, there is no area trapped in the neck.
611: \end{rem}
612:
613: In Subsection 3.2, we prove three important lemmas (lemma 2,4 and 6), which provide a
614: technical foundation for the main theorems. The proof are rather technical,
615: readers are then encouraged to skip Subsection 3.2
616: and read the other Subsections first. In Subsection 3.3,
617: we prove a weak convergence
618: theorem. In Subsection 3.4, we briefly describe the properties of the
619: limit metrics. In Subsection 3.5, we show that a renormalized sequence of
620: metrics at each bubble point will have a weak convergent subsequence.
621:
622:
623: \subsection{Small energy lemmas}
624: In this subsection, the notion of a pseudo bubble point is introduced. It is
625: subsequently used to prove three key lemmas: lemma 2, 4 and 6.
626: Lemma 2 shows that the concentration of total
627: energy (product of curvature energy and area) at each bubble point
628: must be greater than $ 4\pi^2.\;$ Thus, there are at most a finite number
629: of bubble points for any subsequence of metrics. Lemma 4 shows
630: that if a point is not a pseudo bubble point,
631: then the sequence of metrics in a neighborhood
632: of that point is uniformly bounded from above. Lemma 6
633: shows that in any domain, if the metrics are uniformly bounded from above, then
634: either the sequence of metrics approaches $0$ everywhere in its domain, or a
635: subsequence of these metrics weakly converges in $H^{2,2}$ in any compact sub-domain.\\
636:
637: For any $p \in D,$
638: a small disk center at $p$ with radius $r$ will be denoted by $D_{r}(p)$.
639: \[ D_{r}(p) = \{(x,y) \in D | (x-x_p)^2 + (y-y_p)^2 < r^2 \}.\]
640:
641: Define local energy and area functions with respect to any point $p \in D $ as the following:
642:
643: \[\begin{array}{ccc}
644: K_p(r) & = & \displaystyle{{\overline{\lim}}_{ k \rightarrow \infty}}
645: { \int_{D_{r}(p)} {{(\triangle \varphi_k)}^2
646: \over{e^{2 \varphi_k}}}d\,x\;d\,y },\qquad \forall r > 0, \\
647: A_p(r) & = & \displaystyle{{\overline{\lim} }_{ k \rightarrow \infty}}
648: { \int_{D_{r}(p)} e^{2\varphi_k} d\,x\;d\,y,}\qquad \forall r > 0.
649: \end{array}
650: \]
651:
652: In this definition, the limit taken is only an upper limit, since
653: it is not known whether
654: $\{\varphi_n \}$ has any weak convergent subsequence.
655: \begin{defi} The energy and area concentration functions
656: of a sequence of metrics $\{g_k \}$ at any point
657: $\;p \in M,$ are defined as follows:
658: \[
659: K_p = \displaystyle{\lim_{r \rightarrow 0}}\, K_p(r),\qquad
660: A_p = \displaystyle{\lim_{r \rightarrow 0}}\, A_p(r).
661: \]
662: \label{defi:energy:area}
663: \end{defi}
664:
665: Any point $p\in D$ is called a pseudo bubble point
666: if and only if $ A_p >0 $ and $ E_p > 0.\;$ Later, we could show that
667: $A_p > 0$ actually implies $E_p > 0.\;$ At a pseudo bubble point, there
668: exists a subsequence of $\{ \varphi_n\}$ such that this
669: subsequence has a positive amount of area and energy concentrations
670: there. If we pass to
671: this subsequence, the pseudo bubble point becomes a real bubble point.
672:
673: \begin{prop} Let $p$ be a pseudo bubble point of a sequence of metrics
674: $\{\varphi_n, n \in {\bf N}\}, $ there then
675: exists a subsequence of $\{ \varphi_n\}$
676: such that $p$ is a real bubble point with respect to this subsequence.
677: \label{prop:psedo-bubble}
678: \end{prop}
679: {\bf Proof.} The proof is straightforward.
680:
681: \begin{defi} The waist concentration function, $ l_p(\rho,\rho_{0}), $
682: for any
683: $0 < \rho < \rho_{0}$ is defined as:
684:
685: \[
686: l_p(\rho,\rho_{0}) =\displaystyle {{ \underline {\lim}}_{ n\rightarrow \infty}}\;
687: \displaystyle {\min_{\rho \leq r \leq \rho_{0}} }|{\partial D_r}|_{g_{n}} =\displaystyle
688: {{ \underline {\lim}}_{ n\rightarrow \infty}}\;
689: \displaystyle {\min_{\rho \leq r \leq \rho_{0}} } \int_0^{2 \pi}
690: e^{ \varphi_n(r \cos \theta, r \sin \theta)}\, r\, d\,\theta. \]
691:
692: \label{defi:waist}
693: \end{defi}
694:
695: \begin{lem} Let $\{\varphi_n\}$ be a sequence of metrics with finite area $C_1$
696: and finite energy $C_2.\;$ For any $\rho_{0} > 0$, we have
697: $ \displaystyle {{\lim}_{\rho\rightarrow 0}}\; l_p(\rho,\rho_{0}) = 0$.
698: \label{lem:waist}
699: \end{lem}
700:
701: \noindent{\bf Proof.} If the lemma is false, then
702: there exists a number $\epsilon > 0$
703: such that:
704: $\displaystyle {\lim_{\rho\rightarrow 0}} \;l_p(\rho,\rho_{0}) = 2\; \epsilon > 0.\;$ Choose $\rho$ small enough so that:
705: \begin{equation}
706: {{\epsilon^{2}}\over{ 2 \pi}}\;\ln {{\rho_{0}}\over{\rho}} > 2 \cdot C_1.
707: \end{equation}
708:
709: Since $l_p(\rho,\rho_{0}) $ is a monotonely increasing function on its variable
710: $\rho > 0,\; $
711: \[ l_p(\rho,\rho_{0}) \geq 2 \epsilon, \; \forall\; \;0 < \rho <\rho_{0}.\]
712:
713: In other words,
714: \[{ \displaystyle{\underline{\lim}}_{n\rightarrow \infty}}
715: \displaystyle {\min_{\rho \leq r \leq \rho_{0}}}|{\partial D_r}|_{g_{n}}
716: \geq 2\; \epsilon. \]
717:
718: Fixing the pair of numbers $\,\rho,\,\rho_{0},$ there then exists
719: a number $n_0$ such that $ |{\partial D_r}|_{g_{n_0 }} > \epsilon, \; \forall\;
720: \rho \leq r \leq \rho_{0}.\;$ In a local coordinate,
721: \[ |{\partial D_r}|_{g_{n_0}} = \int_{0}^{2\pi} e^{\varphi_{n_0}} r d\,
722: \theta >\epsilon, \; \forall r \in [\rho,\rho_{0}]. \]
723:
724: However,
725: \begin{eqnarray*} 2 \pi \cdot C_1 & \geq & \int_{\rho}^{\rho_{0}}
726: \int_{0}^{2\pi} e^{2 \varphi_{n_0}}\cdot r \cdot d\theta d\,r
727: \cdot \int_0^{ 2\pi}\, 1 \, d\theta \\
728: & \geq & \int_{\rho}^{\rho_{0}}
729: ( \int_{0}^{2\pi} e^{ \varphi_{n_0}} d\theta)^{2} r d\,r \\
730: &\geq & \int_{\rho}^{\rho_{0}}({{\epsilon}\over{r}})^{2} r d\,r \\
731: &\geq & \epsilon^{2} \ln{{\rho_{0}}\over{\rho}} > 4 \pi \cdot C_1.
732: \end{eqnarray*}
733:
734: The last inequality holds true because of inequality (15).
735: Thus, $ 2\pi > 4\pi, $ which is impossible.
736: The lemma is then proved. QED.\\
737:
738: The following theorem is a generalization of the classical
739: isoparametric inequality. It is a key theorem which we
740: will use it over and over again.
741:
742: \begin{theo}(Readers are referred to \cite{Burago80} for
743: further reference). Let $g$ be a
744: metric in an Euclidean disk $D$ such that
745: $\int_D |K_g| d g < \infty.\;$ For any disk
746: $D_1 \subset \subset D,$ we have:
747:
748: \[ \int_{D_1}\; |K_g| d\,g \geq 2\pi -
749: {(\int_{\partial D_1 } d s_g)^2 \over {2 \int_{D_1} d g}}
750: = 2 \pi - {|\partial D_1|_g^2 \over {2 \int_{D_1} d\,g}}.\]
751: \label{theo:lo:ineq}
752: \end{theo}
753:
754: \begin{lem} Let $\{\varphi_n\}$ be a sequence of metrics with finite area $C_1$
755: and finite energy $C_2.\;$
756: If $p$ is a bubble point of $\{\varphi_k \},\;$
757: then the following inequality holds true:
758: \[ \sqrt{ K_p \cdot A_p } \geq 2 \pi.\]
759: \label{lem:bubble}
760: \end{lem}
761:
762: \begin{rem} (a) The best constant in the above estimate is
763: $ 4 \pi.\;$ \\
764: (b)This lemma also proves that $A_p > 0 $ if and only
765: if $K_p > 0. $
766: \end{rem}
767:
768: This lemma implies that there are only a finite number of bubble points.
769: It can also be regarded as a ``small energy lemma.'' In other words, if
770: the total energy $ \sqrt{ K(\Omega) \cdot A(\Omega)}\; $ is small enough
771: ($\leq 2 \pi$), any weak convergent subsequence of metrics does not
772: have any bubble point in any compact sub-domain of $\Omega.\;$
773: \\
774:
775: \noindent{\bf Proof.} Suppose $p $ is a bubble point
776: and $ A_p > 0.\; $ Let $\;\epsilon > 0 $
777: be any small positive number.
778: Recalled that $A_p = \displaystyle{\lim}_{r \rightarrow 0} A_p(r).\;$
779: Since $ A_p(r)$ is a monotonely increased function on its variable $r,$ then
780: ${\lim}_{r\rightarrow 0} A_p(r) \geq A_p > 0.\;$
781: Choose $\rho_{0}$ and for $n$ large enough:
782:
783: \[ A_p \leq A_{\rho_0}(r) = \overline{\lim_{n\rightarrow \infty}} \int_{D_{\rho_{0}}} d\, g_{n} < (1+{\epsilon \over 2}) A(p). \]
784: For $n$ large enough, we have
785: \[
786: \int_{D_{\rho_{0}}} d\, g_{n} < (1+\epsilon ) A(p)
787: \]
788:
789: Lemma~\ref{lem:waist} then implies:
790: \[\displaystyle {\lim_{\rho \rightarrow 0}}\; l_p(\rho,\rho_{0}) = 0,\;\;
791: \forall \; \rho_{0} > 0.\]
792:
793: For any $\epsilon > 0,$ we choose a small number $\rho_{1} < \rho_{0} $
794: such that $l_p(\rho_{1},\rho_{0}) < \epsilon.\; $
795: There exists a positive number $N $
796: which depends only on $\epsilon $ such that
797: (after passing to a subsequence):
798: \begin{equation} \displaystyle {\min_{\rho_{1} \leq r \leq \rho_{0}}}
799: |{\partial D_r}|_{g_{n}} < 2 \epsilon,\qquad \forall\; n > N. \end{equation}
800:
801: There exists a number $\rho_{n} \in [\rho_{1},\rho_{0}]$ such that:
802: \[ |\partial D_{\rho_{n}}|_{g_{n}} < 3\; \epsilon, \qquad \forall \rho_1\leq \rho_n \leq \rho_0. \]
803:
804: Therefore,
805:
806: \[ A_p \leq \int_{D_{\rho_{1}}} d g_{n}\leq
807: \int_{D_{\rho_{n}}} d g_{n} \leq (1+\epsilon) A(p). \]
808:
809: According to Theorem 4, we have:
810: \begin{eqnarray*} \int_{D_{\rho_n}} |K_{g_n}| d g_n
811: & \geq & 2 \pi - {{ |\partial D_{\rho_{n}}
812: |_{g_{n}}^{2}}\over{ 2 \int_{D_{\rho_{n}}} d\,g_{n}}} \\ & > & 2 \pi -
813: {{9 \epsilon ^{2}}\over{ 2 A(p)}} > 0.
814: \end{eqnarray*}
815:
816: The last inequality holds for any small $\epsilon > 0.\;$ Hence, we have:
817: \begin{eqnarray*}
818: \int_{D_{\rho_{n}}} K_{g_{n}}^{2} d\,g_{n} & \geq &
819: {{ ( \int_{D_{\rho_{n}}} K_{g_{n}}
820: d\,g_{n})^{2}}\over {\int_{D_{\rho_{n}}} d\,g_{n}}} \\ & > &
821: {{ ( 2\pi - {{9 \epsilon ^{2}}\over{ 2 A(p)}})^{2}} \over {( 1 + \epsilon ) A(p)}}.
822: \end{eqnarray*}
823:
824: Since $\rho_{n} < \rho_{0} $ ,
825:
826: \[ \int_{D_{\rho_{0}}} K_{g_{n}}^{2} d\,g_{n} >
827: {{( 2\pi - {{9 \epsilon ^{2}}\over{ 2 A(p)}})^{2}} \over {( 1 + \epsilon ) A(p)}},
828: \;\; \forall\; \;n > N. \]
829:
830: In other words,
831: \[ \displaystyle {\underline {\lim}_{n\rightarrow \infty}}\;
832: \int_{D_{\rho_{0}}} K_{g_{n}}^{2} d\,g_{n} >
833: {{( 2\pi - {{9 \epsilon ^{2}}\over{ 2 A(p)}})^{2}} \over {( 1 + \epsilon ) A(p)}}.\]
834:
835: Let $ \epsilon \rightarrow 0,\; $ then let $ \rho_{0} \rightarrow 0$ , we have:
836: \[ K_p = \displaystyle{ \lim_{\rho_{0} \rightarrow 0}}\;
837: \displaystyle{ \underline {\lim}_{n \rightarrow \infty}}\;
838: \int_{D_{\rho_{0}}} K_{g_{n}}^{2} d\,g_{n} \geq {{ 4 \pi^{2}}\over { A(p)}}. \]
839:
840: The lemma is then established. QED.\\
841:
842:
843: \begin{lem} If $ \{\varphi_n\}$ has finite area $C_1$,then
844: $ \displaystyle {{\underline{\lim}}_{n\rightarrow \infty}}\;
845: \displaystyle {\min _{ 0\leq \rho\leq r\leq \rho_{0}}}\;
846: {{1}\over{2\pi}} \int_{0}^{2\pi} \varphi_{n}
847: (r \cos \theta, r \sin \theta)\;
848: d\,\theta$
849: is bounded from above for any interval $[\rho,\rho_{0}]$.
850: \label{lem:metric:cylin}
851: \end{lem}
852:
853: \noindent {\bf Proof.} If this lemma is false, there then exists
854: $ 0 < \rho < \rho_{0}$ such that:
855: \[ \displaystyle {{\underline{\lim}}_{n\rightarrow \infty}}\;
856: \displaystyle {\min _{\rho\leq r\leq \rho_{0}}}\;
857: {{1}\over{2\pi}} \int_{0}^{2\pi} \varphi_{n}
858: (r \cos \theta, r \sin \theta)\;
859: d\,\theta= \infty .\]
860:
861: Passing to a subsequence if necessary, we may assume:
862: \[
863: {{1}\over{2 \pi}} \int_{0}^{2\pi}\varphi_{n}(r\cos \theta,r \sin \theta)
864: d\theta > n,
865: \; \forall\; r\in [ \rho,\rho_{0}] .
866: \]
867:
868: A Schwartz type of inequality implies:
869: \[ {{1}\over{2\pi}} \int_{0}^{2\pi} e^{ 2\varphi_{n}(r\cos \theta,r \sin \theta)} d\theta
870: > e^{{{1}\over{2 \pi}} \int_{0}^{2\pi} 2 \varphi_{n}(r\cos \theta,r \sin \theta) d\theta }
871: > e^{2 n},\; \forall\; r\in[\rho,\rho_{0}].\]
872:
873: The last inequality implies:
874: \[ C_1 > {{1}\over{2\pi}}\int_{\rho}^{\rho_{0}} \int_{0}^{2\pi}
875: e^{ 2\varphi_{n}(r\cos \theta,r \sin \theta)}\; r d\,r d\theta > e^{2 n}
876: \int_{\rho}^{\rho_{0}} r d\,r = e^{2n} (\rho_{0}^{2}-\rho^{2})/2 \rightarrow \infty. \]
877:
878: This is a contradiction. This lemma is proved. QED.\\
879:
880:
881: The following lemma shows
882: that the conformal parameters $\{\varphi_k\}$
883: must have a uniform upper bound, away from the set of bubble points.
884:
885: \begin{lem} Let $\{\varphi_n\}$ be a sequence of metrics with finite area $C_1$
886: and finite energy $C_2.\;$ If $p$ is not a pseudo bubble point of
887: $\{\varphi_k,\; k \in{\bf N}\},$ i.e.,$ A(p) =0,$ there then exists a small
888: neighborhood $ {\cal {O}}(p) $ of $ p $ and a positive constant $ C $
889: such that $ \displaystyle { \sup_{k \in {\bf N}} }
890: \;\displaystyle {\sup_{ q \in { \cal {O}}(p)}} \varphi_k (q) \leq C.\;$
891: \label{lem:regular}
892: \end{lem}
893:
894:
895: \noindent {\bf Proof. } Define a new function:
896: \[ A_{n}(\rho) = \int_{D_{\rho}} d g_{n} = \int_{ x^2 + y^2 \leq \rho^2}
897: \,e^{ 2 \varphi_n} dx dy,\qquad \forall \; n \in {\bf N},
898: \;\forall\; \rho \; > \;0 .\]
899:
900: Choose a small coordinate disk
901: $D_{r_{0}}(0)$ so that: $ 2 \cdot C_2 \cdot A_n(r_0) < \pi^2.\;$
902: If this lemma is false, we could modified the sequence of metrics
903: slightly so that: $\varphi_n(p) \rightarrow \infty.\;$ We want to
904: draw a contradiction from this assumption. \\
905:
906: For any pair of numbers $ r_{1} > r_{2},\; $ consider the following:
907: \begin{eqnarray*}
908: \lefteqn{| \int_{0}^{2\pi}
909: {{\partial\,\varphi_{n}(r_1\cos \theta,r_1 \sin \theta)}\over{\partial\,r}} r_1 d\,
910: \theta - \int_{0}^{2\pi}
911: {{\partial\,\varphi_{n}(r_2\cos \theta,r_2 \sin \theta)}\over{\partial\,r}} r_2
912: d\,\theta | }
913: \\ & & = |\int_{r_{2}}^{r_{1}} \int_{0}^{2\pi} {{\partial \over {\partial r}}}
914: ({{\partial \,\varphi_{n} (r\cos \theta,r \sin \theta)}\over{\partial\,r}} \cdot r)
915: d\,\theta \;d\,r| = |\int_{r_{2}}^{r_{1}}
916: \int_{0}^{2\pi} ( \varphi_n''\cdot r + \varphi_n')
917: d\,\theta \;d\,r| \\ & & = |\int_{r_{2}}^{r_{1}} \int_{0}^{2\pi} \triangle \varphi_{n}(r \cos \theta,r \sin \theta)
918: r d\,\theta \;d\,r | \\ & & \leq (\int^{r_{1}}_{r_{2}}\int_{0}^{2\pi}
919: {{(\triangle \varphi_{n} )^{2}}\over{ e ^{ 2 \varphi_{n} }}} r d\,\theta
920: \;d\,r)^{ {{1}\over{2}}} \; (\int_{r_{2}}^{r_{1}} \int_0^{2\pi} e^{ 2\varphi_{n}}
921: r d\theta\; d\,r )^{{{1}\over{2}}}.
922: \end{eqnarray*}
923:
924: Since the energy of this sequence of metrics is uniformly bounded
925: from above, the previous inequality implies:
926: \begin{eqnarray*} | \int_{0}^{2\pi} {{\partial\,\varphi_{n}(r_{1} \cos \theta,r_1 \sin \theta)}\over{\partial\,r}}\, r_1 d\,
927: \theta - \int_{0}^{2\pi} {{\partial\,\varphi_{n}(r_{2}\cos \theta,r_2 \sin \theta )}\over{\partial\,r}}\, r_2
928: d\,\theta | & \leq \\ C_2^{1\over 2} \cdot \; (\int_{r_{2}}^{r_{1}} \int_0^{2\pi}
929: e^{ 2\varphi_{n} } r d\theta\; d\,r )^{{{1}\over{2}}}. & \end{eqnarray*}
930:
931: Fixing the number $n$, observe that
932: $\displaystyle {\lim_{r\rightarrow 0} } \int_{0}^{2\pi}
933: {{\partial \varphi_{n}(r\cos \theta, r \sin \theta)}\over{\partial r}} r d\theta = 0.\;$
934: Let $ r_{2} \rightarrow 0 $ and $r_1 = r,$ the result is:
935:
936: \[ | \int_{0}^{2\pi} {{\partial\,\varphi_{n}(r \cos \theta,r \sin \theta)}\over{\partial\,r}} r
937: d\,\theta | < C_2^{1\over 2} \cdot \; \sqrt { A_{n}(r )}. \]
938:
939: Define $\psi_{n}(r) = {{1}\over{2\pi}} \int_{0}^{2\pi}
940: \varphi_{n}(r \cos \theta,r \sin \theta ) d\,\theta,\;$ then:
941: \begin{eqnarray*}
942: |\psi_{n}(r) - \psi_{n}(0)| & \leq & \int_{0} ^{ r}
943: {{1}\over{2\pi}} |\int_{0}^{2\pi}{{\partial \varphi_{n}}\over{\partial \rho}}
944: \rho d\,\theta|\; \frac{ d\,\rho}{\rho} \\ & \leq & \int_{0}^{r} {{1}\over{2\pi}}\cdot C_2^{1\over 2} \cdot \sqrt{ A_{n}(\rho)} \cdot
945: {{d\,\rho}\over{\rho}}.
946: \end{eqnarray*}
947:
948: Since $\psi_n(0) = \varphi_n(p)$, therefore ($0 < \alpha < 1 $)
949: \begin{equation} |\psi_{n}(r) - \varphi_n(p)| \leq {{1}\over{2\pi}}\cdot C_2^{1\over 2} \cdot
950: \int_{0}^{r} ( {{A_{n}(\rho)}\over{\rho^{\alpha}}})^{{{1}\over{2}}}
951: \cdot {{\rho^{{{\alpha}\over{2}}}}\over{\rho}} d\,\rho. \label{eq:ineq20} \end{equation}
952:
953:
954: Following lemma 3, $\displaystyle {\lim_{n \rightarrow \infty} } \psi_{n}(r) < \infty.\; $
955: It is a contradiction if the right hand side (RHS) of the previous
956: inequality (\ref{eq:ineq20}) is uniformly bounded from above since $\{\varphi_n(p)\} \rightarrow \infty.\;$ However, the (RHS) of the inequality (\ref{eq:ineq20}) is bounded according
957: to the next lemma (choose $\alpha = 1).\;$ The lemma is then proved. QED.
958:
959: \begin{lem} Let $\{\varphi_n\}$ be a sequence of metrics with finite area $C_1$
960: and finite energy $C_2.\;$ Suppose $A_p = 0.\;$ For any small number $r > 0, $
961: there exists a positive constant $C$ and a number $N$
962: such that ($ 0< \alpha < 2$):
963: \[ {{\int_{D_{\rho}} e^{2\varphi_n} d\,xd\,y} \over{\rho^{\alpha}}} = {{A_{n}(\rho)}\over{\rho^{\alpha}}} < C,
964: \label{lem:no-coni} \]
965: if $n$ is large enough.
966: \end{lem}
967:
968: \noindent {\bf Proof. }Choose a small coordinate disk
969: $D_{r}(0)$ so that: $ 2 \cdot C_2 \cdot A_n(r) < \pi^2 ( 2 -\alpha)^2 $
970: (since $A_p = 0$). Let $ C $ be any number large enough such that:
971: \[ {{A_{n}(r) }\over{r}^{\alpha}} < C , \;\forall \; n \in
972: {\bf N}.\]
973:
974: It is claimed that this lemma holds true for this constant $C.\;$
975: Otherwise, there exists a number $ \rho_{n} < r$, such that
976: $ A_{n}(\rho_{n}) - C \cdot \rho_{n}^{\alpha} > 0 $.
977: Consider the function $ F_{n} (\rho) = A_{n}(\rho) - C \cdot \rho^{\alpha}.\;$
978: We have $ F_n(r) < 0 < F_n(\rho_{n}). $
979: There then exists an interior point $r_{n} \in (\rho_{n},r_{0})$ such that:
980: \[ F_{n}(r_{n}) = 0, \qquad F_{n}'(r_{n}) < 0, \]
981: or,
982: \[ A_n(r_n) = C \cdot r_n^{\alpha},\qquad
983: \int_{0}^{2\pi} e^{2\varphi_{n}} r_{n} d \theta < C \cdot \alpha
984: \cdot r_{n} ^{\alpha -1}. \]
985:
986: Using a Schwartz type inequality, we have:
987:
988: \[
989: ( \int_{0}^{2\pi} e^{ \varphi_{n}} r_{n} d \theta)^{2} <\int_{0}^{2\pi} e^{2\varphi_{n}} r_{n} d \theta \cdot \int_0^{2\pi} \,r_n\,d\theta
990: < 2 \pi \cdot C\cdot \alpha \cdot \; r_{n} ^{\alpha }.
991: \]
992:
993: In other words, $\; |\partial D_{r_{n}}|_{g_{n}} ^{2} < 2 \pi \alpha \cdot
994: C \cdot r_{n}^{\alpha}.\;$ Therefore,
995: \begin{eqnarray*} \int_{D_{r_{n}}} |K_{g_{n}}| d\;g_{n} & > &
996: 2 \pi - {{|\partial D_{r_{n}}|_{g_{n}} ^{2}}\over{ 2 A_{r_{n}}}} \\ & > &
997: 2\pi - {{ 2\pi \alpha \cdot C \cdot r_{n}^{\alpha}} \over { 2 C \cdot r_{n}^{\alpha}}} \\
998: & = & \pi ( 2 - \alpha) > 0.
999: \end{eqnarray*}
1000:
1001: Using a Schwartz type inequality again, we have:
1002:
1003: \[ \int_{D_{r_{n}}} K_{g_{n}}^{2} d\,g_n >
1004: {{( \pi( 2 -\alpha))^{2}}\over{ A_{n}(r_{n})}} \geq {{( \pi( 2 -\alpha))^{2}}\over{ A_{n}(r)}}. \]
1005:
1006: Thus,
1007: \[C_2 \geq \int_{D_{r}} K_{g_{n}}^{2} \;d\,g_n \geq \int_{D_{r_{n}}} K_{g_{n}}^{2}
1008: \;d\,g >{{ \pi^{2} (2 -\alpha)^2 }\over{ A_{n}(r)}} > 2 \cdot C_2, \]
1009: which is a contradiction. The lemmas is then proved. QED.
1010:
1011: \begin{lem} Let $\{\varphi_n\}$ be a sequence of metrics with finite area $C_1$
1012: and finite energy $C_2.\;$ Suppose $ \displaystyle { \sup_{k \in {\bf N}} }\;\;
1013: \displaystyle {\max_{ q \in D} } \;\varphi_k ( q) < C_3.\;$
1014: Let $\Omega \subset D$ be any compact sub-domain of $D.\;$
1015: There exists a constant
1016: $\beta \in (0,1)$ which depends only on $C_1, C_2, C_3,$ and the domains
1017: $\Omega, D $ such that:
1018: \[ \displaystyle{\sup_{\Omega}}\; \varphi_k \leq \beta \cdot
1019: \displaystyle{\inf_{\Omega}}\; \varphi_k + C, \qquad \forall\; k \in {\bf N}.\]
1020:
1021: In particular, either $\{\varphi_{k}\} $ vanishes everywhere
1022: on $D$ or there exists a constant $C$ such that:
1023: \[
1024: \displaystyle {\inf_{q \in \Omega}}\,
1025: {\varphi_k(q)} > - C, \qquad \forall \; k \in {\bf N}. \]
1026: \label{lem:inf+sup}
1027: \end{lem}
1028: \noindent {\bf Proof.}
1029: The conditions in this lemma are:
1030: \[\left\{ \begin{array}{lcl} \displaystyle {\sup_{q\in D}}\;\varphi_{n}(q)
1031: & < & C_3,\\ \int_{D} {{(\triangle \varphi_{n})^{2}}\over{e^{2\varphi_{n}}}} d\,x d\,y
1032: & < & C_2, \\
1033: \int_D e^{ 2\varphi_n} dx\,dy & < & C_1\end{array}\right.
1034: \]
1035:
1036: From the first two inequalities, we imply:
1037: \[ \| \triangle \varphi_{n} \|_{L^{2}(D)} =
1038: \int_{D} (\triangle \varphi_n)^2 d\,xd\,y < C. \]
1039:
1040: Decompose the conformal
1041: parameter functions $\varphi_{n}\;$ as $ \;
1042: \varphi_{n} = u_{n} + v_{n},\;$ where $u_{n},v_{n} $ satisfy the following:
1043: \[\left\{ \begin{array}{lcl} \triangle u_{n} & = & \triangle \varphi_{n},\\
1044: u_{n}|_{\partial{D} }& = & 0; \end{array}\right. \]
1045: and
1046: \[
1047: \left\{ \begin{array}{lcl} \triangle v_{n} & = & 0,\\
1048: v_{n}|_{\partial{D} }& = & \varphi_{n}|_{\partial{D}}. \end{array}\right. \]
1049:
1050: Clearly $ \|u_{n}\|_{H^{2,2}(D)} < C.\; $ This implies that
1051: $ \displaystyle{\max_{ p\in D}}\, |u_{n}(p)| < C,\;\; \forall \; n \in
1052: {\bf N}. \;$
1053: Since $\varphi_{n}$ is bounded from above by the initial assumption,
1054: the harmonic functions $v_n = \varphi_n - u_n $
1055: is bounded from above. For any sub-domain $\Omega \subset D,$ there
1056: exists a constant $\beta \in (0,1)$ such that
1057: $ \displaystyle {\sup_{\Omega}}\,(C -{v}_{n}) \leq {1\over \beta} \cdot
1058: \displaystyle {\inf_{\Omega}}\; ( C-{v}_{n}).\;$ Thus,
1059: \[
1060: \displaystyle {\sup_{\Omega}}\,\varphi_{n} \leq \beta \cdot
1061: \displaystyle {\inf_{\Omega}}\; \varphi_{n} + C.\]
1062: QED.\\
1063:
1064:
1065:
1066: \subsection{Locally weakly convergence}
1067: \begin{prop} Let $\{\varphi_k,k \in {\bf N}\}$ be a sequence of metrics
1068: in $D$ with finite area $C_1$ and energy $C_2.\;$
1069: There exists at most a finite number of bubble points
1070: ( bounded by $ \sqrt{\frac{C_1 \cdot C_2}{4\pi^2}}$) in $D$
1071: for any subsequence of metrics
1072: of $\{\varphi_k \}.\;$
1073: Moreover, there exists a subsequence of $\{\varphi_k\}$
1074: which has a finite number of bubble points and has no additional
1075: pseudo bubble points in $D$.
1076: \label{prop:finite-bubble}
1077: \end{prop}
1078: {\bf Proof}: We first prove that there exists at most
1079: a finite number of bubble points
1080: for any sequence of metrics which satisfies
1081: inequality~\ref{eq:lo:bound} uniformly.
1082: Suppose that $p_{1}, p_{2},\cdots,p_{k}\;$ are all of the bubble points.
1083: On one hand, we have:
1084: \[ \sum_{i=1}^{k} A(p_{i} ) \leq \int_{D} e^{2 \varphi_n} d\,x d\,y \leq C_1. \]
1085:
1086: On the other hand, lemma~\ref{lem:bubble} implies:
1087: \[ K(p_{i}) \geq {{4\pi^{2}}\over{A(p_{i})}}, \qquad \forall \; i = 1,2,\cdots, k. \]
1088:
1089: The total concentrated energy of this sequence of metrics at these bubble points
1090: must be less than the total amount of energy of this sequence of metrics. Thus,
1091: \[ \begin{array} {ccc} C_2 & \geq &
1092: \int_D {{(\triangle \varphi_n)^2}\over e^{2 \varphi_n}} d\,x d\,y
1093: \\ & \geq & \sum_{i=1}^{k} K(p_{i})
1094: \\ & \geq & \sum_{i=1}^{k} {{4\pi^{2}}\over{A(p_{i})}} \\
1095: &\geq & {{ 4 (k \pi)^{2}}\over{\sum_{i=1}^{k} A(p_{i})}} \geq
1096: {{ 4 (k \pi)^{2}}\over{C_1}} .\end{array} \]
1097:
1098: Therefore, $ k \leq \sqrt{ {{C_1 \cdot C_2 }\over{ 4 \pi^{2}}}}.\;$ \\
1099:
1100: Suppose the original sequence of
1101: metrics has $l$ distinct bubble points and has at least one additional
1102: pseudo point $p.\;$ Passing to an appropriate subsequence,
1103: (by proposition~\ref{prop:psedo-bubble}), $p$ is then a bubble point
1104: for this subsequence. This subsequence then has $(l+1)$ distinctive bubble
1105: points. It is claimed that a subsequence of $\{\varphi_n\}$
1106: can be selected so that it has only a finite number of bubble points
1107: and it has no additional pseudo bubble points. Otherwise, we can
1108: keep passing to an appropriate
1109: subsequence to convert any additional pseudo bubble point into a new
1110: bubble point. Eventually, we will obtain
1111: a subsequence of metrics in $D$ which has more than
1112: $\sqrt{ {{C_1 \cdot C_2 }\over{ 4 \pi^{2}}}}$ number of bubble
1113: points. This is a contradiction. The proposition is then proved. QED.\\
1114:
1115: {\bf Proof of Theorem~\ref{theo:lo:weak}.}
1116: Passing to a subsequence of $\{\varphi_n\}$ if necessary,
1117: so that $\{\varphi_n\}$ has exactly $m (\geq 0)$
1118: number of bubble points and has no additional pseudo bubble points.
1119: Denote these bubble points by $\{p_1,p_2,\cdots,p_m\}.\;$
1120: Choose two compact sub-domains $D^1 $ and $D^2$ so that:
1121: \[ \{p_1,p_2,\cdots,p_m\} \subset D^1 \subset D^2 \subset \subset D.\]
1122:
1123: Let $\epsilon > 0 $ be small enough so that
1124: $\{ D_{\epsilon}(p_s), 1\leq s \leq m\}$ are disjoint
1125: disks in $D_1.\;$ Let $D_{i,j}$ denote the following domains (see Figure~\ref{fg:domain} below):
1126: \[ D_{i,j} = D^i \setminus (\displaystyle{\bigcup_{1\leq s \leq m}}
1127: D_{{1\over 2^j} \epsilon} (p_s)),\qquad i = 1,2,\; \forall \,j \in {\bf N}. \]
1128:
1129: \begin{figure}
1130: \centerline{\psfig{figure=new5.eps}}
1131: \caption{Compact sub-domains}
1132: \label{fg:domain}
1133: \end{figure}
1134:
1135: Clearly, $D_{1,j}$ is a compact sub-domain of $D_{2,j+1}.\;$ Fixing
1136: a number $j,$ there exists
1137: a constant $c_j$ independent of $\{\varphi_n\}$ such that:
1138: \begin{equation} \varphi_n(p) \leq c_j, \qquad \forall p \in D_{2,j+1},\;\forall n \in {\bf N}.
1139: \label{eq:djbound} \end{equation}
1140:
1141: If not, there exists a sequence of points $q_k \in D_{2,j+1} $
1142: such that:
1143: \begin{equation} \displaystyle {\lim_{k\rightarrow \infty}} \varphi_k (q_k) = \infty.
1144: \label{eq:lo:unbound}\end{equation}
1145:
1146: Consider a cluster point $q \in \overline{D_{2,j+1}}$ of $\{q_k\}$
1147: such that $ q_k \rightarrow q$ (passing
1148: to a subsequence of $\{q_k\}$ if necessary).
1149: According to the initial assumption, $q$ is not
1150: a pseudo bubble point of $\{ \varphi_n \}.\; $
1151: Lemma~\ref{lem:regular} then implies that there exists a constant $C$ and
1152: an open neighborhood $\cal{O}$ of $p$ such that
1153: $ \displaystyle{\sup_{n} \sup_{q\in \cal {O}}} \,\varphi_n(q) < C.\;$
1154: This contradicts with equation ~(\ref{eq:lo:unbound}).
1155: Therefore, the inequality~(\ref{eq:djbound}) holds true. Thus,
1156: \[ \int_{D_{2,j}} (\triangle \varphi_n)^2 d\,xd\,y \leq e^{2 c_j} \cdot C_2.\]
1157:
1158: According to lemma~\ref{lem:inf+sup},
1159: either $\varphi_n \rightarrow -\infty$ in $D_{2,j+1}$ or
1160: there exists another constant $c_j'$ such that
1161: \[ \varphi_n(p) \geq - c_j', \qquad \forall p \in D_{1,j}, \forall n \in {\bf N}.
1162: \]
1163:
1164: If $\varphi_n \rightarrow -\infty$ in $D_{1,j},$ define
1165: $\varphi_{0,j} \equiv -\infty.\;$ If $\varphi_n \not \rightarrow
1166: -\infty $ in $D_{1,j},$ then $\varphi_n$ are uniformly bounded in
1167: $H^{2,2}(D_{1,j}).\;$ There then exists a function $\varphi_{0,j}
1168: \in H^{2,2}(D_{1,j})$ such that $ \varphi_n \rightharpoonup \varphi_{0,j}
1169: $ in $ H^{2,2}_{loc}(D_{1,j}).\;$ Thus, in either case, we have:
1170: \[ \varphi_n \rightharpoonup \varphi_{0,j} \;{\rm in}\; {\hat{H}}^{2,2}_{loc}(D_{1,j}).\]
1171:
1172: Define $\{\varphi_{0,j}, j\in {\bf N}\}$ successively
1173: in $D_{1,j}$ for $j=1,2,\cdots $ such that:
1174: \[ \varphi_{n,j} \rightharpoonup \varphi_{0,j}
1175: \;{\rm in}\; {\hat{H}}^{2,2}_{loc}(D_{1,j}), \]
1176: where $\{\varphi_{n,j}\}(j > 1)$ is a subsequence of $\{\varphi_{n,j-1}\}.\;$
1177: Consider the diagonal subsequence $\{\varphi_{n,n}\}.\;$
1178: For any fixed $j>0,$ we have:
1179: \[
1180: \varphi_{n,n} \rightharpoonup \varphi_{0,j} \;{\rm in}\; {\hat{H}}^{2,2}_{loc}(D_{1,j}).
1181: \]
1182:
1183: Clearly, for any $ i > j,$ we have $ \varphi_{0,i} \equiv \varphi_{0,j}$
1184: in $D_{1,j}.\;$ In particularly, $\varphi_{0,i} \equiv -\infty
1185: $ if and only if $\varphi_{0,j} \equiv -\infty$ (lemma~\ref{lem:inf+sup}).
1186: Thus, $\{\varphi_{0,j},j\in {\bf N}\}$ defines a metric
1187: $\varphi_0 $ in ${\hat{H}}^{2,2}(D_1 \setminus \{p_1,p_2,\cdots,p_m\})$
1188: by
1189: \[ \varphi_0(p) = \varphi_{0,j}(p),\qquad \forall\; p \in D_1 \setminus \{p_1,p_2,\cdots,p_m\}. \]
1190:
1191: Therefore,
1192: \[ \varphi_{n,n} \rightharpoonup \varphi_0 \;{\rm in}\; {\hat{H}}^{2,2}_{loc}( D_1 \setminus \{p_1,p_2,\cdots,p_m\} ).\]
1193:
1194: Denote $\{\varphi_{n,n}\}$ by $\{\varphi_n\}$ abd define
1195: \[
1196: A_s(j) = \displaystyle {\lim_{n \rightarrow \infty }}
1197: A_c(\varphi_n, D_{ {1\over{ 2^j}}\epsilon}(p_s)),\qquad K_s(j) =
1198: \displaystyle {\lim_{n \rightarrow \infty}}
1199: K(\varphi_n, D_{ {1\over{ 2^j}}\epsilon}(p_s)),\qquad \forall\, 1 \leq s \leq m.
1200: \]
1201:
1202: Then,
1203: \[ A_{p_s} = \displaystyle {\lim_{j \rightarrow \infty}} A_s(j), \qquad
1204: K_{p_s} = \displaystyle {\lim_{j \rightarrow \infty}} K_s(j).\]
1205:
1206: In $D_{1,j}$, we have:
1207: \[
1208: \displaystyle {\lim_{n \rightarrow \infty }} A_c(\varphi_n, D_{1,j}) = A_c(\varphi_0,D_{1,j}), \qquad \displaystyle {\lim_{n \rightarrow \infty }} K_c(\varphi_n, D_{1,j}) \geq K_c(\varphi_0,D_{1,j}).
1209: \]
1210:
1211: Therefore,
1212: \[
1213: \begin{array}{rcc}
1214: \displaystyle {\lim_{n \rightarrow \infty }} A_c(\varphi_n,D_1) & = & A_c(\varphi_0,D_{1,j})
1215: + \sum_{s=1}^{m} A_s(j), \\ \displaystyle {\lim_{n \rightarrow \infty }}
1216: K_c(\varphi_n,D_1) & \geq & K_c(\varphi_0,D_{1,j})
1217: + \sum_{s=1}^{m} K_s(j). \end{array}
1218: \]
1219:
1220: Taking limit as $ j \rightarrow \infty,$ we have
1221: \[
1222: \begin{array}{ccc}
1223: \displaystyle {\lim_{n \rightarrow \infty }} A_c(\varphi_n,D_1) & = & A_c(\varphi_0,D_1\setminus \{p_1,p_2,\cdots,p_m\})
1224: + \sum_{s=1}^{m} A_{p_s}, \\ \displaystyle {\lim_{n \rightarrow \infty }} K_c(\varphi_n,D_1\setminus \{p_1,p_2,\cdots,p_m\}) & \geq & K_c(\varphi_0,D_1\setminus \{p_1,p_2,\cdots,p_m\}) + \sum_{s=1}^{m} K_{p_s}. \end{array}
1225: \]
1226: Let $D_1 $ and $D_2$ approach $D$, and use
1227: a similar diagonalize argument, we can show that the theorem holds
1228: true. QED. \\
1229:
1230: \subsection{Limit of a weak convergence sequence}
1231:
1232: {\bf Proof of theorem~\ref{theo:lo:prop}.} We will prove 2.1, 2.2 and 2.3
1233: separately.\\
1234:
1235: (2.1). Let $ u = - \ln r = -\ln {\sqrt{x^2 +y^2}} $ and $ \theta = tan^{-1} {y \over x}.\;$
1236: The domain $D\setminus \{0\}$ becomes an infinite cylinder $\{ (u, \theta) |
1237: 0\leq u\leq \infty, -\pi \leq \theta \leq \pi \}$ via
1238: this transformation. Let
1239: $\psi (u,\theta) = \varphi ( e^{-u} \cos \theta, e^{-u} \sin \theta) - u.\;$
1240: Then $ \psi $ satisfies the following inequalities:
1241: \begin{equation}
1242: \left\{\begin{array}{ccc} \int_{0}^{\infty} \int_{-\pi}^{\pi}
1243: {(\triangle_{u,\theta} \psi)^2 \over e^{2 \psi}} d\,\theta d\,u & \leq & C_2,
1244: \\ \int_{0}^{\infty} \int_{-\pi}^{\pi}
1245: e^{2 \psi} d\,\theta d\,u \leq & C_1, \end{array} \right.
1246: \label{eq:lo:psi}
1247: \end{equation}
1248: where $ \triangle_{u,\theta} = {{\partial^2}\over{\partial u^2}} + {{\partial^2}\over{\partial \theta^2}}.\;$ To prove theorem 2.1, we only need to show that
1249: $\psi \rightarrow - \infty$ as $ u\rightarrow \infty.\;$ If this
1250: is not true, there then exists a positive number $C$ and a sequence
1251: of points $\{(u_i, \theta_i), i \in {\bf N} \} (u_i \rightarrow \infty)$
1252: such that: $\psi(u_i, \theta_i) > -C.\; $ Consider the open disk $\tilde{D} = \{ (u,\theta) | -1 < u < 1, - {\pi\over 2} < \theta < {\pi\over 2}\}.\;$ Define a new sequence of metrics in $\tilde{D}$ as:
1253: \[ \varphi_i(u, \theta) = \psi(u+u_i, \theta+ \theta_i), \qquad \forall i \in {\bf N},\; \forall (u,\theta) \in \tilde{D}.\]
1254:
1255: Then $\{ \varphi_i(u,\theta), i \in{\bf N} \} $ is a sequence of functions
1256: in $\tilde{D}$ with finite energy and area. According to theorem~\ref{theo:lo:weak}, there
1257: exists a subsequence $\{\varphi_{n_j}, j\in {\bf N}\}$ of $\{\varphi_n\}$
1258: and a metric $\varphi_0 \in {\hat{H}}^{2,2}_{loc}(\tilde{D}\setminus\{q_1,q_2,\cdots,q_l\})$ for some isolated singular points $\{q_1,q_2,\cdots,q_l\}$
1259: such that:
1260: \[ \varphi_{n_j} \rightharpoonup \varphi_0\; {\rm in}\; {\hat{H}}^{2,2}_{loc}(\tilde{D}\setminus\{q_1,q_2,\cdots,q_l\}),\qquad l \geq 0. \]
1261:
1262: The vanishing case ($\varphi_0 \equiv -\infty$) does not
1263: occur because of
1264: \[ \varphi_{n_j}(0,0) = \psi(u_{n_j},\theta_{n_j}) > - C, \qquad \forall j \in {\bf N}.\]
1265:
1266: If there exists at least one bubble point $p \in \tilde{D}$, we have:
1267: \begin{equation} \int_{\tilde{D}} {(\triangle_{u,\theta} \varphi_{n_j})^2 \over { e^{2\varphi_{n_j}}}} d\,u d\, \theta
1268: \cdot \int_{\tilde{D}} e^{2\varphi_{n_j}} d\,u d\, \theta > {1\over 2} E_p \cdot A_p \geq 2 \pi^2.\label{eq:polar1}
1269: \end{equation}
1270:
1271: If there exists no bubble point, then $\varphi_0 \in H^{2,2}(\tilde{D})
1272: $ and $ \varphi_{n_j} \rightharpoonup \varphi_0$
1273: in $H^{2,2}(\tilde{D}).\;$ If $n $ is large enough, then:
1274: \begin{equation}
1275: \int_{\tilde{D}} {(\triangle_{u,\theta} \varphi_{n_j})^2 \over { e^{2\varphi_{n_j}}}} d\,u d\, \theta
1276: \cdot \int_{\tilde{D}} e^{2\varphi_{n_j}} d\,u d\, \theta > {1\over 2} \int_{\tilde{D}} {(\triangle_{u,\theta} \varphi_0)^2 \over { e^{2\varphi_0}}} d\,u d\, \theta
1277: \cdot \int_{\tilde{D}} e^{2\varphi_0} d\,u d\, \theta > 0.
1278: \label{eq:polar2}
1279: \end{equation}
1280:
1281: However,
1282: \[
1283: \int_{\tilde{D}} {(\triangle_{u,\theta} \varphi_{n_j})^2 \over { e^{2\varphi_{n_j}}}}
1284: d\,u d\, \theta \cdot \int_{\tilde{D}} e^{2\varphi_{n_j}} d\,u d\, \theta =
1285: \int_{u_{n_j} -1}^{u_{n_j}+1} \int_{-\pi \over 2}^{\pi\over 2}
1286: {(\triangle_{u,\theta} \psi)^2 \over { e^{2 \psi}}} d\,u d\, \theta \cdot
1287: \int_{u_{n_j}-1}^{u_{n_j}+1}\int_{-{\pi \over 2}}^{\pi\over 2} e^{2\psi} d\,u d\, \theta
1288: \]
1289: \[ \qquad \qquad \rightarrow 0, \qquad {\rm as } \; j \rightarrow \infty.
1290: \]
1291:
1292: The last formula holds true because of inequality~(\ref{eq:lo:psi}). This
1293: contradicts both inequalities~(\ref{eq:polar1}) and~(\ref{eq:polar2}). The
1294: first part of the theorem is then proved. \\
1295:
1296: (2.2). If $\displaystyle {\lim_{n \rightarrow \infty}}{ 1 \over 2 \pi}
1297: \int_{0}^{2\pi} \varphi_r'(r \cos \theta, r \sin \theta)\, r\, d\,\theta $
1298: does not exist, there then exist two numbers $\alpha \neq \alpha'$
1299: and two alternative sequence of numbers $\{\delta_i\}, \{\delta_i'\}$
1300: such that:
1301: \[ \delta_i < \delta_i' < \delta_{i-1} \rightarrow 0, \qquad \forall i \in {\bf N},\]
1302: and
1303: \[ { 1 \over 2 \pi}
1304: \int_{0}^{2\pi} \varphi_r'(\delta_i \cos \theta, \delta_i \sin \theta) \delta_i d\,\theta
1305: \rightarrow \alpha,\qquad { 1 \over 2 \pi}
1306: \int_{0}^{2\pi} \varphi_r'(\delta_i' \cos \theta, \delta_i' \sin \theta) \delta_i' d\,\theta
1307: \rightarrow \alpha'. \]
1308: Clearly,
1309: \[ A_c(\delta_i,\delta_i') = \int_{\delta_i}^{\delta_i'} \int_{0}^{2 \pi} e^{ 2 \varphi} r d\,\theta d\,r \rightarrow 0.\]
1310: However,
1311: \[
1312: \begin{array}{ccl} |\int_{\delta_i}^{\delta_i'} \int_{0}^{2 \pi} \triangle \varphi \cdot r \cdot d \, \theta d\,r | & = & |\int_{\delta_i}^{\delta_i'} \int_{0}^{2 \pi}( \varphi_r' \cdot r )'_r d\,\theta d\, r| \\
1313: & = & |\int_{0}^{2\pi} \varphi_r'(\delta_i \cos \theta, \delta_i \sin \theta) \delta_i d\,\theta - \int_{0}^{2\pi} \varphi_r'(\delta_i' \cos \theta, \delta_i' \sin \theta) \delta_i' d\,\theta | \\
1314: & \rightarrow & | \alpha - \alpha'| > 0. \end{array}
1315: \]
1316:
1317: On the other hand,
1318: \[
1319: \begin{array}{ccc} C_2 \geq K_c(\delta_i, \delta_i') & =& \int_{\delta_i}^{\delta_i'} \int_{0}^{2 \pi} {{(\triangle \varphi)^2}\over{ e^{ 2\varphi}}} r d\theta d\, r \\
1320: & \geq & {{|\int_{\delta_i}^{\delta_i'} \int_{0}^{2 \pi} \triangle \varphi \cdot r \cdot d \, \theta d\,r |^2 }\over A_c(\delta_i, \delta_i') } \rightarrow
1321: {{| \alpha - \alpha'|}\over {0}} = \infty. \end{array} \]
1322:
1323: This is a contradiction. Therefore, $\displaystyle {\lim_{n \rightarrow \infty}}{ 1 \over 2 \pi}
1324: \int_{0}^{2\pi} \varphi_r'(r \cos \theta, r \sin \theta)\, r\, d\,\theta $ does
1325: exist. \\
1326:
1327: (2.3). For any small $r = e^{-u} > 0, $ consider the domain $\tilde{D} = [-1,1] \times S^1.\;$ Let $\tilde{\varphi}(v,\theta) = \psi(v + u, \theta) $ (following the notations in (2.1)), then
1328: \[
1329: - \triangle_{v,\theta}\; \tilde{\varphi} = K(v+u,\theta) \cdot e^{2 \tilde{\varphi}},\qquad \forall\, (v,\theta)\, \in \tilde{D}.
1330: \]
1331:
1332: There then exists a constant $C$ such that:
1333: $\tilde{\varphi} \leq C.\; $ The right hand side is bounded in
1334: $L^2(\tilde{D}).\;$ Let $w$ be the solution of
1335: \[
1336: \left\{
1337: \begin{array}{ccl} - \triangle w & = & K(v+u,\theta) e^{2 \tilde{\varphi}},\\
1338: w|_{\partial D} & = & 0.\end{array} \right.
1339: \]
1340:
1341: Thus, $ ||w||_{L^{\infty}} $ is uniformly bounded from above
1342: (the bound is actually independent of $u$, since $L^2 $ norm of
1343: $ \triangle_{v,\theta}\; \tilde{\varphi} $ in $\tilde{D}$
1344: uniformly converge to $0$ as $u \rightarrow \infty$).
1345: The harmonic function $h = \tilde{\varphi} - w$ is bounded
1346: from below by a constant $ - C.\;$ This follows that there exists a
1347: constant $\beta \in (0,1)$ (independent of $u$) such that:
1348: \[ \displaystyle{\sup_{\theta}}\,(C - h(0,\theta)) \leq {1\over \beta}
1349: \displaystyle{\inf_{\theta}}\,(C - h(0,\theta)),\]
1350: Or,
1351: \[
1352: \displaystyle{\sup_{\theta}}\,\tilde{\varphi}(0,\theta) \leq {1\over \beta}
1353: \displaystyle{\inf_{\theta}}\,\tilde{\varphi}(0,\theta) + C.\]
1354:
1355: In other words,
1356: \[
1357: \displaystyle{\sup_{\theta}}\,(\tilde{\varphi}(r \cos \theta, r \sin \theta)+ \ln r)
1358: \leq {1 \over \beta} \;
1359: \displaystyle{\inf_{\theta}}\,(\tilde{\varphi}(r \cos \theta, r\sin \theta)+ \ln r ) + C.
1360: \]
1361:
1362: Integrating on both sides over $\theta$, we obtain:
1363: \[
1364: {1\over \beta} (\phi(r) + \ln r) +C_3 \leq \varphi(r \cos \theta, r \sin \theta) +\ln r
1365: \leq \beta (\phi(r) + \ln r)+ C_4,
1366: \]
1367: where $C_3, C_4$ are two constants independent of $r$. QED.\\
1368:
1369: \subsection{Bubbles on bubbles}
1370:
1371: \begin{lem} Suppose $D$ is a coordinate disk with radius $\rho >0 $ and
1372: assume \\
1373: $g = e^{2\varphi}(d\,x^{2} + d\,y^{2})$ is a metric on $D $
1374: with finite energy $ C_2.\;$ For any $ \epsilon > 0,$
1375: there then exists a constant $C_{\epsilon} > 0,$ which depends only on
1376: $D, \epsilon $ such that if \\
1377: $\displaystyle{\max_{r \leq \rho}}\;\int_{0}^{2\pi} e^{\varphi(r \cos \theta,r \sin \theta)} r d\,\theta
1378: < \epsilon, $ then the following holds true:
1379: \[ \int_{D} e^{2 \varphi} d\,x d\,y \leq C_{\epsilon},
1380: \qquad {\rm and}\;\;\displaystyle \;{\lim_{\epsilon \rightarrow 0}}\; C_{\epsilon} = 0.\]
1381: \end{lem}
1382:
1383: \noindent {\bf Proof.} If the lemma is false, then there exists a sequence of metrics
1384: $\{\varphi_{n},\;n \in {\bf N} \} $ such that:
1385: \begin{equation} \left\{\begin{array}{rcl} \int\limits_{D}
1386: {{(\triangle \varphi_n)^2} \over{e^{2 \varphi_n}}} d\,x d\,y & < & C,
1387: \\ \int_{D} e^{2 \varphi_n} d\,x d\,y & = & 1, \\
1388: \displaystyle {\max_{ r \leq \rho}}\;\int_{0}^{2\pi}
1389: e^{\varphi_n(r \cos \theta,r \sin \theta)} r d\,\theta
1390: & = & \epsilon_{n} \rightarrow 0. \end{array}\right.
1391: \label{eq:thin}
1392: \end{equation}
1393:
1394: Any circle ($ |z| = \delta > 0 $) must have a zero length in the limit.
1395: Therefore, $\varphi_{n}$ vanishes identically except at the origin ($z=0$). All of
1396: the area concentrates at the origin since the total area is fixed. Let $\varepsilon > 0 $
1397: be a very small positive number and let $ \{\delta_{n}\}$ be
1398: a sequence of numbers such that:
1399: \[ \int_{ r \leq \delta_{n} } e^{2 \varphi_{n}} d\,x\, d\,y
1400: = \varepsilon < {2\pi^2 \over C_2}. \]
1401:
1402: Define a new sequence of metrics $ F_{n} = e^{ 2 w_{n}}(d\,x^{2} + d\,y^{2})$ as:
1403: \[ w_{n} (z) = \varphi_{n} (\delta_{n} \cdot z) + \ln \delta_{n}, \;\forall r \leq 1.\]
1404:
1405: For this new sequence of metrics, we have
1406: \[\begin{array}{ccccl} \int_{D_1(0)} e^{ 2 w_{n}} & = &\int_{ r\leq \delta_{n}}\,e^{ 2 \varphi_{n}}& = & \epsilon, \\
1407: \int_{D_1(0)} {{(\triangle w_{n})^2} \over {e^{ 2 w_{n}}}} d\,x d\,y & = &
1408: \int_{ r\leq \delta_{n}} {{(\triangle \varphi_{n})^2}
1409: \over {e^{ 2 \varphi_{n}}}} d\,x d\,y & \leq & C_2.
1410: \end{array}\]
1411:
1412: By theorem~\ref{theo:lo:weak}, there exists a subsequence which locally
1413: weak converges to a metric except at a set of finite number of
1414: bubble points. Since each circle $|z| =\delta > 0$ has length $0$,
1415: the only bubble point must be the original point $z = 0$ and all
1416: the area concentrated at $0$ must be less than $\varepsilon.$
1417: Lemma~\ref{lem:bubble} implies that
1418: total energy concentration
1419: at the origin must be bigger than ${4\pi^2 \over \varepsilon} \geq 2\cdot C_2,$
1420: which is a contradiction. This lemma is then proved.
1421: QED.\\
1422:
1423: \begin{cor} Suppose $\{\varphi_n\}$ is a sequence of metrics with finite area $C_1$
1424: and energy $C_2.\;$ There exists a constant $\epsilon_{0}$ such that if
1425: \[\displaystyle {\max_{r\leq \rho} }\, |\partial D_r|_{g_n} =
1426: \displaystyle {\max_{r \leq \rho}}\,\int_0^{\pi}
1427: e^{\varphi_{n}(r\cos\theta,r\sin\theta)} \cdot r\, d\,\theta
1428: \leq \epsilon_{0},\qquad \forall n \in {\bf N}, \]
1429:
1430: then the sequence of metrics does not have any bubble points in $D$.
1431: \label{cor:blow}
1432: \end{cor}
1433: {\bf Proof}: Let $C_{\epsilon}$ be the constant defined according to the previous
1434: lemma. We may choose $\epsilon_0$ so small that $C_{\epsilon_0}$ satisfies:
1435: \[ C_{\epsilon} \cdot C_2 \leq 2 \pi^2.\]
1436:
1437: According to the previous lemma, we have:
1438: \[
1439: A_c(\varphi_n, D) \leq C_{\epsilon_0} \leq { 2\pi^2 \over C_2}.
1440: \]
1441:
1442: Thus the number of possible points $m$ must be bounded by
1443: \[m \leq \sqrt{ {C_{\epsilon_0} \cdot C_2 \over {4\pi^2} }} < 1.
1444: \]
1445:
1446: Therefore, this sequence of metrics has no bubble points.
1447: The corollary is then proved. QED.\\
1448:
1449: \noindent {\bf Proof of theorem \ref{theo:lo:bubble}.}
1450: Choose any small positive number $\epsilon_0 \in (0, \varepsilon).\;$ This number
1451: $\varepsilon\; $ serves as a scaling constant (filter).
1452: The sequence of functions can be modified slightly so that
1453: the following holds true:
1454: \begin{equation}
1455: \varphi_{n}(p) = \displaystyle {\max_{q \in D_{r_0}}} \; \varphi_{n} (q).
1456: \label{eq:blow0}
1457: \end{equation}
1458:
1459: Following theorem~\ref{theo:lo:prop}, in a non-vanishing case, we have
1460: $ \displaystyle {\lim_{ r \rightarrow 0 }}\; \displaystyle
1461: {\max_{0\leq \theta \leq 2\pi}}({\varphi_0}(r \cos \theta,r \sin \theta) + \ln r )
1462: = - \infty.\;$ There then exists a number $ r_{1} > 0 $ such that:
1463: \[ \displaystyle {\max_{ 0\leq \theta \leq 2\pi}} ({\varphi_0}
1464: (r\cos \theta, r \sin \theta) + \ln r) \ll \varepsilon, \;\;\forall\; r < r_{1}. \]
1465:
1466: If $ n $ is large enough,
1467: \begin{equation} \displaystyle{\max_{0\leq\theta\leq 2\pi}}
1468: ( {\varphi_{n}}(r_{1} \cos \theta,r_1 \sin \theta) + \ln r_{1})
1469: \ll \varepsilon, \qquad \forall \;n > N,
1470: \label{eq:blow1}
1471: \end{equation}
1472:
1473: or the length of the circle $ |z| = r_1$ is very small:
1474: \begin{equation}
1475: |\partial D_{r_1} |_{g_n} =
1476: \int_0^{2 \pi} e^{\varphi_n(r_1\cos \theta, r_1 \sin \theta)}\, r_1\,
1477: d\,\theta \ll \varepsilon, \qquad n > {\bf N}.
1478: \label{eq:blow2}
1479: \end{equation}
1480:
1481: According to corollary~\ref{cor:blow}, if $\varepsilon$ is small enough,
1482: we can choose $\delta_{n}$ such that:
1483: \begin{equation}
1484: |\partial D_{r} |_{g_n} =
1485: \int_0^{2 \pi} e^{\varphi_n(r \cos \theta, r \sin \theta)} r\, d\,\theta
1486: < \varepsilon,\qquad \forall\; r_{1} \geq r \geq {\delta_{n}},
1487: \label{eq:blow3} \end{equation}
1488: and
1489: \begin{equation}
1490: |\partial D_{\delta_n} |_{g_n} =
1491: \int_0^{2 \pi} e^{\varphi_n(\delta_n \cos \theta, \delta_n\sin \theta)}
1492: \delta_n \,d\,\theta = \varepsilon.
1493: \label{eq:blow4}
1494: \end{equation}
1495:
1496: Re-normalize this sequence of metrics as:
1497: \begin{equation} \phi_{n} (z) =
1498: \varphi_{n}(\delta_{n} \cdot z ) + \ln \delta_{n}, \qquad \forall\, |z| < {1\over \delta_n}.
1499: \label{eq:blow5}
1500: \end{equation}
1501:
1502: For any $n > 0,\; \phi_{n} $ is then defined in the disk $D_{\delta_{n}^{-1}}(0)$.
1503: For any fixed number $r > 0,\; \phi_{n} $ is well defined on $D_{r}(0)$
1504: if $n$ is large enough. Moreover, $\{\phi_{n}\}$ has a finite amount of
1505: energy and area since
1506: \begin{equation}
1507: \left\{ \begin{array} {lcl} \int\limits_{D_{\delta_{n}^{-1}}}
1508: {(\triangle \phi_n)^2 \over{e^{2\phi_n}} } d\,x d\,y
1509: & = & \int\limits_{D_{1}}{(\triangle \varphi_n)^2 \over{e^{2\varphi_n}} }
1510: d\,x d\,y \leq C_2,\\
1511: \int\limits_{D_{\delta_{n}^{-1}}} e^{2 \phi_n} d\,x d\,y & = &
1512: \int\limits_{D_{1}} e^{ 2\varphi_n} d\, d\,y \leq C_1.
1513: \end{array}\right.
1514: \label{eq:blow6}
1515: \end{equation}
1516:
1517: Applying theorem~\ref{theo:lo:weak} successively to $\{\phi_n\}$ in a sequence of disks
1518: $D_{2^j}(0) (j = 1,2,\cdots ).\;$ In disk $D_2(0),$ there exists a subsequence
1519: of $\{ \phi_{1n}, n \in {\bf N}\}$ of $\{\phi_n, n \in {\bf N}\},$
1520: a finite number of bubble points $ s_1
1521: = \{p_{11},p_{12},\cdots,p_{1m_1}\} (m_1 \geq 0) $ with respect to
1522: this subsequence, and a metric $\phi_{0,1}
1523: \in {\hat{H}}^{2,2}_{loc} (D_2(0) \setminus \{p_{11},p_{12},\cdots,p_{1m_1}\}) $ such that:
1524: \[
1525: \phi_{1n} \rightharpoonup \phi_{0,1} \; {\rm in}\; {\hat{H}}^{2,2}_{loc}
1526: (D_2(0) \setminus \{p_{11},p_{12},\cdots,p_{1m_1}\}),
1527: \]
1528:
1529: Consider the sequence $\{\varphi_{1n}\}$
1530: in disk $D_{2^2}(0).\;$ There exists a subsequence $\{\phi_{2n}\}$ of $\{\phi_{1n}\}, $
1531: a finite number of bubble points
1532: $ s_2 = \{ p_{21},p_{22},\cdots, p_{2m_2}\} (m_2 \geq 0) $
1533: with respect to this subsequence, and a metric $\phi_{0,2}
1534: \in {\hat{H}}^{2,2}(D_{2^2}(0) \setminus \{p_{21},p_{22},\cdots,p_{2m_2}\}) $ such that:
1535:
1536: \[
1537: \phi_{2n} \rightharpoonup \phi_{0,2} \;{\rm in}\; {\hat{H}}^{2,2}_{loc}
1538: (D_{2^2}(0) \setminus \{p_{21},p_{22},\cdots,p_{2m_1}\}),
1539: \]
1540:
1541: Clearly, the set
1542: $ s_1 = \{p_{11},p_{12},\cdots,p_{1m_1}\} $ is a subset of
1543: $ s_2 = \{p_{21},p_{22},\cdots,p_{2m_1}\}$ and $ m_2 \geq m_1.\;$
1544: Moreover, $\phi_{0,2} = \phi_{0,1}$ when both functions are
1545: restricted to the smaller domain $D_2$. In particular,
1546: $ \phi_{0,1} \equiv -\infty $ if and only if $\phi_{0,2} \equiv -\infty.\; $
1547: \\
1548:
1549: In general, suppose that for any $i \leq j,$ a subsequence $\{\phi_{in}\}$
1550: had been selected, and a limit metric $\{\phi_{0,i}, i\leq j\}$ had been defined
1551: in ${\hat{H}}^{2,2}(D_{2^i}(0)\setminus s_i) $ where $s_i$ is the set of bubble
1552: points of $\{\phi_{in}\}$ in $D_{2^i}(0)$.
1553: Consider the subsequence $\{\phi_{jn}\}$ in $D_{2^{j+1}}(0).\;$
1554: There exists a subsequence $\{\varphi_{(j+1)n}\}$
1555: of $\{\varphi_{jn}\},$ a finite number of bubble points
1556: $s_{j+1} = \{p_{(j+1)1},p_{(j+1)2},\cdots,p_{(j+1)m_{j+1}}\}$
1557: with respect to this subsequence, and a limit metric
1558: $\phi_{0,j+1} \in {\hat{H}}^{2,2}(D_{2^{j+1}}(0) \setminus \{p_{(j+1)1},p_{(j+1)2},\cdots,p_{(j+1)m_2}\}) $ such that:
1559: \begin{equation}
1560: \phi_{(j+1)n} \rightharpoonup \phi_{0,j+1} \; {\rm in} \; {\hat{H}}^{2,2}_{loc}
1561: (D_{2^{j+1}}(0) \setminus \{p_{(j+1)1},p_{(j+1)2},\cdots,p_{(j+1)m_{j+1}}\}).
1562: \label{eq:blow7}
1563: \end{equation}
1564:
1565: Consider the diagonal subsequence $ \{\phi_{nn}\}$. This is a subsequence
1566: of all the previous subsequences $\{ \phi_{jn}, n\in {\bf N}\} $ for
1567: $j =1,2,\cdots.\;$ Therefore, all of the previous
1568: weak convergent results hold true for this subsequence.
1569: In particularly, the following three statements (for any $ j > i \geq 1$)
1570: hold true:
1571: \begin{enumerate}
1572: \item $ s_i \subset s_j.$
1573: \item $ \phi_{0,i} \equiv -\infty $
1574: if and only if $ \phi_{0,j} \equiv -\infty.$
1575: \item $\phi_{0,j} |_{D_{2^i}} = \phi_{0,i} $ if neither of two metrics vanishes.
1576: \end{enumerate}
1577:
1578: Following proposition 2, $m_j \leq \sqrt{\frac{C_1\cdot C_2}{4\pi^2}} (\forall j).\;$
1579: There then exists a number $N$ such that $ s_j = s_N, \forall j > N.\;$
1580: We may assume that set of bubble points is: $s_N = \{q_1, q_2,\cdots,q_m\}
1581: = \bigcup_{j} s_j (m = m_N \geq 0).\;$ \\
1582:
1583: Define a function
1584: $\phi_0 $ in $S^2 \setminus\{\infty,q_1,q_2,\cdots,q_m\}$ by:
1585: \[ \phi_0 (p) = \phi_{0,j}(p), \qquad \forall p \in
1586: D_{2^j}(0) \setminus\{q_1,q_2,\cdots,q_m\}.\]
1587:
1588: Thus, $ \phi_0 \in {\hat{H}}^{2,2}_{loc}(S^2 \setminus \{\infty,p_1,p_2,\cdots,p_m\})$
1589: and the following statement holds true:
1590: \[ \phi_{nn} \rightharpoonup \phi_0 \;{\rm in} \; {\hat{H}}^{2,2}_{loc}
1591: (S^2 \setminus \{\infty,q_1,q_2,\cdots,q_m\}).\]
1592:
1593: For simplicity, we re-label $\{\phi_{nn}\}$ as $\{\phi_n\}.\;$
1594: Let $r$ be any number large enough so that
1595: $\{q_1,q_2,\cdots,q_m\} \subset D_{r}(0).\;$ Consider the sequence of functions
1596: $ \{\phi_n\} $ in $D_{r}(0)$. Suppose the concentrations of area and energy in
1597: the bubble point $q_i$ are $A_i$ and $ K_i.\;$ According to
1598: theorem~\ref{theo:lo:weak}, we have:
1599: \begin{equation}
1600: \displaystyle{\lim_{n\rightarrow \infty}} \int_{D_{r}(0)} e^{ 2 \phi_n} \, d\,x \,d\,y =
1601: \int_{D_{r}(0)} e^{ 2 \phi_0}\,d\,x\, d\,y + \sum_{i=1}^{m} A_i.
1602: \label{eq:blow:bubb1}
1603: \end{equation}
1604:
1605: On the other hand,
1606: \begin{equation}
1607: \int_{D_{r}} e^{ 2 \phi_n} \,d\,x\, d\,y =
1608: \int_{D_{ \delta_n \cdot r}} e^{ 2 \varphi_n} \,d\,x\, d\,y
1609: \label{eq:blow:bubb2}
1610: \end{equation}
1611:
1612: Choose a sequence of numbers $\{ \epsilon_i \searrow 0, i \in {\bf N}\}.\;$
1613: According to the proof of proposition 1, we may have (passing
1614: to a subsequence if necessary):
1615: \[
1616: A_p(\epsilon_i) = \displaystyle {\lim_{n\rightarrow \infty}}
1617: A_c(\varphi_n, D_{\epsilon_i}(p)),\qquad K_p(\epsilon_i) =
1618: \displaystyle {\lim_{n\rightarrow \infty}}
1619: K(\varphi_n, D_{\epsilon_i}(p)), \qquad \forall i \in {\bf N},
1620: \]
1621: and
1622: \[ A_p = \displaystyle {\lim_{n\rightarrow \infty}}
1623: A_c(\epsilon_i),\qquad K_p = \displaystyle {\lim_{n\rightarrow \infty}} K(\epsilon_i).
1624: \]
1625:
1626: For any fixed $i$, then $ \delta_n \cdot r < \epsilon_i $ if
1627: $n$ is large enough. Equation~(\ref{eq:blow:bubb2}) then implies:
1628: \[
1629: A_c(\phi_n, D_{r}(p) ) = A_c(\varphi_n, D_{\delta_n \cdot r}(p))
1630: \leq A_c(\varphi_n,D_{\epsilon_i}(p)).
1631: \]
1632:
1633: Taking the limit on both sides as $ n\rightarrow \infty,$ the result is:
1634: \[
1635: \displaystyle{\lim_{n \rightarrow \infty}} A_c(\phi_n, D_{r}) \leq
1636: A_p(\epsilon_i), \qquad \forall i \in {\bf N}.
1637: \]
1638:
1639: Taking limit on both sides as $ i \rightarrow \infty$,
1640: \[
1641: \displaystyle{\lim_{n \rightarrow \infty}} A_c(\phi_n, D_{r}) \leq A_p.\]
1642:
1643: Similarly,
1644: \[
1645: \displaystyle{\lim_{n \rightarrow \infty}} K_c(\phi_n, D_{r}) \leq K_p.
1646: \]
1647:
1648: This implies that $ m \leq \sqrt{\frac{K_p \cdot A_p} {4 \pi^2}}.\; $
1649: Applying Theorem~\ref{theo:lo:weak} for $\{\phi_n\}$ in $D_r$, we have:
1650: \[
1651: \begin{array}{ccc} \displaystyle{\lim_{n \rightarrow \infty}} A_c(\phi_n, D_{r})& = &
1652: A_c(\phi_0, D_{r} \setminus \{q_1,q_2,\cdots, q_m\}) + \sum_{i=1}^{m} A_{q_i} \\
1653: \displaystyle{\lim_{n \rightarrow \infty}}
1654: K(\phi_n, D_{r}) & \geq & K_c(\phi_0,
1655: D_{r} \setminus \{q_1,q_2,\cdots, q_m\}) + \sum_{i=1}^{m} K_{q_i}.\end{array}
1656: \]
1657:
1658: Thus,
1659: \[
1660: \begin{array}{ccc} A_p & \geq & A_c(\phi_0, D_{r} \setminus \{q_1,q_2,\cdots, q_m\})
1661: + \sum_{i=1}^{m} A_{q_i} \\ K_p & \geq &
1662: K(\phi_0, D_{r} \setminus \{q_1,q_2,\cdots, q_m\}) +
1663: \sum_{i=1}^{m} K_{q_i}.\end{array}
1664: \]
1665:
1666: Let $r \rightarrow \infty,$ then:
1667: \[
1668: \begin{array}{ccc} A_p & \geq & A_c(\phi_0, S^2 \setminus \{\infty, q_1,q_2,\cdots, q_m\})
1669: + \sum_{i=1}^{m} A_{q_i} \\ K_p & \geq &
1670: K(\phi_0, S^2\setminus \{\infty,q_1,q_2,\cdots, q_m\}) + \sum_{i=1}^{m} K_{q_i}.
1671: \end{array}
1672: \]
1673:
1674: If a vanishing case occurs in $D_{2^j}(0)$ for some $j,$
1675: then it occurs in any disk $D_{2^i}(0).\;$ Observe the following
1676: inequalities:
1677: \[
1678: \begin{array}{ccl} 2 \pi \cdot e^{\phi_{n}(0) } & \geq & \int_0^{2 \pi}
1679: e^{\phi_{n}(\cos \theta,\sin \theta) } d\,\theta \\
1680: & = & \int_0^{2 \pi} e^{\varphi_{n}(\delta_{n}\cos \theta,\delta_{n}\sin \theta) }
1681: \delta_{n} \cdot d\,\theta = \varepsilon.\end{array}
1682: \]
1683:
1684: The first inequality holds true because of equation~(\ref{eq:blow0}). The last two
1685: equalities holds true because of equation~(\ref{eq:blow4}) and~(\ref{eq:blow5}).
1686: According to lemma~\ref{lem:regular} ( in a vanishing case),
1687: the following two statements hold true: (1) $p$ is a bubble point of
1688: $\{\phi_{n}\}$; (2) there exists at least one bubble point in the unit circle. Thus,
1689: in a vanishing case, $m \geq 2.\;$ \\
1690:
1691: QED.
1692:
1693: \section{Geometrical Consequence}
1694:
1695: \subsection{Theorem of weak convergence}
1696: A Riemannian metric is said to be a ``limit metric'' if
1697: it is a weak limit of a sequence of Riemannian metrics
1698: in $H^{2,2}(\Omega).\;$ Lemma 6 implies that a limit metric
1699: vanishes at one point if and only if it vanishes everywhere in its domain.
1700: For the convenience of notations, we add the $ ``0'' $ metric into
1701: $H^{2,2}(\Omega)$ and the resulting space is denoted by ${\hat{H}}^{2,2}(\Omega).\; $
1702: Assume the following:
1703: \[ K(0, \Omega) = A(0, \Omega) = 0, \qquad \mbox{ for any sub-domain}\, \Omega.\]
1704: A sequence of Riemannian metrics $\{g_n\} \in H^{2,2}(\Omega)$
1705: weakly converges to a limit
1706: metric $g_0$ in ${\hat{H}}^{2,2}_{loc}(\Omega) $ if and only if one of the following two
1707: alternatives holds true (mutually exclusive):
1708: \begin{enumerate}
1709: \item (Vanishing case). If $g_0 \equiv 0,$ then $g_n \rightarrow 0$
1710: everywhere.
1711: \item (Non-vanishing case). If $g_0 \neq 0, $ then
1712: $ \varphi_n \rightharpoonup \varphi_0 \; {\rm in}\; H^{2,2}_{loc}(\Omega),$
1713: where $ g_n = e^{ 2\varphi_n} g_{bk}, g_0 = e^{ 2 \varphi_0} g_{bk}; $
1714: and $g_{bk}$ is a smooth background metric in $\Omega.\;$
1715: \end{enumerate}
1716: We are now ready to re-state the theorem~\ref{theo:lo:weak} in
1717: a geometric context:\\
1718:
1719: \noindent {\bf Theorem~\ref{theo:lo:weak}$'$.}
1720: {\it Let $\{g_n,\; n\in {\bf N}\}$ be a sequence of metrics
1721: with a finite area $C_1$ and energy $C_2 $ in a coordinate disk $D.\;$
1722: There exists a subsequence $\{g_{n_j},j \in {\bf N}\}$ of $\{g_n\},$
1723: a finite number of bubble points
1724: $\{p_1,p_2,\cdots,p_m\}( 0\leq m \leq \sqrt{\frac{C_1\cdot C_2}{4\pi^2}} )$
1725: with respect to $\{g_{n_j}, j \in {\bf N}\},$ and a limit metric
1726: $g_0$ in ${\hat{H}}^{2,2}(D \setminus \{p_1,p_2,\cdots,p_m) $
1727: such that:
1728: \[
1729: g_{n_j} \rightharpoonup g_0\;{\rm in}\; {\hat{H}}^{2,2}_{loc}
1730: (D \setminus \{p_1,p_2,\cdots,p_m\}).\]
1731:
1732: If the amount of area and energy concentrations of $\{g_{n_j}\}$
1733: at each point $p_i$ are $A_{p_i}$ and $K_{p_i},$ then:
1734: \begin{eqnarray}
1735: \displaystyle{\lim_{j\rightarrow \infty}} A(g_{n_j},D)
1736: & = & A(g_0,D\setminus\{p_1,p_2,\cdots,p_m\} ) + \sum_{i=1}^{m} A_{p_i} \\
1737: \displaystyle{\lim_{j\rightarrow \infty}} K(g_{n_j},D) &
1738: \geq & K(g_0,D\setminus\{p_1,p_2,\cdots,p_m\} ) +
1739: \sum_{i=1}^{m} K_{p_i}. \end{eqnarray}}\\
1740:
1741: \noindent {\bf Proof.} Re-write
1742: the sequence of metrics in a fixed coordinate system as:
1743: \[ g_n = e^{2 \varphi_n} (d\,x^2 + d\,y^2). \]
1744: Thus, $\{\varphi_n, n\in {\bf N}\}$ is a sequence of metrics with finite area
1745: $C_1$ and energy $C_2.\;$ The rest of the proof is a direct translation
1746: of the proof of theorem~\ref{theo:lo:weak}
1747: on p.~\pageref{theo:lo:weak}. QED.
1748:
1749:
1750: \begin{theo} Let $\{g_n,\; n\in {\bf N}\}$ be a sequence of Riemannian metrics
1751: in $M$ ($ M$ is any open surface)
1752: with a finite area $C_1$ and energy $C_2.\;$ There exists a
1753: subsequence of $\{g_n \}, $ a finite number of
1754: bubble points
1755: $\{p_1,p_2,\cdots,p_m\} ( 0\leq m \leq \sqrt{\frac{C_1\cdot C_2} {4 \pi^2}})$
1756: with respect to this subsequence, and a limit metric $g_0$
1757: such that:
1758: \[ g_n \rightharpoonup g_0 \;{\rm in}\;
1759: {\hat{H}}^{2,2}_{loc}(M \setminus \{p_1,p_2,\cdots,p_m\}).
1760: \]
1761: If the amount of area and energy concentrations at each point $p_i$
1762: are $A_{p_i}$ and $K_{p_i}$ , then:
1763:
1764: \begin{eqnarray}
1765: \displaystyle{\lim_{j\rightarrow \infty}} A(g_{n_j},M) &
1766: = & A(g_0,M\setminus\{p_1,p_2,\cdots,p_m\} ) + \sum_{i=1}^{m} A_{p_i} \\
1767: \displaystyle{\lim_{j\rightarrow \infty}} K(g_{n_j},M)
1768: & \geq & K(g_0,M\setminus\{p_1,p_2,\cdots,p_m\} ) + \sum_{i=1}^{m} K_{p_i}.
1769: \end{eqnarray}
1770: \end{theo}
1771:
1772:
1773: \noindent {\bf Proof}:
1774: Let $\{U_1, U_2, \cdots, U_n,\cdots\}$ be a locally finite covering
1775: of $M$ where each $U_j$ is a coordinate disk.
1776: Consider the restrictions of the sequence of metrics $\{ g_n\}$
1777: in each $U_j.\;$ These metrics have a finite area $C_1$
1778: and a finite energy $C_2.\;$ Apply theorem~\ref{theo:lo:weak} successively to
1779: metrics in each coordinate disk. In $U_1,$ there exists a subsequence $\{g_{1n},
1780: n\in {\bf N}\} $ of $\{g_n\},$ a finite set of bubble points
1781: $S_1 = \{q_{11},q_{12},\cdots,q_{1m_1}\}
1782: (0\leq m_1 \leq\sqrt{\frac{C_1\cdot C_2}{4\pi^2}} )$
1783: with respect to this subsequence, and a limit metric $h_1 $ in
1784: ${\hat{H}}^{2,2}_{loc}(U_1\setminus S_1)$ such that:
1785: \[ g_{1n} \rightharpoonup h_1 \;{\rm in}\; {\hat{H}}^{2,2}_{loc} (U_1 \setminus S_1).\]
1786:
1787: Consider this subsequence $\{g_{1n}\}$ in $U_2.\;$ There exists a subsequence
1788: $\{g_{2n}, n\in {\bf N}\}$ of $ \{g_{1n}, n\in {\bf N}\},$
1789: a finite set of bubble points
1790: $S_2 = \{q_{21},q_{22},\cdots,q_{2m_2}\}
1791: (0 \leq m_2 \leq\sqrt{\frac{C_1\cdot C_2}{4\pi^2}} )$ with respect to this subsequence,
1792: and a limit metric $h_2$ in ${\hat{H}}^{2,2}_{loc}(U_1 \setminus S_1) $ such that:
1793: \[
1794: g_{2n} \rightharpoonup h_2 \;{\rm in}\; {\hat{H}}^{2,2}_{loc} (U_2 \setminus S_2).\]
1795:
1796: In general, if $\{g_{jn}\}$ had been defined in each coordinate disk $U_i ( i \leq j),$
1797: we can select a subsequence $\{g_{(j+1)n}\}$ of $\{g_{jn}\}$
1798: in $U_{j+1}$
1799: so that there is a finite number of bubble points
1800: $S_{j+1}= \{q_{(j+1)1},q_{(j+1)2},\cdots,q_{(j+1)m_{j+1}}\}
1801: (0 \leq m_{j+1} \leq \sqrt{\frac{C_1\cdot c_2}{4\pi^2}} )$
1802: in $U_{j+1}$ with respect to this subsequence, and a limit metric
1803: $h_{j+1}$ in ${\hat{H}}^{2,2}_{loc}(U_{j+1}\setminus S_{j+1})$
1804: such that:
1805: \[
1806: g_{(j+1)n} \rightharpoonup h_{j+1} \;{\rm in}\;
1807: {\hat{H}}^{2,2}_{loc} (U_{j+1} \setminus S_{j+1}).
1808: \]
1809: Consider the diagonal subsequence $\{g_{nn}, n\in {\bf N}\}.\;$
1810: In each coordinate disk $U_j (\forall j)$,
1811: the following holds true:
1812: \[ g_{nn} \rightharpoonup h_{j} \; {\rm in} \;
1813: {\hat{H}}^{2,2}_{loc} (U_{j} \setminus S_{j}).\]
1814:
1815: This set of limit metrics $\{h_j\}$ then defines a limit metric
1816: $g_0$ in $H^{2,2}_{loc} (M \setminus (\bigcup_{j} S_j))$
1817: by:
1818: \[ g_0(p) = h_j (p),\qquad \forall\; p\, \in U_j.\]
1819:
1820: This metrics is well defined since $h_i \equiv h_j$ on $U_i \bigcap U_j$ if
1821: $U_i \bigcap U_j \neq \emptyset.\;$
1822: Thus,
1823: \[
1824: g_{nn} \rightharpoonup g_0 \;{\rm in}\; H^{2,2}_{loc} (M \setminus (\bigcup_{j} S_j)).
1825: \]
1826:
1827: The cardinality of the set $\bigcup_{j} S_j $ must be
1828: bounded by $ \sqrt{\frac{C_1 \cdot C_2}{4 \pi^2}} $
1829: according to the proof of proposition 2. \\
1830:
1831: Re-Label this subsequence as $\{g_n\}.\;$
1832: For any two pair of coordinate
1833: disks $U_i,\; U_j$ where $ U_i \bigcap U_j \not = \emptyset,$ we have:
1834: \begin{eqnarray}
1835: \displaystyle{\lim_{n\rightarrow \infty}} A(g_n,U_j \bigcup U_k)
1836: & = & A(g_0, (U_j \bigcup U_k)\setminus(S_j \bigcup S_k) )
1837: + \sum_{ p \in S_j \bigcup S_k} A_{p},
1838: \\
1839: \displaystyle{\lim_{n\rightarrow \infty}} K(g_n,U_j \bigcup U_k)
1840: & \geq & K(g_0,(U_j \bigcup U_k)\setminus (S_j \bigcup S_k))
1841: + \sum_{ p \in S_j \bigcup S_k} K_{p}.
1842: \end{eqnarray}
1843: These two formulas can be readily generalized to any number of coordinate disks:
1844: \begin{eqnarray}
1845: \displaystyle{\lim_{n\rightarrow \infty}} A(g_n, \bigcup_{k} U_k)
1846: & = & A(g_0, \bigcup_{k} U_k \setminus ( \bigcup_k S_k) )
1847: + \sum_{ p \in \bigcup_k S_k} A_{p},
1848: \\
1849: \displaystyle{\lim_{n\rightarrow \infty}} K(g_n, \bigcup_k U_k)
1850: & \geq & K(g_0, \bigcup_k U_k \setminus ( \bigcup_k S_k))
1851: + \sum_{ p \in \bigcup_k S_k} K_{p}.
1852: \end{eqnarray}
1853: Observed that $M= \bigcup_{k} U_s $ and $ \bigcup_k S_k = \{p_1,p_2,\cdots,p_m\}.\;$
1854: QED.\\
1855:
1856:
1857: \subsection {Blowing up procedure and tenuously connected sum}
1858: Let us re-state theorem~\ref{theo:lo:bubble} in the geometric context.
1859: \\
1860:
1861: \noindent {\bf Theorem ~\ref{theo:lo:bubble}$'$} (Bubbles on bubbles). {\it
1862: Let $\{g_n, n\in{\bf N}\}$ be a sequence of metrics
1863: in $D$ with a finite area $C_1$ and energy $C_2.\;$ Suppose that $ p=0$ is
1864: the only bubble point in $D$ with area concentration $A_p$ and energy
1865: concentration $K_p.\;$ Fix a local $z-$coordinate system centered at $p$
1866: and a scaling constant $\varepsilon.\;$ If $\varepsilon$ is small enough,
1867: we can re-normalize the sequence of metrics by $\tilde{g}_n(z)
1868: = g_n(\epsilon_n \cdot z + z(p_n))$
1869: where $\{\epsilon_n \searrow 0 \}$ is uniquely
1870: determined by the scaling constant $\varepsilon;$
1871: where $p_n \rightarrow p$ is the supremum of mass $g_n$ in $D.\;$
1872: There then exists a subsequence of $\{g_n\}$, a finite number of bubble points
1873: $\{q_1,q_2,\cdots,q_m\} (0\leq m \leq \sqrt{\frac{A_p \cdot K_p} {4\pi^2}})$
1874: in $S^2 \setminus \{\infty\}$ with respect to the corresponding
1875: subsequence of $\{\tilde{g}_n\},$ and a limit metric $\tilde{g}_0$
1876: in ${\hat{H}}^{2,2}(S^{2}\setminus \{\infty, q_1,q_2, \cdots,q_m\})$
1877: such that:
1878: \[\tilde{g}_n \rightharpoonup \tilde{g}_0\; {\rm in }\; {\hat{H}}^{2,2}_{loc}
1879: (S^{2} \setminus \{\infty, q_1,q_2, \cdots,q_m\} ).\]
1880:
1881: If the amount of area and energy concentrations of $\{\tilde{g}_n\}$
1882: at each $q_i $ are $A_{q_i}$ and $K_{q_i}$ respectively,
1883: then:
1884: \begin{eqnarray}
1885: A_p & \geq & A(\tilde{g}_0,S^2 \setminus\{\infty,q_1,q_2,\cdots, q_m\}) + \sum_{i=1}^{m} A_{q_i}, \\
1886: K_p & \geq & K(\tilde{g}_0,S^2 \setminus\{\infty,q_1,q_2,\cdots, q_m\}) + \sum_{i=1}^{m} K_{q_i}.
1887: \end{eqnarray}
1888: }\\
1889:
1890: Let us review the steps taken in the proof of theorem~\ref{theo:lo:bubble}.
1891: For convenience, we use a complex notation. The metric can be expressed as:
1892: \[
1893: g_n(z) = e^{2 \varphi_n(z)} |d\,z|^2, \qquad \forall n \in {\bf N}.
1894: \]
1895:
1896: The first step is to move the supremum of mass of the metric $g_n$ to the center
1897: of the coordinate system. If $\{p_n\}$ is such a sequence of points,
1898: then define
1899: \[
1900: \tilde{g}_n (z) = g_n(z + z(p_n)),\qquad z \in D_{1}(p).
1901: \]
1902:
1903: Re-Label $\{\tilde{g}_n \}$ as $\{g_n\}.\;$ The supremum of
1904: the metric $g_n$ is now at $p,\;\forall n \in {\bf N}.\;$ \\
1905:
1906: Choose a small positive number $\varepsilon < \epsilon_0 $(as in
1907: corollary~\ref{cor:blow}) as a filter.
1908: Following the proof of theorem 3, there then exists
1909: a number $ r_{1} > 0 $ such that if $ n $ is large enough, we have:
1910: \begin{equation} \displaystyle{\max_{0\leq\theta\leq 2\pi}}
1911: ( {\varphi_{n}}(r_{1} \cos \theta,r_1 \sin \theta) + \ln r_{1})
1912: \ll \varepsilon, \qquad \forall \;n > N,
1913: \label{eq: glo:blow1}
1914: \end{equation}
1915: or the length of this circle at $ |z| = r_1$ is very small:
1916: \begin{equation}
1917: \int_0^{2 \pi} e^{\varphi_n(r_1 \cdot z)}\, r_1\, d\,\theta
1918: \ll \varepsilon, \qquad n > {\bf N}.
1919: \label{eq:glo:blow2}
1920: \end{equation}
1921:
1922: Following corollary~\ref{cor:blow},there exists $\delta_{n} > 0 $ such that:
1923: \begin{equation} \int_0^{2 \pi} e^{\varphi_n(r \cdot z)}\, r\, d\,\theta
1924: < \varepsilon,\qquad \forall\; r_{1} \geq r \geq {\delta_{n}},
1925: \label{eq:glo:blow3} \end{equation}
1926: and
1927: \begin{equation}
1928: \int_0^{2 \pi} e^{\varphi_n (\delta_n \cdot z)}\, \delta_n \,d\,\theta = \varepsilon.
1929: \label{eq:glo:blow4}
1930: \end{equation}
1931:
1932: The circle $ |z| = \delta_n$ is the first circle for which the
1933: metric $g_n$ has a length of $\varepsilon\;$ beyond
1934: a thin neck. Hence, the set of concentric circles $\{|z| = \delta_n \}$
1935: is uniquely determined by the filter size $\varepsilon $ once the local coordinate
1936: system is picked. Define a sequence of conformal parameter functions as:
1937: \[ \phi_n(z) = \varphi_n( \delta_n \cdot z) + \ln \delta_n, \qquad n \in {\bf N}.\]
1938:
1939: Thus re-normalize the original sequence of metrics as
1940: \[
1941: \tilde{g}_n(z) = e^{2 \phi_n(z)} |d\,z|^2 = g_n( \delta_n \cdot z), \qquad n \in {\bf N}.
1942: \]
1943:
1944: Theorem~\ref{theo:lo:bubble} then asserts that we could
1945: choose a subsequence $\{\varphi_{n_i}, i\in {\bf N}\}$ of
1946: $\{\varphi_n, n\in {\bf N}\},$ a finite number of bubble
1947: points $\{q_1,q_2,\cdots,q_m\} ( 0\leq m \leq \sqrt{\frac{A_p\cdot K_p}{4\pi^2}})$
1948: such that one of the following two alternatives holds true:
1949: \begin{enumerate}
1950: \item $\phi_{n_j}( z)
1951: \rightarrow -\infty $ in $ S^2\setminus \{\infty,q_1,q_2,\cdots,q_m\}.$
1952: \item There exists a metric $\phi_0 \in H^{2,2}_{loc}( S^2\setminus
1953: \{\infty,q_1,q_2,\cdots,q_m\})$ such that
1954: \[
1955: \phi_{n_j}( z)
1956: \rightharpoonup \phi_0, \;{\rm in}\; H^{2,2}_{loc}( S^2\setminus
1957: \{\infty,q_1,q_2,\cdots,q_m\}).
1958: \]
1959: \end{enumerate}
1960:
1961: Define $\tilde{g}_0 \equiv 0 $ in a vanishing case; and define
1962: $\tilde{g}_0 = e^{ 2 \phi_0} |d\,z|^2 $ in a non-vanishing case. Thus,
1963: \[ \tilde{g}_{n_j} \rightharpoonup \tilde{g}_0\;{\rm in}\; {\hat{H}}^{2,2}_{loc}( S^2\setminus
1964: \{\infty,q_1,q_2,\cdots,q_m\}). \]
1965:
1966: Moreover,
1967: \begin{eqnarray}
1968: \tau_p & = & A_p - A(g_0,S^2 \setminus\{\infty,q_1,q_2,\cdots, q_m\})
1969: - \sum_{i=1}^{m} A_{q_i} \geq 0,\\
1970: K_p & \geq & K(g_0,S^2 \setminus\{\infty,q_1,q_2,\cdots, q_m\})
1971: + \sum_{i=1}^{m} K_{q_i},
1972: \end{eqnarray}
1973:
1974: where $\tau_p$ denote the amount of area lost in the neck during the
1975: re-normalization (blowing up) process.\\
1976:
1977: Choose $r_2 $ big enough, so that $\{q_1,q_2,\cdots,q_m\}
1978: \subset D_{r_2}.\;$ Consider the cylinder bounded
1979: by the two concentric circles $|z| = r_1$ and $|z| = r_2 \cdot \delta_n.\;$
1980: This cylinder is called the ``neck'' of the blowing up process.
1981: By definition, the length of a circle in this cylinder is bounded
1982: above by $\varepsilon.\; $ As $n\rightarrow
1983: \infty,$ the conformal distance between the two boundary circles approaches $\infty,$ while
1984: part of the interior of the neck collapses into a line.
1985: The collapsing can occur either by keeping the
1986: scalar curvature point-wisely bounded, or by keeping the diameter of the neck
1987: bounded. Denoted this neck by $Neck(r_1,r_2).\;$ We can shrink
1988: the size of the neck by letting $r_1 \rightarrow 0$ and $r_2 \rightarrow \infty.$\\
1989:
1990: This blowing up procedure or the re-normalization procedure depends only on the
1991: filter size $\varepsilon> 0 $ once a coordinate system is fixed.
1992: Suppose that $g_0$ is a limit metric in ${\hat{H}}^{2,2}(D\setminus\{p\})$ such that:
1993:
1994: \[ g_n \rightharpoonup g_0\; {\rm in}\;{\hat{H}}^{2,2}_{loc}(D\setminus\{p\}). \]
1995: The surface $(g_0, D\setminus{p})$ and
1996: $(\tilde{g}_0, S^{2} \setminus \{\infty,q_1,q_2,\cdots,q_m\}) $
1997: are called tenuously connected at $p$ and at $z=\infty.\;$
1998: If $\tau_p = 0,$ the connected sum is then efficient. Otherwise
1999: the connected sum is inefficient.\\
2000:
2001: The following is an example of a ``tenuously connected sum'' of two
2002: Riemannian metrics or surfaces:\\
2003:
2004: \noindent {\bf Example 2.} {\it We first construct
2005: a metric in a disk, where the boundary curve is a closed geodesic. We can make
2006: the length of the boundary approaches $0,$ while keeping the area and energy
2007: finite (see Figure~\ref{fg: closed geod.} below).
2008: The following is a sketch of the construction. Suppose that
2009: $g = e^{2\varphi} |d\,z^2|$ is a rotationally symmetric metric defined in
2010: ${\bf R^2}$ (real plane) such that: \label{example}
2011: \[
2012: \varphi(r) = -\ln r - \beta \cdot \ln (\ln r),\qquad \forall\,r > 2,
2013: \]
2014: where $ {1\over 2} < \beta < {3\over 2}.\;$
2015: Let $\epsilon_n = {{\beta} \over{\ln n}},\delta_n = {{\beta} \over{\ln^2 n}},$ and
2016: $ T_n = n + \ln n.\;$ Define a sequence of metrics $\{\varphi_n\} $ in $D_{T_n}$
2017: as the following:
2018: \[
2019: \varphi_n(r) =\left\{\begin{array}{ll} \varphi(r), &\mbox{when $r\leq n$}\\
2020: \varphi(n)+\ln n -\ln r -\epsilon_n (r-n) +{1\over 2}\delta_n
2021: (r-n)^2, &\mbox{when $n \leq r \leq T_n$} \end{array} \right.
2022: \]
2023:
2024: It is straightforward to prove that $|z| = T_n$ is a closed geodesic of $\varphi_n$
2025: and
2026: \[ \displaystyle{\lim_{n\rightarrow \infty}} \varphi_n(r) = \varphi(r), \qquad \mbox{ if $r$ is finite,}\]
2027: and
2028: \[
2029: \displaystyle{\lim_{n\rightarrow \infty}} E_c(\varphi_n,D_{T_n}) = E_c(\varphi,{\bf R^2}),\qquad
2030: \displaystyle{\lim_{n\rightarrow \infty}} A_c(\varphi_n,D_{T_n}) = A_c(\varphi,{\bf R^2}).
2031: \]
2032:
2033: Gluing two identical copy of $\varphi_n$ along
2034: the curve $|z| = T_n,$ we obtain a metric $g_n$ in $S^2.\;$
2035: Clearly, $\{g_n\}$ has finite energy and area, and it weakly
2036: converges to $g$ everywhere except at
2037: near $z=\infty.\;$ If we blow up the sequence near $z=\infty$, we obtain a
2038: new sequence of metrics
2039: which locally weakly converges to $g$ except at $z=0.\;$
2040: Re-label this metric as $g_{\infty}.\;$ Then the limit tree structure of the
2041: weak limit of $(S^2, g_n)$ consists of a root vertex $g$ and a child vertex
2042: $g_{\infty}.\;$ The corresponding metrics $g$ and $ g_{\infty}$ at
2043: the two nodes are tenuously connected. This sequence of metrics, clearly has
2044: no convergent subsequence in the elementary sense, even up to the
2045: M\"obious group. Using a similar mechanism, we could construct examples
2046: of a sequence of metrics which demonstrates a more sophisticated pattern
2047: of limit tree structures.
2048: }\\
2049:
2050: \begin{figure}
2051: \centerline{\psfig{figure=new3.eps}}
2052: \caption{Surface bounded with a small closed geodesic}
2053: \label{fg: closed geod.}
2054: \end{figure}
2055:
2056:
2057: \begin{lem} Let $g $ be a metric in $D\setminus\{p\}$ with a finite energy $C_2.\;$
2058: Suppose $|\partial D_1|_g = \epsilon $ where $D_1 \subset D\setminus\{p\}.\;$
2059: There exists a constant $C_{\epsilon} > 0 $ (independent of
2060: metric $g$) such that:
2061: \[ A(g, D\setminus \{p\}) = \int_{D\setminus \{p\}} d\,g > C_{\epsilon}.\]
2062: \label{lem:leastarea}
2063: \end{lem}
2064: {\bf Proof.} If the lemma is false, there then exists a
2065: sequence of metrics $\{g_n\}$ in $D\setminus\{p\}$
2066: such that:
2067: \[K(g_n,D\setminus\{p\}) < C_2,\qquad A(g_n,D\setminus\{p\})
2068: \rightarrow 0,\qquad |\partial D_1|_{g_n} = \epsilon>0.\]
2069:
2070: According to theorem 5, there exists a subsequence
2071: of $\{g_{n}\},$ a finite bubble points $\{p_1, p_2, \cdots,p_m\} (m \geq 0)$, and
2072: a limit metric $g_0$ in ${\hat{H}}^{2,2}(D\setminus\{p,p_1,p_2,\cdots,p_m\})$
2073: such that:
2074: \[ g_n \rightharpoonup g_0\;{\rm in}\; {\hat{H}}^{2,2}_{loc}(D\setminus\{p,p_1,p_2,\cdots,p_m\}).
2075: \]
2076:
2077: Thus, $m = 0 $ since the product of area and energy of this subsequence approaches
2078: $0.\;$ Moreover, $g_0 \equiv 0 $ in $ D\setminus\{p\} $ since total
2079: area approaches $0.\;$
2080: This is also impossible since $ |\partial D_1|_{g_n} = \epsilon > 0.\;$
2081: The lemma is then proved. QED.\\
2082:
2083: \begin{prop} (continued from theorem 3$'$). $\tilde{g}_0$ is as defined in theorem
2084: 3$'$. If $\tilde{g}_0\neq 0$ and $m=1,$ then
2085: $ \int\limits_{S^2 \setminus\{\infty,p\}} \,d\,\tilde{g}_0
2086: > C_{\varepsilon} > 0,$ where $C_{\varepsilon}$ depends only
2087: on $C_1, C_2$ and the scaling constant $\varepsilon.$
2088: \end{prop}
2089: Proof: If $\tilde{g}_0 \neq 0 $ and $m=1, $ then $p$ must be the only bubble point
2090: and $|\partial D_1 |_{g_0} = \epsilon > 0.\;$ The proposition
2091: then follows the previous lemma. QED.\\
2092:
2093:
2094: \section{Bubble tree}
2095:
2096:
2097: \noindent {\bf Theorem A.}\label{th:theorem A} {\it The limit of any locally weakly convergent
2098: sequence of metrics $\;\{g_k,\;k \in \bf {N}\}\;$ $\in\;$
2099: ${\cal{S}}(g_{0},C_1,C_2,\Omega),\;$
2100: consists of the following 4 objects: (1) A
2101: finite, rooted tree $T$, possibly reduced to just the base vertex $f.$
2102: (2) The base vertex $f \in T $ is a
2103: limit metric in $\Omega$ with a finite number of bubble points $\{p_i\}$
2104: deleted; the edges emanating from the base vertex is $\{p_i\};$
2105: there are three masses associated with each edge: the area
2106: concentration $a_i$, energy concentration $e_i$ and
2107: area loss during the blowing up process $\tau_i$ ($ a_i \cdot e_i \geq 4\pi^2 $). (3)
2108: Any other vertex $f_s$ is a limit metric defined
2109: on $S^{2}\setminus\{\infty,p_{si}\};$ the edges emanating
2110: from this vertexes are $\{p_{si}\}$; and there are three masses associated
2111: with each edge: the area concentration $a_{si}$, energy concentration $e_{si}$ and
2112: area lost during the blowing up process $\tau_{si}.\;$
2113: (4) For each pair of vertexes $f_{s_1} $ and $ f_{s_2}$ bounding
2114: an common edge in $T$, they are tenuously connected at the pair of respective
2115: singular points. The connected sum is efficient if the area loss associated with
2116: that edge is $0.\;$
2117: If the tree $T$ consists of
2118: only the base vertex $f,\;$the sequence of metrics $\{g_k\}$ is then said
2119: to have a weak convergent limit in the elementary sense (up to the M\"obious group).
2120: The number of vertexes whose valence $\neq 2$, is bounded from
2121: above $ (\leq \sqrt{ C_1\cdot C_2}).\;$
2122: The depth of the tree is also finite in a reasonable sense. } \\
2123:
2124: \noindent {\bf Proof of Theorem A}.
2125: The tree structure is constructed from a sequence of metrics $\{g_n\}$
2126: in $\Omega$ as follows (see Figure~\ref{fg:closed} on p.~\pageref{fg:closed}):
2127: First, choose a scaling constant $\varepsilon_0$ as
2128: a filter for re-normalization process. The $\{g_n\}$ locally weakly converges to
2129: $f_0$ on $\Omega \setminus\{p_1,p_2,\cdots,p_m\} $ except a finite number of
2130: bubble points $\{p_1,p_2,\cdots,p_m \}.\;$ The base vertex of the
2131: tree is the metric $f_0$, which we re-labeled as $f,$
2132: and the edges emanating from the base vertex are the points $\{p_i\}.\;$
2133: Each edge has an energy mass $e_i$ and area mass $a_i,$ which are the energy and area
2134: concentrations at the bubble point $p_i.\;$ For each $p_i$, the re-normalization
2135: process gives a new sequence of metrics $\{\tilde{g}_n\}$
2136: which locally weakly converge to a metric $f_i$
2137: in $S^2 \setminus\{\infty,p_{ij},j=1,2,\cdots,m_i\},$ with the
2138: amount of energy and area concentrated at
2139: each point $p_{ij}$ are $e_{ij}$ and $a_{ij}.\;$ We label each edge as $(p_i,e_i,a_i, \tau_i)$
2140: where $\tau_i$ represent the amount of area lost at the bubble point $p_i$
2141: during the blowing up process. If $\tau_i = 0, $ the blowing
2142: up process is then efficient. The edge $(p_i,e_i, a_i, \tau_i)$ ($e_i \cdot a_i \geq 4\pi^2 $ according to lemma 2) terminates
2143: at the vertex $f_i$ which, in turn, is the source of new edges $\{p_{ij}\}, $
2144: and so on. \\
2145:
2146:
2147: At each vertex $f_I = f_{i_1\cdots i_{k-1}i_k}$ of the tree, use
2148: $S_I $ to denote all of the bubble points of this limit metric other than
2149: the point $ z=\infty.\; $ If $f_I$ is not the base vertex, it must have a
2150: parent vertex $f_{I'} = f_{i_1 \cdots i_{k-1}}$. The surface
2151: $(f_{I'}, S^2\setminus\{\infty, S_{I'}\}) $
2152: ( or $ (f, M\setminus\{p_1,p_2,\cdots,p_m\} $ if $I' = \emptyset $)
2153: is tenuously connected to
2154: $( f_{I},S^2\setminus\{\infty, S_{I'}\}).\; $ If there is area loss
2155: during the blowing up process ($\tau_{I'} \not = 0$),
2156: the connected sum is inefficient.
2157: \\
2158:
2159: Each vertex $f_I$ has a special property: if it vanishes in any point
2160: in its domain, then it vanishes everywhere in its domain. In the
2161: case when $f_I \equiv 0,$ we call this a ghost vertex. At each
2162: ghost vertex other than the base vertex, there exists at least two edges
2163: emanating from it. In other other words, the metric has at least two bubble
2164: points.\\
2165:
2166: The ghost vertex does appear, as seen in example 3 below.
2167: However, there exists at most a finite number
2168: of ghost vertexes. Otherwise, consider all the vertexes in the tree that
2169: have at least two edges emanating from them. These vertexes must
2170: be infinitely many since every ghost vertex has at least two edges emanating
2171: from it. There exists an infinite number of edges
2172: where no two edges belong to the same branch of the tree.
2173: Re-Labeling these edges if necessary, we may assume that these
2174: edges are $\{(q_i,e_i,a_i, \tau_i), i \in {\bf N}\}$
2175: where $a_i \cdot e_i \geq 4 \pi^2.\;$ Therefore,
2176: \[ C_1 \geq \sum_i a_i, \qquad C_2\geq \sum_i e_i. \]
2177: Thus,
2178: \[ C_1 \cdot C_2 \geq \sum_i a_i \cdot e_i = \sum_i 4\pi^2.\]
2179: The last inequality implies that the number of these vertexes (include
2180: all the ghost vertexes) must be finite. \\
2181:
2182: For any other vertex which has only one edge emanating from
2183: it, proposition 3 implies that the area of such a vertex is bounded
2184: below by a positive constant $C_{\epsilon}$,
2185: which depends only on $C_1,C_2$ and the scaling constant $\varepsilon.\;$
2186: The number of these vertexes is finite as well. \\
2187:
2188: Therefore, the limit tree has only a finite depth. If we reduce the size
2189: of the filter, a new vertex might be inserted into the tree
2190: structure. However, these
2191: new vertexes have only one edge emanating from it. The underlying surface
2192: is $S^2$ with two opposite points deleted. QED. \\
2193:
2194: \noindent{\bf Example 3}. {\it Let $f= (z-1)(z-2) \cdots(z-m)$ be a
2195: holomorphic function. Choose a simply connected domain $\Omega$
2196: which contains all zero points of $f$ but no zero points of $f'(z).\;$
2197: Thus, $g_n(z) = { 4\cdot n^2 \cdot |f'|^2 \over{ (1 + n^2 \cdot |f|^2 )^2}} \cdot |d\,z|^2$
2198: is a sequence of metrics well defined in $\Omega $ with finite area
2199: and energy (bounded above by $4\pi \cdot m$). Clearly, $g_n$
2200: weakly converges to $0$ everywhere except at $z=1,2,\cdots,m.\;$ At each
2201: point $z=k$, a renormalized sequence of metrics weakly converges to
2202: a metric in $S^2$ with curvature $1.\;$ Thus, the bubble tree
2203: of $(g_n,\Omega)$ consists of 1 ghost base vertex and $m$ first
2204: level vertexes, where each first level vertex represents a metric
2205: with curvature 1 in $S^2$. }\\
2206:
2207: \noindent {\bf Proof of Corollary B.} Suppose $ \{g_k,k \in {\bf N} \}$ is
2208: a sequence of metrics with finite area $C_1$ and energy $C_2$.
2209: If necessary, we pass to a subsequence so that the weak limit of
2210: this sequence has a bubble tree decomposition as
2211: described in theorem A. Consider a generic pair of consecutive vertexes
2212: $(f_I, f_{Ii})$ in the bubble tree, where $p_i$ is a bubble point of
2213: $f_I$ and the re-normalized sequence of metrics at $p_{Ii}$ weakly converges
2214: to $f_{Ii} $ except a few bubble points. Consider the ``neck'' of this blowing up process.
2215: It is a cylinder where the length of each concentric circle
2216: is bounded above by the scaling constant $\epsilon.\;$ We call this cylinder
2217: a ``thin component.'' Now iterate thorough each pair of consecutive vertexes, and obtain
2218: a finite number of ``thin'' components (See Figure~\ref{fg: thin-thick} below). The collection of ``thin
2219: components'' is labeled by $I_{thin}.\;$ For each fix $n$, remove
2220: all of the ``thin components'' from $M.\;$ The resulting surface
2221: is a disjoint union of a finite number of connected components.
2222: Each connected component is called a ``thick component.'' Label
2223: all of the ``thick'' components by $I_{thick}.\;$ Each thick component,
2224: together with the restriction of $g_n$ on it, weakly converges to a surface
2225: with a finite number of disks deleted. The thick component corresponding to the
2226: base vertex is $\Omega$ with a few disk deleted. All of the rest of thick
2227: components are $ S^2$ with a few disks deleted (When a thick component
2228: corresponds to a ghost vertex in the tree decomposition, the limit
2229: metric is $0$). The size of all deleted disks
2230: could be shrinked to $0$ by shrinking the size of corresponding blowing
2231: up ``neck.'' QED. \\
2232:
2233: {\bf Acknowledgments.} The author wishes to thank Professor J. Goodman
2234: for carefully reading my manuscript and many helpful comments he made.
2235: Thanks also go to Professor R. Schoen for many helpful
2236: and stimulating conversations during the course of this work.
2237: The author also wishes to thank his advisor Professor E. Calabi
2238: for his warm encouragement and continue supports during the past two years.\\
2239:
2240: \begin{figure}
2241: \centerline{\psfig{figure=new4.eps}}
2242: \caption{Thin-thick decompsition}
2243: \label{fg: thin-thick}
2244: \end{figure}
2245:
2246:
2247:
2248: %\end{section}
2249: \nocite{Tshioya96}
2250: \bibliography{test}
2251: %
2252:
2253:
2254: \bigskip
2255:
2256: \bigskip
2257:
2258: \bigskip
2259:
2260: \bigskip
2261:
2262:
2263:
2264: \begin{thebibliography}{99}
2265: \bibitem{BrezisM91} {H. Brezis and F. Merle},
2266: Uniform estimates and blow up
2267: behavior for solutions of $\; - \triangle u = {V}(x) e ^{u} \;$
2268: in two dimension,
2269: Comm. Partial Differential Equation,
2270: \textbf{16} (1991),
2271: 1223-1253
2272:
2273: \bibitem{calabi82} E. Calabi ,
2274: Extremal {K}\"ahler metrics,
2275: Seminar on Differential Geometry,
2276: Ann. of Math. Studies,
2277: \textbf{102} (1982),
2278: 259-290, University Press.
2279: \bibitem{cheeger86} Cheeger, J.,
2280: Gromov, M., Collapsing Riemannian manifolds
2281: while keeping their curvature bounded, I, J. Diff. Geom. 23(1986), 309-346
2282:
2283: \bibitem{Burago80} {Yu. d. Burago and V. a. Zalgaller},
2284: {Geometric Inequalities}
2285: {New York},
2286: {Spring-Verlag},
2287: {1980}
2288:
2289: \bibitem{BavardP88}
2290: C.Bavard et P. Pansu,
2291: l'space des surfaces, a corbure et aire bornees,
2292: Ann. Inst. Fourier,
2293: \textbf{38} (1988),
2294: 175-203
2295:
2296: \bibitem{ParkerWolfson} Parker Thomas H. and Hon G. Wolfson,
2297: Pseduo-{H}olomorphic Maps and Bubble Trees,
2298: The Journal of Geometric Analysis,
2299: \textbf{3} (1993),
2300: 63-98.
2301:
2302:
2303:
2304: \bibitem{Gromove81} M. Gromov, J. Lafonatiaine et P. Pansu ,
2305: Structures metriques pour les varietes {R}iemanniennes,
2306: Cedeic/Fernand Nathan,
2307: Paris 1981
2308: \bibitem{LiSh94}Y. Y. Li and I. Shafrir",
2309: Blow up analysis for solutions of $-\triangle u = {V}\,e^u$
2310: in dimension two,
2311: Indiana Univ. Math. Journal,
2312: \textbf{43}(1994), 1255-1270.
2313: \bibitem{MTrojanov:conical91}M.Troyanov,
2314: Prescribing curvature on compact surfaces with conical singularities,
2315: Tran. AMS,
2316: \textbf{324} (1991), 793-821
2317:
2318: \bibitem{Green88} R.Greene and H. H. Wu ,
2319: Lipschitz Convergence of {R}iemannian {M}anifolds,
2320: Pacific J. Math,
2321: \textbf{131} (1988),
2322: 119-141.
2323:
2324: \bibitem{Peter87}
2325: S. Peters,
2326: Convergence of {R}iemannian {M}anifolds,
2327: Compositio Math.,
\textbf{62} (1987) 3-7.
2328:
2329:
2330:
2331:
2332: \bibitem{MTrojanov:concentration91}
2333: M. Trojanov,
2334: Un principle de concentration-compacite pour les suites
2335: de surfaces riemaniennes,
2336: Ann. Inst. Henri Polatexincare,
2337: \textbf{8} (1991), 1-23.
2338:
2339: \bibitem{Uhlenbeck}K. Uhlenbeck,
2340: Connections with ${L}^{p}$ bounds on {C}urvature ,
2341: Commun. Math. Phys,
2342: \textbf{83} (1982),
2343: 31-42.
2344:
2345: \bibitem{chen943}
2346: Xiu xiong, Chen,
2347: {Obstruction to the Existence of Extremal K\"{a}hler metric
2348: in a
2349: Surface with conical singularities},
2350:
2351: to appear in C.A.G.,
2352: (1994)
2353:
2354: \end{thebibliography}
2355: \end{document}