Kodaira Dimension and the Yamabe Problem
Extracted source text ยท arXiv dg-ga/9702012
\documentstyle{article}
\def\Bbb{\bf}
\def\cp{{\Bbb CP}}
\def\bcp{\cp}
\def\kod{\mbox{Kod}}
%%%%%ENVIRONMENTS
\def\bea{\begin{eqnarray*}}
\def\eea{\end{eqnarray*}}
\def\mv{\mbox{Vol}}
\newtheorem{main}{Theorem}
\renewcommand{\themain}{\Alph{main}}
\newtheorem{defn}{Definition}
\newtheorem{thm}{Theorem}
\newtheorem{prop}{Proposition}
\newtheorem{lem}{Lemma}
\newtheorem{cor}{Corollary}
\newtheorem{conj}{Conjecture}
\newenvironment{proof}{\medskip {\bf Proof.}}{\hfill \rule{.5em}{1em}
\\}
\def\qed{\hfill \rule{.5em}{1em}
\\}
\begin{document}
\sloppy
\title{Kodaira Dimension and the Yamabe Problem}
\author{Claude LeBrun\thanks{Supported
in part by NSF grant DMS-9505744.}
\\
SUNY Stony
Brook
}
\date{}
\maketitle
\begin{abstract} The {\em Yamabe invariant} $Y(M)$
of a
smooth compact manifold is roughly the supremum of
the scalar curvatures of unit-volume constant-scalar curvature
Riemannian metrics $g$ on $M$. (To be absolutely
precise, one only considers constant-scalar-curvature
metrics which are {\em Yamabe minimizers}, but this
does not affect the {\sl sign} of the answer.)
If $M$ is the underlying smooth 4-manifold of a
complex algebraic surface $(M,J)$,
it is shown that the sign of $Y(M)$ is
completely determined by the Kodaira dimension
$\kod (M,J)$. More precisely,
$Y(M) < 0$ iff $\kod (M,J)=2$;
$Y(M) = 0$ iff $\kod (M,J)=0$ or $1$;
and $Y(M) > 0$ iff $\kod (M,J)=-\infty$.
\end{abstract}
%\vfill
%\pagebreak
\section{Introduction}
One may define an interesting and natural diffeomorphism invariant
of a compact smooth $n$-manifold $M$ as the supremum
of the scalar curvatures of unit-volume
constant-scalar-curvature
metrics on $M$. A minor refinement of this definition,
which does not change the sign of the invariant but guarantees
that it is finite, restricts the class of allowed
constant-scalar-curvature metrics to the so-called
Yamabe minimizers. We will refer to the resulting
invariant as the {\em Yamabe invariant} $Y(M)$ of
our manifold. For a more precise definition,
see \S \ref{yam} below.
For 2-manifolds, this invariant is easy to compute;
the classical Gauss-Bonnet theorem tells us
that the Yamabe invariant of a smooth compact surface is
just $4\pi \chi$, where $\chi$ is the
Euler characteristic.
Many important complex-analytic invariants of
compact complex curves are thus determined by the Yamabe invariant
of the underlying smooth 2-manifold.
One of the most important invariants of
a compact complex manifold $(M,J)$ is its
{\em Kodaira dimension} $\kod (M)$.
If $M$ has complex dimension $m$, recall that
the {\em canonical line bundle} $K \to M$
is $K=\wedge^{m,0}$, so that the
holomorphic sections of $K$ are exactly the
holomorphic $m$-forms on $(M,J)$.
One now defines the Kodaira dimension as
$$\kod (M) = \limsup \frac{\log h^0 (M,K^{\otimes \ell})}{\log \ell}.$$
This can be shown to coincide with the
maximal complex dimension of the image of $M$
under pluri-canonical maps to complex projective space,
so that $\kod (M)\in \{ -\infty , 0, 1, \ldots, m\}$.
A compact complex $m$-manifold is said to be of {\em general type}
if $\kod M =m$.
For Riemann surfaces,
the trichotomy $\kod (M) = -\infty , 0, 1$, exactly
coincides with that given by $Y(M) > 0, = 0, < 0$.
The purpose of this paper is to show that
much the same thing happens for compact complex surfaces
$(M^4,J)$. Our main result is
\begin{main} \label{main}
Let $M$ be the underlying 4-manifold of a
compact complex surface $(M^4,J)$ with $b_1(M)$ even.
Then
$$
Y(M) = \left\{
\begin{array}{ll} Y(M) > 0 &\mbox{iff } \kod (M,J) = -\infty \\
Y(M) = 0 &\mbox{iff } \kod (M,J) = 0 \mbox{ or } 1 \\
Y(M)< 0 &\mbox{iff } \kod (M,J) = 2.
\end{array}\right.
$$
\end{main}
The hypothesis that $b_1(M)\equiv 0\bmod 2$
is equivalent to requiring that $(M,J)$ admit
a K\"ahler metric. It is also equivalent to
requiring that $(M,J)$ be a deformation of a
complex algebraic surface. Many of the results
contained in this article will also apply in the
non-K\"ahler case, but the overall picture
remains considerably less clear when $b_1$ is odd.
One of the main tools in the proof of the above
result is Seiberg-Witten theory. One of the
most striking consequences of this
theory is that any two diffeomorphic complex
algebraic surfaces must have the same Kodaira
dimension \cite{FM,FQ}. Since the Yamabe invariant is
obviously a diffeomorphism invariant,
Theorem \ref{main} casts
this result in an interesting new light.
On the other hand, the Yamabe invariant
does not by itself distinguish Kodaira
dimension $0$ from Kodaira dimension $1$.
The distinction between these cases instead only
emerges when
one asks a finer question: when is the Yamabe invariant
actually the scalar curvature of some unit-volume
metric?
Finally, it should be emphasized that,
despite the cited analogy between complex
dimensions $1$ and $2$, the phenomenon
explored here does not persist in higher dimensions.
For example, consider a non-singular complex hypersurface of
degree $m+3$ in $\cp_{m+1}$, where $m\geq 3$.
This complex $m$-manifold has ample
canonical bundle, so its Kodaira dimension is $m$.
However, the underlying smooth manifold is
simply connected, non-spin, and has real dimension
$> 4$, so a surgical construction of
Gromov and Lawson \cite{GL} shows that it
admits metrics of positive scalar curvature,
and thus has $Y >0$.
Thus the Yamabe invariant and Kodaira dimension
are correlated only in complex dimensions $1$ and $2$.
\section{Yamabe Invariants}
\label{yam}
This section is expository in nature,
and is intended as a convenient summary of
the Yamabe folklore which will be needed in the
rest of the paper.
Let $M$ be a smooth compact manifold.
A {\em conformal class} on $M$
is by definition a collection of
smooth Riemannian metrics on $M$ of the form
$$[g]=\{ vg ~|~v: M\to {\Bbb R}^+\},$$
where $g$ is some fixed Riemannain metric.
To each such conformal class, one can
associate a number $Y_{[g]}$, called the
{\em Yamabe constant} of the class, by
$$Y_{[g]} = \inf_{{\hat{g}}\in [g] } \frac{\int_M
s_{\hat{g}}~d\mu_{\hat{g}}}{\left(\int_M
d\mu_{\hat{g}}\right)^{\frac{n-2}{n}}},$$
where $s_{\hat{g}}$ and $d\mu_{\hat{g}}$ respectively
denote the scalar curvature and volume measure of
${\hat{g}}$. If $g$ is a Riemannian metric
for which the scalar curvature is everywhere positive,
or everywhere zero, or everywhere negative,
one can show that $Y_{[g]}$ is correspondingly
positive, zero, or negative.
While $Y_{[g]}$ has been defined as
an infimum, we could have instead defined it
as a {\sl minimum}, for a remarkable theorem
\cite{aubin,lp,rick}
of Yamabe, Trudinger, Aubin, and Schoen
asserts that any conformal class
$[g]$ contains a
metric
which actually minimizes the relevant functional.
Such a metric will be called a {\em Yamabe minimizer}.
Any Yamabe minimizer has constant scalar curvature;
conversely, a constant-scalar-curvature metric $g$
with $s_g \leq 0$ is
automatically a Yamabe minimizer.
Aubin's piece of
the proof
hinges on the observation
that any conformal class on any
$n$-manifold automatically
satisfies
$Y_{[g]}\leq n(n-1) (V_n)^{2/n}$, where
$V_n$ is the volume of
the unit $n$-sphere
$S^n\subset {\Bbb R}^{n+1}$.
Given any smooth compact $n$-manifold, we can therefore
\cite{okob,sch}
define an associated real number by
$$Y(M) = \sup_{[g]} Y_{[g]} =
\sup_{[g]}\inf_{{\hat{g}}\in [g] } \frac{\int_M
s_{\hat{g}}~d\mu_{\hat{g}}}{\left(\int_M
d\mu_{\hat{g}}\right)^{\frac{n-2}{n}}}.$$
By construction, this is a diffeomorphism
invariant of $M$, and it will be called the
{\em Yamabe invariant} of $M$ in this paper.
Notice that $Y(M) > 0$ iff $M$ admits a
metric of positive scalar curvature.
If $Y(M) \leq 0$, the invariant
is simply the supremum of
the scalar curvatures of unit-volume
constant-scalar-curvature metrics on $M$,
since any constant-scalar-curvature metric
of non-positive scalar curvature is automatically a
Yamabe minimizer.
The minimax definition of $Y(M)$ is rather unwieldy for
many purposes. Fortunately,
it can often be calculated in a simpler manner.
\begin{lem}
Let $[g]$ be any conformal class on a smooth compact
$n$-manifold $M$. Then
$$\inf_{{\hat{g}}\in [g]} \int_M |s_{\hat{g}}|^{n/2} d\mu_{\hat{g}}
= |Y_{[g]}|^{n/2}.$$
\label{yobo}
\end{lem}
\begin{proof} If $n=2$, the Gauss-Bonnet theorem
tells us that $\int |s|d\mu \geq
|\int s~d\mu |= |4\pi \chi (M)| =
|Y_{[g]}|$, with equality iff $s$ does not
change sign. We may therefore assume henceforth that
$n \geq 3$. We will break this into two cases, depending
on the sign of $Y_{[g]}$.
First suppose that $Y_{[g]} \geq 0$ and $n \geq 3$. The H\"older
inequality tells us that each metric ${\hat{g}}$ satisfies
$$\left(\int_M |s_{\hat{g}}|^{n/2} d\mu_{\hat{g}} \right)^{2/n}\geq
\frac{\int_M s_{\hat{g}}d\mu_{\hat{g}}}{\left(\int_M
d\mu_{\hat{g}} \right)^{1-\frac{2}{n}}},$$
with equality iff $s_{\hat{g}}$ is a non-negative constant. Since the
infimum of the right-hand side over ${\hat{g}}\in [g]$ is $Y_{[g]}$
and is achieved by a metric of constant scalar curvature $s\geq 0$,
the claim follows.
Finally, we come to the case in which $Y_{[g]} < 0$
and $n \geq 3$.
Let $g\in [g]$ be a metric of constant negative scalar
curvature, and express any other metric in $[g]$
as $\hat{g}=
u^\ell g$,
where $\ell = 4/(n-2)$ and $u$ is some smooth
positive function. The scalar curvatures $s=s_g$ and
$\hat{s}=s_{\hat{g}}$ are related by
$$\hat{s}u^{\ell + 1}= su + (n-1)\ell \triangle u$$
where $\triangle=d^\ast d$ is the positive Laplacian of $g$.
Thus
\bea
\int_M \hat{s}u^\ell d\mu &=&
\int_M \left(s+(n-1)\ell \frac{1}{u}d^\ast du\right) d\mu\\
&=& \int_M \left(s-(n-1)\ell
\frac{|d u |^2}{u^2}\right) d\mu\\
&\leq& \int_Ms~d\mu ,
\eea
where $d\mu=d\mu_g$.
The H\"older inequality therefore implies
\bea
\left(\int_M |\hat{s}u^\ell |^{n/2}d\mu\right)^{2/n}
\left(\int_Md\mu\right)^{\frac{n-2}{n}}
&\geq& \int (-\hat{s}u^\ell ) d\mu\\
&\geq & -\int s ~d\mu\\
&=&|Y_{[g]}|
\left(\int_Md\mu\right)^{(n-2)/n}.
\eea
Hence
$$
\int_M |s_{\hat{g}}|^{n/2}d\mu_{\hat{g}}=
\int_M |\hat{s}u^\ell |^{n/2}d\mu \geq |Y_{[g]}|^{n/2},
$$
with equality iff $u$ is constant.
The result follows.
\end{proof}
This leads to a very useful reinterpretation of $Y(M)$;
cf. \cite{and,bcg,okob,sch}.
\begin{prop}\label{func}
Let $M$ be a smooth compact $n$-manifold, $n \geq 3$.
Then
$$
\inf_g \int_M |s_g|^{n/2}d\mu_g = \left\{
\begin{array}{ll} 0&\mbox{if } Y(M)\geq 0\\
|Y(M)|^{n/2} &\mbox{if } Y(M)\leq 0.
\end{array}\right.
$$
Here the infimum on the left-hand side
is taken over all smooth Riemannian metrics
$g$
on $M$.
\end{prop}
\begin{proof}
By Lemma \ref{yobo},
$$
\inf_g \int_M |s_g|^{n/2}d\mu_g = \inf_{[g]} |Y_{[g]}|^{n/2}.$$
If $Y(M) \leq 0$, we have $Y_{[g]} \leq 0$ for all $[g]$,
so the right-hand-side may be rewritten as
$(-\sup Y_{[g]})^{n/2} = |Y(M)|^{n/2}$, and we are done.
On the other hand, any
smooth manifold of dimension $\geq 3$ admits metrics
of negative scalar curvature \cite{bes}, and
$Y_{[g]}$ depends continuously
on $[g]$. Since the space of metrics is
connected, $Y(M) > 0$ implies there is a conformal
class $[g]$ on $M$ with $Y_{[g]}=0$, and the infimum
therefore vanishes.
\end{proof}
\section{Results from Seiberg-Witten Theory}
It is easy to see that
some
complex surfaces of K\"ahler type
carry metrics with positive scalar curvature.
\begin{prop} \label{rule}
Let $M$ be the underlying $4$-manifold
of a K\"ahler-type complex surface with
Kodaira dimension $-\infty$. Then
$Y(M) > 0$.
\end{prop}
Indeed, the Kodaira-Enriques classification \cite{bpv}
tells us that any such complex surface is either $\cp_2$ or else
is obtained from a $\cp_1$-bundle over a complex
curve by blowing up points. Now $\cp_2$ carries the
Fubini-Study metric, which has positive scalar curvature.
Any $S^2$-bundle over another manifold also carries metrics
of positive scalar curvature; namely, consider any Riemannian
submersion metric with round, totally geodesic fibers,
and then rescale the fibers to have very small diameter while
keeping
the metric on the base fixed. Finally, the blowing-up operation
amounts differentiably to taking connect sums with copies
of $\overline{\cp}_2$; all blow-ups of surfaces with
positive-scalar-curvature metrics therefore
themselves admit
positive-scalar-curvature metrics because a result of
Gromov and Lawson \cite{GL} tells us that the
class of manifolds with positive-scalar-curvature metrics
is closed under surgeries in codimension $\geq 3$.
On the other hand, it is a truly remarkable consequence of
Seiberg-Witten theory that the converse of the
above statement is also true:
\begin{thm} \label{sw}
Let $M$ be the underlying 4-manifold of a
complex surface of K\"ahler type with
Kodaira dimension $\geq 0$. Then $Y(M) \leq 0$.
\end{thm}
Remember,
this is just a fancy way of saying that
$M$ does not admit metrics of positive scalar curvature.
When $b^+(M) > 1$, this was first proved by
Witten \cite{witten}. In the $b^+=1$ case, this was
proved for minimal surfaces in \cite{leb1}, and
then in the
non-minimal case by Friedman and Morgan \cite{FM}.
The proof proceeds by first
showing that for every metric
$g$ on $M$, there must be a solution of the Seiberg-Witten
equations
\bea D^{\theta}\Phi &=&0 \\
F_{\theta}^+&=&i \sigma(\Phi) \eea
and an unknown unitary connection $\theta$
on some line bundle $L\to M$ and
an unknown twisted spinor $\Phi\in \Gamma (S_+\otimes L^{1/2})$.
Here $D^\theta$ is the Dirac operator coupled to
$\theta$, $ F_{\theta}^+$ is the projection
of the curvature of $\theta$ to the
bundle $\Lambda^+$ of $g$-self-dual 2-forms,
and $\sigma : S_+\otimes L^{1/2}\to \Lambda^+$
is the real-quadratic map induced by
the natural
isomorphism $\odot^2{\Bbb S}_+=\Lambda^+\otimes {\Bbb C}$.
Moreover, $L$ can either be taken to be the anti-canonical line bundle
of $(M,J)$ or its pull-back via some
diffeomorphism $M\to M$. However, these equations imply
the Weitzenb\"ock formula
$$0=4\nabla^{\theta\ast}\nabla^\theta + s\Phi + |\Phi|^2\Phi,
$$
where $s$ is the scalar curvature of $g$. Taking the
inner product with $\Phi$ and integrating, this tells
us
that
$$ 0 = \int_M \left[
|2\nabla^\theta\Phi|^2+ s |\Phi |^2 + |\Phi |^4
\right]
d\mu ,$$
which is a contradiction if $s > 0$.
Pushing this argument further yields a stronger
result \cite{leb3}
for surfaces of general type:
\begin{thm}\label{gtype}
Let $M$ be the underlying 4-manifold of a
complex surface with Kodaira dimension $2$. Then $Y(M) < 0$.
Moreover, if
$X$ is the minimal model of $M$, then
$Y(M)=Y(X)= -4\pi \sqrt{2c_1^2(X)}.$
\end{thm}
(Recall \cite{bpv}
that the minimal model $X$ is characterized by the
fact that $M$ can be blown down to $X$, but that $X$ cannot be
blown down any further.
The minimal model $X$ of a surface $M$ of general type
is unique, and $c_1^2 (X) > 0$.)
Indeed, applying the
Schwarz inequality to the integrated Weitzenb\"ock formula
tells us that
$$(\int_M s^2~d\mu )^{1/2} (\int_M
|\Phi|^4 ~d\mu)^{1/2} \geq
\int_M (-s)|\Phi|^2~d\mu
\geq \int_M |\Phi|^4 ~d\mu .$$
We must therefore have
$$\int_Ms^2d\mu \geq \int_M |\Phi|^4 ~d\mu =
8\int_M|F_{\theta}^+|^2d\mu$$
for any metric. However \cite{leb2},
$$ \int_M|F_{\theta}^+|^2d\mu \geq 4\pi^2([c_1(L)]^+)^2$$
where $[c_1(L)]^+$ is self-dual part of the
$g$-harmonic deRham representative of $c_1(L)$.
Moreover, for each $g$ one can show \cite{leb3} that
$L$ can be chosen so that $([c_1(L)]^+)^2\geq c_1^2(X)$.
This shows that
$$ \int_M s^2 d\mu \geq 32\pi^2 c_1^2(X)$$
for all metrics on $M$. But by gluing
together K\"ahler-Einstein orbifolds, gravitional instantons, and
standard metrics on the blow-up of ${\Bbb C}^2$ at the
origin, one can also
construct \cite{leb3}
sequences of metrics
which show that this bound is actually sharp:
$$\inf_g \int_M s^2 d\mu = 32\pi^2 c_1^2(X) >0.$$
Thus $Y(M) = -4\pi \sqrt{2c_1^2(X)} < 0$ by
Proposition \ref{func}.
\section{Rational Elliptic Surfaces Revisited}
A complex surface $(M^4,J)$ is said to be {\em elliptic} if it admits
a holomorphic map $M\to \Sigma$ to a complex curve such that
the regular fibers are elliptic curves $T^2$. While this notion
is of greatest importance in the study of surfaces of
Kodaira dimension $0$ and $1$, there are also elliptic
surfaces of Kodaira dimension $-\infty$. One family of
examples of the latter type are the rational elliptic surfaces
treated this section.
The smooth oriented 4-manifold $\cp_2\# 9 \overline{\cp}_2$
may be concretely realized as the blow-up of $\cp_2$
at $9$ points. One choice of these $9$ points is
to take them to be the points of intersection of a generic pair of
elliptic (i.e. cubic) curves in the projective plane.
There is then exactly a $\cp_1$'s worth of cubics passing through the
$9$ points, corresponding to linear combinations of the
defining equations of the original pair of curves.
When $\cp_2$ is blown up at these $9$ points,
this pencil of cubics gives rise to a holomorphic projection
$\cp_2\# 9 \overline{\cp}_2\to \cp_1$, the generic fiber
of which is a smooth elliptic curve $E\approx T^2$.
The term {\em rational elliptic surface} is thus
used to describe
$\cp_2\# 9 \overline{\cp}_2$ equipped
with the deformation class of this projection to
$\cp_1$.
There is a useful way of constructing special
rational elliptic surfaces by analogy to
the Kummer construction of K3's.
Let $\Lambda \cong {\Bbb Z}\oplus {\Bbb Z}$
be any lattice in $\Bbb C$, and let
$E={\Bbb C}/\Lambda$ be the corresponding elliptic curve.
Recall that
each such curve can be realized as a branched cover over
$\cp_1$ because reflection $z\mapsto -z$
through the origin in $\Bbb C$ induces in an involution
of $E$ with $4$ fixed-points\footnote{corresponding the
half-lattice $\frac{1}{2}\Lambda$}; the orbifold quotient
$E/{\Bbb Z}_2$ can then be identified with
a Riemann surface of genus $0$, and this
realizes $E$ as a branched cover
$\wp : E\to \bcp_1$. The involution of $E$ that
occurs in this argument is called the {\em Weierstrass
involution} of $E$, and will be denoted here by
$\Phi : E\to E$. Notice that $\Phi$ is an isometry
of the obvious flat metric on $E$.
Now
let $\Psi : \cp_1 \to \cp_1$ be the involution
$[z_0:z_1]\mapsto [-z_0: z_1]$, which has two fixed
points and amounts to a $180^\circ$
rotation of $S^2$ around an axis. The product
involution
$$\Phi \times \Psi : E\times \cp_1 \to E\times \cp_1$$
then
has 8 fixed points, and
the quotient $E\times \cp_1/{\Bbb Z}_2$ is
a complex orbifold with 8 singularities
modeled on ${\Bbb C}^2/\pm 1$. Let $\tilde{M}$ be
the blow-up of
of $E\times \cp_1$ at the 8 fixed points of
$\Phi \times \Psi$, and let $M = \tilde{M}/{\Bbb Z}_2$.
The non-singular complex surface $M$ is thus
obtained from $E\times \cp_1/{\Bbb Z}_2$
by replacing each of the $8$ singular points with a
$(-2)$-curve.
Now consider the holomorphic map
$\varpi : M \to \cp_1$
induced by following the
second-factor projection $E\times \cp_1\to \cp_1$
with $[z_0: z_1] \mapsto [z_0^2 : z_1^2]$. All
but two fibers of $\varpi$ are copies of $E$;
the only two exceptions are the fibers over the
north and south poles $[0:1]$ and $[1:0]$. Each of these
two exceptional fibers consists of five $(-2)$-curves,
linked together according to the Dynkin diagram
$\tilde{D}_4$:
\begin{center}
\begin{picture}(240,60)(0,3)
\put(100,30){\circle*{3}}
\put(80,50){\circle*{3}}
\put(80,10){\circle*{3}}
\put(120,50){\circle*{3}}
\put(120,10){\circle*{3}}
\put(80,10){\line(1,1){40}}
\put(80,50){\line(1,-1){40}}
\end{picture}
\end{center}
These two exceptional
fibers are of thus of Kodaira type $I_0^*(\tilde{D}_4)$.
Now
$\varpi : M \to \bcp_1$ is deformation
equivalent to the above-described standard model of the
rational elliptic surface.
In fact, $M$ can be blown down to $\cp_1\times \cp_1$
in such a way that the fibers of $\varpi$ are sent to
the elliptic curves of the pencil
$$t_0x_0(x_0-x_1)y_0^2+ t_1x_1(x_0-ax_1)y_1^2=0,$$
where the number $a$ is given in terms of the lattice
$\Lambda$ by the Weierstrass $\jmath$-function.
Notice that for each $[t]\in \cp_1$ this locus is
an element
of the anti-canonical linear system $|{\cal O}(2,2)|$ on
$\cp_1\times\cp_1$, and that each
elliptic curve of the pencil passes through the
points $([0:1],[0:1],)$,
$([1:1],[0:1])$,
$([a:1],[1:0])$, and
$([1:0],[1:0])$, and all are
tangent to the
second-factor $\cp_1$ at these points.
To see this explicitly, let us first
consider $E\times \cp_1$ as a ruled surface
over $E$, and blow it up at the $8$
relevant points to obtain $\tilde{M}$:
\begin{center}
\begin{picture}(240,110)(0,3)
\put(50,110){\line(1,0){160}}
\put(50,110){\line(0,-1){60}}
\put(20,110){\line(0,-1){60}}
\put(45,80){\vector(-1,0){15}}
\put(-5,80){$\cp_1$}
\put(75,110){\line(2,-3){13}}
\put(85,100){\line(0,-1){40}}
\put(75,50){\line(2,3){13}}
\put(105,110){\line(2,-3){13}}
\put(115,100){\line(0,-1){40}}
\put(105,50){\line(2,3){13}}
\put(100,100){\makebox(0,0){$-1$}}
\put(100,60){\makebox(0,0){$-1$}}
\put(123,80){\makebox(0,0){$-2$}}
\put(140,110){\line(2,-3){13}}
\put(150,100){\line(0,-1){40}}
\put(140,50){\line(2,3){13}}
\put(175,110){\line(2,-3){13}}
\put(185,100){\line(0,-1){40}}
\put(175,50){\line(2,3){13}}
\put(122,44){\makebox(0,0){$-4$}}
\put(122,116){\makebox(0,0){$-4$}}
\put(230,80){\makebox(0,0){$\tilde{M}$}}
\put(230,15){\makebox(0,0){$E$}}
\put(50,50){\line(1,0){160}}
\put(210,110){\line(0,-1){60}}
\put(135,40){\vector(0,-1){15}}
\put(50,15){\line(1,0){160}}
\put(75,15){\circle*{3}}
\put(105,15){\circle*{3}}
\put(140,15){\circle*{3}}
\put(175,15){\circle*{3}}
\end{picture}
\end{center}
Dividing by $\Phi\times\Psi$ then gives us a
non-minimal rational ruled surface:
\begin{center}
\begin{picture}(240,110)(0,3)
\put(50,110){\line(1,0){160}}
\put(50,110){\line(0,-1){60}}
\put(20,110){\line(0,-1){60}}
\put(45,80){\vector(-1,0){15}}
\put(35,70){$\varpi$}
\put(-5,80){$\cp_1$}
\put(75,110){\line(2,-3){13}}
\put(85,100){\line(0,-1){40}}
\put(75,50){\line(2,3){13}}
\put(105,110){\line(2,-3){13}}
\put(115,100){\line(0,-1){40}}
\put(105,50){\line(2,3){13}}
\put(100,100){\makebox(0,0){$-2$}}
\put(100,60){\makebox(0,0){$-2$}}
\put(123,80){\makebox(0,0){$-1$}}
\put(140,110){\line(2,-3){13}}
\put(150,100){\line(0,-1){40}}
\put(140,50){\line(2,3){13}}
\put(175,110){\line(2,-3){13}}
\put(185,100){\line(0,-1){40}}
\put(175,50){\line(2,3){13}}
\put(122,44){\makebox(0,0){$-2$}}
\put(122,116){\makebox(0,0){$-2$}}
\put(230,80){\makebox(0,0){${M}$}}
\put(230,15){\makebox(0,0){$\cp_1$}}
\put(50,50){\line(1,0){160}}
\put(210,110){\line(0,-1){60}}
\put(135,40){\vector(0,-1){15}}
\put(50,15){\line(1,0){160}}
\put(75,15){\circle*{3}}
\put(105,15){\circle*{3}}
\put(140,15){\circle*{3}}
\put(175,15){\circle*{3}}
\end{picture}
\end{center}
(Note that
$\varpi$, which is depicted horizontally in the
above picture, is {\em not}
the ruled-surface projection, which is
here
depicted vertically.)
Contracting $8$ exceptional curves in a judicious order,
\begin{center}
\begin{picture}(240,110)(0,3)
\put(50,110){\line(1,0){160}}
\put(50,110){\line(0,-1){60}}
\put(80,110){\line(0,-1){60}}
\put(110,110){\line(0,-1){60}}
\put(145,110){\line(0,-1){60}}
\put(180,110){\line(0,-1){60}}
\put(80,50){\circle*{3}}
\put(145,50){\circle*{3}}
\put(80,54){\circle*{3}}
\put(145,54){\circle*{3}}
\put(110,110){\circle*{3}}
\put(180,110){\circle*{3}}
\put(110,106){\circle*{3}}
\put(180,106){\circle*{3}}
\put(122,44){\makebox(0,0){$0$}}
\put(122,116){\makebox(0,0){$0$}}
\put(240,80){\makebox(0,0){$\cp_1 \times \cp_1$}}
\put(230,15){\makebox(0,0){$\cp_1$}}
\put(50,50){\line(1,0){160}}
\put(210,110){\line(0,-1){60}}
\put(135,40){\vector(0,-1){15}}
\put(50,15){\line(1,0){160}}
\put(151,80){\makebox(0,0){$0$}}
\put(80,15){\circle*{3}}
\put(110,15){\circle*{3}}
\put(145,15){\circle*{3}}
\put(180,15){\circle*{3}}
\end{picture}
\end{center}
we get a rational ruled surface which can be
recognized as
$\bcp_1\times\bcp_1$ by inspecting self-intersections
of holomorphic curves. Our pencil now becomes
the sub-family of the linear system $|{\cal O}(2,2)|$
consisting of curves which are tangent to the second factor at
$4$ points in the claimed special position.
It follows that $\varpi : M \to \cp_1$ is
deformation equivalent to the
standard model of the rational elliptic surface,
since we may first deform our pencil into a generic
sub-pencil of $|{\cal O}(2,2)|$, and then
map $\cp_1\times\cp_1$ birationally
to $\cp_2$ in the usual way, using one
of the
base points of the pencil as our center.
This leads to the following:
\begin{prop} \label{eh}
Let $V$ be the (flat) 4-orbifold obtained from
${\Bbb R}\times T^3$ by dividing by the
involution induced by
$-1: {\Bbb R}^4\to {\Bbb R}^4$,
let $B=({\Bbb R}\times S^1)/{\Bbb Z}_2$ denote the analogous
2-orbifold, and let $\pi : V\to B$ be the
map induced by projection ${\Bbb R}\times T^3\to {\Bbb R}\times S^1$
to the first two coordinates. Then
a smooth model of the rational elliptic
surface $\cp_2\# 9\overline{\cp}_2 \to \cp_1$
may be obtained from $\pi : V\to B$ by
replacing each of the $8$ singular point of $V$ with a
2-sphere of self-intersection $-2$, forgetting the
orbifold structure on $B$, and adding
a smooth fiber at infinity in the
obvious manner.
\end{prop}
\begin{proof}
We have already seen that the rational elliptic
surface can be viewed as a desingularization
of $(S^2\times T^2)/{\Bbb Z}_2$. Now let
$x\in S^2$ be some point which is not fixed by
the involution, and let $x^\prime\neq x$
be its image under the involution.
Identify $S^2-\{x,x^\prime\}$ with
the cylinder ${\Bbb R}\times S^1$ in such a
manner that the involution becomes
simultaneous reflection in both factors. The result follows.
\end{proof}
The orbifold $V$ comes equipped with a family of
flat orbifold metrics. Replacing the
singularities with gravitational instantons yields
the main result of this section:
\begin{prop} Let $\tilde{V}$ denote the complement of
a generic fiber in the rational elliptic
surface $\cp_2\# 9 \overline{\cp}_2\to \cp_1$,
let $\pi: \tilde{V}\to {\Bbb C}$ be the induced elliptic
fibration, normalized so that all critical
values are contained in open unit disk $\Delta\subset {\Bbb C}$.
Let $f$ be any given flat metric on
the 2-torus $T^2$.
Then there is a family $g_t$, $t\in [1,\infty )$
of Riemannian metrics
on $\tilde{V}$
such that
\begin{itemize}
\item the Ricci curvature of $g_t$ converges uniformly to $0$ as
$t\to\infty$;
\item $(\pi^{-1} ({\Bbb C} - \Delta),g_t)$
is isometric to $([0,\infty )\times S^1, dx^2+d\theta^2)
\times (T^2, f/t)$; and
\item $\lim_{j\to \infty} \mbox{ Vol } (\pi^{-1}(\Delta), g_t ) = 0$.
\end{itemize}
\end{prop}
\begin{proof}
The basic idea is to replace the singularities of
of $V$ with Eguchi-Hanson metrics.
The {\em Eguchi-Hanson metric} \cite{eh,leb0} is a complete Ricci-flat
metric on the ${\cal O}(-2)$ line bundle over $\cp_1$,
or in other words on the blow-up of ${\Bbb C}^2/(\pm 1)$
at the origin.
Essentially by introducing polar coordinates on
$({\Bbb R}^4-0)/(\pm 1)$, we may think of this as
the completion of the metric on
$(1,\infty)\times SO(3)$ given by
$$g_{EH}=\frac{dr^2}{1-\frac{1}{r^4}}+ r^2\left( \sigma_1^2+\sigma_2^2+
(1-\frac{1}{r^4})\sigma_3^2\right)$$
where the left-invariant co-frame $\{ \sigma_j\}$ is
orthonormal for the curvature $+1$ bi-invariant metric.
By rescaling and homothety, we can turn this into
a 1-parameter family
$$\frac{dr^2}{1-\frac{A}{r^4}}+ r^2\left( \sigma_1^2+\sigma_2^2+
(1-\frac{A}{r^4})\sigma_3^2\right), ~~ r > \sqrt[4]{A}$$
of Ricci-flat metrics which bubble off to
the flat orbifold metric on ${\Bbb R}^4/(\pm 1)$
as $A\to 0$.
Let us now modify the Eguchi-Hanson metric. Let
$\phi (t)$ be a smooth monotonely non-increasing function
which is identically $1$ on
$[0,1]$ and identically $0$ on $[2,\infty ]$.
For small values of the parameter $A$,
the modified metric
$$\frac{dr^2}{1-\phi(r)\frac{A}{r^4}}+ r^2\left( \sigma_1^2+\sigma_2^2+
(1-\phi(r)\frac{A}{r^4})\sigma_3^2\right), ~~r > \sqrt[4]{A}$$
then agrees with the Euclidean metric for
$r > 2$, is Ricci-flat when $r < 1$, and has
Ricci curvature $< CA$ for some constant $C$ independent
of the small parameter $A$. Making
another homothety and rescaling, the metric
$$\frac{dr^2}{1-\phi(r/\varepsilon)\frac{A\varepsilon^4}{r^4}}+
r^2\left( \sigma_1^2+\sigma_2^2+
(1-\phi(r/\varepsilon)\frac{A\varepsilon^4}{r^4})\sigma_3^2\right),
~~r > \varepsilon \sqrt[4]{A}$$
is Euclidean for $r > 2\varepsilon$, Ricci-flat for $r < \varepsilon$,
and has Ricci curvature $< CA/\varepsilon^2$. Thus the metrics
$$g_\varepsilon=\frac{dr^2}{1-\phi(r/\varepsilon)
\frac{\varepsilon^8}{r^4}}+
r^2\left( \sigma_1^2+\sigma_2^2+
(1-\phi(r/\varepsilon)\frac{\varepsilon^8}{r^4})\sigma_3^2\right),
~~ r> \varepsilon^2$$
have Ricci curvature uniformly $O(\varepsilon^2)$
as $\varepsilon \searrow 0$.
Notice that the volume form of $g_\varepsilon$ coincides with the
Euclidean volume form
$r^3dr\wedge\sigma_1\wedge\sigma_2\wedge\sigma_3$,
but that the appropriate domain of integration is
the $r > \varepsilon^2$
rather than $r > 0$. Thus,
smoothing a flat orbifold singularity modeled
on ${\Bbb R}^4/(\pm 1)$ by gluing in the
modified Eguchi-Hanson metric $g_\varepsilon$
reduces the total volume, namely by $\pi^2\varepsilon^8/2$.
Now let $f$ be any fixed flat metric
on the 2-torus, and let $\imath$ denote the injectivity
radius of $(T^2,f)$, and
set $\hat{\imath}=\min (\imath , \pi)$.
Endow ${\Bbb R}\times T^3$ with the sequence
of flat metrics
$\hat{g_t}=dx^2+d\theta^2+ f/t$, and push these
down as flat orbifold metrics on $V=({\Bbb R}\times T^3)/{\Bbb Z}_2$.
For each $t$, let $\varepsilon_t = \hat{\imath}/4\sqrt{t}$,
so that the balls of radius $2\varepsilon_t$ centered at the
orbifold singularities of $V$ are pairwise disjoint.
On the blow-up $\tilde{V}$ of $V$, we can then define a
metric $g_t$ as the $\hat{g_t}$ on the complement of
these balls, and the modified Eguchi-Hanson metrics
$g_{\varepsilon_t}$ on the blown-up interior of these balls .
Since $\varepsilon_t\to 0$, the Ricci curvature of $g_t$
tends uniformly to zero. Moreover, the volume of
the region $|x| \leq a$ in $(X,g_t)$ is less than
$2\pi a \alpha/t \to 0$, where $\alpha$ is the area of
$(T^2,f)$, so the volume of the pre-image of any
compact set in $\Bbb C$ tends to $0$ as $t\to\infty$.
\end{proof}
\section{Collapsing Elliptic Surfaces}
\begin{defn}
Let $M$ be a smooth manifold. We will say that $M$
{\em collapses with bounded scalar curvature}
if there is a sequence $g_j$ of smooth metrics
on $M$ for which the absolute value
of the scalar curvatures is uniformly bounded,
but with total volume tending to zero:
$$|s_{g_j}| < {\cal B}, ~~\lim_{j\to \infty}\mbox{Vol}(M,g_j) =0.$$
Similarly, we say that $M$ collapses with
bounded Ricci (respectively, sectional)
curvature if there is a sequence $g_j$ of smooth metrics
on $M$ with
uniformly bounded Ricci (respectively, sectional)
curvature and total volume tending to zero.
\end{defn}
This collection of definitions is loosely inspired by the
work of Cheeger and Gromov \cite{cg}, who studied
sequences of metrics $g_j$ with bounded sectional curvature
and {\em injectivity radius} tending to zero at all points.
A standard comparison argument for the volume of
small balls shows that
a sequence of metrics with sectional
curvature bounded above can have volume tending with to zero
only if the point-wise injectivity radii
uniformly tend to zero, too.
Note, however, that a sequence
of metrics $g_j$ with bounded sectional curvature
and injectivity radius tending to zero may have volume
bounded away from zero --- as even happens for
flat metrics on the 2-torus.
Nonetheless, a remarkable result of Cheeger and Gromov \cite{cg2}
asserts
that if a compact manifold admits
a sequence of metrics $g_j$
with bounded sectional curvature and injectivity radius
uniformly tending to zero,
there must be other sequences
$\tilde{g}_j$ with bounded sectional curvature for which the
volume tends to zero as well. Thus the
present definition of collapse with bounded sectional
curvature
is in accord with the definitions used by others.
Collapse with bounded scalar curvature
is directly relevant to the computation
of Yamabe invariants by virtue of the
following result:
\begin{prop} \label{sam}
Let $M$ be a smooth compact $n$-manifold, $n\geq 3$. Then
the following are equivalent:
\begin{description}
\item{(i)} $M$ collapses with bounded scalar curvature;
\item{(ii)} $\inf_{g}\int_M |s_g|^{n/2}d\mu_g =0$;
\item{(iii)} $Y(M)\geq 0$.
\end{description}
\end{prop}
\begin{proof}
The implication $(i)\Rightarrow (ii)$ is trivial, since
$\int_M |s_{g_j}|^{n/2}d\mu_{g_j}\to 0$ for any sequence of
metrics with bounded scalar curvature and volume tending to
zero. The implication $(ii)\Rightarrow (iii)$ follows
from Proposition \ref{func}. Finally,
the implication $(iii)\Rightarrow (i)$
follows from the fact \cite{aubin} that any $n$-manifold, $n \geq 3$,
admits of metrics of constant
negative scalar curvature; if $Y(M)\geq 0$,
there is a thus a sequence of unit-volume metrics with
constant negative scalar curvature tending to $0$ from below,
and rescaling these metrics so they have $s\equiv -1$ then
makes the volumes of the rescaled metrics tend to zero.
\end{proof}
If $M$ is the underlying 4-manifold of a
complex elliptic surface, these various notions of
collapse can be completely understood.
\begin{thm}\label{scol}
Let $M$ be the underlying 4-manifold of a
complex elliptic surface. Then $M$ collapses with bounded
scalar curvature.
\end{thm}
\begin{thm}\label{rcol}
Let $M$ be the underlying 4-manifold of a
complex elliptic surface. Then $M$ collapses with bounded
Ricci curvature $\Leftrightarrow$
$M$ is minimal.
\end{thm}
\begin{thm}\label{kcol}
Let $M$ be the underlying 4-manifold of a
complex elliptic surface. Then $M$ collapses with bounded
sectional curvature $\Leftrightarrow$
$\chi(M)=0$.
\end{thm}
In order to prove these results, we will
need special diffeomorphic models \cite{FM,mats} of elliptic
surfaces. These are built up in stages, starting with
surfaces we shall call {\em twisted products}.
Let $E$ be an elliptic curve, equipped with a compatible flat
metric, let $o\in E$ be a chosen base-point,
and let $G$ be the finite group of
orientation-preserving isometries $E\to E$ which
fix $o$. (Thus $G$ is a cyclic group of order
$2$, $4$, or $6$.) A twisted product surface
is simply the total space $B$ of a fiber-bundle $B\to \Sigma$ over a
compact complex curve, with fiber $E$ and
structure group $G$.
Now let $B\to \Sigma$ be such a twisted product,
and let a finite collection of points $x_j\in \Sigma$
be specified, and assign each of these points
some integer multiplicity
$m_j > 1$. Give
$\Sigma$ an obifold structure by introducinging orbifold
local coordinates of the form $\hat{z}={z}^{1/m_j}$
near $x_j$, where
$z$ is a local complex coordinate on $\Sigma$ with $z(x_j)=0$.
A process called the {\em logarithmic transform}
now allows us to delete the fiber $E_j\cong E$ over each $x_j$,
and replace
it with $E_j/{\Bbb Z}_{m_j}$, where the ${\Bbb Z}_{m_j}$-action
is generated by any specified translation
$\tau_j : E_j\to E_j$ of order $m_j$. Namely, let
$\tau_j(t)$ be a 1-complex-parameter group of translations
of $E_{x_j}$
with $\tau_j (1)=\tau_j$, let $\Delta\subset {\Bbb C}$ be a disk centered
at $0$ and contained in the range of the orbifold coordinate
$\hat{z}$, and glue $(E_j\times \Delta)/\langle (\tau_j, e^{2\pi i/m_j})\rangle$
to $B-E_{x_j}$ by
$[(y,\tilde{z})]\leftrightarrow (\tau_j (-m_j\log (\hat{z})/2\pi i )(y),
z:=\hat{z}^{m_j})$.
With these constructions in mind, we can now
prove
\begin{lem} Let $M$ be the underlying 4-manifold of any
complex elliptic surface with Euler characteristic
$0$. Then there is a compact $2$-orbifold $\Sigma$
and a
smooth submersive map $\pi : M \to \Sigma$,
such that for every orbifold metric $h$ on $\Sigma$
there is
a Riemannian metric $g$ on $M$ for which
$\pi$ becomes a Riemannian submersion
with flat, totally geodesic
fibers. Moreover, $\pi$ can be chosen so
that its fibers over some open disk in $\Sigma$ are also
regular fibers of the original elliptic fibration of $M$.
\end{lem}
\begin{proof} Any complex elliptic surface
with Euler characteristic zero is
$C^\infty$-deformation equivalent to some logarithmic transform of
a twisted-product surface \cite[Proposition 7.2]{FM}.
But this model is an orbi-bundle over the $2$-orbifold $\Sigma$
with structure group equal to the isometry group of
a flat $2$-torus $E$. Starting from any
orbifold metric $h$ on $\Sigma$, now construct a metric
$g$ on $M$ by gluing together local product metrics
with a partition of unity subordinate to a trivializing
cover of $\Sigma$. Each fiber of $\pi: M\to \Sigma$
is then a flat quotient of $E$, and each fiber has vanishing
second fundamental form because the orthogonal space of
the fibers is the horizontal subspace of a connection
with structure group equal to the isometry group of the fiber.
\end{proof}
The ideas of Cheeger and Gromov \cite{cg}
now immediately lead to a
\bigskip
\noindent {\bf Proof of Theorem \ref{kcol}.}
Consider the metrics
$$g_t = \frac{1}{t}g+(1-\frac{1}{t})\pi^*h$$
on $M$ as $t\to\infty$. We have
$$\mbox{Vol}(M,g_t)= \frac{1}{t}\mbox{Vol}(M,g)\to 0,$$
so it suffices to show that the components of
the curvature tensor remain bounded in an orthonormal
frame as $t\to\infty$. But this follows from O'Neill's
Riemannian submersion formulas \cite{bes}. Indeed, the vertical
sectional curvature is $0$ for all $t$, and the
horizontal sectional curvature is given by
$$K(H)=K(\Sigma) - \frac{3}{4}g_t(v,v)\to K(\Sigma),$$
while the sectional curvature of the 2-plane $P$
spanned by a vertical vector and
a horizontal vector is
$$K(P)= \frac{1}{4}g_t(v^{\|},v^{\|}) \to 0.$$
Here $K(\Sigma)$ is
the Gauss curvature of $(\Sigma,h)$, and
$v$ is the ($t$-independent) vertical component of $[w_1,w_2]$,
where $w_1$ and $w_2$ are horizontal vectors which
project to an oriented orthonormal frame on $\Sigma$,
whereas $v^{\|}$ is the
orthogonal projection of $v$ into $P$.
This shows that the point-wise norm of the curvature tensor,
and hence all sectional curvatures,
remain uniformly bounded
as $t\to \infty$. Any elliptic surface $M$
with $\chi (M)=0$ therefore collapses
with bounded sectional curvature.
Conversely, if $M$ collapses with bounded sectional
curvature,
the Gauss-Bonnet formula tells one that
$\chi (M) =0$.
\qed
The next operation we will need in order to
create smooth models of elliptic surfaces is the
{\em fiber sum}. The fiber sum of two elliptic
surfaces is the smooth manifold obtained by
removing a tubular neighborhood of a regular fiber in
each and then then identifying the boundaries
in a manner compatible with the orientations and
given local trivializations.
Any minimal complex elliptic surface $M$ is
diffeomorphic \cite[Corollary 2.17]{FM} to a fiber sum of
an elliptic surface $\check{M}$
with Euler characteristic $0$
and copies of the rational elliptic surface.
\bigskip
\noindent {\bf Proof of Theorem \ref{rcol}.}
Let us thus begin with some elliptic surface $\check{M}$
with $\chi (\check{M})=0$, and let $\pi : \check{M}\to \Sigma$
be a smooth submersion of the type used in the previous proof.
Let $x_1, \ldots , x_k\in \Sigma$ be the points we will use as the
centers of the fiber sum construction. Let $U_1, \ldots , U_k$
be disjoint disks around these points, chosen so that
$M\to \Sigma$ may be trivialized over their closures,
and identify
each with the disk of radius $2$ about $0\in {\Bbb C}$. Choose $h$ so
that its restriction to the annulus $|z|\in [ 1, 2]$
is isometric to the cylinder $S^1\times [1,2]$, and now
construct the family of metrics $g_t$ used above, with
the stipulation $g$ is actually taken to be a product metric on
$\pi^{-1}(U_1), \ldots , \pi^{-1} (U_k)$.
Letting $\tilde{U}_1\subset U_1, \ldots , \tilde{U}_k\subset U_k$
each correspond to
the unit sub-disk $|z| < 1$,
the Riemannian
manifold-with-boundary $(M-\cup \pi^{-1}(\tilde{U}_j),g_t)$ has ends
isometric to $([0,1]\times S^1, dt^2+ d\theta^2) \times (E, f/t)$,
where $f$ is a fixed flat metric on an elliptic curve $E$.
However, these cylindrical ends
precisely match those of the metrics on $\tilde{V}$
constructed in Proposition \ref{eh}, and we can therefore
glue these objects together to obtain metrics with
bounded Ricci curvature and volume tending to
zero on the fiber sum $M$ of $\check{M}$ and $k$
rational elliptic surfaces. Since a diffeomorphic
model of any minimal elliptic surface can be
constructed in this way, it follows that any minimal
elliptic surface collapses
with bounded Ricci curvature, as claimed.
Conversely, suppose
an oriented 4-manifold $M$ admits
a sequence of metrics $g_j$ with bounded Ricci
curvature $r$ and volume tending to zero. We then
have
$$\lim_{j\to\infty} \int_M \left[ \frac{s_{g_j}^2}{24}-
\frac{|\stackrel{\circ}{r}|^2_{g_j}}{2}\right]d\mu_{g_j}
=0,$$
where $\stackrel{\circ}{r}$ denotes the
trace-free part of the Ricci curvature.
But
$$(2\chi + 3\tau)(M)= \frac{1}{4\pi^2}\int_M \left[ |W_+|^2+
\frac{s_{g}^2}{24}-
\frac{|\stackrel{\circ}{r}|^2_{g}}{2}\right]d\mu_{g}
\geq \frac{1}{4\pi^2}\int_M \left[ \frac{s_{g}^2}{24}-
\frac{|\stackrel{\circ}{r}|^2_{g}}{2}\right]d\mu_{g}$$
for any metric on $M$. It therefore follows that
$(2\chi + 3\tau )(M) \geq 0$.
On the other hand, $2\chi + 3\tau = c_1^2 \leq 0$ if
$M$ is an elliptic surface, and equality occurs iff
$M$ is minimal \cite{bpv}. Hence an elliptic
surface $M$ must be
minimal if it collapses with bounded Ricci curvature.
\qed
Finally, any elliptic surface is obtained from
a minimal elliptic surface by {\em blowing up}
an appropriate number of times.
Recall that the blow-up of ${\Bbb C}^2$ at the
origin is by definition the line bundle ${\cal O}(-1)$
over $\cp_1$. A point in an arbitrary complex surface can
similarly be
obtained by replacing a small ball with a
tubular neighborhood of
the zero section in this bundle. This procedure
is then iterated as needed to produce the desired
surface.
\bigskip
\noindent {\bf Proof of Theorem \ref{scol}.}
The idea is to graft Burns metrics onto the previous
examples. The {\em Burns metric} \cite{leb0} is a scalar-flat
K\"ahler metric on
the blow-up of ${\Bbb C}^2$ at the origin.
By introducing polar coordinates, this can be viewed
as the completion of the metric on $(1,\infty)\times S^3$
given by
$$g_{\mbox{\tiny
Burns}}= \frac{dr^2}{1-\frac{1}{r^2}} + r^2\left(\sigma_1^2+
\sigma_2^2+ (1-\frac{1}{r^2})\sigma_3^2\right) ,$$
where $\{\sigma_j\}$ is an $SU(2)$-invariant orthonormal
frame on $S^3$. By repeating the sequence of homotheties,
rescalings, and cut-offs used in the proof of
Proposition \ref{eh}, the metric
$$g_\epsilon =
\frac{dr^2}{1-\phi(\frac{r}{\epsilon})\frac{\epsilon^6}{r^2}}
+ r^2\left(\sigma_1^2+
\sigma_2^2+ (1-\phi(\frac{r}{\epsilon})\frac{\epsilon^6}{r^2}
)\sigma_3^2\right),~~
r > \epsilon^3
$$
is flat for $r > 2\epsilon$, scalar-flat for $r< \epsilon$,
and has scalar curvature uniformly $O(\epsilon^2)$
as $\epsilon \searrow 0$. Moreover, blowing up a
flat region by replacing a small ball by a
copy of $g_\epsilon$ reduces its total volume, namely
by $\pi^2\epsilon^{12}/2$.
Any elliptic surface is an iterated blow-up of a minimal
one. Up to diffeomorphism, we can therefore
model any such surface by blowing up one of
our previous models at distinct points. But the
constructed metrics on the previous models
can be taken to be flat in the neighborhood
of certain fibers
provided the orbifold metric
$h$ on $\Sigma$ is also taken to be flat on certain regions.
So let $\check{g}_j$ be a sequence of metrics with
bounded Ricci curvature and Vol $\to 0$,
each containing a Euclidean ball of radius $\varrho_j$,
on a minimal elliptic surface $\check{M}$.
On any
blow-up $M=\check{M}\# \ell \overline{\cp}_2$,
we can then
produce a sequence of metrics $g_j$,
with bounded scalar curvature and Vol $\to 0$
by replacing $\ell$ disjoint, Euclidean $(2{\epsilon_j})$-balls with
copies of $g_{\epsilon_j}$, where we may, for example,
take $\epsilon_j=\varrho_j/2\ell$.
\qed
By Proposition \ref{sam}, this immediately implies
\begin{cor}
Any complex elliptic surface $M$ has $Y(M) \geq 0$.
\end{cor}
The reader may prefer to deduce this corollary
directly from Theorem
\ref{rcol} and the
fact that $Y(\overline{\cp}_2) > 0$ by
citing a general result of \cite{okob}.
At heart, however, this is much the
same proof as is given above.
Combining this corollary with Theorem \ref{sw} now yields
\begin{thm} \label{y0}
Let $M$ be a K\"ahler-type complex surface
with $\kod (M)=0$ or $1$. Then $Y(M)=0$.
\end{thm}
\begin{proof}
Any surface of Kodaira dimension $0$ or $1$ is
deformation-equivalent to an elliptic surface \cite{bpv,FM},
and so has $Y(M)\geq 0$ by the previous result.
But if $b_1(M)=0\bmod 2$, $M$ is of K\"ahler type,
and Seiberg-Witten theory tells us that
there is no metric of positive scalar curvature
on $M$, so that $Y(M)\leq 0$. Hence $Y(M)=0$, as claimed.
\end{proof}
Combining
this with Proposition
\ref{rule} and Theorem \ref{gtype}
now proves Theorem \ref{main}.
\section{Concluding Remarks}
Theorem \ref{main} does not allow one to distinguish
between Kodaira dimensions $0$ and $1$ by
comparing Yamabe
invariants. Such a distinction arises
immediately, however, if one asks whether
$$Y(M)=\sup_{[g]} Y_{[g]}$$
is actually realized by the Yamabe constant
of some conformal class.
\begin{prop}
Let $M$ be the underlying 4-manifold of
a complex algebraic surface of Kodaira dimension
$\geq 0$. Then $M$ admits a Riemannian metric
of scalar curvature zero iff $M$ is minimal
and has Kodaira dimension $0$. In particular,
the Yamabe invariant $Y(M)$ is unachieved
whenever $\kod (M) =1$.
\end{prop}
\begin{proof} Any minimal K\"ahler-type
complex surface with
Kodaira
dimension $0$ has $c_1=0$ in
real cohomology, and so carries a
Ricci-flat K\"ahler metric by
Yau's solution \cite{yau} of the Calabi
conjecture. Such a metric has
$s=0$, and so realizes $Y(M)=0$.
Conversely \cite{bes}, if a manifold
with $Y(M)=0$ admits a scalar-flat metric,
the metric in question must be Ricci-flat,
since otherwise it could be deformed into
a metric of positive scalar curvature \cite{bes}.
In dimension 4, this implies that
$$2\chi + 3\tau =\frac{1}{4\pi^2}\int_M\left[2|W_+|^2+\frac{s^2}{24}
-\frac{|\stackrel{\circ}{r}|^2}{2}\right]d\mu\geq 0.$$
Our complex surface must therefore satisfy $c_1^2\geq 0$.
If $c_1$ is not a torsion class, the Seiberg-Witten
invariant of $(M,[J])$ is therefore both metric-independent
and non-zero, even if $b^+(M)=1$.
But the Seiberg-Witten
estimate \cite{leb2}
$$\int_Ms^2d\mu \geq 32\pi^2 (c_1^+)^2$$
then forces $c_1$ to be anti-self-dual with
respect to any scalar-flat metric; thus
$c_1^2\leq 0$, with equality iff $c_1=0$
in real cohomology. Hence $c_1$ is a torsion class,
and our surface must be a minimal surface of
Kodaira dimension $0$.
\end{proof}
While this makes a satisfactory distinction between
{\em minimal} surfaces of Kodaira dimensions $0$ and $1$,
it is still unsatisfactory for non-minimal surfaces.
However, it is not hard to see that there are
sequences of unit-volume constant-scalar-curvature
metrics on any blow-up of, say, a K3 surface
which bubble off to any given K\"ahler-Einstein
metric on the minimal model. The metrics in this
sequence have uniformly bounded diameter.
It seems plausible to conjecture that this does
not happen in the case of Kodaira dimension $1$:
\begin{conj}
Let $M$ be the underlying 4-manifold of a
complex algebraic surface of Kodaira dimension
$\geq 0$. Suppose that there is a sequence
of unit-volume constant-scalar-curvature
metrics on $M$ with uniformly bounded diameter and
$s\nearrow 0$. Then $\kod (M) =0$.
\end{conj}
Of course,
beyond simply knowing the sign of $Y(M)$, one would
like to ask to know its actual value. When $b_1(M)$ is even and
$\kod (M) \geq 0$, this question is completely answered by
Theorems \ref{gtype} and \ref{y0}. For
algebraic surfaces with $\kod (M) =-\infty$,
however, the only case in which $Y(M)$ is actually
known is for $M=\cp_2$, where $Y(M)$ is actually realized
by the Fubini-Study metric \cite{leb4}. It is, in particular,
unclear whether blowing up leaves $Y(M)$ unchanged when $\kod (M)
=-\infty$, although this is what happens when the
Kodaira dimension is larger.
There are a number of places in this
paper where we have assumed that the surface in question
has $b_1$ even in order to assure that there is
a non-trivial Seiberg-Witten invariant. However,
this is superfluous in Kodaira dimension $0$, since
Kodaira surfaces are symplectic and so have non-trivial
invariants by the work of Taubes \cite{taubes}.
For Kodaira dimension $1$ the situation is less clear,
but Biquard \cite{biq} has also shown that some non-K\"ahler surfaces
in this class have non-trivial Seiberg-Witten invariants.
Thus it would be surprising if the following
were not true:
\begin{conj}
Let $M$ be the underlying 4-manifold of a compact
complex manifold of Kodaira dimension $0$ or $1$.
Then, even if $b_1(M)$ is odd, one still has $Y(M)=0$.
\end{conj}
Even allowing for the case of $b_1$ odd, the following
is an immediate consequence of the results proven here:
\begin{thm}
Let $(M,J)$ be any compact
complex surface. If $Y(M) < 0$, then
$M$ is either of general type or of type $VII$.
\end{thm}
For all known surfaces of type $VII$,
however, one can check by hand that
$Y \geq 0$. It therefore seems reasonable to make the following
\begin{conj}
Let $(M,J)$ be any compact
complex surface. If $Y(M) < 0$, then
$M$ is of general type.
\end{conj}
Finally, we have accidentally computed the infima of some
other Riemannian functionals by showing that minimal elliptic
surfaces collapse with bounded Ricci curvature.
\begin{prop}
Let $M$ be the underlying 4-manifold of any
complex elliptic surface.
Then
$$\inf_{[g]} \int_M |W_+|^2d\mu =0.$$
Here $W_+$ denotes the self-dual Weyl
tensor, and the infimum is over the set of all
conformal classes of Riemannian metrics on $M$.
As a consequence
$$\inf_{[g]} \int_M |W_-|^2d\mu = -12\pi^2\tau (M).$$
\end{prop}
Indeed, if $M$ is minimal, this follows from
Theorem \ref{rcol} and the Gauss-Bonnet formula
for
$(2\chi + 3\tau )(M) = 0$. Since the Burns metric
is anti-self-dual, it is easy to extend this
to the non-minimal case; cf. \cite{taubes2}.
By Taubes existence theorem \cite{taubes2} for anti-self-dual
metrics, any complex surface, after being blown up
sufficiently many times, admits metrics with
$W_+=0$. The last result raises, once again, the
fascinating question of determining
precisely how many times a given complex surface
must be blown up before this happens.
\vfill
\noindent
{\bf Acknowledgement.}
The author warmly
thanks Michael Anderson, Peter Kronheimer,
and Heberto del Rio
for their helpful comments and suggestions.
\begin{thebibliography}{99}
\bibitem{and} M.T. Anderson,
{\em Extrema of Curvature Functionals on the Space of Metrics
on 3-Manifolds}, preprint, 1995.
\bibitem{aubin} T. Aubin, {\bf Nonlinear analysis on Manifolds.
Mong-Amp\`ere
Equations}, Springer-Verlag, 1982.
\bibitem{bpv} W. Barth, C. Peters and A. Van de Ven,
{\bf Compact Complex Surfaces}, Springer-Verlag, 1984.
\bibitem{bes} A. Besse, {\bf Einstein Manifolds}, Springer-Verlag,
1987.
\bibitem{bcg} G. Besson, G. Courtois, and S. Gallot,
{\em Les Vari\'et\'es Hyperboliques Sont des
Minima Locaux de l'Entropie
Topologique}, {\bf Inv. Math. 117} (1994) 403--445.
\bibitem{biq} O. Biquard, {\em Les \'Equations de Seiberg-Witten
sur une Surface Complexe Non K\"ahl\'erienne}, preprint, 1995.
\bibitem{cg} J. Cheeger and M. Gromov, {\em Collapsing Riemannian
Manifolds While Keeping Their Curvature Bounded I}, {\bf J. Diff.
Geom. 23} (1986), 309--346.
\bibitem{cg2} J. Cheeger and M. Gromov, {\em Collapsing Riemannian
Manifolds While Keeping Their Curvature Bounded II},
{\bf J. Diff.
Geom. 32} (1990) 269--298.
\bibitem{eh} T. Eguchi and A.J. Hanson,
{\em Asymptotically Flat Self-Dual Solutions
to Euclidean Gravity},
{\bf Phys. Lett. 74B} (1978) 249--251.
\bibitem{FM1} R. Friedman and J. Morgan,
{\bf Smooth 4-Manifolds and Complex Surfaces},
Springer-Verlag, 1994.
\bibitem{FM} R. Friedman and J. Morgan,
{\em Algebraic Surfaces and Seiberg-Witten Invariants}, preprint
alg-geom/9502026.
\bibitem{FQ} R.
Friedman and Z. Qin, {\em The Smooth Invariance of
the Kodaira Dimension of a Complex Surface},
{\bf Math. Res. Lett. 1}
(1994) 369--376.
\bibitem{GL} M. Gromov and H.B. Lawson,
{\em The Classification of Simply Connected Manifolds
of Positive Scalar Curvature},
{\bf Ann. of Math. 111} (1980) 423--434.
\bibitem{okob} O. Kobayashi, {\em Scalar Curvature of
a Metric of Unit Volume}, {\bf Math.\ Ann.\ 279} (1987) 253--265.
\bibitem{leb0} C. LeBrun, {\em Counter-Examples
to the Generalized Positive Action Conjecture},
{\bf Comm.\ Math.\ Phys.\
118} (1988) 1205--1208.
\bibitem{leb1} C. LeBrun, {\em On the Scalar Curvature
of Complex Surfaces}, {\bf Geom.\ Func.\ An.\ 5}
(1995) 619--628.
\bibitem{leb2} C. LeBrun, {\em Polarized 4-Manifolds, Extremal
K\"ahler Metrics, and
Seiberg-Witten Theory}, {\bf Math.\ Res. Lett.\ 2} (1995)
653-662.
\bibitem{leb3} C. LeBrun, {\em Four-Manifolds without
Einstein Metrics}, {\bf Math.\ Res.\ Lett.
3} (1996) 133--147.
\bibitem{leb4} C. LeBrun, {\em Yamabe Constants and
the Perturbed Seiberg-Witten Equations}, {\bf Comm. An. Geom.}
to appear.
\bibitem{lp} J. Lee and T. Parker, {\em The Yamabe Problem},
{\bf Bull. Am. Math. Soc. 17} (1987) 37--91.
\bibitem{mats} Y. Matsumoto,
{\em Diffeomorphism Types of Elliptic Surfaces}, {\bf Topology 25}
(1986) 549--563.
\bibitem{rick} R. Schoen, {\em Conformal Deformation of a
Riemannian Metric to Constant Scalar Curvature},
{\bf J. Diff. Geom. 20}
(1984) 478-495.
\bibitem{sch} R. Schoen, {\em Variational Theory for the
Total Scalar Curvature Functional for Riemannian Metrics and
Related Topics}, {\bf Lec. Notes Math. 1365} (1987) 120--154.
\bibitem{taubes2} C.H. Taubes, {\em
The Existence of Anti-Self-Dual Metrics},
{\bf J.\ Diff.\ Geometry 36} (1992) 163--253.
\bibitem{taubes} C.H. Taubes, {\em The Seiberg-Witten Invariants and
Symplectic Forms}, {\bf Math.\ Res.\ Lett.\ 1} (1994) 809--822.
\bibitem{witten} E. Witten, {\em Monopoles and Four-Manifolds},
{\bf Math.\ Res.\ Lett.\ 1} (1994) 809--822.
\bibitem{yau} S.-T. Yau, {\em On the Ricci-Curvature of a Complex
K\"ahler Manifold
and the Complex Monge-Amp\`ere Equations}, {\bf Comment. Pure Appl.
Math.\ 31}
(1978) 339--411.
\end{thebibliography}
\end{document}