ON THE

2
-SCALAR CURVATURE
YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
Abstract. In this paper, we establish an analytic foundation for a fully non-linear
equation

2

1
= f on manifolds with positive scalar curvature and apply it to give a
(rough) classication of such manifolds. A crucial point is a simple observation that this
equation is a degenerate elliptic equation without any condition on the sign of f and it
is elliptic not only for f > 0 but also for f < 0. By dening a Yamabe constant Y
2,1
with respect to this equation, we show that a positive scalar curvature manifold admits
a conformal metric with positive scalar curvature and positive
2
-scalar curvature if and
only if Y
2,1
> 0. We give a complete solution for the corresponding Yamabe problem.
Namely, let g
0
be a positive scalar curvature metric, then in its conformal class there is
a conformal metric with

2
(g) =
1
(g),
for some constant . Using these analytic results, we give a rough classication of the
space of manifolds with positive scalar curvature metrics.
1. Introduction
Let (M, g
0
) be a compact Riemannian manifold with metric g
0
and [g
0
] the conformal
class of g
0
. Let Ric
g
and R
g
denote the Ricci tensor and scalar curvature of a metric g
respectively. The Schouten tensor of the metric g is dened by
S
g
=
1
n 2
_
Ric
g

R
g
2(n 1)
g
_
.
The importance of the Schouten tensor in conformal geometry can be viewed in the fol-
lowing decomposition of the Riemann curvature tensor
Riem
g
= W
g
+S
g
_g,
where _ is the Kulkani-Nomizu product. Note that g
1
W
g
is invariant in a given
conformal class. Therefore, in a conformal class the Schouten tensor is important.
The objective of this paper has two-folds. First, we study a class of fully nonlinear
quotient equations relating the scalar curvature and the recently introduced
2
-scalar
curvature. Then, we apply the analysis established for this class of equations to study
the space of metrics with positive scalar curvature. Let us rst recall the denition of the

2
-scalar curvature.
For a given 1 k n, the
k
-scalar curvature or k-scalar curvature is dened by

k
(g) :=
k
(g
1
S
g
),
where g
1
S
g
is locally dened by (g
1
S
g
)
i
j
=

k
g
ik
(S
g
)
kj
and
k
is the kth elementary
symmetric function. Here for an nn symmetric matrix A we dene
k
(A) =
k
(), where
1
2 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
= (
1
, ,
n
) is the set of eigenvalues of A. It is clear that
1
(g) is a constant multiple
of the scalar curvature R
g
. The
k
-scalar curvature
k
(g), which was rst considered by
Viaclovsky [59], is a natural generalization of the scalar curvature.
In this paper, we focus on the
2
-scalar curvature. One of the reason to restrict ourself
to
2
-scalar curvature is that
2
(g) still has a variational structure. This is an observation
of Viaclovsky [59].
1
The variational structure is very crucial for our paper.
Since [59] and [9], there has been an intensive study for a nonlinear Yamabe problem
related to the
k
-scalar curvature
k
(g), namely, nding a conformal metric g in a given
conformal class [g
0
] satisfying
(1)
k
(g) = c,
where
(2) g [g
0
]
+
k
.
Here
+
k
is a convex open cone -the Garding cone- dened by

+
k
= = (
1
,
2
, ,
n
) R
n
[
j
() > 0, j k.
By g
+
k
we mean that the Schouten tensor S
g
(x)
+
k
for any x M. g
+
1
is equivalent to that g has positive scalar curvature. Note that the condition g
+
k
guarantees that equation (1) is elliptic.
Equation (1) is a fully nonlinear conformal equation, but it admits many nice properties,
which usually are only true for the semilinear equations. For example, the local a priori
estimates was established in [25], a Liouville type theorem was given in [44]. The existence
problem of (1) has been intensively studied. For the existence results see [10], [17], [18],
[21], [26], [28], [30], [32], [44], [45], [46] [53], [58], [61]. See also a recent survey by Viaclovsky
[62] and references therein or the lecture notes of Guan [20]. There are many interesting
applications in geometry, especially in the 4-dimensional case, see for examples [9], [11],
and [63]. See also [21], [22].
However, till now one can only deal with equations like
(3)
k
(g) = f
with condition f 0 or even f > 0. This has something to do with the ellipticity of
the corresponding equation. With this restriction we could not use equation (3) to study
negative
2
-scalar curvature. (cf. the work of Gursky and Viaclovsky [33] for metrics in

and a class of uniformly elliptic fully nonlinear equations [28].) In this paper, instead
of
2
(g) we consider the quotient type equation
(4)

2

1
= f.
When f > 0, it was studied recently in [18]. See also [27], [30], [23] and [53]. The crucial
point of this paper is that we can also deal with the case with negative functions f for (4).
In fact, we have
1
For k > 2,
k
(g) has a variational structure if and only if the underlying manifold is locally conformally
at ([59],[6]).
ON THE
2
-SCALAR CURVATURE 3
Observation. The operator

2

1
is elliptic in the cone
+
1

1
and concave in
+
1
.
For the notation and the proof of this Observation, see Lemma 1. Recall that the
ordinary Yamabe constant is dened
Y
1
([g
0
]) = Y
1
([g
0
]) = inf
gC
1
([g
0
])
_

1
(g)dvol(g)
(vol(g))
n2
n
.
In this paper we will call it the rst Yamabe constant. For simplicity of notation, we
denote
+
k
[g
0
] by (
k
([g
0
]), or even just (
k
if there is no confusion. A conformal class has
positive rst Yamabe constant if and only if this class contains a metric of positive scalar
curvature. This is a simple, but important fact in the study of the metrics of positive
scalar curvature. As one of applications of our study of equation (4), we will prove a
similar result for the
2
-scalar curvature. From now on we consider the conformal class
[g] with (
1
,= and call a metric with positive scalar curvature psc metric. Now we rst
dene a nonlinear eigenvalue for
2
(g), more precisely for a fully nonlinear operator
(5)
2
(
2
u +du du
[u[
2
2
g
0
+S
g
0
).
Dene
(6) (g
0
,
2
) = (M, g
0
,
2
) :=
_

_
inf
gC
1
([g
0
])
_

2
(g)dvol(g)
_
e
4u
dvol(g)
, if n > 4,
_

2
(g)dvol(g), if n = 4,
sup
gC
1
([g
0
])
_

2
(g)dvol(g)
_
e
4u
dvol(g)
, if n = 3.
One can also dene a similar constant (g
0
,
k
) for
k
. When k = 1, one can check that
(g
0
,
k
) is the rst eigenvalue of the conformal Laplacian. Hence we call the constant
(g
0
,
2
) nonlinear eigenvalue of operator (5). Such a nonlinear eigenvalue for fully non-
linear operator was rst considered in [48]. In the context of fully nonlinear conformal
operators, it was rst considered in a preliminary version of [26].
Now we dene the second conformal Yamabe constant for a conformal class with psc
metrics.
Y
2,1
([g
0
]) = Y
2,1
([g
0
]) :=
_

_
inf
gC
1
([g
0
])
_

2
(g)dvol(g)
(
_

1
(g)dvol(g))
n4
n2
, if n > 4,
_

2
(g)dvol(g), if n = 4,
sup
gC
1
([g
0
])
_

2
(g)dvol(g)
_

1
(g)dvol(g), if n = 3.
We emphasize that the inmum (or supremum) in the denition of
2
and Y
2,1
is taken
over (
1
, not over (
2
. It is well-known that when n = 4,
_
M

2
(g)dvol(g) is a constant in a
4 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
given conformal class. The relationship between the sign of (g
0
,
2
) and Y
2,1
(g
0
]) will be
discussed in the next section.
Now we state our main analytic results in this paper .
Theorem 1. The constant (g
0
,
2
) is achieved, provided (g
0
,
2
) 0. More precisely,
we have
1) If (g
0
,
2
) > 0, then (
2
([g
0
]) is not empty and (g
0
,
2
) > 0 is achieved by a
smooth metric g = e
2u
g
0
(
2
satisfying
(7)
2
(g) = e
4u
.
2) If (g
0
,
2
) = 0, then (g
0
,
2
) = 0 is achieved by a C
1,1
metric g = e
2u
g
0
(
1
satisfying

2
(g) = 0.
Here (
1
is the closure of (
1
. Namely, (
1
is the space of metrics with non-negative scalar
curvature. With the help of Theorem 1 and results in [10], [32] and [18] we can solve the
Yamabe problem for

2

1
completely.
Theorem 2. Let (M
n
, g
0
) be a compact Riemannian manifold with g
0

+
1
and n 3.
The followings hold
1) If Y
2,1
([g
0
]) > 0, then (
2
([g
0
]) is not empty. Moreover there is a smooth metric g
in (
2
satisfying
(8)

2
(g)

1
(g)
= 1.
2) If Y
2,1
([g
0
]) = 0, then there is a C
1,1
metric g in (
1
satisfying satisfying
(9)
2
(g) = 0.
3) If Y
2,1
([g
0
]) < 0, then there is a smooth metric g in (
1
satisfying
(10)
2
(g) =
1
(g).
In fact, we can show that Y
2,1
is achieved when n > 3. This will be carried out elsewhere.
We remark that in case 3) though the solution metric g is in (
1
, it is smooth. One can
show that any point x M with
2
(g)(x) = 0 (and hence
1
(g)(x) = 0) is Ricci-at, i.e.,
Ric(x) = 0. We believe that g (
1
.
A direct consequence of the main analytic results is
Corollary 1. Let (M
n
, g
0
) be a closed Riemannian manifold of dimension n > 2 with
positive Yamabe constant Y
1
([g
0
]) > 0. If Y
2,1
([g
0
]) > 0, then there is a metric g [g
0
]
with g
+
2
, i.e., with
R
g
> 0 and
2
(g) > 0.
Note that Y
2,1
[g
0
]) > 0 if and only if (g
0
,
2
) > 0. See Lemma 3 below. When n = 4,
the result was proved by Chang-Gursky-Yang [9]. In another direct proof was given in
[31]. See also [21] for locally conformally at manifolds. Many interesting applications to
geometry were given in these papers, especially in [11]. In 3-dimension, together with the
Hamiltons result [34] obtained by the Ricci ow, the Corollary gives
ON THE
2
-SCALAR CURVATURE 5
Corollary 2. Let (M
3
, g
0
) be a compact 3-dimensional Riemannian manifold. If Y
1
([g
0
]) >
0 and Y
2,1
([g
0
]) > 0, then M is dieomorphic to a quotient of the sphere.
We remark that the method given in [31] and [21] can be used to give another proof of
Corollary 1 at least for n > 4. In another direction, one may ask if the non-emptyness of

+
2
implies the positivity of Y
2,1
. This question is not easy to answer by only using the
methods given in [31] and [21]. In [18], we showed that the non-emptyness of
+
2
implies
the positivity of
inf
gC
2
(
1
_

1
(g)dvol(g))
)
n4
n2
_

2
(g)dvol(g).
It is clear that Y
2,1
is less than or equals to the above constant. With the analysis
established here we can show that the non-emptyness of
+
2
implies the positivity of Y
2,1
.
Theorem 3. Let (M, g) be a compact Riemanniann manifold with positive scalar curva-
ture. The second Yamabe constant is positive if and only if
+
2
is non-empty.
We remark again that the case n = 4 was given in [9]. As mentioned above, it is well-
known that the positivity of the rst Yamabe constant Y
1
is equivalent to the existence of
a conformal metric of positive scalar curvature. Theorem 3 means that Y
2,1
has a similar
property. Motivated by this result, we hope to use
2
-scalar curvature to give a further
classication of the manifolds admitting metric with positive scalar curvature. This is
the second aim of this paper. For the further discussion let us rst recall the following
denition.
(1
+
) Closed connected manifolds with a Riemannian metric whose scalar curvature is
non-negative and not identically 0.
(1
0
) Closed connected manifolds with a Riemannian metric with non-negative scalar
curvature, but not in class (1
0
).
(1

) Closed connected manifolds not in classes (1

+
) or (1
0
).
There is a remarkable result of Kazdan and Warner obtained in 1975.
Theorem A (Trichotomy Theorem) ([40], [41]) Let M
n
be a closed connected manifold
of dimension n 3.
1. If M belongs to class (1
+
), then every smooth function is the scalar curvature
function for some Riemannian metric on M.
2. If M belongs to class (1
0
), then a smooth function f is the scalar curvature function
of some Riemannian metric on M if and only if f(x) < 0 for some point x M,
or else f = 0. If the scalar curvature of some g vanishes identically, then g is
Ricci at.
3. If M belongs to class (1

), then a smooth function f is the scalar curvature function

of some Riemannian metric on M if and only if f(x) < 0 for some point x M.
The analysis used in the proof of Theorem A is based on the analysis for the eigenvalue
problem for the conformal Laplacian operator
(11) u +
n 2
4(n 1)
R
g
u = u.
6 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
And it is closely related to the famous Yamabe equation
(12) u +
n 2
4(n 1)
R
g
u = u
n+2
n2
.
The existence of solutions for (12) is the so-called Yamabe problem, which was solved by
Yamabe, Trudinger, Aubin and Scheon completely.
From Theorem A or an Earlier result of Aubin [2] we know that a negative function can
always be realized as a scalar function of a metric. See also the results of [16] and [49] for
the existence of negative Ricci curvature. The class of (1
0
) is very small and consisting
of very special manifolds, thanks to a result of Futaki [14]. Theorem A also implies that
class (1
+
) is just the class of manifolds which admits a metric of positive scalar curvature.
There are topological obstructions for the manifolds with positive scalar curvature, see
[47] and [37]. This class attracts much attention of geometers for many years, especially
after the work of Gromov-Lawson[19] and Schoen-Yau [52]. The most important problem
in this eld is the Gromov-Lawson-Rosenberg conjecture which was proved by Stolz [54]
in the simply connected case. For this conjecture, see for instance [50] and [56].
Now using
2
-scalar curvature we divide (1
+
) further into 3 subclasses:
(2
+
) Closed connected manifolds admitting a psc metric whose
2
-scalar curvature is
non-negative and not identically 0.
(2
0
) Closed connected manifolds admitting a psc metric with non-negative
2
-scalar
curvature, but not in class (2
+
).
(2

) Closed connected manifolds in (1

+
), but not in classes (2
+
) or (2
0
).
Analog to the Trichotomy Theorem of Kazdan-Warner for the scalar curvature and in
view of the analysis established here, we propose the following
Conjecture (Trichotomy Theorem) Let M
n
(n > 2) be a closed connected manifold
in class (1
+
).
1. If M belongs to class (2
+
), then for every smooth function f there is a psc metric
g such that f
1
(g) is its
2
-scalar curvature.
2. If M belongs to class (2
0
), then a smooth function f admits a psc metric g with

2
(g) = f
1
(g) if and only if f(x) < 0 for some point x M, or else f = 0.
3. If M belongs to class (2

), then a smooth function f admits a psc metric g with

2
(g) = f
1
(g) if and only if f(x) < 0 for some point x M.
Though we could not now prove this conjecture at moment, we have the following results
support this conjecture.
Theorem 4. Any Riemannian manifold M
n
(n 4) in the class (1
+
) admits a metric in
(
1
with non-positive
2
-scalar curvature.
The metric given in Theorem 4 satises
_

2
(g)dvol(g) < 0 and has vanishing Ricci
curvature at points with
1
(g)(x) = 0.
Theorem 5. Let M
n
be a closed connected manifold of dimension n > 3.
1. If M belongs to class (2
+
), then for every constant b there is a metric g (
1
such
that b
1
(g) is its
2
-scalar curvature.
ON THE
2
-SCALAR CURVATURE 7
2. If M belongs to class (2
0
), then a constant b admits a metric g (
1
with
2
(g) =
b
1
(g) if and only if b 0.
3. If M belongs to class (2

), then a constant b admits a metric g (

1
with
2
(g) =
b
1
(g) if and only if b < 0.
The metrics found in Theorem 5 may have points with vanishing
1
-scalar curvature. At
such points its Ricci tensor also vanishes. When b > 0, then the metric has positive scalar
curvature. It is trivial that the sphere belongs to (2
+
). By Corollary 2, in 3-dimension,
only the sphere and its quotients belong to (2
+
) and there is no manifolds in (2
0
). In
4-dimension, many examples of manifolds with metrics in
+
2
can be found in [9]. Those
manifolds certainly belong to (2
+
). One can prove that S
3
S
1
belongs to (2
0
), namely
there is no positive scalar metric on S
3
S
1
with positive
2
-scalar curvature, see Section
7 below. Like class (1
0
), class (2
0
) should be very small. However, when n > 4, we have
no example of manifolds in class (1
+
) which do not belong to (2
+
). This is somewhat
strange. A possible candidate is the manifold S
6
H
3
, where H
3
is a compact quotient
of the hyperbolic space H
3
. It is easy to check that its product metric g
P
satises that

1
(g
P
) > 0 and
2
(g
P
) = 0. However, we believe that that S
6
H
3
has a metric g with

1
(g) > 0 and
2
(g) > 0. See Example 2 in Section 7. Therefore we may ask
Problem. Is there a topological obstruction for the existence of psc metric with positive

2
-scalar curvature?
For the scalar curvature as mentioned above there are topological obstructions. What
we ask is to nd further conditions to distinguish the manifolds between (2
+
), (2
0
) and
(2
+
) for higher dimensional manifolds. There is a similar and related problem proposed by
Stolz in [55] for further obstructions to the existence of metrics with positive Ricci tensor.
For some relationship between the positive Ricci tensor and positive
k
-scalar curvature,
see [24]. With our analysis, the problem to nd a topological obstruction for
2
-scalar
curvature perhaps might be not very dicult.
The paper is organized as follows. In Section 2, we show the Observation and discuss
the relationship between (g
0
,
2
) and Y
2,1
. We introduce a class of perturbated equations
(27) and a Yamabe type ow (30) in Section 3 and establish local priori estimates for these
equations and ows in Section 4. In Section 5 we prove the global existence of the Yamabe
type ow and Theorem 1 and Theorem 2. One of another crucial points of this paper, the
Yamabe type ow preserves the positivity of the scalar curvature will also be proved in this
section. In Section 6, we show Theorem 3. We give the geometric applications in Section
7 by proving Theorems 4 and 5. In the last section, we mention further applications in
similar equations.
2. Some preliminary facts
Let o
n
be the space of n n real symmetric matrices and F a smooth function in o
n
.
By extending F to the whole space of n n real matrices by F(A) = F(
1
2
(A + A
t
)), we
8 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
view F as a function of n n variables w
ij
and dene
F
ij
=
F
w
ij
.
The quotient function

2

1
can be viewed as a function in o
n
as follows. Let W be a
symmetric matrix and
W
=
1
,
2
, ,
n
its the eigenvalues. Then we dene

2

1
(W) =

2
(
W
)

1
(
W
)
. A symmetric matrix is said to be in
+
k
if
W

+
k
. Let
1
be the subspace of
o
n
consisting of symmetric matrices of rank 1.
Lemma 1. For 1 < k n set F =

k

k1
. We have
1) the matrix (F
ij
)(W) is semi-positive denite for W
+
k1
and is positive denite
for W
+
k1

1
.
2) The function F is concave in the cone
+
k1
. When k = 2, for all W
+
1
and for
all R = (r
ij
) o
n
, we have
(13)

ijkl

2
w
ij
w
kl
_

2
(W)

1
(W)
_
r
ij
r
kl
=

ij
(
1
(W)r
ij

1
(R)w
ij
)
2

3
1
(W)
.
Proof. For = (
1
,
2
, ,
n
) and i 1, 2, , n let
i
= 1, 2, ,

i, , n be the
(n 1)-tuple obtained from without the ith-component. A direct calculation gives
(14)
F

i
=

k1
(
i
)
k1
()
k
()
k2
(
i
)

k1
()
2
.
Since
+
k1
,
i

+
k2
. For the proof see for instance [38]. From two identities

k
() =
i

k1
(
i
) +
k
(
i
) and
k1
() =
i

k2
(
i
) +
k1
(
i
),
(14) becomes
(15)
F

i
=

2
k1
(
i
)
k
(
i
)
k2
(
i
)

k1
()
2
.
Note that we set
0
() = 1. By the Newton-McLaughlin inequality
(k 1)(n k)
2
k1
(
i
) k(n k + 1)
k
(
i
)
k2
(
i
),
we have
F

i
0
and equality implies that
i
= 0, 0, , 0. This proves 1).
2) was proved in [38].
ON THE
2
-SCALAR CURVATURE 9
Though Lemma 1 is a rather simple fact
2
, as mentioned in the introduction it is one
of crucial points of our paper. It means that

2

1
is a concave, degenerate elliptic operator
in
+
1
. With this observation in mind, we consider a family of perturbed operators for a
positive function R
+
(16)
F

:
+
1
R
W F

(W) =

2
(W)

1
(W)
.
As a direct consequence of Lemma 1, we have the following:
Lemma 2. The matrix (F
ij

)(W) is positive denite for all W

+
1
and the function F

is strictly concave in
+
1
. Moreover, for all W
+
1
and for all R = (r
ij
) o
n
(R)
(17)

ijkl

2
F

(W)
w
ij
w
kl
r
ij
r
kl

2

3
1
(W)
_

i
r
ii
_
2
.
Lemma 2 means that for any function > 0, the operator F

is elliptic and strictly

concave.
Now we discuss the relationship between the sign of the invariants Y
2,1
([g
0
]) and that
of (g
0
,
2
) dened in the introduction.
Lemma 3. The eigenvalue (g
0
,
2
) is a nite number. Moreover, we have
(1) if n 4, then (g
0
,
2
) > 0 (resp. = 0, < 0 ) if and only if Y
2,1
([g
0
]) > 0 (resp.
= 0, < 0);
(2) if n = 3, then (g
0
,
2
) > 0 (resp. 0 ) if and only if Y
2,1
([g
0
]) > 0 (resp. 0).
Proof. The following formula was given in the proof of the Sobolev inequality in [17]: for
any g = e
2u
g
0
[g
0
] we have
(18)
2
_

2
(g)dvol(g) =
n 4
2
_

1
(g)[u[
2
g
0
e
2u
dvol(g) +
n 4
4
_
[u[
4
g
0
e
4u
dvol(g)
+
_

i,j
T
ij
S(g
0
)
ij
dvol(g) +
_

i,j
S(g
0
)
ij
u
i
u
j
dvol(g)
+
1
2
_

1
(g
0
)[u[
2
g
0
e
4u
dvol(g),
2
We remark that this Lemma might be observed also by other mathematicians. For example in [13]
there is a rather similar formula. We would like to thank Tobias Lamm who told us this reference.
10 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
where T
ij
=
1
(W)g
ij
W
ij
is the rst Newton transformation associated with W. From
this formula, we have
(19)
2
_

2
(g)dvol(g) =
n 4
2
_

1
(g)[u[
2
g
0
e
2u
dvol(g) +
n 4
4
_
[u[
4
g
0
e
4u
dvol(g)
+
_
e
2u

1
(g)
1
(g
0
)dvol(g)
_
e
4u
[S(g
0
)[
2
g
0
dvol(g)
+(4 n)
_

i,j
S(g
0
)
ij
u
i
u
j
dvol(g) +
_

1
(g
0
)[u[
2
g
0
e
4u
dvol(g)
+
_
e
4u
u,
1
(g
0
))
g
0
dvol(g).
We rst consider the case n 5. By (19), there exists some c > 0 such that
(20)
_

2
(g)dvol(g)
n 4
16
_
[u[
4
g
0
e
4u
dvol(g) c
_
e
4u
dvol(g),
provided that g
+
1
. Here we have used the fact

_

i,j
S(g
0
)
ij
u
i
u
j
dvol(g)

1
8
_
[u[
4
g
0
e
4u
dvol(g) + 2 sup
M
[S(g
0
)[
2
g
0
_
e
4u
dvol(g)
and

_
e
4u
u,
1
(g
0
))
g
0

_

1
(g
0
)[u[
2
g
0
e
4u
dvol(g) +c
_
e
4u
dvol(g).
As a consequence, we conclude
(g
0
,
2
) c.
Similarly, in case n = 3, we have for all g (
1
([g
0
])
(21)
_

2
(g)dvol(g) =
1
4
_

1
(g)[u[
2
g
0
e
2u
dvolg
1
8
_
[u[
4
g
0
e
4u
dvol(g)

1
4
_
e
4u
[u[
2
g
0

1
(g
0
)dvol(g)
+
1
2
_
e
4u
(
2
1
(g
0
) [S(g
0
)[
2
g
0
)dvol(g)
+
1
2
_

i,j
S(g
0
)
ij
u
i
u
j
dvol(g)

1
16
_
[u[
4
g
0
e
4u
dvol(g) +c
_
e
4u
dvol(g),
ON THE
2
-SCALAR CURVATURE 11
which yields (g
0
,
2
) c. Thus, we nish the proof of the rst part of Lemma. It is easy
to see that for all g (
1
([g
0
])
(22)
_
M

1
(g)dvol(g) =
_ _
n 2
2
[u[
2
+
1
(g
0
)
_
e
2u
dvol(g) Y
1
([g
0
])(V ol(g))
n2
n
.
Now we consider the case n 5. Suppose (g
0
,
2
) > 0, i.e.,
_
M

2
(g) (g
0
,
2
)
_
M
e
4u
dvol(g)
for any g = e
2u
g
0

+
1
. From (20), (22) and Holders inequality, we deduce
(23)
_

2
(g)dvol(g) c(
_
[u[
4
g
0
e
4u
dvol(g) +
_
e
4u
dvol(g))
c(
_
M

1
(g)dvol(g))
2
(V ol(g))
1
cY
1
([g
0
])
n
n2
(
_
M

1
(g)dvol(g))
n4
n2
,
which implies Y
2,1
([g
0
]) > 0. Inversely, using Holders inequality and (22), we have
(24)
_

2
(g)dvol(g) c(
_
M

1
(g)dvol(g))
n4
n2
cY
1
([g
0
])
n4
n2
(V ol(g))
n4
n
c
_
e
4u
dvol(g).
This gives the desired result. Clearly, we have (g
0
,
2
) < 0 if and only if Y
2,1
([g
0
]) < 0.
Consequently, (g
0
,
2
) = 0 if and only if Y
2,1
([g
0
]) = 0. In the cases n = 4 and n = 3, the
result is trivial.
Remark 1. In the case n = 3, if (g
0
,
2
) = < 0, then Y
2,1
([g
0
]) < 0. To see this, for
any g (
1
([g
0
]), it follows from (21) and (22) that there holds
(25)
_

1
(g)dvol(g)
_

2
(g)dvol(g)
_

1
(g)dvol(g)
_
e
4u
dvol(g)
c
_
e
u
dvol(g
0
)
_
e
u
dvol(g
0
) c < 0.
12 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
Remark 2. In the case n 5, the invariant Y
2,1
([g
0
]) is nite real number. To see this,
for any g (
1
([g
0
]) it follows from the Holders inequality, (19) and (22) that
(26)
_

2
(g)dvol(g)
(
_

1
(g)dvol(g))
n4
n2
c
_
e
4u
dvol(g)
(
_

1
(g)dvol(g))
n4
n2
c
(V ol(g))
n4
n
(
_

1
(g)dvol(g))
n4
n2
c > .
In the case n = 3, we do not know if Y
2,1
([g
0
]) < + or not, although it is always true
that Y
2,1
([g
0
]) > 0 if and only if (g
0
,
2
) > 0, and that (g
0
,
2
) is nite. However, we
believe that it is true.
3. Yamabe type flows
Now we want to consider the existence of the following equation
(27) F

(g) =

2
(g) e
4u

1
= constant,
with g = e
2u
g
0
for > 0. Note that is a positive function. In this paper we will choose
as a small positive constant. Following [26], [17] and [18] we will introduce a suitable
Yamabe type ow to study equation (27).
For any (0, +) and for g = e
2u
g
0
, consider the following perturbed functional
c

(g) :=
_

_
2
n 4
_
M
(
2
(g) e
4u
)dvol(g), if n ,= 4,

_
1
0
_
M
(
2
(g
t
) 2e
4tu
)udvol(g
t
)dt, if n = 4,
where g
t
= e
2tu
g
0
. When = 0, the functional was considered in [59], [9] and [7]. Set
T
1
(g) =
_
M

1
(g)dvol(g) and T
2
(g) =
_
M

2
(g)dvol(g)
From the variational formula given in [59], [9] and [7], we have
(28)
d
dt
c

(g) =
_
(
2
(g) e
4u
)g
1

d
dt
gdvol(g)
and
(29)
d
dt
T
1
(g) =
n 2
2
_

1
(g)g
1

d
dt
gdvol(g).
Now we introduce a Yamabe type ow, which non-increases c

and preserves T
1
.
(30)
du
dt
=
1
2
g
1
d
dt
g = e
2u

2
(g) e
4u

1
(g)
r

(g)e
2u
+s

(g),
ON THE
2
-SCALAR CURVATURE 13
where r

(g) and s

(g) are space constants, given by

(31) r

(g) :=
T
2
(g)
_
M
e
4u
dvol
g
T
1
(g)
and
(32)
_
M

1
(g)
_
e
2u

2
(g) e
4u

1
(g)
r

(g)e
2u
+s

(g)
_
dvol(g) = 0.
Lemma 4. Flow (30) preserves T
1
and non-increases c

. Hence when n 4, then r

is
non-increasing along the ow, and when n = 3, then r

is non-decreasing along the ow.

Proof. By the denition of s

(g) and (29), ow (30) preserves T

1
. By the denition of s

and r

, we can compute as follows

(33)
d
dt
c

(g) =
_
M
(
2
(g) e
4u
)g
1

d
dt
gdvol(g)
= 2
_
e
2u

1
(g)
_
e
2u

2
(g) e
4u

1
(g)
r

(g)e
2u
_
2
dvol(g).
Therefore, the desired result yields.
4. Local estimates
In this section, we will establish a priori estimates for ow (30) and equations (8), (9)
and (10). Local estimates for this class of fully nonlinear conformal equations was rst
given in [25]. Since then there are many extensions. See for instance [12] and the survey
paper [62]. It is important to note that the a priori estimates established below do not
depend on the perturbation > 0.
Given > 0, assume g
0
(
1
([g
0
]). By Lemma 2, (30) is parabolic. By the standard
implicit function theorem we have the short-time existence result. Let T

(0, ] so that
[0, T

) is the maximum interval for the existence of the ow g(t)

+
1
.
Theorem 6. Assume that n 3, > 0 and g
0

+
1
. Let u be a solution of (30) in a
geodesic ball B
R
[0, T] for T < T

and R <
0
, the injectivity radius of M.
(1) Assume that t [0, T], there holds
r

(t) 0.
Then there is a constant C depending only on (B
R
, g
0
) (independent of and T)
such that for any (x, t) B
R/2
[0, T]
(34) [u[
2
+[
2
u[ C.
(2) Assume that t [0, T], there holds
r

(t) > 0.
14 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
Then there is a constant C depending only on (B
R
, g
0
) (independent of and T)
such that for any (x, t) B
R/2
[0, T]
(35) [u[
2
+[
2
u[ C(1 + sup
t[0,T]
r

(t) e
2 inf
(x,t)B
R
[0,T]
u(x,t)
).
In particular, if we assume n 4, we have
(36) [u[
2
+[
2
u[ C(1 +e
2 inf
(x,t)B
R
[0,T]
u(x,t)
).
Proof. Let W = (w
ij
) be an n n matrix with w
ij
=
2
ij
u + u
i
u
j

|u|
2
2
(g
0
)
ij
+ (S
g
0
)
ij
.
Here u
i
and u
ij
are the rst and second derivatives of u with respect to the background
metric g
0
. Recall F

(W) =

2
(W)

1
(W)
. Set
(37)
(F
ij

(W)) :=
_
F

w
ij
(W)
_
=
_

1
(W)T
ij

2
(W)
ij
+
ij

2
1
(W)
_
where (T
ij
) = (
1
(W)
ij
w
ij
) is the rst Newton transformation associated with W, and

ij
is the Kronecker symbol. In view of Corollary 2 we see that (F
ij

) is positive denite
and F

is concave in
+
1
. For the simplicity of notations, we now drop the index , if there
is no confusion. We try to show the local estimates for rst and second order derivatives
together. Let S(TM) denote the unit tangent bundle of M with respect to the background
metric g
0
. We dene a function

G : S(TM) [0, T] R
(38)

G(e, t) = (
2
u +[u[
2
g
0
)(e, e)
Without loss of generality, we assume R = 1. Let C

0
(B
1
) be a cut-o function dened
as in [25] such that
(39)
0, in B
1
,
= 1, in B
1/2
,
[(x)[ 2b
0

1/2
(x), in B
1
,
[
2
[ b
0
, in B
1
.
Here b
0
> 1 is a constant. Since e
2u
g
0

+
1
, to bound [u[ and [
2
u[ we only need to
bound (
2
u +[u[
2
g
0
)(e, e) from above for all e S(TM) and for all t [0, T]. For this
purpose, denote G(e, t) = (x)

G(e, t). Assume (e
1
, t
0
) S(T
x
0
M) [0, T] such that
G(e
1
, t
0
) = max
S(TM)[0,T]
G(e, t), (40)
t
0
> 0, (41)
G(e
1
, t
0
) > nmax
B
1

1
(g
0
). (42)
ON THE
2
-SCALAR CURVATURE 15
Let (e
1
, , e
n
) be a orthonormal basis at point (x
0
, t
0
). It follows from the fact W
+
1
nG(e
1
, t
0
) (u +n[u[
2
) (n[u[
2
+
n 2
2
[u[
2

1
(g
0
)),

3n 2
2
[u[
2

1
n
G(e
1
, t
0
),
so that
G(e
1
, t
0
)
3n2
2
n +
1
n
[u[
2

21
20
[u[
2
.
Consequently, we obtain
(43)
2
11
u(x
0
, t
0
)
1
20
[u[
2
(x
0
, t
0
)
Set for any i ,= j = 1, , n
e

=
1

2
(e
i
e
j
).
We have
(44) G(e

, t
0
) =
1
2
(G(e
i
, t
0
) +G(e
j
, t
0
))
2
ij
u(x
0
, t
0
).
Thus, there holds
(45) [
2
ij
u(x
0
, t
0
)[ G(e
1
, t
0
)
1
2
(G(e
i
, t
0
) +G(e
j
, t
0
)).
On the other hand, we have i = 1, , n
(46) (n 1)G(e
1
, t
0
) +G(e
i
, t
0
) (u +n[u[
2
) (
3n 2
2
[u[
2

1
(g
0
)),
which implies
(47) G(e
i
, t
0
) (
3n 2
2
[u[
2

1
(g
0
)) (n 1)G(e
1
, t
0
).
Together with (45), we deduce
(48) [
2
ij
u(x
0
, t
0
)[ nG(e
1
, t
0
)
3n 2
2
[u[
2
+
1
(g
0
) (n + 1)G(e
1
, t
0
).
(Indeed, at all point (x, t), the estimate [
2
ij
u[ (n+1)G(e
1
, t
0
) holds). Now choose the
normal coordinates around x
0
such that at point x
0

x
1
= e
1
and consider the function on M [0, T]
G(x, t) = (x)(u
11
+[u[
2
)(x, t).
16 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
(without the confusion, we denotes also this function by G). Clearly, (x
0
, t
0
) is a maximum
point of G(x, t) on M [0, T]. At (x
0
, t
0
), we have
0 G
t
= (u
11t
+ 2

l
u
l
u
lt
), (49)
0 = G
j
=

j

G+(u
11j
+ 2

l1
u
l
u
lj
), for any j, (50)
0 (G
ij
) =
_
_

ij
2
i

2
G+(u
11ij
+

l1
(2u
li
u
lj
+ 2u
l
u
lij
))
_
_
. (51)
Recall that (F
ij
) is denite positive. Hence, we have
(52)
0

i,j1
F
ij
G
ij
G
t

i,j1
F
ij

ij
2
i

2
G+

i,j1
F
ij
(u
11ij
+

l1
(2u
li
u
lj
+ 2u
l
u
lij
))
(u
11t
+ 2

l1
u
l
u
lt
).
First, from the denition of , we have
(53)

i,j1
F
ij

ij
2
i

2
G C

i,j1
[F
ij
[
1

G,
and
(54)

i,j1
[F
ij
[

i
F
ii
=
_
n 1
n
2
(W)

2
1
(W)
_
+
n

2
1
(W)
C

i,j1
[F
ij
[
since W is positive denite . Using the facts that
(55) u
kij
= u
ijk
+

m
R
mikj
u
m
,
(56) u
kkij
= u
ijkk
+

m
(2R
mikj
u
mk
Ric
mj
u
mi
Ric
mi
u
mj
Ric
mi,j
u
m
+R
mikj,k
u
m
)
and
(57) (

l
u
2
l
)
11
= 2

l
(u
11l
u
l
+u
2
1l
) +O([u[
2
),
ON THE
2
-SCALAR CURVATURE 17
we have
(58)

i,j1
F
ij
u
11ij

i,j1
F
ij
_
_
w
ij11
(u
11
)
i
u
j
u
i
(u
11
)
j
+

l1
(u
2
1l
+u
11l
u
l
)(g
0
)
ij
_
_
2

i,j1
F
ij
u
i1
u
j1
C(1 +[
2
u[ +[u[
2
)

i,j1
[F
ij
[
and
(59)

i,j,l
F
ij
u
l
u
lij

i,j,l
F
ij
u
l
w
ijl

i,j,l
F
ij
(u
l
u
il
u
j
+u
l
u
i
u
jl
)
+
1
2

i,j
F
ij
u, ([u[
2
))(g
0
)
ij
C(1 +[u[
2
)

i,j1
[F
ij
[.
Combining (58) and (59), we deduce
(60)

i,j1
F
ij
(u
11ij
+ 2

l1
(u
li
u
lj
+u
l
u
lij
))

i,j1
F
ij
(w
ij11
+ 2

l1
w
ijl
u
l
) + 2

i,j1
F
ij

l2
u
li
u
lj
+

i,j,l1
u
2
1l
F
ij
(g
0
)
ij

i,j
F
ij
_
(u
11
+[u[
2
)
i
u
j
+u
i
(u
11
+[u[
2
)
j
u, (u
11
+[u[
2
))(g
0
)
ij

C(1 +[
2
u[ +[u[
2
)

i,j1
[F
ij
[

i,j
F
ij
(w
ij11
+ 2

l
w
ijl
u
l
) +u
2
11

i,j
F
ij
(g
0
)
ij
+

i,j
F
ij
(
i
u
j
+
j
u
i
, u)(g
0
)
ij
)
G

2
C(1 +[
2
u[ +[u[
2
)

i,j1
[F
ij
[.
Now, we want to estimate

i,j,l
F
ij
w
ijl
u
l
and

i,j
F
ij
w
ij11
respectively. Using Lemma
1, there holds
(61)

i,j,l
F
ij
w
ijl
u
l
=

l
F
l
u
l
,
and
(62)

i,j
F
ij
w
ij11
= F
11

i,j,k,m

2
F
w
ij
w
km
w
ij1
w
km1
F
11
.
18 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
Therefore, these estimates give
(63)

i,j1
F
ij
(w
ij11
+ 2

l1
w
ijl
u
l
) F
11
+ 2

l
F
l
u
l
.
Recall from (30) that
(64) F = u
t
+r

(g)e
2u
s

(g),
so that
(65) F
11
= u
11t
+r

(g)e
2u
(2u
11
+ 4u
2
1
),
(66) F
l
= u
lt
+r

(g)e
2u
(2u
l
), l = 1, , n.
Gathering (43), (52), (53), (54), (60) (63), (65) and (66), we obtain
(67)
0 C
_
_

i,j

F
ij

_
_
G

+
_

i
F
ii
_
u
2
11
C
_
_

i,j

F
ij

_
_
(1 +[u[
2
+[
2
u[)
+

i,j
F
ij
(
i
u
j
+
j
u
i
, u)(g
0
)
ij
)
G

+r

(g)e
2u
_
2u
11
4
n

l=2
u
2
l
_
C
_

i
F
ii
_
G

+
_

i
F
ii
_
u
2
11
C
_

i
F
ii
_
G[u[

C
_

i
F
ii
_
( +G) ,
since r

(g) 0 for any t [0, T]. As a consequence, there holds

(68) C
2
CGCG

G+
2
u
2
11
.
Recall (43) holds at point (x
0
, t
0
) so that
(69) G(x
0
, t
0
) = (u
11
+[u[
2
)(x
0
, t
0
) 21(u
11
)(x
0
, t
0
).
Together with (68), we deduce
(70) C
2
CGCG

G+
G
2
21
2
,
which implies
(71) C CG+
G
2
2 21
2
.
This yields
(72) G C.
Here C is a constant independent of T and . Therefore, we have nished the proof of the
rst part of the Theorem. The second part yields from (67) and Lemma 4.
ON THE
2
-SCALAR CURVATURE 19
The same proof gives the local estimates for the elliptic equation.
Corollary 3. Assume n 3, > 0 and g
0

+
1
. Let B
R
be a geodesic ball for R <
0
, the
injectivity radius of M. Assume that e
2u
g
0

+
1
is a solution of the following equation
in B
R
(73)

2
(g) e
4u

1
(g)
= ,
for some constant .
(1) If 0, then there is a constant C depending only on (B
R
, g
0
) (independent of
and T) such that for any (x, t) B
R/2
[0, T]
(74) [u[
2
+[
2
u[ C.
(2) If > 0, then there is a constant C depending only on (B
R
, g
0
) and (independent
of and T) such that for any (x, t) B
R/2
[0, T]
(75) [u[
2
+[
2
u[ C(1 +e
2 inf
(x,t)B
R
[0,T]
u(x,t)
).
Corollary 4. Under the same hypotheses as in Theorem 6 there is a constant C depending
only on g
0
(independent of ) such that for any t [0, T

)
(76) |u|
C(M)
+|
2
u|
C(M)
C.
Corollary 5. Under the same hypotheses as in Theorem 6 with r

0, then there is a
constant C depending only on g
0
(independent of ) such that for any t [0, T

)
(77) |u|
C
2
(M)
C.
Proof. First, for any t [0, T

) we have
(78) T
1
(g(t)) T
1
(g(0)).
Hence the following Sobolev inequality
_
M

1
(g)dvol(g) Y
1
([g
0
])(V ol(g))
n2
n
, g
+
1
,
implies that the volume V ol(g(t)) along the ow is bounded from above, i.e.,
(79)
_
M
e
nu
dvol(g
0
) < C,
for some constant C > 0. On the other hand, applying Corollary 4, we have
_
M
e
2u
dvol(g) C
_
M

1
(g)dvol(g) = CT
1
(g(0)) > 0.
This, together with (79) and Corollary 4, implies u is uniformly bounded. In view of
Corollary 4, we prove the result.
We remark that in this paper we only use (1) of Theorem 6.
20 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
5. Proof of Theorems 1 and 2
Now for > 0 we dene
(80) a

:=
_

_
inf
gC
1
([g
0
])
c

(g)
(
_
M

1
(g)dvol(g))
n4
n2
if n ,= 4;
_
M

2
(g)dvol(g) if n = 4;
If a

is achieved by a metric g = e
2u
g
0
, the g satises
(81)

2
(g) e
4u

1
(g)
= ,
for some constant . Equivalently, we will consider the energy functional c

on the nor-
malized cone

(
1
([g
0
])
(82)

(
1
([g
0
]) :=
_
g (
1
([g
0
]) [
_
M

1
(g)dvol(g) = 1
_
.
The rst step is to solve the perturbed equation (81).
Proposition 1. Let > 0. In the following three cases
1) when n 5, a

< 0,
2) when n = 4, a

< 0, which is equivalent to

_
M

2
(g)dvol(g) < 0,
3) when n = 3, a

> 0,
ow (30) globally converges to a solution of (81). As a direct application, a

is achieved
by a function u

satisfying (81) for = 1. Moreover, such a solution is unique.

Proof. The proof follows closely the proof given in [17]. When n 5, without loss of
generality we choose g
0

+
1
such that c

(g
0
) < 0 and
_
M

1
(g
0
)dvol(g
0
) = 1. Using
Lemma 4, we have r

(g(t)) r

(g(0)) < 0, t [0, T

). If n = 3, or n = 4, by the
hypotheses, we have r

(g(t)) < c < 0, t [0, T

). Applying Corollary 5, the solution u

of ow (30) has a uniform C
2
bound, which is independent of t. We divide the proof into
3 steps.
Step 1. The ow preserves the positivity of the scalar curvature. This is another crucial
point of this paper.
Proposition 2. There is a constant C
0
> 0, independent of T [0, T

) such that

1
(g(t)) > C
0
for any t [0, T].
ON THE
2
-SCALAR CURVATURE 21
Proof. The proof follows closely the proof given in [26] and [17]. Recall
W = (w
ij
) = (
2
ij
u +u
i
u
j

[u[
2
2
(g
0
)
ij
+ (S
g
0
)
ij
),
F

(W) =

2
(W)

1
(W)
.
Hence, F

= u
t
+ r

(g(t))e
2u
s

(g(t)). Without loss of generality, we assume that the

minimum of F

is achieved at (x
0
, t
0
) M (0, T]. Near (x
0
, t
0
), we have
(83)
d
dt
F

ij
A
ij
(
2
g
(u
t
))
ij
=

ij
A
ij
_
(
2
g
(F

))
ij
r

(g)(
2
g
(e
2u
))
ij

,
where
A
ij
:=
F

w
ij
=
(
2
1
(W)
2
(W) +)
ij

1
(W)W
ij

2
1
(W)
is positive denite. To simplify the notations, we drop the index as before. Since (x
0
, t
0
)
is the minimum of F in M [0, T], at this point, we have
dF
dt
0, F
l
= 0 l and
(F
ij
) is non-negative denite. Note that
(
2
g
)
ij
F = F
ij
+u
i
F
j
+u
j
F
i

l
u
l
F
l

ij
= F
ij
,
at (x
0
, t
0
), where F
j
and F
ij
are the rst and second derivatives with respect to the back-
ground metric g
0
. From the positivity of A and (83), we have
(84)
0 F
t

i,j
A
ij
F
ij
r

(g)

i,j
A
ij
(e
2u
)
ij
+u
i
(e
2u
)
j
+u
j
(e
2u
)
i

l
u
l
(e
2u
)
l

ij

= r

(g)e
2u

i,j
A
ij
2w
ij
+ 2u
i
u
j
+ 2S(g
0
)
ij
+[u[
2

ij

(g)e
2u
_
2
2
(W) 2

1
(W)
_
r

(g)e
2u

i,j
A
ij
(2u
i
u
j
+ 2S(g
0
)
ij
+[u[
2

ij
).
Here we have used

i,j
A
ij
w
ij
=

2
(W) +

1
(W)
. On the other hand, we have
(85)

i,j
A
ij
S(g
0
)
ij
=
(
2
1
(W)
2
(W))
1
(g
0
)

2
1
(W)

1

1
(W)

i,j
W
ij
S(g
0
)
ij
+

1
(g
0
)

2
1
(W)
.
22 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
Going back to (84), we have
(86)
0 F
t

i,j
A
ij
F
ij
r

(g)e
2u
_
2
2
(W) 2

1
(W)
+
2(
2
1
(W)
2
(W))
1
(g
0
)

2
1
(W)

1
(W)

i,j
W
ij
S(g
0
)
ij
+
2
1
(g
0
)

2
1
(W)
_
_
,
since (A
ij
) is positive denite and r

(g) < c is negative. Let us use O(1) denote terms

with a uniform bound. One can check
2
(g) = O(1) for |u|
C
2 is uniformly bounded and

i,j
W
ij
S(g
0
)
ij
= O(1). Also the term
2
1
(W)
2
(W) is always non-negative. From (86),
we conclude that there is a positive constant C
2
> 0 (independent of T) such that
(87)
1
(W)(x
0
, t
0
) > C
2
> 0.
Now for any (x, t) M [0, T], we have

2
(W)(x, t)

1
(W)(x, t)
O(1)
so that there is a positive constant C > 0, independent of T, such that

1
(W)(x, t) C
since we have
2
(W)
1
2

2
1
(W) provided
2
(W) 0. This nishes the proof of the
Proposition and this step.
Step 2. Now we can prove equation (81) admits a solution. From Step 1, we know that
the ow is uniformly parabolic. And, Krylovs theory implies u(t) has a uniform C
2,
bound. Hence, T

= . One can also show that u(t) globally converges to u(), which
clearly is a solution of (81) for k = r

(g()). (Note that r

(g(t)) is monotone and bounded

so that r

(g()) exists) (see [26]). So u

= u()
1
2
log [r

(g())[ solves (81) for = 1.

Step 3. The solution to equation (81) for k = 1 is unique.
Assume u
1
and u
2
are two solutions. We consider the function v = u
1
u
2
, which solves
the following second order elliptic equation
(88)

i,j
A
ij
(x)v
ij
+

i
B
i
v
i
+m(x)v = 0,
where
A
ij
(x) =
_
1
0
F
ij
(tW
2
+ (1 t)W
1
)dt,
ON THE
2
-SCALAR CURVATURE 23
W
l
(x) =
_
(u
l
)
ij
+ (u
l
)
i
(u
l
)
j

1
2
[u
l
[
2
(g
0
)
ij
+ (S(g
0
))
ij
_
, for l = 1, 2
and
m(x) = 2
_
1
0
e
2((1t)u
1
(x)+tu
2
(x))
dt < 0.
This a uniformly elliptic equation. Now applying the classical strong maximum principle,
we deduce that
v 0.
Hence we have the uniqueness. It is clear that the unique solution achieves a

. Thus, we
nish the proof.
In the following, we study a nonlinear eigenvalue problem for the operator
(89)
2
(
2
u +du du
[u[
2
2
g
0
+S
g
0
)
in
+
1
. The nonlinear eigenvalue problem for the operator (89) in
+
2
was considered in
the rst version of [26]. In that paper, the nonlinear eigenvalue problem can only be
considered in
+
2
. With our analysis established for
2
/
1
, we can consider the nonlinear
eigenvalue problem for the operator (89) in a larger class
+
1
.
Proposition 3. Let (M
n
, g
0
) be a compact Riemannian manifold with g
0

+
1
and n 3.
Assume that the rst eigenvalue (g
0
,
2
) > 0. Then (
2
([g
0
]) is not empty. Namely there
exists a metric in [g
0
] with

1
(g) > 0 and
2
(g) > 0.
Moreover, there is a regular metric g = e
2u
g
0
(
2
([g
0
]) satisfying
(90)

2
(g) = e
4u
,
for = (g
0
,
2
) > 0.
Proof. We dene a function h
h : R (0, +]
h() = a

:=
_

_
inf
gC
1
([g
0
])
c

(g)
(
_
M

1
(g)dvol(g))
n4
n2
if n ,= 4;
_
M

2
(g)dvol(g) if n = 4;
Thus, h is a non-increasing function if n 4 and a non-decreasing function if n = 3. First,
we consider the case n 4. Dene a set A
(91) A := (0, +)[ h() < 0.
By the assumption that (g
0
,
2
) > 0, it is easy to check that
(92) ((g
0
,
2
), +) A and (0, (g
0
,
2
)) A = .
24 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
Hence for any > (g
0
,
2
) we have h() < 0. By Proposition 1 we have a smooth
metric g
u

= e
2u

g
0
(
1
([g
0
]), which solves equation (81) for = h() and satises
_
M

1
(g
u

)dvol(g
u

) = 1. From Corollary 5, the set of solutions

u

, ((g
0
,
2
), (g
0
,
2
) + 1)
is uniformly bounded in C
2
norm. From equation (81),
1
(g
u

) is uniformly bounded from

below by a positive constant for ((g
0
,
2
), (g
0
,
2
) + 1), since > (g
0
,
2
) > 0 and
also we have

2
(W)

1
(W)

1
2

1
(W), provided that
1
(W) > 0. By the Krylovs result, the set
of solutions u

, ((g
0
,
2
), (g
0
,
2
) + 1) is also uniformly bounded in C
2,
norm
for > 0. Dene
(93)
0
:= lim
(g
0
,
2
)
+
h().
When (g
0
,
2
), u

(by taking a subsequence) converges in C

2
norm to u
0

+
1
,
which is a solution of (81) for = (g
0
,
2
) and =
0
. Clearly,
0
0. We claim
(94)
0
= 0.
Otherwise, c
(g
0
,
2
)
(u
0
) < 0, which implies that there is some small > 0 such that
((g
0
,
2
), (g
0
,
2
)], we have c

(u
0
) < 0. Hence h() < 0 for all ((g
0
,
2
), +),
that is, ((g
0
,
2
) , +) A. This contradicts to (92). Hence we have (94), which is
equivalent to say that

2
(g
u
0
) (g
0
,
2
)e
4u
0
= 0.
This means that u
0

+
2
solves (90) for = (g
0
,
2
). In particular, u
0

+
2
.
The proof for the case n = 3 is similar by consider a set A dened by
A := (0, +)[ h() > 0.
We leave the proof to the interested reader.
Remark 3. In the case n 4, the function h() is Lipschitz continuous. Indeed, for all
<

and for all g (

1
([g
0
]), we have
(95)
0 < c

(u) c

(u) = (

)
_
e
4u
dvol(g) c(

)(V ol(g))
n4
n
cY
1
([g])
4n
n2
(

)
__

1
(g)dvol(g)
_n4
n2
This proves the claim.
Proof of Theorem 2. First we discuss the case Y
2,1
([g
0
]) > 0. Using Lemma 3, we have
(g
0
,
2
) > 0. By Proposition 3, the cone
+
2
is not empty. Hence, from the results in [18]
for the cases n 5, in [10] for n = 4 and [32] for the case n = 3, there is a solution of (8).
Now suppose Y
2,1
([g
0
]) = 0. As in the proof of Proposition 3, we consider the function
h() for all > 0. With the same arguments, for any small positive > 0, equation (90)
ON THE
2
-SCALAR CURVATURE 25
admits the unique solution u

, since h() < 0 > 0 if n 4 and h() > 0 > 0 if

n = 3. This fact is clear for n 4. And in the case n = 3, we have
inf
gC
1
([g
0
])
_

1
(g)dvol(g)
_
e
4u
dvol(g) > 0.
Moreover, as 0, the family of solution u

is bounded in C
2
norm. Thus, as 0,
u

converges in C
1,
norm for any (0, 1) to some C
1,1
function u. Consequently, this
C
1,1
conformal metric g
u
= e
2u
g
0
[g
0
]

+
1
solves equation (9).
Let us consider the last case Y
2,1
([g
0
]) < 0. As in the previous case, there holds h()
h(0) = Y
2,1
([g
0
]) < 0 > 0 if n 4, and h() h(0) = Y
2,1
([g
0
]) > 0 > 0 if n = 3
and the family of solution u

is bounded in C
2
norm as 0. We claim this family is
also bounded in C
2,
norm for > 0. For this purpose, we will write equation (81) in the
new form. Recall w
ij
=
2
ij
u +u
i
u
j

|u|
2
2
(g
0
)
ij
+(S
g
0
)
ij
. We dene a new second order
tensor

W = ( w
ij
) with
w
ij
= w
ij
+(g
0
)
ij
,
where is some small positive constant to be xed later. It is clear that
(96)
1
(

W) =
1
(W) +n
and
(97)
2
(

W) =
2
(W) + 2(n 1)
1
(W) +n(n 1)
2
.
Denote u

the unique solution of (81) with

_

1
(g)dvol(g) = 1 for k = h() and for all
small > 0. Using (96) and (97), u

satises the following equation

(98)

2
(

W) n(n 1)
2
2(n 1)(
1
(

W) n)

1
(

W) n
= h()e
2u

,
that is, u

solves
(99)

2
(

W)

1
(

W)
+
+nh()e
2u

+n(n 1)
2

1
(

W)
= h()e
2u

+ 2(n 1).
Choose a small > 0 such that (0,
0
) and x M, we have
(100) h()e
2u

(x)
+ 2(n 1) < 0.
Now equation (100) is uniformly elliptic and concave for all (0,
0
). From the classical
Krylovs result, the set of solutions u

is also bounded in C
2,
norm for > 0. Passing
to the limit, u

converges in C
2,

norm for any

(0, ) to some C
2,
function u

+
1
,
which is a solution of (10). Finally, writing equation in the form (100) for = 0, the
uniqueness comes from the maximum principle as in the proof of Proposition 1. Therefore,
we nish the proof.
Proof of Theorem 1. When (g
0
,
2
) > 0, the result follows from Proposition 3. When
(g
0
,
2
) = 0, the proof is the same as in the case Y
2,1
([g
0
]) = 0 in the previous proof.
26 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
Proof of Corollary 1. It follows from Proposition 3 directly.
Proof of Corollary 2. From Theorem 2, the cone
+
2
is not empty. Set g (
2
([g
0
]). From
a result due to Gursky-Viaclovsky [29] the Ricci tensor of g is pointwise positive. Thanks
to the result of Hamilton [34] by using the Ricci ow, M is dieomorphic to a compact 3-
dimensional Riemannian manifold with constant curvature. Moreover, M is dieomorphic
to S
3
/, where is a nite isometry subgroup of S
3
in the standard metric.
6. The Yamabe invariants Y
2,1
([g
0
])
In this section, We prove Theorem 3.
Proposition 4. Let (M
n
, g
0
) be a compact Riemannian manifold with g
0

+
1
and n 3.
The followings hold
1) If there is a regular conformal metric g [g
0
]
+
2
satisfying equation (8), then
Y
2,1
([g
0
]) > 0;
2) If there is a C
1,1
conformal metric g [g
0
]

+
1
satisfying equation (9), then
Y
2,1
([g
0
]) = 0;
3) If there is a regular conformal metric g [g
0
]

+
1
solving equation (10), then
Y
2,1
([g
0
]) < 0.
Theorem 3 follows from this Proposition. To prove the Proposition, we need the fol-
lowing lemma.
Lemma 5. Let (M
n
, g
0
) be a compact Riemannian manifold with g
0

+
1
and n 3. At
most one of the followings holds:
1) equation (8) admits a regular solution g [g
0
]
+
2
;
2) equation (9) admits a C
1,1
solution g [g
0
]

+
1
;
3) equation (10) admits a regular solution g [g
0
]

+
1
.
Proof. Let g [g
0
]
+
2
be a solution of equation (8). Without loss of generality, assume
g = g
0
. From a result of Guan-Wang [27], there is no C
1,1
solution of equation (9) in

+
2
.
On the other hand, for any g = e
2u
g
0
(
1
([g
0
]), let x
0
be a minimum point of u. At
this point, we get u = 0 and hessian matrix
2
u is non-negative denite. Thus, the
Schouten tensor at point x
0
is in
+
2
. As a consequence, equation (10) does not admit a
regular solution g [g
0
]

+
1
.
Now suppose g
1
= e
2u
1
g
0
is a regular solution of equation (10) in [g
0
]

+
1
. As above,
equation (8) does not admit a regular solution g [g
0
]
+
2
. As in the proof of Theorem
2, u
1
solves
(101)

2
(

W)

1
(

W)
+
ne
2u
1
+n(n 1)
2

1
(

W)
= e
2u
1
+ 2(n 1),
ON THE
2
-SCALAR CURVATURE 27
where > 0 is a small positive number such that f(x) := e
2u
1
+ (n 1) < 0 and
w
ij
= w
ij
+(g
0
)
ij
. Let g
2
= e
2u
2
g
0
[g
0
]

+
1
be a C
1,1
solution of equation (9). Thus,
u
2
is a subsolution to (101), that is,
(102)

2
(

W)

1
(

W)
+
nf(x)

1
(

W)
e
2u
1
+ 2(n 1)
since u
2
solves
(103)

2
(

W)

1
(

W)
+
n(n 1)
2

1
(

W)
= 2(n 1)
and
(104)
1
(

W) n.
Set v = u
2
u
1
and H(W, x) =

2
(W) +nf(x)

1
(W)
. Denote
H
ij
:=
H(W, x)
w
ij
.
On the other hand, for any c R, u
2
+c is also a C
1,1
solution of equation (9). Without
loss of generality, we could suppose v 0 and max v = 0. From (101) and (102), v is a
subsolution of some uniformly elliptic second order operator, that is,
Lv :=

i,j
A
ij

2
ij
v +

i
B
i

i
v 0,
where
A
ij
=
_
1
0
H
ij
(s

W
2
+ (1 s)

W
1
)ds.
From the strong maximum principle, v 0. This contradiction yields the desired result
and we nish the proof of Lemma.
Corollary 6. Let (M, g
0
) be a compact Riemannian manifold with positive scalar curva-
ture and of dimension n 4. The sign of (g
0
,
2
), equivalently the sign of Y
2,1
([g
0
]), is
a conformal invariant.
Proof of Proposition 4. It is a direct result of Theorem 2 and Lemma 5.
7. Geometric Applications
First we show that any manifold in class (1
+
) admits a psc metric with negative Yamabe
constant Y
2,1
< 0. Therefore we can deform this metric in its conformal class to a psc
metric with negative
2
-scalar curvature. Then we show that Y
2,1
is continuous in a
suitable sense.
Proof of Theorem 4. With the analysis established above, in order to show the Theo-
rem, We need only to show that for any manifold in (1
+
) there is a psc metric g with
_

2
(g)dvol(g) < 0.
28 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
Here we use the well-known construction of Gromov-Lawson [19] for positive scalar
curvature metrics. See also [50]. Let S
p
be an embedded sphere in M with trivial normal
bundle of codimension q = n p 3. Let S
p
D
q
( r) be an embedding into M for some
small constant r > 0. Let r
0
be a constant xed later as small as we want.
By the construction of Gromov-Lawson, we have a manifold (N
1
, h
1
) with an end S
p

S
q1
[0, +) such that N
1
(S
p
S
q1
[1, ], h
1
) is isometric to M S
p
D
q
( r)
and (S
p
S
q1
[2, +), h
1
) is isometric to S
p
S
q1
(r

) [2, +) with the product

metric. Here S
p
is standard sphere with radius 1 and S
q1
(r

) is the standard sphere

with a small radius r

. The crucial point in the construction of Gromov-Lawson is that

the scalar curvature of h
1
is positive. Now we glue on N
1
S
p
S
q1
(r

) [
0
, +) a
product manifold S
p
D
q
, where D
q
is not equipped with the at metric, but a product
metric of S
q1
[0, b] in a neighborhood of the boundary and of positive scalar curvature
on D
q
. The result manifold is of positive scalar curvature and is dieomorphic to M. It
contains a product S
p
S
q1
(r

) [2,
0
] for a small r

and a large
0
.
Now we consider the case q = 3. In this case, its Ricci curvature, written as a diagonal
matrix, is diagn4, , n4, r
2

, r
2

, 0. Its Schouten tensor, also written as a diagonal

matrix and up to a multiple constant independent of r

, is
diagn 4 , , n 4 , r
2

, r
2

,
with
=
1
2(n 1)
((n 3)(n 4) +
2
r
2

).
It can be written as
1
n 1
r
2

diag1, , 1, n 2, n 2, 1 +C,
Where C is a matrix whose entries are independent of r

. To decide the sign of

2
-scalar
curvature for this product, we only need to compute
2
(1, , 1, n 2, n 2, 1) =

1
2
(n 1)(n 2). Hence choosing r

small, we have negative

2
-scalar curvature in this
product part. By choosing
0
large enough, we obtain a metric g on M with
_
M

2
( g) < 0.
Therefore for the conformal class [ g] we have Y
2,1
([ g]) < 0. By Theorem 2, we have a
conformal metric

h with
2
(

h) 0.
When n = 4, we have another proof. In this case,
_
(
2
(g) +
1
16
[W[
2
)dvol(g)
is the Euler characteristic of M up to a constant multiple. Since
_
[W[
2
dvol(g) is invariant
in a conformal class, so is
_

2
(g)dvol(g). In [1] the authors constructed a sequence of
metrics g
i
satisfying that
Y
1
([g
i
]) Y
1
([g
0
]) and
_
[W[
2
dvol(g
i
) ,
ON THE
2
-SCALAR CURVATURE 29
as i . Since R
g
0
> 0, from this sequence we can nd a psc metric g
i
with
_

2
(g
i
)dvol(g
i
) <
0.
Lemma 6. Let n 4. If the space of psc metric is equipped with the C
4,
-topology, then
map g Y
2,1
([g]) is continuous.
Proof. A similar Lemma for the rst Yamabe constant Y
1
was given in [4], see also [5].
For Y
2,1
, the proof becomes complicated. We use solutions of equations considered here.
Let g
j
be a sequence of psc metric converging to g in C
4,
topology. The proof is trivial
for the case n = 4, since
_
M

2
(g)dvol(g) is constant in a conformal class. Now we consider
the cases n 5. By taking a subseqeunce we may assume that lim
j
Y
2,1
([g
j
]) = Y
0
. It
is easy to see that
(105) Y
0
= lim
j
Y
2,1
([g
j
]) Y
2,1
([g]).
To see this, for any > 0 nd g
u
= e
2u
g (
1
([g]) such that
_

2
(g
u
)dvol(g
u
) < Y
2,1
([g]) +
1
2
and
_

1
(g
u
)dvol(g
u
) = 1.
For suciently large j, we have g
u
(
1
([g
j
]). Since
_

2
and
1
are continuous for this
xed function u, we have
Y
2,1
([g
j
]) Y
2,1
([g]) +,
for a large j, and hence (105).
Now we show that Y
0
Y
2,1
([g]). Let Y
2,1
(S
n
) be the second Yamabe constant for the
standard sphere S
n
. Using a method given in [53], one can show that
Y
2,1
(S
n
) Y
2,1
([g]),
for any metric psc metric g. Without loss of generality, we may assume Y
0
< Y
2,1
(S
n
).
Otherwise, we have lim
j
Y
2,1
([g
j
]) = Y
0
= Y
2,1
(S
n
) Y
2,1
and hence we are done.
Consider the following perturbed energy functional
J
,
(g) :=
_
(
2
(g) e
2u
)dvol(g))
(
_
e
2u

1
(g)dvol(g))
n4
n22
for small > 0 and > 0 in the cone (
1
. The analysis established above implies that
the above functional admits a minimizer in (
1
([g]) for any psc metric. It is easy to check
that for given j N, there are small constants
j
> 0 and
j
> 0 such that a minimizer
g
j
= e
2u
j
g
j
(
1
of the perturbed functional J

j
,
j
satises
J

j
,
j
( g
j
) Y
2,1
([g
j
]) +
1
j
.
As a minimizer of J
,
, g
j
= e
2u
j
g
j
satises an equation similar to (8), for which we
have local estimates Since g
j
converges to g in C
4,
, the local estimates and Y
0
< Y
2,1
(S
n
)
imply that g
j
converges to a metric g = e
2 u
g (
1
([g]) in C
0
-topology, and hence in
C
1,1
-topology. If g (
1
, we can show that
lim
j
Y
2,1
([g
j
]) = lim
j
J

j
,
j
( g
j
) Y
2,1
([g]).
30 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
Otherwise, we rst show that
(106) lim
j
_

2
(e
2u
j
g
j
)dvol(e
2u
j
g
j
) = lim
j
_

2
(e
2u
j
g)dvol(e
2u
j
g)
and
(107) lim
j
_

1
(e
2u
j
g
j
)dvol(e
2u
j
g
j
) = lim
j
_

1
(e
2u
j
g)dvol(e
2u
j
g).
This is clear, for u
j
has a uniform C
2
bound and g
j
conevrges to g. Then, we claim that
there exists metrics e
2 u
j
g (
1
([g]) such that
(108) lim
j
_

2
(e
2 u
j
g)dvol(e
2 u
j
g) = lim
j
_

2
(e
2u
j
g)dvol(e
2u
j
g)
and
(109) lim
j
_

1
(e
2 u
j
g)dvol(e
2 u
j
g) = lim
j
_

1
(e
2u
j
g)dvol(e
2u
j
g).
Given j N, we choose t
j
(0, 1) with limt
j
= 1 such that e
2t
j
u
j
g (
1
([g]). To see
this, we compute
(110)
e
2t
j
u
j

1
(e
2t
j
u
j
g) = t
j

g
u
j
t
2
j
n 2
2
[u
j
[
2
g
+
1
(g)
= t
j
e
2u
j

1
(e
2u
j
g
j
) +t
j
(
g
u
j

g
j
u
j
+
1
(g)
1
(g
j
))
+t
j
(
n 2
2
[u
j
[
2
g
j

n 2
2
[u
j
[
2
g
)
+(1 t
j
)(t
j
n 2
2
[u
j
[
2
g
+
1
(g)).
It is clear that
(111)

t
j
(
g
u
j

g
j
u
j
+
1
(g)
1
(g
j
)) +t
j
(
n 2
2
[u
j
[
2
g
j

n 2
2
[u
j
[
2
g
)

C|g g
j
|
C
1([u
j
[
2
g
+
1
(g)).
Thus, we can choose t
j
close to 1 such that e
2t
j
u
j
g (
1
([g]). From (19), (22) and
fact that u
j
has uniform bound C
0
norm, we have (108) and (109), which implies that
Y
0
= lim
j
Y
2,1
([g
j
]) Y
2,1
([g]). Hence we nish the proof of the Lemma.
In fact, the Lemma is true for C
4
-topology.
Proof of Theorem 5. If M belongs to class (2
+
), then there is a psc metric with positive

2
-scalar curvature. By Theorem 3, Y
2,1
([g]) > 0. From Theorem 4 we have another psc
metric g with Y
2,1
([ g]) < 0. Using the previous Lemma we have the third psc metric g
with Y
2,1
([ g]) = 0. Now 1. follows from Theorem 2. The proof for 2. and 3. is the same.
ON THE
2
-SCALAR CURVATURE 31
In order to prove the Conjecture (Trichotomy Thereom) as in [40] and [41], we need an
implicit function theorem for the linearization operator of
2
(g) in L
2,p
. This seems to be
rather dicult for us, at least at the moment.
Example 1. Let us consider the manifold M = S
3
S
1
. It is a locally conformally at
manifold with
1
> 0 and
2
= 0 for the product metric. From
2
2
(M) =
_
M

2
(g) +
1
16
[W[
2
,
we have
_
S
3
S
1

2
(g) =
1
16
_
[W[
2
0 for any metric g on S
3
S
1
. Hence, S
3
S
1
belongs
to class (2
0
).
Example 2. For n > 4 let us consider the locally conformally at manifold S
p
H
np
,
where p > n/2 and H
np
is a compact quotient of the hyperbolic space H
np
with sectional
curvature 1. Let g
0
be the product metric. It is clear that
1
(g
0
) > 0, for p > n/2. Its

2
-scalar curvature, up to a positive constant multiple, is
(p
n
2
)
2

n
4
.
Hence, p >
n+

n
2
if and only if
2
(g
0
) > 0. Now we consider such p with
2
(g
0
) = 0,
namely p =
n+

n
2
. It is clear that n = (2m + 1)
2
for some m N. For example m = 1,
M = S
6
H
3
. In general M
m
:= S
(m+1)(2m+1)
H
m(2m+1)
. We conjecture that on such
a manifold there is a psc metric with positive
2
-scalar curvature. A rough idea to check
this can be made as follows. By Theorem 2, if it is not true, we know that any psc metric
g on M
m
has Y
2,1
([g]) 0, which is equivalent to
inf
gC
1
([g])
_
M

2
( g)dvol( g) 0.
Hence 0 is be a min-max value of
_
M

2
(g)dvol(g) and the product metric g
0
would be
a critical point of
_

2
on the space of all metrics. If this is true, then a result in [39]
implies that g
0
is a metric of constant sectional curvature. This certainly is false. From
this example, we propose a
Conjecture. When n > 4, class (2
0
) is empty.
8. Further applications in fully nonlinear equations
With the method considered in this paper, we can also deal with the following problems.
1. When (M, g
0
) is a locally conformally at manifold with g
0
(
k1
(k > 1), the
method presented here can be used to study the equation
(112)

k
(g)

k1
(g)
= f.
32 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
Similarly we dene a Yamabe type invariant in [21] as follows
Y
k,k1
([g
0
]) = inf
_
M

k
(g)dvol(g)
(
_
M

k1
(g)dvol(g))
n2k
n2k+2
,
for k n/2. We can show that if Y
k,k1
> 0 ( Y
k,k1
= 0 ans Y
k,k1
< 0 resp.), then there
is a conformal metric g with
k
> 0 (
k
= 0 and
k
0 resp.)
2. Let (M
4
, g
0
) be a compact 4-dimensional manifold with positive scalar curvature.
One can consider the equation
(113)

2
s[W[
2

1
= f,
where W is the Weyl tensor and s is a non-negative number. In this case for a xed s
_
(
2
s[W[
2
)dvol(g)
is a constant in a given conformal class. With the method presented here, we can show
that the number
_
(
2
s[W[
2
)dvol(g) is positive, null or negative resp. if and only if
there is a conformal metric with

2
s[W[
2
> 0, = 0 or 0 resp.
The positive and null cases were studied already in [11] as mentioned above. This can be
seen as a generalization of the following classical result: Let (M
2
, g
0
) be a closed surface.
Its Euler characteristic (M
2
) =
1
2
_
M
R
g
dvol(g) is positive, negative or null resp. if and
only if there is a conformal metric g with positive, negative or null scalar curvature resp.
It is clear that equation (113) can also be considered on a higher dimensional manifold.
3. Our methods can also be applied to study the following fully nonlinear equations

k
(
2
u)

k1
(
2
u)
= f
and

k
(
2
u +ug)

k1
(
2
u +ug)
= f,
in the class of (k 1)-admissible functions.
4. Another interesting problem is a generalization of the prescribed scalar curvature
problem. Let M
n
= S
n
and f : S
n
R
1
be a smooth function. We would like to ask if
there is a conformal metric g (
1
such that

2
(g) = f
1
(g).
A Kazdan-Warner type neccesary condition could be obtained as Han [35] did for the
prescribed
k
-scalar curvature problem. Here, and also in the previous problem, the
function f need not be positive.
ON THE
2
-SCALAR CURVATURE 33
Acknowledgement. This project was started several years ago together with Pengfei Guan.
The key observation was discovered by Pengfei Guan and the second named author in
their study of fully nonlinear equations. We would like to thank Pengfeng Guan for many
helpful discussions and encouragement. A part of the work was carried out while the rst
author was visiting McGill University and he would like to thank the department and
Pengfei Guan for warm hospitality.
References
[1] K. Akutagawa, B. Botvinnik, O. Kobayashi and H. Seshadri, The Weyl functional near the Yamabe
invariant. J. Geom. Anal., 13 (2003), 120.
[2] Aubin, Metriques riemanniennes et courbure, J. Dierential Geometry, 4 (1970), 383424.
[3] T. Aubin,

Equations dierentilles non lineaires et probleme de Yamabe concernant la courbure
scalaire, J. Math. Pures Appl., 55 (1976), 269-296.
[4] L. Berard Bergery, Scalar curvature and isometry groups, in Sepctra of Riemannian manifolds, ed. M.
Berger, S. Murakami, T. Ochiai, Kagai, Tokyo, (1983) 928
[5] A. Besse, Einstein manifolds, Ergebnisse der Mathematik und ihrer Grenzgebiete (3) 10 Springer-
Verlag, Berlin, 1987
[6] T. Branson and A. Gover, Variational status of a class of fully nonlinear curvature prescription prob-
lems, Arxiv: math.DG/0610773
[7] S. Brendle and J. Viaclovsky, A variational characterization for
n/2
, Calc. Var. P. D. E., 20 (2004),
399402.
[8] L. Caarelli, L. Nirenberg and J. Spruck, The Dirichlet problem for nonlinear second-order elliptic
equations. III. Functions of the eigenvalues of the Hessian, Acta Math., 155 (1985), 261301.
[9] A. Chang, M. Gursky and P. Yang, An equation of Monge-amp`ere type in conformal geometry, and
four manifolds of positive Ricci curvature, Ann. of Math., 155 (2002), 709-787.
[10] A. Chang, Gursky and P. Yang, An a priori estimates for a fully nonlinear equation on Four-manifolds,
J. DAnalysis Math., 87 (2002), 151186.
[11] A. Chang, Gursky and P. Yang, A conformally invariant sphere theorem in four dimensions, Publ.
Math. Inst. Hautes

Etudes Sci. 98 (2003), 105143.
[12] S. Chen, Local estimates for some fully nonlinear elliptic equations, Int. Math. Res. Not., 2005
(2005), 34033425.
[13] S. Dieter, Nonlinear degenerate curvature ows for weakly convex hypersurfaces. Calc. Var. PDE, 22
(2005), 229251.
[14] A. Futaki, Scalar-at closed manifolds not admitting positive scalar curvature metrics, Invent. Math.,
112 (1993), 2329.
[15] L. Garding, An inequality for hyperbolic polynomials, J. Math. Mech, 8 (1959), 957965.
[16] Z. Gao and S.T. Yau, The existence of negatively Ricci curved metrics on three-manifolds., Invent.
Math., 85 (1986), 637652.
[17] Yuxin Ge and G. Wang, On a fully nonlinear Yamabe problem, Ann. Sci.

Ecole Norm. Sup., 39
(2006) 569598
[18] Yuxin Ge and G. Wang, On a quotient conformal equation, preprint 2006.
[19] M. Gromov and H. B. Lawson, The classication of simply connected manifolds of positive scalar
curvature, Ann. of Math., (2) 111 (1980), 423434.
[20] P. Guan, Topics in Geometric Fully Nonlinear Equations, Lecture Notes,
http://www.math.mcgill.ca/guan/notes.html
[21] P. Guan, C.-S. Lin and G. Wang, Application of The Method of Moving Planes to Conformally
Invariant Equations, Math. Z., 247 (2004), pp. 1-19.
[22] P. Guan, C.-S. Lin and G. Wang, Schouten tensor and some topological properties, Comm. Anal.
Geom., 13 (2005), 887902.
34 YUXIN GE, CHANG-SHOU LIN, AND GUOFANG WANG
[23] P. Guan, C.-S. Lin and G. Wang, local gradient estimates for conformal quotient equations, to appear
in Int. J. Math.
[24] P. Guan, J. Viaclovsky and G. Wang, Some properties of the Schouten tensor and applications to
conformal geometry, Trans A. M. S., 355 (2003), 925-933.
[25] P. Guan and G. Wang, Local estimates for a class of fully nonlinear equations arising from conformal
geometry, Intern. Math. Res. Not., 2003, (2003), 1413-1432.
[26] P. Guan and G. Wang, A fully nonlinear conformal ow on locally conformally at manifolds, J. reine
und angew. Math., 557, (2003), 219-238.
[27] P. Guan and G. Wang, Geometric inequalities on locally conformally at manifolds, Duke Math. J.,
124, (2004), 177-212.
[28] P. Guan and G. Wang, Conformal deformations of the smallest eigenvalue of the Ricci tensor,
Arxiv:math.DG/0505083 to appear in: Amer. J. Math.
[29] M. Gursky and J. Viaclovsky, A new variational characterization of three-dimensional space forms,
Invent. Math., 145 (2001), no. 2, 251278.
[30] M. Gursky and J. Viaclovsky, Volume comparison and the
k
-Yamabe problem, Adv. in Math., 187
(2004), 447-487.
[31] M. Gursky and J. Viaclovsky, A fully nonlinear equation on 4-manifolds with positive scalar curvature,
J. Di. Geom., 63, (2003), 131-154.
[32] M. Gursky and J. Viaclovsky, Prescribing symmetric functions of the eigenvalues of the Ricci tensor,
to appear in Annals Math.
[33] M. Gursky and J. Viaclovsky, Fully nonlinear equations on Riemannian manifolds with negative
curvature, Indiana Univ. Math. J., 52 (2003), 399419.
[34] R. Hamilton, Three-manifolds with positive Ricci curvature, J. Dierential Geom., 17 (1982), 255
306.
[35] Z.-C. Han, A Kazdan-Warner type identity for the
k
curvature, C. R. Math. Acad. Sci. Paris, 342
(2006) 475478.
[36] E. Hebey, Nonlinear analysis on manifolds: Sobolev spaces and inequalities. Courant Lecture Notes in
Math., 5. Courant Inst. of Math. Sci., New York; Amer. Math. So., Providence, RI, 1999.
[37] N. Hitchin, Harmonic spinors, Adv. Math., 14 (1974) 155.
[38] G. Huisken and C. Sinestrari, Convexity estimates for mean curvature ow and singularities of mean
convex surfaces, Acta Math., 183 (1999), 4570.
[39] Z. Hu and H. Li, A new variational characterization of n-dimensional space forms., Trans. Amer.
Math. Soc., 356 (2004), 30053023
[40] J. Kazdan and F. Warner, Scalar curvature and conformal deformation of Riemannian structure, J.
Dierential Geometry, 10 (1975), 113134.
[41] J. Kazdan and F. Warner, Existence and conformal deformation of metrics with prescribed Gaussian
and scalar curvatures, Ann. of Math., (2) 101 (1975), 317331.
[42] N. Krylov, Nonlinear elliptic and parabolic equations of the second order, D. Reidel, (1987)
[43] J. Lee and T. Parker, The Yamabe problem, Bull. Amer. Math. Soc., (N.S.) 17 (1987), 3791.
[44] A. Li and Y. Li, On some conformally invariant fully nonlinear equations, Comm. Pure Appl. Math.,
56 (2003), 14161464.
[45] A. Li and Y. Li, On some conformally invariant fully nonlinear equations. II. Liouville, Harnack and
Yamabe, Acta Math., 195 (2005), 117154.
[46] A. Li and Y. Li, A Liouville type theorem for some conformally invariant fully nonlinear equations,
Geometric analysis of PDE and several complex variables, 321328
[47] A. Lichnerowicz, Spineurs harmoniques, C. R. Acad. Sci. Paris, 257 (1963) 79.
[48] P. L. Lions, Two remarks on the Monge-Ampere, Ann. Mat. Pure Appl., 142 (1985), 263-275
[49] J. Lohkamp, Metrics of negative Ricci curvature, Ann. of Math., (2) 140 (1994), 655683.
[50] J. Rosenberg and S. Stolz, Metrics of positive scalar curvature and connections with surgery, Surveys
on surgery theory, Vol. 2, 353386, Ann. of Math. Stud., 149, Princeton Univ. Press, Princeton, NJ,
2001.
ON THE
2
-SCALAR CURVATURE 35
[51] R. Schoen, Conformal deformation of a Riemannian metric to constant curvature, J. Di. Geom., 20
(1984), 479-495.
[52] R. Schoen ans S. T. Yau, On the structure of manifolds with positive scalar curvature, manu. math.,
28 (1979), 159183.
[53] W.M. Sheng, N.Trudinger, X.-J. Wang, The Yamabe problem for higher order curvatures, ArXiv:
math.DG/0505463.
[54] S. Stolz, Simply connected manifolds of positive scalar curvature, Ann. of Math, (2) 136 (1992),
511540.
[55] S. Stolz, A conjecture concerning positive Ricci curvature and the Witten genus, Math. Ann., 304
(1996), 785800.
[56] S. Stolz, Manifolds of positive scalar curvature, Topology of high-dimensional manifolds, No. 1, 2
(Trieste, 2001), 661709, ICTP Lect. Notes, 9, Abdus Salam Int. Cent. Theoret. Phys., Trieste, 2002.
[57] N. Trudinger, On imbeddings into Orlicz spaces and some applications, J. Math. Mech., 17 (1967),
473483.
[58] N. Trudinger and X-J. Wang, On Harnack inequalities and singularities of admissible metrics in the
Yamabe problem, ArXiv:math.DG/0509341.
[59] J. Viaclovsky, Conformal geometry, contact geometry and the calculus of variations, Duke J. Math.,
101 (2000), no. 2, 283-316.
[60] J. Viaclovsky, Conformally invariant Monge-Amp`ere equations: global solutions, Trans. AMS., 352
(2000), 43714379.
[61] J. Viaclovsky, Estimates and some existence results for some fully nonlinear elliptic equations on
Riemannian manifolds, Comm. Anal. Geom., 10 (2002), 815-847.
[62] J. Viaclovsky Conformal geometry and fully nonlinear equations, Arxiv:DG/0609158
[63] G. Wang,
k
-scalar curvature and eigenvalues of the Dirac operator, Ann. Global Anal. Geom., 30
(2006), 6571.
[64] H. Yamabe, On a deformation of Riemannian structures on compact manifolds, Osaka Math. J., 12
(1960), 2137.
Laboratoire dAnalyse et de Math ematiques Appliqu ees, CNRS UMR 8050, D epartement
de Math ematiques, Universit e Paris XII-Val de Marne,, 61 avenue du G en eral de Gaulle,
94010 Cr eteil Cedex, France
E-mail address: ge@univ-paris12.fr
Department of Mathematics, National Taiwan University and Taida Institute for Math-
ematical Science, National Taiwan University, Taiwan
E-mail address: cslin@math.ntu.edu.tw
Faculty of Mathematics, University Magdeburg, D-39016, Magdebrug, Germany
E-mail address: gwang@math.uni-magdeburg.de