J.

DIFFERENTIAL GEOMETRY
20 (1984) 479-495

CONFORMAL DEFORMATION
OF A RIEMANNIAN
METRIC TO CONSTANT SCALAR CURVATURE

RICHARD SCHOEN

A well-known open question in differential geometry is the question of

whether a given compact Riemannian manifold is necessarily conformally
equivalent to one of constant scalar curvature. This problem is known as the
Yamabe problem because it was formulated by Yamabe [8] in 1960, While
Yamabe's paper claimed to solve the problem in the affirmative, it was found
by N. Trudinger [6] in 1968 that Yamabe's paper was seriously incorrect.
Trudinger was able to correct Yamabe's proof in case the scalar curvature is
nonpositive. Progress was made on the case of positive scalar curvature by T.
Aubin [1] in 1976. Aubin showed that if dim M > 6 and M is not conformally
flat, then M can be conformally changed to constant scalar curvature. Up until
this time, Aubin's method has given no information on the Yamabe problem in
dimensions 3, 4, and 5. Moreover, his method exploits only the local geometry
of M in a small neighborhood of a point, and hence could not be used on a
conformally flat manifold where the Yamabe problem is clearly a global
problem. Recently, a number of geometers have been interested in the confor-
mally flat manifolds of positive scalar curvature where a solution of Yamabe's
problem gives a conformally flat metric of constant scalar curvature, a metric
of some geometric interest. Note that the class of conformally flat manifolds of
positive scalar curvature is closed under the operation of connected sum, and
hence contains connected sums of spherical space forms with copies of
ι n ι
S X S ~.
In this paper we introduce a new global idea into the problem and we solve
it in the affirmative in all remaining cases; that is, we assert the existence of a
positive solution u on M of the equation

2
(0.1) Δw - " Ru + u<n + 2)Λn-2)
Λ
= 0,
4(Λ - 1)

Received September 13, 1984.

480 RICHARD SCHOEN

where R > 0 is the scalar curvature of M. We denote the linear part of the
operator in (0.1) by L, thus

Lu = Δw - - ~ —£w.
4(/i - 1)
The operator L is a conformally invariant operator in that it changes by a
multiplicative factor when the metric of M is multiplied by a positive function.
Observe that the question (0.1) is (a normalized version of) the Euler-Lagrange
equation for the Sobolev quotient Q(φ) for functions φ on M which is given by

fM(\VΨ\2 +(n - 2)RΨ2/4(n - 1)) dv

The Sobolev quotient (?(M) is then defined by

The number Q(M) depends only on the conformal class of M. By choosing

functions φ which are supported near a point of M, it follows easily that
(0.2) β(M)<β(S")
for any n dimensional manifold M. In this paper we show that equality holds
in (0.2) if and only if M is conformally diffeomorphic to Sn with its standard
metric. An argument which is by now standard and which originates in [6]
shows that if Q(M) < Q(Sn), then there exists a minimum for Q(φ) over
functions φ e Cι(M). This minimizing function then becomes a positive
solution of (0.1) on M.
In order to prove that Q(M) < Q(Sn) for a manifold M conformally
different from S", we need only exhibit a function φ on M with Q(ψ) < Q(Sn).
n
Since one can come arbitrarily close to Q(S ) by a function φ which is
supported near a point o e M, it is natural to perturb such a function to make
it nonzero but small away from o. Since the nonlinearity of (0.1) involves a
higher power of the solution, one expects small solutions to be very close to
solutions of Lu = o. The only positive solution of Lu = o defined outside a
point o e M is (a multiple of) the Green's function of L. (Note that L is
invertible because R is positive.) The question of whether one can satisfy the
inequality Q(ψ) < Q(Sn) reduces to the behavior of the Green's function G
near its pole o. If we assume that the metric of M is conformally flat in a
neighborhood of o, then G has an expansion in suitable coordinates near o as
follows:
2
G(x) =\x\ ~" + A + O(\x\).
 CONFORMAL DEFORMATION OF A RIEMANNIAN METRIC 481

The sign of the constant term A in this expansion is then the crucial ingredient.
If A is positive, then one can find a function φ which is a small multiple of G
outside a neighborhood of o and which satisfies Q(φ) < Q(Sn). On the other
hand, it is a theorem of the author and S. T. Yau that A > 0 and A = 0 only if
M is conformally equivalent to Sn. The case n = 3 follows from the positive
mass theorem [3] since the metric g = G4Λn~2)g is scalar flat and asymptoti-
cally Euclidean,
g,, = (l + A\y\2-")δu + 0(bΓ") for \y\ large,
2
where y = |x|~ x. The case « = 4 is a consequence of the positive action
theorem [4]. The higher dimensional case follows from related techniques and
will appear in [5]. In this same paper [5] we have generalized the method of E.
Witten [7] (see also [2]) to prove A > 0 in case M is a spin manifold. Thus we
have Q(M) < Q(Sn) provided M is conformally flat near some point. The
same argument works for an arbitrary three dimensional manifold since the
Green's function has the above expansion generally in three dimensions. By a
delicate perturbation argument (see §2) we are able to handle general compact
manifolds of dimensions 4 and 5. Thus, combined with the results of [1], we
have an affirmative solution of the Yamabe problem for any compact Rieman-
nian manifold of dimension greater than two.
We recently learned that Rui-Tao Dong showed that solutions blow up near
a single point at most.

1. The Sobolev quotient of a conformally flat manifold

In this section we assume M is a compact Riemannian manifold with metric
g, o is a point of M, and g is conformally flat in a neighborhood of o. Let Q(φ)
denote the Sobolev quotient of a function φ on M, and let E(φ) denote the
energy associated with L, that is

O( ) _

By changing the metric g conformally we may assume that g is flat near o. Let
x be Euclidean coordinates centered at o so that g/y = fi/y in the x coordinates.
Observe that the functions

.2
ε 2 + be
482 RICHARD SCHOEN

for ε > 0, are solutions on R" of the equation

Δw ε 4- n(n - 2)u(En + 2)An 2)
- = 0.

Multiplying this equation by uε and integrating by parts gives

f \vufdx = n(n-2)f u2εnΛ"-2)dx.

From here we can express Q(Sn) in terms of uε:

(1.1) Q(S") = - /R"1V«J dx =n{n_2)ίf u

Let G be the positive solution (a multiple of the Green's function) of LG = 0

on M — {o} which behaves like |JC| 2 ~" near 0. Since G is a harmonic function
near o, it has an expansion for |JC| small

(1.2) G(x)=\x\2~n + A+a(x),

where a(x) is a smooth harmonic function (near 0) with α(0) = 0. Let p 0 be a
small radius, and ε 0 > 0 a number to be chosen small relative to p 0 . Let ψ(x)
be a piecewise smooth decreasing function of |x| which satisfies ψ(x) - 1 for
\x\ < p 0 , ψ(x) = 0 for |x| > 2ρ 0 , and |Vψ| < PQ for p 0 < |x| < 2ρ 0 . We now
1

construct a piecewise smooth test function φ on M as follows:

ί
uE(x) for|x|<p 0 ,

εQ(G(x) - χp(x)a(x)) for p 0 < \x\ < 2p 0 ,

e0G(x) ϊorχ^M~B2po(0).
In order for the function φ to be continuous across dBp (0) we must require ε to
satisfy
(1-3) ^ ^ )

We compute E(ψ) as a sum of the energy in BPQ(0) and the energy in

M — Bpo(Q), Using the equation for uε we have, after an integration by parts,
ί I I2 / \ f 1 Λ -Ί\ f 9"*

Λ
/ | V w J dx = n ( n - 2 ) J u ε " l) dx

Using (1.1) and the definition of φ we have

(1.4) ( \vψ\2dx^ Q(S")l

 CONFORMAL DEFORMATION OF A RIEMANNIAN METRIC 483

Evaluating the energy of φ on M - Bpo we have

dυ = dv
M-BP

+ε o ( (IVψαf - 2vG V(ψα)) dx.

Since \a(x)\ < C|JC|, we see that | v(ψα)| < c for p 0 < |JC| < 2ρ0. Therefore we
have for a constant c

Since G satisfies LG = 0, the first term on the right becomes a boundary

integral

M-BOΛ

Combining this with (1.4) we have

(«-2)/2

(1.5)

If M is not conformally equivalent to S", it follows from [3], [4] and [5] that
A > 0. We use this to show that the last two terms in (1.5) are negative if p 0 , e0
are chosen small. For |x| = p 0 we have from (1.2) and (1.3)
\(Λ-2)/2
due / ε Po
ε -2) cε0

r-
ε
2
dr dr lε ^P ε 2
+ Po

ε \2
--(» -2) (pί (( +1 ""
Po/
Using the inequality (t2 + I ) " 1 > 1 - /2, we get

-(n - 2)eQAp-oι cε0,

484 RICHARD SCHOEN

where the second inequality follows from (1.3). Using this in (1.5) we have

EM
0.6)
-(/i -

where α,,^ denotes the volume of S"'1. Since A > 0, by choosing p 0 small and
ε 0 much smaller than ρ0, we have β(φ) < (?(£")• Thus we have shown the
following result.
Theorem 1. If Mn is a compact Riemannian manifold which is conformally
flat in an open set and is not conformally diffeomorphic to Sn, then the Sobolev
quotient of M is strictly less than that of Sn.
The proof given above works for n = 3 without the conformally flat assump-
tion because the function G(x) has the expansion (1.2) generally in this case.
Proposition 1. // M is a compact three-dimensional manifold which is con-
formally different from S3, then the Sobolev quotient of M is strictly less than
that of S3.
Proof. Let x be normal coordinates centered at 0, and note that G(x)
satisfies

G(x)=\x\~l +A + 0(|x|)
for x small. Let wε, ψ, φ be as above, and apply the same argument. Correction
terms must be introduced in Bpo to account for the difference between g and
the Euclidean metric. We see easily that

dv < j \vuf dx

f u\dx < f u6ε dv

Since the error terms are allowable, we see that the above proof succeeds for
n = 3.

2. The Sobolev quotient of a general four and five dimensional manifold

For n = 4, 5 we will remove the restriction that M be conformally flat in an
open set. First observe that there is no loss of generality in assuming that
R = 0 in a neighborhood of a point o e M since this can be accomplished by
multiplication of the metric by a function. Let x denote a normal rectangular
 CONFORMAL DEFORMATION OF A RIEMANNIAN METRIC 485

coordinate system centered at o, and r = \x\, ξ = x/\x\ a corresponding polar

coordinate system. The metric g of M can then be written
g = dr2 + r2hr,
where hr is a metric on Sn~ι with hQ being the standard metric. Given p > 0
with p small, let ξ(r) be a smooth nonincreasing function satisfying ζ(r) = 1
for r < p, f(r) = 0 for r > 2p, |?'(r)l < cp~\ and |{"(r)| ^ cp"2 for all r > 0.
We now define a modified metric pg on M by setting pg = g on Λf — Z?2p, and
p
g = dr2 + r 2 (f(r)/* 0 +(1 - ξ(r))hr) for r < 2p.
p
Thus g is Euclidean in Bp and agrees with g outside B2p. It is easy to check
that the curvature tensor of pg is bounded independent of p. Let Lp denote the
linear operator taken in terms of pg,

Let λ p denote the lowest eigenvalue of L p , and λ the lowest eigenvalue of L.

Since (M, g) is conformal to a metric of positive scalar curvature, we have
λ > 0. Let G denote the multiple of the Green's function of L with pole at o
normalized so that lim ( J c ( ^ 0 |JC|"~ 2 G(JC) = 1. We need the following lemma.
Lemma 1. The eigenvalues λ p converge toλ as p tends to 0, and hence λ p > 0
for p sufficiently small. Thus Lp has a positive solution Gp with pole at o
normalized so that l i m ^ Q \x\n~2Gp(x) = 1. The functions Gp converge (as p 10)
uniformly to G in C2 norm on compact subsets of M — {o}. For n > 4 the
following estimates hold for x e Blp\
G γ \
X) | γ | 2 ~ " I ^< C\X\
_ \X\ r\x\ ,

\x\ )\<c\x\
>(Gp(x)-
for any a e (3,4), where c depends on a but not p. (Actually one can take a = 4
provided n > 4.) /« //ze second inequality above, V denotes the gradient with
respect to g.
Proof. The first statement follows from the fact that pg converges in C 1
norm to g as p i θ and Rp is uniformly bounded. We omit the details. We next
prove the inequalities by observing that if px > 0 is fixed and sufficiently
small, we have A p |jc| a ~ w < -ε(a)\x\2~n and | Δ p | x | 2 ~ w | ^ cι\x\2-~n for x e
Bpi(0). Therefore we can choose c sufficiently large so that for x e Bpι(0),

Δ,(M2-- CM""") > o,

p]-" - cpΓ" < 0.
486 RICHARD SCHOEN

It can then be seen from the maximum principle that

2-n , ,a-n 2-n , .a-n
\x\ - c\x\ < Gp{x) ^\x\ + c\x\
The second inequality then follows from the gradient estimate for elliptic
p
equations noting that the metrics g are all uniformly equivalent to g. The
convergence of Gp to G now follows because we have a uniform upper bound
on Gp and hence on its derivatives on compact subsets of M - {o}. Any limit
of Gp is forced to have the correct singularity from the above inequality and
hence Gp converges to G. This completes a sketch of the proof of Lemma 1.
Since the metrics pg are Euclidean in Bp(0), the function Gp is harmonic in Bp
and hence has an expansion for \x\ small,

Gp(x)=\x\2~n + Ap+O(\x\), Ap>0.

p
Therefore we have inequality (1.6) for (M, g) for p 0 < p. To relate this
information back to (Λf, g) we let φ be as in §1 (constructed from (M, p g)) and
observe
2
(2.1) VolJdB,) dr

since φ = uε in BpQ is a function of r. The following observation about

Vol g (32? r ) is necessary.
Lemma 2. For r small, the asymptotic formula

= on_x -

Proof. Let gtj be the expression of g in terms of rectangular normal

coordinates x, and let g = det(g / y ). By direct calculation

kj

^ k
Therefore, by Taylor's theorem,

1
L, aijkxx u
^ ^ L> ΞkkjjXx ^ x ^ \\x\ )i

where aijk depends on the third derivatives of τ/g at x = 0. Setting r = \x\,

x = rξ,ζ<Ξ S"-1 we have
1
Vol,(ai»Γ) = r -
 CONFORMAL DEFORMATION OF A RIEMANNIAN METRIC 487

where dξ is the volume element for the unit Sn~ι. Then we have from above

where we have used the obvious facts

Direct calculation in normal coordinates shows

thus completing the proof of Lemma 2.

Since we have chosen R(0) = 0, we now have from (2.1), letting v p denote
the gradient with respect to pg,

(2.2) / \Vφ\2dυ^f |v p φ| 2 dx + c f \ Vφ\\x)\x\4 dx.

^PO ^PO ^PO

From the definition of φ we have

2 W
( ) i j dx.
2
(ε + j x | )

The change of coordinatesy = ε~ιx and simple computation gives

for n = 4,5. Using this in (2.2) then gives (for n = 4,5)

(2.3) [ \Vφ\2dv^[ |Vφ| 2 dx + cp%

J J

where we have used (1.3) which says e 0 * ε ( M ~ 2 ) / 2 .

Observe generally that if f(r) is any nonnegative radial function on B2p, it
follows from Lemma 2 that

(2.4) f fdo<[ fdυp + cj /(W)W4 dx9

J J
Bp Bp hp
where dvp denotes the volume element of pg. We use this to estimate the energy
of φ on B2p — BPQ. First observe that for p 0 < |x| < 2ρ 0 we have

f ( y " ( ) |vG p |<c,

488 RICHARD SCHOEN

where c depends on p and V denotes the spherical gradient, that is, the
gradient with respect to r2hr. From (2.4) we then have

\Vφ\2 dv ^ j \vpφ\2 dvp +cεlj \xf 2

" dx + cpoεl.
B
2p0 ~ Bp0 B
2p0 ~ Bp0 B B
2Po ~ Po

For n = 4,5 this gives

(2.5) ί \Vφ\2dv < ί \Vpφ\2dυp + cpoεl

where c depends on p but not on p 0 .

We now estimate the integral over B2p — B2po. This is more delicate because
we need constants independent of p for which we employ Lemma 1. Throughout
the following argument we use cλ to denote a constant independent of both p
and p 0 . For 2p 0 < |x| < 2p we have φ = ε0Gp, so we estimate the square
gradient of Gp. We first do the radial derivative

r n
2(2 — n)r "~zr~\G o ~ ) +(2 — n) r
or v p '

Using (2.4) on the third function on the right we find, by Lemma 1 and the
fact that we are using normal coordinates,

*2p-J

for any a e (3,4). In particular, for n = 4,5 we get (choosing a = 3.5)

From Lemma 1 we have | vG p | < c / " " " " 1 , from which we easily see

/ IVGJ2 dυ < j I V P G / <fop + d p .

β
2p ~ B2p0 β
B2p ~ 2p0

Combining these with (2.3) and (2.5) we have

(2.6) fB \vφ\2dv*ζfB Iv'φl'dv

 CONFORMAL DEFORMATION OF A RIEMANNIAN METRIC 489

Note that Rp does not vanish in B2p - Bp, so for σ e [p, 2p] we estimate its
integral on Bσ. First observe that in rectangular normal coordinates pgij = ζδiJ
4- (1 — ζ)gjj. We see, by direct calculation,

and hence, by Stokes theorem,

Σ (%,/, "W>)
Since R = 0 in ^ σ , we have

Therefore

ΛpΛp = Γ(σ)/
d

Since JC1 are normal coordinates, the first term in the integral vanishes, and, as
in Lemma 2, Taylor's theorem gives

Σ(g,,-δ,,) = | Σ g»jMχjχk + o(r*).

After integration the quadratic term vanishes and since ζ'(σ) = O(σ~ι) we
finally have

(2.7) / Λp<fcp=

We now compute by Lemma 1 for a e (3,4)

2 f I4-2Λ
dυr
2p p Ίp p

On the other hand we have

490 RICHARD SCHOEN

Applying (2.7) we therefore have

Rp\x\
4 2n
dυp J>-n

Thus for n = 4,5 we have shown (again a = 3.5)

which combined with (2.6) shows

E(φ) < Ep{ψ) + cp0ε20 + clP^2ε20.

Combined with (1.6) we then have

(n-2)/n
E(φ)<Q(S")l

n
-nΛ + n/(n-2) . p 2 •r Λ
l/2 p 2
p0 ε 0 ϊ- cpoεo -r cλp ε 0 .
Arguing as above, we can replace dvp by dυ in the integral to obtain
{n-2)/n
(2.8) E(φ) 2nA 2)
- do

cp0ε20

where c depends on p but c: does not. If we can establish the inequality

(2.9) lim Ap > 0,

then we can finish the proof by fixing p small, then fixing p 0 , and finally
choosing ε 0 sufficiently small. The following lemma gives a condition under
which (2.9) holds.
L e m m a 3 . // the metric G4/(n~2)g is not Ricci flat on M - {o}, then (2.9)
holds.
Proof. For notational simplicity we let g denote p g for any p, and G the
corresponding Green's function. Suppose G 4 / ( "~ 2 ) g is not Ricci flat o n M -
{<?}, and let AT be a compact subset of M — {o} on which G4/(n~2)g is not
Ricci flat. From Lemma 1 we see that for p small the metric G4/(n~2)g is not
Ricci flat on K. Let χ be a smooth nonnegative function with compact support
in M — {o} with χ = 1 on K. For a tensor S = S^ with compact support in
M — { o } we introduce the notation
 CONFORMAL DEFORMATION OF A RIEMANNIAN METRIC 491

and let R* denote the scalar curvature of g'. Let ut denote the solution of

(2.10) A,u,-^=^R'ut = 0 onM-{0},

«,(0) = 1.
Such a solution exists for |/| < δ 0 , with δ 0 depending only on g and S. In fact,
we can write ut = Hfi1, where Ht is the normalized Green's function for the
4Λn 2)
metric g + tG' ~ S which exists for |ί| < δ by Lemma 1.
Since this metric is Euclidean near 0 we have for \x\ small

G(x)=\x\2~" + A+O(\x\),
from which it follows that for |x| small

Integrating (2.10) with respect to dυ\ the volume element of g\ and using
Stoke's theorem we find

J
M-{o)

Differentiating the integral on the right and evaluating at t = 0 we find

d
R\dv< (Ric(g°),s)&0,
)
where we have used R = 0, u0 s 1. Taking 5 = -χRic(g°) we make the term
on the right positive; in fact, since we assume G4/(n~2)g is not Ricci flat on K,
and Gp = G is close to G in C 2 norm (Lemma 1) on the support of χ, we find
d r t t

where 8ι is independent of p. Also by Lemma 1 the metrics g* vary smoothly in

r up to any order on the support of χ uniformly in p. Thus there exists ί0 small
so that A — At > δ 2 with δ 2 > 0 independent of p. Thus we have for p small

This establishes (2.9), and proves Lemma 3.

Finally we must analyze the case when G4/{n~2)g is Ricci flat. We will show
that this can only hold if M is conformally equivalent to S". First observe that
4/(n 2)
(M — {0}, G ~ g) is asymptotically Euclidean in the sense that the metric
g = G*Λn-Dg satisfies for \y\ large, y e R",

(2.11)
492 RICHARD SCHOEN

2
where y = |JC|~ JC, X near 0 in normal coordinates. The following result is of
independent interest.
Proposition 2. Let (N, g) be a Riemannian manifold which is asymptotically
n
Euclidean in the sense o/(2.11). If g is Ricci flat, then (N9 g) is isometric to R
with its Euclidean metric.
Proof. The conclusion of Proposition 2 is not difficult if g is sufficiently
near the Euclidean metric at infinity. We use the Ricci flat assumption to
improve the decay of g. A standard argument asserts the existence of harmonic
coordinates v1,- -,vn defined near infinity such that vι = yι + O d j Ί " 1 ) so that
in terms of υ coordinates the metric satisfies (2.11). We simply rename υι to be
y\ so that without loss of generality we can take >>' to be harmonic coordinates.
The Ricci flat condition then implies for all i,j

This shows that Δg /y = O(\y\~β) and hence elliptic theory gives an improve-
ment on the decay of g; in fact, one derives in a standard way

where (A^) is a constant n X n matrix. By rotating the j> coordinates we may

assume (A^) is diagonal. The condition that yj be harmonic implies An
= | Σ / Au. Since this holds for eachy and n > 2, we conclude that Atj = 0 and
hence we have

We may assume that each yJ is a. global harmonic function on N, possibly

linearly dependent away from infinity. The Bochner formula for yι then shows
Δ| Vj>'|2 = 21VV/I 2 since g is Ricci flat. We clearly have

|v/| = 1 + θ(|.y| "), v|v/| =θ(\y\n).

Integrating the Bochner formula over a large ball gives
3 , ,2

where n is the outer normal to dBR. Letting R go to infinity then shows

V v y ' = 0 and hence vy* is a parallel vector field for each /. It follows
immediately that (y1,- ,yn): N -> Rn defines an isometry. This proves
Proposition 2.
Combining Lemma 3 and Proposition 2 with our previous work, and taking
into account the results of Aubin [1] we get our main theorem on the Sobolev
quotient.
 CONFORMAL DEFORMATION OF A RIEMANNIAN METRIC 493

Theorem 2. Let Mn be a compact Riemannian manifold with n > 3. If M is

not conformally diffeomorphic to Sn, then the Soboleυ quotient Q(M) is strictly
less than Q(Sn).

3. Conformal deformation to constant scalar curvature

We now prove our main theorem concerning conformal deformation. The
argument used here is that of Trudinger [6]. Essentially the same argument was
used later by Aubin [1]. We repeat the argument for the sake of completeness
since it is very simple.
Theorem 3. Any compact Riemannian manifold (Mn, g) has a conformally
related metric u4^n~2)g, u > 0, of constant scalar curvature.
Proof. Let a e [1,(« + 2)/(n - 2)], α 0 = (n + 2)/(/i - 2), and consider
the ratio

where Hλ(M) is the Sobolev space of function with L2 first derivatives. It is

elementary from the Sobolev embedding theorem, and was proved by Yamabe
[8] that there exists, for any a e (1, « 0 ), smooth functions ua > 0, fMu« +1dv
= 1, satisfying

We denote this value by Qa(M) so that β α o (M) = Q(M). Moreover, ua

satisfies the Euler-Lagrange equation

(3-1) Δ«α - 4( " w ~ _ 2 1 ) J ^ t t + Qa{M)u'a = 0.

One attempts to take the limit as a ΐ α 0 . Since we have a uniform bound on the
Hλ norm of wα, by weak compactness we can find a weakly convergent
sequence { wα }. The weak form of (3.1) is

LL ( Vr?
() η
for any η e C°°(M). Since Hx is compactly contained in Lp for any p <
2n/(n — 2), it follows easily that the weak Hλ limit u of the sequence ua
satisfies the limiting equation. (Note that one sees immediately that
lima^aoQa(M) = Q(M).) A regularity result of Trudinger [6] then implies
that u is smooth. One need only show that u is nonzero, and this is where
Theorem 2 enters. Given P e M and p > 0 and small, let η be a smooth
4 9 4
RICHARD SCHOEN

function on M which is equal to one in Bp(P) and zero outside B2p(P).

Multiply (3.1) by η2ua and integrate by parts to get
2 2
η u a+Qa(M)ί

This easily implies for any ε > 0

(1 — ε) I |VηwJ dv^c(ε)ρ~2j ul + Qa(M)[ τj2w£ + 1 ,

where c(ε) depends on ε and M. The Sobolev inequality in Blp holds with the
Euclidean Sobolev constant Q{Sn) plus an error term which is of order ρ2
because the metric is Euclidean up to second order. Therefore we have

(1 - ε)(Q(S") - cp2)l ί
(3.2) V»
2 2
<c(ε)p- ί u a + Qa(M)[ η2uϊ+ ι.
J J
M M
Now observe that η 2 w£ +1 = (rιua)2u"~ι and hence

M \JM J \JM

7J
M

where we have used Holder's inequality twice, have normalized g so that

Vol g (M) = 1, jMuaa + l = 1, and the fact that (a - l)n/2 < a + 1. Since our
theorem is trivial if M is conformally diffeomorphic to Sn, we assume that is
not the case, and hence by Theorem 2 we have Q(M) < Q(Sn). In particular,
for a near α 0 we have Qa{M) < Q(Sn). Now we fix ε, p small enough to
absorb the last term on the right of (3.2) to the left to get

/ \2»/(»-2) , \ ( 2J
}
(ηuj dv\ <c uιadv.
J
J M
Since η is one on Bp(P), we can take a finite covering of M by balls of radius p
and sum these inequalities to obtain
V(Λ"2)/Λ

< c I u2adυ.

Since a + 1 < 2n/(n — 2), this implies

cf/ uldυ.
 CONFORMAL DEFORMATION OF A RIEMANNIAN METRIC 495

2
This gives a uniform lower bound on the L norm of ua. Since Hλ is compactly
2
contained in L , the same lower bound holds on w, and hence u is nonzero.
This completes the proof of Theorem 3.

References

[1] T. Aubin, The scalar curvature, Differential Geometry and Relativity, edited by Cahen and
Flato, Reider, 1976.
[2] T. Parker & C. Taubes, On Witten's proofofthe positive energy theorem, Comm. Math. Phys.
84(1982)223-238.
[3] R. Schoen & S. T. Yau, On the proof of the positive mass conjecture in General Relativity,
Comm. Math. Phys. 65 (1979) 45-76.
[4] , Proof of the positive action conjecture in quantum relativity, Phys. Rev. Lett. 42 (1979)
547-548.
[5] , The geometry and topology of manifolds of positive scalar curvature, in preparation.
[6] N. Trudϊnger, Remarks concerning the conformal deformation of Riemannian structures on
compact manifolds, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 3 (1968) 265-274.
[7] E. Witten, A new proof of the positive energy theorem, Comm. Math. Phys. 80 (1981) 381-402.
[8] H. Yamabe, On a deformation of Riemannian structures on compact manifolds, Osaka Math. J.
12 (1960) 21-37.

UNIVERSITY OF CALIFORNIA, SAN DIEGO