Conformal Deformation of A Riemannian Metric To Constant Scalar Curvature
Conformal Deformation of A Riemannian Metric To Constant Scalar Curvature
Conformal Deformation of A Riemannian Metric To Constant Scalar Curvature
DIFFERENTIAL GEOMETRY
20 (1984) 479-495
CONFORMAL DEFORMATION
OF A RIEMANNIAN
METRIC TO CONSTANT SCALAR CURVATURE
RICHARD SCHOEN
2
(0.1) Δw - " Ru + u<n + 2)Λn-2)
Λ
= 0,
4(Λ - 1)
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
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.
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
ί
uE(x) for|x|<p 0 ,
Λ
/ | V w J dx = n ( n - 2 ) J u ε " l) dx
dυ = dv
M-BP
Since \a(x)\ < C|JC|, we see that | v(ψα)| < c for p 0 < |JC| < 2ρ0. Therefore we
have for a constant c
M-BOΛ
(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
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
Since the error terms are allowable, we see that the above proof succeeds for
n = 3.
\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),
= on_x -
kj
^ k
Therefore, by Taylor's theorem,
1
L, aijkxx u
^ ^ L> ΞkkjjXx ^ x ^ \\x\ )i
where dξ is the volume element for the unit Sn~ι. Then we have from above
2 W
( ) i j dx.
2
(ε + j x | )
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
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
From Lemma 1 we have | vG p | < c / " " " " 1 , from which we easily see
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,
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
After integration the quadratic term vanishes and since ζ'(σ) = O(σ~ι) we
finally have
(2.7) / Λp<fcp=
Rp\x\
4 2n
dυp J>-n
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
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
«,(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)
(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
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
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
/ \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.
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.