Universit¨ at Regensburg Mathematik
Equivariant Yamabe problem and Hebey-Vaugon conjecture
Farid Madani
Preprint Nr. 12/2011
FARIDMADANI
Abstrat. IntheirstudyoftheYamabeprobleminthepreseneofisometry
group,E.HebeyandM.Vaugonannounedaonjeture. Thisonjeturegen-
eralizesT.Aubin'sonjeture,whihhasalreadybeenprovenandissuient
to solvethe Yamabeproblem. Inthispaper,wegeneralizeAubin'stheorem
andweprovetheHebeyVaugononjetureinsomenewases.
1. Introdution
Let
(M, g)
be a ompat Riemannian manifold of dimensionn ≥ 3
. Denote byI(M, g)
,C(M, g)
andR g
theisometrygroup,theonformaltransformationsgroup and thesalarurvature,respetively. LetG
beasubgroupoftheisometrygroupI(M, g)
. E.HebeyandM.Vaugon[5℄onsideredthefollowingproblem:HebeyVaugonproblem.Istheresome
G −
invariantmetrig 0
whih minimizesthe funtional
J (g ′ ) = R
M R g ′ dv(g ′ ) ( R
M dv(g ′ )) n− n 2
where
g ′
belongstotheG −
invariant onformal lass ofmetrisg
denedby:[g] G := { g ˜ = e f g/f ∈ C ∞ (M ), σ ∗ ˜ g = ˜ g ∀ σ ∈ G }
Thepositiveanswerwouldhavetwoonsequenes. Therstisthatthereexistsan
I(M, g) −
invariantmetrig 0
onformaltog
suh that thesalarurvatureR g 0
isonstant. Theseondisthat theA.Lihnerowiz's onjeture[7℄, statedbelow,is
true. BytheworksofJ.Lelong-Ferrand[6℄andM.Obata[9℄,weknowthatif
(M, g)
is notonformalto
(S n , g can )
(the unit sphere endowed with its standardmetrig can
), thenC(M, g)
is ompat and there exists a onformalmetrig ′
tog
suhthat
I(M, g ′ ) = C(M, g)
. This impliesthat therst onsequeneis equivalenttothe
A.Lihnerowizonjeture. Foreveryompat Riemannianmanifold
(M, g)
whihisnotonformaltotheunitsphere
S n
endowedwithitsstandardmetri,thereexists ametri
˜ g
onformal tog
for whihI(M, ˜ g) = C(M, g)
,andthe salar ur-vature
R ˜ g
isonstant.TosuhmetrisorrespondfuntionswhihareneessarilysolutionsoftheYamabe
equation. Inotherwords,if
˜ g = ψ n−2 4 g
,ψ
isaG −
invariantsmoothpositivefuntionthen
ψ
satises4(n − 1)
n − 2 ∆ g ψ + R g ψ = R g ˜ ψ n− n+2 2 .
1991MathematisSubjetClassiation. 53A30,53C21,35J20.
Key words and phrases. Conformal metri; Isometry group; Salar urvature; Yamabe
problem.
The lassial Yamabe problem, whih onsists to nd a onformal metri with
onstant salar urvature on a ompat Riemannian manifold, is the partiular
aseof theproblemabovewhen
G = { id }
. Denote byO G (P )
theorbitofP ∈ M
under
G
,W g
theWeyltensorassoiatedtothemanifold(M, g)
andω n
thevolumeoftheunit sphere
S n
. Wedenetheintegerω(P )
atthepointP
asω(P ) = inf { i ∈ N / k∇ i W g (P ) k 6 = 0 } (ω(P ) = + ∞
if∀ i ∈ N , k∇ i W g (P ) k = 0)
HebeyVaugon onjeture. Let
(M, g)
bea ompat Riemannian manifold ofdimension
n ≥ 3
andG
be a subgroup ofI(M, g)
. If(M, g)
is not onformal to(S n , g can )
or if the ation ofG
has no xed point, then the following inequalityholds
(1)
inf
g ′ ∈[g] G J (g ′ ) < n(n − 1)ω 2/n n ( inf
Q∈M cardO G (Q)) 2/n
Remarks1.1. (1) ThisonjetureisthegeneralizationoftheformerT.Aubin's
onjeture[1℄fortheYamabeproblemorrespondingto
G = { id }
,wheretheonstant in the right side of the inequality isequal to
inf g ′ ∈[g can ] J(g ′ )
forS n
. In thisase, the onjetureisompletely proved.(2) Theinequalityisobviousif
inf g ′ ∈[g] G J (g ′ )
isnonpositive,itistheasewhen thereexistsaYamabemetri withnonpositive salar urvature.(3) Iffor any
Q ∈ M
,cardO G (Q) = + ∞
thenthis onjetureisalso obvious.Theonlyresultsknownaboutthisonjeturearegivenin thefollowingtheorem:
Theorem 1.1 (E.Hebeyand M.Vaugon). Let
(M, g)
be a smooth ompat Rie-mannian manifoldof dimension
n ≥ 3
andG
beasubgroup ofI(M, g)
. Wealwayshave :
g ′ inf ∈[g] G J(g ′ ) ≤ n(n − 1)ω n 2/n ( inf
Q∈M cardO G (Q)) 2/n
andinequality (1) holdsifone ofthe following itemsissatised.
(1) Theationof
G
onM
isfree(2)
3 ≤ dim M ≤ 11
(3) Thereexistsapoint
P
withminimalorbit(nite)underG
suhthatω(P ) >
(n − 6)/2
orω(P ) ∈ { 0, 1, 2 }
.Thease
ω = 3
wasstudiedbyA.Rauzy (privateommuniations).Inthisproveweprovethefollowingresults:
Maintheorem.TheHebeyVaugononjetureholdsifthereexistsapoint
P ∈ M
with minimal orbit(nite)for whih
ω(P ) ≤ 15
orif the degreeof theleading partof
R g
isgreateror equaltoω(P ) + 1
,in theneighborhoodof thispointP
.Corollary 1.1. HebeyVaugon onjeture holds for every smooth ompat Rie-
mannian manifold
(M, g)
of dimensionn ∈ [3, 37]
.To provethe main theorem, we need to onstrut a
G −
invariant test funtionφ
suhthat
I g (φ) < n(n − 1)ω 2/n n ( inf
Q∈M cardO G (Q)) 2/n
Thus,allthediultiesareintheonstrutionofasuhfuntion. Forsomeases,
we anuse the test funtions onstruted by T. Aubin [1℄ and R. Shoen [10℄ in
M. Vaugon[5℄. Butthe item3,presentedin Theorem 1.1,usestest funtions dif-
ferentthanT.AubinandR.Shoenones.
WemultiplyT.Aubin'stestfuntion
u ε,P
byafuntionasfollows:(2)
ϕ ε (Q) = (1 − r ω+2 f (ξ))u ε,P (Q)
(3)
u ε,P (Q) =
ε
r 2 + ε 2 n− 2 2
− ε
δ 2 + ε 2 n− 2 2
if
Q ∈ B P (δ)
0
ifQ ∈ M − B P (δ)
for all
Q ∈ M
, wherer = d(Q, P )
is thedistane betweenP
andQ
.(r, ξ j )
is ageodesioordinatessystemintheneighborhoodof
P
andB P (δ)
isthegeodesiballofenter
P
withradiusδ
xedsuientlysmall.f
isafuntiondependingonlyonξ
,hosensuhthat
R
S n− 1 f dσ = 0
. Withoutlossofgenerality,wesupposethatinthe oordinatessystem(r, ξ j )
wehavedet g = 1 + o(r m )
form ≫ 1
. Infat,E.Hebeyand M.Vaugonprovedthat thereexists
g ˜ ∈ [g] G
for whihdet ˜ g = 1 + o(r m )
andinf g ′ ∈[g] G J (g ′ )
doesnotdependontheonformalG −
invariantmetri.2. Computationof
R
M R g ϕ 2 ε dv
Letbe
I a b (ε) = Z δ/ε
0
t b
(1 + t 2 ) a dt
andI a b = lim
ε→0 I a b (ε)
then
I a 2a−1 (ε) = log ε −1 + O(1)
. If2a − b > 1
thenI a b (ε) = I a b + O(ε 2a−b−1 )
andbyintegrationbyparts,weestablishthefollowingrelationships:
(4)
I a b = b − 1
2a − b − 1 I a b−2 = b − 1
2a − 2 I a−1 b−2 = 2a − b − 3
2a − 2 I a−1 b , 4(n − 2)I n n+1 (I n n−2 ) (n−2)/n = n
Using the inequality
(a − b) β ≥ a β − βa β−1 b
for0 < b < a
, wehave forβ ≥ 2
,0 ≤ α < (n − 2)(β − 1) − n
(5)
Z
M
r α u β ε,P dv = ω n−1 I (n−2)β/2 α+n−1 ε α+n−β(n−2)/2 + O(ε n−2 )
This integralappearsfrequentlyinthefollowingomputations,anditallowsusto
neglet the onstant term in the expression of
u ε
, when we hooseδ
suientlysmall and
ε
smallerthanδ
.Denote by
I g
theYamabefuntional denedforallψ ∈ H 1 (M )
by(6)
I g (ψ) = Z
M |∇ g ψ | 2 dv + (n − 2) 4(n − 1)
Z
M
R g ψ 2 dv
k ψ k −2 N
where
N = 2n/(n − 2)
and∇ g
isthegradientofthemetrig
.The seond integral of the funtional
I g
with the salar urvature term needs aspeial onsideration. Let
µ(P)
beanintegerdened asfollows:|∇ β R g (P ) | = 0
forall
| β | < µ(P )
andthereexistsγ ∈ N µ(P)
suhthat|∇ γ R g (P ) | 6 = 0
thenR g (Q) = ¯ R + O(r µ(P)+1 )
where
R ¯ = r µ(P) P
|β|=µ ∇ β R g (P )ξ β
isahomogeneouspolynomialofdegreeµ(P)
,the
β
aremulti-indies.Forsimpliity,wedroptheletter
P
inω(P)
andµ(P )
.Lemma 2.1.
µ ≥ ω
,g ij = δ ij + O(r ω+2 )
andR ¯
S(r) R g = O(r 2ω+2 )
whih impliesthat
R
S(r) Rdσ ¯ = 0
whenµ < 2ω + 2
¯ R
denotestheaverage. Then
Z
M
R g ϕ 2 ε dv = Z
M
R g u 2 ε,P dv − 2 Z
M
f u 2 ε,P R g r ω+2 dv + Z
M
f 2 u 2 ε,P R g r 2ω+4 dv
= ε 2ω+4 ω n−1
¯ Z
S(r)
r −2ω−2 R g dσI n−2 n+2ω+1 (ε) − 2ε ω+µ+4 I n−2 ω+µ+n+1 (ε)ω n−1
¯ Z
S(r)
r −µ f (ξ) ¯ Rdσ(ξ) + O(ε n−2 )
(7)
MoreoverT.Aubin[2℄provedthat:
Theorem2.1. If
µ ≥ ω + 1
thenthere existsC(n, ω) > 0
suhthat¯ Z
S n− 1 (r)
Rdσ = C(n, ω)( − ∆ g ) ω+1 R(P )r 2ω+2 + o(r 2ω+2 ) ( − ∆ g ) ω+1 R(P )
isnegative. ThenI g (u ε,P ) < n(n−2) 4 ω 2/n n−1
.Fromnowuntiltheendofthissetion,wemaketheassumptionthat
µ = ω
. Now,wereallsomeresultsobtainedbyT.Aubin inhispapers[3,4℄:
R ¯
ishomogeneouspolynomialofdegreeω
then∆ E R ¯
ishomogeneousofdegreeω − 2
and
∆ E R ¯ = r −2 (∆ s R ¯ − ω(n + ω − 2) ¯ R)
where
∆ E
istheEulideanLaplaianand∆ s
istheLaplaianonthesphereS n−1
.∆ k−1 E R ¯
ishomogeneousofdegreeω − 2k + 2
and∆ k E R ¯ = r −2 (∆ s − ν k id)∆ k−1 E R ¯ = r −2k
k
Y
p=1
(∆ S − ν p id) ¯ R
with
(8)
ν k = (ω − 2k + 2)(n + ω − 2k)
The sequeneof integers
(ν k ) {1≤k≤[ω/2]}
is dereasing. It willplaytherole of theeigenvaluesoftheLaplaianonthesphere
S n−1
. It isknownthat theeigenvalues of the geometri Laplaian are non-negative and inreasing. Ourν k
are in theoppositeorder.
WeknowbyT.Aubin'spaper[2℄that
∆ [ω/2] E R ¯ = 0
andR
S(r) Rdσ ¯ = 0
,thenq = min { k ∈ N /∆ k E R ¯ = 0 }
is well dened and
r −ω R ¯ ∈ L q
k=1 E k
, withE k
the eigenspae assoiated to thepositiveeigenvalues
ν k
oftheLaplaian∆ s
onthesphereS n−1
. Ifj 6 = k
,thenE k
isorthogonalto
E j
,forthestandardsalarprodutinH 1 2 (S n−1 )
. Moreover,sineR Rdσ ¯ = 0
thereexistϕ k ∈ E k
(eigenfuntionsof∆ s
)suhthat(9)
R ¯ = r ω ∆ s
q
X
k=1
ϕ k = r ω
q
X
k=1
ν k ϕ k
AordingtoLemma 2.1,weansplit themetri
g
inthefollowingway:(10)
g = E + h
where
E
is the Eulidean metri andh
is a symmetri 2-tensor dened in ourgeodesioordinatessystemby
(11)
h ij = r ω+2 ¯ g ij + r 2(ω+2) g ˆ ij + ˜ h ij
andh ir = h rr = 0
where
¯ g
,g ˆ
andh ˜
are symmetri2-tensorsdened onthesphereS n−1
. Wedenoteby
s
the standard metri on the sphere,∇
,∆
are the assoiated gradient andLaplaianon
S n−1
. Bystraightforwardomputations, Aubin[3℄provedthat:Lemma2.2.
R ¯ = ∇ ij ¯ g ij r ω
and¯ Z
S n− 1 (r)
Rdσ = [B/2 − C/4 − (1 + ω/2) 2 Q]r 2(ω+1) + o(r 2(ω+1) )
where
B = ¯ R
S n− ∇ i 1 g ¯ jk ∇ j ¯ g ik dσ
,C = ¯ R
S n− ∇ i 1 g ¯ jk ∇ i g ¯ jk dσ
andQ = ¯ R
S n− 1 g ¯ ij ¯ g ij dσ
Forfurther detailsreferto [8℄.
Theintegrals
Q
,B
andC
aregivenintermsofthetensorg ¯
. Ourgoalistoomputethem usingtheeigenfuntions
ϕ k
above. Letusdeneb ij =
q
X
k=1
1
(n − 2)(ν k + 1 − n) [(n − 1) ∇ ij ϕ k + ν k ϕ k s ij ]
and
a ij
suh that¯ g ij = a ij + b ij
then,aordingto(9),wehekthat(12)
R ¯ = ¯ R b = ∇ ij b ij r ω
andR ¯ a = ∇ ij a ij r ω = 0
If
g ¯ ij = a ij
thenR ¯ = ¯ R a = 0
andµ ≥ ω + 1
. ByTheorem2.1¯ Z
S n−1 (r)
Rdσ = Z ¯
S n−1 (r)
R a dσ < 0
If
g ¯ ij = b ij
then¯ Z
S n− 1 (r)
Rdσ = Z ¯
S n− 1 (r)
R b dσ = [B b /2 − C b /4 − (1 + ω/2) 2 Q b ]r 2(ω+1) + o(r 2(ω+1) )
where
B b
,C b
andQ b
are thesameintegralsdened in Lemma 2.2when theon-sideredtensor
g ¯ ij = b ij
. Weomputethemintermsofϕ k
Q b = Z ¯
S n− 1
¯ b ij ¯ b ij dσ = n − 1 n − 2
q
X
k=1
ν k
ν k − n + 1
¯ Z
S n− 1
ϕ 2 k dσ
B b = − (n − 1)Q b +
q
X
k=1
ν k
¯ Z
S n− 1
ϕ 2 k dσ
C b = − (n − 1)Q b + n − 1 n − 2
q
X
k=1
ν k
¯ Z
S n− 1
ϕ 2 k dσ
Tondtheseexpressions,weusedseveraltimestheidentity
∇ i b ij = − P q
k=1 ∇ j ϕ k
and Stokesformula (more detailsare given in [3,4℄ and [8℄). Inthe generalase,
wededuethat
Lemma2.3. If
µ = ω
andg ¯ ij = a ij + b ij
,whereb ij
isdenedabove,(13)
¯ Z
S n − 1 (r)
Rdσ = Z ¯
S n − 1 (r)
R a + R b dσ ≤ [B b /2 − C b /4 − (1 + ω/2) 2 Q b ]r 2(ω+1) + o(r 2(ω+1) )
and
(14)
B b /2 − C b /4 − (1 + ω/2) 2 Q b =
q
X
k=1
u k
¯ Z
S n − 1
ϕ 2 k dσ
with
(15)
u k =
n − 3
4(n − 2) − (n − 1) 2 + (n − 1)(ω + 2) 2 4(n − 2)(ν k − n + 1)
ν k
u k
isobtainedusingtheexpressionsofQ b
,B b
andC b
above.3. Generalization of T.Aubin's theorem
Theorem3.1. Ifthereexists
P ∈ M
suhthatω(P) ≤ (n − 6)/2
thenthereexistsf ∈ C ∞ (S n−1 )
withvanishingmeanintegral suhthatI g (ϕ ε ) < n(n − 2) 4 ω n−1 2/n
The ase
ω = 0
of thethistheorem hasalreadybeenprovenby T.Aubin [1℄. Healsoprovedthetheoremwhen
µ ≥ ω + 1
(seeTheorem2.1).Fromnowuntiltheendofthispaper,wedroptheletter
P
inω(P)
andµ(P)
.Proof. If
µ ≥ ω + 1
thentheinequalityholdsbyTheorem2.1. Sowesupposethatµ = ω
until theend of theproof. Westartby omputingthe rstintegralof theYamabefuntional(6)with
ψ = ϕ ε
. Usingformula|∇ g ϕ ε | 2 = (∂ r ϕ ε ) 2 +r −2 |∇ s ϕ ε | 2
,weobtain:
Z
M |∇ g ϕ ε | 2 dv = Z
M |∇ g u ε,P | 2 dv + Z δ
0
[∂ r (r (ω+2) u ε,P )] 2 r n−1 dr Z
S n− 1
f 2 dσ+
Z δ 0
u 2 ε,P r n+2ω+1 dr Z
S n− 1
|∇ f | 2 dσ
Thesubstitution
t = r/ε
gives(16)
Z
M |∇ g ϕ ε | 2 dv = (n − 2) 2 ω n−1 I n n+1 (ε) + ε 2ω+4 Z
S n−1
|∇ f | 2 dσI n−2 2ω+n+1 (ε)+
Z
S n− 1
f 2 dσ[(ω − n+4) 2 I n 2ω+n+5 (ε)+2(ω+2)(ω − n+4)I n 2ω+n+3 (ε)+(ω+2) 2 I n 2ω+n+1 (ε)]
For
k ϕ ε k −2 N
, weneedto omputetheTaylorexpansionof:ϕ N ε (Q) = [1 − N r ω+2 f (ξ) + N (N − 1)
2 r 2ω+4 f 2 (ξ) + o(r 2ω+4 )]u N ε,P
Usingthefat that
R
S n− 1 f dσ(ξ) = 0
andformula(5),weonludethatk ϕ ε k N N =
Z δ 0
Z
S n−1
[1 + N(N − 1)
2 r 2(ω+2) f 2 (ξ) + o(r 2ω+4 )]r n−1 u N ε,P drdσ(ξ)
= ω n−1 I n n−1 + N (N − 1) 2 ε 2(ω+2)
Z
S n− 1
f 2 dσI n 2ω+n+3 + o(ε 2ω+4 )
then
(17)
k ϕ ε k −2 N = (ω n−1 I n n−1 ) −2/N 1
− (N − 1)ε 2(ω+2) Z
S n−1
f 2 dσI n 2ω+n+3 /(ω n−1 I n n−1 ) + o(ε 2ω+4 )
ByEqs(16),(17) ,(7)andtherelationship(4),if
n > 2ω + 6
then:I g (ϕ ε ) = n(n − 2)
4 ω 2/n n−1 + (ω n−1 I n n−1 ) −2/N I n−2 n+2ω+1 ε 2ω+4 × (n − 2)ω n−1
4(n − 1)
¯ Z
S(r)
r −2ω−2 R g dσ − n − 2 2(n − 1)
Z
S n− 1
f (ξ) ¯ Rdσ + Z
S n− 1
|∇ f | 2 dσ+
− n(n − 2) 2 − (ω + 2) 2 (n 2 + n + 2) (n − 1)(n − 2)
Z
S n− 1
f 2 dσ
+ o(ε 2ω+4 )
If
n = 2ω + 6
thenI g (ϕ ε ) = n(n − 2)
4 ω 2/n n−1 + (ω n−1 I n n−1 ) −2/N ε 2ω+4 log ε −1 × (n − 2)ω n−1
4(n − 1)
¯ Z
S(r)
r −2ω−2 R g dσ − n − 2 2(n − 1)
Z
S n− 1
f (ξ) ¯ Rdσ+
Z
S n− 1
|∇ f | 2 dσ + (ω + 2) 2 Z
S n− 1
f 2 dσ
+ O(ε 2ω+4 )
Forfurther detailsreferto [8℄.
Let
I S
be the funtional dened for a funtionf
on the sphereS n−1
, with zeromeanintegral,by
I S (f ) = Z ¯
S n − 1
4(n − 1)(n − 2) |∇ f | 2 − [4n(n − 2) 2 − 4(ω + 2) 2 (n 2 + n + 2)]f 2 +
− 2(n − 2) 2 f Rdσ ¯
This impliesthat if
n > 2ω + 6
(18)
I g (ϕ ε ) = n(n − 2)
4 ω n−1 2/n + ω n−1 2/n I n−2 n+2ω+1 ε 2ω+4 4(n − 1)(n − 2)(I n n−1 ) 2/N × { (n − 2) 2 Z ¯
S(r)
r −2ω−2 R g dσ + I S (f ) } + o(ε 2ω+4 )
andif
n = 2ω + 6
(19)
I g (ϕ ε ) = n(n − 2)
4 ω n−1 2/n + ω n−1 2/n I n−2 n+2ω+1 ε 2ω+4 log ε −1 4(n − 1)(n − 2)(I n n−1 ) 2/N × { (n − 2) 2 Z ¯
S(r)
r −2ω−2 R g dσ + I S (f ) } + O(ε 2ω+4 )
Notie that if
k 6 = j
thenI S (ϕ k + ϕ j ) = I S (ϕ k ) + I S (ϕ j )
. Indeed,ϕ k
andϕ j
areorthogonalforthestandardsalarprodutin
H 1 2 (S n−1 )
.I S (c k ν k ϕ k ) =
d k c 2 k − 2(n − 2) 2 c k ν k 2 Z ¯
S n−1
ϕ 2 k dσ
= − (n − 2) 4 d k
ν k 2 Z ¯
S n− 1
ϕ 2 k dσ
where
d k = 4[(n − 1)(n − 2)ν k − n(n − 2) 2 + (ω + 2) 2 (n 2 + n + 2)]
andc k = (n − 2) 2
d k
Using (8) , weanhekeasily that
d k
is positiveforany1 ≤ k ≤ [ω/2]
. Now,letus onsider
f = P q
1 c k ν k ϕ k
. ThenI S (f ) = −
q
X
1
(n − 2) 4 d k
ν 2 k Z ¯
S n− 1
ϕ 2 k dσ
andbyLemma 2.3
(n − 2) 2 Z ¯
S(r)
r −2ω−2 R g dσ + I S (f ) ≤
q
X
1
(u k (n − 2) 2 − (n − 2) 4 d k
ν k 2 ) Z ¯
S n−1
ϕ 2 k dσ + o(1)
Thefollowinglemmaimpliesthat
I g (ϕ ε ) < n(n−2) 4 ω n−1 2/n
Lemma3.1. Forany
k ≤ q ≤ [ω/2]
the following inequality holdsu k − (n − 2) 2
d k
ν k 2 < 0
Proof. Realltheexpressionof
ν k
givenin(8). Thesequene(U k )
denedbyU k := (ν k − n + 1)d k { (n − 2) u k
ν k − (n − 2) 3 d k
ν k }
is polynomial dereasing in
ν k
whenν k ≥ 0
. In fat,U k = P (ν k )
withP
thedereasingpolynomialin
R +
,dened byP (x) = [(n − 1)(n − 2)x − n(n − 2) 2 + (ω + 2) 2 (n 2 + n + 2)] ×
[(n − 3)(x − n + 1) − (n − 1) 2 − (n − 1)(ω + 2) 2 ] − (n − 2) 3 (x 2 − (n − 1)x)
Thederivativeof
P
isP ′ (x) = − 2(n − 2)x − 2n(n − 2) 3 + 2(n 2 − 3n − 2)(ω + 2) 2
Byassumption
ω + 2 ≤ (n − 2)/2
thenP
isdereasinginR +
. HeneU k = P (ν k ) ≤ P (ν ω/2 ) = U ω/2
forall
k ≤ ω/2
. Iteasytohekthatu ω/2
isnegativesoU k ≤ U ω/2 < 0
.4. Proof of the maintheorem
ByRemarks 1.1, we onsideronly thepositivease(i.e.,
inf g ′ ∈[g] G J(g ′ ) > 0
)andtheasewhenthereexists
P ∈ M
suhthatO G (P) = { P i } 1≤i≤m , m = cardO G (P ) = inf
Q∈M cardO G (Q), ω ≤ n − 6
2
andP 1 = P
Let
ϕ ˜ ε,i
beafuntion denedasfollows:(20)
ϕ ˜ ε,i (Q) = (1 − r ω+2 i f i (ξ))u ε,P i (Q)
where
r i = d(Q, P i )
,thefuntionu ε,P i
isdenedasin(3)andf i
isdenedby:(21)
f i (Q) = cr i −ω ∇ ω g R (P i ) (exp −1 P i Q, · · · , exp −1 P i Q)
exp P i
istheexponentialmap. Inageodesioordinatessystem{ r, ξ j }
withoriginP
,induedbytheexponentialmapf 1 = cr −ω R ¯ = c
q
X
k=1
ν k ϕ k
where
R ¯
,ϕ k
andν k
are denedin Setion2. Thus thefuntionsf i
aredened onthesphere
S n−1
. Thehoieoftheonstantc
isimportant.Lemma 4.1. Suppose that
ω ≤ (n − 6)/2
. Ifω ∈ [3, 15]
orifdeg ¯ R ≥ ω + 1
thenthereexists
c ∈ R
suhthat the orresponding funtionsϕ ˜ ε,i
satisfy :(22)
I g ( ˜ ϕ ε,i ) < 1
4 n(n − 2)ω n 2/n
Remarks4.1. (1) Weprovedinequalityof thislemmafor any
ω ≤ (n − 6)/2
,using test funtion
ϕ ε
(see Theorem 3.1 ). We notie that the dierenebetween
ϕ ε
andϕ ˜ ε,i
isontheonstrutionoftheorrespondingfuntionsf
and
f i
respetively. Fromϕ ˜ ε,i
wedene aG −
invariantfuntion (seeproofofthe maintheorem below), this property is notpossible with the funtion
ϕ ε
.(2) For
ω = 16
andn
suientlybig,weanhekthatforanyc ∈ R
,inequality (22)isfalse.Proof. 1. If
deg ¯ R ≥ ω + 1
,thenbyTheorem2.1I g (u ε,P i ) < n(n − 2)
4 ω n 2/n
It issuientto take
c = 0
,heneϕ ˜ ε,i = u ε,P i
.2. If
deg ¯ R = ω
. UsingestimatesgivenintheproofofTheorem3.1(see(18),(19)),itissuienttoshowthat thereexists
c ∈ R
suh that (23)I S (f 1 ) + (n − 2) 2 Z ¯
S(r)
r −2ω−2 R g dσ r < 0
WekeepthenotationsusedintheproofofTheorem3.1. Thus
I S (f 1 ) =
q
X
k=1
I S (cν k ϕ k ) =
d k c 2 − 2(n − 2) 2 c ν k 2 Z ¯
S n− 1
ϕ 2 k dσ
and
¯ Z
S(r)
r −2ω−2 R g dσ r =
q
X
k=1
u k
¯ Z
S n−1
ϕ 2 k dσ
Toproveinequality(23) ,itissuienttoprovethat
(24)
∀ k ≤ q d k
2(n − 2) c 2 − (n − 2)c + (n − 2) u k
2ν k 2 < 0
Theleftsideoftheinequalityaboveisaseonddegreepolynomialwithvariable
c
,his disriminantis:
(25)
∆ k = (n − 2) 2 − d k u k
ν k 2
Using Lemma 3.1, wededuethat forany
k ≤ q
,∆ k > 0
. Hene,thepolynomialaboveadmitstwodierentrootsdenoted
x k < y k
andgivenbyx k = (n − 2) 2 − (n − 2) √
∆ k
d k
, y k = (n − 2) 2 + (n − 2) √
∆ k
d k
Inequality(24)holdsifandonlyif
(26)
q
\
k=1
(x k , y k ) 6 = ∅
Thesequene
(d k ) k≤[ω/2]
dereases. Itiseasytohekthat(27)
∀ k < j ≤ [ ω
2 ] x k < y j
Hene intersetion(26)isnotemptyif
(28)
∀ k < j ≤ [ ω
2 ] x j < y k
Wealsohekthat if
ω
iseven,u ω/2 < 0
,whihimpliesx ω/2 < 0
.i.
Ifω = 3
thenq = 1
,intersetion aboveisnotempty. Itissuienttotakec = (x 1 + y 2 )/2
.ii.
Ifω = 4
thenk ∈ { 1, 2 }
,x 2 < 0
(beauseu 2 < 0
)and0 < x 1 < y 2
. Heneintersetion
]x 1 , y 1 [ ∩ ]x 2 , y 2 [
isnotempty.iii.
If5 ≤ ω ≤ 15
,itissuienttoprove(28)whihisequivalenttoprovethat(29)
∀ k < j ≤ [ ω
2 ] (n − 2)(d j − d k ) + d k p
∆ j + d j
p ∆ k > 0
Notiethat
∆ k
givenby(25)isarationalfrationinn
. Bystraightforward omputations,wehekthat there existsreel numbersa k , b k , e k , h k
ands k
whih dependonk
andω
suhthat∆ k = a k n 2 + b k n + e k + h k
n − 2 + s k
ν k + 1 − n
(30)
p ∆ k > √ a k (n + b k
2a k
)
(31)
Inequality(29)holdsifweuse(31).
Theexpressions of thereel numbers aboveare known expliitly (weused
thesoftwareMaple to ompute them, see [8℄). Forsimpliity, we omitto
givetheseexpressions.
Proof of themain theorem. Theorbitof
P
under theationofG
issupposedto beminimal(i.e.
cardO G (P) = inf Q∈M cardO G (Q)
). Without lossof generality, wesupposethat3 ≤ ω ≤ (n − 6)/2
,beauseifω > (n − 6)/2
orω ≤ 2
,weonludeusing Theorem1.1. Fromfuntions
ϕ ˜ ε,i
dened by(20),wedenethefuntionφ ε
asfollows:
φ ε =
m
X
k=1
˜ ϕ ε,i
φ ε
isG −
invariant. Infat,foranyσ ∈ G
,suhthatσ(P i ) = P j
u ε,P i = u ε,P j ◦ σ
andf i = f j ◦ σ f i
aredenedby(21),wededuethat˜
ϕ ε,i = ˜ ϕ ε,j ◦ σ
The support of
ϕ ˜ ε,i
is inludedin the ballB P i (δ)
. Wehooseδ
suientlysmallsuhthat forallintegers
i 6 = j
in[1, m]
,intersetionB P j (δ) ∩ B P i (δ) = ∅
. ThusI g (φ ε ) = (cardO G (P)) 2/n I g (ϕ ε )
ByLemma4.1,weonludethat
I g (φ ε ) < n(n − 2)
4 ω 2/n n−1 (cardO G (P )) 2/n
It remainsto notiethatif
˜ g = φ 4/(n−2) ε g
thenJ(˜ g) = 4 n − 1
n − 2 I g (φ ε ) < n(n − 1)ω n−1 2/n (cardO G (P )) 2/n
where
ε
issuientlysmallerthanδ
.Proof of theCorollary 1.1. Supposethat theorbitof
P
undertheationofG
isminimal (otherwisetheonjetureis obvious).
If
ω = ω(P) > [(n − 6)/2]
,weonludeusingTheorem 1.1.If
ω ≤ [(n − 6)/2] ≤ 15
,weonludeusingmain theorem.Referenes
1. T.Aubin, Équationsdiérentiellesnon linéaires et problème deYamabe,J.Math. Pureset
appl55(1976),269296.
2. ,Surquelquesproblèmesdeourburesalaire,J.Funt.Anal240(2006),269289.
3. ,Solution omplète de la
C 0
ompaitéde l'ensemble dessolutions de l'équation de Yamabe,J.Funt.Anal.244(2007),579589.4. ,Onthe
C 0
ompatness ofthesetofthesolutionsoftheYamabeequation,Bull.Si.Math(2008).
5. E.Hebey andM.Vaugon,Le problèmede Yamabe équivariant,Bull.Si.Math117(1993),
241286.
6. J.Lelong-Ferrand,Mém.Aad.RoyaleBelgique,ClassedesSienes39(1971).
7. A.Lihnerowiz,Surlestransformationsonformesd'unevariétériemannienneompate,C.
R.Aad.Si.Paris259(1964).
8. F.Madani,LeproblèmedeYamabeavesingularitésetlaonjeturedeHebeyVaugon,Ph.D.
thesis,UniversitéPierreetMarieCurie,2009.ArXiv:0910.0562.
9. M.Obata, Theonjetures on onformal transformations of riemannian manifolds, J.Di.
Geom.6(1971),247258.
10. R.Shoen,Conformaldeformation ofa riemannianmetritoonstantsalarurvature,J.
Dier.Geom20(1984),479495.
Institut de Mathématiques de Jussieu, Université Pierre et Marie Curie, Équipe:
d'AnalyseComplexeetGéométrie,175,rue Chevaleret,75013Paris,Frane.
E-mail address: madanimath.jussieu.fr