HAL Id: hal-00257734
https://hal.archives-ouvertes.fr/hal-00257734
Preprint submitted on 20 Feb 2008
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Eigenvalues estimate for the Neumann problem on bounded domains
Bruno Colbois, Daniel Maerten
To cite this version:
Bruno Colbois, Daniel Maerten. Eigenvalues estimate for the Neumann problem on bounded domains.
2008. �hal-00257734�
hal-00257734, version 1 - 20 Feb 2008
BRUNOCOLBOISANDDANIELMAERTEN
Abstrat. Inthisnote,weinvestigateupperboundsoftheNeumanneigenvalueprob-
lem for the Laplaian of a domainΩ ina given omplete(not ompat a priori) Rie-
mannian manifold (M, g). For this, we use test funtions for the Rayleigh quotient
subordinatedtoafamilyofopensetsonstrutedinageneralmetriway,interestingfor
itself. Asappliations,weprovethatiftheRiiurvatureof(M, g)isboundedbelow Ricg≥ −(n−1)a2,a≥0,thenthereexistonstantsAn>0, Bn>0onlydependingon
thedimension,suhthat
λk(Ω)≤Ana2+Bn
k
V 2/n
,
where λk(Ω) (k ∈ N∗) denotes the kth eigenvalue of the Neumann problem on any
bounded domainΩ ⊂M of volumeV = Vol(Ω, g). Furthermore, this upper boundis learlyinagreementwiththeWeyllaw. Asaorollary,wegetalsoanestimatewhihis
analogoustoBuser'supperboundsofthespetrumofaompatRiemannianmanifold
withlowerboundontheRiiurvature.
1. Introdution
ThegoalofthispaperistogiveupperboundsforthespetrumoftheLaplaianatingon
ompatdomainsofgivenvolumeofaompleteRiemannianmanifoldwithRiiurvature
bounded below, and, as far as possible, to make these estimates optimal withrespet to
theWeyl law.
For ompat Riemannian manifolds without boundary,the following resultwas proved
byP.Buserin[3 ℄(Satz7),[4℄(Thm. 6.2())(seealsoLi-Yauin[13 ℄(Thm.16)). If{λk}∞k=1
denotethe spetrumofthe Laplaianating onfuntions, then:
1.1. Theorem. Let(Mn, g) be a ompat n-dimensionalRiemannian manifoldwithRii urvature bounded below Ricg≥ −(n−1)a2, a≥0, andof volume V.
There exists a onstant Cn ≥ 1 only depending on the dimension, suh that for all k∈N∗, we have
(1.1) λk(M, g) ≤ (n−1)2
4 a2+Cn k
V 2/n
.
1.2. Remarks. (i) In [13℄, the onstant Cn dependsalso on the diameter.
(ii) Indimensionhigherthan2,anormalizationonthevolumeisnotenoughtoontrol
the spetrum: namely, on any ompat manifold of dimension higher than 2, one
an nda metri of given volume, witharbitrarily large rstnonzero eigenvalue
λ2 of the Laplaian,in vertue of the result of B. Colboisand J. Dodziuk [6℄.
Date:20thFebruary2008.
Keywordsandphrases. Neumannspetrum,upperbound,Weyllaw,metrigeometry.
2000 MathematisSubjetClassiation. 35P15,53C99,51F99.
(iii) When Ricg ≥0, we dedue that there exists Cn >1 with λk(M, g) ≤Cn Vk2/n
for all k. However, when Rii is not supposed positive, then the presene of a
termlike
(n−1)2
4 a2 isneessary: bya result ofR. Brooks [2℄, it ispossibletonda
familyofompathyperboli manifoldswithvolumegoingtoinnityanda positive
uniform lower bound on the rst nonzero eigenvalue.
The idea of the proof of Theorem 1.1 is to onsider k disjoint balls of radius r whih
almostover the manifold (M, g),with r around Vk1/n
,and to apply thenCheng's the-
orem [5 ℄. However, suh a theoremdoes not exist on manifolds with boundary, and with
Neumann boundary ondition. A reason for this is that there is no Bishop-Gromov the-
orem: indeed, even for a Eulidean domain, it is not possible to ontrol the volume of a
ballofradius2r withrespettothevolumeofa ballofradius rand sameenter. See also
Example1.4in[4 ℄.
ThisdoesnotmeanthataresultinthespiritofTheorem1.1doesnotexistfordomains.
Namely, P. Kröger [12℄ proved thanks to harmoni analysis, that on bounded Eulidean
domains, the kth eigenvalue of the Neumann problem was bounded by above by some
expressionCn(k/|Ω|)n/2,whereCnonlydependsuponthedimension. Ananalogousresult
an be derived from the muh more general and diult work of N. Korevaar [11℄ ( see
also [10℄), for bounded domains of nonnegative Rii urvature manifolds, and also for
bounded domainsof negative Rii urvature ompat manifolds (in this ase thebound
dependson the diameter).
This naturally leads tothe
Question: What an be said for bounded domains of a omplete Riemannian manifold
withRiiurvaturebounded below ?
Inthisnote,weonsidertheNeumanneigenvalueproblemfortheLaplaianofabounded
domainΩwithsmooth boundary,inagiven omplete(notompat a priori)Riemannian
manifold (M, g). More preisely, we searh for a ouple (λ, u) ∈ R×C∞ Ω
whih is a
solutionof the following boundary elliptiproblem
∆u=λu onΩ
∂u
∂ν = 0 on∂Ω,
where ∆is the nonnegative Laplaianof the metri g and ν theoutward unit normal of
∂Ω. SineΩisboundedwithsmoothboundary,thespetrumof ∆onΩisan unbounded
sequeneofreal numbers (λk(Ω))k∈N∗ whihan be inreasinglyordered
0 =λ1(Ω)< λ2(Ω)≤ · · · ≤λk(Ω)≤λk+1(Ω)≤ · · · .
Thereexiststandardvariationalharaterisationsofthespetrumof∆whihanbefound
forinstane inthe bookof P.Bérard [1℄ (orin[9℄).
The main resultofthis artile isthe following.
1.3.Theorem. Let(Mn, g) be a omplete n-dimensionalRiemannianmanifoldwithRii urvature bounded below Ricg≥ −(n−1)a2, a≥0.
Thereexist onstantsAn>0, Bn >0onlydepending onthedimension, suhthat forall k∈ N∗, V > 0 and for eah bounded domain Ω⊂ M, with smooth boundary and volume
V, we have
(1.2) λk(Ω)≤Ana2+Bn
k V
2/n
.
If the manifold M is ompat, an interesting speial ase isto hoose Ω =M, and we
reover Theorem 1.1 , up to the value of the onstant An whih is not equal to (n−1)2
4 in
ourpaper.
The proof Theorem 1.3 goes in the same spirit as the proof of Theorem 1.1: in order
to bound λk(Ω), we onsider k disjoint sets A1, ..., Ak in Ω of measure of the order of
V ol(Ω)
k ,and introdue testfuntions f1, ..., fk subordinated to these sets. Weestimate the Rayleighquotientofthese fontionsbyadiretalulation,whih givesthetheorem. The
mainimprovement ofthis paperistheonstrution of anadaptedfamilyofsets A1, .., Ak,
moreonvenient for our purpose asballs. Asthis onstrution isinteresting byitself and
willbeused inother ontexts, we present it inarather abstrat(indeed metri) way.
Thepaperisorganisedasfollows: themetrionstrutionofoursetsisdoneinSetion2,
and in Setion 3 we will use them so as to prove Theorem 1.3 by produing some test
funtionsfor the variational haraterisation ofthespetrum.
2. A metri approah
Inthissetion, weformalize thegeometri situationofTheorem1.3(aboundeddomain
inaompletemanifold)inamoregeneralsetting(aboundeddomaininaompletemetri
spae). More preisely, let (X, d) be a omplete, loally ompat metri spae, Y ⊂X a
bounded Borelian subset endowed with the indued distane, and µ a Borelian measure
withsupportinY suhthatµ(Y) =ω,0< ω <∞. Wewillneedinaddition thefollowing
tehnial assumptions:
(H1) Foreahr >0,thereexistsaonstantC(r)>0suhthateahballofradius4r in X maybe overedbyC(r)ballsofradius r. Moreover,r 7→C(r)isan inreasing
funtionof the radius.
(H2) We suppose that the volume of the rballs tends to 0 uniformly on X, namely
r→0limsup{µ(B(x, r)) : x ∈ X} = 0. However, taking (H1) into aount, this vol-
umeonditionisequivalent to lim
r→0sup{2C(r)µ(B(x, r)) :x∈X}= 0whihisthe
(more onvenient) onditionthatwill be usedintheremainder of theartile.
It is important to remark that these hypothesis are quite natural sine they make part
of the metri properties of the Riemannian manifolds that are involved in Theorem 1.3 .
Thesespeimetri propertiesareolleted inthefollowing fundamental example.
2.1.Example. Atypial exampleof aouple (X, Y) satisfyingthe hypothesis(H1),(H2) is
tohooseXasaomplete n-dimensionalRiemannianmanifold(M, g)withRiiurvature
bounded below Ricg ≥ −(n−1)a2, a ≥ 0 (whih are the lass of manifolds involved in
Theorem 1.3), and as Y a bounded domain with smooth boundary in M. The distane d
isthe distane assoiated tothe Riemannian metri g, the measure µ is the restrition to Y of the Riemannianmeasure of g. The existene of the onstant C(a, r) is given by the
lassial Bishop-Gromov inequality thanks to the lower bound on the Rii urvature of g
(see [14 ℄ p. 156). Preisely, for 0< r < R, andfor eah point p∈M, we have
(2.1)
Vol(B(p, R), g)
Vol(B(p, r), g) ≤ va(R) va(r) ,
where va(R) denotes the volume of a ball of radius R in Mna, the simply onneted n
dimensionalmanifoldof onstant setional urvature −a2.
Thisgivesaboundon thenumberofballsofradiusr thatareneessarytoover aballof
radius 4r (this property known as the paking lemma is a onsequene of Inequality (2.1)).
In fat, x B4r a 4rball and onsider {B(xi, r/2)}i∈I a maximal family of disjoint balls
whose enter xi live in B4r; then the orresponding family of rballs {B(xi, r)}i∈I over B4r. In onsequene, we an over a ball of radius 4r with ≤1 +h
va(4r+r/2) va(r/2)
irballs. We
justdene
C(a, r) = max
t≤r
1 +
va(4t+t/2) va(t/2)
.
The inreasingharater of r7→C(a, r) is by denition.
Furthermore, as r−→0, the ratio Vol(B(p,r),g)
va(r) −→1, we obtain Vol((B(p, R), g) ≤va(R) ,
andonsequently µ(B(p, r)) := Vol(B(p, r)∩Y, g) goes uniformly to 0 as r→0.
Weproveinthesequelthat,underourtehnialassumptions,oneanbuildsomesubsets
Aand Dsatisfying ertainvolume onditions.
2.2. Lemma. Let (X, d) be a omplete, loally ompat metri spae, Y ⊂X a bounded
Borelian withthe indued distane, andµ a Borelian measure withsupport inY suh that µ(Y) =ω, 0< ω <∞ and µ(Y \Y) = 0. In addition,we make the hypothesis (H1),(H2).
Let0< α≤ ω2. Thanksto(H2)there existsr >0withsup{2C(r)µ(B(x, r)) :x∈X} ≤α.
Thenthere exist A, D⊂Y suh that A⊂Dand
µ(A)≥α µ(D)≤2C(r)α d(A, Y ∩Dc)≥3r
.
Proof. We xthepositivenumbers r andα. Letus onsideranypositive integerm∈N∗
anddeneanonnegative appliationΨm:Xm =X×X× · · · ×X
| {z }
mtimes
−→Rbytherelation
Ψm:x = xjm
j=17−→µ
[m
j=1
B xj, r
,
whih is simplythe restrition of themeasure µ to Um(r) a partiular lass of open sets
whih isdened by
Um(r) :=
[m
j=1
B xj, r
/ xjm
j=1∈Xm
.
Sine(X, d) isaompleteandloallyompat metrispae,itisalsotheaseofthenite
produtXm when itis endowed withthe produtdistane. Then for eah m ∈N∗ there existssome xmax,m∈Xm (notneessaryunique)suh that
Ψm(xmax,m) = max
Xm Ψm= max
Um(r)µ=µ
[m
j=1
B xjmax,m, r
.
We rst prove that there exists a nite integer k ∈ N∗ suh that Ψk(xmax,k) ≥ α and Ψk−1(xmax,k−1) ≤α. Indeed, onsider the funtion ξ :N∗ −→ R dened by the relation
ξ(m) = Ψm(xmax,m). On one hand, the ondition sup{2C(r)µ(B(x, r)) : x ∈ X} ≤ α
obviously implies ξ(1)≤ 2C(r)α ≤α. Onthe otherhand, sine Suppµ ⊂Y, there existsa
radius R >0 large enough suh thatµ(B(z, R))≥3ω/4, for a ertainz ∈X. Butit an
belearlydeduedfromAssumption(H1)thatB(z, R)anbenitelyoveredbym0 ∈N∗
ballsofradius r (notie thatm0 depends onR). Consequently itturns out
3α 2 ≤ 3ω
4 ≤µ(B(z, R))≤ max
Um0(r)Ψm0 =ξ(m0).
Thereby the funtion ξ : N∗ −→ R satises ξ(1)≤ α and ξ(m0) ≥ 3α2 , whih entails the
existeneofsome k∈N∗ suhthatΨk(xmax,k)≥α and Ψk−1(xmax,k−1)≤α.
We now setUk := S
1≤j≤k
B
xjmax,k, r
and Vk := S
1≤j≤k
B
xjmax,k,4r
. Thenext stepis to
showthat
µ(Vk)≤C(r)µ(Uk) .
Still aording to Assumption (H1), Vk is overed by kC(r) balls of radius r, namely Vk⊂ S
1≤j≤kC(r)
Bj,where the Bj areballs ofradius r. Butit isquitelear that this union
ofrballsan bewritten as S
1≤j≤kC(r)
Bj = S
1≤j≤C(r)
Wj where eahWj ∈Uk(r). Itfollows
µ(Vk)≤µ
[
1≤j≤kC(r)
Bj
= µ
[
1≤j≤C(r)
Wj
≤
C(r)X
j=1
µ(Wj)
≤ C(r) max
Uk(r)µ=C(r)ξ(k) =C(r)µ(Uk) .
We nally dene the sets A := Y ∩Uk and D := Y ∩Vk. We only have to hek that
theysatisfythe propertiesstated inLemma 2.2 . We observe thatµ(A) =µ(Uk) sine the
measureµ issupportedin Y and µ(Y \Y) = 0. Besides, Uk an be written asthe union
ofan element ofUk−1(r)and anelement ofU1(r)sothat µ(A)≤ξ(k−1) +ξ(1)≤α
1 +1
2
.
Stillsine Suppµ=Y, we obtain µ(D) =µ(Vk)≤C(r)µ(Uk) =C(r)µ(A)≤2C(r)α. By
thedenition ofUk andVk,westraightforwardlyhave d(A, Y ∩Dc)≥3r.
In setion 3, we will use the following orollary of Lemma 2.2 to make the proof of
Theorem1.3 . Wegivethereinanexpliiteonstrutionofthedomainsthatwerementioned
attheend of the introdution.
2.3. Corollary. Let (X, d) be a omplete, loally ompat metri spae, Y ⊂X a bounded
Borelian withthe indued distane, andµ a Borelian measure withsupport inY suh that µ(Y) =ω,0< ω <∞ andµ(Y \Y) = 0. In addition, we make the hypothesis (H1),(H2)
asin Lemma 2.2,and take N a positive integer.
Letr >0 suh that 4C2(r)µ(B(x, r))≤ Nω holdsfor all x∈X, andlet α= 2C(r)Nω . Then,
there exist N measurable subsetsA1, ..., AN ⊂Y suh that µ(Ai)≥α and,for eah i6=j, d(Ai, Aj)≥3r.
Proof. Weonstrut thefamily (Aj)Nj=1 byniteindution applyingLemma 2.2.
•j= 1. We set (X1, d1, µ1) = (X, d, µ) and Y1 = Y, whih satisfy the assumptions of Lemma 2.2. Therefore there existA1, D1 suh thatA1 ⊂D1 ⊂Y1 =Y and
µ(A1) ≥ α
µ(D1) ≤ 2C(r)α= Nω d(A1, Y1∩Dc1) ≥ 3r
.
•j= 2. Weset(X2, d2, µ2) = (X, d, µ|Y2)andY2=Dc1∩Y1,whihsatisfytheassumptions of Lemma 2.2 withω2 =µ2(Y2) ≥ω 1−N1
=ω N+1−2N
≥α. Therefore there
existA2, D2 suh thatA2 ⊂D2⊂Y2 =D1c∩Y1 and
µ(A2) ≥ α
µ(D2) ≤ 2C(r)α= Nω d(A2, Y2∩Dc2) ≥ 3r
.
AsA1 ⊂D1 andA2⊂Y1∩D1c we getd(A1, A2)≥d(A1, Y1∩D1c)≥3r thanksto
the asej= 1.
•j≥3. We suppose that we have already onstruted the families (As)j−1s=1 and (Ds)j−1s=1
thatsatisfytheonditions
As⊂Ds⊂Y ∩(D1∪ · · · ∪Ds−1)c =Ys, s≤j−1 d(As, At)≥3r s6=t,
µ(D1∪ · · · ∪Dj−1)≤ω
j−1 N
.
We set (Xj, dj, µj) = (X, d, µ|Yj) and Yj =Y ∩(D1∪ · · · ∪Dj−1)c, whih satisfy
theassumptionsofLemma2.2withωj =µj(Yj)≥ω
1− j−1N
=ω
N+1−j N
≥α
ifj≤N. Therefore thereexist Aj, Dj suh thatAj ⊂Dj ⊂Yj and
µ(Aj) ≥ α
µ(Dj) ≤ 2C(r)α= Nω d(Aj, Yj ∩Djc) ≥ 3r
.
As Aj ⊂ Y ∩(D1∪ · · · ∪Dj−1)c ⊂ Y ∩(D1∪ · · · ∪Ds−1)c = Ys, s < j, and As ⊂ Ds, we get d(Aj, As) ≥d(As, Ys∩Dsc) ≥ 3r thanksto the ase j = s. As
already said,we an proeed this onstrution solonger we have enough volume
to doit, thatis N times.
3. Proof of Theorem 1.3.
Let (Mn, g) be a omplete n-dimensional Riemannian manifold with Rii urvature bounded below Ricg ≥ −(n−1)a2, and Ω ⊂ M a bounded domain of volume V, with
smooth boundary.
We observe rst that, by renormalisation, it is enough to prove the theorem for the
ase a = 1: namely, if Theorem 1.3 is true for a= 1, and if g is a Riemannian metri
with Ricg ≥ −(n−1)t2g, then g0 = t2g satises Ricg0 ≥ −(n−1)g0. Sine we have λk(g0)≤An+Bn
k V(g0)
2/n
,then, beause λk(g) =t2λk(g0) andV(g) =tnV(g0),we get λk(g)≤Ant2+Bn Vk2/n
.
So,letuseprove Theorem1.3for a= 1. AsinExample2.1, letusonsidertheBorelian
measureµwhih istherestrition to thedomain Ωof theRiemannian volume of (M, g).
In order to prove Theorem 1.3, we will usethe lassial variational haraterization of
the spetrum: to estimate λk from above, it sues to onstrut an H1(Ω)-orthogonal family of k test funtions (fj)kj=1, suh as eah fj has ontroled Rayleigh quotient. In
thesequel,we onstrut test funtionswithdisjoint supportrelated tothe sets A1, ..., Ak
arisingfromCorollary 2.3 , sothatit immediatelyimpliesorthogonality inH1(Ω).
3.1. Lemma. Let A⊂M a subset as in Corollary 2.3. Let Ar:= {x∈M :d(x, A)≤r}, r > 0. There exists a funtion f supported in Ar whose restrition to Ω is of Rayleigh
quotient
R(f)≤ 1 r2
µ(Ar\A) µ(A) .
Proof. Letus dene a plateau funtion
f(p) =
1 if p∈A 1−d(p,A)r if p∈(Ar\A)
0 if p∈(Ar)c .
InCorollary2.3, thedomainA isanite unionofmetri ballsandintersetionwithom- plement ofballs. The boundary is not smooth,but thefuntion d(∂A,·) "distane to the boundary of A" iswell knownto be1Lipshitz on M. Aording to Rademaher's theo- rem(seeSetion 3.1.2, page8184in[8 ℄), d(∂A,·) isdierentiable Ln almosteverywhere (sinedVolg is absolutelyontinuouswith respetto Lebesgue's measureLn),and its g
gradient satises|∇d(∂A,·)|g≤1,Ln almosteverywhere. Itomes out thatthegradient off satisesLnalmost everywhere
|∇f(p)|g ≤ 1
r if p∈(Ar\A) 0 if p∈(Ar\A)c .
Weimmediately dedue
R(f) = R
Ω|∇f|2gdVolg R
Ωf2dVolg ≤ 1 r2
µ(Ar\A) µ(A) .