Photocopying permitted by licenseonly the Gordon and Breach Science Publishersimprint.
Printedin India.
Sobolev Inequalities in 2-D Hyperbolic
Space A Borderline Case
FRANCESCO MUGELLIandGIORGIO TALENTI*
Dipartimento di Matematicadell’Universit&,VialeMorgagni 67A, 1-50134Firenze,Italy
(Received30 March1997)
Sobolev inequalitiesin two-dimensionalhyperbolicspaceI[-][ aredealtwith.Here[HI is modeledonthe upperEuclideanhalf-plane equippedwiththePoincar6-Bergmanmetric.
Some borderline inequalities, where the leading exponent equals the dimension, are focused. Thetechniqueinvolves rearrangements of functions, and tools from calculus of variationsand ordinarydifferentialequations.
Keywords: Sobolev inequalities; Hyperbolic space;Calculus of variations 1991 Mathematics SubjectClassification." 26D10, 45J95,58D106
1.
INTRODUCTION
1.1. In apaper where fluid mechanicsblendswithreal andfunctional analysis, 7],Fraenkel suppliedaproof thatif2
<
q<
oosome constant Aexistssuch that(1.1)
for everytest function
.
Throughoutwelet+
2 be{(x,y)"
-oo<
x<
oo,0<
y< o},
Correspondingauthor. E-mail:talenti@udini.math.unifi.it.
195
the upperEuclidean half-plane, and let testqualify any smooth real- valuedfunction defined in
R2+
whosevalue at(x,y)approacheszerofastenough as eithery approaches 0 or
x2+y2
approaches infinity.(For
technicalreasons, our notations and terminology differ slightly from those adoptedin[7].)
Though chiefly concerned with an existence theory for a partial differential equation, Fraenkel had an eye to the sharp form ofhis inequality, and detectedindeedsuchaforminthecasewhereq 2 and q
10/3.
If q--2, the smallestconstantA,whichrenders(1.1)
truefor every testfunctiong),is exactly 1. Ifq-10/3,letA
26/5
15-1/2.7r-1/5; [= 0.471802666130...];
then
(1.1)
holdsif isanytestfunction, and becomesanequalityif is specifiedby79(x y) y2(1 +
x2q_y2)-3/2.
The formerstatement is astraightforward consequence of thefamiliar Hardy’s inequality, the latter rests upon the following change of variable
bl(Xl,X2,
X3,X4,X5) --(X + X + x] + X52)
-1l,
+
3+ ] +
anda Sobolev inequalityinthe Euclideanspace ofdimension5.
Twomoveshelptodisentangle thematter.
One,
whichdidappearin[7]
and turns(1.1)
into3
jc
u2dxdy(1.2)
ismaking the changeofvariable
(x, y) v u(x, y).
Another is realizing that
(1.2)
falls under Sobolev inequalities in the hyperbolic(or
PoincarO)half-plane.Thepresent paperisthesecond ofaseries, devotedpreciselytothese inequalities.Itcontinues 16],which werefertofor preparatory results andabibliography, and focusessomeborderlineSobolev inequalities instances, such as
(1.2),
where the leading exponent equals the dimension.A
motivehereis topointoutthatcertain lineamentsof the hyperbolic half-plane theRiemannianlengthand area, the geodesic polarcoor- dinates, the isoperimetric theorem and thetheoryof rearrangements outlined in1]
and 16]are akeytoFraenkel’s inequality.A
featurewill emergeindeed: if a point(a, b)
is fixed inR2+
ad libitum, then the test functionsthatreallycount in(1.1)
those renderingdxdy
I]qy
-q/2-2dxdyaminimum have thisspecial form
y)= 7.
Here s Riemannian area ofageodesic disk whose radius equals the Riemanniandistancebetween
(a, b)
and(x,
y); vis asmooth real-valued functionthatis defined in [0, [, decays fast enoughat and makess(s + 4re) (v’)
ds+
v2ds}x { Zlvlq ds}
-2/qaminimum. Inconclusion, Fraenkel’s inequalityamountsde
facto
to avariational problem for functions ofa single variable, which can be treated by simple tools ofthecalculus ofvariations and the theory of ordinary differential equations.
1.2. Let
]2
be2+
equippedwithy-e [(dx)
e+ (dy)2], (1.3)
the PoincarO-Bergman metric.
]2
is a Riemannian manifold that models the two-dimensional hyperbolic space and has the followingproperties see[2, Chapter 14], [13,Section 15], [17,Section
9.5]
and [18,Section2.2], for example.The Riemannian length ofa tangent vector to lI-lI2 at a point (x,y) equals y x
(the
Euclideanlength). The geodesics of lI-lI2arethehalf-lines andthehalf-circles orthogonalto thex-axis.The Riemannian distance betweentwopoints(x,
y)and(x2, Y2)
isthelengthof the geodesicarc joining(Xl,Yl)
and(x> Y2),
and obeys[(X1 x2)
2+
yl2+ y]. (1.4)
cosh(distance)
2yy2
The RiemannianareaonEI
2,
3A, isgiven bydad dxdy
y2 (1 5)
The geodesiccircleinlH[
2,
center(a,
b)andradius r, has equation(x- a)
2+ (y b)
2[2 sinh(r/2)]
2by,(1.6a)
hence coincides with the Euclidean circle whose center is
(a,
bcoshr)
and whoseradius is bsinhr. The Riemannian radius, areaand perim- eter ofageodesic disk inlI-lI2 obeyradius-
log[1 + (27r)-1 (area + perimeter)],
area
r[2 sinh(radius/2)] 2,
perimeter
[(area)
2+ 47r(area)]
1/2 2rsinh(radius).
The Laplace-Beltrami operatoronN
2,
/,is givenby(1.6b) (1.6c) (1.6d)
02 02) (1.7)
The curvature of ]I-lI2 is -1. The following statements are closely related tothelast mentioned oneand germaneto thepresentcontext the former appears in [14]; the latter, known as the isoperimetric theorem in H
2,
appears in [4, Section 10] and [5, Chapter 6], for example.Thespectrum of -A
"L2(]-2)
-+L2(]’ 2)
is exactly[1/4, [.IfEis any sufficiently smooth subset ofH
2,
thenthe Riemannian perimeterandthe RiemannianareaofE,obey the following inequality.If thearea ofEisfinite, then
perimeter
_> v/area(area + 47r); (1.S)
moreover, equality holdsin
(1.8)
ifand onlyifEis a disk.1.3. Thearchitecture ofa Sobolev inequalityon a Riemannianmani- fold is affected by the underlying curvature.
A
typical Sobolev inequality in hyperbolic space]2
claims that if p, q and R lie in a suitablerange p>_
1, q>
p, 1/q>_
1/p-1/2
and R_< p-2
then some positiveconstant existssuch thatyP
u2-+- Uy)
2+
x y>(Cnstant){lulqdxdy} y2
2/q+
R{J + lul
pdxdy
}
2/py2 (1.9)
for everytestfunction u.
Observethat
2+ lulp
dx
y2
dy
thenorm
of
u in Lebesgue spaceLP(]I-2),
and similarly{f 2+ Ibllqdxdy} y2
1/qIfH lulq
djI
1/qthenorm
of
u inLq(]2).
Onthe otherhand, 2"p/2dx dyyP
u2+
Uy)
22+
x ythenorm
of u
in(LP(]-2)) 2.
Proof
If u is a smooth scalar field on ]HI2 thenVu,
the covariant derivativeof u,isthe tangentvectorfield toH2whose componentsareand
uy,
and whose Riemannianlength,IVu],
equalsyCu
x2+
ub/x
Inconclusion, inequality
(1.9)
readsIlVull
2(Lp(]i.2))2> (Constant)llull
eLq(]HI2)+Nil
Let
C(p,
q,R)
the largest constant(1.10)
such that inequality
(1.9)
holds for every test function u.A
theorem from[16]
which was derived there as a consequence of the isoperimetric theorem in H2,
and implies the latter results in the following equations:C(p,q,R) _p-2 (1.1 la)
if
<_
p<
oe,q-p and R-O;C(p,
q,R)
47r(1.11b)
if p- 1,q-2 and -oe<R_< 1;
C(p,
q,R) (--1)e/q [sin ()1
if
<
p<
2, q 2p/(2-p)and -oc<
R_< p-2;
(1.11d)
if2
<
p<
oc, q ocand R 0.1.4. In view of observations made in subsection 1.1, the smallest constant Awhichmakes Fraenkel’s inequality
(1.1)
truefor everytest function isgiven byA-2
C(p,
q,3/4),
herep-2and2
<
q<
oo, a case notcovered by equations(1.11).
In the present paper we look precisely into such a case, and investigate
C(2,
q,R). By
definition,C(2,
q,R)
inff=+ (u2x + u2)
dxdy Rf 2+ u2y
-2dxdy{fR2+lulqy-2dxdy}
2/q(1.12)
underthe conditions: uis a testfunction,u
:
0.Alternatively, observe that
(U2x + u2y)
dxdy/(Uxx + Uyy)U
dxdyif u is sufficiently well behaved the left-hand side is the standard Dirichletintegral, theright-handside isthescalarproduct of
(-Au)
and uinL2(]I2).
Deduce thatC(2,q,R)
coincides withthelargestconstant such thatUI[Lq(IHI2
(( m i)b/,/,/)L2(IHI2) (Constant).
2(1.13)
for everytest function u.
Geodesic polar coordinates in ]I-]I2 are introduced in the proof of Lemma 3.2 together with an instrumental variant.
Arguments
of dimensionalanalysis based upon these coordinates,spectralproperties ofthe Laplace-Beltrami operatorinH2,
formula(1.12)
andinequality(1.13)
show thatC(2,
q,R) {
-oo0 ifif qR< >
21/4,
and R_< 1/4. (1.14)
2.
MAIN RESULTS
THEOREM 2.1 Assume 2
<
q<
oeand-oc<
R< 1/4.
Atest
function
u existssuch that u 0andC(2,
q,R) re2+ (u2x + u2)
dxdy Rf + u2y
-2dxdy{f2+lulqy_2dxdy}2/q
both
C(2,
q,R)
anduareprovidedby the followingrecipe.Asmooth real-valued
function defined
in [0,+
oc[,v, existssuch that:(i)v
satisfies
the followingdifferential
equationdds
(sfs + 4rc)v’(s)) + Rvfs) + Iv(s)l
q-2.v(s)
0for
O<
s<
cx.(ii)v
satisfies
thefollowingboundaryconditions-4rrv’(O) Rv(O) + Iv(O)l
q-2.v(O), v(oo)
-O,anddecaysat infinityinsuchaway that
(sv’(s))ds
<
c.(2.1)
(2.2)
(2.3)
(2.4)
(iii)v isstrictly decreasing.Let (a,
b)
beany pointin IHI2.
The followingequationshold:
(2.5)
{ ooC }
l-2/qC(2,
q,R) Iv(s)l
qds(2.6)
and
(71"
U(X,y)
V-y ((x-- a) 2q-(y
for
every(x, y)from]I-]I2 in otherterms,u(x, y)
(2.7)
(2.8a)
s
rr[2 sinh(r/2)]
2Riemannian area ofageodesic diskofradius r, r Riemannian distance between
(a, b)
and(x, y).
(2.8b) (2.8c)
Remarks (i) Equations
(2.7)
and(2.8)
inform that u is radial and radially decreasing the value of u at any point(x,
y) depends only upon theRiemannian distancebetween(a,
b)and(x,
y),and decreases monotonicallyassuchadistance increases.An
appropriateuseof geodesicpolarcoordinatesgives d[s(s + 4rr)v’
Xu(x, y) T, (2.9)
asdetailed intheproofofLemma3.2. Plugging
(2.9)
in(2.2)
resultsinAu -+-
Ru-+- lul
q-2 uo, (2.10)
inthe language of the calculus of variations,
(2.10)
isexactly the Euler equation implied by(1.12), (2.1)
andasuitable normalization.(ii) The curvature of
]}.][2
is a reason why the leading coefficient inEq. (2.2)
is apolynomial ofdegreetwo with two distinct roots.In fact, the genesis of such equation reveals that thecoefficient inquestion 4rrs-(curvature)
s2.
Thoughharderthan equationslike
(4rrs v’)’ + (lower
orderterms)
0,which appear in [10, Sections 6.73 to
6.76]
and when borderline Sobolev inequalitiesinthe Euclidean planeare consideredEq. (2.2)
has certain particular solutions available inclosedform.
Thefunctiondefined by
s
)
-1/(q-2)v(s) (q 2)
-2/(q-2)+ (2.11)
happens to satisfy
(2.2)
if q>
2 and R-(q-3)(q-2)-2.
The functiondefined by
2q
]
1/(q-2)v(s)-
(q-2)
2(1 __)-2/(q-2) (2.12)
satisfies
(2.2)
if q>
2and R 2(q-4)(q-2)-2.
(iii) Some radial solutions to
Eq. (2.10)
result from the previous remarks. Thefunction definedbyu(x,y)
(q_ 2) (x- a)
2+ (y + b)
2(2.13)
satisfies
(2.10)
in the case where q>2 andR=(q-3)(q-2)-2;
the function definedbyu(x, y)
(q 2)
2(x a)
2q-y2
q_b2 (2.14)
satisfies
(2.10)
inthe casewhereq>
2 andR 2(q-4)(q-2)-2.
Constantfactors apart,someofthesesolutionsappearinTheorem 2.2, and one of them appearedin subsection 1.1.
(iv) The function defined by
(2.11)
satisfies boundary conditions(2.3).
Property(2.4)
holds if in additionq_<
4, whereas(2.5)
holds plainly. Thefunction definedby(2.12)
does thesame if2<
q_<
6.Coupling
(2.6), (2.1 !)
and(2.12)
would giveC(2,
q,R) (2r)1-2/q (q 2)-1-2/q (2.15)
inthecasewhere
( )_1
-oo<R<_l/4,
q-2+2,l+x/1-4R
and
C(2,
q,R)(27r)’-2/q(q 2)
-’-2/q2q(q + 2)
-’+2/q(2.16)
inthe casewhere
-oc
<
R<_ 1/4,
q--2+4.(1 + x/1 -4R)
-1Onemight wonder whether formulas
(2.15)
and(2.16)
are correct.Fraenkel’s result tells us that
(2.16)
holds if q=10/3
andR=-3/4.
Theorem2.2showsthat
(2.15)
and(2.16)
identifyC(2,q,R)
properlyin thecasewhereRis zero.(v)
Ifq islarger than2(but
isnot toolarge)andRisbelow 1/4,thenC(2,
q,R)
canbeaccuratelycomputedbyanalgorithm. Such algorithm consists in:(i)solvingEq.(2.2)
subjecttoconditions(2.3)
and(2.4)
viaan appropriate shooting method; (ii) selecting a solution satisfying condition
(2.5);
(iii) checking the uniqueness of this solution via an appropriate test; (iv)using formula(2.6).
Details can befoundin[15],sample valuesaredisplayedin thetable below.
q
C(2,
q,0)
qC(2,
q,0)
3.0 1.8452701 8.0 3.6479107
10/3
2.2655626 8.5 3.60965113.5 2.4484193 9.0 3.5635303
4.0 2.8944050 9.5 3.5119626
4.5 3.2085021 10.0 3.4567447
5.0 3.4207189 11.0 3.3403841
5.5 3.5569171 12.0 3.2216028
6.0 3.6372463 13.0 3.1043799
6.5 3.6767932 14.0 2.9908876
7.0 3.6866772 15.0 2.8823095
7.5 3.6750545
(2.17)
THEOe,EM 2.2 (i)
C(2,3,0)=(27r) 1/3,
i.e. the following inequality(u2+ 2) uy dxdy>(27r) +l u[3dxdyy2 (2.18)
holds
for
every testfunction
u and is sharp. Equality takes place in(2.18) tf
u(x,y)
(ii) C(2,4,
0) (8rc/3) 1/2,
i.e. the following inequalityu
)+u
dxdy_>--
2+
xy2 (2.19)
holds
for
every testfunction
u and & sharp. Equality takes place in(2.19)/f
u(x,y)
X 2_+_
y2
q_3.
KEY LEMMAS
The lemmas fromthissection, culminatinginLemma3.5 below,show that theSobolev inequalitiesinhandamount to a variationalproblem in dimension one.
LEMMA3.1
If
u isanytestfunction, thenareal-valuedfunction defined
in
[0,
oo[,u*,
existssuch that:(i)
u*(c)
0;(ii) u* isdecreasing;
(iii) u* isequidistributedwith u;
(iv) u* isabsolutely continuous, and thefollowing inequality holds
(u
2+ u 2)
dxdy> s(s + 4rr) [--s (s)
2+
xds.
Proof
Let# be thedecreasing right-continuous map from[0,x[
into [0, ]defined by the following formula:/(t)-
Riemannian areaor {(x,y)
and letu*the map from [0,
oe[
into[0, cxz] definedbyu*(s) min{t _>
0"#(t) < s}. (3.2)
(These definitions mimic those introduced by Hardy and Littlewood andelaboratedbyseveral authors:#isthedistributionfunction, u*isthe decreasingrearrangement ofu see[9,Chapter 10],[11],
12], 1], [20], and the referencesquotedtherein.)Clearly, formulas
(3.1)
and(3.2)
imply properties (i)and (ii).The same formulas and a brief reflection yield
{s _>
0:u*(s)> t}
[0,#(t)[ for every nonnegative t. We deduce that the following equation:
length of
{s _>
O"u*(s) > t}
Riemannian area of
{ (x, y)
E lI-lI2lu(x,y)l > t} (3.3)
holds for every nonnegative t.
Property
(iii)is demonstrated.A
version ofa theorem,which iscentral to the presentcontext and is offeredin[1, Sections 3and4]
and [16, Section2],
implies property (iv).LEMMA3.2 Suppose v is a real-valued
function defined
in [0,suppose v is smooth and decays
fast
enough at infinity. Let (a,b) be anypointin]I’]I2,
and letubedefined
byrr
((x- a)
2u(x, y)
v-y -+- (y b)2)). (3.4)
Thenu is a test
function
and the followingpropertieshold:(i)uandv are equidistributed;
2)dx
dyf s(s + 4rr)(v’(s))
2ds.(ii)
f2+ (u
2x+
uyProof
Thereis a notational convenience and no loss of generalityin assuming that a-0and b- 1.The following equationssinhr.sin 0
x coshr sinhr.cos0 Y
coshr sinhr.cost9
(3.5a)
the following others
cos0 X
2nt_
y2
(x 2+(y-1) 2.v/x 2+(y+1)
2sin0 2x
(3.5b)
V/x2 + (y-1)2 V/X2 + (y + l)2
2, 2
coshr-y(l+x +y ),
andthefollowingconstraints
O_<r<oo, -rr<v
<rc (3.5c)
defineasystem of geodesicpolarcoordinates inlIqI2.Hererstands for the Riemanniandistancebetween
(0, 1)
and(x, y);alinewhere0 Constant is ageodesicarcwhose origin is(0, 1)
and whose anglewith they-axis is0.Define
s
rr[2 sinh(r/2)] 2, (3.6a)
theRiemannian area of ageodesicdisk of radiusr. Inother terms, 2rr-coshr 2rr/s,
2rr.sinhr
v/s(s + 4rr),
71"
(X
2 2s=-
+(y-l) ).
Y
(3.6b)
The systemof curvilinear coordinatesmade upbysand 0isespecially convenient inthe presentcontext suchasystem makesadimensional analysis of Sobolev inequalities possible, by the way.
The Poincar6-Bergmanmetricobeys
[(dx)
2-+-(dy) 2] (dr)
2q-(sinhr)2(dO)
2(ds)
2s(s + 47r) (d9)2
s(s + 4rr) +
47r2
(3.7)
Formula
(3.7)
and customaryrules ofdifferential geometry tell us thatd(Riemannian area) (sinh r)
dr dO dsdO,(3.8)
and that any sufficiently smoothfunction
f
obeysIX7fl e- /(sinh r)
-2(0f)2 47r2 (_)2
s(s + 47r) ss + s(s + 47r (3.9)
Incidentally, observe thefollowing formula
Af--(sinhr)- rr
0( Of)
sinhrrr + (sinh r)
-269002f
2O
Of]
47r202f
0-- s(s + 47r)ss + s(s + 47r
002.(3.10)
Owingto
Eq. (3.4),
onemay check thatuis smooth inparticular,IVu(x,y)l O(r)
as r O.Moreover, u(x, y) approacheszero fastas rapproachesinfinity.
Equations
(3.4)
and(3.8)
imply thatRiemannian area of
{(x,y)
ElH[ 2"lu(x,y)l > t}
length of
{s _>
0"Iv(s)l > t}
for every nonnegative property(i)isdemonstrated.
Equations
(3.4), (3.8)
and(3.9)
givethe Dirichletintegral ofu
s(s
--t-47r)(v’(s))
2ds,property (ii) is demonstrated.
A
couple ofdefinitions are involved in Lemma 3.3 and Sections 4 and 5.DEFINITION3.3 Jis thefunctional definedby
J(v) fs(s + 47r)(v’(s))
ds RfCv(s))
ds{ fOlv(s)lq ds}
2/q(3.11)
DEFINITION 3.4 3 denotes the collection of those real-valued func- tionsvdefined in]0, oe[such that:
(i)
v()
0;(ii)visabsolutely continuous, and both
fo s(v’ (s))2ds
andf (sv’ (s))2ds
converge.
LEMMA 3.5 Suppose0
<
q<
cxand-cx<
R<
x. Then the following holds:(i)
If
isanyrealnumber, theneither{v:
0 ve , J(v) <_ } 0
OF
{v:
O ve , J(v) < l}
fq{v:
v,
vdecreases} -
{3.(ii)
inf{J(v)
0 v E3} C(2,
q,R).
Proof
CombineLemmas 3.1 and3.2.(3.12)
4.
ESTIMATING C(2,
q,0)
THEOREM 4.1 Thefollowing inequality
C(2,
q,0) > f
{( (q
87rq
2)
2(r(q/(q r(2q/(q 2))) 2)) 2}-2/q (4.1)
holds
Proof
Let2<
q<
and R 0. Sinceq
7r
)1-2/qSl+2/q
S(S
-t-471) >_
41-3/q"q"2 for every positive s,wehave
)
1-2/qfoe
sJ(v) >_
41-3/q qq 2
+2/q(v’(s))2
ds(cx I-2/q
)<
I(S)I
qds(4.2)
if0 vE3. Applying Lemma4.2wededuce
J(v) > { (q_2)
87rq 2(I’(q/(q-2)))211-2/q I’(2q/(q-2)) (4.3)
if0 vE
.
Applying(ii), Lemma3.5, concludes theproof.LEMMA4.2 Let2
<
q<
c; letvbeareal-valued absolutelycontinuousfunction defined
in]0,c[
suchthatv(c)=
O. Then/o sl+2/q(lt(s))
2ds>
2q-:Vq{ (I’(q/(q 2))
2}
1-2/q(q- 2)I’(2q/(q- 2)
Iv(s)l
qds(4.4)
and equality holds
if
v isspecifiedbyv(s) (1 + s1-2/q)
-2/(q-2).Proof
Formula(4.4)
is a variant ofaninequalityby Bliss[3].
5.
FURTHER
LEMMASThe lemmas from this section prepare a proof of Theorem 2.1, and describepropertiesof functionspace 3and functional J. Lemma 5.2, coupled with a standard theorem of functional analysis, enables to
assert thatif2
<
q<
ooany bounded subset of isrelativelycompact in Lq(O,oo).
Lemma 5.5 enables to show that if-oc<
R< 1/4
therestriction ofJtotheunitsphere ofLq(O,
oo)
islowersemicontinuous.LEMMA 5.1
If
vbelongsto23, then( ) (v’()> d+ -l (v’(t>> at
(v(s)) 2<log 1+
for
every positivesand every positive c. The followingasymptoticshold:v(s)- o(s -1/2)
ass--, as s 0.(5.2)
Proof
Since visabsolutely continuousandv(oo)-
0, wehavev(s) (-v’(t))dt
for every positives thus
V(S) (ct-Jr- t2)
-1/2"(Et-JI- t2)l/2(--v’(t))dt
ifboth sande are positive. Hence Schwarz inequality gives
et
+
2 x et(v’(t))
2dt
+ (tv’(t))
2dtifbothsand carepositive. Inequality
(5.1)
follows.Properties
(5.2)
are anobviousconsequence of(5.1).
LEMMA 5.2
If2 <
q<
oo, thenapositiveconstantCexistssuch thatIv()l ds
{/o L
<
C.(e + h)
’-q/2. es(v’(s))
2ds+ (sv’(s))
2ds(5.3)
and
oolV(S
+ h) v(s)l
qdsh
s(v’(s))
2ds+
s(sv’(s))
2ds_< C.l+s--- . (5.4)
for
every vfrom ,
and everynonnegativeh ands.Proof A
formula from [8, Section4.271]
givesf0 [log(1 q-O]kds--F(k+ 1)(k)
ifk>1.
Formula
(5.5)
impliesIlo (l
ds<_ r(k + 1)(k) (1 + h)
ifh
>_
0 andk>
1, since(1 + h)k-1 fhC Ilog (1 + 01
dsdecreases monotonically as h increases from 0 to c and k
>
1.Lemma 5.1 yields
]v(s) lq
ds_<
log+
ds{f0 f0 /
q/2x
s(v’(s))
2ds+
e-1(sv’(s))
2dsifhisnonnegative and is positive. Therefore
lV(s)lqds
{/o
r(1 + q/2)((q/2) (Ft(S))2
-1(1
q-h/E)
q/2-1s ds
+ (Sl,,t ())2
ds}q/2
ifhisnonnegative andeispositive. Inequality
(5.3)
isdemonstrated.Since
v(s+h) v(s) --Js f+h
v(t)
dt,wehaves+h
v(s
/h) v(s) (t
2/t/e)
-1/2(t
2/t/e)I/2v’(t)
dtdS
if s,h and e arepositive. As
s+h dt
+ h/s
[ h]
t2 + tl
e e logl+eh/(es+l)
-<elgl+(l-eh)s
Schwarz inequality gives
Iv( + h) v( )l
<
log+ (1 q2eh)s t(v’(t))
2
at +
e(tv’(t))
2at
if s,h ande arepositive. Owingto
(5.5),
wededuceolv(s
-+- h) v(s)l
qdsh
lP(1 + q) (q) s(v’(s))
2(sv (S))
2 q/2-< +e---- (
ds+e dsInequality
(5.4)
is demonstrated.LEMMA
5.3Suppose
v is a real-valuedfunction defined
in ]0,suppose v isabsolutelycontinuousand
v(oc)=
O. Then(sv’(s))
2ds>_
- (v(s)) ds. (5.6)
Proof
Passtothelimit ininequality(4.4)
as q 2,or see[9,
Theorem328].
Alternatively, usethe following identityV2
/
(V
/2SVt)
24(SV’)
2/2(SV2)t,
orPlancherel’s theorem forMellintransforms.
Thefollowingdefinition is involved inLemma5.5andsubsection 6.1.
DEFINITION 5.4 Kisthe functional definedby
f0 f0
K(v) s(s + 47r) (v’(s))
2ds R(v(s))
2ds.LEMMA 5.5
If
--oe<
R< 1/4
andvE,
thenfo {
d(s(s + 4r)P’ RP} ds’
v/K(v)
sup vss
provided thetrial
functions
obey:pissmooth and behaves wellnear0and cxK( )
1.Proof
LetQ
be thebilinearsymmetric formdefined byfo fo
Q(v, p) s(s + 47r)v’ (s)p’ (s)
ds Rv(s)g)(s)
ds.Lemma 5.3 tells usthat Q(v,
v) >_
0 for every v from,
ifR< 1/4.
We deducev/Q(v, v) max{Q(v, p).
E, Q(p, ) 1}
for every v from
,
ifR< 1/4. An
integrationbyparts shows thatfo {
d(s(s + 47r)P RP }
dsQ
v,p v-ss
providedv isin
,
issmoothenoughand behaves wellnear0 and andK()=
1.The conclusionfollows.
6.
PROOF
OFTHEOREM
2.16.1.
A
member of,
v,exists such thatIv(s)[
qds(6.1)
and
J(v)
inf{J(g)): 0 E3}. (6.2)
Proof
We havetoshow thattherestrictionofJto the unitsphere of Lq(O,c)
attains aleast value.Twotypical ingredients of the calculusof variations are involved here: compactnessandsemicontinuity.Lemmas 5.2 and 5.3, and a standard theorem
(see
e.g. [6, TheoremIV.8.20])
tell us that the set defined by the following con- ditionsv
, Iv(s)[
qds- 1,K(v) <_
constantis relatively compact with respect to the topology of Lq(O,
cx3)
Lemma 5.3 ensures boundedness with respect to a topology of,
Lemma5.2 ensures sufficient conditionsfor compactnesswithrespect tothe topologyofLq(O,
oc).
Lemma5.5implies thatKislowersemicontinuous withrespecttothe topology of
Lq(o, oo).
Wededuce that the setmentionedabove iscompactwithrespectto the topology of
Lq(O,
o),and therestrictionofKto suchasetattains a leastvalue.SinceJispositivelyhomogeneousofdegreezero and
J(v) K(v) Iv( )l
qdsthe assertion isdemonstrated.
6.2. Function vsatisfies the followingdifferential equation d
(s(s + 4r)v’(s)) + Rv(s) + J(v) Iv(s)[
q-2v(s)
0ds
for 0
<
s<
oc, and thefollowingboundaryconditions(s + 4r)v’(s)
-+ 0 as s O.(6.3)
(6.4)
Proof
The Gateaux differential ofJ,J,
is themap from into the appropriate dual of whose value at vobeysJ’(v)()
-lim,-0
7 [J( + t) J(v)]
for every from
.
Letv belongto and satisfy(6.1). An
inspection shows1J’(v)() s(s/47r)v’’ds-R vds J(v) Ivlq-2vds
2
forevery from
.
The asymptoticbehaviorofv(s)
as s 0 orsis displayedin
(5.2),
Lemma 5.1. Hence anintegrationby parts yields1j’.v...(
)(s(s + 47r)v’(s)
2
/R
v(t)
dt/J(v) Iv(t)[q-2v(t)
dt9’(s)ds
for every from
.
Property (6.2)
impliesJ’(v)--O.
Wededucef0 f0
s(s + 47r)v’(s) +
Rv(t)
dt+ J(v) [v(t)lq-2v(t)
dt 0,(6.5)
correctingvon asetof measure zeroguarantees that
Eq. (6.5)
holds for everypositives.Equation
(6.5)
implies both(6.3)
and(6.4)
and vice versa.6.3. Function vsatisfies the following boundarycondition
-47rv’(0) Rv(O) + J(v). [v(0)l
q-2.v(0). (6.6) Proof
Equation(6.5)
gives/0 /0 (
47r
Iv’(s)l
ds_<
log+
RIv(s)
/J(v). Iv(s)l
q-}
ds.This inequalityincludesthe followinginformation:
Iv’(s)l
ds<
oo,because of formula
(5.5)
and Lemma5.1.Wededuce thatviscontinuous upto0. Dividing both sidesof
(6.5)
bys,then lettings 0 gives(6.6).
6.4. Functionv satisfies
v()
0 and(sv’(s))
2ds< (6.7)
trivially theseconditions are included in the membershipto
.
6.5. Statement (i) from Lemma 3.5 enables us to convert v into a function that simultaneouslyminimizes J andis decreasing.
In
other words,we can assumethatvobeysall the propertieslistedaboveand,in addition, isdecreasing.Observe that v decreases strictly and is smooth. Indeed, any de- creasing solution of
Eq. (6.3)
is either constant or strictly decreasing;anypositivesolution of
(6.3)
is infinitelydifferentiable.6.6. Replacevby
[J(v)]l/(q-2)
v,(6.8)
inotherwords,renormalize vinsuchaway that
{ fooC Iv(s)[
qds}
1-2/qJ(v). (6.9)
The renormalized v satisfies conditions
(2.2)-(2.5).
Condition(2.6)
resultsfrom(6.9)
andstatement(ii), Lemma 3.5.6.7. ApplyingLemma 3.2concludes theproof.
7.
PROOF OF THEOREM
2.27.1.
A
function which satisfiesthe followingdifferentialequationcl
(s(s + 47r)v’(s)) + Iv(s)l
q-2-v(s)
O,ds
(7.1)
for 0
<
s<
oc,decaysatinfinityinsuchaway thatv(oe)-
0 and(sv’(s))2ds <
oe,(7.2)
and obeys the following boundarycondition
-47rv’(0)- Iv(0)l
q-2.v(0) (7.3)
is given by
v(s)- (1
-k-)-1
orv(s) x/. +
-1according towhether q- 3 or q- 4.
Therefore Theorem 2.2 results from Theorem 2.1 and the following statement" if q
<
3_<
c, conditions(7.1)-(7.3),
plusthe followingonev’(s)<O
for0<s<(7.4)
identify vuniquely.
Thesubsequent subsectionsstep towardaproofofthis statement.
7.2. Two remarks areinorder:
(i) Let 3
<_q<
oe; if v satisfies(7.1)
and (7.2), then$21(S)---0
asS--+0o
(ii) Let2
<
q<
oc; if vsatisfies(7.1)
andsv’(s)
--,0 ass 0 then(7.3)
holds.Proof of
(i) Equation(7.1)
giveslim
t(t + 47r). v’(t) s(s + 47r). v’(s) + Iv(t)lq-:v(t)
dt ifsis positive;condition(7.2)
andLemmas 4.2and 5.3imply thatifq>3.
Proof of
(ii) The hypothesesgives(s + 47r). v’(s) Iv(t)lq-2v(t)
dtfor every positive s, and v(s)=o(log(1/s)) as s0. The proof goes ahead asin subsection6.4.
7.3. Ifq
>
3, thechange ofvariables definedby47r
(_)
s- t- log /
v(s)- ,l/(q-2)u(t)
e
- 1’ -
converts the set made up by
(7.1)-(7.4)
into the set consisting of conditions(7.5)-(7.8).
Inotherwords, ourgoal becomes identifyinga sufficiently smooth real-valued function defined in [0, oc[, u, and a nonnegative par.ameter,A,
such that u andX
satisfy the following equationd2u
dt2
(t) + A(sinh t)-21u(t)lq-2u(t)
0(7.5)
for 0
< <
ec, andusatisfies the followingconditions:u(0)-0
and du-(0)-
1,(7.6)
du
d--- (oc)
0,(7.7)
du
dr(t)>0
for0<t<oo.(7.8)
7.4. IfA ispositive and2
<
q< ,
any solutionto(7.5)
and(7.6)
has the following properties:(i)
du and
,u(t)l <_ min{t, (-A)l/q (sinht) 2/q)
for every nonnegative t.
(ii) uisasymptoticallylinear moreprecisely,twoconstantsAandB exist such that
--(t)
db/ A+
A.o(tq-le -2t)
andu(t)
At+
B+
Ao(tq-le -2t)
as t-00.
(iii) uhas finitely manyzeroes and finitely many bend points.
(iv) uis concaveifand onlyifuhasnopositivezeroes.
(v)
uisincreasingifandonlyif uhasnopositivezeroes.Proof of
(i) Let Hbe defined byH(t)
du(t) + (2A/q). (sinh t)
-2.lu(t)l q.
Equation
(7.5)
gives dHdt
(t) (2A/q). lu(t)l
q d(sinh t) -2,
therefore
dH dt
(t) _<
0for every positive t. Initial conditions
(7.6)
giveH(0+) .
Wededuce
+ (2A/q). (sinh t)
-2.lu(t)l
qfor every positivet.
Property
(i) follows.Proof of
(ii) Equation(7.5)
plusinitial conditions(7.6)
give dud-- (t) A (sinhs)-2lu(s)lq-2u(s)
ds.Hencewehave
du
f0
d- (t) A (sinhs)-2]u(s)lq-2u(s)ds +
aremainder,where
[remainder < A (sinhs)-2[u(s)l
q-1ds.Property
(ii) follows, since(i) giveslu(t)l <_
t.Proof of
(iii) Equation(7.5)
plusinitial conditions(7.6)
give dudt(t) >_ A f0’ lu(s)[q-l(sinhs) -
ds.Therefore
du
]0 -
d---- (t) >_ A sq-(sinhs)
-2dsbecauseofproperty(i),consequently
du A q-2
d-(t) >
q-2and
u(t) >_
t- q-1.(q- 2)(q- 1)
We deduce that
u(t)
increases strictly as increases from 0 to [(q-2)/A]
1/(q-2) and is .strictly positive as 0<
<[(q-2)(q-1)/A]I/(q-2) in other words, u has neither positive zeroes nor bend points in some neighborhood of 0.Letz bedefined by
z du
-.
sinht.(7.9)
Equations
(7.5)
and(7.9)
yieldz.cosh
(dz/dt)
sinhA[ulq-2u.
Eliminatingubetweenthe last twoequations gives
d2z {
-1+ (q- 1)(sinh
/l/(q-1) z cosh dz (q-2)/(q-1)}
-.
sinh z O.Thereforezobeys the following equation
d22
dt2 t-
[-1 + Q(t)]z
O.(7.10)
Here
Q(t) (q 1),(sinh t)-21u(t)l q-2, (7.11)
acoefficient which willplaya roleinsubsequent developments too observe that property(i) yields
0
< Q(t) < (q- 1)(sinht)-2t
q-2.(7.12)
Since the coefficient ofz inEq.
(7.10)
approaches -1 fast enough as t--+ec, Sturm comparison theorem (see e.g. [21, Section 20], [19,Chapter
1])
orstandard oscillationtheorems(see
e.g. [19, Chapter2])
guarantee thatEq. (7.10)
is nonoscillatory, i.e. the zeoresofz donot clusteratinfinity.Property
(iii) follows.Observeincidentally thatifHisdefined by
-- (t).sinh +(2A/q). lu(t)l q,
then
Eq. (7.5)
givesH’(t)
du- (t) d (sinh t) 2,
in otherwords, His anincreasingfunction. Therefore
lu(t)l < lu(t2)l < lu(t3)l
ift,t2,t3... are the bend points ofu arranged inincreasing order comparewith SoninandP61yatheorem[21, Section 19].
Properties (iv) and
(v)
are an immediate consequence of(7.5)
and(7.6).
7.5. Letus assume2
<
q<
oc, and examine how the solution uto(7.5)
and(7.6)
depends upon parameterA.Clearly,
u(t)
=_ if 0. On theotherhand, uremains anincreasing function of if is positive and smallenough the inequalitydt
(t) >_ A sq-(sinhs)
-2ds (derivedinthe previous subsection)and the formulasq-
(sinh s)
-2ds 22-qr(q)(q 1)
(appearingin[8,Section
3.527])
tellusthatdu(t)/dt
ispositive for every positive if2q-2
[F(q)(q- 1)]
Let L be Ou/O), the derivative ofu with respect to
. An
inspectionshowsthatLisgiven bythe following formula
(q- 2)AL
w-u,(7.13)
where wobeys
d2w
dt2
(t) + Q(t)w(t)
0(7.14)
for 0
< <
oandw(0)
0, dwd--- (0)
1.(7.15)
Coefficient
Q
is defined as in(7.11).
The following properties are easilyinferredfrom(7.12),(7.14)
and(7.15):
(i)
]w(t)l < (Constant).
for every positive t;(ii)w is asymptotically linear, and
Iw-(asymptote)l
approaches 0 exponentially fastas approaches;
(iii)whas finitely manyzeroes and finitely many bend points.
LF.MMA 7.1 Letusatisfy
(7.5), (7.6),
andletwsatisfy(7.14)
and(7.15).
Then
w(t) < u(t) 0
ispositiveand doesnotexceed thefirst
positivezeroofw.
Proof
Equation(7.5)
readsdZu/dt
2+ P(t)u
0, where (q-1). P(t)
Q(t). On the other hand, u and w obey the same initial conditions.
Theneither thecomparison theorem appearing in [21, Section 20] or Levincomparison theorem seee.g.[19,Chapter 1, Section7] leads to theconclusion.
LEMMA 7.2 Supposeu
satisfies
(7.5),(7.6)
and(7.8).
Then the solution w to(7.14) and(7.15)
has one positivezero at most.Althoughaformalproofeluded theauthors,the truth ofLemma7.2 may be reasonablyinferredfrom the following facts:
(i) Suppose u satisfies
(7.5), (7.6)
and(7.8).
Then the solution w to(7.14)
and(7.15)
cannothave two distinct zeroes in the following set:{t>0" t.cotht>q/2}. (7.16) (Observe
that t.coth is convexandincreases strictly from to as increases from 0 to.
The root to t.cotht-q/2 is 1.287839 if q-3, is 1.915008 if q-4, lies below q/2 and approaches q/2 asymptoticallyas qgrowslarge.)(ii) Suppose usatisfies
(7.5)
and(7.6),
assumew satisfies(7.14)
and(7.15),
and let a and b obey 0<_
a<
b,w(a) w(b)
0 andw(x)
0fora<
x<
b. Then,bq-3(b a) _>
q2. (7.17)
q-1
(iii)
Suppose
q,,
andaneighborhood of 0are specified. Then bothu(t)
andw(t)
canbe computedwithany prescribed accuracy for every from that neighborhood.Numerical tests show thatno morethanone zero ofw occurs aslongas uremains positive relevantinformation canbe foundin[15].
Proof of
(i) Letube anysolution to (7.5),and letzbe definedbyz(t) u(t)
t.u’(t), (7.18)
the height above the origin of the tangent straightline tothegraphofu at
(t, u(t)).
Equations(7.5)
and(7.18)
givet2d( u)- 7
+z-0,dZd____ "t(sinht)-2lulq-2u"
Eliminatingubetween the lasttwo equations gives
d2z ( q)dz
dt2
+ - (q-
2cotht--1),l/(q-l)(sinht)
dt -2/(q-l)dzl(q-2)/(q-l)
---
z O.(7.19)
In other terms,wehave
dt2
+(2cotht-q) 7
dz-d + Q(t)z- o, (7.20)
provided
Q
isdefined by(7.11).
Suppose u obeys
(7.6)
and(7.8)
too. Then u vanishes at 0 and is concave, hencez(t) >
0 and-.(t)
dz>
0(7.21)
for every positive t. Consequently,Eq.(7.20)
readsdt2
--+ [Q(t)+ (t. cotht-) (a
positivecoeff.)]z-
0.(7.22)
Equations
(7.14)
and(7.22),
inequalities(7.21)
and Sturmcomparison theorem lead tothe conclusion.Proof of
(ii) Statement (ii) follows from a variant ofde la Vall6e- Poussintheorem see[21,
Section17].
Itis aneasymatter toshow that if a, band wobeyO
<_
a<
b,w(a) w(b)
O, andw(x) =/:
O fora<x<b thenfab w(t)
d2w (t)
Therefore wehave
dt
> (’/- + V/-d)2
b-a
(7.23)
tQ(t)
dt>_ +
b a Sinceinequality
(7.12)
impliesQ(t) <_ (q- 1),/q-4,
the conclusionfollows.
7.6. Equation(7.13), and Lemmas 7.1 and7.2 tellusthat
L(t) <
0for every positive ifusatisfies
(7.5), (7.6)
and(7.8).
Inother words,the solutionuto(7.5)
and(7.6)
decreases steadilywithrespectto,
aslong as uitself remains anincreasingfunction oft.Thisimplies that thesolution to
(7.5)-(7.8)
isunique, and concludes the proof.References
[1] A. Baernstein, II, Aunified approachto symmetrization, pp. 47-91, in Partial
DifferentialEquationsofEllipticType (A.Alvino,E.Fabes andG. Talenti,Eds.), Symposia Mathematica,Vol.35, CambridgeUniv.Press,1994.
[2] L. Bianchi,LezionidiGeometriaDifferenziale,Vol. 1,Zanichelli, 1927.
[3] G.A. Bliss,Anintegral inequality, J.London Math.Soc.5(1930)40-46.
[4] Yu.D.BuragoandV.A.Zalgaller,GeometricInequalities,Springer-Verlag, 1988.
[5] I. Chavel, Riemannian geometry: A modern introduction, Cambridge Tracts in Mathematics, Vol. 108, CambridgeUniversityPress,1994.
[6] N. DunfordandJ. Schwartz,LinearOperators. Part I."GeneralTheory,Interscience Publ., 1958.
[7] L.E.Fraenkel, On steadyvortexringswith swirlandaSobolev inequality,pp. 13- 26,inProgressinPartialDifferentialEquations." CalculusofVariations,Applications (C.Bandle, J. Bemelmans,M.Chipot, M.GrfiterandJ.SaintJean Paulin,Eds.), PitmanResearchNotesinMath.Vol. 266,1992.
[8] I.S. Gradshteyn and I.M. Ryzhik, Tables ofIntegrals, Series, and Products, AcademicPress,1965.
[9] G.H. Hardy, J.E.LittlewoodandG. P61ya, Inequalities, CambridgeUniv.Press,first edition1934,second edition 1952.
[10] E. Kamke, Differentialgleichungen, L6sungsmethoden und LO’sungen. Band I."
Gew6hnlicheDifferentialgleichungen, Leipzig, 1959.
[11] B. Kawohl,Rearrangements and convexityof level sets inPDE,Lecture Notesin Math.,Vol.1150, Springer-Verlag, 1985.
[12] V.I. Kolyada, Rearrangements of functions and embedding theorems, Russian Math.Surveys,44(5) (1989)73-117.
[13] J. McCleary, Geometryfrom aDifferentiable Viewpoint, Cambridge-Univ. Press, 1994.
[14] H.P. McKean,Anupperbound to thespectrum of Aonamanifoldof negative curvature,J.DifferentialGeometry 4(1970)359-366.
[15] F. Mugelli, Complementary techniquesin Sobolevinequalitiesfor thehyperbolic plane,Report,FlorenceUniv.Math. Dept. (1997).
[16] F. MugelliandG.Talenti, Sobolevinequalitiesin 2-dimensionalhyperbolicspace, InternationalSeriesofNumerical Mathematics, Vol. 123,Birkh/iuserVerlag, 1997.
[17] G. Sansone and J. Gerretsen, Lectures on the Theory ofFunctionsofaComplex
Variable, Vol.2, Wotters-Noordhoff, 1970.
[18] C.L.Siegel,Topicsincomplexfunctionstheory,Vol. 2:Automorphicfunctionsand Abelian integrals, Interscience Tracts in Pure Appl. Math., Vol. 25, Wiley- Interscience, 1971.
[19] C.A.Swanson,Comparison and OscillationTheoryofLinearDifferentialEquations, AcademicPress, 1968.
[20] G.Talenti,Inequalitiesinrearrangementinvariant function spaces, pp.177-231,in Nonlinear Analysis,FunctionSpacesand Applications, Vol. 5(M.Krbec,A.Kufner, B. OpicandJ.Rakosnik,Eds.),Prometheus Publ,Prague,1994.
[21] F.G.Tricomi, Equazionid[.’/’erenziali, Boringhieri, 1961.