Photocopying permittedbylicenseonly licensebyGordon and Breach Science Publishers Printed in Malaysia
On Weighted Hardy and Poincar -
type Inequalities for Differences
V.I. BURENKOV a, W.D. EVANS
a andM.L. GOLDMAN
bSchool ofMathematics, University of Wales, Cardiff, SenghennyddRoad, Cardiff, CF24AG,
UK
Moscow State
Inst.ofRadioEngineering, ElectronicsandAutomation (Tech. Univ.), Pr. Vernadskogo78,Moscow,
Russia(Received7June1996)
Acriterion is obtainedfor theHardy-typeinequality
tf if(x)lPv(x)dxl
1/p <cl{(
v(a)fo
If(x)lPdxt foal tl/P I
+
If(x) f(y)lPw(Ix yl)dxdytobe valid for0< a <A< cxand0<p <cx.Thisweakens a criterionpreviouslyfoundby the firsttwoauthors and comes closetobeing necessaryaswellassufficient.Whenaninequality inthe criterion isreversed, aPoincar6-type inequalityis derived in somecases.
Keywords: Hardy;Poincar6;inequalities;differences.
AMS1991 SubjectClassification" Primary:26D99,Secondary:36B72.
1 INTRODUCTION
In [2]
theproblem
of existence of aboundedlinear extension ofWp
(’)(fl)
into
W
(’)(Rn),
for spaceswithsome"generalized" smoothnessdefinedby functions ) and v respectively,wasinvestigated in the case of domains fl admitting arbitrarily strong degeneration.A
centralroleintheanalysiswas played bythefollowingHardy-type
inequality:forallf Lp(O, a)
2 V.I. BURENKOVetal.
(fo
aIf(x)
pv(x)dx tl/p
<_ Cl{(fo v(a)
aIf(x)
p dx)lip
(foafo
a)l/p]
+ If(x)-
f(y)Ipw(I
x-y I)dxdy(1.1)
wherecl> 0 isindependentoff
anda.It
wasassumed that 0 <a<_A
<_ x, 0 < p < x, w is anon-negative measurablefunction on (0,A)
which is suchthat for all x(0, A)
fx
Av(x) := B + w(t)dt
< cx(1.2)
for some
B
6 [0, cx),and there existsot 6 (1,2)
such thatv(x)
< otv(2x), x 6 (0,A/2). (1.3)
WhenaA
cxz, theinequalitybecomesIf(x)
pv(x)dx
< Cl
f(x)
f(y) pw(I
x y I)dxdy(1.4)
for allf Lp(O, xz).
Specialcases ofthe inequality, and analogousones, werepreviously studied by Yakovlev [8,9]; see also Grisvard [4], Kufner andPersson
[5], Kufner and Triebel[6]
and Triebel [7]; a discussion of these earlier works and further referencesmaybe found in[2]. Necessary
conditions arealsogivenin[2]forthevalidityof(1.1),
and ampleevidence is provided that thesufficiencycondition(1.3)
isclosetobeing necessary.It
is alsoworthyof noteforsubsequentreference that(1.3)
isshown in[2,Remark2.4]
toimplythato
x
<_
c2xv(x),
x (0, A),(1.5)
for somec2 > 0,and this isclearly necessaryfor(1.1)
asthe choicef
1indicates.
In
this paper we adapt the techniques in[2]
to obtain(1.1)
under a weaker condition than(1.3),
thereby narrowingstillfurtherthegapbetween sufficiency and necessary conditions.Moreover,
weprove
that when a condition which, in some sense, is converse to that whichreplaces(1.3)
isassumed,aPoincar6-type inequalityfor differences is obtained.2
A HARDY-TYPE
INEQUALITYTHEOREM2.1
Let
0 < p < cx, 0 <A <_
cxz, 0 <B
< cx, andletw be anon-negative measurablefunction
on(0, A)
which issuch that(1.2)
and(1.5)
aresatisfied for
all x E(0, A).
Furthermore, supposethatthereexist Ix > 0and 0 <’
< 1 such thatv -v <
),v(x)
x E(0, Ao) (2.1)
+1
where
Ao
min{1,Ix}A.
Then,for
alla(0, A]
and allmeasurablef
on(O,a), (1.1)
issatisfied,
whereCl isindependentof f
anda.In
particular,(1.4)
issatisfied for
allf Lp(O, o).
Proof Let
1 < p < cxinitially.As
in[2]
we start with theinequalityIf(x)l
_< If(y)[+ If(x)
f(y)[.For
e (0,a/[Ix +
1]),thisgives(fea/(/z+l) f
+l)x[f (x)lPw(y
x)dydx(f- [y/(tx+l) )liP
<_
If (x)lPw(y
x)dxdy+l)e JO
(foafo
y+ If(x) f(y)lPw(y
x)dxdyand so
If (x)l
pw(t)dtdx
(fea f.y )lip
_< If(y)l w(t)atay
y/(/z+l)
where
+A
foafo
a(1/2) If(x) f (y)lPw(lx
yl)dxdy.4 V.I.BURENKOVetal.
Hence
wehavefa/(/z+l) If (x)IP[v(Ixx) v(a x)]dx )
lip(fe
a lZX)alp
<
If(x)lP[v( v(x)]dx +A
a/(/z+l)
<
If(x)IP[v(
/ZXv(x)]dx
p+l(lZafaa )l/p
+ v((/x ---) + 1) [f(x)lPdx + A.
/(/z+l)
From (2.1)
itfollows that(;x )
v -v(x) <
,v(tzx) +1
andconsequently
O < x <
a/(tx +
l)fe
a/(/x+lIf (x)lP[v(txx) v(a
x)]dx)
1/p<_
YIf (x)lPv(tzx)dx
(lafaa(i
z+ )lip
+ v(i)---z)
/(/.,+1)If(x)lPdx
Also v
(a x) _<
v in[0,a/ (/z + 1)],
and so, sinceot1/p+ /
1/p _<21-1/P(0t + )l/p,
wehaveIf (x)lPv(x)dx
If(x)lP[v(x) v(a x)]dx
(a/(+l))lip
+ lf(x)[Pv(a -x)dx
fea/(/x+l) )
1/p<_
,,1/p if (x)lPv(ixx)dx
((a)foa )lip
+ 2P-lv
(/Z + 1)
aIf(x)lPdx
On
allowingIf
e-- (x)lPv(x)dx
0thisgives+ A. (2.2)
(2.3)
Iza p
lip
_<
21-l/P(1 yl/p)-I
V(/./, -+- 1)
2If(x)
dx+ A
In [1]
itisprovedthat(1.5)
impliesthe existence of 6(0, 1)
andc3 > 0 suchthatx
v(x)
< c3y’
v(y),Hence, if/x
> 1, wehave0 < x < y.
(2.4)
v(x)
<c3x- (txx) v(Ixx)
c31x’ v(txx), (2.5)
while
v(x)
<v(txx)
if 0</x < 1.Similarly(.a)
v <
c3(1 + 1/lz)’v(a).
/z+l
Therefore,from(2.3),for somec > 0 independentof
f
and a,(2.6)
(fa/(/z+l) [f (x)lPv(x)dx )lip
< c{(fo v(a)
aIf (x)lPdx )lip + A }
dO
Finally
(1.1)
followsfrom(fo
aIf (x)lPv(x)dx )lip <__ (f0a/(/x+l) ]f (x)lPv(x)dx )lip
( fa
a+ v(a/[lz+ 11)
/(/x+l)
If(x)lPdx)
1/pand
(2.5).
6 V.I.BURENKOVetal.
The case0 < p < 1 canbe treated similarly,theonly differencebeing that we start with
If(x)l
p< If(Y)l
p+ If(x)
f(y)lpand, after multiplying byw(y x), integratethisinequalityin the sameway as before instead ofapplying
IIL.
COROLLAR’
2.2Suppose
thatfor
some positiveetl,etasatisfyingetl < 1
+ 1/et2 (2.7)
wehave that
v(x)
<etlV([1+ 1//xlx), v(x)
<etav(tzx),
x E(0, A1), (2.8)
whereA1 min{1//x,/z//x + 1}A.
Then(1.5)
and(2.1)
aresatisfied.
Proof It
follows from(2.8)
that eitheretl < 1+ 1//x
or0/2 < /Z. The argumentin[2,
Remark2.41
canbeusedtoshow that(1.5)
issatisfied.For
instance,
suppose
et2 </xand/z
> 1,the other casesbeingsimilar. Thenfo
xv()cl
k=0J/z-k-xv()ct
<_ k=0a
Jg-k- xfx
x’_,(etz/x)
v(t)dt=o
(by(2.8),)
<
(1- ot2//x)-l(1- 1/tx)xv(x/tz)
<c2xv(x)
which is
(1.5). Moreover
v(x/(lz
"4-1))- v(x/l)
< (etl-1)U(X//) <
(etl-1)et2v(x)
and
(2.1)
follows since(etl1)et2
< 1by(2.8).
Remark 2.3 Theorem 2.1 in
[2]
is the special case/z 1,et2 1 of Corollary 1.Note
that(2.8)
reducesto asingle inequalityin one other case, namelyv(x) <_
av x1+
thegoldenratio.whereet < 2
’
xE(0,
1f)
2A(2.9)
Remark 2.4 Since
foafo
aIf(x) f(y)lPw(Ix
yl)dxdy 2fo
aIlAhfll,(O,a_h)W(h)dh,
where
Ahf(X f(x -+-
h)f(x),
theinequality(1.1)
maybe written as(fo
aIf (x)lPv(x)dx
< c8{(fo v(a)
aIf(x)lPdx
(fo
a)lip I
+ IlAhfllP,(O,a_h)W(h)dh
In
[3] Burenkov and Goldman have proved that(1.5)
is necessary and sufficientfor thevalidityof arougher inequalitywhere
(fo
aIf(x)lPv(x)dx
<__ C9{(fo v(a)
aIf (x)lPdx )lip
a
t ]
d-
OOh,p(f)Pw(h)dh
09h,p(f)
supIIAtf[ILp(0,a_t),
0<t<h
the modulus ofcontinuityof
f.
3
A POINCARIP.-TYPE
INEQUALITYTHEOREM3.1
Let
O < p < cxz, O <A
< o,0 <B
< cx and let w be anon- negativemeasurablefunction
on (0,A)
whichsatisfies (1.2)for
allx (0,
A). Suppose
thereexisttz > 0 andy > 1 such thatv(x/[tz +
1])-v(x/lz)
> yv(x), x 6 (0, A0),(3.1)
whereA0
min{ 1, tz}A,
andthatif
lz 5 1 thereexistc4 > 1 andc5 (0,1)
suchthat
V(X)
<_41)(X),
X(0, A) (3.2)
8 V.I. BURENKOVetal.
if
lz < 1 andV(X)
<{ 41)(/ZX), csv(lzx/[lz + 1]),
Xx 6(0, A [A/tz, A) (3.3)
if
lz> 1. Then,for
alla(0,
A]andf
such thatf
bLp(O,
a;v(x)dx) for
somebC
(fo
aIf(x) blP v(x)dx )liP
(foafo
a)lip
< C6
If(x) f(y)lPw(lx
yl)dxdy wherec6 isindependentof f,
b anda.(3.4)
Proof It
isclearly enoughtoprove
thetheoremfor b 0.Let
1 < p< cx:the modificationsnecessaryfor 0 < p < 1 are as in theproofof Theorem 2.1.
On
startingwiththeinequalityIf(Y)l < If(x)l + If(x) f(Y)l
andfollowing the initialsteps of theproofofTheorem 1 with e 0,we obtain
(fo
aIf (x)lP[v(lzx/(tz
/1))
v(x)]dx(foa/(/x+l) )lip
<_
If (x)lP[v(txx) v(a
x)]dx If 0 </z _< 1,then+ . (3.5)
(fo
aIf (x)lP[vOzx/(lz
/1))
v(x)]dx)lip
(fO
a)lip
<
If(x)lPv(Izx)dx + A
and,by
(3.1),
thisgives(fo
af (x) lP v(x)dx )liP <_ If (x)lPV(lx)dx
lip
<_
(l/p_ 1)-IA
since,by
(3.2), f Lp(O,
a;v(tzx)dx).
Thus(3.4)
follows.Let
IX > 1.Ifa <A/IX,
weobtainfrom(3.5)
and(3.1)
that(fo
aif(x)lPv(ixx)dx
<(y1/e 1)-IA,
theleft-handsidebeingfinite sincev(ixx) <
v(x).
Thus(3.4)
followsin this caseby(3.3).
Ifa >A/Ix,
we have from(3.1),(3.3)
and(3.5)
(fo ’
aflzIf (x)lPv(Ixx)dx -t- fA If (x)lP[v (IxX) v(x)]dx
/
(foa/i
x)lip
<
[f(x)[Pv(Ixx)dx -+-
A.Theleft-hand side isagainfinite and
(3.4)
follows from(3.3).
COROLLARY
3.2Suppose
thatfor
somepositiveor1,Or2satisfyingOtl> 1
+ 1/Or2 (3.6)
wehave that
V(x)
>_ OtlV([1-+- 1/Ix]X), v(x)
>_C2V(IxX),
x E(0, A1),
whereA1 min{1/Ix, Ix/Ix + 1}A.
Then(3.1)
issatisfied.
Proof From
(3.7),for x (0, A0),(3.7)
v(x/[Ix + 1])- v(x/Ix) (o 1)v(x/Ix) >
(Otl-1)c2v(x)
and thisyields
(3.1)
since(ol1)ot2
> 1by(3.7).
1+, in
(3.5)
we obtainthe Remark 3.3 On choosing Ix 1 and Ix 2followingtwosufficiencyconditionsfor
(3.1)
tobe valid:v(x)
>otv(2x) for
c >2, x E(O,A/2), (3.8) for
ot > x 6(0, 2A/[
1+ ,/]).
(3.9)
Remark 3.4 The choicef (x)
c7
b in(3.4)
yieldsa contradictionunless c bLp(O,
a;v(x)dx). Hence
it isnecessary thata
v(x)dx
cx.(3.10)
10 V.I. BURENKOVetal.
Remark 3.5
From (3.7)
and(3.8)
itfollowsthatx
a
v()d
< 7XU(X), X (0, A),(3.11)
for somec7 > 0.For
instancesuppose thator2 >/z,/x> 1 andA
cxz, the other casesbeingsimilar. Thenv()
k=0(tz/2)
k=0 kxv() " v()d
_< k=0<_- (1-/d,/ot2)-l(/.z-kx v( -) 1)xv(x).
Acknowledgements
M.L.
Goldman gratefully acknowledges the support of theEPSRC under grant GR/K69681,which madepossibleavisit totheSchoolof Mathematics, Cardiff,wherethiswork wasstarted. Also,thethree authors aregratefulfor supportfrom theINTAS
scheme(No. 94-881).
References
[1] N.K. Bariand S.B. Strechkin. Best approximationsand differential properties oftwo conjugate functions(Russian). Proc. MoscowMath.Soc.,5(1956),483-521.
[2] V.I.Burenkov andW.D. Evans. Weighted Hardy-type inequalitiesfor differencesand the extensionproblemforspaceswithgeneralisedsmoothness.Toappear inJ.London Math.
Soc.
[3] V.I.Burenkov andM.L.Goldman.Necessaryand sufficient conditions for the validity ofa weighted Hardy-type inequalityfor the modulus ofcontinuity. In preparation.
[4] P.Grisvard.Espacesintermediaresentreespacesde Sobolev avec poids.Ann.ScuolaNorm.
Sup.Pisa,23(1969),373-386.
[5] A. Kufner and L.E. Persson. Hardy inequalities of fractional order via interpolation, ResearchReport17,Dept.Math. Lulea Univ. Techn.,ISSN1101-1327,ISRNHLU-TMAT- RES-93/17-SF,(1993),1-14.
[6] A.Kufner andH.Triebel. Generalisations ofHardy’s inequality.Conf Sere.Mat. Univ.
Bari.,156(1978),1-21.
[7] H.Triebel.Interpolation Theory,FunctionSpaces,DifferentialOperators, VEBDeutscher Verlagder Wissenschaften, Berlin, 1978;North-Holland, Amsterdam-New York-Oxford, (1978).
[8] G.N.Yakovlev.Boundary propertiesofacertainclass of functions (Russian).TrudySteklov Inst.Math.,60(1961),325-349.
[9] G.N.Yakovlev.Onthetracesof functions from thespace