• Non ci sono risultati.

The Orlicz version of the Lp Minkowski problem for −n < p < 0

N/A
N/A
Protected

Academic year: 2021

Condividi "The Orlicz version of the Lp Minkowski problem for −n < p < 0"

Copied!
29
0
0

Testo completo

(1)

Contents lists available atScienceDirect

Advances

in

Applied

Mathematics

www.elsevier.com/locate/yaama

The

Orlicz

version

of

the

L

p

Minkowski

problem

for

−n

< p

< 0

Gabriele Bianchia,∗, KárolyJ. Böröczkyb,c, Andrea Colesantia aDipartimentodiMatematicaeInformatica“U.Dini”,UniversitàdiFirenze,

VialeMorgagni67/A,Firenze,I-50134,Italy

b

AlfrédRényiInstituteofMathematics,HungarianAcademyofSciences, Reltanodau.13-15,H-1053Budapest,Hungary

cDepartmentofMathematics,CentralEuropeanUniversity,Nadoru9,H-1051,

Budapest,Hungary

a r t i c l e i n f o a b s t r a c t

Articlehistory:

Received28June2019 Accepted9August2019 Availableonline30August2019

MSC:

primary52A38,35J96

Keywords:

LpMinkowskiproblem

OrliczMinkowskiproblem Monge-Ampèreequation

Given a function f on the unit sphere Sn−1, the Lp

Minkowski problem asks for a convex body K whose Lp

surface area measure has density f with respect to the standard(n−1)-HausdorffmeasureonSn−1.Inthispaperwe dealwiththegeneralizationof thisproblemwhich arisesin theOrlicz-Brunn-MinkowskitheorywhenanOrliczfunction ϕ substitutestheLpnormandp isintherange(−n,0).This

problemisequivalenttosolvetheMonge-Ampereequation ϕ(h) det(∇2h + hI) = f on Sn−1,

whereh isthesupportfunctionoftheconvexbodyK. ©2019ElsevierInc.Allrightsreserved.

Firstandthirdauthorsare supportedinpartbythe GruppoNazionaleper l’Analisi Matematica,la

Probabilità e le loro Applicazioni(GNAMPA) of the Istituto Nazionale di AltaMatematica (INdAM). SecondauthorissupportedinpartbyNKFIHgrants116451,121649and129630.

* Correspondingauthor.

E-mailaddresses:gabriele.bianchi@unifi.it(G. Bianchi),boroczky.karoly.j@renyi.mta.hu

(K.J. Böröczky),andrea.colesanti@unifi.it(A. Colesanti).

https://doi.org/10.1016/j.aam.2019.101937

(2)

1. Introduction

Weworkinthen-dimensionalEuclideanspaceRn,n≥ 2.Aconvex body K inRn is

acompactconvexsetthathasnon-emptyinterior.Given aconvexbodyK,forx∈ ∂K we denote by νK(x) ⊂ Sn−1 the family of all unit exteriornormal vectors to K at x

(theGaußmap).Wecanthendefinethesurfaceareameasure SK ofK,whichisaBorel

measure ontheunitsphereSn−1 ofRn, asfollows:foraBorelsetω⊂ Sn−1 weset SK(ω) =Hn−1



νK−1(ω)=Hn−1({x ∈ ∂K : νK(x)∩ ω = ∅}) (see, e.g.,Schneider[38]).

TheclassicalMinkowskiproblemcanbeformulatedasfollows:givenaBorelmeasure

μ onSn−1,findaconvexbodyK suchthatμ= SK.Thereaderisreferredto[38,Chapter

8] foranexhaustivepresentationof thisproblemanditssolution.

Throughoutthispaperwewillconsider(eitherfortheclassicalMinkowskiproblemor foritsvariants)thecaseinwhichμ hasadensityf withrespecttothe(n−1)-dimensional HausdorffmeasureonSn−1.UnderthisassumptiontheMinkowskiproblemisequivalent tosolve(intheclassicorintheweaksense)adifferentialequationonthesphere.Namely:

det(2h + hI) = f, (1)

where:h isthesupportfunctionofK,∇2h isthematrixformedbythesecondcovariant

derivativesofh withrespecttoalocalorthonormalframeonSn−1 andI istheidentity matrixoforder (n− 1).

Many different types of variations of the Minkowski problem have been considered (wereferforinstanceto [38,Chapters8and9]).Ofparticularinterestforourpurposes is the so called Lp version of theproblem (see [1–11,14–34,36,37,39–50]).At theorigin ofthis newproblem thereisthereplacementoftheusualMinkowskiadditionofconvex bodiesbythep-addition.Asaneffect,thecorresponding differentialequationtakes the form

h1−pdet(2h + hI) = f, (2)

(see[38,Section9.2]).ThestudyoftheLpMinkowskiproblemsdevelopedinasignificant wayinthelastdecades,asapartofthesocalledLpBrunn-Minkowskitheory,which rep-resentsnowasubstantialareaofConvexGeometry.Oneofthemostinterestingaspects ofthisproblemisthatseveralthresholdvaluesoftheparameterp canbeidentified,e.g. p= 1, p= 0, p=−n, across whichthe natureof theproblem changesdrastically. For an account on the literatureand on the stateof the art ofthe Lp Minkowskiproblem (especially forthevaluesp< 1)wereferthereaderto[1] and[2].

Of particular interest here is the range −n < p < 0. In this case Chou and Wang (see [10]) solved thecorresponding problem when the measure μ hasadensity f , and

(3)

f isbounded and bounded awayfrom zero.This result wasslightlygeneralised bythe authorsincollaborationwithYang in[1],where f isallowed tobe inL n

n+p.

Theorem1.1(ChouandWang; Bianchi,Böröczky,Colesanti andYang). Forn≥ 1 and

−n< p< 0, if thenon-negative and non-trivialfunction f is in L n

n+p(S

n−1) then (2)

has a solution in the Alexandrov sense; namely, f dHn−1 = dSK,p for a convex body

K∈ Kn

0.In addition, iff is invariantunder aclosed subgroup G ofO(n), then K can

bechosentobe invariantunder G.

AsafurtherextensionoftheLpMinkowskiproblem,onemayconsider itsOrlicz ver-sion.Formally,thisproblem arisesinthecontextoftheOrlicz-Brunn-Minkowskitheory ofconvexbodies(see[38,Chapter 9]).Inpractice,therelevantdifferentialequationis

ϕ(h) det(∇2h + hI) = f,

where ϕ is a suitable Orlicz function. The Lp Minkowski problem is obtained when ϕ(t)= t1−p, fort≥ 0.

Whenϕ : (0,∞)→ (0,∞) iscontinuousandmonotonedecreasing,thisproblem(under

asymmetryassumption) hasbeen consideredbyHaberl, Lutwak,Yang,Zhang in[17]. Comparingthepreviousassumptionsonϕ withtheLpcase,weseethatthiscorresponds tothevaluesp≥ 1.

Weareinterestedinthecaseinwhichthemonotonicityassumption isreversed, cor-respondingto thevaluesp< 1.Henceweassumethatϕ : [0,∞)→ R iscontinuousand monotoneincreasing,havingtheexampleϕ(t)= t1−p,p< 1,asaprototype.Tocontrol inamorepreciseformthebehaviourofϕ withthatofapowerfunction,weassumethat thereexists p< 1 such that

lim inf t→0+

ϕ(t)

t1−p > 0. (3)

Concerningthebehaviourofϕ at∞ weimposethecondition:



1

1

ϕ(t)dt <∞. (4)

The corresponding Minkowski problem in this setting can be called the Orlicz Lp Minkowskiproblem.Thesolutionofthis problemintherangep∈ (0,1) isdueto Jian, Lu[25]. We alsonote thatOrlicz versionsof theso called Lp dual Minkowskiproblem havebeenconsideredrecentlybyGardner,Hug,Weil,Xing,Ye[13],Gardner,Hug,Xing, Ye[14],Xing,Ye,Zhu [43] andXing, Ye[44].

In this paper we focus on the range of values p ∈ (−n,0). As an extension of the resultscontainedin[1],weestablishthefollowingexistencetheorem(notethat,asusual

(4)

inthecaseofOrliczversionsofMinkowskitypeproblems,wecanonlyprovideasolution upto aconstantfactor).

Theorem 1.2. For n ≥ 2, −n < p < 0 and monotone increasing continuous function ϕ: [0,∞)→ [0,∞) satisfyingϕ(0)= 0, andconditions (3) and(4), if thenon-negative

non-trivial function f is in L n

n+p(S

n−1), then there exists λ > 0 and a convex body

K∈ Kn

0 withV (K)= 1 such that

λϕ(h) det(∇2h + hI) = f

holdsforh= hK intheAlexandrovsense;namely,λϕ(hK)dSK= f dHn−1.Inaddition,

iff isinvariantunderaclosedsubgroupG ofO(n),thenK canbechosentobeinvariant

under G.

Wenote thattheoriginmaylieon∂K forthesolutionK inTheorem 1.2.

We observe that Theorem 1.2 readily yields Theorem 1.1. Indeed if −n < p < 0,

f ∈ L n

n+p(S

n−1), f ≥ 0,f ≡ 0 andλh1−p

K dSK = f dHn−1 for K∈ Kn0 and λ> 0,then

h1−p

K dSK= f dHn−1 forK = λ

1

n−pK.

In Section3 we sketchthe proof of Theorem 1.2 and describe the structure of the paper.

2. Notation

ThescalarproductonRn isdenotedby ·,· ,andthecorrespondingEuclideannorm isdenotedby· .Thek-dimensionalHausdorffmeasurenormalizedinsuchawaythat it coincides with theLebesgue measure on Rk is denotedby Hk. The angle (spherical distance) ofu,v∈ Sn−1 isdenotedby∠(u,v).

We write Kn

0 (K(0)n ) to denote thefamily of convex bodies with o ∈ K (o∈ int K).

Given aconvex bodyK, fora Borelset ω ⊂ Sn−1,νK−1(ω) is theBorel set of x∈ ∂K withνK(x)∩ω = ∅.Apointx∈ ∂K iscalledsmoothifνK(x) consistsofauniquevector, and in this case, we use νK(x) to denote this unique vector, as well. It is well-known thatHn−1-almosteveryx∈ ∂K issmooth(see,e.g.,Schneider[38]);let∂K denotethe familyofsmoothpointsof ∂K.

Foraconvexcompactset K inRn, lethK be itssupportfunction: hK(u) = max{ x, u : x ∈ K} for u ∈ Rn.

Note that if K ∈ Kn0, then hK ≥ 0. If p ∈ R and K ∈ Kn0, then the Lp-surface area

measure isdefinedby

(5)

where forp> 1 the right-handside isassumed to be afinite measure.Inparticular, if p= 1,thenSK,p= SK,andifp< 1 andω⊂ Sn−1 Borel,then

SK,p(ω) =  ν−1K(ω)

x, νK(x) 1−pdHn−1(x).

3. Sketchofthe proofofTheorem 1.2

To sketch the argument leading to Theorem 1.2, first we consider the case when

−n< p≤ −(n− 1) and ϕ(t)= t1−p, andτ

1≤ f ≤ τ2 for someconstants τ2 > τ1> 0.

Wesetψ(t)= 1/ϕ(t)= tp−1 fort> 0,anddefine Ψ: (0,∞)→ (0,∞) by

Ψ(t) =  t ψ(s) ds =−1 pt p,

whichisastrictly convexfunction. GivenaconvexbodyK inRn,weset

Φ(K, ξ) =  Sn−1

Ψ(hK−ξ)f dHn−1;

this is astrictly convex functionof ξ∈ int K. As f > τ1 and p≤ −(n− 1),there is a

(unique)ξ(K)∈ int K suchthat

Φ(K, ξ(K)) = min

ξ∈int KΦ(K, ξ).

ThisstatementisprovedinProposition5.2,buttheconditionsf > τ1andp≤ −(n− 1)

areactuallyusedinthepreparatorystatementLemma5.1.

Usingp>−n andtheBlaschke-Santalóinequality(seeLemma5.4and the prepara-torystatementLemma4.3),oneverifiesthatthereexists aconvexbodyK0 inRn with

V (K0)= 1 maximizingΦ(K,ξ(K)) overallconvexbodiesK inRn withV (K)= 1.

Finallyavariationalargument proves thatthere exists λ0> 0 such thatf dHn−1 =

λ0ϕ(hK0)dSK0. A crucial ingredient (see Lemma6.2)is that,as ψ is C

1 and ψ < 0, Φ(Kt,ξ(Kt)) isadifferentiablefunctionofKtforasuitablevariationKtofK0.

Inthegeneralcase,whenstill keepingtheconditionτ1≤ f ≤ τ2 butallowingany ϕ

whichsatisfiestheassumptionsofTheorem1.2,wemeettwomainobstacles.Ontheone hand,evenifϕ(t)= t1−p but0< t<−(n− 1),it mayhappen thatfor aconvex body

K inRn,theinfimumofΦ(K,ξ) forξ∈ int K isattainedwhenξ tendstotheboundary ofK.Ontheotherhand,thepossiblelackofdifferentiabilityofϕ (orequivalentlyofψ) destroysthevariationalargument.

(6)

Therefore, we approximateψ bysmooth functions, andalso make surethatthe ap-proximating functions are large enough near zero to ensure that the minimum of the analogues ofΦ(K,ξ) as afunctionofξ∈ int K existsforany convexbodyK.

Section4provessomepreparatorystatements,Section5introducesthesuitable ana-logueoftheenergyfunctionΦ(K,ξ(K)),andSection6providesthevariationalformula foranextremalbodyfortheenergyfunction.WeproveTheorem1.2iff isboundedand bounded awayfrom zeroinSection7,andfinally infullstrength inSection8.

4. Somepreliminaryestimates

Inthissection,weprovethesimplebuttechnicalestimatesLemmas4.1and4.3that will beusedinvarioussettings duringthemainargument.

Lemma 4.1. For δ ∈ (0,1), A,ℵ > 0 and˜ q ∈ (−n,0), let ψ : (0, ∞) → (0,∞) satisfy that ψ(t) ≤ ˜ℵtq−1 fort∈ (0,δ] and

δ ψ≤ A.If t∈ (0,δ) and ˜0= max{ ˜ |q|,δAq},then  Ψ(t)=t∞ψ satisfies  Ψ(t)≤ ˜ℵ0tq.

Proof. We observethatift∈ (0,δ),then

 Ψ(t)≤ δ  t ˜ ψ(s) ds + A≤ ˜ℵ δ  t sq−1ds + A = ˜ |q|(tq− δq) + A≤ tqmax  ˜ |q|, A δq . 2

We write Bn to denote theEuclidean unit ballin Rn, and set κ

n =Hn(Bn).For a convex bodyK inRn,letσ(K) denoteitscentroid,whichsatisfies (seeSchneider[38])

−(K − σ(K)) ⊂ n(K − σ(K)). (5)

Next,ifo∈ int K thenthepolarofK is

K ={x ∈ Rn : x, y ≤ 1 ∀y ∈ K} = {tu : u ∈ Sn−1 and 0≤ t ≤ hK(u)−1}.

In particular, the Blaschke-Santaló inequality V (K)V ((K − σ(K))∗) ≤ V (Bn)2 (see Schneider [38])yieldsthat

 Sn−1

h−nK−σ(K)dHn−1≤ nV (B n)2

V (K) . (6)

As apreparationfortheproofofLemma4.3,weneedthefollowingstatementabout absolutely continuous measures.Fort ∈ (0,1) andv ∈ Sn−1, we considerthe spherical strip

(7)

Ξ(v, t) ={u ∈ Sn−1 : | u, v | ≤ t}. Lemma4.2. If f ∈ L1(Sn−1) and f(t) = sup v∈Sn−1  Ξ(v,t) |f| dHn−1 fort∈ (0,1),thenwehave limt→0+ f(t)= 0.

Proof. We observe that f(t) is decreasing, therefore the limit limt→0+ f(t) = δ ≥ 0

exists.Wesuppose thatδ > 0,andseekacontradiction.

Letμ betheabsolutelycontinuousmeasure= |f|dHn−1onSn−1.Accordingtothe definitionof f,foranyk≥ 2,thereexistssomevk ∈ Sn−1suchthatμ(Ξ(vk,k1))≥ δ/2. Let v ∈ Sn−1 be an accumulation point of the sequence {vk}. For any m ≥ 2, there exists αm > 0 such thatΞ(u,2m1 ) ⊂ Ξ(v,m1) if u∈ Sn−1 and ∠(u,v) ≤ αm. Since for any m≥ 2, there exists somek≥ 2m such that∠(vk,v)≤ αm, we haveμ(Ξ(v,m1))

μ(Ξ(vk,1k)) ≥ δ/2. We deduce that μ(v⊥ ∩ Sn−1) = μ  ∩m≥2Ξ(v,m1)  ≥ δ/2, which contradictsμ(v⊥∩ Sn−1)= 0. 2

Lemma4.3. Forδ∈ (0,1),ℵ > 0 and˜ q∈ (−n,0),letΨ : (0, ∞)→ (0,∞) beamonotone decreasingcontinuous functionsuchthat Ψ(t) ≤ ˜ℵtq fort∈ (0,δ] andlim

t→∞Ψ(t) = 0,

and let f be˜ a non-negative function in L n

n+p(S

n−1). Then for any ζ > 0, there exists

a depending on ζ, Ψ, δ, ℵ,˜ q and f such˜ that if K is a convex body in Rn with

V (K)= 1 anddiam K≥ Dζ then

 Sn−1

( Ψ◦ hK−σ(K)) ˜f dHn−1 ≤ ζ.

Proof. We mayassume that σ(K)= o. Let R = maxx∈Kx, and let v ∈ Sn−1 such thatRv∈ K.Itfollowsfrom (5) that −R

nv∈ K.

Sincelimt→∞Ψ(t) = 0 andf is˜ inL1(Sn−1) bytheHölderinequality,wecanchoose

r≥ 1 suchthat  Ψ(r)  Sn−1 ˜ f dHn−1< ζ 2. (7)

WepartitionSn−1 intothetwomeasurable parts

Ξ0={u ∈ Sn−1: hK(u)≥ r}

(8)

Letus estimatetheintegralsoverΞ0 andΞ1.Wededucefrom(7) that  Ξ0 ( Ψ◦ hK) ˜f dHn−1≤ ζ 2. (8)

Nextweclaimthat

Ξ1⊂ Ξ v,nr R . (9)

Forany u∈ Ξ1,wechooseη∈ {−1,1} suchthat u,ηv ≥ 0,thus ηRn v ∈ K yieldsthat

r > hK(u)≥ u,ηRn v .Inturn,weconclude(9).Itfollowsfrom(9) andLemma4.2that fortheL1functionf = ˜f

n n+q,wehave  Ξ1 ˜ fn+qn f nr R . (10)

To estimatethedecreasingfunctionΨ on (0,r), weclaimthatift∈ (0,r) then 

Ψ(t)≤ ℵδ˜ q

rq t

q. (11)

Werecallthatr≥ 1> δ.Inparticular,ift≤ δ,thenΨ(t) ≤ ˜ℵtqyields(11).Ift∈ (δ,r), then usingthatΨ is decreasing,(11) followsfrom

 Ψ(t)≤ Ψ(δ)≤ ˜ℵδ q tq t q ℵδ˜ q rq t q.

Applying first (11), then the Hölder inequality, after that the Blaschke-Santaló in-equality (6) withV (K)= 1 andfinally (10),wededuce that

 Ξ1 ( Ψ◦ hK) ˜f dHn−1 ˜ ℵδq rq  Ξ1 h−|q|K f d˜ Hn−1 ≤ℵδ˜ q rq ⎛ ⎝ Ξ1 h−nK dHn−1 ⎞ ⎠ |q| n ⎛ ⎝ Ξ1 ˜ fn−|q|n dHn−1 ⎞ ⎠ n−|q| n ≤ℵδ˜rqq  nV (Bn)2 |q| n f nr R n+q n .

Therefore afterfixingr≥ 1 satisfying(7),wemaychooseR0> r suchthat

˜ ℵδq rq n |q| nV (Bn) 2|q| n f  nr R0 n+q n 2

(9)

byLemma4.3. Inparticular, ifR≥ R0,then  Ξ1 ( Ψ◦ hK) ˜f dHn−1 ζ 2.

Combining this estimatewith (8) shows that setting = 2R0, ifdiam K ≥ Dζ, then R≥ R0, andhence



Sn−1( Ψ◦ hK)f d˜ Hn−1 ≤ ζ. 2

5. Theextremalproblemrelatedto Theorem1.2whenf is boundedandbounded awayfromzero

For0< τ1< τ2,letthereal functionf onSn−1 satisfy

τ1< f (u) < τ2 for u∈ Sn−1. (12)

Inaddition,letϕ: [0,∞)→ [0,∞) be acontinuousmonotone increasingfunction satis-fyingϕ(0)= 0, lim inf t→0+ ϕ(t) t1−p > 0 and  1 1 ϕ(t)dt <∞.

It will be more convenient to work with the decreasingfunction ψ = 1/ϕ : (0,∞) (0,∞),whichhastheproperties

lim sup t→0+ ψ(t) tp−1, <∞ (13)  1 ψ(t) dt <∞. (14)

WeconsiderthefunctionΨ: (0,∞)→ (0,∞) definedby

Ψ(t) = 

t

ψ(s) ds,

whichreadilysatisfies

Ψ=−ψ, and hence Ψ is convex and strictly monotone decreasing, (15)

lim

t→∞Ψ(t) = 0. (16)

(10)

ψ(t) <ℵtp−1 for t∈ (0, δ). (17) AswepointedoutinSection3,wesmoothenψ usingconvolution.Letη : R→ [0,∞) be anon-negative C∞ “approximationof identity”withsupp η⊂ [−1,0] andRη = 1. Foranyε∈ (0,1),weconsiderthenon-negativeηε(t)=1εη(tε) satisfyingthat



Rηε= 1, supp ηε⊂ [−ε,0],anddefine θε: (0,∞)→ (0,∞) by

θε(t) =  R ψ(t− τ)ηε(τ ) dτ = 0  −ε ψ(t− τ)ηε(τ ) dτ.

As ψ ismonotonedecreasingandcontinuouson(0,∞),thepropertiesofηεyield

θε(t) ≤ ψ(t) for t > 0 and ε ∈ (0, 1)

θε(t1) ≥ θε(t2) for t2> t1> 0 and ε∈ (0, 1)

θεtends uniformly to ψ on any interval with positive endpoints as ε tends to zero. Next,forany t0> 0,thefunctionlt0 onR defined by

lt0(t) =



ψ(t) if t≥ t0

0 if t < t0

is bounded, and hence locally integrable. For the convolution lt0 ∗ ηε, we have that

(lt0∗ ηε)(t)= θε(t) fort> t0 andε∈ (0,1),thus

θε is C1 for each ε∈ (0, 1).

As it is explained in Section 3, we need to modify ψ in a way such that the new functionisoforder atleastt−(n−1) ift> 0 is small.Weset

q = min{p, −(n − 1)},

and hence(17) andδ∈ (0,1) yieldsthat

θε(t)≤ ψ(t) < ℵtq−1 for t∈ (0, δ) and ε ∈ (0, δ). (18)

Nextweconstructθ˜ε: (0,∞)→ (0,∞) satisfying ˜ θε(t) = θε(t)≤ ψ(t) for t ≥ ε and ε ∈ (0, δ) ˜ θε(t) ℵtq−1 for t∈ (0, δ) and ε ∈ (0, δ) ˜ θε(t) = ℵtq−1 for t∈ (0, ε 2] and ε∈ (0, δ) ˜

(11)

Itfollowsthat ˜

θεtends uniformly to ψ on any interval with positive endpoints as ε tends to zero. Toconstructsuitableθ˜ε,firstweobservethattheconditionsabovedetermineθ˜εoutside theinterval (ε

2,ε), and θ˜ε(ε)<ℵεq−1. WritingΔ to denote the degreeonepolynomial

whosegraphisthetangentto thegraphoft→ ℵtq−1 att= ε/2,wehaveΔ(t)<ℵtq−1 fort> ε/2 andΔ(ε)< 0.Thereforewecanchooset0∈ (ε2,ε) suchthatθ˜ε(ε)< Δ(t0)<

ℵεq−1. Wedefine θ˜ε(t) = Δ(t) fort ∈ (ε

2,t0),and construct θ˜ε on(t0,ε) inaway that

˜

θεstaysC1 on(0,∞).Itfollows fromtheway θ˜εisconstructedthatθ˜ε(t)≤ ℵtq−1 also fort∈ [ε

2,ε].

Inordertoensureanegative derivative,weconsiderψε: (0,∞)→ (0,∞) definedby ψε(t) = ˜θε(t) +

ε

1 + t2 (19)

forε∈ (0,δ) andt> 0.This C1 functionψ

εhasthefollowingproperties:

ψε(t) ≤ ψ(t) +1+t12 for t≥ ε and ε ∈ (0, δ)

ψε(t) < 0 for t > 0 and ε∈ (0, δ) ψε(t) < 2ℵtq−1 for t∈ (0, δ) and ε ∈ (0, δ) ψε(t) > ℵtq−1 for t∈ (0,ε2) and ε∈ (0, δ)

ψεtends uniformly to ψ on any interval with positive endpoints as ε tends to zero. (20)

Forε∈ (0,δ),wealsoconsider theC2functionΨε: (0,∞)→ (0,∞) definedby

Ψε(t) = 

t

ψε(s) ds,

andhence(20) yields lim

t→∞Ψε(t) = 0 (21)

Ψε=−ψε, thus Ψεis strictly decreasing and strictly convex. (22)

Forε∈ (0,δ),Lemma4.1and(20) implythatsetting

A =  δ ψ(t) + 1 1 + t2dt, wehave

(12)

Ψε(t)≤ ℵ0tq for0= max{

2 |q|,

A

δq} and t ∈ (0, δ). (23)

Ontheother hand,ifε∈ (0,δ) and t∈ (0,ε

4),then Ψε(t)≥ ε/2  t ℵsq−1ds = |q|(tq− (ε/2)q)≥|q|ℵ (tq− (2t)q) =1tq for1= (1− 2q) |q| > 0. (24)

According to (20), we have limε→0+ψε(t) = ψ(t) and ψε(t) ≤ ψ(t)+ 1

1+t2 for any

t> 0,thereforeLebesgue’sDominatedConvergenceTheoremimplies

lim

ε→0ε(t) = Ψ(t) for any t > 0. (25)

It alsofollowsfrom (20) that ift≥ ε,then

Ψε(t) =  t ψε≤  t ψ(s) + 1 1 + s2ds≤ Ψ(t) + π 2. (26)

Forany convexbodyK andξ∈ int K,weconsider Φε(K, ξ) =  Sn−1ε◦ hK−ξ)f dHn−1=  Sn−1

Ψε(hK(u)− ξ, u )f(u) dHn−1(u).

Naturally,Φε(K) dependsonψ andf ,aswell,butwedonotsignalthesedependences. Weequip Kn

0 withtheHausdorffmetric,whichistheL∞ metriconthespaceof the

restrictions ofsupportfunctionstoSn−1.Forv ∈ Sn−1 and α∈ [0,π2],we consider the sphericalcap

Ω(v, α) ={u ∈ Sn−1 u, v ≥ cos α}. Wewrite π : Rn\{o}→ Sn−1 theradialprojection:

π(x) = x

x.

Inparticular,ifπ isrestrictedtotheboundaryofaK∈ Kn

(0),thenthismapisLipschitz.

Another typical application of the radial projection is to consider, for v ∈ Sn−1, the compositionx→ π(x+ v) asamapv⊥→ Sn−1 where

(13)

ThefollowingLemma5.1isthestatementwhereweapplydirectlythatψ ismodified tobe essentiallytqift is verysmall.

Lemma 5.1. Let ε ∈ (0,δ), and let {Ki} be a sequence of convex bodies tending to a

convexbody K inRn,andletξi∈ int Ki suchthat limi→∞ξi= x0∈ ∂K. Then

lim

i→∞Φε(Ki, ξi) =∞.

Proof. Wemayassumethatlimi→∞ξi= x0= o.Letv∈ Sn−1 beanexteriornormalto

∂K at o, andchoosesomeR > 0 such thatK ⊂ RBn. Thereforewe mayassumethat

Ki− ξi ⊂ (R + 1)Bn,hKi(v)< ε/8 andξi< ε/8 forall ξi, thushKi−ξi(v)< ε/4 for

alli.

Foranyζ∈ (0,ε8),thereexists suchthatifi≥ Iζ,thenξi≤ ζ/2 and y,v ≤ ζ/2 for all y ∈ Ki, and hence y,v ≤ ζ for all y ∈ Ki− ξi. For i ≥ Iζ, any y ∈ Ki− ξi canbe written inthe form y = sv + z where s ≤ ζ andz ∈ v⊥∩ (R + 1)Bn, thus if ∠(v,u)= α∈ [ζ,π2) foru∈ Sn−1,then wehave

hKi−ξi(u)≤ (R + 1) sin α + ζ cos α ≤ (R + 2)α. (28)

Wesetβ = ε

4(R+2),andforζ∈ (0,β),wedefine

Ωζ = Ω(v, β)\Ω(v, ζ).

Inparticular,asΨε(t)≥ ℵ1tqfort∈ (0,ε4) accordingto(24),ifu∈ Ωζ,then(28) implies Ψε(hKi−ξi(u))≥ γ(∠(v, u))

q

forγ =ℵ1(R + 2)q.

Thefunctionx→ π(x+ v) maps = v⊥∩ 

(tan β)Bn\(tan ζ)Bnbijectivelyonto Ωζ,whileβ < 18 and(27) yieldthattheJacobianofthismap isatleast2−non.

Sincef > τ1 and∠(v,π(x+ v))≤ 2x forx∈ Bζ,ifi≥ Iζ,then

Φε(Ki, ξi) =  Sn−1 Ψε(hKi−ξi(u))f (u) dH n−1(u) Ωζ τ1γ(∠(v, u))qdHn−1(u) τ1γ 2n+|q|  xqdHn−1(x) = (n− 1)κn−1τ1γ 2n+|q| tan β tan ζ tq+n−2dt.

Asζ > 0 isarbitrarilysmallandq≤ 1−n,weconcludethatlimi→∞Φε(Ki,ξi)=∞. 2 Nowwesingleouttheoptimalξ∈ int K.

(14)

Proposition5.2. Forε∈ (0,δ) andaconvexbodyK inRn,thereexistsauniqueξ(K)∈ int K such that

Φε(K, ξ(K)) = min

ξ∈int KΦε(K, ξ).

In addition, ξ(K) and Φε(K,ξ(K)) are continuous functionsof K, and Φε(K,ξ(K)) is

translation invariant.

Proof. The firstpartofthis proof,theoneregardingtheexistenceofξ(K)∈ int K and its uniqueness, isvery similar to the proofof [1, Proposition3.2] given by theauthors and Yang for the Lp Minkowski problem. It is very short and we rewrite it here for completeness.

Let ξ12 ∈ int K,ξ1= ξ2,and letλ∈ (0,1).Ifu∈ Sn−1\(ξ1− ξ2),then u,ξ1 =

u,ξ2 ,andhencethestrictconvexityofΨε (see(22))yieldsthat

Ψε(hK(u)− u, λξ1+ (1− λ)ξ2 ) > λΨε(hK(u)− u, ξ1 ) + (1 − λ)Ψε(hK(u)− u, ξ2 ),

thus Φε(K,ξ) isastrictlyconvexfunctionofξ∈ int K byf > τ1.

Letξi∈ int K suchthat

lim

i→∞Φε(K, ξi) =ξ∈int Kinf Φε(K, ξ).

We may assume that limi→∞ξi = x0 ∈ K, and Lemma 5.1 yields x0 ∈ int K. Since

Φε(K,ξ) is a strictly convex and continuous function of ξ ∈ int K, x0 is the unique

minimumpointofξ→ Φε(K,ξ),whichwedenotebyξ(K) (notsignallingthedependence

onε,ψ andf ).

Readilyξ(K) istranslationequivariant,andΦε(K,ξ(K)) istranslationinvariant. Forthecontinuityofξ(K) andΦε(K,ξ(K)),letusconsiderasequence{Ki} ofconvex bodiestendingtoaconvexbodyK inRn.Wemayassumethatξ(K

i) tendstoax0∈ K.

For any y ∈ int K,there exists anI such thaty ∈ int Ki fori ≥ I.SincehKi tends

uniformly tohK onSn−1,wehavethat lim sup

i→∞ Φε(Ki, ξ(Ki))≤ limi→∞i≥I Φε(Ki, y) = Φε(K, y).

AgainLemma5.1impliesthatx0∈ int K.ItfollowsthathKi−ξi(Ki) tendsuniformlyto

hK−x0,thus

Φε(K, x0) = lim

i→∞Φε(Ki, ξ(Ki))≤ limi→∞ i≥I

Φε(Ki, y) = Φε(K, y).

In particular, Φε(K,x0) ≤ Φε(K,y) for any y ∈ int K, thus x0 = ξ(K). In turn, we

(15)

Sinceξ→ Φε(K,ξ)= 

Sn−1Ψε(hK(u)− u,ξ )f(u)dH

n−1(u) ismaximalat ξ(K) int K andΨ=−ψε,wededuce

Corollary5.3. Forε∈ (0,δ) and aconvexbodyK in Rn,wehave  Sn−1 u ψε hK(u)− u, ξ(K) f (u) dHn−1(u) = o.

For aclosed subgroup G ofO(n), we write Kn,G(0) to denote the family of K ∈ Kn

(0)

invariantunderG.

Lemma5.4. Forε∈ (0,δ),there existsaKε∈ Kn(0) with V (Kε)= 1 such that Φε(Kε, ξ(Kε))≥ Φε(K, ξ(K)) for any K∈ Kn(0) with V (K) = 1.

Inaddition,iff isinvariantunderaclosedsubgroupG ofO(n),thenKε canbechosen

tobeinvariantunder G.

Proof. WechooseasequenceKi∈ K(0)n withV (Ki)= 1 fori≥ 1 suchthat lim

i→∞Φ(Ki, ξ(Ki)) = sup{Φ(K, ξ(K)) : K ∈ K

n

(0) with V (K) = 1}.

WritingB1= κ−1/nn Bn todenotetheunitballcentredattheoriginandhavingvolume 1,we mayassumethateachKi satisfies

Φε(Ki, σ(Ki))≥ Φε(Ki, ξ(Ki))≥ Φε(B1, ξ(B1)). (29)

AccordingtoProposition5.2,wemayalsoassumethatσKi= o for eachKi.

Wededuce from Lemma4.3, (21), (23) and(29) that there exists someR > 0 such that Ki ⊂ RBn for any i ≥ 1. According to the Blaschke selection theorem, we may assume that Ki tends to a compact convex set with o ∈ Kε. It follows from the continuity of the volume that V (Kε) = 1, and hence int Kε = ∅. We conclude from Lemma5.2 thatΦε(Kε,ξ(Kε))= limi→∞Φε(Ki,ξ(Ki)).

Iff isinvariantunderaclosedsubgroupG ofO(n),thenweapplythesameargument toconvexbodiesinKn,G(0) insteadofKn(0). 2

SinceΦ(5)< Φ(4),(25) yields some˜δ∈ (0,δ) such thatΨε(4)≥ Φ(5) forε∈ (0,δ).˜ For futurereference, the monotonicityof Ψε, diamκ−1/nn Bn ≤ 4 and(29) yield thatif ε∈ (0,δ),˜ then Φε(Kε, σ(Kε))≥ Φε  κ−1/nn Bn, ξ(κ−1/nn Bn)  Sn−1 Ψε(4)f dHn−1≥ Ψ(5)  Sn−1 f dHn−1. (30)

(16)

6. Variationalformulae andsmoothnessoftheextremalbodywhenf isboundedand boundedawayfromzero

In this section, again let 0 < τ1 < τ2 and let the real function f on Sn−1 satisfy

τ1 < f < τ2. Inaddition, letϕ bethecontinuous functionof Theorem1.2,and weuse

thenotationdevelopedinSection5, sayψ : (0,∞)→ (0,∞) isdefinedbyψ = 1/ϕ. Now thatwe have constructed an extremal body , we want to show thatit sat-isfies therequired differential equation inthe Alexandrov sense by using a variational argument.Thissectionprovidestheformulaethatwewillneed,andensuretherequired smoothnessof.

Concerningthevariationofvolume,akeytoolisAlexandrov’sLemma6.1(seeLemma 7.5.3in[38]).Tostatethis,foranycontinuoush: Sn−1→ (0,∞),wedefinethe Alexan-drov body

[h] ={x ∈ Rn: x, u ≤ h(u) for u ∈ Sn−1}

whichisaconvexbodycontainingtheorigininitsinterior.Obviously,ifK∈ Kn

(0) then

K = [hK].

Lemma 6.1 (Alexandrov). For K ∈ Kn

(0) and a continuous function g : Sn−1 → R,

K(t)= [hK+ tg] satisfies lim t→0 V (K(t))− V (K) t =  Sn−1 g(u) dSK(u).

TohandlethevariationofΦε(K(t),ξ(K(t))) forafamilyK(t) isamoresubtle prob-lem. The next lemma shows essentially that ifwe perturba convex body K in a way such that the support function is differentiable as a function of the parameter t for Hn−1-almostallu∈ Sn−1,thenξ(K) changesalsoinadifferentiableway.Lemma6.2is thepoint oftheproof whereweusethatψε isC1andψε < 0.

Lemma6.2. Forε∈ (0,δ),letc> 0 andt0> 0,andletK(t) beafamilyofconvexbodies

with supportfunctionhtfort∈ [0,t0). Assumethat

(i) |ht(u)− h0(u)|≤ ct foreach u∈ Sn−1 andt∈ [0,t0),

(ii) limt→0+ht(u)−h0(u)

t existsforHn−1-almostallu∈ Sn−1.

Then limt→0+ξ(K(t))−ξ(K(0))

t exists.

Proof. We set K = K(0). We may assume thatξ(K) = o, and hence Proposition5.2 yields that

lim

(17)

ThereexistssomeR > r > 0 suchthatr≤ ht(u)− u,ξ(K(t)) = hK(t)−ξ(K(t))(u)≤ R foru∈ Sn−1 andt∈ [0,t0).SinceψεisC1 on[r,R],wecanwrite

ψε(t)− ψε(s) = ψε(s)(t− s) + η0(s, t)(t− s)

for t,s ∈ [r,R] where limt→sη0(s,t) = 0. Let g(t,u) = ht(u)− hK(u) for u ∈ Sn−1

and t ∈ [0,t0). Since hK(t)−ξ(K(t)) tends uniformly to hK on Sn−1, we deduce that if

t∈ [0,t0),then ψε ht(u)− u, ξ(K(t)) − ψε(hK(u)) = ψε(hK(u))  g(t, u)− u, ξ(K(t)) + e(t, u) (31) where

|e(t, u)| ≤ η(t)|g(t, u) − u, ξ(Kt) | and η(t) = η0(hK(u), ht(u)− u, ξ(K(t)) ). Notethatlimt→0+η(t)= 0 uniformlyinu∈ Sn−1.

Inparticular,(i)yieldsthate(t,u)= e1(t,u)+ e2(t,u) where

|e1(t, u)| ≤ cη(t)t and |e2(t, u)| ≤ η(t)ξ(K(t)). (32)

Itfollowsfrom (31) andfromapplyingCorollary 5.3toK(t) andK that 

Sn−1

ε(hK(u)) 

g(t, u)− u, ξ(K(t)) + e(t, u) f (u)dHn−1(u) = o,

whichcanbe writtenas 

Sn−1

u ψε(hK(u)) g(t, u) f (u)dHn−1(u)

+  Sn−1 u e1(t, u) f (u)dHn−1(u) =  Sn−1

u u, ξ(Kt) ψε(hK(u)) f (u)dHn−1(u)

 Sn−1

u e2(t, u) f (u)dHn−1(u).

Sinceψε(s)< 0 foralls> 0,thesymmetricmatrix A =

 Sn−1

u⊗ u ψε(hK(u)) f (u)dHn−1(u)

(18)

vTAv =  Sn−1

u, v 2ψ

ε(hK(u)) f (u) dHn−1(u) < 0.

Inaddition, A satisfies  Sn−1

u u, ξ(Kt) ψε(hK(u)) f (u)dHn−1(u) = A ξ(Kt).

It followsfrom(32) that ift≥ 0 issmall, then

A−1  Sn−1

u ψε(hK(u)) g(t, u) f (u)dHn−1(u) + ˜e1(t) = ξ(Kt)− ˜e2(t), (33)

where ˜e1(t) ≤ α1η(t)t and ˜e2(t) ≤ α2η(t)ξ(Kt) for constants α12 > 0. Since

η(t) tends to 0 with t, if t ≥ 0 is small, then ξ(K(t))− ˜e2(t)≥ 12ξ(Kt). Adding the estimate g(t,u)≤ ct, we deduce thatξ(K(t))≤ β t for aconstantβ > 0, which in turn yields that limt→0+ ˜ei(t)

t = 0 and ˜ei(0) = 0 for i = 1,2. Since there exists limt→0+g(t,u)−g(0,u)

t = ∂1g(0,u) forH

n−1 almostallu∈ Sn−1, and g(t,u)−g(0,u)

t < c for

allu∈ Sn−1 andt> 0,weconcludethat d dtξ(K(t))   t=0+ = A−1  Sn−1

u ψε(hK(u)) ∂1g(0, u) f (u) dHn−1(u). 2

Corollary 6.3.Under theconditions of Lemma6.2,and settingK = K(0), wehave d dtΦε(K(t), ξ(K(t)))   t=0+ =  Sn−1 ∂thK(t)(u)   t=0+ ψε  hK(u)− u, ξ(K)  f (u) dHn−1(u).

We omit theproof of thisresult sinceit isvery similar to thatof [1, Corollary 3.6], givenbytheauthorsandYangfortheLp Minkowskiproblem,withf (u)dHn−1(u),Ψε, −ψε,Lemma6.2andCorollary5.3replacingrespectivelydμ(u),ϕε,ϕε,Lemma3.5and Corollary 3.3.

Given afamilyK(t) of convex bodiesfor t∈ [0,t0), t0 > 0, to handle thevariation

ofΦε(K(t),ξ(K(t))) atK(0)= K viaapplyingCorollary6.3,weneedtheproperty(see Lemma6.2)thatthereexistsc> 0 suchthat

|hK(t)(u)− hK(u)| ≤ c |t| for any u ∈ Sn−1 and t∈ [0, t0) (34)

lim t→0+

hK(t)(u)− hK(u)

t exists forH

(19)

However,evenifK(t)= [hK+ thC] forK,C∈ Kn(0)andfort∈ (−t0,t0),K mustsatisfy

somesmoothness assumption inorder to ensurethat (35) holds also for thetwo sided limits(problemsoccursayifK isapolytopeandC issmooth).

Werecallthat∂K denotes thesetofsmoothpointsof ∂K.WesaythatK is

quasi-smooth if Hn−1(Sn−1\νK(∂K)) = 0; namely, the set of u ∈ Sn−1 that are exterior

normalsonlyatsingularpointshasHn−1-measurezero.ThefollowingLemma6.4,taken fromBianchi,Böröczky,Colesanti,Yang[1],showsthat(34) and(35) aresatisfiedevenif t∈ (−t0,t0) atleastforK(t)= [hK+ thC] witharbitraryC∈ Kn(0)ifK isquasi-smooth. Lemma 6.4.Let K,C ∈ Kn

(0) be suchthat rC⊂ K for some r > 0.Fort ∈ (−r,r) and

K(t)= [hK+ thC],

(i) if K ⊂ R C for R > 0, then |hK(t)(u)− hK(u)|≤ Rr|t| for any u∈ Sn−1 and

t∈ (−r,r);

(ii) ifu∈ Sn−1 istheexteriornormal atsomesmoothpointz∈ ∂K, then

lim t→0

hK(t)(u)− hK(u)

t = hC(u).

We will need the condition (35) in the following rather special setting taken from Bianchi,Böröczky,Colesanti,Yang[1].

Lemma6.5.LetK beaconvexbodywithrBn⊂ int K forr > 0, letω⊂ Sn−1 beclosed, andif t∈ [0,r),thenlet

K(t) = [hK− 1ω] ={x ∈ K : x, u ≤ hK(u)− t for every u ∈ ω}.

(i) Wehave limt→0+hK(t)(u)−hK(u)

t existsandisnon-positiveforallu∈ S

n−1,and ifu∈ ω,theneven limt→0+

hK(t)(u)−hK(u)

t ≤ −1.

(ii) IfSK(ω)= 0,then limt→0+V (K(t))−V (K)

t = 0.

Proposition6.6. Forε∈ (0,δ),Kε isquasi-smooth.

Proof. We suppose that is not quasi-smooth, and seek a contradiction. It follows thatHn−1(X)> 0 forX = Sn−1\νKε(∂Kε),thereforethereexistsaclosedω⊂ X such

thatHn−1(ω)> 0.SinceνK−1ε(ω)⊂ ∂Kε\∂Kε,wededucethatSKε(ω)= 0.

We may assume that ξ(Kε) = o and rBn ⊂ Kε ⊂ RBn for R > r > 0. As in Lemma6.5,ift∈ [0,r),thenwedefine

K(t) = [hKε− 1ω] ={x ∈ Kε: x, u ≤ hK(u)− t for every u ∈ ω}.

Clearly, K(0) equals . We define α(t) = V (K(t))−1/n, and hence α(0) = 1, and Lemma6.5 (ii)yieldsthatα(0)= 0.

(20)

Weset K(t) = α(t)K(t),andhenceK(0) = Kεand V ( K(t))= 1 for allt∈ [0,r). In addition,weconsiderh(t,u)= hK(t)(u) andh(t,˜ u)= hK(t)(u)= α(t)h(t,u) foru∈ S

n−1 andt∈ [0,r).Since[hKε−thBn]⊂ K(t),Lemma6.4(i)yieldsthat|h(t,u)−h(0,u)|≤ R

r t foru∈ Sn−1 andt∈ [0,r).Henceα(0)= 0 impliesthatthereexistc> 0 andt

0∈ (0,r)

such that |˜h(t,u)− ˜h(0,u)| ≤ ct for u ∈ Sn−1 and t ∈ [0,t0). Applying α(0) = 1,

α(0)= 0 andLemma6.5(i),wededucethat

h(0, u) = lim t→0+ ˜ h(t, u)− ˜h(0, u) t = limt→0+ h(t, u)− h(0, u)

t ≤ 0 exists for all u ∈ S

n−1, h(0, u)≤ −1 for all u ∈ ω.

As ψε is positive and monotone decreasing, f > τ1 and Hn−1(ω) > 0, Corollary 6.3

impliesthat d dtΦε( K(t), ξ( K(t)))   t=0+ =  Sn−1

1h(0, u)˜ · ψε(hK(u)) f (u) dHn−1(u)

≥ −

ω

(−1)ψε(R)τ1dHn−1(u) > 0.

Therefore Φε( K(t),ξ( K(t))) > Φε(Kε,ξ(Kε)) for small t > 0. This contradicts the definitionofand concludestheproof. 2

Forε∈ (0,δ),wedefine λε= 1 n  Sn−1 hKε−ξ(Kε)· ψε(hKε−ξ(Kε))· f dHn−1. (36)

Proposition6.7. Forε∈ (0,δ),ψε(hKε−ξ(Kε))·f dHn−1= λεdSKε asmeasuresonSn−1.

We omit the proof of this result since it is very similar to that of [1, Proposition 6.1], given by the authors and Yang for the Lp Minkowski problem, with −λε, −ψε, Lemma6.1,Lemma6.4,Corollary6.3,and[38] replacingrespectivelyλε,ϕε,Lemma 5.2, Lemma2.3,Corollary 3.6and[35].

7. TheproofofTheorem1.2 whenf isboundedandboundedawayfromzero

Inthissection,againlet0< τ1< τ2,lettherealfunctionf onSn−1 satisfyτ1< f <

τ2,andletϕ bethecontinuousfunctionon[0,∞) ofTheorem 1.2.Weusethenotation

developedinSection5,andhenceψ : (0,∞)→ (0,∞) andψ = 1/ϕ.

To ensurethataconvex bodyis“fat” enoughinLemma7.2and later,thefollowing observationisuseful:

(21)

Lemma 7.1.If K is a convex body in Rn with V (K) = 1 and K ⊂ σ(K)+ RBn for R > 0,then

σ(K) + rBn⊂ K for r = 1

n−1n

−3/2R−(n−1).

Proof. Letz0+ r0Bn bealargest ballinK.Accordingto theSteinhagentheorem [12,

Theorem50], thereexistsv∈ Sn−1 suchthat

| x − z0, v | ≤ c√nr0 for x∈ K,

where c is apositiveuniversal constant.It follows that1= V (K) ≤ c√nr0κn−1Rn−1,

thusr0 n−11 n−1/2R−(n−1).Sinceσ(K)+rn0 Bn⊂ K by−(K−σ(K))⊂ n(K−σ(K)),

wemaychooser = 1

cκn−1n

−3/2R−(n−1). 2

Werecall(compare(36))thatifε∈ (0,δ) andξ(Kε)= o,then λ

ε isdefinedby λε= 1 n  Sn−1 hKεψε(hKε)f dHn−1. (37)

Lemma 7.2. There existR0 > 1, r0 > 0 and λ˜2 > ˜λ1 > 0 depending on f,q,ψ,ℵ such

that ifε∈ (0,δ0) forδ0= min{˜δ,r20} whereδ comes˜ from (30),then ˜λ1≤ λε≤ ˜λ2 and

σ(Kε) + r0Bn⊂ Kε⊂ σ(Kε) + R0Bn.

Proof. Accordingto(23),thereexists0> 0 dependingonq,ψ,ℵ suchthatifε∈ (0,δ)

and t∈ (0,δ),then Ψε(t) ≤ ℵ0tq.In addition, limt→∞Ψε(t)= 0 by(21), therefore we mayapply Lemma4.3.Since (30) provides thecondition

 Sn−1 Ψε(hKε−σ(Kε))f dHn−1≥ Ψ(5)  Sn−1 f dHn−1

forany ε∈ (0,δ),˜ wededuce fromLemma4.3 theexistence ofR0> 0 such thatKε⊂

σ(Kε)+ R

0Bn foranyε∈ (0,δ).˜ Inaddition, theexistenceofr0independent ofε such

thatσ(Kε)+ r

0Bn ⊂ Kε followsfromLemma7.1.

Toestimateλε,weassumeξ(Kε)= o.Letwε∈ Sn−1and ε≥ 0 besuchthatσ(Kε)= εwε,andhencer0wε∈ Kε.ItfollowsthathKε(u)≥ r0/2 holdsforu∈ Ω(wε

3),while

Kε⊂ 2R0Bn,R0> 1 and themonotonicityofψε implythatψε(hKε(u))≥ ψε(2R0)=

ψ(2R0) forallu∈ Sn−1.

Wededucefrom(37) that λε= 1 n  Sn−1 hKεψε(hKε)f dHn−1≥ 1 r0 2 · ψ(2R0)· τ1· H n−1Ωw ε, π 3 = ˜λ1.

(22)

To have asuitable upper bound on λε, we define α∈ (0,π2) with cos α = 2Rr00, and hence Ω(−wε, α) =  u∈ Sn−1: u, wε ≤ −r0 2R0 . A key observation is that if u ∈ Sn−1\Ω(−w

ε,α), then u,wε >−2Rr00 and ε ≤ R0

imply hKε(u)≥ u, wε+ r0u ≥ r0−r0 ε 2R0 ≥ r 0/2, therefore ε<r0 2 yields ψε(hKε(u))≤ ψε(r0/2) = ψ(r0/2). (38)

Anotherobservationisthat⊂ 2R

0Bn implies

hKε(u) < 2R0 for any u∈ Sn−1. (39)

It followsdirectlyfrom(38) and(39) that 

Sn−1\Ω(−wε,α)

hKεψε(hKε)f dHn−1≤ (2R0)ψ(r0/2)τ2n. (40)

However, if u ∈ Ω(−wε,α), then ψε(hKε(u)) can be arbitrary large as ξ(Kε) can

be arbitrary closeto ∂Kε ifε > 0 is small, and hence we transferthe problem to the previous caseu∈ Sn−1\Ω(−wε,α) using Corollary 5.3. Firstapplying u,−wε ≥ 2Rr00 foru∈ Ω(−wε,α),thenCorollary5.3,andafterthat u,wε ≤ 1,f ≤ τ2and(38) implies



Ω(−wε,α)

ψε(hKε(u))f (u) dHn−1(u)≤ 2R0

r0



Ω(−wε,α)

u, −wε ψε(hKε(u))f (u) dHn−1(u)

= 2R0 r0

 Sn−1\Ω(−wε,α)

u, wε ψε(hKε(u))f (u) dHn−1(u)

2R0 r0 · ψ r 0 2 τ2nκn. Now (39) yields  Ω(−wε,α) hKψε(hK)f dHn−1≤ (2R0)2 r0 · ψ r0 2 τ2nκn,

which estimate combinedwith (40) leadsto λε< (2R0)2 r0 + 2R0 ψ(r0 22nκn. Inturn, we concludeLemma7.2. 2

(23)

NowweproveTheorem1.2iff isboundedandbounded awayfrom zero.

Theorem7.3.For0< τ1< τ2,lettherealfunctionf onSn−1satisfyτ1< f < τ2,andlet

ϕ: [0,∞)→ [0,∞) beincreasing andcontinuoussatisfyingϕ(0)= 0,lim inft→0+ ϕ(t)

t1−p >

0, and 1ϕ1 < ∞. Let Ψ(t) = tϕ1. Then there exist λ > 0 and a K ∈ Kn

0 with V (K)= 1 suchthat f dHn−1= λϕ(hK) dSK, asmeasureson Sn−1,and  Sn−1 Ψ(hK−σ(K))f dHn−1≥ Ψ(5)  Sn−1 f dHn−1. (41)

Inaddition, iff is invariantunder aclosedsubgroup G of O(n),then K can bechosen

tobeinvariantunder G.

Proof. We assume that ξ(Kε) = o for all ε ∈ (0,δ

0) where δ0 ∈ (0,δ] comes from

Lemma7.2.UsingtheconstantR0ofLemma7.2,wehavethat

Kε⊂ 2R0Bn and hKε(u) < 2R0 for any u∈ Sn−1 and ε∈ (0, δ0). (42)

We consider the continuous increasing function ϕε : [0,∞)] → [0,∞) defined by ϕε(0)= 0 andϕε(t)= 1/ψε(t) forε∈ (0,δ).Weclaimthat

ϕεtends uniformly to ϕ on [0, 2R0] as ε > 0 tends to zero. (43)

Forany smallζ > 0,there existsδζ > 0 suchthatϕ(t)≤ ζ/2 fort∈ [0,δζ]. Wededuce from (20) thatifε> 0 is small,then |ϕε(t)− ϕ(t)|≤ ζ/2 fort∈ [δζ,2R0]. Howeverϕε ismonotone increasing,thereforeϕε(t),ϕ(t)∈ [0,ζ] for t∈ [0,δζ], completingthe proof of(43).

For any ε ∈ (0,δ0), it follows from Lemma 6.7 thatψε(hKε)f dHn−1 = λεdSKε as

measuresonSn−1.Integratinggϕε(hKε) foranycontinuousrealfunctiong onSn−1,we

deducethat

f dHn−1 = λεϕε(hKε) dSKε (44)

asmeasures onSn−1.

Sinceλ˜1≤ λε≤ ˜λ2forsomeλ˜2> ˜λ1 independentofε accordingtoLemma7.2,(42)

yieldsthe existenceofλ> 0,K ∈ Kn

0 withV (K)= 1 and sequence {ε(m)} tendingto

0 suchthatlimm→∞λε(m)= λ andlimm→∞Kε(m)= K.As hKε(m) tends uniformlyto

hK onSn−1, wededuce thatλε(m)ϕε(m)(hKε(m)) tendsuniformly to λϕ(hK) on Sn−1.

(24)

f dHn−1 = λϕ(hK) dSK.

Wenote thatiff isinvariantunderaclosed subgroupG ofO(n),theneachcanbe chosento beinvariantunderG accordingtoLemma5.4,thereforeK isinvariantunder G in thiscase.

To prove(41),ifε∈ (0,δ0),then(30) providesthecondition

 Sn−1 Ψε(hKε−σ(Kε))f dHn−1≥ Ψ(5)  Sn−1 f dHn−1. (45)

Now Lemma7.2 yields thatthere exists r0 > 0 such that ifε ∈ (0,δ0), then σ(Kε)+

r0Bn ⊂ Kε where0< δ0 r20.Inparticular, ifu∈ Sn−1,thenhKε−σ(Kε)(u)≥ r0,and

hencewededucefrom (26) that

Ψε(hKε−σ(Kε)(u))≤ Ψ(hKε−σ(Kε)(u)) +

π

2. (46)

Since Kε(m)− σ(Kε(m)) tendsto K− σ(K),(25) impliesthatifu∈ Sn−1,then lim

ε→0ε(hK

ε−σ(Kε)(u)) = Ψ(hK−σ(K)(u)). (47)

Combining (45), (46) and (47) with Lebesgue’s Dominated Convergence Theorem,we conclude(41),andinturnTheorem7.3. 2

8. TheproofofTheorem1.2

Let−n< p< 0,letf beanon-negative non-trivialfunctioninL n n+p(S

n−1),and let ϕ: [0,∞)→ [0,∞) beamonotoneincreasingcontinuousfunctionsatisfyingϕ(0)= 0,

lim inf t→0+ ϕ(t) t1−p > 0 (48)  1 1 ϕ(t)dt <∞. (49)

Weassociatecertainfunctionstof andϕ.Foranyintegerm≥ 2,wedefinefmonSn−1 as follows: fm(u) = ⎧ ⎪ ⎨ ⎪ ⎩ m if f (u)≥ m, f (u) if m1 < f (u) < m, 1 m if f (u)≤ 1 m.

In particular,fm≤ ˜f wherethefunctionf : S˜ n−1 → [0,∞) inL n n+p(S

n−1),and hence inL1(Sn−1),isdefinedby

(25)

˜ f (u) =



f (u) if f (u) > 1, 1 if f (u)≤ 1. AsinSection5, using(49),wedefinethefunction

Ψ(t) =  t 1 ϕ for t > 0.

Accordingto(48),thereexist someδ∈ (0,1) andℵ> 1 suchthat 1

ϕ(t) <ℵt

p−1 for t∈ (0, δ). (50)

WededucefromLemma4.1thatthere exists0> 1 depending onϕ suchthat

Ψ(t) <ℵ0tp for t∈ (0, δ). (51)

Form≥ 2,Theorem7.3yieldsaλm> 0 andaconvexbodyKm∈ Kn0 withξ(Km)= o∈ int Km,V (Km)= 1 suchthat

λmϕ(hKm) dSKm = fmdH n−1 (52)  Sn−1 Ψ(hKm−σ(Km))fmdH n−1 ≥ Ψ(5)  Sn−1 fmdHn−1. (53)

Inaddition,iff isinvariantunderaclosedsubgroupG ofO(n),thenfmisalsoinvariant underG,and henceKmcanbechosento beinvariantunderG.

Sincefm≤ ˜f ,andfmconverges pointwiseto f ,Lebesgue’sDominated Convergence theoremyieldstheexistenceofm0> 2 suchthatifm> m0,then

1 2  Sn−1 f <  Sn−1 fm< 2  Sn−1 f. (54) Inparticular,(53) implies  Sn−1 Ψ(hKm−σ(Km)) ˜f dH n−1Ψ(5) 2  Sn−1 f dHn−1. (55)

WededucefromV (Km)= 1,limt→∞Ψ(t)= 0,(51),(55) andLemma4.3 thatthere existsR0> 0 independentofm suchthat

Km⊂ σ(Km) + R0Bn⊂ 2R0Bn for all m > m0. (56)

(26)

σ(Km) + r0Bn ⊂ Km for all m > m0. (57)

To estimateλm frombelow,(56) impliesthat 

Sn−1

ϕ(hKm) dSKm ≤ ϕ(2R0)H

n−1(∂K

m)≤ ϕ(2R0)(2R0)n−1nκn,

and henceitfollows from (52) and (54) the existenceofλ˜1 > 0 independent ofm such

that λm=  Sn−1fmdHn−1  Sn−1ϕ(hKm) dSKm ≥ ˜λ1 for all m > m0. (58)

To haveasuitableupper bound onλmfor any m> m0, wechoosewm∈ Sn−1 and m≥ 0 such thatσ(Km)= mwm.WesetB#m= wm⊥∩ int Bn and considertherelative open set Ξm= (∂Km) mwm+ r0B#m+ (0,∞)wm .

Ifu is anexteriorunitnormalatanx∈ Ξmform> m0,thenx= ( m+ s)wm+ rv for s> 0,r∈ [0,r0) andv∈ wm⊥∩ Sn−1,andhence mwm+ rv∈ Kmyields

u, ( m+ s)wm+ rv = hKm(u)≥ u, mwm+ rv ,

implying that u,wm ≥ 0; or in other words, u ∈ Ω(wm,π2). Since the orthogonal projectionofΞm ontowm⊥ isBm# form> m0,wededucethat

SKm Ωwm, π 2 ≥ Hn−1 m)≥ Hn−1(Bm#) = r0n−1κn−1. (59) Ontheother hand,ifu∈ Ω(wm,π2) form> m0,then mwm+ r0u∈ Km yields

hKm(u)≥ u, mwm+ r0u ≥ r0. (60)

Combining (54),(59) and(60) implies

λm=  Ω(wm,π2) fmdHn−1  Ω(wm,π2)ϕ(hKm) dSKm 2  Sn−1f dHn−1 ϕ(r0)r0n−1κn−1 = ˜λ2 for all m > m0. (61)

Since Km ⊂ 2R0Bn and λ˜1 ≤ λm ≤ ˜λ2 for m > m0 by (56), (58) and (61), there

exists subsequence{Km}⊂ {Km} suchthatKm tendstosomeconvexcompactset K and λm tendsto someλ> 0. Aso∈ Km and V (Km)= 1 forallm, wehaveo ∈ K

(27)

Weclaimthatforanycontinuous functiong : Sn−1 → R,we have  Sn−1 gλϕ(hK) dSK =  Sn−1 gf dHn−1. (62)

Asϕ is uniformlycontinuous on[0,2R0] andhKm tends uniformlyto hK on S

n−1, we deducethatλmϕ(hKm) tendsuniformlyto λϕ(hK) onS

n−1.SinceS Km tendsweakly toSK,wehave lim m→∞  Sn−1 gλmϕ(hKm) dSKm =  Sn−1 gλϕ(hK) dSK.

Ontheotherhand,|gfm|≤ ˜f· maxSn−1|g| forallm≥ 2,andgfmtendspointwisetogf

as m tends to infinity.Therefore Lebesgue’s Dominated Convergence Theorem implies

that lim m→∞  Sn−1 gfmdHn−1=  Sn−1 gf dHn−1,

whichinturn yields(62) by(52).Inturn,we concludeTheorem 1.2by(62). 2 References

[1]G.Bianchi, K.J.Böröczky,A. Colesanti,D.Yang, TheLp-Minkowskiproblem for −n< p< 1,

Adv.Math.341(2019)493–535.

[2] G. Bianchi,K.J.Böröczky,A.Colesanti,SmoothnessintheLp Minkowskiproblemfor p< 1,J.

Geom.Anal.(2019),inpress,https://doi.org/10.1007/s12220-019-00161-y.

[3]K.J.Böröczky,E.Lutwak,D.Yang,G.Zhang,ThelogarithmicMinkowskiproblem,J.Amer.Math. Soc.26(2013)831–852.

[4]S.Campi,P.Gronchi,TheLp-Busemann-Pettycentroidinequality,Adv.Math.167(2002)128–141.

[5]S.Chen,Q.-R.Li,G.Zhu,ThelogarithmicMinkowskiproblemfornon-symmetricmeasures,Trans. Amer.Math.Soc.371(2019)2623–2641.

[6]S.Chen,Q.-R.Li,G.Zhu,OntheLpMonge-Ampèreequation,J.DifferentialEquations263(2017)

4997–5011.

[7]W.Chen,LpMinkowskiproblemwithnotnecessarilypositivedata,Adv.Math.201(2006)77–89.

[8]S.-Y.Cheng,S.-T.Yau,Ontheregularityofthesolutionofthen-dimensionalMinkowskiproblem, Comm.PureAppl.Math.29(1976)495–561.

[9]K.-S.Chou,DeformingahypersurfacebyitsGauss-Kroneckercurvature,Comm.PureAppl.Math. 38(1985)867–882.

[10]K.-S. Chou,X.-J. Wang,TheLp-Minkowski problemandtheMinkowskiproblem incentroaffine

geometry,Adv.Math.205(2006)33–83.

[11]A.Cianchi,E.Lutwak,D.Yang,G.Zhang,AffineMoser-TrudingerandMorrey-Sobolevinequalities, Calc.Var.PartialDifferentialEquations36(2009)419–436.

[12]H.G.Eggleston,Convexity,CambridgeTractsinMathematicsandMathematicalPhysics,vol. 47, CambridgeUniversityPress,NewYork,1958.

[13] R.J.Gardner,D. Hug,W.Weil, S.Xing,D.Ye,GeneralvolumesintheOrlicz-Brunn-Minkowski theoryandarelatedMinkowskiproblemI,Calc.Var.58(2019),https://doi.org/10.1007/s00526 -018-1449-0.

[14]R.J.Gardner,D.Hug,S.Xing,D.Ye,GeneralvolumesintheOrlicz-Brunn-Minkowskitheoryand arelatedMinkowskiproblemII,arXiv:1809.09753,submittedforpublication.

(28)

[15]B. Guan, P. Guan, Convex hypersurfaces of prescribed curvatures, Ann. Math. (2) 156 (2002) 655–673.

[16] P.Guan,C.-S.Lin,Onequationdet(uij+ δiju)= upf onSn,preprint.

[17]C.Haberl, E.Lutwak,D.Yang,G.Zhang,TheevenOrliczMinkowskiproblem,Adv.Math.224 (2010)2485–2510.

[18]C.Haberl,F.Schuster,GeneralLpaffineisoperimetricinequalities,J.DifferentialGeom.83(2009)

1–26.

[19]M.Henk,E.Linke,Cone-volumemeasuresofpolytopes,Adv.Math.253(2014)50–62.

[20]Q. Huang,B.He,On theOrlicz Minkowski problemfor polytopes, DiscreteComput. Geom.48 (2012)281–297.

[21]Y.Huang,Q.Lu,OntheregularityoftheLp-Minkowskiproblem,Adv.inAppl.Math.50(2013)

268–280.

[22]Y.Huang,E.Lutwak,D.Yang,G.Zhang,GeometricmeasuresinthedualBrunn-Minkowskitheory andtheirassociatedMinkowskiproblems,ActaMath.216(2016)325–388.

[23]D. Hug,E. Lutwak,D. Yang, G.Zhang, OntheLp Minkowski problem forpolytopes, Discrete

Comput.Geom.33(2005)699–715.

[24]M.N.Ivaki,AflowapproachtotheL−2Minkowskiproblem,Adv.inAppl.Math.50(2013)445–464. [25]H.Jian,J.Lu,ExistenceofthesolutiontotheOrlicz-Minkowskiproblem,Adv.Math.344(2019)

262–288.

[26]H.Jian,J.Lu,G.Zhu,Mirrorsymmetricsolutionstothecentro-affineMinkowskiproblem,Calc. Var.PartialDifferentialEquations55(2016)1–22.

[27]M.Y.Jiang,Remarksonthe2-dimensionalLp-Minkowskiproblem,Adv.NonlinearStud.10(2010)

297–313.

[28]D.Klain,TheMinkowskiproblemforpolytopes,Adv.Math.185(2004)270–288.

[29]H.Lewy,Ondifferential geometryinthelarge,I.Minkowski’sproblem,Trans.Amer.Math.Soc. 43(1938)258–270.

[30] Q-R.Li,Infinitelymanysolutionsforcentro-affineMinkowskiproblem,Int.Math.Res.Not.(2019), inpress,https://doi.org/10.1093/imrn/rnx284.

[31]J.Lu,X.-J. Wang,Rotationallysymmetricsolutionto theLp-Minkowskiproblem,J.Differential

Equations254(2013)983–1005.

[32]E.Lutwak,V.Oliker,OntheregularityofsolutionstoageneralizationoftheMinkowskiproblem, J.DifferentialGeom.41(1995)227–246.

[33]E.Lutwak,D.Yang,G.Zhang,Lpaffineisoperimetricinequalities,J.DifferentialGeom.56(2000)

111–132.

[34]E.Lutwak,D.Yang,G.Zhang,OntheLp-Minkowskiproblem,Trans.Amer.Math.Soc.356(2004)

4359–4370.

[35]E.Lutwak,D.Yang,G.Zhang,VolumeinequalitiesforsubspacesofLp,J.DifferentialGeom.68

(2004)159–184.

[36]E.Lutwak,D.Yang,G.Zhang,Lp-dualcurvaturemeasures,Adv.Math.329(2018)85–132.

[37]L.Nirenberg,TheWeylandMinkowskiproblemsindifferentialgeometryinthelarge,Comm.Pure Appl.Math.6(1953)337–394.

[38]R.Schneider,ConvexBodies:TheBrunn-MinkowskiTheory,secondexpandededition,Encyclopedia ofMathematicsandItsApplications,CambridgeUniversityPress,Cambridge,2014.

[39]A.Stancu,ThediscreteplanarL0-Minkowskiproblem,Adv.Math.167(2002)160–174.

[40]A.Stancu,Onthenumberofsolutionstothediscretetwo-dimensionalL0-Minkowskiproblem,Adv. Math.180(2003)290–323.

[41]A. Stancu,Centro-affine invariantsfor smooth convexbodies, Int.Math.Res.Not. 2012 (2012) 2289–2320.

[42]T. Wang, On the discrete functionalLp-Minkowski problem, Int.Math. Res. Not. 2015 (2015)

10563–10585.

[43]S.Xing,D.Ye,B.Zhu,ThedualOrlicz-Minkowskiproblem,J.Geom.Anal.28(2018)3829–3855. [44]S.Xing,D.Ye,OnthegeneraldualOrlicz-Minkowskiproblem,IndianaUniv.Math.J.(2019),in

press,arXiv:1802.06331.

[45]G.Xiong, Extremumproblems fortheconevolumefunctionalforconvexpolytopes,Adv.Math. 225(2010)3214–3228.

[46]G.Zhu,ThelogarithmicMinkowskiproblemforpolytopes,Adv.Math.262(2014)909–931. [47]G. Zhu, The centro-affine Minkowski problem for polytopes, J. Differential Geom. 101 (2015)

(29)

[48]G. Zhu, The Lp Minkowski problem for polytopes for 0 < p < 1, J. Funct. Anal. 269 (2015)

1070–1094.

[49]G.Zhu,TheLp-Minkowskiproblemforpolytopesfornegativep,IndianaUniv.Math.J.66(2017)

1333–1350.

[50]G.Zhu,ContinuityofthesolutiontotheLp-Minkowskiproblem,Proc.Amer.Math.Soc.145(2017)

Riferimenti

Documenti correlati

US Uber politics: elite-driven politics of concentrated actors and surrogate representation of dispersed actors—workers and consumers... While the platform economy is disruptive

Nella totale godibilità dello spazio interno, il criterio dell’esposizio- ne appare essere quello dell’esperienza di uno scavo archeologico, nel quale, in logica successione

Here, to make up for the relative sparseness of weather and hydrological data, or malfunctioning at the highest altitudes, we complemented ground data using series of remote sensing

Torres, Multiple positive radial solutions for a Dirichlet problem involving the mean curvature operator in Minkowski space, J.. Bergner, On the Dirichlet problem for the

Our results indicate that, in a setting of free choice between different TV programs, the presence of verbal violence in one program causes subjects to watch

This result strongly suggests that we are observing discrete shifts from part-time to full-time work, as conjectured by Zabalza et al 1980 and Baker and Benjamin 1999, rather

We focused on phase P2 in all the drives, where we inves- tigated the effect of cognitive and sensorimotor distractions on sympathetic arousal and driving behavior.. Sympathetic

shore, the Salento Calcarenite onlaps Oligocene (Castro Limestone) and Miocene (Novaglie Formation) clinos- tratified slope deposits or even Cretaceous rudist