Contents lists available atScienceDirect
Advances in Mathematics
www.elsevier.com/locate/aim
Nonuniqueness of solutions to the L
p-Minkowski problem
✩Huaiyu Jiana,b, Jian Lub,c,∗, Xu-Jia Wangb
aDepartmentofMathematics,TsinghuaUniversity,Beijing100084,China
bCentreforMathematicsandItsApplications,AustralianNationalUniversity, Canberra,ACT0200,Australia
cDepartmentofAppliedMathematics,ZhejiangUniversityofTechnology, Hangzhou310023,China
a r t i c l e i n f o a b s t r a c t
Articlehistory:
Received8July2014
Receivedinrevisedform19May 2015
Accepted21May2015 Availableonline12June2015 CommunicatedbyErwinLutwak
Keywords:
Monge–Ampèreequation Minkowskiproblem Nonuniqueness
Inthispaper weshow byexamplethat there isno uniform estimatefortheLp-Minkowskiproblem.Asaresultweobtain thenonuniquenessofsolutionstotheproblem.
© 2015ElsevierInc.All rights reserved.
1. Introduction
WeconsiderthefollowingLp-Minkowskiproblem,
det(∇2H+HI) =f Hp−1 onSn, (1.1)
✩ ThefirstauthorwassupportedbytheNaturalScienceFoundationofChina(11131005,11271118);the secondauthorwassupportedbytheNaturalScienceFoundationofChina(11401527);andthethirdauthor wassupportedbytheAustralianResearchCouncil(DP120102718andFL130100118).
* Correspondingauthor.
E-mailaddresses:[email protected](H.Y. Jian),[email protected](J. Lu), [email protected](X.-J. Wang).
http://dx.doi.org/10.1016/j.aim.2015.05.010 0001-8708/© 2015ElsevierInc.All rights reserved.
whereH isthesupportfunctionofaconvexbodyK=KHintheEuclideanspaceRn+1, I is the unit matrix, f is agiven positive function on the unit sphere Sn, and p is a constant.Themain resultofthepaperis
Theorem1.1. Foranyp∈(−n−1,0),thereexistsapositivefunctionf ∈C∞(Sn)such that equation (1.1)admits twodifferentsolutions.
To prove Theorem 1.1, we will construct a smooth, positive function f, which is radially symmetricinx1,· · ·,xn, andalso symmetricinxn+1, suchthatequation (1.1) has asolutionwith small volume.By[7] there isanother solutionwhose volume hasa positivelowerbound dependingonlyonn,infSnf and supSnf.Thereforewehavetwo different solutionsforequation (1.1).
The Lp-Minkowski problem was introduced by Lutwak [16] and has been studied by manyauthors [4,5,7,11,14,15,17–21].Oneof themain questionsistheuniquenessof solutions. In particular, a solution to (1.1) is also a self-similar solution to the Gauss curvature flow,ofwhichtheuniquenesshasbeen extensivelystudied.
Denoteq= 1−p.Whenf ≡1,theuniquenesshasbeenobtainedinthecaseq=nfor alln[9]andinthecasen= 2 andq= 1[1].Whenn= 1,q= 1,andf issymmetricwith respect to theorigin, namelyf(θ)=f(θ+π), theuniqueness wasobtained in[10,13].
When q <−n, theuniquenesscanbeobtainedbythemaximum principle[7]. Onecan alsofindsomeuniquenessresultsinthecaseq <0 forsymmetricf in[16].Inthediscrete caseauniquenesswasestablishedfortheL0-Minkowskiproblem in[20].
Ontheotherhand,itiswellknownthatwhenq=n+ 2,allellipsoidswiththevolume oftheunitballaresolutionsof(1.1)withf ≡1.Theuniquenessismoredelicateinthe caseq > n+ 2,atleastwhenn= 1[3].In[7]itwasshownthatthesolutionmaynotbe uniqueifq∈(1,n+ 2) andisverycloseton+ 2.Incontrasttotheuniquenessin[10,13]
for n= 1,q= 1,and symmetricf,asurprising nonuniqueness resultwasdiscovered in [22] fornonsymmetricf,alsointhecasen= 1, q= 1.
The uniqueness of solutions for the case 0 < q < n+ 2 attracted much attentions as itis relatedtothelimitshape ofGauss curvatureflows ofconvexhypersurfacesand has received considerable investigations. See [1,2,6,8,9] and the above discussion. The uniquenessofsolutionshasbeenconjecturedforanumberofspecialcases,includingin particular the case f ≡ 1 and q = 1 [12]. Our theorem above shows that for general positivefunctionf, theremaybe morethanonesolutionto(1.1), forallq∈(1,n+ 2) and all dimensions n. It implies that the limit shape of anisotropic curvature flows is usually notunique.
We will firstprove Theorem 1.1 for the case p∈ (−n−1,−1) in Sections 2 and 3, andthenforthecasep∈[−1,0) inSection4.InSection2weconstructanfεsuchthat equation (1.1)hasasolutionHε ofsmallvolume.TheninSection3weshowthatthere is a solutionto (1.1) of whichthe volumehas apositivelower bound. In Section4we extendtheexampletothecasep∈[−1,0).
2. Asolutionwithsmallvolume
Denoteδ=n+ 2−q∈(0,n+ 1).Let
Mε= diag(1,· · ·,1, ε) = I 0
0 ε
(2.1) beamatrix,where0< ε≤1 isasmallconstantandasbeforeIistheunitn×nmatrix.
Considertheequation
det(∇2h+hI)(x) =|Mεx|−δ, x∈Sn. (2.2) Equation(2.2) is justthe classical Minkowskiproblem.As its right hand sidesatisfies the necessary condition
Snxk|Mεx|−δ = 0 for all1 ≤k ≤n+ 1, there is a solution, whichisuniqueupto translation,totheequation.
Lethε be theuniquesolution to (2.2)suchthatthe centreofthe associated convex bodyKhε is locatedat theorigin.Notethathε isradiallysymmetricinx1,· · ·,xn and symmetricinxn+1.Define
Hε(x) = (detMε)n+q2 ·Mε−1x·hε
Mε−1x Mε−1x
, (2.3)
thenHεisthesupportfunctionofaconvexbodyKHε, andKHε canbeobtainedfrom Khε bymaking thecoordinate transformx→Mε−1xand then makingadilation x→ (detMε)n+q2 x.
Lemma2.1. The functionHε satisfiestheequation
det(∇2Hε+HεI) = ˆhqε Hεq
on Sn, (2.4)
where
ˆhε(x) =hε
Mε−1x Mε−1x
. (2.5)
Proof. Let
uε(x) =Mε−1
x·hε
Mε−1
Mε−1xx
. (2.6)
Bytheinvarianceofthequantityhn+2ε det(∇2hε+hεI) underunimodularaffinetrans- formations,seeProposition 7.1in[7]or formula (2.12)in[15],wehave
det(∇2uε+uεI)(x) = det(∇2hε+hεI)
Mε−1x
|Mε−1x|
· (detMε−1)2
|Mε−1x|n+2.
Here wenote thatMε−1 isnotaunimodulartransformation.Since hε satisfiesequation (2.2),we have
det(∇2uε+uεI)(x) =|Mε−1x|δ· (detMε−1)2
|Mε−1x|n+2
= 1
(detMε)2|Mε−1x|q. (2.7) Therefore bythedefinitionofHε,(2.3),onegets
det(∇2Hε+HεI)(x) = (detMε)n+q2n det(∇2uε+uεI)(x)
= 1
(detMε)n+q2q Mε−1xq
= hˆqε Hεq
. 2
ToestimatethevolumeoftheconvexbodyKHε,oneneedstostudytheconvexbody Khε, or equivalently the support function hε. When δ ∈ (0,n), we have the following uniform estimates.
Lemma 2.2. When 2 < q < n+ 2, there exists a positive constant C, independent of ε∈(0,1],such that
C−1≤hε≤C on Sn. (2.8)
Proof. Onecan easily see thatthe area of∂Khε is uniformly bounded from above. In fact,since
|Mεx| ≥
1−x2n+1 ∀ x∈Sn andε∈(0,1], and notingthatδ=n+ 2−q < n, wehave
area(∂Khε) =
Sn
det(∇2hε+hεI)
=
Sn
|Mεx|−δ
≤
Sn
(1−x2n+1)−δ/2
≤C,
whereC is apositiveconstantdependingonn, δbutindependentof ε.Bytheisoperi- metricinequality,weobtain
vol(Khε)≤Cnarea(∂Khε)n+1n ≤C. (2.9) Assumehε attains its maximum at point xε ∈ Sn, namely hε(xε) = maxSnhε. By convexityandrecallingthathεissymmetric, wehave
hε(x)≥maxhε· xε, x ∀x∈Sn. Observingthat
|Mεx| ≤1 ∀x∈Sn and ε∈(0,1], byequation(2.2),wehave
vol(Khε) = 1 n+ 1
Sn
hεdet(∇2hε+hεI)
= 1
n+ 1
Sn
hε(x)|Mεx|−δ
≥ 1 n+ 1
Sn
hε(x)
≥Cnmaxhε·
Sεn
xε, x
=Cnmaxhε,
whereSεn ={x∈Sn: xε, x>0}. Thereforeweobtainfrom (2.9)that
maxhε≤Cnvol(Khε)≤C. (2.10) Thesecondinequalityof(2.8)isproved.
Toprovethefirstinequalityof(2.8),wemakeuseoftheconceptofminimumellipsoid ofaconvex body.LetEhε betheminimumellipsoid ofKhε.Thenwe have
1
n+ 1Ehε ⊂Khε ⊂Ehε.
Bythesymmetry ofKhε,thecentreofEhε isat theorigin.LetR1,ε≥ · · · ≥Rn+1,ε be thelengthsof thesemi-axisofEhε.Then
R1,ε≤(n+ 1) maxhε, Rn+1,ε≤(n+ 1) minhε.
Hence
vol(Khε)≤vol(Eε)
≤ωn+1Rn1,εRn+1,ε
≤Cn(maxhε)n·minhε,
where ωn+1 isthevolumeoftheunitballinRn+1.By(2.10) itfollowsthat 1≤Cn(maxhε)n−1·minhε.
Thefirstinequalityof(2.8)follows. 2 Now let
fε= ˆhqε. (2.11)
In viewof (2.5)and (2.8), there exist two positiveconstants C1, C2, independent ofε, suchthat
C1≤fε≤C2. (2.12)
From (2.4),Hε isasolutionto
det(∇2Hε+HεI) = fε
Hεq
onSn. (2.13)
By(2.3),wehavethevolumeestimate
vol(KHε) = (detMε)2n+2n+q ·detMε−1·vol(Khε)
= (detMε)n+qδ vol(Khε). (2.14) Note thatdetMε=ε.Hencewhen0< δ < n,thevolumeofKHε canbe assmallaswe want,providedεissufficientlysmall.
Remark. From the transform (2.3) and estimate (2.8), one easily sees thatthere is no uniform upperboundforthesolutionHε, whenε>0 issmall.
3. Avariationalsolution
Inthepaper[7],thevariationalproblem sup
h
yinf∈Kh
J[h(x)−y·x] : vol(Kh) = 1
(3.1)
is studied. Here h is the support function of a non-degenerate convex body Kh with volumevol(Kh).ThefunctionalJ[h] isgivenby
J[h] = 1 q−1
Sn
f hq−1.
Forconveniencewerecallthefollowingexistenceresultin[7].
Lemma3.1.(See[7].)Let1< q < n+2andf ∈L∞(Sn)beapositivefunctionsatisfying (2.12). Then the variational problem (3.1) admits a maximiser H which satisfies, for someLagrangemultiplier λ>0,
det(∇2H+HI) = λf
Hq onSn. (3.2)
Whenf(x)=f(−x),onecanconsidertheproblem(3.1)intherestrictedclassh(x)= h(−x).Onecanfollow thesameargumentsasin[7]toshowthatamaximiser H exists anditscorrespondingconvex bodyis centrallysymmetric.
Bythis lemma,weobtainasolutiontoequation (1.1). Let
H˜ =λ−n+q1 H. (3.3)
ThenH˜ isasolutiontoequation
det(∇2H˜+ ˜HI) = f
H˜q onSn. (3.4)
Thevolumeofthecorresponding convexbodyKH˜ is
vol(KH˜) =λ−n+1n+q. (3.5) Weneed to estimate theLagrange constantλ. It is given as follows. MultiplyingH toboth sidesofequation(3.2)andtakingintegration,wehave
1 = vol(KH) = 1 n+ 1
Sn
Hdet(∇2H+HI)
= λ
n+ 1
Sn
f Hq−1
= (q−1)λ
n+ 1 J[H]. (3.6)
TheBlaschke–Santaloinequalityforcentrallysymmetricconvexbodiesisgivenby V(h)
Sn
1
hn+1dS(x)≤ ωn2 n+ 1,
where h(x) = h(−x) is the support function of any centrally symmetric convex body and ωn isthevolumeoftheunitball inRn.Usingthis inequality,wehave
J(H) = 1 q−1
Sn
f Hq−1
=supf q−1
Sn
1 Hq−1
n+1q−1
Sn
1 n+2−qn+1
≤Cnsupf q−1 .
On the other hand, let h0 be the constant function ωn+1−1/(n+1) which is the support functionoftheballcentred attheoriginwhosevolumeisequalto 1.Fromthevariational characterisation ofH,wehave
J(H)≥H(h0) =Cn supf q−1 . From (3.6),wethusobtain
Cn−1inf
Snf ≤λ−1≤Cnsup
Sn
f.
By(3.5),weseethatthereisapositiveconstantCn dependingonlyonn,suchthat Cn−1(inf
Snf)n+1n+q ≤vol(KH˜)≤Cn(sup
Sn
f)n+1n+q. (3.7) Now we let f =fε be the function in(2.13). Denote the corresponding variational solutionbyH˜ε,whichisthesupportfunctionofaconvexbodyKH˜ε.Byvirtueof(2.12) and (3.7),onehastheestimates
C3≤vol(KH˜ε)≤C4,
where C3,C4 arepositiveconstantsdependingonlyonn,q,C1 andC2.Butrecallthat forthesolutionHεgivenin(2.3),thevolumevol(KHε)→0 asε→0.HenceH˜εandHε are two different solutions.We have therefore obtainedthe nonuniqueness of solutions to (2.13)forsmallε>0,inthecaseq∈(2,n+ 2).
Remark.Asf isrotationallysymmetricwithrespecttothexn+1-axis,onemayconsider thesupremumin(3.1)inthefamilyofrotationallysymmetricconvexbodiesandobtain a rotationally symmetric solution. Therefore for equation (2.13) the two solutions we obtained are both rotationally symmetric with respect to the xn+1-axis, and so are smooth.
Wealsoremarkthattheideaofourproofisbasedontheobservationthatvol(KHε) issmall.Butwhenq= 1 andq=n+ 2,onecanprovethatforanysolutionto(1.1),the volume of the associated convex body hasa positive lower bound. Therefore our con- structionofmorethanonesolution doesnotcoverthecaseq= 1.Infact,asmentioned intheintroduction,whenn= 1 andq= 1,thesolutionisuniquewhenf is symmetric [10,13].
4. Thecase1< q ≤2
WehaveprovedTheorem 1.1forthecase2< q < n+ 2.Inthissectionweshowthat the solutions Hε and H˜ε given in (2.3) and (3.3) are different even when 1 < q ≤ 2, providedεaresufficientlysmall.
RecallthatKhε isradiallysymmetricinx1,· · ·,xnandsymmetricinxn+1.Wedenote R1 =R1,ε =hε(1,0,· · ·,0),Rn+1=Rn+1,ε=hε(0,· · ·,0,1).Namely R1 and Rn+1 are respectivelythevaluesofhεontheequatorandatthenorthpoleofthesphereSn. Lemma4.1. Thereexistsaconstant C >0,independentof ε∈(0,1],suchthat
Rn+1< CR1. (4.1)
Proof. Denote Sη = {x ∈ Sn : |xn+1| < η} and Sηc = Sn\Sη, where η ∈ (0,1) is a constant.LetΓη ={p∈∂Khε :G(p)∈Sη}and Γcη =∂Khε\Γη,where GistheGauss mapof∂Khε, namelyG(p) istheunitouternormalof∂Khε at thepointp.
If(4.1)isnottrue,wehave
area(Γ1/2)>>area(Γc1/2).
Ontheotherhand,byequation (2.2)we have area(Γ1/2) =
S1/2
det(∇2hε+hεI) =
S1/2
|Mεx|−δ. (4.2)
Notethat
sup
Sη
|Mεx|−δ≤inf
Scη|Mεx|−δ ∀η∈(0,1) andε∈(0,1].
Hencetheright handsideof(4.2)
≤Cn
Sc1/2
|Mεx|−δ =Cn
S1/2c
det(∇2hε+hεI) =Cnarea(Γc1/2). (4.3)
Wereachacontradiction. 2
Byequation(2.2),wehaveasε→0 that area(∂Khε) =
Sn
|Mεx|−δ
=
(C+o(1))εq−2 if 1< q <2,
(C+o(1))|logε| ifq= 2, (4.4) where C is apositiveconstantindependentof ε.Onthe otherhand,byLemma 4.1we have
C−1R1n≤area(∂Khε)≤CR1n
.
From (4.4)itfollowsthat
C−1εq−2n ≤R1≤Cεq−2n if 1< q <2,
C−1|logε|1/n≤R1≤C|logε|1/n ifq= 2. (4.5) Observing that
vol(Khε) =CRn1Rn+1, by(2.14) weobtain
vol(KHε) =εn+qδ vol(Khε)
=
Cε(q−1)(n+q−2)
n+q Rn+1 if 1< q <2,
Cεn+qδ |logε|Rn+1 ifq= 2. (4.6) Now considerthesolutionH˜ε givenin(3.3).From (3.7), wehave
vol(KH˜ε)≥C(inf
Snfε)n+1n+q. Notingthatfε= ˆhqε andthatinfSnˆhε= infSnhε,weseethat
vol(KH˜ε)≥C(inf
Snhε)q(n+1)n+q
≥CR
q(n+1) n+q
n+1 . (4.7)
Inorderto provethatH˜ε andHε aretwodifferentsolutions,weneedthat,
vol(KH˜ε)>vol(KHε). (4.8)
By(4.6)and(4.7),itsufficesto have Rn+1>
Cεn+q−2n if 1< q <2,
Cε|logε|n+2n ifq= 2. (4.9) Soitsufficesto givealowerbound forRn+1 tocomplete Theorem 1.1.Infactwehave thefollowing
Lemma4.2. For1< q≤2,wehave Rn+1≈
ε(2−q)(n−1)n if 1< q <2,
|logε|1−nn if q= 2. (4.10) Herethenotation “≈” means theratio of thetwosides has uniform positiveupper and lowerbounds.
Proof. Letuεbeas (2.6),Kuε be thecorrespondingconvexbody.From(2.7),we have area(∂Kuε) =
Sn
1
ε2|Mε−1x|q = (C+o(1))ε−1, ∀q >1. (4.11) Denote
r1=r1,ε=uε(1,0,· · ·,0) =R1, rn+1 =rn+1,ε=uε(0,· · ·,0,1) =ε−1Rn+1. Since
area(∂Kuε)≈r1n+rn1−1rn+1, multiplyingbothsidesbyε,andnoting(4.11), weobtain
1≈εRn1 +Rn−11 Rn+1. Onaccountof(4.5),εRn1 <<1,whichleadsto
1≈Rn1−1Rn+1. (4.12)
By(4.5)again,wehave(4.10). 2
Combining (4.9)and (4.10), weobtain (4.8).Therefore thecase 1< q ≤2 ofTheo- rem 1.1isalsoproved.
References
[1]B.Andrews,Gausscurvatureflow:thefateoftherollingstones,Invent.Math.138(1999)151–161.
[2]B.Andrews,MotionofhypersurfacesbyGausscurvature,PacificJ.Math.195(2000)1–34.
[3]B.Andrews,Classificationoflimitingshapesforisotropiccurveflows,J.Amer.Math.Soc.16(2003) 443–459.
[4]K.Boroczky,E.Lutwak,D.Yang,G.Zhang,ThelogarithmicMinkowskiproblem,J.Amer.Math.
Soc.26(2012)831–852.
[5]W.X.Chen,LpMinkowskiproblemwithnotnecessarilypositivedata,Adv.Math.201(2006)77–89.
[6]K.S. Chou(K. Tso), Deforming a hypersurface by itsGauss–Kronecker curvature, Comm. Pure Appl.Math.38(1985)867–882.
[7]K.S. Chou, X.J. Wang, The Lp-Minkowski problem and theMinkowski problem incentroaffine geometry,Adv.Math.205(2006)33–83.
[8]K.S.Chou,X.-P.Zhu,Anisotropicflowsforconvexplanecurves,DukeMath.J.97(1999)579–619.
[9]B.Chow,DeformingconvexhypersurfacesbythenthrootoftheGaussiancurvature,J.Differential Geom.23(1985)117–138.
[10]C.Dohmen,Y.Giga,Selfsimilarshrinking curvesforanisotropic curvatureflowequations, Proc.
JapanAcad.Ser.A70(1994)252–255.
[11]J.B.Dou,M.J. Zhu, Thetwo dimensionalLp Minkowski problem andnonlinearequationswith negativeexponents,Adv.Math.230(2012)1209–1221.
[12]W.Firey,Shapesofwornstones,Mathematika21(1974)1–11.
[13]M.E. Gage,Evolvingplanecurvesby curvature inrelativegeometries, DukeMath.J. 72(1993) 441–466.
[14] H.Y. Jian, J. Lu,Mirror-symmetric solutionsto thecenteroaffine Minkowski problem, preprint, 2013.
[15]J.Lu,X.-J.Wang,RotationallysymmetricsolutionstotheLp-Minkowskiproblem,J.Differential Equations254(2013)983–1005.
[16]E. Lutwak, The Brunn–Minkowski–Firey theory I: mixed volumesand theMinkowski problem, J. DifferentialGeom.38(1993)131–150.
[17]E.Lutwak,V.Oliker,OntheregularityofsolutionstoageneralizationoftheMinkowskiproblem, J.DifferentialGeom.41(1995)227–246.
[18]E.Lutwak,D.Yang,G.Zhang,OntheLp-Minkowskiproblem,Trans.Amer.Math.Soc.356(2004) 4359–4370.
[19]E.Lutwak,D.Yang,G.Zhang,OptimalSobolevnormsandtheLpMinkowskiproblem,Int.Math.
Res.Not.IMRN2006(2006),Art.ID62987,21 pp.
[20]A.Stancu,Onthenumberofsolutionstothediscretetwo-dimensionalL0-Minkowskiproblem,Adv.
Math.180(2003)290–323.
[21]V. Umanskiy,On solvability of two-dimensional Lp-Minkowski problem,Adv. Math.180 (2003) 176–186.
[22]H.Yagisita,Non-uniquenessofself-similarshrinkingcurvesforananisotropiccurvatureflow,Calc.
Var.PartialDifferentialEquations26(2006)49–55.