www.elsevier.com/locate/anihpc
Some nonlinear differential inequalities and an application to Hölder continuous almost complex structures
Adam Coffman
a,∗, Yifei Pan
b,aaDepartment of Mathematical Sciences, Indiana University – Purdue University Fort Wayne, 2101 E. Coliseum Blvd., Fort Wayne, IN 46805-1499, USA
bSchool of Sciences, Nanchang University, Nanchang 330022, PR China Received 22 July 2009; received in revised form 2 October 2009; accepted 14 October 2009
Available online 2 February 2011
Abstract
We consider some second order quasilinear partial differential inequalities for real-valued functions on the unit ball and find conditions under which there is a lower bound for the supremum of nonnegative solutions that do not vanish at the origin. As a consequence, for complex-valued functionsf (z)satisfying∂f/∂z¯= |f|α, 0< α <1, andf (0)=0, there is also a lower bound for sup|f|on the unit disk. For eachα, we construct a manifold with anα-Hölder continuous almost complex structure where the Kobayashi–Royden pseudonorm is not upper semicontinuous.
©2011 Elsevier Masson SAS. All rights reserved.
MSC:primary 35R45; secondary 32F45, 32Q60, 32Q65, 35B05 Keywords:Differential inequality; Almost complex manifold
1. Introduction
We begin with an analysis of a second order quasilinear partial differential inequality for real-valued functions of nreal variables,
u−B|u|ε0, (1)
whereB >0 andε∈ [0,1)are constants. In Section 2, we use a Comparison Principle argument to show that (1) has
“no small solutions,” in the sense that there is a uniform lower boundM >0 for the supremum of solutionsuwhich are nonnegative on the unit ball and nonzero at the origin.
We also consider a generalization of (1):
uu−B|u|1+ε−C| ∇u|20, (2)
and find conditions under which there is a similar property of no small solutions, in Theorem 2.4.
* Corresponding author.
E-mail addresses:CoffmanA@ipfw.edu (A. Coffman), Pan@ipfw.edu (Y. Pan).
0294-1449/$ – see front matter ©2011 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.anihpc.2011.02.001
As an application of the results on the inequality (1), we show failure of upper semicontinuity of the Kobayashi–
Royden pseudonorm for a family of 4-dimensional manifolds with almost complex structures of regularity C0,α, 0< α <1. This generalizes theα= 12 example of [5]; it is known [6] that the Kobayashi–Royden pseudonorm is upper semicontinuous for almost complex structures with regularityC1,α.
Our construction of the almost complex manifolds in Section 4 is very similar to that of [5]; we give the details for the convenience of the reader, and to show how the argument breaks down asα→1−, due to a shrinking radius of the domain. We also take the opportunity in Section 3 to state some lemmas which allow for a more quantitative description than that of [5].
One of the steps in [5] is a Maximum Principle argument applied to a complex-valued functionh(z)satisfying the equation∂h/∂z¯= |h|1/2, to get the property of no small solutions. The main difference between our paper and [5] is the use of a Comparison Principle in Section 2 instead of the Maximum Principle, and we arrive at this result:
Theorem 1.1.For anyα∈(0,1), supposeh(z)is a continuous complex-valued function on the closed unit disk, and on the set{z: |z|<1, h(z)=0},hhas continuous partial derivatives and satisfies
∂h
∂z¯ = |h|α. (3)
Ifh(0)=0thensup|h|> Sα, where the constantSα>0is defined by:
Sα=
2(1−α) 2−α
1/(1−α)
. (4)
2. Some differential inequalities
LetDRdenote the open ball inRncentered at0 with radiusR >0, and letDRdenote the closed ball.
Lemma 2.1.Given constantsB >0and0ε <1, let M=
B(1−ε)2 2(2ε+n(1−ε))
1
1−ε
>0.
Suppose the functionu:D1→Rsatisfies:
• uis continuous onD1,
• u(x) 0forx∈D1,
• on the open setω= {x∈D1: u(x) =0},u∈C2(ω),
• forx∈ω:
u(x) −B u(x) ε
0. (5)
Ifu(0)=0, thensupx∈D1u(x) > M . Proof. Define a comparison function
v(x) =M|x|1−ε2 ,
sov∈C2(Rn)since 0ε <1. By construction ofM, it can be checked thatvis a solution of this nonlinear Poisson equation on the domainRn:
v(x) −Bv(x) ε≡0.
Suppose, toward a contradiction, thatu(x) Mfor allx∈D1. For a pointx0on the boundary ofω⊆Rn, either
|x0| =1, in which case by continuity,u(x0)M=v(x0), or 0<|x0|<1 andu(x0)=0, so u(x0)v(x0). Since uvon the boundary ofω, the Comparison Principle [4, Theorem 10.1] applies to the subsolutionuand the solution von the domainω. The relevant hypothesis for the Comparison Principle in this case is that the second term expression
of (5),−BXε, is weakly decreasing, which usesB >0 andε0. (To satisfy this technical condition for allX∈R, we define a functionc:R→Rbyc(X)= −BXε forX0, andc(X)=0 forX0. Thencis weakly decreasing inX,vsatisfiesv(x) +c(v(x)) ≡0 andusatisfiesu(x) +c(u(x)) 0.)
The conclusion of the Comparison Principle is thatuv onω, however0∈ωandu(0) > v(0), a contradic- tion. 2
Of course, the constant functionu≡0 satisfies the inequality (5), and so does the radial comparison functionv, so the initial conditionu(0)=0 is necessary.
Example 2.2.In then=1 case,M=(B(12(1−+ε)ε)2)1−ε1 . For pointsc1,c2∈R,c1< c2, define a function
u(x)=
⎧⎪
⎨
⎪⎩
M(x−c2)1−2ε ifxc2,
0 ifc1xc2,
M(c1−x)1−2ε ifxc1.
Thenu∈C2(R), and it is nonnegative and satisfies u=B|u|ε (the case of equality in then=1 version of (5)).
Forc1<0< c2, this gives an infinite collection of solutions of the ODEu=B|u|ε which are identically zero in a neighborhood of 0, so the ODE does not have a unique continuation property. Forc1>0 orc2<0, the functionu satisfiesu(0)=0 and the other hypotheses of Lemma 2.1, and its supremum on(−1,1)exceedsM even though it can be identically zero on an interval not containing 0.
Example 2.3.In the casen=2,B=1,ε=0, (5) becomes the linear inequalityu1 and the number M=14 agrees with Lemma 2 of [5], which was proved there using a Maximum Principle argument.
By applying Lemma 2.1 to the Laplacian of a power ofu, we get the following generalization.
Theorem 2.4.Given constantsB >0,C∈R, andε <1, let M=
⎧⎨
⎩
(2(2(ε−B(1C)−+ε)n(12−ε)))1−1ε ifCε, (B(12n−ε))1−1ε ifCε.
Suppose the functionu:D1→Rsatisfies:
• uis continuous onD1,
• u(x) 0forx∈D1,
• on the open setω= {x∈D1: u(x) =0},u∈C2(ω),
• forx∈ω:
u(x)u( x) Bu(x) 1+ε+C∇u(x) 2.
Ifu(0)=0, thensupx∈D1u(x) > M.
Proof. Letμ=min{ε, C}, soμε <1, and on the setω, u(x)u( x) Bu(x) 1+ε+μ∇u(x) 2.
Consider the functionu1−μonD1, sou1−μ∈C0(D1)∩C2(ω), and on the setω,
u1−μ
=(1−μ)u−μ−1
uu−μ| ∇u|2 (1−μ)u−μ−1Bu1+ε
=(1−μ)B
u1−μ(ε−μ)/(1−μ)
.
Since(1−μ)B >0, andμε <1⇒0ε1−−μμ<1, Lemma 2.1 applies tou1−μ. If(u(0))1−μ=0, then supu1−μ>
(1−μ)B(1−ε1−−μμ)2 2(2ε1−−μμ+n(1−ε1−−μμ))
1
1−ε−μ
1−μ ⇒ supu >
B(1−ε)2 2(2(ε−μ)+n(1−ε))
1
1−ε
. 2
Functions satisfying a differential inequality of the form (1) or (2) also satisfy a Strong Maximum Principle; the only condition isB >0.
Theorem 2.5. Given any open set Ω⊆Rn, and any constantsB >0,C, ε∈R, suppose the functionu:Ω →R satisfies:
• uis continuous onΩ,
• on the setω= {x∈Ω: u(x) > 0},u∈C2(ω),
• on the setω,usatisfies
uu−B|u|1+ε−C| ∇u|20.
Ifu(x0) >0for somex0∈Ω, thenudoes not attain a maximum value onΩ.
Proof. Note that the constant functionu≡0 is the only locally constant solution of the inequality forB >0. IfB=0 then obviously any constant function would be a solution.
Given a functionusatisfying the hypotheses,ωis a nonempty open subset ofΩ. Suppose, toward a contradiction, that there is somex1∈Ω withu(x) u(x1)for allx∈Ω. In particular,u(x1)u(x0) >0, sox1∈ω. Letω1be the connected component ofωcontainingx1.
Forx∈ω1,usatisfies the linear, uniformly elliptic inequality u(x) +
−B
u(x)ε−1 u(x) +
−C∇u(x) u(x)
· ∇u(x) 0,
where the coefficients (defined in terms of the given u) are locally bounded functions of x, and (−B(u(x)) ε−1) is negative for all x∈ω. It follows from the Strong Maximum Principle [4, Theorem 3.5] that since u attains a maximum value at x1, then u is constant onω1. Since the only constant solution is 0, it follows that u(x1)=0, a contradiction. 2
The next lemma shows how an inequality like (5) withn=2 can arise from a first order PDE for a complex-valued function.
Lemma 2.6.Consider constantsα,γ∈Rwith0< α <1. Letω⊆Cbe an open set, and supposeh:ω→Csatisfies:
• h∈C1(ω),
• h(z)=0for allz∈ω,
• ∂h∂¯z= |h|αonω.
Then, the following inequality is satisfied onω:
|h|(1−α)γ
4(1−α)γ −(2−α)2
|h|(1−α)(γ−2). (6)
Remark.The special caseα=12,γ=32is Lemma 1 of [5]; its proof there is a long calculation in polar coordinates, which can be generalized to some other values ofαby an analogous argument. However, usingz,z¯coordinates allows for a shorter calculation.
Proof of Lemma 2.6. We first want to show that h is smooth on ω, applying the regularity and bootstrapping technique of PDE to the equation ∂h/∂z¯= |h|α. We recall the following fact (for a more general statement, see
Theorem 15.6.2 of [1]): for a nonnegative integer , and 0< β <1, ifϕ∈Cloc ,β(ω)andg:ω→Chas first deriva- tives inL2loc(ω)and is a solution of∂g/∂z¯=ϕ, theng∈Cloc+1,β(ω). In our case,ϕ= |h|α∈C1(ω)⊆Cloc0,β(ω)(since h∈C1(ω)and is nonvanishing), andg=hhas continuous first derivatives, so we can conclude thatg=h∈Cloc1,β(ω).
Repeating gives thath∈Cloc2,β(ω), etc.
Since the conclusion is a local statement, it is enough to expressωas a union of open subsetsωk and establish the conclusion on each subset. For eachzk∈ω, there is a sufficiently small diskωk containingzk, where real exponenti- ation ofh(z)is well defined onωk, by choosing a single-valued branch of log to definehr=exp(rlog(h)).
The condition ∂h∂z¯= |h|α can be re-written hz¯=(h)¯ z= |h|α=hα/2h¯α/2.
This leads to
hzz¯=(hz¯)z=
hα/2h¯α/2
z
=α 2
h(α/2)−1h¯α/2hz+hαh¯α−1
= (h)¯ zz¯
,
which is used in a line of the next step. For an arbitrary exponentm∈R, |h|m
z¯z=
hm/2h¯m/2
z¯z
= ∂
∂z m
2hm2−1h¯zh¯m2 +hm2 m
2h¯m2−1(h)¯ z¯
=m 2
∂
∂z
hm2−1+α2h¯m2+α2 +hm2h¯m2−1(h)¯ ¯z
=m 2
m 2 +α
2 −1
hm2+α2−2hzh¯m2+α2 +hm2+α2−1 m
2 +α 2
h¯m2+α2−1(¯h)z
+m
2hm2−1hzh¯m2−1(¯h)z¯+hm2 m
2 −1
h¯m2−2(h)¯ z(h)¯ ¯z+hm2h¯m2−1(h)¯ z¯z
.
=m 2
Re
(m+α−2)|h|m+α−4h¯2hz +
m 2 +α
|h|m+2α−2+m
2|h|m−2|hz|2
.
With the aim of applying Lemma 2.1 to the function|h|m, we consider the expression (8), with real constantsB,ε, andm=0. In line (9), we assign
ε= 1
m(m+2α−2) (7)
to be able to combine like terms, and in line (10), we chooseB=4m−(2−α)2to complete the square.
|h|m
−B
|h|mε
(8)
=4
|h|m
zz¯−B|h|mε
=2m
Re
(m+α−2)|h|m+α−4h¯2hz +
m 2 +α
|h|m+2α−2+m
2|h|m−2|hz|2
−B|h|mε
=
m(m+2α)−B
|h|m+2α−2 (9)
+Re
2m(m+α−2)|h|m+α−4h¯2hz
+m2|h|m−2|hz|2 |h|m−2
m2+2αm−B
|h|2α−2|m||m+α−2||h|α|hz| +m2|hz|2
= |h|m−2
|m+α−2||h|α− |m||hz|2
0. (10)
Considering the form of (7), it is convenient to choosem=(1−α)γ for some constantγ=0. The claim of the lemma follows; theγ=0 case can be checked separately. 2
The parameter γ can be chosen arbitrarily large; to apply Lemma 2.1 to get the “no small solutions” result of Theorem 1.1, we need the RHS coefficient in (6) to be positive, so γ >(24(1−−α)α)2, and also the RHS exponent(1− α)(γ −2)to be nonnegative, so γ2. In contrast, theα=12,γ=32 case appearing in Lemma 1 of [5] has RHS exponent−14. The approach of Theorem 2 of [5] is to use the negative exponent together with the result of Example 2.3 to show that assuminghhas a small solution leads to a contradiction. As claimed, their method can be generalized to apply to other nonpositive exponents, but 4(1(2−−α)α)2 < γ 2 holds only forα <2(√
2−1)≈0.8284.
Proof of Theorem 1.1. Given a continuoush:D1→Csatisfying the hypotheses of Theorem 1.1, on the setω= {z∈D1: h(z)=0},h∈C1(ω), and the conclusion of Lemma 2.6 can be re-written:
|h|(1−α)γ
4(1−α)γ −(2−α)2
|h|(1−α)γ1−2γ. (11) The hypotheses of Lemma 2.1 are satisfied withn=2,u(x, y)= |h(x+iy)|(1−α)γ, andu(0)=0, when the RHS of (11) has a positive coefficient (soγ >(24(1−−α)α)2) and the quantityε=1−γ2 is in[0,1)(forγ 2). The conclusion of Lemma 2.1 is:
sup
z∈D1
h(z)(1−α)γ> M= 1
4·
4(1−α)γ−(2−α)2
· 2
γ 2γ /2
⇒ sup
z∈D1
h(z)>
4(1−α)γ −(2−α)2 γ2
1
2(1−α)
.
We can optimize this lower bound, using elementary calculus to show that the maximum value of 4(1−α)γ−(2−α)2
γ2
is achieved at the critical pointγ = (22(1−−α)α)2 >max{2,(24(1−−α)α)2}, and the lower bound for the sup isSα as appearing in (4). 2
Note thatSα is decreasing for 0< α <1, withS1/2=49,S2/3=18, andSα→0 asα→1−. This theorem is used in the proof of Theorem 4.3.
Example 2.7.As noted by [5], a 1-dimensional analogue of Eq. (3) in Theorem 1.1 is the well-known (for example, [2, §I.9]) ODEu(x)=B|u(x)|α for 0< α <1 andB >0, which can be solved explicitly. By an elementary separa- tion of variables calculation, the solution on an interval whereu=0 is|u(x)| =(±(1−α)(Bx+C))1−α1 . The general solution on the domainRis, forc1< c2,
u(x)=
⎧⎪
⎨
⎪⎩
(1−α)1−1α(B(x−c2))1−1α ifxc2,
0 ifc1xc2,
−(1−α)1−α1 (B(c1−x))1−α1 ifxc1.
Sou∈C1(R), and ifu(0)=0, then sup−1<x<1|u(x)|> ((1−α)B)1−α1 . 3. Lemmas for holomorphic maps
We continue with the DR notation for the open disk in the complex plane centered at the origin. The following quantitative lemmas on inverses of holomorphic functionsDR→Care used in a step of the proof of Theorem 4.3 where we put a mapDr→C2into a normal form, (14).
Lemma 3.1.(See [3, Exercise I.1].) Supposef :D1→D1is holomorphic, withf (0)=0,|f(0)| =δ >0. For any η∈(0, δ), lets=(1δ−−ηδη)η;then the restricted functionf:Dη→D1takes on each valuew∈Ds exactly once. 2
The hypotheses implyδ1 by the Schwarz Lemma.
Lemma 3.2. For a holomorphic map Z1:Dr →D2 with Z1(0)=0, Z1(0)=1, if r > 4
√2
3 then there exists a continuous functionφ:D1→Drwhich is holomorphic onD1and which satisfies(Z1◦φ)(z)=zfor allz∈D1. Remark.It follows from the Schwarz Lemma thatr2, and it follows from the fact thatφis an inverse ofZ1that φ(0)=0 andφ(0)=1.
Proof of Lemma 3.2. Define a new holomorphic functionf :D1→D1by f (z)=1
2·Z1(r·z),
sof (0)=0,f(0)=2r, and Lemma 3.1 applies withδ=r2. If we chooseη=3r8, thens=64−3r12r2 2, and the assumption r >4
√2
3 impliess >12. It follows from Lemma 3.1 that there exists a functionψ:Ds→Dηsuch that(f◦ψ )(z)=z for allz∈D1/2⊆Ds; this inverse functionψis holomorphic onD1/2. The claimed functionφ:D1→Drη⊆Dr is defined byφ(z)=r·ψ (12·z), so forz∈D1,
Z1 φ(z)
=Z1
r·ψ 1
2·z
=2·f
ψ 1
2·z
=2·1
2 ·z=z. 2 4. J-holomorphic disks
For S > 0, consider the bidisk ΩS =D2 ×DS ⊆C2, as an open subset of R4, with coordinates x = (x1, y1, x2, y2)=(z1, z2)and the trivial tangent bundleT ΩS⊆TR4. Consider an almost complex structureJ on ΩS given by a complex structure operator onTxΩSof the following form:
J (x) =
⎛
⎜⎝
0 −1 0 0
1 0 0 0
0 λ 0 −1
λ 0 1 0
⎞
⎟⎠, (12)
whereλ:ΩS→Ris any function.
A differentiable map Z:Dr →ΩS is aJ-holomorphic disk ifdZ◦Jstd=J◦dZ, whereJstd is the standard complex structure onDr⊆C. Letz=x+iybe the coordinate onDr. ForJ of the form (12), ifZ(z)is defined by complex-valued component functions,
Z:Dr→ΩS:Z(z)=
Z1(z), Z2(z)
, (13)
then theJ-holomorphic property implies thatZ1:Dr→D2is holomorphic in the standard way.
Example 4.1.If the functionλ(z1, z2)satisfiesλ(z1,0)=0 for allz1∈D2, then the mapZ:D2→ΩS:Z(z)=(z,0) is aJ-holomorphic disk.
Definition 4.2.The Kobayashi–Royden pseudonorm onΩSis a functionT ΩS→R:(x, v) → (x, v)K, defined on tangent vectorsv∈TxΩSto be the number
glb 1
r: ∃aJ-holomorphicZ:Dr→ΩS, Z(0)= x, dZ(0) ∂
∂x
= v
.
Under the assumption thatλ∈C0,α(ΩS), 0< α <1, it is shown by [6] and [7] that there is a nonempty set ofJ- holomorphic disks throughxwith tangent vectorvas in the definition, so the pseudonorm is a well-defined function.
Further, each such disk satisfiesZ∈C1(Dr).
At this point we pickα∈(0,1)and setλ(z1, z2)= −2|z2|α. LetS=Sα>0 be the constant defined by formula (4) from Theorem 1.1. Then,(ΩS, J )is an almost complex manifold with the following property:
Theorem 4.3.If0=b∈DSthen(0, b), (1,0)K 3
4√ 2. Remark. Since 3
4√
2 ≈0.53, and (0,0), (1,0)K 12 by Example 4.1, the theorem shows that the Kobayashi–
Royden pseudonorm is not upper semicontinuous onT ΩS.
Proof. Consider a J-holomorphic map Z :Dr →ΩS of the form (13), and suppose Z(0)=(0, b)∈ΩS and dZ(0)(∂x∂ )=(1,0). Then the holomorphic functionZ1:Dr→D2satisfiesZ1(0)=0,Z1(0)=1, andZ2∈C1(Dr) satisfiesZ2(0)=b.
Suppose, toward a contradiction, that there exists such a mapZwithb=0 andr >4
√2
3 . Then Lemma 3.2 applies toZ1: there is a re-parametrizationφwhich putsZinto the following normal form:
(Z◦φ):D1→ΩS, z→
Z1 φ(z)
, Z2 φ(z)
= z, f (z)
, (14)
wheref =Z2◦φ:D1→DS satisfiesf ∈C0(D1)∩C1(D1). From the fact thatZ◦φisJ-holomorphic onD1, it follows from the form (12) ofJ that iff (z)=u(x, y)+iv(x, y), thenf satisfies this system of nonlinear Cauchy–
Riemann equations onD1: du
dy = −dv
dx and du dx +λ
z, f (z)
=dv
dy (15)
with the initial conditionsf (0)=b,ux(0)=uy(0)=vx(0)=0 andvy(0)=λ(0, b)= −2|b|α. The system of equa- tions implies
∂f
∂z¯ =1 2
∂
∂x(u+iv)+i ∂
∂y(u+iv)
=1 2
ux−vy+i(vx+uy)
= −1 2λ
z, f (z)
= |f|α. (16)
So, Theorem 1.1 applies, withf =h. The conclusion is that sup
z∈D1
f (z)> Sα,
but this contradicts|f (z)|< S=Sα. 2
The previously mentioned existence theory for J-holomorphic disks shows there are interesting solutions of Eq. (16), and therefore also the inequality (11).
Example 4.4. For 0< α <1, (ΩS, J ), λ(z1, z2)= −2|z2|α as above, a map Z:Dr →ΩS of the form Z(z)= (z, f (z)) isJ-holomorphic iff (x, y)=u(x, y)+iv(x, y)is a solution of (15). Again generalizing theα=12 case of [5], examples of such solutions can be constructed (for smallr) by assumingv≡0 andudepends only onx, so (15) becomes the ODEu(x)−2|u(x)|α=0. This is the equation from Example 2.7; we can conclude thatJ-holomorphic disks inΩSdo not have a unique continuation property.
Acknowledgement
The authors acknowledge the insightful remarks of the referee, whose suggestions led to a shorter proof of Lemma 2.6 and an improvement of Theorem 2.4.
References
[1] K. Astala, T. Iwaniec, G. Martin, Elliptic Partial Differential Equations and Quasiconformal Mappings in the Plane, Princeton Math. Ser., vol. 48, Princeton University Press, Princeton, 2009, MR 2472875 (2010j:30040), Zbl 1182.30001.
[2] G. Birkhoff, G.-C. Rota, Ordinary Differential Equations, Ginn & Co., 1962, MR 0138810 (25 #2253), Zbl 0102.29901.
[3] J. Garnett, Bounded Analytic Functions, rev. first ed., Grad. Texts in Math., vol. 236, Springer, 2007, MR 2261424 (2007e:30049), Zbl 1106.30001.
[4] D. Gilbarg, N. Trudinger, Elliptic Partial Differential Equations of Second Order, Springer CIM, 2001, MR 1814364 (2001k:35004), Zbl 1042.35002.
[5] S. Ivashkovich, S. Pinchuk, J.-P. Rosay, Upper semi-continuity of the Kobayashi–Royden pseudo-norm, a counterexample for Hölderian almost complex structures, Ark. Mat. (2) 43 (2005) 395–401, MR 2173959 (2006g:32038), Zbl 1091.32009.
[6] S. Ivashkovich, J.-P. Rosay, Schwarz-type lemmas for solutions of∂-inequalities and complete hyperbolicity of almost complex manifolds, Ann. Inst. Fourier (Grenoble) 54 (7) (2004) 2387–2435, MR 2139698 (2006a:32032), Zbl 1072.32007.
[7] A. Nijenhuis, W. Woolf, Some integration problems in almost-complex and complex manifolds, Ann. of Math. (2) 77 (1963) 424–489, MR 0149505 (26 #6992), Zbl 0115.16103.