ON HOPF’S LEMMA AND THE STRONG MAXIMUM PRINCIPLE 5 0 0 S.BERTONE,A.CELLINA,ANDE.M.MARCHINI 2 n a Abstract. InthispaperweconsiderHopf’sLemmaandtheStrongMaximum J Principleforsupersolutionsto N 8 2 gi(u2xi)uxixi =0 i=1 X ] undersuitablehypotheses thatallowgi toassumevaluezeroatzero. P A . h t 1. Introduction a m LetΩ RN beaconnected,openandboundedset;wecallΩregularifforevery ⊂ [ z ∂Ω, there exists a tangent plane, continously depending on z. We say that Ω ∈ satisfies the interior ball condition at z if there exists an open ball B Ω with 1 ⊂ z ∂B. On Ω, consider the operator v ∈ 3 N 0 (1) F(u)= g (u2 )u , 5 i xi xixi i=1 1 X 0 where gi :[0,+ ) [0,+ ) are continuous functions. ∞ → ∞ 5 When F is elliptic, two classical results hold. 0 / h Hopf’s Lemma: t LetΩ beregular, letu besuchthat F(u) 0on Ω. Supposethat thereexistsz ∂Ω a ≤ ∈ m such that u(z)<u(x), for all x in Ω. : v If, in addition, Ω satisfies the interior ball condition at z, we have i X ∂u r (z)<0, a ∂ν where ν is the outer unit normal to B at z. The Strong Maximum Principle: LetubesuchthatF(u) 0onΩ,thenifuitattainsminimuminΩ,itisaconstant. ≤ In 1927 Hopf proved the Strong Maximum Principle in the case of second order elliptic partial differential equations, by applying a comparison technique, see [11]. For the class of quasilinear elliptic problems, many contributions have been given, to extend the validity of the previous results, as in [1, 5, 6, 8, 9, 10, 12, 13, 14, 15, 16, 18]. Key words and phrases. Strongmaximumprinciple. 1 2 S.BERTONE,A.CELLINA,ANDE.M.MARCHINI In the case in equation (1) we have g 1, for every i, then F(u)=∆u, and we i ≡ find the classical problem of the Laplacian, see [7, 9]. On the other hand, when there exists i 1,...,N such that g 0 on an i interval I = [0,T] R, the Strong Maximum∈ {Principle}does not hold.≡Indeed, in ⊂ this case, it is always possible to define a function u assuming minimum in Ω and such that N g (u2 )u = 0. For instance, let g (t) = 0, for every t [0,2]. i=1 i xi xixi N ∈ The function P (x2 1)4 if 1 x 1 u(x1,...,xN)= −0 N − oth−erw≤ise N ≤ , (cid:26) satisfies (1) in RN. We are interested in the case when 0 g (t) 1 and it does not exist i such i ≤ ≤ that g 0 on an interval. Since g could assume value zero, the equation (1) is i i ≡ non elliptic. The results known so far, for the validity of Hopf’s Lemma and of the Strong Maximum Principle, suggest that, for possibly non elliptic equations, but arising from a functional having rotational symmetry, this validity shall depend only on the behaviour of the functions g near zero, see [4]. i In this paper, we prove, in section 3, a sufficient condition for the validity of the Hopf’s Lemma and of the Strong Maximum Principle; a remarkable feature of this condition is that it concerns only the behaviour of the function g (t) that goes i fastesttozero,astgoestozero. Hopf’slemmaandtheStrongMaximumPrinciple areessentiallythesameresultaslongaswecanbuildsubsolutionswhoselevellines canhavearbitrarilylargecurvature. Thisneednotbealwayspossibleforproblems not possessing rotational symmetry. This difficulty will be evident in sections 4 and 5. In these sections, a more restricted class of equations is considered, namely when all the functions g , for i = 1,...,N 1, are 1 and only g is allowed to go i N − to zero. In this simpler class of equations we are able to show that the condition (g (t))3/2 N lim >0 t→0+ tgN′ (t) is at once necessaryfor the validity ofHopf’s Lemma and sufficientfor the validity of the Strong Maximum Principle. 2. Preliminary results We impose the following local assumptions. Assumptions (L): There exists t>0 such that: i) on [0,t], for every i=1,...,N 1, − 0 g (t) g (t) 1; N i ≤ ≤ ≤ ii) g is continuous on [0,t]; positive and differentiable on (0,t]; N iii) on (0,t], the function t g (t)+g (t)t is non decreasing. → N N′ ON HOPF’S LEMMA AND THE STRONG MAXIMUM PRINCIPLE 3 Notice that, in case ii) above is violated, the Strong Maximum Principle does nothold;andthatconditioniii)aboveincludesthecaseoftheLaplacian,g (t) 1; i ≡ and, finally, that under these assumptions, g could assume value zero at most for i t=0. Moreover,we can consider the equation N g (u2 )u =0 i xi xixi i=1 X as the Euler-Lagrangeequation associated to the functional N 1 J(u)= L( u)dΩ= f (u2 )u2 dΩ, ZΩ ∇ ZΩ 2 i=1 i xi xi! X whereL( u)isstrictlyconvexin (u ,...,u ):u2 t, for every i=1,...,N . ∇ { x1 xN xi ≤ } Indeed, fix i. Let f be a solution to the differential equation i (2) g (t)=f (t)+5tf (t)+2t2f (t), i i i′ i′′ for t [0,t]. Since ∈ ∂2L (u2 )=f (u2 )+5u2 f (u2 )+2u4 f (u2 )=g (u2 ), ∂u2 xi i xi xi i′ xi xi i′′ xi i xi xi we have that N ∂2L N div L( u)= u = g (u2 )u . ∇∇u ∇ ∂u2 xixi i xi xixi i=1 xi i=1 X X Moreover,the strict convexity of L( u) in (u ,...,u ):u2 t, for every i= ∇ { x1 xN xi ≤ 1,...,N follows by the fact that g is positive in (0,t]. i } Sincewewillneedgeneralcomparisontheoremsthatdependontheglobalprop- erties of the solutions, i.e. on their belonging to a Sobolev space, we will need also a growth assumption on g (assumption (G)) to insure these properties of the i solutions. Assumption (G): Each function f as defined in (2), is bounded and f (u2 )u2 is strictly convex. i i xi xi Any function g satisfying assumptions (L) on [0,t] can be extended so as to i satisfy assumption (G) on [0,+ ). In fact, it is enough to extend g to (t,+ ) i ∞ ∞ by setting g (t)=f (t), for t>t. i i Definition 1. Let Ω be open, and let u W1,2(Ω). The map u is a weak solution ∈ to the equation F(u)=0 if, for every η C (Ω), ∈ 0∞ L( u(x)), η(x) dx =0. h∇ ∇ ∇ i ZΩ u is a weak subsolution (F(u) 0) if, for every η C (Ω), η 0, ≥ ∈ 0∞ ≥ L( u(x)), η(x) dx 0. h∇ ∇ ∇ i ≤ ZΩ 4 S.BERTONE,A.CELLINA,ANDE.M.MARCHINI u is a weak supersolution (F(u) 0) if, for every η C (Ω), η 0, ≤ ∈ 0∞ ≥ L( u(x)), η(x) dx 0. h∇ ∇ ∇ i ≥ ZΩ We say that a function w W1,2(Ω) is such that w 0 if w+ W1,2(Ω). ∈ |∂Ω ≤ ∈ 0 The growth assumption (G) assures that, if u W1,2(Ω), then L( u(x)) ∈ ∇ ∇ ∈ L2(Ω). The strict convexity of L implies the following comparison lemma. Lemma 1. Let Ω be a open and bounded set, let v W1,2(Ω) be a subsolution and ∈ let u W1,2(Ω) be a supersolution to the equation F(u) = 0. If v u , then ∂Ω ∂Ω v u∈a.e. in Ω. | ≤ | ≤ We wish to express the operator N F(v)= g (v2 )v i xi xixi i=1 X in polar coordinates. Set x =ρcosθ ...cosθ cosθ 1 N 1 2 1 − x =ρcosθ ...cosθ sinθ 2 N 1 2 1 − ... x =ρsinθ N N 1 − so that 2 2 x x v x i i ρ i v =v and v =v + 1 . xi ρ ρ xixi ρρ(cid:18) ρ (cid:19) ρ " −(cid:18) ρ (cid:19) # When v is a radial function, F reduces to N 2 2 2 x x v x F(v)= g v2 i v i + ρ 1 i = i=1 i ρ(cid:18) ρ (cid:19) !" ρρ(cid:18) ρ (cid:19) ρ −(cid:18) ρ (cid:19) !# X N 2 2 N 2 2 x x v x x v g v2 i i + ρ g v2 i 1 i . ρρ i=1 i ρ(cid:18) ρ (cid:19) !(cid:18) ρ (cid:19) ρ i=1 i ρ(cid:18) ρ (cid:19) ! −(cid:18) ρ (cid:19) ! X X In general, we do not expect that the equation F(v) = 0 admits radial solutions. Howeverwe will use the expressionof F validfor radialfunctions in orderto reach our results. The following technical lemmas will be used later. Lemma 2. Let n=2,...,N and set t(1 a) h (a)=g − (1 a)+g (ta)a. n N N n 1 − (cid:18) − (cid:19) For every 0 < t t (t defined in assumptions (L)), h (a) h (1/n), for every a n n ≤ ≥ in [0,1]. Proof. Since, on (0,t], the function t g (t)+g (t)t is non decreasing, we have → N N′ that t(1 a) t(1 a) t(1 a) h′n(a)=−gN n −1 −gN′ n −1 n −1 +gN(ta)+gN′ (ta)ta≥0 (cid:18) − (cid:19) (cid:18) − (cid:19) − ON HOPF’S LEMMA AND THE STRONG MAXIMUM PRINCIPLE 5 if and only if a 1/n, so that h (a) h (1/n), for every a [0,1]. n n ≥ ≥ ∈ (cid:3) Lemma 3. For every 0<t t (t defined in assumptions (L)), we have that ≤ N 2 2 x x t i i g t g . N N i=1 (cid:18) ρ (cid:19) !(cid:18) ρ (cid:19) ≥ (cid:18)N(cid:19) X Proof. We prove the claim by induction on N. Let N =2. Set a=sin2θ . Applying Lemma 2 we obtain that 1 2 2 2 2 x x x x 1 1 2 2 g t +g t = N N ρ ρ ρ ρ (cid:18) (cid:19) !(cid:18) (cid:19) (cid:18) (cid:19) !(cid:18) (cid:19) t g (t(1 a))(1 a)+g (ta)a g . N N N − − ≥ 2 (cid:18) (cid:19) Suppose that the claim is true for N 1, i.e. − N 1 2 2 − xi xi t g t g . N N i=1 (cid:18) ρ (cid:19) !(cid:18) ρ (cid:19) ≥ (cid:18)N −1(cid:19) X Let us prove it for N. Set y =ρcosθ ...cosθ cosθ 1 N 2 2 1 − y =ρcosθ ...cosθ sinθ 2 N 2 2 1 − ... y =ρsinθ N 1 N 2 − − and set a=sin2θN 1. Applying Lemma 2 we obtain that − N 2 2 x x i i g t = N ρ ρ i=1 (cid:18) (cid:19) !(cid:18) (cid:19) X N 1 2 2 − yi yi g t (1 a) (1 a)+g (ta)a N N i=1 (cid:18)ρ(cid:19) − !(cid:18)ρ(cid:19) − ≥ X t(1 a) t g − (1 a)+g (ta)a g , N N N N 1 − ≥ N (cid:18) − (cid:19) (cid:18) (cid:19) and the claim is proved. (cid:3) 3. A sufficient condition for the validity of Hopf’s Lemma and of the Strong Maximum Principle Consider the improper Riemann integral ξ g (ζ2/N) ξ g (ζ2/N) N N dζ = lim dζ ζ ζ Z0 ξ→0Zξ as an extended valued function G, b b ξ g (ζ2/N) N G(ξ)= dζ, ζ Z0 where we mean that G(ξ) + whenever the integral diverges. ≡ ∞ 6 S.BERTONE,A.CELLINA,ANDE.M.MARCHINI We wish to prove the following lemma. Lemma 4 (Hopf’s Lemma). Let Ω RN be a connected, open and bounded set. ⊂ Let u W1,2(Ω) C Ω be a weak solution to ∈ ∩ (cid:0) (cid:1) N g (u2 )u 0. i xi xixi ≤ i=1 X In addition to the assumptions (L) and (G) on g , assume that G(ξ) + . Sup- i ≡ ∞ pose that there exists z ∂Ω such that ∈ u(z)<u(x), for all x in Ω and that Ω satisfies the interior ball condition at z. Then ∂u (z)<0, ∂ν where ν is the outer unit normal to B at z. As an example of an equation satisfying the assumptions of the theorem above, considertheLaplaceequation∆u=0. Thefunctionsg 1satisfytheassumptions i ≡ (L) and (G), and ξ 1 G(ξ)= dζ =+ . ζ ∞ Z0 Another example is obtained setting 1 g (t)= N ln(t) | | for 0 t 1/e. The assumptions (L) and (G) are satisfied; moreover, for 0 ≤ ≤ ≤ ξ2/N 1/e, ≤ ξ dζ G(ξ)= =+ . ζ ln(ζ2/N) ∞ Z0 | | Proof of Lemma 4. a) Assume that u(z)=0 and that B =B(O,r). We prove the claim by contradiction. Suppose that ∂u (z) 0, ∂ν ≥ where ν is the outer unit normal to B at z. Let ǫ = min u(x):x B(O,r/2) ; ∈ we have that ǫ>0. Set n o ω =B(O,r) B(O,r/2). \ b) We seek a radial function v W1,2(ω) C(ω) satisfying ∈ ∩ v is a weak solution to F(v) 0 in ω ≥ v >0 in ω (3) v =0 in ∂B(O,r) v ǫ in ∂B(O,r/2) ≤ v (z)<0. ρ ON HOPF’S LEMMA AND THE STRONG MAXIMUM PRINCIPLE 7 Consider the Cauchy problem N 1 ζ ζ = − ′ − ρ g (ζ2/N) N (4) ζ(r/2)= ǫ. −r There exists a unique localsolution ζ of (4), such that ζ(ρ) g (ζ2/N) ρ N 1 2ρ N dζ = − ds= (N 1)ln . ζ − s − − r Z−rǫ Zr/2 (cid:18) (cid:19) We claim that ζ is defined in [r/2,+ ). Indeed, suppose that ζ is defined in ∞ [r/2,τ), with τ < + . Since ζ > 0, ζ is an increasing function, so that τ < + ′ ∞ ∞ if and only if lim ζ(ρ)=0. But ρ τ → ζ(ρ) g (ζ2/N) 2ρ N = lim dζ = lim (N 1)ln , −∞ ρ→τZ−rǫ ζ ρ→τ− − (cid:18) r (cid:19) a contradiction. Hence, the solution ζ of (4) is defined in [r/2,+ ). ∞ Setting v =ζ, since, for every ρ (r/2,r), ρ ∈ ǫ <v (ρ)<0, ρ −r we have that the function ρ v(ρ)= v (s)ds ρ Zr solves the problem v2 v ρ ρ v g + (N 1)=0, ρρ N N! ρ − in particular,v(ρ)>0 and v (ρ)<0, for every ρ (r/2,r), v(r)=0 and v(r/2) ρ ∈ ≤ ǫ. Since v 0 and √t v 0, for every ρ (r/2,r), by the hypotheses on g ρρ ρ i ≥ − ≤ ≤ ∈ and by Lemma 3, we have that N 2 2 N 2 2 x x v x x F(v)=v g v2 i i + ρ g v2 i 1 i ρρi=1 i ρ(cid:18) ρ (cid:19) !(cid:18) ρ (cid:19) ρ i=1 i ρ(cid:18) ρ (cid:19) ! −(cid:18) ρ (cid:19) !≥ X X N 2 2 N 2 x x v x v g v2 i i + ρ 1 i = ρρi=1 N ρ(cid:18) ρ (cid:19) !(cid:18) ρ (cid:19) ρ i=1 −(cid:18) ρ (cid:19) ! X X N x 2 x 2 v v2 v v g v2 i i + ρ(N 1) v g ρ + ρ(N 1). ρρi=1 N ρ(cid:18) ρ (cid:19) !(cid:18) ρ (cid:19) ρ − ≥ ρρ N N! ρ − X The function v solves (3), indeed, v is in C2(ω) and it is such that F(v) 0 and ≥ v >0 in ω, v(r)=0, v(r/2) ǫ and v (z)<0. ρ ≤ c) Since u,v W1,2(ω) C(ω), v is a weaksubsolutionand u is a weak solution ∈ ∩ toF(u)=0,andv u ,applyingLemma1,weobtainthatv u inω. From ∂ω ∂ω | ≤ | ≤ ∂v ∂u v (z)= (z)< (z), ρ ∂ν ∂ν it follows that there exists x0 ω such that v(x0)>u(x0), a contradiction. ∈ 8 S.BERTONE,A.CELLINA,ANDE.M.MARCHINI (cid:3) From Hopf’s Lemma we derive: Theorem 1 (Strong Maximum Principle). Let Ω RN be a connected, open and ⊂ bounded set. Let u W1,2(Ω) C Ω be a weak supersolution to ∈ ∩ N (cid:0) (cid:1) g (u2 )u =0. i xi xixi i=1 X In addition to the assumptions (L) and (G) on g , assume that G(ξ) + . Then, i ≡ ∞ if u attains its minimum in Ω, it is a constant. Proof. a) Assume min u = 0 and set = x Ω : u(x) = 0 . By contradiction, Ω C { ∈ } suppose that the open set Ω = . \C 6 ∅ b)SinceΩisaconnectedset,thereexists andR>0suchthatB(s,R) Ω ∈C ⊂ and B(s,R) (Ω )= . Let p B(s,R) (Ω ). Consider the line ps. Moving ∩ \C 6 ∅ ∈ ∩ \C p along this line, we can assume that B(p,d(p, )) (Ω ) and that there exists C ⊂ \C one point z such that d(p, )= d(p,z). Set r = d(p, ). W.l.o.g. suppose that ∈C C C p=O. c) The set Ω satisfies the interior ball condition at z, hence Hopf’s Lemma \C implies ∂u (z)<0. ∂ν But this is a contradiction: since u attains minimum at z Ω, we have that ∈ Du(z)=0. (cid:3) 4. A necessary condition for the validity of Hopf’s Lemma In this and the following section we consider the operator N 1 − (5) F(u)= u +g(u2 )u , xixi xN xNxN i=1 X We wish to provide a necessary condition for the validity of Hopf’s Lemma in a class of non elliptic equations. Consider the case ξ g(ζ2/N) G(ξ)= dζ <+ . ζ ∞ Z0 Theorem2. Considertheoperator(5),whereg satisfiesassumptions(L)and(G), and on (0,t] (t defined in assumptions (L)), g (t)>0 and g(t)+g (t)t 1. ′ ′ ≤ If (g(t))3/2 lim =0, t→0+ tg′(t) ON HOPF’S LEMMA AND THE STRONG MAXIMUM PRINCIPLE 9 then there exist: an open regular region Ω RN; a radial function u C2(Ω) such ⊂ ∈ that F(u) 0 in Ω and a point z ∂Ω such that u(z) = 0, u(z) u(x) for every ≤ ∈ ≤ x Ω and ∈ ∂u (z)=0 ∂ν where ν is the outer unit normal to Ω at z. If, in addition, we assume that g(t) (6) is bounded in (0,t] tg (t) ′ then Ω satisfies the interior ball condition at z. Remark 1. When lim g (t)t exists, it follows that t 0+ ′ → (g(t))3/2 lim t 0+ tg′(t) → exists, and that g(t)+g (t)t 1, on (0,t]. ′ ≤ Indeed, we have that lim (g(t)+g′(t)t)=0. t 0+ → Otherwise, there exists K >0 such that, when 0<t t, g (t)t K, so that ′ ≤ ≥ t g(t)t= (g(s)+g (s)s)ds Kt, ′ ≥ Z0 and g(t) K. From ≥ ξ2 g(t) dt<+ , t ∞ Z0 it follows that lim g(t)=0, a contradiction. t 0+ → The map 1 g(t)= , ln(t)k | | with k >2, for 0 t 1/e, satisfies the assumption ≤ ≤ (g(t))3/2 lim =0. t→0+ tg′(t) The following lemma is instrumental to the proofs of the main results. Lemma 5. Let g satisfies assumptions (L) and (G). Suppose that for every 0 < t t, ≤ g (t) 0 and g(t)+g (t)t 1. ′ ′ ≥ ≤ Set k (a)=(1 a)+ag(ta) and k (a)= a (1 a)g(ta) 1 2 − − − − For every 0 < t t (t defined in assumptions (L)), k and k are non increasing 1 2 ≤ in [0,1]. 10 S.BERTONE,A.CELLINA,ANDE.M.MARCHINI Proof. Since, for every 0<t t, ≤ g′(t) 0 and g(t)+g′(t)t 1, ≥ ≤ we have that, for every 0 a 1, ≤ ≤ k1′(a)=−1+g(ta)+g′(ta)ta≤0 and k2′(a)=−1+g(ta)−(1−a)g′(ta)t=−1+g(ta)+g′(ta)ta−g′(ta)t≤0. (cid:3) Proof of Theorem 2. a) Let v be a radial function. Setting a = sin2θ , (5) N 1 − reduces to v F(v)=v 1 a+ag(v2a) + ρ N 2+a+(1 a)g(v2a) . ρρ − ρ ρ − − ρ Let a=1, we seek a(cid:0)solution to (cid:1) (cid:0) (cid:1) v (7) v g(v2)+(N 1) ρ =0 ρρ ρ − ρ suchthat v (R(1)+1)=0 and v (ρ)<0, for every ρ [2,R(1)+1). Consider the ρ ρ ∈ Cauchy problem N 1 ζ ζ = − ′ − ρ g(ζ2) (8) ζ(2)= 1. − We areinterestedinanegativesolutionζ. Define R(1)tobe theunique positive real solution to R(1)+1 G( 1) (N 1)ln =0, − − − 2 (cid:18) (cid:19) i.e. G(−1) R(1)=2e N−1 1. − The solution ζ of (8), satisfies ζ(ρ) g(t2) ρ N 1 ρ G(ζ(ρ)) G( 1)= dt= − ds= (N 1)ln . − − t − s − − 2 Zζ(2) Z2 (cid:16) (cid:17) Then, for everyρ (2,R(1)+1), G(ζ(ρ))>0 and ζ(ρ)<0, while ζ(R(1)+1)=0. ∈ Setting v (ρ)=ζ(ρ) and ρ ρ v(ρ)= v (s)ds, ρ ZR(1)+1 we obtain that v solves (7) and, for every ρ (2,R(1)+1), ∈ v (ρ)<v (R(1)+1)=0 and v(ρ)>v(R(1)+1)=0. ρ ρ b) Set, for ρ (1,R(1)], u(ρ)=v(ρ+1). Since, for the function v, we have ∈ v (ρ) v (ρ)g(v2(ρ))+(N 1) ρ =0, ρρ ρ − ρ at ρ+1 we obtain u (ρ) (9) u (ρ)g(u2(ρ))+(N 1) ρ =0. ρρ ρ − ρ+1