BREZIS TYPE THEOREM INVOLVING FRACTIONAL LAPLACIANS MINGFENG ZHAO March 05, 2014 Theorem 1 (Theorem 1 on Page 271 in Brezis[1]). Let n ≥ 1 and p > 1, the only nonnegative solution of ∆u = up in Rn is zero. Remark 1. Let u(x) = ex1 > 0 in Rn , then ∆u = u in Rn . Theorem 2. Let 0 < s < 1, p > 1, and u be a solution of the following problem: Ls (u)(z) := div [y 1−2s ∇u(x, y)] = 0, ∀z = (x, y) ∈ Rn+1 + , ∂u (x, 0) := − lim y 1−2s uy (x, y) = −up (x, 0), ∀x ∈ Rn , (1) y&0 ∂ν s u(x, y) ≥ 0, ∀(x, y) ∈ Rn+1 . + p 2s + b in Then then either u(x, y) ≡ 0 in Rn+1 + , or there exists some a, b > 0 such that 2sa = b , and u(x, y) = ay Rn+1 + . Corollary 1. Let 0 < s < 1 and p > 1, any non-negative bounded solution of (−∆)s u = −up in Rn must be constant. Remark 2. When s = 1/2, Theorem 2 is proved by Lou and Zhu[2]. Indeed, our proof of Theorem 2 follows the same idea as Lou and Zhu[2]. n Remark 3. When s = 1/2, let u(x, y) = ey cos(x1 ) in Rn+1 + , then ∆u(x, y) = 0 and uy (x, 0) = cos(x1 ) = u(x, 0) in R . Conjecture 1. Theorem 2 does not hold for p = 1. n+1 n+1 If u(x, y) ≡ 0 in Rn+1 such + , we are done. If u(x, y) 6≡ 0 in R+ , and there exists some z0 = (x0 , y0 ) ∈ R+ that u(z0 ) = 0. Since u(x, y) ≥ 0 in Rn+1 + , by the Strong Maximum Principle, then y0 = 0 and u(x, y) > u(x0 , 0) for ∂u ∂u all (x, y) ∈ Rn+1 (x0 , 0) < 0, which contradicts with (x0 , 0) = + . By the Hopf’s Lemma for Ls , we know that ∂νs ∂νs −up (x0 , 0) = 0. So the rest case should be u(x, y) > 0 for all (x, y) ∈ Rn+1 + . Consider the s-Kelvin transformation of u as: v(z) = 1 |z|n−2s u z |z|2 , ∀z ∈ Rn+1 + \{0}. It’s easy to see that v(z) → 0 as |z| → ∞. For all i = 1, · · · , n + 1, we have n+1 X z ∂ zj z ∂ 2s−n 2s−n vi (z) = u |z| + |z| uj · |z|2 ∂zi |z|2 ∂zi |z|2 j=1 1 2 MINGFENG ZHAO = = u z |z|2 · (2s − n)|z|2s−n−2 zi + |z|2s−n n+1 X uj j=1 2s−n−2 (2s − n)|z| u z |z|2 z |z|2 zi + |z| 2s−n−2 ui z |z|2 δij |z|2 − 2zi zj |z|4 · z |z|2 − 2|z| 2s−n−4 n+1 X uj j=1 z |z|2 zi zj . So for all i = 1, · · · , n + 1, we have vii (z) = (2s − n)(2s − n − 2)|z|2s−n−4 u 2s−n−2 +(2s − n)|z| u z |z|2 uj n+1 X −2|z| uj j=1 = z |z|2 zj − 2|z| 2s−n−2 +(2s − n)|z| u z |z|2 −2(2s − n − 4)|z|2s−n−6 −2|z|2s−n−4 z · ∇u = z |z|2 (2s − n)(2s − n − 2)|z| +(2s − n)|z|2s−n−2 u −2|z|2s−n−6 n+1 X uj j=1 n+1 X z |z|2 z |z|2 u z |z|2 ui uj z |z|2 zi + |z| zi2 ui n+1 X zj |z|2 uij n+1 X ujk z |z|2 · ∂ ∂zi zk |z|2 z |z|2 ∂ · ∂zi zj |z|2 · zi zj n+1 X uj z |z|2 zi + |z| z |z|2 2s−n−2 · δij |z|2 − 2zi zj |z|4 n+1 X uij j=1 n+1 X ujk z |z|2 · z |z|2 · δij |z|2 − 2zi zj |z|4 δik |z|2 − 2zi zk · zi zj |z|4 2s−n−4 + (2s − n)|z| − 2(2s − n − 4)|z|2s−n−6 z |z|2 n+1 X zi ui z |z|2 − 2(2s − zi + |z|2s−n−4 uii uj j=1 ∂ ∂zi j=1 j,k=1 z |z|2 · zi δij zi2 zj − 2|z|2s−n−4 2s−n−2 j=1 2s−n−4 z |z|2 + (2s − n − 2)|z|2s−n−4 ui z |z|2 z |z|2 zi2 + (2s − n)|z|2s−n−2 zi − 2|z|2s−n−4 zi ui 2s−n−4 zi zj uij n+1 X + (2s − n − 2)|z| n+1 X uj j,k=1 2s−n−4 z |z|2 zi2 zj − 2|z|2s−n−4 j=1 2s−n−4 j=1 (2s − n)(2s − n − 2)|z|2s−n−4 u z |z|2 n+1 X j=1 + (2s − n − 2)|z| j=1 2s−n−4 zi2 + (2s − n)|z|2s−n−2 zi n+1 X −2(2s − n − 4)|z|2s−n−6 z |z|2 n)|z|2s−n−6 zi2 z z |z|2 · ∇u z |z|2 zi2 zj − 2|z|2s−n−6 n+1 X j=1 uij z |z|2 z z z − 2|z|2s−n−4 z · ∇u − 2|z|2s−n−4 zi ui |z|2 |z|2 |z|2 j,k=1 z z z 2 2s−n−4 2s−n−6 2 = (2s − n)(2s − n − 2)|z|2s−n−4 u z + (2s − n)|z| z u − 2(2s − n)|z| z z · ∇u i i i i |z|2 |z|2 |z|2 z z z +(2s − n)|z|2s−n−2 u + (2s − n − 2)|z|2s−n−4 ui zi + |z|2s−n−4 uii 2 2 |z| |z| |z|2 n+1 X z z 2s−n−6 2 −4|z|2s−n−6 zi zj uij − 2(2s − n − 4)|z| z z · ∇u i |z|2 |z|2 j=1 +4|z|2s−n−8 zi2 zj zk ujk BREZIS TYPE THEOREM INVOLVING FRACTIONAL LAPLACIANS n+1 X 3 z z z 2s−n−4 2s−n−4 − 2|z| − 2|z| z · ∇u z u i i |z|2 |z|2 |z|2 j,k=1 z z 2s−n−4 2 2s−n−4 = (2s − n)(2s − n − 2)|z| u zi + (4s − 2n − 4)|z| zi ui |z|2 |z|2 z z z 2s−n−2 2s−n−4 −2(4s − 2n − 4)|z|2s−n−6 zi2 z · ∇u + (2s − n)|z| + |z| u u ii |z|2 |z|2 |z|2 n+1 n+1 X X z z 2s−n−6 2s−n−8 2 −4|z| zi zj uij + 4|z| zi zj zk ujk |z|2 |z|2 j=1 j,k=1 z −2|z|2s−n−4 z · ∇u . |z|2 +4|z|2s−n−8 zi2 zj zk ujk So we get vy (z) = = ∆v(z) = n+1 X z |z|2 y + |z|2s−n−2 uy z |z|2 n+1 X z yzj |z|2 j=1 z z z 2s−n−2 2s−n−4 2s−n−2 y + |z| uy − 2|z| z · ∇u y (2s − n)|z| u |z|2 |z|2 |z|2 (2s − n)|z|2s−n−2 u − 2|z|2s−n−4 uj vii (z) i=1 = = Ls (v)(x, y) z z 2s−n−4 + (4s − 2n − 4)|z| z · ∇ |z|2 |z|2 z z z 2s−n−2 2s−n−4 2s−n−4 + (2s − n)(n + 1)|z| u + |z| ∆u −2(4s − 2n − 4)|z| z · ∇u |z|2 |z|2 |z|2 n+1 n+1 X X z z 2s−n−6 −4|z|2s−n−6 zi zj uij + 4|z| z z u j k jk |z|2 |z|2 i,j=1 j,k=1 z 2s−n−4 −2(n + 1)|z| z · ∇u |z|2 z z z 2s−n−4 2s−n−4 (2s − n)(2s − 1)|z|2s−n−2 u − 2(2s − 1)|z| z · ∇u + |z| ∆u |z|2 |z|2 |z|2 (2s − n)(2s − n − 2)|z|2s−n−2 u := div [y 1−2s ∇v(x, y)] = y 1−2s ∆v(z) + (1 − 2s)y −2s vy (z) = y −2s [y∆v(z) + (1 − 2s)vy (z)] z z z 2s−n−4 2s−n−4 − y2(2s − 1)|z| z · ∇u + y|z| ∆u y −2s y(2s − n)(2s − 1)|z|2s−n−2 u |z|2 |z|2 |z|2 z z z 2s−n−2 2s−n−2 2s−n−4 (1 − 2s)(2s − n)|z| u y + (1 − 2s)|z| uy − 2(1 − 2s)|z| z · ∇u y |z|2 |z|2 |z|2 z z y −2s y|z|2s−n−4 ∆u + (1 − 2s)|z|2s−n−2 uy 2 |z| |z|2 y z z y −2s |z|2s−n−2 · ∆u + (1 − 2s)uy |z|2 |z|2 |z|2 = = = 4 MINGFENG ZHAO = = ∂v (x, 0) ∂νs −2s y y z z |z| · · ∆u + (1 − 2s)uy |z|2 |z|2 |z|2 |z|2 z |z|−n−2s−2 Ls (u) |z|2 −2s−n−2 := − lim y 1−2s vy (x, y) y&0 = = = = = z z z 2s−n−2 2s−n−2 2s−n−4 y + |z| − 2|z| y − lim y (2s − n)|z| u uy z · ∇u y&0 |z|2 |z|2 |z|2 z z 2s−n−4 − lim (2s − n)y 2−2s (2s − n)|z|2s−n−2 u − 2|z| z · ∇u y&0 |z|2 |z|2 z − lim y 1−2s |z|2s−n−2 uy y&0 |z|2 z − lim y 1−2s |z|2s−n−2 uy Since 0 < s < 1 y&0 |z|2 1−2s y z − lim |z|−n−2s u y y&0 |z|2 |z|2 x ∂u , 0 , ∀x ∈ Rn \{0}. |x|−n−2s ∂νs |x|2 1−2s So we get Ls (v)(z) = ∂v (x, 0) ∂vs = = = = ∀z ∈ Rn+1 + , x −n−2s ∂u −|x| ,0 ∂νs |x|2 p x , 0 −|x|−n−2s u |x|2 p 1 x p(n−2s)−n−2s −|x| u ,0 |x|n−2s |x|2 0, −|x|α v p (x, 0), ∀x ∈ Rn . Then v : Rn+1 + \{0} → R satisfies (2) div [y 1−2s ∇v(x, y)] = 0, ∀(x, y) ∈ Rn+1 + , ∂v (x, 0) := − lim y 1−2s vy (x, y) = −|x|α v p (x, 0), y&0 ∂ν s v(x, y) > 0, ∀(x, y) ∈ Rn+1 + \{0}, lim v(z) = 0, |z|→∞ α = p(n − 2s) − (n + 2s). ∀x ∈ Rn \{0}, T n + 2s + +T , that is, α > 0. For any R > 0, let BR = BR (0) Rn+1 ∂Rn+1 Firstly, we will assume p > + , and + , ΓR = ∂BR n − 2s T n+1 n+1 SR = ∂BR (0) R+ . Since v(x, y) > 0 for all (x, y) ∈ R+ \{0}, then min v(z) > 0. z∈S1 BREZIS TYPE THEOREM INVOLVING FRACTIONAL LAPLACIANS Lemma 1. For any 0 < < min 1, min v(z) , we have z∈S1 v(z) ≥ 2s , 1 + 2s ∀x ∈ B1+ \{0}. Proof. For any 0 < r < 1, consider the function ψr (z) = y 2s 2s rn−2s − n−2s + , 1 + 2s |z| 1 + 2s ∀z ∈ Rn+1 + \{0}. By Remark 4, we know that Ls (ψr )(z) = = = = = ∂ψr (x, 0) ∂νs div [y 1−2s ∇ψr (x, y)] Ls (y 2s ) 1 + 2s div [y 1−2s ∇y 2s ] 1 + 2s div [y 1−2s (2s)y 2s−1 ] 1 + 2s div [2s] 1 + 2s ∀z ∈ Rn+1 + , = 0, = − lim y 1−2s Dy ψr (x, y) = = = y&0 lim y 1−2s Dy (y 2s ) 1 + 2s y&0 − lim y 1−2s (2s)y 2s−1 1 + 2s y&0 2s − , ∀x ∈ Rn \{0}. 1 + 2s − For all z ∈ Sr , we have ψr (z) = ≤ = y 2s 2s −+ 1 + 2s 1 + 2s 2s −+ 1 + 2s 1 + 2s Since 0 < y ≤ r < 1 0 ≤ v(z). For all z ∈ S1 , we have ψr (z) = < 2s y 2s − rn−2s + 1 + 2s 1 + 2s 2s + Since 0 < y ≤ 1 1 + 2s 1 + 2s < ≤ v(z). 5 6 MINGFENG ZHAO Let Pr (z) = v(z) − ψr (z) in B1+ \Br+ , then Pr satisfies div [y 1−2s ∇Pr (x, y)] = 0, ∀z ∈ B1+ \Br+ , ∂P 2s r (x, 0) = −|x|α v p (x, 0) + , ∀x ∈ Γ1 \Γr , ∂ν 1 + 2s s S P (z) ≥ 0, ∀z ∈ S S . r r 1 Claim I: Pr (z) ≥ 0 for all B1+ \Br+ . If the Claim I is not true, by the Strong Maximum Principle, there exists some x0 ∈ Γ1 \Γr such that Pr (x0 , 0) = inf Pr (z) < 0, and Pr (z) > Pr (x0 , 0), z∈B1+ \Br+ ∀z ∈ B1+ \Br+ . 2s ∂Pr (x0 , 0) < 0, that is, −|x0 |α v p (x0 , 0) + < 0, which implies that ∂νs 1 + 2s 1 α 2s p · |x0 |− p 1 + 2s By the Hopf’s Lemma for Ls , we know that v(x0 , 0) > ≥ 2s , 1 + 2s Since p > 1 > , α > 0 and |x0 | < 1. On the other hand, we have Pr (x0 , 0) = v(x0 , 0) − > v(x0 , 0) − rn−2s 2s − 1 + 2s |x0 |n−2s 2s 1 + 2s > 0, which contradicts with Pr (x0 , 0) < 0. By the Claim I, we know that v(z) ≥ rn−2s y 2s 2s rn−2s 2s − n−2s + ≥ − n−2s , 1 + 2s |z| 1 + 2s 1 + 2s |z| ∀z ∈ B1+ \Br+ . By taking r & 0 in the above inequality, we know that v(z) ≥ 2s , 1 + 2s ∀z ∈ B1+ \{0}. Remark 4. Let n ≥ 2, define u(x, y) = 1 1 = n−2s , 2 |(x, y)|n−2s [|x| + y 2 ] 2 ∀(x, y) ∈ Rn+1 + . Let r = |(x, y)|, then for all (x, y) 6= 0, we have div [y 1−2s ∇u(x, y)] = y 1−2s ∆u(x, y) + (1 − 2s)y −2s uy (x, y) h i n y = y 1−2s · u00 (r) + · u0 (r) + (1 − 2s)y −2s u0 (r) · r r BREZIS TYPE THEOREM INVOLVING FRACTIONAL LAPLACIANS = y 1−2s [(2s − n)(2s − n − 1)r2s−n−2 + n(2s − n)r2s−n−2 ] + (1 − 2s)y 1−2s (2s − n)r2s−n−2 = y 1−2s (2s − n)(2s − 1)r2s−n−2 + (1 − 2s)y 1−2s (2s − n)r2s−n−2 = uy (x, y) = u0 (r) · = −y 1−2s uy (x, y) 0 y r (2s − n)yr2s−n−2 = −y 2−2s (2s − n)r2s−n−2 = y 2−2s (n − 2s)|(x, y)|2s−n−2 ≥ 0, Since n ≥ 2 ≥ 2s. Now for any λ < 0, we define Σλ = z ∈ Rn+1 : x1 > λ , + Tλ = z ∈ Rn+1 : x1 = λ , + Σλ = Σλ \{0}, zλ = (2λ − x1 , x2 , · · · , xn , y), vλ (z) = v(z λ ), wλ (z) = v(z) − vλ (z). wλ (z) = 0, Ls (wλ )(z) = Ls (v)(2λ − x1 , x2 , · · · , xn , y) = 0, Notice that For all x ∈ ∂Rn+1 + T ∀z ∈ ∂Σλ \ Rn+1 + , ∀z ∈ Σλ . Σλ , it’s easy to see that |x| < |xλ |, which implies that ∂wλ (x, 0) ∂νs = ∂v ∂v λ (x, 0) − (x , 0) ∂νs ∂νs = −|x|α v p (x, 0) + |xλ |α v p (xλ , 0) = −|x|α [v p (x, 0) − vλp (xλ , 0)] + [|xλ |α − |x|α ]v p (xλ , 0) ≥ −|x|α [v p (x, 0) − vλp (xλ , 0)] = −c1 (x, λ)wλ (x, 0), By the Mean Value Theorem. where c1 (x, λ) = p|x|α ξ1p−1 (x, λ) for some positive function ξ1 (x, λ) which is between v(x, 0) and vλ (x, 0). 7 8 MINGFENG ZHAO Lemma 2. For any fixed λ < 0, if there exists some small 0 < r < |λ| such that wλ (z) > 0 for all z ∈ Br+ \{0}, then wλ (z) ≥ 0 for all z ∈ Σλ . Proof. If the Lemma 2 is not true, then there exists some z0 ∈ Σλ such that wλ (z0 ) < 0. Since lim |z|→∞ v(z) = 0, then lim wλ (z) = 0. Since wλ (z) > 0 for all z ∈ Br+ \{0} and wλ (z) = 0 for all z ∈ Tλ , then there exists some |z|→∞ S z1 ∈ Σλ \ Br+ Tλ such that wλ (z1 ) = inf wλ (z) < 0. z∈Σλ T By the Strong Maximum Principle, we know that z1 ∈ ∂Σλ ∂Rn+1 + \{0} and wλ (z) > wλ (z1 ) for all z ∈ Σλ . By the ∂wλ Hopf’s Lemma for Ls , we know that (z1 ) < 0. On the other hand, we know that ∂νs ∂wλ (z1 ) ≥ −c1 (z0 , λ)wλ (z0 ) > 0. ∂νs We get a contradiction. Therefore, we know that wλ (z) ≥ 0 for all z ∈ Σλ . Lemma 3. There exists some Λ −1 such that for all λ ≤ Λ, we have wλ (z) ≥ 0, ∀z ∈ Σλ . Proof. By Lemma 1, there exists some 1 > 0 such that v(z) ≥ 2, Since lim |z|→∞ ∀z ∈ B1+ \{0}. v(z) = 0, then there exists some R0 1 such that |v(z)| ≤ , ∀|z| ≥ R0 . Let Λ = −R0 − 1 −1, for all λ ≤ Λ, then B1+ \{0} ⊂ Σλ . For all z ∈ B1+ \{0}, we know that |z λ | = |(2λ − x1 , · · · , xn , y)| ≥ 2|λ| − |z| ≥ 2|λ| − 1 ≥ |λ| > R0 , which implies that |vλ (z)| = |v(z λ )| ≤ . Hence we get wλ (z) = v(z) − vλ (z) ≥ 2 − = > 0, ∀z ∈ B1+ \{0}. By Lemma 2, we know that wλ (z) ≥ 0 for all z ∈ Σλ . By the Lemma 3, we can define λ0 = sup Lemma 4. λ0 = 0. λ < 0 : wµ (z) ≥ 0 in Σµ , ∀µ ≤ λ . BREZIS TYPE THEOREM INVOLVING FRACTIONAL LAPLACIANS 9 Proof. If the Lemma 4 is not true, that is, λ0 < 0. By the Definition of λ0 , we know that wλ0 (z) ≥ 0 in Σλ0 . If wλ0 (z) ≡ 0 in Σλ0 , that is, v(z) = v(z λ ) in Σλ0 , in particular, we get 0 = = ∂wλ0 (x, 0) ∂νs ∂v ∂v λ0 (x, 0) − (x , 0) ∂νs ∂νs = −|x|α v p (x, 0) + |xλ0 |α v p (xλ , 0) = [|xλ0 |α − |x|α ]v p (x, 0), On the other hand, for all x ∈ ∂Rn+1 + T ∀x ∈ ∂Rn+1 + \ Σλ 0 Σλ0 , since λ0 < 0, then we know that |x| < |xλ0 |. Since v(x, 0) > 0, then the above inequality will give us 0 > 0, contradiction. So the only possibility is that wλ0 (z) 6≡ 0 in Σλ0 . Since wλ0 (z) ≥ 0 T in Σλ0 , by the Strong Maximum Principle, we have wλ0 (z) > 0 in Σλ0 . If there exists some x0 ∈ Σλ0 ∂Rn+1 such + that wλ0 (x0 , 0) = 0, then wλ0 (x, y) > 0 = wλ0 (x0 , 0) for all (x, y) ∈ Σλ0 , by the Hopf’s Lemma for Ls , we have T ∂wλs (x0 , 0) < 0. On the other hand, since wλ0 (x0 , 0) = 0, then v(x0 , 0) = v(xλ0 0 , 0) > 0. Since (x0 , 0) ∈ Σλ0 ∂Rn+1 + , ∂νs ∂wλs then |xλ0 0 | > |x0 |. Hence we get (x0 , 0) = −|x0 |α v p (x0 , 0) + |xλ0 0 |α v p (xλ0 0 , 0) > 0, contradiction. In summary, we ∂νs know that wλ0 (z) > 0 for all Σλ0 \Tλ0 . |λ0 | , 1 , there exists some constant δ > 0 which only depends on λ0 and r such that Claim I: For any 0 < r < min 2 wλ0 (z) > δ for all z ∈ Br+ \{0}. Notice that wλ0 satisfies the following PDEs: Ls (wλ0 )(z) = 0, ∀z ∈ Br+ , ∂w λ0 (x, 0) = −|x|α v p (x, 0) + |xλ0 |α vλp0 (x, 0), ∂ν s w (z) > 0, ∀z ∈ B + \{0}. ∀x ∈ Γr \{0}, r λ0 Since wλ0 (z) > 0 for all z ∈ Br+ \{0}, then min wλ0 (z) > 0. Let 0 < < min 1, min wλ0 (z) . For all z ∈ Br+ \{0}, z∈Sr z∈Sr |λ0 | , then z λ0 ∈ / Br+ , which implies that since 0 < r < 2 C1 := sup vλ0 (z) ∈ (0, ∞). z∈Br+ \{0} For any 0 < t < r and 0 < µ < 1, let ψt,µ (z) = 2sµ tn−2s (1 − µ)y 2s − n−2s + , 1 + 2s |z| 1 + 2s ∀z ∈ Br+ \Bt+ . By the proof of Lemma 1, we know that Ls (ψt,µ ) = 0, ∀z ∈ Br+ \Bt+ , ∂ψt,µ 2s(1 − µ) (x, 0) = − , ∀x ∈ Γr \Γt . ∂νs 1 + 2s 10 MINGFENG ZHAO For all z ∈ Sr , we have ψt,µ (z) = < ≤ ≤ 2sµ 1 + 2s 2sµ 1 + 2s 2sµ 1 + 2s 2s 1 + 2s tn−2s (1 − µ)y 2s + rn−2s 1 + 2s 2s (1 − µ)y + 1 + 2s (1 − µ) + Since 0 < y < r < 1 1 + 2s + Since 0 < µ < 1 1 + 2s − = ≤ wλ0 (z). For all z ∈ St , we have ψt,µ (z) = ≤ = (1 − µ)y 2s 2sµ −+ 1 + 2s 1 + 2s 2s −+ Since 0 < y, µ < 1 1 + 2s 1 + 2s 0 ≤ wλ0 (z). Let Pt,µ (z) = wλ0 (z) − ψt,µ (z) in Br+ \Bt+ , then Pt,µ satisfies Ls (Pt,µ )(z) = 0, ∀z ∈ Br+ \Bt+ , ∂Pt,µ 2s(1 − µ) − |x|α v p (x, 0) + |xλ0 |α vλp0 (x, 0), (x, 0) = ∂ν 1 + 2s s P (z) ≥ 0, ∀z ∈ S S S . t,µ r ∀x ∈ Γr \{0}, t If there exists some point z1 ∈ Br+ \Bt+ such that Pt,µ (z1 ) < 0, by the Strong Maximum Principle, there exists some x0 ∈ Γr \Γt such that Pt,µ (x0 , 0) = inf Pt,µ (z) < 0, and Pt,µ (z) > Pt,µ (x0 , 0), z∈Bt+ \Bt+ Since Pt,µ (x0 , 0) < 0, then wλ0 (x0 , 0) < = < ψt,µ (x0 , 0) 2sµ tn−2s − 1 + 2s |x0 |n−2s 2sµ . 1 + 2s Since x0 ∈ Γr \Γt , then t < |x0 | < r, which implies that ψt,µ (x0 , 0) = 2sµ tn−2s 2sµ 2s − ≤ < < 1. 1 + 2s |x0 |n−2s 1 + 2s 1 + 2s ∀z ∈ Br+ \Bt+ . BREZIS TYPE THEOREM INVOLVING FRACTIONAL LAPLACIANS 11 Since Pt,µ (x0 , 0) < 0, then v(x0 , 0) < ≤ vλ0 (x0 , 0) + ψt,µ (x0 , 0) C1 + 1 = C2 . By the Hopf’s Lemma for Ls , we know that 2s(1 − µ) − |x0 |α v p (x0 , 0) + |xλ0 0 |α vλp0 (x0 , 0) < 0, which implies that 1 + 2s |xλ0 0 |α v p (x0 , 0) − |xλ0 0 |α vλp0 (x0 , 0) > |xλ0 0 |α v p (x0 , 0) + ≥ 2s(1 − µ) , 1 + 2s 2s(1 − µ) − |x0 |α v p (x0 , 0) 1 + 2s Since |xλ0 0 | > |x0 |. By the Mean Value Theorem, there exists some positive number ξ2 (λ0 , x0 ) which is between v(x0 , 0) and vλ0 (x0 , 0) such that 2s(1 − µ) 1 + 2s ≤ |xλ0 0 |α v p (x0 , 0) − |xλ0 0 |α vλp0 (x0 , 0) = |xλ0 0 |α pξ2p−1 (λ0 , x0 )wλ0 (x0 , 0) Since wλ0 (z) > 0 in Σλ0 , then 0 ≤ v(x0 , 0) ≤ vλ0 (x0 , 0) ≤ C1 , which implies that 2s(1 − µ) 1 + 2s ≤ |xλ0 0 |α pC1p−1 wλ0 (x0 , 0) ≤ C3 wλ0 (x0 , 0). In summary, we get 2s(1 − µ) 1 2sµ > wλ0 (x0 , 0) ≥ · . 1 + 2s 1 + 2s C3 In particular, we have C3 µ > (1 − µ). That is, µ > 1 1 . But if we take some fixed µ < , we must have Pt,µ (z) ≥ 0 for all z ∈ Br+ \Bt+ , that is, 1 + C3 1 + C3 wλ0 (z) ≥ ψt,µ (z) = 2sµ tn−2s (1 − µ)y 2s 2sµ tn−2s − n−2s + ≥ − n−2s , 1 + 2s |z| 1 + 2s 1 + 2s |z| ∀z ∈ Br+ \Bt+ . 2sµ , for all z ∈ Br+ \{0}. 1 + 2s By the Definition of λ0 , there exists a strictly decreasing sequence 0 > λk > λ0 such that lim λk = λ0 and k→∞ |λ0 | , 1 , by the Claim I, there exists some small δ > 0 such that inf wλk (z) < 0. For any fixed 0 < r < min 2 z∈Σλ wλ0 (z) ≥ 2δ for all z ∈ Br+ . Since lim λk = λ0 , then there exists some K 1 such that wλk (z) ≥ δ > 0 for all z ∈ Br+ . By taking t & 0, we get wλ0 (z) ≥ δ with δ = k→∞ By Lemma 2, we know that wλk (z) ≥ 0 for all z ∈ Σλk , which contradicts with inf wλk (z) < 0. z∈Σλ In summary, we must have λ0 = 0. 12 MINGFENG ZHAO Proof of Theorem 2. Case I: p > n + 2s . By Lemma 4, we know that n − 2s ∀z ∈ Rn+1 with x1 > 0. + v(x1 , · · · , xn , y) ≥ v(−x1 , · · · , xn , y), Let v(z) = v(−x1 , · · · , xn , y) for all z ∈ Rn+1 + \{0}, then v is also a solution of 2, then v(−x1 , · · · , xn , y) = v(x1 , · · · , xn , y) ≥ v(−x1 , · · · , xn , y) = v(x1 , · · · , xn , y), ∀z ∈ Rn+1 with x1 > 0. + So v is symmetric with respect to x1 = 0. By the rotation, we know that v is radial symmetric with respect to x ∈ Rn . Since we know that v(z) = 1 |z|n−2s u z |z|2 z |z|2 , ∀z ∈ Rn+1 + \{0}. , ∀z ∈ Rn+1 + \{0}. Then u(z) = 1 |z| v n−2s Since v is radially symmetric with respect to x ∈ Rn , then u is also radially symmetric with respect to x ∈ Rn . On the other hand, we know that any translation with respect to x ∈ Rn of u is also a solution of 1, then for each y > 0, u(·, y) is radial symmetric with respect to each x ∈ Rn , which implies that u(·, y) is a constant function d 1−2s 0 [y U (y)] = 0, then there in Rn , that is, u(x, y) = U (y) for some function U on [0, ∞). By (1), we have dy exists some a, b ∈ R such that U (y) = ay 2s + b for all y > 0. Since U (y) > 0 for all y ≥ 0, then a, b > 0. Since ∂U (x, 0) = − lim y 1−2s U 0 (y) = −a · 2s = −bp , then 2sa = bp . y&0 ∂νs n + 2s n + m + 2s Case II: 1 < p ≤ . In this case, there exists some large m ∈ N such that p > . Let u(x, t, y) = u(x, y) n − 2s n − m − 2s n+m+1 , then u is a solution of (2) in Rn+m+1 . By the result of Case I, we know that, u(x, t, y) = for all (x, t, y) ∈ R+ + u(x, y) = ay 2s + b in Rn+m+1 . + Conjecture 2. Let 0 < s < 1 and p, q > 1, then the only solutions of the following problem: Ls (u)(z) = Lz (v)(z) = 0, ∀z = (x, y) ∈ Rn+1 + , ∂u ∂v (3) (x, 0) = −v p (x, 0), (x, 0) = −uq (x, 0) ∀x ∈ Rn , ∂ν ∂ν s s u(x, y), v(x, y) > 0, ∀(x, y) ∈ Rn+1 . + must be the form of u(x, y) = ay 2s + b and v(x, y) = cy 2s + d in Rn+1 for some a, b, c, d > 0 such that 2sa = dp and + 2sc = bq . References [1] H. Brezis. Semilinear equations in RN without condition at infinity. Appl. Math. Optim., 12:271–282, 1984. [2] Yuan Lou and Meijun Zhu. Classifications of nonnegative solutions to some elliptic problems. Differential and Integral Equations, 12:601– 612, 1999. BREZIS TYPE THEOREM INVOLVING FRACTIONAL LAPLACIANS Department of Mathematics, University of Connecticut, 196 Auditorium Road, Unit 3009, Storrs, CT 06269-3009 E-mail address: mingfeng.zhao@uconn.edu 13