ON LIOUVILLE’S THEOREM FOR BIHARMONIC FUNCTIONS MINGFENG ZHAO September 30, 2011 Definition 1. Let Ω be a domain in Rn , then we say that a function u on Ω is sub biharmonic in Ω if u ∈ C 4 (Ω), and ∆2 u(x) = ∆[∆u](x) ≥ 0, for all x ∈ Ω. we say that a function u on Ω is super biharmonic in Ω if u ∈ C 4 (Ω), and ∆2 u(x) ≤ 0, for all x ∈ Ω. we say that a function u on Ω is biharmonic in Ω if u ∈ C 4 (Ω), and ∆2 u(x) = 0, for all x ∈ Ω. Theorem 1 (The Mean Value Theorem for the Sub Biharmonic Functions, the equality (2), in Page 37, in Huilgol[3]). Let u be a sub biharmonic function in BR (x0 ) for some x0 ∈ Rn and R > 0. Then for all 0 < r < R, we have Z 1 r2 · ∆u(x0 ) ≤ · u(x) dS(x) u(x0 ) + 2n |∂Br (x0 )| ∂Br (x0 ) Z r2 1 u(x0 ) + · ∆u(x0 ) ≤ · u(x) dx. 2(n + 2) |Br (x0 )| Br (x0 ) Proof. a. For any 0 < r < R, we consider the function: r2 1 φ(r) = u(x ) + · ∆u(x0 ) − · 2n |∂Br (x0 )| 0 Z u(x) dS(x) ∂Br (x0 ) Z r2 1 · ∆u(x0 ) − · u(x0 + ry) · rn−1 dS(y) 2n nαn · rn−1 ∂B1 (0) Z 1 r2 · ∆u(x0 ) − · u(x0 + ry) dS(y). = u(x0 ) + 2n nαn ∂B1 (0) = u(x0 ) + Also we know that lim φ(r) = u(x0 ) + 0 − r&0 1 · u(x0 ) · nαn = 0. nαn And 0 φ (r) Z = r r · ∆u(x0 ) − · n nαn Z = 1 1 · ∆u(x0 ) − · n nαn ∂B1 (0) d u(x0 + ry) dS(y) dr ∇u(x0 + ry) · y dS(y) ∂B1 (0) 1 Let x = x0 + ry 2 MINGFENG ZHAO = = = r 1 · ∆u(x0 ) − · n nαn Z ∇u(x) · ∂Br (x0 ) r 1 · · ∆u(x0 ) − n nαn · rn−1 Z r 1 · · ∆u(x0 ) − n nαn · rn−1 Z z − x0 1−n ·r dS(x) r Let x = x0 + ry ∇u(x) · n(x) dS(x) ∂Br (x0 ) ∂Br (x0 ) ∂u(x) dS(x) ∂n Z = = = r 1 · ∆u(x) dx · ∆u(x0 ) − n nαn · rn−1 Br (x0 ) # " Z r 1 0 · ∆u(x) dx · ∆u(x ) − n αn · rn Br (x0 ) # " Z r 1 0 · ∆u(x) dx · ∆u(x ) − n αn · rn Br (x0 ) By the Divergence Theorem Since u is sub biharmonic in BR (x0 ), then ∆2 u(x) = ∆[∆u](x) ≥ 0, for all x ∈ BR (x0 ). That is, ∆u is a subharmonic function in BR (x0 ). Then by the Mean Value Theorem for the Subharmonic Function, the Theorem 2.1, in Page 14, in Gilbarg and Trudinger[2], we know that for all 0 < r < R, we have 1 ∆u(x ) ≤ · αn · rn 0 Z ∆u(x) dx. Br (x0 ) Hence, we get that φ0 (r) ≤ 0, for all 0 < r < R. That is, φ(r) is a decreasing function, which implies that φ(r) ≤ lim φ(r) = 0, r&0 for all 0 < r < R. That is, we get u(x0 ) + r2 1 · ∆u(x0 ) ≤ · 2n |∂Br (x0 )| Z u(x) dS(x). ∂Br (x0 ) b. Since, we have known that u(x0 ) + r2 · ∆u(x0 ) ≤ 2n = Z 1 · u(x) dS(x) |∂Br (x0 )| ∂Br (x0 ) Z 1 · u(x) dS(x), nαn rn−1 ∂Br (x0 ) for all 0 < r < R. Both sides are multiplied by n · rn−1 , we get nrn−1 u(x0 ) + rn+1 1 · ∆u(x0 ) ≤ · 2 αn Z u(x) dS(x). ∂Br (x0 ) ON LIOUVILLE’S THEOREM FOR BIHARMONIC FUNCTIONS 3 Then both sides integrate with respect to r from 0 to r, we get Z r Z r Z rZ ∆u(x0 ) 1 nu(x0 ) tn−1 dt + · tn+1 dr ≤ · u(x) dS(x) dt. 2 αn 0 ∂Bt (x0 ) 0 0 Then we can get ∆u(x0 ) rn+2 1 rn + · ≤ · nu(x ) · n 2 n+2 αn 0 Z u(x) dx. Br (x0 ) Both sides are divided by rn , we get r2 1 · · ∆u(x0 ) ≤ 2(n + 2) αn · rn Z r2 1 · ∆u(x0 ) ≤ · 2(n + 2) |Br (x0 )| Z u(x0 ) + u(x) dx. Br (x0 ) Therefore, we get u(x0 ) + u(x) dx. Br (x0 ) As a direct consequence of the Theorem 1, we have the following: Theorem 2. Let x0 ∈ Rn and R > 0. Then a. If u is super biharmonic in BR (x0 ), then for all 0 < r < R, we have Z 1 r2 · ∆u(x0 ) ≥ · u(x) dS(x) u(x0 ) + 2n |∂Br (x0 )| ∂Br (x0 ) Z r2 1 u(x0 ) + · ∆u(x0 ) ≥ · u(x) dx. 2(n + 2) |Br (x0 )| Br (x0 ) b. If u is biharmonic in BR (x0 ), then for all 0 < r < R, we have r2 u(x ) + · ∆u(x0 ) 2n = r2 · ∆u(x0 ) 2(n + 2) = 0 u(x0 ) + Z 1 · u(x) dS(x) |∂Br (x0 )| ∂Br (x0 ) Z 1 · u(x) dx. |Br (x0 )| Br (x0 ) Theorem 3 (The Theorem 2, in Page 37, in Huilgol[3]). Let u be biharmonic in Rn , and lim |x|→∞ u(x) = 0. |x|2 Then u is a linear function on Rn . In particular, we know that if u is a bounded biharmonic function in Rn , then u is a constant function on Rn . Proof. For any x0 ∈ Rn and fix, by the Mean Value Theorem, the Theorem 2, we know that for all r > 0, we have Z r2 1 0 0 u(x ) + · ∆u(x ) = · u(x) dS(x). 2n |∂Br (x0 )| ∂Br (x0 ) That is, we get (1) u(x0 ) 1 1 + · ∆u(x0 ) = · 2 r 2n nαn · rn+1 Z u(x) dS(x). ∂Br (x0 ) 4 MINGFENG ZHAO Since lim |x|→∞ u(x) = 0, then for any > 0, we know that there exists some constants R0 > 0 such that |x|2 |u(x)| ≤ |x|2 , for all |x| ≥ R0 . So when r > R0 + |x0 |, we know that for all x ∈ ∂Br (x0 ), we have |x| = |x − x0 + x0 | ≥ |x − x0 | − |x0 | = r − |x0 | ≥ R0 . Then we have 1 · nαn · rn+1 Z ≤ u(x) dS(x) ∂Br (x0 ) ≤ = ≤ = 1 · nαn · rn+1 Z 1 · nαn · rn+1 Z 1 · nαn · rn+1 Z 1 · nαn · rn+1 Z 1 · nαn · rn+1 Z |u(x)| dS(x) ∂Br (x0 ) |x|2 dx Br (x0 ) ∂Br |x − x0 + x0 |2 dx (x0 ) [|x − x0 | + |x0 |]2 dx ∂Br (x0 ) [r + |x0 |]2 dx ∂Br (x0 ) [r + |x0 |]2 · nαn · rn−1 nαn · rn+1 = = · [r + |x0 |]2 . r2 By taking r % ∞, we get lim sup r%∞ 1 · nαn · rn+1 Z 1 · nαn · rn+1 Z u(x) dS(x) ≤ . ∂Br (x0 ) Then taking & 0, then we get lim sup r%∞ u(x) dS(x) = 0. ∂Br (x0 ) Therefore, by taking r % ∞ in (1), we can conclude that ∆u(x0 ) = 0. Since x0 is arbitrary point in Rn , then we get ∆u(x) = 0, for all x ∈ Rn . For any x0 ∈ Rn and fix, by the Gradient Estimates for the Harmonic Function, the Theorem 2.10, in Page 23, in Gilbarg and Trudinger[2], we know that for any r > R0 + |x0 |, we know that for all x ∈ ∂Br (x0 ), we have |x| = |x − x0 + x0 | ≥ |x − x0 | − |x0 | = r − |x0 | ≥ R0 , we have 2 2n 2 0 |D u(x )| ≤ · sup |u(x)| r x∈Br (x0 ) = 4n2 · sup |u(x)|. r2 x∈Br (x0 ) ON LIOUVILLE’S THEOREM FOR BIHARMONIC FUNCTIONS 5 By the Maximum Principle for the Harmonic Functions, the Theorem 2.2, in Page 15, in Gilbarg and Trudinger[2], we know that |u(x)| = sup x∈Br (x0 ) sup |u(x)|. x∈∂Br (x0 ) But for any x ∈ ∂Br (x0 ), we have |u(x)| ≤ |x|2 = |x − x0 + x0 |2 ≤ · [|x − x0 | + |x0 |]2 = [r + |x0 |]2 . Then we have |D2 u(x0 )| ≤ = 4n2 · [r + |x0 |]2 r2 2 r + |x0 | 2 . 4n · · r By taking r % ∞, we get |D2 u(x0 )| ≤ 4n2 · . By taking & 0, we get D2 u(x0 ) = 0. Since x0 is arbitrary in Rn , then D2 u(x) = 0, for all x ∈ Rn . Therefore, u is a linear function on Rn . In particular, we know that if u is a bounded biharmonic function in Rn , then u is a constant function on Rn . Remark 1. The Theorem 3 is sharp, since u(x) = |x|2 for all x ∈ Rn is biharmonic in Rn , but u is just bounded from below in Rn , and unbounded from above in Rn . Definition 2. Let k ∈ N, and Ω be a domain in Rn , then we say that a function u on Ω is sub k-harmonic in Ω if u ∈ C 2k (Ω), and ∆k u(x) = ∆∆ · · · ∆} u(x) ≥ 0, | {z for all x ∈ Ω. k many times we say that a function u on Ω is super k-harmonic in Ω if u ∈ C 2k (Ω), and ∆k u(x) ≤ 0, for all x ∈ Ω. we say that a function u on Ω is biharmonic in Ω if u ∈ C 2k (Ω), and ∆k u(x) = 0, Theorem 4. Let k ∈ N, x0 ∈ Rn and R > 0. Then for all x ∈ Ω. 6 MINGFENG ZHAO a. If u is sub k-harmonic in BR (x0 ), then for all 0 < r < R, we have 0 u(x ) + k−1 X i=1 0 u(x ) + n2i · i! · k−1 X i=1 2i · i! · r2i Qi−1 i j=1 (n r2i Qi + 2j) · ∆ u(x ) ≤ i j=1 (n + 2j) 0 0 · ∆ u(x ) ≤ Z 1 · |∂Br (x0 )| 1 · |Br (x0 )| u(x) dS(x) ∂Br (x0 ) Z u(x) dx. Br (x0 ) b. If u is super k-harmonic in BR (x0 ), then for all 0 < r < R, we have u(x0 ) + k−1 X i=1 0 u(x ) + n2i · i! · k−1 X i=1 2i · i! · r2i Qi−1 j=1 (n r2i Qi + 2j) · ∆i u(x0 ) ≥ i j=1 (n + 2j) 0 · ∆ u(x ) ≥ Z 1 · |∂Br (x0 )| 1 · |Br (x0 )| u(x) dS(x) ∂Br (x0 ) Z u(x) dx. Br (x0 ) c. If u is k-harmonic in BR (x0 ), then for all 0 < r < R, we have u(x0 ) + k−1 X i=1 u(x0 ) + n2i · i! · k−1 X i=1 r2i Qi−1 j=1 (n r2i Qi 2i · i! · j=1 (n + 2i) + 2j) · ∆i u(x0 ) · ∆i u(x0 ) = = Z 1 · |∂Br (x0 )| 1 · |Br (x0 )| u(x) dS(x) ∂Br (x0 ) Z u(x) dx. Br (x0 ) Proof. We just prove for the case of sub k-harmonic functions, that is, let u be a sub k-harmonic function in BR (x0 ). Claim I: For any 0 < r < R, we have u(x0 ) + k−1 X i=1 0 u(x ) + n2i · i! · k−1 X i=1 2i · i! · r2i Qi−1 j=1 (n r2i Qi + 2j) · ∆i u(x0 ) ≤ i j=1 (n + 2j) 0 · ∆ u(x ) ≤ Z 1 · |∂Br (x0 )| 1 · |Br (x0 )| u(x) dS(x) ∂Br (x0 ) Z u(x) dx. Br (x0 ) We prove the Claim I by induction. When k = 1, that is, u is a subharmonic function in BR (x0 ). We know that u(x0 ) + 1−1 X i=1 u(x0 ) + n2i · i! · 1−1 X i=1 2i · i! · r2i Qi−1 · ∆i u(x0 ) = u(x0 ) r2i Qi · ∆i u(x0 ) = u(x0 ). j=1 (n + 2j) j=1 (n + 2j) By the Mean Value Theorem for the Subharmonic Function, the Theorem 2.1, in Page 14, in Gilbarg and Trudinger[2], we know that for all 0 < r < R, we have u(x0 ) ≤ u(x0 ) ≤ Z 1 · u(x) dS(x) |∂Br (x0 )| ∂Br (x0 ) Z 1 · u(x) dx. |Br (x0 )| Br (x0 ) So the Claim I is true for k = 1. When k = 2, that is, u is a sub biharmonic function in BR (x0 ). We know that 0 u(x ) + 2−1 X i=1 n2i · i! · r2i Qi−1 j=1 (n + 2j) · ∆i u(x0 ) = u(x0 ) + r2 · ∆u(x0 ) 2n ON LIOUVILLE’S THEOREM FOR BIHARMONIC FUNCTIONS u(x0 ) + 2−1 X i=1 2i · i! · r2i Qi j=1 (n + 2j) · ∆i u(x0 ) = u(x0 ) + 7 r2 · ∆u(x0 ). 2(n + 2) By the Theorem 1, we know that u(x0 ) + u(x0 ) + r2 · ∆u(x0 ) ≤ 2n r2 · ∆u(x0 ) ≤ 2(n + 2) Z 1 · u(x) dS(x) |∂Br (x0 )| ∂Br (x0 ) Z 1 · u(x) dx. |Br (x0 )| Br (x0 ) So the Claim I is true for k = 2. Now we assume for k = m ≥ 2, the Claim I is true. Look at k = m + 1, that is, u is ( a sub m + 1-harmonic function in BR x0 ), so ∆m+1 u(x) = ∆m [∆u](x) ≤ 0, for all x ∈ BR (x0 ). That is, ∆u is sub m-harmonic function in BR (x0 ). Then by the assumption for k = m, we have 0 ∆u(x ) + m−1 X i=1 ∆u(x0 ) + (2) n2i · i! · m−1 X i=1 2i · i! · r2i Qi−1 j=1 (n r2i Qi j=1 (n i + 2j) + 2j) 0 · ∆ [∆u](x ) ≤ · ∆i [∆u](x0 ) ≤ Z 1 · |∂Br (x0 )| 1 · |Br (x0 )| ∆u(x) dS(x) ∂Br (x0 ) Z ∆u(x) dx. Br (x0 ) Now let φ(r) = u(x0 ) + m X i=1 = 0 u(x ) + n2i · i! · r2i Qi−1 n2i · i! · r2i Qi−1 m X i=1 j=1 (n + 2j) · ∆i u(x0 ) − 1 · |∂Br (x0 )| Z 1 · ∆ u(x ) − · n−1 nα · n r j=1 (n + 2j) i 0 u(x) dS(x) ∂Br (x0 ) Z u(x0 + ry) · rn−1 dS(y) ∂B1 (0) Let x = x0 + ry = 0 u(x ) + m X i=1 n2i · i! · r2i Qi−1 1 · · ∆ u(x ) − nα n j=1 (n + 2j) i 0 Z u(x0 + ry) dS(y). ∂B1 (0) So we can get lim φ(r) = 0. r&0 And φ0 (r) = m X i=1 = m X i=1 = m X i=1 1 2i · r2i−1 · ∆i u(x0 ) − · Qi−1 nαn n2i · i! · j=1 (n + 2j) Z 2i · r2i−1 1 · ∆i u(x0 ) − · Q i−1 nαn n2i · i! · j=1 (n + 2j) Z 2i · r2i−1 1 · ∆i u(x0 ) − · Qi−1 i nαn n2 · i! · j=1 (n + 2j) Z Let x = x0 + ry ∂B1 (0) d u(x0 + ry) dS(y) dr ∇u(x0 + ry) · y dS(y) ∂B1 (0) ∇u(x) · ∂Br (x0 ) x − x0 n−1 ·r dS(x) r 8 MINGFENG ZHAO m X = i=1 m X = i=1 m X = i=1 2i · r2i−1 1 · · ∆i u(x0 ) − Qi−1 i nαn · rn−1 n2 · i! · j=1 (n + 2j) Z 1 2i · r2i−1 · ∆i u(x0 ) − · Qi−1 n−1 i nα · n r n2 · i! · j=1 (n + 2j) Z 1 2i · r2i−1 · ∆i u(x0 ) − · Qi−1 n−1 i nα · n r n2 · i! · j=1 (n + 2j) Z ∇u(x) · n(x) dS(x) ∂Br (x0 ) ∂Br (x0 ) ∂u(x) dS(x) ∂n u(x) dx Br (x0 ) By the Divergence Theorem r · n = " m X i=1 2i · r2i−2 1 · · ∆i u(x0 ) − Qi−1 nαn · rn−1 2i · i! · j=1 (n + 2j) # Z ∆u(x) dx Br (x0 ) " # Z m X r 1 2i · r2i−2 0 i 0 · · ∆u(x ) + · ∆ u(x ) − ∆u(x) dx Qi−1 i n nαn · rn−1 Br (x0 ) j=1 (n + 2j) i=2 2 · i! · " # Z m−1 2i X 2(i + 1) · r r 1 ∆u(x) dx · ∆u(x0 ) + · · ∆i+1 u(x0 ) − Qi i+1 · (i + 1)! · n nαn · rn−1 Br (x0 ) j=1 (n + 2j) i=1 2 " # Z m−1 2i X r r 1 · ∆u(x0 ) + · ∆i [∆u](x0 ) − · ∆u(x) dx Qi i · i! · n nαn · rn−1 Br (x0 ) 2 (n + 2j) j=1 i=1 = = = ≤ 0, By (2) That is, φ(r) is decreasing function on 0 < r < R, in particular, we get φ(r) ≤ lim φ(r) = 0. r&0 That is, we get m X u(x0 ) + n2i · i! · r2i Qi−1 n2i · i! · r2i Qi−1 i=1 1 · |∂Br (x0 )| Z 1 · ∆ u(x ) ≤ · n−1 nα · n r j=1 (n + 2j) Z j=1 (n + 2j) · ∆i u(x0 ) ≤ u(x) dS(x). ∂Br (x0 ) That is, we get 0 u(x ) + m X i=1 i 0 u(x) dS(x). ∂Br (x0 ) Both sides are multiplied by n · rn−1 , and then integrate with respect to r from 0 to r, we can get n · u(x0 ) · Z r tn−1 dt + n · m X 0 i=1 n2i · i! · 1 Qi−1 j=1 (n + 2j) · ∆i u(x0 ) · Z r t2i+n−1 ≤ 0 1 · αn Z r Z u(x) dS(x)dt. 0 ∂Bt (x0 ) That is, we have u(x0 ) · rn + m X r2i+n 1 ≤ · n + 2i αn Z r2i 1 · ∆ u(x ) · ≤ · Q i−1 n i n + 2i α n·r 2 · i! · j=1 (n + 2j) Z i=1 1 2i · i! · Qi−1 j=1 (n + 2j) · ∆i u(x0 ) · u(x) dx. Bt (x0 ) Both sides are divided by rn , we can get 0 u(x ) + m X i=1 1 i 0 u(x) dx. Bt (x0 ) ON LIOUVILLE’S THEOREM FOR BIHARMONIC FUNCTIONS 9 That is, we have u(x0 ) + m X i=1 2i · i! · r2i Qi j=1 (n + 2j) · ∆i u(x0 ) ≤ 1 · |Br (x0 )| Z u(x) dx. Bt (x0 ) Hence, we get that the Claim I is true for k = m + 1. Therefore, by induction, we can conclude that if u is a sub k-harmonic function in BR (x0 ), then for any 0 < r < R, we have u(x0 ) + k−1 X i=1 u(x0 ) + r2i Qi−1 n2i · i! · j=1 (n k−1 X i=1 r2i Qi 2i · i! · j=1 (n + 2j) + 2j) · ∆i u(x0 ) ≤ · ∆i u(x0 ) ≤ Z 1 · |∂Br (x0 )| 1 · |Br (x0 )| u(x) dS(x) ∂Br (x0 ) Z u(x) dx. Br (x0 ) Theorem 5 (The Remark 2, in Page 38, in Huilgol[3]). Let k ∈ N, and u be a bounded k-harmonic in Rn . Then u is a constant function on Rn . Proof. We prove this Theorem by induction. When k = 1, that is, u is a bounded harmonic function in Rn , then by the Usual Liouville Theorem for the Harmonic Functions, Theorem 8, in Page 30, in Evans[1], we know that u is a constant function on Rn . So the Theorem is true for k = 1. When k = 2, that is, u is a bounded biharmonic function in Rn , then by the Theorem 3, we know that u is a constant function on Rn . So the Theorem is true for k = 2. Now we assume the Theorem is true for k = m, that is, if u is a bounded m-harmonic function, then u is a constant function on Rn . Look at k = m + 1, assume that u is a m + 1-harmonic function in Rn , then by the Theorem 4, we know that for any x0 ∈ Rn and fix, then for all r > 0, we have u(x0 ) + m X i=1 2i · i! · r2i Qi j=1 (n + 2j) · ∆i u(x0 ) = 1 · |Br (x0 )| Z u(x) dx. Br (x0 ) Both sides are divided by r2m , we get m (3) u(x0 ) X r2i−2m 1 + · ∆i u(x0 ) = · Q i 2m n+2m i · i! · r α · r n (n + 2j) 2 j=1 i=1 Z u(x) dx. Br (x0 ) And we note that 1 · αn · rn+2m Z u(x) dx ≤ Br (x0 ) ≤ = = 1 · αn · rn+2m Z 1 · αn · rn+2m Z |u(x)| dx Br (x0 ) kukL∞ (Rn ) dx Br (x0 ) kukL∞ (Rn ) · αn · r n αn · rn+2m kukL∞ (Rn ) . r2m 10 MINGFENG ZHAO By taking r % ∞, we get 1 · r%∞ αn · r n+2m Z lim u(x) dx = 0. Br (x0 ) So for (3), we take r % ∞, we can get 2m · m! · 1 Qm j=1 (n + 2j) · ∆m u(x0 ) = 0. By the arbitrary of x0 ∈ Rn , so we get ∆m u(x) = 0, for all x ∈ Rn . By the assumption for k = m, we know that u is a constant function on Rn . That is, the Theorem is true for k = m + 1. Therefore, by induction, we know that the Theorem is true for all k ∈ N. ON LIOUVILLE’S THEOREM FOR BIHARMONIC FUNCTIONS References [1] Lawrence C. Evans. Partial differential equations. American Mathematical Society, 1998. [2] David Gilbarg and Neil S. Trudinger. Elliptic partial differential equations of second order. Springer, 2001. [3] R. R. Huilgol. On liouville’s theorem for biharmonic functions. SIAM J. Appl. Math., 20:37–39, 1971. Department of Mathematics, University of Connecticut, 196 Auditorium Road, Unit 3009, Storrs, CT 06269-3009 E-mail address: mingfeng.zhao@uconn.edu 11