Symmetry of solutions of minimal gradient graph equations on punctured space
Zixiao Liu, Jiguang Bao*
April 23, 2021
Abstract In this paper, we study symmetry and existence of solutions of minimal gradient graph equations on punctured space Rn \{0}, which include the Monge-Ampere` equation, inverse harmonic Hessian equation and the special Lagrangian equation. This extends the classifica- tion results of Monge-Ampere` equations. Under some conditions, we also give the character- ization of the solvability on exterior Dirichlet problem in terms of their asymptotic behaviors. Keywords: minimal gradient graph equation, optimal symmetry, existence. MSC 2020: 35J60; 35B06.
1 Introduction
We consider the following fully nonlinear elliptic equations
2 n Fτ (λ(D u)) = C0 in R \{0}, (1.1)
2 2 where C0 is a constant, λ(D u) = (λ1, λ2, ··· , λn) are n eigenvalues of Hessian matrix D u, π τ ∈ [0, 2 ] and
1 n X ln λi, τ = 0, n √ i=1 a2 + 1 n λ + a − b X i π ln , 0 < τ < 4 , 2b λi + a + b arXiv:2104.10904v1 [math.AP] 22 Apr 2021 i=1 √ n X 1 π Fτ (λ) := − 2 , τ = 4 , 1 + λi √ i=1 2 n a + 1 X λi + a − b arctan , π < τ < π , b λ + a + b 4 2 i=1 i n X arctan λ , τ = π , i 2 i=1
*Supported in part by National Natural Science Foundation of China (11871102 and 11631002).
1 a = cot τ, b = p|cot2 τ − 1|. Equations (1.1) origin from gradient graph (x, Du(x)) with zero n n mean curvature under various metrics equipped on R × R . In 2010, Warren [32] first proved 2 2 n that if u ∈ C (Ω) is a solution of Fτ (λ(D u)) = C0 in Ω ⊂ R , then the volume of (x, Du(x)) π π π 1 is a maximal for τ ∈ [0, 4 ) and minimal for τ ∈ ( 4 , 2 ] among all homologous, C , space-like n n n-surfaces in (R × R , gτ ), where h π i g = sin τδ + cos τg , τ ∈ 0, , τ 0 0 2 is the linearly combined metric of standard Euclidean metric
n n X X δ0 = dxi ⊗ dxi + dyj ⊗ dyj, i=1 j=1 and the pseudo-Euclidean metric
n n X X g0 = dxi ⊗ dyi + dyj ⊗ dxj. i=1 j=1
If τ = 0, then (1.1) becomes the Monge-Ampere` equation
det D2u = enC0 .
n For the Monge-Ampere` equation on entire R , the classical theorem by Jorgens¨ [22], Calabi [11] 2 n and Pogorelov [30] states that any convex classical solution of det D u = 1 in R must be a quadratic polynomial. See Cheng-Yau [13], Caffarelli [5] and Jost-Xin [23] for different proofs n and extensions. For the Monge-Ampere` equation in exterior domain of R , there are exterior Jorgens-Calabi-Pogorelov¨ type results by Ferrer-Mart´ınez-Milan´ [15] for n = 2 and Caffarelli-Li [7], which state that any locally convex solution must be asymptotic to quadratic polynomials (for n = 2 we need additional ln-term) near infinity. For the Monge-Ampere` equation on punctured n space R \{0}, there are classification results by Jorgens¨ [22] for n = 2 and Jin-Xiong [21] that shows every locally convex function of det D2u = 1, modulo an affine transform, has to be |x| 1 R n n u(x) = 0 (r + c) dr for some c ≥ 0. For further generalization and refined asymptotics on Jorgens-Calabi-Pogorelov¨ type results we refer to [3, 8, 18, 31] and the references therein. π If τ = 2 , then (1.1) becomes the special Lagrangian equation n X 2 arctan λi D u = C0. (1.2) i=1 n For special Lagrangian equation on entire R , there are Bernstein-type results by Yuan [33, 34], which state that any classical solution of (1.2) and ( −KI, n ≤ 4, n − 2 D2u ≥ or C > π, (1.3) −( √1 + (n))I, n ≥ 5, 0 3 2 must be a quadratic polynomial, where I denote the identity matrix, K is any constant and (n) n is a small dimensional constant. For special Lagrangian equation on exterior domain of R , there
2 is an exterior Bernstein-type result by Li-Li-Yuan [25], which states that any classical solution of (1.2) on exterior domain with (1.3) must be asymptotic to quadratic polynomial (for n = 2 we need additional ln-term) near infinity. Furthermore, Chen-Shankar-Yuan [12] proved that a convex viscosity solution of (1.2) must be smooth. π Pn 1 If τ = , then (1.1) is a translated inverse harmonic Hessian equation 2 = 1, 4 i=1 λi(D u) n which is a special form of Hessian quotient equation. For Hessian quotient equation on entire R , there is a Bernstein-type result by Bao-Chen-Guan-Ji [1]. π For general τ ∈ [0, 2 ], Warren [32] proved the Bernstein-type results under suitable semi- convex conditions by the results of Jorgens¨ [22]-Calabi [11]-Pogorelov [30], Flanders [16] and Yuan [33, 34]. In our earlier work [28], we generalized the results to equation (1.1) on exterior domain and provide finer asymptotic expansions. For further generalization with perturbed right hand side we refer to [29] etc. In this paper, we focus on n ≥ 3 and prove a type of classification results as in Jorgens¨ [22] and Jin-Xiong [21]. T n T Hereinafter, we let x denote the transpose of vector x ∈ R , O denote the transpose of n×n matrix O, Sym(n) denote the set of symmetric n × n matrix, σk(λ) denote the kth elementary form of λ and DFτ (λ(A)) denote the matrix with elements being value of partial derivative of Fτ (λ(M)) w.r.t Mij variable at matrix A. A function u(x) is called generalized symmetric with respect to A ∈ Sym(n) if it relies only on the value of xT Ax. As in [6], we say u ∈ C0(Ω) is n 2 a viscosity subsolution (supersolution) of (1.1) in Ω ⊂ R if for any function ψ ∈ C (Ω) with 2 2 2 2 π π D ψ > 0, D ψ > (−a + b)I, D ψ > −I, D ψ > −(a + b)I for τ = 0, τ ∈ (0, 4 ), τ = 4 , π π τ ∈ ( 4 , 2 ) respectively and point x ∈ Ω satisfying ψ(x) = u(x) and ψ ≥ (≤)u in Ω, we have 2 Fτ (λ(D ψ(x))) ≥ (≤)C0. If u is both viscosity subsolution and supersolution of (1.1), we say it is a viscosity solution. By a direct computation (see for instance [19, 32]), if λi > −a − b, for i = 1, 2, ··· , n, then n n X λi + a − b X λi + a nπ arctan = arctan − . (1.4) λ + a + b b 4 i=1 i i=1
Hence replacing u(x) by (a−b) 2 π u(x) + 2 |x| , 0 < τ < 4 , 1 2 π u(x) + 2 |x| , τ = 4 , (1.5) u(x) a 2 π π b + 2b |x| , 4 < τ < 2 , it is equivalent to consider 2 n Gτ (λ(D u)) = C0 in R \{0}, (1.6)
3 where n 1 X ln λi, τ = 0, n √ i=1 2 n a + 1 X λi ln , 0 < τ < π , 2b λ + 2b 4 i=1 i n √ X 1 G (λ) := − 2 , τ = π , τ λ 4 i=1 i √ n ! a2 + 1 X nπ arctan λ − π < τ < π , b i 4 4 2 i=1 n X arctan λ , τ = π . i 2 i=1 Our first main result provides a type of symmetry and relationship between asymptotic at infinity and value at origin.
2 n 0 n Theorem 1.1. Let u ∈ C (R \{0}) be a classical convex solution of (1.6). Then u ∈ C (R ),
2 n Gτ (λ(D u)) ≥ C0 in R (1.7)
n in viscosity sense and there exist c ∈ R, β ∈ R and A ∈ Sym(n) with Gτ (λ(A)) = C0 such that 1 u(x) ≤ xT Ax + βx + c in n. (1.8) 2 R Furthermore, u(x) − βx is symmetric in eigenvector directions of A, i.e.,
n T T u(xe) − βxe = u(x) − βx, ∀ {xe ∈ R :(O xe)i = ±(O x)i, ∀ i = 1, 2, ··· , n}, (1.9) where O is an orthogonal matrix such that OAOT is diagonal.
Remark 1.2. After a change of coordinate y := OT x, u(y) − βy is symmetric with respect to n coordinate planes. Especially, u(y) − βy is central symmetry i.e., u(y) − βy = u(−y) + βy.
Remark 1.3. The solutions of (1.1) and (1.6) have asymptotic behavior and higher order expan- sion at infinity, see for instance earlier results by the authors [28, 29].