1 Introduction
In this paper, we consider the following prescribed mean
curvature problem on the torus 𝕋n:=ℝn/ℤn:
(1.1)-div(∇u1+|∇u|2)=ν(∇u)⋅g(x,u(x)) on 𝕋n,
where ν is the unit normal vector of u, that is,
ν(z)=11+|z|2(-z,1).
The vector field g(x,xn+1):𝕋n×ℝ→ℝn+1 is given, and we
seek a solution u satisfying (1.1).
The left-hand side of (1.1) represents the mean curvature of the graph of u,
and the right-hand side is the normal component of the vector field g on the graph.
In the case of Dirichlet conditions of a bounded domain Ω⊂ℝn,
prescribed mean curvature problems have been studied by numerous researchers.
Bergner [3] solved the Dirichlet problem in the case where the right-hand side of (1.1)
is H=H(x,u,ν(∇u)) under the assumptions of boundedness (|H|<∞),
monotonicity (∂n+1H≥0), and convexity of Ω.
Under the same conditions for the function H,
Marquardt [9] imposed a condition on ∂Ω depending on H
that guarantees the existence of solutions even for a domain Ω that is not necessarily convex.
In [13], we proved the existence of a solution only under the condition that
the Sobolev norm of H is sufficiently small.
In the case of a compact Riemannian manifold,
Aubin [2] solved the linear elliptic problem -∂i[aij(x)∂ju]=H(x)
if the integrated value of H is zero. The assumption of the
integrated value plays an important role in the existence of solutions to elliptic equations
on a compact Riemannian manifold.
Denny [4] solved the quasilinear elliptic problem -div(a(u(x))∇u)=H(x) on the torus 𝕋n with n=2,3.
Prescribed mean curvature problems on the one-dimensional torus
(u′1+(u′)2)′=H(x,u,u′)
have been investigated
for a wide variety of conditions H (we refer to, for example, [5, 7, 8, 10, 11, 14]).
As we noted in [13], the motivation for the present study comes from a singular perturbation problem, and we proved the following in [12]. Suppose a constant ε>0 and functions ϕε∈W1,2 and gε∈W1,p, with p>n+12, satisfy
-εΔϕε+W′(ϕε)ε=ε∇ϕε⋅gε,
∫(ε|∇ϕε|22+W(ϕε)ε)𝑑x+∥gε∥W1,p(Ω~)≤C,
where W is a double-well potential such as W(ϕ)=(1-ϕ2)2.
Then the interface {ϕε=0} converges locally in the Hausdorff distance to a
surface having a mean curvature given by ν⋅g as ε→0. Here, ν is the unit normal vector of the surface, and g is the weak W1,p limit of gε.
If the surface is represented locally as a graph of a function u on 𝕋n,
we can observe that u satisfies (1.1).
In this paper, we prove the existence of solutions to (1.1) assuming
that the Sobolev norm of g is sufficiently small, gn+1 for the (n+1)-st component is monotonous,
and the integrated value of gn+1 is zero.
The following theorem is the main result.
Theorem 1.1.
Fix n+12<p<n+1 and q=npn+1-p.
Then there exists a constant ε1=ε1(n,p)>0
with the following property:
If ε<ε1, and
g=(g1,…,gn,gn+1)=(g′,gn+1)∈W1,p(𝕋n×(-1,1);ℝn+1)
satisfies the relations
(1.2)∥g∥W1,p(𝕋n×(-1,1))<ε23,
(1.3)∂n+1gn+1(x,xn+1)>ε+ε12|∂n+1g′(x,xn+1)|,
(1.4)∫𝕋ngn+1(x,0)=0,
then there exists a function u∈W2,q(Tn) such that
(1.5)-div(∇u1+|∇u|2)=ν(∇u)⋅g(x,u(x)) on 𝕋n.
Moreover, the following inequality holds:
∥u-∫𝕋nu(y)𝑑y∥W2,q(𝕋n)≤ε12.
Assumptions (1.2) and (1.3) guarantee the existence and uniqueness
of solutions to the linearized problem of (1.1) where a given function depends on ∇u.
Equation (1.4) is necessary for the existence of solutions to elliptic equations on the torus.
To the best of our knowledge, prescribed mean curvature problems on the torus in the general dimension have been insufficiently studied. However, we have proved the existence of the solution under natural assumptions.
The following is the method of proof. First, we find the conditions of H for the linearized problem of (1.1), i.e.
-div(∇u1+|∇v|2)=H,
to have a unique solution. If we add a suitable constant term for any v,
the function ν(∇v)⋅g(x,v(x))
satisfies the conditions.
By estimating the norm of this solution with g, the mapping T(v)=u has a fixed point
using a fixed-point theorem, and Theorem 1.1 follows.
2 Proof of Theorem 1.1
A theorem that holds in the Euclidean space also holds on a torus, as we consider a function on a torus to be a periodic function in the Euclidean space.
Let X(𝕋n) be a function space on 𝕋n.
We define a subspace Xave(𝕋n)⊂X(𝕋n) as
Xave:={w∈X:∫𝕋nw=0}.
Theorem 2.1.
Suppose v∈C1(Tn) and H∈Lave2(Tn).
Then there exists a unique function u∈Wave1,2(Tn) such that
∫𝕋n∇u⋅∇ϕ1+|∇v|2=∫𝕋nHϕ
for all ϕ∈W1,2(Tn).
Proof.
We define a function B:Wave1,2(𝕋n)×Wave1,2(𝕋n)→ℝ by
B[w1,w2,v]:=∫𝕋n∇w1⋅∇w21+|∇v|2.
By the Hölder inequality, we obtain
|B[w1,w2,v]|≤∫𝕋n|∇w1||∇w2|
≤∥∇w1∥L2(𝕋n)∥∇w2∥L2(𝕋n)
(2.1)≤∥w1∥W1,2(𝕋n)∥w2∥W1,2(𝕋n).
Using the Poincaré inequality, we have
|B[w,w,v]|≥11+∥v∥C1(𝕋n)2∥∇w∥L2(𝕋n)2
(2.2)≥11+∥v∥C1(𝕋n)2∥∇w∥W1,2(𝕋n)2.
By (2.1), (2.2), and the Lax–Milgram theorem,
for any H∈Lave2(𝕋n),
there exists a unique function
u∈Wave1,2(𝕋n)
such that
(2.3)∫𝕋n∇u⋅∇ψ1+|∇v|2=∫𝕋nHψ
for all ψ∈Wave1,2(𝕋n).
For any ϕ∈W1,2(𝕋n),
we define cϕ:=∫𝕋nϕ and ϕ~:=ϕ-cϕ∈Wave1,2(𝕋n).
By (2.3) and H∈Lave2(𝕋n), we obtain
∫𝕋n∇u⋅∇ϕ1+|∇v|2=∫𝕋n∇u⋅∇ϕ~1+|∇v|2=∫𝕋nHϕ~=∫𝕋nHϕ.
Thus, Theorem 2.1 follows.
∎
We define a mollifier as follows:
η(x):={Cexp(1|x|2-1)for |x|<1,0for |x|≥1,
where the constant C>0 is selected such that ∫ℝn+1η=1. We define
ηλ(x):=1λnη(xλ).
For any f∈L2(𝕋n×(-1,1)) and xn+1∈(-1+λ,1-λ),
fλ(x,xn+1):=∫𝕋n×(-1,1)ηλ(x-y,xn+1-yn+1)f(y,yn+1)𝑑y
=∫Bn+1(0,λ)ηλ(y,yn+1)f(x-y,xn+1-yn+1)𝑑y,
where Bn+1(x,λ) is an open ball with center x and radius λ in 𝕋n×ℝ.
Moreover, for any
g∈W1,p(𝕋n×(-1,1);ℝn+1),
we define gλ:=(gλ1,…,gλn,gλn+1)=(gλ′,gλn+1).
Lemma 2.2.
Fix β1>0 and 0<λ<1. Suppose v∈C1(Tn) satisfies ∥v∥C1(Tn)<β1, and
g∈W1,p(𝕋n×(-1,1);ℝn+1)
satisfies
∂n+1gn+1(x,xn+1)>β1|∂n+1g′(x,xn+1)|.
For any positive constant c0>0, if v(Tn)+c0⊂(-1+λ,1-λ), then
∫𝕋nν(∇v)⋅gλ(x,v)<∫𝕋nν(∇v)⋅gλ(x,v+c0).
Proof.
From the assumptions, we compute
∫𝕋nν(∇v)⋅(gλ(x,v+c0)-gλ(x,v))
=∫𝕋n11+|∇v|2∫vv+c0-∇v⋅∂n+1gλ′(x,t)+∂n+1gλn+1(x,t)dt
≥∫𝕋n11+|∇v|2∫vv+c0-β1|∂n+1gλ′(x,t)|+∂n+1gλn+1(x,t)dt
≥∫𝕋n11+|∇v|2∫vv+c0∫𝕋n×(-1,1)ηλ(x-y,t-yn+1){-β1|∂n+1g′(y,yn+1)|+∂n+1gn+1(y,yn+1)}𝑑t
>0.
Lemma 2.2 follows.
∎
Lemma 2.3.
Suppose g∈W1,p(Tn×(-1,1)
and v∈C1(Tn) with ∥v∥C1(Tn)≤716.
Let q=npn+1-p. Then there exists a constant c1=c1(n,p)>0 such that,
if λ<18,
∥gλ(⋅,v(⋅))∥Lq(𝕋n)≤c1∥g∥W1,p(𝕋n×(-1,1)).
Proof.
By the same proof as in [13, Lemma 2.3], we obtain
(2.4)∥gλ(⋅,v(⋅))∥Lq(𝕋n)≤c2∥gλ∥W1,p(𝕋n×(-78,78)),
where c2=c2(n,p)>0.
Using the Hölder inequality, we obtain
∫𝕋n×(-78,78)|gλ|p𝑑x≤∫𝕋n×(-78,78)(∫Bn+1(x,λ)ηλ1-1p+1p(x-y,xn+1-yn+1)|g(y,yn+1)|𝑑y)p𝑑x
≤∫𝕋n×(-78,78)(∫Bn+1(x,λ)ηλ(x-y,xn+1-yn+1)|g(y,yn+1)|p𝑑y)𝑑x
≤∫𝕋n×(-1,1)|g(y,yn+1)|p(∫Bn+1(y,λ)ηλ(x-y,xn+1-yn+1)𝑑x)𝑑y
(2.5)=∫𝕋n×(-1,1)|g(y,yn+1)|p𝑑y.
We can show that
∥∇gλ∥Lp(𝕋n×(-78,78))≤∥∇g∥Lp(𝕋n×(-1,1))
in the exact same manner, and
Lemma 2.3 follows by (2.4) and (2.5).
∎
Theorem 2.4.
Suppose v∈C1(Tn) and
g∈W1,p(𝕋n×(-1,1);ℝn+1).
Then there exist constants ε2=ε2(n,p)>0 such that, if λ<18, ε<ε2, and ∥v∥C1(Tn)≤ε1/2, then
g satisfies (1.2)–(1.4). Then
there exist a unique function u∈Wave1,2(Tn) and
a unique constant -14<cv<14 such that
(2.6)∫𝕋n∇u⋅∇ϕ1+|∇v|2=∫𝕋nν(∇v)⋅gλ(x,v+cv)ϕ
for all ϕ∈W1,2(Tn).
Proof.
We define
F(t):=∫𝕋nν(∇v)⋅gλ(x,v+t).
The function F is continuous. Suppose that ε<1162. We will consider that
the domain of F is [-14,14].
By the mean value theorem,
there exists a constant c3=c3(n,p)>0 such that
F(14)=∫𝕋n(ν(∇v)-ν(0)+ν(0))⋅gλ(x,v+14)
(2.7)≥-c3∥v∥C1(𝕋n)∥gλ(⋅,v(⋅)+14)∥Lq(𝕋n)+∫𝕋ngλn+1(x,v+14).
By Lemma 2.3 and ∥v+14∥C1(𝕋n)≤516, we obtain
∥gλ(⋅,v(⋅)+14)∥Lq(𝕋n)≤c1∥g∥W1,p(𝕋n×(-1,1)).
By (1.3) and (1.4),
there exists a constant c4=c4(n)>0 such that
∫𝕋ngλn+1(x,v+14)=∫𝕋n∫Bn+1(0,λ)ηλ(y,yn+1)gn+1(x-y,v+14-yn+1)𝑑y𝑑x
>∫𝕋n∫Bn+1(0,λ)ηλ(y,yn+1)gn+1(x-y,116)𝑑y𝑑x
>∫𝕋n∫Bn+1(0,λ)ηλ(y,yn+1)(gn+1(x-y,0)+ε16)𝑑y𝑑x
(2.8)>c416ε.
By (1.2), (2.7)–(2.8), and ∥v∥C1(𝕋n)<ε1/2, if
ε<(c416c1c3)6=:ε2(n,p),
then
F(14)>-c1c3∥v∥C1(𝕋n)∥gλ∥W1,p(𝕋n×(-1,1))+c416ε
>-c1c3ε76+c416ε
>ε(-c1c3ε16+c416)
>0.
Similarly, we can show that
F(-14)<0.
By Lemma 2.2 and the mean value theorem,
there exists a unique constant -14<cv<14 that satisfies F(cv)=0.
By using Theorem 2.1, Theorem 2.4 follows.
∎
Let us define an operator T:𝒜(s)→Wave1,2(𝕋n)×[-14,14] by T(v)=(T1(v),T2(v)):=(u,cv) that satisfies
(2.6), where
𝒜(s):={w∈Wave2,q(𝕋n):∥w∥W2,q(𝕋n)≤s}.
Theorem 2.5.
There exist constants ε3=ε3(n,p)>0 and
c5=c5(n,p)>0 such that,
if λ<18, ε<min{ε2,ε3}, v∈A(ε1/2),
and g∈W1,p(Tn×(-1,1);Rn+1) satisfies (1.2)–(1.4),
then
∥T1(v)∥W2,q(𝕋n)≤c5∥g∥W1,p(𝕋n×(-1,1)).
Proof.
We first assume that v∈C∞(𝕋n)∩𝒜(ε1/2).
Using [6, Corollary 8.11], we obtain T1(v)∈C∞(𝕋n). Thus, we can
rewrite (2.6) as
ΔT1(v)1+|∇v|2+∇T1(v)⋅∇(11+|∇v|2)=-ν(∇v)⋅gλ(x,v+T2(v)).
Using [6, Theorem 9.11],
we find that there exists a constant
c6=c6(n,p)>0 such that
∥T1(v)∥W2,q(𝕋n)≤c6(∥T1(v)∥Lq(𝕋n)+∥ν(∇v)⋅gλ(x,v+T2(v))∥Lq(𝕋n)
(2.9)+∥∇T1(v)⋅∇(11+|∇v|2)∥Lq(𝕋n)).
Using Lemma 2.3, we obtain
(2.10)∥ν(∇v)⋅gλ(x,v+T2(v))∥Lq(𝕋n)≤c1∥g∥W1,p(𝕋n×(-1,1)).
Using the Sobolev inequality,
we find that there exists a constant
c7=c7(n,p)>0 such that
∥∇T1(v)⋅∇(11+|∇v|2)∥Lq(𝕋n)≤∥T1(v)∥C1(𝕋n)∥∇(11+|∇v|2)∥Lq(𝕋n)
(2.11)≤c7∥T1(v)∥W2,q(𝕋n)∥v∥W2,q(𝕋n).
Next, we estimate the term ∥T1(v)∥Lq(𝕋n). If q≤2, then, by (2.2) and
Lemma 2.3, we obtain
∥T1(v)∥Lq(𝕋n)≤c8(n,p)∥T1(v)∥L2(𝕋n)
≤c9(n,p)B[T1(v),T1(v),v]12.
=c9(∫𝕋n∇T1(v)⋅∇T1(v)1+|∇v|2)12
=c9(∫𝕋nν(∇v)⋅gλ(x,v+T2(v))T1(v))12
≤c10(n,p)∥g∥W1,p(𝕋n)12∥T1(v)∥L∞(𝕋n)12
(2.12)≤c11(n,p)∥g∥W1,p(𝕋n)+14c6∥T1(v)∥W2,q(𝕋n).
If q>2, by (2.12) and the Riesz–Thorin theorem, we obtain
∥T1(v)∥Lq(𝕋n)≤∥T1(v)∥L2(𝕋n)1q∥T1(v)∥L2(𝕋n)1-1q
≤c12(n,p)∥g∥W1,p(𝕋n)12q∥T1(v)∥L∞(𝕋n)12q+1-1q
(2.13)≤c13(n,p)∥g∥W1,p(𝕋n)+14c6∥T1(v)∥W2,q(𝕋n).
By (2.9)–(2.13), there exists a constant c14=c14(n,p)>0
such that
∥T1(v)∥W2,q(𝕋n)≤c14(∥g∥W1,p(𝕋n×(-1,1))+∥T1(v)∥W2,q(𝕋n)∥v∥W2,q(𝕋n))+14∥T1(v)∥W2,q(𝕋n).
If ε<116c142, we obtain
(2.14)∥T1(v)∥W2,q(𝕋n)≤2c14∥g∥W1,p(𝕋n×(-1,1)).
For the general case of v∈W2,q(𝕋n),
suppose that {vm}m∈ℕ∈C∞(𝕋n) converges to v in the sense of C1(𝕋n).
By (2.14), there exists a subsequence
{vmk}k∈ℕ⊂{vm}m∈ℕ
such that T1(vmk) converges to a function w∞∈W2,q(𝕋n)
in the sense of C1(𝕋n), and
T2(vmk) converges to a constant d∞∈[-14,14].
For any ϕ∈W1,2(𝕋n), we obtain
∫𝕋nν(∇v)⋅gλ(x,v+d∞)ϕ-ν(∇vmk)⋅gλ(x,vmk+T2(vmk))ϕ
≤∫𝕋n|ϕ||ν(∇v)-ν(∇vmk)||gλ(x,vmk+T2(vmk))|+∫𝕋n|ϕ||∫vmk+T2(vmk)v+d∞∂n+1gλ(x,s)|
(2.15)→0 (k→∞)
and
∫𝕋n∇w∞⋅∇ϕ1+|∇v|2-∇T1(vmk)⋅∇ϕ1+|∇vmk|2
≤∫𝕋n(∇w∞-∇T1(vmk))⋅∇ϕ1+|∇v|2+∫𝕋n(∇T1(vmk)⋅∇ϕ)(11+|∇v|2-11+|∇vmk|2)
(2.16)→0 (k→∞).
By (2.15) and (2.16), we obtain
∫𝕋n∇w∞⋅∇ϕ1+|∇v|2-ν(∇v)⋅gλ(x,v+d∞)ϕ
=limk→∞∫𝕋n∇T1(vmk)⋅∇ϕ1+|∇vmk|2-ν(∇vmk)⋅gλ(x,vmk+T2(vmk))ϕ
(2.17)=0,
that is, T(v)=(w∞,d∞). By (2.14) and (2.17), Theorem 2.5 follows.
∎
Next, we provide the fixed-point theorem, which is needed later ([1, Theorem 1]).
An operator T:X→A is considered weakly sequentially continuous if,
for every sequence {xm}m∈ℕ⊂X and x∞∈X such that
xm weakly converges to x∞, T(xm) weakly converges to T(x∞).
Theorem 2.6.
Let X be a metrizable, locally convex topological vector space and let Ω be a weakly compact convex subset of X. Then any weakly
sequentially continuous map T:Ω→Ω has a fixed point.
We first prove Theorem 1.1 in the case of gλ.
Theorem 2.7.
There exists a constant ε4=ε4(n,p)>0 such that,
if λ<18 and ε<ε4, then
g∈W1,p(𝕋n×(-1,1);ℝn+1)
satisfies (1.2)–(1.4).
Then there exists a function uλ∈W2,q(Tn) such that
(2.18)-div(∇uλ1+|∇uλ|2)=ν(∇uλ)⋅gλ(x,uλ(x)) on 𝕋n.
Proof.
The set W2,q(𝕋n) is a metrizable, locally convex topological vector space, and
the set 𝒜(ε1/2) is a weakly compact convex subset of W2,q(𝕋n).
By (1.2) and Theorem 2.5, if
ε<min{ε2,ε3,c5-6}=:ε4,
we have
∥T1(v)∥W2,q(𝕋n)≤c5∥g∥W1,p(𝕋n×(-1,1))
≤c5ε16ε12
(2.19)≤ε12 for any v∈𝒜(ε12),
that is,
T1(𝒜(ε1/2))⊂𝒜(ε1/2).
Suppose that {vm}m∈ℕ weakly converges to v∞ in the sense of W2,q(𝕋n).
According to Theorem 2.5, there exists a subsequence {vmk}k∈ℕ⊂{vm}m∈ℕ
such that T1(vmk) weakly converges to a function w∞∈W2,q(𝕋n)
in the sense of W2,q(𝕋n), and
T2(vmk) converges to a constant d∞∈[-14,14].
By the same argument (2.15)–(2.17), for any ϕ∈W2,q(𝕋n),
∫𝕋n∇w∞⋅∇ϕ1+|∇v∞|2-ν(∇v∞)⋅gλ(x,v∞+d∞)ϕ=0,
that is, we obtain limk→∞T1(vmk)=T1(v∞)
by the uniqueness of solution of Theorem 2.4.
Therefore, every convergent subsequence of {T1(vm)}
converges to T1(v∞), and T1 is a weakly sequentially continuous map.
Using Theorem 2.6, we obtain a function
vλ∈Wave2,q(𝕋n) satisfying
-div(∇vλ1+|∇vλ|2)=ν(∇vλ)⋅gλ(x,vλ(x)+T2(vλ)) on 𝕋n,
that is, uλ:=vλ+T2(vλ)∈W2,q(𝕋n) satisfying (2.18).
∎
Proof of Theorem 1.1.
Suppose uλ∈W2,q(𝕋n) satisfies (2.18).
By Theorem 2.5, there exists a convergent subsequence
{uλk}k∈ℕ⊂{uλ}0<λ<18
with a limit u∞∈W2,q(𝕋n) in the sense of C1(𝕋n) and λk→0. We show that u∞ satisfies (1.5). For any ϕ∈W1,2(𝕋n), we obtain
∫𝕋n-div(∇uλk1+|∇uλk|2-∇u∞1+|∇u∞|2)ϕ=∫𝕋n(∇uλk1+|∇uλk|2-∇u∞1+|∇u∞|2)⋅∇ϕ
(2.20)→0.
Using Lemma 2.3, we have
∫𝕋nν(∇uλk)⋅gλk(x,uλk)-ν(∇u∞)⋅g(x,u∞)
=∫𝕋n(ν(∇uλk)-ν(∇u∞))⋅gλk(x,uλk)+∫𝕋nν(∇u∞)⋅(gλk(x,uλk)-g(x,uλk))
+∫𝕋nν(∇u∞)⋅(g(x,uλk)-7g(x,u∞))
=c1∥ν(∇uλk)-ν(∇u∞)∥C0(𝕋n)∥gλk∥W1,p(𝕋n×(-1,1))+c1∥gλk-g∞∥W1,p(𝕋n×(-1,1))+∫𝕋n|∫u∞uλk∂n+1g(x,s)|
(2.21)→0.
By (2.20) and (2.21), we obtain
∫𝕋n-div(∇u∞1+|∇u∞|2)ϕ-ν(∇u∞)⋅g(x,u∞)ϕ
=limk→∞∫𝕋n-div(∇uλk1+|∇uλk|2)ϕ-ν(∇uλk)⋅gλk(x,uλk)ϕ
→0.
Thus, u∞ satisfies (1.5)
using the fundamental lemma of the calculus of variations.
By (2.19), we obtain
∥u∞-∫𝕋nu∞(y)𝑑y∥W2,q(𝕋n)≤ε12,
and Theorem 1.1 follows.
∎