Nonlinear Analysis 171 (2018) 32–40

Contents lists available at ScienceDirect

Nonlinear Analysis
www.elsevier.com/locate/na

Superposition principle on the viscosity solutions of infinity

Laplace equations
Guanghao Hong*, Xiaomeng Feng
School of Mathematics and Statistics, Xi’an Jiaotong University, Xi’an, 710049, PR China

article info abstract

Article history: We consider the sum of the solutions of two infinity Laplace equations in disjoint
Received 31 August 2017 variables. We prove that the superposed function is a viscosity solution of the infinity
Accepted 21 January 2018 Laplace equation in the extension domains with the sum of inhomogeneous terms
Communicated by Enzo Mitidieri if one of the solutions is in the sense of viscosity and the other is in the classical
sense. We also construct a counterexample to show that the conclusion may not be
MSC: true if both of the solutions are merely in the viscosity sense.
primary 35J25
© 2018 Elsevier Ltd. All rights reserved.
35J70
secondary 49N60
Keywords:
Superposition principle
Infinity Laplace equation
Viscosity solution

1. Introduction

The infinity Laplace equation

∑
△∞ u(x) := uxi uxj uxi xj = 0
1≤i,j≤n

was introduced by G. Aronsson [1] in the 1960s. R. Jensen [10] proved the equivalence of the infinity Laplace
equation and the absolutely minimizing Lipschitz extension problem. He also proved the existence and
uniqueness of the viscosity solution to the Dirichlet problem:

△∞ u = 0 in Ω , u = g on ∂Ω

for any bounded domain Ω ⊂ Rn and g ∈ C(∂Ω ). Crandall–Evans–Gariepy [2] introduced the property of
comparison with cones and proved that it is a characteristic property of infinity harmonic functions. The
interior regularity for infinity harmonic functions was achieved by Evans, Savin and Smart in [4,14] and [3].
The boundary regularity was studied by Wang–Yu [15], Hong [6,8] and Hong-Liu [9].

* Corresponding author.
E-mail addresses: [email protected] (G. Hong), [email protected] (X. Feng).

https://doi.org/10.1016/j.na.2018.01.011
0362-546X/© 2018 Elsevier Ltd. All rights reserved.
 G. Hong, X. Feng / Nonlinear Analysis 171 (2018) 32–40 33

The inhomogeneous infinity Laplace equation:

△∞ u = f in Ω (1)

was introduced by Lu-Wang [13]. Lindgren [11] proved that the blow-ups are linear if f ∈ C(Ω )∩L∞ (Ω ) and
u is everywhere differentiable if f ∈ C 1 (Ω ) ∩ L∞ (Ω ). Hong [7] proved the boundary differentiability of u at
a differentiable boundary point and Feng–Hong [5] studied the slope estimate and boundary differentiability
of u on the convex domains.
In [11], Lindgren constructed an extension
4
ũ(x1 , . . . , xn+2 ) = u(x1 , . . . , xn ) + 5xn+1 + C|xn+2 | 3
6
and used the conclusion that if △∞ u = f in Rn in the viscosity sense then △∞ ũ = f + 324 C in Rn+2 in
the viscosity sense without a proof. Both of the papers [7] and [5] used the same extension and conclusion.
The purpose of the extension is to make the slope function strictly positive and the inhomogeneous term
bounded away from 0. The conclusion seems obvious but we will see it is not so. In the book [12](Page
58), Lindqvist also used the similar extension and conclusion. The author gave a very short proof of the
conclusion in the footnote, but we do not think the proof is strict enough. The last sentence of the proof
says “The desired inequality follows”, but we cannot see why the inequality follows from the proceeding
deduction. The argument does not involve an analysis on the second order derivatives, the counterexample
in this paper indicates that one should not prove the conclusion without going deep into the analysis on the
second order derivatives.
In this note, we will give a strict proof of the above mentioned conclusion and provide a counterexample
to show that the things are not that simple. We begin by recalling the definition of viscosity solution.

Definition 1. Let u ∈ C(Ω ), x0 ∈ Ω and K ∈ R. We say △∞ u(x0 ) ≥ K in the viscosity sense if

△∞ φ(x0 ) ≥ K whenever φ ∈ C 2 (D) for some neighborhood D of x0 , φ(x0 ) = u(x0 ) and φ(x) ≥ u(x) in D.
We have the similar definition for △∞ u(x0 ) ≤ K in the viscosity sense and we say △∞ u(x0 ) = K in the
viscosity sense if u satisfies both △∞ u(x0 ) ≥ K and △∞ u(x0 ) ≤ K in the viscosity sense.

We will prove the following main conclusion in Section 2.

Theorem 1. Let x′0 ∈ Ω ′ ⊂ Rk , x′′0 ∈ Ω ′′ ⊂ Rn−k , Ω := Ω ′ × Ω ′′ ⊂ Rn and x0 = (x′0 , x′′0 ) ∈ Ω . Assume

that v ∈ C(Ω ′ ) satisfies

△∞ v(x′0 ) ≥ K ′

in the viscosity sense and w ∈ C(Ω ′′ ) is second order differentiable at x′′0 and satisfies

△∞ w(x′′0 ) ≥ K ′′

in the classical sense. Let u(x) := v(x′ ) + w(x′′ ) for x = (x′ , x′′ ) ∈ Ω . Then u ∈ C(Ω ) satisfies

△∞ u(x0 ) ≥ K ′ + K ′′

in the viscosity sense.

In Section 3, we will give a counterexample to show that if △∞ w(x′′0 ) ≥ K ′′ is also merely true in the
viscosity sense then we may not have △∞ u(x0 ) ≥ K ′ + K ′′ in the viscosity sense.
34 G. Hong, X. Feng / Nonlinear Analysis 171 (2018) 32–40

2. Proof of Theorem 1

Proof of Theorem 1. Without loss of generality, we may assume that

x′0 = 0 ∈ B ′ := B1k (0) ⊂ Ω ′ , v(0) = 0,

x′′0 = 0 ∈ B ′′ := B1n−k (0) ⊂ Ω ′′ , w(0) = 0,

x0 = 0 ∈ B := B1n (0) ⊂ B ′ × B ′′ ⊂ Ω , u(0) = v(0) + w(0) = 0.

For Φ ∈ C 2 (B) satisfying

Φ(0) = u(0) = 0 and Φ(x) ≥ u(x) in B,

we have DΦ(0) = pe1 + qek+1 by doing some coordinate transformations. We want to show that

△∞ Φ(0) = DΦ(0) · D2 Φ(0)DΦ(0) = p2 ã + q 2 b̃ + 2pqc̃ ≥ K ′ + K ′′

where ã = Φ11 (0), b̃ = Φk+1,k+1 (0) and c̃ = Φ1,k+1 (0) = Φk+1,1 (0).
Clearly, w(x′′ ) is second order differentiable at 0 and Dw(0) = qek+1 . We denote b := wk+1,k+1 (0). It is
easy to see

b ≤ b̃.

We denote
v(te) − pe1 · te
ae := lim 1 2
2t
t→0

and

a := lim ae
e→e1

where e ∈ span{e1 , . . . , ek } and |e| = 1. For any ε > 0, there exists θ0 > 0 such that
ε
ae ≤ a +
2
for any

e ∈ Cθ0 := {e ∈ S k−1 : dS k−1 (e, e1 ) ≤ θ0 }.

From the definition of ae , there exists te > 0 such that

1 ε
v(te) ≤ pe1 · te + (ae + )t2 for all t ∈ (−te , te ).
2 2
From the continuity of v and the compactness of Cθ0 , we have

t0 := inf {te } > 0.

e∈Cθ
0

Therefore, for any t ∈ (−t0 , t0 ) and e ∈ Cθ0 ,

1
v(te) ≤ pe1 · te + (a + ε)t2 . (2)
2
It is clear that

a ≤ ã.
 G. Hong, X. Feng / Nonlinear Analysis 171 (2018) 32–40 35

In the following we show that

△∞ Φ(0) = p2 ã + q 2 b̃ + 2pqc̃
= p2 a + q 2 b + p2 (ã − a) + q 2 (b̃ − b) + 2pqc̃
≥ K ′ + K ′′ .
Clearly,

△∞ w(0) = q 2 b ≥ K ′′ .

Let

[D2 Φ(0)]k×k := D2 Φ(0)⏐span{e

⏐
1 ,...,ek }

and Λ denotes the maximum eigenvalue of [D2 Φ(0)]k×k . For arbitrary ε > 0, let
1 1 2 2
φε (x′ ) := pe1 · x′ + (a + ε)x21 + M (|x2 | + · · · + |xk | ) (3)
2 2
Λ−a
where M > tan2 θ0
. Then

φε (x′ ) ≥ v(x′ ) in Bt0 (0) \ Cθ0 × (−t0 , t0 ).

Combine (2) and (3), we must have

φε (x′ ) ≥ v(x′ ) in Cθ0 × (−t0 , t0 ).

Therefore,

φε (x′ ) ≥ v(x′ ) in Bt0 (0).

According to the definition of the viscosity subsolution, it follows that

△∞ φε (0) = pe1 · D2 φε (0)pe1 = p2 (a + ε) ≥ K ′ .

Sending ε → 0, we have

p2 a ≥ K ′ .

It remains to show

p2 (ã − a) + q 2 (b̃ − b) + 2pqc̃ ≥ 0,

that is
( )
ã − a c̃
A= (4)
c̃ b̃ − b
is a 2 × 2 nonnegative definite matrix. We prove (4) by contradiction. If A has a negative eigenvalue
−λ0 < 0, then there exists δ > 0 such that for all symmetric matrix B satisfying |B − A| < δ, B has a
negative eigenvalue −λ < − λ20 < 0. By the definition of a, we have for any 0 < ε < λ40 and any θ > 0, there
exist e ∈ Cθ and a sequence {tj } with tj → 0, such that
1
v(tj e) > pe1 · tj e + (a − ε)t2j . (5)
2
Define

v̄(t) := v(te), w̄(s) := w(sek+1 ),

36 G. Hong, X. Feng / Nonlinear Analysis 171 (2018) 32–40

ū(t, s) := v̄(t) + w̄(s) = u(te + sek+1 ),

Φ̄(t, s) := Φ(te + sek+1 ).

By (5), we obtain
1
v̄(tj ) > pe1 · etj + (a − ε)t2j .
2
Clearly,
Φ̄(0, 0) = ū(0, 0), Φ̄(t, s) ≥ ū(t, s),

Φ̄(0, 0) = (pe1 · e, q),

( ) ( )
2
Φee (0) Φe,k+1 (0) Φee (0) Φe,k+1 (0)
D Φ̄(0, 0) = = .
Φk+1,e (0) Φk+1,k+1 (0) Φk+1,e (0) b̃
We denote
( )
Φee (0) − a Φe,k+1 (0)
B= .
Φk+1,e (0) b̃ − b
Let θ be sufficiently small, then |e − e1 | is very small. It follows that
|B − A| < δ.
Then B has a negative eigenvalue −λ < − λ20 < 0 and there exists ξ = (ξ 1 , ξ 2 ), |ξ| = 1 such that
Bξ = −λξ and ξ · Bξ = −λ < 0.
Notice that b̃ − b ≥ 0, then ξ ̸= (0, 1). Set
ξ2
sj := tj and ζj := (tj , sj ).
ξ1
It follows that
2 λ0 2
ζj · Bζj = −λ|ζj | < − |ζj | < 0.
2
λ0
For small enough tj and 0 < ε < 4 < λ2 ,
1 2
Φ̄(tj , sj ) = Φ̄(0, 0) + DΦ̄(0, 0) · ζj + ζj · D2 Φ̄(0, 0)ζj + o(|ζj | )
2
1 1 1 2
= 0 + pe1 · etj + qsj + at2j + bs2j + ζj · Bζj + o(|ζj | )
2 2 2
1 1 2 2
< v̄(tj ) + w̄(sj ) + εt2j − λ|ζj | + o(|ζj | )
2 2
1 2
≤ ū(tj , sj ) + (ε − λ)|ζj |
2
< ū(tj , sj )
which is a contradiction to Φ̄(t, s) ≥ ū(t, s). In the third line of the above inequality, we used that
2
w̄(sj ) + o(|sj | ) ≥ qsj + 12 bs2j for small enough sj . This is true because w is second order differentiable.
2
If we only assume that w solves the inequality in the viscosity sense, then w̄(sj ) + o(|sj | ) ≥ qsj + 21 bs2j is
only true for a special choice of the sequence {sj }. But here {sj } are decided by {tj } (which is a special
choice for v) and ξ so cannot meet a specified requirement of w simultaneously. This is why we must require
one of v and w to be second order differentiable. The counterexample in the next section exposes this fact
from the opposite angle.
Hence A is a nonnegative definite matrix. Theorem 1 is established. □
 G. Hong, X. Feng / Nonlinear Analysis 171 (2018) 32–40 37

Remark 1. Similarly, if v ∈ C(Ω ′ ) satisfies

△∞ v(x′0 ) ≤ K ′
in the viscosity sense and w ∈ C(Ω ′′ ) is second order differentiable at x′′0 and satisfies
△∞ w(x′′0 ) ≤ K ′′
in the classical sense. Then u = v + w ∈ C(Ω ) satisfies
△∞ u(x0 ) ≤ K ′ + K ′′
in the viscosity sense.

Theorem 1 concerns the conclusion on the pointwise situation, whereas we now focus on the whole domain.
It is easy to get the following corollary.

Corollary 1. Let x′ ∈ Ω ′ , x′′ ∈ Ω ′′ , x = (x′ , x′′ ) ∈ Ω := Ω ′ × Ω ′′ . Assume that v(x′ ) ∈ C(Ω ′ ) is a viscosity
solution of
△∞ v(x′ ) = f (x′ ) in Ω ′
and w(x′′ ) ∈ C 2 (Ω ′′ ) is a classical solution of
△∞ w(x′′ ) = g(x′′ ) in Ω ′′ .
Then u(x) = v(x′ ) + w(x′′ ) ∈ C(Ω ) is a viscosity solution of
△∞ u(x) = f (x′ ) + g(x′′ ) in Ω .

3. A counterexample

The following example shows that Theorem 1 cannot in general be extended to merely assume w ∈ C(Ω ′′ )
satisfying △∞ w(x′′0 ) ≥ K ′′ in the viscosity sense. The counterexample is in dimension 2.
Let
1 1
v(x) = x + x2 τ (x) and w(y) = y + y 2 τ (3y)
2 2
where τ (x) ∈ L∞ (−1, 1) ∩ C ∞ ((−1, 1)/{0}), τ (0) = 0, τ (−x) = τ (x) and on (0, 1) τ (x) is given by
⎧ 1.2
⎪ τ (x) = 1, if x = k+1
2
⎪
⎪
⎪
1.2 1.2 + ε
⎪
⎪
⎨ 1 > τ (x) > −2, if
⎪
⎪ < x < k+1
2k+1 2
1.2 + ε 1.2 − ε
τ (x) = −2, if ≤x≤
⎪
⎪
2k+1 2k
⎪
⎪
⎪
−
⎪
⎪
⎪
⎩ − 2 < τ (x) < 1, if 1.2 ε 1.2
<x< k
2k 2
with small enough ε ≪ 0.1 and k = 0, 1, 2, . . .. Note that any point x ∈ (0, 1) belongs to one of the four
kinds of intervals for some k exactly. It is easy to check that
v(x) − x
lim 1 2 = lim τ (x) = 1,
2x
x→0 x→0

and
w(y) − y
lim 1 2 = lim τ (3y) = 1.
2y
y→0 y→0
38 G. Hong, X. Feng / Nonlinear Analysis 171 (2018) 32–40

Thus, for any ϕ ∈ C 2 (−1, 1) satisfying

ϕ(0) = v(0) and ϕ(x) ≥ v(x),

we have ϕ′ (0) = v ′ (0) = 1 and

△∞ ϕ(0) = ϕ′ (0)ϕ′′ (0)ϕ′ (0) ≥ 1.

Similarly, for any ψ ∈ C 2 (−1, 1) satisfying

ψ(0) = w(0) and ψ(y) ≥ w(y),

we have ψ ′ (0) = w′ (0) = 1 and

△∞ ψ(0) = ψ ′ (0)ψ ′′ (0)ψ ′ (0) ≥ 1.

Hence v(x) and w(y) satisfy

△∞ v(0) ≥ 1 and △∞ w(0) ≥ 1

in the viscosity sense respectively.

Let

u(x, y) := v(x) + w(y).

In the following we will show that u does not satisfy

△∞ u(0, 0) ≥ 1 + 1 = 2

in the viscosity sense.

Define
{ }
y 9 10
C= (x, y) : ∈ ( , ) .
x 10 9
We claim that for all (x, y) ∈ C,

either τ (x) = −2 or τ (3y) = −2.

Fix (x, y) ∈ C. In fact, if τ (x) ̸= −2, then there exists l ∈ Z such that
( )
1.2 − ε 1.2 + ε
x∈ , .
2l 2l
It follows that
( ) ( )
27 1.2 − ε 10 1.2 + ε 1.2 + ε 1.2 − ε
3y ∈ · , · ⊂ , l−2 ,
10 2l 3 2l 2l−1 2
i.e.

τ (3y) = −2.

Therefore,
u(x, y) − Du(0, 0) · (x, y) x2 τ (x) + y 2 τ (3y) 1
lim 2 = lim <− .
1 x2 + y 2 3
2 |(x, y)|
(x,y)∈C (x,y)∈C
(x,y)→(0,0) (x,y)→(0,0)
 G. Hong, X. Feng / Nonlinear Analysis 171 (2018) 32–40 39

Let
1
Φ(x, y) = x + y − (x + y)2 + M (x − y)2
12
with M := M (ε) large enough. Using the same argument as (3), we have

Φ(x, y) ≥ u(x, y).

Note that Φ(0, 0) = u(0, 0) and DΦ(0, 0) = Du(0, 0) = (1, 1). Then we have

△∞ Φ(0, 0) = DΦ(0, 0) · D2 Φ(0, 0)DΦ(0, 0)

1 1
⎛ ⎞
− + 2M − − 2M
= (1, 1) ⎝
⎜ 6 6 ⎠ (1, 1)
⎟ T
1 1
− − 2M − + 2M
6 6
2
= − < 2.
3
Therefore, u does not satisfy △∞ u(0, 0) ≥ 2 in the viscosity sense.

Remark 2. The following question is still a challenge for us. If we have two viscosity sub-solutions in the
whole domain rather than at one point and we assume the two right hand side functions are continuous,
that is, v ∈ C(Ω ′ ) and w ∈ C(Ω ′′ ) satisfy △∞ v(x′ ) ≥ f (x′ ) in Ω ′ and △∞ w(x′′ ) ≥ g(x′′ ) in Ω ′′ separately
(both are in the viscosity sense) with f and g continuous, is it true that △∞ (v + w) ≥ f + g in Ω ′ × Ω ′′ in
the viscosity sense? Even if f = g = 0, we do not know the answer. That is, we do not know whether or not
the superposition of two infinity harmonic functions in disjoint variables is an infinity harmonic function.
We think it is a very interesting question.

Acknowledgment

This work is supported by the National Nature Science Foundation of China: NSFC 11301411 and
11671316.

References

[1] G. Aronsson, Extension of functions satisfying Lipschitz conditions, Ark. Mat. 6 (1967) 551–561.
[2] M.G. Crandall, L.C. Evans, R.F. Gariepy, Optimal Lipschitz extensions and the infinity Laplacian, Calc. Var. Partial
Differential Equations 13 (2) (2001) 123–139.
[3] L.C. Evans, O. Savin, C 1,α regularity for infinity harmonic functions in two dimensions, Calc. Var. Partial Differential
Equations 32 (3) (2008) 325–347.
[4] L.C. Evans, C.K. Smart, Everywhere differentiability of infinity harmonic functions, Calc. Var. Partial Differential
Equations 42 (1–2) (2011) 289–299.
[5] X. Feng, G. Hong, Slope Estimate and boundary differentiability for inhomogeneous infinity Laplace equation on convex
domains, preprint.
[6] G. Hong, Boundary differentiability of infinity harmonic functions, Nonlinear Anal.: TMA 93 (2013) 15–20.
[7] G. Hong, Boundary differentiability for inhomogeneous infinity Laplace equations, Electron. J. Differential Equations 72
(2014) 1–6.
[8] G. Hong, Counterexample to C 1 boundary regularity of infinity harmonic functions, Nonlinear Anal.: TMA 104 (2014)
120–123.
[9] G. Hong, D. Liu, Slope estimate and boundary differentiability of infinity harmonic functions on convex domains, Nonlinear
Anal.: TMA 139 (2016) 158–168.
[10] R. Jensen, Uniqueness of Lipschitz extensions minimizing the sup-norm of the gradient, Arch. Ration. Mech. Anal. 123
(1993) 51–74.
40 G. Hong, X. Feng / Nonlinear Analysis 171 (2018) 32–40

[11] E. Lindgren, On the regularity of solutions of the inhomogeneous infinity Laplace equation, Proc. Amer. Math. Soc. 142 (1)
(2014) 277–288.
[12] P. Lindqvist, Notes on the Infinity Laplace Equation, in: SpringerBriefs in Mathematics, BCAM Basque Center for Applied
Mathematics, Springer, Bilbao, Cham, 2016.
[13] G. Lu, P. Wang, Inhomogeneous infinity Laplace equation, Adv. Math. 217 (4) (2008) 1838–1868.
[14] O. Savin, C 1 regularity for infinity harmonic functions in two dimensions, Arch. Ration. Mech. Anal. 176 (3) (2005)
351–361.
[15] C.Y. Wang, Y.F. Yu, C 1 -boundary regularity of planar infinity harmonic functions, Math. Res. Lett. 19 (4) (2012) 823–835.