PROCEEDINGS OF THE

AMERICAN MATHEMATICAL SOCIETY

Volume 136, Number 8, August 2008, Pages 2823–2828
S 0002-9939(08)09561-0
Article electronically published on April 15, 2008

A NEW MAXIMUM PRINCIPLE

OF ELLIPTIC DIFFERENTIAL EQUATIONS
IN DIVERGENCE FORM

DONGSHENG LI AND LIHE WANG

(Communicated by David S. Tartakoff)

Abstract. In this paper will be presented a new maximum principle of el-

liptic differential equations in divergence form which can be regarded as the
counterpart of the Alexandroff-Bakelman-Pucci maximum principle of elliptic
differential equations in nondivergence form.

1. Introduction
The Alexandroff-Bakelman-Pucci maximum principle is an important tool for
studying elliptic differential equations in nondivergence form. In this paper, we
shall find the counterpart of it for elliptic differential equations in divergence form.
Throughout this paper, we assume that Ω ⊂ Rn is a bounded domain, f (x) ∈
L (Ω) and aij (x) ∈ L∞ (Ω), i, j = 1, 2, . . . , n such that for x ∈ Ω a.e., (aij )n×n is
p

symmetric and
(1.1) λI ≤ (aij )n×n ≤ ΛI
in the sense of nonnegative definiteness, where λ ≤ Λ are two positive constants.
We denote by Br the open ball
in Rn centered at the origin with radius r. For
convenience, we extend (aij (x) n×n to be the identity outside Ω and still denote
the extended matrix by (aij (x) n×n . Since Ω is bounded, we assume that

Ω ⊂ Bd
for d > 0. Let χE denote the character function of the set E, and define f + =
max{0, f } and f − = max{0, −f } for any function f .
To compare with our new maximum principle, we first state our Alexandroff-
Bakelman-Pucci maximum principle as the following (cf. [1], [2] for the proof).
Theorem 1.1 (Alexandroff-Bakelman-Pucci maximum principle). Suppose f ∈
Lp (Ω) and u ∈ W 2,p (Ω) is a subsolution of the following:
(1.2) −aij (x)∂ij u(x) = f (x)

Received by the editors August 1, 2005, and, in revised form, January 20, 2007.
2000 Mathematics Subject Classification. Primary 35J25.
Key words and phrases. Maximum principle, elliptic equation in divergence form.
The first author was supported by the NSF of China: 10771166.
The second author was supported by PCSIRT.
c

2008 American Mathematical Society
Reverts to public domain 28 years from publication

2823
2824 DONGSHENG LI AND LIHE WANG

in Ω. If p ≥ n, then there exists a constant C depending only on λ, Λ, p and n such

that
n
  p1
(1.3) sup u ≤ sup u+ + C · d2− p · (f + )p ,
Ω ∂Ω {ψ=Γψ }∩Ω
+
where ψ = u − sup∂Ω u and Γψ is the concave envelope of ψ in B2d (cf. [1] for its
definition).
For the following elliptic differential equation in divergence form,
 
(1.4) −∂i aij (x)∂j u(x) = f (x) in Ω,
the estimate of (1.3) cannot be obtained by exploiting the concave envelope Γψ as
Theorem 1.1 since the concave envelope is appropriate for (1.2) while not for (1.4).
However, if we replace the concave envelope by another proper function, a similar
estimate of (1.3) can still be obtained. In fact, the proper function is the solution
of the following obstacle problem:
⎧
⎨ Find u ∈ Kψ such that
⎪
(1.5)
⎪
J(u) = inf w∈Kψ J(w),
⎩

where 
J(w) = aij ∂i w∂j w
B2d
for any w ∈ W01,2 (B2d ) and
Kψ = {w ∈ W01,2 (B2d ) : w ≥ ψ a.e. in Ω}
for any fixed ψ ∈ W 1,2 (Ω) with ψ ≤ 0 on ∂Ω. For Problem (1.5), we have the
following well-known results (cf. [3], [4]):
Lemma 1.2. (i) There exists a unique Υψ ∈ Kψ such that
J(Υψ ) = inf J(w).
w∈Kψ

(ii) There exists a unique Υ̃ψ ∈ Kψ such that


aij ∂i w∂j (w − Υ̃ψ ) ≥ 0, ∀w ∈ Kψ .
B2d

(iii) There exists a unique Υ̂ψ ∈ Kψ such that


aij ∂i Υ̂ψ ∂j (w − Υ̂ψ ) ≥ 0, ∀w ∈ Kψ .
B2d

(iv) Υψ = Υ̃ψ = Υ̂ψ .

Now we can present our maximum principle for (1.4) as the following:
Theorem 1.3 (Maximum Principle). Suppose f ∈ Lp (Ω) with p > n2 if n ≥ 4 or
p = 2 if n = 1, 2, 3, and u ∈ W 1,2 (Ω) is a subsolution of (1.4). Then there is a
constant C depending only on λ, n and p such that
n
  p1
(1.6) sup u ≤ sup u+ + C · d2− p · (f + )p ,
Ω ∂Ω {ψ=Υψ }∩Ω
+
where ψ = u − sup∂Ω u and Υψ is given by Lemma 1.2.
 MAXIMUM PRINCIPLE OF ELLIPTIC DIFFERENTIAL EQUATIONS 2825

In section 2, we study the regularity of the solution of the obstacle problem (1.5).
In section 3, we shall prove our main result, Theorem 1.3.

2. Regularity of solutions of obstacle problems

Our regularity result for the obstacle problem (1.5) is the following:
Theorem 2.1. Suppose f ∈ L2 (Ω) and ψ ∈ W 1,2 (Ω) with ψ ≤ 0 on ∂Ω such that
 
(2.1) −∂i aij ∂j ψ ≤ f

in the weak sense in Ω. Then we have

 
(2.2) 0 ≤ −∂i aij (x)∂j Υψ (x) ≤ f + (x) · χE

in the weak sense in B2d , where Υψ is given by Lemma 1.2 and E = {x ∈ Ω :

ψ(x) = Υψ (x)}.
 
If aij , f and ψ are smooth and −∂i aij ∂j ψ = f in Ω, Lewy and Stampacchia
[3] showed that
 
−∂i aij (x)∂j Υψ (x) = f + (x) · χE
for x ∈ B2d a.e. The idea used in [3] with some appropriate changes is also applica-
ble for the proof of Theorem 2.1. The first part of (2.2) is well known (cf. [3], [4]),
and then we only need to show the second part of it, which needs the following two
lemmas.
Lemma 2.2. Suppose w ∈ W 1,2 (Ω) such that w ≥ 0 on ∂Ω and
 
(2.3) −∂i aij (x)∂j w(x) ≥ f (x)

in the weak sense in Ω, where f ∈ L2 (Ω). If w(x) ≤ 0 implies f (x) ≥ 0 for x ∈ Ω

a.e., then we have w ≥ 0 a.e. in Ω.
Proof. Since w(x) ≥ 0 on ∂Ω, we have w− ∈ W01,2 (Ω). From (2.3) and the assump-
tion that w(x) ≤ 0 implies f (x) ≥ 0 for x ∈ Ω a.e., it follows that
  
aij ∂i w∂j w− ≥ f w− = f w− ≥ 0,
Ω Ω E

where E = {x ∈ Ω : w(x) ≤ 0}. Thereby,

  
λ |Dw− |2 ≤ aij ∂i w− ∂j w− = aij ∂i (−w)∂j w− ≤ 0.
Ω Ω Ω
−
Hence w = 0 a.e. in Ω.

Lemma 2.3. Suppose f ∈ L2 (Ω) and θ is a Lipschitz function defined in (−∞, +∞)
such that 0 ≤ θ ≤ 1. Then the nonlinear equation
 
(2.4) −∂i aij (x)∂j u(x) = f + (x) · θ(u − ψ) · χΩ

has a solution in W01,2 (B2d ) for any ψ ∈ L2 (Ω).

2826 DONGSHENG LI AND LIHE WANG

Proof. We use the Schauder fixed point theory. For any w ∈ L2 (B2d ), the equation
 
(2.5) −∂i aij (x)∂j u(x) = f + (x) · θ(w − ψ) · χΩ

has a unique weak solution u ∈ W01,2 (B2d ), and we denote u = T w. Since 0 ≤ θ ≤ 1,

there exists a constant C depending only on λ and d such that
||T w||W 1,2 (B2d ) ≤ C||f + ||L2 (Ω) , ∀w ∈ L2 (B2d ).
Thereby the compactness of T from L2 (B2d ) to L2 (B2d ) follows clearly. If we set
D = {w ∈ L2 (B2d ) : ||w||L2 (B2d ) ≤ C||f + ||L2 (Ω) }, then T (D) ⊂ T (L2 (B2d )) ⊂ D.
Since θ is Lipschitz continuous, T is continuous from L2 (B2d ) to W01,2 (B2d ) and
then to L2 (B2d ). Therefore, according to the Schauder fixed point theory, there
exists u ∈ D such that u = T u ∈ W01,2 (B2d ).

Proof of Theorem 2.1. Consider a sequence of functions {θm }∞
m=1 defined by
1
⎧
⎪
⎪ 1 for t ≤ 2m ,
⎪
⎪
⎪
⎨
1 1
θm = 2 − 2mt for 2m ≤t≤ m ,
⎪
⎪
⎪
⎪
1
⎪
0 for t ≥ m ,
⎩

and let um ∈ W01,2 (B2d ) be a solution of (2.4) with θ = θm .

We first claim that um ≥ ψ a.e. in Ω for any m. Indeed, the maximum principle
implies um ≥ 0 a.e. in B2d . In view of ψ ≤ 0 on ∂Ω, we deduce um − ψ ≥ 0 on ∂Ω.
By (2.1) and (2.4) with θ and u replaced by θm and um , respectively,
 
−∂i aij ∂j (um − ψ) ≥ f + (x) · θm (um − ψ) − f
   
in the weak sense in Ω. Since f + · θm (um − ψ) − f (x) ≥ 0 if um − ψ (x) ≤ 0
for x ∈ Ω a.e., by Lemma 2.2, our claim is proved.
It is clear that {um } is bounded in W01,2 (B2d ). Without loss of generality, we
assume that
(2.6) um → ũ weakly in W01,2 (B2d ), strongly in L2 (B2d ) and a.e. in B2d .
We let Ẽ = {x ∈ Ω : ũ(x) = ψ(x)} and claim that
(2.7) lim sup θm (um − ψ) ≤ χẼ a.e. in Ω.
m→∞

In fact, by (2.6) and um ≥ ψ a.e. in Ω, there exists K ⊂ Ω with |K| = 0 such

that um (x) → ũ(x) and um (x) ≥ ψ(x) for all x ∈ Ω\K. If x ∈ Ẽ, then (2.7)
holds clearly since θm ≤ 1. Suppose x ∈ Ω\{Ẽ ∪ K} and then ũ(x) > ψ(x). Since
1
um (x) → ũ(x), there exists m0 such that if m ≥ m0 , um (x) − ψ(x) > m . Thereby
θm (um (x) − ψ(x)) = 0 for m ≥ m0 . Hence, (2.7) holds.
Since for any φ ∈ C01 (B2d ) with φ ≥ 0,
 
aij ∂i um ∂j φ = f + · θm (um − ψ) · χΩ · φ,
B2d B2d

by taking the limit in both sides and in view of (2.6), (2.7), we deduce
 
(2.8) aij ∂i ũ∂j φ ≤ f + · χẼ · φ.
B2d B2d
 MAXIMUM PRINCIPLE OF ELLIPTIC DIFFERENTIAL EQUATIONS 2827

If we can show that ũ = Υψ , then the proof of Theorem 2.1 will be complete.
Indeed, for each m, since θm is decreasing, we have for any v ∈ W01,2 (B2d ),

aij ∂i (v − um )∂j (v − um ) − f + · θm (v − ψ) − θm (um − ψ) · χΩ · (v − um ) ≥ 0.


B2d

Combining with (2.4), where θ and u are replaced by θm and um respectively, we

obtain

(2.9) aij ∂i v∂j (v − um ) − f + · θm (v − ψ) · χΩ · (v − um ) ≥ 0.
B2d
1
For any fixed positive integer m0 , in (2.9), take v − ψ > m0 a.e. in Ω and then

aij ∂i v∂j (v − um ) ≥ 0
B2d

for m > m0 . By (2.6),


(2.10) aij ∂i v∂j (v − ũ) ≥ 0.
B2d

Since m0 is arbitrary, we have that (2.10) holds for any v ∈ Kψ . Then Lemma 1.2
implies ũ = Υψ .

3. Proof of Theorem 1.3

To prove Theorem 1.3, we also need the following theorem, which can be proved
by Moser iteration (cf. [2]) or by the rearrangement technique (cf. [5]).
Theorem 3.1. Suppose f ∈ Lp (Ω) with p > n2 and u ∈ W 1,2 (Ω) is a subsolution
of (1.4). Then there is a constant C depending only on λ, n and p such that
n
sup u ≤ sup u+ + C · r 2− p · ||f ||Lp (Ω) ,
Ω ∂Ω

where r satisfies |Ω| = ωn r and ωn denotes the volume of the unit ball in Rn .
n

Remark 3.2. It is clear that the estimate of u by Theorem 1.3 is more accurate
than by Theorem 3.1.
Proof of Theorem 1.3. Set ψ = u − sup∂Ω u+ , and then (2.1) holds. Let Υψ be
given by Lemma 1.2. From Theorem 2.1, it follows that
 
−∂i aij ∂j Υψ ≤ f + · χ{ψ=Υψ }∩Ω

in the weak sense in B2d . In view of Theorem 3.1, there exists a constant C
depending only on λ, n and p such that
n
sup Υψ ≤ C · d2− p · ||f + · χ{ψ=Υψ }∩Ω ||Lp (Ω) .
Ω

Combining with
sup u − sup u+ ≤ sup ψ ≤ sup Υψ ,
Ω ∂Ω Ω Ω

we have that (1.6) holds.

2828 DONGSHENG LI AND LIHE WANG

References
[1] Caffarelli, L. A., and Cabre, X., Fully nonlinear elliptic equations, Colloquium Publications,
43, Amer. Math. Soc., Providence, RI, 1995. MR1351007 (96h:35046)
[2] Gilbarg, D., and Trudinger, N. S., Elliptic partial differential equations of second order, 2nd
ed., Springer-Verlag, 1983. MR737190 (86c:35035)
[3] Lewy, H., and Stampacchia, G., On the regularity of the solution of a variational inequality,
Comm. on Pure and Appl. Math., 1969, XXII, 153-188. MR0247551 (40:816)
[4] Lions, J.-L., and Stampacchia, G., Variational inequalities, Comm. on Pure and Appl. Math.,
1967, XX, 493-519. MR0216344 (35:7178)
[5] Talenti, G., Elliptic equations and rearrangements, Annali della Scuola Norm. Sup. di Pisa,
1976, 3, 697-718. MR0601601 (58:29170)

College of Science, Xi’an Jiaotong University, Xi’an 710049, China

E-mail address: [email protected]

Department of Mathematics, The University of Iowa, Iowa City, Iowa 52242-1419

E-mail address: [email protected]