c Birkhäuser Verlag, Basel, 2007

NoDEA
Nonlinear differ. equ. appl. 14 (2007) 259—289
1021–9722/07/040259–31
DOI 10.1007/s00030-007-5029-9

Existence, uniqueness and regularity for nonlinear

parabolic equations with nonlocal terms
Nathaël ALIBAUD
Université Montpellier II
Département de mathématiques
CC 051, Place E. Bataillon
34 095 Montpellier cedex 5, France
e-mail: [email protected]

Abstract. We study evolution equations that are fully nonlinear, degener-

ate parabolic, nonlocal and nonmonotone. The major difficulty lies in non-
monotonicity, i.e. in the fact that no comparison principle can be obtained.
This implies that the classical method used to prove existence in the con-
text of fully nonlinear degenerate equations, namely Perron’s one, does not
apply. We thus need to use a fixed point argument and to get suitable a pri-
ori estimates, we need a refined version of classical continuous dependance
estimates. This technical result is of independent interest. We also obtain
results such as uniform or Hölder continuity of the solution.

2000 Mathematics Subject Classification: 35A05, 35B30, 35B65, 35G20,

35K65, 47G20
Key words: nonlinear degenerate parabolic integro-partial differential equa-
tion; nonmonotone equation; viscosity solution; continuous dependance esti-
mate; Hölder estimate.

1 Introduction
In recent years there has been an interest in developing viscosity solutions theory
for parabolic integro-PDEs. Particularly equations that occur in the theory of
optimal control of jump-diffusion (Lévy) processes [7, 14, 16, 2, 11, 13]. The use
of viscosity solutions is appropriately chosen because most of these equations are
degenerate or fully nonlinear. In general, existence is proved by Perron’s method
with the help of a comparison principle. But in few other many applications, like
signal [4], there is no such a comparison principle. The equations are then said to
260 Nathaël Alibaud NoDEA

be nonmonotone. The notion of viscosity solution can still be used but existence
must be proved by classical fixed point methods. Our purpose is to find a general
framework to treat these difficulties that are degeneracy, nonlinearity, presence
of nonlocal terms and nonmonotonicity. For other recent works on nonmonotone
equations, we refer the reader to [1, 3, 12] which study equations that are involved
in the theory of dislocation.
Let us present our mathematical framework. We are interested in exis-
tence, uniqueness and regularity of viscosity solution of fully nonlinear degenerate
parabolic integro-PDEs of the form

∂t u + F (t, x, u, Du, D2 u, g[u]) = 0 in QT , (1.1)

u(0, .) = u0 on RN , (1.2)

where QT :=]0, T [×RN , F : [0, T ] × RN × R × RN × SN × RM → R is a given

functional nonincreasing with respect to (w.r.t. for short) the D2 u-variable and
g[u] is a nonlocal term. Here SN denotes the space of symmetric N ×N real valued
matrices. We investigate the case where no monotonicity assumption w.r.t. the
nonlocality is assumed (see Remark 3.1 for more details); this led us to consider
a class of nonlocal term of order 0 satisfying a Lipschitz condition of the form

sup |g[u] − g[v]|(τ, x) ≤ C sup |u − v|(τ, x), (1.3)

[0,t]×RN [0,t]×RN

where C does not depend on t, u and v. Examples of nonlocal terms satisfying

(1.3) are nonlinear integral operators of the form


M (t, x, z, u(t, x), u(t, x + z))dµt,x (z), (1.4)
RN

where supQT |µt,x |(RN ) < +∞ and M is Lipschitz w.r.t. its two last arguments.
Further examples are given by integral operators in both time and space variables
such as

t
g[u](t) = S(t)v0 + S(t − τ )f (u(τ ))dτ, (1.5)
0

where S(.) is a semigroup generated by a linear operator A, f : R → R is Lips-

chitz and v0 is a given initial condition. Then, the Cauchy problem (1.1)-(1.2) is
equivalent to the following system:

∂t u + F (t, x, u, Du, D2 u, v) = 0 in QT ,
u(0, .) = u0 on RN ,
d
v + Av = f (u) in ]0, T [,
dt
v(0) = v0 .
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 261

When A is the Laplacian operator, for example, the associated semigroup S(.) is
defined by
S(t)v := G(t) ∗ v,
where ∗ designs the convolution product in RN and G(t)(x) is the Green Kernel.
As far as the Hamiltonian F is concerned, we first study the case where it
is Lipschitz w.r.t. the g[u]-variable and next the case where there is a coupling
between g[u] and the derivatives of u. The last one case can be seen as a general-
ization of [4] which treats the nonmonotone equation
 
∂t u − f (DG ∗ u(t, .))tr A(Du)D2 u = 0 in QT ,

for f > 0 Lipschitz-continuous, G which is a Gaussian function and A ≥ 0 bounded

continuous on RN − {0}. In fact, our technics combined with these ones used
to treat the mean curvature flow by the level set method could allow to treat
discontinuous Hamiltonians at Du = 0 and such that

F ∗ (t, x, r, 0, 0, λ) = F∗ (t, x, r, 0, 0, λ).

But, for the sake of clarity we have chosen to present only the continuous case. An
illustrating example of what kind of coupling can be considered is the following
quasilinear Hamiltonian
 
F (t, x, r, p, X, λ) = H(t, x, r, p, λ) − tr t σ(t, x, p, λ) σ(t, x, p, λ)X , (1.6)

where H and σ are Lipschitz w.r.t. (x, λ) respectively locally and globally in p.
Another interest of this paper are so-called continuous dependence estimates
(see Theorem 4.1) for local parabolic equations which allow to obtain lots of needed
a priori estimates. This can be seen as a generalization of results of Souganidis
[15, Proposition 1.4] for first-order equations and of Jakobsen and Karlsen [10,
Theorem 3.1] for second-order equations. Let us recall that some of their appli-
cations are a priori Lipschitz and Hölder estimates. In our setting, their results
are not sufficient, in particular they do not permit to prove that the function is
uniformly continuous. This improved version seems to us of independent interest.
The rest of this paper is organized as follows: in Section 2, we introduce
the definitions and notations that will be used throughout this paper. In Section
3, we state our results and prove them in Section 4. The last one section also
contains our continuous dependance estimate for local equations.

2 Preliminaries
Throughout the paper, we will use the notations that follow. For a, b ∈ R, we let
a ∨ b denote the real max{a, b}. We let a+ denote the real a ∨ 0. Let k be an
integer. For x ∈ Rk , we let |x| denote the Euclidean norm of x. We let QT denote
the cylinder ]0, T [×RN . We let MN denote the space of N ×N real valued matrices
262 Nathaël Alibaud NoDEA

and SN the space of such matrices which are symmetric. For every X, Y ∈ SN ,
we say that X ≤ Y when Xξ, ξ ≤ Y ξ, ξ for all ξ ∈ RN . The notation ., . is
the Euclidean scalar product of RN . Let (E, dE ) be a metric space. The closed
ball of E centered at x and of radius R is denoted by B E (x, R).
Let us now introduce some functional vector spaces. Consider µ ∈]0, 1] and
u : QT → Rk . Define
||u||∞ := sup |u(t, x)|,
t∈[0,T ],x∈RN

|u(t, x) − u(t, y)|

[u]µ := sup ,
t∈[0,T ],x,y∈RN ,x=y |x − y|µ
||u||µ := ||u||∞ + [u]µ .

We let Cb (QT , Rk ) and Cb0,µ (QT , Rk ) denote the spaces of continuous functions
u : QT → Rk such that ||u||∞ < +∞ and ||u||µ < +∞, respectively. We let
BU C(QT , Rk ) denote the space of bounded uniformly continuous functions u.
When k = 1, we let Cb (QT ), Cb0,µ (QT ) and BU C(QT ) denote the preceding
spaces. Consider a function h : O ⊆ Rk → R and a nonnegative real α. We let
ωα (h) denote the modulus of continuity of size α of h. That is to say,
ωα (h) = sup |h(x) − h(y)|.
x,y∈O, |x−y|≤α

We call modulus a function m : R+ → R+ such that m is continuous, nondecreas-

ing, m(0) = 0 and such that m(α1 + α2 ) ≤ m(α1 ) + m(α2 ) for all nonnegative
reals α1 and α2 .
Following [9], we now recall the notion of viscosity solution. This last one
notion can be defined for discontinuous locally bounded functions and Hamilto-
nians. Here, we only need the continuous case. Consider the following general
equation:
∂t u + G(t, x, u, Du, D2 u) = 0 in QT , (2.1)

where the Hamiltonian G : [0, T ] × RN × R × RN × SN → R is continuous and

nonincreasing w.r.t. its last argument. For each bounded continuous function u :
2,+(−)
QT → R and each subset O of QT , we let PO u(t, x) denote the second-order
parabolic superjet (subjet) of u at (t, x) ∈ O relatively to O. Let us recall that
2,+(−)
(a, p, X) ∈ PO u(t, x) if and only if (iff for short)
u(s, y) ≤ (≥)u(t, x) + as + p, y − x
1
+ X(y − x), y − x + o(|s − t| + |y − x|2 )
2
as O  (s, y) → (t, x). We let simply P 2,+(−) u(t, x) denote the semijet
2,+(−)
PQT u(t, x).
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 263

Definition 2.1 Let u belong to C(QT ).

i) The function u is a viscosity subsolution of (2.1) iff for every (t, x) ∈ QT and
(a, p, X) ∈ P 2,+ u(t, x),

a + G(t, x, u, p, X) ≤ 0.

ii) The function u viscosity supersolution of (2.1) iff for every (t, x) ∈ QT and
(a, p, X) ∈ P 2,− u(t, x),

a + G(t, x, u, p, X) ≥ 0.

iii) The function u viscosity solution of (2.1) iff it is both a viscosity sub- and
supersolution of (2.1).

2,+(−)
Remark 2.1 Define the closure P O u(t, x) as the set of (a, p, X) ∈ R × RN ×
2,+(−)
SN such that, there are (tn , xn ) ∈ O and (an , pn , Xn ) ∈ PO u(tn , xn ) such that
u(tn , xn ) → u(t, x) and (tn , xn , an , pn , Xn ) → (t, x, a, p, X). In fact, penalization
technics used in [9] allow to prove that the definitions above are still true by
2,(+)−
replacing QT by O :=]0, T ] × RN and P 2,(+)− u(t, x) by P O u(t, x).
In a similar way, we can define the notion of continuous viscosity semisolution
of the following differential equation:

f˙ + G(f ) = 0 in ]0, T [,

where f : [0, T ] → R is the unknown function, and G ∈ C([0, T ]) is given. Then,

the notion of parabolic semijet is replaced by the notion of first-order Fréchet
semidifferential. Consider t ∈]0, T [. Let us recall that a ∈ ∂ 1,+(−) f (t) iff

f (s) ≤ (≥)f (t) + as + o(|s − t|),

as s → t.
Let us now define the notion of viscosity solution of (1.1) which is used in our
paper. Consider a continuous Hamiltonian F and a nonlocal term g[.] : Cb (QT ) →
C(QT , RM ). Let F (u) : [0, T ] × RN × R × RN × SN → R denote the functional
defined by
F (u) (t, x, r, p, X) = F (t, x, r, p, X, g[u]).
Note that F (u) is continuous when u is bounded continuous. Consider the following
equation in w:

∂t w + F (u) (t, x, w, Dw, D2 w) = 0 in QT . (2.2)

Definition 2.2 A function u ∈ Cb (QT ) is a viscosity solution of (1.1) iff it is a

viscosity solution of (2.2).
264 Nathaël Alibaud NoDEA

3 Main results
Let us state our main results. We first consider nonlocal term uncoupled with the
derivatives of u and next we study the case where there is a coupling. All the
constants appearing in this section are noted C F (resp. C g ) when depending on
F
the Hamiltonian F (resp. the nonlocal term g[.]) and CR when also depending on
a real number R.

3.1 Uncoupled nonlocal term

Let us consider a class of nonlocal terms g[.] that satisfy the following conditions:
(H1) The operator g[.] is well-defined from Cb (QT ) into Cb (QT , RM ) (in partic-
ular, ||g[0]||∞ ≤ C g < +∞).
(H2) If u is uniformly continuous w.r.t. x independently of t, then so is g[u].
(H3) There exists a constant C g ≥ 0 such that for every u, v ∈ Cb (QT ) and
t ∈ [0, T ],

sup ||g[u](τ, .) − g[v](τ, .)||∞ ≤ C g sup ||u(τ, .) − v(τ, .)||∞ .

τ ∈[0,t] τ ∈[0,t]

As far as F is concerned, let us assume the following conditions:

(H4) The Hamiltonian F is continuous and for each R ≥ 0, F is uniformly
continuous w.r.t. the (Du, D2 u)-variable, independently of the others, on
[0, T ] × RN × [−R, R] × B RN (0, R) × B SN (0, R) × B RM (0, R).
(H5) The Hamiltonian F is nondecreasing w.r.t. the u-variable.
There is a modulus mR (.), depending on a real number R, such that for every
ε > 0, t ∈ [0, T ], x, y ∈ RN , r ∈ [−R, R], X, Y ∈ SN and λ ∈ B RM (0, R),
(H6) if
   
3 X 0 3 I −I
− ≤ ≤ (3.1)
ε 0 −Y ε −I I

then
     
x−y x−y |x − y|2
F t, y, r, , Y, λ −F t, x, r, , X, λ ≤ mR |x − y| + .
ε ε ε

(H7) The Hamiltonian F is Lipschitz w.r.t. the g[u]-variable, independently of

the others, with a Lipschitz constant that is denoted by C F .
(H8) supQT |F (t, x, 0, 0, 0, 0)| ≤ C F < +∞.
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 265

Remark 3.1 Assumptions (H4)–(H6) and (H8) are classical when studying local
equations (see [9]). Monotonicity assumptions w.r.t. the nonlocality (see [14, 16,
2, 11] and remarks in Example 3.1 below) are replaced here by (H3) and (H7).
Assumption (H2) is necessary to solve (1.1)–(1.2) after having frozen the nonlocal
part. If we strengthen (H4), then we can actually omit (H2) to solve (1.1)–(1.2) in
the space of bounded continuous functions (to see this, one could combine technics
used in [2] with our technics).

Then we have the following result.

Theorem 3.1 (Existence, uniqueness and BU C-regularity) Assume (H1)–(H8).
Let u0 belong to BU C(RN ). Then there exists a unique u ∈ BU C(QT ) such that
u is a viscosity solution of (1.1)–(1.2). Moreover, for each t ∈ [0, T ]

||u(t, .)||∞ ≤ (||u0 ||∞ + Ct) eγ0 t , (3.2)

where C = C F + C F C g and γ0 = C F C g .

Remark 3.2 This theorem is also a regularity result, since we get the uniform
continuity of the solution. Moreover, the assumptions (in particular (H2) and
(H6)) seem to us quite general to get this.

Example 3.1 Suitable assumptions under which nonlocal terms of the form (1.4)
or (1.5) satisfy (H1)–(H3) can easily be determinate by the readers. Some simple
illustrating examples of such nonlocal terms are:

• convolution operators K ∗ u(t, .), where K ∈ L1 (RN );


• Lévy operators of the form − RN u(t, x + z) − u(t, x)dµ(z), where µ is a
bounded Borel measure (under suitable assumptions to ensure (H2), µ can
also depend on (t, x));
t
• Volterra operators of the form − 0
B(t − s)u(s, x)ds, where B ∈ L1 (0, T ).

Note that if the measure and the kernels above are nonnegative and if F is
nondecreasing w.r.t. the g[u]-variable then (1.1) is monotone (see [16, 2, 11]); but,
in the general case (1.1) can become nonmonotone.

We are now interested in Hölder regularity of u w.r.t. x. For µ ∈]0, 1], let
us consider the following condition:

(H2)’ There exists a constant C g ≥ 0 such that for every u ∈ Cb0,µ (QT ) and
t ∈ [0, T ],
sup [g[u](τ, .)]µ ≤ C g (1 + sup ||u(τ, .)||µ ).
τ ∈[0,t] τ ∈[0,t]
266 Nathaël Alibaud NoDEA

We let also
  
|x−y|2
(H6)’ denote Assumption (H6) for mR |x − y| + F
= CR |x − y|µ
2
 ε

+ |x−y|
ε ,

F
where CR is a nonnegative constant that depends on R. Let us state our regularity
result.
Theorem 3.2 (Hölder regularity) Assume that (H1), (H2)’, (H3)–(H5), (H6)’,
(H7) and (H8) hold true for any µ ∈]0, 1]. Let u be the unique viscosity solution
of (1.1)–(1.2). Define R = ||u||∞ and Rg = C g (1 + R). Then for each t ∈ [0, T ],
 µ

[u(t, .)]µ ≤ 4 [u0 ]µ e 2 γt + CR eγ1 (t)t , (3.3)

F
 F g µ γt  2−µ
2  −1  F
where γ = 2(CR∨R g +1), γ1 (t) = 4C C e 2 and CR = C F C g CR∨Rg +
F g
C R .

3.2 Nonlocal term coupled with the derivatives

Now, we strengthen (H2)’ by:

(H2)” There exists a constant C g ≥ 0 such that for every u ∈ Cb (QT ) and
t ∈ [0, T ],
sup [g[u](τ, .)]1 ≤ C g (1 + sup ||u(τ, .)||∞ ).
τ ∈[0,t] τ ∈[0,t]

Assumptions (H6)’ and (H7) on F are relaxed by:

(H9) There exists a constant C F ≥ 0, and for each R ≥ 0 there exists a constant
CRF
≥ 0 such that, for every ε > 0, t ∈ [0, T ], x, y ∈ RN , |r| ≤ R, X, Y ∈ SN
and |λ|, |µ| ≤ R, if (3.1) holds true then
   
x−y x−y
F t, y, r, , Y, λ − F t, x, r, , X, µ ≤
ε ε
 2
  
|x − y| |x − y| |λ − µ|2
F
CR |x − y| + + C F |λ − µ| + |λ − µ| + .
ε ε ε

Remark 3.3 Let us comment assumption (H9), Under (H6)’ (with µ = 1) and
(H7), we have
   
x−y x−y
F t, y, r, , Y, λ − F t, x, r, , X, µ
ε ε
 2

|x − y|
≤ CRF
|x − y| + + C F |λ − µ| .
ε
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 267

2
The role of the new term C F |λ − µ| |x−y|
ε + C F |λ−µ|
ε that appears in (H9) can be
illustrate by the following simple example of Hamiltonian:

F (t, x, r, p, X, λ) = λ|p| − tr (A(λ)X) ,

where A = tσσ and σ is Lipschitz.

Let us state our last result.

Theorem 3.3 (Existence, uniqueness and Lipschitz regularity) Let us assume
(H1), (H2)”, (H3)–(H5), (H8) and (H9). Then for each u0 ∈ W 1,∞ (RN ) there
exists a unique u ∈ Cb (QT ) such that u is a viscosity solution of (1.1) and (1.2).
Moreover, defining R = ||u||∞ and Rg = C g (1 + R), Estimate (3.2) still holds
true and for each t ∈ [0, T ]
γ
[u(t, .)]1 ≤ (2[u0 ]1 + CR t) e 2 t , (3.4)
 F   F 
where γ = 2 CR∨R F g
g +C C (1 + C
g
+ R + C g R2 ) +1 and CR = CR∨R F g
g +C R .

Example 3.2 Simple assumptions under which quasilinear Hamiltonians of the

form (1.6) satisfy (H9) are the following: there are nonnegative constants Ci
(i = 1, 2) and for each R ≥ 0 there exists CR ≥ 0 such that:

For every t ∈ [0, T ], x ∈ RN , r ∈ R, p ∈ RN , λ ∈ RM ,

|∂λ H(t, x, r, p, λ)| ≤ C1 (1 + |p|) and |∂λ σ(t, x, p, λ)| ≤ C2 .

If |r|, |λ| ≤ R, then |∂x H(t, x, r, p, λ)| ≤ CR (1 + |p|) and |∂x σ(t, x, p, λ)| ≤ CR .

4 Proofs of the results

In this section, we prove the preceding results. This section is organized as follows:
in Subsection 4.1, we state a technical result (Theorem 4.1). This result is proved
in Subsection 4.4. Subsection 4.2 is devoted to the proofs of Theorems 3.1 and
3.2. Theorem 3.3 is proved in Subsection 4.3.

4.1 Continuous dependence estimates for local

parabolic equations
To state our technical result, we have to introduce some notations. Let us consider
equations of the form

∂t ui + Gi (t, x, ui , Dui , D2 ui ) = 0 in QT , (4.1)

268 Nathaël Alibaud NoDEA

where Gi : [0, T ] × RN × R × RN × SN → R is a given functional (i = 1, 2). We

make the following assumptions on Gi :
The Hamiltonian Gi is continuous and for each R ≥ 0, Gi is uniformly
continuous w.r.t. the (Dui , D2 ui )-variables, independently of the others, on
[0, T ] × RN × [−R, R] × B RN (0, R) × B SN (0, R). (4.2)

There exists γ ≥ 0 such that for every t ∈ [0, T ], x ∈ RN , r, s ∈ R, p ∈ RN and

X ∈ SN , if r ≥ s then
Gi (t, x, r, p, X) ≥ Gi (t, x, s, p, X) + γ(r − s). (4.3)
For each R ≥ 0, there exists a modulus mR (.) such that for every ε > 0, t ∈ [0, T ],
x, y ∈ RN , r ∈ [−R, R] and X, Y ∈ SN , if (3.1) holds true then
     
x−y x−y |x − y|2
Gi t, y, r, , Y − Gi t, x, r, , X ≤ mR |x − y| + .
ε ε ε
(4.4)
Finally, we assume that for each R ≥ 0,
sup (G2 (t, x, r, 0, 0) − G1 (t, x, r, 0, 0)) < +∞. (4.5)
(t,x,r)∈QT ×[−R,R]

Let ui be a bounded continuous sub- and supersolution of (4.1) for respec-

tively i = 1 and i = 2, γ be a nonnegative constant (that will be appropriately
chosen when using Theorem 4.1 in Subsection 4.2 and 4.3) and ε be a positive
real. For t ∈ [0, T ], define
 
2 +
γt |x − y|
mε (t) := sup u1 (t, x) − u2 (t, y) − e ,
(x,y)∈RN ×RN 2ε
    
σε (t) := inf∗ sup G2 t, y, r, eγt p, eγt Y − G1 t, x, r, eγt p, eγt X
n∈N dn (t)
ε
+
|x − y|2
− γeγt ,
2ε
where we let dnε (t) denote the set of (x, y, r, p, X, Y ) such that,

 |r| ≤ ||u1 (t, .)||∞ ∨ ||u2 (t, .)||∞ ,

 p = x−yε ,

 Condition (3.1) holds true,
 |x−y|2
4ε ≤ m2ε (t) − mε (t) + n1 .
We also define
 +
    |x − y|2
Σε (t) := sup G2 τ, y, r, eγτ p, eγt Y −G1 τ, x, r, eγt p, eγτ X − γeγτ ,
Dε (t) 2ε
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 269

where we let Dε (t) denote the set of (τ, x, y, r, p, X, Y ) such that



 r ∈ [0, t],

|r| ≤ ||u1 (τ, .)||∞ ∨ ||u2 (τ, .)||∞ ,

 p = x−yε ,

Condition (3.1) holds true.
Then, we have the following result.

Theorem 4.1 Assume that Gi satisfies (4.2)–(4.5) (i = 1, 2). Assume that ui ∈

Cb (QT ) is a sub- and a supersolution of (4.1) for i = 1 and i = 2, respectively.
Then:
i) For each ε > 0 the function σε (.) is measurable and for each t ∈ [0, T ],

t
mε (t) ≤ mε (0) + σε (τ )dτ.
0

ii) If moreover γ > 0, then for each t ∈ [0, T ] mε (t) ≤ mε (0) + γ1 Σε (t).

The proof of this result is given in Subsection 4.4.

4.2 Case of an uncoupled nonlocal term

Proof of Theorem 3.1 We use a contracting fixed point theorem. We let the reader
verify that for u ∈ BU C(QT ) the Hamiltonian F (u) satisfies (4.2)–(4.4). Under
these assumptions, it is well-known that there is a comparison principle between
semicontinuous semisolutions of (2.2). Moreover, supQT |F (u) (t, x, 0, 0, 0)| < +∞
and for u0 ∈ BU C(RN ) there exists a unique bounded continuous solution w of
(2.2) such that (1.2) is satisfied; we let Θu denote this solution. For a proof of
these results we refer the reader to [9].

First step: stability of BU C(QT ) by Θ. Let us prove that for all u ∈ BU C(QT ),
Θu is uniformly continuous. We let mε (.), σε (.) and dnε (.) denote the functions
introduced in Subsection 4.1 for ui = u, Gi = F (u) (i = 1, 2) and γ = 0.
(u)
Let mR (.) denote the modulus deriving from (4.4). For ε > 0, t ∈ [0, T ] and
(x, y, r, p, X, Y ) ∈ dnε (t),
 
(u) |x − y|2
F (u) (t, y, r, p, Y ) − F (u) (t, x, r, p, X) ≤ mR |x − y| + ,
ε
2
where R = ||Θu||∞ . By the definition of dnε (t), we see that |x−y|
4ε ≤ m2ε (t) −
mε (t) + n1 . Thus
 
|x − y|2 √ 1 1
|x − y| + ≤ 2 ε m2ε (t) − mε (t) + + 4 m2ε (t) − mε (t) + .
ε n n
270 Nathaël Alibaud NoDEA

Let αεn (t) denote the right hand side of this inequality and define αε (t) := inf n∈N∗
αεn (t). We get
(u)
F (u) (t, y, r, p, Y ) − F (u) (t, x, r, p, X) ≤ mR (αεn (t)) .

Taking the supremum w.r.t. (x, y, r, p, X, Y ) ∈ dnε (t), we see that for all ε > 0 and
all t ∈ [0, T ],
(u) (u)
σε (t) ≤ inf∗ mR (αεn (t)) ≤ mR (αε (t)) .
n∈N

By the item i) of Theorem 4.1, we find that for all ε > 0 and all t ∈ [0, T ],

T
(u)
mε (t) ≤ mε (0) + mR (αε (τ )) dτ. (4.6)
0

We let Iε denote the integral term of this inequality. Let us recall that for each τ ∈
[0, T ], mε (τ ) is nonnegative and nondecreasing w.r.t. ε. The limit limε→0 > mε (τ )

then exists and limε→0 > (m2ε (τ ) − mε (τ )) = 0. Moreover, we know that mε (τ ) ≤

 
(u)
2||Θu||∞ and we infer that the family of functions mR (αε (.)) is uniformly
ε>0
>
bounded and converges to 0 pointwise as ε → 0. By Lebesgue’s convergence
Theorem, we deduce that limε→0
> Iε = 0. Since (4.6) implies that

|x − y|2
|Θu(t, x) − Θu(t, y)| ≤ mε (0) + Iε + ,
2ε

for all t ∈ [0, T ], all x, y ∈ RN and all ε > 0, we have proved that Θu is uniformly
continuous w.r.t. x independently of t.
The uniform continuity in both time and space variable is now a consequence
of the following result.

Proposition 4.1 For each ν > 0, there exists Cν ≥ 0 that only depends on the
Hamiltonian F , ||u||∞ , ||Θu||∞ and the modulus of continuity w.r.t. the x-variable
of Θu and that is such that, for each α ≥ 0,

sup ωα (Θu(., x)) ≤ inf {ν + Cν α}. (4.7)

x∈RN ν>0

The proof of Proposition 4.1 is well-known for local equations (see [6, Lemma 9.1])
and can easily be adapted to nonlocal equations; for the reader’s convenience, a
sketch of the proof of this result is given in Appendix A. The proof of the fact
that Θ maps BU C(QT ) into itself is complete.

Second step: resolution of (1.1)–(1.2). Define the space

E := {u ∈ BU C(QT ) : u(0, .) = u0 }
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 271

that we endow with the distance dE (u, v) = ||(u1 − u2 )γ ||∞ , where γ := 2C F C g

and

wγ (t, x) := e−γt w(t, x). (4.8)

We know that Θ(E) ⊆ E. Let us prove that Θ : E → E is a contraction. Let ui

γt
belong to E (i = 1, 2). Define v := Θu2 + e2 dE (u1 , u2 ). Let (a, p, X) belong to
P 2,− v(t, x). Using successively (H5), (H7) and (H3), we show that

a + F (u1 ) (t, x, v, p, X) ≥ a + F (t, x, Θu2 , p, X, g[u1 ]),

≥ a + F (t, x, Θu2 , p, X, g[u2 ])
−C F |g[u1 ](t, x) − g[u2 ](t, x)|,
≥ a + F (t, x, Θu2 , p, X, g[u2 ])
−C F C g sup ||u1 (τ, .) − u2 (τ, .)||∞ ,
τ ∈[0,t]
(u2 )
≥ a+F (t, x, Θu2 , p, X, ) − C F C g eγt dE (u1 , u2 ).
 
Since a − C F C g eγt dE (u1 , u2 ), p, X ∈ P 2,− Θu2 (t, x), the viscosity inequalities
applied to Θu2 (supersolution of (2.2) with u = u2 ) imply that v is a supersolution
of (2.2) with u = u1 . By the comparison principle, we infer that Θu1 ≤ v and
(Θu1 − Θu2 )γ ≤ 12 dE (u1 , u2 ). We can argue similarly to get the other inequality.
We conclude that Θ is a contraction. Since the metric space (E, dE ) is complete,
Banach fixed point Theorem implies that the Cauchy problem (1.1)–(1.2) admits
a unique viscosity solution in BU C(QT ).

Third step: proof of (3.2). Estimate (3.2) will be a consequence of the following
result.

Proposition 4.2 If u satisfies (3.2), then Θu also satisfies (3.2).

Proof. Let t ∈ [0, T [ and h > 0 be such that t + h ≤ T . A simple computation

shows that the function

s → v(s) := sup ||Θu(τ, .)||∞ + (s − t) sup |F (u) (τ, x, 0, 0, 0)|

τ ∈[0,t] (τ,x)∈Qt+h

is a supersolution of (2.2) in the domain ]t, t + h[×RN . Since Θu(t, .) ≤ v(t), the
comparison principle implies that

sup Θu ≤ sup ||Θu(τ, .)||∞ + h sup |F (τ, x, 0, 0, 0, g[u])|.

Qt+h τ ∈[0,t] (τ,x)∈Qt+h
272 Nathaël Alibaud NoDEA

Using successively (H7), (H3) and (H1) and (H8), we deduce that for all (τ, x) ∈
Qt+h ,

|F (τ, x, 0, 0, 0, g[u])| ≤ |F (τ, x, 0, 0, 0, 0)| + C F |g[u](τ, x)|,

≤ |F (τ, x, 0, 0, 0, 0)|
 

+C F
||g[0]||∞ + C g
sup ||u(τ , .)||∞ ,
τ  ∈[0,τ ]

≤ C F + C F C g (1 + sup ||u(τ  , .)||∞ ),

τ  ∈[0,t+h]

= C + γ0 sup ||u(τ  , .)||∞ ,

τ  ∈[0,t+h]

where C and γ0 are defined as in (3.2). We get

 
sup Θu ≤ sup ||Θu(τ, .)||∞ + h C + γ0 sup ||u(τ, .)||∞ .
Qt+h τ ∈[0,t] τ ∈[0,t+h]

We can argue similarly to get the other inequality. We deduce that for all 0 ≤ t ≤
t + h ≤ T,
 
sup ||Θu(τ, .)||∞ ≤ sup ||Θu(τ, .)||∞ + h C + γ0 sup ||u(τ, .)||∞ .
τ ∈[0,t+h] τ ∈[0,t] τ ∈[0,t+h]

Since Θu is uniformly continuous, the function

t ∈ [0, T ] → f (t) := sup ||Θu(τ, .)||∞

τ ∈[0,t]

is continuous. Lemma 4.1 then implies that if u satisfies (3.2), then f is a contin-
uous viscosity subsolution of the following equation:

f˙ = C + γ0 g in ]0, T [,

where g(t) is equal to the right hand side of (3.2). The comparison principle then
completes the proof of Proposition 4.2

Let us return to the proof of (3.2). Since {u ∈ E : (3.2) holds true} is a
nonempty closed subspace of E which is stable by Θ, we see that the unique fixed
point of Θ belongs to this subspace. This completes the proof of (3.2) and a
fortiori the proof of Theorem 3.1.

Proof of Theorem 3.2 Consider the space

E µ := {u ∈ E : (3.2) and (3.3) hold true} ,

where R is any L∞ bound on E µ .

Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 273

First step: stability of E µ by Θ. Let us prove that Θ(E µ ) ⊆ E µ . Proposition 4.2

implies that for all u ∈ E µ , Θu satisfies (3.2) and
||Θu||∞ ≤ R. (4.9)
To prove the Hölder continuity of Θu, we have to introduce some notations. For
γ > 0, define
Fγ (t, x, r, p, X, λ) := e−γt F (t, x, eγt r, eγt p, eγt X, λ),
Fγ(u) (t, x, r, p, X) := Fγ (t, x, r, p, X, g[u]).
The function (Θu)γ , defined as in (4.8), is a viscosity solution of the following
equation in w:
∂t w + γw + Fγ(u) (t, x, w, Dw, D2 w) = 0 in QT .
Following [15] and [10], we derive an Hölder estimate on (Θu)γ . Theorem 4.1 item
ii), applied to
Gi (t, x, r, p, X) = γr + Fγ(u) (t, x, r, p, X)
and ui = (Θu)γ (i = 1, 2), implies that for all ε > 0, all t ∈ [0, T ] and all x, y ∈ RN ,
1 |x − y|2
(Θu)γ (t, x) − (Θu)γ (t, y) ≤ mε (0) + Σε (t) + eγt . (4.10)
γ 2ε
We have
 +
|x − y|2
mε (0) = sup u0 (x) − u0 (y) − ,
(x,y)∈RN ×RN 2ε
 +
|x − y|2
≤ sup [u0 ]µ |x − y|µ − ,
(x,y)∈RN ×RN 2ε
2 µ
≤ [u0 ]µ2−µ ε 2−µ . (4.11)
Let us derive an upper bound on Σε (t). For (τ, x, y, r, p, X, Y ) ∈ Dε (t), define
   
I := Fγ(u) τ, y, r, eγτ p, eγτ Y − Fγ(u) τ, x, r, eγτ p, eγτ X , (4.12)
 
x − y γτ
= Fγ τ, y, r, eγτ , e Y, g[u](τ, y)
ε
 
x − y γτ
−Fγ τ, x, r, eγτ , e X, g[u](τ, x) .
ε
Condition (H7) implies that
 
γτ x − y
I ≤ Fγ τ, y, r, e γτ
, e Y, g[u](τ, x)
ε
 
γτ x − y
−Fγ τ, x, r, e , e X, g[u](τ, x) + e−γτ C F |g[u](τ, y) − g[u](τ, x)|.
γτ
ε
274 Nathaël Alibaud NoDEA

By (H1) and (H3), we see that |g[u](τ, x)| ≤ Rg where Rg is defined as in Theorem
3.2. By the definition of Dε (t), we see that eγτ |r| ≤ eγτ ||(Θu)γ (τ, .)||∞ ≤
||Θu||∞ ≤ R (see (4.9)). Moreover, Condition (3.1) holds true and (H6)’ then
implies that
 
|x − y|2
I ≤ e−γτ CR∨R F
g |x − y|µ + e(γ+γ)τ + e−γτ C F |g[u](τ, y) − g[u](τ, x)|,
ε
 
|x − y|2
≤ CR∨R
F
g |x − y|µ + eγτ + C F e−γτ [g[u](τ, .)]µ |x − y|µ .
ε
Using (H2)’, we get
|x − y|2
I ≤ C1 |x − y|µ + C2 eγτ , (4.13)
ε
 
F F
where C1 = CR∨R g +C Rg + C g sups∈[0,t] [uγ (s, .)]µ and C2 = CR∨R
F
g . Taking

the supremum w.r.t. (τ, x, y, r, p, X, Y ) ∈ Dε (t), we get

 +
|x − y|2 |x − y|2
Σε (t) ≤ sup C1 |x − y|µ + C2 eγτ − γeγτ .
ε 2ε
τ ∈[0,t],
x,y∈RN

If we take γ = 2 (C2 + 1), then for all t ∈ [0, T ] and all ε > 0,
 +
r2 2 µ
Σε (t) ≤ sup C1 rµ − ≤ C12−µ ε 2−µ . (4.14)
r>0 ε
Inequalities (4.10) and (4.11) then imply that for all t ∈ [0, T ] and all x, y ∈ RN ,
 2
 
 2
C 2−µ
µ |x − y| 
2
(Θu)γ (t, x) − (Θu)γ (t, y) ≤ inf [u0 ]µ2−µ + 1  ε 2−µ + eγt ,
ε>0  γ 2ε 
 2
 2−µ
2
2 2−µ
C1  µ
≤ 2 [u0 ]µ 2−µ
e 2 γt |x − y|µ ,
+
γ
 
C1 µ
≤ 2 [u0 ]µ + 2−µ e 2 γt |x − y|µ ;
γ 2
that is to say,
 
  C1 µ
(Θu(t, .))γ ≤ 2 [u0 ]µ + 2−µ e 2 γt
µ γ 2
 
F F g
(CR∨R g + C R ) µ 2C F C g µ
≤ 2 [u0 ]µ + 2−µ e 2 γt + 2−µ e 2 γt sup [uγ (τ, .)]µ .
γ 2 γ 2 τ ∈[0,t]
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 275

Recalling that γ > 0 is arbitrary, we can take γ = γ1 (t) > 0 in this inequality,
F
Cg µ
where γ1 (t) is such that 2C 2−µ e 2 γt = 12 . We then have proved that for all
γ1 (t) 2

u ∈ E µ and all t ∈ [0, T ],

 
  F
(CR∨R F g
g + C R ) µ 1
(Θu(t, .))γ1 (t) ≤ 2 [u0 ]µ + 2−µ e 2 γt + sup [uγ (t) (τ, .)]∞ ,
µ γ1 (t) 2 2 τ ∈[0,t] 1
1 µ
 1
≤ 4[u0 ]µ e 2 γt + CR + sup [uγ (t) (τ, .)]∞ , (4.15)
2 2 τ ∈[0,t] 1

where R is any L∞ bound on E µ and γ1 (t) and CR are defined as in Theorem

3.2. A simple computation now completes the proof of the stability of E µ by Θ.

Second step: conclusion. Since E µ is a nonempty closed subspace of E, we deduce

that the unique fixed point of Θ belongs to E µ . Thus, the unique viscosity solution
u of (1.1)–(1.2) is Hölder continuous. Noting now that (4.15) still holds true for
R = max{||u||∞ , ||Θu||∞ } completes the proof of Theorem 3.2.

4.3 Case of a coupled nonlocal term

Proof of Theorem 3.3 We use a contracting fixed point theorem. For each u ∈
Cb (QT ), the Hamiltonian F (u) satisfies (4.2)–(4.4) and sup QT |F (u) (t, x, 0, 0, 0)| <
+∞; hence, we can define Θu as the unique continuous bounded viscosity solution
of (2.2) which satisfies (1.2).

First step: a needed gradient estimate. Let us prove that for all u ∈ Cb (QT ) and
all t ∈ [0, T ],
γ
[Θu(t, .)]1 ≤ (2[u0 ]1 + CR t) e 2 t , (4.16)
where R = max{||u||∞ , ||Θu||∞ } and γ and CR are defined as in Theorem 3.3.
Let us argue similarly as in the preceding proof. We now use the item i) of
Theorem 4.1 for Gi = F (u) and ui = Θu (i = 1, 2). We get
|x − y|2
Θu(t, x) − (Θu) (t, y) ≤ mε (0) + t sup σε (τ ) + eγt . (4.17)
τ ∈[0,t] 2ε

Condition (H9) now gives the following estimate on I (the quantity I being
defined for r = 0 as in (4.12)): for all ε > 0, all t ∈ [0, T ], all τ ∈ [0, t] and
all (x, y, r, p, X, Y ) ∈ dnε (τ ),
 
|x − y|
I ≤ CR∨R F
g |x − y| 1 + eγτ
ε
 
|x − y| |g[u](τ, y) − g[u](τ, x)|
+C F |g[u](τ, y) − g[u](τ, x)| 1 + eγτ + eγτ ,
ε ε
276 Nathaël Alibaud NoDEA

 
|x − y|
≤ F
CR∨R g |x
− y| 1 + e γτ
ε
 
γτ |x − y| γτ [g[u]]1 |x − y|
+C [g[u]]1 |x − y| 1 + e
F
+e ,
ε ε
 F   F 2
 γτ |x − y|2
≤ CR∨R g +C [g[u]]1 |x − y|+ CR∨Rg + C [g[u]]1 + C [g[u]]1 e
F F F
,
ε
where R := max{||u||∞ , ||Θu||∞ } and Rg is defined as in Theorem 3.3. By (H2)”, we
F F g
get (4.13) for µ = 1 and C1 which is now equal to CR∨R g + C R . Inequalities (4.17),

(4.11) and (4.14) and an optimization wrt ∈ then complete the proof of (4.16).

Second step: local-in-time solvability of (1.1)–(1.2). For t∗ ∈]0, T ], define the

space
 
Et1∗ := u ∈ Cb (Qt∗ ) : u(0, .) = u0 and (3.2) holds true for all t ∈ [0, t∗ ]
endowed with the distance of the uniform convergence. For each u ∈ Et1∗ , we
let Θu also denote the unique bounded continuous viscosity solution of (1.1), in
the domain Qt∗ , that satisfies (1.2). By Proposition 4.2, we known that Θ maps
Et1∗ into itself. We leave it to the reader to verify that Proposition 4.2 still holds
true under the assumptions of Theorem 3.3. Let us prove that Θ : E → E is a
contraction for t∗ sufficiently small.
In what follows, we let R denote a L∞ bound on Et1∗ and we let R denote
a bound on Θ(Et1∗ ) for the [.]1 -seminorm. Such a number R exists, thanks to
(4.16). Consider ui ∈ Et1∗ (i = 1, 2). Define N := supt∈[0,t∗ ] {Θu1 (t, .) − Θu2 (t, .)}
and M := supt∈[0,t∗ ] ||u1 (t, .) − u2 (t, .)||∞ . The item i) of Theorem 4.1, applied to
the functions Θui , the Hamiltonians F (ui ) and γ = 0, implies that for all ε > 0,
2
[u0 ]21 R
N ≤ mε (0) + t∗ sup σε (t) ≤ ε + t∗ sup σε (t) ≤ ε + t∗ sup σε (t).
t∈[0,t∗ ] 2 t∈[0,t∗ ] 2 t∈[0,t∗ ]
(4.18)
Let us derive an upper bound on supt∈[0,t∗ ] σε (t). Consider t ∈ [0, t∗ ] and (x, y, r, p,
X, Y ) ∈ dnε (t). Condition (H9) implies that

F (u2 ) (t, y, r, p, Y ) − F (u1 ) (t, x, r, p, X) ≤

   
|x − y| |x − y| |λ − µ|
F
CR∨R g |x − y| 1 + + C F
|λ − µ| 1 + + , (4.19)
ε ε ε
where λ = g[u1 ](t, y) and µ = g[u2 ](t, x). By (H2)” and (H3),
|λ − µ| ≤ [g[u1 ](t, .)]1 |x − y| + |g[u1 ](t, x) − g[u2 ](t, x)|,
≤ C g (1 + sup ||u1 (τ, .)||∞ )|x − y| + C g M,
τ ∈[0,t∗ ]

≤ R |x − y| + C M.
g g
(4.20)
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 277

Using (4.19) and (4.20),

F (u2 ) (t, y, r, p, Y ) − F (u1 ) (t, x, r, p, X)

   
|x − y| |x − y| M
≤ C1 |x − y| 1 + + C2 M 1 + + , (4.21)
ε ε ε

where Ci only depends on C F , C g , CR∨R

F
g and R (i = 1, 2). By the definition of
n
dε (t),

√ 1 √ 1
|x − y| ≤ 2 ε m2ε (t) − mε (t) + ≤ 2R ε + 2 ε . (4.22)
n n

Indeed if m2ε (t) = 0, then (4.22) is immediate (since x = y). In the other case,
that is to say if m2ε (t) > 0, we can deduce from (4.37) that the approximate
supremum m2ε,η (t) (defined in (4.25)) is positive for all η sufficiently small. If
m2ε,η (t) is achieved at some (x, y) (by (4.24), such a maximal point always exists),
1,+
then we see that x−y2ε ∈ ∂ (Θu2 (t, .))(y) and it follows that |x − y| ≤ 2R ε. By
|x−y|2
a simple computation, we get m2ε,η (t) − mε,η (t) ≤ 4ε ≤ R 2 ε. The limit as
> 2
η → 0 in this inequality then gives m2ε (t) − mε (t) ≤ R ε. Inequality (4.22) is
now a immediate consequence of the definition of dnε (t). Taking the supremum
w.r.t. t ∈ [0, t∗ ] and (x, y, r, p, X, Y ) ∈ dnε (t) in (4.21) then gives
 
M
sup σε (t) ≤ C3 ε + C4 M 1 + ,
t∈[0,t∗ ] ε


where Ci only depends on C F , C g , CR∨R
F
g , R and R (i = 3, 4). Inequality (4.18)

implies that for all ε > 0,

    2 
R 2 ∗ M ∗ R M2
N≤ ε + t C3 ε + C4 M 1 + = t C4 M + + t C3 ε + t∗ C4
∗
.
2 ε 2 ε

By taking the infimum w.r.t. ε, there exists a universal constant C ≥ 0 such that
  2  12 
R √ 
sup {Θu1 (t, .) − Θu2 (t, .)} ≤ t∗ C4 + C + t∗ C3 t∗ C4 M.
t∈[0,t∗ ] 2

By exchanging the role of Θu1 and Θu2 , we conclude that

sup ||Θu1 (t, .) − Θu2 (t, .)||∞ ≤

t∈[0,t∗ ]
   12 
∗ R 2 √ 
t C4 + C + t∗ C3 t∗ C4 sup ||u1 (t, .) − u2 (t, .)||∞ .
2 t∈[0,t∗ ]
278 Nathaël Alibaud NoDEA


Since Ci only depends on C F , C g , CR∨R
F
g , R and R (i = 3, 4), we have proved
∗
that Θ is a contraction for t > 0 sufficiently small. Consequently, there exists a
∗ ∗
unique ut ∈ Cb (Qt∗ ) such that ut is solution of (1.1)–(1.2) in Qt∗ .

Third step: global solvability. Define now


I := t ∈]0, T ] : there exists a unique ut ∈ Cb (Qt ) that satisfies (3.2)

such that ut is solution of (1.1)–(1.2) in Qt

and tmax = sup I. By proposition 4.2 and the preceding step, I = ∅. Let us prove
that tmax = T . Let us assume the contrary and let us seek a contradiction. By
construction, there exists a unique utmax ∈ Cb (Qtmax ) such that utmax is solution
of (1.1)–(1.2) in Qtmax . By the gradient estimate of the first step (see (4.16)) and
Proposition 4.1, utmax satisfies (3.2), (3.4) and (4.7). We leave it to the reader to
verify that Proposition 4.1 still holds true under the assumptions of Theorem 3.3.
<
The family (utmax (s, .))s∈[0,tmax [ thus satisfies the Cauchy property for s → tmax
and it follows that utmax ∈ Cb0,1 (Qtmax ). Consider the following Cauchy problem:

∂t u + F (t, x, u, Du, D2 u, g[u]) = 0 in ]tmax , T [×RN , (4.23)

u(tmax , .) = utmax on RN

where g[u] := g[w] with w which is defined as follows: w := utmax on Qtmax

and w = u on [tmax , t] × RN . Arguing as in the second step, we can prove that
there exists t ∈]tmax , T ] such that, there exists a unique u ∈ Cb ([tmax , t] × RN )
solution of (4.23) in ]tmax , t[×RN . For each u ∈ Cb ([tmax , t] × RN ) which satisfies
 ∈ Cb (Qt ) as follows: u
u(tmax , .) = utmax (tmax , .), define u  = utmax on Qtmax and
 = u on [tmax , t] × RN . Since
u
2,(+)− tmax
P 2,(+)− u
(tmax , x) ⊆ PQ u (tmax , x),
tmax

 is solution of (1.1) in Qt .
the function u is solution of (4.23) in ]tmax , t[×RN iff u
We deduce that t ∈ I. Since t > tmax , we get a contradiction and necessarily
tmax = T . The proof of Theorem 3.3 is now complete (note that (4.16) implies
(3.4)).

4.4 Proof of Theorem 4.1

Before proving our continuous dependence estimate, we have to recall two classical
lemmas. The first lemma establishes a relation between the Fréchet subdifferential
and the Clarke generalized derivative of a function of the real variable and the
second lemma is a semicontinuity result on marginal functions.
Here is the first result.
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 279

Lemma 4.1 Let f : [0, T ] → R be a locally bounded function. For each s ∈]0, T [,

f (t + h) − f (t)
sup ∂ 1,+ f (s) ≤ lim sup .
h
t→s
>
h→0

Proof. Let ξ belong to ∂ 1,+ f (s). For each τ sufficiently small, f (s + τ ) ≤ f (s) +
ξτ + o(τ ). If τ < 0, then

f (s + τ ) − f (s) o(τ ) f (t) − f (t + h) o(h)

ξ≤ + = + ,
τ τ −h h
> <
where t + h := s and t := s + τ . Since t → s and h → 0 when τ → 0, the proof of
Lemma 4.1 is complete.

To state the second lemma, we have to introduce some notations and recall
some definitions on multiapplications. Consider E and F two metric spaces, k an
integer, and f a given function from E × F into R. For x ∈ E, define g(x) :=
supy∈d(x) f (x, y) where we let d(x) denote any subset of F depending on x.

Definition 4.1 The multiapplication d : E ⇒ F is said to be nonempty valued

if for each x ∈ E, d(x) is a nonempty subset of F . The multiapplication d is said
to be

i) upper semicontinuous on E (u.s.c. for short) iff for all x ∈ E and all neigh-
bourhood µ of d(x), there exists η > 0 such that for all x ∈ B E (x, η),
d(x ) ⊆ µ (1 );

ii) lower semicontinuous on E (l.s.c. for short) iff for all x ∈ E, all sequence
xm → x and all y ∈ d(x), there exists a sequence ym ∈ d(xm ) such that
ym → y;

iii) continuous on E iff it is both u.s.c. and l.s.c. on E.

Here is the second lemma.

Lemma 4.2 Assume that f is continuous and that d is nonempty valued, u.s.c.
on E and such that Im(d) := ∪x∈E d(x) is relatively compact. Then, g : E → R
is well-defined and upper semicontinuous. If moreover d is l.s.c. on E, then g is
continuous on E.

1 In this paper, we only consider multiapplications of the form d(x) = {y ∈ F : h(x, y) ≤ 0},

where h : E × F → Rk is continuous and Im(d) := ∪x∈E d(x) is relatively compact; note that
such multiapplications are always u.s.c..
280 Nathaël Alibaud NoDEA

For a proof of this result we refer the reader to [5, Theorem 7.3.1]. Let us now
return to the proof of our continuous dependence estimate.
Proof of Theorem 4.1 Let us introduce some notations that will be needed. Define

|x − y|2
ψε (t, x, y) := u1 (t, x) − u2 (t, y) − eγt ,
2ε
and

Ψ(t, x, y, r1 , r2 , p, X, Y ) :=
 
2 +
    γt |x − y|
G2 t, y, r2 , e p, e Y − G1 t, x, r1 , e p, e X − γe
γt γt γt γt
.
2ε

Let us perturb the functions mε (.) and σε (.) the following way: let φ ∈ C 2 (RN )
be nonnegative and such that φ(0) = 0, C φ := ||Dφ||∞ + ||D2 φ||∞ < +∞ and

lim φ(x) = +∞; (4.24)

|x|→+∞

for η ∈]0, 1] and t ∈ [0, T ], define (2 )

+
mε,η (t) := sup (ψε (t, x, y) − η(φ(x) + φ(y))) , (4.25)
(x,y)∈RN ×RN

σε,η (t) := sup Ψ(t, x, y, r, r, p, X, Y ),

dε,η (t)

where we let dε,η (t) denote the set of (x, y, r, p, X, Y ) such that,



 φ(x) + φ(y) ≤ 2(||u1 ||∞η∨||u2 ||∞ ) ,



 |r| ≤ maxi=1,2 sup φ≤ 2(||u1 ||∞η∨||u2 ||∞ ) |ui (t, .)|,

x−y
p= ε ,


 Condition (3.1) holds true,


 |x−y|2
4ε ≤ m2ε,η (t) − mε,η (t).

Let us give some properties on these functions. We admit the result below for a
while until the end of the proof of Theorem 4.1.

Proposition 4.3 The functions mε,η (.) and σε,η (.) are well-defined, real valued
and continuous on [0, T ].

2 The presence of the penalization term ηφ(y) is not necessary, but it will simplify the rest of

the proof.
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 281

The strategy of the proof of Theorem 4.1 is the following: first, we will prove that
there exists a constant R ≥ 0 such that for each η ∈]0, 1],

mε,η (.) is a continuous viscosity subsolution of the following

(4.26)
differential equation in f : f˙ = (σε,η − γf )+ + 2ωR (ηC φ ) in ]0, T [,

where we let ωR (.) denote the modulus deriving from (4.2); this will give us
some approximate continuous dependence estimates, thanks to the comparison
>
principle, and we will conclude by taking the limit as η → 0 in these estimates.

First step: proof of (4.26). By Proposition 4.3 and Lemma 4.1, (4.26) can be seen
as a consequence of the following property: for each s ∈]0, T [,
mε,η (t + h) − mε,η (t)
lim sup ≤ (σε,η (s) − γmε,η (s))+ + 2ωR (ηC φ ). (4.27)
h
t→s
>
h→0

To prove (4.27), we have to perturb again the functions mε,η (.) and σε,η (.) the
following way: define

E := {(t, h) ∈ R2 : 0 ≤ t ≤ t + h ≤ T }

and for every (t, h) ∈ E and λ ∈ R+ , define

+
Mε (t, h, λ) := sup (ψε (τ, x, y) − η(φ(x) + φ(y)) − λ(τ − t)) ,
τ ∈[t,t+h],
(x,y)∈RN ×RN

Σ(t, h, λ) := sup Ψ(τ, x, y, r1 , r2 , p, X, Y ),

D(t,h,λ)

where we let D(t, h, λ) denote the set of (τ, x, y, r1 , r2 , p, X, Y ) such that,



 τ ∈ [t, t + h],


 φ(x) + φ(y) ≤ 2(||u1 ||∞η∨||u2 ||∞ ) ,




 r1 − r2 ≥ Mε (t, h, λ),

|ri | ≤ maxj=1,2 sup[t,t+h]×
φ≤ 2(||u1 ||∞ ∨||u2 ||∞ )  |uj |, (i = 1, 2)


η

 p = x−y
,

 ε

 Condition (3.1) holds true,


 |x−y| 2

4ε ≤ M2ε (t, h, λ) − Mε (t, h, λ).

Here are some properties on these functions, that we admit until the end of the
proof of Theorem 4.1.

Proposition 4.4 The functions Mε (., ., .) and Σ(., ., .) are well-defined, real val-
ued and continuous on E × R+ .
282 Nathaël Alibaud NoDEA

The new domain, on which the supremum Mε (t, h, λ) is computed, is intro-

duced in order to use Equation (4.1). Note the presence of the penalization terms,
2
η(φ(x) + φ(y)) and |x−y|2ε , which are used classically when working with viscosity
solutions (see [9]). Let us recall that there are respectively used to treat unbounded
domain and to split in two the space variables and use Ishii Lemma.
The strategy of the proof of (4.27) is the following: for every (t, h) ∈ E, the
parameter λ will be chosen large enough in order to ensure that

Mε (t, h, λ) ≤ Mε (t, 0, λ) = mε,η (t); (4.28)

mε,η (t+h)−mε,η (t)

hence, we will infer that h ≤ λ and (4.27) will be proved by passing
>
to the limit in this inequality as t → s and h → 0.

Construction of the best λ satisfying (4.28). Let us take

 
2 (||u1 ||∞ ∨ ||u2 ||∞ ) eγT φ 6e
γT
R = max ||u1 ||∞ ∨ ||u2 ||∞ , 2 +C , +C φ
.
ε ε
(4.29)

For every (t, h) ∈ E, define

λ(t, h) := sup d(t, h), (4.30)

where we let d(t, h) denote the set of λ ∈ R+ such that

 
λ − Σ(t, h, λ) − 2ωR ηC φ ≤ 0. (4.31)

We have the following result:

Proposition 4.5 The function λ(., .) is well-defined, real valued and u.s.c on E.

Proof. It is easy to see that Σ(t, h, λ) is nonnegative and bounded independently

of (t, h, λ) ∈ E × R+ . Then, d(., .) is nonempty valued and ∪(t,h)∈E d(t, h) is
relatively compact. The continuity of Σ(., ., .) (see proposition 4.4) implies that
the multiapplication d(., .) is u.s.c. on E (see footnote 1 on page 17). Lemma 4.2
then completes the proof of Proposition 4.5.

Let us prove that for all (t, h) ∈ E,

Mε (t, h, λ(t, h)) ≤ Mε (t, 0, λ(t, h)). (4.32)

When Mε (t, h, λ(t, h)) = 0, (4.32) is immediate (since Mε (., ., .) is nonnegative).

Assume that Mε (t, h, λ(t, h)) > 0. The continuity Mε (., ., .) (see Proposition 4.4)
implies that for all λ > λ(t, h) sufficiently close to λ(t, h), Mε (t, h, λ) > 0. Let λ
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 283

be such a number. By Condition (4.24), there exists (τ , x, y) at which Mε (t, h, λ)

is achieved. Necessarily,

φ(x) + φ(y) ≤ 2(||u1 ||∞η∨||u2 ||∞ ) ,
2 (4.33)
eγτ |x−y|
4ε ≤ M2ε (t, h, λ) − Mε (t, h, λ).

Let us prove that τ = t. Assume the contrary and let us seek a contradiction. We
let O denote the cylinder ]t, t + h] × RN . Ishii Lemma (see [8] or [9, Theorem 8.3])
implies that there are
2,+
(a, eγτ p + ηDφ(x), eγτ X + ηD2 φ(x)) ∈ P O u1 (τ , x),
2,−
(b, eγτ p − ηDφ(y), eγτ Y − ηD2 φ(y)) ∈ P O u2 (τ , y),

with  x−y
p= ε ,
2
γτ |x−y|
 a − b = λ + γe 2ε ,
Condition (3.1) holds true.
Using the viscosity inequalities, we find that
 
a − b ≤ G2 τ , y, u2 , eγτ p − ηDφ(y), eγτ Y − ηD2 φ(y)
 
− G1 τ , x, u1 , eγτ p + ηDφ(x), eγτ X + ηD2 φ(x) .

γτ
By (4.33) and (3.1), we know that |eγτ p| ≤ 2 2(||u1 ||∞ ∨||u ε
2 ||∞ )e
and |X|, |Y | ≤
6
ε . Recalling that γ is nonnegative, the definition of R (see (4.29)) and (4.2) imply
that
     
a − b ≤ G2 τ , y, u2 , eγτ p, eγτ Y − G1 τ , x, u1 , eγτ p, eγτ X + 2ωR ηC φ .

We then deduce that

   
λ ≤ G2 τ , y, u2 , eγτ p, eγτ Y − G1 τ , x, u1 , eγτ p, eγτ X
|x − y|2  
−γeγτ + 2ωR ηC φ ,
2ε  
≤ Ψ(τ , x, y, r1 , r2 , p, X, Y ) + 2ωR ηC φ , (4.34)

where r1 = u1 (τ , x) and r2 = u2 (τ , y). Recalling that Mε (t, h, λ) > 0, we get

+
(ψε (τ , x, y) − η(φ(x) + φ(y)) − λ(τ − t)) > 0.

As a result of this,

ψε (τ , x, y) − η(φ(x) + φ(y)) − λ(τ − t) =

+
(ψε (τ , x, y) − η(φ(x) + φ(y)) − λ(τ − t)) = Mε (t, h, λ).
284 Nathaël Alibaud NoDEA

2
Since eγτ |x−y|
2ε +η(φ(x)+φ(y))+λ(τ −t) ≥ 0, we get r1 −r2 ≥ Mε (t, h, λ). Recalling
that (4.33) and (3.1) hold true, (τ , x, y, r1 , r2 , p, X, Y ) belongs to D(t, h, λ). By
(4.34), λ ≤ Σ(t, h, λ) + 2ωR ηC φ . By (4.30), (4.31) and the fact that λ > λ(t, h),
we get a contradiction. Consequently, τ = t and Mε (t, h, λ) ≤ Mε (t, 0, λ). By the
>
continuity of Mε (., ., .), we can pass to the limit in λ → λ(t, h) to complete the
proof of (4.32).
>
Passing to the limit as t → s and h → 0. By a simple computation, we
deduce from (4.32) that for all (t, h) ∈ E,
mε,η (t + h) ≤ Mε (t, h, λ(t, h)) + λ(t, h)h
≤ Mε (t, 0, λ(t, h)) + λ(t, h)h
= mε,η (t) + λ(t, h)h.
It follows that for all s ∈]0, T [,
mε,η (t + h) − mε,η (t)
lim sup ≤ lim sup λ(t, h) ≤ λ(s, 0),
h
t→s t→s
> >
h→0 h→0

thanks to the upper semicontinuity of λ(., .) (see Proposition 4.5). But, the con-
tinuity of Σ(., ., .) and the definition of λ(., .) imply that
λ(s, 0) = Σ(s, 0, λ(s, 0)) + 2ωR (ηC φ )
(in fact, this holds true not only for (t, h) = (s, 0) but for all (t, h) ∈ E ). Moreover,
(4.3) implies that
+
Σ(s, 0, λ(s, 0)) ≤ (σε,η (s) − γMε (s, 0, λ(s, 0))) = (σε,η (s) − γmε,η (s))+ .
We deduce that λ(s, 0) ≤ (σε,η (s) − γmε,η (s))+ + 2ωR (ηC φ ). This implies (4.27)
and this thus completes the proof of (4.26).
>
Second step: taking limit as η → 0. By the comparison principle, we deduce from
(4.26) that for each t ∈ [0, T ],

t
mε,η (t) ≤ mε,η (0) + σε,η (τ )dτ + 2t ωR (ηC φ ); (4.35)
0

moreover, if γ > 0, then for each t ∈ [0, T ]

1
mε,η (t) ≤ mε,η (0) + sup σε,η (τ )dτ + 2t ωR (ηC φ ). (4.36)
γ τ ∈[0,t]

Indeed, the right hand side of (4.35) (resp. (4.36)) is a classical solution (resp. a
continuous supersolution) of (4.26) as function of the t-variable. Let us pass to
>
the limit as η → 0 in these inequalities.
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 285

For all t ∈ [0, T ],

lim miε,η (t) = miε (t) (i = 1, 2) and lim sup σε,η (t) ≤ σε (t). (4.37)
> >
η →0 η →0

Indeed, let n be a nonnegative integer and let t belong to [0, T ]. For all positive
η sufficiently small, m2ε,η (t) − mε,η (t) ≤ m2ε (t) − mε (t) + n1 . Consequently,
dε,η (t) ⊆ dnε (t) and σε,η (t) ≤ σεn (t). Taking the infimum w.r.t. n ∈ N∗ implies
that lim supη→0 > σε,η (t) ≤ σε (t). The proof of the other limits of (4.37) is left to

the reader. The set dε,η (t) is bounded independently of t and η. We deduce, by
(4.4) and (4.5), that the family of functions (σε,η )η>0 is uniformly bounded on
>
[0, T ]. By (4.37) and Fatou’s Lemma, the limit as η → 0 in (4.35) completes the
proof of the item i) of Theorem 4.1 (for a proof of the measurability of σε (.), see
Appendix B).
If γ > 0, then Uτ ∈[0,t] {r}+dε,η (t) ⊆ Dε (t). Thus, the item ii) of Theorem 4.1
can easily be deduced from (4.36). This completes the proof of Theorem 4.1.

Proof of Propositions 4.3 and 4.4 We only prove the continuity of Σ(., ., .) since
the other proofs are easier or very similar. Let us use Lemma 4.2, with f = Ψ and
d(.) = D(., ., .). It is immediate that Ψ is continuous and we leave to the reader
to verify that the reader verify that ∪(t,h,λ)∈E×R+ D(t, h, λ) is relatively compact
and that D(., ., .) is u.s.c.. Note that for all (t, h, λ) ∈ E × R+ ,

(t, 0, 0, r∞ , −r∞ , 0, 0, 0) ∈ D(t, h, λ),

where r∞ is defined as in (4.40). This proves that D(., ., .) is nonempty valued. Let
us prove that D(., ., .) is l.s.c.. Let (t, h, λ) belong to E × R+ , (tm , hm , λm )n∈N∗ ∈
∗
(E × R+ )N be a sequence which converges to (t, h, λ) and (τ, x, y, r1 , r2 , p, X, Y )
belong to D(t, h, λ). For m ∈ N∗ , define
∞

rm := max sup[t 2(||u1 ||∞ ∨||u2 ||∞ ) |uj |,
j=1,2 m ,tm +hm ]× φ≤ η

Mm := Mε (tm , hm , λm )

and

τm := max{tm , min{τ, tm + hm }},

∞ ∞
r1,m := max{−rm , min{Mm , r1 , rm }},
∞
r2,m := max{−rm , min{r2 , r1,m − M m }}.

|x−yν |2 2
For any ν ∈ [0, 1], define yν := (1 − ν)x + νy. We get 4ε = ν 2 |x−y|
4ε . Define
!
2 4ε
νm := min 1, (M2ε (tm , hm , λm ) − Mε (tm , hm , λm ))
|x − y|2
286 Nathaël Alibaud NoDEA

x−ym
and ym = yνm and pm = ε . Let us prove that
(τm , x, ym , r1,m , r2,m , pm , X, Y ) ∈ D(tm , hm , λm ) (4.38)
and
(τm , x, ym , r1,m , r2,m , pm , X, Y ) → (τ, x, y, r1 , r2 , p, X, Y ).
We only verify that ri,m → ri (i=1,2), which is the most difficult point to prove.
Let us first verify (4.38). We only verify the conditions on ri,m (i = 1, 2) and we
leave the verification of the other conditions to the reader. That is to say, we have
∞
to verify that |ri,m | ≤ rm (i = 1, 2) and that r1,m − r2,m ≥ Mn . Observe that for
all reals a ≤ b and c,
a ≤ max{a, min{c, b}} ≤ b and
(4.39)
if a ≤ c ≤ b, then max{a, min{c, b}} = c.
∞ ∞ ∞
Since Mm ≤ 2rm , we see that −rm + Mm ≤ rm . By (4.39) and the nonnegativity
∞ ∞ ∞ ∞
of Mm , we infer that −rm ≤ −rm + Mm ≤ r1,m ≤ rm . Using now that −rm ≤
∞ ∞ ∞
r1,m − Mm , (4.39) implies that −rm ≤ r2,m ≤ r1,m − Mm ≤ rm − Mm ≤ rm . We
then conclude that (4.38) holds true.
Moreover, we have the following properties:
∞
rm → r∞ := max sup[t,t+h]×
φ≤ 2(||u1 ||∞ ∨||u2 ||∞ )  |uj |, (4.40)
j=1,2 η

Mm → M := Mε (t, h, λ).
This can be seen by using Lemma 4.2 to prove the continuity of r∞ and M , as
functions of the (t, h)-variable. By the continuity of the functions min{., .} and
max{., .}, we get
r1,m → max{−r∞ + M, r1 , r∞ },
" #
r2,m → max −r∞ , min{r2 , lim r1,m − M } .
m
∞
Since |ri | ≤ r and r1 − r2 ≥ M (because (τ, x, y, r1 , r2 , p, X, Y ) ∈ D(t, h, λ)), we
deduce that −r∞ +M ≤ r2 +M ≤ r1 ≤ r∞ and (4.39) implies that limm r1,m = r1
and limm r2,m = r2 . The proof of Propositions 4.3 and 4.4 is now complete.

A Sketch of the proof of Proposition 4.1

Let
supt∈[0,T ] ωβ (Θu(t, .)) − ν
Cν := sup ,
β>0 β2
$
2R
M := 2 ,
Cν
Cν := sup |F (t, x, 0, p, X, λ)|,
K
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 287

where R = ||u||∞ ∨ ||Θu||∞ and K denote the set of (t, x, p, X, λ) such that,


 t ∈ [0, T ],

|p| ≤ Cν M,
 |X| ≤ Cν ,


|λ| ≤ C g (1 + R).
By (H4), (H7) and (H8), we know that Cν is finite. Let s and y be fixed. A simple
computation shows that the function
|x − y|2
v(t, x) := Θu(s, y) + ν + Cν + Cν (t − s)
2
is a viscosity subsolution of (2.2) in ]s, T [×BRN (y, M ). Moreover, for each x
|x − y|2
Θu(s, x) − Θu(s, y) ≤ ω|x−y| (Θu(s, .)) ≤ ν + Cν .
2
It follows that Θu(s, .) ≤ v(s, .). Another simple computation shows that Θu ≤ v
on the domain ]s, T [×∂B RN (y, M ). Using the comparison principle with a Dirich-
let condition (see [9]) and choosing x = y, we deduce that Θu(t, x) ≤ Θu(s, x) +
ν + Cν (s − t). We can argue similarly to obtain the other inequality and prove
that |Θu(t, x) − Θu(s, x)| ≤ ν + Cν |s − t| for all t, s, x. The proof of Proposition
4.1 is complete.

B Proof of the measurability of σε (.)

For t ∈ [0, T ], define σεn (t) := supdnε (t) Ψ(t, x, y, r, r, p, X, Y ). Let us prove that
σεn (.) is measurable. For a set A of a metric space E, define
0 if x ∈ A,
IA (x) :=
−∞ if not.
We let A denote the set of (X, Y ) ∈ S2N such that (3.1) holds true. For t ∈ [0, T ],
we let B(t) (resp. C(t)) denote the set of r ∈ R (resp. (x, y) ∈ R2N ) such that
2
r ≤ ||u1 (t, .)||∞ ∨ ||u1 (t, .)||∞ (resp. |x−y|
4ε ≤ m2ε (t) − mε (t) + n1 ). We see that
 
n x−y
σε (t) = sup(x,y,r,X,Y )∈R2N ×R×A Ψ t, x, y, r, r, , X, Y
ε
!
+IB(t) (r) + IC(t) (x, y) .
 ◦
  ◦ 
For each t ∈ [0, T ], C(t) = C(t) and either B(t) = {0} or B(t) = B(t); this
implies that
  !
x−y
σεn (t) = sup Ψ t, x, y, r, r, , X, Y + IB(t) (r) + IC(t) (x, y) ,
Q2N ×Q×D ε
288 Nathaël Alibaud NoDEA

where D is a countable dense subset of A. Moreover, t → ||u1 (t, .)||∞ and

miε (.) are measurable as l.s.c. functions (i = 1, 2). Consequently, for each
(x, y, r, X, Y ) ∈ Q2N × Q × D, the function
 
x−y
t → Ψ t, x, y, r, r, + IB(t) (r) + IC(t) (x, y)
ε

is measurable. The function σε (.) then is measurable as countable supremum of

measurable functions. Since σε (.) = inf N∗ σεn (.), we have established the measur-
ability of σε (.).

Acknowledgments

The author would like to thank J. Droniou and C. Imbert for the fruitful
discussions we had together.

References
[1] O. ALVAREZ, P. CARDALIAGUET and R. MONNEAU, Existence and
uniqueness for dislocation dynamics with nonnegative velocity, Interfaces Free
Bound 7(4) (2005), 415–434.

[2] O. ALVAREZ and A. TOURIN, Viscosity solutions of nonlinear integro-

differential equations, Ann. Inst. H. Poincaré Anal. Non Linéaire, 13(3)
(1996), 293–317.

[3] O. ALVAREZ, P. HOCH, Y. LE BOUAR and R. MONNEAU, Dislocation

dynamics: short time existence and uniqueness of the solution, Arch. Rational
Mech. Ana. 181(3) (2006), 449.

[4] L. ALVAREZ, P.-L. LIONS and J.-M. MOREL, Image selective smoothing
and edge detection by nonlinear diffusion. II, SIAM J. Numer. Anal., 29(3)
(1992), 845–866.

[5] J.-P. AUBIN, Initiation à l’analyse appliquée, Masson (1994)

[6] G. BARLES, S. BITON and O. LEY, A Geometrical Approach to the Study

of Quasilinear Parabolic Equations in RN Arch. Rational. Mech. Ana., 162(4)
(2002), 287–325.

[7] A. BENSOUSSAN and J.-L. LIONS, Contrôle Impulsionnel et Inéquations

QuasiVariationnelles, Dunod, Paris (1982).

[8] M. G. CRANDALL and H. ISHII, The maximum principle for semicontinuous

functions, Differential Integral Equations 3 (1990), 1001–1014.
Vol. 14, 2007 Existence, uniqueness and regularity for nonlinear parabolic equations 289

[9] M. G. CRANDALL, H. ISHII and P.-L. LIONS, User’s guide to viscosity

solutions of second order partial differential equations, Bull. Amer. Math.
Soc. (N.S.), 27 (1992), 1–67.
[10] E. R. JAKOBSEN and K. H. KARLSEN, Continuous Dependance Estimates
for Viscosity Solutions of Fully Nonlinear Degenerate Parabolic Equations,
Journal of Differential Equations 183 (2002), 497–525.

[11] E. R. JAKOBSEN and K. H. KARLSEN, Continuous dependence estimates

for viscosity solutions of integro-PDEs, Journal of Differential Equations
212(2) (2005), 278–318.

[12] N. FORCADEL, Dislocation dynamics with a mean curvature term: short

time existence and uniqueness, Preprint (2005).
[13] C. IMBERT, A non-local regularization of first order Hamilton-Jacobi equa-
tions, Journal of Differential Equations 211 (2005), 214–246.

[14] A. SAYAH, Equations d’Hamilton-Jacobi du premier ordre avec termes

intégro-différentials, Parties I et II, Comm P.D.E. , 16 (1991), 1057–1093.

[15] P. E. SOUGANIDIS, Existence of viscosity solutions of Hamilton-Jacobi

equations, J. of Differential Equations 56 (1985), 345–390.

[16] N. YAMADA, Viscosity solutions for integro-differential equations of Volterra

type, Gatuko International Series Mathematical Sciences and Applications,
7 (1995), 441–446.

Received 5 October 2005; accepted 17 June 2006

Published Online First 21 September 2007