Weak Dynamic Programming Principle for Viscosity

Solutions
Bruno Bouchard, Nizar Touzi

To cite this version:

Bruno Bouchard, Nizar Touzi. Weak Dynamic Programming Principle for Viscosity Solutions. SIAM
Journal on Control and Optimization, 2011, 49 (3), pp.948-962. �hal-00367355v4�

HAL Id: hal-00367355

https://hal.science/hal-00367355v4
Submitted on 12 Jul 2011

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 WEAK DYNAMIC PROGRAMMING PRINCIPLE
FOR VISCOSITY SOLUTIONS
BRUNO BOUCHARD ∗ AND NIZAR TOUZI †

Abstract. We prove a weak version of the dynamic programming principle for standard stochas-
tic control problems and mixed control-stopping problems, which avoids the technical difficulties
related to the measurable selection argument. In the Markov case, our result is tailor-made for the
derivation of the dynamic programming equation in the sense of viscosity solutions.

Key words. Optimal control, Dynamic programming, discontinuous viscosity solutions.

AMS subject classifications. Primary 49L25, 60J60; secondary 49L20, 35K55.

1. Introduction. Consider the standard class of stochastic control problems in

the Mayer form

V (t, x) := sup E [f (XTν )|Xtν = x] ,

ν∈U

where U is the controls set, X ν is the controlled process, f is some given function,
0 < T ≤ ∞ is a given time horizon, t ∈ [0, T ) is the time origin, and x ∈ Rd is some
given initial condition. This framework includes the general class of stochastic control
problems under the so-called Bolza formulation, the corresponding singular versions,
and optimal stopping problems.
A key-tool for the analysis of such problems is the so-called dynamic programming
principle (DPP), which relates the time−t value function V (t, .) to any later time−τ
value V (τ, .) for any stopping time τ ∈ [t, T ) a.s. A formal statement of the DPP is:
00
V (t, x) = v(t, x) := sup E [V (τ, Xτν )|Xtν = x] .00 (1.1)
ν∈U

In particular, this result is routinely used in the case of controlled Markov jump-
diffusions in order to derive the corresponding dynamic programming equation in the
sense of viscosity solutions, see Lions [10, 11], Fleming and Soner [8], Touzi [15], for
the case of controlled diffusions, and Oksendal and Sulem [12] for the case of Markov
jump-diffusions.
The statement (1.1) of the DPP is very intuitive and can be easily proved in the
deterministic framework, or in discrete-time with finite probability space. However,
its proof is in general not trivial, and requires on the first stage that V be measurable.
When the value function V is known to be continuous, the abstract measurability
arguments are not needed and the proof of the dynamic programming principle is
significantly simplified. See e.g. Fleming and Soner [8], or Kabanov and Klueppelberg
[9] in the context of a special singular control problem in finance. Our objective is to
reduce the proof to this simple context in a general situation where the value function
has no a priori regularity.
The inequality ”V ≤ v” is the easy one but still requires that V be measurable.
Our weak formulation avoids this issue. Namely, under fairly general conditions on
the controls set and the controlled process, it follows from an easy application of the

∗ CEREMADE, Université Paris Dauphine and CREST-ENSAE, [email protected]

† Ecole Polytechnique Paris, Centre de Mathématiques Appliquées, [email protected]
1
2 BOUCHARD B. AND N. TOUZI

tower property of conditional expectations that

V (t, x) ≤ sup E [V ∗ (τ, Xτν )|Xtν = x] ,

ν∈U

where V ∗ is the upper semicontinuous envelope of the function V .

The proof of the converse inequality ”V ≥ v” in a general probability space turns
out to be difficult when the function V is not known a priori to satisfy some continuity
condition. See e.g. Bertsekas and Shreve [2], Borkar [3], and El Karoui [7].
Our weak version of the DPP avoids the non-trivial measurable selection argument
needed to prove the inequality V ≥ v in (1.1). Namely, in the context of a general
control problem presented in Section 2, we show in Section 3 that:

V (t, x) ≥ supν∈U E [ϕ(τ, Xτν )|Xt = x]

for every upper-semicontinuous minorant ϕ of V.

We also show that an easy consequence of this result is that

h i
V (t, x) ≥ sup E V∗ (τnν , Xτνnν )|Xt = x ,
ν∈U

where τnν := τ ∧inf {s > t : |Xsν − x| > n}, and V∗ is the lower semicontinuous envelope
of V .
This result is weaker than the classical DPP (1.1). However, in the controlled
Markov jump-diffusions case, it turns out to be tailor-made for the derivation of the
dynamic programming equation in the sense of viscosity solutions. Section 5 reports
this derivation in the context of controlled jump diffusions.
Finally, Section 4 provides an extension of our argument in order to obtain a weak
dynamic programming principle for mixed control-stopping problems.
2. The stochastic control problem. Let (Ω, F, P) be a probability space sup-
porting a càdlàg Rd -valued process Z with independent increments. Given T > 0, let
F := {Ft , 0 ≤ t ≤ T } be the completion of its natural filtration on [0, T ]. Note that
F satisfies the usual conditions, see e.g. [6]. We assume that F0 is trivial and that
FT = F.
For every t ≥ 0, we set Ft := (Fst )s≥0 , where Fst is the completion of σ(Zr −Zt , t ≤
r ≤ s ∨ t) by null sets of F.
We denote by T the collection of all F−stopping times. For τ1 , τ2 ∈ T with
τ1 ≤ τ2 a.s., the subset T[τ1 ,τ2 ] is the collection of all τ ∈ T such that τ ∈ [τ1 , τ2 ] a.s.
When τ1 = 0, we simply write Tτ2 . We use the notations T[τt 1 ,τ2 ] and Tτt2 to denote
the corresponding sets of Ft -stopping times.
Throughout the paper, the only reason for introducing the filtration F through
the process Z is to guarantee the following property of the filtrations Ft .
remark 2.1. Notice that Fst −measurable random variables are independent of
Ft for all s, t ≤ T , and that Fst is the trivial degenerate σ−algebra for s ≤ t. Similarly,
all Ft −stopping times are independent of Ft .
For τ ∈ T and a subset A of a finite dimensional space, we denote by L0τ (A)
the collection of all Fτ −measurable random variables with values in A. H0 (A) is the
collection of all F−progressively measurable processes with values in A, and H0rcll (A)
is the subset of all processes in H0 (A) which are right-continuous with finite left limits.
In the following, we denote by Br (z) (resp. ∂Br (z)) the open ball (resp. its
boundary) of radius r > 0 and center z ∈ R` , ` ∈ N.
 Weak Dynamic Programming Principle for Viscosity Solutions 3

Througout this note, we fix an integer d ∈ N, and we introduce the sets:

S := [0, T ] × Rd andSo := (τ, ξ) : τ ∈ TT and ξ ∈ L0τ (Rd ) .



We also denote by USC(S) (resp. LSC(S)) the collection of all upper-semicontinuous

(resp. lower-semicontinuous) functions from S to R.
The set of control processes is a given subset Uo of H0 (Rk ), for some integer k ≥ 1,
so that the controlled state process defined as the mapping:
ν
(τ, ξ; ν) ∈ S × Uo 7−→ Xτ,ξ ∈ H0rcll (Rd )for some S withS ⊂ S ⊂ So

is well-defined and satisfies:

ν


θ, Xτ,ξ (θ) ∈ Sfor all(τ, ξ) ∈ S and θ ∈ T[τ,T ] .

A suitable choice of the set S in the case of jump-diffusion processes driven by Brow-
nian motion is given in Section 5 below.
Given a Borel function f : Rd −→ R and (t, x) ∈ S, we introduce the reward
function J : S × U −→ R:
ν


J(t, x; ν) := E f Xt,x (T ) (2.1)

which is well-defined for controls ν in

n o
ν
 
U := ν ∈ Uo : E |f (Xt,x (T ))| < ∞, ∀ (t, x) ∈ S . (2.2)

We say that a control ν ∈ U is t-admissible if it is Ft -progressively measurable, and

we denote by Ut the collection of such processes. The stochastic control problem is
defined by:

V (t, x) := sup J(t, x; ν)for(t, x) ∈ S. (2.3)

ν∈Ut

remark 2.2. The restriction to control processes that are Ft -progressively mea-
surable in the definition of V (t, ·) is natural and consistent with the case where t = 0,
since F0 is assumed to be trivial, and is actually commonly used, compare with e.g.
[16]. It will be technically important in the following. It also seems a-priori necessary
in order to ensure that Assumption A4 below makes sense, see Remark 3.2 and the
proof of Proposition 5.4 below. However, we will show in Remark 5.2 below that it is
not restrictive.
3. Dynamic programming for stochastic control problems. For the pur-
pose of our weak dynamic programming principle, the following assumptions are cru-
cial.

Assumption A For all (t, x) ∈ S and ν ∈ Ut , the controlled state process satisfies:
ν
A1 (Independence) The process Xt,x is Ft -progressively measurable.
A2 (Causality) For ν̃ ∈ Ut , τ ∈ T[t,T ] and A ∈ Fτt , if ν = ν̃ on [t, τ ] and
t

ν ν̃
ν1A = ν̃1A on (τ, T ], then Xt,x 1A = Xt,x 1A .
t
A3 (Stability under concatenation) For every ν̃ ∈ Ut , and θ ∈ T[t,T ]:

ν1[0,θ] + ν̃1(θ,T ] ∈ Ut .
4 BOUCHARD B. AND N. TOUZI

t
A4 (Consistency with deterministic initial data) For all θ ∈ T[t,T ] , we have:
a. For P-a.e ω ∈ Ω, there exists ν̃ω ∈ Uθ(ω) such that
ν ν


E f Xt,x (T ) |Fθ (ω) ≤ J(θ(ω), Xt,x (θ)(ω); ν̃ω ).
t
b. For t ≤ s ≤ T , θ ∈ T[t,s] , ν̃ ∈ Us , and ν̄ := ν1[0,θ] + ν̃1(θ,T ] , we have:
ν̄ ν


E f Xt,x (T ) |Fθ (ω) = J(θ(ω), Xt,x (θ)(ω); ν̃)forP − a.e. ω ∈ Ω.
ν
remark 3.1. Assumption A2 above means that the process Xt,x is defined
(caused) by the control ν pathwise.
remark 3.2. Let θ be equal to a fixed time s in A4-b. If ν̃is allowed to depend 
ν̃
on Fs , then the left-hand side in A4-b does not coincide with E f (Xs,X ν (s)(ω) (T )) .
t,x
Hence, the above identity can not hold in this form.
remark 3.3. In Section 5 below, we show that Assumption A4-a holds with
equality in the jump-diffusion setting. Although we have no example of a control
problem where the equality does not hold, we keep Assumption A4-a under this form
because the proof only needs this requirement.
remark 3.4. Assumption A3 above implies the following property of the controls
set which will be needed later:
t t
A5 (Stability under bifurcation) For ν1 , ν2 ∈ Ut , τ ∈ T[t,T ] and A ∈ Fτ , we have:

ν̄ := ν1 1[0,τ ] + (ν1 1A + ν2 1Ac ) 1(τ,T ] ∈ Ut .

t
To see this, observe that τA := T 1A + τ 1Ac is a stopping time in T[t,T ] (the indepen-
dence of Ft follows from Remark 2.1), and ν̄ = ν1 1[0,τA ] +ν2 1(τA ,T ] is the concatenation
of ν1 and ν2 at the stopping time τA .
Given ν̄ as constructed above, it is clear that this control can be concatenated with
another control ν3 ∈ Ut by following the same argument. Iterating the above property,
t
we therefore see that for 0 ≤ t ≤ T and τ ∈ T[t,T ] , we have the following extension:
for a finite sequence (ν1 , . . . , νn ) of controls in Ut with νi = ν1 on [0, τ ], and for a
partion (Ai )1≤i≤n of Ω with Ai ∈ Fτt for every i ≤ n:
n
X
ν̄ := ν1 1[0,τ ] + 1(τ,T ] νi 1Ai ∈ Ut .
i=1

Our main result is the following weak version of the dynamic programming prin-
ciple which uses the following notation:
V∗ (t, x) := lim inf V (t0 , x0 ), V ∗ (t, x) := lim sup V (t0 , x0 ),(t, x) ∈ S.
(t0 ,x0 )→(t,x) (t0 ,x0 )→(t,x)

Theorem 3.5. Let Assumptions A hold true and assume that V is locally
bounded. Then for every (t, x) ∈ S, and for every family of stopping times {θν , ν ∈
t
Ut } ⊂ T[t,T ] , we have

V (t, x) ≤ sup E V ∗ (θν , Xt,x

ν
(θν )) .
 
(3.1)
ν∈Ut

Assume further that J(.; ν) ∈ LSC(S) for every ν ∈ Uo . Then, for any function
ϕ : S −→ R:
ϕ ∈ USC(S) and V ≥ ϕ =⇒ V (t, x) ≥ sup E ϕ(θν , Xt,x
ν
(θν )) ,
 
(3.2)
ν∈Utϕ
 Weak Dynamic Programming Principle for Viscosity Solutions 5

where Utϕ = ν ∈ Ut : E ϕ(θν , Xt,x (θν ))− < ∞ .

  ν
  
(θν ))+ < ∞ or E ϕ(θν , Xt,xν

Before proceeding to the proof of this result, we report the following consequence.
Corollary 3.6. Let the conditions of Theorem 3.5 hold. For (t, x) ∈ S, let
∞
{θν , ν ∈ Ut } ⊂ T[t,T
t ν
] be a family of stopping times such that Xt,x 1[t,θ ν ] is L −bounded
for all ν ∈ Ut . Then,

sup E V∗ (θν , Xt,xν

(θν )) ≤ V (t, x) ≤ sup E V ∗ (θν , Xt,x
ν
(θν )) .
   
(3.3)
ν∈Ut ν∈Ut

Proof. The right-hand side inequality is already provided in Theorem 3.5. Fix
r > 0. It follows from standard arguments, see e.g. Lemma 3.5 in [13], that we can
find a sequence of continuous functions (ϕn )n such that ϕn ≤ V∗ ≤ V for all n ≥ 1
and such that ϕn converges pointwise to V∗ on [0, T ] × Br (0). Set φN := minn≥N ϕn
for N ≥ 1 and observe that the sequence (φN )N is non-decreasing and converges
pointwise to V∗ on [0, T ] × Br (0). Applying (3.2) of Theorem 3.5 and using the
monotone convergence Theorem, we then obtain:

V (t, x) ≥ lim E φN (θν , Xt,x

ν
(θν )) = E V∗ (θν , Xt,x
ν
(θν )) .
   
N →∞

remark 3.7. Notice that the value function V (t, x) is defined by means of Ut
as the set of controls. Because of this, the lower semicontinuity of J(., ν) required in
the second part of Theorem 3.5 does not imply that V is lower semicontinuous in its
t-variable. See however Remark 5.3 below.
Proof. [Theorem 3.5] 1. Let ν ∈ Ut be arbitrary and set θ := θν . The first
assertion is a direct consequence of Assumption A4-a. Indeed, it implies that, for
P-almost all ω ∈ Ω, there exists ν̃ω ∈ Uθ(ω) such that
ν ν


E f Xt,x (T ) |Fθ (ω) ≤ J(θ(ω), Xt,x (θ)(ω); ν̃ω ) .
ν
Since, by definition, J(θ(ω), Xt,x (θ)(ω); ν̃ω ) ≤ V ∗ (θ(ω), Xt,x
ν
(θ)(ω)), it follows from
the tower property of conditional expectations that
ν ν
(T ) |Fθ ≤ E V ∗ θ, Xt,x ν

  
 

E f Xt,x (T ) = E E f Xt,x (θ) .

2. Let ε > 0 be given. Then there is a family (ν (s,y),ε )(s,y)∈S ⊂ Uo such that:

ν (s,y),ε ∈ Us and J(s, y; ν (s,y),ε ) ≥ V (s, y) − ε,for every(s, y) ∈ S. (3.4)

By the lower-semicontinuity of (t0 , x0 ) 7→ J(t0 , x0 ; ν (s,y),ε ), for fixed (s, y) ∈ S, together

with the upper-semicontinuity of ϕ, we may find a family (r(s,y) )(s,y)∈S of positive
scalars so that, for any (s, y) ∈ S,

ϕ(s, y)−ϕ(t0 , x0 ) ≥ −ε and J(s, y; ν (s,y),ε )−J(t0 , x0 ; ν (s,y),ε ) ≤ ε for (t0 , x0 ) ∈ B(s, y; r(s,y) ),
(3.5)
where, for r > 0 and (s, y) ∈ S,

B(s, y; r) := {(t0 , x0 ) ∈ S : t0 ∈ (s − r, s], |x0 − y| < r} . (3.6)

Note that we do not use here balls of the usual form Br (s, y) and consider the topoly
induced by half-closed intervals on [0, T ]. The fact that t0 ≤ s for (t0 , x0 ) ∈ B(s, y; r)
6 BOUCHARD B. AND N. TOUZI

will play an important role when appealing to Assumption A4-b in Step 3 below.
Clearly, {B(s, y; r) : (s, y) ∈ S, 0 < r ≤ r(s,y) } forms an open covering of (0, T ] × Rd .
It then follows from the Lindelöf covering Theorem, see e.g. [14] Theorem 6.3 Chap.
VIII, that we can find a countable sequence (ti , xi , ri )i≥1 of elements of S × R, with
0 < ri ≤ r(ti ,xi ) for all i ≥ 1, such that S ⊂ {0} × Rd ∪ (∪i≥1 B(ti , xi ; ri )). Set
A0 := {T } × Rd , C−1 := ∅, and define the sequence

Ai+1 := B(ti+1 , xi+1 ; ri+1 ) \ Ci whereCi := Ci−1 ∪ Ai , i ≥ 0.

With this construction, it follows from (3.4), (3.5), together with the fact that V ≥ ϕ,
that the countable family (Ai )i≥0 satisfies

ν
(θ, Xt,x (θ)) ∈ ∪i≥0 Ai P−a.s., Ai ∩Aj = ∅ for i 6= j ∈ N, and J(·; ν i,ε ) ≥ ϕ−3ε on Ai for i ≥ 1,
(3.7)
where ν i,ε := ν (ti ,xi ),ε for i ≥ 1.
t n
3. We now prove (3.2). We fix ν ∈ Ut and θ ∈ T[t,T ] . Set A := ∪0≤i≤n Ai , n ≥ 1.
Given ν ∈ Ut , we define

 n
X 
νsε,n := 1[t,θ] (s)νs + 1(θ,T ] (s) νs 1(An )c (θ, Xt,x
ν
(θ)) + ν
1Ai (θ, Xt,x (θ))νsi,ε ,fors ∈ [t, T ].
i=1

ν
Notice that {(θ, Xt,x (θ)) ∈ Ai } ∈ Fθt as a consequence of Assumption A1. Then,
it follows from the stability under concatenation Assumption A3, Remark 3.4 that
ν ε,n ∈ Ut . By the definition of the neighbourhood (3.6), notice that θ = θ ∧ ti ≤ ti on
ν
{(θ, Xt,x (θ)) ∈ Ai }. Then, using Assumptions A4-b, A2, and (3.7), we deduce that:

h  ε,n  i h  ε,n  i
ν ν ν ν



E f Xt,x (T ) |Fθ 1An θ, Xt,x (θ) = E f Xt,x (T ) |Fθ 1A0 θ, Xt,x (θ)
n
X h  ε,n  i
ν ν


+ E f Xt,x (T ) |Fθ∧ti 1Ai θ, Xt,x (θ)
i=1
 
ν ε,n ν


= V T, Xt,x (T ) 1A0 θ, Xt,x (θ)
n
X
ν
(θ∧ti ); ν i,ε )1Ai θ, Xt,x
ν


+ J(θ∧ti , Xt,x (θ)
i=1
n
X
ν ν



≥ ϕ(θ, Xt,x (θ)) − 3ε 1Ai θ, Xt,x (θ)
i=0
ν ν



= ϕ(θ, Xt,x (θ)) − 3ε 1An θ, Xt,x (θ) ,

which, by definition of V and the tower property of conditional expectations, implies

V (t, x) ≥ J(t, x; ν ε,n )

h h  ε,n  ii
ν
= E E f Xt,x (T ) |Fθ
ν ν ν ν



 


≥ E ϕ θ, Xt,x (θ) − 3ε 1An θ, Xt,x (θ) + E f Xt,x (T ) 1(An )c θ, Xt,x (θ) .
 Weak Dynamic Programming Principle for Viscosity Solutions 7

ν


Since f Xt,x (T ) ∈ L1 , it follows from the dominated convergence theorem that:
ν ν


V (t, x) ≥ −3ε + lim inf E ϕ(θ, Xt,x (θ))1An θ, Xt,x (θ)
n→∞
ν
(θ))+ 1An θ, Xt,x
ν


= −3ε + lim E ϕ(θ, Xt,x (θ)
n→∞
ν
(θ))− 1An θ, Xt,x
ν


− lim E ϕ(θ, Xt,x (θ)
n→∞
ν
 
= −3ε + E ϕ(θ, Xt,x (θ)) ,

where the last equality follows from the left-hand side of (3.7)  and from the mono-
ν
tone convergence theorem, due to the fact that either E ϕ(θ, Xt,x (θ))+ < ∞ or
(θ))− < ∞. The proof of (3.2) is completed by the arbitrariness of
 ν

E ϕ(θ, Xt,x
ν ∈ Ut and ε > 0.
remark 3.8. (Lower-semicontinuity condition I) It is clear from the above proof
that it suffices to prove the lower-semicontinuity of (t, x) 7→ J(t, x; ν) for ν in a subset
Ũo of Uo such that supν∈Ũt J(t, x; ν) = V (t, x). Here Ũt is the subset of Ũo whose
elements are Ft -progressively measurable. In most applications, this allows to reduce
to the case where the controls are essentially bounded or satisfy a strong integrability
condition.
remark 3.9. (Lower-semicontinuity condition II) In the above proof, the lower-
semicontinuity assumption is only used to construct the balls Bi on which J(ti , xi ; ν i,ε )−
J(·; ν i,ε ) ≤ ε. Clearly, it can be alleviated, and it suffices that the lower-semicontinuity
holds in time from the left, i.e.

lim inf J(t0 , x0 ; ν i,ε ) ≥ J(ti , xi ; ν i,ε ).

(t0 ,x0 )→(ti ,xi ), t0 ≤ti

remark 3.10. (The Bolza and Lagrange formulations) Consider the stochastic
control problem under the so-called Lagrange formulation:
"Z #
T
ν ν ν ν



V (t, x) := sup E Yt,x,1 (s)g s, Xt,x (s), νs ds + Yt,x,1 (T )f Xt,x (T ) ,
ν∈Ut t

where
ν ν ν ν


dYt,x,y (s) = −Yt,x,y (s)k s, Xt,x (s), νs ds , Yt,x,y (t) = y > 0 .

Then, it is well known that this problem can be converted into the Mayer formulation
(2.3) by augmenting the state process to (X, Y, Z), where
ν ν ν ν


dZt,x,y,z (s) = Yt,x,y (s)g s, Xt,x (s), νs ds , Zt,x,y,z (t) = z ∈ R ,

and considering the value function

 ν ν ν


V̄ (t, x, y, z) := sup E Zt,x,y,z (T ) + Yt,x,y (T )f Xt,x (T ) = yV (t, x) + z .
ν∈Ut

In particular, V (t, x) = V̄ (t, x, 1, 0). The first assertion of Theorem 3.5 implies
" Z θν #
ν ν ∗ ν ν ν ν ν


V (t, x) ≤ sup E Yt,x,1 (θ )V (θ , Xt,x (θ )) + Yt,x,1 (s)g s, Xt,x (s), νs ds (3.8)
.
ν∈Ut t
8 BOUCHARD B. AND N. TOUZI

Given a upper-semicontinuous minorant ϕ of V , the function ϕ̄ defined by ϕ̄(t, x, y, z) :=

yϕ(t, x) + z is an upper-semicontinuous minorant of V̄ . From the second assertion of
Theorem 3.5, we see that for a family {θν , ν ∈ Ut } ⊂ T[t,T
t
],

V (t, x) ≥ sup E ϕ̄ θν , Xt,x

ν
(θν ), Yt,x,1
ν
(θν ), Zt,x,1,0
ν
(θν )


ν∈Utϕ̄
θν
" Z #
ν
(θν )ϕ(θν , Xt,x
ν
(θν )) + ν ν


= sup E Yt,x,1 Yt,x,1 (s)g s, Xt,x (s), νs ds (3.9)
.
ν∈Utϕ̄ t

remark 3.11. (Infinite Horizon) Infinite horizon problems can be handled sim-
ilarly. Following the notations of the previous Remark 3.10, we introduce the infinite
horizon stochastic control problem:
Z ∞ 
∞ ν ν


V (t, x) := sup E Yt,x,1 (s)g s, Xt,x (s), νs ds .
ν∈Ut t

Then, it is immediately seen that V ∞ satisfies the weak dynamic programming prin-
ciple (3.8)-(3.9).
4. Dynamic programming for mixed control-stopping problems. In this
section, we provide a direct extension of the dynamic programming principle of The-
orem 3.5 to the larger class of mixed control and stopping problems.
In the context of the previous section, we consider a Borel function f : Rd −→ R,
and we assume |f | ≤ f¯ for some continuous function f¯. For (t, x) ∈ S the reward
J¯ : S × Ū × T[t,T ] −→ R:

¯ x; ν, τ ) := E f Xt,x
ν


J(t, (τ ) , (4.1)

which is well-defined for every control ν in

n h i o
Ū := ν ∈ Uo : E sup f¯(Xt,x ν
(s)) < ∞ ∀ (t, x) ∈ S .
t≤s≤T

The mixed control-stopping problem is defined by:

V̄ (t, x) := sup ¯ x; ν, τ ) ,
J(t, (4.2)
t
(ν,τ )∈Ūt ×T[t,T ]

where Ūt is the subset of elements of Ū that are Ft -progressively measurable.

The key ingredient for the proof of (4.6) is the following property of the set of
stopping times TT :

For all θ, τ1 ∈ TTt and τ2 ∈ T[θ,T

t t
] ,we haveτ1 1{τ1 <θ} + τ2 1{τ1 ≥θ} ∈ TT . (4.3)

In order to extend the result of Theorem 3.5, we shall assume that the following
version of A4 holds:
t t
Assumption A4’ For all (t, x) ∈ S, (ν, τ ) ∈ Ūt × T[t,T ] and θ ∈ T[t,T ] , we have:
θ(ω)
a. For P-a.e ω ∈ Ω, there exists (ν̃ω , τ̃ω ) ∈ Ūθ(ω) × T[θ(ω),T ] such that

ν ν



1{τ ≥θ} (ω)E f Xt,x (τ ) |Fθ (ω) ≤ 1{τ ≥θ} (ω)J θ(ω), Xt,x (θ)(ω); ν̃ω , τ̃ω
 Weak Dynamic Programming Principle for Viscosity Solutions 9

t s
b. For t ≤ s ≤ T , θ ∈ T[t,s] , (ν̃, τ̃ ) ∈ Ūs × T[s,T ] , τ̄ := τ 1{τ <θ} + τ̃ 1{τ ≥θ} , and
ν̄ := ν1[0,θ] + ν̃1(θ,T ] , we have for P−a.e. ω ∈ Ω:
ν̄ ν


1{τ ≥θ} (ω)E f Xt,x (τ̄ ) |Fθ (ω) = 1{τ ≥θ} (ω)J(θ(ω), Xt,x (θ)(ω); ν̃, τ̃ ).
Theorem 4.1. Let Assumptions A1, A2, A3 and A4’ hold true. Then for every
(t, x) ∈ S, and for all family of stopping times {θν , ν ∈ Ūt } ⊂ T[t,T t
]:
ν
(τ )) + 1{τ ≥θν } V̄ ∗ (θν , Xt,x
ν
(θν )) . (4.4)
 
V̄ (t, x) ≤ sup E 1{τ <θν } f (Xt,x
t
(ν,τ )∈Ūt ×T[t,T ]

Assume further that the map (t, x) 7−→ J(t, ¯ x; ν, τ ) satisfies the following lower-
semicontinuity property
lim inf ¯ 0 , x0 ; ν, τ ) ≥ J(t,
J(t ¯ x; ν, τ )for every(t, x) ∈ S and (ν, τ ) ∈ Ū × T . (4.5)
0 0
t ↑t,x →x

Then, for any function ϕ ∈ USC(S) with V̄ ≥ ϕ:

ν
(τ )) + 1{τ ≥θν } ϕ(θν , Xt,x
ν
(θν )) , (4.6)
 
V̄ (t, x) ≥ sup E 1{τ <θν } f (Xt,x
(ν,τ )∈Ūtϕ ×T[t,T
t
]

where Ūtϕ = ν ∈ Ūt : E ϕ(θν , Xt,x (θν ))− < ∞ .

  ν
  
(θν ))+ < ∞ or E ϕ(θν , Xt,x
ν

For simplicity, we only provide the proof of Theorem 4.1 for optimal stopping
problems, i.e. in the case where Ū is reduced to a singleton. The dynamic program-
ming principle for mixed control-stopping problems is easily proved by combining the
arguments below with those of the proof of Theorem 3.5.

Proof. (for optimal stopping problems) We omit the control ν from all notations,
¯ x; τ ). Inequality (4.4) follows immediately from
thus simply writing Xt,x (·) and J(t,
the tower property together with Assumptions A4’-a, recall that J¯ ≤ V̄ ∗ .
We next prove (4.6). Arguing as in Step 2 of the proof of Theorem 3.5, we first
observe that, for every ε > 0, we can find a countable family Āi ⊂ (ti −ri , ti ]×Ai ⊂ S,
together with a sequence of stopping times τ i,ε in T[ttii,T ] , i ≥ 1, satisfying Ā0 =
{T } × Rd and
¯ τ i,ε ) ≥ ϕ − 3ε on Āi for i ≥ 1(4.7)
∪i≥0 Āi = S, Āi ∩ Āj = ∅ for i 6= j ∈ N,andJ(·; .
Set Ān := ∪i≤n Āi , n ≥ 1. Given two stopping times θ, τ ∈ T[t,Tt
] , it follows from (4.3)
(and Assumption A1 in the general mixed control case) that
n
!
X
n,ε i,ε
τ := τ 1{τ <θ} + 1{τ ≥θ} T 1(Ān )c (θ, Xt,x (θ)) + τ 1Āi (θ, Xt,x (θ))
i=1
t
is a stopping time in T[t,T ].
We then deduce from the tower property together with
Assumptions A4’-b and (4.7) that
¯ x; τ n,ε )
V̄ (t, x) ≥ J(t,
ν


≥ E f Xt,x (τ ) 1{τ <θ} + 1{τ ≥θ} (ϕ(θ, Xt,x (θ)) − 3ε) 1Ān (θ, Xt,x (θ))
 
+E 1{τ ≥θ} f (Xt,x (T ))1(Ān )c (θ, Xt,x (θ)) .
By sending n → ∞ and arguing as in the end of the proof of Theorem 3.5, we deduce
that
 
V̄ (t, x) ≥ E f (Xt,x (τ )) 1{τ <θ} + 1{τ ≥θ} ϕ(θ, Xt,x (θ)) − 3ε,
t
and the result follows from the arbitrariness of ε > 0 and τ ∈ T[t,T ].
10 BOUCHARD B. AND N. TOUZI

5. Application to controlled Markov jump-diffusions. In this section, we

show how the weak DPP of Theorem 3.5 allows to derive the correponding dynamic
programming equation in the sense of viscosity solutions. We refer to Crandal, Ishii
and Lions [5] and Fleming and Soner [8] for a presentation of the general theory of
viscosity solutions.
For simplicity, we specialize the discussion to the context of controlled Markov
jump-diffusions driven by a Brownian motion and a compound Poisson process. The
same technology can be adapted to optimal stopping and impulse control or mixed
problems, see e.g. [4].
5.1. Problem formulation and verification of Assumption A. We shall
work on the product space Ω := ΩW ×ΩN where ΩW is the set of continuous functions
from [0, T ] into Rd , and ΩN is the set of integer-valued measures on [0, T ] × E with
E := Rm for some m ≥ 1. For ω = (ω 1 , ω 2 ) ∈ Ω, we set W (ω) = ω 1 and N (ω) = ω 2
and define FW = (FtW )t≤T (resp. FN = (FtN )t≤T ) as the smallest right-continuous
filtration on ΩW (resp. ΩN ) such that W (resp. N ) is optional. We let PW be the
Wiener measure on (ΩW , FTW ) and PN be the measure on (ΩN , FTN ) under which N
is a compound Poisson measure with intensity Ñ (de, dt) = λ(de)dt, for some finite
measure λ on E, endowed with its Borel tribe E. We then define the probability
measure P := PW ⊗ PN on (Ω, FTW ⊗ FTN ). With this construction, W and N are
independent under P. Without loss of generality, we can assume that the natural right-
continuous filtration F = (Ft )t≤T induced by (W, N ) is complete. In the following, we
shall slightly abuse notations and sometimes write Nt (·) for N (·, (0, t]) for simplicity.
We let U be a closed subset of Rk , k ≥ 1, and µ : S×U −→ Rd and σ : S×U −→
M be two Lipschitz continuous functions, and β : S × U × E −→ Rd a measurable
d

function, Lipschitz-continuous with linear growth in (t, x, u) uniformly in e ∈ E. Here

Md denotes the set of d-dimensional square matrices.
By Uo , we denote the collection of all square integrable predictable processes with
values U . For every ν ∈ Uo , the stochastic differential equation:
Z
dX(r) = µ (r, X(r), νr ) dr+σ (r, X(r), νr ) dWr + β(r, X(r−), νr , e)N (de, dr), t ≤ r ≤ T,
E
(5.1)
ν ν
has a unique strong solution Xτ,ξ such that Xτ,ξ (τ ) = ξ, for any initial condition
  
(τ, ξ) ∈ S := (τ, ξ) ∈ So : ξ is Fτ − measurable, and E |ξ|2 < ∞ . Moreover, this
solution satisfies
 
ν
(r)|2 < C(1 + E |ξ|2 ),
 
E sup |Xτ,ξ (5.2)
τ ≤r≤T

for some constant C which may depend on ν.

remark 5.1. Clearly, less restrictive conditions could be imposed on β and
N . We deliberately restrict here to this simple case, in order to avoid standard
technicalities related to the definition of viscosity solutions for integro-differential
operators, see e.g. [1] and the references therein.
The following remark shows that in the present case, it is not necessary to restrict
the control processes ν to Ut in the definition of the value function V (t, x).
remark 5.2. Let Ṽ be defined by
ν
 
Ṽ (t, x) := sup E f (Xt,x (T )) .
ν∈U
 Weak Dynamic Programming Principle for Viscosity Solutions 11

The difference between Ṽ (t, ·) and V (t, ·) comes from the fact that all controls in U
are considered in the former, while we restrict to controls independent of Ft in the
latter. We claim that
Ṽ = V ,
so that both problems are indeed equivalent. Clearly, Ṽ ≥ V . To see that the converse
holds true, fix (t, x) ∈ [0, T ) × Rd and ν ∈ U. Then, ν can be written as a measur-
able function of the canonical process ν((ωs )0≤s≤t , (ωs − ωt )t≤s≤T ), where, for fixed
(ωs )0≤s≤t , the map ν(ωs )0≤s≤t : (ωs − ωt )t≤s≤T 7→ ν((ωs )0≤s≤t , (ωs − ωt )t≤s≤T ) can
be viewed as a control independent on Ft . Using the independence of the increments
of the Brownian motion and the compound Poisson process, and Fubini’s Lemma, it
thus follows that
Z h Z
ν(ωs ) i
J(t, x; ν) = E f (Xt,x 0≤s≤t (T )) dP((ωs )0≤s≤t ) ≤ V (t, x)dP((ωs )0≤s≤t )

where the latter equals V (t, x). By arbitrariness of ν ∈ U, this implies that Ṽ (t, x) ≤
V (t, x).
remark 5.3. By the previous remark, it follows that the value function V
inherits the lower semicontinuity of the performance criterion required in the second
part of Theorem 3.5, compare with Remark 3.7. This simplification is specific to the
simple stochastic control problem considered in this section, and may not hold in
other control problems, see e.g. [4]. Consequently, we shall deliberately ignore the
lower semicontinuity of V in the subsequent analysis in order to show how to derive
the dynamic programming equation in a general setting.
Let f : Rd −→ R be a lower semicontinuous function with linear growth, and
define the performance criterion J by (2.1). Then, it follows that U = Uo and,
ν
from (5.2) and the almost sure continuity of (t, x) 7→ Xt,x (T ), that J(., ν) is lower
semicontinuous, as required in the second part of Theorem 3.5.
The value function V is defined by (2.3). Various types of conditions can be for-
mulated in order to guarantee that V is locally bounded. For instance, if f is bounded
from above, this condition is trivially satisfied. Alternatively, one may restrict the
set U to be bounded, so that the linear growth of f implies corresponding bounds for
V . We do not want to impose such a constraint because we would like to highlight
the fact that our methodology applies to general singular control problems. We then
leave this issue as a condition which is to be checked by specific arguments to the case
in hand.
Proposition 5.4. In the above controlled diffusion context, assume further that
V is locally bounded. Then, the value function V satisfies the weak dynamic program-
ming principle (3.1)-(3.2).
Proof. Conditions A1, A2 and A3 from Assumption A are obviously satisfied in
the present context. It remains to check that A4 holds true. For ω ∈ Ω and r ≥ 0,
we denote ω·r := ω.∧r and Tr (ω)(·) := ω·∨r − ωr so that ω· = ω·r + Tr (ω)(·). Fix
t
(t, x) ∈ S, ν ∈ Ut , θ ∈ T[t,T ] , and observe that, by the flow property,
Z  
 ν

 ν(ω θ(ω) +Tθ(ω) (ω))
E f Xt,x (T ) |Fθ (ω) = f Xθ(ω),X ν (θ)(ω) (T )(Tθ(ω) (ω)) dP(Tθ(ω) (ω))
t,x
Z  
ν(ω θ(ω) +Tθ(ω) (ω̃))
= f Xθ(ω),X ν (θ)(ω) (T )(Tθ(ω) (ω̃)) dP(ω̃)
t,x

ν
= J(θ(ω), Xt,x (θ)(ω); ν̃ω )
12 BOUCHARD B. AND N. TOUZI

where, ν̃ω (ω̃) := ν(ω θ(ω) + Tθ(ω) (ω̃)) is an element of Uθ(ω) . This already proves A4-a.
t
As for A4-b, note that if ν̄ := ν1[0,θ] + ν̃1(θ,T ] with ν̃ ∈ Us and θ ∈ T[t,s] , then the
same computations imply
Z  
 ν̄

 ν̃(ω θ(ω) +Tθ(ω) (ω̃))
E f Xt,x (T ) |Fθ (ω) = f Xθ(ω),X ν (θ)(ω) (T )(Tθ(ω) (ω̃)) dP(ω̃),
t,x

ν ν̄
where we used the flow property together with the fact that Xt,x = Xt,x on [t, θ]
ν̄
and that the dynamics of Xt,x depends only on ν̃ after θ. Now observe that ν̃ is
independent of Fs and therefore on ω θ(ω) since θ ≤ s P − a.s. It follows that
Z  
 ν̄

 ν̃(Ts (ω̃))
E f Xt,x (T ) |Fθ (ω) = f Xθ(ω),X ν (θ)(ω) (T )(Tθ(ω) (ω̃)) dP(ω̃)
t,x

ν
= J(θ(ω), Xt,x (θ)(ω); ν̃) .

remark 5.5. It can be similarly proved that A4’ holds true, in the context of
mixed control-stopping problems.
5.2. PDE derivation. We can now show how our weak formulation of the dy-
namic programming principle allows to characterize the value function as a discontin-
uous viscosity solution of a suitable Hamilton-Jacobi-Bellman equation.
Let C 0 denote the set of continuous maps on [0, T ]×Rd endowed with the topology
of uniform convergence on compact sets. To (t, x, p, A, ϕ) ∈ [0, T ]×Rd ×Rd ×Md ×C 0 ,
we associate the Hamiltonian of the control problem:

H(t, x, p, A, ϕ) := inf H u (t, x, p, A, ϕ),

u∈U

where, for u ∈ U ,
1
H u (t, x, p, A, ϕ) := −hµ(t, x, u), pi − Tr [(σσ 0 )(t, x, u)A]
Z 2
− (ϕ(t, x + β(t, x, u, e)) − ϕ(t, x)) λ(de),
E

and σ 0 is the transpose of the matrix σ.

Notice that the operator H is upper-semicontinuous, as an infimum over a family
of continuous maps (note that β is locally bounded uniformly with respect to its last
argument and that λ is finite, by assumption). However, since the set U may be
unbounded, it may fail to be continuous. We therefore introduce the corresponding
lower-semicontinuous envelope:

H∗ (z) := lim
0
inf H(z 0 )forz = (t, x, p, A, ϕ) ∈ S × Rd × Md × C 0 .
z →z

Corollary 5.6. Assume that V is locally bounded. Then:

(i) V ∗ is a viscosity subsolution of

−∂t V ∗ + H∗ (., DV ∗ , D2 V ∗ , V ∗ ) ≤ 0on[0, T ) × Rd .

(ii) V∗ is a viscosity supersolution of

−∂t V∗ + H(., DV∗ , D2 V∗ , V∗ ) ≥ 0on[0, T ) × Rd .

 Weak Dynamic Programming Principle for Viscosity Solutions 13

Proof. 1. We start with the supersolution property. Assume to the contrary that
there is (t0 , x0 ) ∈ [0, T ) × Rd together with a smooth function ϕ : [0, T ) × Rd −→ R
satisfying

0 = (V∗ − ϕ)(t0 , x0 ) < (V∗ − ϕ)(t, x)for all(t, x) ∈ [0, T ) × Rd , (t, x) 6= (t0 , x0 ),

such that

−∂t ϕ + H(., Dϕ, D2 ϕ, ϕ) (t0 , x0 ) < 0.

(5.3)

For ε > 0, let φ be defined by

φ(t, x) := ϕ(t, x) − ε(|t − t0 |2 + |x − x0 |4 ),

and note that φ converges uniformly on compact sets to ϕ as ε → 0. Since H is

upper-semicontinuous and (φ, ∂t φ, Dφ, D2 φ)(t0 , x0 ) = (ϕ, ∂t ϕ, Dϕ, D2 ϕ)(t0 , x0 ), we
can choose ε > 0 small enough so that there exist u ∈ U and r > 0, with t0 + r < T ,
satisfying

−∂t φ + H u (., Dφ, D2 φ, φ) (t, x) < 0for all(t, x) ∈ Br (t0 , x0 ),

(5.4)

where we recall that Br (t0 , x0 ) denotes the ball of radius r and center (t0 , x0 ). Let
(tn , xn )n be a sequence in Br (t0 , x0 ) such that (tn , xn , V (tn , xn )) → (t0 , x0 , V∗ (t0 , x0 )),
and let X·n := Xtun ,xn (·) denote the solution of (5.1) with constant control ν = u and
initial condition Xtnn = xn , and consider the stopping time

θn := inf {s ≥ tn : (s, Xsn ) ∈

/ Br (t0 , x0 )} .

Note that θn < T since t0 + r < T . Applying Itô’s formula to φ(·, X n ), and using
(5.4) and (5.2), we see that
" Z θn #
n
∂t φ − H (., Dφ, D φ, φ) (s, Xs )ds ≤ E φ(θn , Xθnn ) .
u 2 n
   
φ(tn , xn ) = E φ(θn , Xθn ) −
tn

Now observe that ϕ ≥ φ + η on ([0, T ] × Rd) \ Br (t0 , x0 ) for some η > 0. Hence, the
above inequality implies that φ(tn , xn ) ≤ E ϕ(θn , Xθnn ) − η. Since (φ − V )(tn , xn ) →
0, we can then find n large enough so that

V (tn , xn ) ≤ E ϕ(θn , Xθnn ) − η/2for sufficiently largen ≥ 1.

 

On the other hand, it follows from (3.2) that:

V (tn , xn ) ≥ sup E ϕ(θn , Xtνn ,xn (θn )) ≥ E ϕ(θn , Xθnn ) ,

   
ν∈Utn

which is the required contradiction.

2. We now prove the subsolution property. Assume to the contrary that there is
(t0 , x0 ) ∈ [0, T ) × Rd together with a smooth function ϕ : [0, T ) × Rd −→ R satisfying

0 = (V ∗ − ϕ)(t0 , x0 ) > (V ∗ − ϕ)(t, x)for all(t, x) ∈ [0, T ) × Rd , (t, x) 6= (t0 , x0 ),

14 BOUCHARD B. AND N. TOUZI

such that

−∂t ϕ + H∗ (., Dϕ, D2 ϕ, ϕ) (t0 , x0 ) > 0.

(5.5)

For ε > 0, let φ be defined by

φ(t, x) := ϕ(t, x) + ε(|t − t0 |2 + |x − x0 |4 ),

and note that φ converges uniformly on compact sets to ϕ as ε → 0. By the lower-

semicontinuity of H∗ , we can then find ε, r > 0 such that t0 + r < T and

−∂t φ + H u (., Dφ, D2 φ, φ) (t, x) > 0for everyu ∈ U and (t, x) ∈ Br (t0 , x0 ).(5.6)

Since (t0 , x0 ) is a strict maximizer of the difference V ∗ − φ, it follows that

sup (V ∗ − φ) ≤ −2η for some η > 0 . (5.7)

([0,T ]×Rd )\Br (t0 ,x0 )

Let (tn , xn )n be a sequence in Br (t0 , x0 ) such that (tn , xn , V (tn , xn )) → (t0 , x0 , V ∗ (t0 , x0 )).
n
For an arbitrary control ν n ∈ Utn , let X n := Xtνn ,xn denote the solution of (5.1) with
initial condition Xtnn = xn , and set

θn := inf {s ≥ tn : (s, Xsn ) ∈

/ Br (t0 , x0 )} .

Notice that θn < T as a consequence of the fact that t0 + r < T . We may assume
without loss of generality that

|(V − φ)(tn , xn )| ≤ ηfor alln ≥ 1. (5.8)

Applying Itô’s formula to φ(·, X n ) and using (5.6) leads to

" Z θn h #
i
n νn
∂t φ − H (., Dφ, D φ, φ) (s, Xs )ds ≥ E φ(θn , Xθnn ) .
2 n
 
φ(tn , xn ) = E φ(θn , Xθn ) −
tn

In view of (5.7), the above inequality implies that φ(tn , xn ) ≥ E V ∗ (θn , Xθnn ) + 2η,
 

which implies by (5.8) that:

V (tn , xn ) ≥ E V ∗ (θn , Xθnn ) + ηforn ≥ 1.

 

Since ν n ∈ Utn is arbitrary, this contradicts (3.1) for n ≥ 1 fixed.

Acknowledgments. The authors are grateful to Nicole El Karoui and Marcel
Nutz for fruitful comments. This research is part of the Chair Financial Risks of the
Risk Foundation sponsored by Société Générale, the Chair Derivatives of the Future
sponsored by the Fédération Bancaire Française, the Chair Finance and Sustainable
Development sponsored by EDF and Calyon, and the Chair Les particuliers face au
risque sponsored by Groupama.

REFERENCES

[1] G. Barles and C. Imbert, Second-Order Elliptic Integro-Differential Equations: Viscosity So-
lutions’ Theory Revisited, Annales de l’IHP, 25 (2008), pp. 567-585.
[2] D.P. Bertsekas and S.E. Shreve, Stochastic Optimal Control : The Discrete Time Case,
Mathematics in Science and Engineering, 139, Academic Press, 1978.
 Weak Dynamic Programming Principle for Viscosity Solutions 15

[3] V.S. Borkar, Optimal Control of Diffusion Processes, Pitman Research Notes 203. Longman
Sci. and Tech. Harlow, 1989.
[4] B. Bouchard, N.-M. Dang and C.-A. Lehalle, Optimal control of trading algorithms: a
general impulse control approach, to appear in SIAM Journal on Financial Mathematics.
[5] M.G. Crandall, H. Ishii and P.-L. Lions, User’s guide to viscosity solutions of second order
Partial Differential Equations, Amer. Math. Soc., 27 (1992), pp. 1-67.
[6] C. Dellacherie and P.-A. Meyer, Probabilité et Potentiel, Théorie du potentiel, Hermann,
Springer, 1987.
[7] N. El Karoui, Les Aspects probabilistes du contrôle stochastique, Springer Lecture Notes in
Mathematics 876, Springer Verlag, New York, 1981.
[8] W.H. Fleming and H.M. Soner, Controlled Markov Processes and Viscosity Solutions, Second
Edition, Springer, 2006.
[9] Y. Kabanov and C. Klueppelberg, A geometric approach to portfolio optimization in models
with transaction costs, Finance and Stochastics, 8 (2004), pp. 207-227.
[10] P.-L. Lions, Optimal Control of Diffusion Processes and Hamilton-Jacobi-Bellman Equations
I, Comm. PDE., 8 (1983), pp. 1101-1134.
[11] P.-L. Lions, Optimal Control of Diffusion Processes and Hamilton-Jacobi-Bellman Equations,
Part II: Viscosity Solutions and Uniqueness, Comm. PDE., 8 (1983), pp. 1229-1276.
[12] B. Oksendal and A. Sulem, Applied Stochastic Control of Jump Diffusions, Universitext,
Springer (Second edition), 2007.
[13] P. J. Reny, On the Existence of Pure and Mixed Strategy Nash Equilibria in Discontinuous
Games, Econometrica, 67 (1999), pp. 1029-1056.
[14] J. Dugundj, Topology, Allyn and Bacon series in Advanced Mathematics, Allyn and Bacon edt,
1966.
[15] N. Touzi, Stochastic Control Problems, Viscosity Solutions, and Application to Finance,
Quaderni, Edizioni della Scuola Normale Superiore, Pisa, 2002.
[16] J. Yong and X. Zhou, Stochastic Controls: Hamiltonian Systems and HJB Equations,
Springer, New York, 1999.