Large Deviation Principle For Reflected Stochastic Diffe 2023 Statistics P

Statistics and Probability Letters 193 (2023) 109707
Contents lists available at ScienceDirect
Statistics and Probability Letters

journal homepage: www.elsevier.com/locate/stapro
Large deviation principle for Reflected Stochastic Differential

Equations driven by G-Brownian motion in non-convex
domains
∗
Abdoulaye Soumana Hima, Ibrahim Dakaou
Département de Mathématiques, Université Dan Dicko Dankoulodo de Maradi, BP 465, Maradi, Niger
article info a b s t r a c t
Article history: In this paper, we establish a large deviation principle for the solution of Reflected
Received 6 June 2022 Stochastic Differential Equations driven by G-Brownian motion in non-convex domains.
Received in revised form 5 September 2022 Moreover, we prove that the solution converges to the solution of a deterministic
Accepted 6 October 2022
Skorohod equation.
Available online 14 October 2022
© 2022 Elsevier B.V. All rights reserved.
MSC:
60F10
60H10
Keywords:
Large deviations
Skorohod problem
G-Brownian motion
Stochastic differential equations
Non-convex reflecting boundaries
1. Introduction
Large deviation theory is a branch of probability theory that deals with the study of rare events and of small
probabilities. This branch finds application in statistics, operations research, information theory, statistical physics,
financial mathematics and the list goes on. Nowadays, one of the main applications of large deviation theory is risk theory.
Let X s,x,ε be the diffusion process that is the unique solution of the following stochastic differential equation (SDE in
short)
t t
√
∫ ∫
s,x,ε
Xt =x+ β (Xrs,x,ε )dr + ε σ (Xrs,x,ε )dWr , 0 ≤ s ≤ t ≤ T (1.1)
s s
where β is a Lipschitz function defined on Rn with values in Rn , σ is a Lipschitz function defined on Rn with values
in Rn×d , and W is a standard Brownian motion in Rd defined on a complete probability space (Ω , F , P). The existence
and uniqueness of the strong solution X s,x,ε of (1.1) is standard. Thanks to the work of Freidlin and Wentzell (1984), the
s,x
sequence (X s,x,ε )ε>0 converges in probability, as ε goes to 0, to (ϕt )s≤t ≤T solution of the following deterministic equation
∫ t
ϕts,x = x + β (ϕrs,x )dr , 0 ≤ s ≤ t ≤ T
s
and satisfies a large deviation principle (LDP in short).
∗ Corresponding author.
E-mail addresses: [email protected] (A. Soumana Hima), [email protected] (I. Dakaou).
https://doi.org/10.1016/j.spl.2022.109707
0167-7152/© 2022 Elsevier B.V. All rights reserved.
A. Soumana Hima and I. Dakaou Statistics and Probability Letters 193 (2023) 109707
Sheu (1998) established a large deviation principle of reflecting diffusions. More precisely, the author proved that the
solution of stochastic differential equation with a small diffusion coefficient in a nonsmooth domain normally reflected
at boundary satisfies a large deviation principle and converges to a deterministic path in Lp .
Motivated from the study of mathematical finance, various versions of nonlinear expectations are introduced. Shige
Peng constructed a kind of dynamically consistent fully nonlinear expectations through PDE approach. Some authors
contributed to obtain results of large deviation principle under nonlinear expectation.
Gao and Jiang (2010) extended the work of Freidlin and Wentzell (1984) to stochastic differential equations driven by
G-Brownian motion (G-SDEs in short). The authors considered the following G-SDE: for every 0 ≤ t ≤ T ,
∫ t ∫ t ∫ t
x,ε
Xt =x+ bε (Xrx,ε )dr + ε hε (Xrx,ε )d⟨B, B⟩r /ε + ε σ ε (Xrx,ε )dBr /ε
0 0 0
and used discrete time approximation to establish LDP for G-SDEs. Their proof avoids the stopping time technique and
the Girsanov transformation. Their main tool is exponential moment estimates.
Recently, Lin and Hima (2019) generalize the results of Lin (2013a,b) to the multidimensional case when the reflecting
boundary is not necessarily convex. The authors pathwisely constructed an iteration sequence by using the deterministic
result and conducted a fixed-point argument as Lions and Sznitman (1984) to obtain their results on multi-dimensional
reflected stochastic differential equations driven by G-Brownian motion.
Motivated by the aforementioned works, we aim to establish LDP for Reflected G-SDEs. More precisely, we consider
the following reflected stochastic differential equation driven by G-Brownian motion: for every 0 ≤ t ≤ T , x0 ∈ D,
⎧
ε
∫ t ( ε) ∫ t ( ε) ∫ t ( ε) ε
⎨ Xt = x∫0 + 0 b Xr dr + ε 0 ∫h Xr d⟨B, B⟩r + ε 0 σ Xr dBr + Kt
⎪
ε t ε ε t ε
K = 0 nr d |K |r , |K |t = 0 1{Xrε ∈∂ D} d |K |r
⎪ t
nr ∈ NXrε , if Xrε ∈ ∂ D
⎩
The purpose of this paper is to prove that (X ε , K ε ) converges, as ε goes to 0, to (ϕ, K) solution of a deterministic Skorohod
equation and {X ε }ε>0 satisfies a large deviation principle. Compared with the classical large deviation theory, there
are not enough results on large deviation theory and its related applications under nonlinear expectation. We would
like to mention that Hu (2010) obtained Cramér’s upper bound for capacities induced by sublinear expectations; Cao
(2014) established, under sublinear expectations, a new expression of the Gärtner–Ellis theorem, which is a version
of Cramér’s theorem where independence is weakened; Chen and Xiong (2012) obtained the large deviation principle
of diffusion process under sublinear expectations; Dakaou and Hima (2021) established a large deviation principle for
backward stochastic differential equations driven by G-Brownian motion; Liu and Zhang (2021) established large deviation
principle for linear processes generated by real stationary sequences under the sub-linear expectation. On one hand, our
present paper continues to find suitable methods to study the large deviation theory and its application under nonlinear
expectation and on the other hand the paper is the first study of large deviations for reflected SDEs under nonlinear
expectation.
The remaining part of the paper is organized as follows. In Section 2, we present some preliminaries in the G-framework
that are useful in this paper. In Section 3, we give some notations, state the assumptions, and define the Reflected G-SDE
we are concerned with. Section 4 is devoted to the large deviations for Reflected G-SDE.
2. Preliminaries
We review some basic notions and results about G-expectation, G-Brownian motion and G-stochastic integrals (see
Peng, 2019, for more details).
Let Ω be a complete separable metric space, and let H be a linear space of real-valued functions defined on Ω
satisfying: if Xi ∈ H, i = 1, . . . , n, then
ϕ (X1 , . . . , Xn ) ∈ H, ∀ϕ ∈ Cl,Lip (Rn ),
where Cl,Lip (Rn ) is the space of real continuous functions defined on Rn such that for some C > 0 and k ∈ N depending
on ϕ ,
|ϕ (x) − ϕ (y)| ≤ C (1 + |x|k + |y|k )|x − y|, ∀x, y ∈ Rn .
Definition 2.1 (Sublinear expectation space).. A sublinear expectation Ê on H is a functional Ê : H −→ R satisfying the
following properties: for all X , Y ∈ H, we have
1. Monotonicity: if X ≥ Y , then Ê[X ] ≥ Ê[Y ];

2. Constant preservation: Ê[c ] = c;
3. Sub-additivity: Ê[X + Y ] ≤ Ê[X ] + Ê[Y ];
4. Positive homogeneity: Ê[λX ] = λÊ[X ], for all λ ≥ 0.
(Ω , H, Ê) is called a sublinear expectation space.
2
Definition 2.2 (Independence).. Fix the sublinear expectation space (Ω , H, Ê). A random variable Y ∈ H is said to be
independent of (X1 , X2 , . . . , Xn ), Xi ∈ H, if for all ϕ ∈ Cl,Lip (Rn+1 ),
[ ]
Ê [ϕ (X1 , X2 , . . . , Xn , Y )] = Ê Ê [ϕ (x1 , x2 , . . . , xn , Y )] ⏐(x .
⏐
1 ,x2 ,...,xn )=(X1 ,X2 ,...,Xn )
Now we introduce the definition of G-normal distribution.
Definition 2.3 (G-normal distribution).. A random variable X ∈ H is called G-normally distributed, √ noted by X ∼
N (0, [σ 2 , σ 2 ]), 0 ≤ σ ≤ σ , if for any function ϕ ∈ Cl,Lip (R), the function u defined by u(t , x) := Ê[ϕ (x + tX )], (t , x) ∈
[0, ∞) × R, is a viscosity solution of the following G-heat equation:
∂t u − G D2x u = 0, u(0, x) = ϕ (x),
( )
where
1
G(a) := (σ 2 a+ − σ 2 a− ).
2
In multi-dimensional case, the function G(·): Sd −→ R is defined by
1
G(A) = sup tr(AQ),
2 Q∈Σ
where Sd denotes the space of d × d symmetric matrices and Σ is a given nonempty, bounded and closed subset of Sd .
Hence we usually denote X ∼ N (0, Σ ), where Σ := {θθ τ : θ ∈ Θ } and Θ ⊂ Rd×d , θ τ is the transpose of θ .
Throughout this paper, we assume that there exist constants 0 < σ ≤ σ < ∞ such that
Θ ⊂ {θ ∈ Rd×d : σ 2 Id×d ≤ θθ τ ≤ σ 2 Id×d },
where Id×d is d-dimensional unit matrix.
Let Ω := C ([0, ∞), R), which equipped with the raw filtration F generated by the canonical process (Bt )t ≥0 ,
i.e., Bt (ω) = ωt , for (t , ω) ∈ [0, ∞) × Ω . Let ΩT := C ([0, T ], R) and let us consider the function spaces defined by
{
Lip(ΩT ) := ϕ (Bt1 , Bt2 − Bt1 , . . . , Btn − Btn−1 ) : n ≥ 1,
}
0 ≤ t1 ≤ t2 ≤ · · · ≤ tn ≤ T , ϕ ∈ Cl,Lip (Rn ) , for T > 0,
∞
⋃
Lip(Ω ) := Lip(Ωn ).
n=1
Definition 2.4 (G-Brownian motion and G-expectation).. On the sublinear expectation space Ω , Lip(Ω ), Ê , the canonical
( )
process (Bt )t ≥0 is called a G-Brownian motion if the following properties are verified:
1. B0 = 0
2. For each t , s ≥ 0, the increment Bt +s − Bt ∼ N (0, [sσ 2 , sσ 2 ]) and is independent from (Bt1 , . . . , Btn ), for 0 ≤ t1 ≤
· · · ≤ tn ≤ t.
Moreover, the sublinear expectation Ê is called G-expectation.
(√
Remark 2.1. For each λ > 0, λBt /λ t ≥0 is also a G-Brownian motion. This is the scaling property of G-Brownian motion,
)
which is the same as that of the classical Brownian motion.
( [ ])1/p
p
For ξ ∈ Lip(ΩT ) and p ≥ 1, we consider the norm ∥ξ ∥Lp := Ê |ξ |p . Denote by LG (ΩT ) the Banach completion of
G
Lip(ΩT ) under ∥ · ∥Lp . It is easy to check that the conditional G-expectation Êt [·] : Lip(ΩT ) −→ Lip(Ωt ) is a continuous
G
p p
mapping and thus can be extended to Êt [·] : LG (ΩT ) −→ LG (Ωt ).
Theorem 2.1 (Representation Theorem of G-expectation, see Hu and Peng, 2009; Denis et al., 2011).. There exists a weakly
compact set P ⊂ M1 (ΩT ), the set of probability measures on (ΩT , B(ΩT )), such that
Ê[ξ ] := sup EP [ξ ] for all ξ ∈ L1G (ΩT ).
P ∈P
P is called a set that represents Ê.

Let P be a weakly compact set that represents Ê. For this P , we define the capacity of a measurable set A by
Ĉ (A) := sup P(A), A ∈ B ( ΩT ) .
P ∈P
A set A ∈ B(ΩT ) is a polar if Ĉ (A) = 0. A property holds quasi-surely (q.s.) if it is true outside a polar set.
3
An important feature of the G-expectation framework is that the quadratic variation ⟨B⟩ of the G-Brownian motion is
no longer a deterministic process, which is given by
N −1
∑
⟨B⟩t := ( lim (Bt N − Bt N )2 ,
j+1 j
)
δ πtN →0 j=0
where πtN = {t0 , t1 , . . . , tN }, N = 1, 2, . . ., are refining partitions of [0, t ]. By Peng (2019), for all s, t ≥ 0, ⟨B⟩t +s − ⟨B⟩t ∈
[sσ 2 , sσ 2 ], q.s.
Let MG0 (0, T ) be the collection of processes in the following form: for a given partition πTN := {t0 , t1 , . . . , tN } of [0, T ],
N −1
∑
ηt (ω) = ξj (ω) 1[tj ,tj+1 ) (t), (2.1)
j=0
( [(
∫T )p/2 ])1/p
where ξi ∈ Lip(Ωti ), for all i = 0, 1, . . . , N − 1. For p ≥ 1 and η ∈ MG0 (0, T ), let ∥η∥H p := Ê 0
2
|ηs | ds ,
G
( [∫ ])1/p
T p p
∥η∥M p := Ê 0 |ηs |p ds and denote by HG (0, T ), MG (0, T ) the completions of MG0 (0, T ) under the norms ∥ · ∥H p ,
G G
∥ · ∥M p respectively.
G
For η ∈ MG0 (0, T ) of the form (2.1), the Itô integral with respect to G-Brownian motion is defined by the linear mapping
I : MG0 (0, T ) −→ L2G (ΩT ),
∫ T N −1
∑
I (η) := ηt dBt = ξk (Btk+1 − Btk ),
0 k=0
which can be continuously extended to I : HG1 (0, T ) −→ L2G (ΩT ). On the other hand, the stochastic integral with respect
to (⟨B⟩t )t ≥0 is defined by the linear mapping Q : MG0 (0, T ) −→ L1G (ΩT ),
∫ T N −1
∑
Q(η) := ηt d⟨B⟩t = ξk (⟨B⟩tk+1 − ⟨B⟩tk ),
0 k=0
which can be continuously extended to Q : HG1 (0, T ) −→ L1G (ΩT ). ∫·

Finally, we recall that Gao (2009) proves the G-Itô type integral X· = 0 ηr dBr has a continuous Ĉ -modification, for any
η ∈ MG (0, T ).
2
3. Notations, assumptions, and formulation of the problem
Let D be a domain in Rd . For x ∈ ∂ D, we denote

⋃
Nx,r = {n ∈ Rd : |n| = 1, B(x − rn, r) ∩ D = ∅}, r > 0 and Nx = Nx,r ,
r >0
where B(z , r) := {y ∈ Rd : |y − z | < r |}, for z ∈ Rd .
CONDITION (A). Assume that there exists a constant r0 > 0 such that Nx = Nx,r0 ̸ = ∅, for all x ∈ ∂ D.
CONDITION (B). Assume that there exist constants δ > 0 and β ∈ [1, ∞) such that for any x ∈ ∂ D, we can find a unit
vector lx such that
1 ⋃
⟨lx , n⟩ ≥ , for all n∈ Ny ,
β
y∈B(x,δ )∩∂ D
where ⟨·, ·⟩ denotes the usual inner product in Rd .

Throughout this paper, we consider a domain D ⊂ Rd satisfying both Condition (A) and (B). For each x ∈ Rd such
that dist(x, D) < r0 , there exists a unique x ∈ D with |x − x| = dist(x, D). If x ∈
/ D we have x ∈ ∂ D and |xx−x
−x|
∈ Nx (see
e.g. Remark 1.3 in Saisho, 1987). We keep this notation x for the projection of x on D in the remainder of this paper.
We assume furthermore
CONDITION (C). There exists a bounded function Ψ ∈ Cb2 Rd whose first and second derivatives are also bounded, and
( )
there exists δ ′ > 0, such that
1
∀x ∈ ∂ D, ∀y ∈ D, ∀n ∈ Nx , ⟨y − x, n⟩ + ⟨∇ Ψ (x) , n⟩ |y − x|2 ≥ 0.
δ′
We note the bound of Ψ and its derivatives by LΨ .
4
CONDITION (D). CONDITION (A) is satisfied and there exist constants C1 ≥ 0, C2 ∈]0, r0 [ such that
|x − y| ≤ (1 + γ C1 )|x − y|
holds for any x, y ∈ Rd with |x − x| ≤ C2 , |y − y| ≤ C2 , where γ = max{|x − x|, |y − y|}.
Let T > 0, we need the following notations:
• C [0, T ] denotes the space of continuous functions ψ : [0, T ] −→ Rd .

• C0 [0, T ] denotes the space of continuous functions ψ : [0, T ] −→ Rd such that ψ0 = 0.
• C [0, T ] denotes the space of continuous functions Φ : [0, T ] −→ D.
• V [0, T ] denotes the space of functions φ : [0, T ] −→ Rd with bounded variation and φ0 = 0.
3.1. Deterministic Skorohod problem
We recall the solvability result of deterministic Skorohod problems in the domain D satisfying Conditions (A) and (B),
which could be found in Saisho (1987).
Assume that φ is a continuous function taking values in Rd and that φ is of bounded variation over each finite interval.
We denote by |φ|t the total variation of φ over [0, t ], i.e.,
n
∑
|φ|t := sup |φtk − φtk−1 |.
0=t1 <t2 <···<tn =t , n∈N∗
k=1
For a continuous function w defined on [0, T ], T > 0, taking values in Rd with w0 = 0 (i.e. w ∈ C0 [0, T ]) and for
x0 ∈ D, we consider the following Skorohod equation:
ξt = x0 + wt + φt , t ∈ [0, T ] . (3.1)
We call a couple of functions (ξ , φ) solution of (3.1), if it satisfies (3.1) and the following conditions:
• The function ξ is continuous and takes values in D (i.e. ξ ∈ C [0, T ]),

• The function φ is continuous and takes values in Rd with φ0 = 0. Moreover, it is of bounded variation over [0, T ]
(i.e. φ ∈ V [0, T ]) and for all t ∈ [0, T ],
∫ t ∫ t
φt = nr d |φ|r , |φ|t = 1{ξr ∈∂ D} d |φ|r ,
0 0
where nr ∈ Nξr , if ξr ∈ ∂ D.
Theorem 3.1 (Saisho, 1987, Theorem 4.1). Suppose that the domain D ⊂ Rd satisfies Conditions (A) and (B). Then there exists
a unique solution (ξ , φ) for the deterministic Skorohod problem (3.1).
Let w, w ′ ∈ C0 [0, T ] and (ξ , φ), ξ ′ , φ ′ be the unique solutions of the Skorohod equations ξ = x0 + w + φ and
( )
ξ = x0 + w′ + φ ′ , respectively. Then we have the following result.
′
Theorem 3.2 (Saisho, 1987, Proposition 4.1). Under Conditions (B) and (D), we have for each T > 0,
|ξt − ξt′ | ≤ C ⏐w − w′ ⏐t , 0 ≤ t ≤ T ,
⏐ ⏐
where C > 0 is a constant depending only on the constants C1 , C2 , r0 , β, δ, T .
3.2. Reflected G-Brownian motion and reflected G-SDE
We replace the deterministic function w in the Skorohod problem (3.1) by the G-Brownian motion B and consider the
following equation in the ‘‘quasi-sure’’ sense:
Xt = x0 + Bt + Kt , x0 ∈ D, 0 ≤ t ≤ T . (3.2)
A couple of processes (X , K ) solves the Skorohod problem for the G-Brownian motion (3.2), if there exists a polar set A,
such that
• The processes X et K belong to MG2 [0, T ] ; Rd , and for all ω ∈ Ac ,

( )
Xt (ω) = x0 + Bt (ω) + Kt (ω), 0 ≤ t ≤ T ;
• For all ω ∈ Ac , X (ω) is continuous and takes values in D;

5
• For all ω ∈ Ac , K (ω) is continuous and takes values in Rd with K0 (ω) = 0. Moreover, K (ω) is of bounded variation
over [0, T ] and for all t ∈ [0, T ],
∫ t ∫ t
Kt (ω) = nr (ω) d |K |r (ω) , |K |t (ω) = 1{Xr (ω)∈∂ D} d |K |r (ω) ,
0 0
where nr (ω) ∈ NXr (ω) , if Xr (ω) ∈ ∂ D.
Theorem 3.3 (Lin and Hima, 2019, Theorem 3.1). Suppose that the domain D ⊂ Rd satisfies Conditions (A) and (B). Then there
exists a couple (X , K ) ∈ MG2 ([0, T ]; Rd ) × MG2 ([0, T ]; Rd ) which solves the Skorohod problem (3.2) whenever x0 ∈ D. Moreover,
if the problem (3.2) admits two solutions (X , K ) and (X ′ , K ′ ), then the exists a polar set Ã, such that for all ω ∈ Ãc ,
X (ω ) = X ′ (ω ) and K (ω) = K ′ (ω), 0 ≤ t ≤ T .
On the other hand, we consider the following reflected stochastic differential equations driven by G-Brownian motion,
which is formulated as (we use Einstein convention): for 0 ≤ t ≤ T ,
∫ t ∫ t ∫ t
Xt = x0 + b (Xr ) dr + hij (Xr ) d⟨Bi , Bj ⟩r + σj (Xr ) dBjr + Kt , q.s. (3.3)
0 0 0
where b, h and σ satisfy the following assumptions:
(A1) b, hij , σj : D −→ Rd , i, j = 1, 2, . . . , d are deterministic functions.
(A2) b, h, σ are bounded by L0 and uniformly L0 -Lipschitz, i.e., there exists a constant L0 > 0 such that,
|b(x) − b(y)| + |h(x) − h(y)| + |σ (x) − σ (y)| ≤ L0 |x − y|, ∀x, y ∈ D,

where | · | denotes the Hilbert–Schmidt norm for matrices.
A couple of processes (X , K ) solves the Skorohod stochastic differential Eq. (3.3), if there exists a polar set A, such that
• The processes X et K belong to MG2 [0, T ] ; Rd and satisfies (3.3);

( )
• For all ω ∈ Ac , X (ω) takes values in D;

• For all ω ∈ Ac , K (ω) takes values in Rd with K0 (ω) = 0. Moreover, K (ω) is of bounded variation over [0, T ] and for
all t ∈ [0, T ],
∫ t ∫ t
Kt (ω) = nr (ω) d |K |r (ω) , |K |t (ω) = 1{Xr (ω)∈∂ D} d |K |r (ω) ,
0 0
where nr (ω) ∈ NXr (ω) , if Xr (ω) ∈ ∂ D.
Theorem 3.4 (Lin and Hima, 2019, Theorem 3.3). Suppose that the domain D ⊂ Rd satisfies Conditions (A) and (B). Then there
exists a unique couple (X , K ) ∈ MG2 ([0, T ]; Rd ) × MG2 ([0, T ]; Rd ) which solves the Skorohod stochastic differential Eq. (3.3)
whenever x0 ∈ D and the coefficients b, h, σ satisfy Assumptions (A1) and (A2).
Now, in order to establish the main result of this work, replace h and σ by ε h and εσ respectively in (3.3) and let (Xtε )t ≥0
be the first component of the solution of the corresponding Skorohod stochastic differential equation. In the following
section, we will show {X ε }ε>0 satisfies a large deviation principle.
4. Large deviations
We recall the following notations on large deviations under a sublinear expectation.

Let (χ , d) be a Polish space. Let (U ε , ε > 0) be a family of measurable maps from Ω into (χ, d) and let δ (ε ), ε > 0 be
a positive function satisfying δ (ε ) → 0 as ε → 0.
A nonnegative function I on χ is called to be (good) rate function if {x : I (x) ≤ α} (its level set) is (compact) closed
for {all 0 ≤ α < } ∞.
Ĉ (U ε ∈ ·) ε>0 is said to satisfy large deviation principle with speed δ (ε ) and with rate function I if for any measurable
closed subset F ⊂ χ ,
lim sup δ (ε ) log Ĉ (U ε ∈ F ) ≤ − inf I (x),

ε→0 x∈ F
and for any measurable open subset O ⊂ χ ,
lim inf δ (ε ) log Ĉ (U ε ∈ O) ≥ − inf I (x).

ε→0 x∈O
6
4.1. Large deviations for G-SDEs
In Gao and Jiang (2010), for any ε > 0, the authors considered the following random perturbation SDEs driven by
d-dimensional G-Brownian motion B
∫ t ∫ t ∫ t
x,ε ε
Xt =x+ b (Xrx,ε )dr +ε h ε
(Xrx,ε )d
⟨B, B⟩r /ε + ε σ ε (Xrx,ε )dBr /ε
0 0 0
where ⟨B, B⟩ is treated as a d × d-dimensional vector,

bε = (bε1 , . . . , bεn )τ : Rn −→ Rn , σ ε = (σiε,j ) : Rn −→ Rn×d
and hε : Rn −→ Rn×d .
2
Consider the following conditions:
(H1) bε , σ ε and hε are uniformly bounded;
(H2) bε , σ ε and hε are uniformly Lipschitz continuous;
(H3) bε , σ ε and hε converge uniformly to b := b0 , σ := σ 0 and h := h0 respectively.
Define
{
Hd := f ∈ C0 ([0, T ], Rd ); f is absolutely continuous and
∫ T }
2
∥f ∥2H := |f ′ (r)| dr < +∞ ,
0
{ g(t) − g(s) }
A := g ∈ C0 ([0, T ], R d×d
): ∈ Σ for all t > s
t −s
{ ∫ t
= g= g (r)dr ; g : [0, T ] −→ Rd×d Borel measurable and
′ ′
0
}
g ′ (t) ∈ Σ for all t ∈ [0, T ] .
We recall the following result of a joint large deviation principle for G-Brownian motion and its quadratic variation
process.
Theorem 4.1 (Gao and Jiang, 2010, Theorem 2.4). Ĉ ε Bt /ε |t ∈[0,T ] ∈ · satisfies large deviation principle with speed ε and
{ ( )}
ε>0
with rate function
{ ∫
T
1
2 0
infQ∈Σ ⟨f ′ (r), Q−1 f ′ (r)⟩dr , if f ∈ Hd ,
IB (f ) =
+∞, otherwise.
and Ĉ (ε Bt /ε , ε⟨B⟩t /ε ) |t ∈[0,T ] ∈ · ε>0 satisfies large deviation principle with speed ε and with rate function
{ ( )}
{ ∫
T
1
⟨f ′ (r), (g ′ (r))−1 f ′ (r)⟩dr , if (f , g) ∈ Hd × A,
J(f , g) = 2 0
+∞, otherwise.
For any (f , g) ∈ Hd × A, let Ψ (f , g) ∈ C [0, T ] be the unique solution of the following ordinary differential equation
(ODE in short)
∫ t ∫ t
Ψ (f , g)(t) = x + b(Ψ (f , g)(r))dr + σ (Ψ (f , g)(r))f ′ (r)dr
0 0
∫ t
+ h(Ψ (f , g)(r))g ′ (r)dr .
0
By Gao and Jiang (2010), we have the following results.
Theorem 4.2( (Gao and Jiang, 2010, ) (Theorem 3.2).. Let (H1), ) (H2) and (H3) hold. Then for any closed subset F and any open
subset O in C0 ([0, T ], Rd ), ∥ · ∥ × C0 ([0, T ], Rd×d ), ∥ · ∥ × (C0 ([0, T ], Rn ), ∥ · ∥),
x,ε
lim sup ε log Ĉ (ε Bt /ε , ε⟨B⟩t /ε , Xt Î(f , g , ψ ),
( )
− x) |t ∈[0,T ] ∈ F ≤ − inf
ε→0 (f ,g ,ψ )∈F
and
x,ε
lim inf ε log Ĉ (ε Bt /ε , ε⟨B⟩t /ε , Xt Î(f , g , ψ ),
( )
− x) |t ∈[0,T ] ∈ O ≥ − inf
ε→0 (f ,g ,ψ )∈O
7
where
J(f , g), if (f , g) ∈ Hd × A, x + ψ = Ψ (f , g)
{
Î(f , g , ψ ) =
+∞, otherwise.
Corollary 4.3. Let (H1), (H2) and (H3) hold. Then for any closed subset F and any open subset O in C0 ([0, T ], Rn ),
x,ε
lim sup ε log Ĉ (Xt − x) |t ∈[0,T ] ∈ F ≤ − inf Λ(ψ ),
( )
ε→0 ψ∈F
and
x,ε
lim inf ε log Ĉ (Xt − x) |t ∈[0,T ] ∈ O ≥ − inf Λ(ψ ),
( )
ε→0 ψ∈O
where
{ }
Λ(ψ ) = inf J(f , g) : x + ψ = Ψ (f , g) .
4.2. Large deviations for reflected G-SDE
We consider the following Reflected SDE (RGSDE in short): for every 0 ≤ t ≤ T , x0 ∈ D,

⎧
ε
∫ t ( ε) ∫ t ( ε) ∫ t ( ε) ε
⎨ Xt = x∫0 + 0 b Xr dr + ε 0 ∫h Xr d⟨B, B⟩r + ε 0 σ Xr dBr + Kt
⎪
ε t ε ε t ε
K = 0 nr d |K |r , |K |t = 0 1{Xrε ∈∂ D} d |K |r (4.1)
⎪ t
nr ∈ NXrε , if Xrε ∈ ∂ D
⎩
It follows from Lin and Hima (2019) that, under the assumptions (A1) and (A2), the RGSDE (4.1) has a unique solution
(X ε , K ε ).
The objective of this work is to prove that the solution (X ε , K ε ) of RGSDE (4.1) converges, as ε goes to 0, to (ϕ, K)
solution of the following deterministic Skorohod equation:
∫t
⎨ ϕt = x∫0 + 0 b (ϕr ) dr + Kt ∫
⎧
t t
Kt = 0 nr d |K|r , |K|t = 0 1{ϕr ∈∂ D} d |K|r (4.2)
nr ∈ Nϕr , if ϕr ∈ ∂ D
⎩
and {X ε }ε>0 satisfies a large deviation principle. Note that ϕ is a solution of the following problem: a particle starts initially
at x0 ∈ D. It moves according to the velocity field b(x), x ∈ D. Whenever it reaches the boundary, it bounces back normally.
Let ψ ∈ C0 [0, T ], Φ ∈ C [0, T ] and ρ ∈ V [0, T ] such that
⎨ Φt = ∫x0 + ψt + ρt
⎧
t ∫t
ρt = 0 nr d |ρ|r , |ρ|t = 0 1{Φr ∈∂ D} d |ρ|r
nr ∈ NΦr , if Φr ∈ ∂ D
⎩
Let Φ = F (x0 + ψ ). It is known that F is continuous (see Lions and Sznitman, 1984; Saisho, 1987).
For x0 ∈ D and ψ̃ = x0 + ψ a continuous function defined on [0, T ], taking values in Rd with ψ̃0 = x0 , define two
operators:
Γ ψ̃ := ψ̃ + ρ and Υ ψ̃ := ρ.
By the reflection principle, the solution X ε of RGSDE (4.1) is given by
Xtε = (Γ Y ε )(t) and Ktε = (Υ Y ε )(t), 0 ≤ t ≤ T , (4.3)
where Y ε is a solution of the following G-SDE:
∫ t ∫ t ∫ t
Ytε = x0 + b ((Γ Y ε )(r)) dr + ε h ((Γ Y ε )(r)) d⟨B, B⟩r + ε σ ((Γ Y ε )(r)) dBr .
0 0 0
√
Set B̃t = √1ε Bt ε . Thanks to Remark 2.1, B̃ is also a G-Brownian motion. Then, we have Bt = εB̃t /ε , ⟨B, B⟩t = ε⟨B̃, B̃⟩t /ε .
Therefore, Y ε is a solution of the following G-SDE:
∫ t ∫ t ∫ t
Ytε = x0 + bε ((Γ Y ε )(r)) dr + ε hε ((Γ Y ε )(r)) d⟨B̃, B̃⟩r /ε + ε σ ε ((Γ Y ε )(r)) dB̃r /ε
0 0 0
√
where bε := b, hε := ε h and σ ε := εσ .
For any (f , g) ∈ Hd × A, let Ψ̂ (f , g) ∈ C [0, T ] be the unique solution of the following ODE
∫ t
Ψ̂ (f , g)(t) = x0 + b(Ψ̂ (f , g)(r))dr + Kt ,
0
8
where K have already been defined in (4.2). Similar as (4.3), Ψ̂ (f , g) can also be written as
Ψ̂ (f , g)(t) = (Γ Ψ̃ (f , g))(t) and Kt = (Υ Ψ̃ (f , g))(t), 0 ≤ t ≤ T , (4.4)
where Ψ̃ (f , g) satisfies
∫ t
Ψ̃ (f , g)(t) = x0 + b[(Γ Ψ̃ (f , g))(r)]dr .
0
If we define
B̃εt = B̃t − f (ε t)/ε, c ε (r , x) = bε (x) + σ ε (x)f ′ (r) + hε (x)g ′ (r), c(r , x) = b(x).
Then
∫ t ∫ t
ε ε ε
Y t = x0 + c (r , (Γ Y )(r)) dr + ε σ ε ((Γ Y ε )(r)) dB̃εr /ε
0 0
∫ t
+ hε ((Γ Y ε )(r)) d(ε⟨B̃⟩r /ε − g(r))
0
and
∫ t
Ψ̃ (f , g)(t) = x0 + c(r , (Γ Ψ̃ (f , g))(r))dr .
0
For m ≥ 1, set tk = kT /m, k = 0, 1, . . . , m. For a function ψ̃ , put
(Γ ψ̃ )(m) (t) = (Γ ψ̃ )(tk ), t ∈ [tk , tk+1 [, k = 0, 1, . . . , m.
ε,m
Yt = Ytεk , t ∈ [tk , tk+1 [, k = 0, 1, . . . , m.
From now on, unless otherwise stated, we shall use C to denote a constant depending on C1 , C2 , r0 , β, δ, T whose value
may vary from formula to formula.
From Theorem 3.2, for ψ̃ 1 , ψ̃ 2 ∈ C [0, T ] taking values in Rd with ψ̃01 = ψ̃02 = x0 and t ∈ [0, T ], we have
sup |(Γ ψ̃ 1 )(r) − (Γ ψ̃ 2 )(r)| ≤ C sup |ψ̃r1 − ψ̃r2 |. (4.5)

0≤r ≤t 0≤r ≤t
Then, by (4.3) and (4.4),
sup |Xtε − Ψ̂ (f , g)(t)| + sup |Ktε − Kt | ≤ C sup |Ytε − Ψ̃ (f , g)(t)|. (4.6)

0≤t ≤T 0≤t ≤T 0≤t ≤T
We denote by ∥ψ∥ := sup0≤t ≤T |ψt | for any function ψ on [0, T ].

We need the following preliminaries results.
Lemma 4.4. For any η > 0,
lim sup lim sup ε log Ĉ ∥Y ε − Y ε,m ∥ > η = −∞.

( )
m→∞ ε→0
Proof. The proof is similar to that of Lemma 3.1 of Gao and Jiang (2010), so we omit it. □
Lemma 4.5. For any a > 0 and η > 0,
lim sup lim sup ε log sup

µ→0 ε→0 ∥ f ∥ 2 ≤a
H
( ∫ ·  )
ε ε ε 
ε σ ( Γ ) > η, µ
 
Ĉ 
 ( Y )(r) d B̃r /ε  ∥ε B̃·/ε − f ∥ ≤ = −∞.
0
Proof. The proof is similar to that of Lemma 3.2 of Gao and Jiang (2010), so we omit it. □
Lemma 4.6. For any a > 0 and η > 0,

µ,ν→0 ε→0 ∥f ∥2H ≤a,g ∈A
Ĉ ∥Y ε − Ψ̃ (f , g)∥ > η, ∥ε B̃·/ε − f ∥ ≤ µ, ∥ε⟨B̃⟩·/ε − g ∥ ≤ ν = −∞.

( )
9
Proof. For any (f , g) ∈ Hd × A with ∥f ∥2H ≤ a,

∫ t
ε
Yt − Ψ̃ (f , g)(t) = [c ε (r , (Γ Y ε )(r)) − c(r , (Γ Ψ̃ (f , g))(r))]dr
0
∫ t
+ε σ ε ((Γ Y ε )(r)) dB̃εr /ε
0
∫ t
+ hε ((Γ Y ε )(r)) d(ε⟨B̃⟩r /ε − g(r)).
0
∫T √
Then by (A1), (A2), (4.5) and 0 (|f ′ (r)| + |g ′ (r)|)dr ≤ ( Ta + σ 2 T ), there exists a constant C > 0 such that for all t ∈ [0, T ]
and all (f , g) ∈ Hd × A with ∥f ∥2H ≤ a,
⏐∫ t ⏐ ∫ t
⏐ [c ε (r , (Γ Y ε )(r)) − c(r , (Γ Ψ̃ (f , g))(r))]dr ⏐ ≤ C sup |Ysε − Ψ̃ (f , g)(s)|dr
⏐ ⏐
⏐ ⏐
0 0 0≤s≤r
√ √
+ ε L0 ( Ta + σ 2 T )
and
⏐∫ t ⏐
⏐ hε ((Γ Y ε )(r)) d(ε⟨B̃⟩r /ε − g(r))⏐ ≤ L0 ∥ε⟨B̃⟩·/ε − g ∥.
⏐ ⏐
⏐ ⏐
0
Therefore,
√ √
sup |Ysε − Ψ̃ (f , g)(s)| ≤ L0 ∥ε⟨B̃⟩·/ε − g ∥ + ε L0 ( Ta + σ 2 T )
0≤s≤t
 ∫ · 
ε ε ε 
+ ε σ ((Γ Y )(r)) dB̃r /ε 
 

0
∫ t
+C sup |Ysε − Ψ̃ (f , g)(s)|dr
0 0≤s≤r
which implies from Gronwall’s inequality

( √ √ )
∥Y ε − Ψ̃ (f , g)∥ ≤ L0 ∥ε⟨B̃⟩·/ε − g ∥ + ε L0 ( Ta + σ 2 T ) × eC
 ∫ · 
ε ε ε 
ε σ ( Γ )
  C
+ ( Y )(r) d B̃r /ε  × e
0
√ √
Choose ε0 > 0 and ν0 > 0 such that for all 0 < ε ≤ ε0 and 0 < ν ≤ ν0 , L0 ν eC < η/3 and εL0 ( Ta + σ 2 T )eC < η/3.
Then
{ ε
∥Y − Ψ̃ (f , g)∥ > η, ∥ε B̃·/ε − f ∥ ≤ µ, ∥ε⟨B̃⟩·/ε − g ∥ ≤ ν
}
{ ∫ ·  }
ε ε ε 
ε σ ( Γ ) > η/ , µ ,
  −C
⊂  ( Y )(r) dB̃ r /ε  e 3 ∥ε B̃·/ε − f ∥ ≤
0
which allows to conclude by Lemma 4.5. □
Proposition 4.7. For any a > 0 and η > 0,

µ,ν→0 ε→0 ∥f ∥2H ≤a,g ∈A
Ĉ ∥X ε − Ψ̂ (f , g)∥ + ∥K ε − K∥ > η, ∥ε B·/ε − f ∥ ≤ µ, ∥ε⟨B⟩·/ε − g ∥ ≤ ν = −∞.

( )
Proof. In view of (4.6), the proof follows from Lemma 4.6. □

We have the following result of large deviations:
(A2) hold.) Then for any closed subset F and any open subset O in C0 ([0, T ], Rd ), ∥ · ∥ ×
( )
(Theorem 4.8. Let (A1) and
C0 ([0, T ], Rd×d ), ∥ · ∥ × C [0, T ], ∥ · ∥ ,
) (
lim sup ε log Ĉ (ε Bt /ε , ε⟨B⟩t /ε , Xtε − x0 ) |t ∈[0,T ] ∈ F ≤ − inf Î(f , g , ψ ),

( )
ε→0 (f ,g ,ψ )∈F
and
lim inf ε log Ĉ (ε Bt /ε , ε⟨B⟩t /ε , Xtε − x0 ) |t ∈[0,T ] ∈ O ≥ − Î(f , g , ψ ),
( )
inf
ε→0 (f ,g ,ψ )∈O
10
where
J(f , g), if (f , g) ∈ Hd × A, x0 + ψ = Ψ̂ (f , g)
{
Î(f , g , ψ ) =
+∞, otherwise.
Proof. By using the same argument used in the proof of Theorem 3.2 of Gao and Jiang (2010), the proof follows from
Proposition 4.7. □
We immediately have the following result.
Corollary 4.9. Let (A1) and (A2) hold. Then for any closed subset F and any open subset O in C [0, T ],
lim sup ε log Ĉ (Xtε − x0 ) |t ∈[0,T ] ∈ F ≤ − inf Λ(ψ ),

( )
ε→0 ψ∈F
and
lim inf ε log Ĉ (Xtε − x0 ) |t ∈[0,T ] ∈ O ≥ − inf Λ(ψ ),

( )
ε→0 ψ∈O
where
{ }
Λ(ψ ) = inf J(f , g) : x0 + ψ = Ψ̂ (f , g) .
Now we prove that the solution (X ε , K ε ) of RGSDE (4.1) converges, as ε goes to 0, to the solution (ϕ, K) of deterministic
Skorohod equation (4.2).
Theorem 4.10. Let (A1) and (A2) hold. Suppose that the domain D ⊂ Rd satisfies Conditions (A), (B) and (C). For any
ε ∈ (0, 1], there exists a constant C > 0 that depends on Σ , d, δ ′ , LΨ and L0 , independent of ε , such that
[ ] [ ]
Ê sup |Xtε − ϕt |4 + Ê sup |Ktε − Kt |4 ≤ C ε 4 .
0 ≤t ≤T 0≤t ≤T
Proof. By Proposition 4.4 of Lin and Hima (2019), we have

[ ] [ ]
Ê sup |Xsε − ϕs |4 + Ê sup |Ksε − Ks |4
0≤s≤t 0≤s≤t
∫ t( [ ] [ ])
ε
≤C Ê sup |Xs − ϕs | + Ê sup |εσ (ϕs )|
4 4
dr
0 0≤s≤r 0≤s≤r
∫ t( [ ] )
ε
≤C Ê sup |Xs − ϕs |4
dr + C ε 4
0 0≤s≤r
which implies the desired result by Gronwall’s inequality. □
Data availability
No data was used for the research described in the article.
Acknowledgements
The authors would like to thank the associate editor and the anonymous referee for their helpful comments and
suggestions that greatly improved the paper.
References
Cao, X., 2014. An upper bound of large deviations for capacities. Math. Probl. Eng. 2014, 1–6.
Chen, Z., Xiong, J., 2012. Large deviation principle for diffusion processes under a sublinear expectation. Sci. China Math. 55 (11), 2205–2216.
Dakaou, I., Hima, A.S., 2021. Large deviations for backward stochastic differential equations driven by G-Brownian motion. J. Theoret. Probab. 34 (2),
499–521.
Denis, L., Hu, M., Peng, S., 2011. Function spaces and capacity related to a sublinear expectation: application to G-brownian motion paths. Potential
Anal. 34, 139–161.
Freidlin, M.I., Wentzell, A.D., 1984. Random Perturbations of Dynamical Systems. Springer, Berlin.
Gao, F., 2009. Pathwise properties and homeomorphic flows for stochastic differential equations driven by G-brownian motion. Stochastic Process.
Appl. 119, 3356–3382.
Gao, F., Jiang, H., 2010. Large deviations for stochastic differential equations driven by G-brownian motion. Stochastic Process. Appl. 120, 2212–2240.
Hu, F., 2010. On Cramér’s theorem for capacities. C. R. Math. Acad Sci. Paris 348 (17–18), 1009–1013.
Hu, M., Peng, S., 2009. On representation theorem of G-expectations and paths of G-Brownian motion. Acta Math. Appl. Sin. Engl. Ser. 25, 539–546.
11
Lin, Y., 2013a. Équations différentielles stochastiques sous les espérances mathématiques non-linéaire et applications (Ph.D. thesis). Université de
Rennes 1.
Lin, Y., 2013b. Stochastic differential equations driven by G-Brownian motion with reflecting boundary conditions. Electron. J. Probab. 18 (9), 1–23.
Lin, Y., Hima, A.S., 2019. Reflected stochastic differential equations driven by G-Brownian motion in non-convex domains. Stoch. Dyn. 19 (03),
1950025.
Lions, P.L., Sznitman, A.S., 1984. Stochastic differential equations with reflecting boundary conditions. Comm. Pure Appl. Math. 37 (4), 511–537.
Liu, W., Zhang, Y., 2021. Large deviation principle for linear processes generated by real stationary sequences under the sub-linear expectation.
Comm. Statist. Theory Methods 1–15.
Peng, S., 2019. Nonlinear Expectations and Stochastic Calculus under Uncertainty with Robust CLT and G-Brownian Motion, vol. 95. Probability Theory
and Stochastic Modelling, Springer.
Saisho, Y., 1987. Stochastic differential equations for multi-dimensional domain with reflecting boundary. Probab. Theory Related Fields 74, 455–477.
Sheu, S.S., 1998. A large deviation principle of reflecting diffusions. Taiwanese J. Math. 2 (2), 251–256.
12

Large Deviation Principle For Reflected Stochastic Diffe 2023 Statistics P

Uploaded by

Copyright:

Available Formats

Large Deviation Principle For Reflected Stochastic Diffe 2023 Statistics P

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Large Deviation Principle For Reflected Stochastic Diffe 2023 Statistics P

Uploaded by

Copyright:

Available Formats

Statistics and Probability Letters 193 (2023) 109707

Contents lists available at ScienceDirect

Statistics and Probability Letters

Large deviation principle for Reflected Stochastic Differential

and satisfies a large deviation principle (LDP in short).

1. Monotonicity: if X ≥ Y , then Ê[X ] ≥ Ê[Y ];

Now we introduce the definition of G-normal distribution.

P is called a set that represents Ê.

which can be continuously extended to Q : HG1 (0, T ) −→ L1G (ΩT ). ∫·

3. Notations, assumptions, and formulation of the problem

Let D be a domain in Rd . For x ∈ ∂ D, we denote

where B(z , r) := {y ∈ Rd : |y − z | < r |}, for z ∈ Rd .

where ⟨·, ·⟩ denotes the usual inner product in Rd .

Let T > 0, we need the following notations:

• C [0, T ] denotes the space of continuous functions ψ : [0, T ] −→ Rd .

3.1. Deterministic Skorohod problem

• The function ξ is continuous and takes values in D (i.e. ξ ∈ C [0, T ]),

where C > 0 is a constant depending only on the constants C1 , C2 , r0 , β, δ, T .

3.2. Reflected G-Brownian motion and reflected G-SDE

• The processes X et K belong to MG2 [0, T ] ; Rd , and for all ω ∈ Ac ,

Xt (ω) = x0 + Bt (ω) + Kt (ω), 0 ≤ t ≤ T ;

• For all ω ∈ Ac , X (ω) is continuous and takes values in D;

where nr (ω) ∈ NXr (ω) , if Xr (ω) ∈ ∂ D.

X (ω ) = X ′ (ω ) and K (ω) = K ′ (ω), 0 ≤ t ≤ T .

where b, h and σ satisfy the following assumptions:

(A1) b, hij , σj : D −→ Rd , i, j = 1, 2, . . . , d are deterministic functions.

|b(x) − b(y)| + |h(x) − h(y)| + |σ (x) − σ (y)| ≤ L0 |x − y|, ∀x, y ∈ D,

• The processes X et K belong to MG2 [0, T ] ; Rd and satisfies (3.3);

• For all ω ∈ Ac , X (ω) takes values in D;

where nr (ω) ∈ NXr (ω) , if Xr (ω) ∈ ∂ D.

We recall the following notations on large deviations under a sublinear expectation.

lim sup δ (ε ) log Ĉ (U ε ∈ F ) ≤ − inf I (x),

and for any measurable open subset O ⊂ χ ,

lim inf δ (ε ) log Ĉ (U ε ∈ O) ≥ − inf I (x).

4.1. Large deviations for G-SDEs

where ⟨B, B⟩ is treated as a d × d-dimensional vector,

Consider the following conditions:

(H1) bε , σ ε and hε are uniformly bounded;

(H2) bε , σ ε and hε are uniformly Lipschitz continuous;

(H3) bε , σ ε and hε converge uniformly to b := b0 , σ := σ 0 and h := h0 respectively.

By Gao and Jiang (2010), we have the following results.

4.2. Large deviations for reflected G-SDE

We consider the following Reflected SDE (RGSDE in short): for every 0 ≤ t ≤ T , x0 ∈ D,

Ψ̂ (f , g)(t) = (Γ Ψ̃ (f , g))(t) and Kt = (Υ Ψ̃ (f , g))(t), 0 ≤ t ≤ T , (4.4)

For m ≥ 1, set tk = kT /m, k = 0, 1, . . . , m. For a function ψ̃ , put

(Γ ψ̃ )(m) (t) = (Γ ψ̃ )(tk ), t ∈ [tk , tk+1 [, k = 0, 1, . . . , m.

sup |(Γ ψ̃ 1 )(r) − (Γ ψ̃ 2 )(r)| ≤ C sup |ψ̃r1 − ψ̃r2 |. (4.5)

Then, by (4.3) and (4.4),

sup |Xtε − Ψ̂ (f , g)(t)| + sup |Ktε − Kt | ≤ C sup |Ytε − Ψ̃ (f , g)(t)|. (4.6)

We denote by ∥ψ∥ := sup0≤t ≤T |ψt | for any function ψ on [0, T ].

Lemma 4.4. For any η > 0,

lim sup lim sup ε log Ĉ ∥Y ε − Y ε,m ∥ > η = −∞.

Lemma 4.5. For any a > 0 and η > 0,

lim sup lim sup ε log sup

Lemma 4.6. For any a > 0 and η > 0,

lim sup lim sup ε log sup

Ĉ ∥Y ε − Ψ̃ (f , g)∥ > η, ∥ε B̃·/ε − f ∥ ≤ µ, ∥ε⟨B̃⟩·/ε − g ∥ ≤ ν = −∞.

Proof. For any (f , g) ∈ Hd × A with ∥f ∥2H ≤ a,