HAL Id: hal-01818668
https://hal.archives-ouvertes.fr/hal-01818668v2
Preprint submitted on 14 Sep 2018
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Random walk approximation of BSDEs with Hölder continuous terminal condition
Christel Geiss, Céline Labart, Antti Luoto
To cite this version:
Christel Geiss, Céline Labart, Antti Luoto. Random walk approximation of BSDEs with Hölder
continuous terminal condition. 2018. �hal-01818668v2�
Random walk approximation of BSDEs with Hölder continuous terminal condition
Christel Geiss
1, Céline Labart
2, Antti Luoto
3Abstract
In this paper we consider the random walk approximation of the solution of a Markovian BSDE whose terminal condition is a locally Hölder continuous function of the Brownian motion.
We state the rate of the L
2-convergence of the approximated solution to the true one. The proof relies in part on growth and smoothness properties of the solution u of the associated PDE. Here we improve existing results by showing some properties of the second derivative of u in space.
Keywords : Backward stochastic differential equations, numerical scheme, random walk approxi- mation, speed of convergence
MSC codes : 65C30 60H35 60G50 65G99
1 Introduction
Let (Ω, F , P ) be a complete probability space carrying the standard Brownian motion B = (B
t)
t≥0and assume (F
t)
t≥0is the augmented natural filtration. We consider the following backward stochastic differential equation (BSDE for short)
Y
s= g(B
T) + Z
Ts
f (r, B
r, Y
r, Z
r)dr − Z
Ts
Z
rdB
r, 0 ≤ s ≤ T, (1) where f is Lipschitz continuous and g is a locally α-Hölder continuous and polynomially bounded function (see (3)). In this paper we are interested in the L
2-convergence of the numerical approx- imation of (1) by using a random walk. First results dealing with the numerical approximation of BSDEs date back to the late 1990s. Bally (see [2]) was the first to consider this problem by introducing random discretization, namely the jump times of a Poisson process. In his PhD thesis, Chevance (see [17]) proposed the following discretization
y
k= E (y
k+1+ hf (y
k+1)|F
kn), k = n − 1, · · · , 0, n ∈ N
∗and proved the convergence of (Y
tn)
t:= (y
[t/h])
tto Y . At the same time, Coquet, Mackevičius and Mémin [18] proved the convergence of Y
nby using convergence of filtrations, still in the case of
1
Department of Mathematics and Statistics, P.O.Box 35 (MaD), FI-40014 University of Jyvaskyla, Finland christel.geiss@jyu.fi
2
Univ. Grenoble Alpes, Univ. Savoie Mont Blanc, CNRS, LAMA, 73000 Chambéry, France celine.labart@univ-smb.fr
3
Department of Mathematics and Statistics, P.O.Box 35 (MaD), FI-40014 University of Jyvaskyla, Finland
antti.k.luoto@student.jyu.fi
a generator independent from z. The general case (f depends on z, terminal condition ξ ∈ L
2) has been studied by Briand, Delyon and Mémin (see [5]). In that paper the authors define an approximated solution (Y
n, Z
n) based on random walk and prove weak convergence to (Y, Z) using convergence of filtrations. We also refer to [27], [29], [30], [31] for other numerical methods for BSDEs which use a random walk approach. The rate of convergence of this method was left as an open problem.
Introducing instead of random walk an approach based on the dynamic programming equation, Bouchard and Touzi in [8] and Zhang in [35] managed to establish a rate of convergence. However, to be fully implementable, this algorithm requires to have a good approximation of its associated conditional expectation. For this, various methods have been developed (see [24], [19], [15]). For- ward methods have also been introduced to approximate (1) : a branching diffusion method (see [26]), a multilevel Picard approximation (see [34]) and Wiener chaos expansion (see [7]). Many ex- tensions of (1) have also been considered : high order schemes (see [11], [10]), schemes for reflected BSDEs (see [3], [14]), for fully-coupled BSDEs (see [21], [9]), for quadratic BSDEs (see [13]), for BSDEs with jumps (see [23]) and for McKean-Vlasov BSDEs (see [1], [16], [12]).
From a numerical point of view, the random walk is of course not competitive with recent methods listed above. We emphasize that the aim of this paper is to give the convergence rate of the initial method based on random walk, which, to the best of our knowledge, has not been done so far.
As in [5], let us introduce the following approximation of B, based on a random walk:
B
tn=
√ h
[t/h]
X
i=1
ε
i, 0 ≤ t ≤ T,
where h =
Tn(n ∈ N
∗) and (ε
i)
i=1,2,...is a sequence of i.i.d. Rademacher random variables. Consider the following approximated solution (Y
n, Z
n) of (Y, Z)
Y
tnk
= g(B
Tn) + h
n−1
X
m=k
f (t
m+1, B
tnm, Y
tnm, Z
tnm) − √ h
n−1
X
m=k
Z
tnmε
m+1, 0 ≤ k ≤ n − 1. (2) The main result of our paper gives the rate of convergence in L
2-norm of Y
vn− Y
vand Z
vn− Z
vfor each v ∈ [0, T ) (see Theorem 3.1). Basically, we get that the L
2-norm of the error on Y is of order h
α4and the L
2-norm of the error on Z is of order
hα
√ 4
T−v
. The proof of this result is based on several ingredients. In particular, we need some estimates on the bound of the first and second derivatives of the solution of the PDE associated to the BSDE (1). We establish these bounds in the case of a forward backward SDE (FBSDE for short) whose terminal condition satisfies the Hölder continuity condition (3). This result extends Zhang [36, Theorem 3.2].
The rest of the paper is organized as follows. Section 2 introduces notations, assumptions and the representation for Z and Z
nbased on the Malliavin weights. Section 3 states the rate of convergence of the error on Y and Z in L
2-norm, which is the main result of the paper. Section 4 presents numerical simulations and Section 5 recalls some properties of Malliavin weights, of the regularity of solutions to FBSDEs with a locally Hölder continuous terminal condition function and states some properties of the solutions to the PDEs associated to these FBSDEs.
2 Preliminaries
This section is dedicated to notations, assumptions and the representation of Z and Z
nusing the
Malliavin weights.
Notation:
• G
k:= σ(ε
i: 1 ≤ i ≤ k) and G
0= {∅, Ω}. The associated discrete-time random walk (B
ntk)
nk=0is (G
k)
nk=0-adapted.
• k · k
p:= k · k
Lp(P)for p ≥ 1 and for p = 2 simply k · k. constant.
Assumption 2.1.
• g is locally Hölder continuous with order α ∈ (0, 1] and polynomially bounded (p
0≥ 0, C
g> 0) in the following sense
∀(x, y) ∈ R
2, |g(x) − g(y)| ≤ C
g(1 + |x|
p0+ |y|
p0)|x − y|
α. (3)
• The function [0, T ] × R
3: (t, x, y, z) 7→ f (t, x, y, z) satisfies
|f (t, x, y, z) − f (t
0, x
0, y
0, z
0)| ≤ L
f( √
t − t
0+ |x − x
0| + |y − y
0| + |z − z
0|). (4) Notice that (3) implies
|g(x)| ≤ K(1 + |x|
p0+1) =: Ψ(x). (5) In the rest of the paper, the study of the error (Y
n− Y, Z
n− Z) will either rely on (2) or on its integral version:
Y
sn= g(B
Tn) + Z
(s,T]
f (r, B
rn−, Y
rn−, Z
rn−)d[B
n, B
n]
r− Z
(s,T]
Z
rn−dB
nr, 0 ≤ s ≤ T, (6) where the backward equation (6) arises from (2) by setting Y
rn:= Y
tnmand Z
rn:= Z
tnmfor r ∈ [t
m, t
m+1). For n large enough, (6) has a unique solution (Y
n, Z
n), and (Y
tnm, Z
tnm)
n−1m=0is adapted to the filtration (G
m)
n−1m=0. Let us now introduce the Malliavin representations for Z and Z
n. They are the cornerstone of our study of the error on Z .
2.1 Representations for Z and Z
nWe will use the representation (see Ma and Zhang [28, Theorem 4.2]) Z
t= E
tg(B
T)N
Tt+
Z
T tf (s, B
s, Y
s, Z
s)N
stds
!
, 0 ≤ t ≤ T, (7)
where E
t[·] = E [·|F
t], and for all s ∈ (t, T ] we have N
st:= B
s− B
ts − t .
Lemma 2.2. Suppose that Assumption 2.1 holds. Then the process Z
ngiven by (6) has the repre- sentation
Z
tnk= E
kg(B
Tn) B
tnn− B
ntkt
n− t
k+ E
k
h
n−1
X
m=k+1
f (t
m+1, B
tnm, Y
tnm, Z
tnm) B
ntm− B
tnkt
m− t
k
(8)
for k = 0, 1, . . . , n − 1, where E
k[ · ] := E [ · |G
k].
Proof. We multiply equation (2) by ε
k+1and take the conditional expectation with respect to G
k. Since (Y
tnk
, Z
tnk
) is G
k-measurable, it holds for 0 ≤ k ≤ n − 1 that E
kY
tnkε
k+1= E
k(g(B
Tn)ε
k+1) + hE
k n−1X
m=k
f(t
m+1, B
ntm, Y
tnm, Z
tnm)ε
k+1!
− √ hE
kn−1
X
m=k
Z
tnmε
m+1ε
k+1!
=
√ h E
kg(B
Tn) B
tnn− B
tnkt
n− t
k+ h
3/2n−1
X
m=k+1
E
kf (t
m+1, B
tnm, Y
tnm, Z
tnm) B
tnm− B
ntkt
m− t
k− √
hZ
tnk, (9) where the l.h.s. is equal to zero. Indeed, for m ≥ k + 1, we have
E
k(Z
tnmε
m+1ε
k+1) = E
k(Z
tnmε
k+1E
mε
m+1) = 0, and for m = k it holds E
k(Z
tnkε
2k+1) = Z
tnk. Moreover, the fact that B
Tn= √
h P
n−1m=0ε
m+1, where (ε
m)
m=1,2...are i.i.d., yields
E
k(g(B
Tn)ε
k+1) = E
kg(B
Tn)
n−1
X
m=k
ε
k+1n − k
!
= E
kg(B
Tn)
n−1
X
m=k
ε
m+1n − k
!
= √ h E
kg(B
Tn) B
tnn− B
ntkt
n− t
k.
Similarly, for m ≥ k + 1, we get (using [5, Proposition 5.1], where it is stated that both Y
tnmand Z
tnmcan be represented as functions of t
mand B
ntm)
E
kf(t
m+1, B
ntm, Y
tnm, Z
tnm)ε
k+1=
√ h E
kf (t
m+1, B
tnm, Y
tnm, Z
tnm) B
tnm− B
tnkt
m− t
k.
It remains to divide (9) by √
h and rearrange.
3 Main result
This section is devoted to the main result of the paper: the rate of the L
2-convergence of (Y
n, Z
n) to (Y, Z). The proof will rely on the fact that the random walk B
ncan be constructed from the Brownian motion B by Skorohod embedding. Let τ
0:= 0 and define
τ
k:= inf {t > τ
k−1: |B
t− B
τk−1| =
√
h}, k ≥ 1.
Then (B
τk− B
τk−1)
∞k=1is a sequence of i.i.d. random variables with P (B
τk− B
τk−1= ± √
h) =
12, which means that √
hε
k=
dB
τk− B
τk−1. We will use this random walk for our approximation, i.e.
we will require
B
tn=
[t/h]
X
k=1
(B
τk− B
τk−1), 0 ≤ t ≤ T. (10)
Properties satisfied by τ
kand B
τkare stated in Lemma A.1. We will denote by E
τkthe conditional
expectation w.r.t. F
τk.
Theorem 3.1. Let Assumption 2.1 hold. If B
nsatisfies (10) then we have (for sufficiently large n) that
E |Y
v− Y
vn|
2≤ C
0h
α2for v ∈ [0, T ), E |Z
v− Z
vn|
2≤ C
0h
α2T − t
k+ C
1h
α2(T − v)
1−α21
v6=tkfor v ∈ [t
k, t
k+1), k = 0, ..., n − 1, where we have the dependencies C
0= C(T, p
0, L
f, C
g, C
5.3y, C
5.3z, K
f, c
5.4, α), C
1= C(T, p
0, C
5.3z, α) and K
f:= sup
0≤t≤T|f(t, 0, 0, 0)|.
Remark 3.2. Theorem 3.1 implies that sup
v∈[0,T)
E |Y
v− Y
vn|
2≤ C
0h
α2and E Z
T0
|Z
v− Z
vn|
2dv ≤ C(C
0, C
1, β) h
βfor β ∈ (0,
α2).
Proof of Theorem 3.1. Let u : [0, T ) × R → R be the solution of the PDE associated to (1). Since by Theorem 5.4
Y
s= u(s, B
s), Z
s= u
x(s, B
s), a.s.
we introduce
F(s, x) := f (s, x, u(s, x), u
x(s, x)),
so that F(s, B
s) = f (s, B
s, Y
s, Z
s). We first give some properties satisfied by F .
Lemma 3.3. If Assumption 2.1 holds then F is a Lipschitz continuous and polynomially bounded function in x :
|F (t, x
1) − F (t, x
2)| ≤ C(T, L
f, c
2,35.4)(1 + |x
1|
p0+1+ |x
2|
p0+1) |x
1− x
2| (T − t)
1−α2,
|F (t, x)| ≤ C(T, L
f, c
1,25.4, K
f) Ψ(x) (T − t)
1−α2, where Ψ(x) is given in (5).
Proof of Lemma 3.3. Thanks to the mean value theorem and Theorem 5.4-(ii-c) and (iii-b) we have for x
1, x
2∈ R that there exist ξ
1, ξ
2∈ [min{x
1, x
2}, max{x
1, x
2}] such that
|F (t, x
1) − F (t, x
2)| = |f (t, x
1, u(t, x
1), u
x(t, x
1)) − f (t, x
2, u(t, x
2), u
x(t, x
2))|
≤ L
f(|x
1− x
2| + |u(t, x
1) − u(t, x
2)| + |u
x(t, x
1) − u
x(t, x
2)|)
≤ L
f1 + c
25.4Ψ(ξ
1) (T − t)
1−α2+ c
35.4Ψ(ξ
2) (T − t)
1−α2!
|x
1− x
2|
≤ C(T, L
f, c
2,35.4)(1 + |x
1|
p0+1+ |x
2|
p0+1) |x
1− x
2|
(T − t)
1−α2.
The second inequality can be shown similarly.
For the estimate of E |Y
tk− Y
tnk|
2we will use (1) and (2): Since Y
tnkis F
τk-measurable we have kY
tk− Y
tnkk ≤ k E
tkg(B
T) − E
τkg(B
Tn)k
+
E
tkZ
T tkf (s, B
s, Y
s, Z
s)ds − h E
τkn−1
X
m=k
f (t
m+1, B
tnm, Y
tnm, Z
tnm)
. (11) We frequently express conditional expectations with the help of an independent copy of B denoted by ˜ B, for example E
tg(B
T) = ˜ E g(B
t+ ˜ B
T−t).
By (3) and Lemma A.1,
k E
tkg(B
T) − E
τkg(B
nT)k
2= E | E ˜ g(B
tk+ ˜ B
T−tk) − E ˜ g(B
τk+ ˜ B
τ˜n−k)|
2≤ ( E E ˜ (Ψ
1)
4)
12( E E ˜ |B
tk− B
τk+ ˜ B
T−tk− B ˜
˜τn−k|
4α)
12≤ C(C
g, T, p
0)(( E |B
tk− B
τk|
4α)
12+ ( E |B
T−tk− B
τn−k|
4α)
12)
≤ C(C
g, T, p
0)h
α2, (12)
where Ψ
1:= C
g(1 + |B
tk+ ˜ B
T−tk|
p0+ |B
τk+ ˜ B
˜τn−k|
p0). To estimate the other term in (11) we consider the decomposition
E
tkf (s, B
s, Y
s, Z
s) − E
τkf (t
m+1, B
tnm, Y
tnm, Z
tnm)
= ( E
tkf (s, B
s, Y
s, Z
s) − E
tkf (t
m, B
tm, Y
tm, Z
tm)) + ( E
tkF (t
m, B
tm) − E
τkF (t
m, B
τm))
+( E
τkF (t
m, B
τm) − E
τkF (t
m, B
tm)) + ( E
τkf (t
m, B
tm, Y
tm, Z
tm) − E
τkf (t
m+1, B
tnm, Y
tnm, Z
tnm))
=: D
1(s, m) + D
2(m) + ... + D
4(m) so that
E
tkZ
T tkf (s, B
s, Y
s, Z
s)ds − h E
τk n−1X
m=k
f (t
m+1, B
tnm, Y
tnm, Z
tnm)
≤
n−1
X
m=k
Z
tm+1tm
D
1(s, m)ds
+ h
4
X
i=2
kD
i(m)k
! .
For D
1we have by Theorem 5.3 that kD
1(s, m)k ≤ L
f( √
s − t
m+ kB
s− B
tmk + kY
s− Y
tmk + kZ
s− Z
tmk)
≤ C(T, L
f, C
5.3y, C
5.3z, p
0) (T − s)
α−22h
12, (13) where the last inequality follows from kB
s− B
tmk = √
s − t
m≤ h
12for s ∈ [t
m, t
m+1] and kY
s− Y
tmk + kZ
s− Z
tmk ≤ ( E Ψ(B
tm)
2)
12C
5.3yZ
stm
(T − r)
α−1dr
12
+ C
5.3zZ
stm
(T − r)
α−2dr
12
!
≤ C(T, C
5.3y, C
5.3z, p
0) √
s − t
m((T − s)
α−12+ (T − s)
α−22).
We bound D
2using Lemma 3.3 and Lemma A.1. Similar to (12) we conclude (setting Ψ
2:=
1 + |B
tk+ ˜ B
tm−k|
p0+1+ |B
τk+ ˜ B
τ˜m−k|
p0+1) that
kD
2(m)k = E | E
tkF (t
m, B
tm) − E
τkF (t
m, B
τm)|
21 2
≤ C(T, L
f, c
2,35.4)( E E ˜ Ψ
42)
141
(T − t
m)
1−α2(t
kh + t
m−kh)
14≤ C(T, p
0, L
f, c
2,35.4) 1
(T − t
m)
1−α2h
14. For D
3we apply again Lemma 3.3 and Lemma A.1,
kD
3(m)k ≤ kF (t
m, B
tm) − F (t
m, B
τm)k ≤ C(T, L
f, c
2,35.4) 1
(T − t
m)
1−α2kΨ
3|B
tm− B
τm|k
≤ C(T, p
0, L
f, c
2,35.4) 1
(T − t
m)
1−α2h
14, where Ψ
3:= 1 + |B
tm|
p0+1+ |B
τm|
p0+1. For the last term D
4we get
kD
4(m)k ≤ L
f(h
12+ kB
tm− B
tnmk + kY
tm− Y
tnmk + kZ
tm− Z
tnmk).
Finally, using the estimates for the terms D
1(s, m), D
2(m), ..., D
4(m) we arrive at kY
tk− Y
tnkk ≤ C(C
g, T, p
0)h
α4+ C(T, L
f, C
5.3y, C
5.3z, p
0) h
12Z
T tk(T − s)
α−22ds +C(T, p
0, L
f, c
2,35.4)h
14n−1
X
m=k
h
(T − t
m)
1−α2+ hL
fn−1
X
m=k
(kY
tm− Y
tnmk + kZ
tm− Z
tnmk)
≤ C(C
g, T, p
0, L
f, c
2,35.4, C
5.3y, C
5.3z)h
α4+ hL
fn−1
X
m=k
(kY
tm− Y
tnmk + kZ
tm− Z
tnmk). (14) For kZ
tk− Z
tnk
k we exploit the representations (7) and (8) and estimate kZ
tk− Z
tnkk ≤ 1
T − t
kk E
tkg(B
T)(B
T− B
tk) − E
τkg(B
τn)(B
τn− B
τk)k + E
tkZ
T tk+1f (s, B
s, Y
s, Z
s) B
s− B
tks − t
kds
!
− E
τk
h
n−1
X
m=k+1
f (t
m+1, B
tnm, Y
tnm, Z
tnm) B
ntm− B
tnkt
m− t
k
+ E
tkZ
tk+1tk
f (s, B
s, Y
s, Z
s) B
s− B
tks − t
kds . Then, similar to (12), we have for the terminal condition by Lemma A.1 that
k E
tk[g(B
T)(B
T− B
tk)] − E
τk[g(B
τn)(B
τn− B
τk)]k
= k E ˜ [g(B
tk+ ˜ B
T−tk) − g(B
tk)]( ˜ B
T−tk− B ˜
τ˜n−k) + ˜ E [g(B
tk+ ˜ B
T−tk) − g(B
τk+ ˜ B
τ˜n−k)] ˜ B
τ˜n−kk
≤ C(C
g, T, p
0)h
14(T − t
k)
α2+14+ C(C
g, T, p
0)h
α4(T − t
k)
12≤ C(C
g, T, p
0)h
α4(T − t
k)
12.
Here we have used that ˜ E [g(B
tk)( ˜ B
T−tk− B ˜
τ˜n−k)] = 0. The term ˜ E [g(B
tk+ ˜ B
T−tk)−g(B
tk)]( ˜ B
T−tk− B ˜
τ˜n−k) provides us with the factor (T − t
k)
α2((T − t
k)h)
14. For the next term of the estimate of kZ
tk− Z
tnk
k we use for s ∈ [t
m, t
m+1), where m ≥ k + 1, the decomposition E
tkf (s, B
s, Y
s, Z
s)(B
s− B
tk)
s − t
k− E
τkf (t
m+1, B
tnm, Y
tnm, Z
tnm)(B
tnm− B
tnk)
t
m− t
k= E
tkf (s, B
s, Y
s, Z
s)(B
s− B
tk)
s − t
k− E
tkf (t
m, B
tm, Y
tm, Z
tm)(B
tm− B
tk) t
m− t
k+ E
tkF(t
m, B
tm)(B
tm− B
tk)
t
m− t
k− E
τkF (t
m, B
τm)(B
τm− B
τk) t
m− t
k+ E
τk[F (t
m, B
τm) − F (t
m, B
tm)] B
τm− B
τkt
m− t
k+ E
τk[f (t
m, B
tm, Y
tm, Z
tm) − f (t
m+1, B
tnm, Y
tnm, Z
tnm)] B
tnm− B
tnk
t
m− t
k=: T
1(s, m) + T
2(m) + ... + T
4(m).
Then by the conditional Hölder inequality and by (13) as well as by Lemma 3.3 we have kT
1(s, m)k ≤ kD
1(s, m)k kB
s− B
tkk
s − t
k+ kf (t
m, B
tm, Y
tm, Z
tm)k
B
s− B
tks − t
k− B
tm− B
tkt
m− t
k≤ C(T, L
f, C
5.3y, C
5.3z, p
0) (T − s)
α−22h
12√ s − t
k+C(T, L
f, c
1,25.4, K
f) ( E Ψ(B
tm)
2)
12(T − t
m)
1−α2×
kB
s− B
tmk
s − t
k+ kB
tm− B
tkk
1
s − t
k− 1 t
m− t
k≤ C(T, L
f, K
f, C
5.3y, C
5.3z, c
1,25.4, p
0)(T − s)
α−22h
14(s − t
k)
34. Indeed,
kB
s− B
tmk
s − t
k+ kB
tm− B
tkk
1
s − t
k− 1 t
m− t
k≤
√ s − t
ms − t
k+
√ t
m− t
k(s − t
m)
(s − t
k)(t
m− t
k) ≤ C h
14(s − t
k)
34, where the last inequality follows from s− t
m≤ t
m+1−t
m= h and h ≤ t
m− t
k≤ s −t
k. We estimate T
2with the help of Lemma 3.3 and Lemma A.1 as follows :
kT
2(m)k ≤ k D b
2(m)k kB
tm− B
tkk
t
m− t
k+ kF (t
m, B
τm)k kB
tm−k− B
τm−kk t
m− t
k≤ C(T, p
0, L
f, K
f, c
5.4) 1 (T − t
m)
1−α2h
14(t
m− t
k)
34.
Here D b
2(m) := (˜ E |F (t
m, B
tk+ ˜ B
tm−k) − F (t
m, B
τk+ ˜ B
τ˜m−k)|
2)
12which can be estimated as D
2(m).
For T
3the conditional Hölder inequality and Lemma A.1 yield kT
3(m)k ≤ k D b
3(m)k
B
τm− B
τkt
m− t
k≤ C(T, p
0, L
f, c
2,35.4) 1 (T − t
m)
1−α2h
14(t
m− t
k)
12, where D b
3(m) := F (t
m, B
τm) − F (t
m, B
tm) is estimated as D
3(m). Finally,
kT
4(m)k ≤ L
f(h
12+ kB
tm− B
ntmk + kY
tm− Y
tnmk + kZ
tm− Z
tnmk) 1
√ t
m− t
k.
For the estimate of E
tkR
tk+1tk
f (s, B
s, Y
s, Z
s)
Bss−t−Btkk
ds one notices that by the conditional Hölder inequality,
k E
tkf (s, B
s, Y
s, Z
s)
Bss−t−Btkk
k = k E
tk[(f (s, B
s, Y
s, Z
s) − f (s, B
tk, Y
tk, Z
tk))
Bss−t−Btkk
]k
≤ kf(s, B
s, Y
s, Z
s) − f(s, B
tk, Y
tk, Z
tk)k 1
√ s − t
k≤ C(T, L
f, C
5.3y, C
5.3z, p
0) (T − s)
α−22h
12√ s − t
k, where the last inequality follows in the same way as in (13). Consequently, we have
kZ
tk− Z
tnkk ≤ C(C
g, T, p
0)
(T − t
k)
12h
α4+ C(T, L
f, K
f, C
5.3y, C
5.3z, c
1,25.4, p
0) Z
Ttk
ds
(T − s)
1−α2(s − t
k)
34h
14+C(T, p
0, L
f, K
f, c
5.4) h
n−1
X
m=k+1
1 (T − t
m)
1−α2h
14(t
m− t
k)
34+L
fh
n−1
X
m=k+1
(kB
tm− B
tnmk + kY
tm− Y
tnmk + kZ
tm− Z
tnmk) 1
√ t
m−k.
Lemma A.2 enables to bound the second and third term of the r.h.s. by C
h1 4
(T−tk)34−α2
B (
α2,
14), which is bounded by C
hα 4
(T−tk)12−α4
. Thus we get kZ
tk− Z
tnkk ≤ C
0h
α4(T − t
k)
12+ L
fh
n−1
X
m=k+1
(kY
tm− Y
tnmk + kZ
tm− Z
tnmk) 1
√ t
m−k.
Then we use (14) and the above estimate to get kY
tk− Y
tnkk + kZ
tk− Z
tnkk ≤ C
0h
α4(T − t
k)
12+ C(L
f) h
n−1
X
m=k+1
(kY
tm− Y
tnmk + kZ
tm− Z
tnmk) 1
√ t
m−k.
If this inequality is iterated, one gets a shape where the Gronwall lemma applies. Indeed, setting a
m:= (kY
tm− Y
tnmk + kZ
tm− Z
tnmk) one has to consider the double sum
n−1
X
m=k+1
n−1
X
l=m+1
a
lh
√ t
l−m
√ h t
m−k= h
n−1
X
l=k+1
l−1
X
m=k+1
√ h t
m−k√ t
l−m
a
l≤ Ch
n−1
X
l=k+1
a
l.
Consequently,
kY
tk− Y
tnkk + kZ
tk− Z
tnkk ≤ C
0h
α4(T − t
k)
12which gives the bound on the error on Z . Moreover, (14) yields
kY
tk− Y
tnk
k ≤ C
0h
α4. If v ∈ [t
k, t
k+1), we have by Theorem 5.3 that
kY
v− Y
vnk ≤ kY
v− Y
tkk + kY
tk− Y
tnk
k ≤ C(C
5.3y, T, p
0) Z
vtk
(T − r)
α−1dr
12+ kY
tk− Y
tnk