2020 Mathematics Subject Classification. 60L20, 60H10.
Introduction
Let X \mathbf{X} X be a p p p -rough path with 2 < p < 3 2 < p < 3 2 < p < 3 . On a filtered probability space carrying a Brownian motion W W W , we consider the following rough backward stochastic differential equation (rough BSDE)
Y t = ξ + ∫ t T f ( r , Y r , Z r ) d r + ∫ t T H ( Y r ) d X r − ∫ t T Z r d W r , t ∈ [ 0 , T ] . (1) Y_t = \xi+ \int_t^T f(r,Y_r,Z_r)\,\mathrm{d}r + \int_t^T H(Y_r)\,\mathrm{d}\mathbf{X}_r - \int_t^T Z_r\,\mathrm{d}W_r,\qquad t \in[0,T]. \tag*{(1)} Y t = ξ + ∫ t T f ( r , Y r , Z r ) d r + ∫ t T H ( Y r ) d X r − ∫ t T Z r d W r , t ∈ [ 0 , T ] . (1)
This equation was originally introduced by Diehl and Friz [7 ] . Their construction starts with smooth paths X n \mathbf{X}^n X n whose canonical rough path lifts X n \mathbf{X}^n X n converge to X \mathbf{X} X . The corresponding classical BSDEs are solved via a Doss–Sussmann transformation, and the solution to (1) is then defined as the limit of the resulting pairs ( Y n , Z n ) (Y^n,Z^n) ( Y n , Z n ) . Formally, the rough integral is interpreted as
∫ t T H ( Y r ) d X r : = lim n → ∞ ∫ t T H ( Y r n ) d X r n , (2) \int_t^T H(Y_r)\,\mathrm{d}\mathbf{X}_r := \lim_{n\to\infty}\int_t^T H(Y_r^n)\,\mathrm{d}\mathbf{X}_r^n, \tag*{(2)} ∫ t T H ( Y r ) d X r := n → ∞ lim ∫ t T H ( Y r n ) d X r n , (2)
where the limit is independent of the approximating sequence and gives a robust solution map in the rough path topology. It does not, however, provide an intrinsic definition of the integral in (1) . In particular, H ( Y ) H(Y) H ( Y ) is not treated as an integrand within a rough stochastic integration theory, and the dynamics of Y Y Y are characterized only through approximation.
A direct theory for (1) is available in the Young regime. Diehl and Zhang [8 ] consider drivers of finite q q q -variation with q < 2 q < 2 q < 2 and establish well-posedness via a fixed-point argument. In that setting, the integral involving Y Y Y is a pathwise Young integral and the equation is satisfied directly. Song et al. [27 , 28 ] further develop the nonlinear Young regime, providing a probabilistic interpretation of stochastic partial differential equations (SPDEs) driven by space-time noise in the Young regime. A general jump case is studied by Becherer and Sun [2 ] . For genuinely rough drivers, Liang and Tang [22 ] extend the approximation and flow approach of [7 ] to multidimensional BSDEs, under either a smallness condition or a diagonal structural assumption on the nonlinear rough coefficient. They also develop an intrinsic theory for multidimensional equations but with a linear rough drift. Their construction uses stochastic controlled rough paths, a p p p -variation rough stochastic integral, and the stochastic sewing lemma. Stochastic controlled rough paths and stochastic sewing also play a central role in recent developments for rough (forward) stochastic differential equations; see, for example, [11 , 1 ] . A reflected rough BSDE theory is introduced by Li et al. [21 ] , who also study the associated rough PDEs with obstacles, as well as optimal stopping and American option pricing problems.
The aim of the present paper is to solve (1) intrinsically with a nonlinear rough coefficient H ( Y ) H(Y) H ( Y ) appearing in the rough integral, without resorting to Doss–Sussmann transformation or defining the solution through smooth approximations. We show that the pair ( Y , − H ( Y ) ) (Y,-H(Y)) ( Y , − H ( Y )) belongs to an appropriate space of rough semimartingales (RSMs), originally introduced in [16 ] . Within this space, the integral against X \mathbf{X} X in (1) is defined intrinsically as a rough stochastic integral via compensated Riemann sums:
∫ t T H ( Y r ) d X r : = lim ∣ π ∣ → 0 ∑ [ u , v ] ∈ π ( H ( Y u ) δ X u , v − D H ( Y u ) H ( Y u ) X u , v ) , (3) \int_t^T H(Y_r)\,\mathrm{d}\mathbf{X}_r := \lim_{|\pi|\to0}\sum_{[u,v]\in\pi}\bigl(H(Y_u)\,\delta\mathbf{X}_{u,v}-DH(Y_u)H(Y_u)\,\mathbb{X}_{u,v}\bigr), \tag*{(3)} ∫ t T H ( Y r ) d X r := ∣ π ∣ → 0 lim [ u , v ] ∈ π ∑ ( H ( Y u ) δ X u , v − D H ( Y u ) H ( Y u ) X u , v ) , (3)
where the limit is taken in probability. Thus, the solution pair ( Y , Z ) (Y,Z) ( Y , Z ) satisfies (1) as a hybrid rough stochastic integral equation in the precise sense of Definition 4.1.
A notable feature of our work is that we are able to obtain all required estimates in the RSM setting. In particular, neither the construction of (3) nor the well-posedness proof uses the stochastic sewing lemma. This differs from the stochastic controlled rough path approach of Liang and Tang [22 ] and from the rough SDE theory developed in [11 , 1 ] . The present work therefore provides an original alternative to stochastic sewing for natural classes of equations that combine rough signals with (implicit) martingale components.
The distinction is relevant for BSDEs because the martingale integrand Z Z Z is part of the solution and is known only implicitly as a general predictable process. Stochastic controlled rough path methods require suitable time regularity of the martingale component in an L k L^k L k -scale. Such regularity is not available from the definition of the classical BSDE and must normally be established as part of the analysis. A rough semimartingale, by contrast, is specified through a Doob–Meyer-type decomposition and does not impose time regularity on its martingale part. Moreover, its definition is independent of a prescribed L k L^k L k -integrability exponent. This allows the notion of solution to be formulated before any integrability estimates are proved.
The intrinsic formulation also allows the comparison principle and stability estimates to be established directly from the equation. Strict comparison, for example, is not generally preserved under passage to the limit: approximation readily gives non-strict ordering, but strictness may be lost. Similarly, direct difference estimates are difficult when the rough integral is known only through the approximation (2) . Working in the RSM space, we prove a strict comparison theorem, an a priori estimate, existence and uniqueness, and local stability with respect to the terminal condition, the drift coefficient, and the driving rough path.
Our method also reduces the regularity required for the rough coefficient, and we only need
H ∈ C b γ , for some γ > p . (4) H \in C^\gamma_b, \qquad\text{for some } \gamma> p. \tag*{(4)} H ∈ C b γ , for some γ > p . (4)
This is the standard regularity threshold for the corresponding RDE. In contrast, the flow transformation approach in [7 ] requires the stronger condition H ∈ C b γ + 2 H \in C^{\gamma+2}_b H ∈ C b γ + 2 . The additional two derivatives are needed to control the transformed stochastic equation and are not intrinsic to the rough dynamics. Since no transformation is used in the present approach, the strict comparison theorem, the a priori estimate, and the well-posedness result all hold under the natural condition (4) .
We now describe the main estimates. Suppose first that Y Y Y is essentially bounded. To control the RSM norm of ( Y , − H ( Y ) ) (Y,-H(Y)) ( Y , − H ( Y )) , one needs estimates for the remainder in the rough expansion of H ( Y ) H(Y) H ( Y ) . In contrast to the deterministic RDE case, this remainder contains a martingale component, which can be controlled via a recent result on p p p -variation estimates for martingale transforms [16 ] . Its conditional version Lemma A.2 gives the required control of the mixed conditional moments of the p p p -variation norm; see (32) in Proposition 3.5. This is the main input in the composition estimates for rough semimartingales.
An essential bound for Y Y Y is obtained through comparison. The difference of two solutions satisfies a linear rough BSDE. We derive an explicit exponential representation of this linear equation and combine it with a John–Nirenberg inequality for VMOp -var ^{p\text{-var}} p -var processes. This yields the comparison theorem and, by comparison with suitable deterministic RDEs, the required uniform bound. The same representation also yields strict comparison under the corresponding strict ordering assumptions.
The remaining difficulty is the loss of integrability caused by nonlinear composition. As observed in the stochastic controlled rough path setting [11 ] , estimates for H ( Y ) − H ( Y ~ ) H(Y)-H(\widetilde{Y}) H ( Y ) − H ( Y ) naturally require a higher integrability exponent than the one initially available for Y − Y ~ Y-\widetilde{Y} Y − Y . We address this by proving an energy estimate for suitable increasing processes. It implies that the conditional norms ∥ ∥ ⋅ ∥ ∥ p -var ∥ ∞ ; k , ∞ \|\!\| \cdot\|\!\|_{p\text{-var}\,\|\,\infty;k,\infty} ∥ ∥ ⋅ ∥ ∥ p -var ∥ ∞ ; k , ∞ , defined by conditional moments of the p p p -variation norm, are equivalent for different values of k k k . These norms have the same self-improving feature as classical k k k -BMO norms, and we therefore refer to them as BMO-type norms. Their equivalence compensates for the integrability loss in the nonlinear composition estimate. This leads to a local contraction and, by iteration, to existence and uniqueness of the solution on the full time interval. The same estimates yield continuous dependence of the solution on the data.
Finally, we relate rough BSDEs to backward doubly stochastic differential equations (BDSDEs), famously introduced by Pardoux and Peng [24 ] , as a counterpart in the BSDE theory for SPDEs. To this end, we first introduce a backward causality property for processes with rough parameters to describe their measurability with respect to the backward filtration. Then we show that the solution to rough BSDEs has such a measurable version with respect to the σ \sigma σ -algebra generated by the forward Brownian motion and the backward rough path flow, via the measurable selection initiated by Dellacherie and Meyer [4 ] and recently developed by Denkert et al. [6 ] . Randomizing X X X by a Brownian rough path then produces a BDSDE, while the intrinsic rough integral identifies the corresponding backward stochastic integral. This, for the first time, provides a direct connection between the rough BSDE and BDSDE formulations without passing through smooth approximations of the additional Brownian signal. (Applications to semilinear rough and stochastic PDEs are also possible but will be pursued elsewhere.)
In conclusion, our contributions are threefold. Firstly, we introduce the intrinsic notion of solution to rough BSDEs in the framework of rough semimartingales, as well as the corresponding rough stochastic analysis (without reliance on - and complementary to - the work [11 ] ). Secondly, we establish the well-posedness and overcome the problem of degenerating integrability by establishing energy estimates for BMO-type processes. Thirdly, we connect rough BSDEs and BDSDEs via a novel direct randomization argument.
Orginzation of Paper. In Section 2.1, we introduce notations and assumptions. Section 2.2 recalls the RSM framework and the construction of rough stochastic integrals. In Section 3.1, we prove energy estimates for increasing processes and the equivalence of the BMO-type p p p -variation norms. Composition and integration estimates for RSMs are established in Section 3.2, and the RSM seminorm used in the stability analysis is introduced in Section 3.3. Section 4 contains the well-posedness theory. We first prove the comparison theorem in Section 4.1, derive the a p r i o r i a\ priori a p r i or i estimate in Section 4.2, and then establish existence, uniqueness, and stability under the condition (1.4) in Section 4.3. Section 5 studies the connection with BDSDEs, including backward causality in Section 5.1 and Brownian randomization in Section 5.2. Appendix A collects Lépinglé-type inequalities, conditional p p p -variation estimates for martingale transforms, and estimates for RDEs with unbounded drift. The symbolic index is provided in Appendix B.
Mathematical Roadmap. First and importantly, Lemma 3.2 implies that the q q q -variation of a continuous process is a BMO-process. Consequently, the conditional norms ∥ ⋅ ∥ q -var ∣ ∞ ; κ , ∞ \lVert\cdot\rVert_{q\text{-var}\mid\infty;\kappa,\infty} ∥ ⋅ ∥ q -var ∣ ∞ ; κ , ∞ are comparable for all k > 1 k>1 k > 1 . Together with the conditional martingale-transform estimate of Lemma A.2, this underlies two parallel lines of the argument. The first controls a single solution: Proposition 3.5 bounds the remainder of H ( Y ) H(Y) H ( Y ) , Corollary 3.8 converts this into a bound for ∫ H ( Y ) d X \int H(Y)\,dX ∫ H ( Y ) d X , and Propositions 4.7 and 4.8 then show that, once ∥ Y ∥ ∞ \lVert Y\rVert_\infty ∥ Y ∥ ∞ is controlled, both ∥ δ M Y ∥ B M O \lVert\delta M^Y\rVert_{\mathrm{BMO}} ∥ δ M Y ∥ BMO and the RSM seminorm are bounded. The missing bound for ∥ Y ∥ ∞ \lVert Y\rVert_\infty ∥ Y ∥ ∞ comes from a different direction: the comparison theorem (Theorem 4.3), proved via the exponential representation of the linearized equation and the John–Nirenberg inequality, compares Y Y Y with two deterministic RDEs and yields the a p r i o r i a\ priori a p r i or i estimate of Theorem 4.9. The second line of argument controls differences: Proposition 3.10, in which the equivalence of norms compensates for the integrability loss in the nonlinear composition, feeds into Lemma 3.17 and then into the contraction property of Proposition 4.16. Theorem 4.17 combines the contraction on small intervals with the a p r i o r i a\ priori a p r i or i estimate to obtain global well-posedness, and the same difference estimates give the stability result, Theorem 4.19. Section 5 relies on the preceding theory only for well-posedness, stability, and the RDE bound underlying the a p r i o r i a\ priori a p r i or i estimate.
Acknowledgment PKF and HZ acknowledge support from DFG CRC/TRR 388 “Rough Analysis, Stochastic Dynamics and Related Fields”, Projects A07, B04 and B05. Part of this work was carried out during a visit of the first author to Shandong University. JS is partially supported by National Natural Science Foundation of China (Grant Numbers 12521001, 12471142); JS and HZ are supported by the Fundamental Research Funds for the Central Universities. HZ is partially supported by NSF of Shandong (ZR2023MA026).
Preliminaries
Notations and Assumptions
We collect notations that will be used in this article. Let ( E , ∥ ⋅ ∥ E ) (E,\|\cdot\|_E) ( E , ∥ ⋅ ∥ E ) be a Banach space, and let V V V and K K K be Euclidean spaces. We denote by ∣ ⋅ ∣ |\cdot| ∣ ⋅ ∣ the Euclidean norm for both vectors and matrices. Some of the definitions introduced below depend on the underlying time interval or terminal time. For simplicity of notation, we omit this dependence when the interval is [ 0 , T ] [0,T] [ 0 , T ] (with terminal time T T T ) and no confusion arises.
Probability space: Fix a finite time horizon T > 0 T>0 T > 0 . For an integer d ≥ 1 d\ge1 d ≥ 1 , let { W t } t ∈ [ 0 , T ] \{W_t\}_{t\in[0,T]} { W t } t ∈ [ 0 , T ] be a d d d -dimensional standard Brownian motion on the probability space ( Ω , F , P ) (\Omega,\mathcal{F},\mathbb{P}) ( Ω , F , P ) , and let ( F t ) t ∈ [ 0 , T ] (\mathcal{F}_t)_{t\in[0,T]} ( F t ) t ∈ [ 0 , T ] be the augmented filtration generated by W W W . Denote the conditional expectation by E t [ ⋅ ] : = E [ ⋅ ∣ F t ] \mathbb{E}_t[\cdot]:=\mathbb{E}[\cdot\mid\mathcal{F}_t] E t [ ⋅ ] := E [ ⋅ ∣ F t ] .
Partition: For 0 ≤ s ≤ t ≤ T 0\le s\le t\le T 0 ≤ s ≤ t ≤ T , set Δ [ s , t ] : = { ( u , v ) : s ≤ u ≤ v ≤ t } \Delta_{[s,t]}:=\{(u,v):s\le u\le v\le t\} Δ [ s , t ] := {( u , v ) : s ≤ u ≤ v ≤ t } and Δ : = Δ [ 0 , T ] \Delta:=\Delta_{[0,T]} Δ := Δ [ 0 , T ] . Let P [ s , t ] \mathcal{P}_{[s,t]} P [ s , t ] denote the set of all partitions π = { [ t i , t i + 1 ] ; s = t 0 < t 1 < ⋯ < t n = t } \pi=\{[t_i,t_{i+1}];\ s=t_0<t_1<\cdots<t_n=t\} π = {[ t i , t i + 1 ] ; s = t 0 < t 1 < ⋯ < t n = t } of [ s , t ] [s,t] [ s , t ] , and define the mesh size of π \pi π by ∣ π ∣ : = max 0 ≤ i < n ( t i + 1 − t i ) |\pi|:=\max_{0\le i<n}(t_{i+1}-t_i) ∣ π ∣ := max 0 ≤ i < n ( t i + 1 − t i ) .
Control function: For a function w : Δ → R + w:\Delta\to\mathbb{R}^{+} w : Δ → R + , we call w w w a control function if w w w is continuous and satisfies that w ( s , u ) + w ( u , t ) ≤ w ( s , t ) w(s,u)+w(u,t)\le w(s,t) w ( s , u ) + w ( u , t ) ≤ w ( s , t ) for s ≤ u ≤ t s\le u\le t s ≤ u ≤ t . In particular, w ( s , s ) = 0 w(s,s)=0 w ( s , s ) = 0 for all s s s .
Linear map: We denote by L ( V , K ) \mathcal{L}(V,K) L ( V , K ) the space of linear maps from V V V to K K K , endowed with the operator norm ∥ F ∥ L ( V , K ) = sup x ∈ V , ∣ x ∣ ≤ 1 ∣ F ( x ) ∣ \|F\|_{\mathcal{L}(V,K)}=\sup_{x\in V,|x|\le1}|F(x)| ∥ F ∥ L ( V , K ) = sup x ∈ V , ∣ x ∣ ≤ 1 ∣ F ( x ) ∣ . We write V ⊗ K V\otimes K V ⊗ K for the (algebraic) tensor product of V V V and K K K , equipped with the projective norm. In finite dimensions, the spaces L ( V , K ) \mathcal{L}(V,K) L ( V , K ) and K ⊗ V K\otimes V K ⊗ V are canonically isomorphic, and the above norms are equivalent.
∥ f ∥ ∞ ; U : = ess sup r ∈ U ∥ f ( r ) ∥ E . \|f\|_{\infty;U}:=\operatorname*{ess\,sup}_{r\in U}\|f(r)\|_E. ∥ f ∥ ∞ ; U := r ∈ U ess sup ∥ f ( r ) ∥ E .
If U U U is clear from the context, we simply write ∥ f ∥ ∞ \|f\|_\infty ∥ f ∥ ∞ for ∥ f ∥ ∞ ; U \|f\|_{\infty;U} ∥ f ∥ ∞ ; U . Denote by C ( U , E ) C(U,E) C ( U , E ) the space of continuous functions f : U → E f:U\to E f : U → E . Denote by C b ( U , E ) C_b(U,E) C b ( U , E ) the space of functions f ∈ C ( U , E ) f\in C(U,E) f ∈ C ( U , E ) with finite uniform norm.
For ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ and k ≥ 1 k\ge1 k ≥ 1 , denote by L k ( [ s , t ] , E ) L^k([s,t],E) L k ([ s , t ] , E ) the space of measurable functions f : [ s , t ] → E f:[s,t]\to E f : [ s , t ] → E such that
∥ f ∥ L k ( [ s , t ] ) : = { ∫ s t ∣ f r ∣ k d r } 1 / k < ∞ . \|f\|_{L^k([s,t])}:=\left\{\int_s^t |f_r|^k\,\mathrm{d}r\right\}^{1/k}<\infty. ∥ f ∥ L k ([ s , t ]) := { ∫ s t ∣ f r ∣ k d r } 1/ k < ∞.
For λ ≥ 1 \lambda\ge1 λ ≥ 1 , denote by C λ -var ( Δ [ s , t ] , E ) C^{\lambda\text{-var}}(\Delta_{[s,t]},E) C λ -var ( Δ [ s , t ] , E ) the space of continuous functions f ∈ C ( Δ [ s , t ] , E ) f\in C(\Delta_{[s,t]},E) f ∈ C ( Δ [ s , t ] , E ) such that
∥ f ∥ λ -var ; [ s , t ] : = sup π ∈ P [ s , t ] { ∑ [ u , v ] ∈ π ∥ f u , v ∥ E λ } 1 / λ < ∞ . \|f\|_{\lambda\text{-var};[s,t]}:=\sup_{\pi\in\mathcal{P}_{[s,t]}}\left\{\sum_{[u,v]\in\pi}\|f_{u,v}\|_E^\lambda\right\}^{1/\lambda}<\infty. ∥ f ∥ λ -var ; [ s , t ] := π ∈ P [ s , t ] sup ⎩ ⎨ ⎧ [ u , v ] ∈ π ∑ ∥ f u , v ∥ E λ ⎭ ⎬ ⎫ 1/ λ < ∞.
Also denote the λ \lambda λ -variation norm of f f f as a two-parameter function defined by
( ∥ f ∥ λ -var ) s , t : = ∥ f ∥ λ -var ; [ s , t ] . (\|f\|_{\lambda\text{-var}})_{s,t}:=\|f\|_{\lambda\text{-var};[s,t]}. ( ∥ f ∥ λ -var ) s , t := ∥ f ∥ λ -var ; [ s , t ] .
For any α ∈ ( 0 , 1 ] \alpha\in(0,1] α ∈ ( 0 , 1 ] and f ∈ C ( Δ [ s , t ] , E ) f \in C(\Delta_{[s,t]},E) f ∈ C ( Δ [ s , t ] , E ) , we write f ∈ C α ( Δ [ s , t ] , E ) f \in C^\alpha(\Delta_{[s,t]},E) f ∈ C α ( Δ [ s , t ] , E ) if
∥ f ∥ α -H o ¨ l ; [ s , t ] : = sup ( u , v ) ∈ Δ [ s , t ] u ≠ v ∥ f u , v ∥ E ∣ u − v ∣ α < ∞ . \|f\|_{\alpha\text{-Höl};[s,t]} := \sup_{\substack{(u,v)\in\Delta_{[s,t]}\\u\ne v}} \frac{\|f_{u,v}\|_E}{|u-v|^\alpha}<\infty. ∥ f ∥ α -H o ¨ l ; [ s , t ] := ( u , v ) ∈ Δ [ s , t ] u = v sup ∣ u − v ∣ α ∥ f u , v ∥ E < ∞.
For g ∈ C ( [ s , t ] , E ) g \in C([s,t],E) g ∈ C ([ s , t ] , E ) , define the increment operator δ g : Δ [ s , t ] → E \delta g:\Delta_{[s,t]}\to E δ g : Δ [ s , t ] → E by
δ g u , v : = g v − g u , where ( u , v ) ∈ Δ [ s , t ] . \delta g_{u,v}:=g_v-g_u,\qquad\text{where }(u,v)\in\Delta_{[s,t]}. δ g u , v := g v − g u , where ( u , v ) ∈ Δ [ s , t ] .
For a one-parameter function g ∈ C ( [ 0 , T ] , E ) g \in C([0,T],E) g ∈ C ([ 0 , T ] , E ) , we write ∥ δ g ∥ λ -var \|\delta g\|_{\lambda\text{-var}} ∥ δ g ∥ λ -var and similarly for the Hölder norm.
Assume X = ( X , X ) \mathbf{X}=(X,\mathbb{X}) X = ( X , X ) with X ∈ C ( [ 0 , T ] , V ) X\in C([0,T],V) X ∈ C ([ 0 , T ] , V ) and X ∈ C ( Δ , V ⊗ 2 ) \mathbb{X}\in C(\Delta,V^{\otimes2}) X ∈ C ( Δ , V ⊗ 2 ) . For any ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , set
∣ δ X ∣ λ -var ; [ s , t ] : = ∥ δ X ∥ λ -var ; [ s , t ] + ∥ X ∥ λ / 2 -var ; [ s , t ] (5) |\delta\mathbf{X}|_{\lambda\text{-var};[s,t]}:=\|\delta X\|_{\lambda\text{-var};[s,t]}+\|\mathbb{X}\|_{\lambda/2\text{-var};[s,t]} \tag*{(5)} ∣ δ X ∣ λ -var ; [ s , t ] := ∥ δ X ∥ λ -var ; [ s , t ] + ∥ X ∥ λ /2 -var ; [ s , t ] (5)
and
∥ δ X ∥ λ -var ; [ s , t ] : = ∥ δ X ∥ λ -var ; [ s , t ] + ∥ X ∥ λ / 2 -var ; [ s , t ] 1 / 2 . \|\delta\mathbf{X}\|_{\lambda\text{-var};[s,t]}:=\|\delta X\|_{\lambda\text{-var};[s,t]}+\|\mathbb{X}\|_{\lambda/2\text{-var};[s,t]}^{1/2}. ∥ δ X ∥ λ -var ; [ s , t ] := ∥ δ X ∥ λ -var ; [ s , t ] + ∥ X ∥ λ /2 -var ; [ s , t ] 1/2 .
We also set ∣ δ X ∣ α ; [ s , t ] : = ∥ δ X ∥ α -H o ¨ l ; [ s , t ] + ∥ X ∥ 2 α -H o ¨ l ; [ s , t ] |\delta\mathbf{X}|_{\alpha;[s,t]}:=\|\delta X\|_{\alpha\text{-Höl};[s,t]}+\|\mathbb{X}\|_{2\alpha\text{-Höl};[s,t]} ∣ δ X ∣ α ; [ s , t ] := ∥ δ X ∥ α -H o ¨ l ; [ s , t ] + ∥ X ∥ 2 α -H o ¨ l ; [ s , t ] and ∥ δ X ∥ α ; [ s , t ] : = ∥ δ X ∥ α -H o ¨ l ; [ s , t ] + ∥ X ∥ 2 α -H o ¨ l ; [ s , t ] 1 / 2 \|\delta\mathbf{X}\|_{\alpha;[s,t]}:=\|\delta X\|_{\alpha\text{-Höl};[s,t]}+\|\mathbb{X}\|_{2\alpha\text{-Höl};[s,t]}^{1/2} ∥ δ X ∥ α ; [ s , t ] := ∥ δ X ∥ α -H o ¨ l ; [ s , t ] + ∥ X ∥ 2 α -H o ¨ l ; [ s , t ] 1/2 .
A path X ∈ C ( [ 0 , T ] , V ) X\in C([0,T],V) X ∈ C ([ 0 , T ] , V ) is called a Lipschitz path if it is Lipschitz continuous in time. It is termed a smooth path if it possesses continuous derivatives of all orders on the closed interval [ 0 , T ] [0,T] [ 0 , T ] , in which case we write X ∈ C ∞ ( [ 0 , T ] , V ) X\in C^\infty([0,T],V) X ∈ C ∞ ([ 0 , T ] , V ) . For a Lipschitz path X X X , we define its (canonical) lift as the pair X = ( X , X ) \mathbf{X}=(X,\mathbb{X}) X = ( X , X ) , where the second-order component X s , t \mathbb{X}_{s,t} X s , t is given by
X s , t : = ∫ s t δ X s , r ⊗ ∂ r X r d r , ( s , t ) ∈ Δ . \mathbb{X}_{s,t}:=\int_s^t \delta X_{s,r}\otimes\partial_rX_r\,\mathrm{d}r,\qquad(s,t)\in\Delta. X s , t := ∫ s t δ X s , r ⊗ ∂ r X r d r , ( s , t ) ∈ Δ.
Differentiability: For γ ≥ 0 \gamma\ge0 γ ≥ 0 , let ⌊ γ ⌋ \lfloor\gamma\rfloor ⌊ γ ⌋ denote the largest integer less than or equal to γ \gamma γ . Denote by C b γ ( V , K ) C_b^\gamma(V,K) C b γ ( V , K ) the space of ⌊ γ ⌋ \lfloor\gamma\rfloor ⌊ γ ⌋ times continuously differentiable functions F : V → K F:V\to K F : V → K with bounded derivatives up to order ⌊ γ ⌋ \lfloor\gamma\rfloor ⌊ γ ⌋ , and such that D ⌊ γ ⌋ F D^{\lfloor\gamma\rfloor}F D ⌊ γ ⌋ F is Hölder continuous with exponent γ − ⌊ γ ⌋ \gamma-\lfloor\gamma\rfloor γ − ⌊ γ ⌋ . The norm is defined by
∥ F ∥ C b γ : = ∑ i = 0 ⌊ γ ⌋ ∥ D i F ∥ ∞ + ∥ D ⌊ γ ⌋ F ∥ ( γ − ⌊ γ ⌋ ) -H o ¨ l . \|F\|_{C_b^\gamma}:=\sum_{i=0}^{\lfloor\gamma\rfloor}\|D^iF\|_\infty+\|D^{\lfloor\gamma\rfloor}F\|_{(\gamma-\lfloor\gamma\rfloor)\text{-Höl}}. ∥ F ∥ C b γ := i = 0 ∑ ⌊ γ ⌋ ∥ D i F ∥ ∞ + ∥ D ⌊ γ ⌋ F ∥ ( γ − ⌊ γ ⌋) -H o ¨ l .
L k L^k L k -norm: For k ≥ 1 k\ge1 k ≥ 1 (k = ∞ k=\infty k = ∞ resp.), denote by L t k ( V ) : = L k ( F t , V ) L_t^k(V):=L^k(\mathcal{F}_t,V) L t k ( V ) := L k ( F t , V ) the space of V V V -valued F t \mathcal{F}_t F t -measurable random variables ξ \xi ξ such that
∥ ξ ∥ k : = { E [ ∣ ξ ∣ k ] } 1 / k < ∞ ( ∥ ξ ∥ ∞ : = ess sup ω ∈ Ω ∣ ξ ( ω ) ∣ < ∞ resp. ) \|\xi\|_k:=\left\{\mathbb{E}[|\xi|^k]\right\}^{1/k}<\infty\qquad(\|\xi\|_\infty:=\operatorname*{ess\,sup}_{\omega\in\Omega}|\xi(\omega)|<\infty\ \text{resp.}) ∥ ξ ∥ k := { E [ ∣ ξ ∣ k ] } 1/ k < ∞ ( ∥ ξ ∥ ∞ := ω ∈ Ω ess sup ∣ ξ ( ω ) ∣ < ∞ resp. )
Moreover, we denote by L 0 L^0 L 0 the space of measurable random variables.
For ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ and k ≥ 1 k\ge1 k ≥ 1 (k = ∞ k=\infty k = ∞ resp.), define L k ( [ s , t ] × Ω , V ) L^k([s,t]\times\Omega,V) L k ([ s , t ] × Ω , V ) (L ∞ ( [ s , t ] × Ω , V ) L^\infty([s,t]\times\Omega,V) L ∞ ([ s , t ] × Ω , V ) resp.) as the space of progressively measurable processes S S S satisfying
∥ S ∥ L k ( [ s , t ] × Ω ) : = ∥ ∥ S ⋅ ∥ k ∥ L k ( [ s , t ] ) < ∞ \|S\|_{L^k([s,t]\times\Omega)}:=\|\|S_\cdot\|_k\|_{L^k([s,t])}<\infty ∥ S ∥ L k ([ s , t ] × Ω ) := ∥∥ S ⋅ ∥ k ∥ L k ([ s , t ]) < ∞
( ∥ S ∥ L ∞ ( [ s , t ] × Ω ) : = sup r ∈ [ s , t ] ∥ S r ∥ ∞ < ∞ resp. ) \left(\|S\|_{L^\infty([s,t]\times\Omega)}:=\sup_{r\in[s,t]}\|S_r\|_\infty<\infty\quad\text{resp.}\right) ( ∥ S ∥ L ∞ ([ s , t ] × Ω ) := r ∈ [ s , t ] sup ∥ S r ∥ ∞ < ∞ resp. )
where we may also use ∥ ⋅ ∥ ∞ : = ∥ ⋅ ∥ L ∞ ( [ 0 , T ] × Ω ) \|\cdot\|_\infty:=\|\cdot\|_{L^\infty([0,T]\times\Omega)} ∥ ⋅ ∥ ∞ := ∥ ⋅ ∥ L ∞ ([ 0 , T ] × Ω ) for processes as well if there is no confusion.
Denote by H k , ∞ ( [ s , t ] ) \mathcal{H}^{k,\infty}([s,t]) H k , ∞ ([ s , t ]) the space of progressively measurable processes S S S such that
∥ S ∥ H k , ∞ ; [ s , t ] : = { ∥ E [ ∥ S ∥ L k ( [ ⋅ , t ] ) k ] ∥ L ∞ ( [ s , t ] × Ω ) } 1 / k < ∞ . \|S\|_{\mathcal{H}^{k,\infty};[s,t]}:=\left\{\left\|\mathbb{E}\left[\|S\|_{L^k([\cdot,t])}^k\right]\right\|_{L^\infty([s,t]\times\Omega)}\right\}^{1/k}<\infty. ∥ S ∥ H k , ∞ ; [ s , t ] := { E [ ∥ S ∥ L k ([ ⋅ , t ]) k ] L ∞ ([ s , t ] × Ω ) } 1/ k < ∞.
∥ ξ ∣ G ∥ m : = { E [ ∣ ξ ∣ m ∣ G ] } 1 m \left\|\xi\mid\mathcal{G}\right\|_m := \left\{\mathbb{E}\left[|\xi|^m \mid\mathcal{G}\right]\right\}^{\frac{1}{m}} ∥ ξ ∣ G ∥ m := { E [ ∣ ξ ∣ m ∣ G ] } m 1
Define the following ( m , n ) (m,n) ( m , n ) -norm by taking the L n L^n L n -norm of the conditional L m L^m L m -norm:
∥ ξ ∣ G ∥ m , n : = ∥ ∥ ξ ∣ G ∥ m ∥ n . (6) \left\|\xi\mid\mathcal{G}\right\|_{m,n} := \left\|\left\|\xi\mid\mathcal{G}\right\|_m\right\|_n . \tag*{(6)} ∥ ξ ∣ G ∥ m , n := ∥ ∥ ξ ∣ G ∥ m ∥ n . (6)
We have ([11 ] , Proposition 2.3), for n ≥ m n \ge m n ≥ m ,
∥ ξ ∥ m = ∥ ∥ ξ ∣ G ∥ m ∥ m ≤ ∥ ∥ ξ ∣ G ∥ m ∥ n ≤ ∥ ∥ ξ ∣ G ∥ n ∥ n = ∥ ξ ∥ n , for all ξ ∈ L n . \|\xi\|_m = \left\|\left\|\xi\mid\mathcal{G}\right\|_m\right\|_m \le\left\|\left\|\xi\mid\mathcal{G}\right\|_m\right\|_n \le\left\|\left\|\xi\mid\mathcal{G}\right\|_n\right\|_n = \|\xi\|_n,\qquad\text{for all }\xi\in L^n. ∥ ξ ∥ m = ∥ ∥ ξ ∣ G ∥ m ∥ m ≤ ∥ ∥ ξ ∣ G ∥ m ∥ n ≤ ∥ ∥ ξ ∣ G ∥ n ∥ n = ∥ ξ ∥ n , for all ξ ∈ L n .
For a two-parameter process F : Δ × Ω → V F:\Delta\times\Omega\to V F : Δ × Ω → V define
∥ F ∥ ∞ ; m , n : = sup ( s , t ) ∈ Δ ∥ F s , t ∣ F s ∥ m , n , and ∥ F ∥ ∞ ; n : = sup ( s , t ) ∈ Δ ∥ F s , t ∥ n . \|F\|_{\infty;m,n} := \sup_{(s,t)\in\Delta}\|F_{s,t}\mid\mathcal{F}_s\|_{m,n},\qquad\text{and}\qquad\|F\|_{\infty;n} := \sup_{(s,t)\in\Delta}\|F_{s,t}\|_n. ∥ F ∥ ∞ ; m , n := ( s , t ) ∈ Δ sup ∥ F s , t ∣ F s ∥ m , n , and ∥ F ∥ ∞ ; n := ( s , t ) ∈ Δ sup ∥ F s , t ∥ n .
BMO-martingale: For ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , let M [ s , t ] l o c ( V ) \mathcal{M}^{\mathrm{loc}}_{[s,t]}(V) M [ s , t ] loc ( V ) be the space of continuous local martingales on [ s , t ] [s,t] [ s , t ] , with M s = 0 M_s=0 M s = 0 . For M ⋅ − M s ∈ M [ s , t ] l o c ( V ) M_{\cdot}-M_s\in\mathcal{M}^{\mathrm{loc}}_{[s,t]}(V) M ⋅ − M s ∈ M [ s , t ] loc ( V ) , we call M ∈ B M O [ s , t ] ( V ) M\in\mathrm{BMO}_{[s,t]}(V) M ∈ BMO [ s , t ] ( V ) if
∥ δ M ∥ B M O ; [ s , t ] : = ess sup ( u , ω ) ∈ [ s , t ] × Ω { E u [ δ ⟨ M ⟩ u , t ] } 1 2 < ∞ . \|\delta M\|_{\mathrm{BMO};[s,t]} := \operatorname*{ess\,sup}_{(u,\omega)\in[s,t]\times\Omega}\left\{\mathbb{E}_u\left[\delta\langle M\rangle_{u,t}\right]\right\}^{\frac{1}{2}} < \infty. ∥ δ M ∥ BMO ; [ s , t ] := ( u , ω ) ∈ [ s , t ] × Ω ess sup { E u [ δ ⟨ M ⟩ u , t ] } 2 1 < ∞.
BMO and VMO-process: Let S S S be a V V V -valued adapted right continuous process with left limits (RCLL). Then, we call S S S a BMO-process on [ s , t ] [s,t] [ s , t ] if
ϱ s , t ( S ) : = sup s ≤ τ ≤ τ ′ ≤ t ∥ S τ ′ − S τ − ∣ F τ ∥ 1 , ∞ < ∞ , \varrho_{s,t}(S) := \sup_{s\le\tau\le\tau'\le t}\|S_{\tau'}-S_{\tau-}\mid\mathcal{F}_\tau\|_{1,\infty}<\infty, ϱ s , t ( S ) := s ≤ τ ≤ τ ′ ≤ t sup ∥ S τ ′ − S τ − ∣ F τ ∥ 1 , ∞ < ∞ ,
where the supremum is taken over all stopping times τ , τ ′ \tau,\tau' τ , τ ′ , and S τ − S_{\tau-} S τ − is the left limit of S τ ′ S_{\tau'} S τ ′ . When S S S is continuous in time, the stopping times in the supremum can be replaced by deterministic times, which can be referred to [20, Proposition 2.2].
As a corollary of (A.2) in Lemma A.1, for any continuous local martingale M M M , the BMO norm ∥ δ M ∥ B M O ; [ s , t ] \|\delta M\|_{\mathrm{BMO};[s,t]} ∥ δ M ∥ BMO ; [ s , t ] is equivalent to the modulus of mean oscillation ϱ s , t ( M ) \varrho_{s,t}(M) ϱ s , t ( M ) . Hence, a continuous local martingale is a BMO-process on [ s , t ] [s,t] [ s , t ] if and only if it belongs to B M O [ s , t ] \mathrm{BMO}_{[s,t]} BMO [ s , t ] .
For a BMO-process S S S on [ 0 , T ] [0,T] [ 0 , T ] , we call S S S a VMO process if
lim h ↓ 0 sup ( s , t ) ∈ Δ , t − s ≤ h ϱ s , t ( S ) = 0. \lim_{h\downarrow0}\sup_{\substack{(s,t)\in\Delta,\\t-s\le h}}\varrho_{s,t}(S)=0. h ↓ 0 lim ( s , t ) ∈ Δ , t − s ≤ h sup ϱ s , t ( S ) = 0.
For λ ≥ 1 \lambda\ge1 λ ≥ 1 and a VMO process S S S , we call S S S a VMOλ ^\lambda λ -var process if ∥ ϱ ( S ) ∥ λ -var ; [ 0 , T ] < ∞ \|\varrho(S)\|_{\lambda\text{-var};[0,T]}<\infty ∥ ϱ ( S ) ∥ λ -var ; [ 0 , T ] < ∞ .
Lemma 2.1. Assume 1 ≤ m ≤ n ≤ ∞ 1\le m\le n\le\infty 1 ≤ m ≤ n ≤ ∞ , m < ∞ m<\infty m < ∞ , and q ≥ 1 q\ge1 q ≥ 1 . For ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , denote by L ∞ ; m , n C 2 q -var ( [ s , t ] ) L^{\infty;m,n}C^{q\text{-var}}_2([s,t]) L ∞ ; m , n C 2 q -var ([ s , t ]) the set of all two-parameter continuous processes P P P with ∥ ∥ P ∥ q -var ∥ ∞ ; m , n ; [ s , t ] < ∞ \|\|P\|_{q\text{-var}}\|_{\infty;m,n;[s,t]}<\infty ∥∥ P ∥ q -var ∥ ∞ ; m , n ; [ s , t ] < ∞ , and denote by L ∞ ; m , n C q -var ( [ s , t ] ) L^{\infty;m,n}C^{q\text{-var}}([s,t]) L ∞ ; m , n C q -var ([ s , t ]) the set of all one-parameter continuous processes S S S with ∥ S ∥ L ∞ ; m , n C q -var ; [ s , t ] : = ∥ S t ∥ n + ∥ ∥ δ S ∥ q -var ∥ ∞ ; m , n ; [ s , t ] < ∞ \|S\|_{L^{\infty;m,n}C^{q\text{-var}};[s,t]}:=\|S_t\|_n+\|\|\delta S\|_{q\text{-var}}\|_{\infty;m,n;[s,t]}<\infty ∥ S ∥ L ∞ ; m , n C q -var ; [ s , t ] := ∥ S t ∥ n + ∥∥ δ S ∥ q -var ∥ ∞ ; m , n ; [ s , t ] < ∞ . Then, ( L ∞ ; m , n C 2 q -var ( [ s , t ] ) , ∥ ∥ ⋅ ∥ q -var ∥ ∞ ; m , n ; [ s , t ] ) (L^{\infty;m,n}C^{q\text{-var}}_2([s,t]),\|\|\cdot\|_{q\text{-var}}\|_{\infty;m,n;[s,t]}) ( L ∞ ; m , n C 2 q -var ([ s , t ]) , ∥∥ ⋅ ∥ q -var ∥ ∞ ; m , n ; [ s , t ] ) and ( L ∞ ; m , n C q -var ( [ s , t ] ) , ∥ ⋅ ∥ L ∞ ; m , n C q -var ; [ s , t ] ) (L^{\infty;m,n}C^{q\text{-var}}([s,t]),\|\cdot\|_{L^{\infty;m,n}C^{q\text{-var}};[s,t]}) ( L ∞ ; m , n C q -var ([ s , t ]) , ∥ ⋅ ∥ L ∞ ; m , n C q -var ; [ s , t ] ) are two Banach spaces.
Proof. Let { P n } n ≥ 1 \{P^n\}_{n\ge1} { P n } n ≥ 1 be a Cauchy sequence of L ∞ ; m , n C 2 q -var ( [ s , t ] ) L^{\infty;m,n}C^{q\text{-var}}_2([s,t]) L ∞ ; m , n C 2 q -var ([ s , t ]) . Then, there exists a subsequence { n j } j ≥ 1 \{n_j\}_{j\ge1} { n j } j ≥ 1 such that
lim j ′ → ∞ ∑ j ≥ j ′ E [ ∥ P n j − P n j + 1 ∥ q -var ; [ s , t ] ] = 0. \lim_{j'\to\infty}\sum_{j\ge j'}\mathbb{E}\left[\|P^{n_j}-P^{n_{j+1}}\|_{q\text{-var};[s,t]}\right]=0. j ′ → ∞ lim j ≥ j ′ ∑ E [ ∥ P n j − P n j + 1 ∥ q -var ; [ s , t ] ] = 0.
Hence, by Lévy’s theorem, we have
E [ lim j ′ → ∞ ∑ j ≥ j ′ ∥ P n j − P n j + 1 ∥ q -var ; [ s , t ] ] = 0. \mathbb{E}\left[\lim_{j'\to\infty}\sum_{j\ge j'}\|P^{n_j}-P^{n_{j+1}}\|_{q\text{-var};[s,t]}\right]=0. E j ′ → ∞ lim j ≥ j ′ ∑ ∥ P n j − P n j + 1 ∥ q -var ; [ s , t ] = 0.
Therefore, a.s., { P n j } j ≥ 1 \{P^{n_j}\}_{j\ge1} { P n j } j ≥ 1 forms a Cauchy sequence under the norm ∥ ⋅ ∥ q -var ; [ s , t ] \lVert\cdot\rVert_{q\text{-var};[s,t]} ∥ ⋅ ∥ q -var ; [ s , t ] .
Denote by P P P the limit of { P n j } j ≥ 1 \{P^{n_j}\}_{j\ge1} { P n j } j ≥ 1 . By Fatou’s lemma for the conditional expectation and the lower semi-continuity property of ∥ ⋅ ∥ q -var \lVert\cdot\rVert_{q\text{-var}} ∥ ⋅ ∥ q -var , we see that
lim j → ∞ ∥ ∥ P n j − P ∥ q -var ∥ ∞ ; m , n ; [ s , t ] = 0. \lim_{j\to\infty}\left\lVert\lVert P^{n_j}-P\rVert_{q\text{-var}}\right\rVert_{\infty;m,n;[s,t]}=0. j → ∞ lim ∥ ∥ P n j − P ∥ q -var ∥ ∞ ; m , n ; [ s , t ] = 0.
Consequently, the Cauchy property forces the sequence { P n } n ≥ 1 \{P^n\}_{n\ge1} { P n } n ≥ 1 to have a unique limit P P P , which implies the desired Banach space property of ( L ∞ ; m , n C q q -var ( [ s , t ] ) , ∥ ∥ ⋅ ∥ q -var ∥ ∞ ; m , n ; [ s , t ] ) (L^{\infty;m,n}C_q^{q\text{-var}}([s,t]),\lVert\lVert\cdot\rVert_{q\text{-var}}\rVert_{\infty;m,n;[s,t]}) ( L ∞ ; m , n C q q -var ([ s , t ]) , ∥∥ ⋅ ∥ q -var ∥ ∞ ; m , n ; [ s , t ] ) . Similarly, one can prove that ( L ∞ ; m , n C q -var ( [ s , t ] ) , ∥ ⋅ ∥ L ∞ ; m , n C q -var ; [ s , t ] ) (L^{\infty;m,n}C^{q\text{-var}}([s,t]),\lVert\cdot\rVert_{L^{\infty;m,n}C^{q\text{-var}};[s,t]}) ( L ∞ ; m , n C q -var ([ s , t ]) , ∥ ⋅ ∥ L ∞ ; m , n C q -var ; [ s , t ] ) is also a Banach space. □
Below are assumptions that will be used in the paper.
Assumption ( A p a r ) (A_{\mathrm{par}}) ( A par ) .
( p , p ′ ) (p,p') ( p , p ′ ) : p ∈ ( 2 , 3 ) p\in(2,3) p ∈ ( 2 , 3 ) , p ≤ p ′ p\le p' p ≤ p ′ , and 2 p + 1 p ′ > 1 \frac{2}{p}+\frac{1}{p'}>1 p 2 + p ′ 1 > 1 .
( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) : p ∈ ( 2 , 3 ) p\in(2,3) p ∈ ( 2 , 3 ) , p ≤ q ≤ q ′ p\le q\le q' p ≤ q ≤ q ′ , and 1 p + 1 q + 1 q ′ > 1 \frac{1}{p}+\frac{1}{q}+\frac{1}{q'}>1 p 1 + q 1 + q ′ 1 > 1 . In particular, this implies q < 4 q<4 q < 4 .
( γ , p , q , q ′ ) (\gamma,p,q,q') ( γ , p , q , q ′ ) : Assumption ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) holds, and γ > p \gamma>p γ > p satisfies
γ − 1 ≥ q + q ′ q ′ . \gamma-1\ge\frac{q+q'}{q'}. γ − 1 ≥ q ′ q + q ′ .
Throughout this paper, for q ≤ q ′ q\le q' q ≤ q ′ we denote by
q ′ ′ / 2 : = ( 1 / q + 1 / q ′ ) − 1 = q q ′ q + q ′ . q''/2:=(1/q+1/q')^{-1}=\frac{qq'}{q+q'}. q ′′ /2 := ( 1/ q + 1/ q ′ ) − 1 = q + q ′ q q ′ .
Assumption ( A f ) (A_f) ( A f ) .
( f ) (f) ( f ) : Let C f > 0 C_f>0 C f > 0 be a constant. Suppose that f : Ω × [ 0 , T ] × R × R d → R f:\Omega\times[0,T]\times\mathbb{R}\times\mathbb{R}^d\to\mathbb{R} f : Ω × [ 0 , T ] × R × R d → R is a progressively measurable function such that the following bounds hold
∣ f ( ω , t , 0 , 0 ) ∣ ≤ C f and ∣ f ( ω , t , y , z ) − f ( ω , t , y ′ , z ′ ) ∣ ≤ C f ( ∣ y − y ′ ∣ + ∣ z − z ′ ∣ ) \lvert f(\omega,t,0,0)\rvert\le C_f\quad\text{and}\quad\lvert f(\omega,t,y,z)-f(\omega,t,y',z')\rvert\le C_f(\lvert y-y'\rvert+\lvert z-z'\rvert) ∣ f ( ω , t , 0 , 0 )∣ ≤ C f and ∣ f ( ω , t , y , z ) − f ( ω , t , y ′ , z ′ )∣ ≤ C f (∣ y − y ′ ∣ + ∣ z − z ′ ∣)
for all ( ω , t , y , y ′ , z , z ′ ) ∈ Ω × [ 0 , T ] × R × R × R d × R d (\omega,t,y,y',z,z')\in\Omega\times[0,T]\times\mathbb{R}\times\mathbb{R}\times\mathbb{R}^d\times\mathbb{R}^d ( ω , t , y , y ′ , z , z ′ ) ∈ Ω × [ 0 , T ] × R × R × R d × R d .
( f d e t ) (f_{\mathrm{det}}) ( f det ) : Assumption ( f ) (f) ( f ) holds. Assume further that f f f is a deterministic function.
Under Assumption ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) , for simplicity of notation, we denote the parameter set by
Θ : = ( T , d , q , q ′ , p ) . (7) \Theta:=(T,d,q,q',p). \tag*{(7)} Θ := ( T , d , q , q ′ , p ) . (7)
In addition, we write ≲ \lesssim ≲ to represent that an inequality holds up to the right-hand side (RHS) multiplied by a constant that depends only on Θ \Theta Θ ; for a parameter K K K , the notation ≲ K \lesssim_K ≲ K represents that an inequality holds up to the RHS multiplied by a constant that depends only on ( Θ , K ) (\Theta,K) ( Θ , K ) .
Elements of Rough Stochastic Analysis
Throughout this section, p ∈ ( 2 , 3 ) p\in(2,3) p ∈ ( 2 , 3 ) is a fixed number, p ′ , q , q ′ p',q,q' p ′ , q , q ′ are numbers greater than two such that p ′ ≥ p p'\ge p p ′ ≥ p and q ′ ≥ q q'\ge q q ′ ≥ q , and V 1 V_1 V 1 , V 2 V_2 V 2 , and V V V are Euclidean spaces.
Definition 2.2 (p p p -rough path). For any X ∈ C p -var ( [ 0 , T ] , V ) \mathbf{X}\in C^{p\text{-var}}([0,T],V) X ∈ C p -var ([ 0 , T ] , V ) and X ∈ C p / 2 -var ( Δ , V ⊗ 2 ) \mathbb{X}\in C^{p/2\text{-var}}(\Delta,V^{\otimes2}) X ∈ C p /2 -var ( Δ , V ⊗ 2 ) , we call X = ( X , X ) \mathbf{X}=(X,\mathbb{X}) X = ( X , X ) a p p p -rough path (a rough path with finite p p p -variation), denoted by X ∈ C p -var ( [ 0 , T ] , V ) \mathbf{X}\in\mathcal{C}^{p\text{-var}}([0,T],V) X ∈ C p -var ([ 0 , T ] , V ) , if the following Chen’s relation holds:
X s , t − X s , u − X u , t = δ X s , u ⊗ δ X u , t , 0 ≤ s ≤ u ≤ t ≤ T . (8) \mathbb{X}_{s,t}-\mathbb{X}_{s,u}-\mathbb{X}_{u,t}=\delta X_{s,u}\otimes\delta X_{u,t},\qquad0\le s\le u\le t\le T. \tag*{(8)} X s , t − X s , u − X u , t = δ X s , u ⊗ δ X u , t , 0 ≤ s ≤ u ≤ t ≤ T . (8)
We call X \mathbf{X} X a geometric p p p -rough path, denoted by X ∈ C 0 , p -var ( [ 0 , T ] , V ) \mathbf{X}\in\mathcal{C}^{0,p\text{-var}}([0,T],V) X ∈ C 0 , p -var ([ 0 , T ] , V ) , if X ∈ C p -var ( [ 0 , T ] , V ) \mathbf{X}\in\mathcal{C}^{p\text{-var}}([0,T],V) X ∈ C p -var ([ 0 , T ] , V ) and there exists a sequence of smooth paths { X n } n ≥ 1 \{X^n\}_{n\ge1} { X n } n ≥ 1 with their canonical lifts { X n : = ( X n , X n ) } n ≥ 1 \{\mathbf{X}^n:=(X^n,\mathbb{X}^n)\}_{n\ge1} { X n := ( X n , X n ) } n ≥ 1 such that ∣ δ ( X n − X ) ∣ p -var → 0 \lvert\delta(\mathbf{X}^n-\mathbf{X})\rvert_{p\text{-var}}\to0 ∣ δ ( X n − X ) ∣ p -var → 0 as n → ∞ n\to\infty n → ∞ .
Definition 2.3 (1/p p p -Hölder rough path). For X ∈ C p -var ( [ 0 , T ] , V ) \mathbf{X} \in\mathcal{C}^{p\text{-var}}([0,T],V) X ∈ C p -var ([ 0 , T ] , V ) , we call X \mathbf{X} X a 1 / p 1/p 1/ p -Hölder rough path, denoted by X ∈ C 1 / p ( [ 0 , T ] , V ) \mathbf{X} \in\mathcal{C}^{1/p}([0,T],V) X ∈ C 1/ p ([ 0 , T ] , V ) , if ∣ δ X ∣ 1 / p < ∞ \lvert\delta\mathbf{X}\rvert_{1/p}<\infty ∣ δ X ∣ 1/ p < ∞ . We call X \mathbf{X} X a geometric 1 / p 1/p 1/ p -Hölder rough path, denoted by X ∈ C g 0 , 1 / p ( [ 0 , T ] , V ) \mathbf{X}\in\mathcal{C}^{0,1/p}_{g}([0,T],V) X ∈ C g 0 , 1/ p ([ 0 , T ] , V ) , if X ∈ C 1 / p ( [ 0 , T ] , V ) \mathbf{X}\in\mathcal{C}^{1/p}([0,T],V) X ∈ C 1/ p ([ 0 , T ] , V ) and there exists a sequence of smooth paths { X n } n ≥ 1 \{\mathbf{X}^{n}\}_{n\geq1} { X n } n ≥ 1 with their canonical lifts { X n : = ( X n , X n ) } n ≥ 1 \{\mathbf{X}^{n}:=(X^{n},\mathbb{X}^{n})\}_{n\geq1} { X n := ( X n , X n ) } n ≥ 1 such that ∣ δ ( X n − X ) ∣ 1 / p → 0 \lvert\delta(\mathbf{X}^{n}-\mathbf{X})\rvert_{1/p}\to0 ∣ δ ( X n − X ) ∣ 1/ p → 0 as n → ∞ n\to\infty n → ∞ .
Remark 2.4. Define the bracket process
[ X ] t : = δ X 0 , t ⊗ δ X 0 , t − ( X 0 , t + X 0 , t T ) . [\mathbf{X}]_{t}:=\delta X_{0,t}\otimes\delta X_{0,t}-\left(\mathbb{X}_{0,t}+\mathbb{X}_{0,t}^{\mathsf{T}}\right). [ X ] t := δ X 0 , t ⊗ δ X 0 , t − ( X 0 , t + X 0 , t T ) .
By (8) , we have δ [ X ] s , t = δ X s , t ⊗ 2 − 2 Sym ( X s , t ) \delta[\mathbf{X}]_{s,t}=\delta X_{s,t}^{\otimes2}-2\operatorname{Sym}(\mathbb{X}_{s,t}) δ [ X ] s , t = δ X s , t ⊗ 2 − 2 Sym ( X s , t ) , and hence, [ X ] ∈ C p / 2 -var ( [ 0 , T ] , V ⊗ 2 ) [\mathbf{X}]\in\mathcal{C}^{p/2\text{-var}}([0,T],V^{\otimes2}) [ X ] ∈ C p /2 -var ([ 0 , T ] , V ⊗ 2 ) .
Let I ⊂ [ 0 , T ] I\subset[0,T] I ⊂ [ 0 , T ] be an interval and let X = ( X , X ) ∈ C p -var ( I , V ) \mathbf{X}=(X,\mathbb{X})\in\mathcal{C}^{p\text{-var}}(I,V) X = ( X , X ) ∈ C p -var ( I , V ) . We introduce the following three notions as integrands for rough stochastic integrals.
Definition 2.5 (Controlled rough path). Assume ψ : I → V 1 \psi:I\to V_{1} ψ : I → V 1 and ψ ′ : I → L ( V , V 1 ) \psi':I\to\mathcal{L}(V,V_{1}) ψ ′ : I → L ( V , V 1 ) are two deterministic continuous functions. We call ( ψ , ψ ′ ) (\psi,\psi') ( ψ , ψ ′ ) an X X X -controlled rough path with finite ( p , p ′ ) (p,p') ( p , p ′ ) -variation, denoted by ( ψ , ψ ′ ) ∈ D X ( p , p ′ ) -var ( I , V 1 ) (\psi,\psi')\in\mathcal{D}^{(p,p')\text{-var}}_{X}(I,V_{1}) ( ψ , ψ ′ ) ∈ D X ( p , p ′ ) -var ( I , V 1 ) , if
∥ ( ψ , ψ ′ ) ∥ D X ( p , p ′ ) -var ; I : = ∥ δ ψ ∥ p -var ; I + ∥ δ ψ ′ ∥ p ′ -var ; I + ∥ R ψ ∥ p p ′ p + p ′ -var ; I < ∞ , \|(\psi,\psi')\|_{\mathcal{D}^{(p,p')\text{-var}}_{X};I}:=\|\delta\psi\|_{p\text{-var};I}+\|\delta\psi'\|_{p'\text{-var};I}+\|R^{\psi}\|_{\frac{pp'}{p+p'}\text{-var};I}<\infty, ∥ ( ψ , ψ ′ ) ∥ D X ( p , p ′ ) -var ; I := ∥ δ ψ ∥ p -var ; I + ∥ δ ψ ′ ∥ p ′ -var ; I + ∥ R ψ ∥ p + p ′ p p ′ -var ; I < ∞ ,
where the remainder R s , t ψ : = δ ψ s , t − ψ s ′ δ X s , t R^{\psi}_{s,t}:=\delta\psi_{s,t}-\psi'_{s}\delta X_{s,t} R s , t ψ := δ ψ s , t − ψ s ′ δ X s , t for s , t ∈ Δ I s,t\in\Delta_{I} s , t ∈ Δ I .
Definition 2.6 (Rough semimartingale). Let
Y : I × Ω → V 1 , Y ′ : I × Ω → L ( V , V 1 ) Y:I\times\Omega\to V_{1},\qquad Y':I\times\Omega\to\mathcal{L}(V,V_{1}) Y : I × Ω → V 1 , Y ′ : I × Ω → L ( V , V 1 )
be continuous adapted processes. We call ( Y , Y ′ ) (Y,Y') ( Y , Y ′ ) an X X X -controlled rough semimartingale (RSM) with ( p , p ′ ) (p,p') ( p , p ′ ) -variation, and write
( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( I , V 1 ) , (Y,Y')\in\operatorname{RSM}^{(p,p')\text{-var}}_{X}(I,V_{1}), ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( I , V 1 ) ,
if there exist a continuous local martingale
M Y ∈ M I l o c ( V 1 ) M^{Y}\in\mathcal{M}^{\mathrm{loc}}_{I}(V_{1}) M Y ∈ M I loc ( V 1 )
and a two-parameter process P Y P^{Y} P Y such that
δ Y s , t = Y s ′ δ X s , t + P s , t Y + δ M s , t Y ≡ Y s ′ δ X s , t + R s , t Y ( s , t ) ∈ Δ I , \delta Y_{s,t}=Y'_{s}\delta X_{s,t}+P^{Y}_{s,t}+\delta M^{Y}_{s,t}\equiv Y'_{s}\delta X_{s,t}+R^{Y}_{s,t}\qquad(s,t)\in\Delta_{I}, δ Y s , t = Y s ′ δ X s , t + P s , t Y + δ M s , t Y ≡ Y s ′ δ X s , t + R s , t Y ( s , t ) ∈ Δ I ,
and ( Y − M Y , Y ′ ) ∈ D X ( p , p ′ ) -var ( I , V 1 ) (Y-M^{Y},Y')\in\mathcal{D}^{(p,p')\text{-var}}_{X}(I,V_{1}) ( Y − M Y , Y ′ ) ∈ D X ( p , p ′ ) -var ( I , V 1 ) , a.s. We call P Y P^{Y} P Y (resp. R Y R^{Y} R Y ) the pathwise (resp. stochastic) remainder of the RSM ( Y , Y ′ ) (Y,Y') ( Y , Y ′ ) .
Remark 2.7. Given ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( [ 0 , T ] , V ) (Y,Y')\in\operatorname{RSM}^{(p,p')\text{-var}}_{X}([0,T],V) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , V ) , whenever 1 p + 1 p ′ > 1 2 \frac{1}{p}+\frac{1}{p'}>\frac{1}{2} p 1 + p ′ 1 > 2 1 , the decomposition Y = ( Y − M Y ) + M Y Y=(Y-M^{Y})+M^{Y} Y = ( Y − M Y ) + M Y in Definition (10) is unique, as a consequence of eq:1.6 .
Definition 2.8 (Rough stochastic integral). Let X : = ( X , X ) \mathbf{X}:=(X,\mathbb{X}) X := ( X , X ) belong to C p -var ( [ 0 , T ] , V ) \mathcal{C}^{p\text{-var}}([0,T],V) C p -var ([ 0 , T ] , V ) . Let ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( [ 0 , T ] , V 1 ) (Y,Y')\in\operatorname{RSM}^{(p,p')\text{-var}}_{X}([0,T],V_{1}) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , V 1 ) . By eq:1.7 , assuming ( p , p ′ ) (p,p') ( p , p ′ ) , the limit in probability of the following Riemann sum exists:
∫ u v Y r d X r : = ∫ u v ( Y r , Y r ′ ) d X r : = lim ∣ π ∣ → 0 π ∈ P [ u , v ] ∑ [ s , t ] ∈ π Y s δ X s , t + Y s ′ X s , t . (9) \int_{u}^{v}Y_{r}\,dX_{r}:=\int_{u}^{v}(Y_{r},Y'_{r})\,d\mathbf{X}_{r}:=\lim_{\substack{|\pi|\to0\\ \pi\in\mathcal{P}_{[u,v]}}}\sum_{[s,t]\in\pi}Y_{s}\delta X_{s,t}+Y'_{s}\mathbb{X}_{s,t}. \tag*{(9)} ∫ u v Y r d X r := ∫ u v ( Y r , Y r ′ ) d X r := ∣ π ∣ → 0 π ∈ P [ u , v ] lim [ s , t ] ∈ π ∑ Y s δ X s , t + Y s ′ X s , t . (9)
We call ∫ u v Y r d X r \int_{u}^{v}Y_{r}\,dX_{r} ∫ u v Y r d X r a rough stochastic integral.
Remark 2.9. When X X X is Lipschitz continuous and X \mathbf{X} X is the canonical lift of X X X (we refer to eq:14 , Definition 7.2 for the definition), then [ X ] = 0 [\mathbf{X}]=0 [ X ] = 0 . Moreover, for any continuous process Y Y Y , the integral ∫ u v Y r d X r \int_{u}^{v}Y_{r}\,dX_{r} ∫ u v Y r d X r defined by the Riemann sum in (9) is equal to the Lebesgue integral ∫ u v Y r ∂ r X r d r \int_{u}^{v}Y_{r}\partial_{r}X_{r}\,dr ∫ u v Y r ∂ r X r d r , where ∂ r X \partial_{r}X ∂ r X denotes the time derivative of X X X .
Remark 2.10. For ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( [ 0 , T ] , V 1 ) (Y,Y') \in\operatorname{RSM}_{\mathbf{X}}^{(p,p')\text{-var}}([0,T],V_1) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , V 1 ) with p ′ p' p ′ satisfying ( p , p ′ ) (p,p') ( p , p ′ ) , by the integration by parts formula in [16 ] , Section 5.3, the following equality holds:
∫ u v Y r d X r : = M v Y X v − M u Y X u − ∫ u v X r d M r Y + ∫ u v ( Y − M Y ) r d X r , (10) \int_u^v Y_r\mathrm{d}\mathbf{X}_r := M_v^Y X_v-M_u^Y X_u-\int_u^v X_r\mathrm{d}M_r^Y+\int_u^v (Y-M^Y)_r\mathrm{d}\mathbf{X}_r, \tag*{(10)} ∫ u v Y r d X r := M v Y X v − M u Y X u − ∫ u v X r d M r Y + ∫ u v ( Y − M Y ) r d X r , (10)
where ∫ X r d M r Y \int X_r\mathrm{d}M_r^Y ∫ X r d M r Y is an Itô integral, and ∫ ( Y − M Y ) r d X r \int(Y-M^Y)_r\mathrm{d}\mathbf{X}_r ∫ ( Y − M Y ) r d X r is a rough integral (see, e.g., [10 ] ) defined pathwise.
Remark 2.11. For an RSM, its integration against a rough path yields again an RSM. More precisely, given ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( [ 0 , T ] , L ( V , V 1 ) ) (Y,Y') \in\operatorname{RSM}_{\mathbf{X}}^{(p,p')\text{-var}}([0,T],\mathcal{L}(V,V_1)) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , L ( V , V 1 )) with p ′ p' p ′ satisfying ( p , p ′ ) (p,p') ( p , p ′ ) , let Γ t : = ∫ 0 t Y r d X r \Gamma_t := \int_0^t Y_r\mathrm{d}\mathbf{X}_r Γ t := ∫ 0 t Y r d X r be defined as in Definition 2.8. By [16 ] , Theorem 1.7, ( Γ , Y ) ∈ D X ( p , p ) -var ( [ 0 , T ] , V 1 ) (\Gamma,Y) \in\mathcal{D}_{\mathbf{X}}^{(p,p)\text{-var}}([0,T],V_1) ( Γ , Y ) ∈ D X ( p , p ) -var ([ 0 , T ] , V 1 ) a.s. Consequently, M Γ = 0 M^\Gamma=0 M Γ = 0 and P Γ = R Γ P^\Gamma=R^\Gamma P Γ = R Γ .
Rough Stochastic Analysis of Rough Semimartingales
In this section, we establish several estimates for RSMs that will be used in the study of BSDE (1) in Section 4. Recall that p ∈ ( 2 , 3 ) p \in(2,3) p ∈ ( 2 , 3 ) is a fixed number associated with a given rough path X = ( X , X ) \mathbf{X}=(X,\mathbb{X}) X = ( X , X ) . Throughout this section, V V V , V 1 V_1 V 1 , and V 2 V_2 V 2 are Euclidean spaces; and p ′ p' p ′ , q q q , and q ′ q' q ′ are positive numbers satisfying
p ≤ p ′ , p ≤ q ≤ q ′ . (11) p \le p', \qquad p \le q \le q'. \tag*{(11)} p ≤ p ′ , p ≤ q ≤ q ′ . (11)
Energy Estimates and Equivalence of BMO-Type Norms
One of our main tools is the following energy inequality for adapted, non-decreasing processes, the proof of which can be referred to, e.g. [20 ] , Lemma A.2. In the following, we denote A u − : = lim v ↑ u A v A_{u-}:=\lim_{v\uparrow u}A_v A u − := lim v ↑ u A v for u > 0 u>0 u > 0 and A 0 − : = A 0 A_{0-}:=A_0 A 0 − := A 0 .
Lemma 3.1. Let s , t ∈ [ 0 , T ] s,t \in[0,T] s , t ∈ [ 0 , T ] be two fixed times. Assume C C C is a deterministic constant and that { A r } r ∈ [ s , t ] \{A_r\}_{r\in[s,t]} { A r } r ∈ [ s , t ] is an adapted, right continuous, non-decreasing process such that
∥ E τ [ A t − A τ − ] ∥ ∞ ≤ C , for all stopping times τ ∈ [ s , t ] . (12) \left\|\mathbb{E}_\tau[A_t-A_{\tau-}]\right\|_\infty\le C,\qquad\text{for all stopping times }\tau\in[s,t]. \tag*{(12)} ∥ E τ [ A t − A τ − ] ∥ ∞ ≤ C , for all stopping times τ ∈ [ s , t ] . (12)
That is, { A r } r ∈ [ s , t ] \{A_r\}_{r\in[s,t]} { A r } r ∈ [ s , t ] is a BMO-process on [ s , t ] [s,t] [ s , t ] . Then, for every stopping time τ ′ ∈ [ s , t ] \tau'\in[s,t] τ ′ ∈ [ s , t ] and k ≥ 1 k\ge1 k ≥ 1 ,
∥ δ A τ ′ , t | F τ ′ ∥ k , ∞ ≤ ∣ k ! ∣ 1 / k C . (13) \left\|\delta A_{\tau',t}\middle|\mathcal{F}_{\tau'}\right\|_{k,\infty}\le|k!|^{1/k}C. \tag*{(13)} ∥ δ A τ ′ , t ∣ F τ ′ ∥ k , ∞ ≤ ∣ k ! ∣ 1/ k C . (13)
Recall ∥ ∣ δ V ∣ q -var ∥ ∞ ; k , ∞ ; [ s , t ] : = sup ( u , v ) ∈ Δ [ s , t ] ∥ ∣ δ V ∣ q -var ; [ u , v ] | F u ∥ k , ∞ \left\|\left|\delta V\right|_{q\text{-var}}\right\|_{\infty;k,\infty;[s,t]}:=\sup_{(u,v)\in\Delta_{[s,t]}}\left\|\left|\delta V\right|_{q\text{-var};[u,v]}\middle|\mathcal{F}_u\right\|_{k,\infty} ∣ δ V ∣ q -var ∞ ; k , ∞ ; [ s , t ] := sup ( u , v ) ∈ Δ [ s , t ] ∣ δ V ∣ q -var ; [ u , v ] F u k , ∞ . Lemma 3.2 shows that the BMO-type norm ∥ ∣ δ V ∣ q -var ∥ ∞ ; k , ∞ \left\|\left|\delta V\right|_{q\text{-var}}\right\|_{\infty;k,\infty} ∣ δ V ∣ q -var ∞ ; k , ∞ with any k ≥ 1 k\ge1 k ≥ 1 is dominated by the version with k = 1 k=1 k = 1 .
Lemma 3.2. For a one-parameter continuous process V V V , assume ∥ ∣ δ V ∣ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] < ∞ \left\|\left|\delta V\right|_{q\text{-var}}\right\|_{\infty;1,\infty;[s,t]}<\infty ∣ δ V ∣ q -var ∞ ; 1 , ∞ ; [ s , t ] < ∞ . Then its q q q -variation, { ∣ δ V ∣ q -var ; [ s , r ] } r ∈ [ s , t ] \{\left|\delta V\right|_{q\text{-var};[s,r]}\}_{r\in[s,t]} { ∣ δ V ∣ q -var ; [ s , r ] } r ∈ [ s , t ] , is a BMO-process on [ s , t ] [s,t] [ s , t ] . Moreover, we have for every r ∈ [ s , t ] r\in[s,t] r ∈ [ s , t ] and k ≥ 1 k\ge1 k ≥ 1 ,
∥ ∣ δ V ∣ q -var ; [ r , t ] | F r ∥ k , ∞ ≤ ∣ k ! ∣ 1 / k ∥ ∣ δ V ∣ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] . (14) \left\|\left|\delta V\right|_{q\text{-var};[r,t]}\middle|\mathcal{F}_r\right\|_{k,\infty}\le|k!|^{1/k}\left\|\left|\delta V\right|_{q\text{-var}}\right\|_{\infty;1,\infty;[s,t]}. \tag*{(14)} ∣ δ V ∣ q -var ; [ r , t ] F r k , ∞ ≤ ∣ k ! ∣ 1/ k ∣ δ V ∣ q -var ∞ ; 1 , ∞ ; [ s , t ] . (14)
Consequently, it holds
∥ ∣ δ V ∣ q -var ∥ ∞ ; k , ∞ ; [ s , t ] ≤ ∣ k ! ∣ 1 / k ∥ ∣ δ V ∣ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] . \left\|\left|\delta V\right|_{q\text{-var}}\right\|_{\infty;k,\infty;[s,t]}\le|k!|^{1/k}\left\|\left|\delta V\right|_{q\text{-var}}\right\|_{\infty;1,\infty;[s,t]}. ∣ δ V ∣ q -var ∞ ; k , ∞ ; [ s , t ] ≤ ∣ k ! ∣ 1/ k ∣ δ V ∣ q -var ∞ ; 1 , ∞ ; [ s , t ] .
Remark 3.3. Note that the above lemma is the key observation to solve the problem of degenerating integrability when one considers a composition of rough semimartingales with smooth functions (see eq:3.33 in Proposition 3.10). When q > 2 q>2 q > 2 , the above equivalence of norms for continuous martingales also follows from (A.2) in Lemma A.1.
Proof. For r ∈ [ s , t ] r \in[s,t] r ∈ [ s , t ] , denote A u r : = ∥ δ V ∥ q -var ; [ r , u ] A_u^r := \lVert\delta V \rVert_{q\text{-var};[r,u]} A u r := ∥ δ V ∥ q -var ; [ r , u ] . Then, a.s., u ↦ E u [ δ A u , t r ] = E u [ δ A r , t r ] − δ A r , u r u \mapsto\mathbb{E}_u[\delta A_{u,t}^r] = \mathbb{E}_u[\delta A_{r,t}^r] - \delta A_{r,u}^r u ↦ E u [ δ A u , t r ] = E u [ δ A r , t r ] − δ A r , u r is continuous. Moreover, for any stopping time τ ∈ [ r , t ] \tau\in[r,t] τ ∈ [ r , t ] , E τ [ δ A τ , t r ] = E τ [ δ A r , t r ] − A r , τ r = E u [ δ A u , t r ] ∣ u = τ \mathbb{E}_\tau[\delta A_{\tau,t}^r] = \mathbb{E}_\tau[\delta A_{r,t}^r] - A_{r,\tau}^r = \mathbb{E}_u[\delta A_{u,t}^r]\big|_{u=\tau} E τ [ δ A τ , t r ] = E τ [ δ A r , t r ] − A r , τ r = E u [ δ A u , t r ] u = τ . Hence,
∥ E τ [ A t r − A τ − r ] ∥ ∞ = ∥ E τ [ δ A τ , t r ] ∥ ∞ ≤ sup u ∈ [ r , t ] ∥ E u [ δ A u , t r ] ∥ ∞ . \lVert\mathbb{E}_\tau[A_t^r-A_{\tau-}^r]\rVert_\infty= \lVert\mathbb{E}_\tau[\delta A_{\tau,t}^r]\rVert_\infty\leq\sup_{u\in[r,t]}\lVert\mathbb{E}_u[\delta A_{u,t}^r]\rVert_\infty. ∥ E τ [ A t r − A τ − r ] ∥ ∞ = ∥ E τ [ δ A τ , t r ] ∥ ∞ ≤ u ∈ [ r , t ] sup ∥ E u [ δ A u , t r ] ∥ ∞ .
Furthermore, noting that δ A u , t r = ∥ δ V ∥ q -var ; [ r , t ] − ∥ δ V ∥ q -var ; [ r , u ] ≤ ∥ δ V ∥ q -var ; [ u , t ] \delta A_{u,t}^r = \lVert\delta V\rVert_{q\text{-var};[r,t]}-\lVert\delta V\rVert_{q\text{-var};[r,u]} \leq\lVert\delta V\rVert_{q\text{-var};[u,t]} δ A u , t r = ∥ δ V ∥ q -var ; [ r , t ] − ∥ δ V ∥ q -var ; [ r , u ] ≤ ∥ δ V ∥ q -var ; [ u , t ] from the subadditivity of the q q q -variation norm, and that [ r , t ] ⊆ [ s , t ] [r,t]\subseteq[s,t] [ r , t ] ⊆ [ s , t ] , we obtain that for all u ∈ [ r , t ] u\in[r,t] u ∈ [ r , t ] ,
∥ E u [ δ A u , t r ] ∥ ∞ ≤ ∥ E u [ ∥ δ V ∥ q -var ; [ u , t ] ] ∥ ∞ ≤ ∥ ∥ δ V ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] = : C . \lVert\mathbb{E}_u[\delta A_{u,t}^r]\rVert_\infty\leq\lVert\mathbb{E}_u[\lVert\delta V\rVert_{q\text{-var};[u,t]}]\rVert_\infty\leq\lVert\lVert\delta V\rVert_{q\text{-var}}\rVert_{\infty;1,\infty;[s,t]}=:C. ∥ E u [ δ A u , t r ] ∥ ∞ ≤ ∥ E u [∥ δ V ∥ q -var ; [ u , t ] ] ∥ ∞ ≤ ∥∥ δ V ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] =: C .
Consequently, ∥ E τ [ A t r − A τ − r ] ∥ ∞ ≤ C \lVert\mathbb{E}_\tau[A_t^r-A_{\tau-}^r]\rVert_\infty\leq C ∥ E τ [ A t r − A τ − r ] ∥ ∞ ≤ C for any stopping time τ ∈ [ r , t ] \tau\in[r,t] τ ∈ [ r , t ] . By Lemma 3.1 with τ ′ = r \tau'=r τ ′ = r ,
∥ ∥ δ V ∥ q -var ; [ r , t ] ∥ F r ; k , ∞ = ∥ δ A r , t r ∥ F r ; k , ∞ ≤ ∣ k ! ∣ 1 / k C , \lVert\lVert\delta V\rVert_{q\text{-var};[r,t]}\rVert_{\mathcal{F}_r;k,\infty}=\lVert\delta A_{r,t}^r\rVert_{\mathcal{F}_r;k,\infty}\leq|k!|^{1/k}C, ∥∥ δ V ∥ q -var ; [ r , t ] ∥ F r ; k , ∞ = ∥ δ A r , t r ∥ F r ; k , ∞ ≤ ∣ k ! ∣ 1/ k C ,
which implies (14) . The other inequality follows by taking the supremum over time. □
Estimates for Rough Semimartingales
Unless otherwise specified, throughout this subsection, we assume X ∈ C p -var ( [ 0 , T ] , V ) X\in\mathcal{C}^{p\text{-var}}([0,T],V) X ∈ C p -var ([ 0 , T ] , V ) and ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( [ 0 , T ] , V 1 ) (Y,Y')\in\operatorname{RSM}_{X}^{(p,p')\text{-var}}([0,T],V_1) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , V 1 ) .
The following lemma concerns the fact that the composition of G ( ⋅ ) G(\cdot) G ( ⋅ ) with an RSM remains an RSM. The following general integral form turns out to be helpful for later purposes (see Theorem 4.3).
Lemma 3.4. Assume ( p , p ′ ) (p,p') ( p , p ′ ) satisfies (3.1), and γ − 2 ≥ p / p ′ \gamma-2\geq p/p' γ − 2 ≥ p / p ′ . Let G ∈ C γ − 1 ( R , R d ) G\in\mathcal{C}^{\gamma-1}(\mathbb{R},\mathbb{R}^d) G ∈ C γ − 1 ( R , R d ) . Assume both ( Y , Y ′ ) (Y,Y') ( Y , Y ′ ) and ( Y ‾ , Y ‾ ′ ) (\overline{Y},\overline{Y}') ( Y , Y ′ ) belong to RSM X ( p , p ′ ) -var ( [ 0 , T ] , R ) \operatorname{RSM}_{X}^{(p,p')\text{-var}}([0,T],\mathbb{R}) RSM X ( p , p ′ ) -var ([ 0 , T ] , R ) . Denote
α t : = ∫ 0 1 G ( ( 1 − λ ) Y t + λ Y ‾ t ) d λ , α t ′ : = ∫ 0 1 D G ( ( 1 − λ ) Y t + λ Y ‾ t ) ⋅ ( ( 1 − λ ) Y t ′ + λ Y ‾ t ′ ) d λ . (15) \alpha_t:=\int_0^1G((1-\lambda)Y_t+\lambda\overline{Y}_t)\,d\lambda,\qquad
\alpha'_t:=\int_0^1DG((1-\lambda)Y_t+\lambda\overline{Y}_t)\cdot((1-\lambda)Y'_t+\lambda\overline{Y}'_t)\,d\lambda. \tag*{(15)} α t := ∫ 0 1 G (( 1 − λ ) Y t + λ Y t ) d λ , α t ′ := ∫ 0 1 D G (( 1 − λ ) Y t + λ Y t ) ⋅ (( 1 − λ ) Y t ′ + λ Y t ′ ) d λ . (15)
Then, we have ( α , α ′ ) ∈ RSM X ( p , p ′ ) -var ( [ 0 , T ] , R d ) (\alpha,\alpha')\in\operatorname{RSM}_{X}^{(p,p')\text{-var}}([0,T],\mathbb{R}^d) ( α , α ′ ) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , R d ) with the local martingale part given by
M t α = ∫ 0 1 { ∫ 0 t D G ( ( 1 − λ ) Y r + λ Y ‾ r ) [ ( 1 − λ ) d M r Y + λ d M r Y ‾ ] } d λ . (16) M_t^\alpha=\int_0^1\left\{\int_0^tDG((1-\lambda)Y_r+\lambda\overline{Y}_r)\left[(1-\lambda)\,dM_r^Y+\lambda\,dM_r^{\overline{Y}}\right]\right\}\,d\lambda. \tag*{(16)} M t α = ∫ 0 1 { ∫ 0 t D G (( 1 − λ ) Y r + λ Y r ) [ ( 1 − λ ) d M r Y + λ d M r Y ] } d λ . (16)
In particular, when γ = 3 \gamma=3 γ = 3 and Y = Y ‾ Y=\overline{Y} Y = Y , the pair ( G ( Y ) , D G ( Y ) Y ′ ) (G(Y),DG(Y)Y') ( G ( Y ) , D G ( Y ) Y ′ ) is an RSM.
Proof. Noting that p / p ′ ∈ ( 0 , 1 ] p/p'\in(0,1] p / p ′ ∈ ( 0 , 1 ] , it suffices to prove the result further assuming γ ≤ 3 \gamma\leq3 γ ≤ 3 . For λ ∈ [ 0 , 1 ] \lambda\in[0,1] λ ∈ [ 0 , 1 ] and t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] , denote Y t λ : = ( 1 − λ ) Y t + λ Y ‾ t Y_t^\lambda:=(1-\lambda)Y_t+\lambda\overline{Y}_t Y t λ := ( 1 − λ ) Y t + λ Y t , ( Y λ ) t ′ : = ( 1 − λ ) Y t ′ + λ Y ‾ t ′ (Y^\lambda)'_t:=(1-\lambda)Y'_t+\lambda\overline{Y}'_t ( Y λ ) t ′ := ( 1 − λ ) Y t ′ + λ Y t ′ ,
G t λ : = G ( Y t λ ) , ( G λ ) t ′ : = D G ( Y t λ ) ⋅ ( Y λ ) t ′ . (17) G_t^\lambda:=G(Y_t^\lambda),\qquad(G^\lambda)'_t:=DG(Y_t^\lambda)\cdot(Y^\lambda)'_t. \tag*{(17)} G t λ := G ( Y t λ ) , ( G λ ) t ′ := D G ( Y t λ ) ⋅ ( Y λ ) t ′ . (17)
For ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , denote
D s , t λ : = δ G s , t λ − ( G λ ) s ′ δ X s , t − ∫ s t D G ( Y r λ ) d M r Y λ . (18) D_{s,t}^\lambda:=\delta G_{s,t}^\lambda-(G^\lambda)'_s\delta X_{s,t}-\int_s^tDG(Y_r^\lambda)\,dM_r^{Y^\lambda}. \tag*{(18)} D s , t λ := δ G s , t λ − ( G λ ) s ′ δ X s , t − ∫ s t D G ( Y r λ ) d M r Y λ . (18)
Step 1. In this step, we prove that ( G λ , ( G λ ) ′ ) (G^\lambda,(G^\lambda)') ( G λ , ( G λ ) ′ ) is an RSM. Without loss of generality, assume G ( ⋅ ) ∈ C b γ − 1 G(\cdot)\in\mathcal{C}_b^{\gamma-1} G ( ⋅ ) ∈ C b γ − 1 and that every term below is bounded and hence has finite k k k -moments for any given k ≥ 1 k\geq1 k ≥ 1 ; otherwise, the standard localization argument applies. For the simplicity of notations, denote C G : = ∥ G ( ⋅ ) ∥ C b γ − 1 C_G:=\lVert G(\cdot)\rVert_{\mathcal{C}_b^{\gamma-1}} C G := ∥ G ( ⋅ ) ∥ C b γ − 1 .
First, by the boundedness and the ( γ − 2 ) (\gamma-2) ( γ − 2 ) -Hölder continuity of D G DG D G , for t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] it holds that
∥ δ ( G λ ) ′ ∥ p ′ -var ; [ t , T ] ≤ C G ( ∥ δ ( Y λ ) ′ ∥ p ′ -var ; [ t , T ] + ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ ( Y λ ) ′ ∥ ∞ ) , ∥ ( G λ ) ′ ∥ ∞ ≤ C G ∥ ( Y λ ) ′ ∥ ∞ , \lVert\delta(G^\lambda)'\rVert_{p'\text{-var};[t,T]}\leq C_G\left(\lVert\delta(Y^\lambda)'\rVert_{p'\text{-var};[t,T]}+\lVert\delta Y^\lambda\rVert_{p\text{-var};[t,T]}^{\gamma-2}\lVert(Y^\lambda)'\rVert_\infty\right),\qquad\lVert(G^\lambda)'\rVert_\infty\leq C_G\lVert(Y^\lambda)'\rVert_\infty, ∥ δ ( G λ ) ′ ∥ p ′ -var ; [ t , T ] ≤ C G ( ∥ δ ( Y λ ) ′ ∥ p ′ -var ; [ t , T ] + ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥( Y λ ) ′ ∥ ∞ ) , ∥( G λ ) ′ ∥ ∞ ≤ C G ∥( Y λ ) ′ ∥ ∞ ,
where in the first inequality we used the condition p ′ ( γ − 2 ) ≥ p p'(\gamma-2) \ge p p ′ ( γ − 2 ) ≥ p . Clearly ∥ ( G λ ) ′ ′ ∥ ∞ \|(G^\lambda)''\|_\infty ∥ ( G λ ) ′′ ∥ ∞ is uniformly bounded in λ ∈ [ 0 , 1 ] \lambda\in[0,1] λ ∈ [ 0 , 1 ] . Furthermore, noting that Y λ Y^\lambda Y λ (( Y λ ) ′ (Y^\lambda)' ( Y λ ) ′ resp.) is a convex combination of Y Y Y and Y ˉ \bar{Y} Y ˉ (Y ′ Y' Y ′ and Y ˉ ′ \bar{Y}' Y ˉ ′ resp.), we have that for all λ ∈ [ 0 , 1 ] \lambda\in[0,1] λ ∈ [ 0 , 1 ] ,
∥ δ ( G λ ) ′ ′ ∥ p ′ -var ; [ t , T ] ≤ C G ( ∥ δ Y ′ ∥ p ′ -var ; [ t , T ] + ∥ δ Y ˉ ′ ∥ p ′ -var ; [ t , T ] + ( ∥ δ Y ∥ p -var ; [ t , T ] + ∥ δ Y ˉ ∥ p -var ; [ t , T ] ) ( ∥ Y ′ ∥ ∞ + ∥ Y ˉ ′ ∥ ∞ ) ) . (19) \|\delta(G^\lambda)''\|_{p'\text{-var};[t,T]} \le C_G\left(\|\delta Y'\|_{p'\text{-var};[t,T]}+\|\delta\bar{Y}'\|_{p'\text{-var};[t,T]}+\left(\|\delta Y\|_{p\text{-var};[t,T]}+\|\delta\bar{Y}\|_{p\text{-var};[t,T]}\right)(\|Y'\|_\infty+\|\bar{Y}'\|_\infty)\right). \tag*{(19)} ∥ δ ( G λ ) ′′ ∥ p ′ -var ; [ t , T ] ≤ C G ( ∥ δ Y ′ ∥ p ′ -var ; [ t , T ] + ∥ δ Y ˉ ′ ∥ p ′ -var ; [ t , T ] + ( ∥ δ Y ∥ p -var ; [ t , T ] + ∥ δ Y ˉ ∥ p -var ; [ t , T ] ) ( ∥ Y ′ ∥ ∞ + ∥ Y ˉ ′ ∥ ∞ ) ) . (19)
This implies the uniform boundedness ∥ δ ( G λ ) ′ ′ ∥ p ′ -var \|\delta(G^\lambda)''\|_{p'\text{-var}} ∥ δ ( G λ ) ′′ ∥ p ′ -var , due to the finiteness of ∥ δ Y ′ ∥ p ′ -var \|\delta Y'\|_{p'\text{-var}} ∥ δ Y ′ ∥ p ′ -var , ∥ δ Y ˉ ′ ∥ p ′ -var \|\delta\bar{Y}'\|_{p'\text{-var}} ∥ δ Y ˉ ′ ∥ p ′ -var , ∥ δ Y ∥ p -var \|\delta Y\|_{p\text{-var}} ∥ δ Y ∥ p -var , and ∥ δ Y ˉ ∥ p -var \|\delta\bar{Y}\|_{p\text{-var}} ∥ δ Y ˉ ∥ p -var .
Next, by the definition of P λ P^\lambda P λ we have
P s , t λ = ∫ 0 1 [ D G ( Y s λ + θ ⋅ δ Y s , t λ ) − D G ( Y s λ ) ] d θ ⋅ δ M s , t Y λ + ∫ 0 1 D G ( Y s λ + θ ⋅ δ Y s , t λ ) d θ ⋅ P s , t Y λ + ∫ 0 1 [ D G ( Y s λ + θ ⋅ δ Y s , t λ ) − D G ( Y s λ ) ] d θ ⋅ ( ( Y λ ) s ′ δ X s , t − ∫ s t [ D G ( Y r λ ) − D G ( Y s λ ) ] d M r Y λ ) = : K s , t λ ; 1 + K s , t λ ; 2 + K s , t λ ; 3 + K s , t λ ; 4 . (20) \begin{aligned}
P^\lambda_{s,t}={}&\int_0^1\left[DG(Y^\lambda_s+\theta\cdot\delta Y^\lambda_{s,t})-DG(Y^\lambda_s)\right]\,d\theta\cdot\delta M^{Y^\lambda}_{s,t}+\int_0^1 DG(Y^\lambda_s+\theta\cdot\delta Y^\lambda_{s,t})\,d\theta\cdot P^{Y^\lambda}_{s,t}\\
&+\int_0^1\left[DG(Y^\lambda_s+\theta\cdot\delta Y^\lambda_{s,t})-DG(Y^\lambda_s)\right]\,d\theta\cdot\left((Y^\lambda)'_s\delta X_{s,t}-\int_s^t[DG(Y^\lambda_r)-DG(Y^\lambda_s)]\,dM^{Y^\lambda}_r\right)\\
&=:K^{\lambda;1}_{s,t}+K^{\lambda;2}_{s,t}+K^{\lambda;3}_{s,t}+K^{\lambda;4}_{s,t}. \tag*{(20)}
\end{aligned} P s , t λ = ∫ 0 1 [ D G ( Y s λ + θ ⋅ δ Y s , t λ ) − D G ( Y s λ ) ] d θ ⋅ δ M s , t Y λ + ∫ 0 1 D G ( Y s λ + θ ⋅ δ Y s , t λ ) d θ ⋅ P s , t Y λ + ∫ 0 1 [ D G ( Y s λ + θ ⋅ δ Y s , t λ ) − D G ( Y s λ ) ] d θ ⋅ ( ( Y λ ) s ′ δ X s , t − ∫ s t [ D G ( Y r λ ) − D G ( Y s λ )] d M r Y λ ) =: K s , t λ ; 1 + K s , t λ ; 2 + K s , t λ ; 3 + K s , t λ ; 4 . (20)
For K λ ; 1 K^{\lambda;1} K λ ; 1 and K λ ; 3 K^{\lambda;3} K λ ; 3 , since D G ( ⋅ ) DG(\cdot) D G ( ⋅ ) is ( γ − 2 ) (\gamma-2) ( γ − 2 ) -Hölder continuous, we have
∫ 0 1 ∣ D G ( Y s λ + θ ⋅ δ Y s , t λ ) − D G ( Y s λ ) ∣ d θ ≤ ∥ G ( ⋅ ) ∥ C b γ − 1 ∣ δ Y s , t λ ∣ γ − 2 = C G ∣ δ Y s , t λ ∣ γ − 2 . \int_0^1\left|DG(Y^\lambda_s+\theta\cdot\delta Y^\lambda_{s,t})-DG(Y^\lambda_s)\right|d\theta\le\|G(\cdot)\|_{C_b^{\gamma-1}}|\delta Y^\lambda_{s,t}|^{\gamma-2}=C_G|\delta Y^\lambda_{s,t}|^{\gamma-2}. ∫ 0 1 D G ( Y s λ + θ ⋅ δ Y s , t λ ) − D G ( Y s λ ) d θ ≤ ∥ G ( ⋅ ) ∥ C b γ − 1 ∣ δ Y s , t λ ∣ γ − 2 = C G ∣ δ Y s , t λ ∣ γ − 2 .
Thus, noting that p p ′ p + p ′ ≥ p γ − 1 \frac{pp'}{p+p'}\ge\frac{p}{\gamma-1} p + p ′ p p ′ ≥ γ − 1 p , by Hölder’s inequality we have for t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] ,
∥ K λ ; 1 ∥ p p ′ p + p ′ -var ; [ t , T ] ≤ ∥ K λ ; 1 ∥ p γ − 1 -var ; [ t , T ] ≤ C G ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ δ M Y λ ∥ p -var ; [ t , T ] . (21) \|K^{\lambda;1}\|_{\frac{pp'}{p+p'}\text{-var};[t,T]}\le\|K^{\lambda;1}\|_{\frac{p}{\gamma-1}\text{-var};[t,T]}\le C_G\|\delta Y^\lambda\|_{p\text{-var};[t,T]}^{\gamma-2}\|\delta M^{Y^\lambda}\|_{p\text{-var};[t,T]}. \tag*{(21)} ∥ K λ ; 1 ∥ p + p ′ p p ′ -var ; [ t , T ] ≤ ∥ K λ ; 1 ∥ γ − 1 p -var ; [ t , T ] ≤ C G ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ δ M Y λ ∥ p -var ; [ t , T ] . (21)
and similarly,
∥ K λ ; 3 ∥ p p ′ p + p ′ -var ; [ t , T ] ≤ ∥ K λ ; 3 ∥ p γ − 1 -var ; [ t , T ] ≤ C G ∥ ( Y λ ) ′ ∥ ∞ ; [ t , T ] ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ δ X ∥ p -var ; [ t , T ] . (22) \|K^{\lambda;3}\|_{\frac{pp'}{p+p'}\text{-var};[t,T]}\le\|K^{\lambda;3}\|_{\frac{p}{\gamma-1}\text{-var};[t,T]}\le C_G\|(Y^\lambda)'\|_{\infty;[t,T]}\|\delta Y^\lambda\|_{p\text{-var};[t,T]}^{\gamma-2}\|\delta X\|_{p\text{-var};[t,T]}. \tag*{(22)} ∥ K λ ; 3 ∥ p + p ′ p p ′ -var ; [ t , T ] ≤ ∥ K λ ; 3 ∥ γ − 1 p -var ; [ t , T ] ≤ C G ∥ ( Y λ ) ′ ∥ ∞ ; [ t , T ] ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ δ X ∥ p -var ; [ t , T ] . (22)
For K λ ; 2 K^{\lambda;2} K λ ; 2 , by the boundedness of ∥ D G ( ⋅ ) ∥ ∞ \|DG(\cdot)\|_\infty ∥ D G ( ⋅ ) ∥ ∞ , we have
∥ K λ ; 2 ∥ p p ′ p + p ′ -var ; [ t , T ] ≤ ∥ D G ( ⋅ ) ∥ ∞ ⋅ ∥ P Y λ ∥ p p ′ p + p ′ -var ; [ t , T ] . (23) \|K^{\lambda;2}\|_{\frac{pp'}{p+p'}\text{-var};[t,T]}\le\|DG(\cdot)\|_\infty\cdot\|P^{Y^\lambda}\|_{\frac{pp'}{p+p'}\text{-var};[t,T]}. \tag*{(23)} ∥ K λ ; 2 ∥ p + p ′ p p ′ -var ; [ t , T ] ≤ ∥ D G ( ⋅ ) ∥ ∞ ⋅ ∥ P Y λ ∥ p + p ′ p p ′ -var ; [ t , T ] . (23)
To estimate K λ ; 4 K^{\lambda;4} K λ ; 4 , for k ≥ 1 k\ge1 k ≥ 1 , by Lemma A.2 with r = p p ′ p + p ′ r=\frac{pp'}{p+p'} r = p + p ′ p p ′ , r 1 = p γ − 2 r_1=\frac{p}{\gamma-2} r 1 = γ − 2 p , k 1 = 2 k k_1=2k k 1 = 2 k , k 0 = 2 k k_0=2k k 0 = 2 k ,
∥ ∥ K λ ; 4 ∥ p p ′ p + p ′ -var ; [ t , T ] ∥ F t ≲ k , γ ∥ ∥ δ D G ( Y λ ) ∥ p γ − 2 -var ; [ t , T ] ∥ F t 2 k ∥ δ ⟨ M Y λ ⟩ t , T 1 2 ∥ F t 2 k ≤ C G ∥ ∥ δ Y λ ∥ p -var ; [ t , T ] ∥ F t 2 k γ − 2 ∥ δ ⟨ M Y λ ⟩ t , T 1 2 ∥ F t 2 k . (24) \begin{aligned}
\left\|\left\|K^{\lambda;4}\right\|_{\frac{pp'}{p+p'}\text{-var};[t,T]}\right\|_{\mathcal{F}_t}\lesssim_{k,\gamma}\left\|\left\|\delta DG(Y^\lambda)\right\|_{\frac{p}{\gamma-2}\text{-var};[t,T]}\right\|_{\mathcal{F}_t^{2k}}\left\|\delta\left\langle M^{Y^\lambda}\right\rangle_{t,T}^{\frac12}\right\|_{\mathcal{F}_t^{2k}} \\
&\le C_G\left\|\left\|\delta Y^\lambda\right\|_{p\text{-var};[t,T]}\right\|_{\mathcal{F}_t^{2k}}^{\gamma-2}\left\|\delta\left\langle M^{Y^\lambda}\right\rangle_{t,T}^{\frac12}\right\|_{\mathcal{F}_t^{2k}}. \tag*{(24)}
\end{aligned} K λ ; 4 p + p ′ p p ′ -var ; [ t , T ] F t ≲ k , γ δ D G ( Y λ ) γ − 2 p -var ; [ t , T ] F t 2 k δ ⟨ M Y λ ⟩ t , T 2 1 F t 2 k ≤ C G δ Y λ p -var ; [ t , T ] F t 2 k γ − 2 δ ⟨ M Y λ ⟩ t , T 2 1 F t 2 k . (24)
Combining (20) , (21) , (22) , (23) , and (24) , we have that a.s., P λ P^\lambda P λ has finite p p ′ p + p ′ \frac{pp'}{p+p'} p + p ′ p p ′ -variation. This together with the finiteness of ∥ δ G λ ∥ p -var + ∥ δ ( G λ ) ′ ∥ p ′ -var + ∥ ( G λ ) ′ ∥ ∞ \|\delta G^\lambda\|_{p\text{-var}}+\|\delta(G^\lambda)'\|_{p'\text{-var}}+\|(G^\lambda)'\|_\infty ∥ δ G λ ∥ p -var + ∥ δ ( G λ ) ′ ∥ p ′ -var + ∥ ( G λ ) ′ ∥ ∞ and the decomposition (18) implies that ( G λ , ( G λ ) ′ ) ∈ RSM X ( p , p ′ ) -var (G^\lambda,(G^\lambda)')\in\operatorname{RSM}^{(p,p')\text{-var}}_X ( G λ , ( G λ ) ′ ) ∈ RSM X ( p , p ′ ) -var .
Step 2. In this step, we prove that ( α , α ′ ) (\alpha,\alpha') ( α , α ′ ) is an RSM. Since α ′ \alpha' α ′ is given by the integral of ( G λ ) ′ (G^\lambda)' ( G λ ) ′ , which is uniformly bounded in λ ∈ [ 0 , 1 ] \lambda\in[0,1] λ ∈ [ 0 , 1 ] , ∥ α ′ ∥ ∞ \|\alpha'\|_\infty ∥ α ′ ∥ ∞ is bounded. Furthermore, in view of the inequality (19) , Minkowski’s inequality yields
\|\delta\alpha'\|_{p'\text{-var};[t,T]}\le C_G\times\text{RHS of }[](#eq:3.9). \tag*{(25)}
Next, we establish the RSM decomposition of ( α , α ′ ) (\alpha,\alpha') ( α , α ′ ) . Note that the following equality holds:
δ α s , t − α s ′ δ X s , t = ∫ 0 1 P s , t λ d λ + ∫ 0 1 δ M s , t G λ d λ = : P s , t + δ M s , t . \delta\alpha_{s,t}-\alpha'_s\delta X_{s,t}=\int_0^1 P^\lambda_{s,t}\,d\lambda+\int_0^1\delta M^{G^\lambda}_{s,t}\,d\lambda=:P_{s,t}+\delta M_{s,t}. δ α s , t − α s ′ δ X s , t = ∫ 0 1 P s , t λ d λ + ∫ 0 1 δ M s , t G λ d λ =: P s , t + δ M s , t .
where P s , t : = ∫ 0 1 P s , t λ d λ P_{s,t}:=\int_0^1P^\lambda_{s,t}\,d\lambda P s , t := ∫ 0 1 P s , t λ d λ and M t : = ∫ 0 1 M t G λ d λ M_t:=\int_0^1M^{G^\lambda}_t\,d\lambda M t := ∫ 0 1 M t G λ d λ . By Fubini’s theorem, we have, noting M t G λ = ∫ 0 t D G ( Y r λ ) d M r Y λ M^{G^\lambda}_t=\int_0^tDG(Y^\lambda_r)dM^{Y^\lambda}_r M t G λ = ∫ 0 t D G ( Y r λ ) d M r Y λ from (18) ,
M t = ∫ 0 1 E t [ ∫ 0 T D G ( Y r λ ) d M r Y λ ] d λ = E t [ ∫ 0 1 ∫ 0 T D G ( Y r λ ) d M r Y λ d λ ] = E t [ M T ] . M_t=\int_0^1\mathbb{E}_t\left[\int_0^T DG(Y^\lambda_r)dM^{Y^\lambda}_r\right]d\lambda=\mathbb{E}_t\left[\int_0^1\int_0^T DG(Y^\lambda_r)dM^{Y^\lambda}_r\,d\lambda\right]=\mathbb{E}_t[M_T]. M t = ∫ 0 1 E t [ ∫ 0 T D G ( Y r λ ) d M r Y λ ] d λ = E t [ ∫ 0 1 ∫ 0 T D G ( Y r λ ) d M r Y λ d λ ] = E t [ M T ] .
Therefore, M t M_t M t is a continuous martingale, and then ( s , t ) ↦ P s , t (s,t)\mapsto P_{s,t} ( s , t ) ↦ P s , t is continuous a.s. Moreover, similar to ∥ δ α ′ ∥ p ′ -var \lVert\delta\alpha'\rVert_{p'\text{-var}} ∥ δ α ′ ∥ p ′ -var , Minkowski’s inequality again implies the finiteness of the p p ′ p + p ′ \frac{pp'}{p+p'} p + p ′ p p ′ -variation of P P P . Consequently, we get that ( α , α ′ ) ∈ RSM X ( p , p ′ ) -var (\alpha,\alpha')\in\operatorname{RSM}^{(p,p')\text{-var}}_X ( α , α ′ ) ∈ RSM X ( p , p ′ ) -var with M α = M M^\alpha=M M α = M and P α = P P^\alpha=P P α = P . □
Below we obtain upper bounds for the pathwise remainder P G ( Y ) P^{G(Y)} P G ( Y ) as well as for M G ( Y ) M^{G(Y)} M G ( Y ) . Recall q ′ ′ 2 : = q q ′ q + q ′ \frac{q''}{2}:=\frac{qq'}{q+q'} 2 q ′′ := q + q ′ q q ′ .
Proposition 3.5. Assume that (11) and q < min { 4 , q ′ } q<\min\{4,q'\} q < min { 4 , q ′ } hold. Let G ∈ C b 2 ( V 1 , V 2 ) G\in C_b^2(V_1,V_2) G ∈ C b 2 ( V 1 , V 2 ) with ∥ G ∥ C b 2 ≤ L \lVert G\rVert_{C_b^2}\leq L ∥ G ∥ C b 2 ≤ L for some constant L > 0 L>0 L > 0 . Fix k ≥ 1 k\geq1 k ≥ 1 and set k 0 : = 2 k q − 2 k_0:=\frac{2k}{q-2} k 0 := q − 2 2 k . Then, we have that for ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ ,
∥ ∥ P G ( Y ) ∥ q ′ ′ 2 -var ; [ s , t ] ∥ F s ∥ k ≲ k , L ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ 4 − q 2 ∥ δ M Y ∥ 1 2 , s , t ∥ F s ∥ k 0 + ∥ ∥ P Y ∥ q ′ ′ 2 -var ; [ s , t ] ∥ F s ∥ k + ∥ δ X ∥ p -var ; [ s , t ] ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ q q ′ ∥ Y ′ ∥ ∞ , (26) \left\lVert\lVert P^{G(Y)}\rVert_{\frac{q''}{2}\text{-var};[s,t]}\right\rVert_{\mathcal{F}_s\lVert_k}\lesssim_{k,L}\lVert\delta Y\rVert_{q\text{-var};[s,t]}\lVert\mathcal{F}_s\lVert_{\frac{4-q}{2}}\lVert\delta M^Y\rVert_{\frac{1}{2},s,t}\lVert\mathcal{F}_s\lVert_{k_0}+\lVert\lVert P^Y\rVert_{\frac{q''}{2}\text{-var};[s,t]}\rVert_{\mathcal{F}_s\lVert_k}
+\lVert\delta X\rVert_{p\text{-var};[s,t]}\lVert\delta Y\rVert_{q\text{-var};[s,t]}\lVert\mathcal{F}_s\lVert_{\frac{q}{q'}}\lVert Y'\rVert_{\infty}, \tag*{(26)} ∥ P G ( Y ) ∥ 2 q ′′ -var ; [ s , t ] F s ∥ k ≲ k , L ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ 2 4 − q ∥ δ M Y ∥ 2 1 , s , t ∥ F s ∥ k 0 + ∥∥ P Y ∥ 2 q ′′ -var ; [ s , t ] ∥ F s ∥ k + ∥ δ X ∥ p -var ; [ s , t ] ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ q ′ q ∥ Y ′ ∥ ∞ , (26)
and
∥ δ ⟨ M G ( Y ) ⟩ s , t 1 2 ∥ F s ∥ k ≲ k , L ∥ δ ⟨ M Y ⟩ s , t 1 2 ∥ F s ∥ k . (27) \left\lVert\delta\langle M^{G(Y)}\rangle^{\frac12}_{s,t}\right\rVert_{\mathcal{F}_s\lVert_k}\lesssim_{k,L}\left\lVert\delta\langle M^Y\rangle^{\frac12}_{s,t}\right\rVert_{\mathcal{F}_s\lVert_k}. \tag*{(27)} δ ⟨ M G ( Y ) ⟩ s , t 2 1 F s ∥ k ≲ k , L δ ⟨ M Y ⟩ s , t 2 1 F s ∥ k . (27)
Proof. For ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , by standard calculus, we obtain
G ( Y t ) − G ( Y s ) − D G ( Y s ) Y s ′ δ X s , t = ∫ 0 1 D G ( Y s + λ ⋅ δ Y s , t ) d λ ⋅ δ Y s , t − ∫ 0 1 D G ( Y s ) d λ ( Y s ′ δ X s , t ) = ∫ 0 1 D G ( Y s + λ ⋅ δ Y s , t ) d λ ( δ M s , t Y + P s , t Y ) + ∫ 0 1 [ D G ( Y s + λ ⋅ δ Y s , t ) − D G ( Y s ) ] d λ ( Y s ′ δ X s , t ) , (28) \begin{aligned}
G(Y_t)-G(Y_s)-DG(Y_s)Y'_s\delta X_{s,t}
=\int_0^1DG(Y_s+\lambda\cdot\delta Y_{s,t})\,d\lambda\cdot\delta Y_{s,t}-\int_0^1DG(Y_s)\,d\lambda(Y'_s\delta X_{s,t}) \\
&=\int_0^1DG(Y_s+\lambda\cdot\delta Y_{s,t})\,d\lambda(\delta M^Y_{s,t}+P^Y_{s,t})+\int_0^1[DG(Y_s+\lambda\cdot\delta Y_{s,t})-DG(Y_s)]\,d\lambda(Y'_s\delta X_{s,t}), \tag*{(28)}
\end{aligned} G ( Y t ) − G ( Y s ) − D G ( Y s ) Y s ′ δ X s , t = ∫ 0 1 D G ( Y s + λ ⋅ δ Y s , t ) d λ ⋅ δ Y s , t − ∫ 0 1 D G ( Y s ) d λ ( Y s ′ δ X s , t ) = ∫ 0 1 D G ( Y s + λ ⋅ δ Y s , t ) d λ ( δ M s , t Y + P s , t Y ) + ∫ 0 1 [ D G ( Y s + λ ⋅ δ Y s , t ) − D G ( Y s )] d λ ( Y s ′ δ X s , t ) , (28)
where the last step is due to δ Y s , t = δ M s , t Y + P s , t Y + Y s ′ δ X s , t \delta Y_{s,t}=\delta M^Y_{s,t}+P^Y_{s,t}+Y'_s\delta X_{s,t} δ Y s , t = δ M s , t Y + P s , t Y + Y s ′ δ X s , t . For brevity, we denote M : = M G ( Y ) M:=M^{G(Y)} M := M G ( Y ) and P : = P G ( Y ) P:=P^{G(Y)} P := P G ( Y ) .
By (16) in Lemma 3.4 with Y = Y ˉ Y=\bar Y Y = Y ˉ , we have M t = ∫ 0 t D G ( Y r ) d M r Y ˉ M_t=\int_0^tDG(Y_r)dM^{\bar Y}_r M t = ∫ 0 t D G ( Y r ) d M r Y ˉ , t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] . For ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , by the definition of P P P , we get
\begin{aligned}
P_{s,t}:=\operatorname{RHS\ of\ }[](#eq:3.17)-\delta M_{s,t} \\
&=\int_0^1[DG(Y_s+\lambda\cdot\delta Y_{s,t})-DG(Y_s)]\,d\lambda\cdot\delta M^Y_{s,t}+\int_0^1DG(Y_s+\lambda\cdot\delta Y_{s,t})\,d\lambda\cdot P^Y_{s,t} \\
&+\int_0^1[DG(Y_s+\lambda\cdot\delta Y_{s,t})-DG(Y_s)]\,d\lambda(Y'_s\delta X_{s,t})-\int_s^t[DG(Y_r)-DG(Y_s)]\,dM^Y_r
=:K^1_{s,t}+K^2_{s,t}+K^3_{s,t}+K^4_{s,t}.
\end{aligned}
For K 1 K^1 K 1 , since D G ( ⋅ ) DG(\cdot) D G ( ⋅ ) and D 2 G ( ⋅ ) D^2G(\cdot) D 2 G ( ⋅ ) are bounded, we have for all ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ and ( u , v ) ∈ Δ [ s , t ] (u,v)\in\Delta_{[s,t]} ( u , v ) ∈ Δ [ s , t ] ,
∫ 0 1 ∣ D G ( Y u + λ ⋅ δ Y u , v ) − D G ( Y u ) ∣ d λ ≤ 2 q − 2 2 ∥ D G ( ⋅ ) ∥ ∞ q − 2 2 ∥ D 2 G ( ⋅ ) ∥ ∞ 4 − q 2 ⋅ ∣ δ Y u , v ∣ 4 − q 2 = : C G ∣ δ Y u , v ∣ 4 − q 2 , \int_0^1\left|DG(Y_u+\lambda\cdot\delta Y_{u,v})-DG(Y_u)\right|d\lambda\leq2^{\frac{q-2}{2}}\lVert DG(\cdot)\rVert_{\infty}^{\frac{q-2}{2}}\lVert D^2G(\cdot)\rVert_{\infty}^{\frac{4-q}{2}}\cdot|\delta Y_{u,v}|^{\frac{4-q}{2}}=:C_G|\delta Y_{u,v}|^{\frac{4-q}{2}}, ∫ 0 1 ∣ D G ( Y u + λ ⋅ δ Y u , v ) − D G ( Y u ) ∣ d λ ≤ 2 2 q − 2 ∥ D G ( ⋅ ) ∥ ∞ 2 q − 2 ∥ D 2 G ( ⋅ ) ∥ ∞ 2 4 − q ⋅ ∣ δ Y u , v ∣ 2 4 − q =: C G ∣ δ Y u , v ∣ 2 4 − q ,
where C G : = 2 q − 2 2 ∥ D G ( ⋅ ) ∥ ∞ q − 2 2 ∥ D 2 G ( ⋅ ) ∥ ∞ 4 − q 2 C_G := 2^{\frac{q-2}{2}}\|DG(\cdot)\|_{\infty}^{\frac{q-2}{2}}\|D^2G(\cdot)\|_{\infty}^{\frac{4-q}{2}} C G := 2 2 q − 2 ∥ D G ( ⋅ ) ∥ ∞ 2 q − 2 ∥ D 2 G ( ⋅ ) ∥ ∞ 2 4 − q . Hence, we have, with Q : = ( 1 2 − 1 q + 1 q ′ ) − 1 > 2 Q := \left(\frac{1}{2}-\frac{1}{q}+\frac{1}{q'}\right)^{-1}>2 Q := ( 2 1 − q 1 + q ′ 1 ) − 1 > 2 ,
∥ K 1 ∥ q ′ ′ 2 -var ; [ s , t ] ≤ C G ∥ ∣ δ Y ∣ 4 − q 2 ⋅ δ M Y ∥ q ′ ′ 2 -var ; [ s , t ] ≤ C G ∥ δ Y ∥ q -var ; [ s , t ] 4 − q 2 ∥ δ M Y ∥ Q -var ; [ s , t ] , (29) \|K^1\|_{\frac{q''}{2}\text{-var};[s,t]}\leq C_G\left\|\left|\delta Y\right|^{\frac{4-q}{2}}\cdot\delta M^Y\right\|_{\frac{q''}{2}\text{-var};[s,t]}\leq C_G\|\delta Y\|_{q\text{-var};[s,t]}^{\frac{4-q}{2}}\|\delta M^Y\|_{Q\text{-var};[s,t]}, \tag*{(29)} ∥ K 1 ∥ 2 q ′′ -var ; [ s , t ] ≤ C G ∣ δ Y ∣ 2 4 − q ⋅ δ M Y 2 q ′′ -var ; [ s , t ] ≤ C G ∥ δ Y ∥ q -var ; [ s , t ] 2 4 − q ∥ δ M Y ∥ Q -var ; [ s , t ] , (29)
where the second inequality follows from Hölder’s inequality. For K 2 K^2 K 2 , it is straightforward to get
∥ K 2 ∥ q ′ ′ 2 -var ; [ s , t ] ≤ ∥ D G ( ⋅ ) ∥ ∞ ⋅ ∥ P Y ∥ q ′ ′ 2 -var ; [ s , t ] . \|K^2\|_{\frac{q''}{2}\text{-var};[s,t]}\leq\|DG(\cdot)\|_{\infty}\cdot\|P^Y\|_{\frac{q''}{2}\text{-var};[s,t]}. ∥ K 2 ∥ 2 q ′′ -var ; [ s , t ] ≤ ∥ D G ( ⋅ ) ∥ ∞ ⋅ ∥ P Y ∥ 2 q ′′ -var ; [ s , t ] .
For K 3 K^3 K 3 , using the same argument leading to (29) , we obtain
∥ K 3 ∥ q ′ ′ 2 -var ; [ s , t ] ≲ ∥ D G ( ⋅ ) ∥ ∞ q ′ − q q ′ ∥ D 2 G ( ⋅ ) ∥ ∞ q q ′ ⋅ ∥ Y ′ ∥ ∞ ; [ s , t ] q q ′ ∥ δ Y ∥ q -var ; [ s , t ] q q ′ ∥ δ X ∥ q -var ; [ s , t ] . (30) \|K^3\|_{\frac{q''}{2}\text{-var};[s,t]}\lesssim\|DG(\cdot)\|_{\infty}^{\frac{q'-q}{q'}}\|D^2G(\cdot)\|_{\infty}^{\frac{q}{q'}}\cdot\|Y'\|_{\infty;[s,t]}^{\frac{q}{q'}}\|\delta Y\|_{q\text{-var};[s,t]}^{\frac{q}{q'}}\|\delta X\|_{q\text{-var};[s,t]}. \tag*{(30)} ∥ K 3 ∥ 2 q ′′ -var ; [ s , t ] ≲ ∥ D G ( ⋅ ) ∥ ∞ q ′ q ′ − q ∥ D 2 G ( ⋅ ) ∥ ∞ q ′ q ⋅ ∥ Y ′ ∥ ∞ ; [ s , t ] q ′ q ∥ δ Y ∥ q -var ; [ s , t ] q ′ q ∥ δ X ∥ q -var ; [ s , t ] . (30)
Recall k 0 = 2 k q − 2 k_0=\frac{2k}{q-2} k 0 = q − 2 2 k . Taking conditional k k k -moments in (29) –(30) and applying Hölder’s inequality, we have
∥ ∥ K 1 + K 2 + K 3 ∥ q ′ ′ 2 -var ; [ s , t ] | F s ∥ k ≲ k , L ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k 4 − q 2 ∥ ∥ δ M Y ∥ Q -var ; [ s , t ] | F s ∥ k 0 + ∥ δ X ∥ q -var ; [ s , t ] ∥ ∥ δ Y ∥ q -var ; [ s , t ] | F s ∥ k q q ′ ∥ Y ′ ∥ ∞ ≲ k ∥ ∥ δ Y ∥ q -var ; [ s , t ] | F s ∥ k 4 − q 2 ∥ δ ⟨ M Y ⟩ s , t 1 2 | F s ∥ k 0 + ∥ ∥ P Y ∥ q ′ ′ 2 -var ; [ s , t ] | F s ∥ k + ∥ δ X ∥ q -var ; [ s , t ] ∥ ∥ δ Y ∥ q -var ; [ s , t ] | F s ∥ k q q ′ ∥ Y ′ ∥ ∞ , (31) \begin{aligned}
\left\|\|K^1+K^2+K^3\|_{\frac{q''}{2}\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_k
&\lesssim_{k,L}\|\delta Y\|_{q\text{-var};[s,t]}\|\mathcal{F}_s\|_k^{\frac{4-q}{2}}\left\|\|\delta M^Y\|_{Q\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_{k_0}\\
&\quad+\|\delta X\|_{q\text{-var};[s,t]}\left\|\|\delta Y\|_{q\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_k^{\frac{q}{q'}}\|Y'\|_{\infty}\\
&\lesssim_k\left\|\|\delta Y\|_{q\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_k^{\frac{4-q}{2}}\left\|\delta\langle M^Y\rangle_{s,t}^{\frac{1}{2}}\middle|\mathcal{F}_s\right\|_{k_0}
+\left\|\|P^Y\|_{\frac{q''}{2}\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_k\\
&\quad+\|\delta X\|_{q\text{-var};[s,t]}\left\|\|\delta Y\|_{q\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_k^{\frac{q}{q'}}\|Y'\|_{\infty}, \tag*{(31)}
\end{aligned} ∥ K 1 + K 2 + K 3 ∥ 2 q ′′ -var ; [ s , t ] F s k ≲ k , L ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k 2 4 − q ∥ δ M Y ∥ Q -var ; [ s , t ] F s k 0 + ∥ δ X ∥ q -var ; [ s , t ] ∥ δ Y ∥ q -var ; [ s , t ] F s k q ′ q ∥ Y ′ ∥ ∞ ≲ k ∥ δ Y ∥ q -var ; [ s , t ] F s k 2 4 − q δ ⟨ M Y ⟩ s , t 2 1 F s k 0 + ∥ P Y ∥ 2 q ′′ -var ; [ s , t ] F s k + ∥ δ X ∥ q -var ; [ s , t ] ∥ δ Y ∥ q -var ; [ s , t ] F s k q ′ q ∥ Y ′ ∥ ∞ , (31)
where the second inequality follows from Lépingle’s inequality (Lemma A.1).
For the term K 4 K^4 K 4 , applying Lemma A.2 with r = q ′ ′ 2 r=\frac{q''}{2} r = 2 q ′′ , r 1 = 2 q 4 − q r_1=\frac{2q}{4-q} r 1 = 4 − q 2 q , k 1 = 2 k 4 − q k_1=\frac{2k}{4-q} k 1 = 4 − q 2 k and k 0 = 2 k q − 2 k_0=\frac{2k}{q-2} k 0 = q − 2 2 k , we have
∥ ∥ K 4 ∥ q ′ ′ 2 -var ; [ s , t ] | F s ∥ k ≲ k ∥ ∥ ∂ D G ( Y ⋅ ) ∥ 2 q 4 − q -var ; [ s , t ] | F s ∥ k 1 ∥ δ ⟨ M Y ⟩ s , t 1 2 | F s ∥ k 0 \left\|\|K^4\|_{\frac{q''}{2}\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_k\lesssim_k\left\|\|\partial DG(Y_\cdot)\|_{\frac{2q}{4-q}\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_{k_1}\left\|\delta\langle M^Y\rangle_{s,t}^{\frac{1}{2}}\middle|\mathcal{F}_s\right\|_{k_0} ∥ K 4 ∥ 2 q ′′ -var ; [ s , t ] F s k ≲ k ∥ ∂ D G ( Y ⋅ ) ∥ 4 − q 2 q -var ; [ s , t ] F s k 1 δ ⟨ M Y ⟩ s , t 2 1 F s k 0
≲ k C G ∥ ∥ δ Y ∥ q -var ; [ s , t ] | F s ∥ k 4 − q 2 ∥ δ ⟨ M Y ⟩ s , t 1 2 | F s ∥ k 0 , (32) \lesssim_k C_G\left\|\|\delta Y\|_{q\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_k^{\frac{4-q}{2}}\left\|\delta\langle M^Y\rangle_{s,t}^{\frac{1}{2}}\middle|\mathcal{F}_s\right\|_{k_0}, \tag*{(32)} ≲ k C G ∥ δ Y ∥ q -var ; [ s , t ] F s k 2 4 − q δ ⟨ M Y ⟩ s , t 2 1 F s k 0 , (32)
where the second inequality is due to the fact that D G DG D G is both Lipschitz and bounded. Combining (31) and (32) , we get the desired (26) .
Finally, recalling M t = ∫ 0 t D G ( Y r ) d M r Y M_t=\int_0^t DG(Y_r)dM^Y_r M t = ∫ 0 t D G ( Y r ) d M r Y , we get ∥ δ ⟨ M ⟩ s , t 1 / 2 ∣ F s ∥ k ≤ ∥ D G ( ⋅ ) ∥ ∞ ∥ δ ⟨ M Y ⟩ s , t 1 / 2 ∣ F s ∥ k \|\delta\langle M\rangle_{s,t}^{1/2}\mid\mathcal{F}_s\|_k\leq\|DG(\cdot)\|_{\infty}\|\delta\langle M^Y\rangle_{s,t}^{1/2}\mid\mathcal{F}_s\|_k ∥ δ ⟨ M ⟩ s , t 1/2 ∣ F s ∥ k ≤ ∥ D G ( ⋅ ) ∥ ∞ ∥ δ ⟨ M Y ⟩ s , t 1/2 ∣ F s ∥ k and then (27) follows. □
The following proposition establishes an estimate for the rough stochastic integral and, in particular, implies that we can improve the regularity of the integral from RSM X ( p , p ′ ) -var \operatorname{RSM}_{X}^{(p,p')\text{-var}} RSM X ( p , p ′ ) -var to RSM X ( p , p ) -var \operatorname{RSM}_{X}^{(p,p)\text{-var}} RSM X ( p , p ) -var .
Proposition 3.6. Assume ( p , p ′ ) (p,p') ( p , p ′ ) holds. For ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( [ 0 , T ] , L ( V , V 1 ) ) (Y,Y')\in\operatorname{RSM}_{X}^{(p,p')\text{-var}}([0,T],\mathcal{L}(V,V_1)) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , L ( V , V 1 )) , denote the rough stochastic integral Γ t : = ∫ 0 t Y r d X r \Gamma_t:=\int_0^tY_r\,dX_r Γ t := ∫ 0 t Y r d X r , t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] , as in Definition 2.8.
Then, ( Γ , Y ) (\Gamma,Y) ( Γ , Y ) belongs to RSM ( p , p ) -var ( [ 0 , T ] , V 1 ) \operatorname{RSM}^{(p,p)\text{-var}}([0,T],V_1) RSM ( p , p ) -var ([ 0 , T ] , V 1 ) . Moreover, assuming ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) and k ≥ 1 k\geq1 k ≥ 1 , the remainder R s , t Γ : = δ Γ s , t − Y s δ X s , t R^\Gamma_{s,t}:=\delta\Gamma_{s,t}-Y_s\delta X_{s,t} R s , t Γ := δ Γ s , t − Y s δ X s , t satisfies the following estimate for all ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ :
∥ ∥ R Γ ∥ p 2 -var ; [ s , t ] | F s ∥ k ≲ k ∥ X ∥ p -var ; [ s , t ] ( ∥ Y ′ ′ ∥ ∞ + ∥ ∥ δ Y ′ ∥ q ′ -var ; [ s , t ] | F s ∥ k ) + ∥ δ X ∥ p -var ; [ s , t ] ( ∥ ∥ P Y ∥ q ′ ′ 2 -var ; [ s , t ] | F s ∥ k + ∥ δ M Y ∥ B M O ; [ s , t ] ) . \left\|\|R^\Gamma\|_{\frac{p}{2}\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_k\lesssim_k\|X\|_{p\text{-var};[s,t]}\left(\|Y''\|_{\infty}+\left\|\|\delta Y'\|_{q'\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_k\right)
+\|\delta X\|_{p\text{-var};[s,t]}\left(\left\|\|P^Y\|_{\frac{q''}{2}\text{-var};[s,t]}\middle|\mathcal{F}_s\right\|_k+\|\delta M^Y\|_{\mathrm{BMO};[s,t]}\right). ∥ R Γ ∥ 2 p -var ; [ s , t ] F s k ≲ k ∥ X ∥ p -var ; [ s , t ] ( ∥ Y ′′ ∥ ∞ + ∥ δ Y ′ ∥ q ′ -var ; [ s , t ] F s k ) + ∥ δ X ∥ p -var ; [ s , t ] ( ∥ P Y ∥ 2 q ′′ -var ; [ s , t ] F s k + ∥ δ M Y ∥ BMO ; [ s , t ] ) .
Remark 3.7. The expression on the right-hand side is closely related to the RSM seminorm ( Y , Y ′ ) X ; q , q ′ ; k , ∞ (Y,Y')_{X;q,q';k,\infty} ( Y , Y ′ ) X ; q , q ′ ; k , ∞ which is introduced in (48) in Section 3.3.
Proof. By [16 ] , we have ( Γ , Y ) ∈ R S M X ( p , p ) -var ( [ 0 , T ] , V 1 ) (\Gamma,Y)\in\mathrm{RSM}_{X}^{(p,p)\text{-var}}([0,T],V_1) ( Γ , Y ) ∈ RSM X ( p , p ) -var ([ 0 , T ] , V 1 ) ; see Remark 2.11 for details. Note that
R s , t Γ = ∫ s t δ Y s , r d X r = Ξ s , t 1 + Ξ s , t 2 + Y s ′ X s , t , (33) R_{s,t}^{\Gamma}=\int_s^t\delta Y_{s,r}\,dX_r=\Xi_{s,t}^{1}+\Xi_{s,t}^{2}+Y_s'X_{s,t}, \tag*{(33)} R s , t Γ = ∫ s t δ Y s , r d X r = Ξ s , t 1 + Ξ s , t 2 + Y s ′ X s , t , (33)
where
Ξ s , t 1 : = ∫ s t δ Y s , r d X r − Y s ′ X s , t − ∫ s t δ M s , r Y d X r and Ξ s , t 2 : = ∫ s t δ M s , r Y d X r . \Xi_{s,t}^{1}:=\int_s^t\delta Y_{s,r}\,dX_r-Y_s'X_{s,t}-\int_s^t\delta M_{s,r}^{Y}\,dX_r
\qquad\text{and}\qquad
\Xi_{s,t}^{2}:=\int_s^t\delta M_{s,r}^{Y}\,dX_r. Ξ s , t 1 := ∫ s t δ Y s , r d X r − Y s ′ X s , t − ∫ s t δ M s , r Y d X r and Ξ s , t 2 := ∫ s t δ M s , r Y d X r .
Recall that δ Y s , r = δ ( Y − M Y ) s , r + δ M s , r Y \delta Y_{s,r}=\delta(Y-M^{Y})_{s,r}+\delta M_{s,r}^{Y} δ Y s , r = δ ( Y − M Y ) s , r + δ M s , r Y . Hence, denoting A s , t : = ( Y − M Y ) s δ X s , t + Y s ′ X s , t A_{s,t}:=(Y-M^{Y})_s\delta X_{s,t}+Y_s'X_{s,t} A s , t := ( Y − M Y ) s δ X s , t + Y s ′ X s , t ,
Ξ s , t 1 = ∫ s t δ ( Y − M Y ) s , r d X r + ∫ s t δ M s , r Y d X r − Y s ′ X s , t − ∫ s t δ M s , r Y d X r = ∫ s t ( Y − M Y ) r d X r − A s , t . \Xi_{s,t}^{1}=\int_s^t\delta(Y-M^{Y})_{s,r}\,dX_r+\int_s^t\delta M_{s,r}^{Y}\,dX_r-Y_s'X_{s,t}-\int_s^t\delta M_{s,r}^{Y}\,dX_r=\int_s^t(Y-M^{Y})_r\,dX_r-A_{s,t}. Ξ s , t 1 = ∫ s t δ ( Y − M Y ) s , r d X r + ∫ s t δ M s , r Y d X r − Y s ′ X s , t − ∫ s t δ M s , r Y d X r = ∫ s t ( Y − M Y ) r d X r − A s , t .
Since ( ( Y − M Y ) , Y ′ ) ∈ D X ( p , p ′ ) -var ((Y-M^{Y}),Y')\in\mathcal{D}_{X}^{(p,p')\text{-var}} (( Y − M Y ) , Y ′ ) ∈ D X ( p , p ′ ) -var a.s., by the definition of rough integrals (see, e.g., [10 , 15 ] ),
∫ s t ( Y − M Y ) r d X r = lim ∣ π ∣ → 0 ∑ [ u , v ] ∈ π A u ∨ s , v ∧ t a.s. \int_s^t(Y-M^{Y})_r\,dX_r=\lim_{|\pi|\to0}\sum_{[u,v]\in\pi}A_{u\vee s,v\wedge t}\quad\text{a.s.} ∫ s t ( Y − M Y ) r d X r = ∣ π ∣ → 0 lim [ u , v ] ∈ π ∑ A u ∨ s , v ∧ t a.s.
Note Ξ s , t 1 = ∫ s t ( Y − M Y ) r d X r − A s , t \Xi_{s,t}^{1}=\int_s^t(Y-M^{Y})_r\,dX_r-A_{s,t} Ξ s , t 1 = ∫ s t ( Y − M Y ) r d X r − A s , t . Then, since ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) holds, the sewing lemma (see [15 ] , Theorem 2.5) yields a universal constant C > 0 C>0 C > 0 such that, a.s., the following estimate holds
∥ Ξ 1 ∥ p 2 -var ; [ s , t ] ≤ C ( ∥ δ Y ′ ∥ q ′ -var ; [ s , t ] ∥ X ∥ p 2 -var ; [ s , t ] + ∥ P Y ∥ q ′ ′ 2 -var ; [ s , t ] ∥ δ X ∥ p -var ; [ s , t ] ) . (34) \|\Xi^{1}\|_{\frac{p}{2}\text{-var};[s,t]}
\leq C\left(\|\delta Y'\|_{q'\text{-var};[s,t]}\|X\|_{\frac{p}{2}\text{-var};[s,t]}+\|P^{Y}\|_{\frac{q''}{2}\text{-var};[s,t]}\|\delta X\|_{p\text{-var};[s,t]}\right). \tag*{(34)} ∥ Ξ 1 ∥ 2 p -var ; [ s , t ] ≤ C ( ∥ δ Y ′ ∥ q ′ -var ; [ s , t ] ∥ X ∥ 2 p -var ; [ s , t ] + ∥ P Y ∥ 2 q ′′ -var ; [ s , t ] ∥ δ X ∥ p -var ; [ s , t ] ) . (34)
For the term Ξ s , t 2 \Xi_{s,t}^{2} Ξ s , t 2 , by (10) we get Ξ s , t 2 = ∫ s t δ X r , t d M r Y = δ X s , t δ M s , t Y − ∫ s t δ X s , r d M r Y \Xi_{s,t}^{2}=\int_s^t\delta X_{r,t}\,dM_r^{Y}=\delta X_{s,t}\delta M_{s,t}^{Y}-\int_s^t\delta X_{s,r}\,dM_r^{Y} Ξ s , t 2 = ∫ s t δ X r , t d M r Y = δ X s , t δ M s , t Y − ∫ s t δ X s , r d M r Y . Then, Lemma A.2 and Lemma A.1 yield
∥ ∥ Ξ 2 ∥ p 2 -var ; [ s , t ] F s ∥ k ≲ ∥ ∥ δ M s , ⋅ Y ∥ ∞ ; [ s , t ] F s ∥ k ∥ δ X ∥ p -var ; [ s , t ] ≲ k ∥ ∥ δ M Y ∥ B M O ; [ s , t ] ∥ δ X ∥ p -var ; [ s , t ] . (35) \|\|\Xi^{2}\|_{\frac{p}{2}\text{-var};[s,t]}\mathcal{F}_s\|_k
\lesssim\|\|\delta M^Y_{s,\cdot}\|_{\infty;[s,t]}\mathcal{F}_s\|_k\|\delta X\|_{p\text{-var};[s,t]}
\lesssim_k\|\|\delta M^Y\|_{\mathrm{BMO};[s,t]}\|\delta X\|_{p\text{-var};[s,t]}. \tag*{(35)} ∥∥ Ξ 2 ∥ 2 p -var ; [ s , t ] F s ∥ k ≲ ∥∥ δ M s , ⋅ Y ∥ ∞ ; [ s , t ] F s ∥ k ∥ δ X ∥ p -var ; [ s , t ] ≲ k ∥∥ δ M Y ∥ BMO ; [ s , t ] ∥ δ X ∥ p -var ; [ s , t ] . (35)
The p 2 \frac{p}{2} 2 p -variation of the last term Y s ′ X s , t Y_s'X_{s,t} Y s ′ X s , t in (33) is plainly bounded by ∥ Y ′ ∥ ∞ ; [ s , t ] ∥ X ∥ p 2 -var ; [ s , t ] \|Y'\|_{\infty;[s,t]}\|X\|_{\frac{p}{2}\text{-var};[s,t]} ∥ Y ′ ∥ ∞ ; [ s , t ] ∥ X ∥ 2 p -var ; [ s , t ] . Combining this with (33) , (34) , (35) , we get the desired inequality. □
The following corollary provides the estimate for the rough stochastic integral term ∫ G ( Y r ) d X r \int G(Y_r)dX_r ∫ G ( Y r ) d X r under the rough semimartingale norm.
Corollary 3.8. Assume ( p , p ′ ) (p,p') ( p , p ′ ) and ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) hold with q < q ′ q<q' q < q ′ . For G ∈ C b 2 ( V 2 , L ( V , V 1 ) ) G\in C_b^2(V_2,\mathcal{L}(V,V_1)) G ∈ C b 2 ( V 2 , L ( V , V 1 )) and ( Y , Y ′ ) ∈ R S M X ( p , p ′ ) -var ( [ 0 , T ] , V 2 ) (Y,Y')\in\mathrm{RSM}_{X}^{(p,p')\text{-var}}([0,T],V_2) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , V 2 ) with ∥ Y ′ ∥ ∞ + ∥ G ∥ C b 2 ≤ K \|Y'\|_{\infty}+\|G\|_{C_b^2}\leq K ∥ Y ′ ∥ ∞ + ∥ G ∥ C b 2 ≤ K , set Λ t : = ∫ 0 t G ( Y r ) d X r \Lambda_t:=\int_0^tG(Y_r)dX_r Λ t := ∫ 0 t G ( Y r ) d X r . Then, for any k ≥ 1 k\geq1 k ≥ 1 , there is a constant C k , K > 0 C_{k,K}>0 C k , K > 0 depending only on ( k , K ) (k,K) ( k , K ) such that for all ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ ,
∥ ∥ R Λ ∥ p 2 -var ; [ s , t ] F s ∥ k + ∥ ∥ δ Λ ∥ p -var ; [ s , t ] F s ∥ k ≤ C k , K ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∣ δ X ∣ p -var ; [ s , t ] + ∥ ∥ δ Y ′ ∥ q ′ -var ; [ s , t ] F s ∥ k ) \|\|R^\Lambda\|_{\frac{p}{2}\text{-var};[s,t]}\mathcal{F}_s\|_k+\|\|\delta\Lambda\|_{p\text{-var};[s,t]}\mathcal{F}_s\|_k
\leq C_{k,K}|\delta X|_{p\text{-var};[s,t]}\left(1+|\delta X|_{p\text{-var};[s,t]}+\|\|\delta Y'\|_{q'\text{-var};[s,t]}\mathcal{F}_s\|_k\right) ∥∥ R Λ ∥ 2 p -var ; [ s , t ] F s ∥ k + ∥∥ δ Λ ∥ p -var ; [ s , t ] F s ∥ k ≤ C k , K ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∣ δ X ∣ p -var ; [ s , t ] + ∥∥ δ Y ′ ∥ q ′ -var ; [ s , t ] F s ∥ k )
+ ( 1 + ∣ δ X ∣ p -var ; [ s , t ] ) ⋅ ∥ ∥ δ Y ∥ q -var ; [ s , t ] F s ∥ k + ∥ δ M Y ∥ B M O ; [ s , t ] + ∥ ∥ P Y ∥ q ′ ′ 2 -var ; [ s , t ] F s ∥ k . (36) \qquad+(1+|\delta X|_{p\text{-var};[s,t]})\cdot\|\|\delta Y\|_{q\text{-var};[s,t]}\mathcal{F}_s\|_k+\|\delta M^Y\|_{\mathrm{BMO};[s,t]}+\|\|P^Y\|_{\frac{q''}{2}\text{-var};[s,t]}\mathcal{F}_s\|_k. \tag*{(36)} + ( 1 + ∣ δ X ∣ p -var ; [ s , t ] ) ⋅ ∥∥ δ Y ∥ q -var ; [ s , t ] F s ∥ k + ∥ δ M Y ∥ BMO ; [ s , t ] + ∥∥ P Y ∥ 2 q ′′ -var ; [ s , t ] F s ∥ k . (36)
Proof. As explained at the beginning of this subsection, ( G ( Y ) , D G ( Y ) Y ′ ) ∈ R S M X ( p , p ′ ) -var (G(Y),DG(Y)Y')\in\mathrm{RSM}_{X}^{(p,p')\text{-var}} ( G ( Y ) , D G ( Y ) Y ′ ) ∈ RSM X ( p , p ′ ) -var . Denote the corresponding Gubinelli derivative by G ( Y ) ′ = D G ( Y ) Y ′ G(Y)'=DG(Y)Y' G ( Y ) ′ = D G ( Y ) Y ′ . Then, by Proposition 3.5 (where q < q ′ q<q' q < q ′ is required), we obtain
∥ ∥ P G ( Y ) ∥ q ′ ′ 2 -var ; [ s , t ] F s ∥ k ≲ k , K ∥ ∥ δ Y ∥ q -var ; [ s , t ] F s ∥ k 2 q ′ q ′ ′ ∥ δ M Y ∥ B M O ; [ s , t ] \|\|P^{G(Y)}\|_{\frac{q''}{2}\text{-var};[s,t]}\mathcal{F}_s\|_k
\lesssim_{k,K}\|\|\delta Y\|_{q\text{-var};[s,t]}\mathcal{F}_s\|_k^{\frac{2q'}{q''}}\|\delta M^Y\|_{\mathrm{BMO};[s,t]} ∥∥ P G ( Y ) ∥ 2 q ′′ -var ; [ s , t ] F s ∥ k ≲ k , K ∥∥ δ Y ∥ q -var ; [ s , t ] F s ∥ k q ′′ 2 q ′ ∥ δ M Y ∥ BMO ; [ s , t ]
+ ∥ ∥ P Y ∥ q ′ ′ 2 -var ; [ s , t ] F s ∥ k + ∣ δ X ∣ p -var ; [ s , t ] ∥ ∥ δ Y ∥ q -var ; [ s , t ] F s ∥ k q q ′ . \qquad+\|\|P^Y\|_{\frac{q''}{2}\text{-var};[s,t]}\mathcal{F}_s\|_k+|\delta X|_{p\text{-var};[s,t]}\|\|\delta Y\|_{q\text{-var};[s,t]}\mathcal{F}_s\|_k^{\frac{q}{q'}}. + ∥∥ P Y ∥ 2 q ′′ -var ; [ s , t ] F s ∥ k + ∣ δ X ∣ p -var ; [ s , t ] ∥∥ δ Y ∥ q -var ; [ s , t ] F s ∥ k q ′ q .
and ∥ δ M G ( Y ) ∥ B M O ≲ K ∥ δ M Y ∥ B M O \lVert\delta M^{G(Y)}\rVert_{\mathrm{BMO}} \lesssim_K \lVert\delta M^Y\rVert_{\mathrm{BMO}} ∥ δ M G ( Y ) ∥ BMO ≲ K ∥ δ M Y ∥ BMO . This, together with Proposition [6 ] yields
∥ ∥ R Λ ∥ p 2 -var ; [ s , t ] ∥ F s ∥ k ≲ k ∣ δ X ∣ p -var ; [ s , t ] ( ∥ G ( Y ) ′ ∥ ∞ + ∥ ∥ δ G ( Y ) ′ ∥ q ′ -var ; [ s , t ] ∥ F s ∥ k + ∥ ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k 4 − q 2 ∥ δ M Y ∥ B M O ; [ s , t ] + ∥ ∥ P Y ∥ q ′ ′ 2 -var ; [ s , t ] ∥ F s ∥ k + ∣ δ X ∣ p -var ; [ s , t ] ∥ ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k q q ′ + ∥ δ M Y ∥ B M O ; [ s , t ] ) ≲ K ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∥ ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k + ∥ ∥ δ Y ′ ∥ q ′ -var ; [ s , t ] ∥ F s ∥ k + ∥ ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k 4 − q 2 ∥ δ M Y ∥ B M O ; [ s , t ] + ∥ ∥ P Y ∥ q ′ ′ 2 -var ; [ s , t ] ∥ F s ∥ k + ∣ δ X ∣ p -var ; [ s , t ] ∥ ∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k q q ′ + ∥ δ M Y ∥ B M O ; [ s , t ] ) . \begin{aligned}
\lVert\lVert R^\Lambda\rVert_{\frac{p}{2}\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k
&\lesssim_k |\delta X|_{p\text{-var};[s,t]}\Bigl(\lVert G(Y)'\rVert_\infty+\lVert\lVert\delta G(Y)'\rVert_{q'\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k\\
&\qquad+\lVert\lVert\delta Y\rVert_{q\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k^{\frac{4-q}{2}}\lVert\delta M^Y\rVert_{\mathrm{BMO};[s,t]}\\
&\qquad+\lVert\lVert P^Y\rVert_{\frac{q''}{2}\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k+|\delta X|_{p\text{-var};[s,t]}\lVert\lVert\delta Y\rVert_{q\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k^{\frac{q}{q'}}+\lVert\delta M^Y\rVert_{\mathrm{BMO};[s,t]}\Bigr)\\
&\lesssim_K|\delta X|_{p\text{-var};[s,t]}\Bigl(1+\lVert\lVert\delta Y\rVert_{q\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k+\lVert\lVert\delta Y'\rVert_{q'\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k\\
&\qquad+\lVert\lVert\delta Y\rVert_{q\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k^{\frac{4-q}{2}}\lVert\delta M^Y\rVert_{\mathrm{BMO};[s,t]}+\lVert\lVert P^Y\rVert_{\frac{q''}{2}\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k\\
&\qquad+|\delta X|_{p\text{-var};[s,t]}\lVert\lVert\delta Y\rVert_{q\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k^{\frac{q}{q'}}+\lVert\delta M^Y\rVert_{\mathrm{BMO};[s,t]}\Bigr).
\end{aligned} ∥∥ R Λ ∥ 2 p -var ; [ s , t ] ∥ F s ∥ k ≲ k ∣ δ X ∣ p -var ; [ s , t ] ( ∥ G ( Y ) ′ ∥ ∞ + ∥∥ δ G ( Y ) ′ ∥ q ′ -var ; [ s , t ] ∥ F s ∥ k + ∥∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k 2 4 − q ∥ δ M Y ∥ BMO ; [ s , t ] + ∥∥ P Y ∥ 2 q ′′ -var ; [ s , t ] ∥ F s ∥ k + ∣ δ X ∣ p -var ; [ s , t ] ∥∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k q ′ q + ∥ δ M Y ∥ BMO ; [ s , t ] ) ≲ K ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∥∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k + ∥∥ δ Y ′ ∥ q ′ -var ; [ s , t ] ∥ F s ∥ k + ∥∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k 2 4 − q ∥ δ M Y ∥ BMO ; [ s , t ] + ∥∥ P Y ∥ 2 q ′′ -var ; [ s , t ] ∥ F s ∥ k + ∣ δ X ∣ p -var ; [ s , t ] ∥∥ δ Y ∥ q -var ; [ s , t ] ∥ F s ∥ k q ′ q + ∥ δ M Y ∥ BMO ; [ s , t ] ) .
Noting ∣ a ∣ ( 4 − q ) / 2 ∣ b ∣ ≤ ∣ a ∣ + ∣ b ∣ 2 / ( q − 2 ) |a|^{(4-q)/2}|b|\leq|a|+|b|^{2/(q-2)} ∣ a ∣ ( 4 − q ) /2 ∣ b ∣ ≤ ∣ a ∣ + ∣ b ∣ 2/ ( q − 2 ) and ∣ a ∣ q / q ′ ≤ 1 + ∣ a ∣ |a|^{q/q'}\leq1+|a| ∣ a ∣ q / q ′ ≤ 1 + ∣ a ∣ , we get
∥ ∥ R Λ ∥ p 2 -var ; [ s , t ] ∥ F s ∥ k ≲ k , K RHS of (3.25) . (37) \lVert\lVert R^\Lambda\rVert_{\frac{p}{2}\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_k\lesssim_{k,K}\text{RHS of (3.25)}. \tag*{(37)} ∥∥ R Λ ∥ 2 p -var ; [ s , t ] ∥ F s ∥ k ≲ k , K RHS of (3.25) . (37)
Then, the desired estimate (3.25) follows from (3.26) and the fact δ Λ s , t = G ( Y s ) δ X s , t + R s , t Λ \delta\Lambda_{s,t}=G(Y_s)\delta X_{s,t}+R^\Lambda_{s,t} δ Λ s , t = G ( Y s ) δ X s , t + R s , t Λ . □
Remark 3.9. The condition 0 < ( 4 − q ) / 2 < 1 0<(4-q)/2<1 0 < ( 4 − q ) /2 < 1 , equivalently q ∈ ( 2 , 4 ) q\in(2,4) q ∈ ( 2 , 4 ) , appearing in (3.15) of Proposition [5 ] , is crucial for the proof of Corollary [8 ] . Combined with Young’s inequality, this condition leads to the exponent 2 / ( q − 2 ) 2/(q-2) 2/ ( q − 2 ) in the BMO norm of M Y M^Y M Y in (3.25). The additional requirement that 2 / ( q − 2 ) < 2 2/(q-2)<2 2/ ( q − 2 ) < 2 , namely q > 3 q>3 q > 3 , in turn is the critical threshold in the proof of Proposition [4 ] .
The following two propositions provide quantitative estimates for the difference of two RSMs. Let X = ( X , X ) \mathbf{X}=(X,\mathbb{X}) X = ( X , X ) and X ˉ = ( X ˉ , X ˉ ) \bar{\mathbf{X}}=(\bar{X},\bar{\mathbb{X}}) X ˉ = ( X ˉ , X ˉ ) belong to C p -var ( [ s , t ] , V ) \mathcal{C}^{p\text{-var}}([s,t],V) C p -var ([ s , t ] , V ) , and
( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( [ s , t ] , V 1 ) , ( Y ˉ , Y ˉ ′ ) ∈ RSM X ˉ ( p , p ′ ) -var ( [ s , t ] , V 1 ) . (38) (Y,Y')\in\operatorname{RSM}_{\mathbf{X}}^{(p,p')\text{-var}}([s,t],V_1),\qquad(\bar{Y},\bar{Y}')\in\operatorname{RSM}_{\bar{\mathbf{X}}}^{(p,p')\text{-var}}([s,t],V_1). \tag*{(38)} ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ s , t ] , V 1 ) , ( Y ˉ , Y ˉ ′ ) ∈ RSM X ˉ ( p , p ′ ) -var ([ s , t ] , V 1 ) . (38)
Proposition 3.10. Assume (3.1) and (3.27) hold. Let G ∈ C b γ ( V 1 , V 2 ) G\in C_b^\gamma(V_1,V_2) G ∈ C b γ ( V 1 , V 2 ) for γ ∈ ( 2 , 3 ] \gamma\in(2,3] γ ∈ ( 2 , 3 ] satisfying γ − 1 q ≥ 2 q ′ ′ \frac{\gamma-1}{q}\geq\frac{2}{q''} q γ − 1 ≥ q ′′ 2 or equivalently q ≤ q ′ ( γ − 2 ) q\leq q'(\gamma-2) q ≤ q ′ ( γ − 2 ) (recalling 2 q ′ ′ = q + q ′ q q ′ \frac{2}{q''}=\frac{q+q'}{qq'} q ′′ 2 = q q ′ q + q ′ ). Then for every k ≥ 1 k\geq1 k ≥ 1 , we have
∥ ∥ P G ( Y ) − P G ( Y ˉ ) ∥ q ′ ′ 2 -var ; [ s , t ] ∥ F s ∥ k , ∞ ≲ k , γ ∥ G ∥ C b γ ( L 1 + L 2 + L 3 + L 4 ) , (39) \lVert\lVert P^{G(Y)}-P^{G(\bar{Y})}\rVert_{\frac{q''}{2}\text{-var};[s,t]}\lVert\mathcal{F}_s\rVert_{k,\infty}\lesssim_{k,\gamma}\lVert G\rVert_{C_b^\gamma}(L_1+L_2+L_3+L_4), \tag*{(39)} ∥∥ P G ( Y ) − P G ( Y ˉ ) ∥ 2 q ′′ -var ; [ s , t ] ∥ F s ∥ k , ∞ ≲ k , γ ∥ G ∥ C b γ ( L 1 + L 2 + L 3 + L 4 ) , (39)
where L i L_i L i , i = 1 , 2 , 3 , 4 i=1,2,3,4 i = 1 , 2 , 3 , 4 , are defined as follows:
L 1 : = ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) ⋅ ( ∥ ∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 1 + ∥ ∥ δ Y ˉ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 1 ) , L 2 : = ∥ ∥ δ ( Y − Y ˉ ) ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ⋅ ( ∥ ∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] + ∥ ∥ δ Y ˉ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ) , L 3 : = ∥ ∥ P Y − P Y ˉ ∥ q ′ ′ 2 -var ∥ ∞ ; k , ∞ + ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) ⋅ ∥ ∥ P Y ˉ ∥ q ′ ′ 2 -var ∥ ∞ ; k , ∞ , L 4 : = ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) ( ∥ ∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] + ∥ ∥ δ Y ˉ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ) + ∥ ∥ δ ( Y − Y ˉ ) ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ∥ δ M Y ˉ ∥ B M O ; [ s , t ] + ∥ ∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ∥ δ ( M Y − M Y ˉ ) ∥ B M O ; [ s , t ] . \begin{aligned}
L_1&:=\lVert Y-\bar{Y}\rVert_{L^\infty([s,t]\times\Omega)}\cdot\Bigl(\lVert\lVert\delta Y\rVert_{q\text{-var}}\lVert_{\infty;1,\infty;[s,t]}^{\gamma-1}+\lVert\lVert\delta\bar{Y}\rVert_{q\text{-var}}\lVert_{\infty;1,\infty;[s,t]}^{\gamma-1}\Bigr),\\
L_2&:=\lVert\lVert\delta(Y-\bar{Y})\rVert_{q\text{-var}}\lVert_{\infty;1,\infty;[s,t]}\cdot\Bigl(\lVert\lVert\delta Y\rVert_{q\text{-var}}\lVert_{\infty;1,\infty;[s,t]}+\lVert\lVert\delta\bar{Y}\rVert_{q\text{-var}}\lVert_{\infty;1,\infty;[s,t]}\Bigr),\\
L_3&:=\lVert\lVert P^Y-P^{\bar{Y}}\rVert_{\frac{q''}{2}\text{-var}}\lVert_{\infty;k,\infty}+\lVert Y-\bar{Y}\rVert_{L^\infty([s,t]\times\Omega)}\cdot\lVert\lVert P^{\bar{Y}}\rVert_{\frac{q''}{2}\text{-var}}\lVert_{\infty;k,\infty},\\
L_4&:=\lVert Y-\bar{Y}\rVert_{L^\infty([s,t]\times\Omega)}\Bigl(\lVert\lVert\delta Y\rVert_{q\text{-var}}\lVert_{\infty;1,\infty;[s,t]}+\lVert\lVert\delta\bar{Y}\rVert_{q\text{-var}}\lVert_{\infty;1,\infty;[s,t]}\Bigr)\\
&\qquad+\lVert\lVert\delta(Y-\bar{Y})\rVert_{q\text{-var}}\lVert_{\infty;1,\infty;[s,t]}\lVert\delta M^{\bar{Y}}\rVert_{\mathrm{BMO};[s,t]}+\lVert\lVert\delta Y\rVert_{q\text{-var}}\lVert_{\infty;1,\infty;[s,t]}\lVert\delta(M^Y-M^{\bar{Y}})\rVert_{\mathrm{BMO};[s,t]}.
\end{aligned} L 1 L 2 L 3 L 4 := ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) ⋅ ( ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 1 + ∥∥ δ Y ˉ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 1 ) , := ∥∥ δ ( Y − Y ˉ ) ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ⋅ ( ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] + ∥∥ δ Y ˉ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ) , := ∥∥ P Y − P Y ˉ ∥ 2 q ′′ -var ∥ ∞ ; k , ∞ + ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) ⋅ ∥∥ P Y ˉ ∥ 2 q ′′ -var ∥ ∞ ; k , ∞ , := ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) ( ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] + ∥∥ δ Y ˉ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ) + ∥∥ δ ( Y − Y ˉ ) ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ∥ δ M Y ˉ ∥ BMO ; [ s , t ] + ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ∥ δ ( M Y − M Y ˉ ) ∥ BMO ; [ s , t ] .
Remark 3.11. The above proposition shows that to bound the ( k , ∞ ) (k,\infty) ( k , ∞ ) norm of the RSM remainder after composition, it suffices to calculate ( 1 , ∞ ) (1,\infty) ( 1 , ∞ ) norms of the RSM, ( k , ∞ ) (k,\infty) ( k , ∞ ) norms of its remainder, and BMO norms of its martingale part. Thanks to this proposition, the problem of degenerating integrability, as shown in [11 ] , is overcome.
Proof. For ( s , t ) ∈ Δ (s,t) \in\Delta ( s , t ) ∈ Δ , we have
( P G ( Y ) − P G ( Y ˉ ) ) s , t = δ G ( Y ) s , t − D G ( Y s ) Y s ′ δ X s , t − δ M s , t G ( Y ) − ( δ G ( Y ˉ ) s , t − D G ( Y ˉ s ) Y ˉ s ′ δ X s , t − δ M s , t G ( Y ˉ ) ) . (40) \left(P^{G(Y)}-P^{G(\bar Y)}\right)_{s,t}
= \delta G(Y)_{s,t}-DG(Y_s)Y'_s\delta X_{s,t}-\delta M^{G(Y)}_{s,t}-\left(\delta G(\bar Y)_{s,t}-DG(\bar Y_s)\bar Y'_s\delta X_{s,t}-\delta M^{G(\bar Y)}_{s,t}\right). \tag*{(40)} ( P G ( Y ) − P G ( Y ˉ ) ) s , t = δ G ( Y ) s , t − D G ( Y s ) Y s ′ δ X s , t − δ M s , t G ( Y ) − ( δ G ( Y ˉ ) s , t − D G ( Y ˉ s ) Y ˉ s ′ δ X s , t − δ M s , t G ( Y ˉ ) ) . (40)
Define Y ^ t : = Y t − Y ˉ t \hat Y_t:=Y_t-\bar Y_t Y ^ t := Y t − Y ˉ t , M ^ : = M Y − M Y ˉ \hat M:=M^Y-M^{\bar Y} M ^ := M Y − M Y ˉ , and P ^ s , t : = P s , t Y − P s , t Y ˉ \hat P_{s,t}:=P^Y_{s,t}-P^{\bar Y}_{s,t} P ^ s , t := P s , t Y − P s , t Y ˉ . Then, plugging Y s ′ δ X s , t = δ Y s , t − P s , t Y − δ M s , t Y Y'_s\delta X_{s,t}=\delta Y_{s,t}-P^Y_{s,t}-\delta M^Y_{s,t} Y s ′ δ X s , t = δ Y s , t − P s , t Y − δ M s , t Y , Y ˉ s ′ δ X s , t = δ Y ˉ s , t − P s , t Y ˉ − δ M s , t Y ˉ \bar Y'_s\delta X_{s,t}=\delta\bar Y_{s,t}-P^{\bar Y}_{s,t}-\delta M^{\bar Y}_{s,t} Y ˉ s ′ δ X s , t = δ Y ˉ s , t − P s , t Y ˉ − δ M s , t Y ˉ , δ M s , t G ( Y ) = ∫ s t D G ( Y r ) d M r Y \delta M^{G(Y)}_{s,t}=\int_s^t DG(Y_r)dM^Y_r δ M s , t G ( Y ) = ∫ s t D G ( Y r ) d M r Y , and δ M s , t G ( Y ˉ ) = ∫ s t D G ( Y ˉ r ) d M r Y ˉ \delta M^{G(\bar Y)}_{s,t}=\int_s^t DG(\bar Y_r)dM^{\bar Y}_r δ M s , t G ( Y ˉ ) = ∫ s t D G ( Y ˉ r ) d M r Y ˉ into (40) , we obtain
( P G ( Y ) − P G ( Y ˉ ) ) s , t = ∫ 0 1 [ D G ( Y s + λ ⋅ δ Y s , t ) − D G ( Y s ) ] d λ ⋅ δ Y s , t − ∫ 0 1 [ D G ( Y ˉ s + λ ⋅ δ Y ˉ s , t ) − D G ( Y ˉ s ) ] d λ ⋅ δ Y ˉ s , t + D G ( Y s ) P s , t Y − D G ( Y ˉ s ) P s , t Y ˉ + D G ( Y s ) δ M s , t Y − D G ( Y ˉ s ) δ M s , t Y ˉ − ( δ M s , t G ( Y ) − δ M s , t G ( Y ˉ ) ) = ∫ 0 1 [ D G ( Y s + λ ⋅ δ Y s , t ) − D G ( Y ˉ s + λ ⋅ δ Y ˉ s , t ) − D G ( Y s ) + D G ( Y ˉ s ) ] d λ ⋅ δ Y s , t + ∫ 0 1 [ D G ( Y ˉ s + λ ⋅ δ Y ˉ s , t ) − D G ( Y ˉ s ) ] d λ ⋅ δ Y ^ s , t + D G ( Y s ) P ^ s , t + ( D G ( Y s ) − D G ( Y ˉ s ) ) P s , t Y ˉ − ∫ s t δ ( D G ( Y ) − D G ( Y ˉ ) ) s , r d M r Y ˉ − ∫ s t δ D G ( Y ) s , r d M ^ r . (41) \begin{aligned}
\left(P^{G(Y)}-P^{G(\bar Y)}\right)_{s,t}
={}&\int_0^1[DG(Y_s+\lambda\cdot\delta Y_{s,t})-DG(Y_s)]d\lambda\cdot\delta Y_{s,t}\\
&-\int_0^1[DG(\bar Y_s+\lambda\cdot\delta\bar Y_{s,t})-DG(\bar Y_s)]d\lambda\cdot\delta\bar Y_{s,t}
+DG(Y_s)P^Y_{s,t}-DG(\bar Y_s)P^{\bar Y}_{s,t}\\
&+DG(Y_s)\delta M^Y_{s,t}-DG(\bar Y_s)\delta M^{\bar Y}_{s,t}
-\left(\delta M^{G(Y)}_{s,t}-\delta M^{G(\bar Y)}_{s,t}\right)\\
={}&\int_0^1[DG(Y_s+\lambda\cdot\delta Y_{s,t})-DG(\bar Y_s+\lambda\cdot\delta\bar Y_{s,t})-DG(Y_s)+DG(\bar Y_s)]d\lambda\cdot\delta Y_{s,t}\\
&+\int_0^1[DG(\bar Y_s+\lambda\cdot\delta\bar Y_{s,t})-DG(\bar Y_s)]d\lambda\cdot\delta\hat Y_{s,t}
+DG(Y_s)\hat P_{s,t}+\left(DG(Y_s)-DG(\bar Y_s)\right)P^{\bar Y}_{s,t}\\
&-\int_s^t\delta(DG(Y)-DG(\bar Y))_{s,r}dM^{\bar Y}_r-\int_s^t\delta DG(Y)_{s,r}d\hat M_r. \tag*{(41)}
\end{aligned} ( P G ( Y ) − P G ( Y ˉ ) ) s , t = = ∫ 0 1 [ D G ( Y s + λ ⋅ δ Y s , t ) − D G ( Y s )] d λ ⋅ δ Y s , t − ∫ 0 1 [ D G ( Y ˉ s + λ ⋅ δ Y ˉ s , t ) − D G ( Y ˉ s )] d λ ⋅ δ Y ˉ s , t + D G ( Y s ) P s , t Y − D G ( Y ˉ s ) P s , t Y ˉ + D G ( Y s ) δ M s , t Y − D G ( Y ˉ s ) δ M s , t Y ˉ − ( δ M s , t G ( Y ) − δ M s , t G ( Y ˉ ) ) ∫ 0 1 [ D G ( Y s + λ ⋅ δ Y s , t ) − D G ( Y ˉ s + λ ⋅ δ Y ˉ s , t ) − D G ( Y s ) + D G ( Y ˉ s )] d λ ⋅ δ Y s , t + ∫ 0 1 [ D G ( Y ˉ s + λ ⋅ δ Y ˉ s , t ) − D G ( Y ˉ s )] d λ ⋅ δ Y ^ s , t + D G ( Y s ) P ^ s , t + ( D G ( Y s ) − D G ( Y ˉ s ) ) P s , t Y ˉ − ∫ s t δ ( D G ( Y ) − D G ( Y ˉ ) ) s , r d M r Y ˉ − ∫ s t δ D G ( Y ) s , r d M ^ r . (41)
For the first term on the RHS of the above equation, we have,
∣ ∫ 0 1 [ D G ( Y ˉ s + λ δ Y ˉ s , t ) − D G ( Y s + λ δ Y s , t ) − D G ( Y ˉ s ) + D G ( Y s ) ] d λ ∣ ≤ ∣ ∫ 0 1 ∫ 0 1 [ D 2 G ( Y ˉ s + λ ⋅ δ Y ˉ s , t + τ ( Y ^ s + λ ⋅ δ Y ^ s , t ) ) − D 2 G ( Y ˉ s + τ Y ^ s ) ] d τ ⋅ ( Y ^ s + λ ⋅ δ Y ^ s , t ) d λ ∣ + ∣ ∫ 0 1 ∫ 0 1 [ D 2 G ( Y ˉ s + τ Y ^ s ) d τ ⋅ ( Y ^ s + λ ⋅ δ Y ^ s , t − Y ^ s ) d λ ∣ ≤ ∥ G ∥ C b γ ( ∣ δ Y ^ s , t ∣ γ − 2 + ∣ δ Y ˉ s , t ∣ γ − 2 ) ∥ Y ^ ∥ ∞ + ∥ G ∥ C b 2 ∣ δ Y ^ s , t ∣ . (42) \begin{aligned}
&\left|\int_0^1[DG(\bar Y_s+\lambda\delta\bar Y_{s,t})-DG(Y_s+\lambda\delta Y_{s,t})-DG(\bar Y_s)+DG(Y_s)]d\lambda\right|\\
\le{}&\left|\int_0^1\int_0^1[D^2G(\bar Y_s+\lambda\cdot\delta\bar Y_{s,t}+\tau(\hat Y_s+\lambda\cdot\delta\hat Y_{s,t}))-D^2G(\bar Y_s+\tau\hat Y_s)]d\tau\cdot(\hat Y_s+\lambda\cdot\delta\hat Y_{s,t})d\lambda\right|\\
&+\left|\int_0^1\int_0^1[D^2G(\bar Y_s+\tau\hat Y_s)d\tau\cdot(\hat Y_s+\lambda\cdot\delta\hat Y_{s,t}-\hat Y_s)d\lambda\right|\\
\le{}&\|G\|_{C_b^\gamma}\left(|\delta\hat Y_{s,t}|^{\gamma-2}+|\delta\bar Y_{s,t}|^{\gamma-2}\right)\|\hat Y\|_\infty+\|G\|_{C_b^2}|\delta\hat Y_{s,t}|. \tag*{(42)}
\end{aligned} ≤ ≤ ∫ 0 1 [ D G ( Y ˉ s + λ δ Y ˉ s , t ) − D G ( Y s + λ δ Y s , t ) − D G ( Y ˉ s ) + D G ( Y s )] d λ ∫ 0 1 ∫ 0 1 [ D 2 G ( Y ˉ s + λ ⋅ δ Y ˉ s , t + τ ( Y ^ s + λ ⋅ δ Y ^ s , t )) − D 2 G ( Y ˉ s + τ Y ^ s )] d τ ⋅ ( Y ^ s + λ ⋅ δ Y ^ s , t ) d λ + ∫ 0 1 ∫ 0 1 [ D 2 G ( Y ˉ s + τ Y ^ s ) d τ ⋅ ( Y ^ s + λ ⋅ δ Y ^ s , t − Y ^ s ) d λ ∥ G ∥ C b γ ( ∣ δ Y ^ s , t ∣ γ − 2 + ∣ δ Y ˉ s , t ∣ γ − 2 ) ∥ Y ^ ∥ ∞ + ∥ G ∥ C b 2 ∣ δ Y ^ s , t ∣. (42)
and for the second one,
∣ ∫ 0 1 [ D G ( Y ˉ s + λ δ Y ˉ s , t ) − D G ( Y ˉ s ) ] d λ ∣ ≤ ∥ G ∥ C b 2 ∣ δ Y ˉ s , t ∣ . (43) \left|\int_0^1[DG(\bar Y_s+\lambda\delta\bar Y_{s,t})-DG(\bar Y_s)]d\lambda\right|
\le\|G\|_{C_b^2}|\delta\bar Y_{s,t}|. \tag*{(43)} ∫ 0 1 [ D G ( Y ˉ s + λ δ Y ˉ s , t ) − D G ( Y ˉ s )] d λ ≤ ∥ G ∥ C b 2 ∣ δ Y ˉ s , t ∣. (43)
For the d M Y ˉ dM^{\bar Y} d M Y ˉ and d M ^ d\hat M d M ^ terms, denote
Ξ s , t : = ∫ s t δ D G ( Y ) s , r d M ^ r + ∫ s t δ ( D G ( Y ) − D G ( Y ˉ ) ) s , r d M r Y ˉ . \Xi_{s,t}:=\int_s^t\delta DG(Y)_{s,r}d\hat M_r+\int_s^t\delta(DG(Y)-DG(\bar Y))_{s,r}dM^{\bar Y}_r. Ξ s , t := ∫ s t δ D G ( Y ) s , r d M ^ r + ∫ s t δ ( D G ( Y ) − D G ( Y ˉ ) ) s , r d M r Y ˉ .
By Lemma A.5 we get (noting q ≤ q ′ ( γ − 2 ) q\le q'(\gamma-2) q ≤ q ′ ( γ − 2 ) )
∥ δ ( D G ( Y ) − D G ( Y ˉ ) ) ∥ q ′ − v a r ; [ s , t ] ≤ ∥ D 2 G ∥ C b γ − 2 ∥ Y ^ ∥ L ∞ ( [ s , t ] × Ω ) ( ∥ δ Y ∥ q − v a r ; ∞ ; 1 , ∞ ; [ s , t ] γ − 2 + ∥ δ Y ˉ ∥ q − v a r ; ∞ ; 1 , ∞ ; [ s , t ] γ − 2 + ∥ δ Y ^ ∥ q − v a r ; ∞ ; 1 , ∞ ; [ s , t ] ) . \|\delta(DG(Y)-DG(\bar Y))\|_{q'-\mathrm{var};[s,t]}
\le\|D^2G\|_{C_b^{\gamma-2}}\|\hat Y\|_{L^\infty([s,t]\times\Omega)}
\left(\|\delta Y\|_{q-\mathrm{var};\infty;1,\infty;[s,t]}^{\gamma-2}
+\|\delta\bar Y\|_{q-\mathrm{var};\infty;1,\infty;[s,t]}^{\gamma-2}
+\|\delta\hat Y\|_{q-\mathrm{var};\infty;1,\infty;[s,t]}\right). ∥ δ ( D G ( Y ) − D G ( Y ˉ )) ∥ q ′ − var ; [ s , t ] ≤ ∥ D 2 G ∥ C b γ − 2 ∥ Y ^ ∥ L ∞ ([ s , t ] × Ω ) ( ∥ δ Y ∥ q − var ; ∞ ; 1 , ∞ ; [ s , t ] γ − 2 + ∥ δ Y ˉ ∥ q − var ; ∞ ; 1 , ∞ ; [ s , t ] γ − 2 + ∥ δ Y ^ ∥ q − var ; ∞ ; 1 , ∞ ; [ s , t ] ) .
Hence, by Lemma A.2 and 3.2, we have
∥ Ξ ∥ q ′ 2 − v a r ; [ s , t ] ∥ F s ∥ k , ∞ ≲ k ∥ δ ( D G ( Y ) − D G ( Y ˉ ) ) ∥ q ′ − v a r ; [ s , t ] ∥ F s ∥ 2 k , ∞ ∥ δ M Y ˉ ∥ B M O ; [ s , t ] + ∥ ∥ δ D G ( Y ) ∥ q − v a r ; [ s , t ] ∥ F s ∥ 2 k , ∞ . × ∥ δ M ^ ∥ B M O ; [ s , t ] \begin{aligned}
\|\Xi\|_{\frac{q'}{2}-\mathrm{var};[s,t]}\|\mathcal F_s\|_{k,\infty}
\lesssim_k
\|\delta(DG(Y)-DG(\bar Y))\|_{q'-\mathrm{var};[s,t]}\|\mathcal F_s\|_{2k,\infty}\|\delta M^{\bar Y}\|_{\mathrm{BMO};[s,t]}
+\|\|\delta DG(Y)\|_{q-\mathrm{var};[s,t]}\|\mathcal F_s\|_{2k,\infty}. \\
&\times\|\delta\widehat{M}\|_{\mathrm{BMO};[s,t]}
\end{aligned} ∥Ξ ∥ 2 q ′ − var ; [ s , t ] ∥ F s ∥ k , ∞ ≲ k ∥ δ ( D G ( Y ) − D G ( Y ˉ )) ∥ q ′ − var ; [ s , t ] ∥ F s ∥ 2 k , ∞ ∥ δ M Y ˉ ∥ BMO ; [ s , t ] + ∥∥ δ D G ( Y ) ∥ q − var ; [ s , t ] ∥ F s ∥ 2 k , ∞ . × ∥ δ M ∥ BMO ; [ s , t ]
\begin{aligned}
\lesssim_k \|\|\delta(DG(Y)-DG(\overline{Y}))\|_{q'\text{-var}}\|_{\infty;1,\infty;[s,t]}\|\delta M^{\overline{Y}}\|_{\mathrm{BMO};[s,t]}+\|\|\delta DG(Y)\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]} \\
&\times\|\delta\widehat{M}\|_{\mathrm{BMO};[s,t]} {#eq:3.33} \\
&\leq\|G\|_{C_b^\gamma}\left(\|\widehat{Y}\|_{L^\infty([s,t]\times\Omega)}\left(\|\|\delta Y\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}^{\gamma-2}+\|\|\delta\overline{Y}\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}^{\gamma-2}\right)+\|\|\delta\widehat{Y}\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}\right) \\
&\times\|\delta M^{\overline{Y}}\|_{\mathrm{BMO};[s,t]}+\|\|\delta Y\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}\|\delta\widehat{M}\|_{\mathrm{BMO};[s,t]}.
\end{aligned}
Combining (41) , (42) , (43) , and eq:3.33 , by taking conditional expectation at time s s s and using Lemma (12) again, we have
∥ ∥ P G ( Y ) − P G ( Y ‾ ) ∥ q ′ ′ 2 -var ∥ ∞ ; k , ∞ ; [ s , t ] \left\|\|P^{G}(Y)-P^{G}(\overline{Y})\|_{\frac{q''}{2}\text{-var}}\right\|_{\infty;k,\infty;[s,t]} ∥ P G ( Y ) − P G ( Y ) ∥ 2 q ′′ -var ∞ ; k , ∞ ; [ s , t ]
≲ k ∥ G ∥ C b γ ( ∥ Y ^ ∥ L ∞ ( [ s , t ] × Ω ) ( ∥ ∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 1 + ∥ ∥ δ Y ‾ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 1 ) \lesssim_k \|G\|_{C_b^\gamma}\left(\|\widehat{Y}\|_{L^\infty([s,t]\times\Omega)}\left(\|\|\delta Y\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}^{\gamma-1}+\|\|\delta\overline{Y}\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}^{\gamma-1}\right)\right. ≲ k ∥ G ∥ C b γ ( ∥ Y ∥ L ∞ ([ s , t ] × Ω ) ( ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 1 + ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 1 )
+ ∥ ∥ δ Y ^ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ( ∥ ∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] + ∥ ∥ δ Y ‾ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ) + ∥ ∥ P ^ ∥ q ′ ′ 2 -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ Y ^ ∥ L ∞ ( [ s , t ] × Ω ) ∥ ∥ P Y ^ ∥ q ′ ′ 2 -var ∥ ∞ ; k , ∞ ; [ s , t ] \begin{aligned}
\left.+\|\|\delta\widehat{Y}\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}\left(\|\|\delta Y\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}+\|\|\delta\overline{Y}\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}\right)\right. \\
&+\|\|\widehat{P}\|_{\frac{q''}{2}\text{-var}}\|_{\infty;k,\infty;[s,t]}+\|\widehat{Y}\|_{L^\infty([s,t]\times\Omega)}\|\|P^{\widehat{Y}}\|_{\frac{q''}{2}\text{-var}}\|_{\infty;k,\infty;[s,t]}
\end{aligned} + ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ( ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] + ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ) + ∥∥ P ∥ 2 q ′′ -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ Y ∥ L ∞ ([ s , t ] × Ω ) ∥∥ P Y ∥ 2 q ′′ -var ∥ ∞ ; k , ∞ ; [ s , t ]
+ ( ∥ Y ^ ∥ L ∞ ( [ s , t ] × Ω ) ( ∥ ∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 2 + ∥ ∥ δ Y ‾ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 2 ) + ∥ ∥ δ Y ^ ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ) \left.+\left(\|\widehat{Y}\|_{L^\infty([s,t]\times\Omega)}\left(\|\|\delta Y\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}^{\gamma-2}+\|\|\delta\overline{Y}\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}^{\gamma-2}\right)+\|\|\delta\widehat{Y}\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}\right)\right. + ( ∥ Y ∥ L ∞ ([ s , t ] × Ω ) ( ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 2 + ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] γ − 2 ) + ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] )
× ∥ δ M Y ‾ ∥ B M O ; [ s , t ] + ∥ ∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ∥ δ M ^ ∥ B M O ; [ s , t ] ) . \left.\times\|\delta M^{\overline{Y}}\|_{\mathrm{BMO};[s,t]}+\|\|\delta Y\|_{q\text{-var}}\|_{\infty;1,\infty;[s,t]}\|\delta\widehat{M}\|_{\mathrm{BMO};[s,t]}\right). × ∥ δ M Y ∥ BMO ; [ s , t ] + ∥∥ δ Y ∥ q -var ∥ ∞ ; 1 , ∞ ; [ s , t ] ∥ δ M ∥ BMO ; [ s , t ] ) .
which is exactly the desired estimate (39) . □
The next proposition provides the estimate for the difference of two rough stochastic integrals.
Proposition 3.12. Assume that ( p , p ′ ) (p,p') ( p , p ′ ) and ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) hold, and that (38) holds with V 1 V_1 V 1 replaced by L ( V , V 1 ) \mathcal{L}(V,V_1) L ( V , V 1 ) . Denote rough stochastic integrals by
I t X : = ∫ 0 t Y r d X r and I t X ‾ : = ∫ 0 t Y ‾ r d X ‾ r , for t ∈ [ 0 , T ] . I_t^X:=\int_0^t Y_r\,dX_r\qquad\text{and}\qquad I_t^{\overline{X}}:=\int_0^t \overline{Y}_r\,d\overline{X}_r,\qquad\text{for }t\in[0,T]. I t X := ∫ 0 t Y r d X r and I t X := ∫ 0 t Y r d X r , for t ∈ [ 0 , T ] .
Then, for k ≥ 1 k\geq1 k ≥ 1 , it holds that for all ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ ,
∥ ∥ R I X − R I X ‾ ∥ p 2 -var ∥ ∞ ; k , ∞ ; [ s , t ] \|\|R^{I^X}-R^{I^{\overline{X}}}\|_{\frac{p}{2}\text{-var}}\|_{\infty;k,\infty;[s,t]} ∥∥ R I X − R I X ∥ 2 p -var ∥ ∞ ; k , ∞ ; [ s , t ]
\begin{aligned}
\lesssim\|\|P^Y-P^{\overline{Y}}\|_{\frac{q''}{2}\text{-var}}\|_{\infty;k,\infty;[s,t]}\|\delta\overline{X}\|_{p\text{-var};[s,t]}+\|\|P^Y\|_{\frac{q''}{2}\text{-var}}\|_{\infty;k,\infty;[s,t]}\|\delta(X-\overline{X})\|_{p\text{-var};[s,t]} \\
&+\left(\|Y'-\overline{Y}'\|_\infty+\|\|\delta(Y'-\overline{Y}')\|_{q'\text{-var}}\|_{\infty;k,\infty;[s,t]}\right)\|\delta\overline{X}\|_{\frac{p}{2}\text{-var};[s,t]} \\
&+\left(\|Y'\|_\infty+\|\|\delta Y'\|_{q'\text{-var}}\|_{\infty;k,\infty;[s,t]}\right)\|X-\overline{X}\|_{\frac{p}{2}\text{-var};[s,t]}+\|\delta(M^Y-M^{\overline{Y}})\|_{\mathrm{BMO};[s,t]}\|\delta\overline{X}\|_{p\text{-var};[s,t]} \\
&+\|\delta M^Y\|_{\mathrm{BMO};[s,t]}\|\delta(X-\overline{X})\|_{p\text{-var};[s,t]}. {#eq:3.34}
\end{aligned}
Proof. Note that for ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ ,
(R^{I^X}-R^{I^{\overline{X}}})_{s,t}=\delta(I^X-I^{\overline{X}})_{s,t}-Y_s\delta X_{s,t}+\overline{Y}_s\delta\overline{X}_{s,t}=A^1_{s,t}+A^2_{s,t}+Y'_sX_{s,t}-\overline{Y}'_s\overline{X}_{s,t}, {#eq:3.35}
where
A s , t 1 : = ∫ s t ( Y − M Y ) r d X r − ∫ s t ( Y ‾ − M Y ‾ ) r d X ‾ r − ( ( Y − M Y ) s δ X s , t − ( Y ‾ − M Y ‾ ) s δ X ‾ s , t + Y s ′ X s , t − Y ‾ s ′ X ‾ s , t ) A^1_{s,t}:=\int_s^t(Y-M^Y)_r\,dX_r-\int_s^t(\overline{Y}-M^{\overline{Y}})_r\,d\overline{X}_r-\left((Y-M^Y)_s\delta X_{s,t}-(\overline{Y}-M^{\overline{Y}})_s\delta\overline{X}_{s,t}+Y'_sX_{s,t}-\overline{Y}'_s\overline{X}_{s,t}\right) A s , t 1 := ∫ s t ( Y − M Y ) r d X r − ∫ s t ( Y − M Y ) r d X r − ( ( Y − M Y ) s δ X s , t − ( Y − M Y ) s δ X s , t + Y s ′ X s , t − Y s ′ X s , t )
and
A s , t 2 : = ∫ s t δ M s , r Y d X r − ∫ s t δ M s , r Y ‾ d X ‾ r . A^2_{s,t}:=\int_s^t\delta M^Y_{s,r}\,dX_r-\int_s^t\delta M^{\overline{Y}}_{s,r}\,d\overline{X}_r. A s , t 2 := ∫ s t δ M s , r Y d X r − ∫ s t δ M s , r Y d X r .
Define
Ξ s , t : = ( Y − M Y ) s δ X s , t − ( Y ‾ − M Y ‾ ) s δ X ‾ s , t + Y s ′ X s , t − Y ‾ s ′ X ‾ s , t . \Xi_{s,t}:=(Y-M^Y)_s\delta X_{s,t}-(\overline{Y}-M^{\overline{Y}})_s\delta\overline{X}_{s,t}+Y'_sX_{s,t}-\overline{Y}'_s\overline{X}_{s,t}. Ξ s , t := ( Y − M Y ) s δ X s , t − ( Y − M Y ) s δ X s , t + Y s ′ X s , t − Y s ′ X s , t .
It follows that
A s , t 1 = ∫ s t ( Y − M Y ) r d X r − ∫ s t ( Y ‾ − M Y ‾ ) r d X ‾ r − Ξ s , t . A^1_{s,t}=\int_s^t(Y-M^Y)_r\,dX_r-\int_s^t(\overline{Y}-M^{\overline{Y}})_r\,d\overline{X}_r-\Xi_{s,t}. A s , t 1 = ∫ s t ( Y − M Y ) r d X r − ∫ s t ( Y − M Y ) r d X r − Ξ s , t .
Next, we apply the sewing lemma to estimate A s , t 1 A^1_{s,t} A s , t 1 . For s ≤ u ≤ t s \le u \le t s ≤ u ≤ t , note that
δ Ξ s , u , t = − P s , u Y δ X u , t − δ Y s , u ′ X ‾ u , t + P s , u Y ‾ δ X ‾ u , t + δ Y ‾ s , u ′ X ‾ u , t = − ( P Y − P Y ‾ ) s , u δ X ‾ u , t − P s , u Y δ ( X − X ‾ ) u , t − δ ( Y ′ − Y ‾ ′ ) s , u X ‾ u , t − δ Y s , u ′ ( X u , t − X ‾ u , t ) . \begin{aligned}
\delta\Xi_{s,u,t}=-P^Y_{s,u}\delta X_{u,t}-\delta Y'_{s,u}\overline{X}_{u,t}+P^{\overline{Y}}_{s,u}\delta\overline{X}_{u,t}+\delta\overline{Y}'_{s,u}\overline{X}_{u,t} \\
&=-(P^Y-P^{\overline{Y}})_{s,u}\delta\overline{X}_{u,t}-P^Y_{s,u}\delta(X-\overline{X})_{u,t}-\delta(Y'-\overline{Y}')_{s,u}\overline{X}_{u,t}-\delta Y'_{s,u}(X_{u,t}-\overline{X}_{u,t}).
\end{aligned} δ Ξ s , u , t = − P s , u Y δ X u , t − δ Y s , u ′ X u , t + P s , u Y δ X u , t + δ Y s , u ′ X u , t = − ( P Y − P Y ) s , u δ X u , t − P s , u Y δ ( X − X ) u , t − δ ( Y ′ − Y ′ ) s , u X u , t − δ Y s , u ′ ( X u , t − X u , t ) .
Hence, we obtain
∣ δ Ξ s , u , t ∣ ≤ ∥ P Y − P Y ‾ ∥ q ′ ′ 2 -var ; [ s , t ] ∥ δ X ‾ ∥ p -var ; [ s , t ] + ∥ P Y ∥ q ′ ′ 2 -var ; [ s , t ] ∥ δ ( X − X ‾ ) ∥ p -var ; [ s , t ] \lvert\delta\Xi_{s,u,t}\rvert\leq\lVert P^Y-P^{\overline{Y}}\rVert_{\frac{q''}{2}\text{-var};[s,t]}\lVert\delta\overline{X}\rVert_{p\text{-var};[s,t]}+\lVert P^Y\rVert_{\frac{q''}{2}\text{-var};[s,t]}\lVert\delta(X-\overline{X})\rVert_{p\text{-var};[s,t]} ∣ δ Ξ s , u , t ∣ ≤ ∥ P Y − P Y ∥ 2 q ′′ -var ; [ s , t ] ∥ δ X ∥ p -var ; [ s , t ] + ∥ P Y ∥ 2 q ′′ -var ; [ s , t ] ∥ δ ( X − X ) ∥ p -var ; [ s , t ]
+ ∥ δ ( Y ′ − Y ‾ ′ ) ∥ q ′ -var ; [ s , t ] ∥ X ‾ ∥ p -var ; [ s , t ] + ∥ δ Y ′ ∥ q ′ -var ; [ s , t ] ∥ X − X ‾ ∥ p -var ; [ s , t ] . (44) \qquad+\lVert\delta(Y'-\overline{Y}')\rVert_{q'\text{-var};[s,t]}\lVert\overline{X}\rVert_{p\text{-var};[s,t]}+\lVert\delta Y'\rVert_{q'\text{-var};[s,t]}\lVert X-\overline{X}\rVert_{p\text{-var};[s,t]}. \tag*{(44)} + ∥ δ ( Y ′ − Y ′ ) ∥ q ′ -var ; [ s , t ] ∥ X ∥ p -var ; [ s , t ] + ∥ δ Y ′ ∥ q ′ -var ; [ s , t ] ∥ X − X ∥ p -var ; [ s , t ] . (44)
Note A s , t 1 = ∫ s t ( Y − M Y ) r d X r − ∫ s t ( Y ‾ − M Y ‾ ) r d X ‾ r − Ξ s , t A^1_{s,t}=\int_s^t(Y-MY)_r\,dX_r-\int_s^t(\overline{Y}-M\overline{Y})_r\,d\overline{X}_r-\Xi_{s,t} A s , t 1 = ∫ s t ( Y − M Y ) r d X r − ∫ s t ( Y − M Y ) r d X r − Ξ s , t . Since ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) holds, by the sewing lemma (see, e.g., [15 ] , Theorem 2.5), (44) implies that a.s.
\lvert A^1_{s,t}\rvert\lesssim\text{RHS of [](#eq:3.36)}. \tag*{(45)}
Moreover, noting that A s , t 2 = ∫ s t δ X ‾ s , r d ( M Y − M Y ‾ ) r + ∫ s t δ ( X − X ‾ ) s , r d M Y r A^2_{s,t}=\int_s^t\delta\overline{X}_{s,r}\,d(MY-M\overline{Y})_r+\int_s^t\delta(X-\overline{X})_{s,r}\,dMY_r A s , t 2 = ∫ s t δ X s , r d ( M Y − M Y ) r + ∫ s t δ ( X − X ) s , r d M Y r from integration by parts, by Lépingle’s inequality for martingale transforms, we have
∥ ∥ A 2 ∥ p 2 -var ; [ s , t ] | F s ∥ k , ∞ \left\lVert\left\lVert A^2\right\rVert_{\frac{p}{2}\text{-var};[s,t]}\middle|\mathcal{F}_s\right\rVert_{k,\infty} A 2 2 p -var ; [ s , t ] F s k , ∞
≲ k ∥ δ ( M Y − M Y ‾ ) ∥ B M O ; [ s , t ] ∥ δ X ‾ ∥ p -var ; [ s , t ] + ∥ δ M Y ∥ B M O ; [ s , t ] ∥ δ ( X − X ‾ ) ∥ p -var ; [ s , t ] . (46) \lesssim_k\lVert\delta(MY-M\overline{Y})\rVert_{\mathrm{BMO};[s,t]}\lVert\delta\overline{X}\rVert_{p\text{-var};[s,t]}+\lVert\delta MY\rVert_{\mathrm{BMO};[s,t]}\lVert\delta(X-\overline{X})\rVert_{p\text{-var};[s,t]}. \tag*{(46)} ≲ k ∥ δ ( M Y − M Y ) ∥ BMO ; [ s , t ] ∥ δ X ∥ p -var ; [ s , t ] + ∥ δ M Y ∥ BMO ; [ s , t ] ∥ δ ( X − X ) ∥ p -var ; [ s , t ] . (46)
In addition, we have
∣ Y s ′ X s , t − Y ‾ s ′ X ‾ s , t ∣ ≤ ∣ Y s ′ X ‾ s , t − Y ‾ s ′ X ‾ s , t ∣ + ∣ Y s ′ X s , t − Y s ′ X ‾ s , t ∣ \lvert Y'_sX_{s,t}-\overline{Y}'_s\overline{X}_{s,t}\rvert\leq\lvert Y'_s\overline{X}_{s,t}-\overline{Y}'_s\overline{X}_{s,t}\rvert+\lvert Y'_sX_{s,t}-Y'_s\overline{X}_{s,t}\rvert ∣ Y s ′ X s , t − Y s ′ X s , t ∣ ≤ ∣ Y s ′ X s , t − Y s ′ X s , t ∣ + ∣ Y s ′ X s , t − Y s ′ X s , t ∣
≤ 2 ( ∥ Y ′ − Y ‾ ′ ∥ ∞ ∥ X ‾ ∥ p -var ; [ s , t ] + ∥ Y ′ ∥ ∞ ∥ X − X ‾ ∥ p -var ; [ s , t ] ) . (47) \leq2\left(\lVert Y'-\overline{Y}'\rVert_\infty\lVert\overline{X}\rVert_{p\text{-var};[s,t]}+\lVert Y'\rVert_\infty\lVert X-\overline{X}\rVert_{p\text{-var};[s,t]}\right). \tag*{(47)} ≤ 2 ( ∥ Y ′ − Y ′ ∥ ∞ ∥ X ∥ p -var ; [ s , t ] + ∥ Y ′ ∥ ∞ ∥ X − X ∥ p -var ; [ s , t ] ) . (47)
Finally, by eq:3.35 we have ( R I X − R I X ‾ ) s , t = A s , t 1 + A s , t 2 + Y s ′ X s , t − Y ‾ s ′ X ‾ s , t (R^{I X}-R^{I\overline{X}})_{s,t}=A^1_{s,t}+A^2_{s,t}+Y'_sX_{s,t}-\overline{Y}'_s\overline{X}_{s,t} ( R I X − R I X ) s , t = A s , t 1 + A s , t 2 + Y s ′ X s , t − Y s ′ X s , t . Together with (45) , (46) , and (47) , this implies eq:3.34 . □
Seminorms
We now introduce some seminorms that will be used for the fixed point argument in the next section. As discussed in the introduction, one key difficulty in the analysis of rough BSDEs, compared with rough SDEs, lies in establishing the regularity of Z Z Z . To address this issue, we introduce the RSM seminorm ∣ ∣ ∣ ⋅ ∣ ∣ ∣ X ; q , q ′ ; k , ∞ \lvert\!\lvert\!\lvert\cdot\rvert\!\rvert\!\rvert_{X;q,q';k,\infty} ∣ ∣ ∣ ⋅ ∣ ∣ ∣ X ; q , q ′ ; k , ∞ . This quantity arises naturally in rough stochastic integral estimates (see Remark 3.7) and in the proof of the comparison principle (Theorem 4.3).
For ( p ′ , q , q ′ ) (p',q,q') ( p ′ , q , q ′ ) satisfying (11) , X = ( X , X ) ∈ C p -var ( [ 0 , T ] , R d ) \mathbf{X}=(X,\mathbb{X})\in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^d) X = ( X , X ) ∈ C p -var ([ 0 , T ] , R d ) , ( Y , Y ′ ) ∈ R S M X ( p , p ′ ) -var ( [ 0 , T ] , R ) (Y,Y')\in\mathrm{RSM}^{(p,p')\text{-var}}_{\mathbf{X}}([0,T],\mathbb{R}) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , R ) , and k ≥ 1 k\geq1 k ≥ 1 , define the seminorm (recall q ′ ′ / 2 : = q q ′ / ( q + q ′ ) q''/2:=qq'/(q+q') q ′′ /2 := q q ′ / ( q + q ′ ) )
∣ ∣ ∣ ( Y , Y ′ ) ∣ ∣ ∣ X ; q , q ′ ; k , ∞ : = ∥ δ Y ′ ∥ q ′ -var ; ∞ ; k , ∞ + ∥ Y ′ ∥ ∞ + ∥ ∥ P Y ∥ q ′ ′ 2 -var ; ∞ ; k , ∞ + ∥ δ M Y ∥ B M O . (48) \lvert\!\lvert\!\lvert(Y,Y')\rvert\!\rvert\!\rvert_{\mathbf{X};q,q';k,\infty}:=\lVert\delta Y'\rVert_{q'\text{-var};\infty;k,\infty}+\lVert Y'\rVert_\infty+\lVert\!\lVert P^Y\rVert_{\frac{q''}{2}\text{-var};\infty;k,\infty}+\lVert\delta MY\rVert_{\mathrm{BMO}}. \tag*{(48)} ∣ ∣ ∣( Y , Y ′ )∣ ∣ ∣ X ; q , q ′ ; k , ∞ := ∥ δ Y ′ ∥ q ′ -var ; ∞ ; k , ∞ + ∥ Y ′ ∥ ∞ + ∥ ∥ P Y ∥ 2 q ′′ -var ; ∞ ; k , ∞ + ∥ δ M Y ∥ BMO . (48)
Here, recall ( ∥ ⋅ ∥ q -var ) s , t : = ∥ ⋅ ∥ q -var ; [ s , t ] : Δ [ 0 , T ] → R (\lVert\cdot\rVert_{q\text{-var}})_{s,t}:=\lVert\cdot\rVert_{q\text{-var};[s,t]}:\Delta_{[0,T]}\to\mathbb{R} (∥ ⋅ ∥ q -var ) s , t := ∥ ⋅ ∥ q -var ; [ s , t ] : Δ [ 0 , T ] → R is a two-parameter function. For an arbitrary interval I ⊂ [ 0 , T ] I\subset[0,T] I ⊂ [ 0 , T ] , the notation ∣ ∣ ∣ ( Y , Y ′ ) ∣ ∣ ∣ X ; q , q ′ ; k , ∞ ; I \lvert\!\lvert\!\lvert(Y,Y')\rvert\!\rvert\!\rvert_{\mathbf{X};q,q';k,\infty;I} ∣ ∣ ∣( Y , Y ′ )∣ ∣ ∣ X ; q , q ′ ; k , ∞ ; I is defined analogously by restricting the underlying objects to I I I with, for example, ∥ Y ′ ∥ ∞ \lVert Y'\rVert_\infty ∥ Y ′ ∥ ∞ replaced by ∥ Y ′ ∥ L ∞ ( I × Ω ) \lVert Y'\rVert_{L^\infty(I\times\Omega)} ∥ Y ′ ∥ L ∞ ( I × Ω ) .
Lemma 3.13. Assume (11) holds and p ≥ q ′ ′ / 2 p\geq q''/2 p ≥ q ′′ /2 . Let ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , and suppose X ∈ C p -var ( [ s , t ] , R d ) \mathbf{X}\in\mathcal{C}^{p\text{-var}}([s,t],\mathbb{R}^d) X ∈ C p -var ([ s , t ] , R d ) and ( Y , Y ′ ) ∈ R S M X ( p , p ′ ) -var ( [ s , t ] , R ) (Y,Y')\in\mathrm{RSM}^{(p,p')\text{-var}}_{\mathbf{X}}([s,t],\mathbb{R}) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ s , t ] , R ) . Then, for any k ≥ 1 k\geq1 k ≥ 1 , it holds
∥ ∥ δ Y ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k ( 1 + ∥ δ X ∥ p -var ; [ s , t ] ) ∣ ∣ ∣ ( Y , Y ′ ) ∣ ∣ ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] . (49) \lVert\!\lVert\delta Y\rVert_{p\text{-var}}\rVert_{\infty;k,\infty;[s,t]}\lesssim_k\left(1+\lVert\delta X\rVert_{p\text{-var};[s,t]}\right)\lvert\!\lvert\!\lvert(Y,Y')\rvert\!\rvert\!\rvert_{\mathbf{X};q,q';k,\infty;[s,t]}. \tag*{(49)} ∥ ∥ δ Y ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k ( 1 + ∥ δ X ∥ p -var ; [ s , t ] ) ∣ ∣ ∣( Y , Y ′ )∣ ∣ ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] . (49)
∥ ∥ δ Y ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ Y ∥ L ∞ ( [ s , t ] × Ω ) ≲ k ∥ Y t ∥ ∞ + ( 1 + ∥ δ X ∥ p -var ; [ s , t ] ) ∣ ∣ ∣ ( Y , Y ′ ) ∣ ∣ ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] . (50) \lVert\!\lVert\delta Y\rVert_{p\text{-var}}\rVert_{\infty;k,\infty;[s,t]}+\lVert Y\rVert_{L^\infty([s,t]\times\Omega)}\lesssim_k\lVert Y_t\rVert_\infty+\left(1+\lVert\delta X\rVert_{p\text{-var};[s,t]}\right)\lvert\!\lvert\!\lvert(Y,Y')\rvert\!\rvert\!\rvert_{\mathbf{X};q,q';k,\infty;[s,t]}. \tag*{(50)} ∥ ∥ δ Y ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ Y ∥ L ∞ ([ s , t ] × Ω ) ≲ k ∥ Y t ∥ ∞ + ( 1 + ∥ δ X ∥ p -var ; [ s , t ] ) ∣ ∣ ∣( Y , Y ′ )∣ ∣ ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] . (50)
Moreover, for another X ‾ ∈ C p -var ( [ s , t ] , R d ) \overline{X} \in\mathcal{C}^{p\text{-var}}([s,t],\mathbb{R}^{d}) X ∈ C p -var ([ s , t ] , R d ) and ( Y ‾ , Y ‾ ′ ) ∈ RSM X ‾ ( p , p ′ ) -var ( [ s , t ] , R ) (\overline{Y},\overline{Y}') \in\operatorname{RSM}_{\overline{X}}^{(p,p')\text{-var}}([s,t],\mathbb{R}) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ s , t ] , R ) , it holds
∥ ∥ δ ( Y − Y ‾ ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ Y − Y ‾ ∥ L ∞ ( [ s , t ] × Ω ) ≲ k ∥ ( Y − Y ‾ ) t ∥ ∞ + ( 1 + ∥ δ X ∥ p -var ; [ s , t ] ) ∥ Y , Y ′ ; Y ‾ , Y ‾ ′ ∥ X , X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ Y ‾ ′ ∥ L ∞ ( [ s , t ] × Ω ) ∥ δ ( X − X ‾ ) ∥ p -var ; [ s , t ] . (51) \begin{aligned}
\lVert\lVert\delta(Y-\overline{Y})\rVert_{p\text{-var}}\rVert_{\infty;k,\infty;[s,t]}+\lVert Y-\overline{Y}\rVert_{L^{\infty}([s,t]\times\Omega)}
\lesssim_{k}&\lVert(Y-\overline{Y})_{t}\rVert_{\infty}\\
&+(1+\lVert\delta X\rVert_{p\text{-var};[s,t]})\lVert Y,Y';\overline{Y},\overline{Y}'\rVert_{X,\overline{X};q,q';k,\infty;[s,t]}\\
&+\lVert\overline{Y}'\rVert_{L^{\infty}([s,t]\times\Omega)}\lVert\delta(X-\overline{X})\rVert_{p\text{-var};[s,t]}. \tag*{(51)}
\end{aligned} ∥∥ δ ( Y − Y ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ Y − Y ∥ L ∞ ([ s , t ] × Ω ) ≲ k ∥( Y − Y ) t ∥ ∞ + ( 1 + ∥ δ X ∥ p -var ; [ s , t ] ) ∥ Y , Y ′ ; Y , Y ′ ∥ X , X ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ Y ′ ∥ L ∞ ([ s , t ] × Ω ) ∥ δ ( X − X ) ∥ p -var ; [ s , t ] . (51)
Proof. By the equation δ Y = Y ′ δ X + P Y + δ M Y \delta Y=Y'\delta X+P^{Y}+\delta M^{Y} δ Y = Y ′ δ X + P Y + δ M Y , for ( u , v ) ∈ Δ [ s , t ] (u,v)\in\Delta_{[s,t]} ( u , v ) ∈ Δ [ s , t ] we have
∥ ∥ δ Y ∥ p -var ; [ u , v ] ∥ F u ; k ≤ ∥ Y ′ ∥ L ∞ ( [ u , v ] × Ω ) ∥ δ X ∥ p -var ; [ u , v ] + ∥ ∥ P Y ∥ p -var ; [ u , v ] ∥ F u ; k + ∥ ∥ δ M Y ∥ p -var ; [ u , v ] ∥ F u ; k . (52) \lVert\lVert\delta Y\rVert_{p\text{-var};[u,v]}\rVert_{\mathcal{F}_{u};k}
\leq\lVert Y'\rVert_{L^{\infty}([u,v]\times\Omega)}\lVert\delta X\rVert_{p\text{-var};[u,v]}+\lVert\lVert P^{Y}\rVert_{p\text{-var};[u,v]}\rVert_{\mathcal{F}_{u};k}+\lVert\lVert\delta M^{Y}\rVert_{p\text{-var};[u,v]}\rVert_{\mathcal{F}_{u};k}. \tag*{(52)} ∥∥ δ Y ∥ p -var ; [ u , v ] ∥ F u ; k ≤ ∥ Y ′ ∥ L ∞ ([ u , v ] × Ω ) ∥ δ X ∥ p -var ; [ u , v ] + ∥∥ P Y ∥ p -var ; [ u , v ] ∥ F u ; k + ∥∥ δ M Y ∥ p -var ; [ u , v ] ∥ F u ; k . (52)
For the RHS of (52) , noting p ≥ q ′ ′ / 2 p\geq q''/2 p ≥ q ′′ /2 so that ∥ P Y ∥ p -var ; [ u , v ] ≤ ∥ P Y ∥ q ′ ′ / 2 -var ; [ u , v ] \lVert P^{Y}\rVert_{p\text{-var};[u,v]}\leq\lVert P^{Y}\rVert_{q''/2\text{-var};[u,v]} ∥ P Y ∥ p -var ; [ u , v ] ≤ ∥ P Y ∥ q ′′ /2 -var ; [ u , v ] ; and by Lépingle’s inequality (Lemma A.1) we get ∥ ∥ δ M Y ∥ p -var ; [ u , v ] ∥ F u ; k ≲ k ∥ δ M Y ∥ B M O ; [ s , t ] \lVert\lVert\delta M^{Y}\rVert_{p\text{-var};[u,v]}\rVert_{\mathcal{F}_{u};k}\lesssim_{k}\lVert\delta M^{Y}\rVert_{\mathrm{BMO};[s,t]} ∥∥ δ M Y ∥ p -var ; [ u , v ] ∥ F u ; k ≲ k ∥ δ M Y ∥ BMO ; [ s , t ] . Hence, (49) holds. Furthermore, for u ∈ [ s , t ] u\in[s,t] u ∈ [ s , t ] noting that ∥ Y u ∥ ∞ ≤ ∥ Y t ∥ ∞ + ∥ ∥ δ Y u , t ∥ F u ∥ k , ∞ \lVert Y_{u}\rVert_{\infty}\leq\lVert Y_{t}\rVert_{\infty}+\lVert\lVert\delta Y_{u,t}\rVert_{\mathcal{F}_{u}}\rVert_{k,\infty} ∥ Y u ∥ ∞ ≤ ∥ Y t ∥ ∞ + ∥∥ δ Y u , t ∥ F u ∥ k , ∞ from the inequality ∣ Y u ∣ = ∣ E u [ Y u ] ∣ ≤ ∣ Y t ∣ + ∣ E u [ δ Y u , t ] ∣ |Y_{u}|=|\mathbb{E}_{u}[Y_{u}]|\leq|Y_{t}|+|\mathbb{E}_{u}[\delta Y_{u,t}]| ∣ Y u ∣ = ∣ E u [ Y u ] ∣ ≤ ∣ Y t ∣ + ∣ E u [ δ Y u , t ] ∣ , (49) yields (50) . (51) can be proved similarly. □
The following Lemma 3.14, in contrast to Proposition 3.5 where G ∈ C b 2 G\in C^{2}_{b} G ∈ C b 2 is assumed, concerns a weaker condition G ∈ C b γ − 1 G\in C^{\gamma-1}_{b} G ∈ C b γ − 1 , and gives estimates on the composition of G G G and an RSM. Since G G G may not be twice differentiable, the interpolation argument cannot be applied as we did in Proposition 3.5. However, the lower regularity condition enables us to prove the comparison principle in Theorem 4.3 under the condition H ∈ C b γ H\in C^{\gamma}_{b} H ∈ C b γ , in which case G = D H G=DH G = D H .
Lemma 3.14. Assume (11) holds with p ′ ≤ q ′ p'\leq q' p ′ ≤ q ′ . Let X ∈ C p -var ( [ 0 , T ] , R d ) X\in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^{d}) X ∈ C p -var ([ 0 , T ] , R d ) , and let γ \gamma γ satisfy γ − 2 ≥ p / p ′ \gamma-2\geq p/p' γ − 2 ≥ p / p ′ , and G ∈ C b γ − 1 G\in C^{\gamma-1}_{b} G ∈ C b γ − 1 . Assume ( Y , Y ′ ) , ( Y ‾ , Y ‾ ′ ) ∈ RSM X ( p , p ′ ) -var ( [ 0 , T ] , R ) (Y,Y'),(\overline{Y},\overline{Y}')\in\operatorname{RSM}_{X}^{(p,p')\text{-var}}([0,T],\mathbb{R}) ( Y , Y ′ ) , ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , R ) . Let ( α , α ′ ) (\alpha,\alpha') ( α , α ′ ) be defined in (15) . Denote C G : = ∥ G ∥ C b γ − 1 C_{G}:=\lVert G\rVert_{C^{\gamma-1}_{b}} C G := ∥ G ∥ C b γ − 1 . Then, for every k ≥ 1 k\geq1 k ≥ 1 ,
∥ ( α , α ′ ) ∥ X ; q , q ′ ; k , ∞ ≲ k , γ C G ( 1 + ∥ δ X ∥ p -var ; [ 0 , T ] γ − 1 ) ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 + ∥ ( Y ‾ , Y ‾ ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 ) . (53) \lVert(\alpha,\alpha')\rVert_{X;q,q';k,\infty}
\lesssim_{k,\gamma} C_{G}(1+\lVert\delta X\rVert_{p\text{-var};[0,T]}^{\gamma-1})
\left(1+\lVert(Y,Y')\rVert_{X;q,q';k,\infty}^{\gamma-1}
+\lVert(\overline{Y},\overline{Y}')\rVert_{X;q,q';k,\infty}^{\gamma-1}\right). \tag*{(53)} ∥( α , α ′ ) ∥ X ; q , q ′ ; k , ∞ ≲ k , γ C G ( 1 + ∥ δ X ∥ p -var ; [ 0 , T ] γ − 1 ) ( 1 + ∥( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 + ∥( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 ) . (53)
Moreover, assume further that ( q , q ′ ) (q,q') ( q , q ′ ) satisfies ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) . Then, it follows that for all ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ ,
∥ ∥ δ ( ∫ s ⋅ α r d X r ) ∥ p -var ; [ s , t ] ∥ F s ; k , ∞ ≲ k , γ C G ∥ δ X ∥ p -var ; [ s , t ] ( 1 + ∥ δ X ∥ p -var ; [ 0 , T ] γ − 1 ) × ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 + ∥ ( Y ‾ , Y ‾ ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 ) . (54) \begin{aligned}
\left\lVert\left\lVert\delta\left(\int_{s}^{\cdot}\alpha_{r}\,dX_{r}\right)\right\rVert_{p\text{-var};[s,t]}\right\rVert_{\mathcal{F}_{s};k,\infty}
\lesssim_{k,\gamma}& C_{G}\lVert\delta X\rVert_{p\text{-var};[s,t]}(1+\lVert\delta X\rVert_{p\text{-var};[0,T]}^{\gamma-1})\\
&\times\left(1+\lVert(Y,Y')\rVert_{X;q,q';k,\infty}^{\gamma-1}
+\lVert(\overline{Y},\overline{Y}')\rVert_{X;q,q';k,\infty}^{\gamma-1}\right). \tag*{(54)}
\end{aligned} δ ( ∫ s ⋅ α r d X r ) p -var ; [ s , t ] F s ; k , ∞ ≲ k , γ C G ∥ δ X ∥ p -var ; [ s , t ] ( 1 + ∥ δ X ∥ p -var ; [ 0 , T ] γ − 1 ) × ( 1 + ∥( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 + ∥( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 ) . (54)
Proof. Without loss of generality, we assume γ ≤ 3 \gamma\leq3 γ ≤ 3 . By Lemma 3.4 we have ( α , α ′ ) ∈ RSM X ( p , p ′ ) -var (\alpha,\alpha')\in\operatorname{RSM}_{X}^{(p,p')\text{-var}} ( α , α ′ ) ∈ RSM X ( p , p ′ ) -var . We now prove (53) .
Let Y t λ : = ( 1 − λ ) Y t + λ Y ‾ t Y^{\lambda}_{t}:=(1-\lambda)Y_{t}+\lambda\overline{Y}_{t} Y t λ := ( 1 − λ ) Y t + λ Y t , ( Y λ ) t ′ : = ( 1 − λ ) Y t ′ + λ Y ‾ t ′ (Y^{\lambda})'_{t}:=(1-\lambda)Y'_{t}+\lambda\overline{Y}'_{t} ( Y λ ) t ′ := ( 1 − λ ) Y t ′ + λ Y t ′ , and denote G λ , ( G λ ) ′ G^{\lambda},(G^{\lambda})' G λ , ( G λ ) ′ by (17) . By (15) in Lemma 3.4, we have ∥ α ′ ∥ ∞ ≤ C G ( ∥ Y ′ ∥ ∞ + ∥ Y ‾ ′ ∥ ∞ ) \lVert\alpha'\rVert_{\infty}\leq C_{G}(\lVert Y'\rVert_{\infty}+\lVert\overline{Y}'\rVert_{\infty}) ∥ α ′ ∥ ∞ ≤ C G (∥ Y ′ ∥ ∞ + ∥ Y ′ ∥ ∞ ) . Furthermore for ∥ δ α ′ ∥ q ′ -var \lVert\delta\alpha'\rVert_{q'\text{-var}} ∥ δ α ′ ∥ q ′ -var , by the proof leading to (19) with p ′ p' p ′ replaced by q ′ q' q ′ , we have
∥ δ ( G λ ) ′ ∥ q ′ -var ; [ t , T ] ≤ C G ( ∥ δ Y ′ ∥ q ′ -var ; [ t , T ] + ∥ δ Y ‾ ′ ∥ q ′ -var ; [ t , T ] + ( ∥ δ Y ∥ p -var ; [ t , T ] γ − 2 + ∥ δ Y ‾ ∥ p -var ; [ t , T ] γ − 2 ) ( ∥ Y ′ ∥ ∞ + ∥ Y ‾ ′ ∥ ∞ ) ) . (55) \begin{aligned}
\lVert\delta(G^{\lambda})'\rVert_{q'\text{-var};[t,T]}
\leq C_{G}\big(&\lVert\delta Y'\rVert_{q'\text{-var};[t,T]}+\lVert\delta\overline{Y}'\rVert_{q'\text{-var};[t,T]}\\
&+(\lVert\delta Y\rVert_{p\text{-var};[t,T]}^{\gamma-2}+\lVert\delta\overline{Y}\rVert_{p\text{-var};[t,T]}^{\gamma-2})(\lVert Y'\rVert_{\infty}+\lVert\overline{Y}'\rVert_{\infty})\big). \tag*{(55)}
\end{aligned} ∥ δ ( G λ ) ′ ∥ q ′ -var ; [ t , T ] ≤ C G ( ∥ δ Y ′ ∥ q ′ -var ; [ t , T ] + ∥ δ Y ′ ∥ q ′ -var ; [ t , T ] + (∥ δ Y ∥ p -var ; [ t , T ] γ − 2 + ∥ δ Y ∥ p -var ; [ t , T ] γ − 2 ) (∥ Y ′ ∥ ∞ + ∥ Y ′ ∥ ∞ ) ) . (55)
Then applying Minkowski’s inequality, noting α ′ = ∫ 0 1 ( G λ ) ′ d λ \alpha'=\int_{0}^{1}(G^{\lambda})'\,d\lambda α ′ = ∫ 0 1 ( G λ ) ′ d λ , we get
\lVert\delta\alpha'\rVert_{q'\text{-var};[t,T]}\leq C_{G}\times\text{RHS of }[](#eq:3.47).
This, together with (49) in Lemma 3.13 yields
∥ α ′ ∥ ∞ + ∥ ∥ δ α ′ ∥ q ′ -var ∥ ∞ ; k , ∞ ≲ γ C G ( 1 + ∥ δ X ∥ p -var γ − 2 ) ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 + ∥ ( Y ‾ , Y ‾ ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 ) . (56) \lVert\alpha'\rVert_{\infty}+\lVert\lVert\delta\alpha'\rVert_{q'\text{-var}}\rVert_{\infty;k,\infty}
\lesssim_{\gamma} C_{G}(1+\lVert\delta X\rVert_{p\text{-var}}^{\gamma-2})
\left(1+\lVert(Y,Y')\rVert_{X;q,q';k,\infty}^{\gamma-1}
+\lVert(\overline{Y},\overline{Y}')\rVert_{X;q,q';k,\infty}^{\gamma-1}\right). \tag*{(56)} ∥ α ′ ∥ ∞ + ∥∥ δ α ′ ∥ q ′ -var ∥ ∞ ; k , ∞ ≲ γ C G ( 1 + ∥ δ X ∥ p -var γ − 2 ) ( 1 + ∥( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 + ∥( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 ) . (56)
Then for M α M^\alpha M α , noting ∥ δ M α ∥ B M O ≤ ∫ 0 1 ∥ δ M G λ ∥ B M O d λ ≤ C G ∫ 0 1 ∥ δ M Y λ ∥ B M O d λ \|\delta M^\alpha\|_{\mathrm{BMO}} \le\int_0^1 \|\delta M^{G^\lambda}\|_{\mathrm{BMO}}\,d\lambda\le C_G\int_0^1 \|\delta M^{Y^\lambda}\|_{\mathrm{BMO}}\,d\lambda ∥ δ M α ∥ BMO ≤ ∫ 0 1 ∥ δ M G λ ∥ BMO d λ ≤ C G ∫ 0 1 ∥ δ M Y λ ∥ BMO d λ , we have
∥ δ M α ∥ B M O ≤ C G ( ∥ δ M Y ∥ B M O + ∥ δ M Y ˉ ∥ B M O ) ≤ C G ( ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ + ∥ ( Y ˉ , Y ˉ ′ ) ∥ X ; q , q ′ ; k , ∞ ) . (57) \|\delta M^\alpha\|_{\mathrm{BMO}} \le C_G\left(\|\delta M^Y\|_{\mathrm{BMO}}+\|\delta M^{\bar Y}\|_{\mathrm{BMO}}\right) \le C_G\left(\|(Y,Y')\|_{X;q,q';k,\infty}+\|(\bar Y,\bar Y')\|_{X;q,q';k,\infty}\right). \tag*{(57)} ∥ δ M α ∥ BMO ≤ C G ( ∥ δ M Y ∥ BMO + ∥ δ M Y ˉ ∥ BMO ) ≤ C G ( ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ + ∥ ( Y ˉ , Y ˉ ′ ) ∥ X ; q , q ′ ; k , ∞ ) . (57)
Now we estimate P α P^\alpha P α . Denote P λ : = P G λ P^\lambda:=P^{G^\lambda} P λ := P G λ and define K λ ; i K^{\lambda;i} K λ ; i , i = 1 , 2 , 3 , 4 i=1,2,3,4 i = 1 , 2 , 3 , 4 , by (3.10). For K λ ; 1 K^{\lambda;1} K λ ; 1 , K λ ; 2 K^{\lambda;2} K λ ; 2 and K λ ; 3 K^{\lambda;3} K λ ; 3 , one can repeat the procedure in Step 1 of Lemma 3.4, but compute the q ′ 2 \frac{q'}{2} 2 q ′ -variation instead, and obtain (noting q ′ 2 ≥ p p ′ p + p ′ ≥ p γ − 1 \frac{q'}{2}\ge\frac{pp'}{p+p'}\ge\frac{p}{\gamma-1} 2 q ′ ≥ p + p ′ p p ′ ≥ γ − 1 p due to q ≥ p q\ge p q ≥ p and q ′ ≥ p ′ q'\ge p' q ′ ≥ p ′ )
∑ i = 1 3 ∥ K λ ; i ∥ q ′ 2 -var ; [ t , T ] ≤ C G ( ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ δ M Y λ ∥ p -var ; [ t , T ] + ∥ P Y λ ∥ q ′ 2 -var ; [ t , T ] + ∥ ( Y λ ) ′ ∥ ∞ ; [ t , T ] ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ δ X ∥ p -var ; [ t , T ] ) . (58) \sum_{i=1}^{3}\|K^{\lambda;i}\|_{\frac{q'}{2}\text{-var};[t,T]} \le C_G\left(\|\delta Y^\lambda\|_{p\text{-var};[t,T]}^{\gamma-2}\|\delta M^{Y^\lambda}\|_{p\text{-var};[t,T]}+\|P^{Y^\lambda}\|_{\frac{q'}{2}\text{-var};[t,T]}+\|(Y^\lambda)'\|_{\infty;[t,T]}\|\delta Y^\lambda\|_{p\text{-var};[t,T]}^{\gamma-2}\|\delta X\|_{p\text{-var};[t,T]}\right). \tag*{(58)} i = 1 ∑ 3 ∥ K λ ; i ∥ 2 q ′ -var ; [ t , T ] ≤ C G ( ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ δ M Y λ ∥ p -var ; [ t , T ] + ∥ P Y λ ∥ 2 q ′ -var ; [ t , T ] + ∥ ( Y λ ) ′ ∥ ∞ ; [ t , T ] ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ δ X ∥ p -var ; [ t , T ] ) . (58)
To estimate K λ ; 4 K^{\lambda;4} K λ ; 4 , similar to (3.14), by Lemma A.2 with r = q ′ 2 r=\frac{q'}{2} r = 2 q ′ , r 1 = p γ − 2 r_1=\frac{p}{\gamma-2} r 1 = γ − 2 p , k 1 = k 0 = 2 k k_1=k_0=2k k 1 = k 0 = 2 k ,
∥ ∥ K λ ; 4 ∥ q ′ 2 -var ; [ t , T ] ∥ F t ∥ k ≲ k , γ ∥ ∥ δ D G ( Y λ ) ∥ p γ − 2 -var ; [ t , T ] ∥ F t ∥ k 1 ∥ δ ⟨ M Y λ ⟩ t , T 1 2 ∥ F t ∥ k 0 ≤ C G ∥ ∥ δ Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k γ − 2 ∥ δ ⟨ M Y λ ⟩ t , T 1 2 ∥ F t ∥ k 0 . (59) \begin{aligned}
\|\|K^{\lambda;4}\|_{\frac{q'}{2}\text{-var};[t,T]}\|\mathcal{F}_t\|_k\lesssim_{k,\gamma}\|\|\delta DG(Y^\lambda)\|_{\frac{p}{\gamma-2}\text{-var};[t,T]}\|\mathcal{F}_t\|_{k_1}\|\delta\langle M^{Y^\lambda}\rangle^{\frac12}_{t,T}\|\mathcal{F}_t\|_{k_0} \\
&\le C_G\|\|\delta Y^\lambda\|_{p\text{-var};[t,T]}\|\mathcal{F}_t\|_k^{\gamma-2}\|\delta\langle M^{Y^\lambda}\rangle^{\frac12}_{t,T}\|\mathcal{F}_t\|_{k_0}. \tag*{(59)}
\end{aligned} ∥∥ K λ ; 4 ∥ 2 q ′ -var ; [ t , T ] ∥ F t ∥ k ≲ k , γ ∥∥ δ D G ( Y λ ) ∥ γ − 2 p -var ; [ t , T ] ∥ F t ∥ k 1 ∥ δ ⟨ M Y λ ⟩ t , T 2 1 ∥ F t ∥ k 0 ≤ C G ∥∥ δ Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k γ − 2 ∥ δ ⟨ M Y λ ⟩ t , T 2 1 ∥ F t ∥ k 0 . (59)
The k k k -th conditional moment of the first term on the RHS of (3.50) is bounded via Hölder’s inequality, noting 1 k 1 + 1 k 0 = 1 k \frac{1}{k_1}+\frac{1}{k_0}=\frac{1}{k} k 1 1 + k 0 1 = k 1 :
∥ ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ δ M Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k ≤ ∥ ∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ F t ∥ k 1 ∥ ∥ δ M Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k 0 ≤ ∥ ∥ δ Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k 1 γ − 2 ∥ ∥ δ M Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k 0 \begin{aligned}
\|\|\delta Y^\lambda\|_{p\text{-var};[t,T]}^{\gamma-2}\|\delta M^{Y^\lambda}\|_{p\text{-var};[t,T]}\|\mathcal{F}_t\|_k
\le\|\|\delta Y^\lambda\|_{p\text{-var};[t,T]}^{\gamma-2}\|\mathcal{F}_t\|_{k_1}\|\|\delta M^{Y^\lambda}\|_{p\text{-var};[t,T]}\|\mathcal{F}_t\|_{k_0} \\
&\le\|\|\delta Y^\lambda\|_{p\text{-var};[t,T]}\|\mathcal{F}_t\|_{k_1}^{\gamma-2}\|\|\delta M^{Y^\lambda}\|_{p\text{-var};[t,T]}\|\mathcal{F}_t\|_{k_0}
\end{aligned} ∥∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ δ M Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k ≤ ∥∥ δ Y λ ∥ p -var ; [ t , T ] γ − 2 ∥ F t ∥ k 1 ∥∥ δ M Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k 0 ≤ ∥∥ δ Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k 1 γ − 2 ∥∥ δ M Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k 0
≲ k ∥ ∥ δ Y λ ∥ p -var ∥ F t ∥ k , ∞ ∥ ∥ δ M Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k 0 , \lesssim_k \|\|\delta Y^\lambda\|_{p\text{-var}}\|\mathcal{F}_t\|_{k,\infty}\|\|\delta M^{Y^\lambda}\|_{p\text{-var};[t,T]}\|\mathcal{F}_t\|_{k_0}, ≲ k ∥∥ δ Y λ ∥ p -var ∥ F t ∥ k , ∞ ∥∥ δ M Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k 0 ,
where the last inequality is due to ∥ ∥ δ Y λ ∥ p -var ∥ ∞ ; k 1 , ∞ ≲ k ∥ ∥ δ Y λ ∥ p -var ∥ ∞ ; k , ∞ \|\|\delta Y^\lambda\|_{p\text{-var}}\|_{\infty;k_1,\infty}\lesssim_k\|\|\delta Y^\lambda\|_{p\text{-var}}\|_{\infty;k,\infty} ∥∥ δ Y λ ∥ p -var ∥ ∞ ; k 1 , ∞ ≲ k ∥∥ δ Y λ ∥ p -var ∥ ∞ ; k , ∞ from Lemma 3.2.
Thus, combining (3.50) and (3.51), and applying Lemma A.1 to M Y λ M^{Y^\lambda} M Y λ , we get
∥ ∥ P λ ∥ q ′ 2 -var ; [ t , T ] ∥ F t ∥ k ≲ k , γ C G ( ∥ ∥ δ Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k γ − 2 ∥ δ ⟨ M Y λ ⟩ t , T 1 2 ∥ F t ∥ k 0 + ∥ ∥ P Y λ ∥ q ′ 2 -var ; [ t , T ] ∥ F t ∥ k + ∥ ( Y λ ) ′ ∥ ∞ ∥ ∥ δ Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k γ − 2 ∥ δ X ∥ p -var ; [ t , T ] ) . (60) \|\|P^\lambda\|_{\frac{q'}{2}\text{-var};[t,T]}\|\mathcal{F}_t\|_k\lesssim_{k,\gamma}C_G\left(\|\|\delta Y^\lambda\|_{p\text{-var};[t,T]}\|\mathcal{F}_t\|_k^{\gamma-2}\|\delta\langle M^{Y^\lambda}\rangle^{\frac12}_{t,T}\|\mathcal{F}_t\|_{k_0}+\|\|P^{Y^\lambda}\|_{\frac{q'}{2}\text{-var};[t,T]}\|\mathcal{F}_t\|_k+\|(Y^\lambda)'\|_\infty\|\|\delta Y^\lambda\|_{p\text{-var};[t,T]}\|\mathcal{F}_t\|_k^{\gamma-2}\|\delta X\|_{p\text{-var};[t,T]}\right). \tag*{(60)} ∥∥ P λ ∥ 2 q ′ -var ; [ t , T ] ∥ F t ∥ k ≲ k , γ C G ( ∥∥ δ Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k γ − 2 ∥ δ ⟨ M Y λ ⟩ t , T 2 1 ∥ F t ∥ k 0 + ∥∥ P Y λ ∥ 2 q ′ -var ; [ t , T ] ∥ F t ∥ k + ∥ ( Y λ ) ′ ∥ ∞ ∥∥ δ Y λ ∥ p -var ; [ t , T ] ∥ F t ∥ k γ − 2 ∥ δ X ∥ p -var ; [ t , T ] ) . (60)
Since ( Y λ , ( Y λ ) ′ , P Y λ , M Y λ ) (Y^\lambda,(Y^\lambda)',P^{Y^\lambda},M^{Y^\lambda}) ( Y λ , ( Y λ ) ′ , P Y λ , M Y λ ) is a convex combination of ( Y , Y ′ , P Y , M Y ) (Y,Y',P^Y,M^Y) ( Y , Y ′ , P Y , M Y ) and ( Y ˉ , Y ˉ ′ , P Y ˉ , M Y ˉ ) (\bar Y,\bar Y',P^{\bar Y},M^{\bar Y}) ( Y ˉ , Y ˉ ′ , P Y ˉ , M Y ˉ ) , and since the sum ∥ ∥ δ Y ∥ p -var ∥ ∞ ; k , ∞ + ∥ ∥ δ Y ˉ ∥ p -var ∥ ∞ ; k , ∞ \|\|\delta Y\|_{p\text{-var}}\|_{\infty;k,\infty}+\|\|\delta\bar Y\|_{p\text{-var}}\|_{\infty;k,\infty} ∥∥ δ Y ∥ p -var ∥ ∞ ; k , ∞ + ∥∥ δ Y ˉ ∥ p -var ∥ ∞ ; k , ∞ is dominated by the product of ( 1 + ∣ δ X ∣ p -var ) (1+|\delta X|_{p\text{-var}}) ( 1 + ∣ δ X ∣ p -var ) and ( ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ + ∥ ( Y ˉ , Y ˉ ′ ) ∥ X ; q , q ′ ; k , ∞ ) (\|(Y,Y')\|_{X;q,q';k,\infty}+\|(\bar Y,\bar Y')\|_{X;q,q';k,\infty}) ( ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ + ∥ ( Y ˉ , Y ˉ ′ ) ∥ X ; q , q ′ ; k , ∞ ) (see Lemma 3.13), estimate (3.52) implies that
∥ ∥ P λ ∥ q ′ 2 -var ∥ ∞ ; k , ∞ ≲ k , γ C G ( 1 + ∣ δ X ∣ p -var ; [ 0 , T ] γ − 1 ) ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 + ∥ ( Y ˉ , Y ˉ ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 ) . \|\|P^\lambda\|_{\frac{q'}{2}\text{-var}}\|_{\infty;k,\infty}\lesssim_{k,\gamma}C_G(1+|\delta X|_{p\text{-var};[0,T]}^{\gamma-1})\left(1+\|(Y,Y')\|_{X;q,q';k,\infty}^{\gamma-1}+\|(\bar Y,\bar Y')\|_{X;q,q';k,\infty}^{\gamma-1}\right). ∥∥ P λ ∥ 2 q ′ -var ∥ ∞ ; k , ∞ ≲ k , γ C G ( 1 + ∣ δ X ∣ p -var ; [ 0 , T ] γ − 1 ) ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 + ∥ ( Y ˉ , Y ˉ ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 ) .
Therefore, applying Minkowski’s inequality to P s , t α = ∫ 0 1 P s , t λ d λ P^\alpha_{s,t}=\int_0^1P^\lambda_{s,t}\,d\lambda P s , t α = ∫ 0 1 P s , t λ d λ yields
∥ ∥ P α ∥ q ′ 2 -var ∥ ∞ ; k , ∞ ≲ k , γ C G ( 1 + ∣ δ X ∣ p -var ; [ 0 , T ] γ − 1 ) ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 + ∥ ( Y ˉ , Y ˉ ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 ) . \|\|P^\alpha\|_{\frac{q'}{2}\text{-var}}\|_{\infty;k,\infty}\lesssim_{k,\gamma}C_G(1+|\delta X|_{p\text{-var};[0,T]}^{\gamma-1})\left(1+\|(Y,Y')\|_{X;q,q';k,\infty}^{\gamma-1}+\|(\bar Y,\bar Y')\|_{X;q,q';k,\infty}^{\gamma-1}\right). ∥∥ P α ∥ 2 q ′ -var ∥ ∞ ; k , ∞ ≲ k , γ C G ( 1 + ∣ δ X ∣ p -var ; [ 0 , T ] γ − 1 ) ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 + ∥ ( Y ˉ , Y ˉ ′ ) ∥ X ; q , q ′ ; k , ∞ γ − 1 ) .
This together with (3.48) and (3.49) implies the desired estimate (3.45).
Finally, assume further ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) (hence ( p , p ′ ) (p,p') ( p , p ′ ) holds from p ′ ≤ q ′ p'\le q' p ′ ≤ q ′ ) and set Γ t : = ∫ 0 t α r d X r \Gamma_t:=\int_0^t\alpha_r\,dX_r Γ t := ∫ 0 t α r d X r . Noting δ Γ s , t = R s , t Γ + α s ′ δ X s , t \delta\Gamma_{s,t}=R^\Gamma_{s,t}+\alpha'_s\delta X_{s,t} δ Γ s , t = R s , t Γ + α s ′ δ X s , t , the estimate (3.46) follows directly from (3.45) and Proposition 3.6. □ \square □
The next proposition shows that the seminorm ∥ ⋅ ∥ X ; q , q ′ ; k , ∞ \|\cdot\|_{X;q,q';k,\infty} ∥ ⋅ ∥ X ; q , q ′ ; k , ∞ over a large interval is dominated by the sum of the corresponding seminorms over small intervals.
Proposition 3.15. Assume (11) holds. Let ( s , t ) ∈ Δ (s,t) \in\Delta ( s , t ) ∈ Δ , and suppose X ∈ C p -var ( [ s , t ] , R d ) \mathbf{X} \in\mathcal{C}^{p\text{-var}}([s,t],\mathbb{R}^{d}) X ∈ C p -var ([ s , t ] , R d ) and ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( [ s , t ] , R ) (Y,Y') \in\operatorname{RSM}_{\mathbf{X}}^{(p,p')\text{-var}}([s,t],\mathbb{R}) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ s , t ] , R ) . Then, we have
∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ ( 1 + ∣ δ X ∣ q -var ; [ s , t ] q ) ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , u ] + ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ u , t ] . (61) \left|(Y,Y')\right|_{\mathbf{X};q,q';k,\infty;[s,t]} \leq\left(1+\left|\delta X\right|_{q\text{-var};[s,t]}^{q}\right)\left|(Y,Y')\right|_{\mathbf{X};q,q';k,\infty;[s,u]}+\left|(Y,Y')\right|_{\mathbf{X};q,q';k,\infty;[u,t]}. \tag*{(61)} ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ ( 1 + ∣ δ X ∣ q -var ; [ s , t ] q ) ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , u ] + ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ u , t ] . (61)
Proof. Using the subadditivity of q ′ q' q ′ -variation, we get that for s ≤ a ≤ u ≤ b ≤ t s \leq a \leq u \leq b \leq t s ≤ a ≤ u ≤ b ≤ t ,
∣ ∥ Y ′ ∥ q ′ -var ; [ a , b ] F a ∣ k , ∞ ≤ ∣ ∥ Y ′ ∥ q ′ -var ; [ a , u ] F a ∣ k , ∞ + ∣ ∥ Y ′ ∥ q ′ -var ; [ u , b ] F u ∣ k , ∞ , \left|\left\|Y'\right\|_{q'\text{-var};[a,b]}\mathcal{F}_{a}\right|_{k,\infty}\leq\left|\left\|Y'\right\|_{q'\text{-var};[a,u]}\mathcal{F}_{a}\right|_{k,\infty}+\left|\left\|Y'\right\|_{q'\text{-var};[u,b]}\mathcal{F}_{u}\right|_{k,\infty}, ∥ Y ′ ∥ q ′ -var ; [ a , b ] F a k , ∞ ≤ ∥ Y ′ ∥ q ′ -var ; [ a , u ] F a k , ∞ + ∥ Y ′ ∥ q ′ -var ; [ u , b ] F u k , ∞ ,
which implies
∣ ∥ Y ′ ∥ q ′ -var ∣ ∞ ; k , ∞ ; [ s , t ] ≤ ∣ ∥ Y ′ ∥ q ′ -var ∣ ∞ ; k , ∞ ; [ s , u ] + ∣ ∥ Y ′ ∥ q ′ -var ∣ ∞ ; k , ∞ ; [ u , t ] . \left|\left\|Y'\right\|_{q'\text{-var}}\right|_{\infty;k,\infty;[s,t]}\leq\left|\left\|Y'\right\|_{q'\text{-var}}\right|_{\infty;k,\infty;[s,u]}+\left|\left\|Y'\right\|_{q'\text{-var}}\right|_{\infty;k,\infty;[u,t]}. ∥ Y ′ ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] ≤ ∥ Y ′ ∥ q ′ -var ∞ ; k , ∞ ; [ s , u ] + ∥ Y ′ ∥ q ′ -var ∞ ; k , ∞ ; [ u , t ] .
For the pathwise remainder P Y P^{Y} P Y , by Definition (10) we have that for u ∈ [ r , v ] ⊂ [ a , b ] u \in[r,v] \subset[a,b] u ∈ [ r , v ] ⊂ [ a , b ] ,
P r , v Y = P r , u Y + P u , v Y + δ Y r , u ′ δ X u , v , P^{Y}_{r,v}=P^{Y}_{r,u}+P^{Y}_{u,v}+\delta Y'_{r,u}\delta X_{u,v}, P r , v Y = P r , u Y + P u , v Y + δ Y r , u ′ δ X u , v ,
and hence by Hölder’s inequality,
∣ ∥ P Y ∥ q ′ ′ 2 -var ; [ a , b ] F a ∣ k , ∞ ≤ ∣ ∥ P Y ∥ q ′ ′ 2 -var ; [ a , u ] F a ∣ k , ∞ + ∣ ∥ P Y ∥ q ′ ′ 2 -var ; [ u , b ] F u ∣ k , ∞ + ∣ ∥ Y ′ ∥ q ′ -var ; [ a , u ] F a ∣ k , ∞ ∣ X ∣ q -var ; [ s , t ] . \left|\left\|P^{Y}\right\|_{\frac{q''}{2}\text{-var};[a,b]}\mathcal{F}_{a}\right|_{k,\infty}
\leq\left|\left\|P^{Y}\right\|_{\frac{q''}{2}\text{-var};[a,u]}\mathcal{F}_{a}\right|_{k,\infty}
+\left|\left\|P^{Y}\right\|_{\frac{q''}{2}\text{-var};[u,b]}\mathcal{F}_{u}\right|_{k,\infty}
+\left|\left\|Y'\right\|_{q'\text{-var};[a,u]}\mathcal{F}_{a}\right|_{k,\infty}\left|X\right|_{q\text{-var};[s,t]}. P Y 2 q ′′ -var ; [ a , b ] F a k , ∞ ≤ P Y 2 q ′′ -var ; [ a , u ] F a k , ∞ + P Y 2 q ′′ -var ; [ u , b ] F u k , ∞ + ∥ Y ′ ∥ q ′ -var ; [ a , u ] F a k , ∞ ∣ X ∣ q -var ; [ s , t ] .
Therefore, we have
∣ ∥ P Y ∥ q ′ ′ 2 -var ∣ ∞ ; k , ∞ ; [ s , t ] ≤ ∣ ∥ P Y ∥ q ′ ′ 2 -var ∣ ∞ ; k , ∞ ; [ s , u ] + ∣ ∥ P Y ∥ q ′ ′ 2 -var ∣ ∞ ; k , ∞ ; [ u , t ] + ∣ ∥ Y ′ ∥ q ′ -var ; [ s , u ] ∣ ∞ ; k , ∞ ∣ X ∣ q -var ; [ s , t ] . \left|\left\|P^{Y}\right\|_{\frac{q''}{2}\text{-var}}\right|_{\infty;k,\infty;[s,t]}
\leq\left|\left\|P^{Y}\right\|_{\frac{q''}{2}\text{-var}}\right|_{\infty;k,\infty;[s,u]}
+\left|\left\|P^{Y}\right\|_{\frac{q''}{2}\text{-var}}\right|_{\infty;k,\infty;[u,t]}
+\left|\left\|Y'\right\|_{q'\text{-var};[s,u]}\right|_{\infty;k,\infty}\left|X\right|_{q\text{-var};[s,t]}. P Y 2 q ′′ -var ∞ ; k , ∞ ; [ s , t ] ≤ P Y 2 q ′′ -var ∞ ; k , ∞ ; [ s , u ] + P Y 2 q ′′ -var ∞ ; k , ∞ ; [ u , t ] + ∥ Y ′ ∥ q ′ -var ; [ s , u ] ∞ ; k , ∞ ∣ X ∣ q -var ; [ s , t ] .
Finally, noting the subadditivity of BMO norms, the desired (61) follows. □
Lemma 3.16. Assume (11) . Let ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , and suppose X ∈ C p -var ( [ s , t ] , R d ) \mathbf{X}\in\mathcal{C}^{p\text{-var}}([s,t],\mathbb{R}^{d}) X ∈ C p -var ([ s , t ] , R d ) . Denote
∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] : = ∥ Y t ∥ ∞ + ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] . \left\|(Y,Y')\right\|_{\mathbf{X};q,q';k,\infty;[s,t]}:=\left\|Y_{t}\right\|_{\infty}+\left|(Y,Y')\right|_{\mathbf{X};q,q';k,\infty;[s,t]}. ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] := ∥ Y t ∥ ∞ + ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] .
Then, the space RSM X ( p , p ′ ) -var ( [ s , t ] , R ) \operatorname{RSM}_{\mathbf{X}}^{(p,p')\text{-var}}([s,t],\mathbb{R}) RSM X ( p , p ′ ) -var ([ s , t ] , R ) , ∥ ⋅ ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] \left\|\cdot\right\|_{\mathbf{X};q,q';k,\infty;[s,t]} ∥ ⋅ ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] forms a Banach space.
Proof. Let { ( Y n , ( Y n ) ′ ) } n ≥ 1 \{(Y^{n},(Y^{n})')\}_{n\geq1} {( Y n , ( Y n ) ′ ) } n ≥ 1 be a Cauchy sequence in this space. By Lemma 2.1 and the Banach property of BMO martingales (see, e.g., eq:ref-18 , Chapter 2.1) there exists ( Y ′ , P , M ) (Y',P,M) ( Y ′ , P , M ) such that
lim n → ∞ ∥ ( Y n ) ′ − Y ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∣ ∥ δ ( ( Y n ) ′ − Y ′ ) ∥ q ′ -var ∣ ∞ ; k , ∞ ; [ s , t ] + ∣ ∥ P Y n − P ∥ q ′ ′ 2 -var ∣ ∞ ; k , ∞ ; [ s , t ] + ∥ δ ( M Y n − M ) ∥ B M O ; [ s , t ] = 0. (62) \lim_{n\to\infty}\left\|(Y^{n})'-Y'\right\|_{L^{\infty}([s,t]\times\Omega)}
+\left|\left\|\delta((Y^{n})'-Y')\right\|_{q'\text{-var}}\right|_{\infty;k,\infty;[s,t]}
+\left|\left\|P^{Y^{n}}-P\right\|_{\frac{q''}{2}\text{-var}}\right|_{\infty;k,\infty;[s,t]}
+\left\|\delta(M^{Y^{n}}-M)\right\|_{\mathrm{BMO};[s,t]}=0. \tag*{(62)} n → ∞ lim ∥ ( Y n ) ′ − Y ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ δ (( Y n ) ′ − Y ′ ) ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] + P Y n − P 2 q ′′ -var ∞ ; k , ∞ ; [ s , t ] + δ ( M Y n − M ) BMO ; [ s , t ] = 0. (62)
Furthermore, let Y r : = lim n → ∞ Y t n − Y r ′ δ X r , t − P r , t − δ M r , t Y_{r}:=\lim_{n\to\infty}Y_{t}^{n}-Y'_{r}\delta X_{r,t}-P_{r,t}-\delta M_{r,t} Y r := lim n → ∞ Y t n − Y r ′ δ X r , t − P r , t − δ M r , t , we see that δ Y u , v = ( Y ′ ) u δ X u , v + P u , v + δ M u , v \delta Y_{u,v}=(Y')_{u}\delta X_{u,v}+P_{u,v}+\delta M_{u,v} δ Y u , v = ( Y ′ ) u δ X u , v + P u , v + δ M u , v . Hence, to prove ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var (Y,Y')\in\operatorname{RSM}_{\mathbf{X}}^{(p,p')\text{-var}} ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var , it suffices to check that a.s., ∥ δ Y ∥ p -var ; [ s , t ] + ∥ δ Y ′ ∥ p ′ -var ; [ s , t ] + ∥ P ∥ p p ′ / ( p + p ′ ) -var ; [ s , t ] < ∞ \|\delta Y\|_{p\text{-var};[s,t]}+\|\delta Y'\|_{p'\text{-var};[s,t]}+\|P\|_{pp'/(p+p')\text{-var};[s,t]}<\infty ∥ δ Y ∥ p -var ; [ s , t ] + ∥ δ Y ′ ∥ p ′ -var ; [ s , t ] + ∥ P ∥ p p ′ / ( p + p ′ ) -var ; [ s , t ] < ∞ .
Indeed, by the convergence in (62) , after passing to a subsequence if necessary, ( ( Y n ) ′ , P Y n ) ((Y^{n})',P^{Y^{n}}) (( Y n ) ′ , P Y n ) converges to ( Y ′ , P ) (Y',P) ( Y ′ , P ) in the uniform norm a.s. Hence, by the lower semi-continuity of p ′ p' p ′ -variation norm (and, respectively, the p p ′ / ( p + p ′ ) pp'/(p+p') p p ′ / ( p + p ′ ) -variation norm) established in eq:ref-14 , Lemma 5.12, we see that a.s. Y ′ Y' Y ′ and P P P have finite p ′ p' p ′ -variation and p p ′ / ( p + p ′ ) pp'/(p+p') p p ′ / ( p + p ′ ) -variation, respectively. Furthermore, by Lemma A.1, after passing to a further subsequence, M Y n M^{Y^{n}} M Y n converges to M M M in the uniform norm a.s. By the equation δ Y u , v = Y u ′ δ X u , v + P u , v + δ M u , v \delta Y_{u,v}=Y'_{u}\delta X_{u,v}+P_{u,v}+\delta M_{u,v} δ Y u , v = Y u ′ δ X u , v + P u , v + δ M u , v , it follows that Y n Y^{n} Y n also converges to Y Y Y in the uniform norm a.s., which in turn implies that Y Y Y has finite p p p -variation a.s. Therefore, ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ( [ s , t ] , R ) (Y,Y')\in\operatorname{RSM}_{\mathbf{X}}^{(p,p')\text{-var}}([s,t],\mathbb{R}) ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var ([ s , t ] , R ) with P Y = P P^{Y}=P P Y = P and M Y = M M^{Y}=M M Y = M . Finally, applying (62) again we obtain
lim n → ∞ ∣ ( Y n , ( Y n ) ′ ; Y , Y ′ ) ∣ X , X ; q , q ′ ; k , ∞ ; [ s , t ] = 0. \lim_{n\to\infty}\left|(Y^{n},(Y^{n})';Y,Y')\right|_{\mathbf{X},\mathbf{X};q,q';k,\infty;[s,t]}=0. n → ∞ lim ∣ ( Y n , ( Y n ) ′ ; Y , Y ′ ) ∣ X , X ; q , q ′ ; k , ∞ ; [ s , t ] = 0.
Clearly lim n → ∞ ∥ Y t − Y t n ∥ ∞ = 0 \lim_{n \to\infty}\|Y_t-Y_t^n\|_{\infty}=0 lim n → ∞ ∥ Y t − Y t n ∥ ∞ = 0 , which completes the proof. □
Given two RSMs ( Y , Y ′ ) (Y,Y') ( Y , Y ′ ) and ( Y ‾ , Y ‾ ′ ) (\overline{Y},\overline{Y}') ( Y , Y ′ ) controlled by different rough paths X \mathbf{X} X and X ‾ \overline{\mathbf{X}} X , consider ( H ( Y ) , H ( Y ) ′ ) (H(Y),H(Y)') ( H ( Y ) , H ( Y ) ′ ) and ( H ‾ ( Y ‾ ) , H ‾ ( Y ‾ ) ′ ) (\overline{H}(\overline{Y}),\overline{H}(\overline{Y})') ( H ( Y ) , H ( Y ) ′ ) being the corresponding composition of different nonlinear functions. The following Lemma 3.17 establishes an upper bound for the difference between the rough stochastic integral of ( H ( Y ) , H ( Y ) ′ ) (H(Y),H(Y)') ( H ( Y ) , H ( Y ) ′ ) and ( H ‾ ( Y ‾ ) , H ‾ ( Y ‾ ) ′ ) (\overline{H}(\overline{Y}),\overline{H}(\overline{Y})') ( H ( Y ) , H ( Y ) ′ ) . In particular, this estimate is locally Lipschitz continuous with respect to ∣ Y , Y ′ ; Y ‾ , Y ‾ ′ ∣ X , X ‾ ; q , q ′ ; k , ∞ \lvert Y,Y';\overline{Y},\overline{Y}'\rvert_{\mathbf{X},\overline{\mathbf{X}};q,q';k,\infty} ∣ Y , Y ′ ; Y , Y ′ ∣ X , X ; q , q ′ ; k , ∞ .
To facilitate the statement of Lemma 3.17, let R 1 > 0 R_1>0 R 1 > 0 be a constant such that
∥ ξ ∥ ∞ + ∥ ξ ‾ ∥ ∞ + ∥ H ∥ C b γ + ∥ H ‾ ∥ C b γ + ∣ δ X ∣ p -var ; [ 0 , T ] + ∣ δ X ‾ ∣ p -var ; [ 0 , T ] ≤ R 1 . \|\xi\|_{\infty}+\|\overline{\xi}\|_{\infty}+\|H\|_{C_b^\gamma}+\|\overline{H}\|_{C_b^\gamma}+|\delta\mathbf{X}|_{p\text{-var};[0,T]}+|\delta\overline{\mathbf{X}}|_{p\text{-var};[0,T]}\leq R_1. ∥ ξ ∥ ∞ + ∥ ξ ∥ ∞ + ∥ H ∥ C b γ + ∥ H ∥ C b γ + ∣ δ X ∣ p -var ; [ 0 , T ] + ∣ δ X ∣ p -var ; [ 0 , T ] ≤ R 1 .
and denote Θ 1 : = ( Θ , C f , R 1 ) \Theta_1:=(\Theta,C_f,R_1) Θ 1 := ( Θ , C f , R 1 ) . Recall the parameter q ′ ′ / 2 = q q ′ / ( q + q ′ ) q''/2=qq'/(q+q') q ′′ /2 = q q ′ / ( q + q ′ ) .
Lemma 3.17. Given ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , assume ( γ , p , q , q ′ ) (\gamma,p,q,q') ( γ , p , q , q ′ ) , H , H ‾ ∈ C b γ H,\overline{H}\in C_b^\gamma H , H ∈ C b γ , and X , X ‾ ∈ C p -var ( [ 0 , T ] , R d ) \mathbf{X},\overline{\mathbf{X}}\in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^d) X , X ∈ C p -var ([ 0 , T ] , R d ) . Suppose that ( Y , Y ′ ) (Y,Y') ( Y , Y ′ ) and ( Y ‾ , Y ‾ ′ ) (\overline{Y},\overline{Y}') ( Y , Y ′ ) satisfy (3.27) with p ′ = p p'=p p ′ = p and Y t , Y ‾ t ∈ L ∞ Y_t,\overline{Y}_t\in L^\infty Y t , Y t ∈ L ∞ . Denote
I v X : = ∫ s v H ( Y r ) d X r , I v X ‾ : = ∫ s v H ‾ ( Y ‾ r ) d X ‾ r , v ∈ [ s , t ] . (63) I_v^{\mathbf{X}}:=\int_s^v H(Y_r)d\mathbf{X}_r,\qquad I_v^{\overline{\mathbf{X}}}:=\int_s^v \overline{H}(\overline{Y}_r)d\overline{\mathbf{X}}_r,\qquad v\in[s,t]. \tag*{(63)} I v X := ∫ s v H ( Y r ) d X r , I v X := ∫ s v H ( Y r ) d X r , v ∈ [ s , t ] . (63)
Assume further that for some constant C > 0 C>0 C > 0 ,
∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] + ∣ ( Y ‾ , Y ‾ ′ ) ∣ X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] ≤ C . (64) \lvert(Y,Y')\rvert_{\mathbf{X};q,q';k,\infty;[s,t]}+\lvert(\overline{Y},\overline{Y}')\rvert_{\overline{\mathbf{X}};q,q';k,\infty;[s,t]}\leq C. \tag*{(64)} ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] + ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ C . (64)
Then, the following estimates hold, with the constant depending implicitly on ( k , γ , Θ 1 , C ) (k,\gamma,\Theta_1,C) ( k , γ , Θ 1 , C ) :
∥ R X − R X ‾ ∥ p 2 -var ∞ ; k , ∞ ; [ s , t ] + ∥ δ ( I X − I X ‾ ) ∥ p -var ∞ ; k , ∞ ; [ s , t ] ≲ ∣ δ X ∣ p -var ; [ s , t ] ( ∥ Y t − Y ‾ t ∥ ∞ + ∣ Y , Y ′ ; Y ‾ , ( Y ‾ ) ′ ∣ X , X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ H − H ‾ ∥ C b γ − 1 ) + ∣ δ ( X − X ‾ ) ∣ p -var ; [ s , t ] . (65) \left\lVert R^{\mathbf{X}}-R^{\overline{\mathbf{X}}}\right\rVert_{\frac{p}{2}\text{-var}\,\infty;k,\infty;[s,t]}+\left\lVert\delta(I^{\mathbf{X}}-I^{\overline{\mathbf{X}}})\right\rVert_{p\text{-var}\,\infty;k,\infty;[s,t]}
\lesssim|\delta\mathbf{X}|_{p\text{-var};[s,t]}\left(\|Y_t-\overline{Y}_t\|_{\infty}
+\lvert Y,Y';\overline{Y},(\overline{Y})'\rvert_{\mathbf{X},\overline{\mathbf{X}};q,q';k,\infty;[s,t]}+\|H-\overline{H}\|_{C_b^{\gamma-1}}\right)+|\delta(\mathbf{X}-\overline{\mathbf{X}})|_{p\text{-var};[s,t]}. \tag*{(65)} R X − R X 2 p -var ∞ ; k , ∞ ; [ s , t ] + δ ( I X − I X ) p -var ∞ ; k , ∞ ; [ s , t ] ≲ ∣ δ X ∣ p -var ; [ s , t ] ( ∥ Y t − Y t ∥ ∞ + ∣ Y , Y ′ ; Y , ( Y ) ′ ∣ X , X ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ H − H ∥ C b γ − 1 ) + ∣ δ ( X − X ) ∣ p -var ; [ s , t ] . (65)
Proof. Without loss of generality, we assume γ ≤ 3 \gamma\leq3 γ ≤ 3 . By Proposition 3.12, we have
∥ R X − R X ‾ ∥ p 2 -var ∞ ; k , ∞ ; [ s , t ] ≲ Θ 1 ∣ δ ( X − X ‾ ) ∣ p -var ; [ s , t ] + ∣ δ X ∣ p -var ; [ s , t ] ∣ H ( Y ) , H ( Y ) ′ ; H ‾ ( Y ‾ ) , H ‾ ( Y ‾ ) ′ ∣ X , X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] , (66) \left\lVert R^{\mathbf{X}}-R^{\overline{\mathbf{X}}}\right\rVert_{\frac{p}{2}\text{-var}\,\infty;k,\infty;[s,t]}
\lesssim_{\Theta_1}|\delta(\mathbf{X}-\overline{\mathbf{X}})|_{p\text{-var};[s,t]}+|\delta\mathbf{X}|_{p\text{-var};[s,t]}\lvert H(Y),H(Y)';\overline{H}(\overline{Y}),\overline{H}(\overline{Y})'\rvert_{\mathbf{X},\overline{\mathbf{X}};q,q';k,\infty;[s,t]}, \tag*{(66)} R X − R X 2 p -var ∞ ; k , ∞ ; [ s , t ] ≲ Θ 1 ∣ δ ( X − X ) ∣ p -var ; [ s , t ] + ∣ δ X ∣ p -var ; [ s , t ] ∣ H ( Y ) , H ( Y ) ′ ; H ( Y ) , H ( Y ) ′ ∣ X , X ; q , q ′ ; k , ∞ ; [ s , t ] , (66)
where
H ( Y ) ′ : = D H ( Y ) Y ′ and H ‾ ( Y ‾ ) ′ : = D H ‾ ( Y ‾ ) Y ‾ ′ . H(Y)':=DH(Y)Y'\quad\text{and}\quad\overline{H}(\overline{Y})':=D\overline{H}(\overline{Y})\overline{Y}'. H ( Y ) ′ := D H ( Y ) Y ′ and H ( Y ) ′ := D H ( Y ) Y ′ .
Thus, in order to prove (65) , it suffices to estimate ∣ H ( Y ) , H ( Y ) ′ ; H ‾ ( Y ‾ ) , H ‾ ( Y ‾ ) ′ ∣ X , X ‾ ; q , q ′ ; k , ∞ \lvert H(Y),H(Y)';\overline{H}(\overline{Y}),\overline{H}(\overline{Y})'\rvert_{\mathbf{X},\overline{\mathbf{X}};q,q';k,\infty} ∣ H ( Y ) , H ( Y ) ′ ; H ( Y ) , H ( Y ) ′ ∣ X , X ; q , q ′ ; k , ∞ , which will be split into three steps.
Step 1. In this step, we estimate P H ( Y ) − P H ‾ ( Y ‾ ) P^{H(Y)}-P^{\overline{H}(\overline{Y})} P H ( Y ) − P H ( Y ) . We first provide some upper bounds for both the uniform norm and the ∥ ∥ ⋅ ∥ q -var ∥ ∞ ; k , ∞ \lVert\lVert\cdot\rVert_{q\text{-var}}\rVert_{\infty;k,\infty} ∥∥ ⋅ ∥ q -var ∥ ∞ ; k , ∞ norm of Y − Y ‾ Y-\overline{Y} Y − Y , Y Y Y , and Y ‾ \overline{Y} Y . By Lemma 3.13 we have
∥ Y − Y ‾ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ δ ( Y − Y ‾ ) ∥ q -var ∞ ; k , ∞ ; [ s , t ] ≲ Θ 1 ∥ Y t − Y ‾ t ∥ ∞ + ∣ Y , Y ′ ; Y ‾ , Y ‾ ′ ∣ X , X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] + ∣ δ ( X − X ‾ ) ∣ p -var ; [ s , t ] . (67) \|Y-\overline{Y}\|_{L^\infty([s,t]\times\Omega)}+\|\delta(Y-\overline{Y})\|_{q\text{-var}\,\infty;k,\infty;[s,t]}
\lesssim_{\Theta_1}\|Y_t-\overline{Y}_t\|_{\infty}+\lvert Y,Y';\overline{Y},\overline{Y}'\rvert_{\mathbf{X},\overline{\mathbf{X}};q,q';k,\infty;[s,t]}+|\delta(\mathbf{X}-\overline{\mathbf{X}})|_{p\text{-var};[s,t]}. \tag*{(67)} ∥ Y − Y ∥ L ∞ ([ s , t ] × Ω ) + ∥ δ ( Y − Y ) ∥ q -var ∞ ; k , ∞ ; [ s , t ] ≲ Θ 1 ∥ Y t − Y t ∥ ∞ + ∣ Y , Y ′ ; Y , Y ′ ∣ X , X ; q , q ′ ; k , ∞ ; [ s , t ] + ∣ δ ( X − X ) ∣ p -var ; [ s , t ] . (67)
Moreover, since ∥ Y ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] \|Y\|_{L^\infty([s,t]\times\Omega)}+\|\lVert\delta Y\rVert_{q\text{-var}}\|_{\infty;k,\infty;[s,t]} ∥ Y ∥ L ∞ ([ s , t ] × Ω ) + ∥ ∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] is dominated by ∥ Y t ∥ ∞ + ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] \|Y_t\|_{\infty}+\lvert(Y,Y')\rvert_{\mathbf{X};q,q';k,\infty;[s,t]} ∥ Y t ∥ ∞ + ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] by upper bound given by (64) , we have
∥ Y ∥ L ∞ ( [ s , t ] × Ω ) + ∥ Y ‾ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ Y ‾ ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ Θ 1 , C ∥ Y t ∥ ∞ + ∥ Y ‾ t ∥ ∞ + ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] + ∣ ( Y ‾ , Y ‾ ′ ) ∣ X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] ≲ Θ 1 , C 1. (68) \|Y\|_{L^\infty([s,t]\times\Omega)}+\|\overline{Y}\|_{L^\infty([s,t]\times\Omega)}+\|\lVert\delta Y\rVert_{q\text{-var}}\|_{\infty;k,\infty;[s,t]}+\|\lVert\delta\overline{Y}\rVert_{q\text{-var}}\|_{\infty;k,\infty;[s,t]}
\lesssim_{\Theta_1,C} \|Y_t\|_{\infty}+\|\overline{Y}_t\|_{\infty}+\lvert(Y,Y')\rvert_{\mathbf{X};q,q';k,\infty;[s,t]}+\lvert(\overline{Y},\overline{Y}')\rvert_{\overline{\mathbf{X}};q,q';k,\infty;[s,t]}\lesssim_{\Theta_1,C}1. \tag*{(68)} ∥ Y ∥ L ∞ ([ s , t ] × Ω ) + ∥ Y ∥ L ∞ ([ s , t ] × Ω ) + ∥ ∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ Θ 1 , C ∥ Y t ∥ ∞ + ∥ Y t ∥ ∞ + ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] + ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] ≲ Θ 1 , C 1. (68)
Now we estimate P H ( Y ) − P H ( Y ˉ ) P^{H}(Y)-P^{H}(\bar{Y}) P H ( Y ) − P H ( Y ˉ ) and P H ( Y ˉ ) − P H ˉ ( Y ˉ ) P^{H}(\bar{Y})-P^{\bar{H}}(\bar{Y}) P H ( Y ˉ ) − P H ˉ ( Y ˉ ) , respectively. By Proposition (20) (noting γ ≤ 3 \gamma\leq3 γ ≤ 3 ) and the inequality ∥ ∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ ≤ 1 + ∥ ∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ γ − 1 \lVert\lVert\delta Y\rVert_{q\text{-var}}\rVert_{\infty;k,\infty}\leq1+\lVert\lVert\delta Y\rVert_{q\text{-var}}\rVert_{\infty;k,\infty}^{\gamma-1} ∥∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ ≤ 1 + ∥∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ γ − 1 , noting 2 q ′ ≤ max { γ − 1 q , 2 q } \frac{2}{q'}\leq\max\left\{\frac{\gamma-1}{q},\frac{2}{q}\right\} q ′ 2 ≤ max { q γ − 1 , q 2 } ,
∥ ∥ P H ( Y ) − P H ( Y ˉ ) ∥ ∥ q ′ ′ 2 -var ; ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 1 ( 1 + ∥ ∥ δ Y ∥ ∥ q -var ; ∞ ; k , ∞ ; [ s , t ] γ − 1 + ∥ ∥ δ Y ˉ ∥ ∥ q -var ; ∞ ; k , ∞ ; [ s , t ] γ − 1 + ∥ ∥ P Y ˉ ∥ ∥ q ′ ′ 2 -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ M Y ˉ ∥ B M O ; [ s , t ] + ∥ ∥ δ Y ˉ ∥ ∥ q -var ; ∞ ; k , ∞ ; [ s , t ] γ − 2 ∥ M Y ˉ ∥ B M O ; [ s , t ] ) ( ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y − Y ˉ ) ∥ ∥ q -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ P Y − P Y ˉ ∥ ∥ q ′ ′ 2 -var ; [ s , t ] ; ∞ ; k , ∞ + ∥ δ ( M Y − M Y ˉ ) ∥ B M O ; [ s , t ] ) . (69) \begin{aligned}
\lVert\lVert P^{H}(Y)-P^{H}(\bar{Y})\rVert\rVert_{\frac{q''}{2}\text{-var};\infty;k,\infty;[s,t]}
&\lesssim_{\gamma,\Theta_1}
\left(1+\lVert\lVert\delta Y\rVert\rVert_{q\text{-var};\infty;k,\infty;[s,t]}^{\gamma-1}
+\lVert\lVert\delta\bar{Y}\rVert\rVert_{q\text{-var};\infty;k,\infty;[s,t]}^{\gamma-1}\right.\\
&\quad\left.+\lVert\lVert P^{\bar{Y}}\rVert\rVert_{\frac{q''}{2}\text{-var};\infty;k,\infty;[s,t]}
+\lVert M^{\bar{Y}}\rVert_{\mathrm{BMO};[s,t]}\right.\\
&\quad\left.+\lVert\lVert\delta\bar{Y}\rVert\rVert_{q\text{-var};\infty;k,\infty;[s,t]}^{\gamma-2}\lVert M^{\bar{Y}}\rVert_{\mathrm{BMO};[s,t]}\right)
\left(\lVert Y-\bar{Y}\rVert_{L^\infty([s,t]\times\Omega)}\right.\\
&\quad\left.+\lVert\lVert\delta(Y-\bar{Y})\rVert\rVert_{q\text{-var};\infty;k,\infty;[s,t]}
+\lVert\lVert P^{Y}-P^{\bar{Y}}\rVert\rVert_{\frac{q''}{2}\text{-var};[s,t];\infty;k,\infty}
+\lVert\delta(M^{Y}-M^{\bar{Y}})\rVert_{\mathrm{BMO};[s,t]}\right). \tag*{(69)}
\end{aligned} ∥∥ P H ( Y ) − P H ( Y ˉ )∥ ∥ 2 q ′′ -var ; ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 1 ( 1 + ∥∥ δ Y ∥ ∥ q -var ; ∞ ; k , ∞ ; [ s , t ] γ − 1 + ∥∥ δ Y ˉ ∥ ∥ q -var ; ∞ ; k , ∞ ; [ s , t ] γ − 1 + ∥∥ P Y ˉ ∥ ∥ 2 q ′′ -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ M Y ˉ ∥ BMO ; [ s , t ] + ∥∥ δ Y ˉ ∥ ∥ q -var ; ∞ ; k , ∞ ; [ s , t ] γ − 2 ∥ M Y ˉ ∥ BMO ; [ s , t ] ) ( ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( Y − Y ˉ )∥ ∥ q -var ; ∞ ; k , ∞ ; [ s , t ] + ∥∥ P Y − P Y ˉ ∥ ∥ 2 q ′′ -var ; [ s , t ] ; ∞ ; k , ∞ + ∥ δ ( M Y − M Y ˉ ) ∥ BMO ; [ s , t ] ) . (69)
Then, by (68) and (69) , it follows that
∥ ∥ P H ( Y ) − P H ( Y ˉ ) ∥ ∥ q ′ ′ 2 -var ; ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 1 , C ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y − Y ˉ ) ∥ ∥ q -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ P Y − P Y ˉ ∥ ∥ q ′ ′ 2 -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ δ ( M Y − M Y ˉ ) ∥ B M O ; [ s , t ] . (70) \begin{aligned}
\lVert\lVert P^{H}(Y)-P^{H}(\bar{Y})\rVert\rVert_{\frac{q''}{2}\text{-var};\infty;k,\infty;[s,t]}
&\lesssim_{\gamma,\Theta_1,C}\lVert Y-\bar{Y}\rVert_{L^\infty([s,t]\times\Omega)}\\
&\quad+\lVert\lVert\delta(Y-\bar{Y})\rVert\rVert_{q\text{-var};\infty;k,\infty;[s,t]}
+\lVert\lVert P^{Y}-P^{\bar{Y}}\rVert\rVert_{\frac{q''}{2}\text{-var};\infty;k,\infty;[s,t]}\\
&\quad+\lVert\delta(M^{Y}-M^{\bar{Y}})\rVert_{\mathrm{BMO};[s,t]}. \tag*{(70)}
\end{aligned} ∥∥ P H ( Y ) − P H ( Y ˉ )∥ ∥ 2 q ′′ -var ; ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 1 , C ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( Y − Y ˉ )∥ ∥ q -var ; ∞ ; k , ∞ ; [ s , t ] + ∥∥ P Y − P Y ˉ ∥ ∥ 2 q ′′ -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ δ ( M Y − M Y ˉ ) ∥ BMO ; [ s , t ] . (70)
Therefore, combining (67) and (70) , we get
LHS of ( 3.62 ) ≲ γ , Θ 1 , C ∥ Y t − Y ˉ t ∥ ∞ + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ δ ( X − X ˉ ) ∥ p -var ; [ s , t ] . (71) \operatorname{LHS\ of\ }(3.62)\lesssim_{\gamma,\Theta_1,C}\lVert Y_t-\bar{Y}_t\rVert_\infty+\lVert Y,Y';\bar{Y},\bar{Y}'\rVert_{X,\bar{X};q,q';k,\infty;[s,t]}+\lVert\delta(X-\bar{X})\rVert_{p\text{-var};[s,t]}. \tag*{(71)} LHS of ( 3.62 ) ≲ γ , Θ 1 , C ∥ Y t − Y ˉ t ∥ ∞ + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ δ ( X − X ˉ ) ∥ p -var ; [ s , t ] . (71)
Denote H ^ : = H − H ˉ \widehat{H}:=H-\bar{H} H := H − H ˉ and H ^ ( Y ˉ ) ′ : = D H ^ ( Y ˉ ) Y ˉ ′ \widehat{H}(\bar{Y})':=D\widehat{H}(\bar{Y})\bar{Y}' H ( Y ˉ ) ′ := D H ( Y ˉ ) Y ˉ ′ . Then, Lemma (24) (letting p ′ = q ′ p / q p'=q'p/q p ′ = q ′ p / q therein so that γ − 2 ≥ p / p ′ \gamma-2\geq p/p' γ − 2 ≥ p / p ′ ) yields, in view of (64) ,
∥ ∥ P H ^ ( Y ˉ ) ∥ ∥ q ′ ′ 2 -var ; ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 1 , C ∥ H ^ ∥ C b γ − 1 . (72) \lVert\lVert P^{\widehat{H}}(\bar{Y})\rVert\rVert_{\frac{q''}{2}\text{-var};\infty;k,\infty;[s,t]}
\lesssim_{\gamma,\Theta_1,C}\lVert\widehat{H}\rVert_{C_b^{\gamma-1}}. \tag*{(72)} ∥∥ P H ( Y ˉ )∥ ∥ 2 q ′′ -var ; ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 1 , C ∥ H ∥ C b γ − 1 . (72)
Thus, by (71) and (72) we get, noting P H ( Y ) − P H ˉ ( Y ˉ ) = P H ( Y ) − P H ( Y ˉ ) + P H ^ ( Y ˉ ) P^{H}(Y)-P^{\bar{H}}(\bar{Y})=P^{H}(Y)-P^{H}(\bar{Y})+P^{\widehat{H}}(\bar{Y}) P H ( Y ) − P H ˉ ( Y ˉ ) = P H ( Y ) − P H ( Y ˉ ) + P H ( Y ˉ ) ,
∥ ∥ P H ( Y ) − P H ( Y ˉ ) ∥ ∥ q ′ ′ 2 -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ P H ^ ( Y ˉ ) ∥ ∥ q ′ ′ 2 -var ; ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 1 , C ∥ Y t − Y ˉ t ∥ ∞ + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ δ ( X − X ˉ ) ∥ p -var ; [ s , t ] + ∥ H ^ ∥ C b γ − 1 . (73) \begin{aligned}
\lVert\lVert P^{H}(Y)-P^{H}(\bar{Y})\rVert\rVert_{\frac{q''}{2}\text{-var};\infty;k,\infty;[s,t]}
+\lVert\lVert P^{\widehat{H}}(\bar{Y})\rVert\rVert_{\frac{q''}{2}\text{-var};\infty;k,\infty;[s,t]}
&\lesssim_{\gamma,\Theta_1,C}\lVert Y_t-\bar{Y}_t\rVert_\infty\\
&\quad+\lVert Y,Y';\bar{Y},\bar{Y}'\rVert_{X,\bar{X};q,q';k,\infty;[s,t]}\\
&\quad+\lVert\delta(X-\bar{X})\rVert_{p\text{-var};[s,t]}+\lVert\widehat{H}\rVert_{C_b^{\gamma-1}}. \tag*{(73)}
\end{aligned} ∥∥ P H ( Y ) − P H ( Y ˉ )∥ ∥ 2 q ′′ -var ; ∞ ; k , ∞ ; [ s , t ] + ∥∥ P H ( Y ˉ )∥ ∥ 2 q ′′ -var ; ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 1 , C ∥ Y t − Y ˉ t ∥ ∞ + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ δ ( X − X ˉ ) ∥ p -var ; [ s , t ] + ∥ H ∥ C b γ − 1 . (73)
Step 2. In this step, we estimate M H ( Y ) − M H ˉ ( Y ˉ ) M^{H}(Y)-M^{\bar{H}}(\bar{Y}) M H ( Y ) − M H ˉ ( Y ˉ ) . Note that for all ( u , v ) ∈ Δ [ s , t ] (u,v)\in\Delta_{[s,t]} ( u , v ) ∈ Δ [ s , t ] ,
δ ( M H ( Y ) − M H ˉ ( Y ˉ ) ) u , v = ∫ u v D H ( Y r ) d ( M r Y − M r Y ˉ ) + ∫ u v [ D H ( Y r ) − D H ( Y ˉ r ) + D H ^ ( Y ˉ r ) ] d M r Y ˉ . \delta(M^{H}(Y)-M^{\bar{H}}(\bar{Y}))_{u,v}
=\int_u^v DH(Y_r)\,\mathrm{d}(M_r^Y-M_r^{\bar{Y}})
+\int_u^v[DH(Y_r)-DH(\bar{Y}_r)+D\widehat{H}(\bar{Y}_r)]\,\mathrm{d}M_r^{\bar{Y}}. δ ( M H ( Y ) − M H ˉ ( Y ˉ ) ) u , v = ∫ u v D H ( Y r ) d ( M r Y − M r Y ˉ ) + ∫ u v [ D H ( Y r ) − D H ( Y ˉ r ) + D H ( Y ˉ r )] d M r Y ˉ .
Then, it follows that
∥ δ ( M H ( Y ) − M H ˉ ( Y ˉ ) ) 1 2 , u , v ∣ F u ∥ 2 = ∥ δ ( ∫ s ⋅ D H ( Y r ) d ( M r Y − M r Y ˉ ) + ∫ 0 ⋅ [ D H ( Y r ) − D H ( Y ˉ r ) + D H ^ ( Y ˉ r ) ] d M r Y ˉ ) 1 2 , u , v | F u ∥ 2 ≤ ∥ D H ∥ ∞ ∥ δ ⟨ M Y − M Y ˉ ⟩ 1 2 , u , v ∣ F u ∥ 2 + ( ∥ D 2 H ∥ ∞ ∥ Y − Y ˉ ∥ ∞ + ∥ D H ^ ∥ ∞ ) ∥ δ ⟨ M Y ˉ ⟩ 1 2 , u , v ∣ F u ∥ 2 . (74) \begin{aligned}
\lVert\delta(M^{H}(Y)-M^{\bar{H}}(\bar{Y}))_{\frac12,u,v}\mid\mathcal{F}_u\rVert_2
&=\left\lVert\delta\left(\int_s^\cdot DH(Y_r)\,\mathrm{d}(M_r^Y-M_r^{\bar{Y}})
+\int_0^\cdot[DH(Y_r)-DH(\bar{Y}_r)+D\widehat{H}(\bar{Y}_r)]\,\mathrm{d}M_r^{\bar{Y}}\right)_{\frac12,u,v}\middle|\mathcal{F}_u\right\rVert_2\\
&\leq\lVert DH\rVert_\infty\lVert\delta\langle M^Y-M^{\bar{Y}}\rangle_{\frac12,u,v}\mid\mathcal{F}_u\rVert_2\\
&\quad+\left(\lVert D^2H\rVert_\infty\lVert Y-\bar{Y}\rVert_\infty+\lVert D\widehat{H}\rVert_\infty\right)
\lVert\delta\langle M^{\bar{Y}}\rangle_{\frac12,u,v}\mid\mathcal{F}_u\rVert_2. \tag*{(74)}
\end{aligned} ∥ δ ( M H ( Y ) − M H ˉ ( Y ˉ ) ) 2 1 , u , v ∣ F u ∥ 2 = δ ( ∫ s ⋅ D H ( Y r ) d ( M r Y − M r Y ˉ ) + ∫ 0 ⋅ [ D H ( Y r ) − D H ( Y ˉ r ) + D H ( Y ˉ r )] d M r Y ˉ ) 2 1 , u , v F u 2 ≤ ∥ D H ∥ ∞ ∥ δ ⟨ M Y − M Y ˉ ⟩ 2 1 , u , v ∣ F u ∥ 2 + ( ∥ D 2 H ∥ ∞ ∥ Y − Y ˉ ∥ ∞ + ∥ D H ∥ ∞ ) ∥ δ ⟨ M Y ˉ ⟩ 2 1 , u , v ∣ F u ∥ 2 . (74)
Hence, by the boundedness of D H DH D H and D 2 H D^2H D 2 H , (64) , and (67) , the estimate (74) yields
∥ δ ( M H ( Y ) − M H ˉ ( Y ˉ ) ) ∥ B M O ; [ s , t ] ≲ Θ 1 ∥ δ ( M Y − M Y ˉ ) ∥ B M O ; [ s , t ] + ( ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ D H ^ ∥ ∞ ) ∥ δ M Y ˉ ∥ B M O ; [ s , t ] ≲ Θ 1 , C ∥ δ ( M Y − M Y ˉ ) ∥ B M O ; [ s , t ] + ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ D H ^ ∥ ∞ ≲ Θ 1 ∥ Y t − Y ˉ t ∥ ∞ + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ δ ( X − X ˉ ) ∥ p -var ; [ s , t ] + ∥ D H ^ ∥ ∞ . (75) \begin{aligned}
\lVert\delta(M^{H}(Y)-M^{\bar{H}}(\bar{Y}))\rVert_{\mathrm{BMO};[s,t]}
&\lesssim_{\Theta_1}\lVert\delta(M^Y-M^{\bar{Y}})\rVert_{\mathrm{BMO};[s,t]}
+\left(\lVert Y-\bar{Y}\rVert_{L^\infty([s,t]\times\Omega)}+\lVert D\widehat{H}\rVert_\infty\right)
\lVert\delta M^{\bar{Y}}\rVert_{\mathrm{BMO};[s,t]}\\
&\lesssim_{\Theta_1,C}\lVert\delta(M^Y-M^{\bar{Y}})\rVert_{\mathrm{BMO};[s,t]}
+\lVert Y-\bar{Y}\rVert_{L^\infty([s,t]\times\Omega)}+\lVert D\widehat{H}\rVert_\infty\\
&\lesssim_{\Theta_1}\lVert Y_t-\bar{Y}_t\rVert_\infty
+\lVert Y,Y';\bar{Y},\bar{Y}'\rVert_{X,\bar{X};q,q';k,\infty;[s,t]}\\
&\quad+\lVert\delta(X-\bar{X})\rVert_{p\text{-var};[s,t]}+\lVert D\widehat{H}\rVert_\infty. \tag*{(75)}
\end{aligned} ∥ δ ( M H ( Y ) − M H ˉ ( Y ˉ )) ∥ BMO ; [ s , t ] ≲ Θ 1 ∥ δ ( M Y − M Y ˉ ) ∥ BMO ; [ s , t ] + ( ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) + ∥ D H ∥ ∞ ) ∥ δ M Y ˉ ∥ BMO ; [ s , t ] ≲ Θ 1 , C ∥ δ ( M Y − M Y ˉ ) ∥ BMO ; [ s , t ] + ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) + ∥ D H ∥ ∞ ≲ Θ 1 ∥ Y t − Y ˉ t ∥ ∞ + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ δ ( X − X ˉ ) ∥ p -var ; [ s , t ] + ∥ D H ∥ ∞ . (75)
Step 3. In this step, we estimate H ( Y ) ′ − H ˉ ( Y ˉ ) ′ H(Y)' - \bar{H}(\bar{Y})' H ( Y ) ′ − H ˉ ( Y ˉ ) ′ . By the triangular inequality, we have
∥ H ( Y ) ′ − H ˉ ( Y ˉ ) ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( H ( Y ) ′ − H ˉ ( Y ˉ ) ′ ) ∥ q ′ -var ; ∞ ; k , ∞ ; [ s , t ] ≤ ∥ ( D H ( Y ) − D H ( Y ˉ ) ) Y ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ D H ^ ( Y ˉ ) Y ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ D H ˉ ( Y ˉ ) ( Y ′ − Y ˉ ′ ) ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( ( D H ( Y ) − D H ( Y ˉ ) ) Y ′ ) ∥ q ′ -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ ( D H ^ ( Y ˉ ) Y ′ ) ∥ q ′ -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ ( D H ˉ ( Y ˉ ) ( Y ′ − Y ˉ ′ ) ) ∥ q ′ -var ; ∞ ; k , ∞ ; [ s , t ] . (76) \begin{aligned}
\|H(Y)' - \bar{H}(\bar{Y})'\|_{L^\infty([s,t]\times\Omega)} &+ \|\|\delta(H(Y)' - \bar{H}(\bar{Y})')\|_{q'\text{-var};\infty;k,\infty;[s,t]} \\
&\leq\|(DH(Y)-DH(\bar{Y}))Y'\|_{L^\infty([s,t]\times\Omega)}
+\|D\hat{H}(\bar{Y})Y'\|_{L^\infty([s,t]\times\Omega)} \\
&\quad+\|D\bar{H}(\bar{Y})(Y'-\bar{Y}')\|_{L^\infty([s,t]\times\Omega)} \\
&\quad+\|\|\delta((DH(Y)-DH(\bar{Y}))Y')\|_{q'\text{-var};\infty;k,\infty;[s,t]} \\
&\quad+\|\|\delta(D\hat{H}(\bar{Y})Y')\|_{q'\text{-var};\infty;k,\infty;[s,t]} \\
&\quad+\|\|\delta(D\bar{H}(\bar{Y})(Y'-\bar{Y}'))\|_{q'\text{-var};\infty;k,\infty;[s,t]}. \tag*{(76)}
\end{aligned} ∥ H ( Y ) ′ − H ˉ ( Y ˉ ) ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( H ( Y ) ′ − H ˉ ( Y ˉ ) ′ ) ∥ q ′ -var ; ∞ ; k , ∞ ; [ s , t ] ≤ ∥ ( D H ( Y ) − D H ( Y ˉ )) Y ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ D H ^ ( Y ˉ ) Y ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ D H ˉ ( Y ˉ ) ( Y ′ − Y ˉ ′ ) ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ (( D H ( Y ) − D H ( Y ˉ )) Y ′ ) ∥ q ′ -var ; ∞ ; k , ∞ ; [ s , t ] + ∥∥ δ ( D H ^ ( Y ˉ ) Y ′ ) ∥ q ′ -var ; ∞ ; k , ∞ ; [ s , t ] + ∥∥ δ ( D H ˉ ( Y ˉ ) ( Y ′ − Y ˉ ′ )) ∥ q ′ -var ; ∞ ; k , ∞ ; [ s , t ] . (76)
For the first three terms on the RHS of (76) , by the boundedness of ∥ D 2 H ∥ ∞ + ∥ D 2 H ˉ ∥ ∞ \|D^2H\|_\infty+\|D^2\bar{H}\|_\infty ∥ D 2 H ∥ ∞ + ∥ D 2 H ˉ ∥ ∞ ,
∥ ( D H ( Y ) − D H ( Y ˉ ) ) Y ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ D H ^ ( Y ˉ ) Y ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ D H ˉ ( Y ˉ ) ( Y ′ − Y ˉ ′ ) ∥ L ∞ ( [ s , t ] × Ω ) ≤ ∥ D 2 H ∥ ∞ ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) ∥ Y ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ D H ^ ∥ ∞ ∥ Y ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ D H ˉ ∥ ∞ ∥ ( Y ′ − Y ˉ ′ ) ∥ L ∞ ( [ s , t ] × Ω ) . (77) \begin{aligned}
&\|(DH(Y)-DH(\bar{Y}))Y'\|_{L^\infty([s,t]\times\Omega)}
+\|D\hat{H}(\bar{Y})Y'\|_{L^\infty([s,t]\times\Omega)}
+\|D\bar{H}(\bar{Y})(Y'-\bar{Y}')\|_{L^\infty([s,t]\times\Omega)} \\
&\leq\|D^2H\|_\infty\|Y-\bar{Y}\|_{L^\infty([s,t]\times\Omega)}
\|Y'\|_{L^\infty([s,t]\times\Omega)}
+\|D\hat{H}\|_\infty\|Y'\|_{L^\infty([s,t]\times\Omega)} \\
&\quad+\|D\bar{H}\|_\infty\|(Y'-\bar{Y}')\|_{L^\infty([s,t]\times\Omega)}. \tag*{(77)}
\end{aligned} ∥ ( D H ( Y ) − D H ( Y ˉ )) Y ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ D H ^ ( Y ˉ ) Y ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ D H ˉ ( Y ˉ ) ( Y ′ − Y ˉ ′ ) ∥ L ∞ ([ s , t ] × Ω ) ≤ ∥ D 2 H ∥ ∞ ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) ∥ Y ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ D H ^ ∥ ∞ ∥ Y ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ D H ˉ ∥ ∞ ∥ ( Y ′ − Y ˉ ′ ) ∥ L ∞ ([ s , t ] × Ω ) . (77)
For the other three terms on the RHS of (76) , by the inequality ∣ δ ( a ⋅ b ) u , v ∣ ≤ ∣ δ a u , v ∣ ⋅ ∣ b u ∣ + ∣ δ b u , v ∣ ⋅ ∣ a v ∣ |\delta(a\cdot b)_{u,v}|\leq|\delta a_{u,v}|\cdot|b_u|+|\delta b_{u,v}|\cdot|a_v| ∣ δ ( a ⋅ b ) u , v ∣ ≤ ∣ δ a u , v ∣ ⋅ ∣ b u ∣ + ∣ δ b u , v ∣ ⋅ ∣ a v ∣ and Lemma (76) , we get for all ( u , v ) ∈ Δ [ s , t ] (u,v)\in\Delta_{[s,t]} ( u , v ) ∈ Δ [ s , t ] ,
∥ ∥ δ ( ( D H ( Y ) − D H ( Y ˉ ) ) Y ′ ) ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ + ∥ ∥ δ ( D H ^ ( Y ˉ ) Y ′ ) ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ + ∥ ∥ δ ( D H ˉ ( Y ˉ ) ( Y ′ − Y ˉ ′ ) ) ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ≲ ∥ D H ∥ C b 1 [ ( ∥ ∥ δ Y ∥ q -var ; [ u , v ] ∥ F u ∥ k , ∞ γ − 2 + ∥ ∥ δ Y ˉ ∥ q -var ; [ u , v ] ∥ F u ∥ k , ∞ γ − 2 ) ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y − Y ˉ ) ∥ q -var ; [ u , v ] ∥ F u ∥ k , ∞ ∥ Y ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) ∥ ∥ δ Y ′ ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ] + ∥ D H ^ ∥ C b 1 ( ∥ ∥ δ Y ˉ ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ∥ Y ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ Y ′ ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ∥ Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) ) + ∥ D H ˉ ∥ C b 1 ( ∥ ∥ δ Y ˉ ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ∥ Y ′ − Y ˉ ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y ′ − Y ˉ ′ ) ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ∥ Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) ) . (78) \begin{aligned}
&\|\|\delta((DH(Y)-DH(\bar{Y}))Y')\|_{q'\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty}
+\|\|\delta(D\hat{H}(\bar{Y})Y')\|_{q'\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty} \\
&\quad+\|\|\delta(D\bar{H}(\bar{Y})(Y'-\bar{Y}'))\|_{q'\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty} \\
&\lesssim\|DH\|_{C_b^1}
\left[
\left(\|\|\delta Y\|_{q\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty}^{\gamma-2}
+\|\|\delta\bar{Y}\|_{q\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty}^{\gamma-2}\right)
\|Y-\bar{Y}\|_{L^\infty([s,t]\times\Omega)}\right. \\
&\qquad\left.+\|\|\delta(Y-\bar{Y})\|_{q\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty}\|Y'\|_{L^\infty([s,t]\times\Omega)}
+\|Y-\bar{Y}\|_{L^\infty([s,t]\times\Omega)}
\|\|\delta Y'\|_{q'\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty}
\right] \\
&\quad+\|D\hat{H}\|_{C_b^1}
\left(\|\|\delta\bar{Y}\|_{q'\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty}\|Y'\|_{L^\infty([s,t]\times\Omega)}
+\|\|\delta Y'\|_{q'\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty}\|\bar{Y}\|_{L^\infty([s,t]\times\Omega)}\right) \\
&\quad+\|D\bar{H}\|_{C_b^1}
\left(\|\|\delta\bar{Y}\|_{q'\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty}\|Y'-\bar{Y}'\|_{L^\infty([s,t]\times\Omega)}
+\|\|\delta(Y'-\bar{Y}')\|_{q'\text{-var};[u,v]}\|\mathcal{F}_u\|_{k,\infty}\|\bar{Y}\|_{L^\infty([s,t]\times\Omega)}\right). \tag*{(78)}
\end{aligned} ∥∥ δ (( D H ( Y ) − D H ( Y ˉ )) Y ′ ) ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ + ∥∥ δ ( D H ^ ( Y ˉ ) Y ′ ) ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ + ∥∥ δ ( D H ˉ ( Y ˉ ) ( Y ′ − Y ˉ ′ )) ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ≲ ∥ D H ∥ C b 1 [ ( ∥∥ δ Y ∥ q -var ; [ u , v ] ∥ F u ∥ k , ∞ γ − 2 + ∥∥ δ Y ˉ ∥ q -var ; [ u , v ] ∥ F u ∥ k , ∞ γ − 2 ) ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( Y − Y ˉ ) ∥ q -var ; [ u , v ] ∥ F u ∥ k , ∞ ∥ Y ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) ∥∥ δ Y ′ ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ] + ∥ D H ^ ∥ C b 1 ( ∥∥ δ Y ˉ ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ∥ Y ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ Y ′ ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ∥ Y ˉ ∥ L ∞ ([ s , t ] × Ω ) ) + ∥ D H ˉ ∥ C b 1 ( ∥∥ δ Y ˉ ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ∥ Y ′ − Y ˉ ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( Y ′ − Y ˉ ′ ) ∥ q ′ -var ; [ u , v ] ∥ F u ∥ k , ∞ ∥ Y ˉ ∥ L ∞ ([ s , t ] × Ω ) ) . (78)
Then, combining (76) , (77) , and (78) , and applying Young’s inequality ∣ a ∣ γ − 2 + ∣ a ∣ ≤ 2 + 2 ∣ a ∣ ( γ − 2 ) ∨ 1 |a|^{\gamma-2}+|a|\leq2+2|a|^{(\gamma-2)\vee1} ∣ a ∣ γ − 2 + ∣ a ∣ ≤ 2 + 2∣ a ∣ ( γ − 2 ) ∨ 1 , noting that γ − 2 q ≥ 1 q ′ \frac{\gamma-2}{q}\geq\frac{1}{q'} q γ − 2 ≥ q ′ 1 and q ≤ q ′ q\leq q' q ≤ q ′ , we have
∥ H ( Y ) ′ − H ˉ ( Y ˉ ) ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( H ( Y ) ′ − H ˉ ( Y ˉ ) ′ ) ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ Θ 1 ( 1 + ∥ Y t ∥ ∞ + ∥ Y ˉ t ∥ ∞ + ∥ ∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] ( γ − 2 ) ∨ 1 + ∥ ∥ δ Y ˉ ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] ( γ − 2 ) ∨ 1 ) × ( 1 + ∥ Y ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ Y ′ ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] ) × ( ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y − Y ˉ ) ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ D H ^ ∥ C b 1 + ∥ Y ′ − Y ˉ ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y ′ − Y ˉ ′ ) ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] ) ≲ γ , Θ 1 , C ( ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y − Y ˉ ) ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ D H ^ ∥ C b 1 + ∥ Y ′ − Y ˉ ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y ′ − Y ˉ ′ ) ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] ) ≲ Θ 1 ∥ Y t − Y ˉ t ∥ ∞ + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; q , q ′ ; k , ∞ ; [ s , t ] + ∣ δ ( X − X ˉ ) ∣ p -var ; [ s , t ] + ∥ D H ^ ∥ C b 1 . (79) \begin{aligned}
&\|H(Y)'-\bar{H}(\bar{Y})'\|_{L^\infty([s,t]\times\Omega)}
+\|\|\delta(H(Y)'-\bar{H}(\bar{Y})')\|_{q'\text{-var}}\|_{\infty;k,\infty;[s,t]} \\
&\lesssim_{\Theta_1}
\left(1+\|Y_t\|_\infty+\|\bar{Y}_t\|_\infty
+\|\|\delta Y\|_{q\text{-var}}\|_{\infty;k,\infty;[s,t]}^{(\gamma-2)\vee1}
+\|\|\delta\bar{Y}\|_{q\text{-var}}\|_{\infty;k,\infty;[s,t]}^{(\gamma-2)\vee1}\right) \\
&\quad\times\left(1+\|Y'\|_{L^\infty([s,t]\times\Omega)}
+\|\|\delta Y'\|_{q'\text{-var}}\|_{\infty;k,\infty;[s,t]}\right) \\
&\quad\times\left(\|Y-\bar{Y}\|_{L^\infty([s,t]\times\Omega)}
+\|\|\delta(Y-\bar{Y})\|_{q\text{-var}}\|_{\infty;k,\infty;[s,t]}
+\|D\hat{H}\|_{C_b^1}\right. \\
&\qquad\left.+\|Y'-\bar{Y}'\|_{L^\infty([s,t]\times\Omega)}
+\|\|\delta(Y'-\bar{Y}')\|_{q'\text{-var}}\|_{\infty;k,\infty;[s,t]}\right) \\
&\lesssim_{\gamma,\Theta_1,C}
\left(\|Y-\bar{Y}\|_{L^\infty([s,t]\times\Omega)}
+\|\|\delta(Y-\bar{Y})\|_{q\text{-var}}\|_{\infty;k,\infty;[s,t]}
+\|D\hat{H}\|_{C_b^1}\right. \\
&\qquad\left.+\|Y'-\bar{Y}'\|_{L^\infty([s,t]\times\Omega)}
+\|\|\delta(Y'-\bar{Y}')\|_{q'\text{-var}}\|_{\infty;k,\infty;[s,t]}\right) \\
&\lesssim_{\Theta_1}
\|Y_t-\bar{Y}_t\|_\infty+\|Y,Y';\bar{Y},\bar{Y}'\|_{X,\bar{X};q,q';k,\infty;[s,t]}
+|\delta(X-\bar{X})|_{p\text{-var};[s,t]}+\|D\hat{H}\|_{C_b^1}. \tag*{(79)}
\end{aligned} ∥ H ( Y ) ′ − H ˉ ( Y ˉ ) ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( H ( Y ) ′ − H ˉ ( Y ˉ ) ′ ) ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ Θ 1 ( 1 + ∥ Y t ∥ ∞ + ∥ Y ˉ t ∥ ∞ + ∥∥ δ Y ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] ( γ − 2 ) ∨ 1 + ∥∥ δ Y ˉ ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] ( γ − 2 ) ∨ 1 ) × ( 1 + ∥ Y ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ Y ′ ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] ) × ( ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( Y − Y ˉ ) ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ D H ^ ∥ C b 1 + ∥ Y ′ − Y ˉ ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( Y ′ − Y ˉ ′ ) ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] ) ≲ γ , Θ 1 , C ( ∥ Y − Y ˉ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( Y − Y ˉ ) ∥ q -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ D H ^ ∥ C b 1 + ∥ Y ′ − Y ˉ ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( Y ′ − Y ˉ ′ ) ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] ) ≲ Θ 1 ∥ Y t − Y ˉ t ∥ ∞ + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; q , q ′ ; k , ∞ ; [ s , t ] + ∣ δ ( X − X ˉ ) ∣ p -var ; [ s , t ] + ∥ D H ^ ∥ C b 1 . (79)
where the second inequality is due to (64) and (68) ; and the last is due to (67) .
Finally, combining (73) , (75) , and (79) , which were established in the above three steps, we obtain
∥ H ( Y ) , H ( Y ) ′ ; H ‾ ( Y ‾ ) , H ‾ ( Y ‾ ) ′ ∥ X , X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] ≲ γ , Θ 1 , C ∥ Y t − Y ‾ t ∥ ∞ + ∥ Y , Y ′ ; Y ‾ , Y ‾ ′ ∥ X , X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] + ∣ δ ( X − X ‾ ) ∣ p -var ; [ s , t ] + ∥ H ^ ∥ C b γ − 1 . (80) \left\|H(Y),H(Y)';\overline{H}(\overline{Y}),\overline{H}(\overline{Y})'\right\|_{X,\overline{X};q,q';k,\infty;[s,t]}
\lesssim_{\gamma,\Theta_1,C}\|Y_t-\overline{Y}_t\|_\infty+\left\|Y,Y';\overline{Y},\overline{Y}'\right\|_{X,\overline{X};q,q';k,\infty;[s,t]}+|\delta(X-\overline{X})|_{p\text{-var};[s,t]}+\|\widehat{H}\|_{C_b^{\gamma-1}}. \tag*{(80)} H ( Y ) , H ( Y ) ′ ; H ( Y ) , H ( Y ) ′ X , X ; q , q ′ ; k , ∞ ; [ s , t ] ≲ γ , Θ 1 , C ∥ Y t − Y t ∥ ∞ + Y , Y ′ ; Y , Y ′ X , X ; q , q ′ ; k , ∞ ; [ s , t ] + ∣ δ ( X − X ) ∣ p -var ; [ s , t ] + ∥ H ∥ C b γ − 1 . (80)
Hence, by (66) and (80) , ∥ ∥ R I X − R I X ‾ ∥ p / 2 -var ∥ ∞ ; k , ∞ ; [ s , t ] \left\|\left\|R^{I X}-R^{I\overline{X}}\right\|_{p/2\text{-var}}\right\|_{\infty;k,\infty;[s,t]} R I X − R I X p /2 -var ∞ ; k , ∞ ; [ s , t ] is bounded by the RHS of (65) .
To conclude the proof, we prove the estimate (65) . Denote ( H ( Y ) δ X ) u , v : = H ( Y u ) δ X u , v (H(Y)\delta X)_{u,v}:=H(Y_u)\delta X_{u,v} ( H ( Y ) δ X ) u , v := H ( Y u ) δ X u , v and ( H ‾ ( Y ‾ ) δ X ‾ ) u , v : = H ‾ ( Y ‾ u ) δ X ‾ u , v (\overline{H}(\overline{Y})\delta\overline{X})_{u,v}:=\overline{H}(\overline{Y}_u)\delta\overline{X}_{u,v} ( H ( Y ) δ X ) u , v := H ( Y u ) δ X u , v . Note that
∥ ∥ H ( Y ) δ X − H ‾ ( Y ‾ ) δ X ‾ ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≤ ∣ δ X ‾ ∣ p -var ; [ s , t ] ∥ H ( Y ) − H ‾ ( Y ‾ ) ∥ L ∞ ( [ s , t ] × Ω ) + ∣ δ ( X − X ‾ ) ∣ p -var ; [ s , t ] ∥ H ( Y ) ∥ L ∞ ( [ s , t ] × Ω ) ≲ Θ 1 ∣ δ X ‾ ∣ p -var ; [ s , t ] ( ∥ H ( Y t ) − H ‾ ( Y ‾ t ) ∥ ∞ + ∥ H ( Y ) , H ( Y ) ′ ; H ‾ ( Y ‾ ) , H ‾ ( Y ‾ ) ′ ∥ X , X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] ) + ∣ δ ( X − X ‾ ) ∣ p -var ; [ s , t ] ≲ γ , Θ 1 , C ∣ δ X ‾ ∣ p -var ; [ s , t ] ( ∥ Y t − Y ‾ t ∥ ∞ + ∥ Y , Y ′ ; Y ‾ , Y ‾ ′ ∥ X , X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ H ^ ∥ C b γ − 1 ) + ∣ δ ( X − X ‾ ) ∣ p -var ; [ s , t ] . (81) \begin{aligned}
\left\|\left\|H(Y)\delta X-\overline{H}(\overline{Y})\delta\overline{X}\right\|_{p\text{-var}}\right\|_{\infty;k,\infty;[s,t]}
&\leq|\delta\overline{X}|_{p\text{-var};[s,t]}\|H(Y)-\overline{H}(\overline{Y})\|_{L^\infty([s,t]\times\Omega)}\\
&\quad+|\delta(X-\overline{X})|_{p\text{-var};[s,t]}\|H(Y)\|_{L^\infty([s,t]\times\Omega)}\\
&\lesssim_{\Theta_1}|\delta\overline{X}|_{p\text{-var};[s,t]}
\left(\|H(Y_t)-\overline{H}(\overline{Y}_t)\|_\infty+
\left\|H(Y),H(Y)';\overline{H}(\overline{Y}),\overline{H}(\overline{Y})'\right\|_{X,\overline{X};q,q';k,\infty;[s,t]}\right)\\
&\quad+|\delta(X-\overline{X})|_{p\text{-var};[s,t]}\\
&\lesssim_{\gamma,\Theta_1,C}|\delta\overline{X}|_{p\text{-var};[s,t]}
\left(\|Y_t-\overline{Y}_t\|_\infty+
\left\|Y,Y';\overline{Y},\overline{Y}'\right\|_{X,\overline{X};q,q';k,\infty;[s,t]}+\|\widehat{H}\|_{C_b^{\gamma-1}}\right)\\
&\quad+|\delta(X-\overline{X})|_{p\text{-var};[s,t]}. \tag*{(81)}
\end{aligned} H ( Y ) δ X − H ( Y ) δ X p -var ∞ ; k , ∞ ; [ s , t ] ≤ ∣ δ X ∣ p -var ; [ s , t ] ∥ H ( Y ) − H ( Y ) ∥ L ∞ ([ s , t ] × Ω ) + ∣ δ ( X − X ) ∣ p -var ; [ s , t ] ∥ H ( Y ) ∥ L ∞ ([ s , t ] × Ω ) ≲ Θ 1 ∣ δ X ∣ p -var ; [ s , t ] ( ∥ H ( Y t ) − H ( Y t ) ∥ ∞ + H ( Y ) , H ( Y ) ′ ; H ( Y ) , H ( Y ) ′ X , X ; q , q ′ ; k , ∞ ; [ s , t ] ) + ∣ δ ( X − X ) ∣ p -var ; [ s , t ] ≲ γ , Θ 1 , C ∣ δ X ∣ p -var ; [ s , t ] ( ∥ Y t − Y t ∥ ∞ + Y , Y ′ ; Y , Y ′ X , X ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ H ∥ C b γ − 1 ) + ∣ δ ( X − X ) ∣ p -var ; [ s , t ] . (81)
where the second and the last inequalities follow from the fact that
∥ H ( Y ) − H ‾ ( Y ‾ ) ∥ L ∞ ( [ s , t ] × Ω ) ≤ ∥ H ( Y t ) − H ‾ ( Y ‾ t ) ∥ ∞ + ∥ H ( Y ) , H ( Y ) ′ ; H ‾ ( Y ‾ ) , H ‾ ( Y ‾ ) ′ ∥ X , X ‾ ; q , q ′ ; k , ∞ ; [ s , t ] \|H(Y)-\overline{H}(\overline{Y})\|_{L^\infty([s,t]\times\Omega)}
\leq\|H(Y_t)-\overline{H}(\overline{Y}_t)\|_\infty+
\left\|H(Y),H(Y)';\overline{H}(\overline{Y}),\overline{H}(\overline{Y})'\right\|_{X,\overline{X};q,q';k,\infty;[s,t]} ∥ H ( Y ) − H ( Y ) ∥ L ∞ ([ s , t ] × Ω ) ≤ ∥ H ( Y t ) − H ( Y t ) ∥ ∞ + H ( Y ) , H ( Y ) ′ ; H ( Y ) , H ( Y ) ′ X , X ; q , q ′ ; k , ∞ ; [ s , t ]
and (80) , respectively. Furthermore, noting δ ( I X − I X ‾ ) u , v = ( R I X − R I X ‾ ) u , v + H ( Y u ) δ X u , v − H ‾ ( Y ‾ u ) δ X ‾ u , v \delta(I^X-I^{\overline{X}})_{u,v}=(R^{I X}-R^{I\overline{X}})_{u,v}+H(Y_u)\delta X_{u,v}-\overline{H}(\overline{Y}_u)\delta\overline{X}_{u,v} δ ( I X − I X ) u , v = ( R I X − R I X ) u , v + H ( Y u ) δ X u , v − H ( Y u ) δ X u , v , we have
∥ ∥ δ ( I X − I X ‾ ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≤ ∥ ∥ ( R I X − R I X ‾ ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ H ( Y ) δ X − H ‾ ( Y ‾ ) δ X ‾ ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] . \begin{aligned}
\left\|\left\|\delta(I^X-I^{\overline{X}})\right\|_{p\text{-var}}\right\|_{\infty;k,\infty;[s,t]}
\leq{}&
\left\|\left\|(R^{I X}-R^{I\overline{X}})\right\|_{p\text{-var}}\right\|_{\infty;k,\infty;[s,t]}\\
&+\left\|\left\|H(Y)\delta X-\overline{H}(\overline{Y})\delta\overline{X}\right\|_{p\text{-var}}\right\|_{\infty;k,\infty;[s,t]}.
\end{aligned} δ ( I X − I X ) p -var ∞ ; k , ∞ ; [ s , t ] ≤ ( R I X − R I X ) p -var ∞ ; k , ∞ ; [ s , t ] + H ( Y ) δ X − H ( Y ) δ X p -var ∞ ; k , ∞ ; [ s , t ] .
Therefore, since ∥ ∥ R I X − R I X ‾ ∥ p / 2 -var ∥ ∞ ; k , ∞ ; [ s , t ] \left\|\left\|R^{I X}-R^{I\overline{X}}\right\|_{p/2\text{-var}}\right\|_{\infty;k,\infty;[s,t]} R I X − R I X p /2 -var ∞ ; k , ∞ ; [ s , t ] is bounded by the RHS of (65) , by (81) , we obtain the desired (65) . □
Rough BSDE
In this section, we establish the well-posedness of the rough BSDE (1) . In Section 4.1, we establish a comparison principle for rough BSDEs. In Section 4.2, we first derive an RSM a priori estimate for the solution, by clarifying how the RSM seminorm ∥ ⋅ ∥ X ; q , q ′ ; k , ∞ \|\cdot\|_{X;q,q';k,\infty} ∥ ⋅ ∥ X ; q , q ′ ; k , ∞ , defined in (48) , depends on the BMO norm of M Y M^Y M Y and the essential supremum of Y Y Y . These key estimates are subsequently used in Section 4.3 to study the existence and stability of solutions.
Throughout this section, p ∈ ( 2 , 3 ) p\in(2,3) p ∈ ( 2 , 3 ) is a fixed number, p ′ p' p ′ , q q q , q ′ q' q ′ are numbers such that p ≤ p ′ p\leq p' p ≤ p ′ , p ≤ q p\leq q p ≤ q , and q ≤ q ′ q\leq q' q ≤ q ′ .
We first introduce the notion of solutions to rough BSDE (1) . In particular, no integrability condition on the solution is required in the definition.
Definition 4.1. Let X ∈ C p -var ( [ 0 , T ] , R d ) X\in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^d) X ∈ C p -var ([ 0 , T ] , R d ) and H ∈ C 2 ( R , R d ) H\in C^2(\mathbb{R},\mathbb{R}^d) H ∈ C 2 ( R , R d ) . For a continuous adapted process Y Y Y and a progressively measurable process Z Z Z , we call ( Y , Z ) (Y,Z) ( Y , Z ) a solution to Eq. (1) with the parameter ( ξ , f , H , X ) (\xi,f,H,X) ( ξ , f , H , X ) , if ( Y , − H ( Y ) ) ∈ R S M X ( p , p ′ ) -var ( [ 0 , T ] , R ) (Y,-H(Y))\in\mathrm{RSM}_{X}^{(p,p')\text{-var}}([0,T],\mathbb{R}) ( Y , − H ( Y )) ∈ RSM X ( p , p ′ ) -var ([ 0 , T ] , R ) with some ( p , p ′ ) (p,p') ( p , p ′ ) satisfying ( p , p ′ ) (p,p') ( p , p ′ ) , ∫ 0 T ∣ Z r ∣ 2 d r < ∞ \int_0^T|Z_r|^2\,dr<\infty ∫ 0 T ∣ Z r ∣ 2 d r < ∞ a.s., and the equation (1) holds a.s.
We note that M t Y = ∫ 0 t Z r d W r M_t^Y = \int_0^t Z_r dW_r M t Y = ∫ 0 t Z r d W r . This relation is in force throughout this section, as follows from Remark 2.7.
We also note that if ( Y , Z ) (Y,Z) ( Y , Z ) is a solution to Eq. (1) with some p ′ ≥ p p' \ge p p ′ ≥ p , the regularity of the solution can be improved to ( Y , − H ( Y ) ) ∈ RSM X ( p , p ′ ) -var (Y,-H(Y)) \in\operatorname{RSM}_{X}^{(p,p')\text{-var}} ( Y , − H ( Y )) ∈ RSM X ( p , p ′ ) -var . Indeed, by a standard localization argument (see, e.g., [3 ] , Lemma 4.7), one may apply Corollary 3.8 with G = H G=H G = H , and assume that the RHS of (36) is finite. Since ( Y , Z ) (Y,Z) ( Y , Z ) satisfies Eq. (1) , it holds that
δ Y s , t = − δ Λ s , t − ∫ s t f ( r , Y r , Z r ) d r + δ M s , t Y , P s , t Y = − R s , t Y − ∫ s t f ( r , Y r , Z r ) d r . \delta Y_{s,t}=-\delta\Lambda_{s,t}-\int_s^t f(r,Y_r,Z_r)\,dr+\delta M^Y_{s,t},\qquad P^Y_{s,t}=-R^Y_{s,t}-\int_s^t f(r,Y_r,Z_r)\,dr. δ Y s , t = − δ Λ s , t − ∫ s t f ( r , Y r , Z r ) d r + δ M s , t Y , P s , t Y = − R s , t Y − ∫ s t f ( r , Y r , Z r ) d r .
Note that by (36) the processes Λ \Lambda Λ and R Λ R^\Lambda R Λ have finite p p p -variation and finite p / 2 p/2 p /2 -variation, respectively. Thus, Y Y Y and H ( Y ) H(Y) H ( Y ) have finite p p p -variation, and the pathwise remainder P Y P^Y P Y has finite p / 2 p/2 p /2 -variation. This implies that ( Y , − H ( Y ) ) (Y,-H(Y)) ( Y , − H ( Y )) belongs to RSM X ( p , p ′ ) -var \operatorname{RSM}_{X}^{(p,p')\text{-var}} RSM X ( p , p ′ ) -var .
Comparison Theorem
In this subsection, we prove the comparison principle for rough BSDEs solved in the sense of Definition 4.1, comparing two solutions corresponding to different terminal values and f f f -terms. As a corollary, the uniqueness of solutions is obtained.
We first relate the norm ∥ ⋅ ∣ F s ∥ 1 , ∞ \lVert\cdot\mid\mathcal{F}_s\rVert_{1,\infty} ∥ ⋅ ∣ F s ∥ 1 , ∞ to the VMOp -var ^{p\text{-var}} p -var process. The John–Nirenberg inequality for VMOp -var ^{p\text{-var}} p -var processes, proved in [20 ] , is essential for the proof of the comparison.
Lemma 4.2. Assume S S S is an adapted continuous process. For p ≥ 1 p\ge1 p ≥ 1 and a control function w w w , assume further that ∥ δ S s , t ∣ F s ∥ 1 , ∞ ≤ w ( s , t ) 1 / p \lVert\delta S_{s,t}\mid\mathcal{F}_s\rVert_{1,\infty}\le w(s,t)^{1/p} ∥ δ S s , t ∣ F s ∥ 1 , ∞ ≤ w ( s , t ) 1/ p for all ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ .
Then, S S S is a VMOp -var ^{p\text{-var}} p -var process, and (recall ϱ ( ⋅ ) \varrho(\cdot) ϱ ( ⋅ ) is the modulus of mean oscillation)
∥ ϱ ( S ) ∥ p -var ; [ 0 , T ] ≤ 2 w ( 0 , T ) 1 p . \lVert\varrho(S)\rVert_{p\text{-var};[0,T]}\le2w(0,T)^{\frac{1}{p}}. ∥ ϱ ( S ) ∥ p -var ; [ 0 , T ] ≤ 2 w ( 0 , T ) p 1 .
Proof. By [20 ] , Proposition 2.2, we have ϱ s , t ( S ) ≤ 2 sup u ∈ [ s , t ] ∥ δ S u , t ∣ F u ∥ 1 , ∞ \varrho_{s,t}(S)\le2\sup_{u\in[s,t]}\lVert\delta S_{u,t}\mid\mathcal{F}_u\rVert_{1,\infty} ϱ s , t ( S ) ≤ 2 sup u ∈ [ s , t ] ∥ δ S u , t ∣ F u ∥ 1 , ∞ . Consequently,
∥ ϱ ( S ) ∥ p -var ≤ 2 sup π ∈ P [ 0 , T ] { ∑ [ s , t ] ∈ π sup u ∈ [ s , t ] ∥ δ S u , t ∣ F u ∥ 1 , ∞ p } 1 p ≤ 2 sup π ∈ P [ 0 , T ] { ∑ [ s , t ] ∈ π w ( s , t ) } 1 p ≤ 2 w ( 0 , T ) 1 p , \lVert\varrho(S)\rVert_{p\text{-var}}\le2\sup_{\pi\in\mathcal{P}_{[0,T]}}\left\{\sum_{[s,t]\in\pi}\sup_{u\in[s,t]}\lVert\delta S_{u,t}\mid\mathcal{F}_u\rVert_{1,\infty}^{p}\right\}^{\frac{1}{p}}\le2\sup_{\pi\in\mathcal{P}_{[0,T]}}\left\{\sum_{[s,t]\in\pi}w(s,t)\right\}^{\frac{1}{p}}\le2w(0,T)^{\frac{1}{p}}, ∥ ϱ ( S ) ∥ p -var ≤ 2 π ∈ P [ 0 , T ] sup ⎩ ⎨ ⎧ [ s , t ] ∈ π ∑ u ∈ [ s , t ] sup ∥ δ S u , t ∣ F u ∥ 1 , ∞ p ⎭ ⎬ ⎫ p 1 ≤ 2 π ∈ P [ 0 , T ] sup ⎩ ⎨ ⎧ [ s , t ] ∈ π ∑ w ( s , t ) ⎭ ⎬ ⎫ p 1 ≤ 2 w ( 0 , T ) p 1 ,
which completes the proof. □
Theorem 4.3 (Comparison theorem). Assume ( γ , p , q , q ′ ) (\gamma,p,q,q') ( γ , p , q , q ′ ) holds, H : R → R d H:\mathbb{R}\to\mathbb{R}^d H : R → R d is a function such that D H ∈ C b γ − 1 ( R , R d ) DH\in C_b^{\gamma-1}(\mathbb{R},\mathbb{R}^d) D H ∈ C b γ − 1 ( R , R d ) , and assume X ∈ C p -var ( [ 0 , T ] , R d ) \mathbf{X}\in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^d) X ∈ C p -var ([ 0 , T ] , R d ) . Let ξ \xi ξ and ξ ˉ \bar{\xi} ξ ˉ be two terminal values belonging to L ∞ L^\infty L ∞ , and let f f f and f ˉ \bar{f} f ˉ be two functions satisfying ( f ) (f) ( f ) . Suppose ( Y , Z ) (Y,Z) ( Y , Z ) and ( Y ˉ , Z ˉ ) (\bar{Y},\bar{Z}) ( Y ˉ , Z ˉ ) are solutions to Eq. (1) with terminal values ξ \xi ξ and ξ ˉ \bar{\xi} ξ ˉ and generators f f f and f ˉ \bar{f} f ˉ , respectively, such that
∥ ( Y , − H ( Y ) ) ∥ X ; q , q ′ ; 1 , ∞ + ∥ ( Y ˉ , − H ( Y ˉ ) ) ∥ X ; q , q ′ ; 1 , ∞ < ∞ . \lVert(Y,-H(Y))\rVert_{\mathbf{X};q,q';1,\infty}+\lVert(\bar{Y},-H(\bar{Y}))\rVert_{\mathbf{X};q,q';1,\infty}<\infty. ∥( Y , − H ( Y )) ∥ X ; q , q ′ ; 1 , ∞ + ∥( Y ˉ , − H ( Y ˉ )) ∥ X ; q , q ′ ; 1 , ∞ < ∞.
Then, if ξ ≤ ξ ˉ \xi\le\bar{\xi} ξ ≤ ξ ˉ and f ( r , Y ˉ r , Z ˉ r ) ≤ f ˉ ( r , Y ˉ r , Z ˉ r ) f(r,\bar{Y}_r,\bar{Z}_r)\le\bar{f}(r,\bar{Y}_r,\bar{Z}_r) f ( r , Y ˉ r , Z ˉ r ) ≤ f ˉ ( r , Y ˉ r , Z ˉ r ) a.s. for every r ∈ [ t , T ] r\in[t,T] r ∈ [ t , T ] , we have Y t ≤ Y ˉ t Y_t\le\bar{Y}_t Y t ≤ Y ˉ t a.s. Moreover, if ξ < ξ ˉ \xi<\bar{\xi} ξ < ξ ˉ a.s., then the comparison is strict; namely, Y t < Y ˉ t Y_t<\bar{Y}_t Y t < Y ˉ t a.s.
Remark 4.4. The strict comparison property is not immediate for solutions constructed via smooth approximation approaches, such as those in [7 ] , Theorem 3 and [22 ] , Definition 3.1. By contrast, the RSM framework allows us to establish the strict inequality directly.
Proof. Assume d = 1 d=1 d = 1 without loss of generality. Let R ∈ ( 0 , ∞ ) R\in(0,\infty) R ∈ ( 0 , ∞ ) be a constant such that
∥ D H ∥ C b γ − 1 + C f + ∥ ( Y , − H ( Y ) ) ∥ X ; q , q ′ ; 1 , ∞ + ∥ ( Y ˉ , − H ( Y ˉ ) ) ∥ X ; q , q ′ ; 1 , ∞ + ∥ δ X ∥ p -var + T ≤ R . \lVert DH\rVert_{C_b^{\gamma-1}}+C_f+\lVert(Y,-H(Y))\rVert_{\mathbf{X};q,q';1,\infty}+\lVert(\bar{Y},-H(\bar{Y}))\rVert_{\mathbf{X};q,q';1,\infty}+\lVert\delta X\rVert_{p\text{-var}}+T\le R. ∥ D H ∥ C b γ − 1 + C f + ∥( Y , − H ( Y )) ∥ X ; q , q ′ ; 1 , ∞ + ∥( Y ˉ , − H ( Y ˉ )) ∥ X ; q , q ′ ; 1 , ∞ + ∥ δ X ∥ p -var + T ≤ R .
For t ∈ [ 0 , T ] t \in[0,T] t ∈ [ 0 , T ] , define
{ α t : = ∫ 0 1 D H ( Y t + λ ( Y ‾ t − Y t ) ) d λ , α t ′ : = − ∫ 0 1 D 2 H ( Y t + λ ( Y ‾ t − Y t ) ) ( ( 1 − λ ) H ( Y t ) + λ H ( Y ‾ t ) ) d λ , β t : = f ( t , Y t , Z t ) − f ( t , Y ‾ t , Z ‾ t ) Y t − Y ‾ t 1 { Y t − Y ‾ t ≠ 0 } , and θ t : = f ( t , Y ‾ t , Z t ) − f ( t , Y ‾ t , Z ‾ t ) Z t − Z ‾ t 1 { Z t − Z ‾ t ≠ 0 } . \left\{
\begin{aligned}
\alpha_t &:= \int_0^1 DH(Y_t+\lambda(\overline{Y}_t-Y_t))\,d\lambda,\\
\alpha'_t &:= -\int_0^1 D^2H(Y_t+\lambda(\overline{Y}_t-Y_t))((1-\lambda)H(Y_t)+\lambda H(\overline{Y}_t))\,d\lambda,\\
\beta_t &:= \frac{f(t,Y_t,Z_t)-f(t,\overline{Y}_t,\overline{Z}_t)}{Y_t-\overline{Y}_t}\mathbf{1}_{\{Y_t-\overline{Y}_t\ne0\}}, \quad\text{and}\quad
\theta_t := \frac{f(t,\overline{Y}_t,Z_t)-f(t,\overline{Y}_t,\overline{Z}_t)}{Z_t-\overline{Z}_t}\mathbf{1}_{\{Z_t-\overline{Z}_t\ne0\}}.
\end{aligned}
\right. ⎩ ⎨ ⎧ α t α t ′ β t := ∫ 0 1 D H ( Y t + λ ( Y t − Y t )) d λ , := − ∫ 0 1 D 2 H ( Y t + λ ( Y t − Y t )) (( 1 − λ ) H ( Y t ) + λ H ( Y t )) d λ , := Y t − Y t f ( t , Y t , Z t ) − f ( t , Y t , Z t ) 1 { Y t − Y t = 0 } , and θ t := Z t − Z t f ( t , Y t , Z t ) − f ( t , Y t , Z t ) 1 { Z t − Z t = 0 } .
One can assume q ′ > q q' > q q ′ > q without loss of generality, since one can always choose a constant q ~ ′ > q ′ \widetilde{q}' > q' q ′ > q ′ so that ( γ , p , q , q ~ ′ ) (\gamma,p,q,\widetilde{q}') ( γ , p , q , q ′ ) still satisfies condition ( γ , p , q , q ′ ) (\gamma,p,q,q') ( γ , p , q , q ′ ) . According to (f) and the condition D H ∈ C b γ − 1 DH \in C_b^{\gamma-1} D H ∈ C b γ − 1 , we see that α \alpha α , β \beta β , θ \theta θ , and α ′ \alpha' α ′ are bounded adapted processes. Furthermore, letting p ′ : = q ′ p q p' := \frac{q'p}{q} p ′ := q q ′ p , then γ − 1 ≥ p + p ′ p ′ \gamma-1 \ge\frac{p+p'}{p'} γ − 1 ≥ p ′ p + p ′ . Consequently, since D H ∈ C b γ − 1 ( R , R d ) DH \in C_b^{\gamma-1}(\mathbb{R},\mathbb{R}^d) D H ∈ C b γ − 1 ( R , R d ) and ∥ ( Y , − H ( Y ) ) ∥ X ; q , q ′ ; 1 , ∞ + ∥ ( Y ‾ , − H ( Y ‾ ) ) ∥ X ; q , q ′ ; 1 , ∞ ≤ R \|(Y,-H(Y))\|_{\mathbb{X};q,q';1,\infty}+\|(\overline{Y},-H(\overline{Y}))\|_{\mathbb{X};q,q';1,\infty}\le R ∥ ( Y , − H ( Y )) ∥ X ; q , q ′ ; 1 , ∞ + ∥ ( Y , − H ( Y )) ∥ X ; q , q ′ ; 1 , ∞ ≤ R , by Lemma 3.4 we have that ( α , α ′ ) ∈ R S M X ( p , p ′ ) -var (\alpha,\alpha')\in\mathrm{RSM}^{(p,p')\text{-var}}_{\mathbb{X}} ( α , α ′ ) ∈ RSM X ( p , p ′ ) -var . Moreover, it follows from (3.46) in Lemma 3.14 that for some constant C ^ γ , R > 0 \widehat{C}_{\gamma,R}>0 C γ , R > 0 ,
∥ ∥ δ ( ∫ u ⋅ α r d X r ) ∥ p -var ; [ u , v ] | F u ∥ 1 , ∞ ≤ C ^ γ , R ∣ δ X ∣ p -var ; [ u , v ] (82) \left\|\left\|\delta\left(\int_u^\cdot\alpha_r\,dX_r\right)\right\|_{p\text{-var};[u,v]}\middle|\mathcal{F}_u\right\|_{1,\infty}
\le\widehat{C}_{\gamma,R}|\delta X|_{p\text{-var};[u,v]} \tag*{(82)} δ ( ∫ u ⋅ α r d X r ) p -var ; [ u , v ] F u 1 , ∞ ≤ C γ , R ∣ δ X ∣ p -var ; [ u , v ] (82)
for all ( u , v ) ∈ Δ (u,v)\in\Delta ( u , v ) ∈ Δ .
Therefore, ( Y − Y ‾ , Z − Z ‾ ) (Y-\overline{Y},Z-\overline{Z}) ( Y − Y , Z − Z ) solves the following linear BSDE for t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] :
( Y − Y ‾ ) t = ξ ^ + ∫ t T [ β r ( Y − Y ‾ ) r + θ r ( Z − Z ‾ ) r + f ^ r ] d r + ∫ t T α r ( Y − Y ‾ ) r d X r − ∫ t T ( Z − Z ‾ ) r d W r , (Y-\overline{Y})_t=\widehat{\xi}+\int_t^T[\beta_r(Y-\overline{Y})_r+\theta_r(Z-\overline{Z})_r+\widehat{f}_r]\,dr+\int_t^T\alpha_r(Y-\overline{Y})_r\,dX_r-\int_t^T(Z-\overline{Z})_r\,dW_r, ( Y − Y ) t = ξ + ∫ t T [ β r ( Y − Y ) r + θ r ( Z − Z ) r + f r ] d r + ∫ t T α r ( Y − Y ) r d X r − ∫ t T ( Z − Z ) r d W r ,
where ξ ^ : = ξ − ξ ‾ \widehat{\xi}:=\xi-\overline{\xi} ξ := ξ − ξ and f ^ r : = ( f − f ‾ ) ( r , Y ‾ r , Z ‾ r ) \widehat{f}_r:=(f-\overline{f})(r,\overline{Y}_r,\overline{Z}_r) f r := ( f − f ) ( r , Y r , Z r ) . For ( t , r ) ∈ Δ (t,r)\in\Delta ( t , r ) ∈ Δ , define
S t r : = ∫ t r ( β s − 1 2 ∣ θ s ∣ 2 ) d s + 1 2 tr ( ∫ t r α s α s T d [ X ] s ) + ∫ t r θ s d W s + ∫ t r α s d X s . S_t^r:=\int_t^r\left(\beta_s-\frac{1}{2}|\theta_s|^2\right)\,ds+\frac{1}{2}\operatorname{tr}\left(\int_t^r\alpha_s\alpha_s^{\mathsf{T}}\,d[X]_s\right)+\int_t^r\theta_s\,dW_s+\int_t^r\alpha_s\,dX_s. S t r := ∫ t r ( β s − 2 1 ∣ θ s ∣ 2 ) d s + 2 1 tr ( ∫ t r α s α s T d [ X ] s ) + ∫ t r θ s d W s + ∫ t r α s d X s .
Then, an application of the rough Itô formula in [3 ] to ( Y ⋅ − Y ‾ ⋅ ) e S t ⋅ (Y_\cdot-\overline{Y}_\cdot)e^{S_t^\cdot} ( Y ⋅ − Y ⋅ ) e S t ⋅ yields
( Y t − Y ‾ t ) = e S t T ξ ^ + ∫ t T e S t r f ^ r d r − ∫ t T [ e S t r ( Z r − Z ‾ r ) + θ r e S t r ( Y r − Y ‾ r ) ] d W r . (83) (Y_t-\overline{Y}_t)=e^{S_t^T}\widehat{\xi}+\int_t^T e^{S_t^r}\widehat{f}_r\,dr-\int_t^T[e^{S_t^r}(Z_r-\overline{Z}_r)+\theta_re^{S_t^r}(Y_r-\overline{Y}_r)]\,dW_r. \tag*{(83)} ( Y t − Y t ) = e S t T ξ + ∫ t T e S t r f r d r − ∫ t T [ e S t r ( Z r − Z r ) + θ r e S t r ( Y r − Y r )] d W r . (83)
Note ∥ ∫ u v α r d X r | F u ∥ 1 , ∞ ≤ ∥ ∥ δ ( ∫ u ⋅ α r d X r ) ∥ p -var ; [ u , v ] | F u ∥ 1 , ∞ \left\|\int_u^v\alpha_r\,dX_r\middle|\mathcal{F}_u\right\|_{1,\infty}\le\left\|\left\|\delta\left(\int_u^\cdot\alpha_r\,dX_r\right)\right\|_{p\text{-var};[u,v]}\middle|\mathcal{F}_u\right\|_{1,\infty} ∫ u v α r d X r F u 1 , ∞ ≤ δ ( ∫ u ⋅ α r d X r ) p -var ; [ u , v ] F u 1 , ∞ for all ( u , v ) ∈ Δ (u,v)\in\Delta ( u , v ) ∈ Δ . By (82) and Lemma 4.2, ∫ t ⋅ α r d X r \int_t^\cdot\alpha_r\,dX_r ∫ t ⋅ α r d X r is a V M O p -var \mathrm{VMO}^{p\text{-var}} VMO p -var process with ∥ ϱ ( ∫ t ⋅ α r d X r ) ∥ p -var ; [ t , T ] ≤ 2 C ^ γ , R ∣ δ X ∣ p -var ; [ t , T ] \|\varrho(\int_t^\cdot\alpha_r\,dX_r)\|_{p\text{-var};[t,T]}\le2\widehat{C}_{\gamma,R}|\delta X|_{p\text{-var};[t,T]} ∥ ϱ ( ∫ t ⋅ α r d X r ) ∥ p -var ; [ t , T ] ≤ 2 C γ , R ∣ δ X ∣ p -var ; [ t , T ] . Furthermore, according to the estimate for Young integrals shown in [14 ] , it holds ∥ ∫ u v α r α r T d [ X ] r | F u ∥ 1 , ∞ ≤ C ~ γ , R ∣ δ X ∣ p -var ; [ u , v ] \left\|\int_u^v\alpha_r\alpha_r^{\mathsf{T}}\,d[X]_r\middle|\mathcal{F}_u\right\|_{1,\infty}\le\widetilde{C}_{\gamma,R}|\delta X|_{p\text{-var};[u,v]} ∫ u v α r α r T d [ X ] r F u 1 , ∞ ≤ C γ , R ∣ δ X ∣ p -var ; [ u , v ] for some constant C ~ γ , R > 0 \widetilde{C}_{\gamma,R}>0 C γ , R > 0 . Hence, by Lemma 4.2 again, ∫ t ⋅ α r α r T d [ X ] r \int_t^\cdot\alpha_r\alpha_r^{\mathsf{T}}\,d[X]_r ∫ t ⋅ α r α r T d [ X ] r is a V M O p -var \mathrm{VMO}^{p\text{-var}} VMO p -var process with ∥ ϱ ( ∫ t ⋅ α r α r T d [ X ] r ) ∥ p -var ; [ t , T ] ≤ 2 C ~ γ , R ∣ δ X ∣ p -var ; [ t , T ] \|\varrho(\int_t^\cdot\alpha_r\alpha_r^{\mathsf{T}}\,d[X]_r)\|_{p\text{-var};[t,T]}\le2\widetilde{C}_{\gamma,R}|\delta X|_{p\text{-var};[t,T]} ∥ ϱ ( ∫ t ⋅ α r α r T d [ X ] r ) ∥ p -var ; [ t , T ] ≤ 2 C γ , R ∣ δ X ∣ p -var ; [ t , T ] . Similarly, by the boundedness of β , θ \beta,\theta β , θ , one can prove that ∫ t ⋅ ( β s − 1 2 ∣ θ s ∣ 2 ) d s + ∫ t ⋅ θ s d W s \int_t^\cdot(\beta_s-\frac{1}{2}|\theta_s|^2)\,ds+\int_t^\cdot\theta_s\,dW_s ∫ t ⋅ ( β s − 2 1 ∣ θ s ∣ 2 ) d s + ∫ t ⋅ θ s d W s is a V M O p -var \mathrm{VMO}^{p\text{-var}} VMO p -var process, with the p p p -variation of its modulus of mean oscillation dominated by R R R .
Consequently, S t S^t S t is a V M O p -var \mathrm{VMO}^{p\text{-var}} VMO p -var process. Hence, by the John–Nirenberg inequality shown in [20 ] , for some constant C γ , R > 0 C_{\gamma,R}>0 C γ , R > 0 it holds that for all r ∈ [ t , T ] r\in[t,T] r ∈ [ t , T ] , ∥ e S t r ∥ 2 ≤ C γ , R \|e^{S_t^r}\|_2\le C_{\gamma,R} ∥ e S t r ∥ 2 ≤ C γ , R . Therefore, by (83) , ∫ t T [ e S t r ( Z r − Z ‾ r ) + θ r e S t r ( Y r − Y ‾ r ) ] d W r \int_t^T[e^{S_t^r}(Z_r-\overline{Z}_r)+\theta_re^{S_t^r}(Y_r-\overline{Y}_r)]\,dW_r ∫ t T [ e S t r ( Z r − Z r ) + θ r e S t r ( Y r − Y r )] d W r is uniformly integrable and thus a martingale. Taking conditional expectation of (83) , we have
( Y − Y ‾ ) t = E t [ e S t T ξ ^ + ∫ t T e S t r f ^ r d r ] . (Y-\overline{Y})_t=\mathbb{E}_t\left[e^{S_t^T}\widehat{\xi}+\int_t^T e^{S_t^r}\widehat{f}_r\,dr\right]. ( Y − Y ) t = E t [ e S t T ξ + ∫ t T e S t r f r d r ] .
This yields the (strict) comparison result, noting that e S t r > 0 e^{S_t^r}>0 e S t r > 0 and f ^ r ≥ 0 \widehat{f}_r\ge0 f r ≥ 0 a.s. □ \square □
Remark 4.5. The theory of BSDEs also includes supersolutions, which correspond to dynamics of the form (1.1) augmented by an additional increasing process. We refer to [9 ] for further discussion of supersolutions of BSDEs. As for the comparison theorem, we note that Theorem 4.3 extends without difficulty to the case where Y ‾ t \overline{Y}_t Y t is a supersolution.
Corollary 4.6. Assume ( γ , p , q , q ′ ) (\gamma,p,q,q') ( γ , p , q , q ′ ) holds, H : R → R d H:\mathbb{R}\to\mathbb{R}^d H : R → R d is a function such that D H ∈ C b γ − 1 ( R , R d ) DH\in C_b^{\gamma-1}(\mathbb{R},\mathbb{R}^d) D H ∈ C b γ − 1 ( R , R d ) , and X ∈ C p -var ( [ 0 , T ] , R d ) X\in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^d) X ∈ C p -var ([ 0 , T ] , R d ) . Suppose that ξ ∈ L ∞ \xi\in L^\infty ξ ∈ L ∞ and that f f f satisfies (f). Then, Eq. eq:1 admits at most one solution ( Y , Z ) (Y,Z) ( Y , Z ) satisfying ∥ ( Y , − H ( Y ) ) ∥ X ; q , q ′ ; 1 , ∞ < ∞ \lVert(Y,-H(Y))\rVert_{X;q,q';1,\infty}<\infty ∥( Y , − H ( Y )) ∥ X ; q , q ′ ; 1 , ∞ < ∞ .
Proof. Given any two solutions ( Y ( 1 ) , Z ( 1 ) ) (Y^{(1)},Z^{(1)}) ( Y ( 1 ) , Z ( 1 ) ) and ( Y ( 2 ) , Z ( 2 ) ) (Y^{(2)},Z^{(2)}) ( Y ( 2 ) , Z ( 2 ) ) , Theorem 4.3 yields that Y ( 1 ) = Y ( 2 ) Y^{(1)}=Y^{(2)} Y ( 1 ) = Y ( 2 ) . Consequently, by Eq. eq:1 we get
∫ 0 t ( Z r ( 1 ) − Z r ( 2 ) ) d W r = ∫ 0 t ( f ( r , Y r ( 1 ) , Z r ( 1 ) ) − f ( r , Y r ( 2 ) , Z r ( 2 ) ) ) d r . \int_0^t (Z_r^{(1)}-Z_r^{(2)})\,dW_r=\int_0^t\left(f(r,Y_r^{(1)},Z_r^{(1)})-f(r,Y_r^{(2)},Z_r^{(2)})\right)\,dr. ∫ 0 t ( Z r ( 1 ) − Z r ( 2 ) ) d W r = ∫ 0 t ( f ( r , Y r ( 1 ) , Z r ( 1 ) ) − f ( r , Y r ( 2 ) , Z r ( 2 ) ) ) d r .
This implies Z ( 1 ) = Z ( 2 ) Z^{(1)}=Z^{(2)} Z ( 1 ) = Z ( 2 ) , since a continuous local martingale of finite variation is constant. □
The Rough Semimartingale a priori Estimate
In this subsection, assuming that ( Y , Z ) (Y,Z) ( Y , Z ) is a solution to Eq. eq:1 , and that ∥ ( Y , − H ( Y ) ) ∥ X ; q , q ′ ; k , ∞ \lVert(Y,-H(Y))\rVert_{X;q,q';k,\infty} ∥( Y , − H ( Y )) ∥ X ; q , q ′ ; k , ∞ is finite, we establish upper bounds for ∥ Y ∥ ∞ \lVert Y\rVert_\infty ∥ Y ∥ ∞ and ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ .
Throughout this subsection, we assume ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) holds, X ∈ C p -var ( [ 0 , T ] , R d ) X\in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^d) X ∈ C p -var ([ 0 , T ] , R d ) , H ∈ C b 2 ( R , R d ) H\in C_b^2(\mathbb{R},\mathbb{R}^d) H ∈ C b 2 ( R , R d ) , ξ ∈ L ∞ \xi\in L^\infty ξ ∈ L ∞ , and assumption (f) is satisfied. Let ( Y , Z ) (Y,Z) ( Y , Z ) be a solution to Eq. eq:1 in the sense of Definition 4.1, and thus (see the discussion under Definition 4.1) ( Y , − H ( Y ) ) ∈ RSM X ( p , p ) -var (Y,-H(Y))\in\operatorname{RSM}_X^{(p,p)\text{-var}} ( Y , − H ( Y )) ∈ RSM X ( p , p ) -var .
Proposition 4.7. Assume q < q ′ q<q' q < q ′ . For any k > 1 k>1 k > 1 , assume ( Y , Z ) (Y,Z) ( Y , Z ) is a solution to Eq. eq:1 with
∥ ( Y , − H ( Y ) ) ∥ X ; q , q ′ ; k , ∞ < ∞ , (84) \lVert(Y,-H(Y))\rVert_{X;q,q';k,\infty}<\infty, \tag*{(84)} ∥( Y , − H ( Y )) ∥ X ; q , q ′ ; k , ∞ < ∞ , (84)
Assume ∥ H ∥ C b 2 + C f + ∥ δ X ∥ p -var + T ≤ K \lVert H\rVert_{C_b^2}+C_f+\lVert\delta X\rVert_{p\text{-var}}+T\le K ∥ H ∥ C b 2 + C f + ∥ δ X ∥ p -var + T ≤ K for some positive constant K K K . Then, there exists a constant C k , K > 0 C_{k,K}>0 C k , K > 0 depending only on ( k , K ) (k,K) ( k , K ) such that with θ : = 2 q − 2 > 1 \theta:=\frac{2}{q-2}>1 θ := q − 2 2 > 1 ,
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ≤ C k , K ( 1 + ∥ ξ ∥ ∞ + ∥ δ M Y ∥ B M O θ ) . (85) \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty}\le C_{k,K}\left(1+\lVert\xi\rVert_\infty+\lVert\delta M^Y\rVert_{\mathrm{BMO}}^\theta\right). \tag*{(85)} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ≤ C k , K ( 1 + ∥ ξ ∥ ∞ + ∥ δ M Y ∥ BMO θ ) . (85)
Proof. For t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] , denote
I t X : = ∫ 0 t H ( Y r ) d X r , I t : = ∫ 0 t f ( r , Y r , Z r ) d r . (86) I_t^X:=\int_0^t H(Y_r)\,dX_r,\qquad I_t:=\int_0^t f(r,Y_r,Z_r)\,dr. \tag*{(86)} I t X := ∫ 0 t H ( Y r ) d X r , I t := ∫ 0 t f ( r , Y r , Z r ) d r . (86)
We present the proof in four steps to facilitate readability.
Step 1. First, we establish upper bounds for the p p p -variations of M Y M^Y M Y and Y Y Y . Note that
δ M s , T Y = ∫ s T Z r d W r = ξ − Y s + δ I s , T X + δ I s , T , s ∈ [ 0 , T ] . \delta M^Y_{s,T}=\int_s^T Z_r\,dW_r=\xi-Y_s+\delta I^X_{s,T}+\delta I_{s,T},\qquad s\in[0,T]. δ M s , T Y = ∫ s T Z r d W r = ξ − Y s + δ I s , T X + δ I s , T , s ∈ [ 0 , T ] .
By Lemma A.1 and Doob’s maximal inequality, noting k > 1 k>1 k > 1 , we have
∥ ∥ δ M Y ∥ p -var ; [ s , T ] ∥ F s ∣ k ≲ k ∥ ∥ M Y − M s Y ∥ ∞ ; [ s , T ] ∥ F s ∣ k ≲ k ∥ δ M s , T Y ∥ F s ∣ k ≤ ∥ ξ − Y s ∥ F s ∣ k + ∥ δ I s , T X ∥ F s ∣ k + ∥ δ I s , T ∥ F s ∣ k . (87) \begin{aligned}
\lVert\lVert\delta M^Y\rVert_{p\text{-var};[s,T]}\rVert_{\mathcal{F}_s\mid k}\lesssim_k\lVert\lVert M^Y-M_s^Y\rVert_{\infty;[s,T]}\rVert_{\mathcal{F}_s\mid k}\lesssim_k\lVert\delta M^Y_{s,T}\rVert_{\mathcal{F}_s\mid k} \\
&\le\lVert\xi-Y_s\rVert_{\mathcal{F}_s\mid k}+\lVert\delta I^X_{s,T}\rVert_{\mathcal{F}_s\mid k}+\lVert\delta I_{s,T}\rVert_{\mathcal{F}_s\mid k}. \tag*{(87)}
\end{aligned} ∥∥ δ M Y ∥ p -var ; [ s , T ] ∥ F s ∣ k ≲ k ∥∥ M Y − M s Y ∥ ∞ ; [ s , T ] ∥ F s ∣ k ≲ k ∥ δ M s , T Y ∥ F s ∣ k ≤ ∥ ξ − Y s ∥ F s ∣ k + ∥ δ I s , T X ∥ F s ∣ k + ∥ δ I s , T ∥ F s ∣ k . (87)
Note ξ − Y s = ξ − E s [ ξ ] + E s [ ξ − Y s ] \xi-Y_s=\xi-\mathbb{E}_s[\xi]+\mathbb{E}_s[\xi-Y_s] ξ − Y s = ξ − E s [ ξ ] + E s [ ξ − Y s ] and E s [ ξ − Y s ] = − E s [ δ ( I X + I ) s , T ] \mathbb{E}_s[\xi-Y_s]=-\mathbb{E}_s[\delta(I^X+I)_{s,T}] E s [ ξ − Y s ] = − E s [ δ ( I X + I ) s , T ] . Hence, we have
∥ ξ − Y s ∥ F s ∣ k ≤ ∥ ξ − E s [ ξ ] ∥ F s ∣ k + ∥ E s [ δ ( I X + I ) s , T ] ∥ F s ∣ k ≤ 2 ∥ ξ ∥ ∞ + ∥ δ ( I X + I ) s , T ∥ F s ∣ k . (88) \lVert\xi-Y_s\rVert_{\mathcal{F}_s\mid k}\le\lVert\xi-\mathbb{E}_s[\xi]\rVert_{\mathcal{F}_s\mid k}+\lVert\mathbb{E}_s[\delta(I^X+I)_{s,T}]\rVert_{\mathcal{F}_s\mid k}\le2\lVert\xi\rVert_\infty+\lVert\delta(I^X+I)_{s,T}\rVert_{\mathcal{F}_s\mid k}. \tag*{(88)} ∥ ξ − Y s ∥ F s ∣ k ≤ ∥ ξ − E s [ ξ ] ∥ F s ∣ k + ∥ E s [ δ ( I X + I ) s , T ] ∥ F s ∣ k ≤ 2 ∥ ξ ∥ ∞ + ∥ δ ( I X + I ) s , T ∥ F s ∣ k . (88)
Then, combining (87) and (88) , we have
∥ ∥ δ M Y ∥ p -var ; [ s , T ] ∥ F s ∣ k ≲ k ∥ ξ ∥ ∞ + ∥ ∥ δ I X ∥ p -var ; [ s , T ] ∥ F s ∣ k + ∥ ∥ δ I ∥ p -var ; [ s , T ] ∥ F s ∣ k . (89) \lVert\lVert\delta M^Y\rVert_{p\text{-var};[s,T]}\rVert_{\mathcal{F}_s\mid k}\lesssim_k\lVert\xi\rVert_\infty+\lVert\lVert\delta I^X\rVert_{p\text{-var};[s,T]}\rVert_{\mathcal{F}_s\mid k}+\lVert\lVert\delta I\rVert_{p\text{-var};[s,T]}\rVert_{\mathcal{F}_s\mid k}. \tag*{(89)} ∥∥ δ M Y ∥ p -var ; [ s , T ] ∥ F s ∣ k ≲ k ∥ ξ ∥ ∞ + ∥∥ δ I X ∥ p -var ; [ s , T ] ∥ F s ∣ k + ∥∥ δ I ∥ p -var ; [ s , T ] ∥ F s ∣ k . (89)
This, together with the fact that δ Y = − δ I X − δ I + δ M Y \delta Y=-\delta I^X-\delta I+\delta M^Y δ Y = − δ I X − δ I + δ M Y , yields
∥ ∥ δ Y ∥ p -var ; [ s , T ] ∥ F s ∣ k ≲ k ∥ ξ ∥ ∞ + ∥ ∥ δ I X ∥ p -var ; [ s , T ] ∥ F s ∣ k + ∥ ∥ δ I ∥ p -var ; [ s , T ] ∥ F s ∣ k . (90) \lVert\lVert\delta Y\rVert_{p\text{-var};[s,T]}\rVert_{\mathcal{F}_s\mid k}\lesssim_k\lVert\xi\rVert_\infty+\lVert\lVert\delta I^X\rVert_{p\text{-var};[s,T]}\rVert_{\mathcal{F}_s\mid k}+\lVert\lVert\delta I\rVert_{p\text{-var};[s,T]}\rVert_{\mathcal{F}_s\mid k}. \tag*{(90)} ∥∥ δ Y ∥ p -var ; [ s , T ] ∥ F s ∣ k ≲ k ∥ ξ ∥ ∞ + ∥∥ δ I X ∥ p -var ; [ s , T ] ∥ F s ∣ k + ∥∥ δ I ∥ p -var ; [ s , T ] ∥ F s ∣ k . (90)
Step 2. In this step, we estimate the terms involving ∥ δ I X ∥ p -var \lVert\delta I^{X}\rVert_{p\text{-var}} ∥ δ I X ∥ p -var and ∥ δ I ∥ p -var \lVert\delta I\rVert_{p\text{-var}} ∥ δ I ∥ p -var on the RHS of (89) and (90) . By Corollary 3.8 (where q < q ′ q<q' q < q ′ is required), we have for all ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , (recall q ′ ′ / 2 : = q q ′ / ( q + q ′ ) q''/2:=qq'/(q+q') q ′′ /2 := q q ′ / ( q + q ′ ) )
∥ ∥ I X ∥ p / 2 -var ; [ s , t ] | F s ∥ k + ∥ ∥ δ I X ∥ p -var ; [ s , t ] | F s ∥ k ≲ k , K ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∥ ∥ δ Y ∥ q -var ; [ s , t ] | F s ∥ k + ∥ δ M Y ∥ B M O ; [ s , t ] θ + ∥ ∥ P Y ∥ q ′ ′ / 2 -var ; [ s , t ] | F s ∥ k ) . (91) \left\lVert\left\lVert I^{X}\right\rVert_{p/2\text{-var};[s,t]}\middle|\mathcal{F}_{s}\right\rVert_{k}+\left\lVert\left\lVert\delta I^{X}\right\rVert_{p\text{-var};[s,t]}\middle|\mathcal{F}_{s}\right\rVert_{k}
\lesssim_{k,K}|\delta X|_{p\text{-var};[s,t]}\left(1+\left\lVert\left\lVert\delta Y\right\rVert_{q\text{-var};[s,t]}\middle|\mathcal{F}_{s}\right\rVert_{k}+\left\lVert\delta M^{Y}\right\rVert_{\mathrm{BMO};[s,t]}^{\theta}+\left\lVert\left\lVert P^{Y}\right\rVert_{q''/2\text{-var};[s,t]}\middle|\mathcal{F}_{s}\right\rVert_{k}\right). \tag*{(91)} I X p /2 -var ; [ s , t ] F s k + δ I X p -var ; [ s , t ] F s k ≲ k , K ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∥ δ Y ∥ q -var ; [ s , t ] F s k + δ M Y BMO ; [ s , t ] θ + P Y q ′′ /2 -var ; [ s , t ] F s k ) . (91)
In addition, by Assumption ( f ) (f) ( f ) and the fact ∣ Y r ∣ ≤ ∥ ξ ∥ ∞ + ∥ δ Y ∥ q -var ; [ s , T ] |Y_{r}|\leq\|\xi\|_{\infty}+\lVert\delta Y\rVert_{q\text{-var};[s,T]} ∣ Y r ∣ ≤ ∥ ξ ∥ ∞ + ∥ δ Y ∥ q -var ; [ s , T ] for r ∈ [ s , t ] r\in[s,t] r ∈ [ s , t ] , we get
∥ ∥ δ I ∥ p / 2 -var ; [ s , t ] | F s ∥ k ≤ ∥ ∫ s t ∣ f ( r , Y r , Z r ) ∣ d r | F s ∥ k ≤ ∥ ∫ s t ∣ f ( r , 0 , 0 ) ∣ d r | F s ∥ k + ∥ ∫ s t ∣ f ( r , Y r , Z r ) − f ( r , 0 , 0 ) ∣ d r | F s ∥ k \begin{aligned}
\left\lVert\left\lVert\delta I\right\rVert_{p/2\text{-var};[s,t]}\middle|\mathcal{F}_{s}\right\rVert_{k}\leq\left\lVert\int_{s}^{t}|f(r,Y_{r},Z_{r})|\mathrm{d}r\middle|\mathcal{F}_{s}\right\rVert_{k} \\
&\leq\left\lVert\int_{s}^{t}|f(r,0,0)|\mathrm{d}r\middle|\mathcal{F}_{s}\right\rVert_{k}+\left\lVert\int_{s}^{t}|f(r,Y_{r},Z_{r})-f(r,0,0)|\mathrm{d}r\middle|\mathcal{F}_{s}\right\rVert_{k}
\end{aligned} ∥ δ I ∥ p /2 -var ; [ s , t ] F s k ≤ ∫ s t ∣ f ( r , Y r , Z r ) ∣ d r F s k ≤ ∫ s t ∣ f ( r , 0 , 0 ) ∣ d r F s k + ∫ s t ∣ f ( r , Y r , Z r ) − f ( r , 0 , 0 ) ∣ d r F s k
≲ k , K ( t − s ) ( 1 + ∥ ξ ∥ ∞ + ∥ ∥ δ Y ∥ q -var ; [ s , T ] | F s ∥ k ) + ( t − s ) 1 / 2 ∥ δ M Y ∥ B M O ; [ s , t ] θ . (92) \lesssim_{k,K}(t-s)\left(1+\|\xi\|_{\infty}+\left\lVert\left\lVert\delta Y\right\rVert_{q\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}\right)+(t-s)^{1/2}\left\lVert\delta M^{Y}\right\rVert_{\mathrm{BMO};[s,t]}^{\theta}. \tag*{(92)} ≲ k , K ( t − s ) ( 1 + ∥ ξ ∥ ∞ + ∥ δ Y ∥ q -var ; [ s , T ] F s k ) + ( t − s ) 1/2 δ M Y BMO ; [ s , t ] θ . (92)
Step 3. Now, we prove the desired bound (85) for small T T T . Plugging both (91) and (92) into (89) and (90) , noting that ∥ δ M Y ∥ B M O ≤ 1 + ∥ δ M Y ∥ B M O \lVert\delta M^{Y}\rVert_{\mathrm{BMO}}\leq1+\lVert\delta M^{Y}\rVert_{\mathrm{BMO}} ∥ δ M Y ∥ BMO ≤ 1 + ∥ δ M Y ∥ BMO by Young’s inequality, we get
∥ ∥ δ M Y ∥ p -var ; [ s , T ] | F s ∥ k + ∥ ∥ δ Y ∥ p -var ; [ s , T ] | F s ∥ k ≲ k , K 1 + ∥ ξ ∥ ∞ + ( ∣ T − s ∣ 1 / 2 + ∣ δ X ∣ p -var ; [ s , T ] ) ( 1 + ∥ ∥ δ Y ∥ q -var ; [ s , T ] | F s ∥ k + ∥ δ M Y ∥ B M O ; [ s , T ] θ + ∥ ∥ P Y ∥ q ′ ′ / 2 -var ; [ s , T ] | F s ∥ k ) . (93) \left\lVert\left\lVert\delta M^{Y}\right\rVert_{p\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}+\left\lVert\left\lVert\delta Y\right\rVert_{p\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}
\lesssim_{k,K}1+\|\xi\|_{\infty}+\left(|T-s|^{1/2}+|\delta X|_{p\text{-var};[s,T]}\right)\left(1+\left\lVert\left\lVert\delta Y\right\rVert_{q\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}+\left\lVert\delta M^{Y}\right\rVert_{\mathrm{BMO};[s,T]}^{\theta}+\left\lVert\left\lVert P^{Y}\right\rVert_{q''/2\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}\right). \tag*{(93)} δ M Y p -var ; [ s , T ] F s k + ∥ δ Y ∥ p -var ; [ s , T ] F s k ≲ k , K 1 + ∥ ξ ∥ ∞ + ( ∣ T − s ∣ 1/2 + ∣ δ X ∣ p -var ; [ s , T ] ) ( 1 + ∥ δ Y ∥ q -var ; [ s , T ] F s k + δ M Y BMO ; [ s , T ] θ + P Y q ′′ /2 -var ; [ s , T ] F s k ) . (93)
Next, noting ∥ δ Y ′ ∥ p -var = ∥ δ H ( Y ) ∥ p -var ≤ ∥ D H ∥ ∞ ∥ δ Y ∥ p -var \lVert\delta Y'\rVert_{p\text{-var}}=\lVert\delta H(Y)\rVert_{p\text{-var}}\leq\lVert DH\rVert_{\infty}\lVert\delta Y\rVert_{p\text{-var}} ∥ δ Y ′ ∥ p -var = ∥ δ H ( Y ) ∥ p -var ≤ ∥ D H ∥ ∞ ∥ δ Y ∥ p -var , we have
\left\lVert\left\lVert\delta Y'\right\rVert_{p\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}=\left\lVert\left\lVert\delta H(Y)\right\rVert_{p\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}\lesssim_{k,K}\text{RHS of }[](#eq:4.12).
Moreover, by Remark 2.11, we have that ( I X , H ( Y ) ) ∈ D X ( p , p ) -var (I^{X},H(Y))\in\mathcal{D}_{X}^{(p,p)\text{-var}} ( I X , H ( Y )) ∈ D X ( p , p ) -var a.s. with P Y = − R I X − δ I P^{Y}=-RI^{X}-\delta I P Y = − R I X − δ I . Hence, combining (91) , (92) , and (93) , we get
\begin{aligned}
\lVert H(Y)\rVert_{\infty}+\left\lVert\left\lVert\delta H(Y)\right\rVert_{p\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}+\left\lVert\left\lVert P^{Y}\right\rVert_{p/2\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k} \\
&+\left\lVert\left\lVert\delta M^{Y}\right\rVert_{p\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}\lesssim_{k,K}\text{RHS of }[](#eq:4.12). \tag*{(94)}
\end{aligned}
Taking the essential supremum over [ 0 , T ] × Ω [0,T]\times\Omega [ 0 , T ] × Ω in (94) , the finiteness of ∥ ( Y , − H ( Y ) ) ∥ X ; q , q ′ ; k , ∞ \lVert(Y,-H(Y))\rVert_{X;q,q';k,\infty} ∥( Y , − H ( Y )) ∥ X ; q , q ′ ; k , ∞ assumed in (84) yields that ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ < ∞ \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty}<\infty ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ < ∞ .
Furthermore, by (52) in Lemma 3.13,
\left\lVert\left\lVert\delta Y\right\rVert_{q\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}\leq\left\lVert\left\lVert\delta Y\right\rVert_{p\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}\leq\left(1+\lVert\delta X\rVert_{p\text{-var};[s,T]}\right)\times\text{LHS of }[](#eq:4.13). \tag*{(95)}
Therefore, by (94) and (95) , noting ∥ ∥ P Y ∥ q ′ ′ / 2 -var ; [ s , T ] | F s ∥ k ≤ ∥ ∥ P Y ∥ p / 2 -var ; [ s , T ] | F s ∥ k \left\lVert\left\lVert P^{Y}\right\rVert_{q''/2\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k}\leq\left\lVert\left\lVert P^{Y}\right\rVert_{p/2\text{-var};[s,T]}\middle|\mathcal{F}_{s}\right\rVert_{k} P Y q ′′ /2 -var ; [ s , T ] F s k ≤ P Y p /2 -var ; [ s , T ] F s k , we get
\begin{aligned}
\text{LHS of }[](#eq:4.13)\lesssim_{k,K}(1+|\delta X|_{p\text{-var};[0,T]})(1+\|\xi\|_{\infty})+(|T|^{1/2}+|\delta X|_{p\text{-var};[0,T]})\lVert\delta M^{Y}\rVert_{\mathrm{BMO}}^{\theta} \\
&+(|T|^{1/2}+|\delta X|_{p\text{-var};[0,T]})(1+|\delta X|_{p\text{-var};[0,T]})\times\text{LHS of }[](#eq:4.13). \tag*{(96)}
\end{aligned}
By taking essential supremum over ( s , ω ) ∈ [ 0 , T ] × Ω (s,\omega)\in[0,T]\times\Omega ( s , ω ) ∈ [ 0 , T ] × Ω in (96) , and applying Lemma A.1,
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ≤ C k , K ( 1 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ( 1 + ∥ ξ ∥ ∞ ) + ( ∣ T ∣ 1 / 2 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ∥ δ M Y ∥ B M O θ + ( ∣ T ∣ 1 / 2 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ( 1 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ . (97) \begin{aligned}
\lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty}\leq C_{k,K}(1+|\delta X|_{p\text{-var};[0,T]})(1+\|\xi\|_{\infty})+(|T|^{1/2}+|\delta X|_{p\text{-var};[0,T]})\lVert\delta M^{Y}\rVert_{\mathrm{BMO}}^{\theta} \\
&+(|T|^{1/2}+|\delta X|_{p\text{-var};[0,T]})(1+|\delta X|_{p\text{-var};[0,T]})\lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty}. \tag*{(97)}
\end{aligned} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ≤ C k , K ( 1 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ( 1 + ∥ ξ ∥ ∞ ) + ( ∣ T ∣ 1/2 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ∥ δ M Y ∥ BMO θ + ( ∣ T ∣ 1/2 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ( 1 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ . (97)
where C k , K > 0 C_{k,K}>0 C k , K > 0 is a constant independent of ξ \xi ξ . Denote ε : = ( C k , K ( 1 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ) − 1 \varepsilon:= \left(C_{k,K}(1+|\delta X|_{p\text{-var};[0,T]})\right)^{-1} ε := ( C k , K ( 1 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ) − 1 , and choose T T T sufficiently small such that ( ∣ T ∣ 1 2 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ≤ 1 2 ε \left(|T|^{\frac{1}{2}}+|\delta X|_{p\text{-var};[0,T]}\right)\le\frac{1}{2}\varepsilon ( ∣ T ∣ 2 1 + ∣ δ X ∣ p -var ; [ 0 , T ] ) ≤ 2 1 ε . Then, it follows from (97) that
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ≤ ( 2 C k , K + 1 ) ( 1 + ∥ ξ ∥ ∞ ) + ∥ δ M Y ∥ B M O θ . (98) \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty}\le(2C_{k,K}+1)(1+\lVert\xi\rVert_\infty)+\lVert\delta M^Y\rVert_{\mathrm{BMO}}^\theta. \tag*{(98)} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ≤ ( 2 C k , K + 1 ) ( 1 + ∥ ξ ∥ ∞ ) + ∥ δ M Y ∥ BMO θ . (98)
Step 4. Now we establish an estimate for ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ on an arbitrary interval [ 0 , T ] [0,T] [ 0 , T ] . Let C k , K C_{k,K} C k , K be the constant chosen in (97) , and let ε \varepsilon ε be defined as above. Since w ^ ( s , t ) : = ∣ t − s ∣ + ∥ δ X ∥ p -var ; [ s , t ] p + ∥ X ∥ p / 2 -var ; [ s , t ] p / 2 \widehat{w}(s,t):=|t-s|+\lVert\delta X\rVert_{p\text{-var};[s,t]}^p+\lVert\mathbb{X}\rVert_{p/2\text{-var};[s,t]}^{p/2} w ( s , t ) := ∣ t − s ∣ + ∥ δ X ∥ p -var ; [ s , t ] p + ∥ X ∥ p /2 -var ; [ s , t ] p /2 defines a control function, it follows from [15 ] (see also [21 ] ) that there exists a partition π = { [ t i , t i + 1 ] ; i = 0 , 1 , … , N − 1 } ∈ P [ 0 , T ] \pi=\{[t_i,t_{i+1}];\ i=0,1,\ldots,N-1\}\in\mathcal{P}_{[0,T]} π = {[ t i , t i + 1 ] ; i = 0 , 1 , … , N − 1 } ∈ P [ 0 , T ] such that w ^ ( t i , t i + 1 ) ≤ min { ∣ ε / 6 ∣ 2 , ∣ ε / 6 ∣ p , ∣ ε / 6 ∣ p / 2 } \widehat{w}(t_i,t_{i+1})\le\min\{|\varepsilon/6|^2,|\varepsilon/6|^p,|\varepsilon/6|^{p/2}\} w ( t i , t i + 1 ) ≤ min { ∣ ε /6 ∣ 2 , ∣ ε /6 ∣ p , ∣ ε /6 ∣ p /2 } for all i i i . Consequently, ∣ t i + 1 − t i ∣ 1 2 + ∣ δ X ∣ p -var ; [ t i , t i + 1 ] ≤ ε / 2 |t_{i+1}-t_i|^{\frac{1}{2}}+|\delta X|_{p\text{-var};[t_i,t_{i+1}]}\le\varepsilon/2 ∣ t i + 1 − t i ∣ 2 1 + ∣ δ X ∣ p -var ; [ t i , t i + 1 ] ≤ ε /2 . Moreover, by the superadditivity of the control function w ^ \widehat{w} w , the size N N N of the partition is bounded by a constant depending only on T T T , ∣ δ X ∣ p -var ; [ 0 , T ] |\delta X|_{p\text{-var};[0,T]} ∣ δ X ∣ p -var ; [ 0 , T ] and ε \varepsilon ε (and hence only on K K K and ε \varepsilon ε ).
Then, by repeating the procedure in Step 1–Step 3 on each interval [ t i , t i + 1 ] ∈ π [t_i,t_{i+1}]\in\pi [ t i , t i + 1 ] ∈ π in place of [ 0 , T ] [0,T] [ 0 , T ] , we obtain, as in (98) ,
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ; [ t i , t i + 1 ] ≤ ( 2 C k , K + 1 ) ( 1 + ∥ Y t i + 1 ∥ ∞ ) + ∥ δ M Y ∥ B M O θ . (99) \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty;[t_i,t_{i+1}]}\le(2C_{k,K}+1)(1+\lVert Y_{t_{i+1}}\rVert_\infty)+\lVert\delta M^Y\rVert_{\mathrm{BMO}}^\theta. \tag*{(99)} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ; [ t i , t i + 1 ] ≤ ( 2 C k , K + 1 ) ( 1 + ∥ Y t i + 1 ∥ ∞ ) + ∥ δ M Y ∥ BMO θ . (99)
For ∥ Y t i + 1 ∥ ∞ \lVert Y_{t_{i+1}}\rVert_\infty ∥ Y t i + 1 ∥ ∞ , by (50) in Lemma 3.13, there exists a constant C ^ k , K > 0 \widehat{C}_{k,K}>0 C k , K > 0 independent of i i i such that
∥ Y t i + 1 ∥ ∞ ≤ C ^ k , K ( ∥ ξ ∥ ∞ + ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ; [ t i + 1 , T ] ) . (100) \lVert Y_{t_{i+1}}\rVert_\infty\le\widehat{C}_{k,K}\left(\lVert\xi\rVert_\infty+\lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty;[t_{i+1},T]}\right). \tag*{(100)} ∥ Y t i + 1 ∥ ∞ ≤ C k , K ( ∥ ξ ∥ ∞ + ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ; [ t i + 1 , T ] ) . (100)
Combining (99) and (100) , it holds, for some constant C ~ k , K > 0 \widetilde{C}_{k,K}>0 C k , K > 0 independent of i i i ,
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ; [ t i , t i + 1 ] ≤ C ~ k , K ( 1 + ∥ ξ ∥ ∞ ) + C ~ k , K ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ; [ t i + 1 , T ] + ∥ δ M Y ∥ B M O θ . (101) \begin{aligned}
\lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty;[t_i,t_{i+1}]} &\le\widetilde{C}_{k,K}(1+\lVert\xi\rVert_\infty)\\
&\quad+\widetilde{C}_{k,K}\lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty;[t_{i+1},T]}+\lVert\delta M^Y\rVert_{\mathrm{BMO}}^\theta. \tag*{(101)}
\end{aligned} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ; [ t i , t i + 1 ] ≤ C k , K ( 1 + ∥ ξ ∥ ∞ ) + C k , K ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ; [ t i + 1 , T ] + ∥ δ M Y ∥ BMO θ . (101)
Furthermore, by Proposition 3.15, we see that
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ; [ t i , T ] ≲ K ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ; [ t i , t i + 1 ] + ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ; [ t i + 1 , T ] . \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty;[t_i,T]}\lesssim_K\lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty;[t_i,t_{i+1}]}+\lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty;[t_{i+1},T]}. ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ; [ t i , T ] ≲ K ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ; [ t i , t i + 1 ] + ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ; [ t i + 1 , T ] .
Then by (101) , there exists a constant C k , K > 0 C_{k,K}>0 C k , K > 0 such that for i = 0 , 1 , … , N − 1 i=0,1,\ldots,N-1 i = 0 , 1 , … , N − 1 ,
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ; [ t i , T ] ≤ C ~ k , K ( 1 + ∥ ξ ∥ ∞ + ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ; [ t i + 1 , T ] ) + ∥ δ M Y ∥ B M O θ . (102) \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty;[t_i,T]}\le\widetilde{C}_{k,K}\left(1+\lVert\xi\rVert_\infty+\lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty;[t_{i+1},T]}\right)+\lVert\delta M^Y\rVert_{\mathrm{BMO}}^\theta. \tag*{(102)} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ; [ t i , T ] ≤ C k , K ( 1 + ∥ ξ ∥ ∞ + ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ; [ t i + 1 , T ] ) + ∥ δ M Y ∥ BMO θ . (102)
In particular, for i = N − 1 i=N-1 i = N − 1 , this yields
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ; [ t N − 1 , T ] ≤ ( 2 C k , K + 1 ) ( 1 + ∥ ξ ∥ ∞ ) + ∥ δ M Y ∥ B M O θ . \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty;[t_{N-1},T]}\le(2C_{k,K}+1)(1+\lVert\xi\rVert_\infty)+\lVert\delta M^Y\rVert_{\mathrm{BMO}}^\theta. ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ; [ t N − 1 , T ] ≤ ( 2 C k , K + 1 ) ( 1 + ∥ ξ ∥ ∞ ) + ∥ δ M Y ∥ BMO θ .
The desired (85) then follows by iterating (102) . □
In the previous result, we show that the RSM seminorm ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ is dominated by 1 + ∥ ξ ∥ ∞ + ∥ δ M Y ∥ B M O θ 1+\lVert\xi\rVert_\infty+\lVert\delta M^Y\rVert_{\mathrm{BMO}}^\theta 1 + ∥ ξ ∥ ∞ + ∥ δ M Y ∥ BMO θ for some θ > 1 \theta>1 θ > 1 . Next in Proposition 4.8, we show that ∥ δ M Y ∥ B M O \lVert\delta M^Y\rVert_{\mathrm{BMO}} ∥ δ M Y ∥ BMO can be bounded by the variant depending on ∥ Y ∥ ∞ \lVert Y\rVert_\infty ∥ Y ∥ ∞ and system parameters.
Proposition 4.8. Let ( Y , Z ) (Y,Z) ( Y , Z ) be a solution to Eq. (1) satisfying (84) with k > 1 k>1 k > 1 . Assume ∥ Y ∥ ∞ + ∥ H ∥ C b 2 + C f + ∣ δ X ∣ p -var ; [ 0 , T ] + T ≤ K ^ \lVert Y\rVert_\infty+\lVert H\rVert_{C_b^2}+C_f+|\delta X|_{p\text{-var};[0,T]}+T\le\widehat{K} ∥ Y ∥ ∞ + ∥ H ∥ C b 2 + C f + ∣ δ X ∣ p -var ; [ 0 , T ] + T ≤ K , where K ^ > 0 \widehat{K}>0 K > 0 is a constant. Then, there exists a constant C k , K ^ > 0 C_{k,\widehat{K}}>0 C k , K > 0 depending only on ( k , K ^ ) (k,\widehat{K}) ( k , K ) s.t.
∥ δ M Y ∥ B M O ≤ C k , K ^ . (103) \lVert\delta M^Y\rVert_{\mathrm{BMO}}\le C_{k,\widehat{K}}. \tag*{(103)} ∥ δ M Y ∥ BMO ≤ C k , K . (103)
Proof. Without loss of generality, one can choose a sufficiently larger q ′ q' q ′ so that q < q ′ q<q' q < q ′ and conditions ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) and (84) still hold. By Proposition 4.7 (where q < q ′ q<q' q < q ′ is required due to the application of Corollary 3.8), we have ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ < ∞ \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty}<\infty ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ < ∞ . Therefore, we may assume further without loss of generality that q > 3 q>3 q > 3 , hence that q ′ ∈ ( q , 4 ) q'\in(q,4) q ′ ∈ ( q , 4 ) , and that q ′ q' q ′ is chosen sufficiently close to q q q , such that ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) and (84) hold. Note that this implies θ = 2 q − 2 ∈ ( 1 , 2 ) \theta= \frac{2}{q-2} \in(1,2) θ = q − 2 2 ∈ ( 1 , 2 ) . Denote the process I t [ X ] : = tr ( ∫ 0 t H ( Y r ) H ( Y r ) ⊤ d [ X ] r ) I^{[X]}_t := \operatorname{tr}\left(\int_0^t H(Y_r)H(Y_r)^\top\mathrm{d}[X]_r\right) I t [ X ] := tr ( ∫ 0 t H ( Y r ) H ( Y r ) ⊤ d [ X ] r ) . By the rough Itô formula shown in [3 ] , we have
∣ Y t ∣ 2 + ∥ Z ∥ L 2 ( [ t , T ] ) 2 = ∣ ξ ∣ 2 + 2 ∫ t T Y r H ( Y r ) d X r − 2 ∫ t T Y r Z r d W r + 2 ∫ t T Y r f ( r , Y r , Z r ) d r − δ I t , T [ X ] . (104) |Y_t|^2+\|Z\|_{L^2([t,T])}^2=|\xi|^2+2\int_t^T Y_rH(Y_r)\mathrm{d}X_r-2\int_t^T Y_rZ_r\mathrm{d}W_r+2\int_t^T Y_rf(r,Y_r,Z_r)\mathrm{d}r-\delta I^{[X]}_{t,T}. \tag*{(104)} ∣ Y t ∣ 2 + ∥ Z ∥ L 2 ([ t , T ]) 2 = ∣ ξ ∣ 2 + 2 ∫ t T Y r H ( Y r ) d X r − 2 ∫ t T Y r Z r d W r + 2 ∫ t T Y r f ( r , Y r , Z r ) d r − δ I t , T [ X ] . (104)
Taking the conditional expectation E t [ ⋅ ] \mathbb{E}_t[\cdot] E t [ ⋅ ] on (104) , for each t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] we get
E t [ ∥ Z ∥ L 2 ( [ t , T ] ) 2 ] ≤ E t [ ∣ ξ ∣ 2 + 2 ∫ t T Y r H ( Y r ) d X r + 2 ∫ t T Y r f ( r , Y r , Z r ) d r − δ I t , T [ X ] ] . (105) \mathbb{E}_t\left[\|Z\|_{L^2([t,T])}^2\right]\leq\mathbb{E}_t\left[|\xi|^2+2\int_t^T Y_rH(Y_r)\mathrm{d}X_r+2\int_t^T Y_rf(r,Y_r,Z_r)\mathrm{d}r-\delta I^{[X]}_{t,T}\right]. \tag*{(105)} E t [ ∥ Z ∥ L 2 ([ t , T ]) 2 ] ≤ E t [ ∣ ξ ∣ 2 + 2 ∫ t T Y r H ( Y r ) d X r + 2 ∫ t T Y r f ( r , Y r , Z r ) d r − δ I t , T [ X ] ] . (105)
Let κ ( ⋅ ) \kappa(\cdot) κ ( ⋅ ) be a smooth cut-off function s.t. κ ( y ) = 1 \kappa(y)=1 κ ( y ) = 1 if ∣ y ∣ ≤ K ^ |y|\leq\widehat K ∣ y ∣ ≤ K and κ ( y ) = 0 \kappa(y)=0 κ ( y ) = 0 if ∣ y ∣ > K ^ + 1 |y|>\widehat K+1 ∣ y ∣ > K + 1 . Denote H ~ ( y ) : = κ ( y ) y ⋅ H ( y ) \widetilde H(y):=\kappa(y)y\cdot H(y) H ( y ) := κ ( y ) y ⋅ H ( y ) , f ~ ( t , y , z ) : = κ ( y ) y ⋅ f ( t , κ ( y ) y , z ) \widetilde f(t,y,z):=\kappa(y)y\cdot f(t,\kappa(y)y,z) f ( t , y , z ) := κ ( y ) y ⋅ f ( t , κ ( y ) y , z ) .
Then, by the condition ∥ H ∥ C b 2 < ∞ \|H\|_{C_b^2}<\infty ∥ H ∥ C b 2 < ∞ and Assumption ( f ) (f) ( f ) , it follows for all t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] ,
∥ H ~ ∥ C b 2 ≲ K ^ 1 , ∣ f ~ ( t , y , z ) ∣ ≲ K ^ 1 + ∣ z ∣ . (106) \|\widetilde H\|_{C_b^2}\lesssim_{\widehat K}1,\qquad|\widetilde f(t,y,z)|\lesssim_{\widehat K}1+|z|. \tag*{(106)} ∥ H ∥ C b 2 ≲ K 1 , ∣ f ( t , y , z ) ∣ ≲ K 1 + ∣ z ∣. (106)
Furthermore, since K ^ \widehat K K is an upper bound of ∥ Y ∥ ∞ \|Y\|_\infty ∥ Y ∥ ∞ , the inequality (105) leads to:
E t [ δ ⟨ M Y ⟩ t , T ] = E t [ ∥ Z ∥ L 2 ( [ t , T ] ) 2 ] ≤ ∥ ξ ∥ ∞ 2 + 2 E t [ δ ( I ~ X + I ~ ) t , T − δ I t , T [ X ] ] , (107) \mathbb{E}_t\left[\delta\langle M^Y\rangle_{t,T}\right]=\mathbb{E}_t\left[\|Z\|_{L^2([t,T])}^2\right]\leq\|\xi\|_\infty^2+2\mathbb{E}_t\left[\delta(\widetilde I^X+\widetilde I)_{t,T}-\delta I^{[X]}_{t,T}\right], \tag*{(107)} E t [ δ ⟨ M Y ⟩ t , T ] = E t [ ∥ Z ∥ L 2 ([ t , T ]) 2 ] ≤ ∥ ξ ∥ ∞ 2 + 2 E t [ δ ( I X + I ) t , T − δ I t , T [ X ] ] , (107)
where
I ~ t X : = ∫ 0 t H ~ ( Y r ) d X r , I ~ t : = ∫ 0 t f ~ ( r , Y r , Z r ) d r , t ∈ [ 0 , T ] . \widetilde I^X_t:=\int_0^t\widetilde H(Y_r)\mathrm{d}X_r,\qquad\widetilde I_t:=\int_0^t\widetilde f(r,Y_r,Z_r)\mathrm{d}r,\qquad t\in[0,T]. I t X := ∫ 0 t H ( Y r ) d X r , I t := ∫ 0 t f ( r , Y r , Z r ) d r , t ∈ [ 0 , T ] .
To estimate I ~ X \widetilde I^X I X , using Corollary 3.8 3.8 3.8 , we have for t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] ,
∥ δ I ~ t , T X ∣ F t ∥ k ≲ k , K ^ ∣ δ X ∣ p -var ( 1 + ∥ ∥ δ Y ∥ q -var ; [ t , T ] ∣ F t ∥ k + ∥ δ M Y ∥ B M O θ + ∥ ∥ P Y ∥ q ′ ′ 2 -var ; [ t , T ] ∣ F t ∥ k ) . (108) \left\|\delta\widetilde I^X_{t,T}\mid\mathcal{F}_t\right\|_k\lesssim_{k,\widehat K}|\delta X|_{p\text{-var}}\left(1+\|\|\delta Y\|_{q\text{-var};[t,T]}\mid\mathcal{F}_t\|_k+\|\delta M^Y\|_{\mathrm{BMO}}^\theta+\|\|P^Y\|_{\frac{q''}{2}\text{-var};[t,T]}\mid\mathcal{F}_t\|_k\right). \tag*{(108)} δ I t , T X ∣ F t k ≲ k , K ∣ δ X ∣ p -var ( 1 + ∥∥ δ Y ∥ q -var ; [ t , T ] ∣ F t ∥ k + ∥ δ M Y ∥ BMO θ + ∥∥ P Y ∥ 2 q ′′ -var ; [ t , T ] ∣ F t ∥ k ) . (108)
In addition, by ( 3.42 ) (3.42) ( 3.42 ) in Lemma 3.13 3.13 3.13 and Proposition 4.7 4.7 4.7 (where it requires k > 1 k>1 k > 1 ), we have
∥ ∥ δ Y ∥ q -var ; [ t , T ] ∣ F t ∥ k ≲ k , K ^ 1 + ∥ ( Y , − H ( Y ) ) ∥ X ; q , q ′ ; k , ∞ ≲ k , K ^ 1 + ∥ δ M Y ∥ B M O θ . (109) \|\|\delta Y\|_{q\text{-var};[t,T]}\mid\mathcal{F}_t\|_k\lesssim_{k,\widehat K}1+\|(Y,-H(Y))\|_{\mathbb{X};q,q';k,\infty}\lesssim_{k,\widehat K}1+\|\delta M^Y\|_{\mathrm{BMO}}^\theta. \tag*{(109)} ∥∥ δ Y ∥ q -var ; [ t , T ] ∣ F t ∥ k ≲ k , K 1 + ∥ ( Y , − H ( Y )) ∥ X ; q , q ′ ; k , ∞ ≲ k , K 1 + ∥ δ M Y ∥ BMO θ . (109)
Moreover, by Proposition 4.7 4.7 4.7 again, we also have
∥ ∥ P Y ∥ q ′ ′ 2 -var ; [ t , T ] ∣ F t ∥ k ≲ k , K ^ 1 + ∥ δ M Y ∥ B M O θ . (110) \|\|P^Y\|_{\frac{q''}{2}\text{-var};[t,T]}\mid\mathcal{F}_t\|_k\lesssim_{k,\widehat K}1+\|\delta M^Y\|_{\mathrm{BMO}}^\theta. \tag*{(110)} ∥∥ P Y ∥ 2 q ′′ -var ; [ t , T ] ∣ F t ∥ k ≲ k , K 1 + ∥ δ M Y ∥ BMO θ . (110)
Hence, combining (108) , (109) , and (110) , it follows that
∥ ∥ δ I ~ t , T X ∣ F t ∥ k ≲ k , K ^ 1 + ∥ δ M Y ∥ B M O θ . (111) \|\|\delta\widetilde I^X_{t,T}\mid\mathcal{F}_t\|_k\lesssim_{k,\widehat K}1+\|\delta M^Y\|_{\mathrm{BMO}}^\theta. \tag*{(111)} ∥∥ δ I t , T X ∣ F t ∥ k ≲ k , K 1 + ∥ δ M Y ∥ BMO θ . (111)
On the other hand, for the term I ~ \widetilde I I , by (106) and Hölder’s inequality, we obtain that
∥ ∥ δ I ~ t , T ∣ F t ∥ k = ∥ ∫ t T f ~ ( r , Y r , Z r ) d r ∣ F t ∥ k ≲ k , K ^ 1 + ∥ ∥ Z ∥ L 2 ( [ t , T ] ) ∣ F t ∥ k ≲ k 1 + ∥ δ M Y ∥ B M O . (112) \|\|\delta\widetilde I_{t,T}\mid\mathcal{F}_t\|_k=\left\|\left.\int_t^T\widetilde f(r,Y_r,Z_r)\mathrm{d}r\right|\mathcal{F}_t\right\|_k\lesssim_{k,\widehat K}1+\|\|Z\|_{L^2([t,T])}\mid\mathcal{F}_t\|_k\lesssim_k1+\|\delta M^Y\|_{\mathrm{BMO}}. \tag*{(112)} ∥∥ δ I t , T ∣ F t ∥ k = ∫ t T f ( r , Y r , Z r ) d r F t k ≲ k , K 1 + ∥∥ Z ∥ L 2 ([ t , T ]) ∣ F t ∥ k ≲ k 1 + ∥ δ M Y ∥ BMO . (112)
In addition, by the estimate for Young integrals (see, e.g., [8 ] ) and (109) ,
δ I t , T [ X ] ≲ K ^ ( 1 + E t [ ∥ δ Y ∥ q -var ; [ t , T ] ] ) ∥ δ [ X ] ∥ p 2 -var ; [ t , T ] ≲ k , K ^ 1 + ∥ δ M Y ∥ B M O θ . (113) \delta I^{[X]}_{t,T}\lesssim_{\widehat K}\left(1+\mathbb{E}_t[\|\delta Y\|_{q\text{-var};[t,T]}]\right)\|\delta[X]\|_{\frac{p}{2}\text{-var};[t,T]}\lesssim_{k,\widehat K}1+\|\delta M^Y\|_{\mathrm{BMO}}^\theta. \tag*{(113)} δ I t , T [ X ] ≲ K ( 1 + E t [ ∥ δ Y ∥ q -var ; [ t , T ] ] ) ∥ δ [ X ] ∥ 2 p -var ; [ t , T ] ≲ k , K 1 + ∥ δ M Y ∥ BMO θ . (113)
Combining (107) , (111) , (112) , and (113) , we see that, with θ < 2 \theta<2 θ < 2 ,
∥ δ M Y ∥ B M O 2 = ess sup ( t , ω ) ∈ [ 0 , T ] × Ω E t [ δ ⟨ M Y ⟩ t , T ] ≲ k , K ^ 1 + ∥ δ M Y ∥ B M O θ , \|\delta M^Y\|_{\mathrm{BMO}}^2=\operatorname*{ess\,sup}_{(t,\omega)\in[0,T]\times\Omega}\mathbb{E}_t\left[\delta\langle M^Y\rangle_{t,T}\right]\lesssim_{k,\widehat K}1+\|\delta M^Y\|_{\mathrm{BMO}}^\theta, ∥ δ M Y ∥ BMO 2 = ( t , ω ) ∈ [ 0 , T ] × Ω ess sup E t [ δ ⟨ M Y ⟩ t , T ] ≲ k , K 1 + ∥ δ M Y ∥ BMO θ ,
and then obtain (103) via Young’s inequality. □ \square □
The following Theorem 4.9 gives an estimate for ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ . Recall that Θ \Theta Θ is defined in (7) , and define Θ 2 : = ( Θ , C f , R 2 ) \Theta_2 := (\Theta,C_f,R_2) Θ 2 := ( Θ , C f , R 2 ) .
Theorem 4.9. Assume H ∈ C b γ H \in C_b^\gamma H ∈ C b γ for some γ > p \gamma> p γ > p , and assume further that ( f d e t ) (f_{\mathrm{det}}) ( f det ) holds. Suppose that for some constant R 2 > 0 R_2 > 0 R 2 > 0 ,
∥ ξ ∥ ∞ + ∥ H ∥ C b 2 + ∣ δ X ∣ p -var ; [ 0 , T ] + T ≤ R 2 . \lVert\xi\rVert_\infty+ \lVert H \rVert_{C_b^2} + \lvert\delta X \rvert_{p\text{-var};[0,T]} + T \le R_2. ∥ ξ ∥ ∞ + ∥ H ∥ C b 2 + ∣ δ X ∣ p -var ; [ 0 , T ] + T ≤ R 2 .
Let ( Y , Z ) (Y,Z) ( Y , Z ) be a solution to Eq. (1) satisfying (84) with k > 1 k > 1 k > 1 . Then, there is a constant C γ , Θ 2 > 0 C_{\gamma,\Theta_2} > 0 C γ , Θ 2 > 0 depending only on ( γ , Θ 2 ) (\gamma,\Theta_2) ( γ , Θ 2 ) such that
∥ Y ∥ ∞ + ∥ δ M Y ∥ B M O ≤ C γ , Θ 2 . (114) \lVert Y \rVert_\infty+ \lVert\delta M^Y \rVert_{\mathrm{BMO}} \le C_{\gamma,\Theta_2}. \tag*{(114)} ∥ Y ∥ ∞ + ∥ δ M Y ∥ BMO ≤ C γ , Θ 2 . (114)
Moreover, we have that for some constant C k , γ , Θ 2 > 0 C_{k,\gamma,\Theta_2} > 0 C k , γ , Θ 2 > 0 depending only on ( k , γ , Θ 2 ) (k,\gamma,\Theta_2) ( k , γ , Θ 2 ) ,
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ ≤ C k , γ , Θ 2 . (115) \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty} \le C_{k,\gamma,\Theta_2}. \tag*{(115)} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ≤ C k , γ , Θ 2 . (115)
Proof. Without loss of generality, one can choose a sufficiently larger q ′ q' q ′ so that q < q ′ q < q' q < q ′ and conditions ( p , q , q ′ ) (p,q,q') ( p , q , q ′ ) and (84) still hold.
We first establish an upper bound for ∥ Y ∥ ∞ \lVert Y \rVert_\infty ∥ Y ∥ ∞ . Consider the following (deterministic) RDEs:
{ Y ‾ t = C ‾ + ∫ t T f ( r , Y ‾ r , 0 ) d r + ∫ t T H ( Y ‾ r ) d X r , Y ‾ t = C ‾ + ∫ t T f ( r , Y ‾ r , 0 ) d r + ∫ t T H ( Y ‾ r ) d X r , \begin{cases}
\overline{Y}_t = \overline{C} + \int_t^T f(r,\overline{Y}_r,0)\,dr + \int_t^T H(\overline{Y}_r)\,dX_r,\\
\underline{Y}_t = \underline{C} + \int_t^T f(r,\underline{Y}_r,0)\,dr + \int_t^T H(\underline{Y}_r)\,dX_r,
\end{cases} { Y t = C + ∫ t T f ( r , Y r , 0 ) d r + ∫ t T H ( Y r ) d X r , Y t = C + ∫ t T f ( r , Y r , 0 ) d r + ∫ t T H ( Y r ) d X r ,
where C ‾ : = ∥ ξ ∥ ∞ \overline{C} := \lVert\xi\rVert_\infty C := ∥ ξ ∥ ∞ and C ‾ : = − ∥ ξ ∥ ∞ \underline{C} := -\lVert\xi\rVert_\infty C := − ∥ ξ ∥ ∞ . Noting H ∈ C b γ H \in C_b^\gamma H ∈ C b γ , by Proposition A.4, the above two equations admit unique solutions Y ‾ \overline{Y} Y and Y ‾ \underline{Y} Y , respectively, such that ( Y ‾ , − H ( Y ‾ ) ) (\overline{Y},-H(\overline{Y})) ( Y , − H ( Y )) , ( Y ‾ , − H ( Y ‾ ) ) ∈ D X ( p , p ) -var (\underline{Y},-H(\underline{Y})) \in\mathcal{D}^{(p,p)\text{-var}}_X ( Y , − H ( Y )) ∈ D X ( p , p ) -var .
Since Y ‾ \overline{Y} Y (resp. Y ‾ \underline{Y} Y ) is deterministic, ( Y ‾ , 0 ) (\overline{Y},0) ( Y , 0 ) (resp. ( Y ‾ , 0 ) (\underline{Y},0) ( Y , 0 ) ) uniquely solves BSDE (1) with ξ \xi ξ replaced by C ‾ \overline{C} C (resp. C ‾ \underline{C} C ). Thus, we have
∥ ( Y ‾ , − H ( Y ‾ ) ) ∥ X ; p , p ; k , ∞ + ∥ ( Y ‾ , − H ( Y ‾ ) ) ∥ X ; p , p ; k , ∞ ≤ ∥ ( Y ‾ , − H ( Y ‾ ) ) ∥ D X ( p , p ) -var + ∥ ( Y ‾ , − H ( Y ‾ ) ) ∥ D X ( p , p ) -var + 2 ∥ H ( ⋅ ) ∥ ∞ < ∞ . \lVert(\overline{Y},-H(\overline{Y}))\rVert_{X;p,p;k,\infty}+\lVert(\underline{Y},-H(\underline{Y}))\rVert_{X;p,p;k,\infty}
\le\lVert(\overline{Y},-H(\overline{Y}))\rVert_{\mathcal{D}^{(p,p)\text{-var}}_X}+\lVert(\underline{Y},-H(\underline{Y}))\rVert_{\mathcal{D}^{(p,p)\text{-var}}_X}+2\lVert H(\cdot)\rVert_\infty<\infty. ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ + ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ ≤ ∥( Y , − H ( Y )) ∥ D X ( p , p ) -var + ∥( Y , − H ( Y )) ∥ D X ( p , p ) -var + 2 ∥ H ( ⋅ ) ∥ ∞ < ∞.
Therefore, we can apply Theorem 4.3, noting C ‾ ≤ ξ ≤ C ‾ \underline{C} \le\xi\le\overline{C} C ≤ ξ ≤ C , to get
Y ‾ t ≤ Y t ≤ Y ‾ t . (116) \underline{Y}_t \le Y_t \le\overline{Y}_t. \tag*{(116)} Y t ≤ Y t ≤ Y t . (116)
In addition, by the a priori estimate for RDEs shown in Proposition A.4, we have
∣ Y ‾ t ∣ ∨ ∣ Y ‾ t ∣ ≤ C γ , Θ 2 , for all t ∈ [ 0 , T ] , \lvert\overline{Y}_t\rvert\vee\lvert\underline{Y}_t\rvert\le C_{\gamma,\Theta_2},\quad\text{for all }t\in[0,T], ∣ Y t ∣ ∨ ∣ Y t ∣ ≤ C γ , Θ 2 , for all t ∈ [ 0 , T ] ,
for some constant C γ , Θ 2 C_{\gamma,\Theta_2} C γ , Θ 2 . This together with (116) yields ∥ Y ∥ ∞ ≤ C γ , Θ 2 \lVert Y\rVert_\infty\le C_{\gamma,\Theta_2} ∥ Y ∥ ∞ ≤ C γ , Θ 2 .
Finally, by Propositions 4.8 and then 4.7 (noting k > 1 k > 1 k > 1 and q < q ′ q < q' q < q ′ are required), the boundedness of ∥ Y ∥ ∞ \lVert Y\rVert_\infty ∥ Y ∥ ∞ implies (114) and (115) . □ \square □
Remark 4.10. In the above theorem, the function f f f can be random, provided that it is uniformly bounded from above and below by two deterministic uniformly Lipschitz continuous functions. In contrast, Theorem 4.3 does not permit a comparison in the coefficient H ( ⋅ ) H(\cdot) H ( ⋅ ) , which serves as the integrand in the rough integral. Consequently, it is nontrivial to extend H ( ⋅ ) H(\cdot) H ( ⋅ ) to a stochastic (controlled) field, as the forward case considered in [11 , 1 ] .
Remark 4.11. In this paper, we focus on 1-dimensional equations. The main difficulty in the multi-dimensional case is that the a priori estimate (see Theorem 4.9) relies heavily on the comparison theorem (Theorem 4.3) which is not applicable to multi-dimensional cases in general. To see this, Proposition 4.7 and 4.8 show that once the essential supremum of the modulus of the solution Y Y Y is bounded, the RSM seminorm is bounded as well. However, since Z Z Z is also unknown in the equation, it is difficult to obtain a priori estimates for Y Y Y using the classical RDE approach (e.g. Friz and Hairer [10 ] , Proposition 8.2). We refer to Liang and Tang [22 ] for some special multidimensional cases via Doss–Sussmann transformation.
Proposition 4.12 shows that the solutions to classical BSDEs satisfy condition (84) .
Proposition 4.12. Let X ∈ C p -var ( [ 0 , T ] , R d ) X \in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^{d}) X ∈ C p -var ([ 0 , T ] , R d ) be the canonical lift of a smooth path X : [ 0 , T ] → R d X:[0,T]\to\mathbb{R}^{d} X : [ 0 , T ] → R d . Suppose that ξ ∈ L ∞ \xi\in L^{\infty} ξ ∈ L ∞ , and that ( f ) (f) ( f ) holds. Then, there exists a unique solution to Eq. (1) such that for any k ≥ 1 k \ge1 k ≥ 1 ,
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ < ∞ . (117) \|(Y,-H(Y))\|_{X;p,p;k,\infty}<\infty. \tag*{(117)} ∥ ( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ < ∞. (117)
Proof. The well-posedness of Eq. (1) follows from the classical BSDE theory. Indeed, by [30 ] , Theorem 7.3.3, there exists a unique pair ( Y , Z ) (Y,Z) ( Y , Z ) such that Y Y Y is continuous, adapted and bounded, Z Z Z is progressively measurable and quadratically integrable, and the equation (1) holds. Moreover, clearly Y Y Y is a semimartingale, and the rough stochastic integral ∫ H ( Y r ) d X r \int H(Y_r)\,\mathrm{d}X_r ∫ H ( Y r ) d X r coincides with the Lebesgue integral ∫ H ( Y r ) ∂ r X r d r \int H(Y_r)\partial_rX_r\,\mathrm{d}r ∫ H ( Y r ) ∂ r X r d r . Thus, ( Y , − H ( Y ) ) ∈ RSM X ( p , p ) -var (Y,-H(Y)) \in\operatorname{RSM}_{X}^{(p,p)\text{-var}} ( Y , − H ( Y )) ∈ RSM X ( p , p ) -var , and ( Y , Z ) (Y,Z) ( Y , Z ) solves Eq. (1) in the sense of Definition 4.1.
Furthermore, by [30 ] , Theorem 7.2.1, the process M t Y = ∫ 0 t Z r d W r M_t^Y=\int_0^t Z_r\,\mathrm{d}W_r M t Y = ∫ 0 t Z r d W r is a BMO martingale. Let I X , I I^X,\mathcal{I} I X , I be defined by (86) . Then, for ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ , we have P s , t Y = − δ ( I X + I ) s , t + H ( Y s ) δ X s , t P_{s,t}^Y=-\delta(I^X+\mathcal{I})_{s,t}+H(Y_s)\delta X_{s,t} P s , t Y = − δ ( I X + I ) s , t + H ( Y s ) δ X s , t , and hence
∥ ∥ P Y ∥ p 2 -var ∥ ∞ ; k , ∞ ≤ sup s ∈ [ 0 , T ] ∥ ∫ s T ∣ H ( Y r ) − H ( Y s ) ∣ ⋅ ∣ ∂ r X r ∣ d r + ∫ s T ∣ f ( r , Y r , Z r ) ∣ d r ∣ F s ∥ k , ∞ \left\|\|P^Y\|_{\frac{p}{2}\text{-var}}\right\|_{\infty;k,\infty}
\leq\sup_{s\in[0,T]}\left\|\left.\int_s^T |H(Y_r)-H(Y_s)|\cdot|\partial_rX_r|\,\mathrm{d}r+\int_s^T |f(r,Y_r,Z_r)|\,\mathrm{d}r\right|\mathcal{F}_s\right\|_{k,\infty} ∥ P Y ∥ 2 p -var ∞ ; k , ∞ ≤ s ∈ [ 0 , T ] sup ∫ s T ∣ H ( Y r ) − H ( Y s ) ∣ ⋅ ∣ ∂ r X r ∣ d r + ∫ s T ∣ f ( r , Y r , Z r ) ∣ d r F s k , ∞
≲ ∥ H ∥ ∞ sup r ∈ [ 0 , T ] ∣ ∂ r X r ∣ T + C f ( T + ∥ Y ∥ ∞ T + ∥ δ M Y ∥ B M O T 1 2 ) < ∞ . \lesssim\|H\|_{\infty}\sup_{r\in[0,T]}|\partial_rX_r|T+C_f\left(T+\|Y\|_{\infty}T+\|\delta M^Y\|_{\mathrm{BMO}}T^{\frac12}\right)<\infty. ≲ ∥ H ∥ ∞ r ∈ [ 0 , T ] sup ∣ ∂ r X r ∣ T + C f ( T + ∥ Y ∥ ∞ T + ∥ δ M Y ∥ BMO T 2 1 ) < ∞.
Similarly, it also holds ∥ ∥ δ ( I X + I ) ∥ p -var ∥ ∞ ; k , ∞ < ∞ \left\|\|\delta(I^X+\mathcal{I})\|_{p\text{-var}}\right\|_{\infty;k,\infty}<\infty ∥ δ ( I X + I ) ∥ p -var ∞ ; k , ∞ < ∞ . Noting that δ Y s , t = − δ I s , t X − δ I s , t + δ M s , t Y \delta Y_{s,t}=-\delta I^X_{s,t}-\delta\mathcal{I}_{s,t}+\delta M^Y_{s,t} δ Y s , t = − δ I s , t X − δ I s , t + δ M s , t Y , and that ∥ ∥ δ M Y ∥ p -var ∥ ∞ ; k , ∞ ≲ k ∥ δ M Y ∥ B M O < ∞ \left\|\|\delta M^Y\|_{p\text{-var}}\right\|_{\infty;k,\infty}\lesssim_k\|\delta M^Y\|_{\mathrm{BMO}}<\infty ∥ δ M Y ∥ p -var ∞ ; k , ∞ ≲ k ∥ δ M Y ∥ BMO < ∞ (see Lemma A.1), we get ∥ ∥ δ Y ∥ p -var ∥ ∞ ; k , ∞ < ∞ \left\|\|\delta Y\|_{p\text{-var}}\right\|_{\infty;k,\infty}<\infty ∥ ∥ δ Y ∥ p -var ∥ ∞ ; k , ∞ < ∞ , and hence ∥ ∥ δ H ( Y ) ∥ p -var ∥ ∞ ; k , ∞ < ∞ \left\|\|\delta H(Y)\|_{p\text{-var}}\right\|_{\infty;k,\infty}<\infty ∥ ∥ δ H ( Y ) ∥ p -var ∥ ∞ ; k , ∞ < ∞ . Consequently, the desired (117) is obtained, which, combined with Corollary 4.6, also implies that ( Y , Z ) (Y,Z) ( Y , Z ) is the unique solution to Eq. (1) . □
Well-posedness of Rough BSDEs
In this section, we establish the well-posedness of Eq. (1) in the sense of Definition 4.1. We first define the solution map, and then prove the invariance and the contraction of the solution map on small intervals. Next, a solution is constructed via Picard’s iteration, and then the local well-posedness together with the a priori estimate of solutions, allows us to extend the unique solution globally. After proving the well-posedness of the solution, we establish the stability of the solution with respect to coefficients. Throughout this subsection, we assume ( γ , p , q , q ′ ) (\gamma,p,q,q') ( γ , p , q , q ′ ) .
To facilitate the statement of Lemma 4.13, let R > 0 R>0 R > 0 be a constant such that
∥ H ∥ C b γ − 1 + ∣ δ X ∣ p -var ; [ 0 , T ] + T ≤ R , \|H\|_{\mathcal{C}^{\gamma-1}_b}+|\delta X|_{p\text{-var};[0,T]}+T\leq R, ∥ H ∥ C b γ − 1 + ∣ δ X ∣ p -var ; [ 0 , T ] + T ≤ R ,
and denote Θ ‾ : = ( Θ , C f , R ) \overline{\Theta}:=(\Theta,C_f,R) Θ := ( Θ , C f , R ) .
Lemma 4.13. Let ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ and k ≥ 1 k\geq1 k ≥ 1 . Assume ( γ , p , q , q ′ ) (\gamma,p,q,q') ( γ , p , q , q ′ ) , ( f ) (f) ( f ) , X = ( X , X ) ∈ C p -var ( [ s , t ] , R d ) \mathbf{X}=(X,\mathbb{X})\in\mathcal{C}^{p\text{-var}}([s,t],\mathbb{R}^{d}) X = ( X , X ) ∈ C p -var ([ s , t ] , R d ) , ξ ∈ L t ∞ \xi\in L^\infty_t ξ ∈ L t ∞ , and H ∈ C b γ − 1 H\in C_b^{\gamma-1} H ∈ C b γ − 1 . Let ( Y , Y ′ ) ∈ RSM X ( p , p ) -var ( [ s , t ] , R ) (Y,Y')\in\operatorname{RSM}_{X}^{(p,p)\text{-var}}([s,t],\mathbb{R}) ( Y , Y ′ ) ∈ RSM X ( p , p ) -var ([ s , t ] , R ) , Z Z Z be a progressively measurable process with ∫ s t ∣ Z r ∣ 2 d r < ∞ \int_s^t|Z_r|^2\,\mathrm{d}r<\infty ∫ s t ∣ Z r ∣ 2 d r < ∞ a.s. Assume ∥ Y t ∥ ∞ < ∞ \|Y_t\|_\infty<\infty ∥ Y t ∥ ∞ < ∞ , ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] < ∞ \|(Y,Y')\|_{X;q,q';k,\infty;[s,t]}<\infty ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] < ∞ , and δ M s , t Y = ∫ s t Z r d W r \delta M^Y_{s,t}=\int_s^t Z_r\,\mathrm{d}W_r δ M s , t Y = ∫ s t Z r d W r . Then, there exists a unique triplet ( Y ~ , − H ( Y ) , Z ~ ) (\widetilde{Y},-H(Y),\widetilde{Z}) ( Y , − H ( Y ) , Z ) such that ( Y ~ , − H ( Y ) ) ∈ RSM X p , p ( [ s , t ] , R ) (\widetilde{Y},-H(Y))\in\operatorname{RSM}_{X}^{p,p}([s,t],\mathbb{R}) ( Y , − H ( Y )) ∈ RSM X p , p ([ s , t ] , R ) , and Z ~ \widetilde{Z} Z is a progressively measurable, ∫ s t ∣ Z ~ r ∣ 2 d r < ∞ \int_s^t|\widetilde{Z}_r|^2\,\mathrm{d}r<\infty ∫ s t ∣ Z r ∣ 2 d r < ∞ a.s., ∥ ( Y ~ , − H ( Y ) ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] < ∞ \|(\widetilde{Y},-H(Y))\|_{X;q,q';k,\infty;[s,t]}<\infty ∥ ( Y , − H ( Y )) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] < ∞ , δ M s , t Y ~ = \delta M^{\widetilde{Y}}_{s,t}= δ M s , t Y = ∫ s t Z ~ r d W r \int_s^t \widetilde{Z}_r\,dW_r ∫ s t Z r d W r , and
Y ~ v = ξ + ∫ v t f ( r , Y r , Z r ) d r + ∫ v t H ( Y r ) d X r − ∫ v t Z ~ r d W r , v ∈ [ s , t ] . (118) \widetilde{Y}_v=\xi+\int_v^t f(r,Y_r,Z_r)\,dr+\int_v^t H(Y_r)\,dX_r-\int_v^t \widetilde{Z}_r\,dW_r,\qquad v\in[s,t]. \tag*{(118)} Y v = ξ + ∫ v t f ( r , Y r , Z r ) d r + ∫ v t H ( Y r ) d X r − ∫ v t Z r d W r , v ∈ [ s , t ] . (118)
Moreover, we have
∥ ( Y ~ , − H ( Y ) ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ ˉ 1 + ∥ ξ ∥ ∞ + ∥ Y t ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] p / q ′ . (119) \left\lVert(\widetilde{Y},-H(Y))\right\rVert_{X;q,q';k,\infty;[s,t]}
\lesssim_{k,\gamma,\bar{\Theta}}1+\lVert\xi\rVert_{\infty}+\lVert Y_t\rVert_{\infty}
+\left(\lvert\delta X\rvert_{p\text{-var};[s,t]}+(t-s)^{\frac{1}{2}}\right)
\left\lVert(Y,Y')\right\rVert_{X;q,q';k,\infty;[s,t]}^{\gamma^{-1}}
+\left\lVert(Y,Y')\right\rVert_{X;q,q';k,\infty;[s,t]}^{p/q'}. \tag*{(119)} ( Y , − H ( Y )) X ; q , q ′ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ ˉ 1 + ∥ ξ ∥ ∞ + ∥ Y t ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] p / q ′ . (119)
Proof. For simplicity, we denote I v X : = ∫ s v H ( Y r ) d X r I_v^X:=\int_s^v H(Y_r)\,dX_r I v X := ∫ s v H ( Y r ) d X r and I v : = ∫ s v f ( r , Y r , Z r ) d r I_v:=\int_s^v f(r,Y_r,Z_r)\,dr I v := ∫ s v f ( r , Y r , Z r ) d r , v ∈ [ s , t ] v\in[s,t] v ∈ [ s , t ] . By Lemma 3.2, and by conditions ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] < ∞ \left\lVert(Y,Y')\right\rVert_{X;q,q';k,\infty;[s,t]}<\infty ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] < ∞ and ξ ∈ L t ∞ \xi\in L_t^\infty ξ ∈ L t ∞ ,
∥ ξ ∥ 2 + ∥ ∥ δ I ∥ 1 -var ; [ s , t ] + ∥ δ I X ∥ p -var ; [ s , t ] ∥ 2 ≲ k ∥ ξ ∥ ∞ + ∥ ∥ δ I ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ I X ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] \lVert\xi\rVert_2+\left\lVert\left\lVert\delta I\right\rVert_{1\text{-var};[s,t]}+\left\lVert\delta I^X\right\rVert_{p\text{-var};[s,t]}\right\rVert_2
\lesssim_k\lVert\xi\rVert_\infty+\left\lVert\left\lVert\delta I\right\rVert_{1\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}+\left\lVert\left\lVert\delta I^X\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]} ∥ ξ ∥ 2 + ∥ δ I ∥ 1 -var ; [ s , t ] + δ I X p -var ; [ s , t ] 2 ≲ k ∥ ξ ∥ ∞ + ∥ ∥ δ I ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] + δ I X p -var ∞ ; k , ∞ ; [ s , t ]
≲ k , γ , Θ ˉ ∥ ξ ∥ ∞ + ( ∣ t − s ∣ + ∣ t − s ∣ 1 2 ) ( 1 + ∥ Y ∥ L ∞ ( [ s , t ] × Ω ) + ∥ δ M Y ∥ B M O ; [ s , t ] ) \lesssim_{k,\gamma,\bar{\Theta}}\lVert\xi\rVert_\infty+\left(\lvert t-s\rvert+\lvert t-s\rvert^{\frac{1}{2}}\right)\left(1+\lVert Y\rVert_{L^\infty([s,t]\times\Omega)}+\lVert\delta M^Y\rVert_{\mathrm{BMO};[s,t]}\right) ≲ k , γ , Θ ˉ ∥ ξ ∥ ∞ + ( ∣ t − s ∣ + ∣ t − s ∣ 2 1 ) ( 1 + ∥ Y ∥ L ∞ ([ s , t ] × Ω ) + ∥ δ M Y ∥ BMO ; [ s , t ] )
+ ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) \qquad+\lvert\delta X\rvert_{p\text{-var};[s,t]}\left(1+\left\lVert(Y,Y')\right\rVert_{X;q,q';k,\infty;[s,t]}^{\gamma^{-1}}\right) + ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 )
≲ Θ ˉ ∥ ξ ∥ ∞ + ∥ Y t ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ∣ t − s ∣ 1 2 ) ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) . (120) \lesssim_{\bar{\Theta}}\lVert\xi\rVert_\infty+\lVert Y_t\rVert_\infty+\left(\lvert\delta X\rvert_{p\text{-var};[s,t]}+\lvert t-s\rvert^{\frac{1}{2}}\right)\left(1+\left\lVert(Y,Y')\right\rVert_{X;q,q';k,\infty;[s,t]}^{\gamma^{-1}}\right). \tag*{(120)} ≲ Θ ˉ ∥ ξ ∥ ∞ + ∥ Y t ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ∣ t − s ∣ 2 1 ) ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) . (120)
where we apply the following inequalities,
∥ δ I ∥ 1 -var ; [ v , t ] ≤ C f ∣ ∫ v t ( 1 + ∣ Y r ∣ + ∣ Z r ∣ ) d r ∣ \left\lVert\delta I\right\rVert_{1\text{-var};[v,t]}\leq C_f\left\lvert\int_v^t(1+\lvert Y_r\rvert+\lvert Z_r\rvert)\,dr\right\rvert ∥ δ I ∥ 1 -var ; [ v , t ] ≤ C f ∫ v t ( 1 + ∣ Y r ∣ + ∣ Z r ∣) d r
≲ Θ ˉ ∣ t − v ∣ ( 1 + ∣ Y t ∣ + ∥ Y ∥ p -var ; [ v , t ] ) + ∣ t − v ∣ 1 2 ( ∫ v t ∣ Z r ∣ 2 d r ) 1 2 , (121) \lesssim_{\bar{\Theta}}\lvert t-v\rvert(1+\lvert Y_t\rvert+\lVert Y\rVert_{p\text{-var};[v,t]})+\lvert t-v\rvert^{\frac{1}{2}}\left(\int_v^t\lvert Z_r\rvert^2\,dr\right)^{\frac{1}{2}}, \tag*{(121)} ≲ Θ ˉ ∣ t − v ∣ ( 1 + ∣ Y t ∣ + ∥ Y ∥ p -var ; [ v , t ] ) + ∣ t − v ∣ 2 1 ( ∫ v t ∣ Z r ∣ 2 d r ) 2 1 , (121)
and by (3.46) in Lemma 3.14 (letting p ′ = p q ′ / q p'=pq'/q p ′ = p q ′ / q therein so that γ − 2 ≥ p / p ′ \gamma-2\geq p/p' γ − 2 ≥ p / p ′ ),
∥ ∥ δ I X ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ ˉ ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) . \left\lVert\left\lVert\delta I^X\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}
\lesssim_{k,\gamma,\bar{\Theta}}\lvert\delta X\rvert_{p\text{-var};[s,t]}
\left(1+\left\lVert(Y,Y')\right\rVert_{X;q,q';k,\infty;[s,t]}^{\gamma^{-1}}\right). δ I X p -var ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ ˉ ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) .
Hence, since ∥ ξ ∥ 2 + ∥ δ ( I + I X ) v , t ∥ 2 ≤ ∥ ξ ∥ 2 + ∥ δ I ∥ 1 -var ; [ s , t ] + ∥ δ I X ∥ p -var ; [ s , t ] \lVert\xi\rVert_2+\left\lVert\delta(I+I^X)_{v,t}\right\rVert_2\leq\lVert\xi\rVert_2+\left\lVert\delta I\right\rVert_{1\text{-var};[s,t]}+\left\lVert\delta I^X\right\rVert_{p\text{-var};[s,t]} ∥ ξ ∥ 2 + δ ( I + I X ) v , t 2 ≤ ∥ ξ ∥ 2 + ∥ δ I ∥ 1 -var ; [ s , t ] + δ I X p -var ; [ s , t ] , (120) yields
∥ ξ ∥ 2 + sup v ∈ [ s , t ] ∥ δ I v , t + δ I v , t X ∥ 2 < ∞ . \lVert\xi\rVert_2+\sup_{v\in[s,t]}\left\lVert\delta I_{v,t}+\delta I^X_{v,t}\right\rVert_2<\infty. ∥ ξ ∥ 2 + v ∈ [ s , t ] sup δ I v , t + δ I v , t X 2 < ∞.
Then, by the martingale representation theorem for L 2 L^2 L 2 -integrable random variable (see, e.g., [17 ] , Problem 4.17), there exists a unique progressively measurable process Z ~ \widetilde{Z} Z such that a.s.,
∫ v t Z ~ r d W r = ξ + δ I s , t + δ I s , t X − E v [ ξ + δ I s , t + δ I s , t X ] \int_v^t\widetilde{Z}_r\,dW_r=\xi+\delta I_{s,t}+\delta I^X_{s,t}-\mathbb{E}_v\left[\xi+\delta I_{s,t}+\delta I^X_{s,t}\right] ∫ v t Z r d W r = ξ + δ I s , t + δ I s , t X − E v [ ξ + δ I s , t + δ I s , t X ]
for all v ∈ [ s , t ] v\in[s,t] v ∈ [ s , t ] . Let
Y ~ v = E v [ ξ + δ I v , t + δ I v , t X ] , v ∈ [ s , t ] . (122) \widetilde{Y}_v=\mathbb{E}_v\left[\xi+\delta I_{v,t}+\delta I^X_{v,t}\right],\qquad v\in[s,t]. \tag*{(122)} Y v = E v [ ξ + δ I v , t + δ I v , t X ] , v ∈ [ s , t ] . (122)
Then ( Y ~ , Z ~ ) (\widetilde{Y},\widetilde{Z}) ( Y , Z ) satisfies (118) , and hence as explained under Definition 4.1, one can directly check that ( Y ~ , − H ( Y ) ) ∈ R S M X ( p , p ) -var ( [ s , t ] , R ) (\widetilde{Y},-H(Y))\in\mathrm{RSM}_X^{(p,p)\text{-var}}([s,t],\mathbb{R}) ( Y , − H ( Y )) ∈ RSM X ( p , p ) -var ([ s , t ] , R ) with δ M s , t Y ~ = ∫ s t Z ~ r d W r \delta M_{s,t}^{\widetilde{Y}}=\int_s^t\widetilde{Z}_r\,dW_r δ M s , t Y = ∫ s t Z r d W r .
Now we prove (119) . By the lower bound of the BDG inequality and the estimate (120) , we get
∥ δ M Y ~ ∥ B M O ; [ s , t ] ≲ sup r ∈ [ s , t ] ∥ sup v ∈ [ r , t ] ∣ δ M r , v Y ~ ∣ ∥ F r ; 2 , ∞ \left\lVert\delta M^{\widetilde{Y}}\right\rVert_{\mathrm{BMO};[s,t]}
\lesssim\sup_{r\in[s,t]}\left\lVert\sup_{v\in[r,t]}\lvert\delta M_{r,v}^{\widetilde{Y}}\rvert\right\rVert_{\mathcal{F}_r;2,\infty} δ M Y BMO ; [ s , t ] ≲ r ∈ [ s , t ] sup v ∈ [ r , t ] sup ∣ δ M r , v Y ∣ F r ; 2 , ∞
≲ k ∥ ξ ∥ ∞ + ∥ ∥ δ I ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ I X ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] \lesssim_k\lVert\xi\rVert_\infty+\left\lVert\left\lVert\delta I\right\rVert_{1\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}+\left\lVert\left\lVert\delta I^X\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]} ≲ k ∥ ξ ∥ ∞ + ∥ ∥ δ I ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] + δ I X p -var ∞ ; k , ∞ ; [ s , t ]
≲ k , γ , Θ ˉ ∥ ξ ∥ ∞ + ∥ Y t ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ∣ t − s ∣ 1 2 ) ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) , (123) \lesssim_{k,\gamma,\bar{\Theta}}\lVert\xi\rVert_\infty+\lVert Y_t\rVert_\infty+\left(\lvert\delta X\rvert_{p\text{-var};[s,t]}+\lvert t-s\rvert^{\frac{1}{2}}\right)\left(1+\left\lVert(Y,Y')\right\rVert_{X;q,q';k,\infty;[s,t]}^{\gamma^{-1}}\right), \tag*{(123)} ≲ k , γ , Θ ˉ ∥ ξ ∥ ∞ + ∥ Y t ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ∣ t − s ∣ 2 1 ) ( 1 + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) , (123)
where the second inequality is due to Doob’s maximum inequality so that
∥ sup v ∈ [ r , t ] ∣ δ M r , v Y ~ ∣ | F r ∥ 2 , ∞ ≲ ∥ δ M r , t Y ~ | F r ∥ 2 , ∞ = ∥ δ ( Y ~ + I + I X ) r , t | F r ∥ 2 , ∞ . \left\lVert\sup_{v\in[r,t]} \left|\delta M^{\widetilde{Y}}_{r,v}\right| \middle| \mathcal{F}_r \right\rVert_{2,\infty}
\lesssim\left\lVert\delta M^{\widetilde{Y}}_{r,t} \middle| \mathcal{F}_r \right\rVert_{2,\infty}
= \left\lVert\delta(\widetilde{Y}+\mathcal{I}+\mathcal{I}^{X})_{r,t} \middle| \mathcal{F}_r \right\rVert_{2,\infty}. v ∈ [ r , t ] sup δ M r , v Y F r 2 , ∞ ≲ δ M r , t Y F r 2 , ∞ = δ ( Y + I + I X ) r , t F r 2 , ∞ .
and by the equation δ Y ~ r , t = ξ − E r [ ξ ] − E r [ δ ( I + I X ) r , t ] \delta\widetilde{Y}_{r,t}=\xi-\mathbb{E}_{r}[\xi]-\mathbb{E}_{r}[\delta(\mathcal{I}+\mathcal{I}^{X})_{r,t}] δ Y r , t = ξ − E r [ ξ ] − E r [ δ ( I + I X ) r , t ] and Lemma 3.2 so that
∥ δ ( Y ~ + I + I X ) r , t | F r ∥ 2 , ∞ ≲ k ∥ ξ ∥ ∞ + ∥ ∥ δ I ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ I X ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] . \left\lVert\delta(\widetilde{Y}+\mathcal{I}+\mathcal{I}^{X})_{r,t} \middle| \mathcal{F}_r \right\rVert_{2,\infty}
\lesssim_{k}\|\xi\|_{\infty}+\left\lVert\!\left\lVert\delta\mathcal{I}\right\rVert_{1\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}+\left\lVert\!\left\lVert\delta\mathcal{I}^{X}\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}. δ ( Y + I + I X ) r , t F r 2 , ∞ ≲ k ∥ ξ ∥ ∞ + ∥ ∥ δ I ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] + δ I X p -var ∞ ; k , ∞ ; [ s , t ] .
Furthermore, by Proposition 3.6 and Lemma 3.14,
∥ ∥ R I X ∥ p 2 -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ ‾ ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) . \left\lVert\!\left\lVert R\mathcal{I}^{X}\right\rVert_{\frac{p}{2}\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}
\lesssim_{k,\gamma,\overline{\Theta}}|\delta X|_{p\text{-var};[s,t]}\left(1+\lvert(Y,Y')\rvert_{X;q,q';k,\infty;[s,t]}^{\gamma-1}\right). R I X 2 p -var ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ ∣ δ X ∣ p -var ; [ s , t ] ( 1 + ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) .
Consequently, noting P ~ Y = − R I X − δ I \widetilde{P}^{Y}=-R\mathcal{I}^{X}-\delta\mathcal{I} P Y = − R I X − δ I , the above estimate together with (121) implies
∥ ∥ P ~ Y ∥ p 2 -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ ‾ ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ( 1 + ∥ Y t ∥ ∞ + ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) . (124) \left\lVert\!\left\lVert\widetilde{P}^{Y}\right\rVert_{\frac{p}{2}\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}
\lesssim_{k,\gamma,\overline{\Theta}}\left(|\delta X|_{p\text{-var};[s,t]}+(t-s)^{\frac{1}{2}}\right)\left(1+\|Y_t\|_{\infty}+\lvert(Y,Y')\rvert_{X;q,q';k,\infty;[s,t]}^{\gamma-1}\right). \tag*{(124)} P Y 2 p -var ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) ( 1 + ∥ Y t ∥ ∞ + ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) . (124)
Moreover, since ∥ ∥ Y ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] \left\lVert\!\left\lVert Y\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]} ∥ Y ∥ p -var ∞ ; k , ∞ ; [ s , t ] is dominated by ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] \lvert(Y,Y')\rvert_{X;q,q';k,\infty;[s,t]} ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] (see Lemma 3.13), by the standard interpolation argument (noting p ≤ q ′ p\leq q' p ≤ q ′ ) we get
∥ H ( Y ) ∥ ∞ + ∥ ∥ H ( Y ) ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ ∥ H ∥ ∞ + ∥ H ∥ ∞ ( q ′ − p ) / q ′ ∥ D H ∥ ∞ p / q ′ ∥ ∥ Y ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] p / q ′ ≲ Θ ‾ 1 + ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] p / q ′ . (125) \|H(Y)\|_{\infty}+\left\lVert\!\left\lVert H(Y)\right\rVert_{q'\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}
\lesssim\|H\|_{\infty}+\|H\|_{\infty}^{(q'-p)/q'}\|DH\|_{\infty}^{p/q'}\left\lVert\!\left\lVert Y\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}^{p/q'}
\lesssim_{\overline{\Theta}}1+\lvert(Y,Y')\rvert_{X;q,q';k,\infty;[s,t]}^{p/q'}. \tag*{(125)} ∥ H ( Y ) ∥ ∞ + ∥ H ( Y ) ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] ≲ ∥ H ∥ ∞ + ∥ H ∥ ∞ ( q ′ − p ) / q ′ ∥ D H ∥ ∞ p / q ′ ∥ Y ∥ p -var ∞ ; k , ∞ ; [ s , t ] p / q ′ ≲ Θ 1 + ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] p / q ′ . (125)
Then the desired estimate (119) follows by combining (123) , (124) , and (125) .
Finally, Eq. (118) yields that Y ~ v \widetilde{Y}_v Y v must be given by (122) , and the uniqueness follows. □
The solution map Φ [ s , t ] ξ \Phi^{\xi}_{[s,t]} Φ [ s , t ] ξ then can be well-defined as follows.
Definition 4.14. Under the same assumptions as in Lemma 4.13, define the solution map
Φ [ s , t ] ξ ( Y , Y ′ , Z ) : = ( Y ~ , − H ( Y ) , Z ~ ) . \Phi^{\xi}_{[s,t]}(Y,Y',Z):=(\widetilde{Y},-H(Y),\widetilde{Z}). Φ [ s , t ] ξ ( Y , Y ′ , Z ) := ( Y , − H ( Y ) , Z ) .
The following Proposition 4.15 shows that the solution Φ [ s , t ] ξ ( ⋅ ) \Phi^{\xi}_{[s,t]}(\cdot) Φ [ s , t ] ξ ( ⋅ ) is invariant within the ball with respect to the seminorm ∣ ⋅ ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] \lvert\cdot\rvert_{X;q,q';k,\infty;[s,t]} ∣ ⋅ ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] when ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 / 2 |\delta X|_{p\text{-var};[s,t]}+(t-s)^{1/2} ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1/2 is smaller than a specific constant θ \theta θ . In particular, θ \theta θ can be chosen uniformly with respect to the upper bound of ξ \xi ξ .
Proposition 4.15. Assume that ( Y , Y ′ , Z ) (Y,Y',Z) ( Y , Y ′ , Z ) satisfies the same condition as in Lemma 4.13. Assume further that p < q ′ p<q' p < q ′ . Denote by ( Y ~ , − H ( Y ) , Z ~ ) = Φ [ s , t ] ξ ( Y , Y ′ , Z ) (\widetilde{Y},-H(Y),\widetilde{Z})=\Phi^{\xi}_{[s,t]}(Y,Y',Z) ( Y , − H ( Y ) , Z ) = Φ [ s , t ] ξ ( Y , Y ′ , Z ) . Let Θ ‾ \overline{\Theta} Θ be the same as in Lemma 4.13. Then, there exists K > 0 K>0 K > 0 such that for every K ′ ≥ K K'\geq K K ′ ≥ K , there exists θ > 0 \theta>0 θ > 0 such that if ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 / 2 ≤ θ |\delta X|_{p\text{-var};[s,t]}+(t-s)^{1/2}\leq\theta ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1/2 ≤ θ and ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ K ′ \lvert(Y,Y')\rvert_{X;q,q';k,\infty;[s,t]}\leq K' ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ K ′ , it holds that
∣ ( Y ~ , − H ( Y ) ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ K ′ . (126) \lvert(\widetilde{Y},-H(Y))\rvert_{X;q,q';k,\infty;[s,t]}\leq K'. \tag*{(126)} ∣( Y , − H ( Y )) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ K ′ . (126)
In particular, we can choose K : = C ( 1 + ∥ ξ ∥ ∞ ) K:=C(1+\|\xi\|_{\infty}) K := C ( 1 + ∥ ξ ∥ ∞ ) and θ : = 1 2 C − 1 ∣ K ′ ∣ − ( γ − 2 ) \theta:=\frac{1}{2}C^{-1}|K'|^{-(\gamma-2)} θ := 2 1 C − 1 ∣ K ′ ∣ − ( γ − 2 ) , where C C C is a constant depending only on ( k , γ , Θ ‾ ) (k,\gamma,\overline{\Theta}) ( k , γ , Θ ) .
Proof. Noting p < q ′ p<q' p < q ′ , for every ρ > 0 \rho>0 ρ > 0 we have
∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] p / q ′ ≲ ρ 1 + ρ ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] . \lvert(Y,Y')\rvert_{X;q,q';k,\infty;[s,t]}^{p/q'}
\lesssim_{\rho}1+\rho\lvert(Y,Y')\rvert_{X;q,q';k,\infty;[s,t]}. ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] p / q ′ ≲ ρ 1 + ρ ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] .
Thus, by choosing ρ \rho ρ sufficiently small, one can apply (119) in Lemma 4.13 to determine a constant C 1 > 0 C_{1}>0 C 1 > 0 such that
∣ ( Y ~ , − H ( Y ) ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ C 1 ( 1 + ∥ ξ ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 / 2 ) ∣ ( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) . \lvert(\widetilde{Y},-H(Y))\rvert_{X;q,q';k,\infty;[s,t]}
\leq C_{1}\left(1+\|\xi\|_{\infty}+\left(|\delta X|_{p\text{-var};[s,t]}+(t-s)^{1/2}\right)\lvert(Y,Y')\rvert_{X;q,q';k,\infty;[s,t]}^{\gamma-1}\right). ∣( Y , − H ( Y )) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ C 1 ( 1 + ∥ ξ ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1/2 ) ∣( Y , Y ′ ) ∣ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ) .
Set K : = 2 C 1 ( 1 + ∥ ξ ∥ ∞ ) K := 2C_1(1+\|\xi\|_\infty) K := 2 C 1 ( 1 + ∥ ξ ∥ ∞ ) and θ : = 1 2 C 1 − 1 ∣ K ′ ∣ − ( γ − 2 ) \theta:= \frac{1}{2}C_1^{-1}|K'|^{-(\gamma-2)} θ := 2 1 C 1 − 1 ∣ K ′ ∣ − ( γ − 2 ) . If ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ≤ θ (|\delta X|_{p\text{-var};[s,t]}+(t-s)^{\frac{1}{2}})\leq\theta ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) ≤ θ , we see that for K ′ ≥ K K'\geq K K ′ ≥ K the condition (126) implies
C 1 ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ≤ K ′ / 2. C_1\left(|\delta X|_{p\text{-var};[s,t]}+(t-s)^{\frac{1}{2}}\right)\|(Y,Y')\|_{X;q,q';k,\infty;[s,t]}^{\gamma-1}\leq K'/2. C 1 ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] γ − 1 ≤ K ′ /2.
Consequently, noting C 1 ( 1 + ∥ ξ ∥ ∞ ) ≤ K / 2 ≤ K ′ / 2 C_1(1+\|\xi\|_\infty)\leq K/2\leq K'/2 C 1 ( 1 + ∥ ξ ∥ ∞ ) ≤ K /2 ≤ K ′ /2 , it holds ∥ ( Y ~ , − H ( Y ) ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ K ′ \|(\widetilde Y,-H(Y))\|_{X;q,q';k,\infty;[s,t]}\leq K' ∥ ( Y , − H ( Y )) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ K ′ .
The following Proposition 4.16 shows that the twice composition of the solution map is a contraction on small intervals. The proof relies heavily on the equivalence of norms proved in Lemma 3.2.
Proposition 4.16. Assume ( γ , p , q , q ′ ) (\gamma,p,q,q') ( γ , p , q , q ′ ) , ξ ∈ L t ∞ \xi\in L^\infty_t ξ ∈ L t ∞ and ( f ) (f) ( f ) holds. Let k ≥ 1 k\geq1 k ≥ 1 , H ∈ C b γ ( R , R d ) H\in C_b^\gamma(\mathbb{R},\mathbb{R}^d) H ∈ C b γ ( R , R d ) , and X ∈ C p -var ( [ s , t ] , R d ) X\in\mathcal{C}^{p\text{-var}}([s,t],\mathbb{R}^d) X ∈ C p -var ([ s , t ] , R d ) . Let R 0 > 0 R_0>0 R 0 > 0 be a constant satisfying
∥ ξ ∥ ∞ + ∥ H ∥ C b γ + ∣ δ X ∣ p -var ; [ s , t ] ≤ R 0 . \|\xi\|_\infty+\|H\|_{C_b^\gamma}+|\delta X|_{p\text{-var};[s,t]}\leq R_0. ∥ ξ ∥ ∞ + ∥ H ∥ C b γ + ∣ δ X ∣ p -var ; [ s , t ] ≤ R 0 .
Denote Θ 0 : = ( Θ , C f , R 0 ) \Theta_0:=(\Theta,C_f,R_0) Θ 0 := ( Θ , C f , R 0 ) . For ( Y , Y ′ ) , ( Y ‾ , Y ‾ ′ ) ∈ R S M X ( p , p ) -var (Y,Y'),(\overline Y,\overline Y')\in\mathrm{RSM}_X^{(p,p)\text{-var}} ( Y , Y ′ ) , ( Y , Y ′ ) ∈ RSM X ( p , p ) -var with Y t = Y ‾ t Y_t=\overline Y_t Y t = Y t and
∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ ( Y ‾ , Y ‾ ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] < ∞ \|(Y,Y')\|_{X;q,q';k,\infty;[s,t]}+\|(\overline Y,\overline Y')\|_{X;q,q';k,\infty;[s,t]}<\infty ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] < ∞
Let Z Z Z (Z ‾ \overline Z Z resp.) be the unique progressively measurable process with ∥ Z ∥ L 2 ( [ s , t ] ) < ∞ \|Z\|_{L^2([s,t])}<\infty ∥ Z ∥ L 2 ([ s , t ]) < ∞ (∥ Z ‾ ∥ L 2 ( [ s , t ] ) < ∞ \|\overline Z\|_{L^2([s,t])}<\infty ∥ Z ∥ L 2 ([ s , t ]) < ∞ resp.) a.s. such that δ M s , t Y = ∫ s t Z r d W r \delta M^Y_{s,t}=\int_s^t Z_r\,dW_r δ M s , t Y = ∫ s t Z r d W r (δ M s , t Y ‾ = ∫ s t Z ‾ r d W r \delta M^{\overline Y}_{s,t}=\int_s^t\overline Z_r\,dW_r δ M s , t Y = ∫ s t Z r d W r resp.). Denote the i i i -fold composition of Φ [ s , t ] ξ \Phi^\xi_{[s,t]} Φ [ s , t ] ξ with itself by ( Φ [ s , t ] ξ ) ∘ i (\Phi^\xi_{[s,t]})^{\circ i} ( Φ [ s , t ] ξ ) ∘ i and
{ ( Y ( i ) , ( Y ( i ) ) ′ , Z ( i ) ) : = ( Φ [ s , t ] ξ ) ∘ i ( Y , Y ′ , Z ) , ( Y ‾ ( i ) , ( Y ‾ ( i ) ) ′ , Z ‾ ( i ) ) : = ( Φ [ s , t ] ξ ) ∘ i ( Y ‾ , Y ‾ ′ , Z ‾ ) , i = 0 , 1 , 2. \left\{
\begin{aligned}
(Y^{(i)},(Y^{(i)})',Z^{(i)})&:=(\Phi^\xi_{[s,t]})^{\circ i}(Y,Y',Z),\\
(\overline Y^{(i)},(\overline Y^{(i)})',\overline Z^{(i)})&:=(\Phi^\xi_{[s,t]})^{\circ i}(\overline Y,\overline Y',\overline Z),\quad i=0,1,2.
\end{aligned}
\right. ⎩ ⎨ ⎧ ( Y ( i ) , ( Y ( i ) ) ′ , Z ( i ) ) ( Y ( i ) , ( Y ( i ) ) ′ , Z ( i ) ) := ( Φ [ s , t ] ξ ) ∘ i ( Y , Y ′ , Z ) , := ( Φ [ s , t ] ξ ) ∘ i ( Y , Y ′ , Z ) , i = 0 , 1 , 2.
Then, for every K ^ > 0 \widehat K>0 K > 0 and ε ∈ ( 0 , 1 ) \varepsilon\in(0,1) ε ∈ ( 0 , 1 ) , there exists θ ^ > 0 \widehat\theta>0 θ > 0 depending on ( k , γ , Θ 0 , K ^ , ε ) (k,\gamma,\Theta_0,\widehat K,\varepsilon) ( k , γ , Θ 0 , K , ε ) such that if
{ sup i = 0 , 1 , 2 ( ∥ ( Y ( i ) , ( Y ( i ) ) ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ ( Y ‾ ( i ) , ( Y ‾ ( i ) ) ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] ) ≤ K ^ , ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 / 2 ≤ θ ^ , \left\{
\begin{aligned}
\sup_{i=0,1,2}\bigl(&\|(Y^{(i)},(Y^{(i)})')\|_{X;q,q';k,\infty;[s,t]}\\
&+\|(\overline Y^{(i)},(\overline Y^{(i)})')\|_{X;q,q';k,\infty;[s,t]}\bigr)&\leq\widehat K,\\
|\delta X|_{p\text{-var};[s,t]}+(t-s)^{1/2}&\leq\widehat\theta,
\end{aligned}
\right. ⎩ ⎨ ⎧ i = 0 , 1 , 2 sup ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1/2 ∥ ( Y ( i ) , ( Y ( i ) ) ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] + ∥ ( Y ( i ) , ( Y ( i ) ) ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ s , t ] ) ≤ θ , ≤ K ,
the following contraction property for ( Φ [ s , t ] ξ ) ∘ 2 (\Phi^\xi_{[s,t]})^{\circ2} ( Φ [ s , t ] ξ ) ∘ 2 holds:
∥ ( Y ( 2 ) , ( Y ( 2 ) ) ′ ; Y ‾ ( 2 ) , ( Y ‾ ( 2 ) ) ′ ) ∥ X , X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ ( 1 − ε ) ∥ ( Y , Y ′ ; Y ‾ , Y ‾ ′ ) ∥ X , X ; q , q ′ ; k , ∞ ; [ s , t ] . \|(Y^{(2)},(Y^{(2)})';\overline Y^{(2)},(\overline Y^{(2)})')\|_{X,X;q,q';k,\infty;[s,t]}
\leq(1-\varepsilon)\|(Y,Y';\overline Y,\overline Y')\|_{X,X;q,q';k,\infty;[s,t]}. ∥ ( Y ( 2 ) , ( Y ( 2 ) ) ′ ; Y ( 2 ) , ( Y ( 2 ) ) ′ ) ∥ X , X ; q , q ′ ; k , ∞ ; [ s , t ] ≤ ( 1 − ε ) ∥ ( Y , Y ′ ; Y , Y ′ ) ∥ X , X ; q , q ′ ; k , ∞ ; [ s , t ] .
Proof. For simplicity of notations, denote
{ P ( i ) : = P Y ( i ) , I v ( i ) : = ∫ s v H ( Y r ( i ) ) d X r , I v ( i ) : = ∫ s v f ( r , Y r ( i ) , Z r ( i ) ) d r , P ‾ ( i ) : = P Y ‾ ( i ) , I ‾ v ( i ) : = ∫ s v H ( Y ‾ r ( i ) ) d X r , I ‾ v ( i ) : = ∫ s v f ( r , Y ‾ r ( i ) , Z ‾ r ( i ) ) d r , i = 0 , 1 , 2. \left\{
\begin{aligned}
P^{(i)}&:=P^{Y^{(i)}},& I_v^{(i)}&:=\int_s^v H(Y_r^{(i)})\,dX_r,& \mathcal I_v^{(i)}&:=\int_s^v f(r,Y_r^{(i)},Z_r^{(i)})\,dr,\\
\overline P^{(i)}&:=P^{\overline Y^{(i)}},& \overline I_v^{(i)}&:=\int_s^v H(\overline Y_r^{(i)})\,dX_r,& \overline{\mathcal I}_v^{(i)}&:=\int_s^v f(r,\overline Y_r^{(i)},\overline Z_r^{(i)})\,dr,\quad i=0,1,2.
\end{aligned}
\right. ⎩ ⎨ ⎧ P ( i ) P ( i ) := P Y ( i ) , := P Y ( i ) , I v ( i ) I v ( i ) := ∫ s v H ( Y r ( i ) ) d X r , := ∫ s v H ( Y r ( i ) ) d X r , I v ( i ) I v ( i ) := ∫ s v f ( r , Y r ( i ) , Z r ( i ) ) d r , := ∫ s v f ( r , Y r ( i ) , Z r ( i ) ) d r , i = 0 , 1 , 2.
We first estimate P ( i + 1 ) − P ‾ ( i + 1 ) P^{(i+1)}-\overline P^{(i+1)} P ( i + 1 ) − P ( i + 1 ) . On the one hand, by Lemma 3.17, we have for i = 0 , 1 i=0,1 i = 0 , 1 ,
∥ ∥ R ( i ) − R ‾ ( i ) ∥ ∥ p -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ ( I ( i ) − I ‾ ( i ) ) ∥ ∥ p -var ; ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ^ ∣ δ X ∣ p -var ; [ s , t ] ∥ ( Y ( i ) , ( Y ( i ) ) ′ ; Y ‾ ( i ) , ( Y ‾ ( i ) ) ′ ) ∥ X , X ; q , q ′ ; k , ∞ ; [ s , t ] . (127) \begin{aligned}
\bigl\|\!\bigl\|R^{(i)}-\overline R^{(i)}\bigr\|\!\bigr\|_{p\text{-var};\infty;k,\infty;[s,t]}
+\bigl\|\!\bigl\|\delta(I^{(i)}-\overline I^{(i)})\bigr\|\!\bigr\|_{p\text{-var};\infty;k,\infty;[s,t]}
\lesssim_{k,\gamma,\Theta_0,\widehat K}
|\delta X|_{p\text{-var};[s,t]}
\|(Y^{(i)},(Y^{(i)})';\overline Y^{(i)},(\overline Y^{(i)})')\|_{X,X;q,q';k,\infty;[s,t]}. \tag*{(127)}
\end{aligned} R ( i ) − R ( i ) p -var ; ∞ ; k , ∞ ; [ s , t ] + δ ( I ( i ) − I ( i ) ) p -var ; ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ∣ δ X ∣ p -var ; [ s , t ] ∥ ( Y ( i ) , ( Y ( i ) ) ′ ; Y ( i ) , ( Y ( i ) ) ′ ) ∥ X , X ; q , q ′ ; k , ∞ ; [ s , t ] . (127)
On the other hand, a similar estimate also holds for I ( i ) − I ‾ ( i ) \mathcal I^{(i)}-\overline{\mathcal I}^{(i)} I ( i ) − I ( i ) . Indeed, since ( Y ( i ) − Y ‾ ( i ) ) t = 0 (Y^{(i)}-\overline Y^{(i)})_t=0 ( Y ( i ) − Y ( i ) ) t = 0 we have sup r ∈ [ u , t ] ∣ Y r ( i ) − Y ‾ r ( i ) ∣ ≤ ∥ Y ( i ) − Y ‾ ( i ) ∥ p -var ; [ u , t ] \sup_{r\in[u,t]}|Y_r^{(i)}-\overline Y_r^{(i)}|\leq\|Y^{(i)}-\overline Y^{(i)}\|_{p\text{-var};[u,t]} sup r ∈ [ u , t ] ∣ Y r ( i ) − Y r ( i ) ∣ ≤ ∥ Y ( i ) − Y ( i ) ∥ p -var ; [ u , t ] for u ∈ [ s , t ] u\in[s,t] u ∈ [ s , t ] ; and by Lemma 3.13 we have
∥ ∥ Y ( i ) − Y ‾ ( i ) ∥ ∥ p -var ; ∞ ; k , ∞ ; [ s , t ] ≤ ∥ ( Y ( i ) , ( Y ( i ) ) ′ ; Y ‾ ( i ) , ( Y ‾ ( i ) ) ′ ) ∥ X , X ; q , q ′ ; k , ∞ ; [ s , t ] . \bigl\|\!\bigl\|Y^{(i)}-\overline Y^{(i)}\bigr\|\!\bigr\|_{p\text{-var};\infty;k,\infty;[s,t]}
\leq\|(Y^{(i)},(Y^{(i)})';\overline Y^{(i)},(\overline Y^{(i)})')\|_{X,X;q,q';k,\infty;[s,t]}. Y ( i ) − Y ( i ) p -var ; ∞ ; k , ∞ ; [ s , t ] ≤ ∥ ( Y ( i ) , ( Y ( i ) ) ′ ; Y ( i ) , ( Y ( i ) ) ′ ) ∥ X , X ; q , q ′ ; k , ∞ ; [ s , t ] .
Thus, according to inequalities ∣ f ( r , y , z ) − f ( r , y ˉ , z ˉ ) ∣ ≤ C f ( ∣ y − y ˉ ∣ + ∣ z − z ˉ ∣ ) |f(r,y,z)-f(r,\bar{y},\bar{z})|\leq C_f(|y-\bar{y}|+|z-\bar{z}|) ∣ f ( r , y , z ) − f ( r , y ˉ , z ˉ ) ∣ ≤ C f ( ∣ y − y ˉ ∣ + ∣ z − z ˉ ∣ ) and ∫ u t ( ∣ Y r ( i ) − Y ˉ r ( i ) ∣ + ∣ Z r ( i ) − Z ˉ r ( i ) ∣ ) d r ≤ ( t − s ) ∥ Y ( i ) − Y ˉ ( i ) ∥ p -var ; [ u , t ] + ( t − s ) 1 / 2 { ∫ u t ∣ Z r ( i ) − Z ˉ r ( i ) ∣ 2 d r } 1 / 2 \int_u^t(|Y_r^{(i)}-\bar{Y}_r^{(i)}|+|Z_r^{(i)}-\bar{Z}_r^{(i)}|)\,\mathrm{d}r\leq(t-s)\|Y^{(i)}-\bar{Y}^{(i)}\|_{p\text{-var};[u,t]}+(t-s)^{1/2}\{\int_u^t|Z_r^{(i)}-\bar{Z}_r^{(i)}|^2\,\mathrm{d}r\}^{1/2} ∫ u t ( ∣ Y r ( i ) − Y ˉ r ( i ) ∣ + ∣ Z r ( i ) − Z ˉ r ( i ) ∣ ) d r ≤ ( t − s ) ∥ Y ( i ) − Y ˉ ( i ) ∥ p -var ; [ u , t ] + ( t − s ) 1/2 { ∫ u t ∣ Z r ( i ) − Z ˉ r ( i ) ∣ 2 d r } 1/2 , it holds that
∥ ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ^ ( t − s ) 1 2 ∣ Y ( i ) , ( Y ( i ) ) ; Y ˉ ( i ) , ( Y ˉ ( i ) ) ∣ X , X ; q , q ′ ; k , ∞ ; [ s , t ] . (128) \bigl\|\bigl\|\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})\bigr\|_{1\text{-var}}\bigr\|_{\infty;k,\infty;[s,t]}
\lesssim_{k,\gamma,\Theta_0,\widehat{K}}(t-s)^{\frac{1}{2}}\bigl|Y^{(i)},(Y^{(i)});\bar{Y}^{(i)},(\bar{Y}^{(i)})\bigr|_{X,X;q,q';k,\infty;[s,t]}. \tag*{(128)} δ ( I ( i ) − I ˉ ( i ) ) 1 -var ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ( t − s ) 2 1 Y ( i ) , ( Y ( i ) ) ; Y ˉ ( i ) , ( Y ˉ ( i ) ) X , X ; q , q ′ ; k , ∞ ; [ s , t ] . (128)
Therefore, noting that P ( i + 1 ) − P ˉ ( i + 1 ) = − R I ( i ) + R I ˉ ( i ) − δ ( I ( i ) − I ˉ ( i ) ) P^{(i+1)}-\bar{P}^{(i+1)}=-R\mathcal{I}^{(i)}+R\bar{\mathcal{I}}^{(i)}-\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)}) P ( i + 1 ) − P ˉ ( i + 1 ) = − R I ( i ) + R I ˉ ( i ) − δ ( I ( i ) − I ˉ ( i ) ) , we have
∥ ∥ P ( i + 1 ) − P ˉ ( i + 1 ) ∥ p 2 -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ^ ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ∣ Y ( i ) , ( Y ( i ) ) ; Y ˉ ( i ) , ( Y ˉ ( i ) ) ∣ X , X ; q , q ′ ; k , ∞ ; [ s , t ] . (129) \bigl\|\bigl\|P^{(i+1)}-\bar{P}^{(i+1)}\bigr\|_{\frac{p}{2}\text{-var}}\bigr\|_{\infty;k,\infty;[s,t]}
\lesssim_{k,\gamma,\Theta_0,\widehat{K}}\left(|\delta X|_{p\text{-var};[s,t]}+(t-s)^{\frac{1}{2}}\right)\bigl|Y^{(i)},(Y^{(i)});\bar{Y}^{(i)},(\bar{Y}^{(i)})\bigr|_{X,X;q,q';k,\infty;[s,t]}. \tag*{(129)} P ( i + 1 ) − P ˉ ( i + 1 ) 2 p -var ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) Y ( i ) , ( Y ( i ) ) ; Y ˉ ( i ) , ( Y ˉ ( i ) ) X , X ; q , q ′ ; k , ∞ ; [ s , t ] . (129)
Denote M ( i ) : = M Y ( i ) M^{(i)}:=M^{Y^{(i)}} M ( i ) := M Y ( i ) and M ˉ ( i ) : = M Y ˉ ( i ) \bar{M}^{(i)}:=M^{\bar{Y}^{(i)}} M ˉ ( i ) := M Y ˉ ( i ) , i = 0 , 1 , 2 i=0,1,2 i = 0 , 1 , 2 . Next, we estimate M ( i + 1 ) − M ˉ ( i + 1 ) M^{(i+1)}-\bar{M}^{(i+1)} M ( i + 1 ) − M ˉ ( i + 1 ) and Y ( i + 1 ) − Y ˉ ( i + 1 ) Y^{(i+1)}-\bar{Y}^{(i+1)} Y ( i + 1 ) − Y ˉ ( i + 1 ) for i = 0 , 1 i=0,1 i = 0 , 1 .
We first estimate M ( i + 1 ) − M ˉ ( i + 1 ) M^{(i+1)}-\bar{M}^{(i+1)} M ( i + 1 ) − M ˉ ( i + 1 ) . Since for r ∈ [ s , t ] r\in[s,t] r ∈ [ s , t ] it holds ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) r = E r [ δ ( I ( i ) − I ˉ ( i ) ) r , t + δ ( I ( i ) − I ˉ ( i ) ) r , t ] (Y^{(i+1)}-\bar{Y}^{(i+1)})_r=\mathbb{E}_r[\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})_{r,t}+\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})_{r,t}] ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) r = E r [ δ ( I ( i ) − I ˉ ( i ) ) r , t + δ ( I ( i ) − I ˉ ( i ) ) r , t ] and Y t ( i ) − Y ˉ t ( i ) = 0 Y_t^{(i)}-\bar{Y}_t^{(i)}=0 Y t ( i ) − Y ˉ t ( i ) = 0 , we have
δ ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) r , t = E r [ δ ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) r , t ] = − E r [ δ ( I ( i ) − I ˉ ( i ) ) r , t + δ ( I ( i ) − I ˉ ( i ) ) r , t ] . \delta(Y^{(i+1)}-\bar{Y}^{(i+1)})_{r,t}
=\mathbb{E}_r[\delta(Y^{(i+1)}-\bar{Y}^{(i+1)})_{r,t}]
=-\mathbb{E}_r\left[\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})_{r,t}+\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})_{r,t}\right]. δ ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) r , t = E r [ δ ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) r , t ] = − E r [ δ ( I ( i ) − I ˉ ( i ) ) r , t + δ ( I ( i ) − I ˉ ( i ) ) r , t ] .
Hence, it holds that
δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) r , t = δ ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) r , t + δ ( I ( i ) − I ˉ ( i ) ) r , t + δ ( I ( i ) − I ˉ ( i ) ) r , t = − E r [ δ ( I ( i ) − I ˉ ( i ) + I ( i ) − I ˉ ( i ) ) r , t ] + δ ( I ( i ) − I ˉ ( i ) + I ( i ) − I ˉ ( i ) ) r , t . \begin{aligned}
\delta(M^{(i+1)}-\bar{M}^{(i+1)})_{r,t}
&=\delta(Y^{(i+1)}-\bar{Y}^{(i+1)})_{r,t}+\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})_{r,t}+\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})_{r,t}\\
&=-\mathbb{E}_r\left[\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)}+\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})_{r,t}\right]+\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)}+\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})_{r,t}.
\end{aligned} δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) r , t = δ ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) r , t + δ ( I ( i ) − I ˉ ( i ) ) r , t + δ ( I ( i ) − I ˉ ( i ) ) r , t = − E r [ δ ( I ( i ) − I ˉ ( i ) + I ( i ) − I ˉ ( i ) ) r , t ] + δ ( I ( i ) − I ˉ ( i ) + I ( i ) − I ˉ ( i ) ) r , t .
This, together with Lemma 3.2 yields that
∥ ∥ δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k ∥ ∥ δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) ∥ B M O ∥ ≲ ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ ∞ ; 2 , ∞ ; [ s , t ] + ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ ∞ ; 2 , ∞ ; [ s , t ] . (130) \bigl\|\bigl\|\delta(M^{(i+1)}-\bar{M}^{(i+1)})\bigr\|_{p\text{-var}}\bigr\|_{\infty;k,\infty;[s,t]}
\lesssim_k \bigl\|\bigl\|\delta(M^{(i+1)}-\bar{M}^{(i+1)})\bigr\|_{\mathrm{BMO}}\bigr\|
\lesssim\bigl\|\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})\bigr\|_{\infty;2,\infty;[s,t]}+\bigl\|\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})\bigr\|_{\infty;2,\infty;[s,t]}. \tag*{(130)} δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) p -var ∞ ; k , ∞ ; [ s , t ] ≲ k δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) BMO ≲ δ ( I ( i ) − I ˉ ( i ) ) ∞ ; 2 , ∞ ; [ s , t ] + δ ( I ( i ) − I ˉ ( i ) ) ∞ ; 2 , ∞ ; [ s , t ] . (130)
where the last inequality is due to the BDG inequality and Doob’s maximum inequality:
∥ δ ⟨ M ( i + 1 ) − M ˉ ( i + 1 ) ⟩ r , t 1 / 2 ∥ 2 ∣ F r ≲ ∥ sup v ∈ [ r , t ] ∣ δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) r , v ∣ ∣ F r ∥ 2 \bigl\|\delta\langle M^{(i+1)}-\bar{M}^{(i+1)}\rangle_{r,t}^{1/2}\bigr\|_2\bigm|\mathcal{F}_r
\lesssim\bigl\|\sup_{v\in[r,t]}|\delta(M^{(i+1)}-\bar{M}^{(i+1)})_{r,v}|\bigm|\mathcal{F}_r\bigr\|_2 δ ⟨ M ( i + 1 ) − M ˉ ( i + 1 ) ⟩ r , t 1/2 2 F r ≲ v ∈ [ r , t ] sup ∣ δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) r , v ∣ F r 2
≲ ∥ δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) r , t ∣ F r ∥ 2 \lesssim\bigl\|\delta(M^{(i+1)}-\bar{M}^{(i+1)})_{r,t}\bigm|\mathcal{F}_r\bigr\|_2 ≲ δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) r , t F r 2
≲ ∥ δ ( I ( i ) − I ˉ ( i ) + I ( i ) − I ˉ ( i ) ) r , t ∣ F r ∥ 2 . \lesssim\bigl\|\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)}+\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})_{r,t}\bigm|\mathcal{F}_r\bigr\|_2. ≲ δ ( I ( i ) − I ˉ ( i ) + I ( i ) − I ˉ ( i ) ) r , t F r 2 .
In the RHS of (130) , note that ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ ∞ ; 2 , ∞ ; [ s , t ] \|\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})\|_{\infty;2,\infty;[s,t]} ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ ∞ ; 2 , ∞ ; [ s , t ] plus ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ ∞ ; 2 , ∞ ; [ s , t ] \|\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})\|_{\infty;2,\infty;[s,t]} ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ ∞ ; 2 , ∞ ; [ s , t ] is bounded by ∥ ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ p -var ∥ ∞ ; 2 , ∞ ; [ s , t ] \|\|\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})\|_{p\text{-var}}\|_{\infty;2,\infty;[s,t]} ∥∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ p -var ∥ ∞ ; 2 , ∞ ; [ s , t ] plus ∥ ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ 1 -var ∥ ∞ ; 2 , ∞ ; [ s , t ] \|\|\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})\|_{1\text{-var}}\|_{\infty;2,\infty;[s,t]} ∥∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ 1 -var ∥ ∞ ; 2 , ∞ ; [ s , t ] . Furthermore, by the equivalence of norms proved in Lemma 3.2, the latter one is dominated by ∥ ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] \|\|\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})\|_{p\text{-var}}\|_{\infty;k,\infty;[s,t]} ∥∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] plus ∥ ∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] \|\|\delta(\mathcal{I}^{(i)}-\bar{\mathcal{I}}^{(i)})\|_{1\text{-var}}\|_{\infty;k,\infty;[s,t]} ∥∥ δ ( I ( i ) − I ˉ ( i ) ) ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] . Then, by (130) and estimates (127) and (128) , we get
∥ ∥ δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) ∥ B M O ; [ s , t ] ≲ k , γ , Θ 0 , K ^ ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ∣ Y ( i ) , ( Y ( i ) ) ; Y ˉ ( i ) , ( Y ˉ ( i ) ) ∣ X , X ; q , q ′ ; k , ∞ ; [ s , t ] . (131) \begin{aligned}
&\bigl\|\bigl\|\delta(M^{(i+1)}-\bar{M}^{(i+1)})\bigr\|_{p\text{-var}}\bigr\|_{\infty;k,\infty;[s,t]}+\bigl\|\delta(M^{(i+1)}-\bar{M}^{(i+1)})\bigr\|_{\mathrm{BMO};[s,t]}\\
&\qquad\lesssim_{k,\gamma,\Theta_0,\widehat{K}}\left(|\delta X|_{p\text{-var};[s,t]}+(t-s)^{\frac{1}{2}}\right)\bigl|Y^{(i)},(Y^{(i)});\bar{Y}^{(i)},(\bar{Y}^{(i)})\bigr|_{X,X;q,q';k,\infty;[s,t]}. \tag*{(131)}
\end{aligned} δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) p -var ∞ ; k , ∞ ; [ s , t ] + δ ( M ( i + 1 ) − M ˉ ( i + 1 ) ) BMO ; [ s , t ] ≲ k , γ , Θ 0 , K ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) Y ( i ) , ( Y ( i ) ) ; Y ˉ ( i ) , ( Y ˉ ( i ) ) X , X ; q , q ′ ; k , ∞ ; [ s , t ] . (131)
Note δ ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) = δ ( − I ( i ) + I ˉ ( i ) − I ( i ) + I ˉ ( i ) + M ( i + 1 ) − M ˉ ( i + 1 ) ) \delta(Y^{(i+1)}-\bar{Y}^{(i+1)})=\delta(-\mathcal{I}^{(i)}+\bar{\mathcal{I}}^{(i)}-\mathcal{I}^{(i)}+\bar{\mathcal{I}}^{(i)}+M^{(i+1)}-\bar{M}^{(i+1)}) δ ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) = δ ( − I ( i ) + I ˉ ( i ) − I ( i ) + I ˉ ( i ) + M ( i + 1 ) − M ˉ ( i + 1 ) ) . The above estimate together with (127) and (128) yields
∥ ∥ δ ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ^ ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ∣ Y ( i ) , ( Y ( i ) ) ; Y ˉ ( i ) , ( Y ˉ ( i ) ) ∣ X , X ; q , q ′ ; k , ∞ ; [ s , t ] . (132) \bigl\|\bigl\|\delta(Y^{(i+1)}-\bar{Y}^{(i+1)})\bigr\|_{p\text{-var}}\bigr\|_{\infty;k,\infty;[s,t]}
\lesssim_{k,\gamma,\Theta_0,\widehat{K}}\left(|\delta X|_{p\text{-var};[s,t]}+(t-s)^{\frac{1}{2}}\right)\bigl|Y^{(i)},(Y^{(i)});\bar{Y}^{(i)},(\bar{Y}^{(i)})\bigr|_{X,X;q,q';k,\infty;[s,t]}. \tag*{(132)} δ ( Y ( i + 1 ) − Y ˉ ( i + 1 ) ) p -var ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) Y ( i ) , ( Y ( i ) ) ; Y ˉ ( i ) , ( Y ˉ ( i ) ) X , X ; q , q ′ ; k , ∞ ; [ s , t ] . (132)
For ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ (Y^{(i+1)})'-(\bar{Y}^{(i+1)})' ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ , since ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ = − H ( Y ( i ) ) + H ( Y ˉ ( i ) ) (Y^{(i+1)})'-(\bar{Y}^{(i+1)})'=-H(Y^{(i)})+H(\bar{Y}^{(i)}) ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ = − H ( Y ( i ) ) + H ( Y ˉ ( i ) ) , by Lemma A.5 (noting H ∈ C 0 γ H\in C^\gamma_0 H ∈ C 0 γ so that H ∈ C 0 2 H\in C^2_0 H ∈ C 0 2 ) and the inequality ∣ ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ ∣ ≲ Θ 0 ∣ Y ( i ) − Y ˉ ( i ) ∣ \lvert(Y^{(i+1)})'-(\bar{Y}^{(i+1)})'\rvert\lesssim_{\Theta_0}\lvert Y^{(i)}-\bar{Y}^{(i)}\rvert ∣( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ ∣ ≲ Θ 0 ∣ Y ( i ) − Y ˉ ( i ) ∣ we have
∥ ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ δ ( ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ ) ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] ≲ Θ 0 ( 1 + ∥ δ Y ( i ) ∥ q -var ∞ ; k , ∞ ; [ s , t ] + ∥ δ Y ˉ ( i ) ∥ q -var ∞ ; k , ∞ ; [ s , t ] ) ∥ Y ( i ) − Y ˉ ( i ) ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y ( i ) − Y ˉ ( i ) ) ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] . (133) \begin{aligned}
\lVert(Y^{(i+1)})'-(\bar{Y}^{(i+1)})'\rVert_{L^\infty([s,t]\times\Omega)}
&+\lVert\delta((Y^{(i+1)})'-(\bar{Y}^{(i+1)})')\rVert_{q'\text{-var}_{\infty;k,\infty;[s,t]}}\\
&\lesssim_{\Theta_0}\left(1+\lVert\delta Y^{(i)}\rVert_{q\text{-var}_{\infty;k,\infty;[s,t]}}+\lVert\delta\bar{Y}^{(i)}\rVert_{q\text{-var}_{\infty;k,\infty;[s,t]}}\right)\lVert Y^{(i)}-\bar{Y}^{(i)}\rVert_{L^\infty([s,t]\times\Omega)}\\
&\quad+\lVert\lVert\delta(Y^{(i)}-\bar{Y}^{(i)})\rVert_{q'\text{-var}}\rVert_{\infty;k,\infty;[s,t]}. \tag*{(133)}
\end{aligned} ∥( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ δ (( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ ) ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] ≲ Θ 0 ( 1 + ∥ δ Y ( i ) ∥ q -var ∞ ; k , ∞ ; [ s , t ] + ∥ δ Y ˉ ( i ) ∥ q -var ∞ ; k , ∞ ; [ s , t ] ) ∥ Y ( i ) − Y ˉ ( i ) ∥ L ∞ ([ s , t ] × Ω ) + ∥∥ δ ( Y ( i ) − Y ˉ ( i ) ) ∥ q ′ -var ∥ ∞ ; k , ∞ ; [ s , t ] . (133)
For the RHS of the above inequality, by (3.41) in Lemma 3.13 we get
∥ ∥ δ Y ( i ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ Y ˉ ( i ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ Θ 0 , K ^ 1. (134) \lVert\lVert\delta Y^{(i)}\rVert_{p\text{-var}}\rVert_{\infty;k,\infty;[s,t]}+\lVert\lVert\delta\bar{Y}^{(i)}\rVert_{p\text{-var}}\rVert_{\infty;k,\infty;[s,t]}\lesssim_{\Theta_0,\widehat K}1. \tag*{(134)} ∥∥ δ Y ( i ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥∥ δ Y ˉ ( i ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ Θ 0 , K 1. (134)
Note Y t ( i ) = Y ˉ t ( i ) Y_t^{(i)}=\bar{Y}_t^{(i)} Y t ( i ) = Y ˉ t ( i ) . Then, combining (4.52), (4.53), and (3.43) in Lemma 3.13, we have
∥ ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ δ ( ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ ) ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 0 , K ^ ∥ Y ( i ) , ( Y ( i ) ) ′ ; Y ˉ ( i ) , ( Y ˉ ( i ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] . \begin{aligned}
\lVert(Y^{(i+1)})'-(\bar{Y}^{(i+1)})'\rVert_{L^\infty([s,t]\times\Omega)}
&+\lVert\delta((Y^{(i+1)})'-(\bar{Y}^{(i+1)})')\rVert_{q'\text{-var}_{\infty;k,\infty;[s,t]}}\\
&\lesssim_{\gamma,\Theta_0,\widehat K}\lVert Y^{(i)},(Y^{(i)})';\bar{Y}^{(i)},(\bar{Y}^{(i)})'\rVert_{\mathbb{X},q,q';k,\infty;[s,t]}.
\end{aligned} ∥( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ δ (( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ ) ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 0 , K ∥ Y ( i ) , ( Y ( i ) ) ′ ; Y ˉ ( i ) , ( Y ˉ ( i ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] .
Consequently, by (4.48) and (4.50) we have
∥ Y ( i + 1 ) , ( Y ( i + 1 ) ) ′ ; Y ˉ ( i + 1 ) , ( Y ˉ ( i + 1 ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ^ ∥ Y ( i ) , ( Y ( i ) ) ′ ; Y ˉ ( i ) , ( Y ˉ ( i ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] . (135) \lVert Y^{(i+1)},(Y^{(i+1)})';\bar{Y}^{(i+1)},(\bar{Y}^{(i+1)})'\rVert_{\mathbb{X},q,q';k,\infty;[s,t]}
\lesssim_{k,\gamma,\Theta_0,\widehat K}
\lVert Y^{(i)},(Y^{(i)})';\bar{Y}^{(i)},(\bar{Y}^{(i)})'\rVert_{\mathbb{X},q,q';k,\infty;[s,t]}. \tag*{(135)} ∥ Y ( i + 1 ) , ( Y ( i + 1 ) ) ′ ; Y ˉ ( i + 1 ) , ( Y ˉ ( i + 1 ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ∥ Y ( i ) , ( Y ( i ) ) ′ ; Y ˉ ( i ) , ( Y ˉ ( i ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] . (135)
However, it is not obvious to extract a small interval factor in the above estimate. Nevertheless, we can estimate ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ (Y^{(i+1)})'-(\bar{Y}^{(i+1)})' ( Y ( i + 1 ) ) ′ − ( Y ˉ ( i + 1 ) ) ′ by Y ( i − 1 ) − Y ˉ ( i − 1 ) Y^{(i-1)}-\bar{Y}^{(i-1)} Y ( i − 1 ) − Y ˉ ( i − 1 ) -terms times a small interval factor.
Indeed, since Y t ( 1 ) − Y ˉ t ( 1 ) = 0 Y_t^{(1)}-\bar{Y}_t^{(1)}=0 Y t ( 1 ) − Y ˉ t ( 1 ) = 0 , by the same proof as in Lemma 3.13, it holds that ∥ Y ( 1 ) − Y ˉ ( 1 ) ∥ L ∞ ( [ s , t ] × Ω ) ≤ ∥ ∥ δ ( Y ( 1 ) − Y ˉ ( 1 ) ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] \lVert Y^{(1)}-\bar{Y}^{(1)}\rVert_{L^\infty([s,t]\times\Omega)}\leq\lVert\lVert\delta(Y^{(1)}-\bar{Y}^{(1)})\rVert_{p\text{-var}}\rVert_{\infty;k,\infty;[s,t]} ∥ Y ( 1 ) − Y ˉ ( 1 ) ∥ L ∞ ([ s , t ] × Ω ) ≤ ∥∥ δ ( Y ( 1 ) − Y ˉ ( 1 ) ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] . Then, (4.52) and (4.53) yield, noting p ≤ q ≤ q ′ p\leq q\leq q' p ≤ q ≤ q ′ ,
∥ ( Y ( 2 ) ) ′ − ( Y ˉ ( 2 ) ) ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ δ ( ( Y ( 2 ) ) ′ − ( Y ˉ ( 2 ) ) ′ ) ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 0 , K ^ ∥ ∥ δ ( Y ( 1 ) − Y ˉ ( 1 ) ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] . \begin{aligned}
\lVert(Y^{(2)})'-(\bar{Y}^{(2)})'\rVert_{L^\infty([s,t]\times\Omega)}
&+\lVert\delta((Y^{(2)})'-(\bar{Y}^{(2)})')\rVert_{q'\text{-var}_{\infty;k,\infty;[s,t]}}\\
&\lesssim_{\gamma,\Theta_0,\widehat K}
\lVert\lVert\delta(Y^{(1)}-\bar{Y}^{(1)})\rVert_{p\text{-var}}\rVert_{\infty;k,\infty;[s,t]}.
\end{aligned} ∥( Y ( 2 ) ) ′ − ( Y ˉ ( 2 ) ) ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ δ (( Y ( 2 ) ) ′ − ( Y ˉ ( 2 ) ) ′ ) ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] ≲ γ , Θ 0 , K ∥∥ δ ( Y ( 1 ) − Y ˉ ( 1 ) ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] .
This, together with the estimate of Y ( 1 ) − Y ˉ ( 1 ) Y^{(1)}-\bar{Y}^{(1)} Y ( 1 ) − Y ˉ ( 1 ) shown in (4.51) (letting i = 0 i=0 i = 0 ), yields
∥ ( Y ( 2 ) ) ′ − ( Y ˉ ( 2 ) ) ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ δ ( ( Y ( 2 ) ) ′ − ( Y ˉ ( 2 ) ) ′ ) ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ^ ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] . (136) \begin{aligned}
\lVert(Y^{(2)})'-(\bar{Y}^{(2)})'\rVert_{L^\infty([s,t]\times\Omega)}
&+\lVert\delta((Y^{(2)})'-(\bar{Y}^{(2)})')\rVert_{q'\text{-var}_{\infty;k,\infty;[s,t]}}\\
&\lesssim_{k,\gamma,\Theta_0,\widehat K}
\left(\lvert\delta X\rvert_{p\text{-var};[s,t]}+(t-s)^{\frac12}\right)
\lVert Y,Y';\bar{Y},\bar{Y}'\rVert_{\mathbb{X},q,q';k,\infty;[s,t]}. \tag*{(136)}
\end{aligned} ∥( Y ( 2 ) ) ′ − ( Y ˉ ( 2 ) ) ′ ∥ L ∞ ([ s , t ] × Ω ) + ∥ δ (( Y ( 2 ) ) ′ − ( Y ˉ ( 2 ) ) ′ ) ∥ q ′ -var ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] . (136)
Finally, combining (4.48), (4.50), and (4.55), we obtain that
∥ Y ( 2 ) , ( Y ( 2 ) ) ′ ; Y ˉ ( 2 ) , ( Y ˉ ( 2 ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ^ ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 / 2 ) × ( ∥ Y ( 1 ) , ( Y ( 1 ) ) ′ ; Y ˉ ( 1 ) , ( Y ˉ ( 1 ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] ) . \begin{aligned}
\lVert Y^{(2)},(Y^{(2)})';\bar{Y}^{(2)},(\bar{Y}^{(2)})'\rVert_{\mathbb{X},q,q';k,\infty;[s,t]}
&\lesssim_{k,\gamma,\Theta_0,\widehat K}
\left(\lvert\delta X\rvert_{p\text{-var};[s,t]}+(t-s)^{1/2}\right)\\
&\quad\times\left(\lVert Y^{(1)},(Y^{(1)})';\bar{Y}^{(1)},(\bar{Y}^{(1)})'\rVert_{\mathbb{X},q,q';k,\infty;[s,t]} \right.\\
&\qquad\left. +\lVert Y,Y';\bar{Y},\bar{Y}'\rVert_{\mathbb{X},q,q';k,\infty;[s,t]}\right).
\end{aligned} ∥ Y ( 2 ) , ( Y ( 2 ) ) ′ ; Y ˉ ( 2 ) , ( Y ˉ ( 2 ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 0 , K ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1/2 ) × ( ∥ Y ( 1 ) , ( Y ( 1 ) ) ′ ; Y ˉ ( 1 ) , ( Y ˉ ( 1 ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] ) .
This, together with (4.54) (letting i = 0 i=0 i = 0 ), yields that for some constant C k , γ , Θ 0 , K ^ > 0 C_{k,\gamma,\Theta_0,\widehat K}>0 C k , γ , Θ 0 , K > 0 ,
∥ Y ( 2 ) , ( Y ( 2 ) ) ′ ; Y ˉ ( 2 ) , ( Y ˉ ( 2 ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] ≤ C k , γ , Θ 0 , K ^ ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 / 2 ) × ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] . \begin{aligned}
\lVert Y^{(2)},(Y^{(2)})';\bar{Y}^{(2)},(\bar{Y}^{(2)})'\rVert_{\mathbb{X},q,q';k,\infty;[s,t]}
&\leq C_{k,\gamma,\Theta_0,\widehat K}\left(\lvert\delta X\rvert_{p\text{-var};[s,t]}+(t-s)^{1/2}\right)\\
&\quad\times\lVert Y,Y';\bar{Y},\bar{Y}'\rVert_{\mathbb{X},q,q';k,\infty;[s,t]}.
\end{aligned} ∥ Y ( 2 ) , ( Y ( 2 ) ) ′ ; Y ˉ ( 2 ) , ( Y ˉ ( 2 ) ) ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] ≤ C k , γ , Θ 0 , K ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1/2 ) × ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , q , q ′ ; k , ∞ ; [ s , t ] .
Setting θ ^ : = ( 1 − ε ) C k , γ , Θ 0 , K ^ − 1 \hat{\theta}:=(1-\varepsilon)C_{k,\gamma,\Theta_0,\widehat K}^{-1} θ ^ := ( 1 − ε ) C k , γ , Θ 0 , K − 1 , we get the desired contraction property. □
In the following Theorem 4.17, we prove the well-posedness of Eq. (1) via Picard’s iteration. By the contraction property of the solution map on small intervals, we extend the solution to the whole interval, via the telescopic argument which is valid due to the a a a priori estimate for the solution established in Theorem 4.9.
Theorem 4.17. Assume ( γ , p , q , q ′ ) (\gamma,p,q,q^{\prime}) ( γ , p , q , q ′ ) , ξ ∈ L ∞ \xi\in L^{\infty} ξ ∈ L ∞ and ( f d e t ) (f_{\mathrm{det}}) ( f det ) holds. Let H ∈ C b γ ( R , R d ) H \in C^{\gamma}_{b}(\mathbb{R},\mathbb{R}^{d}) H ∈ C b γ ( R , R d ) and X ∈ C p -var ( [ 0 , T ] , R d ) X \in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^{d}) X ∈ C p -var ([ 0 , T ] , R d ) . Then, there exists a unique solution ( Y , Z ) (Y,Z) ( Y , Z ) to Eq. (1) with ( Y , − H ( Y ) ) ∈ RSM X ( p , p ) -var (Y,-H(Y)) \in\operatorname{RSM}^{(p,p)\text{-var}}_{X} ( Y , − H ( Y )) ∈ RSM X ( p , p ) -var and ∥ ( Y , − H ( Y ) ) ∥ X ; q , q ′ ; 1 , ∞ < ∞ \lVert(Y,-H(Y))\rVert_{X;q,q^{\prime};1,\infty}<\infty ∥( Y , − H ( Y )) ∥ X ; q , q ′ ; 1 , ∞ < ∞ .
Moreover, for every k ≥ 1 k \ge1 k ≥ 1 ,
∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; k , ∞ < ∞ . (137) \lVert(Y,-H(Y))\rVert_{X;p,p;k,\infty}<\infty. \tag*{(137)} ∥( Y , − H ( Y )) ∥ X ; p , p ; k , ∞ < ∞. (137)
In particular, ∥ Y ∥ ∞ + ∥ δ ( ∫ 0 ⋅ Z r d W r ) ∥ B M O < ∞ \lVert Y\rVert_{\infty}+\lVert\delta(\int_{0}^{\cdot}Z_{r}\,\mathrm{d}W_{r})\rVert_{\mathrm{BMO}}<\infty ∥ Y ∥ ∞ + ∥ δ ( ∫ 0 ⋅ Z r d W r ) ∥ BMO < ∞ .
Proof. Uniqueness is already proved in Corollary 4.6 by the comparison principle, so it suffices to construct a solution satisfying (137) . Since (137) is independent of ( q , q ′ ) (q,q^{\prime}) ( q , q ′ ) , we can assume p < q ′ p<q^{\prime} p < q ′ without loss of generality for the application of Proposition 4.15. In addition, we fix k > 1 k>1 k > 1 , where the condition k > 1 k>1 k > 1 is also required in Theorem 4.9 which will be used later.
For t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] , denote ( Y ( i ) , Z ( i ) ) (Y^{(i)},Z^{(i)}) ( Y ( i ) , Z ( i ) ) by
( Y ( i ) , ( Y ( i ) ) ′ , Z ( i ) ) : = Φ [ t , T ] ξ ( Y ( i − 1 ) , ( Y ( i − 1 ) ) ′ , Z ( i − 1 ) ) , i ≥ 1 ; ( Y ( 0 ) , ( Y ( 0 ) ) ′ , Z ( 0 ) ) : = ( 0 , 0 , 0 ) . (Y^{(i)},(Y^{(i)})^{\prime},Z^{(i)}):=\Phi^{\xi}_{[t,T]}(Y^{(i-1)},(Y^{(i-1)})^{\prime},Z^{(i-1)}),\quad i\ge1;\qquad(Y^{(0)},(Y^{(0)})^{\prime},Z^{(0)}):=(0,0,0). ( Y ( i ) , ( Y ( i ) ) ′ , Z ( i ) ) := Φ [ t , T ] ξ ( Y ( i − 1 ) , ( Y ( i − 1 ) ) ′ , Z ( i − 1 ) ) , i ≥ 1 ; ( Y ( 0 ) , ( Y ( 0 ) ) ′ , Z ( 0 ) ) := ( 0 , 0 , 0 ) .
Our goal is to construct a solution ( Y , Z ) (Y,Z) ( Y , Z ) to (1) by taking the limit of ( Y ( i ) , Z ( i ) ) (Y^{(i)},Z^{(i)}) ( Y ( i ) , Z ( i ) ) . For the reader’s convenience, we divide the construction into three steps.
Step 1. In this step, we prove the convergence of { ( Y ( i ) , ( Y ( i ) ) ′ , Z ( i ) ) } i ≥ 1 \{(Y^{(i)},(Y^{(i)})^{\prime},Z^{(i)})\}_{i\ge1} {( Y ( i ) , ( Y ( i ) ) ′ , Z ( i ) ) } i ≥ 1 on [ t , T ] [t,T] [ t , T ] , assuming ∣ δ X ∣ p -var ; [ t , T ] + ( T − t ) 1 / 2 |\delta X|_{p\text{-var};[t,T]}+(T-t)^{1/2} ∣ δ X ∣ p -var ; [ t , T ] + ( T − t ) 1/2 is small enough.
By the a priori estimate established in Theorem 4.9, there exists a constant L L L such that for any time t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] and any solution ( Y ^ , Z ^ ) (\widehat{Y},\widehat{Z}) ( Y , Z ) on [ t , T ] [t,T] [ t , T ] with ∥ ( Y ^ , − H ( Y ^ ) ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] < ∞ \lVert(\widehat{Y},-H(\widehat{Y}))\rVert_{X;q,q^{\prime};k,\infty;[t,T]}<\infty ∥( Y , − H ( Y )) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] < ∞ , it holds
sup r ∈ [ t , T ] ∥ Y ^ r ∥ ∞ ≤ L . (138) \sup_{r\in[t,T]}\lVert\widehat{Y}_{r}\rVert_{\infty}\le L. \tag*{(138)} r ∈ [ t , T ] sup ∥ Y r ∥ ∞ ≤ L . (138)
In particular, ∥ ξ ∥ ∞ ≤ L \lVert\xi\rVert_{\infty}\le L ∥ ξ ∥ ∞ ≤ L . Let C > 0 C>0 C > 0 be the constant obtained in Proposition 4.15 (where p < q ′ p<q^{\prime} p < q ′ is needed, and C C C is independent of ξ \xi ξ ). Denote K 1 K_{1} K 1 as follows:
K 1 : = C ( 1 + L ) . K_{1}:=C(1+L). K 1 := C ( 1 + L ) .
Noting K 1 ≥ C ( 1 + ∥ ξ ∥ ∞ ) K_{1}\ge C(1+\lVert\xi\rVert_{\infty}) K 1 ≥ C ( 1 + ∥ ξ ∥ ∞ ) , Proposition 4.15 yields that if assuming further
∣ δ X ∣ p -var ; [ t , T ] + ( T − t ) 1 / 2 ≤ C − 1 K 1 − ( γ − 2 ) / 2 = : θ 1 , (139) |\delta X|_{p\text{-var};[t,T]}+(T-t)^{1/2}\le C^{-1}K_{1}^{-(\gamma-2)/2}=:\theta_{1}, \tag*{(139)} ∣ δ X ∣ p -var ; [ t , T ] + ( T − t ) 1/2 ≤ C − 1 K 1 − ( γ − 2 ) /2 =: θ 1 , (139)
then for ( ψ ~ , ψ ~ ′ , z ~ ) : = Φ [ t , T ] ξ ( ψ , ψ ′ , z ) (\widetilde{\psi},\widetilde{\psi}^{\prime},\widetilde{z}):=\Phi^{\xi}_{[t,T]}(\psi,\psi^{\prime},z) ( ψ , ψ ′ , z ) := Φ [ t , T ] ξ ( ψ , ψ ′ , z ) it holds:
∥ ( ψ ~ , ψ ~ ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] ≤ K 1 whenever ∥ ( ψ , ψ ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] ≤ K 1 . (140) \lVert(\widetilde{\psi},\widetilde{\psi}^{\prime})\rVert_{X;q,q^{\prime};k,\infty;[t,T]}\le K_{1}\quad\text{whenever}\quad\lVert(\psi,\psi^{\prime})\rVert_{X;q,q^{\prime};k,\infty;[t,T]}\le K_{1}. \tag*{(140)} ∥( ψ , ψ ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] ≤ K 1 whenever ∥( ψ , ψ ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] ≤ K 1 . (140)
Since ∥ ( Y ( 0 ) , ( Y ( 0 ) ) ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] = 0 ≤ K 1 \lVert(Y^{(0)},(Y^{(0)})^{\prime})\rVert_{X;q,q^{\prime};k,\infty;[t,T]}=0\le K_{1} ∥( Y ( 0 ) , ( Y ( 0 ) ) ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] = 0 ≤ K 1 , assuming (139) holds, we have
sup i ≥ 0 ∥ ( Y ( i ) , ( Y ( i ) ) ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] ≤ K 1 . (141) \sup_{i\ge0}\lVert(Y^{(i)},(Y^{(i)})^{\prime})\rVert_{X;q,q^{\prime};k,\infty;[t,T]}\le K_{1}. \tag*{(141)} i ≥ 0 sup ∥( Y ( i ) , ( Y ( i ) ) ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] ≤ K 1 . (141)
Now we show the contraction of ( Φ [ t , T ] ξ ) ⊗ 2 (\Phi^{\xi}_{[t,T]})^{\otimes2} ( Φ [ t , T ] ξ ) ⊗ 2 . Denote by B [ t , T ] ξ ; q , q ′ ( K 1 ) \mathcal{B}^{\xi;q,q^{\prime}}_{[t,T]}(K_{1}) B [ t , T ] ξ ; q , q ′ ( K 1 ) the set as follows:
B [ t , T ] ξ ; q , q ′ ( K 1 ) : = { ( ψ , ψ ′ , z ) ; ( ψ , ψ ′ ) ∈ RSM X ( p , p ) -var ( [ t , T ] ) , ∥ ( ψ , ψ ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] ≤ K 1 , ψ T = ξ , \mathcal{B}^{\xi;q,q^{\prime}}_{[t,T]}(K_{1}):=\left\{(\psi,\psi^{\prime},z);\;(\psi,\psi^{\prime})\in\operatorname{RSM}^{(p,p)\text{-var}}_{X}([t,T]),\ \lVert(\psi,\psi^{\prime})\rVert_{X;q,q^{\prime};k,\infty;[t,T]}\le K_{1},\ \psi_{T}=\xi,
\right. B [ t , T ] ξ ; q , q ′ ( K 1 ) := { ( ψ , ψ ′ , z ) ; ( ψ , ψ ′ ) ∈ RSM X ( p , p ) -var ([ t , T ]) , ∥( ψ , ψ ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] ≤ K 1 , ψ T = ξ ,
z is progressively measurable with ∥ z ∥ L 2 ( [ t , T ] ) < ∞ a.s., and δ M u , v ψ = ∫ u v z r d W r } . \left.
z\text{ is progressively measurable with }\lVert z\rVert_{L^{2}([t,T])}<\infty\ \text{a.s., and }\delta M^{\psi}_{u,v}=\int_{u}^{v}z_{r}\,\mathrm{d}W_{r}\right\}. z is progressively measurable with ∥ z ∥ L 2 ([ t , T ]) < ∞ a.s., and δ M u , v ψ = ∫ u v z r d W r } .
By (141) , we have ( Y ( i ) , ( Y ( i ) ) ′ , Z ( i ) ) ∈ B [ t , T ] ξ ; q , q ′ ( K 1 ) (Y^{(i)},(Y^{(i)})',Z^{(i)}) \in\mathcal{B}_{[t,T]}^{\xi;q,q'}(K_1) ( Y ( i ) , ( Y ( i ) ) ′ , Z ( i ) ) ∈ B [ t , T ] ξ ; q , q ′ ( K 1 ) for every i ≥ 1 i \ge1 i ≥ 1 . Moreover, since the map Φ [ t , T ] ξ \Phi_{[t,T]}^\xi Φ [ t , T ] ξ is invariant in B t , T ξ ; q , q ′ ( K 1 ) \mathcal{B}_{t,T}^{\xi;q,q'}(K_1) B t , T ξ ; q , q ′ ( K 1 ) under the condition (139) , by Proposition 4.16 there exists θ 2 > 0 \theta_2 > 0 θ 2 > 0 depending on K 1 K_1 K 1 such that if assuming further
∣ δ X ∣ p -var ; [ t , T ] + ( T − t ) 1 / 2 ≤ θ 1 ∧ θ 2 , (142) \lvert\delta X \rvert_{p\text{-var};[t,T]} + (T-t)^{1/2} \le\theta_1 \wedge\theta_2, \tag*{(142)} ∣ δ X ∣ p -var ; [ t , T ] + ( T − t ) 1/2 ≤ θ 1 ∧ θ 2 , (142)
then the map ( Φ [ t , T ] ξ ) ∘ 2 (\Phi_{[t,T]}^\xi)^{\circ2} ( Φ [ t , T ] ξ ) ∘ 2 is a contraction in B [ t , T ] ξ ; q , q ′ ( K 1 ) \mathcal{B}_{[t,T]}^{\xi;q,q'}(K_1) B [ t , T ] ξ ; q , q ′ ( K 1 ) .
Assume (142) holds. Then, by the completeness of B [ t , T ] ξ ; q , q ′ ( K 1 ) \mathcal{B}_{[t,T]}^{\xi;q,q'}(K_1) B [ t , T ] ξ ; q , q ′ ( K 1 ) shown in Lemma 3.16 and the contraction property of ( Φ [ t , T ] ξ ) ∘ 2 (\Phi_{[t,T]}^\xi)^{\circ2} ( Φ [ t , T ] ξ ) ∘ 2 , there exists ( Y , Y ′ , Z ) (Y,Y',Z) ( Y , Y ′ , Z ) , the unique fixed point of ( Φ [ t , T ] ξ ) ∘ 2 (\Phi_{[t,T]}^\xi)^{\circ2} ( Φ [ t , T ] ξ ) ∘ 2 in B [ t , T ] ξ ; q , q ′ ( K 1 ) \mathcal{B}_{[t,T]}^{\xi;q,q'}(K_1) B [ t , T ] ξ ; q , q ′ ( K 1 ) , such that ( Y , Y ′ ) ∈ RSM X ( p , p ) -var ( [ t , T ] , R ) (Y,Y') \in\operatorname{RSM}_{X}^{(p,p)\text{-var}}([t,T],\mathbb{R}) ( Y , Y ′ ) ∈ RSM X ( p , p ) -var ([ t , T ] , R ) and
lim i → ∞ ∥ ( Y ( i ) , ( Y ( i ) ) ′ ; Y , Y ′ ) ∥ X , X ; q , q ′ ; k , ∞ ; [ t , T ] = 0. \lim_{i\to\infty}\lVert(Y^{(i)},(Y^{(i)})';Y,Y')\rVert_{X,X;q,q';k,\infty;[t,T]}=0. i → ∞ lim ∥( Y ( i ) , ( Y ( i ) ) ′ ; Y , Y ′ ) ∥ X , X ; q , q ′ ; k , ∞ ; [ t , T ] = 0.
Clearly, since ( Y , Y ′ , Z ) ∈ B [ t , T ] ξ ; q , q ′ ( K 1 ) (Y,Y',Z) \in\mathcal{B}_{[t,T]}^{\xi;q,q'}(K_1) ( Y , Y ′ , Z ) ∈ B [ t , T ] ξ ; q , q ′ ( K 1 ) , we have ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] ≤ K 1 \lVert(Y,Y')\rVert_{X;q,q';k,\infty;[t,T]} \le K_1 ∥( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] ≤ K 1 .
Step 2. In this step, we prove the existence of the solution ( Y , Z ) (Y,Z) ( Y , Z ) on small intervals.
In the step 1, we show that ( Y , Y ′ , Z ) (Y,Y',Z) ( Y , Y ′ , Z ) is the unique fixed point of ( Φ [ t , T ] ξ ) ∘ 2 (\Phi_{[t,T]}^\xi)^{\circ2} ( Φ [ t , T ] ξ ) ∘ 2 in B [ t , T ] ξ ; q , q ′ ( K 1 ) \mathcal{B}_{[t,T]}^{\xi;q,q'}(K_1) B [ t , T ] ξ ; q , q ′ ( K 1 ) . Now we show that it is also a fixed point of Φ [ t , T ] ξ \Phi_{[t,T]}^\xi Φ [ t , T ] ξ in B [ t , T ] ξ ; q , q ′ ( K 1 ) \mathcal{B}_{[t,T]}^{\xi;q,q'}(K_1) B [ t , T ] ξ ; q , q ′ ( K 1 ) .
Denote ( Y ~ , Y ~ ′ , Z ~ ) : = Φ [ t , T ] ξ ( Y , Y ′ , Z ) (\widetilde{Y},\widetilde{Y}',\widetilde{Z}) := \Phi_{[t,T]}^\xi(Y,Y',Z) ( Y , Y ′ , Z ) := Φ [ t , T ] ξ ( Y , Y ′ , Z ) . Since ( Y , Y ′ , Z ) ∈ B [ t , T ] ξ ; q , q ′ ( K 1 ) (Y,Y',Z) \in\mathcal{B}_{[t,T]}^{\xi;q,q'}(K_1) ( Y , Y ′ , Z ) ∈ B [ t , T ] ξ ; q , q ′ ( K 1 ) , by (140) ( Y ~ , Y ~ ′ , Z ~ ) (\widetilde{Y},\widetilde{Y}',\widetilde{Z}) ( Y , Y ′ , Z ) also belongs to B [ t , T ] ξ ; q , q ′ ( K 1 ) \mathcal{B}_{[t,T]}^{\xi;q,q'}(K_1) B [ t , T ] ξ ; q , q ′ ( K 1 ) . Furthermore, the triplet ( Φ [ t , T ] ξ ) ∘ 3 ( Y , Y ′ , Z ) (\Phi_{[t,T]}^\xi)^{\circ3}(Y,Y',Z) ( Φ [ t , T ] ξ ) ∘ 3 ( Y , Y ′ , Z ) , which is equal to ( Y ~ , Y ~ ′ , Z ~ ) (\widetilde{Y},\widetilde{Y}',\widetilde{Z}) ( Y , Y ′ , Z ) , is also a fixed point of ( Φ [ t , T ] ξ ) ∘ 2 (\Phi_{[t,T]}^\xi)^{\circ2} ( Φ [ t , T ] ξ ) ∘ 2 . Therefore, the uniqueness of the fixed point of ( Φ [ t , T ] ξ ) ∘ 2 (\Phi_{[t,T]}^\xi)^{\circ2} ( Φ [ t , T ] ξ ) ∘ 2 yields
( Y ~ , Y ~ ′ , Z ~ ) = ( Y , Y ′ , Z ) . (\widetilde{Y},\widetilde{Y}',\widetilde{Z})=(Y,Y',Z). ( Y , Y ′ , Z ) = ( Y , Y ′ , Z ) .
Consequently, ( Y , Y ′ , Z ) (Y,Y',Z) ( Y , Y ′ , Z ) is a fixed point of Φ [ t , T ] ξ \Phi_{[t,T]}^\xi Φ [ t , T ] ξ , and hence ( Y , Z ) (Y,Z) ( Y , Z ) solves Eq. (1) on [ t , T ] [t,T] [ t , T ] .
Since ( Y , Z ) (Y,Z) ( Y , Z ) solves Eq. (1) with ∥ ( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] < ∞ \lVert(Y,Y')\rVert_{X;q,q';k,\infty;[t,T]}<\infty ∥( Y , Y ′ ) ∥ X ; q , q ′ ; k , ∞ ; [ t , T ] < ∞ , the a priori estimate proved in Theorem 4.9 implies ∥ ( Y , Y ′ ) ∥ X ; p , p ; k , ∞ ; [ t , T ] < ∞ \lVert(Y,Y')\rVert_{X;p,p;k,\infty;[t,T]}<\infty ∥( Y , Y ′ ) ∥ X ; p , p ; k , ∞ ; [ t , T ] < ∞ , which meets our requirement (137) .
Step 3. Now we apply a backward telescoping argument to prove the existence of a solution on an arbitrary interval [ 0 , T ] [0,T] [ 0 , T ] .
We claim that there exists a partition π = { [ t i , t i + 1 ] ; i = 0 , 1 , … , N − 1 } \pi=\{[t_i,t_{i+1}];i=0,1,\ldots,N-1\} π = {[ t i , t i + 1 ] ; i = 0 , 1 , … , N − 1 } of [ 0 , T ] [0,T] [ 0 , T ] such that, on each subinterval [ t i , t i + 1 ] [t_i,t_{i+1}] [ t i , t i + 1 ] , we can construct a solution as in Steps 1 and 2. To see this, since θ 1 \theta_1 θ 1 , K 1 K_1 K 1 , and θ 2 \theta_2 θ 2 are chosen independent of the terminal value, and since for any terminal value Y v Y_v Y v the condition K 1 ≥ C ( 1 + ∥ Y v ∥ ∞ ) K_1 \ge C(1+\lVert Y_v\rVert_\infty) K 1 ≥ C ( 1 + ∥ Y v ∥ ∞ ) always holds so that Proposition 4.15 always applies at any terminal time v v v , we can choose the partition fine enough so that the following condition holds:
∣ δ X ∣ p -var ; [ t i , t i + 1 ] + ∣ t i + 1 − t i ∣ 1 / 2 ≤ θ 1 ∧ θ 2 , \lvert\delta X\rvert_{p\text{-var};[t_i,t_{i+1}]}+\lvert t_{i+1}-t_i\rvert^{1/2}\le\theta_1\wedge\theta_2, ∣ δ X ∣ p -var ; [ t i , t i + 1 ] + ∣ t i + 1 − t i ∣ 1/2 ≤ θ 1 ∧ θ 2 ,
and then the arguments in Steps 1 and 2 apply for each subinterval [ t i , t i + 1 ] [t_i,t_{i+1}] [ t i , t i + 1 ] .
Finally, the global existence of the solution follows by iterating the procedures in Steps 1 and 2 on each subinterval, proceeding backward from the terminal time T T T to the initial time 0 0 0 , which can be done in finitely many steps. □ \square □
Remark 4.18. Given the a priori estimate (138) , the existence and uniqueness of the solution follow directly from Steps 1–3 of Theorem 4.17. Indeed, on the one hand, uniqueness is a direct consequence of the contractivity of the solution map. On the other hand, the a priori estimate ensures that the solution obtained by Picard iteration on a sufficiently small time interval can be successively extended to the whole interval [ 0 , T ] [0,T] [ 0 , T ] , thereby yielding global existence. This Picard iteration approach also applies to the multidimensional case.
We emphasize that in the scalar case, the a priori estimate is obtained via the comparison theorem. For the multidimensional case, however, the comparison theorem does not hold in general. Hence establishing the a priori estimate is highly nontrivial. More discussion can be seen in Remark 4.11.
We have the following quantitative stability result for the equation. Denote Θ 1 ′ : = ( Θ , C f , R 1 ′ ) \Theta'_1 := (\Theta,C_f,R'_1) Θ 1 ′ := ( Θ , C f , R 1 ′ ) .
Theorem 4.19. Assume both ( ξ , f , H , X ) (\xi,f,H,\mathbf{X}) ( ξ , f , H , X ) and ( ξ ˉ , f ˉ , H ˉ , X ˉ ) (\bar{\xi},\bar{f},\bar{H},\bar{\mathbf{X}}) ( ξ ˉ , f ˉ , H ˉ , X ˉ ) satisfy the conditions in Theorem 4.17. Let ( Y , Z ) (Y,Z) ( Y , Z ) (( Y ˉ , Z ˉ ) ( \bar{Y},\bar{Z}) ( Y ˉ , Z ˉ ) resp.) be the unique solution to (1.1) with the coefficient ( ξ , f , H , X ) (\xi,f,H,\mathbf{X}) ( ξ , f , H , X ) (the coefficient ( ξ ˉ , f ˉ , H ˉ , X ˉ ) (\bar{\xi},\bar{f},\bar{H},\bar{\mathbf{X}}) ( ξ ˉ , f ˉ , H ˉ , X ˉ ) resp.) such that
∥ ( Y , − H ( Y ) ) ∥ X ; q , q ′ ; 1 , ∞ + ∥ ( Y ˉ , − H ˉ ( Y ˉ ) ) ∥ X ˉ ; q , q ′ ; 1 , ∞ < ∞ . \lVert(Y,-H(Y))\rVert_{\mathbf{X};q,q';1,\infty}+\lVert(\bar{Y},-\bar{H}(\bar{Y}))\rVert_{\bar{\mathbf{X}};q,q';1,\infty}<\infty. ∥( Y , − H ( Y )) ∥ X ; q , q ′ ; 1 , ∞ + ∥( Y ˉ , − H ˉ ( Y ˉ )) ∥ X ˉ ; q , q ′ ; 1 , ∞ < ∞.
Let R 1 ′ > 0 R'_1>0 R 1 ′ > 0 be a constant such that
∥ ξ ∥ ∞ + ∥ ξ ˉ ∥ ∞ + ∥ H ∥ C b γ − 1 + ∥ H ˉ ∥ C b γ − 1 + ∣ δ X ∣ p -var + ∣ δ X ˉ ∣ p -var + T ≤ R 1 ′ . \lVert\xi\rVert_\infty+\lVert\bar{\xi}\rVert_\infty+\lVert H\rVert_{C_b^{\gamma-1}}+\lVert\bar{H}\rVert_{C_b^{\gamma-1}}+\lvert\delta X\rvert_{p\text{-var}}+\lvert\delta\bar{X}\rvert_{p\text{-var}}+T\leq R'_1. ∥ ξ ∥ ∞ + ∥ ξ ˉ ∥ ∞ + ∥ H ∥ C b γ − 1 + ∥ H ˉ ∥ C b γ − 1 + ∣ δ X ∣ p -var + ∣ δ X ˉ ∣ p -var + T ≤ R 1 ′ .
Then, for any k ≥ 1 k\geq1 k ≥ 1 the following local Lipschitz inequality holds up to a multiplicative constant depending on ( k , γ , Θ 1 ′ ) (k,\gamma,\Theta'_1) ( k , γ , Θ 1 ′ ) ,
∥ Y , − H ( Y ) ; Y ˉ , − H ˉ ( Y ˉ ) ∥ X , X ˉ ; p , p ; k , ∞ ≲ ∥ ξ − ξ ˉ ∥ ∞ + ∥ f − f ˉ ∥ ∞ + ∥ H − H ˉ ∥ C b γ − 1 + ∣ δ ( X − X ˉ ) ∣ p -var . (143) \lVert Y,-H(Y);\bar{Y},-\bar{H}(\bar{Y})\rVert_{\mathbf{X},\bar{\mathbf{X}};p,p;k,\infty}
\lesssim\lVert\xi-\bar{\xi}\rVert_\infty+\lVert f-\bar{f}\rVert_\infty+\lVert H-\bar{H}\rVert_{C_b^{\gamma-1}}+\lvert\delta(X-\bar{X})\rvert_{p\text{-var}}. \tag*{(143)} ∥ Y , − H ( Y ) ; Y ˉ , − H ˉ ( Y ˉ ) ∥ X , X ˉ ; p , p ; k , ∞ ≲ ∥ ξ − ξ ˉ ∥ ∞ + ∥ f − f ˉ ∥ ∞ + ∥ H − H ˉ ∥ C b γ − 1 + ∣ δ ( X − X ˉ ) ∣ p -var . (143)
In particular, ∥ Y − Y ˉ ∥ ∞ + ∥ δ ( ∫ 0 ⋅ ( Z r − Z ˉ r ) d W r ) ∥ B M O \lVert Y-\bar{Y}\rVert_\infty+\lVert\delta(\int_0^\cdot(Z_r-\bar{Z}_r)\,dW_r)\rVert_{\mathrm{BMO}} ∥ Y − Y ˉ ∥ ∞ + ∥ δ ( ∫ 0 ⋅ ( Z r − Z ˉ r ) d W r ) ∥ BMO is also bounded by the right-hand side of the above inequality.
Proof. Without loss of generality, we assume k > 1 k>1 k > 1 to apply Theorem 4.9. Recall Y ′ = − H ( Y ) Y'=-H(Y) Y ′ = − H ( Y ) and Y ˉ ′ = − H ˉ ( Y ˉ ) \bar{Y}'=-\bar{H}(\bar{Y}) Y ˉ ′ = − H ˉ ( Y ˉ ) . For simplicity of notation, denote
L : = ∥ H − H ˉ ∥ C b γ − 1 + ∣ δ ( X − X ˉ ) ∣ p -var ; [ 0 , T ] + ∥ f − f ˉ ∥ ∞ , I t X : = ∫ 0 t H ( Y r ) d X r , I t : = ∫ 0 t f ( r , Y r , Z r ) d r , I t X ˉ : = ∫ 0 t H ˉ ( Y ˉ r ) d X ˉ r , I ˉ t : = ∫ 0 t f ˉ ( r , Y ˉ r , Z ˉ r ) d r . \begin{aligned}
L&:=\lVert H-\bar{H}\rVert_{C_b^{\gamma-1}}+\lvert\delta(X-\bar{X})\rvert_{p\text{-var};[0,T]}+\lVert f-\bar{f}\rVert_\infty,\\
I_t^X&:=\int_0^t H(Y_r)\,dX_r,\quad I_t:=\int_0^t f(r,Y_r,Z_r)\,dr,\\
I_t^{\bar{X}}&:=\int_0^t \bar{H}(\bar{Y}_r)\,d\bar{X}_r,\quad\bar{I}_t:=\int_0^t \bar{f}(r,\bar{Y}_r,\bar{Z}_r)\,dr.
\end{aligned} L I t X I t X ˉ := ∥ H − H ˉ ∥ C b γ − 1 + ∣ δ ( X − X ˉ ) ∣ p -var ; [ 0 , T ] + ∥ f − f ˉ ∥ ∞ , := ∫ 0 t H ( Y r ) d X r , I t := ∫ 0 t f ( r , Y r , Z r ) d r , := ∫ 0 t H ˉ ( Y ˉ r ) d X ˉ r , I ˉ t := ∫ 0 t f ˉ ( r , Y ˉ r , Z ˉ r ) d r .
By the a priori estimate made by Theorem 4.9, we have for some constant K ^ \hat{K} K ^ depending only on k k k , γ \gamma γ , and Θ 1 ′ \Theta'_1 Θ 1 ′ ,
∥ ( Y , Y ′ ) ∥ X ; p , p ; k , ∞ ; [ 0 , T ] + ∥ ( Y ˉ , Y ˉ ′ ) ∥ X ˉ ; p , p ; k , ∞ ; [ 0 , T ] ≤ K ^ . \lVert(Y,Y')\rVert_{\mathbf{X};p,p;k,\infty;[0,T]}+\lVert(\bar{Y},\bar{Y}')\rVert_{\bar{\mathbf{X}};p,p;k,\infty;[0,T]}\leq\hat{K}. ∥( Y , Y ′ ) ∥ X ; p , p ; k , ∞ ; [ 0 , T ] + ∥( Y ˉ , Y ˉ ′ ) ∥ X ˉ ; p , p ; k , ∞ ; [ 0 , T ] ≤ K ^ .
First, we estimate P Y − P Y ˉ P^Y-P^{\bar{Y}} P Y − P Y ˉ . By Lemma 3.17, we have (noting the dependence of K ^ \hat{K} K ^ is absorbed by ( k , γ , Θ 1 ′ ) (k,\gamma,\Theta'_1) ( k , γ , Θ 1 ′ ) and
∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; q , q ′ ; k , ∞ ; [ s , t ] ≤ ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] , ∥ R X − R X ˉ ∥ p / 2 -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ δ ( X − X ˉ ) ∥ p -var ; ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 1 ′ ∣ δ X ∣ p -var ; [ s , t ] ( ∥ Y t − Y ˉ t ∥ ∞ + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] + ∥ H − H ˉ ∥ C b γ − 1 + ∣ δ ( X − X ˉ ) ∣ p -var ; [ s , t ] ) . \begin{aligned}
\lVert Y,Y';\bar{Y},\bar{Y}'\rVert_{\mathbf{X},\bar{\mathbf{X}};q,q';k,\infty;[s,t]}
&\leq\lVert Y,Y';\bar{Y},\bar{Y}'\rVert_{\mathbf{X},\bar{\mathbf{X}};p,p;k,\infty;[s,t]},\\
\lVert R^X-R^{\bar{X}}\rVert_{p/2\text{-var};\infty;k,\infty;[s,t]}
+\lVert\delta(X-\bar{X})\rVert_{p\text{-var};\infty;k,\infty;[s,t]}
&\lesssim_{k,\gamma,\Theta'_1}\lvert\delta X\rvert_{p\text{-var};[s,t]}\left(\lVert Y_t-\bar{Y}_t\rVert_\infty\right.\\
&\qquad\left.+\lVert Y,Y';\bar{Y},\bar{Y}'\rVert_{\mathbf{X},\bar{\mathbf{X}};p,p;k,\infty;[s,t]}+\lVert H-\bar{H}\rVert_{C_b^{\gamma-1}}+\lvert\delta(X-\bar{X})\rvert_{p\text{-var};[s,t]}\right).
\end{aligned} ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; q , q ′ ; k , ∞ ; [ s , t ] ∥ R X − R X ˉ ∥ p /2 -var ; ∞ ; k , ∞ ; [ s , t ] + ∥ δ ( X − X ˉ ) ∥ p -var ; ∞ ; k , ∞ ; [ s , t ] ≤ ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] , ≲ k , γ , Θ 1 ′ ∣ δ X ∣ p -var ; [ s , t ] ( ∥ Y t − Y ˉ t ∥ ∞ + ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] + ∥ H − H ˉ ∥ C b γ − 1 + ∣ δ ( X − X ˉ ) ∣ p -var ; [ s , t ] ) .
Moreover, by the same calculation leading to (4.47), but with the difference of ( Y − Y ˉ ) t (Y-\bar{Y})_t ( Y − Y ˉ ) t and f − f ˉ f-\bar{f} f − f ˉ added, we have
∥ ∥ δ ( I − I ˉ ) ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 1 ′ ( t − s ) ∥ ( Y − Y ˉ ) t ∥ ∞ + ( t − s ) 1 2 ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] + ∥ f − f ˉ ∥ ∞ . \begin{aligned}
\lVert\lVert\delta(I-\bar{I})\rVert_{1\text{-var}}\rVert_{\infty;k,\infty;[s,t]}
&\lesssim_{k,\gamma,\Theta'_1}(t-s)\lVert(Y-\bar{Y})_t\rVert_\infty\\
&\quad+(t-s)^{\frac12}\lVert Y,Y';\bar{Y},\bar{Y}'\rVert_{\mathbf{X},\bar{\mathbf{X}};p,p;k,\infty;[s,t]}+\lVert f-\bar{f}\rVert_\infty.
\end{aligned} ∥∥ δ ( I − I ˉ ) ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 1 ′ ( t − s ) ∥( Y − Y ˉ ) t ∥ ∞ + ( t − s ) 2 1 ∥ Y , Y ′ ; Y ˉ , Y ˉ ′ ∥ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] + ∥ f − f ˉ ∥ ∞ .
Therefore, noting that P Y − P Y ˉ = − R I X + R I ˉ X ˉ − δ ( I − I ˉ ) P^{Y}-P^{\bar{Y}}=-R^{I}X+R^{\bar{I}}\bar{X}-\delta(I-\bar{I}) P Y − P Y ˉ = − R I X + R I ˉ X ˉ − δ ( I − I ˉ ) , we have
∥ P Y − P Y ˉ ∥ p 2 -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 1 ′ ∥ Y t − Y ˉ t ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ∣ Y , Y ′ ; Y ˉ , Y ˉ ′ ∣ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] + L . (144) \left\lVert P^{Y}-P^{\bar{Y}}\right\rVert_{\frac{p}{2}\text{-var}\,\lVert\infty;k,\infty;[s,t]}
\lesssim_{k,\gamma,\Theta'_1}\left\lVert Y_t-\bar{Y}_t\right\rVert_{\infty}+\left(\left\lvert\delta X\right\rvert_{p\text{-var};[s,t]}+(t-s)^{\frac{1}{2}}\right)\left\lvert Y,Y';\bar{Y},\bar{Y}'\right\rvert_{X,\bar{X};p,p;k,\infty;[s,t]}+L. \tag*{(144)} P Y − P Y ˉ 2 p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 1 ′ Y t − Y ˉ t ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) Y , Y ′ ; Y ˉ , Y ˉ ′ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] + L . (144)
Next, we estimate M Y − M Y ˉ M^{Y}-M^{\bar{Y}} M Y − M Y ˉ and Y − Y ˉ Y-\bar{Y} Y − Y ˉ . Note that for r ∈ [ s , t ] r\in[s,t] r ∈ [ s , t ] ,
δ ( M Y − M Y ˉ ) r , t = − E r [ Y t − Y ˉ t + δ ( I X − I X ˉ + I − I ˉ ) r , t ] + Y t − Y ˉ t + δ ( I X − I X ˉ + I − I ˉ ) r , t . \delta(M^{Y}-M^{\bar{Y}})_{r,t}=-\mathbb{E}_{r}\left[Y_t-\bar{Y}_t+\delta(I^{X}-I^{\bar{X}}+I-\bar{I})_{r,t}\right]+Y_t-\bar{Y}_t+\delta(I^{X}-I^{\bar{X}}+I-\bar{I})_{r,t}. δ ( M Y − M Y ˉ ) r , t = − E r [ Y t − Y ˉ t + δ ( I X − I X ˉ + I − I ˉ ) r , t ] + Y t − Y ˉ t + δ ( I X − I X ˉ + I − I ˉ ) r , t .
Then, since I X − I X ˉ I^{X}-I^{\bar{X}} I X − I X ˉ and I − I ˉ I-\bar{I} I − I ˉ are already estimated as follows:
∥ ∥ δ ( I X − I X ˉ ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ ( I − I ˉ ) ∥ 1 -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 1 ′ ∥ Y t − Y ˉ t ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ∣ Y , Y ′ ; Y ˉ , Y ˉ ′ ∣ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] + L . (145) \left\lVert\left\lVert\delta(I^{X}-I^{\bar{X}})\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}+\left\lVert\left\lVert\delta(I-\bar{I})\right\rVert_{1\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}
\lesssim_{k,\gamma,\Theta'_1}\left\lVert Y_t-\bar{Y}_t\right\rVert_{\infty}+\left(\left\lvert\delta X\right\rvert_{p\text{-var};[s,t]}+(t-s)^{\frac{1}{2}}\right)\left\lvert Y,Y';\bar{Y},\bar{Y}'\right\rvert_{X,\bar{X};p,p;k,\infty;[s,t]}+L. \tag*{(145)} δ ( I X − I X ˉ ) p -var ∞ ; k , ∞ ; [ s , t ] + δ ( I − I ˉ ) 1 -var ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 1 ′ Y t − Y ˉ t ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) Y , Y ′ ; Y ˉ , Y ˉ ′ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] + L . (145)
we can follow the same calculation as (130) –(131) to obtain
∥ ∥ δ ( M Y − M Y ˉ ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ δ ( M Y − M Y ˉ ) ∥ B M O ; [ s , t ] ≲ k , γ , Θ 1 ′ RHS of (4.64) . (146) \left\lVert\left\lVert\delta(M^{Y}-M^{\bar{Y}})\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}+\left\lVert\delta(M^{Y}-M^{\bar{Y}})\right\rVert_{\mathrm{BMO};[s,t]}\lesssim_{k,\gamma,\Theta'_1}\text{RHS of (4.64)}. \tag*{(146)} δ ( M Y − M Y ˉ ) p -var ∞ ; k , ∞ ; [ s , t ] + δ ( M Y − M Y ˉ ) BMO ; [ s , t ] ≲ k , γ , Θ 1 ′ RHS of (4.64) . (146)
Note δ ( Y − Y ˉ ) = δ ( − I X + I ˉ X ˉ − I + I ˉ + M Y − M Y ˉ ) \delta(Y-\bar{Y})=\delta(-I^{X}+\bar{I}^{\bar{X}}-I+\bar{I}+M^{Y}-M^{\bar{Y}}) δ ( Y − Y ˉ ) = δ ( − I X + I ˉ X ˉ − I + I ˉ + M Y − M Y ˉ ) . The above estimate together with (4.64) yields
∥ ∥ δ ( Y − Y ˉ ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 1 ′ RHS of (4.64) . (147) \left\lVert\left\lVert\delta(Y-\bar{Y})\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}\lesssim_{k,\gamma,\Theta'_1}\text{RHS of (4.64)}. \tag*{(147)} δ ( Y − Y ˉ ) p -var ∞ ; k , ∞ ; [ s , t ] ≲ k , γ , Θ 1 ′ RHS of (4.64) . (147)
Now, for Y ′ − Y ˉ ′ = − H ( Y ) + H ( Y ˉ ) − ( H − H ˉ ) ( Y ˉ ) Y'-\bar{Y}'=-H(Y)+H(\bar{Y})-(H-\bar{H})(\bar{Y}) Y ′ − Y ˉ ′ = − H ( Y ) + H ( Y ˉ ) − ( H − H ˉ ) ( Y ˉ ) , by Lemma A.5 (noting H ∈ C b γ H\in C_b^\gamma H ∈ C b γ so that H ∈ C b 2 H\in C_b^2 H ∈ C b 2 ) and inequalities ∣ Y ′ − Y ˉ ′ ∣ ≲ Θ 1 ′ ∣ Y − Y ˉ ∣ + ∥ H − H ˉ ∥ ∞ \lvert Y'-\bar{Y}'\rvert\lesssim_{\Theta'_1}\lvert Y-\bar{Y}\rvert+\lVert H-\bar{H}\rVert_{\infty} ∣ Y ′ − Y ˉ ′ ∣ ≲ Θ 1 ′ ∣ Y − Y ˉ ∣ + ∥ H − H ˉ ∥ ∞ and ∥ δ ( H − H ˉ ) ( Y ˉ ) ∥ p -var ≤ ∥ D ( H − H ˉ ) ∥ ∞ ∥ δ Y ˉ ∥ p -var \left\lVert\delta(H-\bar{H})(\bar{Y})\right\rVert_{p\text{-var}}\leq\lVert D(H-\bar{H})\rVert_{\infty}\left\lVert\delta\bar{Y}\right\rVert_{p\text{-var}} δ ( H − H ˉ ) ( Y ˉ ) p -var ≤ ∥ D ( H − H ˉ ) ∥ ∞ δ Y ˉ p -var , we have
∥ Y ′ − Y ˉ ′ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y ′ − Y ˉ ′ ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ≲ Θ 1 ′ ( 1 + ∥ ∥ δ Y ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ ∥ δ Y ˉ ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ) ∥ Y − Y ˉ ∥ L ∞ ( [ s , t ] × Ω ) + ∥ ∥ δ ( Y − Y ˉ ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ H − H ˉ ∥ C b 1 ( 1 + ∥ ∥ δ Y ˉ ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] ) ≲ k , γ , Θ 1 ′ ∥ Y t − Y ˉ t ∥ ∞ + ∥ ∥ δ ( Y − Y ˉ ) ∥ p -var ∥ ∞ ; k , ∞ ; [ s , t ] + ∥ H − H ˉ ∥ C b 1 . (148) \begin{aligned}
\left\lVert Y'-\bar{Y}'\right\rVert_{L^\infty([s,t]\times\Omega)}+\left\lVert\left\lVert\delta(Y'-\bar{Y}')\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}
&\lesssim_{\Theta'_1}\left(1+\left\lVert\left\lVert\delta Y\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}+\left\lVert\left\lVert\delta\bar{Y}\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}\right)\left\lVert Y-\bar{Y}\right\rVert_{L^\infty([s,t]\times\Omega)}\\
&\quad+\left\lVert\left\lVert\delta(Y-\bar{Y})\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}+\left\lVert H-\bar{H}\right\rVert_{C_b^1}\left(1+\left\lVert\left\lVert\delta\bar{Y}\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}\right)\\
&\lesssim_{k,\gamma,\Theta'_1}\left\lVert Y_t-\bar{Y}_t\right\rVert_{\infty}+\left\lVert\left\lVert\delta(Y-\bar{Y})\right\rVert_{p\text{-var}}\right\rVert_{\infty;k,\infty;[s,t]}+\left\lVert H-\bar{H}\right\rVert_{C_b^1}. \tag*{(148)}
\end{aligned} Y ′ − Y ˉ ′ L ∞ ([ s , t ] × Ω ) + δ ( Y ′ − Y ˉ ′ ) p -var ∞ ; k , ∞ ; [ s , t ] ≲ Θ 1 ′ ( 1 + ∥ δ Y ∥ p -var ∞ ; k , ∞ ; [ s , t ] + δ Y ˉ p -var ∞ ; k , ∞ ; [ s , t ] ) Y − Y ˉ L ∞ ([ s , t ] × Ω ) + δ ( Y − Y ˉ ) p -var ∞ ; k , ∞ ; [ s , t ] + H − H ˉ C b 1 ( 1 + δ Y ˉ p -var ∞ ; k , ∞ ; [ s , t ] ) ≲ k , γ , Θ 1 ′ Y t − Y ˉ t ∞ + δ ( Y − Y ˉ ) p -var ∞ ; k , ∞ ; [ s , t ] + H − H ˉ C b 1 . (148)
By (4.66), the RHS of the above estimate is, in turn, bounded by the RHS of (4.64). Combining (4.63), (4.65), and (4.67), for some constant C > 0 C>0 C > 0 independent of t t t we have
∣ Y , Y ′ ; Y ˉ , Y ˉ ′ ∣ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] ≤ C ( ∥ Y t − Y ˉ t ∥ ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 2 ) ∣ Y , Y ′ ; Y ˉ , Y ˉ ′ ∣ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] + L ) . \left\lvert Y,Y';\bar{Y},\bar{Y}'\right\rvert_{X,\bar{X};p,p;k,\infty;[s,t]}
\leq C\left(\left\lVert Y_t-\bar{Y}_t\right\rVert_{\infty}+\left(\left\lvert\delta X\right\rvert_{p\text{-var};[s,t]}+(t-s)^{\frac{1}{2}}\right)\left\lvert Y,Y';\bar{Y},\bar{Y}'\right\rvert_{X,\bar{X};p,p;k,\infty;[s,t]}+L\right). Y , Y ′ ; Y ˉ , Y ˉ ′ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] ≤ C ( Y t − Y ˉ t ∞ + ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 2 1 ) Y , Y ′ ; Y ˉ , Y ˉ ′ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] + L ) .
Hence, the following inequality holds whenever ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1 / 2 ) ≤ ( 2 C ) − 1 \left(\left\lvert\delta X\right\rvert_{p\text{-var};[s,t]}+(t-s)^{1/2}\right)\leq(2C)^{-1} ( ∣ δ X ∣ p -var ; [ s , t ] + ( t − s ) 1/2 ) ≤ ( 2 C ) − 1 :
∣ Y , Y ′ ; Y ˉ , Y ˉ ′ ∣ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] ≤ 2 C ∥ Y t − Y ˉ t ∥ ∞ + 2 C L . \left\lvert Y,Y';\bar{Y},\bar{Y}'\right\rvert_{X,\bar{X};p,p;k,\infty;[s,t]}\leq2C\left\lVert Y_t-\bar{Y}_t\right\rVert_{\infty}+2CL. Y , Y ′ ; Y ˉ , Y ˉ ′ X , X ˉ ; p , p ; k , ∞ ; [ s , t ] ≤ 2 C Y t − Y ˉ t ∞ + 2 C L .
In particular, noting Y T − Y ˉ T = ξ − ξ ˉ Y_T-\bar{Y}_T=\xi-\bar{\xi} Y T − Y ˉ T = ξ − ξ ˉ , it holds ∣ Y , Y ′ ; Y ˉ , Y ˉ ′ ∣ X , X ˉ ; p , p ; k , ∞ ; [ t , T ] ≤ 2 C ∥ ξ − ξ ˉ ∥ ∞ + 2 C L \left\lvert Y,Y';\bar{Y},\bar{Y}'\right\rvert_{X,\bar{X};p,p;k,\infty;[t,T]}\leq2C\left\lVert\xi-\bar{\xi}\right\rVert_{\infty}+2CL Y , Y ′ ; Y ˉ , Y ˉ ′ X , X ˉ ; p , p ; k , ∞ ; [ t , T ] ≤ 2 C ξ − ξ ˉ ∞ + 2 C L whenever ( ∣ δ X ∣ p -var ; [ t , T ] + ( T − t ) 1 / 2 ) ≤ ( 2 C ) − 1 \left(\left\lvert\delta X\right\rvert_{p\text{-var};[t,T]}+(T-t)^{1/2}\right)\leq(2C)^{-1} ( ∣ δ X ∣ p -var ; [ t , T ] + ( T − t ) 1/2 ) ≤ ( 2 C ) − 1 .
Since for each t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] , by (3.43) in Lemma 3.13 the difference ∥ Y t − Y ˉ t ∥ ∞ \left\lVert Y_t-\bar{Y}_t\right\rVert_{\infty} Y t − Y ˉ t ∞ is dominated by ∥ ξ − ξ ˉ ∥ ∞ + ∣ Y , Y ′ ; Y ˉ , Y ˉ ′ ∣ X , X ˉ ; p , p ; k , ∞ ; [ t , T ] \left\lVert\xi-\bar{\xi}\right\rVert_{\infty}+\left\lvert Y,Y';\bar{Y},\bar{Y}'\right\rvert_{X,\bar{X};p,p;k,\infty;[t,T]} ξ − ξ ˉ ∞ + Y , Y ′ ; Y ˉ , Y ˉ ′ X , X ˉ ; p , p ; k , ∞ ; [ t , T ] . Therefore, the desired estimate can be verified via a backward telescoping argument, in the same way as in Step 4 of Proposition 4.7. □
Remark 4.20. In this paper, we assume that H ( ⋅ ) H(\cdot) H ( ⋅ ) is independent of time t t t . Under the assumption that H H H is deterministic, indeed, one can also consider time-dependent X \mathbf{X} X -controlled vector fields of the type considered in, e.g., [23 , 3 ] . We restrict ourselves to the time-independent case mainly to avoid lengthy but routine calculations. However, the genuinely random vector fields case, such as H ( W t , Y t ) H(W_t,Y_t) H ( W t , Y t ) , is substantially more delicate, as explained in Remark 4.10.
Remark 4.21. Under higher regularity assumptions on H ( ⋅ ) H(\cdot) H ( ⋅ ) , Eq. eq:1 can be solved through Doss–Sussmann transformation and Itô–Wentzell formula. We sketch the argument of the well-posedness by this method under assumptions H ( ⋅ ) ∈ C b 5 ( R , R d ) H(\cdot) \in C_b^5(\mathbb{R},\mathbb{R}^d) H ( ⋅ ) ∈ C b 5 ( R , R d ) and X ∈ C g 0 , p -var ( [ 0 , T ] , R d ) \mathbf{X} \in C_g^{0,p\text{-var}}([0,T],\mathbb{R}^d) X ∈ C g 0 , p -var ([ 0 , T ] , R d ) .
For T ′ ∈ [ 0 , T ] T' \in[0,T] T ′ ∈ [ 0 , T ] , let ϕ T ′ ( t , y ) \phi^{T'}(t,y) ϕ T ′ ( t , y ) be the unique solution to the following RDE:
ϕ T ′ ( t , y ) = y + ∫ t T ′ H ( ϕ T ′ ( r , y ) ) d X r , ( t , y ) ∈ [ 0 , T ′ ] × R . \phi^{T'}(t,y)=y+\int_t^{T'} H(\phi^{T'}(r,y))dX_r,\qquad(t,y)\in[0,T']\times\mathbb{R}. ϕ T ′ ( t , y ) = y + ∫ t T ′ H ( ϕ T ′ ( r , y )) d X r , ( t , y ) ∈ [ 0 , T ′ ] × R .
Then, by [14 ] , Proposition 11.11, ϕ T ′ ( t , y ) \phi^{T'}(t,y) ϕ T ′ ( t , y ) forms a flow of C 3 C^3 C 3 -diffeomorphisms.
Thanks to the C 3 C^3 C 3 -smoothness of ϕ T ′ \phi^{T'} ϕ T ′ , the approximation solution ( Y , Z ) (Y,Z) ( Y , Z ) to Eq. eq:1 obtained by [7 ] has an explicit representation through classical quadratic BSDEs. Indeed, further assume that ξ ∈ L ∞ \xi\in L^\infty ξ ∈ L ∞ and ( f ) (f) ( f ) holds. Denote f ~ T ′ \tilde f^{T'} f ~ T ′ as follows:
f ~ T ′ ( t , y , z ) : = ( f ( t , ϕ T ′ ( t , y ) , ∂ y ϕ T ′ ( t , y ) z ) + 1 2 ∂ y y 2 ϕ T ′ ( t , y ) ∣ z ∣ 2 ) / ∂ y ϕ T ′ ( t , y ) , ( t , y , z ) ∈ [ 0 , T ′ ] × R 1 + d . \tilde f^{T'}(t,y,z):=\left(f(t,\phi^{T'}(t,y),\partial_y\phi^{T'}(t,y)z)+\frac{1}{2}\partial_{yy}^2\phi^{T'}(t,y)|z|^2\right)/\partial_y\phi^{T'}(t,y),\qquad(t,y,z)\in[0,T']\times\mathbb{R}^{1+d}. f ~ T ′ ( t , y , z ) := ( f ( t , ϕ T ′ ( t , y ) , ∂ y ϕ T ′ ( t , y ) z ) + 2 1 ∂ y y 2 ϕ T ′ ( t , y ) ∣ z ∣ 2 ) / ∂ y ϕ T ′ ( t , y ) , ( t , y , z ) ∈ [ 0 , T ′ ] × R 1 + d .
Then, by the proof of [7 ] , Theorem 3, there exists π \pi π , a partition on [ 0 , T ] [0,T] [ 0 , T ] , s.t. for each [ u , v ] ∈ π [u,v]\in\pi [ u , v ] ∈ π ,
Y t = ϕ v ( t , Y ~ t v ) , Z t = ∂ y ϕ v ( t , Y ~ t v ) Z ~ t v , t ∈ [ u , v ] . Y_t=\phi^v(t,\widetilde Y_t^v),\qquad Z_t=\partial_y\phi^v(t,\widetilde Y_t^v)\widetilde Z_t^v,\qquad t\in[u,v]. Y t = ϕ v ( t , Y t v ) , Z t = ∂ y ϕ v ( t , Y t v ) Z t v , t ∈ [ u , v ] .
Here, ( Y ~ v , Z ~ v ) (\widetilde Y^v,\widetilde Z^v) ( Y v , Z v ) is the unique solution to the following quadratic BSDE in L ∞ ( [ u , v ] × Ω , R ) × L 2 ( [ u , v ] × Ω , R d ) L^\infty([u,v]\times\Omega,\mathbb{R})\times L^2([u,v]\times\Omega,\mathbb{R}^d) L ∞ ([ u , v ] × Ω , R ) × L 2 ([ u , v ] × Ω , R d ) :
Y ~ t v = Y v + ∫ t v f ~ v ( r , Y ~ r v , Z ~ r v ) d r − ∫ t v Z ~ r v d W r , t ∈ [ u , v ] . \widetilde Y_t^v=Y_v+\int_t^v\widetilde f^v(r,\widetilde Y_r^v,\widetilde Z_r^v)dr-\int_t^v\widetilde Z_r^v dW_r,\qquad t\in[u,v]. Y t v = Y v + ∫ t v f v ( r , Y r v , Z r v ) d r − ∫ t v Z r v d W r , t ∈ [ u , v ] .
Note that for [ T ′ , T ] ∈ π [T',T]\in\pi [ T ′ , T ] ∈ π , Y T = ξ Y_T=\xi Y T = ξ is given so that ( Y ~ T , Z ~ T ) (\widetilde Y^T,\widetilde Z^T) ( Y T , Z T ) can be solved on [ T ′ , T ] [T',T] [ T ′ , T ] . Consequently, the next terminal value Y T ′ = ϕ T ( T ′ , Y ~ T ′ T ) Y_{T'}=\phi^T(T',\widetilde Y_{T'}^T) Y T ′ = ϕ T ( T ′ , Y T ′ T ) can be obtained, and hence one can repeat this procedure on subintervals up to time 0.
To see that ( Y , Z ) (Y,Z) ( Y , Z ) satisfies Definition 4.1, it suffices to show that ( Y , − H ( Y ) ) (Y,-H(Y)) ( Y , − H ( Y )) is an RSM, and that the integral equation Eq. eq:1 holds. Since H ( ⋅ ) ∈ C b 5 H(\cdot)\in C_b^5 H ( ⋅ ) ∈ C b 5 , by [3 ] , Example 3.16, for each [ u , v ] ∈ π [u,v]\in\pi [ u , v ] ∈ π ,
( ϕ v , − H ( ϕ v ) , ∂ y ϕ v , D H ( ϕ v ) H ( ϕ v ) , ∂ y y 2 ϕ v , 0 ) ∈ D X 3 / p Lip y 3 ( R ; R ) , (\phi^v,-H(\phi^v),\partial_y\phi^v,DH(\phi^v)H(\phi^v),\partial_{yy}^2\phi^v,0)\in D_X^{3/p}\operatorname{Lip}_y^3(\mathbb{R};\mathbb{R}), ( ϕ v , − H ( ϕ v ) , ∂ y ϕ v , D H ( ϕ v ) H ( ϕ v ) , ∂ y y 2 ϕ v , 0 ) ∈ D X 3/ p Lip y 3 ( R ; R ) ,
where the controlled field space D X 3 / p Lip y 3 ( R ; R ) D_X^{3/p}\operatorname{Lip}_y^3(\mathbb{R};\mathbb{R}) D X 3/ p Lip y 3 ( R ; R ) is defined in [3 ] , Definition 3.6. Clearly Y ~ v \widetilde Y^v Y v is a semimartingale and hence ( Y ~ v , 0 ) (\widetilde Y^v,0) ( Y v , 0 ) forms an RSM. Therefore, by the rough stochastic Itô–Wentzell formula [3 ] , Theorem 4.16, ( Y t , − H ( Y t ) ) = ( ϕ v ( t , Y ~ t v ) , − H ( ϕ v ( t , Y ~ t v ) ) ) , t ∈ [ u , v ] (Y_t,-H(Y_t))=(\phi^v(t,\widetilde Y_t^v),-H(\phi^v(t,\widetilde Y_t^v))),\ t\in[u,v] ( Y t , − H ( Y t )) = ( ϕ v ( t , Y t v ) , − H ( ϕ v ( t , Y t v ))) , t ∈ [ u , v ] , is an RSM, and it holds
Y t = Y v + ∫ t v f ( r , Y r , Z r ) d r + ∫ t v H ( Y r ) d X r − ∫ t v Z r d W r , t ∈ [ u , v ] . Y_t=Y_v+\int_t^v f(r,Y_r,Z_r)dr+\int_t^v H(Y_r)dX_r-\int_t^v Z_r dW_r,\qquad t\in[u,v]. Y t = Y v + ∫ t v f ( r , Y r , Z r ) d r + ∫ t v H ( Y r ) d X r − ∫ t v Z r d W r , t ∈ [ u , v ] .
This proves that ( Y , Z ) (Y,Z) ( Y , Z ) is a solution to Eq. eq:1 .
Finally, we show the uniqueness of the solution. Indeed, let { X n = ( X n , X n ) } n ≥ 1 \{\mathbf{X}^n=(X^n,\mathbb{X}^n)\}_{n\geq1} { X n = ( X n , X n ) } n ≥ 1 be a sequence of rough paths canonically lifted by smooth paths { X n } n ≥ 1 \{X^n\}_{n\geq1} { X n } n ≥ 1 , and assume ∣ δ ( X n − X ) ∣ p -var → 0 |\delta(\mathbf{X}^n-\mathbf{X})|_{p\text{-var}}\to0 ∣ δ ( X n − X ) ∣ p -var → 0 . By [7 ] , Theorem 3, we see that
Y n → Y uniformly on [ 0 , T ] a.s. and Z n → Z in L 2 ( [ 0 , T ] × Ω , R d ) . Y^n\to Y\quad\text{uniformly on }[0,T]\quad\text{a.s. and}\quad Z^n\to Z\text{ in }L^2([0,T]\times\Omega,\mathbb{R}^d). Y n → Y uniformly on [ 0 , T ] a.s. and Z n → Z in L 2 ([ 0 , T ] × Ω , R d ) .
Here, ( Y n , Z n ) ∈ L ∞ ( [ 0 , T ] × Ω , R ) × L 2 ( [ 0 , T ] × Ω , R d ) (Y^n,Z^n) \in L^\infty([0,T]\times\Omega,\mathbb{R}) \times L^2([0,T]\times\Omega,\mathbb{R}^d) ( Y n , Z n ) ∈ L ∞ ([ 0 , T ] × Ω , R ) × L 2 ([ 0 , T ] × Ω , R d ) is the unique solution to Eq. (1) with X X X replaced by X n X^n X n . Furthermore, by Propositions 4.12 and Theorem 4.9, we have
sup n ∥ ( Y n , − H ( Y n ) ) ∥ X n ; p , p ; 2 , ∞ < ∞ . \sup_n \left\lVert(Y^n,-H(Y^n)) \right\rVert_{\mathbb{X}^n;p,p;2,\infty} < \infty. n sup ∥ ( Y n , − H ( Y n )) ∥ X n ; p , p ; 2 , ∞ < ∞.
Then, due to the convergence of ( Y n , Z n ) (Y^n,Z^n) ( Y n , Z n ) , Fatou’s lemma yields that ∥ ( Y , − H ( Y ) ) ∥ X ; p , p ; 2 , ∞ < ∞ \left\lVert(Y,-H(Y)) \right\rVert_{\mathbb{X};p,p;2,\infty} < \infty ∥ ( Y , − H ( Y )) ∥ X ; p , p ; 2 , ∞ < ∞ . Consequently, one can apply Corollary 4.6 to see that ( Y , Z ) (Y,Z) ( Y , Z ) is the unique solution to Eq. (1) .
As discussed in [3 ] , Remark 3.17, due to the non-explosion condition obtained in [14 ] , Chapter 11, one may relax the C b 5 C_b^5 C b 5 assumption on H ( ⋅ ) H(\cdot) H ( ⋅ ) to C b γ + 2 C_b^{\gamma+2} C b γ + 2 with γ > p \gamma> p γ > p . However, it is still stronger than the condition H ( ⋅ ) ∈ C b γ H(\cdot) \in C_b^\gamma H ( ⋅ ) ∈ C b γ required in Theorem 4.17.
Relation to BDSDEs
Backward Causality of Rough BSDE Solutions
We record a simple but important causality property of rough BSDEs. Unlike forward rough differential equations, where the solution up to time t t t depends only on the past rough path X [ 0 , t ] X_{[0,t]} X [ 0 , t ] , the solution of a rough BSDE on [ 0 , T ] [0,T] [ 0 , T ] depends, at time t t t , only on the future increments of the rough path on [ t , T ] [t,T] [ t , T ] .
For 0 ≤ t ≤ T 0 \le t \le T 0 ≤ t ≤ T , denote by ⋆ \star ⋆ the group product (also denoted by ⊗ \otimes ⊗ in some literature, see e.g. [14 ] , Chapter 7.2.1) and write
θ t X : = ( X u − 1 ⋆ X v ) t ≤ u ≤ v ≤ T : = ( δ X u , v , X u , v ) t ≤ u ≤ v ≤ T . \theta_t X := (X_u^{-1} \star X_v)_{t \le u \le v \le T} := (\delta X_{u,v},\mathbb{X}_{u,v})_{t \le u \le v \le T}. θ t X := ( X u − 1 ⋆ X v ) t ≤ u ≤ v ≤ T := ( δ X u , v , X u , v ) t ≤ u ≤ v ≤ T .
for the translated restriction of X X X to the time interval [ t , T ] [t,T] [ t , T ] . We shall say that a family of processes { ( Y X , Z X ) } X \{(Y^X,Z^X)\}_X {( Y X , Z X ) } X is X X X -backward causal if, for every t ∈ [ 0 , T ] t \in[0,T] t ∈ [ 0 , T ] and every two rough paths X , X ~ X,\widetilde{X} X , X satisfying
θ t X = θ t X ~ , \theta_t X = \theta_t \widetilde{X}, θ t X = θ t X ,
one has
Y s X = Y s X ~ for all s ∈ [ t , T ] a.s. , Z s X = Z s X ~ for Leb-a.e. s ∈ [ t , T ] a.s. Y_s^X = Y_s^{\widetilde{X}} \quad\text{for all } s \in[t,T]\ \text{a.s.}, \qquad Z_s^X = Z_s^{\widetilde{X}} \quad\text{for Leb-a.e. }s \in[t,T]\ \text{a.s.} Y s X = Y s X for all s ∈ [ t , T ] a.s. , Z s X = Z s X for Leb-a.e. s ∈ [ t , T ] a.s.
Equivalently, the restriction of the solution to [ t , T ] [t,T] [ t , T ] is a function only of the terminal value ξ \xi ξ , the coefficients f , H f,H f , H , the Brownian motion W W W , and the rough path increments θ t X \theta_t X θ t X .
The X X X -backward causal property is immediate from the uniqueness. Indeed, let ( Y X , Z X ) (Y^X,Z^X) ( Y X , Z X ) be the solution of (1) driven by X X X . For s ∈ [ t , T ] s \in[t,T] s ∈ [ t , T ] , the same pair restricted to [ t , T ] [t,T] [ t , T ] satisfies
Y s X = ξ + ∫ s T f ( r , Y r X , Z r X ) d r + ∫ s T H ( Y r X ) d X r − ∫ s T Z r X d W r . Y_s^X = \xi+ \int_s^T f(r,Y_r^X,Z_r^X)\,\mathrm{d}r + \int_s^T H(Y_r^X)\,\mathrm{d}X_r - \int_s^T Z_r^X\,\mathrm{d}W_r. Y s X = ξ + ∫ s T f ( r , Y r X , Z r X ) d r + ∫ s T H ( Y r X ) d X r − ∫ s T Z r X d W r .
The rough integral over [ s , T ] ⊂ [ t , T ] [s,T] \subset[t,T] [ s , T ] ⊂ [ t , T ] is defined entirely from the increments ( δ X u , v , X u , v ) (\delta X_{u,v},\mathbb{X}_{u,v}) ( δ X u , v , X u , v ) , t ≤ u ≤ v ≤ T t \le u \le v \le T t ≤ u ≤ v ≤ T . Hence, if θ t X = θ t X ~ \theta_t X = \theta_t \widetilde{X} θ t X = θ t X , then ( Y X , Z X ) ∣ [ t , T ] (Y^X,Z^X)|_{[t,T]} ( Y X , Z X ) ∣ [ t , T ] and ( Y X ~ , Z X ~ ) ∣ [ t , T ] (Y^{\widetilde{X}},Z^{\widetilde{X}})|_{[t,T]} ( Y X , Z X ) ∣ [ t , T ] solve the same rough BSDE on [ t , T ] [t,T] [ t , T ] , with the same terminal condition ξ \xi ξ , the same coefficients f , H f,H f , H , and the same driving restricted rough path. By Corollary 4.6, these two restrictions coincide.
Thus the solution can be represented in the following triangular form:
( Y s X , Z s X ) s ∈ [ t , T ] = Φ [ t , T ] ( ξ , f , H , W , θ t X ) . (Y_s^X,Z_s^X)_{s\in[t,T]} = \Phi_{[t,T]}(\xi,f,H,W,\theta_t X). ( Y s X , Z s X ) s ∈ [ t , T ] = Φ [ t , T ] ( ξ , f , H , W , θ t X ) .
where Φ [ t , T ] \Phi_{[t,T]} Φ [ t , T ] denotes the solution map of the rough BSDE on the interval [ t , T ] [t,T] [ t , T ] , mapping the terminal condition, coefficients, Brownian motion, and the shifted rough path to the corresponding solution pair. In particular, the solution is independent of the values of X X X before time t t t . This is the precise sense in which rough BSDEs are backward causal with respect to their rough drivers.
Randomization and Relation to BDSDEs
In this section, we show that the solution of rough BSDE eq:1 can be randomized and identified with a solution of the following backward doubly stochastic differential equation (BDSDE):
Y t = ξ + ∫ t T f ( r , Y r , Z r ) d r + ∫ t T H ( Y r ) ∘ d B ← r − ∫ t T Z r d W r , t ∈ [ 0 , T ] , (149) Y_t = \xi+ \int_t^T f(r,Y_r,Z_r)\,dr + \int_t^T H(Y_r) \circ d\overleftarrow{B}_r - \int_t^T Z_r\,dW_r,\quad t \in[0,T], \tag*{(149)} Y t = ξ + ∫ t T f ( r , Y r , Z r ) d r + ∫ t T H ( Y r ) ∘ d B r − ∫ t T Z r d W r , t ∈ [ 0 , T ] , (149)
where B is another Brownian motion that is independent of W, and
∫ t T H ( Y r ) ∘ d B ← r : = ∫ t T H ( Y r ) d B ← r + 1 2 ∫ t T D H ( Y r ) H ( Y r ) d r \int_t^T H(Y_r) \circ d\overleftarrow{B}_r := \int_t^T H(Y_r)\,d\overleftarrow{B}_r + \frac{1}{2}\int_t^T DH(Y_r)H(Y_r)\,dr ∫ t T H ( Y r ) ∘ d B r := ∫ t T H ( Y r ) d B r + 2 1 ∫ t T D H ( Y r ) H ( Y r ) d r
is a (backward) Stratonovich integral. Here, ∫ H ( Y r ) d B ← r \int H(Y_r)d\overleftarrow{B}_r ∫ H ( Y r ) d B r is a backward Itô integral, as defined in [19 ] , Chapter 3.4, which is an Itô integral with respect to the backward filtration F t , T ← : = σ { δ B s , T , s ∈ [ t , T ] } \mathcal{F}_{t,T}^{\leftarrow} := \sigma\{\delta B_{s,T},\ s \in[t,T]\} F t , T ← := σ { δ B s , T , s ∈ [ t , T ]} .
Since we are working with random rough paths lifted from Brownian motions, we assume our rough path X (denoted by X ∣ [ s , t ] \mathbf{X}_{|[s,t]} X ∣ [ s , t ] for its restriction on [ s , t ] [s,t] [ s , t ] ) to take values in the space of geometric rough paths C g 0 , 1 / p ( [ s , t ] , R d ) \mathcal{C}^{0,1/p}_{g}([s,t],\mathbb{R}^{d}) C g 0 , 1/ p ([ s , t ] , R d ) , which we equip with the Borel σ \sigma σ -algebra C [ s , t ] 1 / p \mathcal{C}^{1/p}_{[s,t]} C [ s , t ] 1/ p for any $0 \leq s < t \leq T. W e w r i t e . We write . W e w r i t e \mathcal{C}^{1/p}w h e n when w h e n s = 0a n d and an d t = T. F u r t h e r m o r e , w e d e n o t e b y . Furthermore, we denote by . F u r t h er m or e , w e d e n o t e b y \mathbb{P}t h e p r e d i c t a b l e the predictable t h e p r e d i c t ab l e \sigma− a l g e b r a o n -algebra on − a l g e b r a o n [0,T] \times \Omega generated by W, and thus a process is predictably measurable if and only if it is measurable with respect to \mathbb{P}. I n t h e f o l l o w i n g , . In the following, . I n t h e f o l l o w in g , \mathcal{B}(M)d e n o t e s t h e B o r e l denotes the Borel d e n o t es t h e B or e l \sigma$-algebra on a metric space M.
Proposition 5.1. Assume ( γ , p , q , q ′ ) (\gamma,p,q,q') ( γ , p , q , q ′ ) , ξ ∈ L ∞ \xi\in L^\infty ξ ∈ L ∞ , and ( f d e t ) (f_{\mathrm{det}}) ( f det ) hold. Let H ∈ C b γ ( R , R d ) H \in C_b^\gamma(\mathbb{R},\mathbb{R}^d) H ∈ C b γ ( R , R d ) . For any X = ( X , X ) ∈ C g 0 , 1 / p \mathbf{X} = (X,\mathbb{X}) \in\mathcal{C}^{0,1/p}_{g} X = ( X , X ) ∈ C g 0 , 1/ p , denote by ( Y X , Z X ) (Y^{\mathbf{X}},Z^{\mathbf{X}}) ( Y X , Z X ) the unique solution to Eq. eq:1 on [ 0 , T ] [0,T] [ 0 , T ] obtained by Theorem 4.17. Then, there exists a version ( Y ~ , Z ~ ) ( t , ω ; X ) (\widetilde{Y},\widetilde{Z})(t,\omega;\mathbf{X}) ( Y , Z ) ( t , ω ; X ) of ( Y X , Z X ) ( t , ω ) (Y^{\mathbf{X}},Z^{\mathbf{X}})(t,\omega) ( Y X , Z X ) ( t , ω ) in the sense that
(1) the map ( t , ω , X ) ↦ ( Y ~ , Z ~ ) ( t , ω ; X ) (t,\omega,\mathbf{X}) \mapsto(\widetilde{Y},\widetilde{Z})(t,\omega;\mathbf{X}) ( t , ω , X ) ↦ ( Y , Z ) ( t , ω ; X ) is B ( [ 0 , T ] ) ⊗ F T ⊗ C 1 / p \mathcal{B}([0,T]) \otimes\mathcal{F}_T \otimes\mathcal{C}^{1/p} B ([ 0 , T ]) ⊗ F T ⊗ C 1/ p -measurable, and ( F t ⊗ C 1 / p ) t (\mathcal{F}_t \otimes\mathcal{C}^{1/p})_t ( F t ⊗ C 1/ p ) t -adapted;
(2) for any X ∈ C g 0 , 1 / p \mathbf{X} \in\mathcal{C}^{0,1/p}_{g} X ∈ C g 0 , 1/ p , it holds that Z ~ ( t , ω ; X ) = Z X ( t , ω ) \widetilde{Z}(t,\omega;\mathbf{X}) = Z^{\mathbf{X}}(t,\omega) Z ( t , ω ; X ) = Z X ( t , ω ) a.e.-d t × d P dt \times d\mathbb{P} d t × d P , and that Y ~ ( t , ω ; X ) = Y X ( t , ω ) \widetilde{Y}(t,\omega;\mathbf{X}) = Y^{\mathbf{X}}(t,\omega) Y ( t , ω ; X ) = Y X ( t , ω ) for all t ∈ [ 0 , T ] t \in[0,T] t ∈ [ 0 , T ] , a.s.;
(3) ( Y ~ , Z ~ ) ( ⋅ ; X ) (\widetilde{Y},\widetilde{Z})(\cdot;\mathbf{X}) ( Y , Z ) ( ⋅ ; X ) is X \mathbf{X} X -backward causal in the sense that for any t ∈ [ 0 , T ) t \in[0,T) t ∈ [ 0 , T ) , and ( Y ~ , Z ~ ) (\widetilde{Y},\widetilde{Z}) ( Y , Z ) restricted on [ t , T ] [t,T] [ t , T ] , we have
Z ~ ( ⋅ ; X ) = Z ~ ( ⋅ ; θ t X ) a.e.-Leb [ t , T ] × d P ; \widetilde{Z}(\cdot;\mathbf{X}) = \widetilde{Z}(\cdot;\theta_t\mathbf{X})\quad\text{a.e.-Leb}_{[t,T]} \times d\mathbb{P}; Z ( ⋅ ; X ) = Z ( ⋅ ; θ t X ) a.e.-Leb [ t , T ] × d P ;
Y ~ ( s , ⋅ ; X ) = Y ~ ( s , ⋅ ; θ t X ) for all s ∈ [ t , T ] , a.s. \widetilde{Y}(s,\cdot;\mathbf{X}) = \widetilde{Y}(s,\cdot;\theta_t\mathbf{X})\quad\text{for all }s \in[t,T],\ \text{a.s.} Y ( s , ⋅ ; X ) = Y ( s , ⋅ ; θ t X ) for all s ∈ [ t , T ] , a.s.
Proof. The joint measurability can be established by adapting the argument of [12 , 2 ] , which proceeds in two steps: first, one proves the existence of a jointly measurable version when X \mathbf{X} X is smooth; second, one obtains a jointly measurable version of the original solution by an approximation argument. However, here we take a different approach motivated by [5 ] , Theorem IV.30. We equip the space C 1 / p \mathcal{C}^{1/p} C 1/ p with the filtration C : = ( C t ) t \mathcal{C} := (\mathcal{C}_t)_t C := ( C t ) t and σ \sigma σ -algebra C t : = σ ( { X s ∨ ⋅ − 1 ⋆ X T ∣ X ∈ C 1 / p , T − t ≤ s ≤ T } ) \mathcal{C}_t := \sigma(\{\mathbf{X}_{s\vee\cdot}^{-1}\star\mathbf{X}_T \mid\mathbf{X} \in\mathcal{C}^{1/p}, T-t \leq s \leq T\}) C t := σ ({ X s ∨⋅ − 1 ⋆ X T ∣ X ∈ C 1/ p , T − t ≤ s ≤ T }) .
Consider the mapping
( t , X ) ↦ ( Y X , Z X ) ( T − t ∨ ⋅ ) ∈ L 0 ( [ 0 , T ] × Ω , P , L e b × P ; R 1 + d ) , (t,\mathbf{X}) \mapsto(Y^{\mathbf{X}},Z^{\mathbf{X}})(T-t\vee\cdot) \in L^0([0,T] \times\Omega,\mathbb{P},\mathrm{Leb} \times\mathbb{P};\mathbb{R}^{1+d}), ( t , X ) ↦ ( Y X , Z X ) ( T − t ∨ ⋅ ) ∈ L 0 ([ 0 , T ] × Ω , P , Leb × P ; R 1 + d ) ,
with L 0 ( [ 0 , T ] × Ω , P , L e b × P ; R 1 + d ) L^0([0,T] \times\Omega,\mathbb{P},\mathrm{Leb} \times\mathbb{P};\mathbb{R}^{1+d}) L 0 ([ 0 , T ] × Ω , P , Leb × P ; R 1 + d ) equipped with the weak topology and the trivial filtration ( P t ) t : = ( P t ) t (\mathcal{P}_t)_t := (\mathbb{P}_t)_t ( P t ) t := ( P t ) t .
By Theorem 4.19, we see that the above mapping is continuous, and thus ( B ( [ 0 , T ] ) ⊗ C 1 / p ) / B ( L 0 ) (\mathcal{B}([0,T]) \otimes\mathcal{C}^{1/p})/\mathcal{B}(L^0) ( B ([ 0 , T ]) ⊗ C 1/ p ) / B ( L 0 ) -measurable. Furthermore, it is easy to check that the mapping is C-progressively measurable.
Moreover, noting that P \mathbb{P} P is generated by the Brownian motion W W W , and thus it is separable. Then according to [6 ] with ( Ω , F , F ) (\Omega,\mathcal{F},\mathbb{F}) ( Ω , F , F ) and ( Ω ′ , G , G , P ) (\Omega',\mathcal{G},\mathbb{G},\mathbb{P}) ( Ω ′ , G , G , P ) therein taken as ( C g 0 , 1 / p , C 1 / p , C ) (\mathcal{C}^{0,1/p}_{g},\mathcal{C}^{1/p},\mathcal{C}) ( C g 0 , 1/ p , C 1/ p , C ) and ( [ 0 , T ] × Ω , P , ( P ) t , L e b × P ) ([0,T]\times\Omega,\mathcal{P},(\mathcal{P})_t,\mathrm{Leb}\times\mathbb{P}) ([ 0 , T ] × Ω , P , ( P ) t , Leb × P ) respectively, there exists a ( B ( [ 0 , T ] ) ⊗ C 1 / p ⊗ P ) / B ( R 1 + d ) (\mathcal{B}([0,T])\otimes\mathcal{C}^{1/p}\otimes\mathcal{P})/\mathcal{B}(\mathbb{R}^{1+d}) ( B ([ 0 , T ]) ⊗ C 1/ p ⊗ P ) / B ( R 1 + d ) -measurable, C ⊗ ( P ) t \mathcal{C}\otimes(\mathcal{P})_t C ⊗ ( P ) t -adapted process
( Y ‾ , Z ‾ ) : [ 0 , T ] × C g 0 , 1 / p × [ 0 , T ] × Ω → R 1 + d , (\overline{Y},\overline{Z}):[0,T]\times\mathcal{C}^{0,1/p}_{g}\times[0,T]\times\Omega\to\mathbb{R}^{1+d}, ( Y , Z ) : [ 0 , T ] × C g 0 , 1/ p × [ 0 , T ] × Ω → R 1 + d ,
such that for any ( t , X ) (t,X) ( t , X ) ,
( Y ‾ , Z ‾ ) ( t , X , ⋅ ) = ( Y X , Z X ) ( ( T − t ) ∨ ⋅ , ⋅ ) in L 0 ( [ 0 , T ] × Ω , P , L e b × P ) . (\overline{Y},\overline{Z})(t,X,\cdot)=(Y^{X},Z^{X})((T-t)\vee\cdot,\cdot)\quad\text{in }L^{0}([0,T]\times\Omega,\mathcal{P},\mathrm{Leb}\times\mathbb{P}). ( Y , Z ) ( t , X , ⋅ ) = ( Y X , Z X ) (( T − t ) ∨ ⋅ , ⋅ ) in L 0 ([ 0 , T ] × Ω , P , Leb × P ) .
Now let { ε n } n ≥ 1 \{\varepsilon_n\}_{n\geq1} { ε n } n ≥ 1 be a sequence of strictly positive numbers that converges to 0 0 0 . We take
( Y ~ , Z ~ ) ( t , ω , X ) : = lim sup n → ∞ 1 ε n ∫ t t + ε n ( Y ‾ , Z ‾ ) ( T − t , X , s , ω ) d s , (150) (\widetilde{Y},\widetilde{Z})(t,\omega,X):=\limsup_{n\to\infty}\frac{1}{\varepsilon_n}\int_t^{t+\varepsilon_n}(\overline{Y},\overline{Z})(T-t,X,s,\omega)\,\mathrm{d}s, \tag*{(150)} ( Y , Z ) ( t , ω , X ) := n → ∞ lim sup ε n 1 ∫ t t + ε n ( Y , Z ) ( T − t , X , s , ω ) d s , (150)
and thus ( Y ~ , Z ~ ) (\widetilde{Y},\widetilde{Z}) ( Y , Z ) is ( C T − t ⊗ F t + W ) t (\mathcal{C}_{T-t}\otimes\mathcal{F}^{W}_{t+})_t ( C T − t ⊗ F t + W ) t -adapted. Moreover, for any fixed X X X , and s ≥ t s\geq t s ≥ t ,
( Y ‾ , Z ‾ ) ( T − t , X , s , ω ) = ( Y , Z ) X ( t ∨ s , ω ) = ( Y X , Z X ) ( s , ω ) , d s ⊗ d P -a.s. (\overline{Y},\overline{Z})(T-t,X,s,\omega)=(Y,Z)^{X}(t\vee s,\omega)=(Y^{X},Z^{X})(s,\omega),\quad\mathrm{d}s\otimes\mathrm{d}\mathbb{P}\text{-a.s.} ( Y , Z ) ( T − t , X , s , ω ) = ( Y , Z ) X ( t ∨ s , ω ) = ( Y X , Z X ) ( s , ω ) , d s ⊗ d P -a.s.
which implies, by Lebesgue’s differentiation theorem, ( Y ~ , Z ~ ) ( t , ω , X ) = ( Y , Z ) X ( t , ω ) (\widetilde{Y},\widetilde{Z})(t,\omega,X)=(Y,Z)^{X}(t,\omega) ( Y , Z ) ( t , ω , X ) = ( Y , Z ) X ( t , ω ) , d t ⊗ d P \mathrm{d}t\otimes\mathrm{d}\mathbb{P} d t ⊗ d P -a.s. Thus it is easy to check all the three conditions from the proposition hold. □
In the following, we replace the rough path X \mathbf{X} X by the randomized rough path B \mathbf{B} B . For simplicity of notation, we work with the product space. Consider
( Ω , G , P ) = ( Ω ′ , G ′ , P ′ ) ⊗ ( Ω ′ ′ , G ′ ′ , P ′ ′ ) . (151) (\Omega,\mathcal{G},\mathbb{P})=(\Omega',\mathcal{G}',\mathbb{P}')\otimes(\Omega'',\mathcal{G}'',\mathbb{P}''). \tag*{(151)} ( Ω , G , P ) = ( Ω ′ , G ′ , P ′ ) ⊗ ( Ω ′′ , G ′′ , P ′′ ) . (151)
where ( Ω ′ , G ′ , P ′ ) (\Omega',\mathcal{G}',\mathbb{P}') ( Ω ′ , G ′ , P ′ ) supports a d d d -dimensional Brownian motion W W W , P ′ \mathcal{P}' P ′ is the predictable σ \sigma σ -algebra, and F ′ \mathcal{F}' F ′ is the augmented filtration of F W \mathcal{F}^{W} F W ; ( Ω ′ ′ , G ′ ′ , P ′ ′ ) (\Omega'',\mathcal{G}'',\mathbb{P}'') ( Ω ′′ , G ′′ , P ′′ ) is a probability space with a d d d -dimensional Brownian motion B B B . Moreover, we denote by F t , T ′ ′ : = σ ( B s − B t ∣ s ∈ [ t , T ] ) \mathcal{F}^{\prime\prime}_{t,T}:=\sigma(B_s-B_t\mid s\in[t,T]) F t , T ′′ := σ ( B s − B t ∣ s ∈ [ t , T ]) , F t ′ ′ : = F 0 , t ′ ′ \mathcal{F}^{\prime\prime}_t:=\mathcal{F}^{\prime\prime}_{0,t} F t ′′ := F 0 , t ′′ , and F t : = F t ′ ∨ F t , T ′ ′ \mathcal{F}_t:=\mathcal{F}'_t\vee\mathcal{F}^{\prime\prime}_{t,T} F t := F t ′ ∨ F t , T ′′ the standard non-monotone family of σ \sigma σ -algebras for BDSDEs. In the following, we write ω = ( ω ′ , ω ′ ′ ) \omega=(\omega',\omega'') ω = ( ω ′ , ω ′′ ) .
In view of [10 ] , for each ( s , t ) ∈ Δ T (s,t)\in\Delta_T ( s , t ) ∈ Δ T , define B s , t I t o ^ : = ∫ s t δ B s , r d B r \mathbb{B}^{\mathrm{It\hat{o}}}_{s,t}:=\int_s^t\delta B_{s,r}\,\mathrm{d}B_r B s , t It o ^ := ∫ s t δ B s , r d B r as an Itô integral and B s , t S t r a t o : = ∫ s t δ B s , r ∘ d B r = B s , t I t o ^ + 1 2 ( t − s ) I \mathbb{B}^{\mathrm{Strato}}_{s,t}:=\int_s^t\delta B_{s,r}\circ\mathrm{d}B_r=\mathbb{B}^{\mathrm{It\hat{o}}}_{s,t}+\frac{1}{2}(t-s)I B s , t Strato := ∫ s t δ B s , r ∘ d B r = B s , t It o ^ + 2 1 ( t − s ) I as a Stratonovich integral. It is well-understood that B S t r a t o ( ω ′ ′ ) = ( B ( ω ′ ′ ) , B S t r a t o ( ω ′ ′ ) ) ∈ ⋂ α ∈ ( 1 / 3 , 1 / 2 ) C g 0 , α \mathbf{B}^{\mathrm{Strato}}(\omega'')=(B(\omega''),\mathbb{B}^{\mathrm{Strato}}(\omega''))\in\bigcap_{\alpha\in(1/3,1/2)}\mathcal{C}^{0,\alpha}_{g} B Strato ( ω ′′ ) = ( B ( ω ′′ ) , B Strato ( ω ′′ )) ∈ ⋂ α ∈ ( 1/3 , 1/2 ) C g 0 , α for all ω ′ ′ ∈ N 1 c \omega''\in N_1^c ω ′′ ∈ N 1 c , where N 1 N_1 N 1 is a P ′ ′ \mathbb{P}'' P ′′ -null set of Ω ′ ′ \Omega'' Ω ′′ . Moreover, for any t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] , the lifting mapping
B : = B S t r a t o : [ 0 , T ] × Ω ′ ′ → ( C g 0 , α , C 1 / p ) \mathbf{B}:=\mathbf{B}^{\mathrm{Strato}}:[0,T]\times\Omega''\to(\mathcal{C}^{0,\alpha}_{g},\mathcal{C}^{1/p}) B := B Strato : [ 0 , T ] × Ω ′′ → ( C g 0 , α , C 1/ p )
( t , ω ′ ′ ) ↦ B S t r a t o ( ω ′ ′ ) ⋅ ∧ t (152) (t,\omega'')\mapsto\mathbf{B}^{\mathrm{Strato}}(\omega'')_{\cdot\wedge t} \tag*{(152)} ( t , ω ′′ ) ↦ B Strato ( ω ′′ ) ⋅∧ t (152)
is progressively measurable w.r.t. ( F t ′ ′ ) t (\mathcal{F}^{\prime\prime}_t)_t ( F t ′′ ) t . For the backward Stratonovich integral and rough integral, we have
Proposition 5.2. Suppose that ( Z , Z ′ ) (Z,Z') ( Z , Z ′ ) defined on [ 0 , T ] × Ω ′ × C g 0 , 1 / p [0,T]\times\Omega'\times\mathcal{C}^{0,1/p}_{g} [ 0 , T ] × Ω ′ × C g 0 , 1/ p is B ( [ 0 , T ] ) ⊗ F t ⊗ C 1 / p \mathcal{B}([0,T])\otimes\mathcal{F}_t\otimes\mathcal{C}^{1/p} B ([ 0 , T ]) ⊗ F t ⊗ C 1/ p -measurable and ( F t ′ ⊗ C 1 / p ) t (\mathcal{F}'_t\otimes\mathcal{C}^{1/p})_t ( F t ′ ⊗ C 1/ p ) t -adapted satisfying the following conditions:
(1) *for any X ∈ C g 0 , 1 / p X\in\mathcal{C}^{0,1/p}_{g} X ∈ C g 0 , 1/ p , ( Z , Z ′ ) X ( ⋅ ) : = ( Z , Z ′ ) ( ⋅ ; X ) ∈ R S M X ( p , p ′ ) -var ( Ω ′ ) (Z,Z')^{X}(\cdot):=(Z,Z')(\cdot\,;X)\in\mathrm{RSM}^{(p,p')\text{-var}}_{X}(\Omega') ( Z , Z ′ ) X ( ⋅ ) := ( Z , Z ′ ) ( ⋅ ; X ) ∈ RSM X ( p , p ′ ) -var ( Ω ′ ) , where p ′ ≥ p ≥ 2 p'\geq p\geq2 p ′ ≥ p ≥ 2 such that 2 / p + 1 / p ′ > 1 2/p+1/p'>1 2/ p + 1/ p ′ > 1 ;
(2) ( Z , Z ′ , M Z ) X (Z,Z',M^{Z})^{X} ( Z , Z ′ , M Z ) X is X X X -backward causal.* Then for any t ∈ [ 0 , T ] t \in[0,T] t ∈ [ 0 , T ] ,
( ∫ t T ( Z , Z ′ ) X ( ω ′ ) d X ) ∣ X = B ( ω ′ ′ ) = ∫ t T Z r B ( ω ′ ′ ) ( ω ′ ) d B ← r ( ω ′ ′ ) − 1 2 ∫ t T ( Z ′ ) r B ( ω ′ ′ ) ( ω ′ ) d r = : ∫ t T Z r B ( ω ′ ′ ) ( ω ′ ) ∘ d B ← r ( ω ′ ′ ) , P -a.s. (153) \begin{aligned}
\left.\left(\int_t^T (Z,Z')^{\mathbf X}(\omega')d\mathbf X\right)\right|_{\mathbf X=\mathbf B(\omega'')}=\int_t^T Z_r^{\mathbf B(\omega'')}(\omega')d\overleftarrow{B}_r(\omega'')-\frac{1}{2}\int_t^T (Z')_r^{\mathbf B(\omega'')}(\omega')dr \\
&=:\int_t^T Z_r^{\mathbf B(\omega'')}(\omega')\circ d\overleftarrow{B}_r(\omega''),\quad\mathbb P\text{-a.s.} \tag*{(153)}
\end{aligned} ( ∫ t T ( Z , Z ′ ) X ( ω ′ ) d X ) X = B ( ω ′′ ) = ∫ t T Z r B ( ω ′′ ) ( ω ′ ) d B r ( ω ′′ ) − 2 1 ∫ t T ( Z ′ ) r B ( ω ′′ ) ( ω ′ ) d r =: ∫ t T Z r B ( ω ′′ ) ( ω ′ ) ∘ d B r ( ω ′′ ) , P -a.s. (153)
Proof. Let ( Z ˉ , Z ˉ ′ ) ( ω ′ , ω ′ ′ ) : = ( Z , Z ′ ) B ( ω ′ ′ ) ( ω ′ ) (\bar Z,\bar Z')(\omega',\omega'') := (Z,Z')^{\mathbf B(\omega'')}(\omega') ( Z ˉ , Z ˉ ′ ) ( ω ′ , ω ′′ ) := ( Z , Z ′ ) B ( ω ′′ ) ( ω ′ ) . Since ( Z , Z ′ ) X (Z,Z')^{\mathbf X} ( Z , Z ′ ) X is ( F t ′ ⊗ C 1 / p ) t (\mathcal F'_t\otimes\mathcal C^{1/p})_t ( F t ′ ⊗ C 1/ p ) t -adapted, continuous, and X \mathbf X X -backward causal, Z ˉ \bar Z Z ˉ is ( F t ) t (\mathcal F_t)_t ( F t ) t -adapted, and thus the backward Stratonovich integral on the left-hand side of (153) , is well-defined in the sense that
∫ t T Z ˉ s ∘ d B ← s = lim ∣ P ∣ → 0 ∑ [ u , v ] ∈ P ( Z ˉ v δ B u , v − 1 2 Z ˉ u ′ ( v − u ) ) , in P -probability , \int_t^T \bar Z_s\circ d\overleftarrow{B}_s=\lim_{|\mathcal P|\to0}\sum_{[u,v]\in\mathcal P}\left(\bar Z_v\delta B_{u,v}-\frac{1}{2}\bar Z'_u(v-u)\right),\quad\text{in }\mathbb P\text{-probability}, ∫ t T Z ˉ s ∘ d B s = ∣ P ∣ → 0 lim [ u , v ] ∈ P ∑ ( Z ˉ v δ B u , v − 2 1 Z ˉ u ′ ( v − u ) ) , in P -probability ,
where P \mathcal P P is any interval partition of [ t , T ] [t,T] [ t , T ] .
On the other hand, by a modification of [13 ] , Theorem 3.4, there exists an F T ⊗ C 1 / p \mathcal F_T\otimes\mathcal C^{1/p} F T ⊗ C 1/ p -measurable, X \mathbf X X -backward causal version of ∫ t T ( Z , Z ′ ) X d X \int_t^T (Z,Z')^{\mathbf X}d\mathbf X ∫ t T ( Z , Z ′ ) X d X such that, for every X \mathbf X X , it coincides with ∫ t T ( Z , Z ′ ) X d X \int_t^T (Z,Z')^{\mathbf X}d\mathbf X ∫ t T ( Z , Z ′ ) X d X P ′ \mathbb P' P ′ -almost surely. We still denote this version by ∫ t T ( Z , Z ′ ) X d X \int_t^T (Z,Z')^{\mathbf X}d\mathbf X ∫ t T ( Z , Z ′ ) X d X . Then by [12 ] , Corollary 8.7, we have
∫ t T ( Z , Z ′ ) X d X ∣ X = B = lim ∣ P ∣ → 0 ∑ [ u , v ] ∈ P ( Z ˉ u δ B u , v + Z ˉ u ′ B u , v I t o ^ ) , in P -probability . \left.\int_t^T (Z,Z')^{\mathbf X}d\mathbf X\right|_{\mathbf X=\mathbf B}=\lim_{|\mathcal P|\to0}\sum_{[u,v]\in\mathcal P}\left(\bar Z_u\delta B_{u,v}+\bar Z'_u\mathbb B^{\mathsf{It\hat{o}}}_{u,v}\right),\quad\text{in }\mathbb P\text{-probability}. ∫ t T ( Z , Z ′ ) X d X X = B = ∣ P ∣ → 0 lim [ u , v ] ∈ P ∑ ( Z ˉ u δ B u , v + Z ˉ u ′ B u , v It o ^ ) , in P -probability .
To complete the proof, we only need to show the difference of right-hand sides of the above two identities
∑ [ u , v ] ∈ P ( δ Z ˉ u , v δ B u , v − Z ˉ u ′ B u , v − 1 2 Z ˉ u ′ ( v − u ) ) = ∑ [ u , v ] ∈ P [ Z ˉ u ′ ( δ B u , v δ B u , v − ( v − u ) ) ] − ∑ [ u , v ] ∈ P Z ˉ u ′ B I t o ^ u , v + ∑ [ u , v ] ∈ P R u , v Z ˉ δ B u , v = : ∑ [ u , v ] ∈ P [ Z ˉ u ′ ( δ B u , v δ B u , v − ( v − u ) ) ] − ∑ [ u , v ] ∈ P Z ˉ v ′ B I t o ^ u , v + ∑ [ u , v ] ∈ P R u , v Z ˉ δ B u , v + ∑ [ u , v ] ∈ P Δ u , v (154) \begin{aligned}
\sum_{[u,v]\in\mathcal P}\left(\delta\bar Z_{u,v}\delta B_{u,v}-\bar Z'_u\mathbb B_{u,v}-\frac{1}{2}\bar Z'_u(v-u)\right) \\
&=\sum_{[u,v]\in\mathcal P}\left[\bar Z'_u(\delta B_{u,v}\delta B_{u,v}-(v-u))\right]-\sum_{[u,v]\in\mathcal P}\bar Z'_u\mathbb B^{\mathsf{It\hat{o}}_{u,v}+\sum_{[u,v]\in\mathcal P}R^{\bar Z}_{u,v}\delta B_{u,v}} \\
&=:\sum_{[u,v]\in\mathcal P}\left[\bar Z'_u(\delta B_{u,v}\delta B_{u,v}-(v-u))\right]-\sum_{[u,v]\in\mathcal P}\bar Z'_v\mathbb B^{\mathsf{It\hat{o}}_{u,v}+\sum_{[u,v]\in\mathcal P}R^{\bar Z}_{u,v}\delta B_{u,v}+\sum_{[u,v]\in\mathcal P}\Delta_{u,v}} \tag*{(154)}
\end{aligned} [ u , v ] ∈ P ∑ ( δ Z ˉ u , v δ B u , v − Z ˉ u ′ B u , v − 2 1 Z ˉ u ′ ( v − u ) ) = [ u , v ] ∈ P ∑ [ Z ˉ u ′ ( δ B u , v δ B u , v − ( v − u )) ] − [ u , v ] ∈ P ∑ Z ˉ u ′ B It o ^ u , v + ∑ [ u , v ] ∈ P R u , v Z ˉ δ B u , v =: [ u , v ] ∈ P ∑ [ Z ˉ u ′ ( δ B u , v δ B u , v − ( v − u )) ] − [ u , v ] ∈ P ∑ Z ˉ v ′ B It o ^ u , v + ∑ [ u , v ] ∈ P R u , v Z ˉ δ B u , v + ∑ [ u , v ] ∈ P Δ u , v (154)
converges to zero in P \mathbb P P -probability as ∣ P ∣ → 0 |\mathcal P|\to0 ∣ P ∣ → 0 . It is standard to show ∑ [ u , v ] ∈ P Δ u , v → 0 \sum_{[u,v]\in\mathcal P}\Delta_{u,v}\to0 ∑ [ u , v ] ∈ P Δ u , v → 0 in P \mathbb P P -probability. Moreover, since Z ˉ ′ \bar Z' Z ˉ ′ is ( F t ) t (\mathcal F_t)_t ( F t ) t -adapted, we have that the first summation of (154) converges to zero. Similarly, the second summation also converges to zero. It remains to show the third term converges to zero. Indeed, noting that R u , v Z ˉ = P u , v Z ˉ + δ M u , v Z ˉ R^{\bar Z}_{u,v}=P^{\bar Z}_{u,v}+\delta M^{\bar Z}_{u,v} R u , v Z ˉ = P u , v Z ˉ + δ M u , v Z ˉ , the third summation equals
∑ [ u , v ] ∈ P P u , v Z ˉ δ B u , v + ∑ [ u , v ] ∈ P δ M u , v Z ˉ δ B u , v . (155) \sum_{[u,v]\in\mathcal P}P^{\bar Z}_{u,v}\delta B_{u,v}+\sum_{[u,v]\in\mathcal P}\delta M^{\bar Z}_{u,v}\delta B_{u,v}. \tag*{(155)} [ u , v ] ∈ P ∑ P u , v Z ˉ δ B u , v + [ u , v ] ∈ P ∑ δ M u , v Z ˉ δ B u , v . (155)
For the first summation of (155) , since P Z ( ω ′ ; X ) ∈ C p p ′ p + p ′ -var P^Z(\omega';\mathbf X)\in C^{\frac{pp'}{p+p'}\text{-var}} P Z ( ω ′ ; X ) ∈ C p + p ′ p p ′ -var , P ′ \mathbb P' P ′ -a.s., we have
∑ [ u , v ] ∈ P P u , v Z ( ω ′ ; X ) δ X u , v → 0 , P ′ -a.s. \sum_{[u,v]\in\mathcal P}P^Z_{u,v}(\omega';\mathbf X)\delta X_{u,v}\to0,\quad\mathbb P'\text{-a.s.} [ u , v ] ∈ P ∑ P u , v Z ( ω ′ ; X ) δ X u , v → 0 , P ′ -a.s.
and thus the term involving P Z ˉ ( ω ) = P Z ( ω ′ ; X ) ∣ X = B ( ω ′ ′ ) P\bar Z(\omega)=P^Z(\omega';\mathbf X)|_{\mathbf X=\mathbf B(\omega'')} P Z ˉ ( ω ) = P Z ( ω ′ ; X ) ∣ X = B ( ω ′′ ) times δ B \delta B δ B converges to zero P \mathbb P P -a.s. For the second summation of (155) , without loss of generality, assume that δ M Z ( ω ′ ; X ) \delta M^Z(\omega';\mathbf X) δ M Z ( ω ′ ; X ) is a square-integrable martingale on ( Ω ′ , F ′ , P ′ ) (\Omega',\mathcal F',\mathbb P') ( Ω ′ , F ′ , P ′ ) ; otherwise, one may apply a localization argument first. By the Burkholder-Davis-Gundy inequality, we have
E ∣ ∑ [ u , v ] ∈ P δ M u , v Z δ X u , v ∣ 2 ≲ ∑ [ u , v ] ∈ P ∣ δ X u , v ∣ 2 E ∣ δ ⟨ M Z ⟩ u , v ∣ ≲ sup [ u , v ] ∈ P ∣ δ X u , v ∣ 2 E ∣ δ ⟨ M Z ⟩ 0 , T ∣ → 0. \mathbb{E}\left|\sum_{[u,v]\in\mathcal{P}}\delta M^Z_{u,v}\delta X_{u,v}\right|^2 \lesssim\sum_{[u,v]\in\mathcal{P}}|\delta X_{u,v}|^2\mathbb{E}|\delta\langle M^Z\rangle_{u,v}| \lesssim\sup_{[u,v]\in\mathcal{P}}|\delta X_{u,v}|^2\mathbb{E}|\delta\langle M^Z\rangle_{0,T}| \to0. E [ u , v ] ∈ P ∑ δ M u , v Z δ X u , v 2 ≲ [ u , v ] ∈ P ∑ ∣ δ X u , v ∣ 2 E ∣ δ ⟨ M Z ⟩ u , v ∣ ≲ [ u , v ] ∈ P sup ∣ δ X u , v ∣ 2 E ∣ δ ⟨ M Z ⟩ 0 , T ∣ → 0.
It follows that the second summation of (155) converges to zero in P \mathbb{P} P -probability as well. □
Now we are ready to establish the equivalence between rough BSDEs and BDSDEs. In view of [24 ] , we call a pair ( Y , Z ) (Y,Z) ( Y , Z ) a square-integrable solution to (149) if ( Y , Z ) (Y,Z) ( Y , Z ) satisfies the equation and ( Y , Z ) ∈ S 2 ( [ 0 , T ] × Ω ) × L 2 ( [ 0 , T ] × Ω ) (Y,Z)\in S^2([0,T]\times\Omega)\times L^2([0,T]\times\Omega) ( Y , Z ) ∈ S 2 ([ 0 , T ] × Ω ) × L 2 ([ 0 , T ] × Ω ) , where S 2 S^2 S 2 is the set of continuous adapted processes y y y satisfying ∥ y ∥ S 2 2 : = E [ sup t ∈ [ 0 , T ] ∣ y t ∣ 2 ] < ∞ \lVert y\rVert_{S^2}^2:=\mathbb{E}[\sup_{t\in[0,T]}|y_t|^2]<\infty ∥ y ∥ S 2 2 := E [ sup t ∈ [ 0 , T ] ∣ y t ∣ 2 ] < ∞ .
Theorem 5.3. Under the same assumption as in Proposition 5.1 but with the probability space ( Ω , F , P ) (\Omega,\mathcal{F},\mathbb{P}) ( Ω , F , P ) there replaced by ( Ω ′ , F ′ , P ′ ) (\Omega',\mathcal{F}',\mathbb{P}') ( Ω ′ , F ′ , P ′ ) . Let ( Y X , Z X ) (Y^X,Z^X) ( Y X , Z X ) be the jointly measurable, X X X -backward causal version obtained in Proposition 5.1. Denote by ( Y ˉ , Z ˉ ) ( ω ) : = ( Y B ( ω ′ ′ ) ( ω ′ ) , Z B ( ω ′ ′ ) ( ω ′ ) ) (\bar{Y},\bar{Z})(\omega):=(Y^{B(\omega'')}(\omega'),Z^{B(\omega'')}(\omega')) ( Y ˉ , Z ˉ ) ( ω ) := ( Y B ( ω ′′ ) ( ω ′ ) , Z B ( ω ′′ ) ( ω ′ )) . Then ( Y ˉ , Z ˉ ) (\bar{Y},\bar{Z}) ( Y ˉ , Z ˉ ) is the unique L 2 L^2 L 2 -integrable solution to BDSDE (149) .
Proof. We complete the proof in two steps. First, we show that ( Y ˉ , Z ˉ ) (\bar{Y},\bar{Z}) ( Y ˉ , Z ˉ ) satisfies (149) P \mathbb{P} P -almost surely. We then prove that ( Y ˉ , Z ˉ ) ∈ S 2 × L 2 (\bar{Y},\bar{Z})\in S^2\times L^2 ( Y ˉ , Z ˉ ) ∈ S 2 × L 2 . It therefore follows from [24 ] , Theorem 1.1 that ( Y ˉ , Z ˉ ) (\bar{Y},\bar{Z}) ( Y ˉ , Z ˉ ) is the unique square-integrable solution.
Step 1. Since ( Y ˉ , Z ˉ ) (\bar{Y},\bar{Z}) ( Y ˉ , Z ˉ ) is ( F t ) (\mathcal{F}_t) ( F t ) -adapted, we may argue exactly as in the proof of [12 ] , Proposition 8.2 to obtain
∫ t T Z r X d W r ∣ X = B ( ω ′ ′ ) = ∫ t T Z ˉ r d W r , ∫ t T f ( r , Y r X , Z r X ) d r ∣ X = B ( ω ′ ′ ) = ∫ t T f ( r , Y ˉ r , Z ˉ r ) d r . \left.\int_t^T Z_r^X\,dW_r\right|_{X=B(\omega'')}=\int_t^T\bar{Z}_r\,dW_r,\qquad\left.\int_t^T f(r,Y_r^X,Z_r^X)\,dr\right|_{X=B(\omega'')}=\int_t^T f(r,\bar{Y}_r,\bar{Z}_r)\,dr. ∫ t T Z r X d W r X = B ( ω ′′ ) = ∫ t T Z ˉ r d W r , ∫ t T f ( r , Y r X , Z r X ) d r X = B ( ω ′′ ) = ∫ t T f ( r , Y ˉ r , Z ˉ r ) d r .
To show that ( Y ˉ , Z ˉ ) (\bar{Y},\bar{Z}) ( Y ˉ , Z ˉ ) satisfies (149) , it suffices to verify the following identity
∫ t T H ( Y r X ) d X r ∣ X = B ( ω ′ ′ ) = ∫ t T H ( Y ˉ r ) ∘ d B ← r , \left.\int_t^T H(Y_r^X)\,dX_r\right|_{X=B(\omega'')}=\int_t^T H(\bar{Y}_r)\circ d\overleftarrow{B}_r, ∫ t T H ( Y r X ) d X r X = B ( ω ′′ ) = ∫ t T H ( Y ˉ r ) ∘ d B r ,
which indeed follows from Proposition 5.2 by properly choosing ( p , p ′ ) (p,p') ( p , p ′ ) .
Step 2. Now we prove the square-integrability of ( Y B , Z B ) (Y^B,Z^B) ( Y B , Z B ) . By the proof of Theorem 4.9, we have that Y X Y^X Y X can be dominated by the solutions of RDEs. By Proposition A.4, the uniform norm of these solutions of RDEs is bounded by a constant multiple of 1 + ∣ δ X ∣ 1 / p 1+|\delta X|_{1/p} 1 + ∣ δ X ∣ 1/ p ; hence the same bound applies to Y X Y^X Y X . Consequently, we have the following estimate for the randomization Y B Y^B Y B :
ess sup ω ′ ∈ Ω ′ { sup t ∈ [ 0 , T ] ∣ Y t B ( ω ′ ′ ) ( ω ′ ) ∣ } ≲ 1 + ∣ δ B ∣ 1 / p ( ω ′ ′ ) , for all ω ′ ′ ∈ Ω ′ ′ . \operatorname*{ess\,sup}_{\omega'\in\Omega'}\left\{\sup_{t\in[0,T]}|Y_t^{B(\omega'')}(\omega')|\right\}\lesssim1+|\delta B|_{1/p}(\omega''),\qquad\text{for all }\omega''\in\Omega''. ω ′ ∈ Ω ′ ess sup { t ∈ [ 0 , T ] sup ∣ Y t B ( ω ′′ ) ( ω ′ ) ∣ } ≲ 1 + ∣ δ B ∣ 1/ p ( ω ′′ ) , for all ω ′′ ∈ Ω ′′ .
Then, the exponential integrability of ∣ δ B ∣ 1 / p |\delta B|_{1/p} ∣ δ B ∣ 1/ p (see, e.g., [14 ] , Corollary 13.14) implies that
E [ sup t ∈ [ 0 , T ] ∣ Y t B ∣ 2 ] < ∞ . \mathbb{E}\left[\sup_{t\in[0,T]}|Y_t^B|^2\right]<\infty. E [ t ∈ [ 0 , T ] sup ∣ Y t B ∣ 2 ] < ∞.
For the integrability of ∫ 0 T ∣ Z r B ∣ 2 d r \int_0^T|Z_r^B|^2\,dr ∫ 0 T ∣ Z r B ∣ 2 d r , by Itô’s formula (see, e.g., [26 ] , Lemma 2.1), it holds that
E W [ ∣ Y 0 B ∣ 2 + ∫ 0 T ∣ Z r B ∣ 2 d r ] = E W [ ∣ ξ ∣ 2 + ∫ 0 T ( Y r B ( 2 f ( r , Y r B , Z r B ) + D H ( Y r B ) H ( Y r B ) ) + ∣ H ( Y r B ) ∣ 2 ) d r + 2 ∫ 0 T Y r B H ( Y r B ) d B ← r ] . \mathbb{E}^{W}\left[|Y_0^B|^2+\int_0^T|Z_r^B|^2\,dr\right]
=\mathbb{E}^{W}\left[|\xi|^2+\int_0^T\left(Y_r^B\left(2f(r,Y_r^B,Z_r^B)+DH(Y_r^B)H(Y_r^B)\right)+|H(Y_r^B)|^2\right)\,dr+2\int_0^T Y_r^BH(Y_r^B)\,d\overleftarrow{B}_r\right]. E W [ ∣ Y 0 B ∣ 2 + ∫ 0 T ∣ Z r B ∣ 2 d r ] = E W [ ∣ ξ ∣ 2 + ∫ 0 T ( Y r B ( 2 f ( r , Y r B , Z r B ) + D H ( Y r B ) H ( Y r B ) ) + ∣ H ( Y r B ) ∣ 2 ) d r + 2 ∫ 0 T Y r B H ( Y r B ) d B r ] .
By ( f ) (f) ( f ) and Young’s inequality, we have that $2|yf(r,y,z)|\leq 2C_f(|y|+|y|^2+|yz|)\leq \widetilde{C}(1+|y|^2)+\frac{1}{2}|z|^2f o r s o m e c o n s t a n t for some constant f or so m eco n s t an t \widetilde{C}>0. F u r t h e r m o r e , w e a l s o h a v e . Furthermore, we also have . F u r t h er m or e , w e a l so ha v e |\xi|^2\leq\sup_{s\in[0,T]}|Y_s|^2, , , |yDH(y)H(y)|\leq C ^ + ∣ y ∣ 2 \widehat{C}+|y|^2 C + ∣ y ∣ 2 , and ∣ H ( y ) ∣ 2 ≤ C ^ |H(y)|^2\leq\widehat{C} ∣ H ( y ) ∣ 2 ≤ C for some constant C ^ > 0 \widehat{C}>0 C > 0 . Hence, for some constant C 0 > 0 C_0>0 C 0 > 0
1 2 E W [ ∫ 0 T ∣ Z r B ∣ 2 d r ] ≤ C 0 + C 0 E W [ sup s ∈ [ 0 , T ] ∣ Y s B ∣ 2 ] + 2 E W [ ∫ 0 T Y r B H ( Y r B ) d B ← r ] . (156) \frac{1}{2}\mathbb{E}^{W}\left[\int_0^T|Z_r^B|^2\,\mathrm{d}r\right]\leq C_0+C_0\mathbb{E}^{W}\left[\sup_{s\in[0,T]}|Y_s^B|^2\right]+2\mathbb{E}^{W}\left[\int_0^T Y_r^B H(Y_r^B)\,\mathrm{d}\overleftarrow{B}_r\right]. \tag*{(156)} 2 1 E W [ ∫ 0 T ∣ Z r B ∣ 2 d r ] ≤ C 0 + C 0 E W [ s ∈ [ 0 , T ] sup ∣ Y s B ∣ 2 ] + 2 E W [ ∫ 0 T Y r B H ( Y r B ) d B r ] . (156)
Since H H H is bounded, we have that sup s ∈ [ 0 , T ] ∣ Y s B H ( Y s B ) ∣ 2 \sup_{s\in[0,T]}|Y_s^B H(Y_s^B)|^2 sup s ∈ [ 0 , T ] ∣ Y s B H ( Y s B ) ∣ 2 is integrable. Then, by Itô’s isometry, ∫ t T Y r B H ( Y r B ) d B ← r \int_t^T Y_r^B H(Y_r^B)\,\mathrm{d}\overleftarrow{B}_r ∫ t T Y r B H ( Y r B ) d B r is a backward square-integrable martingale with respect to F t , T B \mathcal{F}_{t,T}^B F t , T B , and hence
E [ ∫ 0 T Y r B H ( Y r B ) d B ← r ] = 0. \mathbb{E}\left[\int_0^T Y_r^B H(Y_r^B)\,\mathrm{d}\overleftarrow{B}_r\right]=0. E [ ∫ 0 T Y r B H ( Y r B ) d B r ] = 0.
Consequently, taking E B [ ⋅ ] \mathbb{E}^{B}[\cdot] E B [ ⋅ ] in (156) and applying Fubini’s theorem, we obtain
E [ ∫ 0 T ∣ Z r B ∣ 2 d r ] ≲ 1 + E [ sup s ∈ [ 0 , T ] ∣ Y s B ∣ 2 ] < ∞ , \mathbb{E}\left[\int_0^T|Z_r^B|^2\,\mathrm{d}r\right]\lesssim1+\mathbb{E}\left[\sup_{s\in[0,T]}|Y_s^B|^2\right]<\infty, E [ ∫ 0 T ∣ Z r B ∣ 2 d r ] ≲ 1 + E [ s ∈ [ 0 , T ] sup ∣ Y s B ∣ 2 ] < ∞ ,
where the last inequality follows from the square-integrability of sup s ∈ [ 0 , T ] ∣ Y s B ∣ \sup_{s\in[0,T]}|Y_s^B| sup s ∈ [ 0 , T ] ∣ Y s B ∣ .
Appendix A. Auxiliary Results on Rough and Rough Stochastic Analysis
In this section, we collect some results in rough and rough stochastic analysis.
The following Lépingle’s inequality for p p p -variation characterizes the k k k -moments of p p p -variation of martingales by its quadratic process.
Lemma A.1. Assume p > 2 p>2 p > 2 , k ≥ 1 k\geq1 k ≥ 1 , and t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] . Let M M M be a continuous martingale such that E [ δ ⟨ M ⟩ t , T k / 2 ] < ∞ \mathbb{E}[\delta\langle M\rangle_{t,T}^{k/2}]<\infty E [ δ ⟨ M ⟩ t , T k /2 ] < ∞ . Then, there exists a positive number C C C depending only on p p p and k k k such that
C − 1 E t [ δ ⟨ M ⟩ t , T k / 2 ] ≤ E t [ ∥ δ M ∥ p -var ; [ t , T ] k ] ≤ C E t [ δ ⟨ M ⟩ t , T k / 2 ] . (157) C^{-1}\mathbb{E}_t[\delta\langle M\rangle_{t,T}^{k/2}]\leq\mathbb{E}_t[\|\delta M\|_{p\text{-var};[t,T]}^k]\leq C\mathbb{E}_t[\delta\langle M\rangle_{t,T}^{k/2}]. \tag*{(157)} C − 1 E t [ δ ⟨ M ⟩ t , T k /2 ] ≤ E t [ ∥ δ M ∥ p -var ; [ t , T ] k ] ≤ C E t [ δ ⟨ M ⟩ t , T k /2 ] . (157)
Consequently, there exists a positive constant C ^ \widehat{C} C depending only on p p p and k k k such that
C ^ − 1 ∥ δ M ∥ B M O ≤ ∥ ∥ δ M ∥ p -var ∥ ∞ ; k , ∞ ≤ C ^ ∥ δ M ∥ B M O . (158) \widehat{C}^{-1}\|\delta M\|_{\mathrm{BMO}}\leq\|\|\delta M\|_{p\text{-var}}\|_{\infty;k,\infty}\leq\widehat{C}\|\delta M\|_{\mathrm{BMO}}. \tag*{(158)} C − 1 ∥ δ M ∥ BMO ≤ ∥∥ δ M ∥ p -var ∥ ∞ ; k , ∞ ≤ C ∥ δ M ∥ BMO . (158)
Proof. The proof of (157) can be found in [14 ] , Theorem 14.12. For (158) , by [18 ] , Corollary 2.1 it follows that there exists a constant C ~ \widetilde{C} C depending only on k k k such that
C ~ − 1 ∥ δ M ∥ B M O k ≤ ess sup ( t , ω ) ∈ [ 0 , T ] × Ω E t [ δ ⟨ M ⟩ t , T k / 2 ] ≤ C ~ ∥ δ M ∥ B M O k . (159) \widetilde{C}^{-1}\|\delta M\|_{\mathrm{BMO}}^k\leq\operatorname*{ess\,sup}_{(t,\omega)\in[0,T]\times\Omega}\mathbb{E}_t[\delta\langle M\rangle_{t,T}^{k/2}]\leq\widetilde{C}\|\delta M\|_{\mathrm{BMO}}^k. \tag*{(159)} C − 1 ∥ δ M ∥ BMO k ≤ ( t , ω ) ∈ [ 0 , T ] × Ω ess sup E t [ δ ⟨ M ⟩ t , T k /2 ] ≤ C ∥ δ M ∥ BMO k . (159)
Hence, by taking essential supremum to (157) and applying (159) , we have (158) .
The following Lemma extends [16 ] , Theorem 1.1 from the usual L k L^k L k -norm to the conditional L k L^k L k -norm, which is essential for estimating the remainder of the composition of the nonlinear function H ( ⋅ ) H(\cdot) H ( ⋅ ) and ( Y , Y ′ ) ∈ R S M X ( p , p ′ ) -var (Y,Y')\in\mathrm{RSM}_{X}^{(p,p')\text{-var}} ( Y , Y ′ ) ∈ RSM X ( p , p ′ ) -var .
Lemma A.2. Let k ≥ 1 k\geq1 k ≥ 1 , 1 < k 1 ≤ ∞ 1<k_1\leq\infty 1 < k 1 ≤ ∞ , 1 ≤ k 0 < ∞ 1\leq k_0<\infty 1 ≤ k 0 < ∞ and 0 < r , r 1 < ∞ 0<r,r_1<\infty 0 < r , r 1 < ∞ . Suppose 1 / r < 1 / r 1 + 1 / 2 1/r<1/r_1+1/2 1/ r < 1/ r 1 + 1/2 , and 1 / k = 1 / k 1 + 1 / k 0 1/k=1/k_1+1/k_0 1/ k = 1/ k 1 + 1/ k 0 . Let ( F t ) 0 ≤ t ≤ T (F_t)_{0\leq t\leq T} ( F t ) 0 ≤ t ≤ T be a continuous adapted process and M t M_t M t be a continuous martingale. Denote by
A s , t : = ∫ s t δ F s , r d M r . A_{s,t}:=\int_s^t\delta F_{s,r}\,\mathrm{d}M_r. A s , t := ∫ s t δ F s , r d M r .
Then, there is a deterministic constant C C C depending only on r , r 1 , k , k 1 , k 0 r,r_1,k,k_1,k_0 r , r 1 , k , k 1 , k 0 such that
∥ ∥ A ⋅ , ⋅ ∥ r -var ; [ s , t ] ∥ F s ; k ≤ C ∥ ∥ δ F ∥ r 1 -var ; [ s , t ] ∥ F s ; k 1 ∥ ∥ δ M s , ⋅ ∥ ∞ ; [ s , t ] ∥ F s ; k 0 . \|\|A_{\cdot,\cdot}\|_{r\text{-var};[s,t]}\|_{\mathcal{F}_s;k}\leq C\|\|\delta F\|_{r_1\text{-var};[s,t]}\|_{\mathcal{F}_s;k_1}\|\|\delta M_{s,\cdot}\|_{\infty;[s,t]}\|_{\mathcal{F}_s;k_0}. ∥∥ A ⋅ , ⋅ ∥ r -var ; [ s , t ] ∥ F s ; k ≤ C ∥∥ δ F ∥ r 1 -var ; [ s , t ] ∥ F s ; k 1 ∥∥ δ M s , ⋅ ∥ ∞ ; [ s , t ] ∥ F s ; k 0 .
In addition, for r 1 = ∞ r_1=\infty r 1 = ∞ and 1 / r < 1 / 2 1/r<1/2 1/ r < 1/2 , we have
∥ ∥ A ⋅ , ⋅ ∥ r -var ; [ s , t ] ∥ k ≤ C ∥ ∥ F ∥ ∞ ; [ s , t ] ∥ F s ; k 1 ∥ ∥ δ M s , ⋅ ∥ ∞ ; [ s , t ] ∥ F s ; k 0 . \|\|A_{\cdot,\cdot}\|_{r\text{-var};[s,t]}\|_k\leq C\|\|F\|_{\infty;[s,t]}\|_{\mathcal{F}_s;k_1}\|\|\delta M_{s,\cdot}\|_{\infty;[s,t]}\|_{\mathcal{F}_s;k_0}. ∥∥ A ⋅ , ⋅ ∥ r -var ; [ s , t ] ∥ k ≤ C ∥∥ F ∥ ∞ ; [ s , t ] ∥ F s ; k 1 ∥∥ δ M s , ⋅ ∥ ∞ ; [ s , t ] ∥ F s ; k 0 .
Proof. Assume at first that the underlying probability space is Borel which guarantees the existence of a regular conditional probability P s ( ω ) = P ( ⋅ ∣ F s ) \mathbb{P}_s(\omega)=\mathbb{P}(\cdot\mid\mathcal{F}_s) P s ( ω ) = P ( ⋅ ∣ F s ) . Obviously M M M restricted to [ s , t ] [s,t] [ s , t ] is a ( P s , ( F r : r ≥ s ) ) (\mathbb{P}_s,(\mathcal{F}_r:r\ge s)) ( P s , ( F r : r ≥ s )) -martingale, almost surely, so that the conditional estimate follows immediately from the unconditional one. One can also avoid the use of regular conditional probabilities. Indeed, most arguments in the proof of [16 ] can be adapted to versions involving conditional expectations. The key point is that the vector-valued BDG inequality shown in [16 ] remains valid in the conditional setting. The extension to conditional estimates relies solely on standard scalar probabilistic arguments (specifically, localization via F s \mathcal{F}_s F s -measurable indicator functions, which amounts to replacing each martingale ( h n k : n ≥ 0 ) (h_n^k:n\ge0) ( h n k : n ≥ 0 ) , in notation of [16 ] , by h n k 1 O h_n^k1_O h n k 1 O , time-0 measurable set O O O , which yields immediately the required conditional form of [16 ] .)
Lemma A.3 (RDEs with BV drifts). Let p ∈ ( 2 , 3 ) p\in(2,3) p ∈ ( 2 , 3 ) and X ∈ C p -var ( [ 0 , T ] , R d ) \mathbf{X}\in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^d) X ∈ C p -var ([ 0 , T ] , R d ) . Let A ∈ C 1 -var ( [ 0 , T ] , R e ) A\in\mathcal{C}^{1\text{-var}}([0,T],\mathbb{R}^e) A ∈ C 1 -var ([ 0 , T ] , R e ) , with A 0 = 0 A_0=0 A 0 = 0 , V ∈ C b 2 ( R e ; L ( R d , R e ) ) V\in\mathcal{C}_b^2(\mathbb{R}^e;\mathcal{L}(\mathbb{R}^d,\mathbb{R}^e)) V ∈ C b 2 ( R e ; L ( R d , R e )) , and assume ( Y , V ( Y ) ) ∈ D X ( p , p ) -var ( [ 0 , T ] , R e ) (Y,V(Y))\in\mathcal{D}_{\mathbf{X}}^{(p,p)\text{-var}}([0,T],\mathbb{R}^e) ( Y , V ( Y )) ∈ D X ( p , p ) -var ([ 0 , T ] , R e ) satisfies
Y t = Y 0 + A t + ∫ 0 t V ( Y r ) d X r . Y_t=Y_0+A_t+\int_0^t V(Y_r)\,\mathrm{d}\mathbf{X}_r. Y t = Y 0 + A t + ∫ 0 t V ( Y r ) d X r .
Then, with ρ ( s , t ) : = ∥ δ X ∥ p -var ; [ s , t ] p + ∥ δ A ∥ 1 -var ; [ s , t ] \rho(s,t):=\lVert\delta\mathbf{X}\rVert_{p\text{-var};[s,t]}^p+\lVert\delta A\rVert_{1\text{-var};[s,t]} ρ ( s , t ) := ∥ δ X ∥ p -var ; [ s , t ] p + ∥ δ A ∥ 1 -var ; [ s , t ] ,
∥ δ Y ∥ p -var ; [ s , t ] ≤ C ( ρ ( s , t ) 1 / p + ρ ( s , t ) ) , \lVert\delta Y\rVert_{p\text{-var};[s,t]}\le C\left(\rho(s,t)^{1/p}+\rho(s,t)\right), ∥ δ Y ∥ p -var ; [ s , t ] ≤ C ( ρ ( s , t ) 1/ p + ρ ( s , t ) ) ,
where C C C depends only on p , d , e p,d,e p , d , e , and ∥ V ∥ C b 2 \lVert V\rVert_{\mathcal{C}_b^2} ∥ V ∥ C b 2 .
Proof. Equipping ( A , X ) (A,\mathbf{X}) ( A , X ) with its canonical joint rough-path lift, the mixed iterated integrals being defined by Riemann–Stieltjes integration. By basic Riemann–Stieltjes estimates, whenever ρ ( s , t ) ≤ 1 \rho(s,t)\le1 ρ ( s , t ) ≤ 1 , this lift is controlled by C p ρ C_p\rho C p ρ in p p p -variation, with C p > 0 C_p>0 C p > 0 depending only on p p p . Hence the standard local RDE estimate (see, e.g., [14 ] , Lemma 10.7), applied to the vector fields ( I e , V ( ⋅ ) ) (I_e,V(\cdot)) ( I e , V ( ⋅ )) , yields constants η ∈ ( 0 , 1 ] \eta\in(0,1] η ∈ ( 0 , 1 ] and C 0 C_0 C 0 , depending only on p , d , e p,d,e p , d , e and ∥ V ∥ C b 2 \lVert V\rVert_{\mathcal{C}_b^2} ∥ V ∥ C b 2 , such that
∥ δ Y ∥ p -var ; [ u , v ] ≤ C 0 ρ ( u , v ) 1 / p whenever ρ ( u , v ) ≤ η . \lVert\delta Y\rVert_{p\text{-var};[u,v]}\le C_0\rho(u,v)^{1/p}\qquad\text{whenever }\rho(u,v)\le\eta. ∥ δ Y ∥ p -var ; [ u , v ] ≤ C 0 ρ ( u , v ) 1/ p whenever ρ ( u , v ) ≤ η .
Since ρ \rho ρ is a control function, a greedy partition of [ s , t ] [s,t] [ s , t ] at level η \eta η has at most 1 + η − 1 ρ ( s , t ) 1+\eta^{-1}\rho(s,t) 1 + η − 1 ρ ( s , t ) intervals. Also, for ( u , v ) ∈ Δ [ s , t ] (u,v)\in\Delta_{[s,t]} ( u , v ) ∈ Δ [ s , t ] satisfying ρ ( u , v ) ≤ η \rho(u,v)\le\eta ρ ( u , v ) ≤ η , it holds that
ρ ( u , v ) 1 / p ( 1 + η − 1 ρ ( s , t ) ) ≤ ρ ( u , v ) 1 / p + η ( 1 − p ) / p ρ ( s , t ) ≤ ( 1 + η ( 1 − p ) / p ) ( ρ ( s , t ) 1 / p + ρ ( s , t ) ) . \rho(u,v)^{1/p}\left(1+\eta^{-1}\rho(s,t)\right)\le\rho(u,v)^{1/p}+\eta^{(1-p)/p}\rho(s,t)\le\left(1+\eta^{(1-p)/p}\right)\left(\rho(s,t)^{1/p}+\rho(s,t)\right). ρ ( u , v ) 1/ p ( 1 + η − 1 ρ ( s , t ) ) ≤ ρ ( u , v ) 1/ p + η ( 1 − p ) / p ρ ( s , t ) ≤ ( 1 + η ( 1 − p ) / p ) ( ρ ( s , t ) 1/ p + ρ ( s , t ) ) .
Therefore, summing the local estimates proves
∥ δ Y ∥ p -var ; [ s , t ] ≤ C 0 ( 1 + η ( 1 − p ) / p ) ( ρ ( s , t ) 1 / p + ρ ( s , t ) ) . \lVert\delta Y\rVert_{p\text{-var};[s,t]}\le C_0\left(1+\eta^{(1-p)/p}\right)\left(\rho(s,t)^{1/p}+\rho(s,t)\right). ∥ δ Y ∥ p -var ; [ s , t ] ≤ C 0 ( 1 + η ( 1 − p ) / p ) ( ρ ( s , t ) 1/ p + ρ ( s , t ) ) .
The following proposition offers an estimate for RDEs with unbounded drift, seemingly unavailable in the existing literature, including [14 ] , Chapter 12, [29 ] , and [25 ] .
Proposition A.4. Assume p ∈ ( 2 , 3 ) p\in(2,3) p ∈ ( 2 , 3 ) , γ > p \gamma>p γ > p , and let X ∈ C p -var ( [ 0 , T ] , R d ) \mathbf{X}\in\mathcal{C}^{p\text{-var}}([0,T],\mathbb{R}^d) X ∈ C p -var ([ 0 , T ] , R d ) . Let F ( t , ⋅ ) F(t,\cdot) F ( t , ⋅ ) be a uniformly Lipschitz function with sup t ∈ [ 0 , T ] ∣ F ( t , 0 ) ∣ < ∞ \sup_{t\in[0,T]}|F(t,0)|<\infty sup t ∈ [ 0 , T ] ∣ F ( t , 0 ) ∣ < ∞ and H ∈ C b γ ( R e ; L ( R d , R e ) ) H\in\mathcal{C}_b^\gamma(\mathbb{R}^e;\mathcal{L}(\mathbb{R}^d,\mathbb{R}^e)) H ∈ C b γ ( R e ; L ( R d , R e )) . Then, the backward RDE
ψ t = y + ∫ t T F ( r , ψ r ) d r + ∫ t T H ( ψ r ) d X r (160) \psi_t=y+\int_t^T F(r,\psi_r)\,\mathrm{d}r+\int_t^T H(\psi_r)\,\mathrm{d}\mathbf{X}_r \tag*{(160)} ψ t = y + ∫ t T F ( r , ψ r ) d r + ∫ t T H ( ψ r ) d X r (160)
admits a unique solution with ( ψ , − H ( ψ ) ) ∈ D X ( p , p ) -var ( [ 0 , T ] , R e ) (\psi,-H(\psi))\in\mathcal{D}_{\mathbf{X}}^{(p,p)\text{-var}}([0,T],\mathbb{R}^e) ( ψ , − H ( ψ )) ∈ D X ( p , p ) -var ([ 0 , T ] , R e ) . Moreover, if
∣ y ∣ + sup t ∈ [ 0 , T ] ∣ F ( t , 0 ) ∣ + sup t ∈ [ 0 , T ] ∥ δ F ( t , ⋅ ) ∥ 1 -H o ¨ l + ∥ H ∥ C b 2 + T ≤ M , |y|+\sup_{t\in[0,T]}|F(t,0)|+\sup_{t\in[0,T]}\lVert\delta F(t,\cdot)\rVert_{1\text{-Höl}}+\lVert H\rVert_{\mathcal{C}_b^2}+T\le M, ∣ y ∣ + t ∈ [ 0 , T ] sup ∣ F ( t , 0 ) ∣ + t ∈ [ 0 , T ] sup ∥ δ F ( t , ⋅ ) ∥ 1 -H o ¨ l + ∥ H ∥ C b 2 + T ≤ M ,
then for some constant C p , d , e , M C_{p,d,e,M} C p , d , e , M depending only on p p p , d d d , e e e and M M M , it holds that
∥ δ ψ ∥ p -var ; [ 0 , T ] ≤ C p , d , e , M ( 1 + ∥ ∥ ∥ δ X ∥ ∥ ∥ p -var ; [ 0 , T ] p ) . (161) \lVert\delta\psi\rVert_{p\text{-var};[0,T]} \le C_{p,d,e,M}\left(1+\lVert\!\lVert\!\lVert\delta X\rVert\!\rVert\!\rVert_{p\text{-var};[0,T]}^{p}\right). \tag*{(161)} ∥ δ ψ ∥ p -var ; [ 0 , T ] ≤ C p , d , e , M ( 1 + ∥ ∥ ∥ δ X ∥ ∥ ∥ p -var ; [ 0 , T ] p ) . (161)
Proof. We first prove the estimate (161) . Assume ( ψ , − H ( ψ ) ) ∈ D X ( p , p ) -var ( [ 0 , T ] , R e ) (\psi,-H(\psi))\in\mathcal{D}_{X}^{(p,p)\text{-var}}([0,T],\mathbb{R}^{e}) ( ψ , − H ( ψ )) ∈ D X ( p , p ) -var ([ 0 , T ] , R e ) is a solution to Eq. (160) . Set A t : = − ∫ 0 t F ( r , ψ r ) d r A_t:=-\int_{0}^{t}F(r,\psi_r)\,dr A t := − ∫ 0 t F ( r , ψ r ) d r . Then, by applying Lemma A.3 backwardly on [ t , T ] [t,T] [ t , T ] with vector field V : = − H V:=-H V := − H , it yields
∥ δ ψ ∥ p -var ; [ t , T ] ≤ C ‾ p , d , e , M ( ρ ( t , T ) 1 / p + ρ ( t , T ) ) , where ρ ( t , T ) : = ∥ ∥ ∥ δ X ∥ ∥ ∥ p -var ; [ t , T ] p + ∫ t T ∣ F ( r , ψ r ) ∣ d r , \lVert\delta\psi\rVert_{p\text{-var};[t,T]}\le\overline{C}_{p,d,e,M}\left(\rho(t,T)^{1/p}+\rho(t,T)\right),
\qquad\text{where }\rho(t,T):=\lVert\!\lVert\!\lVert\delta X\rVert\!\rVert\!\rVert_{p\text{-var};[t,T]}^{p}+\int_{t}^{T}|F(r,\psi_r)|\,dr, ∥ δ ψ ∥ p -var ; [ t , T ] ≤ C p , d , e , M ( ρ ( t , T ) 1/ p + ρ ( t , T ) ) , where ρ ( t , T ) := ∥ ∥ ∥ δ X ∥ ∥ ∥ p -var ; [ t , T ] p + ∫ t T ∣ F ( r , ψ r ) ∣ d r ,
and C ‾ p , d , e , M > 0 \overline{C}_{p,d,e,M}>0 C p , d , e , M > 0 is a constant depending only on p p p , d d d , e e e , and M M M . Using u 1 / p ≤ 1 + u u^{1/p}\le1+u u 1/ p ≤ 1 + u , ∣ F ( r , ψ r ) ∣ ≤ M ( 1 + ∣ ψ r ∣ ) |F(r,\psi_r)|\le M(1+|\psi_r|) ∣ F ( r , ψ r ) ∣ ≤ M ( 1 + ∣ ψ r ∣ ) , and ∣ ψ r ∣ ≤ ∥ δ ψ ∥ p -var ; [ r , T ] + ∣ y ∣ |\psi_r|\le\lVert\delta\psi\rVert_{p\text{-var};[r,T]}+|y| ∣ ψ r ∣ ≤ ∥ δ ψ ∥ p -var ; [ r , T ] + ∣ y ∣ , we obtain that for some constant C ~ p , d , e , M > 0 \widetilde{C}_{p,d,e,M}>0 C p , d , e , M > 0 ,
∥ δ ψ ∥ p -var ; [ t , T ] ≤ C ~ p , d , e , M ( 1 + ∥ ∥ ∥ δ X ∥ ∥ ∥ p -var ; [ t , T ] p ) + C ~ p , d , e , M ∫ t T ∥ δ ψ ∥ p -var ; [ r , T ] d r . \lVert\delta\psi\rVert_{p\text{-var};[t,T]}\le\widetilde{C}_{p,d,e,M}\left(1+\lVert\!\lVert\!\lVert\delta X\rVert\!\rVert\!\rVert_{p\text{-var};[t,T]}^{p}\right)+\widetilde{C}_{p,d,e,M}\int_{t}^{T}\lVert\delta\psi\rVert_{p\text{-var};[r,T]}\,dr. ∥ δ ψ ∥ p -var ; [ t , T ] ≤ C p , d , e , M ( 1 + ∥ ∥ ∥ δ X ∥ ∥ ∥ p -var ; [ t , T ] p ) + C p , d , e , M ∫ t T ∥ δ ψ ∥ p -var ; [ r , T ] d r .
The desired estimate follows from Gronwall’s inequality (in backward form).
After obtaining the a priori estimate for the solution, one can prove the well-posedness of the RDE via a truncation argument. Indeed, when F F F is bounded, the condition H ∈ C b γ H\in C_b^\gamma H ∈ C b γ is sufficient to guarantee well-posedness for rough SDEs (see [11 ] ), and hence for RDEs. Define the truncated function F n ( t , y ) : = F ( t , π n ( y ) ) F^n(t,y):=F(t,\pi_n(y)) F n ( t , y ) := F ( t , π n ( y )) with
π n ( y ) : = { y , ∣ y ∣ ≤ n , n y / ∣ y ∣ , ∣ y ∣ > n . \pi_n(y):=
\begin{cases}
y, & |y|\le n,\\
ny/|y|, & |y|>n.
\end{cases} π n ( y ) := { y , n y /∣ y ∣ , ∣ y ∣ ≤ n , ∣ y ∣ > n .
Then, the a priori estimate ensures that the corresponding truncated solution is, in fact, a solution for sufficiently large n n n ; the well-posedness then follows.
Lemma A.5. Assume γ ∈ ( 2 , 3 ] \gamma\in(2,3] γ ∈ ( 2 , 3 ] and G : R → R G:\mathbb{R}\to\mathbb{R} G : R → R satisfies D G ∈ C b γ − 2 ( R , R ) DG\in C_b^{\gamma-2}(\mathbb{R},\mathbb{R}) D G ∈ C b γ − 2 ( R , R ) . Then for any functions y , y ˉ : [ 0 , T ] → R y,\bar{y}:[0,T]\to\mathbb{R} y , y ˉ : [ 0 , T ] → R , we have for all ( s , t ) ∈ Δ (s,t)\in\Delta ( s , t ) ∈ Δ ,
∣ δ ( G ( y ) − G ( y ˉ ) ) s , t ∣ ≤ ∥ D G ∥ C b γ − 2 ( ( ∣ δ y s , t ∣ γ − 2 + ∣ δ y ˉ s , t ∣ γ − 2 ) ∣ ( y − y ˉ ) t ∣ + ∣ δ ( y − y ˉ ) s , t ∣ ) . (162) |\delta(G(y)-G(\bar{y}))_{s,t}|\le\lVert DG\rVert_{C_b^{\gamma-2}}\left((|\delta y_{s,t}|^{\gamma-2}+|\delta\bar{y}_{s,t}|^{\gamma-2})|(y-\bar{y})_t|+|\delta(y-\bar{y})_{s,t}|\right). \tag*{(162)} ∣ δ ( G ( y ) − G ( y ˉ ) ) s , t ∣ ≤ ∥ D G ∥ C b γ − 2 ( ( ∣ δ y s , t ∣ γ − 2 + ∣ δ y ˉ s , t ∣ γ − 2 ) ∣ ( y − y ˉ ) t ∣ + ∣ δ ( y − y ˉ ) s , t ∣ ) . (162)
Moreover, for 1 ≤ q ≤ q ′ 1\le q\le q' 1 ≤ q ≤ q ′ , assume γ − 1 ≥ ( q + q ′ ) / q ′ \gamma-1\ge(q+q')/q' γ − 1 ≥ ( q + q ′ ) / q ′ . Then, it holds
∥ δ ( G ( y ) − G ( y ˉ ) ) ∥ q ′ -var ; [ s , t ] ≤ ∥ D G ∥ C b γ − 2 ( ( ∥ δ y ∥ q -var ; [ s , t ] γ − 2 + ∥ δ y ˉ ∥ q -var ; [ s , t ] γ − 2 ) sup r ∈ [ s , t ] ∣ ( y − y ˉ ) r ∣ + ∥ δ ( y − y ˉ ) ∥ q ′ -var ; [ s , t ] ) . (163) \lVert\delta(G(y)-G(\bar{y}))\rVert_{q'\text{-var};[s,t]}
\le\lVert DG\rVert_{C_b^{\gamma-2}}\left(\left(\lVert\delta y\rVert_{q\text{-var};[s,t]}^{\gamma-2}+\lVert\delta\bar{y}\rVert_{q\text{-var};[s,t]}^{\gamma-2}\right)\sup_{r\in[s,t]}|(y-\bar{y})_r|+\lVert\delta(y-\bar{y})\rVert_{q'\text{-var};[s,t]}\right). \tag*{(163)} ∥ δ ( G ( y ) − G ( y ˉ )) ∥ q ′ -var ; [ s , t ] ≤ ∥ D G ∥ C b γ − 2 ( ( ∥ δ y ∥ q -var ; [ s , t ] γ − 2 + ∥ δ y ˉ ∥ q -var ; [ s , t ] γ − 2 ) r ∈ [ s , t ] sup ∣ ( y − y ˉ ) r ∣ + ∥ δ ( y − y ˉ ) ∥ q ′ -var ; [ s , t ] ) . (163)
Proof. By the triangular inequality and Newton-Leibniz’s formula, we have
∣ δ ( G ( y ) − G ( y ˉ ) ) s , t ∣ ≤ ∫ 0 1 ∣ D G ( y ˉ t + λ ( y t − y ˉ t ) ) − D G ( y ˉ s + λ ( y s − y ˉ s ) ) ∣ d λ ⋅ ∣ ( y − y ˉ ) t ∣ + ∫ 0 1 ∣ D G ( y ˉ s + λ ( y s − y ˉ s ) ) ∣ d λ ⋅ ∣ δ ( y − y ˉ ) s , t ∣ ≤ ∥ D G ∥ C b γ − 2 ( ∣ δ y s , t ∣ + ∣ δ y ˉ s , t ∣ ) γ − 2 ∣ ( y − y ˉ ) t ∣ + ∥ D G ∥ ∞ ∣ δ ( y − y ˉ ) s , t ∣ , \begin{aligned}
|\delta(G(y)-G(\bar{y}))_{s,t}|
&\le\int_{0}^{1}|DG(\bar{y}_t+\lambda(y_t-\bar{y}_t))-DG(\bar{y}_s+\lambda(y_s-\bar{y}_s))|\,d\lambda\cdot|(y-\bar{y})_t|\\
&\quad+\int_{0}^{1}|DG(\bar{y}_s+\lambda(y_s-\bar{y}_s))|\,d\lambda\cdot|\delta(y-\bar{y})_{s,t}|\\
&\le\lVert DG\rVert_{C_b^{\gamma-2}}\left(|\delta y_{s,t}|+|\delta\bar{y}_{s,t}|\right)^{\gamma-2}|(y-\bar{y})_t|+\lVert DG\rVert_{\infty}|\delta(y-\bar{y})_{s,t}|,
\end{aligned} ∣ δ ( G ( y ) − G ( y ˉ ) ) s , t ∣ ≤ ∫ 0 1 ∣ D G ( y ˉ t + λ ( y t − y ˉ t )) − D G ( y ˉ s + λ ( y s − y ˉ s )) ∣ d λ ⋅ ∣ ( y − y ˉ ) t ∣ + ∫ 0 1 ∣ D G ( y ˉ s + λ ( y s − y ˉ s )) ∣ d λ ⋅ ∣ δ ( y − y ˉ ) s , t ∣ ≤ ∥ D G ∥ C b γ − 2 ( ∣ δ y s , t ∣ + ∣ δ y ˉ s , t ∣ ) γ − 2 ∣ ( y − y ˉ ) t ∣ + ∥ D G ∥ ∞ ∣ δ ( y − y ˉ ) s , t ∣ ,
which yields (162) . Moreover, taking q ′ q' q ′ -variation to both sides of (162) , and noting 1 / q ′ ≤ ( γ − 2 ) / q 1/q'\le(\gamma-2)/q 1/ q ′ ≤ ( γ − 2 ) / q , it then follows (163) .
Appendix B. Symbolic Index
Table 1.
References Table 2.
Institut für Mathematik, Technische Universität Berlin and Weierstraß Institut, Berlin, Germany
Email address: friz@math.tu-berlin.de
Research Center for Mathematics and Interdisciplinary Sciences; Frontiers Science Center for Nonlinear Expectations, Ministry of Education, Shandong University, Qingdao 266237, China
Email address: txjsong@sdu.edu.cn
Research Center for Mathematics and Interdisciplinary Sciences; Frontiers Science Center for Nonlinear Expectations (Ministry of Education), Shandong University, Qingdao 266237, China
Email address: huilinzhang@sdu.edu.cn
Institute for Theoretical Sciences, Westlake Institute for Advanced Study, Westlake University, Hangzhou, Zhejiang, 310030, China
Email address: zhk01815@163.com