Penalisation techniques for one-dimensional reflected rough differential equations (Q2203628)

This paper treats the penalization techniques and their application to one-dimensional reflected rough differential equations. \[ S_{[0,T]} := \{ (s,t) \in [0, T]^2; \quad s \leqslant t \} \] is the simplex by which the 2-parameter functions are indexed. For simplicity, we set \(I := [0, T]\) for the time interval. For \(\beta \in ( 1/3, 1/2)\) (resp. \(p \in (2,3)\) ), rough paths are denoted by \begin{align*} \mathbf{X} &= ( (X_t)_{ t \in I}, \quad ( {\mathbb X}_{(s,t) \in S_I } ) \in C^{\beta} ( I; {\mathbb R}^d) \times C^{2 \beta} ( I; {\mathbb R}^d \otimes {\mathbb R}^d ), \\ & ( \text{resp.} \quad \mathbf{X} \in {\mathcal V}^p (I; {\mathbb R}^d ) \times {\mathcal V}^{p/2} ( I; {\mathbb R}^d \otimes {\mathbb R}^d ) ), \end{align*} where \({\mathbb X}_{s,t}\) is a supplementary two-parameters path of the path \(X\). For a finite-dimensional vector space \(F\), \({\mathcal V}^p\) denotes the set of continuous paths \(f\) : \(I \to F\) such that \(\Vert \delta f \Vert_p \equiv \Vert f \Vert_p < \infty\) for \(\delta f_{s,t}\) \(:=\) \(f_t - f_s\) and \(\Vert g \Vert_p\) is the \(p\)-variation semi-norm. \(C_g^{\beta} \equiv C_g^{\beta}(I; {\mathbb R}^d)\) (resp. \({\mathcal V}_g^p \equiv {\mathcal V}_g^p (I; {\mathbb R}^d)\)) is the space of geometric \(\beta\)-Hölder rough paths (resp. \(p\)-rough paths) with the homogeneous rough path norm \[ \vert\Vert\mathbf{X} \vert\Vert_{\beta} = \Vert X \Vert_{\beta} + \sqrt{ \Vert {\mathbb X} \Vert_{2 \beta} }. \] The authors consider the Skorokhod problem associated to \(\sigma\) and \(L\): \[ Y_t = y_0 + \int_0^t \sigma( Y_s) d \text{X}_s + K_t \quad \text{and} \quad Y_t \geq L_t, \quad \text{ for all } t \in I, \] where \(K\) is the non-decreasing path only when \(Y\) hits \(L\), and \(L\) : \([0, T] \to {\mathbb R}\) is a moving boundary. Hereafter, the problem is denoted by \(SP(\sigma,\mathbf{X}, L)\). If \(\mathbf{X}\) is a geometric rough path, the rough integral of \(Y\) against \(\mathbf{X}\) is classically defined by \[\int_0^T Y_s d\mathbf{X}_s = \lim_{m \to \infty} \sum_{ \pi_m = ( t_i^m)} Y_{t_i^m} \delta X_{t_i^m, t_{i+1}^m} + Y_{t_i^m}' {\mathbb X}_{t_i^m, t_{i+1}^m}, \] where the path \(Y'\) is the Gubinelli derivative of \(Y\), and \(( \pi_m)_{m \in {\mathbb N} }\) is an increasing sequence of subdivisions of \(I = [0, T]\) such that \[ \lim_{m \to \infty} \max_i ( t_{i+1}^m - t_i^m ) = 0 \quad \text{and} \quad t_0^m =0, \quad t_m^m = T. \] For \(\mathbf{X} \in C_g^{\beta} ( I; {\mathbb R}^d)\), \((Y, K)\) is said to solve the Skorokhod problem \(SP( \sigma, \mathbf{X}, L)\) or \((Y, K)\) is said to be a solution to the reflected RDE (rough differential equation) with diffusion coefficient \(\sigma\) started from \(y_0 \geq L_0\) and reflected on the path \(L\), if \par\noindent (i) \((Y, \sigma(Y)) \in {\mathcal V}_X^p\) and \((Y,K)\) satisfies Equation (4), in the sense that both sides are equal, where the integral \(\int_0^{\cdot} \sigma(Y_s) d \mathbf{X}_s\) is understood in the sense of (5); \par\noindent (ii) \(Y_t \geq L_t\) for \(\text{ for all } t \in [0, T]\); \par\noindent (iii) \(K\) is non-decreasing; \par\noindent (iv) \(\int_0^t (Y_s - L_s) d K_s = 0\) for \(\text{ for all } t \in [0, T]\) or equivalently, \(\int_0^t 1_{ \{ Y_s \not= L_s \} } d K_s = 0\). Here are the main results. Theorem 1. Let \(\mathbf{X}= (X, {\mathbb X} ) \in C_g^{\beta}( I; {\mathbb R}^d )\) for some \(\beta \in (1/3, 1) \setminus \{ 1/2 \}\), let \(L \in C^{\alpha}(I; {\mathbb R})\) with \(\alpha > \frac{1}{2} \vee (1- \beta)\), and assume that \(\sigma \in C_b^3({\mathbb R}, ( {\mathbb R}^d )' )\). Then there is at most one solution to the problem \(SP(\sigma, \mathbf{X}, L)\) with initial condition \(y_0 \geq L_0\). Since the penalization terms \(n(\cdot)_-\) in the sequence of penalized RDEs are not differentiable, the authors replace them by smooth functions \(\psi_n\). This allows us to consider the penalized paths defined by \[ Y_t^n = y_0 + \int_0^t \psi_n ( Y_s^n - L_s) ds + \int_0^t \sigma( Y_s^n ) d\mathbf{X}_s, \quad t \in [0, T]. \] Theorem 2. Assume that \(\sigma \in C_b^4 ( {\mathbb R}, ( {\mathbb R}^d)' )\) and \(y_0 \geq L_0\). For each \(n \in {\mathbb N}\), \(Y^n\) denotes the solution to (7). Then \[\left( Y_t^n, \int_0^t \psi_n (Y_s^n - L_s) ds \right)_{ t \in [0, T]}, \quad n \in {\mathbb N} \] converges uniformly on \([0, T]\) to \((Y,K)\), the solution to the Skorekhod problem \(SP(\sigma, \mathbf{X}, L )\), and \(Y \in C^{\beta}\). The next result is about a rate of convergence of the sequence of penalized processes to the reflected solution. Namely, Theorem 3. Under the same assumptions as in Theorem 2, the penalized solution \(Y^n\) converges to \(Y\) with the following rate: for some \(C > 0\), \[ 0 \leqslant Y^t - y_t^n \leqslant C \cdot n^{- \beta}, \qquad \text{ for all } n \in {\mathbb N}^*, \quad \text{ for all } t \in [0, T]. \] Moreover, a Doss-Sussmann representation for RDEs with a drift growing at most linearly is also derived. Finally, by the use of the penalization method, absolute continuity of the law at time \(t > 0\) of some reflected Gaussian RDE with respect to the Lebesgue measure, is proved as well. \par For other related works, see e.g. [\textit{A. Deya} et al., Stochastic Processes Appl. 129, No. 9, 3261--3281 (2019; Zbl 1479.60208)] for one-dimensional reflected rough differential equations, [\textit{A. Deya} et al., J. Funct. Anal. 276, No. 12, 3577--3645 (2019; Zbl 1411.60094)] for a priori estimates for rough PDEs with application to rough conservation laws, [\textit{S. Riedel} and \textit{M. Scheutzow}, J. Differ. Equations 262, No. 1, 283--312 (2017; Zbl 1357.34096)] for rough differential equations with unbounded drift term.