--- title: Linearly Generalized Skorokhod Reflection Mapping url: https://www.emergentmind.com/topics/linearly-generalized-skorokhod-reflection-mapping type: topic --- # Linearly Generalized Skorokhod Reflection Mapping Searching arXiv for recent and foundational papers on linearly generalized Skorokhod reflection mapping. The linearly generalized Skorokhod reflection mapping is a generalization of the classical Skorokhod reflection map in which the constraining term enters through a linear coupling, typically via a reflection matrix, a linear operator, or a linear drift term, while preserving nonnegativity and a complementarity or minimality condition. In the literature summarized here, this notion appears in finite-dimensional orthants driven by jump processes, in infinite-dimensional fluid limits for Jackson networks, and in one-dimensional linearly drifted reflection maps arising from queueing recursions [2512.24491] [2605.16868] [2510.27226]. The subject sits between classical pathwise reflection theory, convex-domain Skorokhod problems, and modern applications to queueing networks, reflected diffusions, and stochastic systems with operator-valued constraints. ## 1. Classical one-sided reflection and its characterizations The classical one-sided Skorokhod problem at the lower barrier \(0\) starts from a continuous path \(X:[0,\infty)\to\mathbb{R}\) and seeks a pair \((Q,Y)\) such that \[ Q(t)=X(t)+Y(t)\ge 0,\qquad t\ge 0, \] where \(Y\) is continuous, nondecreasing, \(Y(0)=0\), and increases only when the reflected process is at the boundary: \[ \int_0^\infty Q(s)\,dY(s)=0. \] For a continuous function \(X\), the unique solution is \[ Y(t)=-\inf_{0\le s\le t}\bigl(X(s)\wedge 0\bigr), \] so that \[ Q(t)=X(t)\vee \sup_{0\le s\le t}\bigl(X(t)-X(s)\bigr). \] This is the one-sided Skorokhod reflection map in its standard form [1003.5638]. A bounded-variation reformulation writes the driving input as a signed Borel measure \(X=A-C\), where \(A\) and \(C\) are locally finite nonnegative Borel measures without atoms. In that setting, \[ Q^*(t):=\sup_{0\le s\le t}X(s,t] =X(t)-\inf_{0\le s\le t}X(s), \] and the paper proves that \(Q^*\) is not merely a reflected path but the unique maximal solution of the nonlinear integral equation \[ Q(t)=\int_0^t \mathbf{1}\bigl(Q(s)>C(s,t]\bigr)\,dA(s). \] This gives an exact characterization of one-sided reflection as a fixed-point problem rather than only an explicit supremum formula [1003.5638]. The multidimensional deterministic precursor is the Skorokhod problem in a convex domain \(D\), where one seeks \[ X=w+\phi, \] with \(X\in D(\mathbb{R}_+,\overline D)\), \(\phi\) of bounded variation, \(\phi(0)=0\), and \[ \phi(t)=\int_0^t \vec n(s)\,d|\phi|(s), \] where \(\vec n(t)\) is a unit inward normal vector. In this formulation the correction acts only on the boundary and in inward normal directions, providing the geometric template from which matrix-based and operator-based linear generalizations are built [2009.03643]. ## 2. Finite-dimensional linear coupling in the positive orthant A direct finite-dimensional linearly generalized Skorokhod reflection mapping is studied for an \(n\)-dimensional process constrained to remain in \(R_+^n\), driven by an exogenous jump process \(Z=(Z^1,\dots,Z^n)\). The state equation is \[ X_t^i=X_{0-}^i+c_it-Z_t^i+L_t^i-\sum_{j\neq i}q_{ij}L_t^j,\quad 1\le i\le n, \] where \(Q=(q_{ij})\) is a nonnegative matrix with \(q_{ii}=0\), and \(L=(L^1,\dots,L^n)\) is the reflection process [2512.24491]. The one-dimensional Skorokhod formula is recalled in the form \[ L_t:=\sup_{s\le t}(Y_s)_-, \] as the smallest nondecreasing process such that \(Y+L\ge 0\). The linear generalization arises because reflection in coordinate \(i\) depends linearly on the entire vector \(L\) through the matrix \(Q\): \[ L_t^i=\sup_{s\le t}\bigg(X_{0-}^i+c_it-Z_s^i-\sum_{j\neq i}q_{ij}L_s^j\bigg)_-. \] At a jump time \(\tau\), the increment must satisfy the coupled system \[ \Delta L^i_\tau=\bigg(X_{\tau-}^i-\Delta Z_\tau^i-\sum_{j\neq i}q_{ij}\Delta L_\tau^j\bigg)_-,\quad 1\le i\le n. \] In this formulation, the classical reflection map is generalized by allowing cross-coupling through the linear operator \(I-Q\) [2512.24491]. The main existence theorem is stated in cone-duality form. The cone \[ C=\{u\in R^n:u\ge 0,\, u^\top (I-Q)\le 0\} \] and its dual \[ C^*=\{y\in R^n:u^\top y \ge 0 \text{ for all }u\in C\} \] govern continuation across jumps. A solution on \([0,\tau)\) can be extended to \([0,\tau]\) if and only if \[ X_{\tau-}-\Delta Z_\tau \in C^*. \] When this holds, there exists a unique minimal jump \(\Delta L_\tau\) with respect to the partial order on \(R_+^n\). Consequently, for any initial conditions and driving processes, there exists a unique minimal strong solution on a maximal interval \([0,\tau^*)\), where \(\tau^*\) is the first jump time at which the dual-cone condition fails [2512.24491]. The same jump condition admits a fixed-point interpretation through a monotone operator \(\Gamma\), and Knaster–Tarski yields a complete lattice of fixed points; iteration from \(z=0\) gives the minimal fixed point, hence the minimal jump. This places the linearly generalized map simultaneously in lattice-theoretic and linear-programming frameworks. The LP formulation minimizes \(\|\Delta L_\tau\|_1\) subject to \[ \Delta L_\tau \ge 0,\qquad (I-Q)\Delta L_{\tau}\ge -(X_{\tau-}-\Delta Z_\tau), \] and Farkas’s lemma converts feasibility into the dual-cone criterion [2512.24491]. A notable special case is the sub-stochastic regime \(\rho(Q)<1\). Then \(C=\emptyset\) and \(C^*=R^n\), so continuation always holds. This recovers Reiman’s all-time existence result as a special case, whereas arbitrary nonnegative matrices with zero diagonal require the maximal-stopping-time theory [2512.24491]. ## 3. Operator-valued and infinite-dimensional reflection maps An infinite-dimensional extension replaces a finite reflection matrix by an integral operator. In a growing open Jackson network, the usual finite-dimensional reflection matrix \(I-P^T\) is replaced in the limit by the operator \(\mathbf{1}-\boldsymbol{G}^T\), where \(\boldsymbol{G}\) is induced by a nonnegative kernel \(G(u,v)\) on \([0,1]^2\). The reflected path and regulator live in \[ \mathcal{D}_T(L_1)=D([0,T],L_1[0,1]). \] The infinite-dimensional Skorokhod problem is \[ Z(t)=X(t)+(\mathbf{1}-\boldsymbol{F})Y(t)\ge 0, \] with \[ dY(t)\ge 0,\qquad Y(0)=0, \] and coordinatewise complementarity \[ Z_u(t)\,dY_u(t)=0,\qquad \forall u\in[0,1]. \] For the network limit, \(\boldsymbol{F}=\boldsymbol{G}^T\) [2605.16868]. The reflection operator class is defined by \[ F\ge 0,\qquad \|\boldsymbol{F}\|_{op}\le 1,\qquad \rho(\boldsymbol{F})<1, \] where \[ \|\boldsymbol{F}\|_{op}=\sup_{v\in[0,1]}\int_0^1 |F(u,v)|\,du. \] The feasible regulator set is \[ \mathbf{\Psi}_F(X)=\{W\in\mathcal{D}_T^\uparrow(L_1): X+(\mathbf{1}-\boldsymbol{F})W\ge 0\}, \] and the reflected pair is defined by \[ Y=\Psi_F(X):=\inf \mathbf{\Psi}_F(X),\qquad Z=\Phi_F(X):=X+(\mathbf{1}-\boldsymbol{F})Y, \] with componentwise infimum \[ Y_u(t)=\inf\{W_u(t): W\in\mathbf{\Psi}_F(X)\}. \] This is presented as the infinite-dimensional analogue of a linearly generalized Skorokhod reflection map [2605.16868]. A central auxiliary operator is \[ \pi_{X,F}(W)(t)=\sup_{0\le s\le t}\bigl[-X(s)+\boldsymbol{F}W(s)\bigr]^+, \] and the theory shows that \[ \mathbf{\Psi}_F(X)=\{W\in\mathcal{D}_T^\uparrow(L_1): W\ge \pi_{X,F}(W)\}, \] with \(\Psi_F(X)\) the unique fixed point of \(\pi_{X,F}\). Existence is obtained by constructing \[ \widetilde W(t)= (\mathbf{1}-\boldsymbol{F})^{-1}\sup_{0\le s\le t}[-X(s)]^+, \] uniqueness by monotone iteration \(x^{n+1}=\pi_{X,F}(x^n)\), and complementarity by a minimality argument [2605.16868]. The spectral radius condition \(\rho(\boldsymbol{F})<1\) is structural. It yields the Neumann-series inverse \[ (\mathbf{1}-\boldsymbol{F})^{-1}=\sum_{n=0}^\infty \boldsymbol{F}^{(n)}, \] and for any \(\gamma\in(0,1)\), some finite \(k\) satisfies \[ \|\boldsymbol{F}^{(k)}\|_{op}\le \gamma,\qquad \|(\mathbf{1}-\boldsymbol{F})^{-1}\|_{op}\le \frac{k}{1-\gamma}. \] These bounds lead to Lipschitz continuity: \[ \|\Psi_F(X^1)-\Psi_F(X^2)\|_{T,1}\le \frac{k}{1-\gamma}\|X^1-X^2\|_{T,1}, \] and \[ \|\Phi_F(X^1)-\Phi_F(X^2)\|_{T,1}\le \left(1+\frac{2k}{1-\gamma}\right)\|X^1-X^2\|_{T,1}. \] The same framework also quantifies stability under perturbations of the operator itself [2605.16868]. This development is explicitly connected to the classical finite-dimensional linearly generalized Skorokhod map of Whitt, with the novelty that a reflection matrix becomes an operator acting on \(L_1[0,1]\)-valued paths. The resulting theory covers heterogeneous, possibly asymmetric networks and continuum-index limits [2605.16868]. ## 4. Linear-drift reflection and queueing recursions A one-dimensional version with linear feedback appears in the study of queues whose interarrival and service times depend linearly and randomly on customer waiting times. The waiting-time sequence satisfies the modified Lindley recursion \[ W_{i+1}^n=\bigl(C_i^n W_i^n+X_i^n\bigr)^+, \] equivalently \[ W_{i+1}^n=C_i^nW_i^n+X_i^n+\Psi_i^n, \qquad \Psi_i^n=-\min(C_i^nW_i^n+X_i^n,0), \] with regulator \[ L_i^n=\sum_{j=0}^i \Psi_j^n. \] In continuous-time indexing, the regulator satisfies \[ L^n(0)=0,\qquad L^n(t)\ge 0,\qquad \int_0^t W^n(s)\,dL^n(s)=0. \] Thus the queueing recursion is represented as a linear recursion plus reflection [2510.27226]. The paper defines the linearly generalized reflection map \[ (\mathcal R_\theta,\mathcal R'_\theta)(x)=(z,l), \] where \((z,l)\) solve \[ z(t)=x(t)-\int_0^t \theta z(s)\,ds+l(t)\ge 0, \] with \[ l(0)=0,\qquad l \text{ nondecreasing},\qquad z(t)\,dl(t)=0. \] For \(\theta=0\), this reduces to the standard Skorokhod reflection map: \[ (\mathcal R_0,\mathcal R'_0)=(\mathcal R,\mathcal R'). \] The linear generalization is therefore not a matrix coupling but a reflection law with linear drift feedback [2510.27226]. A useful representation is \[ (\mathcal R_\theta,\mathcal R'_\theta)(x)=(\mathcal R,\mathcal R')(\mathcal M(x)), \] where \(\mathcal M(x)=u\) is the unique solution of \[ u(t)=x(t)-\int_0^t \theta\,\mathcal R(u)(s)\,ds. \] The paper states that \(\mathcal M_\theta\) is well-defined and Lipschitz continuous on \(\mathcal D_T\), and by this representation \(\mathcal R_\theta\) and \(\mathcal R'_\theta\) are also Lipschitz. The classical explicit formula remains available at the \(\theta=0\) level: \[ l(t)=\sup_{0\le s\le t}[-x(s)]^+,\qquad z(t)=x(t)+\sup_{0\le s\le t}[-x(s)]^+. \] This identifies the linearly generalized map as a drift-perturbed reflection operator built on the standard half-line map [2510.27226]. In moderate deviations, reflection appears exactly when the stable equilibrium lies at the boundary. When the fluid equilibrium is positive, \(\bar W^*=\mu/\theta>0\), the rate function is determined by the unreflected linear map \(\mathcal M_\theta\). When the stable equilibrium is \(0\), the rate function is expressed through \[ \phi=\mathcal R_\theta(w_0+\psi_1+r\mathfrak e), \] and the reflected term survives in the limiting variational problem. The paper states that this dichotomy is parallel to the corresponding diffusion limits: positive equilibrium yields an Ornstein–Uhlenbeck-type limit, whereas zero equilibrium yields a reflected OU process or reflected Brownian motion when \(\theta=0\) [2510.27226]. ## 5. Extended Skorokhod maps, oblique reflection, and linearized derivative problems The linearly generalized perspective is closely connected with the extended Skorokhod map in convex polyhedral domains. For a convex polyhedral domain \[ G \doteq \bigcap_{i=1}^N \{x\in\mathbb{R}^J:\langle x,n_i\rangle \ge c_i\}, \] with face-dependent reflection directions \(d_i\) satisfying \[ \langle d_i,n_i\rangle = 1, \] the extended Skorokhod problem seeks \((Z,Y)\) such that \[ Z(t)=X(t)+Y(t),\qquad Z(t)\in G, \] and for all \(0\le s0\}\,dl(s)=0, \] where \(F\) may have jumps. This framework includes jump reflection, sticky or delayed reflection through the time change \[ A(t)=t+p\,l(t),\qquad z(t)=y(A^{-1}(t)), \] and switching approximations near the boundary. It is a generalized Skorokhod map, but its defining feature is the boundary regulator \(F\), not linear coupling through a reflection matrix or operator [2303.02771]. A second distinction concerns nonlinear-expectation reflection in BSDEs. There the generalized reflection operator is \[ L_t(X)=\inf\{x\geq 0:\mathcal{E}[l(t,x+X)]\geq 0\}, \] and the reflection process \(K\) is characterized by the Skorokhod-type condition \[ \int_0^T \mathcal{E}[l(t,Y_t)]\,dK_t=0. \] The paper explicitly states that this framework is not linear in general; linear behavior appears only in special cases. Thus it supports a generalized Skorokhod reflection interpretation, but not a fundamentally linear one [2511.17049]. A related backward formulation appears in reflected BSDEs with resistance, where time reversal and the Skorokhod equation yield an explicit representation of the compensator \(K\). The solution is then obtained as a fixed point of a mapping built from the Skorohod reflection formula. This is again a generalized reflection-operator viewpoint, but the generalization is stochastic and backward rather than a linearly coupled orthant map in the finite- or infinite-dimensional queueing sense [1103.2078]. Taken together, these works indicate that the linearly generalized Skorokhod reflection mapping is best understood as one branch within a larger reflection-operator landscape. Its distinctive feature is the preservation of Skorokhod minimality or complementarity under linear coupling, whether through \(I-Q\), \(\mathbf{1}-\boldsymbol{F}\), or a linear drift-reflection law. Other generalized reflection constructions may share fixed-point, minimality, and pathwise-continuity mechanisms, but they are not linear in the same sense. Source: https://www.emergentmind.com/topics/linearly-generalized-skorokhod-reflection-mapping