A posteriori error analysis for a class of integral equations and variational inequalities (Q707582)

The main purpose of this paper is the application of the theory of integral equations and variational inequalities with integral operators to the following problems: {\parindent4mm \begin{itemize}\item[1*] Option pricing with jump processes (is modeled by an integral operator and leads to a parabolic equation (European option) or a parabolic variational inequality (American option)) \item[2*] Boundary integral equations and inequalities (they are obtained from elliptic partial differential equations in a bounded or exterior domain, perhaps with unilateral constraint on the boundary-Signorini problem) \item[3*] Functional Laplacian (boundary control problems, surface frame propagation and quasi-geostrophic flow models are all problems governed by the fractional Laplacian). \end{itemize}} A linear integral operator \(\mathcal{I}\) of order \(\rho \in (0,2)\) on a bounded domain \(\Omega \subset \mathbb R^d\), is considered. The Sobolev space \(\widetilde{H}^s(\Omega) = \{ u|_\Omega : u \in H^s(\mathbb R^d), u|_{\mathbb R^d \backslash \Omega}=0\}\) is defined, and the integral operators of the form \(\mathcal{I} u = K*u\) with \(K(x) = (g(x) / \|x\|^{d+\rho})\) are considered, where the convolution integral is defined by a suitable regularization and \(\rho = 2s\). The operator \(\mathcal{A}\) is either an integral operator \(\mathcal{A} = \mathcal{I}\) of order \(2s \in (0,2)\), or an integro-differential operator \(\mathcal{A} = \mathcal{D} + \mathcal{I}\) of order \(2s=2\) where \(\mathcal{D}\) is a second-order differential operator (\(2s\) always indicate the order of \(\mathcal{A}\)). The energy space is \(\mathcal V = \widetilde{H}^s(\Omega)\), and \(\mathcal{A}: \mathcal V \rightarrow \mathcal V^*\) is a continuous and coercive operator. The parabolic equation: \(u_t(t) + \mathcal{A} u(t) = f(t)\) with initial condition \(u(0) = u_o\) is examined. The weak form is proposed: \[ \langle u_t(t) + \mathcal{A} u (t) - f(t), v\rangle_\mathcal V = 0 \quad \forall v \in \mathcal V.\tag{1} \] For the obstacle constraint \(u \geq \chi\) the variational equality (1) becomes the parabolic variational inequality \[ \langle u_t(t) + \mathcal{A} u (t) - f(t), u(t) - v\rangle_\mathcal V \leq 0 \quad \forall v \in \mathcal V\text{ with }v \geq \chi.\tag{2} \] For the space discretization the piecewise linear elements in space over a mesh does not depend on time, but may be graded, is used. For the time discretization, the implicit Euler method where the time partition may also be nonuniform is proposed. Main result: One main objective is to construct a computable a posteriori error estimator: the estimator uses the computed discrete solution, operator and data, and gives an upper bound of the error, measured in the norm of \(\mathcal V\) in the time independent case, and measured in \(L_2 (0,T, \mathcal V)\) in the time-dependent case. The behavior of the proposed error estimators with numerical experiments are illustrated. The authors' original motivation is the pricing of American options on a single asset with possible barriers. In this case the singularities which reduce the convergence rates are located at: the barriers (endpoints of the interval), the free boundary, \(t=0\) (maturity) where the pay off function (initial condition) is nonsmooth and the performance of the estimator under these circumstances are shown (these issues also occur in the equality case and the time independent case). The authors prove that the estimators give an upper bound to the actual error and with the numerical experiments provide additional evidence that: The space and time estimators converge with the correct rates in space and time. The ratio of estimator to the actual error (effectivity index) is of moderate size. The node-based error estimators provide reliable information about the local errors. In the time-independent case an adaptive algorithm driven authors estimators converges with the optimal rate despite the presence of singularities at boundaries and the free boundary.