Quasi-Monte Carlo finite element methods for elliptic PDEs with lognormal random coefficients (Q495544)

The authors analyze a class of numerical methods for the diffusion problem with random coefficients of the form \[ -\nabla\big(a(x,\omega)\cdot\nabla u(x,\omega)\big)=f(x)\text{ for almost all }\omega\in\Omega\text{ and }x\in D, \] subject to the homogeneous Dirichlet condition \(u=0\) on \(\partial D\), where \(D\) is a bounded (spatial) domain in \(\mathbb R_{d}\) and \((\Omega,A,P)\) is a probability space. The authors focus on the lognormal case, assuming that \[ a(x,\omega)=a_{\star}(x)+a_0(x)\exp\big(Z(x,\omega)\big), \] where \(Z\) is a zero mean Gaussian random field and \(a_{\star},a_0\) are given functions continuous on \(\bar{D}\). The goal is to find \(u(\cdot,\omega)\in H_0^1(D)\) such that \[ B(\omega;u,v)=<f,v> \quad \forall v\in H_0^1(D) \text{ and for almost all }\omega\in\Omega, \] where \[ B(\omega;u,v)=\int_D a(x,\omega)\nabla w(x)\cdot \nabla v(x)\,dx, \quad w,x\in H^1(D),\;f\in \big(H_0^1(D)\big)^\prime. \] Motived by applications in uncertainty quantification, the authors are interested in an expected value of a linear function of the solution. They demonstrate the good convergence of their quasi-Monte Carlo finite element method for the considered class of partial differential equation (PDE) problems both theoretically and numerically.