On the positivity of the discrete Green's function for unstructured finite element discretizations in three dimensions (Q6142549): Difference between revisions

Harmonic functions have been well studied across many scientific disciplines including mathematics, stochastic processes and mathematical physics for many years and have many applications. Some important results concerning harmonic functions include many regularity results, Liouville's Theorem, weak and strong maximum principles and their mean value property is used often. An important and well known result for nonnegative harmonic functions is the Harnack inequality. The theorem essentially states that any two values of a non-negative harmonic function are comparable with a constant independent of the particular function itself. The Harnack inequality arises in many problems in analysis for example fully non-linear elliptic problems. This problem has been extended by Moser in 1961, for uniformly elliptic equations in divergence form with bounded measurable coefficients under the assumption that the eigenvalues of the matrix operator are bounded from above and below. In 1980, Krylov and Safonov extended this result to elliptic equations in non-divergence form with bounded measurable coefficients. The Harnak inequality appears in probability, graph theory, and infinite dimensional operators for example. In this interesting paper, the authors study criteria for the non-negativity of the discrete Green's function. Their main result shows that the discrete Green's function cannot be uniformly bounded at the singularity. In addition, the authors establish pointwise upper bounds and establish some \(L_p\) estimates for the discrete Green's function. The authors use techniques used by Leykekhman and Pruitt in two dimensions. In the case of three dimensions, the authors use techniques which for example require the stiffness matrix to be a \(M\)-matrix which gives positivity of the discrete Green's function, but with significant constraints on the triangulation that many mesh refinement schemes do not follow. The paper is well written with an excellent reference list.