A critical analysis of some popular methods for the discretisation of the gradient operator in finite volume methods

Abstract

Finite volume methods (FVMs) constitute a popular class of methods for the numerical simulation of fluid flows. Among the various components of these methods, the discretisation of the gradient operator has received less attention despite its fundamental importance with regards to the accuracy of the FVM. The most popular gradient schemes are the divergence theorem (DT) (or Green-Gauss) scheme, and the least-squares (LS) scheme. Both are widely believed to be secondorder accurate, but the present study shows that in fact the common variant of the DT gradient is second-order accurate only on structured meshes whereas it is zeroth-order accurate on general unstructured meshes, and the LS gradient is second-order and first-order accurate, respectively. This is explained through a theoretical analysis and is confirmed by numerical tests. The schemes are then used within a FVM to solve a simple diffusion equation on unstructured grids generated by several methods; the results reveal that the zeroth-order accuracy of the DT gradient is inherited by the FVM as a whole, and the discretisation error does not decrease with grid refinement. On the other hand, use of the LS gradient leads to second-order accurate results, as does the use of alternative, consistent, DT gradient schemes, including a new iterative scheme that makes the common DT gradient consistent at almost no extra cost. The numerical tests are performed using both an in-house code and the popular public domain PDE solver OpenFOAM. This is the accepted version of the article published in:

A Finite Volume approximation of the Navier-Stokes equations with nonlinear filtering stabilization

“…In OpenFOAM R solvers, the face center pressure values q j are typically obtained from the cell center values by means of a linear interpolation scheme. Such scheme is rigorously secondorder accurate only on structured meshes [46].…”

Theoretical study of the flow in a fluid damper containing high viscosity silicone oil: Effects of shear-thinning and viscoelasticity

“…Concerning the discretisation of the momentum and continuity equations, we used the same method as in our previous work [22], which employs central differencing with least-squares gradients [54] in space and (Table 4) near a piston corner. a three time level implicit scheme in time, which are second-order accurate.…”

A finite volume method for the simulation of elastoviscoplastic flows and its application to the lid-driven cavity case

“…Nevertheless, firstorder accurate gradients suffice for second-order accuracy of the differentiated variables, i.e. are compatible with second-order accurate FVMs [44].…”

“…The other components of these diffusion terms are discretised in exactly the same way for both of them and cancel out, leaving only D τ + f and D τ − f . Since both of these diffusion terms are discretised by central differences they contribute O(1) to the truncation error [44] but O(h 2 ) to the discretisation error. In particular, consider that Eq.…”

“…6a, "CU") or non-uniform ( Fig. 6b, "CN") Cartesian grids, we also employed grids obtained from the CU ones by randomly perturbing their vertices as described in [44]. In particular, from a N × N CU grid (Fig.…”