Discussion Forum

I think that your result reflects the inaccuracy of Comsol in fluid dynamics. The same applies to electrochemistry: 3D models are more or less inaccurate. Richard Compton has written a paper of this.

ABSTRACT Electrochemical simulation via the solution of Fick’s Laws is a widely used technique to corroborate experimental results with well defined theory. This paper analyses use of ‘off-the-shelf’ finite element (FEM) software COMSOL Multiphysics in one, two and three-dimensional quantitative problems and under homogeneous and heterogeneous kinetic systems. Conclusions indicate that two-dimensional problems are within an order of magnitude of accuracy of finite difference simulations and analytical solutions, as long as the problem is well defined in the software and care is taken with regards to appropriate meshing and boundary conditions. Three-dimensional simulations relating to microdiscs result in steady-state current values not quantitatively compatible with experimental observations or analytical solutions.

Journal of Electroanalytical Chemistry 638 (2010) 76–83

Computational fluid dynamics is a very hard task, harder than electrochemistry.

Stating that COMSOL Multiphysics gives inaccurate solutions for fluid dynamics and electrochemistry problems is an inaccurate statement in itself. COMSOL Multiphysics is not one technology, it contains different finite element methods, boundary element methods, meshing techniques, and time stepping methods. So it contains a wide range of different numerical techniques suitable for solving a wide range of problems.

The group that wrote this nine-year old paper became avid users of COMSOL Multiphysics. In fact, one of the authors of the paper, Edmund Dickinson, worked for COMSOL for many years after leaving university and he also made valuable contributions to our development in electrochemistry. The results of some of the boundary fluxes in this paper were obtained through interpolation. This way of computing boundary fluxes would have benefitted from a structured boundary layer mesh, a technique that was not available at that time in COMSOL Multiphysics but is available now. Another option would have been to use the reaction forces to compute accurate boundary fluxes. This technique was available at that time but was not very easy to find.

I believe that it has been convincingly shown over the years, in a number of scientific papers, that our technology is state-of-the-art for the numerical methods that we offer. With a proper discretization, meshing, solver settings, and tolerances, we are as good as the numerical methods allow us to be. Our products in electrochemistry are by far the most popular tools for modeling and simulations among scientists specialized in this field. In fluid mechanics there are also a number of extensive research papers and reports that show that we are able to solve difficult problems in 2D and 3D. See, for example:

https://www.researchgate.net/publication/287217415/download

We offer finite element methods, discontinuous Galerkin methods, boundary element methods, method of lines, explicit time stepping, and implicit time stepping for electrochemistry and fluid mechanics as well as other fields. These methods have been specifically developed by the numerical analysis community for solving such problems. While there will always be debates regarding the suitability of particular methods for particular problems, there is nothing to suggest that the methods implemented in COMSOL Multiphysics are in general less accurate than other methods. To choose between the different methods, and we offer many possibilities, is an art. We devote a lot of efforts to provide the best default settings for different problems. However, every problem is unique and a skilled user can always fine-tune the defaults for the problem at hand. I am sure that our support can help Rita with her specific problem.

Best regards,

Ed Fontes, COMSOL

Rita - With reference to the original question, it is certainly the case in my experience that the default "Normal" mesh is not sufficient to achieve < 1% accuracy for predicting maximum velocities in 3D Poiseuille flow problems in cylinders, when pipes are appreciably long. This is because COMSOL's default mesh sizing is based on the total size of the geometry, whereas for long pipes, the cross-sectional mesh is much more important. Consider that in an established laminar flow, the velocity field is only a function of the cross-sectional coordinates. In such a case, a structured mesh such as a Swept mesh is likely to give more accurate results. I created several such benchmarks while working with COMSOL UK - I expect that COMSOL Support can help here, as suggested by Ed at COMSOL.

Lasse - As a co-author of the paper you cite, I do not believe its conclusions are relevant to the original question.

The conclusions of Cutress et al. 2010 were reached based on the capabilities of hardware that was already getting out-of-date in 2010 when we did the study - 32-bit OS and hence max 4 GB RAM. Its conclusions also applied to COMSOL Multiphysics 3.5a and were revised later with respect to improved capabilities in later versions: https://www.sciencedirect.com/science/article/pii/S1388248113004840. In the latter study we rigorously checked analytical steady-state currents and demonstrated that good accuracy could be achieved in the 3D case, provided the mesh was refined appropriately. I also feel that conclusions applied specifically to analytical measures in one field (electroanalytical chemistry) should not be projected to fluid flow. I don't think it's meaningful to say that one field of computational physics is 'harder' than the other - it depends what problem you are trying to solve and what accuracy you actually require.

Edmund Dickinson National Physical Laboratory, Teddington, UK