Computational Mechanics

Mesh Convergence: Validating FEA Against Analytical and Hand-Coded Solutions

A cantilever beam analysed in Abaqus across three mesh densities and element types, verified against closed-form solutions and independently reimplemented in MATLAB using Gauss quadrature to test integration schemes.

By Musa Ibne Mannan · Erik Jonsson School of Engineering and Computer Science, The University of Texas at Dallas

Why this study exists

Finite element software will return an answer for any mesh you give it. Whether that answer is correct is a separate question, and it is the one that separates competent analysis from expensive-looking output. A cantilever beam under shear loading has a known closed-form solution, which makes it the right vehicle for the underlying question: how do you establish that a finite element result can be trusted?

The study took a 10 m long, 1 m high cantilever beam with a 2000 N shear load and attacked it three ways — commercial FEA, analytical theory, and hand-written numerical code — then compared them.

Abaqus analysis

Static structural analysis was run using three element formulations to isolate the effect of element choice from the effect of mesh density:

• CPS4 — 4-node linear plane-stress elements
• CPS8 — 8-node quadratic plane-stress elements
• C3D8 — 8-node linear brick elements for the full 3D case

Each was solved at coarse, medium and fine mesh densities, with deflection and stress compared against the analytical solution at every step.

Convergence results

The error reduction with mesh refinement was substantial, and the coarse-mesh figures are worth dwelling on.

A coarse mesh produced stress predictions at the critical mid-span location that were wrong by more than 94 percent. The simulation ran without error, produced plausible contour plots, and was almost entirely incorrect. Refining to a fine mesh brought that to roughly 8.6 percent. The stress distribution itself matched theory throughout, showing maximum concentration near the fixed end at the wall — so the qualitative picture was right even when the numbers were badly wrong. That is exactly the trap: a result that looks correct because the pattern is correct.

Independent verification in MATLAB

Comparing Abaqus to theory checks the mesh but not the solver's behaviour on the discretization scheme itself. To probe that, I wrote custom MATLAB code implementing 2×2 Gauss quadrature for Q4 elements, solving the plane-stress problem independently, with a Q9 element extension.

This produced a finding the commercial solver alone would not have exposed. The 2×2 Gauss quadrature scheme gave a distinctly superior and more accurate stress distribution than the simpler one-point quadrature method. One-point integration is computationally cheaper and adequate for smooth fields, but it under-integrates the element and degrades badly where stress gradients are steep — which is precisely where structural analysis needs accuracy, near constraints and load introduction points.

The transferable lesson

Three independent routes to the same problem — commercial FEA, closed-form theory, and hand-coded numerics — is what converts a simulation into a validated result. The mesh convergence study establishes that the discretization is adequate; the analytical comparison establishes the model is correctly posed; the independent implementation establishes that the numerical machinery behaves as assumed. In failure analysis, where a simulation may support a conclusion about why something broke, that chain of verification is not academic diligence. It is the difference between an opinion and an engineering finding.