The bar and the meshes
Shading shows the exact stress P/A(x), darker where the bar is thinner. Filled dots are element end nodes; open dots are the mid-side nodes of the quadratic elements.
Results
| Model | Elements | Unknowns | u(L) | Error in u(L) | Energy-norm error | Max stress error |
|---|
Max stress error is the largest |σh − σ| along the bar divided by the largest exact stress, P/A(L). With exact integration, the energy-norm error is √(1 − uh(L)/u(L)).
Displacement u(x)
Stress and axial force
Convergence
Things to try
- Reproduce p. 104. At m = 2/3 with one element each, the linear element gives u(L) = 1.500 (−9.0%) and the quadratic element gives 1.636 (−0.7%). The exact value is 1.648.
- Set m = 0. The bar is prismatic, the exact field is linear, and every model is exact.
- Look at the sign of the errors. Every element solution is too stiff, so uh(L) ≤ u(L). The assumed fields minimize the total potential energy only within their own family.
- Compare at the same number of unknowns. Two linear elements and one quadratic element both have two unknowns. Which is closer?
- Switch the stress plot to axial force. The exact N equals P everywhere. The element solutions satisfy equilibrium only at the nodes, so N varies inside each element.
- Try the graded mesh with m near 0.9. Short elements at the thin end capture the steep stress there.
- Look at the exact element. One element is enough: its shape functions solve the governing equation (p. 103), so it reproduces the exact displacement and stress along the whole bar. Its stiffness uses the log-mean area where the linear element uses the arithmetic mean.
Element stiffness
Linear element
k = (E Ā / h) [1 −1; −1 1]
Ā = (Aa + Ab)/2, the arithmetic mean of the end areas.Exact element
k = (E Ã / h) [1 −1; −1 1]
à = (Aa − Ab) / ln(Aa/Ab), the log-mean of the end areas. Shape functions N = ln(A/Aa)/ln(Ab/Aa).Quadratic element
kij = ∫ E A(x) Ni′ Nj′ dx
Lagrange shape functions on nodes 0, h/2, h. One element over the whole bar gives the 3 × 3 matrix on p. 103.