Finite Element Methods · Chapter 10 of 10 · Intermediate
Convergence, Errors, and Modeling
An FEA result is a number with an error attached. This closing chapter makes that error visible: refine the mesh until the answer settles, balance the reactions, and check it against theory.
Readiness check
This closing chapter quantifies trust. Tick only what you can do closed-notes.
- Recall that finer meshes are more accurate.
- Compute a percentage error.
- Recall the cantilever deflection PL³/3EI.
- Sum reaction forces and moments.
- Distinguish a numerical error from a modeling error.
The core idea
FEA results carry discretization, element, and modeling error. Refining the mesh until the answer stops changing controls the first; balancing reactions and comparing to theory catches the rest.
refine mesh → result convergesΣ reactions = applied loadcompare to a closed-form benchmarkA finite element model is an approximation, and three errors separate it from reality: discretization error from a finite mesh, element error from the assumed shape functions, and modeling error from idealised geometry, materials, and boundary conditions. Discretization error shrinks as the mesh is refined, either by smaller elements (h-refinement) or higher-order elements (p-refinement); a convergence study repeats the analysis on finer meshes until the quantity of interest stops changing. But a converged answer can still be wrong if the model is wrong, so two further checks are essential: the reactions must sum to the applied load (equilibrium, confirming the boundary conditions), and the result must agree with a closed-form benchmark where one exists. Some geometries, like sharp re-entrant corners, produce stress singularities that never converge, a trap to recognise rather than refine.
The skills, taught in order
Five skills name the error sources, the refinement strategies, the convergence study, the validation checks, and the singularity trap.
10.1 Sources of error
An FEA result deviates from reality through discretization error (a finite mesh), element error (the assumed shape functions cannot capture every field), and modeling error (idealised geometry, materials, loads, and supports). Naming the dominant source decides how to reduce it.
| Check | What it confirms |
|---|---|
| Mesh convergence | discretization error is small |
| Reaction balance | boundary conditions and equilibrium |
| Hand-calc comparison | the model setup is correct |
| Stress continuity | the mesh resolves the gradients |
10.2 h and p refinement
Discretization error is reduced two ways: h-refinement uses more, smaller elements; p-refinement uses higher-order elements with richer shape functions. Both improve accuracy, and a good study refines systematically rather than randomly.
10.3 The convergence study
Running the analysis on successively finer meshes and plotting the quantity of interest shows whether it has converged. When refining further changes the result by a negligible amount, the discretization error is acceptable and the mesh is adequate.
10.4 Equilibrium and hand-check validation
Two checks validate the model. The reactions must sum to the applied load, confirming the boundary conditions and equilibrium. And the result must match a closed-form benchmark, a beam deflection or an axial stress, wherever one exists, confirming the setup.
10.5 Stress singularities and traps
At a sharp re-entrant corner or a point load, the theoretical stress is infinite, so refining the mesh makes the stress climb without limit. Recognising such singularities, rather than chasing a converged stress that does not exist, is part of sound modeling.
Engineering connection: this verify-against-theory discipline, opened in Chapter 1, is what turns every element in the course into a defensible engineering result.
Worked example 1: a convergence study
A cantilever's tip deflection has the closed-form value 0.4167 mm. A coarse solid mesh gives 0.40 mm and a refined mesh gives 0.415 mm. Find the percent error of each and comment on convergence.
- ProblemFind the percent error of each mesh in Figure 1 and judge convergence.
- Given / findExact 0.4167 mm, coarse 0.40 mm, fine 0.415 mm. Find the two percent errors.
- AssumptionsThe closed-form value is the reference; the meshes are systematically refined.
- ModelPercent error is the absolute difference from exact, divided by exact, times 100.
- Equationserror = |FEA − exact|/exact × 100
- SolveCoarse: |0.40 − 0.4167|/0.4167 × 100 = 4.0 percent. Fine: |0.415 − 0.4167|/0.4167 × 100 = 0.41 percent.
- CheckThe error falls by roughly a factor of ten as the mesh refines, and the fine result is within half a percent of theory, so the solution has effectively converged. A further refinement should change it negligibly.
- ConclusionA convergence study turns a single, untrustworthy number into a result with a known, small error. Refining until the answer settles is the minimum standard.
Worked example 2: an equilibrium validation check
A cantilever of length 0.4 m carries a 500 N tip load. Find the reaction force and reaction moment that the fixed support must supply, and explain how this validates an FEA model.
- ProblemFind the reaction force and moment for the cantilever in Figure 2 and its validation role.
- Given / findP = 500 N, L = 0.4 m. Find R and M, and how they validate FEA.
- AssumptionsStatic equilibrium; the support carries all the applied load.
- ModelVertical equilibrium gives R = P; moment equilibrium about the support gives M = P·L.
- EquationsR = PM = P·L
- SolveR = 500 N. M = 500 × 0.4 = 200 N·m.
- CheckThe FEA reactions, summed over the constrained nodes, must equal 500 N and 200 N·m. If they do not, the boundary conditions or load application are wrong, regardless of how converged or colorful the contour plot is.
- ConclusionThe reaction balance is a cheap, decisive validation: it confirms the loads and supports before any stress result is trusted. Paired with a hand-calc and a convergence study, it completes the verification triangle.
Misconceptions and diagnostics
| Mistake | Symptom | Diagnostic question | Correction |
|---|---|---|---|
| Trusting one mesh | A result with no error estimate | "Did the answer change on refinement?" | Run a convergence study before reporting. |
| Chasing a singular stress | Stress climbs without limit at a corner | "Is there a sharp re-entrant corner?" | Recognise the singularity; do not refine it forever. |
| Skipping the reaction check | Boundary-condition error undetected | "Do reactions sum to the applied load?" | Balance the reactions as a validation. |
| Converged but invalid | A precise answer to the wrong problem | "Is the model itself correct?" | Compare to a hand calculation, not just to a finer mesh. |
Practice ladder
An exact stress is 120 MPa and the FEA gives 114 MPa. Find the percent error.
Show answer
error = |114 − 120|/120 × 100 = 5.0 percent.
For the Worked Example 2 cantilever, what reaction moment would a 800 N tip load require?
Show answer
M = P·L = 800 × 0.4 = 320 N·m, and the reaction force would be 800 N.
Three meshes give a stress of 180, 195, and 199 MPa. Has the result converged, and what would you report?
Show answer
The last two differ by 4 MPa (about 2 percent), much less than the first gap of 15 MPa, so it is nearly converged. Report about 200 MPa, ideally with one more refinement to confirm.
For a real FEA you would run, describe the convergence study, the equilibrium check, and the hand calculation you would use to validate it.
What good work looks like
A systematic mesh refinement to convergence, a reaction-balance check, and a closed-form benchmark, with a stated awareness of any stress singularities.
Working with AI, and proving it yourself
Use AI as an examiner, not a solver
Portfolio task
Validate an FEA result end to end: a mesh-convergence study with percent errors, a reaction-balance check, and a comparison to a closed-form benchmark, noting any singularities.
Retrieval and spaced review
Closed notes. Answer out loud, then reveal.
1. Name the three error sources.
Discretization, element, and modeling error.
2. What are h and p refinement?
h uses smaller elements; p uses higher-order elements.
3. What does a convergence study show?
Whether the quantity of interest stops changing as the mesh refines.
4. What does the reaction check confirm?
That the boundary conditions are right and equilibrium holds.
5. Why do some stresses never converge?
Sharp re-entrant corners and point loads create theoretical singularities.
Textbook mapping
| Item | Mapping |
|---|---|
| Primary source | Logan, A First Course in the Finite Element Method, Chapter 7 (Practical Considerations in Modeling) |
| Cross-reference | Hutton, Ch. 9 · CFD (verification and validation) |
| Core topics | 10.1 Error sources · 10.2 h and p refinement · 10.3 Convergence study · 10.4 Equilibrium and hand-check · 10.5 Singularities |
| Engineering connection | Closes the verify-against-theory discipline opened in Chapter 1. |
| Read next | Return to the FEM hub and integrate the workflow. |