Finite Element Method FEM ★
Definition
The classical numerical method that discretizes a domain into finite elements and builds an approximate solution with shape-function interpolation inside each element. The de facto standard for solid mechanics and structural analysis, and the source of most Physics-AI training data.
What it is
FEM converts the strong form of a PDE into a weak (variational) form and looks for an approximate solution in a finite-dimensional subspace: the domain is meshed into elements (triangles, tetrahedra, hexahedra), polynomial shape functions interpolate inside each element, and the assembly yields a large sparse linear (or nonlinear) system.
Key points
- Weak form plus integration by parts lowers the smoothness requirement on the solution — this is why FEM handles complex geometry and discontinuous materials.
- Unstructured-mesh friendly: meshes can hug arbitrary geometric boundaries, its biggest advantage over spectral and finite-difference methods.
- Controllable error: mature a priori/a posteriori error estimates and adaptive mesh refinement (h/p adaptivity).
- It is the data generator: most structural and heat-transfer datasets and cases on this site come from FEM solvers.
Common pitfalls / boundaries
- "Finer mesh is always better" is wrong — separate discretization error from model error; the latter cannot be removed by refinement.
- FEM and Physics-AI are complements, not substitutes: AI accelerates or inverts, FEM verifies and generates data.