Finite Element Methods: A Priori, A Posteriori, and Adaptivity¶
Job: develop the Galerkin / Céa, the a priori (Bramble–Hilbert / Scott–Zhang), the a posteriori (\(\eta_K\) / Dörfler adaptivity), and the mixed / Stokes (inf-sup / Babuška–Brezzi) — with the Poisson FEM as the worked example.
Core¶
The Galerkin framework and Céa¶
| Galerkin / Céa / \(V_h\) | Description |
|---|---|
| Galerkin | \(a(u_h,v_h)=\ell(v_h)\) \(\forall v_h\in V_h\) (the weak form, \(V_h\subset H^1_0\)) |
| Céa lemma | \(\|u-u_h\|_{H^1}\le C\inf_{v\in V_h}\|u-v\|_{H^1}\) (the coercive) |
| \(V_h\) | the \(P_k\) / \(Q_k\) finite element space (the mesh, the degree) |
The a priori error estimates¶
| A priori / Bramble–Hilbert / \(P_k\) | Description |
|---|---|
| A priori | \(\|u-u_h\|_{H^1}\le Ch^k\|u\|_{H^{k+1}}\) (\(P_k\) elements, \(u\in H^{k+1}\)) |
| Bramble–Hilbert | the interpolation error $|u-I_h u|_{H^1}\le Ch^k |
| Scott–Zhang | the \(H^1\)-stable interpolation (the \(I_h:H^1\to V_h\), the non-nodal) |
The a posteriori and the adaptivity¶
| A posteriori / \(\eta_K\) / Dörfler | Description |
|---|---|
| A posteriori | \(\eta_K\) the element indicator, \(\|u-u_h\|\le C(\sum\eta_K^2)^{1/2}\) (the reliability) |
| Dörfler | the bulk chasing (the marking the \(\eta_K\) large, the refinement, the convergence) |
| Adaptivity | the AFEM (the solve \(\to\) estimate \(\to\) mark \(\to\) refine, the convergence) |
The mixed / the Stokes / the inf-sup¶
| Mixed / inf-sup / Stokes / Taylor–Hood | Description |
|---|---|
| Mixed FEM | \(V_h\times Q_h\), the inf-sup / the Babuška–Brezzi (the stable) |
| Stokes | the velocity-pressure, the Taylor–Hood (\(P_2\)-\(P_1\)), the inf-sup |
See it¶
Poisson FEM a priori \(H^1\) / a posteriori \(\eta_K\) and the adaptivity. The Poisson \(-\Delta u=f\) on \(\Omega\), the \(V_h\subset H^1_0\) the \(P_1\), the \(\|u-u_h\|_{H^1}\le Ch\|u\|_{H^2}\) the a priori (the Bramble–Hilbert). The a posteriori: the \(\eta_K=h_K\|f+\Delta u_h\|_{L^2(K)}+\frac12 h_K^{-1/2}\|[\partial_n u_h]\|_{L^2(\partial K)}\) the residual, the \(\|u-u_h\|\le C(\sum\eta_K^2)^{1/2}\) the reliability. The adaptivity: the Dörfler the bulk, the marking the \(\eta_K\) large, the refinement, the convergence. The key point: the Poisson the FEM a priori / a posteriori, the adaptivity the Dörfler.
Check¶
Show that the Céa lemma gives the a priori error, and explain the Bramble–Hilbert.
The Céa: the \(a(u-u_h,v_h)=0\) (the Galerkin orthogonality), the coercive \(a(v,v)\ge\alpha\|v\|^2\), the \(\alpha\|u-u_h\|^2\le a(u-u_h,u-u_h)=a(u-u_h,u-v_h)\le M\|u-u_h\|\|u-v_h\|\), the \(\|u-u_h\|\le(M/\alpha)\inf\|u-v_h\|\) (the Céa). The Bramble–Hilbert: the \(\|u-I_h u\|_{H^1}\le Ch^k|u|_{H^{k+1}}\) (the interpolation, the \(P_k\), the \(H^{k+1}\)). So the Céa + Bramble–Hilbert the a priori.
Explain the a posteriori \(\eta_K\) and the Dörfler bulk marking, and the AFEM convergence.
The \(\eta_K\): the \(h_K\|f+\Delta u_h\|_{L^2(K)}\) the interior residual, the \(\frac12 h_K^{-1/2}\|[\partial_n u_h]\|_{L^2(\partial K)}\) the edge jump, the \(\|u-u_h\|\le C(\sum\eta_K^2)^{1/2}\) the reliability (the a posteriori). The Dörfler: the mark the \(\eta_K\) the bulk (\(\sum_{\mathrm{marked}}\eta_K^2\ge\theta\sum_{\mathrm{all}}\eta_K^2\), the \(\theta\) the bulk parameter). The AFEM: the solve \(\to\) estimate \(\to\) mark \(\to\) refine, the convergence (the Cascon–Kreuzer–Nochetto–Siebert the contraction). So the \(\eta_K\) the reliability, the Dörfler the bulk, the AFEM the convergence.
Pitfalls¶
- The Céa the coercive (the \(a(v,v)\ge\alpha\|v\|^2\), the PDE / elliptic) — the Céa the coercive, the non-coercive the Banach–Nečas–Babuška.
- The Bramble–Hilbert the \(H^{k+1}\) (the \(u\in H^{k+1}\), the \(P_k\)) — the \(H^{k+1}\) the regularity, the \(P_k\) the degree.
- The a posteriori the reliability (the \(\eta_K\), the \(\|u-u_h\|\le C(\sum\eta_K^2)^{1/2}\)) — the reliability the upper, the efficiency the lower.
- The mixed / Stokes the inf-sup (the Babuška–Brezzi, the Taylor–Hood) — the inf-sup the stable, the \(P_2\)-\(P_1\) the Taylor–Hood.
Refs¶
- Brenner–Scott, The Mathematical Theory of Finite Element Methods (2008); the Céa / the Bramble–Hilbert
- Babuška–Rheinboldt, "Error estimates for adaptive FEM" (1978); the a posteriori
- Dörfler, "A convergent adaptive algorithm for Poisson" (1996); the bulk marking
- Brezzi–Fortin, Mixed and Hybrid Finite Element Methods (1991); the inf-sup
Next¶
NA index · Foundations · Geometric Integration · Iterative & Multigrid · High-Dimensional & UQ · PDE / elliptic