FEEC convergence theorem (Arnold-Falk-Winther)

Intuition Beginner

FEEC's convergence theorem explains why compatible finite elements work.

The main message is simple: if the finite element spaces preserve the differential complex and there is a stable projection from smooth fields to mesh fields, then the numerical solution converges for the right structural reason.

This is stronger than checking each element family one at a time. FEEC gives one theorem that covers scalar elliptic problems, flux problems, curl problems, and Hodge Laplacian problems in a shared language.

The theorem says the mesh approximation is stable because it respects the geometry of the PDE.

Visual Beginner

The discrete problem converges because the projection, complex, and stability estimates fit together.

Worked example Beginner

For a scalar Poisson problem, conforming FEM works because the finite element space sits inside the weak energy space and approximates the true solution.

For a curl or divergence problem, that is not enough. The finite element spaces also need to preserve the sequence of differential operators.

FEEC supplies the common rule: build a finite element subcomplex, add a bounded commuting projection, and then prove stability and convergence through the complex.

Check your understanding Beginner

Formal definition Intermediate+

The continuous Hodge Laplacian problem on $k$-forms uses the Hilbert complex $$ H\Lambda^{k-1}\xrightarrow{d}H\Lambda^k\xrightarrow{d}H\Lambda^{k+1}. $$ In mixed form, one solves for variables such as $$ (\sigma,u)\in H\Lambda^{k-1}\times H\Lambda^k, $$ where $\sigma$ represents the codifferential part of $u$.

FEEC chooses finite-dimensional spaces $$ \Lambda_h^{k-1}\subset H\Lambda^{k-1}, \qquad \Lambda_h^k\subset H\Lambda^k $$ inside a finite element subcomplex. It also assumes a bounded cochain projection $$ \pi_h^\bullet\Lambda^\bullet\to \Lambda_h^\bullet $$ commuting with $d$.

The FEEC convergence theorem states, in representative form, that the discrete mixed problem is stable and that the error is bounded by best-approximation terms plus data/projection terms. In a simplified smooth case, one obtains an estimate of the form $$ |u-u_h|{H\Lambda^k} +|\sigma-\sigma_h|{H\Lambda^{k-1}} \leq C h^r\bigl(|u|{H^{r+1}}+|\sigma|{H^r}\bigr), $$ with conventions depending on the family, degree, domain regularity, harmonic components, and norms used.

The constant $C$ must be independent of mesh size over a shape-regular mesh family.

Counterexamples to common slips

• FEEC convergence is not just polynomial approximation. Stability of the complex is essential.
• The bounded projection is not optional; it is the tool that transfers continuous estimates to the discrete complex.
• Harmonic forms and cohomology cannot be ignored in Hodge Laplacian problems.

Key theorem with proof Intermediate+

Theorem (FEEC convergence pattern). Let ${\Lambda }_h^{\bullet }$ be a finite element subcomplex of a Hilbert de Rham complex, and suppose there exists a uniformly bounded cochain projection onto it. Then the discrete mixed Hodge Laplacian problem is stable, and its solution converges quasi-optimally to the continuous solution, with error controlled by approximation in the finite element spaces and harmonic-projection terms.

Proof. The finite element spaces form a subcomplex by 24.03.04, so the discrete exterior derivative satisfies the same $d^2=0$ structure as the continuous derivative. This gives a well-defined discrete Hilbert complex with closed, exact, and harmonic parts.

The bounded cochain projection from 24.03.05 transfers continuous Poincare-type inequalities to the discrete complex. In particular, discrete forms orthogonal to the discrete kernel are controlled by their discrete derivative with constants independent of mesh size. This is the discrete Poincare-Friedrichs estimate.

That estimate supplies the stability needed for the mixed Hodge Laplacian saddle-point problem, in the same role that the Babuška-Brezzi condition plays for mixed methods 24.01.04. Stability gives a uniform bound for the discrete solution.

Finally, Galerkin orthogonality and approximation by ${\Lambda }_h^{\bullet }$ compare the discrete solution with suitable projected or interpolated continuous forms. The projection commutes with $d$, so the derivative terms are approximated in compatible spaces. Combining stability and approximation gives the FEEC a priori estimate. $\square$

Bridge. This theorem is the load-bearing synthesis of 24.01.02, 03.04.15, 24.03.03, 24.03.04, and 24.03.05. It is the reason FEEC can treat mixed Poisson, Maxwell, and Hodge Laplacian discretizations through one framework rather than separate ad hoc arguments.

Exercises Intermediate+

Advanced results Master

The FEEC convergence theorem is best understood as a Hilbert-complex version of stable Galerkin theory. The continuous Hodge Laplacian has a decomposition into exact, coexact, and harmonic parts. A stable discretization must approximate all three without corrupting the complex structure.

The finite element subcomplex gives the discrete analogue of $d^2=0$. The bounded cochain projection gives a stable comparison map from the continuous complex to the discrete one. Together, they imply a discrete Poincare inequality: on the correct orthogonal complement of the kernel, the discrete norm is controlled by the discrete derivative.

That discrete Poincare inequality is the central stability step. It prevents spurious modes and gives uniform well-posedness for the mixed discrete Hodge Laplacian. In saddle-point language, it supplies the FEEC analogue of the inf-sup and kernel-coercivity estimates.

The error estimate then follows a Galerkin pattern. Orthogonality reduces the error to approximation terms, but the approximation must be compatible with $d$. This is why commuting projection matters: it approximates the form and its derivative in a way that respects the complex.

Harmonic forms introduce additional terms. On topologically nontrivial domains, the continuous and discrete harmonic spaces must be compared. FEEC controls this through projection and cohomology compatibility, ensuring that the numerical method does not create false topology or erase real topology.

Synthesis. The FEEC convergence theorem is the payoff for the whole construction. Polynomial differential forms provide approximating spaces; the subcomplex axiom preserves differential identities; bounded cochain projections transfer stability; Hilbert-complex Hodge theory handles kernels and harmonic forms. The result is a unified convergence theorem for compatible discretizations of elliptic complexes.

Full proof set Master

Proposition 1 (bounded projection transfers closed/exact structure). A bounded cochain projection maps closed forms to closed discrete forms and exact forms to exact discrete forms with uniform norm control.

Proof. Closed and exact preservation follow from 24.03.05. Uniform norm control follows from boundedness: $$ |\pi_h^k\omega|\leq C|\omega|. $$ The constant is independent of mesh size, so the transfer remains stable under refinement. $\square$

Proposition 2 (discrete Poincare estimate, proof pattern). If a bounded cochain projection exists and the continuous complex has a Poincare inequality, then the discrete complex has a mesh-uniform Poincare inequality on the corresponding discrete complement.

Proof. Suppose no uniform discrete inequality existed. Then one could choose discrete forms with unit norm, orthogonal to the discrete kernel, whose derivatives tend to zero. Boundedness of the projection and finite element compactness arguments produce a limiting continuous form in the continuous kernel. The commuting projection and orthogonality force the limit to contradict the unit norm or complement condition. Thus a uniform discrete Poincare constant must exist. $\square$

Proposition 3 (quasi-optimal FEEC error pattern). Stability plus Galerkin orthogonality bounds the discrete Hodge Laplacian error by best compatible approximation terms.

Proof. Stability of the discrete mixed problem gives an estimate of the discrete error in terms of the residual against discrete test functions. Galerkin orthogonality cancels the residual for the discrete solution. Insert a projected approximation ${\pi }_{h}u$ and ${\pi }_h\sigma$; the commuting property ensures derivative terms are approximated in the correct adjacent spaces. The resulting bound is controlled by the approximation errors of $u$, $\sigma$, and their derivatives, plus harmonic comparison terms. Polynomial approximation estimates then turn those terms into powers of $h$ under smoothness assumptions. $\square$

Connections Master

• Sobolev spaces of differential forms 24.01.02. FEEC convergence is measured in $H{\Lambda }^k$ and related graph norms.

• Babuška-Brezzi condition 24.01.04. The theorem generalizes mixed stability ideas to Hilbert complexes.

• Hodge Laplacian 03.04.15. The mixed Hodge Laplacian is the model elliptic complex problem.

• Discrete de Rham complex 24.03.04. The subcomplex axiom supplies the discrete chain structure.

• Bounded cochain projection 24.03.05. The projection supplies the stability bridge between continuous and discrete complexes.

Historical & philosophical context Master

Arnold, Falk, and Winther's 2006 Acta Numerica article gave the definitive FEEC synthesis of finite element spaces, homological techniques, and stability theory ^{[Arnold-Falk-Winther]}. Their 2010 Bulletin article made the narrative explicit: Hodge theory explains numerical stability ^{[Arnold-Falk-Winther Bulletin]}.

Before FEEC, many stability results were proved separately for Lagrange, Raviart-Thomas, Nédélec, and related elements. FEEC showed that these proofs share a common skeleton: finite element subcomplexes, bounded cochain projections, and Hilbert-complex estimates.

The philosophical contribution is that convergence is not only an approximation phenomenon. It is a structure-preservation phenomenon. The numerical spaces must approximate the solution and preserve the algebraic topology of the differential operators.

Bibliography Master

@article{ArnoldFalkWinther2006Convergence,
  author = {Arnold, Douglas N. and Falk, Richard S. and Winther, Ragnar},
  title = {Finite element exterior calculus, homological techniques, and applications},
  journal = {Acta Numerica},
  volume = {15},
  pages = {1--155},
  year = {2006}
}

@article{ArnoldFalkWinther2010Convergence,
  author = {Arnold, Douglas N. and Falk, Richard S. and Winther, Ragnar},
  title = {Finite element exterior calculus: from Hodge theory to numerical stability},
  journal = {Bulletin of the American Mathematical Society},
  volume = {47},
  pages = {281--354},
  year = {2010}
}

@book{Arnold2018Convergence,
  author = {Arnold, Douglas N.},
  title = {Finite Element Exterior Calculus},
  series = {CBMS-NSF Regional Conference Series in Applied Mathematics},
  volume = {93},
  publisher = {SIAM},
  year = {2018}
}

@book{BoffiBrezziFortin2013FEEC,
  author = {Boffi, Daniele and Brezzi, Franco and Fortin, Michel},
  title = {Mixed Finite Element Methods and Applications},
  series = {Springer Series in Computational Mathematics},
  volume = {44},
  publisher = {Springer},
  year = {2013}
}