Sobolev spaces of differential forms $H{\Lambda }^k$

Intuition Beginner

Some PDEs do not only ask whether a scalar function has weak derivatives. They ask whether a field has controlled rotation, controlled flux, or both.

In three-dimensional vector calculus, this leads to spaces for fields whose curl is square-integrable and spaces whose divergence is square-integrable.

Differential forms put these cases into one sequence. A scalar potential, a line-integral field, a flux field, and a volume density are all forms of different degrees. The exterior derivative moves from one degree to the next.

Sobolev spaces of differential forms measure a form and its exterior derivative together. This is the language FEEC uses to keep gradient, curl, divergence, and topology in one framework.

Visual Beginner

The same pattern can be read as a sequence of form spaces connected by the exterior derivative.

Worked example Beginner

In a heat equation, the main unknown may be a scalar temperature. In electromagnetism, the main unknown may be an electric field whose circulation matters. In fluid flow, the main unknown may be a flux field whose outflow matters.

These look like different types of unknowns, but FEEC treats them as members of one family. The degree of the form says what kind of geometric quantity it measures.

What this tells us: the right function space depends not only on smoothness, but also on which derivative the PDE uses.

Check your understanding Beginner

Formal definition Intermediate+

Let $\Omega \subset {\mathbb{R}}^n$ be a bounded domain. The space $L^2{\Lambda }^k(\Omega )$ consists of measurable differential $k$-forms whose coefficient functions are square-integrable. With the Euclidean metric, its inner product is $$ \langle \omega,\eta\rangle_{L^2\Lambda^k} =\int_\Omega \langle \omega(x),\eta(x)\rangle_{\Lambda^k},dx. $$

The Sobolev space of differential forms $$ H\Lambda^k(\Omega) $$ is the domain of the weak exterior derivative: $$ H\Lambda^k(\Omega)={\omega\in L^2\Lambda^k(\Omega): d\omega\in L^2\Lambda^{k+1}(\Omega)}. $$ Its graph norm is $$ |\omega|{H\Lambda^k}^2 =|\omega|{L^2\Lambda^k}^2+|d\omega|_{L^2\Lambda^{k+1}}^2. $$

In three-dimensional vector proxy notation, the sequence corresponds to $$ H^1(\Omega) \xrightarrow{\nabla} H(\operatorname{curl};\Omega) \xrightarrow{\operatorname{curl}} H(\operatorname{div};\Omega) \xrightarrow{\operatorname{div}} L^2(\Omega). $$ FEEC writes this uniformly as $$ H\Lambda^0\xrightarrow{d}H\Lambda^1\xrightarrow{d}H\Lambda^2\xrightarrow{d}H\Lambda^3. $$

Counterexamples to common slips

• $H{\Lambda }^k$ does not require every weak partial derivative of every coefficient to lie in $L^2$; it requires the exterior derivative to lie in $L^2$.
• $H(\mathrm{curl})$ and $H(\mathrm{div})$ have different boundary traces. Tangential and normal continuity are not interchangeable.
• The notation $H{\Lambda }^k$ is not a finite element space. It is the continuous Hilbert space that finite element spaces approximate.

Key theorem with proof Intermediate+

Theorem (Hilbert complex property). The spaces $H{\Lambda }^k(\Omega )$ with the weak exterior derivative form a Hilbert complex: $$ \cdots\to H\Lambda^{k-1}(\Omega)\xrightarrow{d}H\Lambda^k(\Omega)\xrightarrow{d}H\Lambda^{k+1}(\Omega)\to\cdots, $$ and $d\circ d=0$ in the weak sense.

Proof. Each $L^2{\Lambda }^k(\Omega )$ is a Hilbert space. The operator $d$ with domain $H{\Lambda }^k(\Omega )$ is closed because if ${\omega }_j\to \omega$ in $L^2{\Lambda }^k$ and $d{\omega }_j\to \mu$ in $L^2{\Lambda }^{k+1}$, then testing against smooth compactly supported forms and passing to the limit in the weak exterior-derivative identity shows that $\mu =d\omega$. Therefore $H{\Lambda }^k$ is complete in the graph norm.

For smooth forms, $d(d\omega )=0$. For a general $\omega \in H{\Lambda }^k$ with $d\omega \in H{\Lambda }^{k+1}$, approximate in the graph norm by smooth forms where available, or use the distributional identity directly. The distributional exterior derivative still satisfies $d^2=0$, since the test-form identity reduces to the smooth identity on compactly supported tests. Hence consecutive maps compose to zero. $\square$

Bridge. This unit refines 24.01.01 by replacing scalar weak derivatives with the weak exterior derivative. It depends on 03.04.02 and 03.04.04 for the smooth form language. The foundational reason is that many PDEs control only the geometrically relevant derivative: gradient for scalar potentials, curl for circulation fields, and divergence for flux fields. FEEC uses the single notation $H{\Lambda }^k$ so that all of these spaces sit inside one Hilbert complex.

Exercises Intermediate+

Advanced results Master

The space $H{\Lambda }^k$ is the natural domain of the exterior derivative as an unbounded closed operator on $L^2{\Lambda }^k$. This operator-theoretic viewpoint is what lets FEEC use Hilbert-complex arguments rather than treating scalar, curl, and divergence problems separately.

Boundary traces depend on degree. In vector proxy notation on a three-dimensional Lipschitz domain, $H(\mathrm{curl})$ has a tangential trace, while $H(\mathrm{div})$ has a normal trace. These are weak traces, not ordinary pointwise boundary values. They determine which interelement continuities are required of conforming finite element spaces.

This distinction explains the classical finite element families. Lagrange elements are built for $H^1$ conformity, Nédélec edge elements for $H(\mathrm{curl})$ conformity, and Raviart-Thomas or Brezzi-Douglas-Marini face elements for $H(\mathrm{div})$ conformity. FEEC recognizes these as degree-specific pieces of a discrete de Rham complex.

The Hilbert complex has cohomology. Closed forms are forms with $d\omega =0$, exact forms are forms of the form $d\eta$, and their quotient records topological information. FEEC stability depends on building discrete spaces that approximate not only functions but also the complex structure and its cohomology.

Synthesis. Sobolev spaces of differential forms are the continuous stage on which FEEC acts. They preserve the geometric derivative relevant to the PDE, carry the correct boundary traces, and assemble the gradient-curl-divergence chain into a single Hilbert complex. Once this continuous complex is fixed, finite element design becomes the search for discrete subspaces that respect it.

Full proof set Master

Proposition 1 (graph norm completeness). The space $H{\Lambda }^k(\Omega )$ is complete in the norm $$ |\omega|{H\Lambda^k}^2=|\omega|{L^2\Lambda^k}^2+|d\omega|_{L^2\Lambda^{k+1}}^2. $$

Proof. Let $({\omega }_j)$ be Cauchy in the graph norm. Then ${\omega }_j$ is Cauchy in $L^2{\Lambda }^k$ and $d{\omega }_j$ is Cauchy in $L^2{\Lambda }^{k+1}$. Let the limits be $\omega$ and $\mu$. The weak derivative identity passes to the limit against compactly supported smooth test forms, so $\mu =d\omega$. Hence $\omega \in H{\Lambda }^k$ and ${\omega }_j\to \omega$ in graph norm. $\square$

Proposition 2 (complex property). The weak exterior derivative satisfies $d\circ d=0$.

Proof. Distributional derivatives commute with testing against smooth compactly supported forms using the sign conventions of exterior calculus. Applying the weak exterior derivative twice transfers the expression to $d^2$ on the test form, which vanishes in the smooth exterior algebra. Therefore $d(d\omega )=0$ distributionally whenever both terms are defined. $\square$

Proposition 3 (three-dimensional proxy). Under the Euclidean vector proxy in dimension three, $H{\Lambda }^0$, $H{\Lambda }^1$, $H{\Lambda }^2$, and $H{\Lambda }^3$ correspond to $H^1$, $H(\mathrm{curl})$, $H(\mathrm{div})$, and $L^2$.

Proof. A zero-form is a scalar function, and $d$ is the gradient. A one-form corresponds to a vector proxy whose exterior derivative corresponds to curl. A two-form corresponds to a flux proxy whose exterior derivative corresponds to divergence. A three-form is a volume density, so no further derivative is present in the three-dimensional de Rham sequence. The $L^2$ and graph-norm conditions give the stated spaces. $\square$

Connections Master

• Sobolev spaces $H^s$ and $W^{k,p}$ 24.01.01. This unit adapts weak-derivative Sobolev ideas to differential forms.

• Differential forms 03.04.02. The degree of a form records the geometric kind of quantity being approximated.

• Exterior derivative 03.04.04. The operator $d$ replaces separate gradient, curl, and divergence notation.

• Mixed FEM for Poisson 24.02.02. Flux variables naturally live in divergence-conforming spaces.

• Bounded cochain projection 24.03.05. FEEC projections must commute with $d$ on the $H{\Lambda }^k$ complex.

Historical & philosophical context Master

Computational electromagnetism made it unavoidable that not all vector fields should be approximated by componentwise $H^1$ elements. Edge elements and face elements encode different continuity requirements, matching curl and divergence rather than full gradients.

Bossavit and Hiptmair helped popularize the differential-forms interpretation of electromagnetic finite elements ^{[Bossavit] [Hiptmair]}. Arnold, Falk, and Winther then placed these insights into the general FEEC framework, where $H{\Lambda }^k$ is the continuous Hilbert-complex level and polynomial differential forms supply the discrete level ^{[Arnold-Falk-Winther]}.

The philosophical point is that PDE discretisation should respect geometry. A method for fluxes should preserve flux continuity; a method for circulation should preserve tangential circulation. Sobolev spaces of forms make these requirements visible before any mesh is chosen.

Bibliography Master

@article{ArnoldFalkWinther2006FEECForms,
  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}
}

@book{Arnold2018FEECForms,
  author = {Arnold, Douglas N.},
  title = {Finite Element Exterior Calculus},
  series = {CBMS-NSF Regional Conference Series in Applied Mathematics},
  volume = {93},
  publisher = {SIAM},
  year = {2018}
}

@article{Hiptmair2002Electromagnetism,
  author = {Hiptmair, Ralf},
  title = {Finite elements in computational electromagnetism},
  journal = {Acta Numerica},
  volume = {11},
  pages = {237--339},
  year = {2002}
}

@book{Bossavit1998Electromagnetism,
  author = {Bossavit, Alain},
  title = {Computational Electromagnetism},
  publisher = {Academic Press},
  year = {1998}
}