an encyclopedia of finite element definitions

How to define a finite element

Written by Matthew W. Scroggs
↓ Cite this page ↓

This page describes how finite elements are defined in the DefElement database.

Cell subentities

Throughout this website, the subentities of a reference element will be referred to as described here.

The vertices, edges, faces, and volumes of a reference element have dimension 0, 1, 2, and 3 (respectively). The (topological) dimension \(d\) is the dimension of the reference element itself. (When using the finite element method, the topological dimensions may differ from the geometric dimension: for example, when meshing a 2D manifold in 3D space, the topological and geometric dimensions are 2 and 3 (respectively).)

The codimension of an entity is given by subtracting the dimension of the entity from the topological dimension of the reference element. Entities of codimension 1, 2, and 3 are called facets, ridges and peaks (respectively). The usual names given to entities of reference elements of topological dimensions 0 to 3 are shown below.

Topological dimensionEntities by dimensionEntities by codimension
012301 (facets)2 (ridges)3 (peaks)
0 (a vertex)point---cell---
1 (an interval)pointsedge--cellpoints--
2 (a polygon)pointsedgesface-celledgespoints-
3 (a polyhedron)pointsedgesfacesvolumecellfacesedgespoints

Ciarlet finite elements

The Ciarlet definition[1] defines a finite element by a triple \((R,\mathcal{V},\mathcal{L})\), where

The basis functions \(\{\phi_0,...,\phi_{n-1}\}\) of the finite element space are defined by

\[l_i(\phi_j) = \begin{cases}1&i=j\\0&i\not=j\end{cases}\]

The correct continuity is enforced by ensuring that the same global degree-of-freedom is associated with corresponding functionals on the shared subentities of neighbouring cells.

Example: Order 1 Lagrange space on a triangle

An order 1 Lagrange space on a triangle is defined by:

The functionals \(l_0\) to \(l_2\) are defined as point evaluations at the three vertices of the triangle:

\[l_0:v\mapsto v(0,0)\] \[l_1:v\mapsto v(1,0)\] \[l_2:v\mapsto v(0,1)\]

It follows from these definitions that the basis functions of the finite element spaces are linear functions that are equal 1 to at one of the vertices, and equal to 0 at the other two. These are:

\[\phi_0(x,y)=1-x-y\] \[\phi_1(x,y)=x\] \[\phi_2(x,y)=y\]

Integral moments

It is common to use integral moment functionals when defining finite elements. Given a mesh entity \(e\) and a finite element space \((e,\mathcal{V}_e,\mathcal{L}_e)\) defined on \(e\), the integral moment functionals \(l_1,...,l_{n_e}\) are defined by \[l_i:v\mapsto \int_e v\phi_i,\] where \(\phi_1,...,\phi_{n_e}\) are the basis functions of the finite element space on \(e\).

For vector-valued spaces, integral moment functional can be defined using a vector-valued space on \(e\) and taking the dot product inside the integral, \[l_i:\boldsymbol{v}\mapsto \int_e \boldsymbol{v}\cdot\boldsymbol{\phi}_i.\] Alternatively, an integral moment can be taken with a scalar-valued space by taking the dot product with a fixed vector \(\boldsymbol{a}\), \[l_i:\boldsymbol{v}\mapsto \int_e \boldsymbol{v}\cdot\boldsymbol{a}\,\phi_i.\] Typically, \(\boldsymbol{a}\) will be tangent to an edge, normal to a facet, or a unit vector in one of the coordinate directions.

Example: Order 1 Nédélec (second kind) space on a triangle

The functionals that define an order 1 Nédélec (second kind) space on a triangle[2] are tangential integral moments with order 1 Lagrange spaces on the edges of the triangle. For example, the two functionals on the edge \(e_0\) of the triangle between \((1,0)\) and \((0,1)\) are \[l_0:\boldsymbol{v}\to\int_{e_0}\boldsymbol{v}\cdot\left(\begin{array}{c}-\frac1{\sqrt2}\\\frac1{\sqrt{2}}\end{array}\right)(1-s_0),\] \[l_1:\boldsymbol{v}\to\int_{e_0}\boldsymbol{v}\cdot\left(\begin{array}{c}-\frac1{\sqrt2}\\\frac1{\sqrt{2}}\end{array}\right)s_0,\] where \(s_0\) varies from 0 (at \((1,0)\)) to 1 (at \((0,1)\)) along \(e_0\).

The basis functions of this space are:

Mapping finite elements

In order to maintain desired properties when mapping finite elements from a reference element to an actual mesh, an appropriate mapping must be defined.[3][4] (For elements with a mixture of functional types, a more complex approach is required.[5])

Let \(F\) be a transformation that maps the reference element to a cell in the mesh, and let \(\boldsymbol{x}\) be a point in the cell.

The Jacobian, \(\mathbf{J}\), of the transformation \(F\) is \(\displaystyle\frac{\mathrm{d}F}{\mathrm{d}x}\) for 1D reference elements, \(\displaystyle\left( \begin{array}{cc} \frac{\partial F_1}{\partial x}&\frac{\partial F_1}{\partial y}\\ \frac{\partial F_2}{\partial x}&\frac{\partial F_2}{\partial y} \end{array} \right)\) for 2D reference elements, or \(\displaystyle\left( \begin{array}{ccc} \frac{\partial F_1}{\partial x}&\frac{\partial F_1}{\partial y}&\frac{\partial F_1}{\partial z}\\ \frac{\partial F_2}{\partial x}&\frac{\partial F_2}{\partial y}&\frac{\partial F_2}{\partial z}\\ \frac{\partial F_3}{\partial x}&\frac{\partial F_3}{\partial y}&\frac{\partial F_3}{\partial z} \end{array} \right)\) for 3D reference elements.

Scalar-valued basis functions

The simplest mapping—used to map scalar basis functions, \(\phi\)—is defined by $$\left(\mathcal{F}\phi\right)(\boldsymbol{x}) :=\phi(F^{-1}(\boldsymbol{x})).$$ The term \(F^{-1}(\boldsymbol{x})\) is the point on the reference element corresponding to the point \(\boldsymbol{x}\), so this mapping maps a value of the function on the reference to the same value at the corresponding point.

Vector-valued basis functions

For vector-valued basis functions, \(\boldsymbol{\phi}\), the covariant Piola (\(\mathcal{F}^\text{curl}\)) and contravariant Piola (\(\mathcal{F}^\text{div}\)) mappings are defined: $$\left(\mathcal{F}^\text{curl}\boldsymbol{\phi}\right)(\boldsymbol{x}) :=\mathbf{J}^{-T}\boldsymbol{\phi}(F^{-1}(\boldsymbol{x}))$$ $$\left(\mathcal{F}^\text{div}\boldsymbol{\phi}\right)(\boldsymbol{x}) :=\frac1{\det \mathbf{J}}\mathbf{J}\boldsymbol{\phi}(F^{-1}(\boldsymbol{x}))$$ The covariant Piola mapping preserves the tangential component of basis functions on edges and facets, and are typically used to map H(curl) elements. The contravariant Piola mapping preserves the normal component of basis functions on facets, and are typically used to map H(div) elements.

Matrix-valued basis functions

For matrix-valued basis functions, \(\mathbf{\Phi}\), the double covariant Piola (\(\mathcal{F}^\text{curl curl}\)) and double contravariant Piola (\(\mathcal{F}^\text{div div}\)) mappings are defined: $$\left(\mathcal{F}^\text{curl curl}\mathbf{\Phi}\right)(\boldsymbol{x}) :=\mathbf{J}^{-T}\mathbf{\Phi}(F^{-1}(\boldsymbol{x}))\mathbf{J}^{-1}$$ $$\left(\mathcal{F}^\text{div div}\mathbf{\Phi}\right)(\boldsymbol{x}) :=\frac1{\left(\det \mathbf{J}\right)^2}\mathbf{J}\mathbf{\Phi}(F^{-1}(\boldsymbol{x}))\mathbf{J}^T$$

Variants of finite elements

For many elements, there are a number of different choices that could be made for the functionals in \(\mathcal{L}\) that define the element that give rise to an element with the same key properties. For example, when defining a Lagrange element on a triangle, there are many possible choices for where exactly to locate the point evaluation functionals. We refer to a pair of elements as variants of each other if:

Commonly used variants of elements are shown on each element's page.


Throughout this website, the notation given here in this section is used.

\(R\)A reference element
\(\mathcal{V}\)A polynomial set
\(\mathcal{L}\)A dual basis
\(l_i\)A functional in the dual basis
\(\phi_i\)A scalar basis function
\(\boldsymbol{\phi}_i\)A vector basis function
\(\mathbf{\Phi}_i\)A matrix basis function
\(F\)A map from a reference to a cell in a mesh
\(\mathcal{F}\)A mapping
\(v_i\)The \(i\)th vertex
\(e_i\)The \(i\)th edge
\(f_i\)The \(i\)th face
\(c_i\)The \(i\)th volume
\(d\)Geometric dimension
\(r\)Exterior derivative order