Dual spaces¶
Mathematical background¶
If \(V\) is a vector space, then define the (anti-)dual space \(V^*\) to be the space of bounded conjugate linear functionals \(V \to K\). Therefore, the dual space is the space containing functions from \(V\) to \(K\), where \(K\) can either be \(\mathbb{R}\) or \(\mathbb{C}\).
If \(\{\phi_i\}\) is a basis for a finite vector space \(V\) then there exists \(\{\phi_i^*\}\) a basis for \(V^*\) such that:
The basis \(\{\phi_i^*\}\) is termed the dual basis. Where it is necessary to make the distinction, we will refer to the space to which a dual space is dual as the primal space and its basis as the primal basis.
Since UFL function spaces are finite-dimensional Hilbert spaces which result from the discretisation of infinite-dimensional Hilbert spaces, all of the function spaces with which we are concerned are reflexive, ie \((V^*)^*\) is isomorphic to V under the canonical map. That is, we can identify \((V^*)^*\) and \(V\):
A form defined over an unknown \(a\) in the primal space \(V\) is a known object in the dual space. For example:
with basis coefficients \(h_i = \displaystyle\int_\Omega \phi_i \text{ d}x\).
Dual objects in UFL¶
For an arbitrary FunctionSpace, V, the corresponding dual space \(V^*\) can be obtained by calling the dual() method:
from firedrake import *
mesh = UnitSquareMesh(10, 10)
V = FunctionSpace(mesh, "CG", 1)
V_star = V.dual()
A Coefficient defines a known function c in V. A Function is a subclass of Coefficient.
Consequently,
c = Function(V)
f_0 = c * dx
is a symbolic expression for the integral of c over the domain and represents a scalar value. f_0 is a Python object of type Form, once assembled, it is a scalar object.
Conversely, Argument defines a placeholder symbol a for an unknown function in V. TestFunction and TrialFunction are syntactic sugar for Argument(V, 0) and Argument(V, 1) respectively.
a = TrialFunction(V)
f_1 = a * dx
represents the integration of the unknown function a over the domain. It’s therefore a linear 1-form, or a function in the dual space \(V^* = V \rightarrow K\). f_1 is also a Python object of type Form. When assembled, it is an object of type Cofunction:
cf = assemble(f_1) # type Cofunction
cf is a known object in the dual space, and the dual equivalent of Coefficient. The more consistent name Cocoefficient was rejected as confusing and risible. Cofunction objects can be combined with symbolic Form objects:
v = TestFunction(V)
a = v * dx
b = assemble(a)
res = a + b
c = assemble(res)
Furthermore, we will want to express unknown objects in the dual space. For example, in order to represent interpolation from a space \(U\) to a space \(V\), it is convienent to reframe this as a problem involving the dual space:
Using the reflexivity of the function space \(U\). This form therefore has two arguments, one in the primal space \(V\) and one in the dual space \(U^*\). Therefore, we need to represent arguments in the dual space - we will call these coarguments. The details of interpolation will be discussed in its own section.
A Coargument can be constructed by either calling Argument on a dual space object or calling Coargument on a dual space.
v = Argument(V, 1) # type Argument
u = Argument(V.dual(), 2) # type Coargument
w = Coargument(V.dual(), 3) # type Coargument
There is a further dual-related type avalilable in UFL. In Cofunction, we have represented an assembled 1-form. However, commonly we also assemble 2-forms. Matrix allows an analogous use, and assembled 2-forms can be naturally combined with 2-forms that have not yet been assembled:
mesh = UnitSquareMesh(10,10)
V = FunctionSpace(mesh, "Lagrange", 1)
u = TrialFunction(V)
v = TestFunction(V)
a = (u*v - inner(grad(u),grad(v)) ) * dx
M = assemble(a) # type Matrix
res = assemble(M + a)
Operations supported symbolically, such as the adjoint and action, are also supported on the dual space equivalent.
mesh = UnitSquareMesh(10,10)
V = FunctionSpace(mesh, "Lagrange", 1)
U = FunctionSpace(mesh, "Lagrange", 1)
u = TrialFunction(U)
v = TestFunction(V)
a = u * v * dx
a = assemble(a) # type Matrix
adj = adjoint(a)
b = Matrix(V, U.dual())
u = Coefficient(U)
u_a = Argument(U, 0)
u_form = u_a * dx
primal_action = action(a, u)
dual_action = action(b, u_form)
In summary, this table describes the dual types corresponding to primal finite element spaces, and to known and unknown functions in those spaces:
Primal quantity |
Dual quantity |
|---|---|