The ModelInterface defines all the different parts needed to compute the weak form of a second order elliptic operator of the form.

\[ \begin{split} -\nabla\cdot (A(x, u, \nabla u)\,\nabla u) + \nabla\cdot(b(x, u)\,u) + c(x, u)\,u &= f(x) + \Psi\quad \text{ in }\Omega,\\ u &= g_D \text{ on }\Gamma_D,\\ (A(x, u, \nabla u)\nabla u)\cdot\nu + \alpha(x, u)\,u &= g_N \text{ on }\Gamma_R,\\ (A(x, u, \nabla u)\nabla u)\cdot\nu &= g_N \text{ on }\Gamma_N,\\ \end{split} \]

In particular, \(\Psi\in H^{-1}\) models a functional, \(f\in L^2\) is the usual "right hand side".

The methods in this class provide all factors of the integrands not involving test functions. The multiplication by the test-functions is added in the class EllipticOperator. The weak formulation then reads:

\[ \int_\Omega (A(x,u,\nabla u)\,\nabla u)\cdot\nabla\phi\,dx + \int_\Omega (\nabla\cdot(b(x, u)\,u) + c(x, u)\,u)\,\phi\,dx + \int_{\Gamma_R} \alpha(x, u)\,u\,\phi\,do - \int_{\Gamma_N} g_N\,\phi\,do - \int_\Omega f(x)\,\phi\,dx - \langle\Psi,\,\phi\rangle = 0. \]

As the ModelInterface is potentially non-linear one has the option to leave the ModelInterface::BulkForcesFunctionType and ModelInterface::NeumannBoundaryFunctionType at zero and instead stuff the "right-hand-side" functions into the ModelInterface::robinFlux() and ModelInterface::source() methods. In principle even the functional \(\Psi\) could be expressed by a \(L^2\) scalar-product (don't shout at me: as the FEM-space is finite dimensional this is of course possible and one way to implement such a functional).

Nevertheless the different forces and other right-hand-side components are there and will be used by EllipticFemScheme and ParabolicFemScheme. Generally, the implementation checks for zero-objects (see ZeroExpression, ZeroGridFunction, ZeroFunctional, ZeroModel); the decision about which component has to be taken into account is taken at compile-time.