Suppose I have a nonlinear system of PDEs. I am actually interested in Navier-Stokes, but, for the sake of simplicity and example, suppose I had

$$ \frac{\partial f}{\partial t} - f \frac{\partial g}{\partial x} = 0 $$

$$ \frac{\partial g}{\partial t} - g \frac{\partial f}{\partial x} = 0 $$

I assume that the domain $\Omega$ is rectangular, say, and that some kind of appropriate boundary conditions have been specified on the boundary $\partial \Omega$.

How could I go about using a Galerkin method to solve for both $f$ and $g$? I understand the derivation of the method for one function fairly well, but I do not see how to generalize it to multiple unknown functions.


So first, represent your system as the following in your case:

\begin{align} \frac{\partial q_k}{\partial t} - q_k \nabla \cdot F_k(\boldsymbol{q}) &= 0 & \forall k \in \lbrace 1, 2 \rbrace \end{align}

where $\boldsymbol{q} = [q_1, q_2]^T = [f, g]^T$, $F_1(\boldsymbol{q}) = q_2 \hat{e}_1$, and $F_2(\boldsymbol{q}) = q_1 \hat{e}_1$. Then you just perform the normal Galerkin for each equation using a set of weight functions for that equation. Let us assume we have some space of functions defined for each function/equation, call them $\mathcal{V}_k$. Then we can proceed with the Galerkin projection like so:

\begin{align} \int_{\Omega} w_j \left(\frac{\partial q_k}{\partial t} - q_k \nabla \cdot F_k(\boldsymbol{q})\right) d\Omega &= 0 &\forall w_j \in \mathcal{V}_k \\ % \int_{\Omega} w_j \frac{\partial q_k}{\partial t} d\Omega &= \int_{\Omega} w_j q_k \nabla \cdot F_k(\boldsymbol{q}) d\Omega &\forall w_j \in \mathcal{V}_k \\ % &= \int_{\Omega} \nabla \cdot \left( w_j q_k F_k(\boldsymbol{q}) \right) - F_k(\boldsymbol{q}) \cdot \nabla \left( w_j q_k \right)d\Omega &\forall w_j \in \mathcal{V}_k \\ % &= \int_{\Gamma} w_j q_k F_k(\boldsymbol{q}) \cdot \hat{n}\; d\Gamma - \int_{\Omega} F_k(\boldsymbol{q}) \cdot \nabla \left( w_j q_k \right)d\Omega &\forall w_j \in \mathcal{V}_k \\ &= G_k(\boldsymbol{q}, w_j) &\forall w_j \in \mathcal{V}_k \end{align}

From here you can discretize things further with your approximate representation for the functions $q_k(x)$ using your space of functions $\mathcal{V}_k$ and come up with the equations you need in terms of any coefficients you are solving for.

  • $\begingroup$ Quick clarification -- should one of those unit vectors be $\hat{e}_2$? Or are both $\hat{e}_1$? $\endgroup$
    – emprice
    Nov 3 '17 at 0:36
  • $\begingroup$ @nosuchthingasstars nope. Those unit vectors are tied to the fact your sample PDEs only have spatial derivatives in the variable $x$, which is in the first unit direction. $\endgroup$
    – spektr
    Nov 3 '17 at 2:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.