I'm trying to solve a coupled set of PDEs using Finite element method. But I require the weak form of the equations to do it. Can anyone shed some light onto how to derive the weak form of these equations.

$$\frac{\partial f}{\partial t} + \frac{\partial{fg}}{\partial x} = 0 \\ \frac{\partial{fg}}{\partial t} + \frac{\partial{f^2 g}}{\partial x} = \frac{3}{R} \frac{\partial }{\partial x}(f \frac{\partial g}{\partial x}) + C f $$

It is an open boundary problem related to fluid flow, but initial conditions are defined. feel free to assume convenient initial values for solving.

f and g are respectively the area of cross-section and velocity of a fluid. They are both functions of x and t

  • $\begingroup$ What kind of functions are $f$ and $g$? (i.e. domain and codomain, regularity...) $\endgroup$ – marco trevi Sep 28 '14 at 11:27
  • $\begingroup$ I hope the edit makes the nature of f and g clear enough. $\endgroup$ – newtonian Sep 28 '14 at 11:44
  • $\begingroup$ I would multiply by a suitable test function $\phi$, integrate over the domain and use Gauss-Green theorem to get rid of the higher derivatives...not sure though how to do it on a system of equations. $\endgroup$ – marco trevi Sep 28 '14 at 12:35

Assuming $t>0$ and $x \in [a,b] \subset \mathbb{R},$ we have:

$$ \int^b_a \left( \frac{\partial f}{\partial t} + \frac{\partial \, f g}{\partial x}\right)\omega_1 \mathrm{d}x = 0,$$ where $\omega_1$ is a test function. On the other hand, for the second equation we have to apply 1st Green's identity, which yields:

$$\begin{align} \int^a_b \left( \frac{\partial \, f g}{\partial t} + \frac{\partial \, f^2 g}{\partial x} \right) \omega_2 \,\mathrm{d}x & = \\ \frac{3}{R} \left( \left. f \frac{\partial g}{\partial x} \omega_2\right|_{x=b} - \left. f \frac{\partial g}{\partial x} \omega_2 \right|_{x=a} - \int^a_b f \frac{\partial g}{\partial x} \frac{\partial \omega_2}{\partial x} \, \mathrm{d}x\right) + \int_a^b C f \omega_2 \, \mathrm{d}x &, \end{align}$$ where $\omega_2$ is another test function. The particular choice of what kind of function they would be is up to you. For example, you can set $\omega_i$ as the set of piecewise linear functions (hat functions) on a partition of $[a,b]$.

Now, information of what happens in the boundary is required for the formulation to be complete.

Two considerations:

  • Note that your system of PDEs is not linear, i.e., contains terms as $f^2$ or $f g$. It is then required to use a method for solving non-linear equations, such as Newton-Raphson. You can always sum the two equations above to come up with something like: $$ \mathsf{F}(f,g; \omega_1, \omega_2) = 0, $$ which can be solved with the iteration: $$ - \mathbf{J}_{\mathsf{F}}(\mathbf{f};\mathbf{\omega}) \Delta \mathbf{f} = \mathsf{F}(\mathbf{f};\omega), \quad \mathbf{f} = (f,g), \quad \omega = (\omega_1,\omega_2), $$ $\mathbf{J}$ is the matrix jacobian of $\mathsf{F}$ with respect to $f$ and $g$ and $\Delta \mathbf{f}$ is the difference between iterations.

  • The FEM approach is all about spatial discretization of the equations. Note that you must apply a time-marching method, such as implicit/backward Euler, to march in time.

Hope this helps!


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.