5
$\begingroup$

I'm trying to apply the finite-volume method (FVM), with which I'm not so familiar, so a simple 1D PDE equation. I already asked this question in the math stackexchange, but was told that it could be a better fit here.

The equation I want to solve is, to simplify, $$\frac{\partial U}{\partial t} + A\left(x\right) \frac{\partial}{\partial x}\left(\frac{F(U)}{A(x)}\right) + \frac{\partial G(U)}{\partial x} = 0.$$

where $U$ is my unknown, $F$ and $G$ some flux functions computed from $U$ and $A$ a function of $x$, that is strictly positive and very smooth. Most of the time, I can actually assume $A$=constant, and therefore the equation becomes $$\frac{\partial U}{\partial t} + \frac{\partial(G(U)+F(U))}{\partial x} = 0,$$ which is straightforward to solve using FVM.

However, I'm not sure how to handle the case where $A$ is not constant. My idea was, when applying the integration step of the finite volume method, to write $$\int_{\text{cell}} A\left(x\right) \frac{\partial}{\partial x}\left(\frac{F(U)}{A(x)}\right) \mathrm{d}\Omega \approx \bar{A} \int_{\text{cell}} \frac{\partial}{\partial x}\left(\frac{F(U)}{A(x)}\right) \mathrm{d}\Omega$$ where $\bar{A}$ is the mean value of $A$ over the considered cell. Once fully discretized, this term would therefore end up looking like

$$\bar{A}_{i} \left( \frac{1}{\Delta x} \left( \frac{F_{i+1/2}}{A_{i+1/2}} - \frac{F_{i-1/2}}{A_{i-1/2}} \right) \right).$$

Does it make sense, or is it completely wrong? Is there a better way to proceed?

I tried changing the coordinate, but I'm not familiar enough with that, and of course I'd like to keep the $\dfrac{\partial G(U)}{\partial x}$ term as it is.

Thanks a lot for any inputs and advices on how to deal with this kind of problem.

$\endgroup$
  • $\begingroup$ How well do you know A, e.g. is it algebraic? $\endgroup$ – origimbo Feb 9 '18 at 14:46
  • $\begingroup$ In a general way, I know the values of $A$ on the cell-centers. For the cell faces, I planned to used $$ A_{i+1/2} = \frac{1}{2} \left( A_i + A_{i+1} \right)$$ $\endgroup$ – merrihurruz Feb 9 '18 at 14:48
5
$\begingroup$

Your proposed discretisation appears to be consistent, but wouldn't normally be interpreted as a finite volume discretisation. Indeed, it looks a lot more like a finite difference method with a slightly unusual stencil. Further, this looks like a nonlinear hyperbolic, rather than elliptic PDE, which fits with your desire to use FVM.

Generically, you can build yourself a finite volume model by integrating the equation over a fixed cell. In this case we can use integration by parts to write that in $n$ dimensions, and for a cell $C$ bounded by a contour, $\delta C$.

$$\frac{\partial}{\partial t} \int_C U d\mathbf{x} = -\int_{\delta C} \mathbf{F}(U)+\mathbf{G}(U)\cdot\mathbf{n} dS +\int_C \frac{\mathbf{F}(U)}{A}\cdot\nabla A d\mathbf{x}. $$

Comparing this to the constant $A$ case, you have an additional source/sink for your tracer, which in 1D looks like $$F(U)\frac{1}{A}\frac{\partial A}{\partial x},$$ which you can discrete using normal cell centred finite differences for the derivative. This actually impacts more on your time stepping scheme, since for sinks ($\frac{\partial F}{\partial U}\frac{\partial A}{\partial x}$ negative) you may want to consider an implicit method in the case that $U$ needs to stay bounded.

$\endgroup$

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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