I have a system of non-linear PDEs that I expect to have shocks as well as the appearance of Gibbs phenomena (spurious oscillations that form near the shock) for 2nd-order methods or higher. I have read that an approach is to use a so-called "high-resolution scheme" that includes a flux limiter function. Essentially, the scheme is employed by using a low resolution (first order) method near discontinuous regions and a higher order method elsewhere.

Before this can be done, the two chosen methods must be written in conservation form, which is stated as:

$$u_{m}^{n+1} = u_{m}^{n} - \frac{\Delta t}{\Delta x}\left(F_{m+\frac{1}{2}}^{n+\frac{1}{2}} - F_{m-\frac{1}{2}}^{n+\frac{1}{2}}\right)$$ The book also mentions that the flux $F$ is a function that depends on $u_{m}$ and some neighbors in space, and that the form of $F$ does not change from one point to the next.

I am a little confused by this specification. How "strict" is this form, exactly? It is unclear to me which parts of this definition are necessary.

Take for example, the leapfrog method.

$$u_{m}^{n+1} = u_{m}^{n-1} - \frac{\Delta t}{\Delta x}\left(f(u_{m+1}^{n})-f(u_{m-1}^{n})\right)$$

At first, this seems like a perfect candidate for a 2nd order method, but it uses $u_{m}^{n-1}$ instead of $u_{m}^{n}$. Does this mean that it is not flux conservative?

Which methods can be put in conservation form? I am planning to use Lax-Friedrichs for the first order, but I am still looking for a suitable higher order method that is easy to extend to non-linear problems.

  • $\begingroup$ If you are asking specifically about finite volume methods, then choosing a time stepping scheme and choosing a spatial discretization scheme are highly coupled tasks, unless the spatial scheme can be written in a semi-discrete form. In this case, you can use whatever time stepping scheme you want as long as you obey the CFL condition $\endgroup$ – whpowell96 May 19 '20 at 2:09
  • $\begingroup$ @whpowell96. So, for example, a method of lines Runge-Kutta type method for time, and a central difference spatial will be flux conservative? $\endgroup$ – user8384493 May 19 '20 at 15:15
  • $\begingroup$ If the underlying differential equation is in the conservation form then you can express it in a finite-volume conservation form. $\endgroup$ – Maxim Umansky May 20 '20 at 3:27

The key feature to a conservative method is simply that the changes due to the fluxes cancel out (i.e., the flux leaving one cell is entering another), so the total mass is constant. Using the form you wrote for a standard conservative method, if we sum $u$ on a grid with $M$ cells, we have:

$$\sum_{m=1}^M u^{n+1}_m = \sum_{m=1}^M u^n_m - \frac{\Delta t}{\Delta x} \left(F^{n+1/2}_{M+1/2} - F^{n+1/2}_{-1/2}\right)$$

If the fluxes at the boundaries are zero (or if the domain is periodic), then the total mass at step $n+1$ is the same as at $n$.

For your leapfrog flux-differencing method, the same analysis shows that (if the flux at the boundaries is zero) $\sum_m u^{n+1}_m = \sum_m u^{n-1}_m$. So all the even steps will have the same total mass and all the odd steps will have the same total mass, but the mass for even and odd steps might be different (depending on whether you use a conservative method for the first step).


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.