I'm a bit confused on how to Runge-Kutta time stepping methods for finite volume problems. With finite volume PDEs, we have essentially have discretized equations of the form $$\frac{\partial\bar{u}}{\partial t} = F(u(t))$$ where $\bar{u}$ is the average over a cell, and $F$ is usually some numerical flux function. However, for problems that implement RK4 they are usually stated as an initial value problem of the form: $$\frac{\partial y}{\partial t} = F(y(t))$$

I am trying to solve the viscid Burger's equation: $$\frac{\partial u}{\partial t} = \frac{\partial}{\partial x}(\frac{\partial u}{\partial x} + u^2)$$

via finite volume methods and am wondering if I can advance in time using Runge-Kutta methods.


Yes, you can! The operation which consists in approximating the field values $u(t)$ by knowing the cell averages $\bar{u}(t)$ is called reconstruction, and is classically based on interpolation. Thus, a method of lines \begin{equation} \frac{\partial \bar{u}}{\partial t} = F(u(t)) \end{equation} rewrites as \begin{equation} \frac{\partial \bar{u}}{\partial t} = L(\bar{u}(t)) \, , \end{equation} where the operator $L$ combines $F$ and a reconstruction procedure.

For example, ENO and WENO finite-volume methods combine a Runge-Kutta method and a reconstruction procedure, in order to achieve high-order integration (see e.g. Shu, 2009).

  • $\begingroup$ Thanks for the response! In fact, I've read this paper and have employed ENO reconstruction methods to reconstruct $u$ from $\bar{u}$, but got confused because her "spatial discretization operator" $L$ seems to be a function of $\bar{u}$ (pg. 95 of the pdf in the link) while mine is a function of $u$, which I thought would invalidate the use of Runge-Kutta. $\endgroup$ – hijasonno Feb 22 '17 at 18:38

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.