next up previous
Next: Pade approximation to exp(x) Up: Brief notes on CFD, Previous: Kutta condition for lifting

Convergence to entropy solution for scalar conservation law

Initial value problem:


\begin{displaymath}
\left. \begin{array}{r}
u_t + f(u)_x = 0 \\
u(x,0) = u_o(x)
\end{array} \right\}
\end{displaymath} (1)

with $f$ being strictly convex and $u_o \in BV({\mathbb{R}})$.

Viscous form of update scheme:


\begin{displaymath}
u^{n+1}_i = u^n_i - \frac{\lambda}{2}( f^n_{i+1} - f^n_{i-1}...
...i + 1/2} \Delta u^n_{i+1/2} - Q^n_{i-1/2} \Delta u^n_{i-1/2} )
\end{displaymath} (2)

with $\lambda = \Delta t /\Delta x$ being fixed. The numerical flux function is given by


\begin{displaymath}
f^n_{i+1/2} = \frac{1}{2}(f^n_i + f^n_{i+1}) - \frac{1}{2\lambda} Q^n_{i+1/2}
\Delta u^n_{i+1/2}
\end{displaymath} (3)

Some standard schemes

  1. Lax-Friedrichs


    \begin{displaymath}
Q^{LF}=1
\end{displaymath}

  2. Engquist-Osher


    \begin{displaymath}
Q^{EO}_{i+1/2} = \lambda \int^1_0 \vert f'(u_i + \theta (u_{i+1} - u_i)) \vert \textrm{d}
\theta
\end{displaymath}

  3. Godunov


    \begin{displaymath}
Q^{G}_{i+1/2} = \lambda \max_{(s-u_i)(u_{i+1}-s) \ge 0} \frac{ f(u_i) +
f(u_{i+1}) - sf(s) }{u_{i+1} - u_i}
\end{displaymath}

  4. Murman-Roe


    \begin{displaymath}
Q^{MR}_{i+1/2} = \lambda\left\vert \int^1_0 f'(u_i + \theta ...
... f'(u_i) \vert & \mbox{ if } u_i = u_{i+1}
\end{array} \right.
\end{displaymath}


\begin{displaymath}
0 \le Q^{MR} \le Q^G \le Q^{EO} \le Q^{LF} \le 1
\end{displaymath}

Tadmor's convergence result:


\begin{displaymath}
\left. \begin{array}{r}
Q^G_{i+1/2} \le Q_{i+1/2} \le 1\\
\lambda \sup\vert f'(s)\vert \le \frac{1}{2}
\end{array} \right\}
\end{displaymath} (4)

The restriction of CFL number was increased to 1 by Aiso.

Theorem (Aiso [1]): Fix an arbitrary positive number $\epsilon \le 1$. Suppose that the flux function $f$ is strictly convex. If the numerical viscosity satisfies


\begin{displaymath}
\max\{ Q^{MR}_{1+1/2}, \epsilon (u_{i+1} - u_i) \} \le Q_{i+1/2} \le 1
\end{displaymath} (5)

then difference approximation (2) converges to the entropy solution under CFL $\le$ 1.

References

  1. Aiso H, Admissibility of difference approximation for scalar conservation laws, Hiroshima Math Journal, 23(1):15-61, 1993.

  2. Aiso H, A general class of difference approximation for scalar conservation laws converging to the entropy solution and including high resolution ones, 15'th ICNMFD, Monterey, 1996.


next up previous
Next: Pade approximation to exp(x) Up: Brief notes on CFD, Previous: Kutta condition for lifting
Praveen. C, last updated on 18-February-2005