1: \begin{abstract}

2: %% Text of abstract.  (Recent JCP letters include abstracts.)

3: Some properties of a Local discontinuous Galerkin (LDG) algorithm are

4: demonstrated for the problem of evaluting a second derivative $g =

5: f_{xx}$ for a given $f$.  (This is a somewhat unusual problem, but it is

6: useful for understanding the initial transient response of an algorithm

7: for diffusion equations.)  LDG uses an auxiliary variable to break this

8: up into two first order equations and then applies techniques by analogy

9: to DG algorithms for advection algorithms.  This introduces an asymmetry

10: into the solution that depends on the choice of upwind directions for

11: these two first order equations.  When using piecewise linear basis

12: functions, this LDG solution $g_h$ is shown not to converge in an

13: $L_2$ norm because the slopes in each cell diverge.  However, when LDG

14: is used in a time-dependent diffusion problem, this error in the second

15: derivative term is transient and rapidly decays away, so that the

16: overall error is bounded.  I.e., the LDG approximation $f_h(x,t)$ for a

17: diffusion equation $\partial f / \partial t = f_{xx}$ converges to the

18: proper solution (as has been shown before), even though the initial rate

19: of change $\partial f_h / \partial t$ does not converge.  We also show

20: results from the Recovery discontinuous Galerkin (RDG) approach, which

21: gives symmetric solutions that can have higher rates of convergence for

22: a stencil that couples the same number of cells.

23: \end{abstract}

24: