Stiffness and absolute stability

Definitions

Absolute stability

Forward Euler vs. Backward Euler

The Curtiss-Hirschfelder example

The test problem [Curtiss & Hirschfelder, 1952]

(1)
\begin{align} \dot y = -50 ( y - \cos(t)) \end{align}

with different initial conditions $y(0) = y_0$ is a classical example of a stiff system. Indeed, it exhibits two widely different time scales, one corresponding to the slow manifold $y^S(t) = \cos(t)$ and the strongly damped mode $e^{-50t}$, as illustrated by the solution curves (light gray) in the figures below.

The absolute stability requirement for the FE method applied to this problem leads to the stepsize restriction $h \le 2/50$. The figure below shows the results of integrating this problem with the FE method and a fixed stepsize $h=2.01/50$.
CurtissHirschfelder_expl_unstable.png

Using a stepsize within the region of absolute stability, leads to a stable method (note that, for accuracy requirements, one may require a smaller stepsize)
CurtissHirschfelder_expl_stable.png

On the other hand, the BE method is stable even with a stepsize $h=0.5$, as shown in the following figure (a CVODE solution with relative and absolute tolerances of $10^{-4}$ is also shown).
CurtissHirschfelder_impl.png

BDF and Adams-Moulton regions of absolute stability

The absolute stability regions for BDF of orders 1 to 6 are shown in the figure below. Note that the stability regions are outside the shaded area for each method.

BDFstab.gif

The absolute stability regions for Adams-Moulton methods of orders 1-9 are shown in the figure below. Note that the first two methods — backward Euler (p=1) and trapezoidal (p=2) — are A-stable.

AMstab.gif
Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-Share Alike 2.5 License.