This preview shows page 1. Sign up to view the full content.
Unformatted text preview: 1 2 3 : : : N ; 1: Had we solved the wave equation (9.1.2) instead of the heat equation (9.1.1) using a
piecewiselinear nite element basis, we would have found the discrete system Mc + Kc = 0 (9.2.5) with p in (9.2.4c) replaced by c2.
The resulting initial value problems (IVPs) for the ordinary di erential equations
(ODEs) (9.2.4a) or (9.2.5) typically have to be integrated numerically. There are several
excellent software packages for solving IVPs for ODEs. When such ODE software is used
with a nite element or nite di erence spatial discretization, the resulting procedure is
called the method of lines. Thus, the nodes of the nite elements appear to be \lines"
in the time direction and, as shown in Figure 9.2.2 for a onedimensional problem, the
temporal integration proceeds along these lines. 8 Parabolic Problems
t x
0 = x0 x1 xj xN1 xN = 1 Figure 9.2.2: \Lines" for a method of lines integration of a onedimensional problem.
Using the ODE software, solutions are calculated in a series of time steps (0 t1],
(t1 t2], : : : . Methods fall into two types. Those that only require knowledge of the solution at time tn in order to obtain a solution at time tn+1 are called onestep methods.
Correspondingly, methods that require information about the solution at tn and several
times prior to tn are called multistep methods. Excellent texts on the subject are available
2, 6, 7, 8]. Onestep methods are RungeKutta methods while the common multistep
methods are Adams or backward di erence methods. Software based on these methods
automatically adjusts the time steps and may also automatically vary the order of accuracy of a class of methods in order to satisfy a prescribed local error tolerance, minimize
computational cost, and maintain numerical e ciency.
The choice of a onestep or multistep method will depend on several factors. Generally, RungeKutta methods are preferred when time integration is simple relative to the
spatial solution. Multistep methods become more e cient for complex nonlinear problems. Implicit RungeKutta methods may be e cient for problems with highfrequency
oscillations. The ODEs that arise from the nite element discretization of parabolic
problems are \sti " 2, 8] so backward di erence methods are the preferred multistep
methods.
Most ODE software 2, 7, 8] addresses rstorder IVPs of the explicit form
_
y(t) = f (t y(t)) y(0) = y0: (9.2.6) Secondorder systems such as (9.2.5) would have to be written as a rstorder system by,
e.g., letting
_
d=c 9.2. SemiDiscrete Galerkin Problems
and, hence, obtaining 9 _
c=d:
_
;Kc
Md Unfortunately, systems having the form of (9.2.4a) or the one above are implicit and
would require inverting or lumping M in order to put them into the standard explicit
form (9.2.6). Inverting M is not terribly di cult when M is constant or independent
of t however, it would be ine cient for nonlinear problems and impossible when M is
singular. The latter case can occur when, e.g., a heat conduction and a potential problem
are solved simultaneously.
Codes for di erentialalgebraic equations (DAEs) dir...
View Full
Document
 Spring '14
 JosephE.Flaherty
 The Land

Click to edit the document details