The Computable Differential Equation Lecture ... - Bruce E. Shapiro
The Computable Differential Equation Lecture ... - Bruce E. Shapiro
The Computable Differential Equation Lecture ... - Bruce E. Shapiro
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
184 CHAPTER 9. DIFFERENTIAL ALGEBRAIC EQUATIONS<br />
<strong>The</strong>re is no solution for (x n , y n ) in terms of the variables at earlier times. This can<br />
be seen be rewriting the system as<br />
x n − t n y n = h(x n−1 − t n y n−1 ) (9.171)<br />
x n − t n y n = f(t n ) (9.172)<br />
or in matrix form<br />
( ) ( )<br />
1 −tn xn<br />
=<br />
1 −t n y n<br />
( )<br />
h(xn−1 − t n y n−1 )<br />
f(t n )<br />
<strong>The</strong> matrix of coefficients is singular hence there is no solution.<br />
(9.173)<br />
9.6 BDF Methods for DAEs<br />
While the numerical solution of initial value problems for ordinary differential equations<br />
has a long history going back over one hundred years, the numerical solution of<br />
DAEs is mostly still an open research issue (see [5] for the more common methods)<br />
with significant interest only dating back to the 1970’s. Since many of the scientific<br />
problems of interest turn out to be stiff, early research focused first on BDF methods<br />
and later on some RK methods that are known to be better for stiff problems.<br />
<strong>The</strong> first DAE algorithms published were BDF methods. 5 <strong>The</strong> solution of linear<br />
and index-1 systems are now understood but methods for higher index systems still<br />
form the basis of much current methods.<br />
In general variable step sizes can cause problems with DAEs. For example, it is<br />
not possible to solve an index-3 system using implicit Euler with variable step-size<br />
because of the following theorem because it will have O(1) errors.<br />
<strong>The</strong>orem 9.11. <strong>The</strong> global error for a k-step BDF applied to an index-ν system is<br />
O(h n ) where n = min(k, k − ν + 2)<br />
Furthermore, all multistep (as well as Runge-Kutta) methods are unstable for<br />
higher index (≥ 3) systems, and in some cases fail for specific index-1 and index-2<br />
systems. Some convergence results for semi-implicit and fully-implicit index-1 systems,<br />
semi-explicit index-2 systems, and index-3 systems in Hessenberg form have<br />
been worked out.<br />
We can write the general k-step BDF method for a scalar ODE (see 6.105) as<br />
k∑<br />
a j y k−j = hb 0 Dy n (9.174)<br />
j=0<br />
where D is the differential operator Dy = y ′ . <strong>The</strong>n the general BDF method<br />
becomes<br />
0 = F (t n , y n , Dy n ) (9.175)<br />
5 C.W.Gear (1971) “Simultaneous Numerical Solution of <strong>Differential</strong>-Algebraic <strong>Equation</strong>s,” IEEE<br />
Transactions on Circuit <strong>The</strong>ory, 18:89-95<br />
Math 582B, Spring 2007<br />
California State University Northridge<br />
c○2007, B.E.<strong>Shapiro</strong><br />
Last revised: May 23, 2007