Stiff Differential Equations
Stiff Differential Equations
MAT 461/561
Spring Semester 2009-10
Lecture 9 Notes
The exact solution is () = 100 , which rapidly decays to zero as increases. If we solve this
problem using Eulers method, with step size = 0.1, then we have
+1 = 100 = 9 ,
which yields the exponentially growing solution = (9) . On the other hand, if we choose
= 103 , we obtain the computed solution = (0.9) , which is much more accurate, and
correctly captures the qualitative behavior of the exact solution, in that it rapidly decays to zero.
The ODE in the preceding example is a special case of the test equation
= , (0) = 1, Re < 0.
1
The exact solution to this problem is () = . However, as increases in magnitude, the problem
becomes increasingly sti. By applying a numerical method to this problem, we can determine how
small must be, for a given value of , in order to obtain a qualitatively accurate solution.
When applying a one-step method to the test equation, the computed solution has the form
+1 = () ,
(, ) = (0 0 ) + (1 1 )1 + + (1 1 ) + ( ).
2
The exponents range from 0 to the multiplicity of minus one, so if the roots are all
distinct, all are equal to zero. In order to ensure that the numerical solution decays to zero as
increases, we must have < 1 for = 1, 2, . . . , . Otherwise, the solution will either converge
to a nonzero value, or grow in magnitude.
Example Consider the 3-step Adams-Bashforth method
+1 = + [23 161 + 52 ].
12
Applying this method to the test equation yields the characteristic polynomial
( )
23 4 5
(, ) = + 1 2 + .
3
12 3 12
Let = 100. If we choose = 0.1, so that = 10, then (, ) has a root approximately
equal to 18.884, so is too large for this method. On the other hand, if we choose = 0.005,
so that = 1/2, then the largest root of (, ) is approximately 0.924, so is suciently
small to produce a qualitatively correct solution.
Next, we consider the 2-step Adams-Moulton method
+1 = + [5+1 + 8 1 ].
12
In this case, we have
( ) ( )
5 2 2 1
(, ) = 1 + 1 + .
12 3 12
3
An example of an A-stable one-step method is the Backward Euler method
+1 = + (+1 , +1 ),