Sie sind auf Seite 1von 3

Picard iteration

The Theorem on Existence and Uniqueness states that if both f (t, y) and f /y are continuous in some region around the point (t0 , y0 ) then there is a unique solution to the initial value problem y = f (t, y) y(t0 ) = y0 (1)

valid in some interval around t0 . In other words, if the slope eld is suciently smooth at each point, then there is a unique integral curve passing thru any given point. How do we prove such a theorem? There two methods, but both use a sequence of approximate solutions, and prove that these approximations converge, at least in a small interval around t0 . One method is due to Euler, and is quite simple to use in practice: one simply connects the dots in the slope eld. The disadvantage to this method is that it only gives an approximation at the dots. In other words, Eulers method only approximates the values of the solution at a nite list of points. It does not give us a formula for an approximate function at every point. However, Eulers method has the advantage that its accuracy can be improved with only minor modications. For most applications some version of an improved Euler method is ideal. A second method is due to Picard. The Picard method gives a sequence of functions which converge to the solution. Picards method is far less ecient computationally than Eulerian methods, but it introduces an important technique that will be useful for the error analysis of Eulerian methods. An approximation method is useless without an estimate of the error. Picards method begins by transforming the pair of conditions that are the initial value problem into a single integral equation. Estimates with integrals are fairly straightforward. Specically, the initial value problem (1) is equivalent to the single equation
t

y(t) = y0 +
t0

f (s, y(s)) ds.

(2)

Indeed, by the Fundamental Theorem of Calculus, if y satises equation (2) then y (t) = f (t, y(t)); and if we take t = t0 in equation (2) then we nd that y(t0 ) = y0 + 0. Picards method starts with the denition of what it means to be a solution: if you guess that a function (t) is a solution, then you can check your guess by substituting it into the right-hand side of equation (2) and comparing it to the left-hand side, which is simply (t) itself. The new idea is that the process of checking each guess produces a new guess which, even if it is not the correct solution, is a better approximation than the one you started with. In this way we obtain an iterative solution, with each new approximation computed from the previous one by the right-hand side of equation (2). This should be reminiscent of Newtons method. In fact, the proof that Picards method produces a convergent sequence is similar to the proof for Newtons method. Lets look at an example: y = ty 2 , y(0) = 1. This becomes the integral equation
t

y(t) = 1 +
0

s y(s)2 ds.

We turn this into an iterative scheme by substituting a guess i (t) for the value of y on the right, and regarding the result as the denition of i+1 (t), which substitutes for the value of y on the left.
t

i+1 (t) = 1 +
0

s i (s)2 ds.

Thus, from 0 we obtain 1 ; from which, in turn, we obtain 2 ; from which we obtain 3 ; and so forth. Where do we start? What should be 0 (t)? Here we take a clue from the initial value: if we set 0 (t) = y0 then at least we get the second condition from (1) correct. So, here are the rst four Picard iterates in our example:

0 (t) = 1 1 s(1)2 ds = 1 + t2 . 2 0 t 1 1 2 (t) = 1 + s (1 + s2 )2 ds = 1 + t2 2 2 0 t 1 1 1 3 (t) = 1 + s (1 + s2 s4 + s6 )2 ds 2 4 24 0 1 2 1 4 1 6 1 1 = 1 + t t + t t8 + t10 2 4 8 24 96 1 (t) = 1 +


t

1 4 1 t + t6 . 4 24

1 12 1 14 t + t . 576 8064

Here is a plot of the rst six Picard iterates, together with a plot of the exact solution (the dashed line).
1.5

0.5

0.5

1.5

1.5

0.5

0.5

1.5

Note that the approximations seem to diverge for |t| somewhere between 1 and 1.5. The exact solution (which can be found by separating variables) is 1/(1 + t2 /2), which is valid for all t. However, in this example the Picard solutions appear to provide what is tantamount to a power series representation, which we know will converge only when t2 /2 < 1 that is, when |t| < 2. This corroborates the observations from the graph.

Further reading
Read section 2.8 for more on Picards Method and the Existence and Uniqueness Theorem. In particular, study example 1 on pages 107109. What is the dierence between the two graphs on pages 108 and 109?

Reading quiz
1. State the Theorem on Existence and Uniqueness for rst-order equations. 2. What is the interval of validity? 3. In what sense is Picards method iterative? 4. What integral equation is equivalent to a rst-order initial value problem. 5. What is the formula for the Picard iterates? 2

Exercises
For each of the following initial value problems, nd the equivalent integral equation, then compute the rst four or ve Picard iterates. Graph all ve of these approximations together on the same coordinate axes. (You may use computer to do the graphing.) On what interval do the iterates seem to be converging? Finally, nd the exact solution, and determine its interval of validity. How does the interval of validity compare to the interval of convergence? 1. y = t2 y t, y(0) = 0. 2. y = 1 y 3 , y(1) = 3. 3. y = (2t + 1)y 2 , y(0) = 1.

Das könnte Ihnen auch gefallen