21 Linear Differential Equation Systems

Today: Just when you thought it was safe! back to linear differential equations. The solution by eigenvectors.

At this point you have seen those essential parts of linear algebra which are needed for solving linear systems of differential equations of the form x0 = Ax Here, A may be an n by n matrix and x an n by 1 column vector function of t. As in any differential equation system x0 = f(x), the right side represents a vector field which exists in space, and the mission of any curve x(t) which intends to solve the system is to be sure that its tangent velocity vector x0(t) agrees at all times with the given vector field. In this lecture we deal only with vector fields of the form f(x) = Ax, and these may be solved rather completely using our new eigenvector knowledge.
Figure 21.1 A vector field in 3 dimensions. We think of it as unchanging unless we have a need for explicit time dependence in the field. But there are also two curves shown, which you might think of dynamically. They are doing their very best to follow the arrows, not only in direction but also to have the correct speed at every point they go through. If they succeed in this, they are solutions to an ODE. The field shown is more complicated than a linear Ax field.

For example, we know that sometimes you find an eigenvector v for which Av is a positive multiple of v. In such a case, the vector field points away 83 from the origin and so any solution passing through point v must be headed away from the origin. Similarly negative eigenvalues give rise to solutions headed for the origin. The effects of complex eigenvalues are a little harder to guess. So lets proceed with the analysis in the following way. Suppose by analogy with many things which have gone before, that we are looking for a solution of the form x(t) = cert, where r is a number as usual, but now c must be a vector for the assumption to make any sense. Then substitute into the differential equation to get x0 = rcert = Ax = Acert These will agree if Ac = rc Now stop and take a deep breath, and think about what you learned before about linear algebra. This equation is the same except for notation as Av = v isnt it? We have reached the following fact: If Av = v, then one solution to x0 = Ax is vet.
Example: From the example on page 76, _ 12 0 3 __ 1 0 _ =1 _ 1 0 _ So one solution to x0 = x + 2 y0 = 3y is _ 1 0

_ e1t i.e., x = et, y = 0. Check these in the differential equation to be sure.

Note that we certainly dont stop with one solution. For each eigenvector you get a solution.
Example: Also for the same example we found that _ 1 2 _ is an eigenvector with eigenvalue 3. Consequently we have another solution _ 1 2 _ e3t, or x = e3t, y = 2e3t. Check these too.

In general we hope for a system x0 = Ax where A is n by n, that there will be n such solutions. Usually there are, but not always. More on this 84 in a moment. The most important thing now is to point out how you can combine these solutions which we have found. Our equation x0 = Ax is linear in the strictest sense of the word: linear combinations of solutions are solutions, i.e., if x0 1 = Ax1 and x0 2 = Ax2, then for any constants c1 and c2 it is true that (c1x1 + c2x2)0 = A(c1x1 + c2x2). This is because differentiation and multiplication by A are both linear operations. You should write out the one-line proof of this yourself if you are not sure about it.
Example: x(t) = c1 _ 1 0 _ et + c2 _ 1 2 _ e3t is a solution to x0 = _ 12 0 3 _ x for any choice of constants c1 and c2. In order to meet a given initial condition, we have to solve for the cj . For example if we want to have x(0) = _ 2 .5 2 _ then it is necessary to solve _ 2 .5 2 _ = c1 _ 1 0 _ +c2 _

1 2 _ . The answer is c2 = 1, c1 = 1.5 and we get the solution x(t) = 1 _ 1 0 _ et + 1.5 _ 1 2 _ e3t.

Back to the question of finding n solutions for the case of n by n matrices A. Certainly if we find n eigenvectors vj we can form solutions like x(t) = c1v1e1t + + cnvnent The question is whether these are all of the solutions to the differential equation. Equivalently, is it possible to meet all the possible initial conditions using x(0) = c1v1 + + cnvn? Note that this is a system of linear algebraic equations in the unknown cj , like we saw in the previous sample. What is needed here is that the vj should be linearly independent. This concept is slightly beyond the level of these notes, but basically it means that the vectors really point in n different directions, or that none of them may be written as a linear combination of the other ones. You will study this concept in a linear algebra course; were only doing an introduction to it here.
Example: Let A = 2 4 300 000 050 3 5. This was problem 10 on page 79. We have Av1 = 3v1 and Av2 = 0 where v1 = 2 4 1 0 0 3 5, v2 = 2 4 0 0 1 3 5, and there are no other eigenvectors except for multiples of these. So some solutions to x0 = Ax are

85
x = c1v1e3t + c2v2. This is one of the exceptional cases where you do not get n(= 3) eigenvectors, nor the general solution to the differential equation. See if you can find another form of solution to these differential equations; for this system it is actually easier to write the equations out and deal with them explicitly, rather than to think about eigenvectors. Problems 1. Solve x0 = 2x y0 = 3y x(0) = 5

y(0) = 0 2. Solve the system of problem 1 with the initial conditions x(0) = 0 y(0) = 5 3. Sketch the phase plane for the system of problem 1. 4. Solve the system of problem 1 with initial conditions x(0) = 1 y(0) = 1 Sketch your solution onto the phase plane, or run pplane on it. Show by eliminating t between your formulas for x(t) and y(t) that x = y
2

. This should help your sketch. 5. Solve x0 = 2x y0 = 3y Show that solutions lie on curves x = cy2 3 , or lie along the positive or negative axes, or are zero. Sketch the phase plane. 6. Sketch the phase plane for the system x0 = _ 12 0 3 _ x which was solved in the text. Does it look more like the phase plane for problem 3 or problem 5, qualitatively? 7. Solve x0 = x + 2y y0 = 4x y and sketch the phase plane. 8. Solve x0 = x y
3

86
y0 = x + 5y and sketch the phase plane. 9. You are given that a matrix A has eigenvectors u = 2 4 1 0 0 3 5, v = 2 4 0 3 1 3 5, w = 2 4 0 2 2 3 5, and that Au = 5u, Av = 4v, and Aw = 3w. Solve x0 = Ax, if x(0) = 2 4 1 6 2 3 5. Note that x(0) = u + 2v.

10. Whats rong with this? x0 = Ax x(0) = _ 3 6 _ and A _ 1 2 _ = _ 7 14 _ is given. So x = 3e7t + C.

22 Systems with Complex Eigenvalues

Todays lecture consists entirely of one example having complex eigenvalues. It is not really that hard, but we want to be careful and complete, the first time we work through one of these. Our system is x0 = Ax where A= 12 2 1 We know what has to be done to solve this; you find eigenvalues then eigenvectors of A, then the solution x(t) is in the form of linear combinations of the vet things. This is what we have done before, and here we will find a bit of complex arithmetic to go along with it. Here we go. First plot the phase plane to see what to expect. You can plot a little bit of it by hand, or you can run pplane on our system: y0 = y + 2z z0 = 2y + z 87
2 1 0 1 2 3 4 4 3 2 1 0 1 2 y z

Figure 22.1 The phase plane for our example. (figure made by pplane)

Now let the analysis begin. First we find the eigenvalues of A. We have det(A I) = det( 12 2 1 ) = 2 2 + 5 This is zero by the quadratic formula if = 2 p
44(5) 2 = 2 p 16 2=1

2i.

So we have our eigenvalues. Now lets look for an eigenvector for 1+2i. We must solve (A(1+2i)I)v = 0. Writing a and b for the components of v, this says 2ia + 2b = 0 2a 2ib = 0 The first equation says b = ia. The simplest choice we can make is to take a = 1, b = i. Now something strange happens. If you check these numbers in the second equation, you will find that they work there too! In fact, you can just forget about the second equation, because it is a multiple of the first one. I know it might not look like a multiple of the first one, but that is because it is an imaginary multiple of the first one. Anyway we have found that 12 2 1 1 i = (1 + 2i) 1 i Read that through again to check it, because there are lots of places you can make mistakes in complex arithmetic. 88 Next let us state the solution we have found so far, and expand it so we can see what it looks like. 1 i e(1+2i)t = et(cos(2t) + i sin(2t)) 1 i = et cos(2t) sin(2t) + iet sin(2t) cos(2t) Thats a mess, isnt it? It is neater to leave it in the original complex exponential form, but we expanded it out so you can see exactly what the real and imaginary parts are. Now we need a second solution, so we can go through all this again with the other eigenvalue 12i. The only thing which changes, if you read through what we just did, is that you must replace i by i wherever it occurs. This happens to work because A is a real matrix. The general solution to the system can then be expressed as x(t) = c1

 1 i e(1+2i)t + c2 1 i e(12i)t There is an alternate to this, sometimes preferred because it emphasises real solutions. This goes back to our discussion in Lecture 12 about real and imaginary parts of solutions to real equations. The same statement holds here, that if x = u + vi is a solution to x0 = Ax where u, v, and A are real, then u and v are also solutions. You can prove this yourself pretty easily. Taking the real and imaginary parts we found for the first solution above allows us to express solutions in the alternate form x(t) = a1et cos(2t) sin(2t) + a2et sin(2t) cos(2t) The relations between the constants are a1 = c1 + c2 and a2 = i(c1 c2). It is important to think about the phase plane and see whether it matches the solutions we found. You can see that we got the increasing exponential et, and this has to do with the fact that solutions are moving away from the origin in Figure 22.1. Also we got sines and cosines, which oscillate positive and negative. That has to do with the rotational motion in the picture.
Problems 1. Solve x0 = Ax if A = _ 2 1 1 2 _ . Find the solution if x(0) = _ 1 0 _ . 2. Solve x0 = _ 01 9 0 _ x. What is the second order equation for the first component x1 in this case?

89
3. Write out the system corresponding to x00 = x using x0 = y and solve it, find a conservation law for the equation and plot its level curves, and plot the phase plane for the system. How are these plots related? 4. Solve x0 = 3x y0 = z z0 = 9y

and plot some solutions. 5. Whats rong with this? x = _ 1 i _ e(1+2i)t = _ 1 i _ et(cos(2t) + i sin(2t)), so re(x) = et cos(2t), im(x) = et sin(2t), and x = et(c1 cos(2t) + c2 sin(2t)).