Boundary Value Problems Part 1

ACM 100c
Boundary value problems and the Sturm-Liouville ODE - part 1
Dan Meiron
Caltech
April 4, 2012
D. Meiron (Caltech)
ACM 100c - Methods of Applied Mathematics
April 4, 2012
1 / 20
Boundary value problems

So far we have explored linear (and nonlinear) ODEs but we have focused on the initial value problems. That is we looked at equations of the form dx = A(z)x dz where A is an n n matrix In order to consider some unique solution we applied n conditions at some initial point typically taken to be z = 0: x(z = 0) = x 0 . Under these conditions, we showed the existence and uniqueness of solutions provided A(z) obeyed various mild smoothness criteria.
D. Meiron (Caltech) ACM 100c - Methods of Applied Mathematics April 4, 2012 2 / 20

In many applications, we are asked to solve the same ODE dx = A(z)x, dz But we want the solution over some xed domain z0 < z < z1 rather than some neighborhood of z0 Most importantly not all n conditions are given at z = z0 . Instead, some are applied at z = z0 and some at z = z1 . Of course, for any hope of uniqueness in the linear case we must have a total of n such conditions. Such problems are called boundary value problems in contrast to the initial value problems we have studied up till now. In the cases we will study we will write the system as an nth order ODE where n is typically 2 But the theory we will present holds for the system as well.

The theory of existence and uniqueness for boundary value problems is considerably more complicated. This is to be expected because many of the guarantees we had for initial value problems are not present for boundary value problems. Consider a simple example of a second order homogeneous ODE y + p(z)y + q(z)y = 0 Suppose we know the general solution: y (z) = c1 y1 (z) + c2 y2 (z). Well assume that the coefcient functions are nice and smooth for any z So we can be assured the functions y1 (z) and y2 (z) are also similarly nice and smooth in the region z0 z z1 .
z0 z z1

Suppose we ask for a solution subject to the following conditions: y (z = z0 ) = a, y (z = z1 ) = b.
This is different from what we have done previously We are now asking that the solution satisfy two conditions as before. But they both involve the value of the solution at the two boundary end points. Plugging these conditions in, we get a 2 2 system to solve for c1 and c2 : c1 y1 (z0 ) + c2 y2 (z0 ) = a, c1 y1 (z1 ) + c2 y2 (z1 ) = b. Whether this system has a solution depends on the values the solutions take on at the boundary.

Now consider solving this 2 2 linear system: c1 y1 (z0 ) + c2 y2 (z0 ) = a, c1 y1 (z1 ) + c2 y2 (z1 ) = b. For example, suppose neither a or b are zero. In that case we will have a solution as long as y1 (z0 ) y2 (z0 ) = 0, y1 (z1 ) y2 (z1 ) In turn this clearly depends on what happens at the boundary and the values the solutions take on there. Suppose, on the other hand we had a = b = 0. Then, in general, we would expect the trivial solution y (z) = 0 This is because if the above determinant did not vanish, we would have to take c1 = c2 = 0 which is just the trivial solution

On the other hand, it might be that in some cases we did get that y1 (z0 ) y2 (z0 ) = 0. y1 (z1 ) y2 (z1 ) This might happen depending on the equation and the locations of the boundary In that case we would get nontrivial solutions but they would not be unique. We see then that such problems are harder to analyze. They seem to depend on matrices such as y1 (z0 ) y2 (z0 ) y1 (z1 ) y2 (z1 ). which are really about global information as regards the solution.

In contrast, linear initial value problems depend on the Wronskian determinant. For example for a second order ODE initial value problem y + p(z)y + q(z)y = 0 the Wronskian is given by W (z) = y1 (z0 ) y2 (z0 ) = 0. y1 (z0 ) y2 (z0 ) z z0
Abels theorem guarantees that this never vanishes as long as the matrix coefcients of a linear system are smooth. Boundary value problems (BVP) turn up in many applications and we will explore quite a few of these in ACM 100c. One particular class of BVP the Sturm-Liouville ODE and its associated boundary value problems is very important. We will motivate this next.
Example of a BVP - the heat equation

To motivate the importance of BVPs and also give a preview of a partial differential equation (PDE) we consider the solution of the heat equation in one space dimension. The heat equation is a partial differential equation that describes the evolution of heat in a uniform medium. The equation is given by (x, t) 2 (x, t) =D t x 2 We want to solve it in the domain 0 x 1, t > 0.
D. Meiron (Caltech)
April 4, 2012
9 / 20

We wont worry for now about where this equation came from well derive it later. Here x describes the length along some ideal rod of material that extends from x = 0 to x = 1. (x, t) is the temperature of the rod D is called the diffusivity, a measure of how well heat diffuses through the rod. Given some initial distribution of heat, we want to see how it evolves for t > 0. So at t = 0 we have (x, t = 0) = 0 (x). This is the initial condition for the problem.

We also need to say what happens at the edges of the rod during the time we are interested in getting the evolution of the temperature. The PDE wont have a unique solution unless we do this. Intuitively this makes sense since the ends could be insulating or perhaps connected to some heat bath that maintains a constant temperature. To specify what happens at the ends we specify boundary conditions at x = 0 and x = 1. We will assume that the temperature of the rod is xed at = 0 by some mechanism: (x = 0, t) = 0, (x = 1, t) = 0, t > 0.
D. Meiron (Caltech)
April 4, 2012
11 / 20
Solving the PDE

One very useful approach to such problems is called the method of separation of variables. We will motivate this approach in great detail later on. But for now we will just assume that (x, t) = T (t)X (x), or that the solution can be written in separable form If we substitute this into the PDE (well omit the details right now) it turns out to be possible to get such a solution if T (t) and X (x) satisfy the following ODEs: d 2 X (x) CX = 0, dx 2 dT (t) DCT = 0 dt where C is a constant called the separation constant.
Solving the PDE

Now lets try to solve the ODEs d 2 X (x) CX = 0, dx 2 dT (t) DCT = 0 dt There are many types of solutions to these equations as we will discuss later. But were interested in the ones that make physical sense. The diffusivity D is always positive (D > 0). We can solve the ODE for the time variable t to give T (t) = a exp(DCt), where a is some arbitrary constant and C is again the separation constant.
Solving the PDE

Now, physically, we dont expect that the temperature will grow exponentially without somehow externally heating the rod (which were not doing). So we try to obtain solutions in which C < 0 meaning our temperature decays in time which seems reasonable for this type of problem. So we set C = 2 , Here is hopefully real or at least 2 has a positive real part. Making this substitution in the ODE for X (x) we get the following boundary value problem: d 2 X (x) + 2 X = 0 dx 2
D. Meiron (Caltech)
X (0) = X (1) = 0.
April 4, 2012 14 / 20
Solving the boundary value problem

We get the following boundary value problem: d 2 X (x) + 2 X = 0 X (0) = X (1) = 0. dx 2 Now it looks like were stuck We have a homogeneous ODE and homogeneous boundary conditions. It seems the only solution is X (x) = 0. This is, in fact, true for almost any value of But recall from our previous discussion that one can nd nontrivial solutions to even a homogeneous boundary value problem This is because in the system we typically solve c1 y1 (z0 ) + c2 y2 (z0 ) = 0, c1 y1 (z1 ) + c2 y2 (z1 ) = 0, it might happen that the matrix has zero determinant

In that case we could get nontrivial solutions although they wont be unique. So we ask whether its possible to play with so that solutions exist. Up till now, we have not said anything about . We can solve the ODE, d 2 X (x) + 2 X = 0 X (0) = X (1) = 0. dx 2 since its a linear constant coefcient ODE to get the general solution: X (x) = c1 sin(x) + c2 cos(x) We want X (x) to satisfy the boundary conditions. At x = 0 in order to have X (0) vanish, we must have c2 = 0. Using this and trying to satisfy the other boundary condition at x = 1 gives us the equation c1 sin() = 0.

Now normally, the only solution to this equation c1 sin() = 0 is c1 = 0 We only get the trivial solution X (x) = 0. But we note that if we set so that the sin() vanishes then c1 could be arbitrary. We dont yet know what this means but at least we get some kind of solution. From the properties of the sine function we know that sin(n ) = 0 where n = n, n = 1, 2, 3, .
We have a set of solutions of the following type: n (x, t) = Bn exp(n2 2 Dt) sin(nx) where the Bn are arbitrary constants.
n = 1, 2, 3, .
The general solution

We see that we went from no solutions to a countable innity of solutions. The heat equation is a linear homogeneous PDE because each term (like the time and space derivatives) appear linearly and there is no inhomogeneous term So it seems to be like a linear homogeneous ODE for which we know we can use the principle of superposition of solutions Because the whole problem is linear we can see that a more general solution of this heat equation is a superposition of the solutions we just found:
(x, t) =
n=1
Bn exp(n2 2 t) sin(nx).
D. Meiron (Caltech)
April 4, 2012
18 / 20

Clearly, the sum
(x, t) =
n=1
Bn exp(n2 2 t) sin(nx).
satises the boundary conditions, because the sines vanish at x = 0, 1 But there is also an initial condition to satisfy. At t = 0 we have some starting distribution of heat in the rod: (x, 0) = 0 (x). In order to satisfy this condition we substitute t = 0 into
(x, t) =
n=1
Bn exp(n2 2 t) sin(nx) to get
0 (x) =
n=1
D. Meiron (Caltech)
Bn sin(nx)
April 4, 2012 19 / 20

So we would have a solution that satises all the conditions if we could gure out the coefcients Bn in the expression
0 (x) =
n=1
Bn sin(nx)
As promising as this looks, there are some unanswered questions: How does one determine Bn ? If you can determine Bn is there only one choice that works? Even if there is a unique choice of Bn can you show the series converges to 0 (x) as n ? If it converges at t = 0 does it converge for t > 0? The answers to these questions will take up the next few lectures.

Boundary Value Problems Part 1

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Boundary Value Problems Part 1

Uploaded by

Copyright:

Available Formats

ACM 100c

Boundary value problems and the Sturm-Liouville ODE - part 1

ACM 100c - Methods of Applied Mathematics