# Doubt about convergence test on differential equations

1. Jun 9, 2015

### jonjacson

I will try to explain this with an analogy.

Let's have this equation:

x^2 =9

And let's assume I don't know algebraic methods to solve it, so I create a list using excel with different values. And I see that if I put x=4 it doesn't work, if I put x=5 it is even worse and so on. But If I put 3.9 I see it gets closer to 9, if I put 3.8, 3.7 it gets closer and closer so finally I found x=3 just by trial and error.

My question is, let's have a differential equation:

dy/dx + (dy/dx)^2 = 0

ANd let's say I want to use the previous method to find a dy/dx in a very specific point, let's say the point (0,1). The question is, Can I be sure it will converge?

With an algebraic equation even if I don't know a method to get the general solution using radicals I am able to get solutions just by substituting values and watching if they work or not. Can I do the same with differential equations?

Just by using different straight lines with different slopes and testing if they work on a very specific point, Can I be sure the values will converge to a solution?

Maybe not and I create an immense list of values with excel and I don't see the data geting closer to any slope, i don't know.

2. Jun 9, 2015

### DEvens

http://en.wikipedia.org/wiki/Differential_equation

In principle, yes. In practice, it takes some clever methods. And there may be easier approaches.

The problem is, your example of x^2 = 9 has a single definite answer. (Well... I suppose -3 works also.) And there is an easy to understand single measure of how good a candidate is.

For the differential equation you then pose, the answer is not a single value but a function. Actually, a family of functions, the specific one of which is selected by specifying some boundary conditions. You are really trying to find a curve that satisfies this equation. A single point is not enough. For example, a single point can't define the slope, and your equation involves the slope.

So in general it is difficult to define a simple function that will show you how far you are away from the true answer. But you can certainly define a function that shows how far away from 0 your value of dy/dx + (dy/dx)^2 is, and then do some clever iterations to make it closer to 0. It will require some cleverness to be sure you are doing it at all possible points on your curve at the same time. But there are a variety of approaches. Here are some of the most basic ones.

http://en.wikipedia.org/wiki/Numerical_methods_for_ordinary_differential_equations

I think you may have chosen an equation that is a little too simple for your purposes. Essentially what you have written is that the square of the slope is the negative of the slope. There are basically only two choices. The slope is zero or the slope is -1. If the slope is zero you have a horizontal line. Any horizontal line will do. If the slope is -1 you have a line with a -45 degree angle. Any such line will do. So you pick the particular case with boundary conditions. For example, if you required the curve to go through the points (1,0) and (2,0) it has to be one particular horizontal line. Or if you required it to go through (-1,1) and (1,-1) then it is the line at -45 degrees through the origin.

Hmmm... Now that I think of it, if you do not require the curve to have a smooth derivative you can get very interesting results by allowing the slope to flip between 0 and -1. That would be an interesting curve. One might reject such a curve if the slope had some physical interpretation.

3. Jun 9, 2015