Problem with estimating error

Hi,

We are trying to calculate the error in a double pendulum with Runge Kutta method.

We first run the program for an h = 0.001 and saved the 4 values inside of a reference vector. We then normalized the vector and compared it's norm against results for bigger h.

The problem is that although h is divided by two in every step, the error seems to be growing.

Why is that happening?

Pastebin

[Walter Roberson]

Heisenberg Uncertainty Principle. At some point, reducing the step size must increase the error, as otherwise you would be able to pinpoint both the position and the momentem simultaneously.

[am]

Hi, this method does not work for even big h such as 0.1.

Can you please look at our code?

[Walter Roberson]

0.1 Planck lengths is too small to work with. There are theoretical reasons to believe that space itself might not exist in any continuous form at scales that small.

[Walter Roberson]

Undefined function or variable t in the RKstep call.

You might be changing the h, but you are not resetting the time.

[am]

I moved in all the vectors and the t, the error is now getting smaller!

But it has not accurace O(h⁴) as expected?

[am]

Additionally, if I choose a smaller h, than I have a bigger error. Is it what you meant with Heinseberg uncertainity principle?