relative error in numerical methods

6.3k Views Asked by At

I have a question about a task I am trying to complete. I have created a code to produce a determinant of a $100\times 100$ matrix. Now I want to compute the relative error between my solution and the solution obtained using

numpy.linalg.det.

So I know that the relative error is calculated by

$$\left| \left| \frac{x-x_0}{x} \right| \right|$$

where $x$ is the absolute value and $x_0$ is the approximation. My question is what is the absolute value and what is the approximation, my code I have generated or using numpy.linalg.det

Both my answer and using numpy.linalg.det are almost equal.

1

There are 1 best solutions below

0
On

The definition of relative error you quoted is (apparently) meant to represent the relative error between an exact value $x$ and an approximation $x_0.$

More generally, the formula still works for other kinds of relative errors; just set $x$ to the value that you want the error to be relative to.

I think the usual practice is, if one of the numbers is somehow considered to be more accurate or reliable, you set $x$ to that number. In your case, since one of the numbers comes from a well-known numeric programming software package that (presumably) has been vetted by many people, and the other number is something you just came up with yourself, I would use numpy's value as $x$ and your value as $x_0.$ If the numbers are really very close, however, it does not make much difference which one you put in the denominator.