Computer Limitations

Mathematics in real life is continuous, not so on a computer. Article based on FastAI Computational Linear Algebra course1.

Course page

Source: https://tzeny.ddns.net:4430/Tzeny/fastai_numerical-linear-algebra/blob/master/nbs/1.%20Why%20are%20we%20here.ipynb

Floats

800px,Gaps between floating point numbers depending on the interval in which the numbers reside
800px,Gaps between floating point numbers depending on the interval in which the numbers reside

Floating point numbers have a maximum and a minimum: 1.79 * 10308and as small as 2.23 * 10−308.

There are uneven gaps between numbers.

Machine epsilon

The error between 1 and the next larger number: ϵmachin**e = 2−53 ≈ 1.11 * 10−16 (as defined by IEEE)

I will stop maintaing course notes for this course as the Jupyter Notebooks are more than adequate.

  1. https://www.youtube.com/watch?list=PLtmWHNX-gukIc92m1K0P6bIOnZb-mg0hY&v=8iGzBMboA0I