1
$\begingroup$

From what I remember from a lecture I had of a course I'm attending called "introduction to computational science", floating-point numbers are distributed logarithmically. What does it mean? And how can I visualize it?

I've a slide where it is said:

We assume that all binary numbers are normalized. Between powers of 2, the floating point numbers are equidistant.

I think this is related to the logarithmic distribution or spacing of the floating point numbers, but I don't understand exactly what it means.

I've also below the statement taken from the same slide this picture:

enter image description here

Apart from the range between $0$ and $0.25$, it seems that the number or density of floating point numbers is decreasing, if this picture actually represents the distribution of floating-point numbers.

Why is that?

Why there's this exception between $0$ and $0.25$?

$\endgroup$
2
$\begingroup$

This statement is referring to the fact that there are the same number of floating point numbers representable between $1$ and $2$ as between $2^k$ and $2^{k+1}$ for any $k$. This is because the mantissa of $x$ and $2x$ have the same representation as a floating point number, only the exponent of $2$ changes.

$\endgroup$
3
  • $\begingroup$ Indeed. They are the range from $2^k$ to $2^{k+1}$ for $k=0$. $\endgroup$ – robjohn Apr 9 '16 at 14:30
  • 1
    $\begingroup$ There are $3$ ticks between $0.25$ and $0.5$; $3$ ticks between $0.5$ and $1$; $3$ ticks between $1$ and $2$; $3$ ticks between $2$ and $4$; $3$ ticks between $4$ and $8$; etc. It looks logarithmic. $\endgroup$ – robjohn Apr 10 '16 at 13:13
  • 1
    $\begingroup$ How many do you think there would need to be between $0$ and $0.25$? One has to stop somewhere before infinity. Sort of like double precision numbers stop at $2^{-1022}$. $\endgroup$ – robjohn Apr 10 '16 at 13:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.