How to measure LSBits-of-difference between floats (if possible)

[Jarrod_S.]

Hi everyone,

I just wanted to throw in a small tidbit regarding functional tolerances. A new constant was added in LabVIEW 8 to the numerical palette, the so-called Machine Epsilon. This is a double-precision numeric representing the positive non-zero number with the least absolute value possible. It comes out to around 2.2E-16.

I thought it might be useful for some of these tests.

[Xu]

Hi Altenbach,

Thank you for response, but ... 32 is too small to make 2 and 1.99989998 equal in my VI. Here 32 means 4.48415509E-44 in single floating point number. Try tolerance=1000, for example.

I think denormalized and normalized are always different. Is there any example for identical denormalized and normalized?

Thanks again,

Xu

[altenbach]

Yes, you're right. Sorry. 😄

[Xu]

No sorry.... Thanks,

[Dynamik]

Hi XU,

I was too ignorant to appreciate your solution when you posted it.    Finally got around to exploring/testing your VI, but still didn't understand floats well enough to trust your solution!  Finally ended-up here:

http://www.cygnus-software.com/papers/comparingfloats/comparingfloats.htm

My conclusions is: You're smart!

Huge thanks for the perfect solution.

[altenbach]

Ditto.

Anyway, using LabVIEW 8 which supports 64bit integers, we can now easily adapt it to support comparison of DBLs. 😄

Here's my attempt (LabVIEW 8.0), please test if it works!

[Dynamik]

Hi Altenbach,

      Saved your VI - but can't test it until I break-down and purchase 8.0 (really, I'm waiting for 8.1  )

Anyway, thanks for the code (and for the stars  )