unexpected results when using ReferenceIntersector , and the Elevation of Levels .

[jonathan.taVCBM2]

Hi all, 

I am trying to figure out if a column (vertical ones only at the moment) reaches a floor or a beam on both ends.

I am finding the expected Floors or Beams but the distance is slightly (0.0000000000040000) shorter than the column. There seem to be no difference if the column is attached at it's top and base to elements or not.

I obtained the elevation parameters of the relevant levels and saw the following : 

Level 2 300.0000000000010000
Level 1 -0.0000000000024364

While the Levels property show 0.0 or 300.0

On top of that - the distance between the intersected floor and beam differ from calculated height of the column , by a similar small number . Since all elements are set in relation to the levels (see screenshot)  this really shouldn't happen. 

The only way I can think is to round all the values to 0.0001 format using Math.Round(value , 4) , I am not too thrilled with this solution.  Any other ideas how to go about this ? 

Thanks , Jonathan

[jeremy_tammik · Accepted solution]

Every floating point calculation performed on a digital computer that has a problem dealing with very small calculation errors such as you describe is fundamentally flawed. 

Every floating point calculation performed on a digital computer needs to be implemented in such a way that it can gracefully handle such errors.

It will help a lot to study this problem in more depth before trying to address your issue directly.

You can start by reading the short article about floating point arithmetic accuracy problems on Wikipedia:

https://en.wikipedia.org/wiki/Floating-point_arithmetic#Accuracy_problems

The Building Coder has frequently discussed such problems and one simple approach to address them called 'fuzz':

https://www.google.com/search?q=fuzz&as_sitesearch=thebuildingcoder.typepad.com

Once you have taken a look at the material suggested above, you will be able to easily resolve your problem yourself.

Good luck and have fun,

Jeremy

[jonathan.taVCBM2]

Thanks Jeremy, 
I understand the gist of the real numbers accuracy and hardware limitations, I am familiar with the XYZ/UV IsAlmostEqualTo methods , I'll read more on the other fuzzy comparisons suggested in these links, pretty sure I'll have some fun.
Thanks again,  Jonathan.