Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Thanks to comments, it should be a plane but why does it look a bit like a fractal? Does my code overlook something or some err in plotting tool? I used Python and GNUplot.

enter image description here

Apparently an animated illusion.

And generated the points with Python and generated graph with GnuPlot like:

$ python copyPasteTheCodeToFile.py > .data
    $ gnuplot -e "set terminal png; set grid; splot '.data'"

Python code

import itertools

def main():
   density = 100.0
   mySet= [x/density for x in range(int(density))]

   for (x,y,z) in [(x,y,z) for x,y,z in itertools.product(mySet, repeat=3) if x+y+z==1]:
      points = points + "%s\t%s\t%s\n"%(str(x),str(y),str(z))

   print points


can you spot an err? I cannot hence I thought A) mathematical err, B) algoritmic err or C) outside-party/program-err, cannot really say which one.


It is a floating point error! The error is the conditional x+y+z==1. Since computers don't evaluate floats like 1.0 + 1.0 to 2.0 but something slightly different, the conditional fails with some right points. The fix is:

$abs(x+y+z-1) < \epsilon$

where the epsilon $\epsilon$ is some very small number like 1e-10. It is still open why there are just the holes in the grid. Why not other points? And why is it vertically symmetric? Is it so with every computer? Does the pattern vary between computers? Anyway I am still investigating why there this specific pattern with this conditional. We know now why there are the patters but we don't know why these patterns. I am running the commands with bulk i86 comp.

share|improve this question
The equation $x + y + z = 1$ forms a plane. Your graphing device gives the illusion that it is "fractal-like", but it is really just a solid plane. –  JavaMan Apr 1 '11 at 19:07
@DJC: thank you, been really perplexed with this one but cannot spot the err. Added info about tools used. I am very unsure whether the problem is algorithmic, mathematical or in tool. Hopefully this is the best place to ask. –  hhh Apr 1 '11 at 19:17
I don't know anything about Python or GNUplot, so maybe this should be turned into a separate question to receive more attention. Maybe still, it might be better for a CS forum. I Would turn it into a new question, and then if this is not appropriate for math.se someone can let us know. –  JavaMan Apr 1 '11 at 21:22
@DJC: I don't know. It is probably not a mathematical question, anymore. Fine, it should be plane so it becomes either Programming, Cross-validaded.SO or some-else-many-thing-related-category question but does it have mathematical value in terms of fractals? Can you do such plane with holes in fractals? If so how? Because it may be an illusion, I cannot firmly say which site I should forward this to. –  hhh Apr 1 '11 at 21:30
I'm a little out of my league here. Maybe someone else will know what appropriate actions should be taken with your new question. –  JavaMan Apr 1 '11 at 21:31

2 Answers 2

up vote 3 down vote accepted

I suspect that the error here comes down to the use of floating-point arithmetic (which has a rather fascinating mathematical structure all its own). Your code is iterating over all of the points $(x, y, z)$ and doing a precise compare of the sum to 1.0; but, for instance, it may well be the case that the triad $(30.0/100.0, 30.0/100.0, 40.0/100.0)$ won't be in your set simply because the floating point representations of .3 and .4 aren't exact and so the sum $.3 + .3 + .4$ comes out to, say, $1.00000001$. While I can't speak specifically to the fractal structure you're seeing, I suspect it's loosely related to the representation of the Sierpinski Gasket in terms of the set of all $(x, y)$ with $x$ AND $y == 0$; see http://ecademy.agnesscott.edu/~lriddle/ifs/siertri/binary.htm for more details on this. The easiest way to fix this would be to just iterate over all points $(x, y)$ and define $z$ in terms of $x$ and $y$ by $z=1.0-(x+y)$; this also has the advantage of speeding up your generation a hundredfold (by eliminating the loop over $z$). More generally (for instance, in circumstances where there isn't an explicit equation for $z$ in terms of $x$ and $y$), floating point compares should almost never be exact, and a test for $(1.0-\epsilon)\lt x+y+z \lt (1.0+\epsilon)$ (for some small $\epsilon$) is a better bet.

EDIT to answer the new parts of the question that came in while I was answering: in general, floating-point representations are likely to be the same across platforms; the IEEE has a standard specification for floating-point numbers and the arithmetic using them, which most platforms support. There are corner cases where specific platforms might differ, but they're unlikely to be exercised by simple additions like this.

share|improve this answer
you are right, I noticed the err and updated the question while you was apparently writing. There is a wave style horizontal patterns. Then there are like Sierpinski Gasket holes but I am unsure what is it really. I am trying to get better precision now with decimal.decimal class and perhaps trying partial fractions/surds to get very high precision to really see what is going on. Ideas why just the holes? And why such patterns in such locations? I think the vertical symmetry is clear (mirror) because the situations are pretty much the same on the sides but other things open. –  hhh Apr 1 '11 at 22:45

I suspect what you are seeing are nothing more than Moiré fringes; artifacts of the plotting method used by your computing environment.

share|improve this answer
not totally true. There is no Moire fringes on 2D plane, found it out (sorry just deleted that picture). The floating point err creates the holes while the 3D -perspective apparently creates the Moire fringes (plotting env. illusion). The 2D holes are perfect triangles, fascinating, no idea why they are created just so perfectly. –  hhh Apr 4 '11 at 2:19

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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