EXTREMELY WEIRD arithmetic bug

Hi there,

Basically I'm adding 0.001 to a number over and over again. The series starts with zero so the resulting number should ALWAYS be a multiple of 0.01, but instead it eventually degenerates into numbers like 5.874000000000296 - obviously NOT a multiple of 0.001!

Can you explain why? Can it REALLY BE that Matlab just can't cope with numbers of that precision? It's a mathematics package!!!

Run the following code to reproduce the bug. I'm running 7.10.0(R2010a).

Just run log=testcase() and then inspect the latter half of log in the variable editor- when you double-click on a cell to expand it you will be able to see the lack of precision.

function [ log ] = testcase( )
%RUN LIKE THIS: log=testcase()
%Then inspect the log in the variable editor.

    runTime=20;

    t=0;

dx=0.001;

i=1;

log(i,1)=t;

it=2;

for i=2:runTime/dx

      t=t+dx;

log(i,1)=t;

end

end

Answer 1

[Arnaud Miege]

Have a look at Why does 1 - 2/3 - 1/3 not equal zero?.

Arnaud

[louise]

I'm being an idiot. It's the floating point problem. I know all about the floating point problem.

SIGH.

I was stupid enough to assume that Matlab was using some futuristic high-precision arithmetic. Nope.

As usual, the problem is BAD ASSUMPTIONS.

Thanks for your fast help.