Yes, have a look at
E = eps(b)
When you convert to single you are going to get the nearest single value to a within E. To see what you are dealing with, look at this:
a = (999999999-32):(999999999+34); b = single(a);
fprintf('%10.0f %10.0f \n',[a;double(b)])
Now when you do
c = b-a
MATLAB does the equivalent of this:
c2 = single(double(b)-a)
Also have a look at this:
