"but something goes wrong because if I convert it back (from cartesian to polar) I don't get the same results"
Nothing goes wrong, you just have two different representations of the same angle.
Your input angles are given over the range 0 .. 2*pi, but cart2pol documentation clearly states that its ouput "...value of the angle is in the range [-pi pi]", so we would certainly not expect to have the same values.
You can convert to 0..2*pi using mod:
theta2 = mod(theta2,2*pi);
Note that due to floating point error we would still not expect those values to be exactly the same, but close enough.
