What is the method to calculate error?

I was fooling around with Comsol Multiphysics. I have found very interesting feature included in Comsol. Recently I was testing second harmonic generation module in it.

The error is calculated and plotted with respect to the iteration number. In my sense, if the iteration number increase the error should be decreased (attached pic). But I have found some alternative in this method.

Could you please help me out?

Thanks in advance.