Guilin Sun
Ansys Employee
This is due to numerical errors, such as limited mesh size, source injection, PML etc. Discrete FDTD computation is mesh dependent, and can approach theoretical result when the mesh size approach zero and PML is sufficiently thick to absorb all energy.  

You can use the montor in between the source and the front surface.

The error is already on the order of 10^-4, quite small. Please refer this link for more information about converging test