Do autograd gradients not flow through FieldProjectionCartesianMonitor?

[samuelkim16]

I am trying to do adjoint optimization of a structure, somewhat drawing from the tutorials on Inverse design of a metalens and Modeling a metalens (specifically the section on field projection to a focal plane). Because the point of optimization is far away from the structure, I would like to use FieldProjectionCartesianMonitor to project the near fields to a position far away from the structure. However, I am calculating the gradient norm (np.linalg.norm(gradient)) to be 0, and I get the warning:

20:33:42 Eastern Daylight Time WARNING: Adjoint simulation for task 'lens'      
                               contains no sources. This can occur if the       
                               objective function does not depend on the        
                               simulation's output. If this is unexpected,      
                               please review your setup or contact customer     
                               support for assistance.

If I replace the monitor with a normal field monitor, then I get gradients again, so I believe the simulation is set up correctly.

The other possibility is that I am accidentally cutting off the gradients in the objective function, which looks something like this:

def prod_power(sim_data, freq):
    focal_field_data = sim_data["focal_plane_proj"].fields_cartesian.sel(f=freq, z=focal_length, x=0, y=0)
    focal_power_ex = anp.abs(focal_field_data.Ex)**2
    focal_power_ey = anp.abs(focal_field_data.Ey)**2
    power_prod = focal_power_ex * focal_power_ey
    return power_prod.item()