modified mpid to use filtering on the phase mask parameters. Automatic differentiation is used to correctly compute gradients. Moreover, the optimization can down be carried out according to different strategies, which can either be: average the objectives, use an epigraph reformulation. The epigraph formulation needs further attention because it does not always leed to good results. A possible solution might be a combination of the two strategies: at the beginning use an average cost function to set up the problem and then switch to an epigraph. This would be like an hot start strategy, since I have noticed the epigraph being not totally capable of handling the problem well if it starts from scratch. On the other hand, the use of the filtering action can even leed to better results than without it, although it is necessary to tune the filter radius. I have seen that even a 1um radius is generally sufficient to remove outliers in the final phase mask, which was the main reason we decided to implement the filter in the first place. In the end, the all defined classes assume we interface to them using dimensions in micrometers. Conversion to meters is handled internally.

sorter_to_gauss.py

@@ -6,20 +6,20 @@ if not os.path.exists("results"):
 
 #Space grid definition
 N = 1025
-unit_cell_pitch = 700e-9
+unit_cell_pitch = 0.700 #i.e., 700 [nm]
 
 n0 = 1.45
 n1 = 1.0
 n2 = 1.45
 mat_refr_index = np.array([n0, n1, n2])
 
-d0 = 476e-6 #propagation distance: source - MS1 [m]
-d1 = 510e-6 #propagation distance: MS1 - MS2 [m]
-d2 = d0 #propagation distance: MS2 - target [m]
+d0 = 476 #propagation distance: source - MS1 [um]
+d1 = 510 #propagation distance: MS1 - MS2 [um]
+d2 = d0 #propagation distance: MS2 - target [um]
 d = np.array([d0, d1, d2])
 
-ms_max_available_diameter = 122e-6 #[um]
-output_fiber_core_diameter = 19e-6 #[um]
+ms_max_available_diameter = 122 #[um]
+output_fiber_core_diameter = 19 #[um]
 opt_max_eval = 140
 
 space_grid = pid.SpaceGrid(
@@ -29,7 +29,7 @@ space_grid = pid.SpaceGrid(
     materials_refr_index=mat_refr_index,
 )
 
-wavelength = 1.55e-6
+wavelength = 1.55
 
 #define the sources
 source1 = pid.Source(input_field='fiber_modes/LP11_field_distribution_lda_1550_N_1025.npy', 
@@ -43,13 +43,13 @@ source2 = pid.Source(input_field='fiber_modes/LP21_field_distribution_lda_1550_N
 
 
 target = pid.Target(input_field='fiber_modes/LP01_field_distribution_lda_1550_N_1025.npy', 
-                    center=(20e-6,0),
+                    center=np.array([20,0]),
                     associated_source=source1, 
                     wavelength=wavelength, 
                     space_grid=space_grid, 
                     field_norm=True)
 target2 = pid.Target(input_field='fiber_modes/LP01_field_distribution_lda_1550_N_1025.npy', 
-                     center=(-20e-6,0),
+                     center=np.array([-20,0]),
                      associated_source=source2, 
                      wavelength=wavelength, 
                      space_grid=space_grid, 
@@ -69,15 +69,17 @@ sim = pid.Simulation(
     n_cpu=1
 )
 
-opt_param_min = 0
-opt_param_max = 1
+opt_param_min = 0.0
+opt_param_max = 1.0
 
 sim.set_optimization(opt_param_min=opt_param_min, 
                      opt_param_max=opt_param_max,
                      ms_max_avail_diameter=ms_max_available_diameter,
                      phase_mask_init_size=output_fiber_core_diameter,
                      opt_max_eval=opt_max_eval,
-                     conic_filter_radius=1.5e-6)
+                     strategy = 'mean',
+                     conic_filter_radius = 1.0, #i.e. 700 [nm]
+                    )
 
 sim.run()