Line in epsilon-matrix for non-integer geometric_center

[HelgeGehring]

Hey, for a simulation (with the latest git-version) I found a strange output for the epsilon values. I've reduced it to a minimal-example:

import numpy as np
import meep as mp
import matplotlib.pyplot as plt

# geometry = [mp.Prism([mp.Vector3(*x) for x in [(70., 0.), (70., 1.), (77., 1.), (77., 0.)]],
#                     height=1, material=mp.Medium(epsilon=2))]
geometry = [mp.Block(center=mp.Vector3(74, 0), size=mp.Vector3(7, 2), material=mp.Medium(epsilon=2))]

sim = mp.Simulation(cell_size=mp.Vector3(79, 15), resolution=15,
                    geometry_center=mp.Vector3(35.5, 0), geometry=geometry,
                    boundary_layers=[mp.PML(2)], subpixel_maxeval=0)

sim.init_sim()

epsilon = sim.get_epsilon()
print(np.unique(epsilon))
plt.imshow(epsilon[1100:])
plt.colorbar()
plt.show()

print(np.unique(sim.get_epsilon()))

As output I get (where the second line is not always the same):

[1.         1.04347826 1.2        1.5        1.84615385 2.        3.        ]
[1.         1.04347826 1.2        1.41176471 1.5        2.        ]

and in the plot there's an additional horizontal line:

also when getting the epsilon the second time, there's still this line.

What I noticed:

• It happens with both geometries in the script
• The position of the line depends on the PML-thickness and is quite close to the PML
• If I change the geometric_center to integer numbers, everything's alright
• Without subpixel_maxeval=0 the problem remains. I also found an example where this line is not constant and the numbers get far higher

[oskooi]

There is a known bug in geometry_center (see e.g. #1335) which has not yet been resolved. Running your example as-is, the values output from get_epsilon() that I am seeing also includes nans.

In your example, if you remove geometry_center from the Simulation object constructor and shift your Block object accordingly, the results are as expected:

geometry = [mp.Block(center=mp.Vector3(74-35.5, 0), size=mp.Vector3(7, 2), material=mp.Medium(epsilon=2))]

output

Using MPI version 3.1, 1 processes
-----------
Initializing structure...
time for choose_chunkdivision = 0.000194522 s
Working in 2D dimensions.
Computational cell is 79 x 15 x 0 with resolution 15
     block, center = (38.5,0,0)
          size (7,2,0)
          axes (1,0,0), (0,1,0), (0,0,1)
          dielectric constant epsilon diagonal = (2,2,2)
time for set_epsilon = 0.0873263 s
-----------
[1.         1.04347826 1.2        2.        ]
[1.         1.04347826 1.2        2.        ]

[HelgeGehring]

Thanks for the quick reply! For integer numbers it seems to work alright for me. I'll keep an eye on that and if I encounter any problem I'll just avoid using geometric_center. Thanks!