epsilon_input_file with .hdf5 file

[dalarev]

Hi everybody,

Trying to import custom 2D geometry using epsilon_input_file with an hdf5 file; hoping to get some guidance on the proper usage.

MWE:

%matplotlib inline
from matplotlib import pyplot as plt
import meep as mp
import h5py

sim = mp.Simulation(
    cell_size=mp.Vector3(1,1),
    resolution=64,
    epsilon_input_file='A.hdf5',
)

plt.figure(dpi=100)
sim.plot2D()
plt.show()

This produces a blank plot (no geometry). I've uploaded the source hdf5 file here. It is simply a cylinder of radius 0.1a with eps = 2.5 and an air background.

I've tried both .hdf5 and .h5 files, but they produce the same results. I generated this array in Julia and exported with HDF5.jl.

I've confirmed that the file contains the expected geometry/dielectric values by visualizing with jupyterlab-h5web.

I am just becoming familiar with hdf5 files, but I suspect the issue may be in the "structure" of the file: when I visualize using jupyterlab-h5web, I see a root and a child; perhaps this is in conflict with MEEP's guidance of being the "first/only" dataset in the file.

I'll also mention that if I give the epsilon_input_file some bogus name like does_not_exist.hdf5, no error is reported and the result is the same blank plot.

Looking forward to your inputs. Thank you!

[dalarev]

The issue persists even when I generate an .h5 file from MEEP geometry.

MWE is below and notebook with outputs is here.

%matplotlib inline
from matplotlib import pyplot as plt
import numpy as np
import meep as mp

## Test: create MEEP geometry, output_epsilon, and try to read this MEEP-generated hdf5

resolution = 64
cell_size = mp.Vector3(1,1,0)

geometry = [mp.Cylinder(0.2, material=mp.Medium(epsilon=2.0))]

sim = mp.Simulation(resolution=resolution,
                    cell_size=cell_size,
                    geometry=geometry,
                   )
sim.plot2D()
sim.run(mp.at_beginning(mp.output_epsilon), until=0)

sim = mp.Simulation(resolution=resolution,
                    cell_size=cell_size,
                    epsilon_input_file="eps-000000000.h5",
                   )
sim.plot2D()
plt.show()

[oskooi]

I think the issue is with the formatting of your HDF5 file. As a test, I used h5utils and was not able to read it:

$ h5ls A.hdf5
A.hdf5: unable to open file

Can you try creating your HDF5 file using h5py?

[dalarev]

Thanks @oskooi.

The HDF5 file works fine for me using h5py. See notebook output.

Regardless, in my 2nd post, I generate the h5 file from within MEEP, which should be self-consistent but still fails.

Thanks for the help.

[dalarev]

Creating the HDF5 file using h5py is also producing a blank plot: notebook with output here.

At this point, the only remaining user error I can think of is an incorrect value epsilon_input_file.

Appreciate your support with this.

[oskooi]

Looks like a bug in #1827 whereby loading the geometry via the epsilon_input_file feature still requires you to manually initialize the meep.Simulation object before invoking plot2D.

sim = mp.Simulation(resolution=resolution,
                    cell_size=cell_size,
                    epsilon_input_file="eps-000000000.h5",
                   )

# manually initialize the sim object
sim.init_sim() 

sim.plot2D()
plt.show()

With this change, the simulation output does show that the HDF5 file was successfully read.

Initializing structure...
read in 64x64x1 epsilon-input-file "eps-000000000.h5"
time for choose_chunkdivision = 8.069e-06 s
read in 64x64x1 epsilon-input-file "eps-000000000.h5"
Working in 2D dimensions.
Computational cell is 1 x 1 x 0 with resolution 64
time for set_epsilon = 0.00823663 s
-----------
read in 64x64x1 epsilon-input-file "eps-000000000.h5"

Thanks for catching this. Will try to post a fix soon. In the meantime, the workaround is quite straightforward.

[dalarev]

Awesome, thanks for the notes!

I also just learned that defining default_material as a NumPy array works, too.

Thanks again!