Import mode profile from HFSS to use as source in MEEP

[dalarev]

Hi,

As the title states, I'm wondering if it's possible to import a mode profile from an HFSS simulation for use as a source in MEEP.

It may be possible to compute the mode profile directly in MEEP - if this is the case, may I ask for a link to the relevant documentation? This is for a circular, corrugated, metallic waveguide which could have several properties that could make it an expensive FDTD computation.

Is this possible? What exactly would I need from HFSS to achieve this? A 2D cross-section of complex field values, I assume.

Appreciate your thoughts and comments.

Cheers!

[smartalecH]

So meep uses MPB to compute modes needed for mode sources. MPB requires all materials to be hermitian (so no loss, gain, metals, etc). It also does not support cylindrical coordinates.

There is an effort to generalize the current EigenmodeSource to allow for arbitrary mode solvers.

But if you've already solved for the mode profile using something like HFSS, and can export it as a numpy array, the meep Source object can then take that numpy array and use that as the source's spatial profile. It will take care of all the necessary interpolation under the hood. Just be sure to create a source for each field component, to specify the proper size, ensure the mode is sufficiently sampled, etc.

[oskooi]

the meep Source object can then take that numpy array and use that as the source's spatial profile.

The amp_data property of the Source object simply specifies the amplitude profile of the current source which is not necessarily that useful.

A useful feature that currently does not exist but would be nice to support would be to read in a field profile from a file or Numpy array and then use it to form an equivalent current using the procedure described in Section 4.2.1 "Principle of Equivalence" of our book chapter: https://arxiv.org/abs/1301.5366.

[hammy4815]

@dalarev ,

Yep, that's correct. If you continue to have trouble formatting the field parameter, you can take a look at the script I'm attaching here as a working example. It uses the get_equiv_sources function to create equivalent sources for the analytic solution to a 2-D gaussian beam and then runs meep with them.

eq-sources-example.py.zip

[dalarev]

Thanks @hammy4815.

I was able to get some sims running using an .fld file from a dipole in HFSS, but I'm seeing some discrepancies.

I need a bit more time to review my code. Yours will be helpful for comparisons.

[dalarev]

Question about fields and the size of the field components that fill it.

My source data is in list format, so I use griddata to interpolate it onto a regular grid. This uses np.meshgrid, which, as you know, returns a (row, column) size rather than (Nx, Ny).

So, when I initialize field = np.ndarray(shape=(6,Nx,Ny,1)), should I fill in the field components by using the transpose? For example, field[0,:,:,0] = Ex.T, where Ex is my interpolated field component? I've tried this and a few other combinations, but there appears to be some strange orientation error in my code.

I hope this makes sense. The full script is here, if that helps.

Appreciate the guidance!

edit: actually, I think the one-line fill was incorrect, so I changed it to a loop where I flip the i,j of the field component:

for i in range(Nx):
     for j in range(Ny):
          field[0,i,j,0] = Ex[j,i]
          ...

But I still think there's some small mistake somewhere because I'm not quite matching the HFSS results yet.

[hammy4815]

Hi @dalarev ,

The way you are filling your array and passing into the eq_source function looks correct to me.

Do you have enough points? You seem to be interpolating onto a 500x500 point grid from only about 2000 datapoints. I'm not sure if this is the issue however, considering your points seem to be chosen at the high information points in your field.

You seem to be looking at time domain fields, I wonder if you try looking at the meep DFT fields at your frequency point over a long enough simulation if it might look more like your starting DFT fields that you provided to eq_source.

[dalarev]

Thanks for the guidance, @hammy4815.

It's easy enough to try a higher-res interpolation grid, so I will give that a shot. I did notice that this parameter changed the results somewhat, so it needs further looking into.

Good idea re: the DFT fields! It is true that I am only interested in results at one frequency, so this makes more sense. I will refer to the docs for guidance on how to use these functions.

I very much appreciate the guidance!

[stevengj]

I assume regardless of what mode you are sourcing (TE, TM, or some hybrid HE mode), both complex E and H field profiles are required from HFSS?

Technically, you can use just the E fields (to make a magnetic current –n×E) or just the H fields (to make an electric current n×H). This still launches the waveguide mode, but the mode will propagate in both directions away from the source instead of just forwards. (It corresponds to taking the one-way source and averaging it with ± its mirror flip).

[dalarev]

Yes, that's exactly what I'm seeing, but I wasn't 100% on the math behind it - thank you!

[stevengj]

See the section on the principal of equivalence in https://arxiv.org/abs/1301.5366

[dalarev]

Hi everybody, hoping to get your input on an intriguing result using get_equiv_sources. I won't spam code but can provide it as needed.

Trying to validate MEEP get_equiv_sources fields with those of a dipole in HFSS. I export near-fields from HFSS and use get_equiv_sources to calculate equivalent currents to define the MEEP source. Long story short: when using get_equiv_sources, MEEP matches the HFSS field profile in shape but underestimates the peak magnitudes; however, when using the HFSS fields directly, without calculating equivalent currents, the amplitudes more closely match up.

This is summarized in the image below. Again, I don't want to spam with detail, but I'm hoping to get any thoughts you may have with this information. Appreciate your input in advance!