Enable Eigenmode Features with Dispersive Materials #919

smartalecH · 2019-06-14T02:17:57Z

The finished version of #475.

Modifies the get_eigenmode routines such that dispersive materials are now supported. Specifically, I modified the get_chi1inv() function, which now evaluates the (real) inverse permittivity by summing all of the susceptibilities at a given point in a given direction for a specific field component at a specific frequency. It should be backward-compatible, as the default frequency is 0 (i.e. only the DC frequency is evaluated in this case).

As mentioned, it only evaluates the real permittivity, since MPB's iterative solver can (currently) only solve for real eigenvalues.

That being said, this method still enables a lot of important features, one of them being S parameter extraction for dispersive materials (provided the materials are mostly dielectric).

Closes #415.

I've implemented a test that compares the actual permittivity, the return value of chi1(), and the "effective index" of a uniform medium (i.e. just the refractive index). All three should be the same.

I check this for Silicon, SIlver, and Lithium Niobate so that we get a range of susceptibilities.

…dispersion merging

smartalecH · 2019-06-14T15:12:40Z

It looks like the serial implementation fails because the new routine doesn't like materials that have been rotated. It's able to pull the epsilon at those values just fine (using sim.fields.get_chi1inv()). But when MPB checks the dielectrics, it throws the "invalid dielectric function" error.

stevengj · 2019-06-14T19:29:50Z

If MPB is giving that error it must be that your ε matrix is not positive-definite.

smartalecH · 2019-06-18T16:26:05Z

This latest iteration fixes the original bug, which means that the routine can now use the get_eigenmode() routines with dispersive materials.

In addition, I modified the array_slice and the h5fields routines so that the user can optionally specify a frequency and they will evaluate the permittivity/permeability at that frequency. This is essential for the visualization module and useful for other post-processing applications.

The array_slice routines seem to be working fine (I've implemented additional tests within dispersive_eigenmode.py). The h5 routines fail when executed in parallel. For some reason, there's a communication error with the get_chi1inv_at_pt() function in monitor.cpp. This is the same function used for the get_eigenmode routines and the array_slice routines (with no issues there).

I'm a little stumped, so if anyone else with more experience with parallel h5 generation could look at it, that would be great.

stevengj · 2019-06-19T18:11:25Z

src/monitor.cpp

+      if (eps.imag() == 0.0){
+        res = 1.0 / eps.real();
+      }else{
+        res = 1.0 / (std::sqrt(eps).real() * std::sqrt(eps).real());


What's going on here? Why not just res = real(1.0 / eps)?

What we really want is real(1/n) ^2, which is not the same as real(1.0 / eps) if eps is complex.

smartalecH · 2019-06-19T18:11:36Z

Output:

Initializing structure...
Working in 2D dimensions.
Computational cell is 16 x 8 x 0 with resolution 10
     block, center = (0,0,0)
          size (1e+20,1,1)
          axes (1,0,0), (0,1,0), (0,0,1)
          dielectric constant epsilon diagonal = (1,1,1)
-----------
Initializing structure...
Working in 2D dimensions.
Computational cell is 16 x 8 x 0 with resolution 10
     block, center = (0,0,0)
          size (1e+20,1,1)
          axes (1,0,0), (0,1,0), (0,0,1)
          dielectric constant epsilon diagonal = (1,1,1)
time for set_epsilon = 0.0751069 s
lorentzian susceptibility: frequency=0.000905797, gamma=0
lorentzian susceptibility: frequency=0.88125, gamma=0
time for set_epsilon = 0.0731909 s
lorentzian susceptibility: frequency=0.000905797, gamma=0
lorentzian susceptibility: frequency=3.31657, gamma=0
lorentzian susceptibility: frequency=0.88125, gamma=0
lorentzian susceptibility: frequency=3.31657, gamma=0
-----------
-----------
creating output file "./test_animation-eps-000000.00.h5"...
creating output file "./test_animation-eps-000000.00.h5"...
HDF5-DIAG: Error detected in HDF5 (1.10.4) thread 0:
  #000: H5F.c line 444 in H5Fcreate(): unable to create file
    major: File accessibilty
    minor: Unable to open file
  #001: H5Fint.c line 1364 in H5F__create(): unable to open file
    major: File accessibilty
    minor: Unable to open file
  #002: H5Fint.c line 1615 in H5F_open(): unable to lock the file
    major: File accessibilty
    minor: Unable to open file
  #003: H5FD.c line 1640 in H5FD_lock(): driver lock request failed
    major: Virtual File Layer
    minor: Can't update object
  #004: H5FDsec2.c line 941 in H5FD_sec2_lock(): unable to lock file, errno = 35, error message = 'Resource temporarily unavailable'
    major: File accessibilty
    minor: Bad file ID accessed
meep: error on line 455 of ../../src/h5file.cpp: error opening HDF5 output file
--------------------------------------------------------------------------
Primary job  terminated normally, but 1 process returned
a non-zero exit code. Per user-direction, the job has been aborted.
--------------------------------------------------------------------------
--------------------------------------------------------------------------
mpirun detected that one or more processes exited with non-zero status, thus causing
the job to be terminated. The first process to do so was:

  Process name: [[39506,1],0]
  Exit code:    1
--------------------------------------------------------------------------

src/monitor.cpp

…sion merge

python/geom.py

python/tests/dispersive_eigenmode.py

src/monitor.cpp

stevengj · 2019-06-25T18:02:05Z

Note that this is not really doing the interpolation from the Yee grid correctly, so it is making a first-order error in principle. In practice, that error only shows up at the boundaries between materials, where it is hard to be accurate anyway, so let's just accept the 1st-order error for now and leave it to another PR to do the interpolation more correctly.

smartalecH · 2019-06-25T21:49:16Z

The PR should be ready for merging. Here's a summary of the features:

Enable get_eigenmode routines with dispersive, anisotropic materials (excluding gyromagnetic materials, for now).
get_array and output_h5 routines now accept an omega parameter and can evaluate the permittivity/permeability at that frequency. Defaults to 0, which performs the previous behavior.
Visualization module can accept an omega parameter to plot the dielectric at that frequency. Defaults to the center frequency of the first source in the sources list.
Small bug fixes in the visualization module (with updated tests)
Added a rotate function to the Medium class.

As discussed, a better interpolation scheme is needed to improve the accuracy around the material boundaries.

stevengj · 2019-06-26T15:37:20Z

src/monitor.cpp

-  Vinv[2 + 3*0] = 1/det * (V[1 + 3*0]*V[2 + 3*1] - V[1 + 3*1]*V[2 + 3*0]);
-  Vinv[2 + 3*1] = 1/det * (V[0 + 3*1]*V[2 + 3*0] - V[0 + 3*0]*V[2 + 3*1]);
-  Vinv[2 + 3*2] = 1/det * (V[0 + 3*0]*V[1 + 3*1] - V[0 + 3*1]*V[1 + 3*0]);
+  if (det.real() == 0 && det.imag() == 0) abort("meep: Matrix is singular, aborting.\n");


doesn't det == 0.0 work?

I was just being safe -- do you want it changed?

det == 0.0 seems clearer and is safe.

(You can't do det == 0, on the other hand, because C++ doesn't let you mix complex<double> with scalars of other types.)

Just fixed with 8590db1

stevengj · 2020-06-20T02:27:47Z

src/monitor.cpp

+        // Account for conductivity term
+        if (conductivity[cc][dd]) {
+          double conductivityCur = conductivity[cc][dd][idx];
+          eps = std::complex<double>(1.0, (conductivityCur/omega)) * eps;


This line looks off to me — it seems like you are doing an element-wise product rather than a matrix multiplication?

Since σ can only be a diagonal matrix currently, a matrix multiplication would multiply by conductivity[cc][cc] (the diagonal element from row cc), no?

stevengj · 2020-09-09T02:07:36Z

What are the step_func_args that were added to some functions, like output_epsilon? They don't seem to be used.

* first stab * begin with test files * bug fixes * update test * found one bug * bug fixes, h5 output, and array-slice output * fix serial bugs * more serial fixes * fix mu issue * disable h5 support * big fix * generalize matrix inverse and fix parallel bug * fix memory issue * add tutorial * minor fixes * refactor * check1 * case2 * add docs * cleanup * check3 * fix test * 1 more fix! * updates * det fix

smartalecH added 6 commits May 13, 2019 15:42

first stab

4582204

Merge branch 'dispersion' of https://github.com/smartalecH/meep into …

c08123e

…dispersion merging

begin with test files

8ede31f

bug fixes

b546a9c

update test

43b6adb

merge

e0ef6ed

smartalecH added 4 commits June 15, 2019 14:31

found one bug

23cd5f6

bug fixes, h5 output, and array-slice output

a4ac0f2

fix serial bugs

b0f65c3

more serial fixes

78d9053

fix mu issue

939a99e

stevengj reviewed Jun 19, 2019

View reviewed changes

src/monitor.cpp Outdated Show resolved Hide resolved

smartalecH added 13 commits June 19, 2019 16:41

disable h5 support

6155d59

big fix

6eac066

generalize matrix inverse and fix parallel bug

c4bc10b

fix memory issue

97a7b0a

add tutorial

8a1e635

Merge branch 'master' of https://github.com/NanoComp/meep into disper…

1f8d91b

…sion merge

minor fixes

81f3cc8

refactor

041ea9c

check1

a3e4987

case2

8eebb18

add docs

9a33f4c

cleanup

2d6aa85

check3

6046280

smartalecH added 2 commits June 25, 2019 10:55

fix test

176834a

1 more fix!

6ad454a