Poynting flux Adjoint #2191
Poynting flux Adjoint #2191
Conversation
…unction. The Poynting Flux adjoint still fails when given field components of zero. The test file test_poynting_flux.py does the same simulation for adjoint and non-adjoint calls now. I'll need to add it into the actual test cases next, however.
…or in step_db.cpp - Assertion 'changed_materials' failed. Creating a pull request will let me difftool code changes to debug the problem.
…dded some more TODOs
… Off by a factor of ten in initial tests, debugging today.
…ine sources (one for each tangential H or E component) to a series of point sources to accomodate end-to-end interpolation - we needed middle to middle. Unfortunately, this did not seem to improve the error appreciably.
Codecov Report
@@ Coverage Diff @@
## master #2191 +/- ##
==========================================
- Coverage 73.09% 71.05% -2.04%
==========================================
Files 17 17
Lines 4906 4975 +69
==========================================
- Hits 3586 3535 -51
- Misses 1320 1440 +120
|
|
Ideally this would be done in C++ directly from the In particular, it should be very similar to Line 1462 in ea43f83 dJ comes not from the user but rather from the Fourier field you are multiplying with (e.g. the Ex adjoint source's "dJ` comes from Hy etcetera).
Note that the Line 537 in ea43f83 |
… dJ. Works well in 2D, fails in 3D. Multiple frequencies untested.
…ro (due to symmetry, etc.). Tested with eigenmode and point sources radiating toward line monitors. 3D returns the wrong gradient entirely - I should be able to fix it since it should just be the format of the input. There are a couple cases in 2D simulations where the gradient is somewhat wrong.
|
Works ok in 2D (see below), fails in 3D (still sticking with a purely Python implementation using FourierFields). I can resolve 3D on my own - I expect the derivative's format isn't what FourierFields expects when we place the adjoint source. The accuracy doesn't improve monotonically in 2D. The gradient's inaccurate in a few cases. Note sure why, I'll explore it some more once it works in 3D. The same resolutions cause similar errors with different sources. Plots from 2D: Eigenmode source resolution is 10
Eigenmode source resolution is 22
Eigenmode source resolution is 26 Eigenmode source resolution is 30
Eigenmode source resolution is 42
point source resolution is 22 point source resolution is 26 |
…'s place_adjoint_source function. This works in 2D, but it returns 9 separate sources in 3D. The extra sources are handled and collated correctly now.
Closes #1307, #1299, #1398
Initial Implementation of Poynting Flux Adjoint. Uses
add_DFT_fieldsobjects. Forward run is accurate. Throws an error instep_db.cppon adjoint run:"Assertion 'changed_materials' failed."Using this pull request to help debug code changes.