Allow raw refractive index input (no manual conjugation required)#101
Open
jhpark94 wants to merge 5 commits intokc-ml2:mainfrom
Open
Allow raw refractive index input (no manual conjugation required)#101jhpark94 wants to merge 5 commits intokc-ml2:mainfrom
jhpark94 wants to merge 5 commits intokc-ml2:mainfrom
Conversation
Three changes for correct lossy material support: 1. convolution_matrix.py: eps = conj(n)^2 internally, so users pass raw n (standard exp(-iwt) convention, Im(n)>0 = loss). 2. transfer_method.py (transfer_1d_1): Substrate boundary uses eigenvalue-consistent q = sqrt(kx^2 - eps) instead of 1j*conj(kz)/n^2. Fixes incorrect Fresnel for lossy substrates. 3. transfer_method.py (transfer_1d_3): Detect same-material interfaces (V_i @ G ≈ -I) and skip boundary matching, applying propagation only. Fixes uniform lossy layer on same lossy substrate (e.g., Au on Au). Verified against analytical results: - Air -> Au, TM: R=0.981936 (Fresnel exact) - Au(20nm)/Glass, TM: R=0.911680, T=0.050702 (thin-film exact) - Au(20nm)/Glass, TE: R=0.911680, T=0.050702 (exact) - Si(200nm)/Glass, TM: R=0.602732, T=0.397268 (no regression) - Au on Au, all thicknesses: R=0.981936 (Fresnel exact) - Au grating on Au: R+T=0.945 (physical, T>0) Breaking change: users should pass raw n in ucell (remove conj()). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
jhpark94
force-pushed
the
fix/exp-minus-iwt
branch
from
d98feee to
9a2a994
Compare
Same three changes as numpy backend: 1. convolution_matrix: eps = conj(n)^2 internally 2. transfer_*_1: eigenvalue-consistent substrate boundary 3. transfer_*_3: same-material interface detection JAX uses jax.lax.cond for JIT compatibility. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…hold Replace the heuristic V_i@G ≈ -I check (atol=0.1) with explicit eps comparison: compare the layer's diagonal eps (from convolution matrix) with the eps below (substrate or previous layer). This is more robust, interpretable, and doesn't require tuning a threshold. Applied to all three backends (numpy, torch, JAX). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- transfer_1d_conical_3, transfer_2d_3: add same_material parameter - _base.py solve_1d_conical, solve_2d: add eps comparison logic - Fix torch transfer_1d_1 type_complex casting (eps_bot) - Applied to all three backends (numpy, torch, JAX) Verified (numpy backend): - Oblique incidence (0-60°, TE/TM): all match Fresnel - 1D conical: Si/Glass R+T=1, Au/Glass R+T<1 - 2D grating: Si/Glass R+T=1, Au patch on Au T>0 - Au uniform on Au (2D): R=0.981936 = Fresnel Note: torch backend has a pre-existing issue with 1D dielectric (R=1.659 for Si/Glass) that exists in the original meent code and is not caused by this PR. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Three bugs fixed:
1. sqrt branch cut: torch loses -0j imaginary sign from kx.conj(),
causing sqrt(-real+0j) to select wrong branch (+1.5j instead of
-1.5j). This flips the G matrix sign for propagating orders,
producing R=1.659 instead of 0.603 for Si/Glass. Fixed by adding
-1e-20j perturbation to force correct branch selection.
2. same_material T calculation: eigenvalue ordering from torch.linalg.eig
differs from numpy, so T = T @ X applied propagation phases to wrong
diffraction orders. Fixed by transforming X to physical basis:
T = T @ W @ X @ W^{-1}.
3. TM de_ti in torch transfer_1d_4: used kz_bot instead of kz_bot.conj(),
giving wrong/negative transmittance for lossy substrates.
All three fixes applied consistently across numpy, torch, and jax backends
for 1D, conical, and 2D solvers.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Author
Update: Three additional torch backend bugs fixedBug 1: sqrt branch cut (critical)
Impact: Si/Glass TM gave R=1.659 instead of correct R=0.603. Bug 2: same_material eigenvalue ordering
Impact: Si/Si gave T=0.000 instead of correct T=0.694. Bug 3: TM de_ti missing
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Users can now pass standard refractive indices directly — no manual
conj(n)needed. Meent handles the exp(+iωt) convention internally.Previously, users had to:
Now:
Changes
convolution_matrix.py:eps = np.conj(layer)**2— applies conjugation internally instead of expecting pre-conjugated inputtransfer_method.py: Substrate boundary uses eigenvalue-consistentq = sqrt(kx² - eps)instead of1j * conj(kz) / n², fixing incorrect results for lossy substratesVerification
Breaking change
Users who previously passed
conj(n)inucellwill now get double conjugation. They should remove the manualconj()call.Scope
Closes #100
🤖 Generated with Claude Code