tmm-faster 0.1.3

A lightweight Python library with a C++ core for very fast calculations of reflection and transmission spectra in optical thin-film systems using the transfer-matrix method.

tmm_faster

tmm_faster is a lightweight Python library with a C++ core for very fast calculations of reflection and transmission spectra in optical thin-film systems using the transfer-matrix method (TMM).

Features

• Calculation of reflection and transmission: Simultaneous calculation of both $s$ and $p$ as well as averaged polarization of isotropic multilayer systems using the transfer-matrix method.
• Include incoherent layers: Fully coherent stacks are calculated with calc_coherent, stacks containing incoherent layers with calc_incoherent (for an example, see Figure 1).
• High performance: C++ core with OpenMP parallelization for rapid calculation of multiple wavelengths and/or incidence angles. Small datasets (<10,000 wavelengths/angles) are up to 130x faster than tmm_fast, large datasets are still 1.7x faster (or 3.2x faster for both polarizations), see Table 1.

Number of wavelengths	tmm_faster	tmm_fast (CPU)	tmm_fast (CUDA)
1	0.02 ms	2.2 ms (110x)	7.2 ms (360x)
10	0.03 ms	3.9 ms (130x)	10.2 ms (340x)
100	0.1 ms	12.8 ms (128x)	24.5 ms (245x)
1,000	0.8 ms	33.9 ms (42x)	178.8 ms (224x)
10,000	9.4 ms	18.3 ms (1.9x)	33.5 ms (3.5x)
100,000	76.1 ms	130.8 ms (1.7)	203.2 ms (2.7x)
1,000,000	710.9 ms	1308.6 ms (1.8x)	1959.3 ms (2.7x)

Table 1: Runtimes for coherent TMM calculation, 10 layers, random n, k and d, 45° angle
(on Intel Core i9-13900H, NVIDIA RTX 4000)

Installation

With pip (Python 3.10+ required)

pip install tmm_faster

Build from source (C++17 compiler required)

git clone [https://github.com/clembron/tmm_faster.git](https://github.com/clembron/tmm_faster.git)
cd tmm_faster
python setup.py build_ext --inplace

Get started

See examples/example.py:

import tmm_faster
import numpy as np

wavelengths = np.linspace(200,1200,1000)
d = [np.inf, 5, 500, 5e4, 200, np.inf] 
c = ['i', 'c', 'c', 'i', 'c', 'i']
angles = np.linspace(0,90,1000)

n_list = [1.0, 0.05 + 3.0j, 2.4 + 0.001j, 1.5, 2.4 + 0.001j, 1.0]
n_array = np.array([n_list for _ in wavelengths])

res_incoherent = tmm_faster.calc_incoherent(n_array, d, c, angles, wavelengths)
res_coherent = tmm_faster.calc_coherent(n_array, d, angles, wavelengths)

Figure 1: Reflection spectra (s-polarization) for an example stack with only coherent layers (left) and the same stack but with the thickest layer in the middle being treated as incoherent (right).

References

• tmm https://github.com/sbyrnes321/tmm
• tmm_fast https://github.com/MLResearchAtOSRAM/tmm_fast

Disclaimer

This project was developed with the assistance of Gemini (Google AI). All physical models and core logic were verified by me.