PyStratify against tmmax

← Verification · Thin films · Plane wave

Reference: tmmax (JAX transfer matrix, float32)
Kind: code: another group's code
Calculated: R, T of five stacks (single layers, two layers, a 5-pair mirror, metal-dielectric) × 3 wavelengths × 3 angles × s, p
Agreement: 2e-6 (float32; observed ≤ 4.2e-7). tmmax 1.1.4 is wrong for two or more coherent layers (up to 0.79 in R): its scan step builds a (2, 2, 1) array that matmul broadcasts — reported upstream (#45) and fixed on tmmax's main 2026-02-11, unreleased on PyPI; with the step fixed it agrees, shipped multilayers are shown, not asserted

100 nm of n = 2.1 in air, normal incidence: R against tmmax (JAX, float32) as shipped tmmax computes in single precision (its scan carries complex64), so agreement stops near 1e-7; single layers are asserted as shipped, to 2e-6.

R, PyStratify R, tmmax 1.1.4 as shipped

Hover to read values.

largest |R − R_tmmax| (float32): 1.48×10⁻⁷
PyStratify the lines, the reference the circles

A five-pair quarter-wave mirror (n = 2.35 / 1.46) on glass, normal incidence: R against tmmax, as shipped and with its scan step fixed tmmax 1.1.4, the version on PyPI, is wrong for two or more layers: its scan step broadcasts the 2×2 matrices as a stack of column vectors (reported upstream, bahremsd/tmmax #44, #45; fixed on its main branch, not released). With the step rebuilt (tmmax_films.py) it agrees to float32; as shipped it is off by up to 0.79. The test asserts the fixed step on multilayers.

R, PyStratify R, tmmax 1.1.4 with the scan step fixed R, tmmax 1.1.4 as shipped

Hover to read values.

largest |R − R_tmmax| (fixed step, float32): 1.12×10⁻⁶
PyStratify the lines, the reference the circles

Run it yourself: tmmax_films.py · test_tmmax_films.py · figures.py · TMMAX_FILMS.md on GitHub, PyStratify 97e52a9