Aluminum mirror coatings achieve optimal specular performance past an opaque thickness of 40 nm, according to research published in Nature on Thursday.
The study implemented a numerically validated transfer-matrix framework alongside an optical sensitivity analysis for a single absorbing aluminum layer on a glass substrate. Researchers found that the engine agrees with an independent transfer-matrix implementation to machine precision over 400–1200 nm. Conclusions remained consistent across a choice of five aluminum optical-constant datasets.
Reflectance and Surface Roughness
Surface roughness decreases specular reflectance due to the Debye–Waller factor. This introduces a characteristic inverse-square wavelength scaling that affects the blue end of the spectrum.
Calculations show that a thickness of 36.5 nm of aluminum reaches within 0.5 percentage points of the physically expected boundary optimum. That optimum consists of an opaque, smooth film delivering a band-averaged reflectance of 91.5 percent.

Deep-Learning Surrogate Models
The team developed a physics-informed deep-learning surrogate model linking film thickness and surface roughness to the reflectance-deficit spectrum. The model achieved a coefficient of determination of 0.9999 with held-out data.
Neural network architecture handles metallic reflectance when paired with a loss function that constrains the magnitude of the spectrum, according to the published findings. A surrogate-driven search successfully recovered the expected boundary optimum, serving as an end-to-end consistency check for the pipeline.
Nature, the British weekly scientific journal founded in the autumn of 1869 by Norman Lockyer and Alexander MacMillan, operates editorial offices in London, the United States, continental Europe, and Asia under Springer Nature. It publishes peer-reviewed research across the natural sciences and, according to the 2022 Journal Citation Reports, holds one of the highest impact factors among multidisciplinary science journals at 50.5. Fewer than 8% of submitted papers are accepted for publication.
Researchers applied physics-informed machine learning to predict the composition and bulk density of Al2O3–TiB2 composites, as reported by Wiley on October 26, 2026. The materials were synthesized through furnace SHS and laser ignition chemical synthesis from the Al–TiO2–B2O3–H3BO3 reactive system.
| Model / Synthesis | Metric | Result |
|---|---|---|
| Model 1 (V1, V2) | Hold-out testing R2 | 0.994 |
| Model 2 | Test R2 | 0.927 |
| LIChemS Composites | Experimental bulk density | 3.71 g/cc |
| Model 2 (ML) | Predicted bulk density | 3.68 g/cc |