Work

Sensitivity analysis and physics-aware deep learning for aluminum mirror coatings: when network architecture outweighs loss-function design

Year: 2026
Type: article
Abstract: Aluminum is the most prevalent material for broadband mirror coatings, yet the design parameters affecting its specular performance, such as film thickness and surface roughness, are not well quantified. Additionally, the pace of developing surrogate models for metallic reflectance has been slower than for dielectric models. This study addresses these gaps by implementing a numerically validated transfer-matrix framework for a single absorbing aluminum layer on a glass substrate, accompanied by an optical sensitivity analysis. The engine agrees with an independent transfer-matrix implementation to machine precision over 400–1200 nm, and its conclusions are robust to the choice among five aluminum optical-constant datasets. Results indicate reflectance saturates past an opaque thickness of approximately 40 nm, and the p -polarized response shows a pseudo-Brewster minimum that does not reach zero. Surface roughness decreases specular reflectance due to the Debye–Waller factor, with a characteristic $$1/\lambda ^{2}$$ scaling affecting the blue end of the spectrum. We then present a physics-informed deep-learning surrogate model that links thickness and roughness to the reflectance-deficit spectrum, achieving a coefficient of determination of 0.9999 with held-out data. Our findings suggest that the architecture of the neural network is crucial for addressing metallic reflectance, contrasting with traditional approaches, provided that the loss function constrains the magnitude of the spectrum. A surrogate-driven search of the bounded design space, verified by exact transfer-matrix calculations, recovers the physically expected boundary optimum (an opaque, smooth film with a band-averaged reflectance of 91.5%), which we use as an end-to-end consistency check of the pipeline; the same calculations show that 36.5 nm of aluminum already reaches within 0.5 percentage points of this value.
Language: English
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7236736892
FWCI:
Cites: 45
Cited by:
Related to: 10