Toward Generalized Dynamic Metasurface Holography with Learned Real-Momentum-Locked Evolution

Abstract Dynamic holography is pivotal for next-generation displays with critical applications in augmented reality, virtual reality, and beyond. While metasurfaces have emerged as powerful approaches for dynamic holography, existing works remain largely confined to intensity-only modulation, restricting the available channels for display and information storage. Herein, we realize a generalized platform applicable to major types of dynamic holography (monochromatic, color, 3D, and vectorial) by introducing a learned metasurface and aperture synthesis framework (LMAS). This paradigm extends Fourier duality from the intrinsic static real-momentum mapping relationship into the dynamical regime and establishes a learned mapping between real-space aperture displacement and momentum-space holographic evolution via gradient optimization. Crucially, the aperture’s inherent field-agnostic nature provides a universal physical interface that unifies diverse optical degrees of freedom, allowing us to demonstrate four distinct types of dynamic meta-holography. The LMAS approach establishes metasurfaces as a generalized, compact, and scalable architecture for holographic evolution within a minimalist system configuration, offering broad potential in advanced displays, optical storage, and information multiplexing.