Optimization of computer-generated holograms with diffraction-engineered initialization

Computer-generated hologram (CGH) enables precise wavefront modulation, which is widely used in holographic displays and laser processing. Gradient-based hologram optimization methods integrate physical models and loss functions with high flexibility. Their performance is highly sensitive to initialization, but the initial phase is generally unrelated to the target field, resulting in a lack of specific guidance and suboptimal performance. In this work, we propose a gradient-based optimization framework with diffraction-engineered initialization. Unlike stochastic or uniform phase initialization, the initial phase is obtained via inverse diffraction calculation from the target field, incorporating physical prior information into the initialization. This physics-informed initial estimate effectively guides the optimization toward a more favorable solution space. Numerical and optical experiments demonstrate that the proposed method enhances convergence speed and reconstruction quality.