In this talk, we will discuss our recent findings: a new dimensionality-independent and gradient-free sampler called Geometric Optics Approximation Sampling, based on a reflector antenna system. The method constructs a reflecting surface that redirects rays from a source with a simpler measure to achieve a desired distribution defined by a complex target measure. By dual re-simulating or ray tracing the reflector antenna system, we can generate numerous independent samples from the target measure. We employ the supporting paraboloid method to derive the reflecting surface without requiring gradient information about the target measure's density, using low-discrepancy or random sequences for discretization. We define a geometric optics approximation measure as the pushforward of the source measure, proving its stability concerning the target domain and establishing error bounds under Wasserstein metrics. The efficacy of our sampler is demonstrated through various examples and numerical experiments, particularly in Bayesian inverse problems, confirming our theoretical results.