高能暂现源 · 每日文献综述

Gravitational wave (GW) microlensing induced by compact dark matter (DM) offers an unparalleled opportunity to explore the fundamental nature of dark matter, as it enables the detection of optically invisible compact objects that are otherwise inaccessible to electromagnetic observations. Conventional approaches for identifying such lensed GW signals suffer from inherent drawbacks: matched-filtering algorithms struggle with the complex, parameter-sensitive diffraction patterns of wave-optics microlensing due to template bank limitations. Additionally, weak lensed signals can be easily obscured by background noise, leading to challenges in maintaining high detection reliability in low signal-to-noise ratio (SNR) regimes. In this work, we introduce the Wavelet Convolution Detector (WCD), a deep learning framework tailored to identify wave-optics diffraction imprints in lensed GW signals. The WCD integrates multi-scale wavelet analysis into residual convolutional blocks, enabling efficient extraction of subtle time-frequency interference structures characteristic of microlensing. To ensure generalization to realistic astrophysical scenarios, the model is trained on a physically motivated synthetic dataset that incorporates realistic distributions of compact DM masses, lens redshifts, and lensing probabilities. Evaluated on simulated binary black hole (BBH) events injected into Gaussian noise mimicking third-generation GW detector environments, the WCD achieves a test accuracy of 92.2% and an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.966. At a stringent false positive rate of 3%, the model maintains a true positive rate of 86.5%, demonstrating robust performance in distinguishing subtle wave-optics imprints from stochastic noise. This makes the WCD a scalable, efficient tool for large-scale blind searches of compact DM candidates via GW microlensing in the upcoming era of third-generation GW detectors.