A centroid location catalog of P-wave microseisms is crucial for understanding the origins of microseisms. Although a back-projection method is feasible for locating the centroids, the computational cost is still expensive for making a global catalog over ten years. Contrary, although the computational cost of beamforming is low, it cannot distinguish P from PP waves. To combine the advantages of both methods, we develop the auto-focusing method as a natural extension of beamforming. In the first step of this method, we estimated the slowness vector based on conventional beamforming and the epicentral distance inferred from the wavefront curvature by maximizing the beam power. In the second step, we iteratively update the values based on the perturbation theory. In the third step, based on the classified phase according to the estimated epicentral distance and the slowness, we infer the source location from the slowness vector with corrections for a global 3-D P-wave velocity structure. We also infer the centroid-single-force (CSF) from the beam power. We applied this method to the vertical components of seismic records at approximately 780 Hi-net stations in Japan from 2004 to 2020. We also compare the CSF catalog with a synthetic CSF catalog based on a numerical ocean wave model: WAVEWATCH III. Both catalogs generally show similar temporal-spatial patterns of centroids. The amplitudes of CSF are consistent with each other, although the seismic signal-to-noise ratio limits the detected events. Exceptionally, significant activities in the Gulf of Carpentaria cannot be explained by the ocean wave model.