Pansharpening aims to fuse a coarse spatial resolution multispectral （MS） image with a fine spatial resolution panchromatic （PAN） image to generate a fused image with fine spatial and spectral resolution. Existing deep learning-based pansharpening methods still face difficulties in simultaneously extracting local high-frequency details and preserving global spectral consistency. Consequently， fused images are prone to spectral distortion when spatial details are enhanced. Moreover， the spatial resolution gap between PAN and MS images often causes feature misalignment during the fusion process， resulting in visual artifacts such as local edge ghosting and color distortion. To address these issues， this paper proposes a Nested Multi-Scale Fusion Network （NMSFusion）， which performs synergistic modeling at both local and global levels. At the local level， a Multi-Scale Gated Block （MSGB） is utilized to extract high-frequency details from the PAN image. At the global level， an Artifact-Free Residual Fusion Module （ARFM） is designed to coordinate global semantics， ensuring spectral consistency between the fused images and the original MS images. Additionally， an Adaptive Coordinate Encoding Module （ACEM） is introduced to alleviate the feature misalignment caused by the resolution gap. Finally， conventional deep networks with random weight initialization tend to disrupt the original color distribution， leading to convergence difficulties and spectral distortion in the early training stages. To tackle this， we propose a plug-and-play Identity Initialization Mechanism （IIM）. By constraining the initial weight of network layers， IIM forces the initial network output to equal the interpolated MS image， thereby providing a reasonable initial state for optimization and promoting stable convergence. Experimental results on three remote sensing datasets demonstrate that NMSFusion outperforms nine representative pansharpening methods in both quantitative metrics and visual assessments. Furthermore， ablation studies demonstrate that the proposed modules not only synergistically enhance the fusion performance but also guarantee an efficient and lightweight network architecture.