This paper presents a novel approach for unsupervised graph domain adaptation (UGDA), called A2GNN, which focuses on the propagation process in graph neural networks (GNNs). The authors argue that the propagation process plays a crucial role in adapting to different graph domains, and that the generalization capability of GNNs has been largely overlooked in previous studies. Through both empirical and theoretical analysis, they show that removing propagation layers in the source graph and stacking multiple propagation layers in the target graph can lead to a tighter target risk bound. The proposed A2GNN model is simple yet effective, and is evaluated on real-world datasets, demonstrating superior performance compared to existing state-of-the-art methods. Theoretical analysis shows that the generalization gap of multi-layer GNNs depends on their propagation layers, and that the proposed A2GNN achieves a tighter error bound. The results indicate that the propagation process is essential for graph domain adaptation, and that the asymmetric architecture of A2GNN is effective in improving performance. The paper also provides a comprehensive theoretical analysis of the error bound for multi-layer GNNs, and shows that the proposed A2GNN achieves a tighter error bound. The experiments demonstrate that A2GNN outperforms recent state-of-the-art baselines on node classification tasks with different gains. The paper concludes that the propagation process is crucial for graph domain adaptation, and that the proposed A2GNN is a simple yet effective method for unsupervised graph domain adaptation.This paper presents a novel approach for unsupervised graph domain adaptation (UGDA), called A2GNN, which focuses on the propagation process in graph neural networks (GNNs). The authors argue that the propagation process plays a crucial role in adapting to different graph domains, and that the generalization capability of GNNs has been largely overlooked in previous studies. Through both empirical and theoretical analysis, they show that removing propagation layers in the source graph and stacking multiple propagation layers in the target graph can lead to a tighter target risk bound. The proposed A2GNN model is simple yet effective, and is evaluated on real-world datasets, demonstrating superior performance compared to existing state-of-the-art methods. Theoretical analysis shows that the generalization gap of multi-layer GNNs depends on their propagation layers, and that the proposed A2GNN achieves a tighter error bound. The results indicate that the propagation process is essential for graph domain adaptation, and that the asymmetric architecture of A2GNN is effective in improving performance. The paper also provides a comprehensive theoretical analysis of the error bound for multi-layer GNNs, and shows that the proposed A2GNN achieves a tighter error bound. The experiments demonstrate that A2GNN outperforms recent state-of-the-art baselines on node classification tasks with different gains. The paper concludes that the propagation process is crucial for graph domain adaptation, and that the proposed A2GNN is a simple yet effective method for unsupervised graph domain adaptation.