The paper introduces ViSNet, an equivariant geometry-enhanced graph neural network designed to improve the representation and prediction of molecular structures and properties. ViSNet addresses the limitations of existing methods by efficiently extracting geometric features and reducing computational costs. The proposed model outperforms state-of-the-art approaches on multiple molecular dynamics (MD) benchmarks, including MD17, revised MD17, and MD22, and achieves excellent performance on QM9 and Molecule3D datasets for chemical property prediction. ViSNet's key contributions include: 1. **Runtime Geometry Calculation (RGC)**: This module implicitly extracts various geometric features (angles, dihedral torsion angles, and improper angles) with linear time complexity, significantly reducing computational overhead. 2. **Vector-Scalar Interactive Message Passing (ViS-MP)**: This mechanism enables efficient interaction between vector and scalar representations, fully utilizing geometric information to enhance the model's geometric representation. 3. **State-of-the-Art Performance**: ViSNet outperforms other algorithms on multiple benchmarks, demonstrating superior accuracy in predicting energy, forces, and other quantum chemical properties. 4. **High-Fidelity MD Simulations**: ViSNet can perform MD simulations with high fidelity, as evidenced by consistent interatomic distance distributions and potential energy surfaces. 5. **Interpretability**: ViSNet provides reasonable interpretability by mapping geometric representations to molecular structures, showing the ability to discriminate different molecular substructures. ViSNet's effectiveness is validated through comprehensive evaluations, including ablation studies and comparisons with state-of-the-art methods. The paper also discusses the computational efficiency and scalability of ViSNet, making it a promising tool for molecular modeling and simulation.The paper introduces ViSNet, an equivariant geometry-enhanced graph neural network designed to improve the representation and prediction of molecular structures and properties. ViSNet addresses the limitations of existing methods by efficiently extracting geometric features and reducing computational costs. The proposed model outperforms state-of-the-art approaches on multiple molecular dynamics (MD) benchmarks, including MD17, revised MD17, and MD22, and achieves excellent performance on QM9 and Molecule3D datasets for chemical property prediction. ViSNet's key contributions include: 1. **Runtime Geometry Calculation (RGC)**: This module implicitly extracts various geometric features (angles, dihedral torsion angles, and improper angles) with linear time complexity, significantly reducing computational overhead. 2. **Vector-Scalar Interactive Message Passing (ViS-MP)**: This mechanism enables efficient interaction between vector and scalar representations, fully utilizing geometric information to enhance the model's geometric representation. 3. **State-of-the-Art Performance**: ViSNet outperforms other algorithms on multiple benchmarks, demonstrating superior accuracy in predicting energy, forces, and other quantum chemical properties. 4. **High-Fidelity MD Simulations**: ViSNet can perform MD simulations with high fidelity, as evidenced by consistent interatomic distance distributions and potential energy surfaces. 5. **Interpretability**: ViSNet provides reasonable interpretability by mapping geometric representations to molecular structures, showing the ability to discriminate different molecular substructures. ViSNet's effectiveness is validated through comprehensive evaluations, including ablation studies and comparisons with state-of-the-art methods. The paper also discusses the computational efficiency and scalability of ViSNet, making it a promising tool for molecular modeling and simulation.