Abstract: Virtual surgery has become an important interdisciplinary field integrating medicine and computer science, supporting intelligent healthcare. Its goal is to reproduce surgical procedures in a safe virtual environment to improve training, optimize planning, and assist intraoperative decision-making. Physical simulation plays a key role, governing tissue deformation, instrument interaction, and haptic feedback, and directly affecting fidelity and real-time performance. This paper categorizes mainstream methods into four families: mesh-based continuum, constraint-based variational projection, particle/hybrid, and data-driven approaches. It analyzes their characteristics through the same challenges, including topology modification, contact and friction, multiphysics coupling, and multi-rate interaction. A task-oriented framework maps these methods to preoperative planning, real-time training, and intraoperative navigation, comparing their performance in accuracy, efficiency, and stability. Finally, limitations and future directions are outlined, including physics-data integration, multi-rate architectures, and standardized clinical validation.