分子动力学Molecular dynamics

原子受力 F = −∇E 由力场（经典 MD）或量子化学计算（从头算 MD，如卡尔-帕里内洛方法）给出。用韦尔莱算法等以约 1–2 fs 的时间步长积分运动方程。通过恒温器、恒压器控制系综（NVE、NVT、NPT），用周期性边界条件模拟宏观体系，用埃瓦尔德求和处理长程静电作用。根据各态历经假设，时间平均等于系综平均，从轨迹可计算径向分布函数、扩散系数、自由能等。

Forces F = −∇E come from a force field (classical MD) or quantum-chemical calculation (ab initio MD, such as the Car–Parrinello method). Equations of motion are integrated with algorithms such as Verlet at time steps of about 1–2 fs. Thermostats and barostats control the ensemble (NVE, NVT, NPT), periodic boundary conditions mimic bulk systems, and Ewald summation handles long-range electrostatics. By the ergodic hypothesis time averages equal ensemble averages, so trajectories yield radial distribution functions, diffusion coefficients, free energies and more.

1957 年阿尔德和温赖特用硬球模型进行了首次分子动力学模拟。1964 年拉赫曼首次模拟了液态氩。1977 年麦卡蒙、格林和卡普拉斯完成了第一个蛋白质（牛胰蛋白酶抑制剂）的分子动力学模拟，时长约 9.2 ps。1985 年卡尔和帕里内洛提出从头算分子动力学。专用超级计算机 Anton（2008 年起）把蛋白质模拟推进到毫秒尺度。

Alder and Wainwright ran the first molecular dynamics simulation with hard spheres in 1957, and Rahman first simulated liquid argon in 1964. In 1977 McCammon, Gelin and Karplus completed the first protein simulation, of bovine pancreatic trypsin inhibitor, for about 9.2 ps. Car and Parrinello introduced ab initio MD in 1985, and the special-purpose supercomputer Anton, from 2008, pushed protein simulations to the millisecond scale.

分子动力学用于研究蛋白质折叠与构象变化、药物与靶点结合自由能、离子通道传输、膜与胶束结构、液体和溶液性质、材料的力学和热学性能、电池电解质中的离子迁移，以及晶体成核和相变。机器学习势能函数正在使接近量子精度的大尺度模拟成为可能。

MD is used to study protein folding and conformational change, drug–target binding free energies, ion-channel transport, membrane and micelle structure, liquids and solutions, mechanical and thermal properties of materials, ion transport in battery electrolytes, and crystal nucleation and phase transitions. Machine-learned potentials are making large-scale simulations with near-quantum accuracy possible.