Grossman Jeffrey C, Mitas Lubos
Lawrence Livermore National Laboratory, 7000 East Avenue L-415, Livermore, California 94550, USA.
Phys Rev Lett. 2005 Feb 11;94(5):056403. doi: 10.1103/PhysRevLett.94.056403. Epub 2005 Feb 10.
A method is presented to treat electrons within the many-body quantum Monte Carlo (QMC) approach "on-the-fly" throughout a molecular dynamics (MD) simulation. Our approach leverages the large (10-100) ratio of the QMC electron to MD ion motion to couple the stochastic, imaginary-time electronic and real-time ionic trajectories. This continuous evolution of the QMC electrons results in highly accurate total energies for the full dynamical trajectory at a fraction of the cost of conventional, discrete sampling. We show that this can be achieved efficiently for both ground and excited states with only a modest overhead to an ab initio MD method. The accuracy of this dynamical QMC approach is demonstrated for a variety of systems, phases, and properties, including optical gaps of hot silicon quantum dots, dissociation energy of a single water molecule, and heat of vaporization of liquid water.
本文提出了一种方法,可在分子动力学(MD)模拟过程中,通过多体量子蒙特卡罗(QMC)方法“即时”处理电子。我们的方法利用了QMC电子运动与MD离子运动之间较大的(10 - 100)比率,来耦合随机的虚时电子轨迹和实时离子轨迹。QMC电子的这种连续演化,能以传统离散采样成本的一小部分,为完整的动力学轨迹提供高精度的总能量。我们表明,对于基态和激发态,仅需在从头算MD方法上增加适度的开销,就能高效实现这一点。这种动力学QMC方法的准确性,在各种系统、相和性质上都得到了证明,包括热硅量子点的光学带隙、单个水分子的解离能以及液态水的汽化热。