• No results found

Temperature initialization and rescaling

Ii • (Oi (Oi X / • (j)i = N i (4.21)

4.3.5 Temperature initialization and rescaling

The initial kinetic energy o f an MD simulation is set by sampling the particles’ velocities from the Maxwell-Boltzmann distribution at the required temperature. The effective temperature o f the system is then calculated from the ensemble average o f its kinetic energy. However, the initial configuration is usually far from equilibrium and has a high potential energy. As the simulation proceeds, much o f the potential energy is converted to kinetic energy, thus raising the temperature o f the system.

It is therefore necessary to scale the kinetic energy o f the particles during the equilibration period. This is typically achieved by multiplying the linear and angular velocities at periodic intervals by a factor of:

s = (4,36)

where g is the number o f degrees o f freedom, T is the required temperature and is the instantaneous value o f the kinetic energy. By repeatedly setting the ‘instantaneous’ temperature to the correct value, the kinetic energy is made to approach its required value.

An MD simulation w ith scaling does not generate equilibrium particle trajectories, so the scaling must be switched o ff before any calculations o f thermodynamic averages are performed. The simulation can then proceed along the A, V,E ensemble, where the total number o f particles, the simulation cell volume and the total energy is held constant, or along an A ,F,T ensemble where the temperature, rather than the total energy, is fixed. The latter is achieved by thermostat methods:

the Nosé-Hoover method^^’^^ which couples the system to a heat bath, and the Gaussian thermostat^^ which replaces the Newton-Euler equations by variants in which the kinetic energy is conserved. The N^VJE ensemble was used in the simulations o f liquid ammonia and lithium-ammonia solutions, which are presented in Chapter 6.

4.3.6 Structural and dynamical properties from MD simulations

The MOLDY code enables configurations o f the system to be saved at periodic intervals. Particular information such as particle positions and molecule quaternions may then be extracted from these binary dump files, in order that time-averaged radial distribution functions and the mean square displacements o f different particle types may be calculated.

4.3.6.1 Radial distribution functions

The radial distribution function (RDF) is one o f the most important structural quantities which characterize a liquid system. In the case o f a molecular system, the partial RDF for atoms a and p is defined as:^^

where p is the molecular number density, V is the volume o f the simulation cell, N is the total number o f particles, r is the vector between the centre o f mass o f molecules i and j , and ria,rjp are the vectors jfrom the centres o f mass o f molecules i j to the atomic sites or,/? respectively. The angled brackets denote a spherical average in addition to the usual configurational average.

In the simulation, the RDF is evaluated

where fihis{b) is the accumulated number o f atom pairs per bin, b is the number o f the histogram bin, 5r is the bin width (hence r = 6& ) and r is the number o f timesteps over which the atom pair distances have been counted.

4.3.6.2 Mean square displacements and diffusion coefficients

The mean square displacement vs. time o f a particular species can be calculated from the particle positions extracted fi-om the binary dump files. For a species o f N particles, the mean square displacement is calculated via:

2 1 2

<1 r{t) - r(0) p > = — 2 ^ 2 ^ k „ (< + io ) - ^ (<o ) I (4.39) ^ «=1 tr,

where (t) is the position o f particle n at time t, and Nt is the total number o f timesteps over which the sum is performed for a particular t. The diffusion coefficient o f that species can then be calculated firom the gradient o f a plot o f the mean square displacement against time, using the Einstein relation:

{ \ r { t ) - r { Q ) Ÿ ) = 6Dt (4.40)

The radial distribution functions and diffusion coefficients extracted firom the MD simulations serve as a basis for comparison with experimental data, as detailed in Chapter 6. The experimental data can also be used as a stringent test for the potential functions used in the MD simulations.

4.4 References

[1] R. L. M cGreevy and L. Pusztai, Mol. Sim. 1, 359 (1988).

[2] L. Pusztai and R. L. McGreevy, J. Phys. Cond. Matter 10, 525 (1998).

[3] V. M. Nield, M. A. Howe, R. L. McGreevy, J. Phys. Cond. Matter 3, 7519 (1991).

[4] J. Swenson, R. L. McGreevy, L. Boijesson and J. D. Wicks, Solid State Ionics 105, 55 (1998).

[5] R. L. McGreevy, RMC Manual.

[6] N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller and E. Teller, J. Phys. Chem. 21, 1087 (1953).

[7] J. D. Wicks and R. L. McGreevy, J. Non-Cryst. Solids 193, 23 (1995). [8] A. K. Soper, Chem. Phys. 202, 295 (1996).

9] A. K. Soper, Chem. Phys. 258, 121 (2000). 10] A. K. Soper, Mol. Phys. 99, 1503 (2001).

11] A. K. Soper and D. T. Bowron, EPSR - A U ser’s Guide (2000).

12] M. D. Allen and D. J. Tildesley, Computer Simulation o f Liquids, OUP, Oxford, (1987).

13] C. G. Gray and K. E. Gubbins, Theory o f Molecular Fluids - Volume 1: Fundamentals, Clarendon Press, Oxford, (1984).

14] D. T. Bowron, A. K. Soper and J. L. Finney, J. Chem. Phys. 114, 6203 (2001).

15] K. Refson, Moldy U ser’s Manual, Revision 2.25.2.6fo r release 2.16, (2001). 16] D. Frenkel and B. Smit, Understanding Molecular Simulation - from

Algorithms to Applications, Academic Press (2002).

17] A. R. Leach, Molecular Modelling: Principles and Applications, Prentice Hall, Harlow, (2001).

18] L. Verlet, Phys. Rev. 165, 201 (1967). 19] D. Beeman, J. Comp. Phys. 20, 130 (1976).

20] B. Quentrec and C. Brot, J. Comp. Phys. 13, 430 (1975). 21] W. G. Hoover, Phys. Rev. A. 31, 1695 (1985).

22] S. Nose, Mol. Phys. 52, 255 (1984).

23] J. P. Hansen and I. R. McDonald, Theory o f Simple Liquids, 2"^ ed.. Academic Press, London (1986).

CHAPTER 5

RESULTS I:

Related documents