19.02.2013 Views

Reviews in Computational Chemistry Volume 18

Reviews in Computational Chemistry Volume 18

Reviews in Computational Chemistry Volume 18

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

The polarizable po<strong>in</strong>t dipole models have been used <strong>in</strong> molecular<br />

dynamics (MD) simulations s<strong>in</strong>ce the 1970s. 48 For these simulations, the<br />

forces, or spatial derivatives of the potential, are needed. From Eq. [<strong>18</strong>], the<br />

force 23 on atomic site k is<br />

F k ¼ r kU <strong>in</strong>d ¼ XN<br />

i ¼ 1<br />

l ir kE 0 i<br />

X<br />

þ lk ðrkTkiÞ li ½20Š<br />

All contributions to the forces from terms <strong>in</strong>volv<strong>in</strong>g derivatives with respect to<br />

the dipoles are zero from the extremum condition of Eq. [19]. 23,49<br />

F<strong>in</strong>d<strong>in</strong>g the <strong>in</strong>ducible dipoles requires a self-consistent method, because<br />

the field that each dipole feels depends on all of the other <strong>in</strong>duced dipoles.<br />

There exist three methods for determ<strong>in</strong><strong>in</strong>g the dipoles: matrix <strong>in</strong>version, iterative<br />

methods, and predictive methods. We describe each of these <strong>in</strong> turn.<br />

The dipoles are coupled through the matrix equation,<br />

A l ¼ E 0<br />

Polarizable Po<strong>in</strong>t Dipoles 97<br />

where the diagonal elements of the matrix, Aii, are a 1<br />

i and the off-diagonal<br />

elements Aij are Tij. For a system with N dipoles, solv<strong>in</strong>g for each of them<br />

<strong>in</strong>volves <strong>in</strong>vert<strong>in</strong>g the N N matrix, A—an OðN3Þ operation that is typically<br />

too computationally expensive to perform at each step of an OðNÞ or OðN2Þ simulation. Consequently, this method has been used only rarely. 31 Note that<br />

s<strong>in</strong>ce Eq. [21] for l is l<strong>in</strong>ear, there is only one solution for the dipoles.<br />

In the iterative method, an <strong>in</strong>itial guess for the field is made by, for example,<br />

just us<strong>in</strong>g the static field, E0 , or by us<strong>in</strong>g the dipoles from the previous<br />

time step of the MD simulation. 48,49 The dipole moments result<strong>in</strong>g from this<br />

field are evaluated us<strong>in</strong>g Eq. [3], which can be iterated to self-consistency.<br />

Typical convergence limits on the dipoles range from 1 10 2 D to<br />

1 10 6 D. 21,27,34–36,50 Long simulations require very strict convergence limits<br />

or mild thermostatt<strong>in</strong>g 50 to prevent problems due to poor energy conservation.<br />

Alternatively, the energy Upol can be monitored for convergence. 19,51 The<br />

level of convergence, and therefore the number of iterations required, varies<br />

considerably. Between 2 and 10 iterations are typically required. For some calculations,<br />

<strong>in</strong>clud<strong>in</strong>g free energy calculations, a high level of convergence may<br />

be necessary. 38 The iterative method is the most common method for f<strong>in</strong>d<strong>in</strong>g<br />

the dipoles.<br />

The predictive methods determ<strong>in</strong>e l for the next time step based on<br />

<strong>in</strong>formation from previous time steps. Ahlström et al. 23 used a first-order predictor<br />

algorithm, which uses the l values from the two previous times steps to<br />

predict l at the next time step,<br />

i 6¼ k<br />

½21Š<br />

l iðtÞ ¼2l iðt tÞ l iðt 2 tÞ ½22Š

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!