28.02.2013 Views

Handbook of Solvents - George Wypych - ChemTech - Ventech!

Handbook of Solvents - George Wypych - ChemTech - Ventech!

Handbook of Solvents - George Wypych - ChemTech - Ventech!

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

470 Jacopo Tomasi, Benedetta Mennucci, Chiara Cappelli<br />

three-body effects can be partially included in a pair-wise approximation, leading it to give<br />

a good description <strong>of</strong> liquid properties. This can be achieved by defining an “effective” pair<br />

potential, able to represent all the many-body effects. To do this, eq. [8.86] can be re-written<br />

as:<br />

eff<br />

∑ 1() i ∑∑<br />

2 ( ij )<br />

[8.88]<br />

V ≈ v r + v r<br />

i<br />

i<br />

j> i<br />

The most widely used effective potential in computer simulations is the simple<br />

Lennard-Jones 12-6 potential (see eq. [8.70]), but other possibilities are also available; see<br />

Section 8.6 for discussion on these choices.<br />

8.7.2.1 Car-Parrinello direct QM simulation<br />

Until now we have focused the attention to the most usual way <strong>of</strong> determining a potential interaction<br />

to be used in simulations. It is worth mentioning a different approach to the problem,<br />

in which the distribution <strong>of</strong> electrons is not treated by means <strong>of</strong> an “effective”<br />

interaction potential, but is treated ab initio by density functional theory (DFT). The most<br />

popular method is the Car-Parrinello (CP) approach, 68,69 in which the electronic degrees <strong>of</strong><br />

freedom are explicitly included in the description and the electrons (to which a fictitious<br />

mass is assigned) are allowed to relax during the course <strong>of</strong> the simulation by a process called<br />

“simulated-annealing”. In that way, any division <strong>of</strong> V into pair-wise and higher terms is<br />

avoided and the interatomic forces are generated in a consistent and accurate way as the<br />

simulation proceeds. This point constitutes the main difference between a CP simulation<br />

and a conventional MD simulation, which is preceded by the determination <strong>of</strong> the potential<br />

and in which the process leading to the potential is completely separated from the actual<br />

simulation. The forces are computed using electronic structure calculations based on DFT,<br />

so that the interatomic potential is parameter-free and derived from first principles, with no<br />

experimental input.<br />

Let us consider a system for which the BO approximation holds and for which the motion<br />

<strong>of</strong> the nuclei can be described by classical mechanics. The interaction potential is given<br />

by:<br />

()<br />

VR = ψ | Hψ<br />

� | [8.89]<br />

0 0<br />

where H is the Hamiltonian <strong>of</strong> the system at fixed R positions and ψ 0 is the corresponding<br />

instantaneous ground state. Eq. [8.89] permits to define the interaction potential from first<br />

principles.<br />

In order to use eq. [8.89] in a MD simulation, calculations <strong>of</strong> ψ 0 for a number <strong>of</strong> configurations<br />

<strong>of</strong> the order <strong>of</strong> 10 4 are needed. Obviously this is computationally very demanding,<br />

so that the use <strong>of</strong> certain very accurate QM methods (for example, the configuration<br />

interaction (CI)) is precluded. A practical alternative is the use <strong>of</strong> DFT. Following Kohn and<br />

Sham, 70 the electron density ρ(r) can be written in terms <strong>of</strong> occupied single-particle<br />

orthonormal orbitals:<br />

() r =∑ i()<br />

r<br />

ρ ψ<br />

i<br />

2<br />

[8.90]

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

Saved successfully!

Ooh no, something went wrong!