CASINO manual - Theory of Condensed Matter
CASINO manual - Theory of Condensed Matter
CASINO manual - Theory of Condensed Matter
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
13.2 The ensemble <strong>of</strong> configurations<br />
The f distribution is represented by an ensemble <strong>of</strong> electron configurations, which are propagated<br />
according to rules derived from the Green’s function <strong>of</strong> Eq. (17). G D represents a drift-diffusion<br />
process while G B represents a branching process. The branching process leads to fluctuations in the<br />
population <strong>of</strong> configurations and/or fluctuations in their weights.<br />
We will introduce labels for the different configurations α present at each time step m. From now<br />
on R represents the electron coordinates <strong>of</strong> a particular configuration in the ensemble and i labels a<br />
particular electron.<br />
In the casino implementation <strong>of</strong> DMC, electrons can moved one at a time (electron-by-electron,<br />
dmc method= 1, default) or all at once (configuration-by-configuration, dmc method= 2). The<br />
former is much more efficient and is the standard procedure for large systems, but most <strong>of</strong> the algorithms<br />
described in the literature are for moving all electrons at once.<br />
13.3 Drift and diffusion<br />
We now discuss the practical implementation <strong>of</strong> the drift-diffusion process using the electron-byelectron<br />
algorithm in which electrons are moved one at a time, as used in casino.<br />
To implement the drift-diffusion step, each electron i in each configuration α is moved from r ′ i (α) to<br />
r i (α) in turn according to<br />
r i = r ′ i + χ + τv i (r 1 , . . . , r i−1 , r ′ i, . . . , r ′ N), (22)<br />
where χ is a three-dimensional vector <strong>of</strong> normally distributed numbers with variance τ and zero mean.<br />
v i (R) denotes those components <strong>of</strong> the total drift vector V(R) due to electron i.<br />
Hence each electron i is moved from r ′ i to r i with a transition probability density <strong>of</strong><br />
(<br />
t i (r 1 , . . . , r i−1 , r i ← r ′ i, r ′ i+1, . . . , r ′ 1<br />
N) =<br />
(2πτ) exp (ri − r ′ i − τv i(r 1 , . . . , r i−1 , r ′ i , . . . , )<br />
r′ N ))2 .<br />
3/2 2τ<br />
(23)<br />
For a complete sweep through the set <strong>of</strong> electrons, the transition probability density for a move from<br />
R ′ = (r ′ 1, . . . , r ′ N ) to R = (r 1, . . . , r N ) is simply the probability that each electron i moves from r ′ i to<br />
r i . So the transition probability density for the configuration move is<br />
T (R ← R ′ ) =<br />
N∏<br />
t i (r 1 , . . . , r i−1 , r i ← r ′ i, r ′ i+1, . . . , r ′ N ). (24)<br />
i=1<br />
In the limit <strong>of</strong> small time steps, the drift velocity V is constant over the (small) configuration move.<br />
Evaluating the product in this case, we find that the transition probability density is<br />
T (R ← R ′ ) = G D (R ← R ′ , τ), (25)<br />
so that the drift-diffusion process is described by the drift-diffusion Green’s function G D .<br />
At finite time steps, however, the approximation that the drift velocity is constant leads to the violation<br />
<strong>of</strong> the detailed balance condition.<br />
We may enforce the detailed balance condition on the DMC Green’s function by means <strong>of</strong> a Metropolisstyle<br />
accept/reject step introduced by Ceperley et al. [27]. This has been shown to greatly reduce<br />
time-step errors [28]. The move <strong>of</strong> the ith electron <strong>of</strong> a configuration is accepted with probability<br />
{<br />
min 1, t i(r 1 , . . . , r i−1 , r ′ i ← r i, r ′ i+1 , . . . , r′ N )Ψ2 (r 1 , . . . , r i , r ′ i+1 , . . . , r′ N )<br />
t i (r 1 , . . . , r i−1 , r i ← r ′ i , r′ i+1 , . . . , r′ N )Ψ2 (r 1 , . . . , r i−1 , r ′ i , . . . , r′ N ) }<br />
{ [ (<br />
= min 1, exp r ′ i − r i + τ )<br />
2 (v i(r 1 , . . . , r i−1 , r ′ i, . . . , r ′ N ) − v i (r 1 , . . . , r i , r ′ i+1, . . . , r ′ N )<br />
· (v<br />
i (r 1 , . . . , r i−1 , r ′ i, . . . , r ′ N ) + v i (r 1 , . . . , r i , r ′ i+1, . . . , r ′ N ) ) ]<br />
× Ψ2 (r 1 , . . . , r i , r ′ i+1 , . . . , r′ N ) }<br />
Ψ 2 (r 1 , . . . , r i−1 , r ′ i , . . . , r′ N )<br />
≡ a i (r 1 , . . . , r i−1 , r i ← r ′ i, r ′ i+1, . . . , r ′ N). (26)<br />
124