310 Monte Carlo
A
B
.yj∆t
.
xj∆t
.
xj∆t
Accepted
Rejected
Fig. 7.7The shooting algorithm. The original path Y(T) is shown as the solid line. A point
yj∆trandomly chosen from this path is used to determine the shooting point xj∆t. Two
example shooting paths are shown as dashed lines. The long dashed line is a successful path
that is accepted, while the short dashed line shows an unsuccessful path that is rejected.
with time step ∆tbut with velocities reversed at the shooting point. Combining eqn.
(7.7.11) with (7.7.10), we obtain for the acceptance probability
Λ[X(T)|Y(T)] =hA(x 0 )hB(xn∆t)min
[
1 ,
f(x 0 )
f(y 0 )
(n− 1
∏
k=0
T(x(k+1)∆t|xk∆t)
T(y(k+1)∆t|yk∆t)
)
×
τ(yj∆t|xj∆t)
τ(xj∆t|yj∆t)
n∏− 1
k=j
T(y(k+1)∆t|yk∆t)
T(x(k+1)∆t|xk∆t)
j∏− 1
k=0
T(yk∆t|y(k+1)∆t)
T(xk∆t|x(k+1)∆t)
=hA(x 0 )hB(xn∆t)
×min
[
1 ,
f(x 0 )
f(y 0 )
τ(yj∆t|xj∆t)
τ(xj∆t|yj∆t)
j∏− 1
k=0
T(yk∆t|y(k+1)∆t)T(x(k+1)∆t|xk∆t)
T(xk∆t|x(k+1)∆t)T(y(k+1)∆t|yk∆t)
]
. (7.7.12)
Although eqn. (7.7.12) might seem rather involved, consider what happens when the
trajectories are generated by molecular dynamics, with the trial probability given by
eqn. (7.7.2). Since a symmetric Trotter factorization of the classical propagator is time
reversible, as discussed in Section 3.10, the ratioT(yk∆t|ry(k+1)∆t)/T(y(k+1)∆t|yk∆tis
unity, as is the ratioT(x(k+1)∆t|xk∆t)/T(xk∆t|x(k+1)∆t), and the acceptance criterion
simplifies to
Λ[X(T)|Y(T)] =hA(x 0 )hB(xn∆t)min
[
1 ,
f(x 0 )
f(y 0 )
τ(yj∆t|xj∆t)
τ(xj∆t|yj∆t)
]
. (7.7.13)
Finally, suppose the new shooting point xj∆tis generated from the old point yj∆t
using the following rule;
xj∆t= yj∆t+ ∆, (7.7.14)