Numerical evaluation 479
×exp
{
−β
∑P
k=1
[
p^2 k
2 m′k
+
1
2
mkω^2 Pu^2 k+
1
P
U(xk(u))
]}
. (12.6.9)
In eqn. (12.6.9), the parametersmkare defined to be
m 1 = 0, mk=
k
k− 1
m, k= 2,...,P, (12.6.10)
andm′ 1 =m,m′k=mk. The notationxk(u) indicates the inverse transformation in
eqn. (12.6.6) or (12.6.7). In order to evaluate eqn. (12.6.9), we canemploy a classical
Hamiltonian of the form
H ̃cl(u,p) =
∑P
k=1
[
p^2 k
2 m′k
+
1
2
mkω^2 Pu^2 k+
1
P
U(xk(u))
]
, (12.6.11)
which leads to the equations of motion
u ̇k=
pk
m′k
p ̇k=−mkω^2 Puk−
1
P
∂U
∂uk
. (12.6.12)
From the chain rule, the forces on the staging variables can be expressed recursively
as
1
P
∂U
∂u 1
=
1
P
∑P
l=1
∂U
∂xl
1
P
∂U
∂uk
=
1
P
[
∂U
∂xk
+
k− 2
k− 1
∂U
∂uk− 1
]
. (12.6.13)
The recursive staging force calculation is performed starting withk= 2 and using
the first expression for∂U/∂u 1. Eqns. (12.6.12) need to be thermostatted to ensure
that the canonical distribution is generated. The presence of thehigh-frequency force
on each staging variable combined with the 1/Pfactor that attenuates the potential-
energy derivatives leads to a weak coupling between these two forces. Therefore, it is
important to have as much thermalization as possible in order to achieve equiparti-
tioning of the energy. It is, therefore, strongly recommended (Tuckermanet al., 1993)
that path-integral molecular dynamics calculations be carried out using the “massive”
thermostatting mechanism described in Section 4.10. This protocolrequires that a
separate thermostat be attached to each Cartesian componentof every staging vari-
able. Thus, for the single-particle one-dimensional system described by eqns. (12.6.12),
if Nos ́e–Hoover chain thermostats of lengthMare employed, the actual equations of
motion would be