478 The Feynman path integral
coordinate transformation capable of uncoupling the harmonic term in eqn. (12.6.3),
then we can replace the single fictitious massm′ in the classical Hamiltonian with
a set of massesm′ 1 ,...,m′P such that only one harmonic frequency remains. Finally,
we can then adjust the time step for stable integration of motion having that charac-
teristic frequency and/or employ RESPA to the problem. This will ensure adequate
sampling of all modes of the cyclic chain. In fact, we have already seen an example
of such a transformation in Section 4.5. Eqns. (4.5.38) and (4.5.39) illustrate how a
simple transformation uncouples the harmonic term for a model polymer that, when
made cyclic, is identical to the discrete path integral. The one-dimensional analog of
this transformation appropriate for eqn. (12.6.2) is
u 1 =x 1
uk=xk−
(k−1)xk+1+x 1
k
, k= 2,...,P, (12.6.5)
the inverse of which is
x 1 =u 1
xk=uk+
k− 1
k
xk+1+
1
k
u 1 , k= 2,...,P. (12.6.6)
Note that, as in eqn. (4.5.39), eqn. (12.6.6) is defined recursively. Becausex 1 =u 1 ,
the recursion can be seeded by starting with thek=Pterm and working backwards
tok= 2. The inverse can also be expressed in closed form as
x 1 =u 1
xk=u 1 +
∑P
l=k
k− 1
l− 1
ul, k= 2,...,P. (12.6.7)
The transformation defined in eqns. (12.6.5), (12.6.6), and (12.6.7)is known as a
staging transformationbecause of its connection to a particular path-integral Monte
Carlo algorithm (Ceperley and Pollock, 1984), which we will discuss in the next section.
The staging transformation was first introduced for path-integral molecular dynamics
by Tuckermanet al.(1993). The variablesu 1 ,...,uPare known asstaging variables, as
distinguished from the original variables, which are referred to asprimitive variables.
We now proceed to develop a molecular dynamics scheme in terms of the staging
variables. When the harmonic coupling term is evaluated using these variables, the
result is
∑P
k=1
(xk−xk+1)^2 =
∑P
k=2
k
k− 1
u^2 k, (12.6.8)
which is completely separable. Since the Jacobian of the transformation is 1, as we
showed in eqn. (4.5.45), the discrete partition function becomes
QP(L,T) =
∫
dp 1 ···dpP
∫
du 1 ···duP