482 The Feynman path integral
(1/P)
∑P
k=1∂U/∂xk. It can be shown that, for the staging transformation, the force
on the modeu 1 is also the average force; however this mode is not physically the same
as the centroid.
In a seminal paper by Feynman and Kleinert (1986), it was shown that the path
centroid could be used to capture approximate quantum effects in asystem. Consider
eqn. (12.6.9) with the variablesu 1 ,...,uPrepresenting the normal modes. If we inte-
grate over the variablesu 2 ,...,uPand the corresponding momentap 1 ,...,pP, then the
result can, in the spirit of Section 8.10, be written as
QP(L,T)∝
∫
dp 1 du 1 exp
{
−β
[
p^21
2 m 1
+W(u 1 )
]}
, (12.6.22)
whereW(u 1 ) is the potential of mean force on the centroid given by
W(u 1 ) =−kTln
{∫
du 2 ···duP
×exp
[
−β
∑P
k=1
(
1
2
mkω^2 Pu^2 k+
1
P
U(xk(u))
)]}
(12.6.23)
up to an additive constant. Although we cannot determineW(u 1 ) for an arbitrary
potentialU(x), Feynman and Kleinert were able to derive an analytical expressionfor
W(u 1 ) for a harmonic oscillator using the functional integral techniquesdiscussed in
Section 12.4. In particular, they showed how to derive the parameters of a harmonic
oscillator potentialUho(x;u 1 ) that minimize the expectation value〈U(x)−Uho(x;u 1 )〉.
The parameters of the potential depend on the position of the centroid so that the
harmonic potential takes the general formUho(x;u 1 ) = (1/2)Ω^2 (u 1 )(x−u 1 )^2 +L(u 1 ).
That is, the frequency and vertical shift depend on the centroidu 1. The optimization
procedure leads to a simple potential functionW ̃(u 1 ) of the centroid that can then be
used in eqn. (12.6.22) to obtain approximate quantum equilibrium and thermodynamic
properties.
We close this section by showing how the path-integral molecular dynamics pro-
tocol extends toN Boltzmann particles inddimensions. Since most path-integral
calculations fall into this category, we will limit our discussion to these. Excellent
descriptions of path-integral algorithms for bosons and fermionscan be found in the
literature (Ceperley, 1995; Miura and Okazaki, 2000). The discreteN-particle partition
function for Boltzmann particles follows directly from eqn. (12.5.10). After introduc-
ingdNPmomentum integrations as in eqn. (12.6.2), the discrete partition function
becomes
QP(N,V,T) =
∏N
i=1
(
miP
2 πβ ̄h^2
)dP/ 2 ∫ ∏N
i=1
dr(1)i ···dr(iP)dp(1)i ···dp(iP) (12.6.24)
×exp
{
−β
∑P
k=1
[N
∑
i=1
p(k)
2
i
2 m′i
+
∑N
i=1
1
2
miω^2 P
(
r(ik+1)−r(ik)
) 2
+
1
P
U
(
r( 1 k),...,r(Nk)