352 Free energy calculations
the reaction coordinates which, for finiteM, leads to a sum of force terms at different
times ∆t/M. This sum is in the form of a trapezoidal rule for a numerical integration
in time. Thus, when the limitM→ ∞is taken, these sums become continuous time
integrals.
Physically, eqn. (8.10.13) tells us that the force driving the slow reaction coordi-
nates is a time average over the motion of the 3N−nadiabatically decoupled fast
variables. If thenmasses assigned to the reaction coordinates are very large, the
remaining variables will follow the slow reaction coordinates approximately instanta-
neously and sample large regions of their phase space at roughly fixed values of the
reaction coordinates. In this limit, the time integrals in eqn. (8.10.13)can be replaced
by configuration-space integrals, assuming that the motion of thefast variables is
ergodic:
2
∆t
∫τ+∆t/ 2
τ
dt Fα[X,Yadb(Y(τ),Y ̇(τ),ΓY(τ),X;t)]
=
∫
dY Fα(X,Y)e−βV ̃(X,Y)
∫
dYe−βV ̃(X,Y)
=
∂
∂qα
1
β
lnZY(q 1 ,...,qn;β). (8.10.14)
Here
ZY(q 1 ,...,qn;β) =ZY(X;β) =
∫
dYe−β
V ̃(X,Y)
(8.10.15)
is the configurational partition function at fixed values of the reaction coordinatesX=
(q 1 ,...,qn). Eqn. (8.10.14) defines an effective potential, the potential of mean force,
on which the reaction coordinates move. Thus, we can define an effective Hamiltonian
for the reaction coordinates as
Heff(X,PX) =
∑n
α=1
p^2 α
2 m′α
−
1
β
lnZY(q 1 ,...,qn;β). (8.10.16)
Since we assume the dynamics to be adiabatically decoupled, thermostats applied to
this Hamiltonian yield the canonical distribution ofHeff(X,PX) at temperatureTq:
Padb(X) =Cn
[∫
dnpexp
{
−βq
∑n
α=1
p^2 α
2 m′α
}]
×exp
{
−βq
(
−
1
β
lnZY(q 1 ,...,qn)
)}
. (8.10.17)
From eqn. (8.10.17), we see that
Padb(X)∝[ZY(q 1 ,...,qn)]βq/β. (8.10.18)