Isokinetic ensemble 205
a single kinetic-energy constraint based on the virial theorem, since the number of
degrees of freedom isdN−1 rather thandN.
A simple yet effective integrator for the isokinetic equations can be obtained by
applying the Liouville operator approach. As usual, we begin by writingthe total
Liouville operator
iL=
∑N
i=1
[
pi
mi
·∇ri+
(
Fi−
[∑N
j=1(Fj·pj)/mj
2 K
]
pi
)
·∇pi
]
(4.12.14)
as the sum of two contributionsiL=iL 1 +iL 2 , where
iL 1 =
∑N
i=1
pi
mi
·∇ri
iL 2 =
∑N
i=1
(
Fi−
[∑N
j=1(Fj·pj)/mj
2 K
]
pi
)
·∇pi. (4.12.15)
The approximate evolution of an isokinetic system over a time ∆tis obtained by acting
with a Trotter factorized operator exp(iL∆t) = exp(iL 2 ∆t/2) exp(iL 1 ∆t) exp(iL 2 ∆t/2)
on an initial condition{p(0),r(0)}. The action of each of the operators in this factor-
ization can be evaluated analytically (Zhang, 1997; Minaryet al., 2003). The action
of exp(iL 2 ∆t/2) can be determined by first solving the coupled first-order differential
equations
dpi,α
dt
=Fi,α−
[∑N
j=1(Fj·pj)/mj
2 K
]
pi,α
=Fi,α−h ̇(t)pi,α (4.12.16)
withr 1 ,...,rN(and henceFi,α) held fixed. Here, we explicitly index both the spatial
components (α= 1,...,d) and particle numbersi = 1,...,N. The solution to eqn.
(4.12.16) can be expressed as
pi,α(t) =
pi,α(0) +Fi,αs(t)
s ̇(t)
, (4.12.17)
wheres(t) is a general integrating factor:
s(t) =
∫t
0
dt′exp[h(t′)]. (4.12.18)
By substituting into the time derivative of the constraint condition
∑N
i=1pi·p ̇i/mi= 0,
we find thats(t) satisfies a differential equation of the form