, from eq. (4.4). First, we rewrite eq. (4.4) as P(t) -
F+(t) + Py(t)
(4.19)
P(t) = O(i12,P*),
where t (4.20)
and y(t) = y for t > Tb. We have written P* explicitly on the right-hand side of eq. (4.19) to indicate that the ordering in 1 is valid when P* is 0( 1). The solution of eq. (4.19) is P(t) = exp(--i2iy(r) X (F+
Again
(T’)
we note that
+
dT} P(0) + / eXP{-A2jty(T) o(A2,
P*)}
eq. (4.21)
dT}
d+. is valid
(4.21) no matter
whether
the bath
is in
P. MAZUR
252
AND
I. OPPENHEIM
equilibrium or not. The averages discussed below, however, are taken assuming the bath to be in equilibrium with respect to the fixed Brownian particle. Taking the average of eq. (4.21) yields
= exp{-A2yt)
P(0) + U(1, P*).
(4.23)
For times t such that Tb I t ( Z(L2y)-1, where I is a small integer, eq. (4.23) reduces to the usual Brownian motion result
= exp{--A2yt}
(4.24)
P(0).
For very long times, t > Z(lsy)-1, however, the last term on the right-hand side of eq. (4.23) may dominate. This term, however, is small compared to (AM”)* = 0 (A-r) w h’ICh is a measure of the equilibrium fluctuations in P. The average of
is obtained from eq. (4.21) as
= exp{-2A2 + idr’ = exp{-221s it) P(0) P(0) + MkT I{ 1 + {e+yt = exp{-21syt)
{y (T) dT) P(0) P(0)
0
I’dT” exp(-As
iy(~)
x
dT -
1s iy(~)
d7)
i Y(T) d+ P (0) + 13 0( 1, P*), (4.25) 0
where we have used eq. (3.14). For t > Tb, eq. (4.25) becomes
e--2A8ytj (4.26)
P(0) + 1) O(1, P*),
where we have used eq. (3.17). For times t such that 76 < t < Z(A2y)-l eq. (4.26) reduces to the usual Brownian motion result
P(0)P(0)+ MkT I{1 - e-2A*yt),
where we have neglected terms of order 1-r compared For very long times, t > Z(Lsy)-1, eq. (4.26) becomes
(4.27)
to terms of order Ad2.
(4.28)
we have neglected terms of 0( 1) compared to terms of 0(A-2). Eq. gives the equilibrium value for
MOLECULAR
5. Friction that
THEORY
OF BROWNIAN
constant for a macroscopic
expression
(4.18)
for the friction
MOTION
253
body. In this section we shall show constant
occurring
in the Langevin
equation describing the stochastic motion of a heavy particle is identical to the friction coefficient, which characterizes the drag on a macroscopic body of mass M moving through the fluid with prescribed slowly varying velocity Ve(t). Consider a system body with position this system is
of N point particles &(t)
and velocity
of mass m in interaction
Ve(t) = A&).
Ro@)) = Bo(rN, PN) + igm
Q’?PN, where fJo(rN,pN)
=
v
with a
The Hamiltonian
- Ro@)I),
for
(5.1)
+utrN)j
U(rN) is the short-range potential of interaction between the point particles which is translationally invariant; and 4(lrr - Ra(t) I) is the short-range potential of interaction between the body and the ith bath particle. We now define a generating function for a time-dependent canonical transformation by N G
=
Z i=l
(Q
-
Roti))
.Pt,
w
where PC is the transformed equations are
i3G
Rc= - api
= rz -
momentum
$
(5.4
vector of particle i. belonging to the new
= Ao(RN, PN) + 2 +(IRtj) i
The new Hamiltonian
i. The transformation
Ro(t),
with Rt the transformed position The transformed Hamiltonian momenta is H = A+
of particle
coordinates
2I Pi* Vo@). i
and
(5.6)
H is of the form
H = Ho(RN, PN) - z Pt. V,(t), i
(5.7)
where Ho(RN, PN) = Bo(RN,
PN) + z +(IRfl)
i
(5.8)
P. MAZUR
254
AND
I. OPPENHEIM
is the Hamiltonian of the N bath particles in an external potential by the fixed body. It is therefore identical with the Hamiltonian
created defined
in eq. (2.3). We can now compute the linear response of any bath dynamical function to the prescribed motion of the body. We shall assume that at t --f - 00, the body was at rest, and the bath in equilibrium distribution function of the bath is given by
f (RN,
PN) = e--BHo/f dRN dPN eePHo.
with it. Thus at t + -00,
the
(5.9)
In particular we are interested in the average value of the force exerted by the bath on the body to terms linear in Vo(t). The force on the body is given by (see eq. (2.6)) (5.10) By standard
methods
we obtain
~@‘Fo(T)>. Vo(t - 7) d7,
(5.11)
0
where the symbol < >t denotes the average value at time t, the symbol < > is defined in eq. (2.8), and Pa(t) is defined by eq. (3.4). If Vo(t) varies sufficiently slowly (5.7) becomes 0%
= -
j3 r
d7. Vo(t)
0 =
-qlPo(t),
(5.12)
where Pa(t) = M Vo(t), and where Asy is given by eq. (4.16). Comparison of eq. (5.12) with eq. (4.15) demonstrates the equivalence of the friction constant occurring in the Langevin equation and the friction constant characterizing the drag on a macroscopic body. This equivalence has also been derived by Lebowitz and RCsiboisa) by a different method. 6. Remarks. The techniques utilized here have enabled us to derive a simple equation of motion for the heavy particle which is valid to order A2 for all times. For t > 70, this equation reduces to the LangeVin equation. We have restricted our considerations to situations in which the magnitude of the reduced momentum, P*, of the Brownian particle is of the same order in A, i.e. La, as the magnitude of the momentum, p, of a bath particle. The Langevin equation, however, is valid for a much larger range of values of P*. In fact, it follows from eqs. (4.1) and (3.3) that the left-hand side of eq. (4.1) is of lower order in I than the right-hand side as long as O(P*) - o(~&p), - 1 < e. If E I - 1, the expansion in eq. (3.3) is not useful.
MOLECULAR
THEORY
OF BROWNIAN
It is possible to extend our considerations
MOTION
255
to treat Brownian
motion in an
external field of force and Brownian motion of particles with internal structure. The equations obtained are the phenomenological equations which have been discussed e.g. by Chandrasekhara) and by Pragerra), respectively. The equations for nonspherical-Brownian particles exhibit a coupling between the various components of the center-of-mass momentum and the angular momentum. Acknowledgement. hospitality extended
The authors over the years.
thank
the reciprocal
institutions
for
APPENDIX
In this appendix, we study the
behavior of the correlation function (2.11). We shall see that the analysis complicated than that presented in (2.77, and
La) = n((P*/m) - FR + F-V,.>,
to obtain
@F(t)> =
(A. 1)
0
where Fe(t) is defined by eq. (3.4). Iteration of eq. (A.l) results in
; n=l
x
7”-1
1%TdTi jld?s . . . j dTn 0
0
0
x eiLO(k-l-d (01 + 02) eiLoTn F), where t has been replaced
(A4
by TO,
and 0s = F-V,,.
(A.9
The P* dependence of eq. (A.2) is explicit and resides completely in the operator 01 and 0s. Because of symmetry and since the factors of P* and Vp. can be moved outside of the average, the terms in eq. (A.2) containing
P. MAZUR
256
AND
I. OPPENHEIM
odd numbers of factors (01 + 0s) must vanish. It is also clear that only those terms with proper ordering of the operators Or and 0s can survive. In particular, each 0s must have at least one more 01 to its right than it has 02’s to its right. Thus, there can be at most rt/2 0s operators in each term of eq. (A.2) and the last factor of 01 + 0s on the right reduces to 01. It is clear that for short times
eih(TJ-l-TJ)
o1
eih(TJ-TJ+l)
the
B(a’)>. (-4.5)
Since we need consider only those correlations in eq. (A.2) which contain an even number of factors 02, i = 1 or 2, there are two cases of eq. (A.5) that will be of interest. In the first case, A contains an even number of factors of 02 and B an odd number of Og. In this case, application of (3.5) leads to ci =
(A (a)) (01 eiLa(rJ-‘j+l) B(d)> [ (B(o’)>
= 0 = 0
73'>
7b
(A.6a)
Tj+l >
Tb.
(A.6b)
Tj-1 -
Tj -
The result for eq. (A.6a) is zero because A contains an odd number of factors of 0~ and F. The result for eq. (A.6b) is zero because
c
= 1
eiLo(7J-1-7f) Ol>(B(o’)>
Tj-1 = 0
ri -
Tj >
Tj+l
>
76
(A.7a)
76.
(A.i’b)
The result for eq. (A.7a) need not be zero since A is odd in 0s and contains one F. The result for eq. (A.7b) is still zero for the same reason that (A.6b) is zero and also because B is even in Og and contains one F. Next, we consider the correlation function C2 = (/I@)
eih(TJ-l-TJ)
o2
&-h(5J-7J+1)
B(a’)>. (A.4
With A even in 02 and B odd, application cs =
<02 eiLa(rJ-rJ+l) B(d)>
of eq. (3.5) leads to
= 0
[
Tj-1 -
75 >
Tj - Tj+l >
76 Tb.
(A.9a) (A.9b)
The result for eq. (A.9a) is zero for the same reason as eq. (A.6a). The result for eq. (A.9b) need not be zero. For A odd in Or and B even, C2 becomes cs =
<02 eiLoCTj-‘j+l)B(d)> [ = 0
TJ-1 - Tj >
rb
(A. 1Oa)
73 - Tj+l >
Tb,
(A. lob)
where (A. 1Oa) need not be zero and (A. lob) is zero since B contains an even of Oa and one F.
number of factors
MOLECULAR
To summarize
THEORY
the results
OF BROWNIAN
of eqs. (A.6)-(A.lO),
times TJ-1, ~5, T~+I mzlst be close together occurs with an even number of factors will be of lower order in TO for TO > Tb each n in eq. (A.2). Because of this and is limited the leading term of order PZin An
fdT1(dT2...y-i+,
o2
we note that
257
the three
for a nonzero result only when Or
of 02 to its left. Thus, these terms than the other cases considered for because the number of factors of 0s eq. (A.2) is of the form eiL~(~~-~a)
O2
O1
0
0
...
MOTION
eiLo(sn-t-~n)
o1
eua+nF),
(A. 11)
with alternating factors of Or and Oz. Here and below we always assume To > Tb. In order to obtain this result we start our considerations on the right. The first 01 must be Or since there are no factors of P* to its right. The second Oa must be 0s or else eqs. (A.6) would apply. The third 0~ must be Or because there are no free P* factors to its right. The argument proceeds in this fashion until the last Oa on the left. The leading contribution to eq. (A.1 1) is obtained by allowing the largest possible lattitude to the time variables consistent with retaining a nonzero integrand. For most of the range of integration, the integrand of eq. (A. 11) breaks up, according to eq. (3.5), into
eiL~(~e4-~~-~l
02)
03) ~01
. . .
03)
eiLOra F>.
(A. 12)
Substitution (2.6) yields
of eqs. (A.3) and (A.4) into eq. (A. 12) and use of eqs. (2.15) and
(-
- 73)) . . .
p/m)*n
-
Tn-1))’
(A.13)
At this point we use the fact that, for reasons of symmetry, (A. 14)
The dominant
contribution
to the nth term in
(n-W2
x{
,po.
723+1)>)
(%)>
1,
where we recall that n is even. The dominant contribution easily obtained using eq. (3.5) and the fact that To > Tb.
(A. 15) to eq. (A. 15) is We use the fact
258
MOLECULAR
THEORY
OF BROWNIAN
MOTION
that oj
=~(F.F,,(O))
do 3 01,
T 2 TZ,(see (3.5)). The time integrations
(A. 16) in (A. 15) result in (A. 17)
for n = 2 and (A.18) for n > 2. Substitution
of (A.17) and (A.18) into eq. (A.2) yields
I = (-yn)P)k IX k=l k!
I1 + ‘($)}-jJ
(A. 19)
where we have used eq. (A.14) and y is defined by P y =3mOL. The terms of ~(Q/To) in eq. (A.19) are a proper estimate of all the approximations made in proceeding from eq. (A.2) to eq. (A.19). The summation of the leading terms in eq. (A. 19) can be easily performed to yield
TO)k
I -
(ml/?) y2 I2 I exp{--yT0L2} (A.21)
A4 Tb),
where k is some arbitrary
integer.
REFERENCES
1) Lebowitz, J. and Rubin, E., Phys. Rev. 131 (1963) 2381.
2) 3) 4) 5) 6) 7) 8) 9) 10)
Resibois, P. and Davis, R., Physica 30 (1964) 1077. Lebowitz, J. and Resibois, P., Phys. Rev. 139A (1965) 1101. Mori, H., Progr. theor. Phys. 33 (1965) 423. Kubo, R., Repts. Progr. Phys, 24 (1966) part 1 255. Zwanzig, R., in Lectures in Theoretical Physics, edited by W. E. Brittin, B. W. Downs and J. Downs, Interscience Publ., Inc. (New York, 1961) vol. III. 106. Mazur, P. and Oppenheim, I., J. Phys. Sot. Japan 26 (1969) suppl. 35. Terwiel, R. and Mazur, P., Physica 32 (1966) 1813; 36 (1967) 289; Resibois, P., Brocas, J. and Decan, G., J. math. Phys. 10 (1969) 964. Chandrasekhar, S., Rev. mod. Phys. 15 (1943) 1. Prager, S., J. them. Phys. 23 (1955) 2404.