Solvent effect on activated rate processes: On the validity of the GLE approach
Chemical Physics 152 ( 199 1) 153- I67 North-Holland
Solvent effect on activated rate processes: on the validity of the GLE approach * Gilles Tarjus ...
Chemical Physics 152 ( 199 1) 153- I67 North-Holland
Solvent effect on activated rate processes: on the validity of the GLE approach * Gilles Tarjus ’ and Daniel Kivelson of Chemistry and Biochemistry University of CaliJbmia. Los Angeles, CA 90024, USA
Department
Received 10 September 1990
The solvent effect on the prefactor of activated rate processes is usually described in terms of a Langevin equation or a generalized Langevin equation (GLE). The validity of such GLEs in the presence of a strong “external” potential (activation energy barrier) has been verified only for unphysical limiting models. Starting from a microscopic description of the system, we obtain a set of conditions under which a GLE is valid. For effective particles moving along a reaction coordinate, A’, this GLE is applicable to those particles which are initially at fixed position, Xc, and a different GLE with a different friction, C”(Xo, t), is required for particles in different initial positions. Thus, though a GLE may be suitable for describing the reactive motion over a restricted region (piecewise validity), it is of doubtful validity when applied to the whole reaction path.
to the solvent. The friction and random force are related by a fluctuation-dissipation relation
1. Introduction
The dynamic effect of the solvent on activated rate processes in solution was first addressed and consistently treated by Kramers [ I]. In Kramers’ approach, the reaction system is idealized as an effective Brownian particle moving across a one-dimensional potential barrier, V(X), where X represents a reaction coordinate. The influence of the solvent on the reaction dynamics is described stochastically through a Fokker-Planck equation for the distribution of position X and velocity k of the Brownian particle or, equivalently, by a Langevin equation: Mji(t)=-~~(X(t))-M~~(t)+R(t))
(1)
where A4is the mass of the effective Brownian particle, the “external force” ( - V,) is given by -V,(X)=-av(x)/ax,
(2)
and [and R ( t ) are the friction constant and the random force, respectively, associated with the coupling * ’
Supported in part by the National Science Foundation and by the Centre National de Recherches Scientitiques. Permanent address: Laboratoire de Physique Thkorique des Liquids, Universitt Pierre et Marie Curie, 4 Place Jussieu, 7552 Paris Cedex 05, France
030 I-O IO4/9 1 /$03.50
=2W-’
d(t, -tz) f,
(3)
where j?= (k,T) -’ and ( ) indicates an equilibrium ensemble average. The random force satisfies the relations =O ,
(R(t)& (R(f))=O,
(4) (5)
and is, moreover, assumed to be a Gaussian process. Using this description, Kramers was able to find the steady-state escape rate out of the potential well. Kramers’ theory is Markovian since it assumes that the characteristic time scale of the solvent fluctuations is very short compared to the time scale of the Brownian particle motion. Arguing that this latter assumption breaks down for the rapid motion of the effective particle near the top of the barrier, Grote and Hynes [ 2 ] proposed a generalization of Kramers’ theory. Their description starts from a generalized Langevin equation ( GLE ) :
G. Tarjus. D. Kivelson / Validity of the GLE approach
154
M/Y(t)= - v,(x(t)) -A4
I
dt, [(t-t,)k(t,)+R(t),
(6)
0
ict,-b)
(7)
3
in which the random force satisfies eqs. (4,5 ) and is assumed to be a Gaussian process. In the GroteHynes treatment only the motion in the vicinity of the barrier top, X=0, is relevant, and the potential V(X) is approximated by V(X) = Vo- jMf.$P
)
(8)
where W, is a frequency related to the curvature of the top of the potential barrier. Combining eqs. (48) with the Gaussian property of R(t), Grote and Hynes derived an expression for the barrier crossing rate constant; the transmission coefficient characterizing deviations from transition state theory is predicted to be AX/o,, where I, is the divergent solution of the GLE for the region near the barrier peak and is given by ~X=&(~X+e(~X))>o
>
(9)
where [(A,) is the Laplace transform of c( ). See also refs. [ 3-8 1. Other authors [ 9- 15 ] have used the GLE approach to calculate the rate constants without restricting its validity to the vicinity of the barrier top, a procedure which, as we shall discuss, we find questionable. Though the description of the motion of the effective particle along the reaction path in terms of a GLE seems reasonable, it is not rigorous, and one encounters difficulties in its use. First of all, it has not been generally proven that in the presence of an arbitrary potential V(X), which we call the external potential, eqs. (6,7) can hold simultaneously with eqs. (4,5). Secondly, the random force R ( t ) is usually not wellspecified and may itself be dependent upon the dynamics of the effective particle in the external field; therefore, in using a GLE one must make additional assumptions of questionable validity, such as the Gaussian nature of R ( t ) and a model description of the friction c(t) in terms of the properties of the liquid bath. So before considering an extension of the functional form of [(X, t), we should focus on the basic applicability of a GLE. The derivation of a Langevin or generalized Lant
gevin equation from exact microscopic equations is a problem that has received much attention [ 16-27 1. Here we reexamine the question, focusing on activated rate processes in solution. For GLE to be of practical value, ( 1) the dynamics entering in the time-dependent friction c(t), should, at the very least, be independent of the externaljeld, V( A’). Additionally, (2 ) the analysis is simplified if the dynamics of the friction i(t) are independent of the motion of the effective particle. Though not a requirement ( 3 ) the GLE can be used most effectively if the dynamics of i(t), in contrast to its value at t =O, depend only upon the dynamics of the pure bath and are independent of the very presence of the stationary tagged particle. In the generalized Stokes-Einstein-Debye treatment of translational and rotational diffusion this is the case; there is no external potential V(X), the friction [depends upon the properties of the bath through the coefficient of shear viscosity q of the pure solvent or bath, and the coupling of the particle motion to the bath is described by a coupling parameter which depends upon the static boundary conditions between particle and bath, but is independent of the particle dynamics. If these conditions do not hold, i.e. if the friction [ depends upon the external field and upon the particle motion, the usefulness of the Langevin approach and the very concept of friction may be of questionable value. It should be noted that the introduction of a time-dependent friction i(t), as in the Grote-Hynes theory, need not necessarily violate any of these requirements; such a time-independent friction can still depend upon the “pure bath” properties through the viscosity, but through a frequency-dependent and k-dependent viscosity, and the static coupling parameter can also be k-dependent. Zwanzig [ 161 and later Lindenberg and co-workers [ 25-271 have shown that, indeed, for a very special model the GLE holds exactly, irrespective of the magnitude of V(X) and of the random (coupling) force R (t ). See also ref. [ 23 1. This model is one in which the bath consists of harmonic oscillators, and in which the potential coupling the effective particle to the bath is bilinear in X and the coordinates (B) of the bath. This appears to be the only case in which the GLE satisfying the above conditions ( 1) and (2 ) holds exactly [ 28 1. On the other hand, Kim and Oppenheim [ 2 1] have shown that for sufficiently small
G. Tarjus. D. Kivelson / Validity of the GLE approach
external field (I’,), for high bath relaxation frequency (rk)-‘, and for large tagged particle mass (M), the GLE also holds. We have sought general conditions under which a GLE is valid and useful. Our studies recapture the exact conditions of Lindenberg and co-workers, as well as the Brownian diffusion limit of Kim and Oppenheim. We also obtain criteria indicating how far a system can deviate from these limits and still be usefully described by a GLE. And, under conditions for which a GLE is valid, we examine the effect on the friction C(t) of the X dependence ofR(X, B).
2. Our approach In this section we outline our approach to the problem, introducing a number of parameters of special interest. In the next four sections we present our results in considerable detail. After that, in section 8 we carry out the step-by-step derivations which lead us to the aforementioned results; by deferring the rather involved procedural details to a late stage in the article we hope to motivate some readers to follow these details, and to save those readers, who either have confidence in our algebra or who lose interest after reading our results, the time and effort needed to read through it all. We then end up with a short discussion of “barrier crossing” in section 9 and a summary of our “conclusions” in section 10. In studying reaction dynamics in liquids in terms of friction, we follow the standard approach and attempt to isolate a reaction coordinate X which is largely decoupled from both the other modes involving the intramolecular motions of the reacting species and those associated with the solvent. As in the Kramers theory, we treat the reaction as the motion of an effective particle moving across an activation barrier along the reaction coordinate, all the other motions being collectively taken as the “bath”, and one regards friction as the result of the coupling of the reactive motion to the bath. It may be that no single, well-separated reaction coordinate can be identified, in which case we can extend the concept of reacting modes to include a set Xof reaction coordinates which are largely decoupled from all other motions. If this cannot be done with a small set of such reaction coordinates, alternative approaches must be
155
sought. We shall focus on the case where one reaction coordinate X suffices. The “bath modes” are not true normal modes of the solvent (modes taken in the generalized hydrodynamic sense or perhaps normal modes such as those introduced by Seeley and Keyes [ 291 and Xu and Stratt [ 301) because they are evaluated in the presence of the effective particle; although motion along the reaction coordinate is explicitly excluded from the “bath modes”, it follows, as we shall discuss, that the bath modes depend upon the presence, but not the motions, of the effective particle. In the gas phase the bath consists entirely of internal modes; in liquids the bath consists of both internal and solvent modes. At least at low frequencies, the behavior of the solvent modes can be treated as that of a hydrodynamic bath and related to wellknown macroscopic liquid properties such as viscosity q(w) and dielectric permeability E( 0); thus the frictional effects can be associated with these properties. At high frequencies, such as those associated with passage over the peak of the activation barrier, the bath may lose its collective, long wavelength behavior, and the relevant frictional properties may not be clearly associated with the macroscopic solvent properties. And, of course, the friction arising from coupling to internal modes probably has little to do with properties such as solvent viscosity and dielectric permeability. Although our formal treatment encompasses friction from both solvent and intramolecular sources, our discussion here will focus largely upon solvent friction. It is useful to describe the bath in terms of two relaxation times, a short time 7; related to the Einstein frequency, and a long time 7% related to the diffusive processes; rf, is a collective property associated with the bulk behavior of the solvent whereas 7: is strongly dependent upon the short-range interaction between the effective particle and the nearby solvent. This isF an oversimplified picture but does seem to incorporate the principal features observed in MD simulations. It should be noted that the bath dynamics, along with these characteristic times, are those for the bath in the presence of a stationary effective particle. Other times which are of interest are il;’ and o;‘, the latter given by eq. (8 ) or more generally by
(10)
G. Tarjus, D. Kivelson / Validity of the GLE approach
156
and the former being approximated by eq. (9); o, is thus related to the curvature of the activation barrier at its peak (X=0), and A, is the actual frequency of barrier-peak passage in the presence of friction. Still another significant time is T, the smallest relevant time after which one is no longer interested in the details of the dynamics affecting i(t) and the response of the bath; this time is given by r= minimum
of r%or I;’
.
(11)
To understand this, note that if, as required, c(t) is independent of the external potential and of the motion of the effective particle, then in a time t> rg the friction is totally relaxed and we need no longer follow its temporal behavior; on the other hand, we are not at all interested in the dynamics after barrier-peak passage has been achieved, and the time for this is 1; ’ . Two interesting limits in which we can examine the GLE are the Kramers-Langevin limit in which the relevant times are very long so that ?=T% )
L-(MP)-l/‘exP(>r)-l
x
1+ F [
-x r3,.
by Hynes
(13)
We find it useful to examine the GLE for a tagged (effective) particle initially placed at a given position X0 along the reaction coordinate. This position might be at the peak of the barrier (XC 0) or any place else, even at the minimum of a potential well. Restating the GLE in this way does not restrict its applicability because once a GLE of this kind is derived, it can be solved for any initial position, and one can then integrate the solutions over any desired distribution of initial positions. To fix the initial position in this way, we introduce an average (A) Oin a lim-
x
,X,,(PM)l/2
1.
(15)
Similar considerations apply to the case where X0 is close to the minimum of a potential well. It is convenient to start with a Hamiltonian of the form
H=[fA4x2+W(X0,6X)]+
C fm,b:+U(X,B), Y
(12)
and the frozen solvent limit, first mentioned and co-workers [ 8 1, in which 74;’
reactive motion in this way because the interaction of the particle with the surrounding bath may be strongly dependent upon the position along X, and an average over all X might mask the interesting dynamics near the barrier peak. The fact that the resulting GLE is localized in this way leads to the concept of piecewise validity, i.e. segmental validity of the GLE. Thus the dynamic length L, the distance traveled by the effective particle in time r, must be short for a GLE to hold; as we shall show, for X0 close to the peak of the barrier,
(16) where W is the gas phase potential, 6X=X-X0, the bath coordinates B= {b,}, and U( X, B) represents all the potential energy contributions over and above W. Next we wish to transform the bath coordinates to get the best normal modes with the tagged particle frozen at some position Xc,. (The actual process for doing this may be a bit vague since in a liquid there is no unique configuration about which the system will make small oscillations, and, consequently, no unique set of modes [ 291. ) The transformed coordinates 6, are linear combinations of the b,‘s, and the new masses M, and the frequencies Q, are functions of the frozen-particle position X0. We can now rewrite U(X, B) as