860 © The Kinetic Theory of Dense Phases @) = JL xe, 1 de, is therefore x9 at all times, and the ‘ean squared displacement from this postion ix Foes? ec) de (@-x0)?)= = Piemne ee |. Cap? 20. eo ‘This is the diffusion law first derived by Einstein. (See Eqs. 20.17 and 20.18 for the three-dimensional equivalents of Eqs. 29.32 and 29.34, respectively.) Itis clear that Eqs. 29.28 and 29.34 are of the samme form; they also describe the same process. Diffusion of particles through a medium occurs by motions in random direetions, ‘uncorrelated with one another after periods of time corre- sponding to many molecular collisions. In the random-walk problem we start with unit probability at the origin of the ‘one-dimensional coordinate system and ask for the proba- bility distribution of positions in space after N steps. To cor- respond with our intuition about diffusion in an isotropic medium, we choose the « prior? probabilities p and q to be equal. Thus, on the average, we have (m1) = (m2) = NI2, and hence (n) = (i) ~ (na) = 0. The average displacement from the origin is zero, For any one sequence of steps, however, the net displacement from the origin may differ from zero, with a probability distribution given by P(ny: N), Eq. 29.29. ‘We can alter that equation by substituting 9.5) so that Eq, 29.28 becomes (3) en or (m2)=4Npg, 936) for p= 4, But for constant step size /, the value of m (the net, number of steps in the positive direction) is simply Av, ‘where Aci the net distance traversed. Thus, inthe isotropic, case (p = q= }), the mean squared displacement for the random-walk problem is ee (an? )= NP =, e337 where is the interval between displacements. By making the identification D =z, we find the solutions of the diffu sion and random-walk problems to be identical. If we con- sider a random walk in which the step size is not constant, then Eq. 29.37 stil holds with the constant step size 1 replaced by the average stop size. ‘Equation 29.34 relates the mean squared displacement of a particle in a given time # to the diffusion coefficient D. ‘This equality (or rather its three-dimensional equivalent) can be used directly o determine the diffusion coefficients of visible colloidal or aerosol particles in thei respective media from observations of Brownian motion. Since we have not had to specify any restriction about the mass of the diffusing pasti- cle, the equation is generally valid, provided only thatthe time ‘exceeds the time over which correlations of molecular veloc- ity can persist. This is the same as saying that the equation holds if the diffusion occurs by a random-walk process. We shall have more to say about correlations (and thei relation to ‘wansport coefficients) inthe next section; see also the iscus- sion of probability aftereffects in Appendix 16A. ‘The concept of a random-walk process has been applied to a variety of physical and chemical processes besides dif- fusion. As an example. consider the growth of a polymer by addition of monomer segments. Suppose that we are inter- ested in the configuration of the polymer, which consists of a given number of monomer units. If we view each monomer segment simply as a vector, then the configuration of the polymer is obtained by addition of the vectors representing monomer segments. The sequence of these vectors will resemble a Brownian-motion trajectory like that in Fig. 29.4 ‘The vector addition process, if not constrained by the fixed ‘bond angles between the monomer segments, is a random- walk process in three dimensions, provided that we neglect the probability of two seements of the chain occupying the same space (excluded-volume effect). The rms separation between the ends of the chain, which provides a measure of the “size” of the polymer molecule, can thus be estimated with simple Brownian-motion methods. A realistic calcula tion requires that one conect for the constraints just men- tioned (bond angles and excluded volume). An approximate bt useful way to deal with fixed bond angles comes from the recognition that after a succession of some bonds, six of seven for simple polymers, bond angle correlation is lost ‘The polymer configuration still can be described by a ran- dom walk, but the step size is not a bond length but a new length called the statistical segment length. Another example suitable for analysis by a random-walk process is the dissociation of a molecule. Consider a diatomic molecule, say Ch, colliding with Ar atoms. By these collisions the Ch molecule may be excited from the {ground vibrational state to one of the higher vibrational states. Deexcitation by collision may, of course, also take place. Collisions thus provide the mechanism for a random walk over the vibrational and rotational energy levels of Cla On reaching the dissociation energy, the Cl molecule ‘breaks apart. The rate of dissociation may then be calculated from the average time for the first passage of the random 29.3 Stochastic Approach walk across a designated energy level, yy to Transport ‘The kinetic theory of transport processes in a dilute gas, as presented in Section 288, is totally dependent on thedynamics of a eworbody mechanical system, in part such details as the scattering angle and the impact parame- ter. We did not even attempt to follow the complicated route by which the transport coeftcients (defined in Table 28.1) can be calculated. The full calculation requires expressing the gradients in a fluid in terms of deviations from the ‘Maxwell-Boltzmann velocity distribution. These deviations must satisfy the general form of the Boltzmann equation, the spatially homogeneous form of which we gave as Eq, 28.28. ‘And this equation in turn depends on ocatloring eroze Gee tions, that is, on collision dynamics. We short-circuited this intricate procedure in Section 28.8 by resorting to a simple argument in terms of mean free paths, and thus ofthe hard sphere model. But neither the simple nor the complex pro- cedure can be extended to dense fluids. The mean free path ‘cannot be ineaningfully defined when the molecules are in ‘continuous interaction, and the Boltmann equation ikewise fails because its derivation assumes independent binary co Tisions: There are no analytic solutions for n-particle scat tering problems with n > 2, The prospects for developing a theory Took bleak, but in the very complexity ofthe problem lies the origin of a simpler approach. Consider a macroscopic. Brownian particle immersed in A fluid. 1 undergoes zigzag random motion due to collisions ‘with 2 number of molecules of the fluid at onetime. If there are interactions with many particles, perhaps a statistical (Gtochastic) rather than an exact mechanical analysis is pos- sible. We now briefly outline such a stochastic approach in ‘order to present some ofits basic ideas Consider « Brownian particle moving with a large initial velocity through a fluid; the particles velocity decreases with time, very much like that of a marble shot into molasses. ‘The decrease in velocity is due to molecular “friction,” the net effect of many random collisions of the particle with the ambient fluid, The fretional deceleration, in many systems, can be taken to be proportional fo the velocity; ths is « use- ful approximation, and probably sufficient. Yet the frietional deceleration cannot be absolutely smooth: since the force on the particle is due to random collisions with the molecules of | the fluid, it fluctuates and only on the average is directly ‘opposed to the particle's velocity. Indeed, such a random, fluctuating force would be acting on the particle even if it ‘were held in place (with zero velocity). We therefore write the equation of motion ofthe Brownian paricte as ae m oot fn, (29.38) dt where m 1s the particle's mass; Vis its velocity; is the 90- called friction coefficient, representing the mean frictional force per unit velocity; and id is a fluctuating, statistical force that represents the random component of the parti- cle’s interactions with the fluid. This is known as the Langevin equation (after Paul Langevin, a French physicist who worked on the problem of Brovinian motion at the beginning of the twentieth century), Since the average of ‘the interactions with the medium leads by definition to the Stochastic Approach to Transport © 861 ne Figure 29:7 Schematic of a randomly fluctuating {orce (in one dimension), Over a suthicietly long Orme, 7) should average to 2e0, frictional term G9. the average of the deviations from this average force must vanish: 2939 (F())=0. __ A plot of such a fluctuating force might resemble Fig. 29.7. Tf we take an ensemble average of By. 29.38, that is, an sverige over the motions of many independent Brownian particles, we have av) mi) ew); 40 dt $v); ean from this equation we see that a large initial velocity will, om the average, decay to zero with a relaxation time mi initial kinetic energy of the Brownian particle is thereby ‘converted to random thermal motion ofthe fuid (heal), This conversion, called [issipation] occurs because of collisions of the particle with the molernles f the retin, We ‘ise thatthe friction coefficient mst be related t0 the fluc- tuating force, and we shall return to this point shortly. > Letus next use the Langevin equation to derive the mean squared displacement of « Brownian particle, From ou? problem, we know that the mean squared displacement at time #is equal 1o 2D, where Dis the diffusion coefficient. The combination ofthis result, ‘withthe Langevin equation yields an important relationship between the diffusion coefficient and the friction coefficient . We begin with Eq. 29.38, multiplying both sides by the coordinate variable x to obtain ‘cussion of the random-wi wiettein —— mm where i= df= v The fist term ean be writen 4s mofo n[ Sea? | 294 which makes the ensemble average of Eq, 29.41 n( So) Slit). vay at4 — | «& B62_* The Kinetic Theory of Dense Phases Note that we have omitted the fluctuating term (xf(1)). This term can be written as (x)4fl0)), since we assume the ‘forces on the particle to be independent ofits position in the fluid: and since (f(8)) = 0. the product alse vanishes. The term (2) should equal kg7 fora particle with one degree of freedom (by equiparttion of energy), and we can thus rewrite Eq, 29.42 in the form 4 mg a= Fad) + kD. (2948) (Note that throughout this argument we have assumed the cquivalence of ensemble averages and time averages: See Section 15.1 on ergodic theory.) The solution of Eq, 29.4 is simply cniya ae Gm 4 Ka Kaiya a a 909) ‘To determine the integration constant A one needs an initia. condition, for which we choose (without restriction of gen- cerahty) x= 0 at ¢=0: we thus obtain bar g Now the solution of Eq. 29.44 can be checked by differenti- ation of Eg 2045 and use of Bq, 2946. Rewriting (i) as A 9.46) we have 1d iy me bitm ra" (1-e-$im), ean ‘on integrating this equation, we finally arrive atthe desired ‘expression for the mean equared dieplacement of a Brown ian particle, 9.48) 2keT[ _m, aT) fy e-cum re [: zine »} Let us take two Timiting cases that provide insight into the meaning of this equation. For periods much shorter than, the relaxation time, that i, for << mg; we can expand the ‘exponential through the wer nr, > Thus the root-mean-square displacement at time ¢ is kein)", where the factor (Tim) is the rms speed, We recover the expected fact tial for very short times the Brownian particle undergoes free motion. The time scale for “very shor is determined by the condition ¢<< ml, that i, by the friction coefficient. At the ether extreme, for periods tong compared to the relaxation time (r-<< ml). Bq, 29.48 simply reduces to w cox) o (Over such a long period the Brownian particle suffers myn ads of collisions, so many that the correlation of interactions disappears and the displacement is 1 sequence of uncorre- fated events, a random walk. We can thut equate Eq. 29.50 (0 our result forthe mean squared displacement in a random ‘walk, Eq. 29.34 to obtain the important relationship oy Bsa, on 32 # oy ‘connecting the friction coefficient and the diffusion coef- ficient D. We have derived this result in one dimension, but valid in three dimensions as well. Both of these coefi- cients, in tum, can be related to the viscosity ofthe fluid ‘medium. As was shown orginally by Stokes, the frictional force ona large spherical particle moving through a conti ‘uum fluid with velocity ¥ ic 95%) fact used frequently for particles ‘of molecular size. Substituting the value of ¢thus obtained into Eq, 29.52 yields the Stokes-Einstein law of diffusion, ky os) It is time now to give some mumenical estimates. Con- sider a Brownian particle, say, a colloidal particle of radius 100 nm, in water. The viscosity of water is almost exactly 10° Pa's (= 10° kg/m s), 90 by Stokes’s law the friction coefficient is $=67(10-> kgm sH107 m)=1.9%10- kg/s. Introducing the value of Boltzmann’s constant, we obtain for the diffusion coefficient of the particle ky _138%10-2 x30 K ¢ 19X10 kyls =2.2x10- ms,Autocorrelation Functions and Transport Coefficients If the density ofthe particle is close to that of water, its mass, is about 4 x 10" kg; the relaxation time is then f T9KI0 kes “Phws, it is only for £<« 10" that the velocity of the Brown- jan particle is essentially constant. For 1 >> 10°, on the ‘other hand, the partcle’s motion is deseribed by a random: walk oF diffusion process, a process random in position and velocity [Note that according to Eq, 29.40 the average velocity of «Brownian particle at first approaches zero exponentially with a time constant m/g; that is, for @ particle of the size described here a given initial velocity is reduced by a factor of e every 2x 10~ s, until vs so small (for ¢>> m/g ) that the motion is random. Inthe later limit the velocity should have its thermal value, as given, say, by Eq. 28.49; for the particle in our example this yields 1,38%10-2 JK x30 K two) for the rms speed at 300 K. However, this is merely the aver age speed benseen collisions; by Eq. 29.51 the net distance traversed in a second of random motion should be only about 20 0 =(2x2.2%10-P m2Asx1 )!2 = 2x10 m, or 10 times the particle's diameter, We can use Langevin's equation to obtain yet another important relationship. Suppose that we have a charged Brownian particle in a fluid. Ifthe charge on the particle is 4, then in an electric field E the force on the particle is gE. -vin equation must be amended to read (still dv me %E. Crates Fina 2955) ‘We again take an ensemble average and consider the stationary-stare condition defined by zero acceleration, dofdt = 0; Eq, 29.55 then becomes. QE =E{v). ‘The[mobilty ulof a charged particle isthe ratio between its velocity and the electric field; the ensemble average of ‘mobility is thus, 930) © 863 From these two equations and Eq, 29.52 we obtain the (uy _ —+ ad 29 58 D kyT es tion to diffusion in space (translational diffusion) discussed s0 far, there exists another diffusion, rotational diffusion, for molecules or larger aggregates such as a col- loidal particle with nonspherical structure. Consider as an example a dipolar molecule, say CgHsNHi, as a solute in a solution; we neglect the details of the structure of the sol- vent. The dipoles, and hence the aniline molecules, can be aligned in an imposed electric field. On removal ofthe field, the orientation of cach dipole will change doc to collisions with the solvent, and there occurs Brownian motion in the angle of the dipole with respect to the inital orientation defined by the imposed field. ‘Tis Brownian motion turns into rotation diffusion described by rotational diffusion coefficient, Do = kyTI6V,, where Vis the molecular volume, The oricntalional correlation function, (CEU + 5)) (See next section for a discussion of such functions), decays exponentially in time, since at large times all angles between ©° and 180° ate equally probable, with » deeay constant given by 1 = Viiky7: Rotational diffusion can be measured with time-resolved fluorescence depolarization, light scat- tering, dielectric relaxation, and transient grating methods. 29.4 Autocorrelation Functions and Transport Coefficients In our discussion of transport processes, we have several ‘times noted the existence of multiple time scales for events ina given system. There is a molecular time scale concerned with momentum relaxation (say, in a dilute gas) oF the local ‘equilibration of any quantity such as energy density. We expect that in a dense system, where a Brownian particle interacts with many molecules at once, the force on the par- ticle must vary rapidly. But one can expect that overa petiod of 10-10"? s, during which the particle can collide with ‘many molecules,’ this force will essentially awerage en zero; thus the random force f() should be uncorrelated over any time longer than 10-!s, Note that this is an interval du {ng which an initial velocity imparted to the large Brownian particle hardly changes at all. According to our previous cal- culation, the velocity relaxation time for the motion of & Brownian particle is ofthe order of 10°; the random force ccan thus be assumed uncorrelated on the time scale of the particle's motion. and Eq. 29.39 applies. On a still longer As we shal se later, molecules calculations confi the aut that tne conelaions of molecular velocity decay in bout tors.