14 Fundamentals of Statistical Mechanics 2. Phase Space(*) A geometrical interpretation of Hamilton’s equations is obtained by considering a set of values of the dynamical variables p; and q (i = 1,2,...,n) as the coordinates of a point P in a 2n-dimensional space, called the “phase space.” (P1,91,P2,925--+)Pns9n) = point P (21) in the 2n-dimensional phase space ° For a system with a single degree of freedom where n = 1, phase space can be visualized and represented by a plane (Figure 2.1); although this is not possible for larger n, this representation serves to symbolize the general case. The notation p,q shall be generally understood to stand for the whole set pi,q; (i = 1,2,...,n). Ph q qo at) Figure 2.1 Phase space for a system with a single degree of freedom, n=1, Let Po represent the point in phase space corresponding to the values Po, qo at the time ¢ = to. Then, by forward or backward integration of Hamilton’s equations, the coordinates p,q of the point P at any other time ¢ are then determined; with ¢ as a continuous variable, the functions p(t), g(t) can be understood to describe a “curve” in phase space, called the phase orbit. From the point of view of pure mechanics, the knowledge of these functions or, geometrically, of the motion along the phase orbit, represents, in fact, all that can be said about an individual system. The phase orbit is not necessarily a closed curve, and since a given set of initial values po, qo completely specifies the subsequent motion of a given mechanical system, phase orbits cannot cross.Fundamentals of Statistical Mechanics 15 It would be necessary, however, to have complete information about the system in the sense that all the values po, go are known, or, equivalently, all the values p,q at some other time, in order to determine the motion along a given phase orbit. In practice, this is not normally the case for a macroscopic system where one measures only a few quantities. In thermodynamics, one may not describe a system by more than, say, its pressure and volume or temperature. Suppose the system is composed of N & 10? particles. This leaves the very large number 2n = 2 x 3N of data, necessary for the mechanical description, practically unknown and, hence, open to guessing. One is therefore reduced to statistical methods, that is, to a statement of probabilities. In dealing with a mechanical system, it is then necessary, in principle, to consider all possible sets of values p,q that the system may assume at a given time. Before introducing these statistical methods, however, we prove an important result of pure mechanics, which is known as Liouville’s Theorem. 3. Liouville’s Theorem(*) We have seen that the dynamical variables pj and g with i = 1,2,...,n may be regarded as the coordinates in a 2n-dimensional space, called the phase space. Every set of these variables is thus represented by a point and the quantity n dd = J] dpida (3.1) i=1 represents the volume element in the phase space. A mechanical system found at the point P(p;,q;) at the time t will follow a definite phase orbit so that at a later time ¢’ it will be found at another point P’(p,, q,) with the set (p), @/,) obtained from the initial set (p;,q:) by forward integration of Hamilton’s equations. Since the new values are determined by the initial values in this fashion, we can write Pe = Pe(Pirdi) 5k =1,2,...,0 yo (3.2) % = U(Pir 9) " These equations can be considered as a transformation of the set of variables (p:,q:) to the set (p),q/,) or, geometrically, as a mapping of the phase space at the time ¢ to that at the time ¢/. The volume element dX and the corresponding volume element d)’, obtained through this transformation (Figure 3.1), are related by the expression dv = J(t,t')dd (3.3)16 Fundamentals of Statistical Mechanics where J(t,t’) is the jacobian determinant of the transformation eA 9p, 9a Oe Oe BO | i alpl.@l)| | J(t,t) = yes (pi Al ae jn Ban | (3.4) First, let t/ = t, so that p, = pe, 9% = qk, and therefore Op, 84, Pk = big Stk = by, Opi 84: (3.5) or _ 9 Me _ 9 Ogi Op: where 6,4 is the Kronecker delta defined by bin = 1 fori=k (3.6) =0 fori#k . In this case all terms on the diagonal of J are equal to unity and all others are zero. Therefore J(t,t) = 1 (3.7) Next, let t’ = t + dt, and keep only first-order terms in dt. One then has from Hamilton’s equations op = WE OH Pees at ~~ Onn 38) i = ith 2 OH © Kk = ato; Opn and thus oH Ph = Pk - 5 -dt Uk 3.9 OH (3.9) % = % + =—dt Ik qk Ops These relations may be differentiated to give Oph, OH Og, &H Pe = 6,-~———dt 5 t= dt ap; Spite” Op pide (3.10) 2 2 . om PH 5 Oh 5 OH 84: 8g:09% 04 8q:OpkFundamentals of Statistical Mechanics 17 q Figure 3.1 The time development of the volume element in phase space as described by Liouville’s Theorem (schematic). In evaluating the determinant in Eq.(3.4), each term containing a non- diagonal element as a factor contains at least another non-diagonal element as a factor. Since all these elements are proportional to dt, these terms are therefore at least of second order in dt and do not contribute to first order. To that order there remains then only the contribution from the diagonal elements I(:- Sir at) (1+ + ee anon) k 2 2 #145 (se oe ae (3.11) k OprOGe = OPKAGK Therefore a J(t,t+ dt) = J(t,t) + gle Dlermeat it (3.12) and since Eq.(3.7) states that J(t,t) = 1 we have the result 2 jin t)leae =0 (3.13) are oe . On the other hand, besides ¢ and ¢’, consider an arbitrary third time t” and the corresponding volume element dX” at the time t” (Figure 3.2) and use Eq.(3.3) with arbitrary t and t’. Then, going from t to t” we have dX" = J(t,t")da (3.14)18 Fundamentals of Statistical Mechanics and from ¢” to t’ dn’ = J(t’,t!)dX” (3.15) ' ‘aN Figure 3.2 Intermediate time t” and volume element dA” used in proof of Liouville’s Theorem. Therefore dN = J(t,t")J(t",t!)dd = I(t, t')dd (3.16) or J(t,t!) = (t,t) I(t, t’) (3.17) Hence, upon differentiation with respect to the final time ¢/ as (t,t! a te ) = Het) (3.18) Now on the right side let t” = t’, and this relation becomes aJ(t,t!) a Be = Mbt GI Eevae (3.19) Since Eq.(3.13) holds for any time t, it also holds for ¢” a gel, t)wae =0 (3.20) Equation (3.19) thus implies On _ gp ilht) =0 (3.21)Fundamentals of Statistical Mechanics 19 This relation now holds for arbitrary t’ and provides a first-order differential equation for the jacobian determinant. For t’ = ¢ one has the initial condition J(t, t) = 1 of Eq.(3.7), and hence J(t,t!) =1 (3.22) for arbitrary ¢’. It follows from Eq.(3.3) that dy’ = dX (3.23) One therefore concludes that the magnitude of the volume element in phase space does not change in the course of its motion along the phase orbits. The same result holds for the volume of any closed finite region R in phase space followed along the phase orbits (Figure 3.3). To every volume element dd within that region at the time t, there corresponds an equal volume element dX’ within the region R’ at the time t’ obtained from R by the equations of motion. Integrating the righthand side of Eq.(3.23) over R corresponds to the integration of the lefthand side over R’ so that dy’ = [ ay (3.24) R R or Vv=2 (3.25) where © and Q represent the volume contained within R and R’ respectively. In abbreviated form, this result is stated by Liouville’s Theorem: The phase volume is a constant of the motion. This theorem can be illustrated for the simplest case of a mass point in free, one-dimensional motion. The hamiltonian is p -z (3.26) Q’ | Q R l ——— ee Figure 3.3 Invariance of finite volume elements in phase space followed along the phase orbits (schematic).20 Fundamentals of Statistical Mechanics and Hamilton’s equations givet ; oH P= - ay = 0 __ OH» (3.27) 91> S ~ im Integration of these equations yields p = constant (3.28) qg= Pe + constant m which provides a parametric representation of the phase orbit. In the previous notation we have , pP=P 3.29) d=q+2e-8 oe The jacobian of this transformation can be immediatly evaluated as | om _\1 ttl _ Itt) =| oe 32. = \0 Tio (3.30) eq 8g which illustrates our general result. Consider further the rectangle abcd at the time t and the corresponding parallelogram a'i/c'd’ at the time t’ shown in Figure 3.4. Evidently Gt = ga + P20 -1) 3 Pa=Pb w= athe -t) 5 pe=pe we = ato ee) m qa = ga+ P(e -2) Thus — dar = 4% (3.32) Wd — We! = Wd- Ge Sw + We use the customary notation where a dot above a symbol denotes the total time derivative.Fundamentals of Statistical Mechanics 21 t 9 Figure 3.4 Phase volume for free one-dimensional motion of a mass point. Since a’b’c'd’ has the same width w as abcd and the same height po—pa, the areas 2 and 1 are equal, illustrating our general result. The extension to arbitrary regions R and R’ (see Figure 3.5) is immediately obtained by dividing the regions into infinitesimally thin horizontal strips and applying the above. l 4 Figure 3.5 Extension of result in Figure 3.4 to a region of arbitrary shape. Another example is the one-dimensional harmonic oscillator with hamiltonian 2 Pp 1 wg? H= — = . am + ymug (3.33)22 Fundamentals of Statistical Mechanics Hamilton’s equations give . OH 2 b= -G, = mtg _ oH pp (3.34) 4=% =m A combination of these relations yields | (p+ imug) = p+ imug = — Pq +i at San isles elie kee (3.35) = iw(p + imwg) and hence, by integration pi +imuwg’ = (p+ imug)e('-) (3.36) The real and imaginary parts of this relation are p = pcosa — mug sina : . (3.37) sina + g cosa with a = w(t! -2) (3.38) the jacobian can be evaluated as !) = & oe =| cosa eal = J(t,t) = ee ae | | —mw sin a cos | = cos*a+sin?a = 1 (3.39) |8q | in accord with our previous analysis. Further, upon change of scale q > mwg nee) (3.40) in phase space one has for the area of an arbitrary region in phase space 2 > mw ae (3.41) Now it is evident from Eq.(3.37) that the region R’ with area mw’, which evolves from the region R with area mw, is obtained from R through aFundamentals of Statistical Mechanics 23 rotation around the origin by the angle a (Figure 3.6). Since this does not change the area, one has mu = muQ (3.42) and hence (3.43) mw’ ———> moq Figure 3.6 Motion of the phase volume along the phase orbits for a one-dimensional harmonic oscillator.