Analytical Mechanics This page intentionally left blank
Download 10.87 Mb. Pdf ko'rish
|
x
H =
H H < E H = E H = E + D ds Fig. 15.1 In other words, ω(E, V ) is the derivative with respect to E of the measure of the region defined by the inequality H ≤ E. It is easy to express ω(E, V ) through the Hamiltonian. Indeed, at every point of H = E the thickness ds of the microcanonical set (Fig. 15.1) can be obtained from ∆ =
X H | ds, to first order in ∆ . In this approximation we therefore have Ω (E, V,
∆ ) =
∆ H =E |∇ X H | −1 d Σ (recall Remark 8.12), from which it follows that ω(E, V ) = H =E |∇ X H | −1 d Σ . (15.21) The next definition follows naturally from these considerations. D efinition 15.7 The microcanonical set is the ergodic statistical set corres- ponding to a closed isolated system, described by the manifold of constant energy E =
Σ E with measure dρ(X) = |∇ X H | −1 d Σ , (15.22) where d
Σ denotes the volume element on the manifold Σ E
Note that the name ‘density of states’ assigned to the function ω can be misleading (a better name would be microcanonical partition function). In reality the role of density on E is played by the function ρ defined by (15.22), which also 626 Statistical mechanics: Gibbs sets 15.5 shows how to compute the average of a function f (X) on the microcanonical set: f = 1 ω H =E f |∇H| d Σ . Remark 15.7 When the Hamiltonian depends only on the momentum variables, H = H(P), the volume V can be factorised in (15.21) as the result of the integration dQ: ω(E, V ) = V N H (P )=E |∇ P H | −1 d Σ P . (15.23)
We shall see that a correct normalisation of ρ and ω requires the further division by N !. Remark 15.8 The procedure leading to the formula (15.22) is rather abstract. We can how- ever deduce the same formula by reasoning only in mechanical terms, and this shows the natural relation between the Hamiltonian flow in the space Γ of an
isolated system and the density of the microcanonical set. This approach relies only on the condition of ergodicity. Assume that a partition of E in subsets of comparable measure and diameter has been defined. Following the Hamiltonian flow, if the system is ergodic, all cells are crossed. Thus if we perform a sampling X 1 , X 2 , . . . , X n , . . . of the trajectory along a time sequence, the probability of finding points of the sample in a specified cell is proportional to the permanence time in the cell during the Hamiltonian flow. Evidently, the average time of per- manence in a cell centred at X is proportional to | ˙X(X)|
−1 . Since ˙ X = I∇
H, we find that the probability density associated with this cell is proportional to |∇ X
| −1 . Remark 15.9 Note that the elastic collisions with the walls and between the particles, which can be represented as singularities of the Hamiltonian, do not contribute to (15.21). However, we must stress that the ergodicity of the system is precisely due to the collisions. Indeed, in the absence of any collision the projection P of the point X onto the space of momenta would be constant. We have already remarked that the function ω(E, V ) carries global, and hence macroscopic, information about the system. Its significance is due to the fact that it can be used to define the entropy of the system: S(E, V ) = k log ω(E, V ) ω 0 , (15.24)
where k is the Boltzmann constant and ω 0 is an arbitrary constant making quantities dimensionless. Note that the logarithm makes the entropy an extensive quantity, as required by the thermodynamical formalism. Indeed, the partition 15.6 Statistical mechanics: Gibbs sets 627 function of a system obtained by the union of isoenergetic systems of identical particles is the product of the partition functions of the component systems. Starting from (15.24) we can also define the temperature: T = ∂S
−1 (15.25)
and the pressure: P = T
∂S ∂V = ∂S ∂V ∂S ∂E −1 (15.26) in such a way that from the expression for the total differential of S: dS =
∂S ∂E dE + ∂S ∂V dV (15.27) we can deduce the first principle of thermodynamics for our system, by simply multiplying by the temperature T : T dS = dE + P dV. (15.28) It is a significant success of this theory to have been able to express the entropy S(E, V ) through the Hamiltonian H(P, Q), and hence through the microscopic mechanical properties of the system. Remark 15.10 We stress that from the physical point of view, the definition of thermodynamical quantities for a system with which the microcanonical set is associated has essen- tially theoretical interest, since by definition the system itself is not accessible to measurement. However, the definitions (15.24)–(15.26) are self-consistent and will be used in subsequent considerations. 15.6 Maxwell–Boltzmann distribution and fluctuations in the microcanonical set Consider a system in equilibrium, closed and isolated, and described by the microcanonical set. We now want to prove two important facts, highlighting the relation between the formalism of the microcanonical set and the kinetic theory (Chapter 14). P roposition 15.1 To every distribution function in the space µ it is possible to associate a volume in the microcanonical set. After the appropriate normal- isation, this volume can be interpreted as the probability of the corresponding distribution. T heorem 15.7 The Maxwell–Boltzmann distribution in the space µ is the dis- tribution to which there corresponds the maximum volume in the microcanonical set, and it is therefore the most probable. 628 Statistical mechanics: Gibbs sets 15.6 Proof of Proposition 15.1 Introduce a partition of the accessible region of the space µ in ‘cells’ (for example, cubic cells) of equal volume ω. This volume must be small with respect to the total volume, but sufficiently large that we can find in each cell representative points of a large enough number of molecules. Since the accessible region in the space µ is bounded, the cells are a finite number: ω 1 , . . . , ω K . Indicate by n i the
number of representative points in the cell ω i (occupation number ). The numbers n i are subject to the characteristic conditions of the microcanonical set: K i =1 n i = N, (15.29) K i =1 ε i n i = E, (15.30) where ε
i = p
2 i /2m and p i is the momentum corresponding to the cell ω i (because
of the non-zero dimensions of the cell the momentum is not defined exactly, and therefore we have a finite variation in the energy, and hence the discretised microcanonical set in the space Γ has non-zero thickness). A K-tuple (n 1 , . . . , n K )
i ) = n
i /ω. To each prescribed distribution of the N points in the cells ω 1 , . . . , ω K (hence to each microscopic state) there corresponds exactly one specific cell of volume ω N in the space Γ . However to a K-tuple of occupation numbers there correspond more than one microscopic state, and hence a larger volume in the space Γ . Indeed, interchanging two particles with representative points in two distinct cells the occupation numbers do not change, but the representative point in the space Γ does. On the other hand, nothing changes if we permute particles inside the same cell. Since N ! is the total number of permutations and n i ! are those inside the cell ω i , which do not change the position of the representative volume element in the space Γ , we find that the total volume in Γ space corresponding to a prescribed sequence of occupation numbers (n 1 , . . . , n K ) is
Ω (n 1 , . . . , n K ) = N ! n 1 !n 2 ! · · · n K ! ω N . (15.31) Proof of Theorem 15.7 We seek the sequence (n 1 , . . . , n K ) maximising Ω and therefore expressing the most probable macroscopic state (with respect to the microcanonical distribution). Recall that n i 1. Using Stirling’s formula, we obtain log n! ≈ n log n, from which it follows that log Ω
1 , . . . , n K ) =
− K i =1 n i log n i + constant. (15.32) Considering now the variables n i as continuous variables, we seek the maximum of the function (15.32) taking into account the constraints expressed by (15.29) 15.6 Statistical mechanics: Gibbs sets 629 and (15.30). Hence we seek to find the extrema of the function Λ = − K i =1 n i log n i − λ
1 K i =1 n i − λ 2 K i =1 n i ε i , where λ
1 , λ
2 are the Lagrange multipliers. This procedure yields the K equations log n i
1 + ε
i λ 2 = 0, i = 1, . . . , K, (15.33) to which we again associate the conditions (15.29) and (15.30). Note that ∂ 2 Λ ∂n i ∂n j = −δ ij 1 n i , and therefore the extremum is a maximum. Redefining the parameters λ 1 , λ 2 we can write the solutions of (15.33) in the form n
= ce −βε
i , i = 1, . . . , K. (15.34) Equation (15.34) is simply a discretised version of the Maxwell–Boltzmann dis- tribution, setting ε i = p 2 i /2m (or ε i = p
2 i /2m + Φ (q i ) when there are external forces). It is easy to recognise that the continuous limit of the distribution associated with (n 1 , . . . , n K ) is precisely the Maxwell–Boltzmann one, and this automatically leads to the determination of the constants c and β in (15.34). We have therefore proved that the Maxwell–Boltzmann distribution is the most probable in the macroscopic equilibrium state associated with the microcanonical set. It is important to realise that the Maxwell–Boltzmann distribution is also the distribution ‘by far’ most probable, in the sense that the fluctuations around it are very small. For this we must evaluate the relative differences n 2
− n 2 i N 2 , (15.35) where
· represents the average on the microcanonical set. T heorem 15.8 The fluctuations (15.35) around the Maxwell–Boltzmann distri- bution tend to zero, for N → ∞, at least as N −1 .
Recall the discussion of Section 15.3 about the comparison between average and most probable value. It is then sufficient to compute n 2 i
i 2 . We follow a very elegant method, that can be found in Huang (1987). If in place of (15.31) we consider the function ˜ Ω
1 , . . . , n K , η
1 , . . . , η K ) = N !
η n 1 1 · · · η
n K K n 1 !n 2 ! · · · n K ! ω N , (15.36) 630 Statistical mechanics: Gibbs sets 15.6 where η
i are parameters varying between 0 and 1, using as before the Lagrange multipliers technique, we find that the maximum is attained for values of the occupation number n i = ˜
n i (η 1 , . . . , η K ) given by ˜ n i = η i n i = cη
i e −βn i , i = 1, . . . , K, (15.37) and that ˜ n i
i and ˜
Ω = Ω when η 1 = . . . = η K = 1. The parameters η i can
therefore be considered as volume reduction factors for the cells ω i . We note now that, for every (n 1 , . . . , n K ) and (η
1 , . . . , η K ), we have η i ∂ ∂η i ˜ Ω (n 1 , . . . , n K , η 1 , . . . , η K ) = n
i ˜ Ω (n 1 , . . . , n K , η
1 , . . . , η K ).
i , which by definition is n i
n n i Ω (n)
n Ω (n) , (15.38)
can also be expressed as n i = n η i ∂/∂η
i ˜ Ω (n, η) n ˜ Ω (n, η)
η = (1,...,1) , (15.39) where n = (n 1 , . . . , n K ), η = (η 1 , . . . , η K ) and
n denotes the sum over all sequences of occupation numbers. For the computation of n 2 i we note that η i ∂ ∂η i η i ∂ ∂η i ˜ Ω = n
2 i ˜ Ω (n,
η), and therefore n 2
= n n 2 i Ω (n) n Ω (n) = ⎡ ⎣ η i ∂/∂η i η i ∂/∂η
i n ˜ Ω (n,
η) n ˜ Ω (n,
η) ⎤ ⎦ η 1 =...=η K =1 . (15.40) 15.7 Statistical mechanics: Gibbs sets 631 The expression in square brackets can be rewritten and elaborated as follows: η i ∂ ∂η i ⎛ ⎝ 1 n ˜ Ω (n, η) η i ∂ ∂η i n ˜ Ω (n, η) ⎞ ⎠ − ⎛ ⎝η i ∂ ∂η i 1 n ˜ Ω (n, η) ⎞ ⎠ η i ∂ ∂η i n ˜ Ω (n, η) = η
i ∂ ∂η i ⎛ ⎝ n n i ˜ Ω (n, η) n ˜ Ω (n,
η) ⎞ ⎠ + ⎛ ⎝ 1 n ˜ Ω (n, η) η i ∂ ∂η i n ˜ Ω (n,
η) ⎞ ⎠ 2 . (15.41) The last term is simply n i 2 , after setting η 1 = . . . = η K = 1, while we identify the average of n i weighted by ˜ Ω (n,
η) with ˜n i , given by (15.37). Therefore the first term in (15.41) is simply η i (∂/∂η i )˜ n i = ˜
n i and reduces to n i for η
1 = . . . = η K = 1.
Finally from (15.40) and (15.41) we find n 2 i − n
i 2 = n i . (15.42) Recalling that n i /N < 1, from this it follows that n 2 i N 2 − n i 2 N 2 1/2 < 1 √ N , (15.43) proving the claim. Taking into account that N can be of the order of 10 23 , equation (15.43) implies that the fluctuations around the Maxwell–Boltzmann distribution are in reality extremely small, and hence that the probability that the system takes a state very different from this particular distribution is very small. 15.7
Gibbs’ paradox We leave aside the study of the general properties of the function S(E, V ) defined by (15.24), and we only compute in a simple way the entropy of a perfect gas, hence assuming that the Hamiltonian of the system is of the form H = N
=1 p 2 i 2m . (15.44) In this case, it is possible to use formula (15.23). In addition, |∇ p
| 2 = N i =1 (p i /m) 2 = (2/m) H. The integral (15.23) must be computed on the sphere 632 Statistical mechanics: Gibbs sets 15.7 P
= 2mE in R 3N . Denoting by χ n the measure of the spherical surface of radius 1 in R n χ n = 2π
n/ 2 / Γ (n/2) , we have ω(E, V ) = V N m 2 1/2
E −1/2
χ 3N (2mE) (3N −1)/2 . (15.45) Using the asymptotic expression Γ (z + 1) ≈ √ 2πz z +(1/2)
e −z , we can write χ 3N ≈ √ 2 (2π/3N ) (3N −1)/2 e (3N /2)−1 . Substituting into (15.45), taking the logarithm, taking into account that N 1 and retaining only the terms in E, V , N , we obtain, because of (15.24), the following expression for the entropy: S(E, V ) = N k log ˜ λV E
3/2 + 3 2 kN,
(15.46) where ˜
λ > 0 is the same constant appearing in expression (14.46) of the previous chapter.
The evident difference between the two expressions is that in (15.46) there appears V instead of V /N . While this does not affect the validity of the expres- sions ∂S/∂E = 1/T , ∂S/∂V = P/T , it nevertheless makes (15.46) unacceptable as a state function of the system. Indeed, if we consider two systems with the same particle density (n = N 1 /V 1 = N
2 /V 2 ), and the same average energy per particle (ε = E
1 /N 1 = E 2 /N 2 ), we want the entropy of their union to be the sum of the entropies S 1 , S 2 . As written, equation (15.46) does not have this property, and yields the paradoxical consequence that it is not possible to partition the system into two or more parts with identical ratios E i /N
, V i /N i and then reassemble it, again obtaining the starting entropy. This is Gibbs’ paradox. This difficulty was immediately evident to Gibbs himself and he had no choice but to correct (15.46) by inserting V /N in place of V : S(E, V ) = kN log V N V
0 + 3 2 kN log
E N E
0 + C,
(15.47) where C is a constant and E 0 , V
0 are constants that arise from making the variables dimensionless. Recalling the approximation log N ! ≈ N log N for N 1, we realise that this correction is equivalent to dividing the function ω(E, V ) by N !. Therefore it amounts to a renormalisation of the density of states (Boltzmann counting), corresponding to considering the particles to be indistinguishable (the natural point of view of quantum mechanics), so that any permutation gives rise to the same state. The kinetic theory does not lead to the same paradox because, as we observed, it admits the interchange of particles. Boltzmann renormalisation (which we henceforth adopt systematically) puts back on track classical statistical mechanics, by introducing in it the concept
15.7 Statistical mechanics: Gibbs sets 633 of non-individuality of the particles. The criticism aimed at classical statistical mechanics on the basis of Gibbs’ paradox is, in our view, scarcely motivated, as the need for considering the particles as indistinguishable does not deprive the theory of its elegance and deep significance. However the theory presents more important limitations when one wants to study the contribution of each degree of freedom to the energy of the system. We shall discuss this topic in the next section.
To this end, it is useful to conclude this section with a few additional considerations on the definition of entropy. Equation (15.24) is not the only possibility of defining the entropy starting from the structure of the microcanonical set. If we assume that the set {H ≤ E} is bounded and denote by B(E, V ) its measure (recall that then ω(E, V ) = (∂/∂E) B(E, V )), we can show that another equivalent definition of entropy is given by S(E, V ) = k log B(E, V )
B 0 . (15.48) To verify this fact we denote by S ω the entropy defined by (15.24) and by S B the entropy defined by (15.48). Define T ω = (∂S
ω /∂E)
−1 and T
B = (∂S
B /∂E)
−1 , and note that T B = (kω/B)
−1 . In addition, S ω
B = k log
cω B = k log c kT B (c constant with dimension of energy), and differentiating with respect to E we have
1 T ω − 1 T B = −k 1 T B ∂T B ∂E , and hence T B
ω = 1
− k ∂T B ∂E . (15.49) Now it is sufficient to recall that k ∂T B /∂E = O (1/N) (and precisely 2/3N in the case of a perfect gas) and conclude that in the thermodynamical limit (when N → ∞) we can assume T ω = T
B , and that this equality is satisfied in practice for N finite (of the characteristic order of magnitude 10 23 ). In particular, we have obtained the following expression for the temperature: T =
k ω(E, V )
B(E, V ) −1 . (15.50) Example 15.1 It is
to verify
the validity
of equations (15.50) for
the Hamiltonian (15.44). It is enough to recall that the volume of the sphere of 634 Statistical mechanics: Gibbs sets 15.8 radius R in R n is 1/nR
n χ n (χ n is the measure of the unit spherical sur- face). Therefore (n = 3N , R = √ 2mE) B = 1 3 N (2mE) 3N/2 χ 3N , and hence ω/B = ∂/∂E log B = 3 2
Example 15.2 We now deduce the heat capacity at constant volume C V and that at con- stant pressure C P from the expression (15.47) for the entropy of a perfect monatomic gas, recalling that dQ = T dS = dE + P dV . For constant volume we then have C V = ∂E/∂T = 3 2 N k. If, on the other hand, we impose P = constant, we must use the substitution V /N = kT /P in (15.47), and hence S = kN log kT /P + 3 2
dS = kN 1/T
+ 3 2 (1/E) 3 2 N k dT , hence T dS = (kN + C V ) dT , and finally C p = kN + C V = 5 2 N k.
15.8 Equipartition of the energy (prescribed total energy) Recall that the average of a quantity f on the microcanonical set, f = lim
∆→0 1 ω ∆ E Download 10.87 Mb. Do'stlaringiz bilan baham: |
ma'muriyatiga murojaat qiling