Analytical Mechanics This page intentionally left blank
Download 10.87 Mb. Pdf ko'rish
|
+∆ f dX,
(15.51) is identified with f = 1
H =E f |∇ X H | d Σ . (15.52)
Hence we compute X i ∂H/∂X j , i, j = 1, . . . , 6N , assuming ∂H/∂X j ≡ 0.
According to (15.52) we have X i ∂H ∂X j = 1 ω H =E X i n j d Σ , (15.53) denoting by n j the jth component of the unit vector n = ∇ X H/ |∇ X H | normal to H = E. It is convenient to notice that if H(X) does not depend on one of the coordinates X j the contribution of the latter to (15.53) is absent. Therefore, when some coordinates can be ignored, the set {H = E} is implicitly replaced by its projection onto the subspace of coordinates X i for which ∂H/∂X i ≡ 0.
In addition we assume that the latter set is bounded. Denoting by e j the unit vector in the jth direction in the space Γ , the integral in (15.53) is simply the flow through H = E of the vector X i e j . Therefore, if n j is non-zero, since n is oriented towards the exterior of the set {H < E} we have X i
∂X j = 1 ω H X · (X i e j ) dX. (15.54)
15.8 Statistical mechanics: Gibbs sets 635 Finally, since ∇ X · (X i e j ) = δ ij , we arrive at X i ∂H ∂X j = δ ij B(E, V )
ω(E, V ) = kT δ
ij (15.55)
(recall the hypothesis ∂H/∂X j ≡ 0), where we used (15.50) in the last step. We have proved the so-called equipartition theorem. T heorem 15.9 (Equipartition) For a system described by the microcanonical set with Hamiltonian H, the average value of X i (∂H/∂X j ) is zero if i = / j, and has value kT if i = j, as long as ∂H/∂X j ≡ 0.
Remark 15.11 If the Hamiltonian H of the system contains terms of the kind p 2 i
p i (∂/∂p i )(p
2 i /2m) = 2 (p 2 i /2m), each one contributes to the average value of the energy with a factor 1 2 kT . Analogously, if the Hamiltonian H contains terms of the kind
1 2 mω 2 i q 2 i we have that q i ∂/∂q
i 1 2 mω 2 i q 2 i = mω 2 i q 2 i , and their contribution to the average value of the energy is equal to 1 2
a system whose Hamiltonian is a quadratic form: H =
1 2 6N i =1 A i X 2 i ≥ 0, A
i ≥ 0,
(15.56) or can be reduced to it via a canonical transformation. 2 Note that by (15.56) we have 6N i =1 X i ∂H ∂X i = 2H. Then using equation (15.55), we can conclude that H =
1 2 lkT, (15.57) where l is the number of non-zero coefficients in expression (15.56). The the- orem of equipartition of energy states precisely that each (non-zero) term in the Hamiltonian (8.6) contributes to the average energy by the quantity 1 2
Remark 15.12 For a perfect monatomic gas, the equipartition theorem implies that the average value ε of the energy for each molecule is equal to 3 2 kT . Recall that in the kinetic theory, this same result was assumed as the definition of temperature. Here we deduced it from the definition of entropy and from the concept of the average of a set. The fact that the kinetic energy of the system is expressible as 3 2
2 The restriction that A i
in the subspace of coordinates that cannot be ignored. 636 Statistical mechanics: Gibbs sets 15.9 The equipartition theorem is the critical issue of classical statistical mechanics. This fact becomes clear when the internal degrees of freedom are increased, to include systems with great complexity, such as black bodies, to which infinitely many degrees of freedom (but not infinite energy) can be attributed. A profound and enlightening discussion of this problem is presented by Gallavotti (1995, pp. 65–85). 15.9
Closed systems with prescribed temperature. Canonical set In Section 15.2 we announced that we would consider two typical situations for closed systems: systems with fixed energy (described by the microcanonical set) and systems with fluctuating energy, but with prescribed average. We anticipated that this corresponds to fixing the temperature of the system. Physically this is done by connecting the system with a thermostat. The statistical set describing the latter thermodynamical state was called by Gibbs the canonical set. Therefore, while for the microcanonical set the independent thermodynamical variables are E, V , for the canonical set they are T, V . The union of the system under consideration, S (with N particles) and of the thermostat T, must be assumed at constant energy, but because of the large size of the thermostat, the fluctuations of the energy in S can in principle be as large as the energy of the entire system S + T. Hence the set E associated with S can span a region in the space Γ of
(where dim Γ = 6N ). The problem is now to define the density ρ(X). D efinition 15.8 The canonical set is the statistical set corresponding to a closed system in thermal equilibrium with a thermostat at temperature T . It is obtained by setting E = Γ
ρ(X) = 1 N !h 3N exp
− H(X)
kT , (15.58) called the canonical or Boltzmann distribution, where h is a constant with the same dimensions as the action [h] = [p][q] and H(X) is the Hamiltonian (15.1) of the system under consideration. Typically, h is identified with Planck’s constant. Remark 15.13 The introduction of h −3N is necessary for ρ dX to be dimensionless. The factor 1/N ! is consistent with the rule of ‘Boltzmann counting’, already discussed. There exist in the literature derivations of (15.58) based on heuristic consid- erations (for example considering the canonical formalism that can be applied to the thermostat, when we can expect that the fluctuations of the energy are very small; this fact, not rigorously true, is largely justified a posteriori). We prefer to motivate (15.58) in a way that highlights the relation with the kinetic theory. To draw this parallel, we must limit ourselves to the case in which the
15.9 Statistical mechanics: Gibbs sets 637 interactions between particles are only of collisional type (short range). Since in the canonical set the fixed thermodynamical parameter is the temperature, it is convenient to start by recalling (Remark 14.3) that the equilibrium solution of the Boltzmann equation has the form f (p, q) = ce −βh(p,q) , β = 1 kT , (15.59) where h(p, q) = p 2 /2m+
Φ (q), and
Φ is the potential energy of the external forces. Recall also that the internal forces influence the dynamics of the system in the space µ (through the interaction kernel), while the structure of the equilibrium depends only on the fact that the interactions are binary and preserve the energy (note that they can be modelled as singularities of the Hamiltonian and therefore they do not contribute to (15.59)). We then examine again the partition in cells ω 1 , . . . , ω K (of volume ω) of the space µ, introduced in Section 15.6, and the corresponding discretisation of the space
Γ in cells of volume ω N . For a fixed cell Ω in the space Γ , by projection on the component subspaces µ 1 , µ 2 , . . . , µ N we can reconstruct the corresponding sequence of occupation numbers (n 1 , . . . , n K ) in the cells ω 1 , . . . , ω K . We know that this correspondence is not one to one, but now we are only interested in the map
Ω → (n
1 , . . . , n K ), and to obtain through this map information about the probability of finding a sampling of the representative point X ∈ Γ precisely in the cell
Ω . This probability is the product of the probabilities of finding, in the space µ, n 1 points in the cell ω 1 , n
2 points in the cell ω 2 , and so on. Denoting by h i the values of the Hamiltonian h(p, q) corresponding to the cells ω i , according to (15.59) such a product is proportional to exp
−β K i =1 n i h i = exp( −βH(X)), (15.60)
where H = K i =1 n i h i . This result, and the procedure followed to obtain it, offers direction to remark on some particularly significant aspects. (1) The distribution function f (p, q) in the space µ can be normalised to a probability distribution by simply dividing it by N . Indeed, (f /N ) dp dq = 1. Denoting by f i the value of f in the cell ω i of the partition of the space µ, considering the product f n 1 1 . . . f
n K K , to obtain the probability of finding X ∈ Γ in the cell Ω corresponding to the sequence (n 1 , . . . , n K ), we should have divided it by N n 1 +···+n K = N N . For N
1 this is in practice N !. This simple consideration shows that the factor 1/N ! appears naturally in the expression for the density, multiplying the exponential e −βH
, and it confirms that in the formalism of Boltzmann it is implicitly the correct count of states, since taking the factors f i corresponds to labelling the cells, but not the particles. 638 Statistical mechanics: Gibbs sets 15.9 (2) The reader will not have missed the rather puzzling fact that in Section 15.6, imposing the constraints of the microcanonical set K i =1 n i = N K i =1 n i h i = E we obtained as most probable distribution in the space µ, using N 1, precisely the same distribution we used to deduce the density of the canonical set! However, there is a subtle conceptual difference: in the microcanonical set- ting we must interpret β as 3 2
−1 , while in the canonical case, β = 1/kT . This does not have practical implications, since once we have correctly defined the temperature in the microcanonical case, the two expressions coincide. The essential coincidence of the equilibrium distributions in the space µ for the two cases carries significant physical information: for N 1 the mean quadratic fluctuation of H in the canonical set is so small that the important region of the space Γ for the canonical set is a ‘thin’ shell (Section 15.3) around the manifold H = H ρ . This fact, which we prove below, implies that for N 1, in spite of the significant formal differences, the two are essentially identifiable and the choice of one or the other formalism is only a question of computational convenience. We stress that this is only true for N 1. Indeed, the canonical set has a unique characteristic: it is orthodic (hence it gives exact information about some averages) even when the number of particles is small (even for N = 1). However the fluctuations are then large. We shall return to this in an appropriate section. (3) We finally remark that when there are internal forces, not of collisional type (hence acting at a distance), we must take their total potential energy into account in the expression for H. D efinition 15.9 The measure of the canonical set in the space Γ is given by Z(V, N, T ) = Γ 1 N !h 3N exp ( −βH(X)) dx (15.61)
and it is called the (canonical) partition function. The extension of the integration to the whole space Γ assumes the introduction of potential barriers simulating reflective walls and it requires implicitly the convergence of the integral. It follows from the definition that the canonical partition function expresses the density of states. Normalising the canonical distribution, and considering (1/Z) ρ(X) we obtain a probability distribution. If f (X) is an observable quantity, its average value in the canonical set is given by f = 1
3N Γ f (X) exp ( −βH(X)) dX. (15.62)
15.9 Statistical mechanics: Gibbs sets 639 P
equal to H =
− ∂ ∂β log Z(V, N, T ). (15.63)
Proof Since Z = C Γ e
dX, where C is independent of β, we have ∂ ∂β log Z = − Γ He −βH
dX Γ e −βH dX = − H . Example 15.3 Consider the Hamiltonian of a perfect monatomic gas H =
N i =1 p 2 i 2m . The partition function of the system takes the value Z(V, N, T ) = V N N !h 3N χ 3N ∞ 0 p 3N −1
e −βp
2 / 2m dp, where χ
3N denotes the measure of the spherical surface of radius 1 in R 3N .
∞ 0 p 3N −1 e −βp 2 / 2m dp = 1 2 (2mkT ) 3N /2
∞ 0 t 3N −2/2 e −t dt = 1 2 (2mkT ) 3N /2
Γ (3N /2) , and hence Z(V, N, T ) = V N
3N χ 3N 1 2 (2mkT ) 3N /2 Γ 3N 2 = V N N !h
3N (2πmkT )
3N /2 , (15.64) where we have used the fact that χ 3N = 2π 3N/2 / Γ (3N /2). Therefore the average value of the energy is H = 3
N kT, in agreement with our previous results. 640 Statistical mechanics: Gibbs sets 15.10 Note also that this is a typical case where H can be interpreted as ‘by far most probable value’ of the energy, following the discussion of Section 15.3. Indeed, if for N 1 we compute ξ( H ) =
1 Z H = H ρ |∇ X H | d Σ ≈ β e (3πN ) −1/2 , we realise that equation (15.5) is satisfied with δ = kT e(3πN ) 1/2 , and hence with δ/ H = O(N
−1/2 ). If we then compute (15.16), i.e. F (E) = EZξ(E), we see that the function to be maximised in order to find the most probable value E is simply E 3N/2 e
, from which it follows that E = 3 2 N kT = H . 15.10
Equipartition of the energy (prescribed temperature) The equipartition theorem, proved for the microcanonical set, can be extended to the canonical set. We assume H(X) → +∞ for |X| → +∞. T heorem 15.10 (Equipartition) The average of the product X i ∂H/∂X
j in the
canonical set is δ ij kT , as long as ∂H/∂X j ≡ 0.
Proof It is enough to notice that Γ X
∂H ∂X j e −βH
dX = − 1 β Γ X i ∂ ∂X j e −βH dX. By integration by parts, since there are no boundary terms, we find X i
∂X j = Γ e −βH dX −1 Γ X i ∂H ∂X j e −βH dX = δ
ij kT.
As a corollary we then obtain that in the canonical case as well, the quadratic Hamiltonians such as (15.56) have average 1 2
coefficients of the form. We have hinted already at the difficulties of the equipartition principle in the microcanonical set for more complex systems than those considered so far. Such difficulties also arise for the canonical set. However it is surprising that for the latter the equipartition principle is still valid (with some restrictions) for systems with few particles (even just one). These are systems with a ‘simple’ Hamiltonian, but that do not follow the associated Hamiltonian flow, but rather a motion subject to fluctuations in the energy at a prescribed temperature, which also determines the average energy.
15.10 Statistical mechanics: Gibbs sets 641 A coupling with a thermostat is therefore implicit, and with it the system, although it has few degrees of freedom, is in thermodynamical equilibrium. Speak- ing of a system with few particles ‘at temperature T ’ we refer to the temperature of the thermostat. This concept is illustrated in the following example. Example 15.4 Consider a system of N 1 particles at temperature T and the subset constituted by only one of the particles, in statistical equilibrium with the rest of the system. Consider the simplest possible case, when the Hamiltonian of the single particle is H = p 2 /2m and let us try to apply the canonical formalism to this single particle. Define the density ρ =
1 h 3 e −βp
2 / 2m (note that we have used N = 1) and, assuming that the system is contained in a cube of side L, compute the partition function Z = L
h 3 R 3 e −βp 2 / 2m dp = L 3 h 3 ∞ 0 4πp
2 e −βp 2 / 2m dp = L 3 h 3 2πm β 3/2
. Formula (15.63) yields H = −
∂β log Z =
∂ ∂β 3 2 log β
= 3 2 kT, which is precisely the expected result. What is essentially different from the case of systems with many particles is the fact that the mean quadratic fluctuation is not small: we find p 4
2 = 15 4 (kT )
2 , so that ( H 2 − H
2 )/ H
2 = 2 3 . Check that for a point in the plane inside a square of side L the function Z becomes Z = L 2 /h 2 (2πm/β) (hence H = kT , ( H 2 − H
2 )/ H
2 = 1), while for a point on an interval of length L, Z = L/h (2πm/β) 1/2
Download 10.87 Mb. Do'stlaringiz bilan baham: |
ma'muriyatiga murojaat qiling