U = γkt N. is indeed an exact one form. Thus there exists a function U(S, V, N) so that. du = T ds pdv. U S = T, U V = p

Transcription

1 A short introduction to statistical mechanics Thermodynamics consists of the application of two principles, the first law which says that heat is energy hence the law of conservation of energy must apply the second law of Thermodynamics, which says that there is no machine whose sole purpose is to do work by extracting heat from a reservoir. While it is not our aim to discuss Thermodynamics we treat the ideal gas in some detail. The ideal gas law says that pv = knt where p is the pressure, V the volume, T the absolute temperature N the number of particles (in moles) in the gas. The constant k is Boltzmann s constant. There is another equation which relates the temperature T to the energy contained in the gas U = γkt N where γ is a constant that depends on the type of the gas. The first law says du = dq pdv + µdn which is a confusing formula. (Here µ is the chemical potential. The formula suggests that there exists a function U(Q, V, N)) so that its gradient is given by (1, p, µ). This is not the case! There is, however, the second law which states that there is an integrating factor 1/T so that ds = 1 T dq is indeed an exact one form. Thus there exists a function U(S, V, N) so that du = T ds pdv. In particular S = T, V = p N = µ. Note the sloppy notation which is at the root of all the confusion in thermodynamics. One does not make the distinction between U as a function of T U as a function of S. both function are represented by the same symbol. Returning to the ideal case, we note that good variables would be U, V, N not S, V, N. Since S > 0 we can invert this relation consider S(U, V, N) hence S = 1 T 1

2 S V = p T S N = µ T. There is the fundamental requirement that the S, U, V N all scale the same way as we enlarge the systems. If we have twice as many particles, twice the interal energy twice the volume then the entropy should be multiplied by two. In other words S(λU, λv, λn) = λs(u, V, N). Differentiating the above relation at λ = 1 gets us For the ideal gas hence S(U, V, N) = U T + pv T µn T. 1 T = γkn U, p T = kn V, S(U, V, N) = γkn log U + kn log V + NC(N) for some unknown function C(N). By scaling or λ(γkn log U + kn log V + NC(N)) = γkλn log(λu) + kλn log(λv ) + λnc(λn) Differentiating this at λ = 1 yields hence C(N) = (γ + 1)k log λ + C(λN) 0 = (γ + 1)k + C (N)N C(N) = (1 + γ)k log N up to some constant. Hence, the entropy for an ideal gas is given by S(U, V, N) = γkn log U + kn log V (1 + γ)kn log N. Gas with a piston Imagine a a cylinder of length one (in some units) insulated against heat transfer filled with gas. Inside the cylinder is an ideal insulating piston at position 0 < x < 1. Half of the gas is on the eft of the piston where it has Volume Ax, (A is the cross section ) temperature T half of the gas is on the right of the piston where it has volume A(1 x) temperature T. Thus, the pressure on the left will be larger than on the right. The

3 piston is fixed with a pin. At some moment we pull the pin let the piston move freely which it does without friction. Eventually it will come to rest the question is where. Call the position y. The pressures on both sides must be the same, hence T L y = T R (1 y). Moreover, by the first law the energy must be conserved, i.e., Therefore T L + T R = T. T L = T y, T R = T (1 y). We know from the second law that the entropy must have gone up. Thus the entropy at the end of the process is greater than the entropy of the system before. The total entropy before the process is given by (up to irrelevant constants) whereas the entropy after is γ log T + log x + log(1 x) γ log(t y) + γ log(t (1 y) + log y + log(1 y). Hence applying the second law we get the inequality A straightforward calculation leads to the log(x(1 x)) γ log + (1 + γ) log(y(1 y)) y(1 y) (x(1 x)) 1 1+γ 4 γ 1+γ Note that when x = 1/ we get correctly y = 1/. What this result is saying that the given macroscopic quantities do not suffice the calculate the exact outcome of this experiment. The Second Law, however, yields bounds for the position of the piston. These inequalities are determined entirely by the macroscopic quantities. It might be that shocks all sorts of nasty things might develop by releasing the piston, but the second law sets absolute bounds on the outcome. The statistical theory of entropy We consider a system of N interacting particles in a box. The system is specified by giving the positions X = (x 1,..., x N ) the momenta P = (p 1,..., p N ). If we give a probability distribution ρ(x, P ) then the Boltzmann entropy is given by S(ρ) = k dxdp ρ(x, P ) log(ρ(x, P )).. 3

4 Assuming that the dynamics of the system is given by the Hamiltonian H(X, P ) one might ask what is the state with maximal entropy, given that the system has total energy U. In other words, we have to maximize S(ρ) given that H(X, P )ρ(x, P )dxdp = U that ρ(x, P )dxdp = 1. To solve this problem we start with the elementary inequality x log x e α (α + 1)x which holds for all x > 0 all α. There is equality if only if Next we have pointwise x = e α. kρ log ρ ke γh+δ k(γh + δ 1)ρ where γ δ are constants. Again, there is equality if only if ρ = e γh+δ. We now use the free constants to fix the correct side conditions γh+δ = 1 γh+δ H = U. The first says that we choose γ such that e δ = γh, the other means that we have to choose γ such that γh H γh = U. (1) 4

5 This is an equation for γ we have to show that there is a solution. Differentiate the left side of (1) with respect to γ get γh H γh γh + [ H γh ] since the left side of (1) is the expectation value of H with respect to the probability distribution ρ = H ρ e γh γh we can write this as H ρ + ( H ρ ) = (H H ρ ) ρ < 0. This means that the left side of (1) is a strictly decreasing function of γ. Clearly, any allowed value for U must be in the range of H. As γ tends to the left side of (1) converges to the minimal value of H. If we take our system of the form H = N p j + V (x 1,..., x N ) j=1 where the particles are constrained in a box, then we see that as γ 0 the left side of (1) tends to. Thus, in this case for any U in the range of H the equation (1) has a unique solution. We call this solution β. Hence the optimizing ρ is given by ρ canon = e βh βh. This is called the canonical ensemble Z = βh is called the partition function. Further we get S(ρ canon ) = kβ H canon + k log Z. () We shall see that this formula has a nice interpretation. For the case of an ideal gas the Hamiltonian is given by H = 1 m 5 N j=1 p j

6 the particles are confined to a volume V. Now β ρ canon = ( mπ )3N/ e β N m j=1 p j N j=1 χ V (x j ) V The energy is i.e., γ = 3/. we see that β = 1 kt Z = ( mπ β )3N/ V N. U = H canon = 3 N 1 β is consistent. Further using () (3) S(ρ canon ) = 3 Nk + kn[3 log(mπ ) + log V ] β recalling that we have to write S in terms of the variable U, V, N we can eliminate β using (3) get S = 3 kn log U + kn log V 3 kn log N + 3 kn(1 log((4mπ/3)). The last term is just an additive constant is irrelevant for thermodynamics. All the other terms fit our expectation except the the penultimate term. We have the serious problem that the entropy is not extensive. We should really have (1 + 3 )kn log N instead of 3 kn log N. Note that we can get out of this quagmire of replace the partition function by Z = 1 βh N! since this term will yield an addition term of the order of N log N for the entropy that would render the entropy extensive. This problem does not show up in quantum statistical mechanics! The way out in the classical case is to undest that the particles are indistinguishable. The configuration (x 1, x, x 3,,x N ) cannot be distinguished from the configuration (x, x 1, x 3,,x N ). In other words we have to consider the density ρ canon = e βh βh χ S N 6

7 where χ SN is the characteristic function of a fundamental domain of the permutation group. Such a set is somewhat complicated to describe when the underlying space is R 3 but since its contribution to the partition function is 1/N! times the whole one. We shall see that the partition function contains in some sense all the information of our thermodynamic system. If the underlying space is one dimensional then a fundamental domain is given by the region x 1 < x < < x N 1 < x N. In this formalism the natural variables are not U, V, N but T, V, N. The way out of that is to perform a Legendre transform. One performs this best by starting from the internal energy U(S, V, N). Solve the equation S = T to get S(T, V, N). Then consider the Free Energy function F (T, V, N) = U(S(T, V, N), V, N) T S(T, V, N) note that Returning to () we see that F T = S, F V = p F N = µ. F = 1 β log Z, where Z = 1 N! βh. Thus, the log of the partition function yields the free energy. 7