Lecture 17

Equipartition Theorem

The Boltzmann distribution is plug and play. There are just two ingredients:

  1. What are the energies of each configuration/state.
  2. How do we count configurations

In certain examples with discrete variables, it is not ambiguous. Each configuration has an energy, and we just sum over all possible configurations

Z = \sum_{s} e^{ - \beta E(s)}

This is in fact all there is in a fully quantum treatment. In that case, each configuration is called a quantum state, and it is a discrete sum because energy is quantized.

But we still get a lot from classical mechanics. Basically all that we learned about kinetic theory will come from classical thinking.

Let’s apply the Boltzmann distribution to a particle moving in space. The energy, which we know from classical mechanics, is kinetic plus potential:

E = \frac{1}{2} m |{\bf v}|^{2} + U({\bf r}) = \frac{1}{2m} | {\bf p}|^{2} + U({\bf r}) which can be expressed here in terms of momentum or velocity.

We can define the partition function for this as well:

Z = \sum_{{\bf r}, {\bf p}} e^{ - \beta E({\bf r}, {\bf p})} But obviously there is an issue with the sum - it is not discrete! Rather, we must replace the sum over configurations with an integral over the configuration, which in this case are the 3D (real-valued) position and momentum ({\bf r}, {\bf p}). Therefore, the partition function should be written

Z = \int d^{D} {\bf p} \, d^{D} {\bf r} e^{ - \beta E({\bf r}, {\bf p})} Next, we need to make this dimensionless, because the partition function is not supposed to have any dimensions. But each component of the integration measure dp_{x} dr_{x} has units [{\rm mass}] \times [ {\rm meters} ^{2}] / [ {\rm second}], which is incidentally the units of angular momentum, or what is called the action in classical mechanics (and path integral formulations of quantum mechanics). Therefore, we have to introduce a constant which has these units. The right choice turns out to be the Planck constant h. We couldn’t have known this ahead of time, but comparing it with the fully quantum calculation (which is more complicated) confirms that this is the right choice. Since the integral is D dimensional, we thus need a normalizing factor of h^{D} to make this dimensionless.

Free Particle in a Box:

Let’s start with the free particle in a box. U = 0 within a finite volume, and U = \infty outside this volume. This will make the partition function

Z =\frac{1}{h^{D}}\left( \int d^{D} {\bf x} \right) \left( \int d^{D} {\bf p} e^{ - \beta |{\bf p}|^{2}/2m}\right)

The first term becomes just the volume V: \int d^{D} {\bf x} = V

The second term is more complicated, so let’s spend some more time on it. The first thing to do is to rescale variables so that we make the \beta and m dependence completely explicit. This is accomplished by defining

{\bf u} = \sqrt{\frac{\beta}{2m}} {\bf p} or in components

u_{1} = \sqrt{\frac{\beta}{2m}} p_{1}, \quad u_{2} = \sqrt{\frac{\beta}{2m}} p_{2}, ...., u_{D} = \sqrt{\frac{\beta}{2m}} p_{D} (if it helps, just assume D = 3, consistent with your reality).

Then the integration measure becomes

d^{D}{\bf p} = dp_{1} dp_{2}..., dp_{D} = \left( \sqrt{ \frac{2 m}{\beta}} du_{1}\right) ...\left( \sqrt{ \frac{2 m}{\beta}} du_{D}\right) = \left( \frac{2m}{\beta}\right)^{D/2} d^{D}{\bf u}

Using the result for the spatial integral, and using this change of variables, the partition function becomes

Z = \frac{V}{h^{D}} \left( \frac{2 m}{\beta}\right)^{D/2} \left(\int d^{D}{\bf u} e^{ - | {\bf u}|^{2}}\right)

The important thing to note here is that the integral in parentheses does not depend on any important physical variables - it is just a number! And in this particular instance, it is possible to evaluate, since it is a product of D gaussian integrals:

\int d^{D}{\bf u} e^{ - | {\bf u}|^{2}} = \int du_{1} e^{ - u_{1}^{2}} \int du_{2} e^{ - u_{2}^{2}} ... \int du_{D} e^{ - u_{D}^{2}} = \pi^{D/2} Since \int_{-\infty}^{\infty} du e^{ - u^{2}} = \sqrt{\pi}.

So the single-particle partition function of a free particle in a box of volume V is

Z = \frac{V}{h^{D}} \left( \frac{2 m \pi}{\beta}\right)^{D/2}

From this, we are able to get the average energy of a single particle:

\langle E \rangle = - \frac{\partial}{\partial \beta} \ln Z = \frac{D}{2} \frac{1}{\beta} = \frac{D}{2} k T

Thus the particle has an average energy of kT/2 per dimension, which are the translational degrees of freedom.

Exercises:

Particle in a Gravitational Field:

Consider the case of a particle confined to a box in the x and y directions, so that 0 \le x \le L and 0 \le y \le L, but confined by a gravitational field in the z direction. The gravitational potential energy is then U = m g z, and 0 \le z \le \infty. Let us find the average thermal energy of the particle in such a field.

The partition function will factorize into separate spatial and momentum parts. Let’s write this:

Z(\beta) = Z_{p}(\beta) Z_{x}(\beta) Where

Z_{x}(\beta) =\int dx dy dz e^{ - \beta m g z} = L^{2}\int_{0}^{\infty} dz e^{ - \beta m g z} I will rescale the integral (though I could also just evaluate it) to find the \beta dependence. That is, after all, all I need to know to compute the average energy. Define u = \beta m g z. then dz = du/(\beta m g), and

Z_{x}(\beta) = L^{2} \frac{1}{\beta m g} \int_{0}^{\infty} e^{ - u} du = \frac{1}{\beta} Z_{x}(1) If you didn’t know how to evaluate this integral, it wouldn’t matter. It is captured in the term Z_{x}(1).
Next, for the momentum part, we have from before

Z_{p}(\beta) = \frac{1}{\beta^{D/2}} Z_{p}(1)

Therefore combining these, we can write the partition function as

Z(\beta) = \frac{1}{\beta^{D/2 + 1}} Z(1)

and

\ln Z(\beta) = - \left( \frac{D}{2} + 1\right) \ln \beta + \ln Z(1)

The average energy is

\langle E \rangle = - \frac{\partial}{\partial \beta} \ln Z(\beta) = \left( \frac{D}{2} + 1\right) \frac{1}{\beta} = \left( \frac{D}{2} + 1\right) k T

Particle in Polynomial Potential (Anharmonic oscillator):

Consider now a 1D particle trapped in a potential U(x) = a x^{n}. What is the average thermal energy?

The partition function now is

Z = \int \frac{d x dp}{h} \exp \left( - \beta \left( \frac{p^{2}}{2m} + a x^{n}\right) \right) = \frac{1}{h} Z_{x}(\beta) Z_{p}(\beta) where

Z_{x}(\beta) = \int_{-\infty}^{\infty} dx \exp \left( - \beta a x^{n} \right), \quad Z_{p}(\beta) = \int_{-\infty}^{\infty} dp \exp \left( - \beta \frac{p^{2}}{2m} \right)

To extract the \beta dependence of Z_{x}, we must define a new variable X^{n} = \beta a x^{n}, or X = ( \beta a)^{1/n} x, then dx = \left( \beta a \right)^{-1/n} dX and

Z_{x}(\beta) = \int \frac{dX}{(\beta a)^{1/n}} e^{ - X^{n}} = \frac{1}{\beta^{1/n}} Z_{x}(1) It’s anyone’s guess how to evaluate this integral, but we know the important part, which is how it scales with \beta. Putting it together, I get

Z(\beta) = \frac{1}{\beta^{1/n + 1/2}} Z(1)

And the average energy is

\langle E \rangle = \left(\frac{1}{2} + \frac{1}{n} \right) k T

Notice that we recover the equipartition result for quadratic degrees of freedom when n = 2.

Ultra-relativistic gas

If we have a relativistic dispersion in D dimensions:

E= c | {\bf p}| Find the equation of state in the form of a relation between pressure, volume, and total internal energy.

The partition function is

Z = \int \frac{d^{D}{\bf p} d^{D} {\bf x}}{h^{D}} \exp \left( - \beta c p\right) = V \int \Omega(D) p^{D-1} dp e^{ - \beta c p} = V \Omega(D) \frac{1}{(\beta c)^{D}} \int du \, u^{D-1} e^{ - u}

Which means Z_{1}(\beta) = \frac{1}{\beta^{D}} Z_{1}(1) and Z_{N}(\beta) = \beta^{ - N D} Z_{N}(1)

The average energy U = \langle E \rangle is

U = - \partial_{\beta} \ln Z = - \partial_{\beta} \left( - DN \ln (\beta)\right) = DN k T Next, defining F = - k T \ln Z_{N}

We get the pressure from

P = - \frac{\partial}{\partial V} F = \frac{Nk T}{V} So that

PV = N k T But we also know that N k T = U/D, so putting this together

PV = \frac{1}{D} U

The familiar result for an ultrarelativistic gas comes from setting D = 3.