The Partition Function

Any good undergrad statistical mechanics class analyzes the partition function. However, Russ Hobbie and I don’t introduce the partition function in Intermediate Physics for Medicine and Biology. Why not? It’s a long story.

Russ and I do discuss the Boltzmann factor. Suppose you have a system that is in thermal equilibrium with a reservoir at absolute temperature T. Furthermore, suppose your system has discrete energy levels with energy E_i, where i is just an integer labeling the levels. The probability P_i of the system being in level i, is proportional to the Boltzmann factor, exp(–E_i/k_BT),

P_i = C exp(–E_i/k_BT),

where exp is the exponential function, k_B is the Boltzmann constant and C is a constant of proportionality. How do you find C? Any probability must be normalized: the sum of the probabilities must equal one,

Σ P_i = 1 ,

where Σ indicates a summation over all values of i. This means

Σ C exp(–E_i/k_BT) = 1,

or

C = 1/[Σ exp(–E_i/k_BT)] .

The sum in the denominator is the partition function, and is usually given the symbol Z,

Z = Σ exp(–E_i/k_BT) .

In terms of the partition function, the probability of being in state P_i is simply

P_i = (1/Z) exp(–E_i/k_BT) .

Here is what Daniel Schroeder writes in his excellent textbook An Introduction to Thermal Physics,

The quantity Z is called the partition function, and turns out to be far more useful than I would have suspected. It is a “constant” in that it does not depend on any particular state s [he uses “s” rather than “i” to count states], but it does depend on temperature. To interpret it further, suppose once again that the ground state has energy zero. Then the Boltzmann factor for the ground state is 1, and the rest of the Boltzmann factors are less than 1, by a little or a lot, in proportion to the probabilities of the associated states. Thus, the partition function essentially counts how many states are accessible to the system, weighting each one in proportion to its probability.

To see why it’s so useful, let’s define β as 1/k_BT. The Boltzmann factor is then

exp(–βE_i)

and the partition function is

Z = Σ exp(–βE_i) .

The average energy, E>, is

E> = [Σ E_i exp(–βE_i)]/[Σ exp(–βE_i)] .

The denominator is just Z. The numerator can be written as the negative of the derivative of Z with respect to β, dZ/dβ (try it and see). So, the average energy is

E> = – (1/Z) dZ/dβ .

I won’t go on, but there are other quantities that are similarly related to the partition function. It’s surprisingly useful.

Is the partition function hidden in IPMB? You might recognize it in Eq. 3.37, which determines the average kinetic energy of a particle at temperature T (the equipartition of energy theorem) It looks a little different, because there’s a continuous range of energy levels, so the sum is disguised as an integral. You can see it again in Eq. 18.8, when evaluating the average value of the magnetic moment during magnetic resonance imaging. The partition function’s there, but it’s nameless.

Why didn’t Russ and I introduce the partition function? In the Introduction of IPMB Russ wrote: “Each subject is approached in as simple a fashion as possible. I feel that sophisticated mathematics, such as vector analysis or complex exponential notation, often hides physical reality from the student.” Like Russ, I think that the partition function is a trick that makes some equations more compact, but hides the essential physics. So we didn’t use it.