APPLICATION OF MAXWELL-BOLTZMANN STATISTICS

In (58) of the previous notes, we have the MB statistics or distribution

ns = Agsexp(-es/kT) (1)
This is only for a set of discrete energy levels e s. In this section, we shall see how (1)
can be applied to a variety of situations. For instance, how can we use this distribution for an
ideal gas which does not have truly discrete energy levels.

PHASE SPACE
Let us now consider an ideal gas which is monatomic and the only energy possessed
by each particle is the ke by virtue of its momentum p
e = p2/2m = (px2 + py2 + pz2)/2m (2)
where m is the mass, p is the total momentum magnitude and p x etc. are the
momentum components in the Cartician frame. In Statistical Mechanics, the space
containing statistical states is called the “phase space” , where
 = dxdydz×dpxdpydpz (3)
Thus, the phase space has six dimensions, three for the real space and three for the
momentum space. For an ideal gas of volume V, we know that the states are uniform
throughout this volume V and thus we can integrate the first (spatial) part
dxdydz = V (4)
Therefore the phase space element now only contains the momentum variable,
 = Vdpxdpydpz (5)
We now treat the gas as having a pseudo discrete energy levels system with energy
and momentum given in (2). In this case, we can replace g s and es in (1) by g and e.
Further, let B be the size of the phase space occupied by one state. The relationship
between the weight(degeneracy) g and the phase space element in (5) is
g = /B = (V/B)dpxdpydpz (6)
The MB distribution of (1) now becomes
dn = ns = A(V/B)dpxdpydpzexp(-e/kT) (7)
It should be noted that B is not known in MB classical statistics. However, as we
shall see later, B has a definite value in quantum statistics.

EXAMPLE 1, IDEAL GAS IN POLAR COORDINATES

For an ideal gas, p is isotropic which means that whatever the direction of the
momentum, (2) shows that the ke e only depends on the magnitude of the momentum.
e = p2/2m (8)

1
 Fig. 4.

For this non-directional problem, we do not consider the Cartesian components p x etc.
In this case, we replace dp x… by a spherical shell element as in fig. 4 with a mean radius p
and thickness dp in momentum space. The surface area of the shell is 4p2 and the volume is
therefore 4p2dp. The MB statistics from (7) is now given by
dn = (AV/B)4p2dpexp(-p2/2mkT) (9)
This should be regarded as the first step for the expression of dn, as p is the first
variable directly related to the phase space. Using the Mathematical technique of changing
variable as in (9) of previous notes, we shall be able to express these expressions in energy
space.
Mathematically (9) is often written in distribution function form, Thus
dn = n(p)dp = (AV/B)4p2exp(-p2/2mkT)dp (10)
2 2
or n(p) = (AV/B)4p exp(-p /2mkT) (11)

2
 EXAMPLE 2, DENSITY OF STATES IN MOMENTUM SPACE
Often, (9) is written in another form
dn = g(p)Aexp(-p2/2mkT)dp (12)
Comparing (11) and (12), we see that g(p) is just one part of the distribution function
n(p) and the other part that has been separated is “A” times the Boltzmann factor exp(-
p2/2mkT). Comparing (12) and (10), we have
g(p) = (V/B)4p2 (13)
g(p) is known as the density of states evaluated at the momentum magnitude p.
On a closer look at (12), the number of particles or systems with momentum range
between p and p + dp depends on three things, namely, the density of states at p, the
probability that these states are occupied (which is exp(-p2/2mkT)) and also the width of the
range dp. This viewpoint is equally applicable in quantum statistics later.

DENSITY OF STATES IN ENERGY SPACE

The result of (13) is easy to obtain as p is part of the phase space. In reality however,
the density of states, hence the MB distribution, is more amenable to measurements in e
space.
For this, we just have to change the variable from p to e as in (9) of previous notes
g(e) = g(p)/f’ (14)
e = p2/2m (15)
f’ = de/dp = p/m (16)
3/2 ½
g(e) = g(p)m/p = (V/B)2(2m) e (17)
The whole distribution for an ideal gas is
dn = g(e)Aexp(-e/kT)de = (V/B)2(2m)3/2e½Aexp(-e/kT)de (18)
Fig. 5 shows the density of states in both p and e space. Note that the former is  p2
and the latter is  square root of e. These are from (13) and (17) respectively.

3
Fig. 5.

4
 THE VALUE OF A FOR AN IDEAL GAS
We have already found the value of the multiplier b. We recall that the other
undetermined multiplier is
A = exp(a) (19)
For an ideal gas, A may be found by simply conserving the total number of molecules
N. This is achieved by integrating (18) for e from 0 to infinity, thus
N = dn = (AV/B)2(2m)3/2e½exp(-e/kT)de (20)
For the integration of this, see (2.55) of my supplementary notes. The value of A
from (20) is therefore
A = (NB/V)(2mkT)-3/2 (22)
This result shows us that A depends on m and T as well as N/V. From this, it is
important to point out that N/V is the particle number per unit volume, i.e. the concentration.
From this point on, we put
n = N/V (23)
-3/2
A = (nB)(2mkT) (24)
With this value of A and putting V = 1, the MB distribution for unit volume of the gas
in (18) now becomes
dn = n(e)de = 2n(kT)-3/2e½exp(-e/kT)de (25)
Note the following:
1. (25) is for unit volume of the gas.
2. The distribution (25) no longer contains the arbitrary parameter B which is the size
of one state in phase space.
3. n(e) is a strong function of T. In fact, more traditionally, the ideal gas MB
distribution of (25) is expressed in speed space v
where e = ½mv2 (26)
Using changing of variable as in (14),
n(v) = 4n(m/2kT)3/2v2exp(-mv2/2kT) (27)
This is shown in fig. 6 for different T. We see that the peak shifts to the high speed
(high energy) side as T increases. The speed corresponding to the peak of n(v) is known as
the most probable velocity vm which means that the sub-group n(vm)dv has the highest
population.
Another feature is about the height of the peaks. We note that the curve is more
spread out at high T. The conservation of n means that the area under the curves must be the
same. This implies that the higher T peaks must be lower in order to compensate for the
broadening of the distribution.

5
 Fig. 6. Maxwell distribution in speed space

EXAMPLE 3, DOPPLER EFFECT OF GAS MOLECULES EMITTING LIGHT

(CARTESIAN TREATMENT)
The MB distribution in the form of (25) or (27) is not very useful if one needs to find
the distribution of a single speed component, say vx along the x-direction. We really need to
go back and use the Cartesian treatment. With this technique, we can find the distribution in
vx.
Changing the variable again
px = mvx (27a)
For simplicity, let us write
x = vx etc. (28)
The energy in (2) becomes
e = ½m(x2 + y2 + z2) (29)
The distribution in (7) becomes
dn = n(x,y,z)dxdydz = (AV/B)m3exp(-mx2/2kT)dx
×exp(-my2/2kT)dy
×exp(-mz2/2kT)dz (30)
Students are advised very strongly to compare this Cartesian form with the polar form
in (10). In this form, the total distribution is simply the product of three separable Boltzmann
factors.
As an exercise, let us find the vx or x-direction distribution regardless the values of y
or z.
n(x)dx = [dndydz]dx = (AV/B)m3exp(-mx2/2kT)dx
×exp(-my2/2kT)dy
×exp(-mz2/2kT)dz (31)

6
 To find n(x) for all y and z, we integrate for all positive and negative values of y and z
velocities. This means that the two integrals are evaluated from minus infinity to plus infinity.
On integration, these integrals are actually constants. Thus (31) becomes
n(vx)dvx = n(x)dx = c(T)exp(-mvx2/2kT)dvx (32)
The temperature dependence of c(T) comes from A as in (24). This distribution is
shown in fig. 7. Notice that this is very different from fig. 6. Here, the peak of this new
distribution occurs at zero x velocity.

Fig. 7. 1D distribution function along the x-velocity.

Now consider an assembly of gas molecules emitting light of a well defined

wavelength 0 (e.g. that arising from atomic optical transition in a gas discharge lamp, in
which molecules are effectively moving light emitters. In this case, the wavelength of light
as observed by an external observer will suffer the Doppler Effect as shown in fig. 7a. If the
pressure of the gas is low, we can ignore collision effects and the simple Maxwell distribution
for independent particles may be used.

7
 Fig. 7a Observer receiving radiation from a moving gas molecule.

As we already know, the Doppler shift of wavelength only depends on the velocity
component along the line of sight, e.g. the x velocity component (x). Using again x for the
velocity component along the line of sight and taking 0 as the original (un-Doppler shifted)
wavelength, the Doppler shifted wavelength is given by
 = 0(1 – x/c) (32a)
where c is the velocity of light in vacuum.
From the above, we can see that the spectrum is no longer a delta function but will
now have a different  depending on x. Its value will now have a spread of values depending
on the spread of x. In other words, the intensity I() (called the spectrum) is now a broad
profile as in fig. 7b rather than a delta function.

Fig. 7b Spectral intensity for a spectral line of 0 subject to Doppler broadening.

The intensity for a given wavelength depends on how many molecules are associated
with the corresponding speed x, i.e. how many such emitters. Thus, from (32)
I(x)dx  n(x)dx or
I(x) = Dexp(-mx2/2kT) (32b)
where D does not contain x.
8
 This equation gives us an intensity function in x velocity space. To change this into
the spectrum in wavelength space, we again need to change the variable from x to .
On differentiation,
d/dx = -0/c (32c)
I()d = I(x)/(d/dx)d
= (-Dc/0)exp[-(mc2/202kT)(0 - )2]d (32d)
or I() = I0exp[-(mc /20 kT)(0 - ) ]
2 2 2
(32e)
The minus sign has been dropped as it only represents an inversion of axis when
changing from x to . I0 is the intensity of the peak in fig. 7b at the unshifted wavelength
position.
Experimentally, one would measure and plot the intensity as a function of wavelength.
The profile of fig. 7b and hence (32e) will provide experimental confirmation of the MB
distribution as the result tells us about the x velocity distribution of the moving light emitting
molecules.
Further, (32e) shows us that the intensity profile is a Gaussian function whose width
is directly related to the temperature T of the gas. This is a very convenient method for
measuring the absolute T of the hot gas.
we put  = d at which the intensity is equal to 1/e (0.37) of the maximum intensity I0.
The spectral line width is the range of wavelength around the peak with an intensity greater
than or equal to 1/e (0.37) of the maximum intensity.
0 - d = w = half line width (32f)
Thus,
line width = 2w = 2×(0 - d) (32g)
We now write
2 = mc2/202kT (32h)
Therefore
I(d) = I0/e = I0exp[-2(0 - d)2] = I0exp(-2w2) (32i)
or e = exp(2w2) (32j)
Thus
w = 1/ (32k)
From (32g), the line width is thus given by
2w = 2/ = (8kT/mc2)½0 (32l)
This result shows that the doppler broadening of the spectrum is proportional to the
square root of the absolute temperature of the gas and inversely proportional to the square
root of the mass m of the molecules.

EXAMPLE 4, MOST PROBABLE VELOCITY AND AVERAGE PROPERTIES

The most probable velocity (speed) vm is corresponding to the peak of the distribution
function in fig. 6. We simply set

dn(v)/dv = 0 (33)
where n(v) was given in (27). The value is
vm = (2kT/m) ½ (34)
For the molecules between v and v+dv, the average value of any property function x(v) is
x = x(v)dn/dn = x(v)n(v)dv/n(v)dv (35)
Note that the underline denotes averaging.
This average is only for the subgroup in the range dv. If we want to find the average over all
the molecules, we need to integrate for all speed from 0 to infinity. Thus
x = x(v)n(v)dv/n(v)dv = x(v)n(v)dv/n (36)
In fact, x(v) can be any function of v, e.g. the speed v or the speed square v2. It is left to the
students as an exercise to show the following:
v = (8kT/m)½ (37)

9
(v2) = 3kT/m (38)
Note that
1. (38) is consistent with the average ke in kinetic theory
e = ½(v2)m = 3kT/2 (39)
Thus, the ke is derived from the average of the speed square rather than square of the average
speed.
2. (37) and (38) show that
v2 <> (v2) (40)
3. vm is again different from either of the average properties in (40).

THE PARTITION FUNCTION

Summing over all the energy levels for (1)
n = Agsexp(-es/kT) = AZ (41)
where we have put
Z = gsexp(-es/kT) (42)
Z is known as the partition function. As we shall see, it contains much
thermodynamical information enabling us to find a number of parameters.

ENTROPY AND THE PARTITION FUNCTION

You recall that, after applying Stirling’s approximation, the weight w was given in
(21). Using the equal sign for large numbers, we get
logw = nlogn + [nsloggs - nslogns] (43)
On re-arranging, we have
logw = nlogn + nslog(gs/ns) (44)
Taking natural log of the MB distribution (1) and re-arranging gives
log(gs/ns) = es/kT - logA (45)
We now have
logw = nlogn + [(nses/kT) - nslogA] = nlog(n/A) + e/kT (46)
where we have put
n = ns and e = nses (47)
From (41)
n/A = Z (48)
We are now in the position of writing the entropy
S = klogw = knlogZ + e/T (49)
This shows us that the partition function Z gives direct value about the entropy.
Looking at its definition in (42), it simply sums up all the information about the energy levels
and the associated states.

THE FREE ENERGY F

From Thermodynamics, F is defined as
F = U - TS (50)
but in Statistical Mechanics, we use e instead of U, thus
F = e - TS (51)
Re-writing (49) gives
-nkTlogZ = e - TS (52)
Hence from (51) and (52), we have
F = -nkTlogZ (53)
Again showing the usefulness of the partition function.

FURTHER INFORMATION FROM Z

From the Thermodynamics notes, we have
P = -(F/V)T (54)
S = -(F/T)V (55)
10
Using the value of F in (53)
P = nkT[logZ/V]T (56)
S = [(nkTlogZ)/T]V (57)
We now see that Z can give us quite a number of thermodynamical parameters.

Last updated
22 January 2015

11