Unit 2

The canonical ensemble.

2.1 The canonical ensemble

Let’s consider two systems A1 and A2 in pure thermal contact, but the combined
system A1 ∪ A2 is isolated. We discussed in Section 1.6.4 of Unit 1 how to obtain
the probability distribution of one of the systems, for example, A1 , from the micro-
canonical ensemble for the combined system. For example, the marginal probability
density of the system A1 to be in a microstate (q, p) ≡ (q1 , . . . , qf1 , p1 , . . . , pf1 ), where
f1 is the number of degrees of freedom of A1 , is given by Eq. (1.96), i.e.
1
ρ1 (q, p) = Qg Ω2 (E − H1 (q, p)) (2.1)
hf1 ( i=1 Ni,1 !) Ω(E)

where we have introduced the corrections due to the indistinguishability between Ni,1
particles of the same species i (i = 1, . . . , g) in the system A1 to avoid the Gibbs’s
paradox, and Ω is defined by Eq. (1.147). On the other hand, the probability density
for E1 reads
Ω1 (E1 )Ω2 (E − E1 )
ω(E1 ) = (2.2)
Ω(E)
We know that ω(E1 ) has a very sharp maximum at E1 = Ẽ1 , which is obtained from
the condition  
∂ ln Ω1 (E1 ) 
 ∂ ln Ω2 (E2 ) 
= (2.3)
∂E1 ∂E2
 
 
E1 =Ẽ1 E2 =E−Ẽ1

or equivalently that β1 = β2 = β, where β = 1/(kB T ). Thus, this condition is

equivalent to the fact that the temperatures T1 and T2 of both systems must be

1
The canonical ensemble. 2

the same T1 = T2 = T . Now let’s consider that A2 is much larger than A1 , in the
sense that the number of degrees of freedom of A2 , f2 , is much larger than f1 . Since
Ω ∼ E νf , where ν is a number of order of 1, then the condition Eq. (2.3) leads to
the following relationship

ν1 f 1 ν2 f 2 Ẽ1 f1
∼ → ∼ (2.4)
Ẽ1 Ẽ2 Ẽ2 f2

Thus, if f2  f1 , then Ẽ2 ≈ E, where E is the total energy of the combined

system. As a consequence, the microstates of system A1 which contribute most to
the marginal probability density are those that satisfy

H1 (q, p)  H2 (Q, P ) ≈ H1 (q, p) + H2 (Q, P ) = E (2.5)

This fact allows us to Taylor expand ln Ω2 (E − H1 (q, p)) around H1 (q, p) = 0.

Truncating the expansion in the first order term, we get that
 
∂ ln Ω2 (E2 )
ln Ω2 (E −H1 (q, p)) ≈ ln Ω2 (E)− H1 (q, p) = ln Ω2 (E)−βH1 (q, p)
∂E2 E2 =E
(2.6)
where β is defined for A2 , but by condition Eq. (2.3) it is also the value for the
system A1 . Strictly speaking, the definition of β for the A2 system corresponds to
the evaluation of the derivative at Ẽ2 , but this value does not differ significatively
(in macroscopic terms) from E, so we can neglect the effect of the variation of E2 in
the evaluation of β2 = β. For this reason, the system A2 is said to act as a thermal
bath, thermal reservoir or thermostat for A1 . However, the converse is not true,
and if the same A2 system is put into pure thermal contact with other system much
larger than it, A2 will not act as a thermal bath for this new system. In this sense,
the concept of thermal bath is a relative one which makes sense when comparing
two systems. Substituting Eq. (2.6) into Eq. (2.1), we get that
1 Ω2 (E) −βH1 (q,p) 1 1 −βH1 (q,p)
ρ1 (q, p) = Qg e ≡ f1 Q g e (2.7)
hf1 ( i=1 Ni,1 !) Ω(E) h ( i=1 Ni,1 !) Z

where e−βH1 is the Boltzmann factor and Z is the so-called (canonical) partition
function, which can be obtained from the normalization condition of Eq. (2.7) as
Z
1
Z ≡ Z(T, X1 , N1 ) = f1 Qg dqdpe−βH1 (q,p) (2.8)
h i=1 Ni,1 !
If
P the system has a discrete energy spectrum, pi,1 = exp(−βEi )/Z, where Z =
i exp(−βEi ) and the sum is over all the microstates of the system A1 . We see that
The canonical ensemble. 3

the partition function depends on the temperature T , as well as external extensive

parameters X1 = {X1,1 , . . .} and the number of particles N1 ≡ {N1,1 , . . . , Ng,1 }
of system A1 . With these definitions, the statistical properties of system A1 only
depend on A2 via the factor β. As a consequence, we will focus only on the system
A1 , which will be in thermal equilibrium with the thermal bath characterized by
a temperature T , but keeping constant the values of X1 and N1 of the system A1 .
Hereafter we will drop the subscript 1 which identifies the system of interest. So,
the macrostate of the system A is determined by the values of T , X and N , and the
corresponding Gibbs’s ensemble for this situation is the canonical ensemble. The
set of microstates compatible with this macrostate are those with the same values
of X and N than the macrostate regardless their microscopic energy. On the other
hand, the probability distribution of microstates is given by Eq. (2.7), which depends
on the additional parameter β = 1/(kB T ). Note that, as it is a function of the energy
of our system, it is a stationary solution of Liouville’s equation (as the probability
distribution of the microcanonical ensemble). In this ensemble, the averages of a
microscopically defined quantity A(q, p) (which provide their macroscopic values for
the canonical macrostate) are given by
Z Z
1
A = dqdpA(q, p)ρ(q, p) = f Q g dqdpA(q, p)e−βH(q,p)
h ( i=1 Ni !) Z
−βH(q,p)
R
dqdpA(q, p)e
= R (2.9)
dqdpe−βH(q,p)

2.2 Evaluation of thermodynamic properties in

the canonical ensemble
Examples of properties that can be evaluated in this way are the probability density
of the energy of the system

Ω(E)e−βE
ω(E) ≡ δ(H(q, p) − E) = (2.10)
Z
where Ω(E) is the (microcanonical) number of microstates associated to the energy
E. Note that this expression is the same as Eq. (2.2) when expression Eq. (2.6)
is substituted into it. An interesting connection between Ω and Z comes from the
normalization condition of Eq. (2.10)
Z ∞
Z= dEΩ(E)e−βE (2.11)
E0
The canonical ensemble. 4

From a mathematical point of view, this relates Z with the Laplace transform of
Ω(E)
Z(β) = e−βE0 L(Ω̃) (2.12)
where Ω̃(E) = Ω(E + E0 ). So, Ω can be obtained in terms of the inverse Laplace
transform of Z (as a function of β).
If A is macroscopic (and so its energy E), the integrand of Eq. (2.11) has a sharp
maximum at the energy E = Ẽ, since Ω(E) ∼ E νf and the Boltzmann factor is a
rapidly decreasing function of the energy. So, we can approximate Eq. (2.11) as

Z = Ω(Ẽ)e−β Ẽ nσE (2.13)

where n is of order of unity. The value of Ẽ is obtained from the maximum condition

∂ ln Ω(E) 
=β (2.14)
∂E 

E=Ẽ

which we note that it is equivalent to fix the microcanonical temperature of the

system to the bath temperature.
The internal energy of the system is defined as
Z
1
E ≡ H(q, p) = f Q g dqdpH(q, p)e−βH(q,p) (2.15)
h ( i=1 Ni !) Z

This property can be evaluated by noticing that

∂e−βH(q,p)
Z Z
∂Z 1 1
= f Qg dqdp = − f Qg dqdpH(q, p)e−βH(q,p)
∂β h ( i=1 Ni !) ∂β h ( i=1 Ni !)
(2.16)
Thus, Eq. (2.15) can be recast as

1 ∂Z ∂ ln Z
E=− =− (2.17)
Z ∂β ∂β
Unlike in the microcanonical ensemble, the internal energy is not constant but a
fluctuating quantity, with a probability distribution given by Eq. (2.10). We ar-
gued above that this probability distribution has a sharp maximum, which must be
equal to E macroscopically (this may not be true if the system is not macroscopic,
since in general the mean and the most-probable value or mode of a probability
distribution do not coincide). We can go further and evaluate the variance of E,
The canonical ensemble. 5

2
σE2 = (H(q, p) − E)2 = H(q, p)2 − E . To calculate this quantity, we evaluate the
derivative of E with respect to β
" 2 #
∂ 2 ln Z 1 ∂ 2Z
  
∂E ∂ 1 ∂Z 1 ∂Z
=− =− =− − 2 (2.18)
∂β ∂β 2 ∂β Z ∂β Z ∂β 2 Z ∂β

2
The second term in the last bracket is easily identified with E . On the other hand,
the first term can be obtained by differentiating Eq. (2.16)

∂ 2Z ∂e−βH(q,p)
Z
1
= − Qg dqdpH(q, p)
∂β 2 hf ( i=1 Ni !) ∂β
Z
1
= Qg dqdp(H(q, p))2 e−βH(q,p) (2.19)
hf ( i=1 Ni !)

Thus, the first term in the last bracket of Eq. (2.18) can be identified as H(q, p)2 .
So,
2 ∂E ∂E
σE2 = H(q, p)2 − E = − = kB T 2 = kB T 2 CX,N (2.20)
∂β ∂T
where CX,N is the heat capacity for X and N constants. Note that CX,N ≥ 0 since
the variance of the energy is a positive quantity. As the heat capacity is an extensive
property, the variance of the energy scales with√ the number of degrees of freedom
f , so its standard deviation σE is of order of f . On the other hand, the internal
energy E, which is also an extensive property, is proportional to f . So, the ratio
between σE and E scales as √
σE f 1
∼ ∼√ (2.21)
E f f
This result state that the mode (i.e. the most probable value of the energy) Ẽ will
coincide macroscopically with the mean value E, since their difference is of order of
σE  E. In this sense, for macroscopic systems the canonical ensemble probability
distribution converges to the microcanonical probability distribution for an energy
E = E, which on the other hand it is the energy at which the (microcanonical)
temperature is the same as the one in the thermal bath. This fact is a special case
of the equivalence between equilibrium ensembles: the macroscopic properties ob-
tained as averages of microscopically defined properties have the same values if they
are evaluated in different ensembles, provided the parameters of these ensembles are
selected to be consistent each other (for example, the energy in the microcanonical
ensemble must have a value at which the temperature is the same as the selected
value for the canonical ensemble). Although the equivalence is strictly true in the
The canonical ensemble. 6

thermodynamic limit, i.e. when f → ∞, it is an excellent approximation for macro-

scopic systems. However, this equivalence breaks down for the fluctuations. For
example, in the canonical ensemble the energy fluctuations are connected with the
heat capacity via Eq. (2.20), but in the microcanonical ensemble they vanish (as
the energy is constant).
Another quantities of interest are the generalized forces, which can be obtained
as
∂H
Yα ≡− (2.22)
∂Xα
which can also be obtained as a derivative of the partition function
1 1 ∂Z 1 ∂ ln Z
Yα = = (2.23)
β Z ∂Xα β ∂Xα
In the case that Xα is the volume V , its conjugated generalized force Y α is the
pressure p.
The results obtained up to know identify the partition function as a key quan-
tity to get the thermodynamic properties of the system in the canonical ensemble,
analogous to the number of microstates or the phase volume in the microcanonical
ensemble. In the next section we will explore in more detail its importance and its
connection with the Thermodynamics.

2.3 The partition function and its connection with

the Thermodynamics
In order to give a thermodynamic interpretation to the partition function, let’s see
how it changes when the temperature and the external extensive parameters vary
quasi-statically. Then
  X  ∂ ln Z 
∂ ln Z
d ln Z = dβ + dXα
∂β X α
∂X α β,X γ6=α
X
= −Edβ + β Y α dXα (2.24)
α

where in the last expression we used Eqs. (2.17) and (2.23). If we add d(βE) to
both sides of Eq. (2.24), we get that
X
d(ln Z + βE) = βdE + Edβ − Edβ + β Y α dXα
α
X dS
= β(dE + Y α dXα ) = (2.25)
α
kB
The canonical ensemble. 7

where we used the expression of the first law of the thermodynamics for reversible
processes. Thus, the entropy in the canonical ensemble can be defined as
S = kB (ln Z + βE) (2.26)
and from this expression
E − T S ≡ F = −kB T ln Z (2.27)
where F is the Helmholtz free energy. Note that F , as Z, is a function of T ,
N and X. This thermodynamic potential provides the maximum energy that can
be transformed into work along an isothermal path, since d¯W ≤ −dF . With the
definition Eq. (2.27), Eqs. (2.17), (2.23) and (2.26) can be recast as
   
∂F ∂F
E = F + TS , S = − , Yα =− (2.28)
∂T X,N ∂Xα T,N
in agreement with the thermodynamic relationships.
The expression Eq. (2.27) is consistent with the results in the microcanonical
ensemble. From Eq. (2.13), and taking into account that Ẽ ≈ E and that σE  E,
we obtain that
S
ln Z ≈ ln Ω(E) − βE = − βE (2.29)
kB
where we used the microcanonical expression for the entropy.
The additivity of F can be shown in a similar way as we did it for the entropy
in Unit 1. We will consider the following cases of interactions between two subsys-
tems A1 and A2 , with the combined system to be in contact with a thermal bath
characterized by a temperature T :
1. Pure thermal interactions, no matter transfer. In this case, the partition
function of the combined system can be written as
Z
1
Z = f1 +f2 Qg Qg dqdpdQdP e−β(H1 (q,p)+H2 (Q,P ) (2.30)
h ( i=1 Ni,1 !) ( i=1 Ni,2 !)
where we assumed that the Hamiltonian of the combined system can be ap-
proximated as
H(q, Q, p, P ) = H1 (q, p) + H2 (Q, P ) (2.31)
Now, we notice that we can factorize the integral in Eq. (2.30) as
 Z 
1 −βH1 (q,p)
Z = dqdpe
hf1 ( gi=1 Ni,1 !)
Q
 Z 
1 −βH2 (Q,P )
× dQdP e
hf2 ( gi=1 Ni,2 !)
Q

= Z1 × Z2 (2.32)
The canonical ensemble. 8

which is the factorization property of the partition functions of weakly in-

teracting systems. This expression is exact and valid for any non-interacting
systems, no matter how large they are. This result contrasts with the ex-
pression Ω(E) ≈ Ω1 (Ẽ1 )Ω2 (Ẽ2 ), which is only valid for macroscopic systems,
otherwise the composition rule must be used. This fact makes mathematically
simpler to study systems in the canonical ensemble than in the microcanon-
ical. Finally, this expression is also valid Rfor systems
P with a discrete energy
−f −1
spectrum (just by substituting h (N !) dqdp by i , i.e. the sum over the
microstates).
If we take the logarithm in both sides of Eq. (2.32) we get that

ln Z = ln Z1 + ln Z2 (2.33)

which yields to the additivity of the Helmholtz free energies F = F1 + F2 since

both subsystems must be at the same temperate in equilibrium.

2. General interaction, no matter transfer. By simplicity, we will consider

that A1 and A2 interact mechanically, so their volumes V1 and V2 can change,
but with the constraint that V = V1 + V2 is constant. Analogously to the
previous case, we can obtain the partition function of the combined system as
Z V
Z(T, V, N ) = dV Z1 (T, V1 , N1 )Z2 (T, V − V1 , N2 ) (2.34)
0

Now we consider that both subsystems are macroscopic. Then, the integrand
in this expression has a sharp maximum for V1 = Ṽ1 , so

Z(T, V, N ) ≈ Z1 (T, Ṽ1 , N1 )Z2 (T, V − Ṽ1 , N2 )nσV (2.35)

where n ∼ 1 and σV  Ṽ1 . The value of V1 is obtained from the condition

 
∂ ln Z1 (T, V1 , N1 )  ∂ ln Z1 (T, V2 , N1 ) 
= (2.36)
∂V ∂V2
 
 
V1 =Ṽ1 V2 =V −Ṽ1

which by Eq. (2.23) is equivalent to the condition of equal pressures p1 = p2 .

Thus, we now obtain that

ln Z(T, V, N ) ≈ ln Z1 (T, Ṽ1 , N1 ) + ln Z2 (T, V − Ṽ1 , N2 ) (2.37)

which again leads to the additivity of the Helmholtz free energies.

The canonical ensemble. 9

3. Open systems. Now the two systems can exchange particles, but the com-
bined system has constant number of particles Ni , for each species of identical
particles i = 1, . . . , g. In this case, the partition function of the combined
system can be written as
N1 Ng
X X
Z(T, X, {Ni }) = ... Z1 (T, X1 , {Ni,1 })Z2 (T, X1 , {Ni − Ni,1 })
N1,1 =0 Ng,1 =0
(2.38)
As in the microcanonical ensemble, the presence of the factorials of the number
of particles of each species in the definition of the partitionQ
functions remedi-
ate the Gibbs’s paradox (if they were not included, a factor i Ni !/[Ni,1 !(Ni −
Ni,1 )!] should be added in front of the integral, as it was done in the micro-
canonical ensemble). Now we assume that all the values of N are macroscopic.
Then, the terms which we are adding have a sharp maximum for {Ni } = {Ñi },
so we can approximate

Z(T, X, {Ni }) ≈ Z1 (T, X1 , {Ñi,1 })Z2 (T, X1 , {Ni − Ñi,1 })∆N1 . . . ∆Ng (2.39)

where ∆Ni  Ñi . Taking the logarithm of this expression, we get that

ln Z(T, X, {Ni }) ≈ ln Z1 (T, X1 , {Ñi,1 }) + ln Z2 (T, X1 , {Ni − Ñi,1 }) (2.40)

which confirms the additivity of the Helmholtz free energies. This expression,
however, is not true if any of the species has a microscopic number of particles.
An extreme case would be when all the particles are distinguishable, so g = N
and Ni = 1 for i = 1, . . . , g.
The values of Ñi are obtained from the maximum condition for the terms in
the sum Eq. (2.38). Treating these variables as continuous, this condition
reads
 
∂ ln Z(T, X1 , {Ni,1 }) 
 ∂ ln Z(T, X2 , {Ni,2 }) 
= (2.41)
∂Ni,1 ∂Ni,2
 
 
Ni,1 =Ñi,1 Ni,2 =Ni −Ñi,1

We define the chemical potential of species i, µi , as

   
∂ ln Z ∂F
µi = −kB T = (2.42)
∂Ni T,X,{Nj6=i } ∂Ni T,X,{Nj6=i }

so Eq. (2.41) implies that the chemical potentials of each species in both
subsystems must be equal in equilibrium.
The canonical ensemble. 10

2.4 The monoatomic three-dimensional ideal gas

As an example, we will consider a monoatomic ideal gas of N indistinguishable
particles enclosed in a three-dimensional volume V , in equilibrium with a thermal
bath of temperature T . The Hamiltonian of the system is
N
X |p |2 i
H(q, p) = (2.43)
i=1
2m

The partition function of the system is

|p |2 +...+|p |2
Z
1 1 N
Z = 3N d3 r1 . . . d3 rN d3 p1 . . . d3 pN e−β 2m (2.44)
h N!
We can factorize this expression as
 Z   Z 
1 1 |p |2 1 |p |2
3 3 1
−β 2m 3 3 N
−β 2m
Z= d r 1 d p1 e × ... × d r N d pN e (2.45)
N ! h3 h3

Notice that each term in brackets is equal (they differ just in the labelling of the
dummy integration variables), so we can write the total partition function as

ζN
Z= (2.46)
N!
where ζ is the partition function of a single particle. This expression can be easily
obtained as
Z ∞  Z ∞  Z ∞ 
βp2
Z Z
1 3 3 − 2mp2 V βp2
− 2mx − 2my
βp2
− 2mz
ζ = 3 d r d pe = 3 dpx e dpy e dpz e
h h −∞ −∞ −∞
 3
V 2πm 2 V
= 3 ≡ 3 (2.47)
h β Λ
√
where Λ ≡ h/ 2πmkB T is called the thermal de Broglie wavelength, and for the
integration in momenta we used the formula
Z ∞  
h −a2 x2 1 h+1
I(h) = dxx e = h+1 Γ (2.48)
0 2a 2
so
∞ ∞
Z Z r   r
βp2
− 2mx
βp2
− 2mx 2m 1 2πm
dpx e =2 dpx e = Γ = (2.49)
−∞ 0 β 2 β
The canonical ensemble. 11

Thus, the logarithm of the partition function can be expressed, for large N , as
   
V 3 3 2πm
ln Z = N ln − ln β + ln +1 (2.50)
N 2 2 h2

Note that, as in the microcanonical ensemble, the factor (N !)−1 solves the Gibbs’s
paradox, so the Helmholtz free energy F = −kB T ln Z is an additive quantity.
From the expression Eq. (2.50), we can get

• The equation of state. The pressure p can be obtained from Eq. (2.23) as
 
1 ∂ ln Z N
p= = → pV = N kB T (2.51)
β ∂V T,N Vβ

which is the equation of state of the ideal gas.

• The energy equation. The internal energy E can be obtained from Eq.
(2.17) as  
∂ ln Z 3N 3
E=− = = N kB T (2.52)
∂β V,N 2β 2
As a consequence, the heat capacity of the gas is
 
∂E 3
CV = = N kB (2.53)
∂T V,N 2

• The chemical potential. The chemical potential µ is obtained as

  " 
2
 32 #  
∂ ln Z N h N 3
µ = −kB T = kB T ln = kB T ln Λ
∂N T,V V 2πmkB T V
(2.54)

• The entropy. The entropy can be obtained from Eq. (2.26) as

   
V 3 3 2πkB m 5
S = kB (ln Z + βE) = N kB ln + ln T + ln + (2.55)
N 2 2 h2 2

Note that all these expressions are the same as those derived from the microcanonical
ensemble, which is a confirmation of the equivalence between equilibrium ensembles.
The canonical ensemble. 12

2.5 The generalized equipartition theorem

Let’s calculate the following averages in the canonical ensemble
∂H −βH
R
∂H dqdpxi ∂x j
e
xi = R (2.56)
∂xj dqdpe−βH

where x1 , . . . , xf denote q1 , . . . , qf and xf +1 , . . . , x2f represent p1 , . . . , pf . We focus

on the numerator of expression Eq. (2.56)

1 ∂e−βH
Z Z  
∂H −βH
dqdpxi e = dx1 . . . dx2f xi − (2.57)
∂xj β ∂xj

Integrating by parts, we get that

Z ∞ Z ∞
∂e−βH  −βH xj =+∞ −βH ∂xi
dxj xi = xi e x =−∞
− dx je (2.58)
−∞ ∂xj j
−∞ ∂xj

where we assume that all the coordinates can be extended from −∞ to ∞. This can
always be done for Cartesian coordinates (where the confinement of the system on a
volume V is due to an external potential which is infinite far away from the volume),
but there are other choices of generalized coordinates, as the angular variables, where
this cannot be done and the results which we will obtain here may not be applicable.
Now we note that the first term in Eq. (2.58) vanishes, since for any function of the
phase space which can represent a property of the system (as it is xi ), the condition

lim f (x1 , . . . , x2f )e−βH = 0 (2.59)

xj →±∞

must be satisfied, otherwise the average f (x1 , . . . , x2f ) would diverge. On the other
hand, ∂xi /∂xj = δi,j , where δi,j is the Kronecker delta. Substituting these expres-
sions into Eq. (2.58), and back to Eq. (2.57), the Eq. (2.56) can be recast as

∂H
xi = kB T δi,j (2.60)
∂xj

which is known as the generalized equipartition theorem. Due to the equivalence of

equilibrium ensembles, this result will be also valid in the microcanonical ensemble.
If we apply this result to the generalized momentum pi , we get that

∂H
pi = kB T (2.61)
∂pi
The canonical ensemble. 13

which is the equipartition theorem. The physical interpretation of this theorem is

the following: from Classical Mechanics, the Hamiltonian H is defined from the
Lagrangian L as
X X ∂H
H= pi q̇i − L = pi −L (2.62)
i i
∂pi
So,
1 X ∂H H +L (K + U ) + (K − U )
pi = = =K (2.63)
2 i ∂pi 2 2
where K is the kinetic energy and U the potential energy. So, each term in the left-
hand side of Eq. (2.63) can be regarded as the contribution to the kinetic energy of
the degree of freedom i, and by the equipartition theorem it has in average a value
kB T /2.
Finally, the application of this theorem to the generalized coordinate qi , i.e.
∂H
qi = kB T (2.64)
∂qi
is the so-called virial theorem, which takes its name from the virial, a quantity
introduced by Clausius, which has the following expression
1X 1 X ∂H
− qi ṗi = qi (2.65)
2 i 2 i ∂qi

The virial theorem then states that, in average, each term in the virial has a value
kB T /2.
Finally, we will apply the generalized equipartition theorem to some simple sit-
uations:
1. A system of N monodimensional independent harmonic oscillators.
In this case, the Hamiltonian can be written as
X p2 X1
i
H(q, p) = + Ki qi2 (2.66)
i
2m i i
2

where mi and Ki are the mass and the elastic constant of the oscillator i,
respectively. Applying both the equipartition and the virial theorem we get
that
∂H p2 p2 kB T
pi = i = kB T → i = (2.67)
∂pi mi 2mi 2
∂H Ki qi2 kB T
qi = Ki qi2 = kB T → = (2.68)
∂qi 2 2
The canonical ensemble. 14

So, the average energy of each harmonic oscillator will be

p2i Ki qi2
i = + = kB T (2.69)
2mi 2
and for the system of N oscillators
N
X
E= i = N kB T (2.70)
i=1

2. The three-dimensional monoatomic ideal gas. In this case, the Hamil-

tonian of the system only has the translational kinetic energy contribution
N  2 3N
p2iy p2iz p2j

X p ix
X
H=K= + + = (2.71)
i=1
2m 2m 2m j=1
2m

By the equipartition theorem

kB T
E = 3N (2.72)
2
which is the expression we obtained in the previous Section.
If we apply the virial theorem to the ideal gas, at first instance we get a
paradox, since ∂H/∂qi ≡ 0 but qi ∂H/∂qi = kB T . There are two ways to solve
this apparently contradictory result. First, the accesible volume V is large
but finite, so the values of the Cartesian coordinates are restricted and the
generalized equipartition theorem cannot be applied for the spacial coordinates
directly. However, there is another way to solve it that provides a physical
meaning to the virial. Let’s suppose that the gas is enclosed in a cubic box
of side L. The gas is confined due to a wall-fluid potential which only acts on
the particles which are close to the boundaries. This term does not affect the
macroscopic properties of the gas, so we can assume it vanishes except for the
fact that the accesible volume to the gas is the interior of the box. Now, we
can evaluate the following expression:
N  
X ∂H
zi = −L × (−FL ) − 0 × F0 = LFL (2.73)
i=1
∂zi

where F0 and FL are the modulus of the (mean) forces beween the gas and
the lower and upper walls, respectively. In this expression we assume that
The canonical ensemble. 15

the interactions between the gas and the walls act only on the particles on
the faces of the cube. A similar argument can be applied along the X and Y
directions. Now, the force that the wall exerts on the gas must be the same
as the force that the gas exerts on the wall in modulus, which is pA = pL2 ,
where p is the gas pressure. So, we get that

N  
3
X ∂H
pL = pV = zi = N kB T (2.74)
i=1
∂zi

which is the equation of state of the ideal gas.

3. The Brownian motion. Let’s consider a particle of mass M suspended in a

fluid in equilibrium at a temperature T . We neglect the effect of any external
field, so the only forces which act on the particle are exerted by the molecules
of the fluid. In equilibrium, and by symmetry reasons, we expect that

vx = vy = vz = 0 (2.75)

However, this result does not mean that the particle is not moving. In fact,
the equipartition theorem states that

1 3 3kB T
M v 2 = kB T → v 2 = (2.76)
2 2 M
So, the motion of the particle is random, with all the directions equally prob-
able, but the modulus of the velocity, also random, satisfies Eq. (2.76). This
motion is named after the botanist Robert Brown, who first observed it ex-
perimentally.