Canonical Ensemble
Canonical Ensemble
Canonical Ensemble
Contents
Outline
In this chapter, we will establish the equilibrium statistical distribution for systems
maintained at a constant temperature T , through thermal contact with a heat bath.
The resulting distribution is also called Boltzmanns distribution.
The canonical distribution also leads to definition of the partition function and an
expression for Helmholtz free energy, analogous to Boltzmanns Entropy formula.
We will study energy fluctuation at constant temperature, and witness another fluctuationdissipation theorem (FDT) and finally establish the equivalence of micro canonical
ensemble and canonical ensemble in the thermodynamic limit. (We first met a manifestation of FDT in diffusion as Einsteins relation.)
Reading Assignment: Reif 6.1-6.7, 6.10
Temperature
For an isolated system, with fixed N number of particles, V volume, E total energy, it
is most conveniently described by the microcanonical (N V E) ensemble, which is a uniform
distribution between two constant energy surfaces.
const
E H({qi }, {pi }) E + E
mc ({qi }, {pi }) =
(1)
0
otherwise
Statistical mechanics also provides the expression for entropy S(N, V, E) = kB ln .
In thermodynamics, S(N, V, E) can be transformed to a more convenient form (by Legendre
transform) of Helmholtz free energy A(N, V, T ), which correspond to a system with constant
N, V and temperature T .
Q: Does the transformation from N, V, E to N, V, T have a meaning in statistical
mechanics?
A: The ensemble of systems all at constant N, V, T is called the canonical N V T ensemble.
Q: What is the probability distribution function for the canonical ensemble? Is it uniform?
A: The distribution is not uniform. It is called Boltzmanns distribution, which we will
develop below.
Q: How do we define temperature in statistical mechanics?
A: In thermodynamics, temperature is established through the zeroth law transitivity
of thermal equilibrium.
This means that temperature is a property that emerges in the thermodynamic limit,
N , where N is the number of particles. The existence of temperature is
independent of the type of the material (e.g. gas, liquid, or solid).
On the other hand, a system with say, N < 10, particles does not have a well defined
temperature (in the usual thermodynamic sense), unless it is in thermal contact with
another system, i.e., a thermostat, which has a large number of particles and a welldefined temperature.
The thermostat is considered to be many many times larger than the system of interest, so that the heat exchange with the system has a negligible effect on the energy per
particle and temperature of the thermostat.
Through thermal contact with the thermostat, the system will eventually settle down
to an equilibrium state with temperature equal to that of the thermostat.
The only requirement for the thermostat is that it is much larger than the system of
interest. You could imagine the rest of the universe as the thermostat. The only
problem there is that the universe is not in thermal equilibrium.
For convenience, the ideal gas is usually used as a model for the thermostat (because its
analytic expression for S(N, V, E) is known). But this is not necessary. The property
of the system of interest will not change if we use a different type of thermostat.
Boltzmanns distribution
Because the system can exchange heat with the thermostat, its total energy is no longer
conserved. Hence, we should no longer expect its energy to be confined as
E H({qi }, {pi }) E + E
Instead, any energy is allowed.
Q: How can we obtain the equilibrium density distribution c ({qi }, {pi }) for the canonical
ensemble?
3
A: The key idea is that, the (system + thermostat) can be considered an isolated system,
and can be described by the microcanonical ensemble.
Let H be the Hamiltonian of the system and HB be the Hamiltonian of the heat bath.
H + HB = E is conserved. Hence the system and heat bath together can be described by
the microcanonical ensemble when they have reached thermal equilibrium.
const
E H({qi }, {pi }) + HB ({qiB }, {pB
B
B
i }) E + E
mc ({qi }, {pi }; {qi }, {pi }) =
0
otherwise
(2)
B
B
where {qi }, {pi } represent the systems degrees of freedom, and {qi }, {pi } represent the heat
baths degrees of freedom. If we only want to know the statistical distribution concerning
the systems degrees of freedom, all we need to do is to integrate out the degrees of freedom
corresponding to the heat bath.
c ({qi }, {pi }) =
Z 3N
YB
B
B
dqjB dpB
j mc ({qi }, {pi }, {qi }, {pi })
(3)
j=1
This is very similar to the study of momentum distribution of one gas molecule in a gas tank
of N molecules (see Microcanonical ensemble notes).
({qi }, {pi }) is proportional to the number of ways the thermostat molecules can rearrange
(4)
Recall that the entropy SB of the heat bath is a function of its energy EB .
EB = E H({qi }, {pi })
SB (EB ) = kB ln B (EB )
(5)
H({qi }, {pi })
(6)
SB (EB ) SB (E)
EB
The higher order terms are neglected.
SB
We recognize that T1 E
, where T is the temperature of the thermostat. T must also
B
be the temperature of the system at thermal equilibrium. Therefore,
H({qi }, {pi })
SB (EB ) = const
T
SB (EB )
H({qi }, {pi })
B (EB ) = exp
= const exp
kB
kB T
H({qi }, {pi })
c ({qi }, {pi }) = const exp
kB T
(7)
(8)
(9)
This is the canonical distribution, also called Boltzmanns distribution or Boltzmanns law.
1
H({qi }, {pi })
c ({qi }, {pi }) = exp
(10)
kB T
Z
The term in the exponential is called Boltzmanns factor.
Z is a normalization factor
Z
Z Y
3N
i=1
H({qi }, {pi })
dqi dpi exp
kB T
(11)
We can compare this result with the momentum distribution of one gas molecule in the
microcanonical ensemble (see Microcanonical ensemble notes).
Do not say its just a normalization factor. We will see that Z is actually very important.
(volume of phase
In the miocrocanonical ensemble, we also have a nomalization constant
space)
1
1
E H({qi }, {pi }) E + E
mc ({qi }, {pi }) =
(12)
0
otherwise
S(N, V, E) = kB ln
N !h3N
(13)
In other words,
S = kB ln ,
N !h3N
(14)
H
H
1
ln Z
dq dp exp
Z
kB T
kB T
= kB
Z
1
=
dq dp c (q, p) H(q, p) + kB ln Z
T
where
(15)
Notice that the first term simply contains the ensemble average of energy.
Z
E hHi
dq dp c (q, p) H(q, p)
(16)
+ kB ln Z,
kB T ln Z = T S E
(17)
T
Recall in thermodynamics, the Helmholtz free energy is defined as A E T S.
Therefore,
A = kB T ln Z
(18)
S=
Z!
N !h3N
Z
N ! h3N
(19)
Notice that energy E is the ensemble average of the Hamiltonian. But the free energy
A cannot be written as an ensemble average. A is proportional to the log of the
normalization constant A = kB ln Z. (Analogous to Boltzmanns entropy expression
for the canonical ensemble S = kB ln .)
In the canonical ensemble, the system acquire a temperature by having a thermal contact
with a thermostat (heat bath) with temperature T . Thus the system is no longer isolated
any more. Its total energy, i.e., Hamiltonian H({qi }, {pi }) is no longer conserved. In other
words, we should expect some fluctuation of total energy in the canonical ensemble.
On the other hand, fluctuations are not considered in thermodynamics. At constant N, V, T ,
the appropriate thermodynamics potential is A(N, V, T ), from which we can compute a
definite value for energy E = A + T S, with S (A/T )N,V .
Hence, in thermodynamics, we expect the system to simultaneously have a definite temperature T and total energy E. In statistical mechanics, if the system have a well defined
temperature, its total energy E must fluctuate.
Q: How do we reconcile this difference between statistical mechanics and thermodynamics?
A: (1) The total energy in thermodynamics should be identified as the ensemble average of
the Hamiltonian in statistical mechanics.
1
E = hHi
Z
Z Y
3N
i=1
1
(20)
kB T
(2) We expect the statistical fluctuation of the Hamiltonian to diminish in the thermodynamic limit (N ). Q: Can we show this?
Define E as the standard deviation of the Hamiltonian in the canonical ensemble, where
(E)2 hH 2 i hHi2
(21)
Z
=
i=1
Z Y
3N
(22)
(23)
1 Z
=
ln Z
(24)
i=1
E hHi =
Since
Z
=
Z Y
3N
Z
=
2
Z Y
3N
hH 2 i hHi2
hH 2 i hHi2
2
(E)
Notice that CV =
E
T N,V
(25)
(26)
i=1
i=1
1
hH i =
Z
2
Z Y
3N
dqi dpi eH H 2 =
i=1
1 2 Z
Z 2
(27)
!2
1 Z
Z
!
Z
1 Z
+
=
1 2 Z
=
Z 2
1 Z
Z
1
=
Z
= hHi
(28)
= hH i hHi = E = kB T 2
E
T
(29)
(30)
p
E = N kB T 2 cV
E N as N
(31)
(32)
E N as N
E
1
0 as N
E
N
(33)
(34)
Therefore, the relative energy fluctuation diminishes in the thermodynamic limit. In this
limit, the difference between canonical and microcanonical ensemble vanishes!
Notice that the variance of H, (E)2 hH 2 i hHi2 , is a measure of the energy fluctuation
of the system.
At the same time, the heat capacity CV E
measures the response of the system to external
T
stimuli How much energy is needed if we want to raise the temperature by one degree.
8
Thus (E)2 = kB T 2 CV is an example of the fluctuation-dissipation theorem (FDT). Spontaneous fluctuations at thermal equilibrium and response to external stimuli are not independent of each other.
We have encountered another fluctuation-dissipation theorem ( =
chapter.
D
)
kB T
in the Diffusion
Other fluctuation-dissipation theorems can be derived similarly. For example, the spontaneous volume fluctuation at constant pressure is related to compressibility see future notes
on the (N, P, T ) ensemble.
Because (E)2 = kB T 2 CV , (E)2 0, and kB T 2 0, we must have CV
Hence the heat capacity is always positive.
E
T N,V
0.
Examples
5.1
Ideal Gas
Lets obtain the explicit expression for A(T, V, N ) from A = kB T ln Z and compare it with
the results from the microcanonical ensemble + Legendre transform.
Z =
H({qi }, {pi }) =
Z =
A(T, V, N ) =
Z Y
3N
1
H({qi }, {pi })
dqi dpi exp
N ! h3N
kB T
i=1
X |pi |2
+ U ({qi })
2m
i
3N
Z
VN
p2
VN
dp exp
=
(2mkB T )3N/2
3N
3N
N! h
2mkB T
N! h
"
#
3/2 !
V 2mkB T
kB T ln Z = kB T N ln
+1
N
h2
(35)
(36)
(37)
(38)
5.2
Z=
exp(H), where H =
N
X
ni
(39)
i=1
{ni }
notice the N !h13N correction is no longer needed because the problem is already discrete and
molecules are distinguishable.
P
The sum {ni } over all possible combinations seems impossible to obtain. But it can be
transformed to a much simpler form.
!
N
N
X
Y
eH = exp
ni =
eni
(40)
i=1
Z =
N
XY
eni =
{ni } i=1
1 + e + e
i=1
N
Y
i=1
ni =0,1,2
eni
2 N
(41)
(42)
e
A
= N kB T ln 1 + e + e2 + N kB T
T
1
2
e
+ 2e
E = A + TS = N
1+e
+ e2
S =
+ e2 kB2T 2
kB T 2
+ e + e2
(43)
(44)
(ln Z) =
N ln(1 + e + e2 )
2
e
+ 2e
= N
1+e
+ e2
E =
10
(45)
E
T
(46)
N
has a very long expression. But the main point here is that its analytic expression exists.
As T 0, e
k T
B
0, E 0, S 0 (third law), A 0, CV ?
11