Virial Theorem (Classical systems)

Consider a stable system of discrete particles, bound by potential forces (forces characterized
exclusively by potential difference). The virial theorem provides a general relation between the total
kinetic energy and the total potential energy of the system. The theorem states
"
1
〈𝐾. 𝐸〉 = − +〈𝐹⃗! ∙ 𝑟⃗! 〉
2
!#$

Here K.E stands for the total kinetic energy of N particles, 𝐹⃗! represents the force on the ith particle,
which is located at position 𝑟⃗! , and the angular brackets represent the time average. The significance of
the virial theorem is that it allows the average total kinetic energy to be calculated even for very
complicated systems that defy an exact solution, such as those considered in statistical mechanics.

Consider
𝜕𝐻
𝜕𝐻 ∫ 𝑑𝑞! 𝑑𝑝! 𝑥! 𝜕𝑥 𝑒 & ( )
%
〈𝑥! 〉 =
𝜕𝑥% ∫ 𝑑𝑞! 𝑑𝑝! 𝑒 & ( )
where 𝑥! and 𝑥% are any of the 6N generalized coordinates, for example, 𝑝 or 𝑞.
𝜕𝐻 & ( ) 1 𝜕𝑒 & ( )
8 𝑑𝑞! 𝑑𝑝! 𝑥! 𝑒 = − 8 𝑑𝑞! 𝑑𝑝! 𝑥!
𝜕𝑥% 𝛽 𝜕𝑥%
Integrating by parts we get
(%)
7!
*
𝜕𝑒 & ( ) 𝜕𝑥! & ( )
8 𝑑𝑞! 𝑑𝑝! 𝑥! = 8 !+,-./0,!+. 𝑑𝑞! 𝑑𝑝! 𝑥! 𝑒 & ( ) : − 8 𝑑𝑞! 𝑑𝑝! 𝑒
𝜕𝑥% 12-/ 033 411/5!+0,-6 𝜕𝑥%
-74-8, 7! (#)
7!

($) (;)
Here 𝑥% and 𝑥% are the extremal values of 𝑥% . The boundary integral vanishes because 𝐻 becomes

infinite at the extremal values of any coordinates. For example, if 𝑥% is momentum 𝑝, then the extremal
8%
values of 𝑝 = ± ∞ and 𝐻 = ; < → ∞. For example, if 𝑥% is spatial coordinate 𝑞, then the extremal

values are at boundary of system, where the potential energy confining the particle to the volume V
becomes infinite. This implies,
𝜕𝐻 & ( ) 1 𝜕𝑥! & ( )
8 𝑑𝑞! 𝑑𝑝! 𝑥! 𝑒 = 8 𝑑𝑞! 𝑑𝑝! 𝑒
𝜕𝑥% 𝛽 𝜕𝑥%
=7
But =7 & = 𝛿!% . Thus, we get
!

𝜕𝐻 1 ∫ 𝑑𝑞! 𝑑𝑝! 𝑒 & ( )

〈𝑥! 〉 = 𝛿!%
𝜕𝑥% 𝛽 ∫ 𝑑𝑞! 𝑑𝑝! 𝑒 & ( )
 𝜕𝐻
〈𝑥! 〉 = 𝑘# 𝑇 𝛿!"
𝜕𝑥"
Is the called the Virial theorem.

If 𝑥! = 𝑥! = 𝑝! then
𝜕𝐻
〈𝑝! 〉 = 〈𝑝! 𝑞̇ ! 〉 = 𝑘> 𝑇
𝜕𝑝!
If 𝑥! = 𝑥! = 𝑞! then
𝜕𝐻
〈𝑞! 〉 = − 〈𝑞! 𝑝̇ ! 〉 = 𝑘> 𝑇
𝜕𝑞!
=) =)
Where we have used Hamiltonian equations of motion =8 = 𝑞̇ ! and =? = − 𝑝̇! .
& &

If we consider 3N degrees of freedom, then

@"
〈+ 𝑝! 𝑞̇ ! 〉 = 3𝑁 𝑘> 𝑇
!#$
@"

− 〈+ 𝑞! 𝑝̇ ! 〉 = 3𝑁 𝑘> 𝑇
!#$

In simple terms,
2〈𝑇𝑜𝑡𝑎𝑙 𝐾. 𝐸 〉 = − 〈𝑇𝑜𝑡𝑎𝑙 𝑃. 𝐸〉

@" A' B
Here, 3𝑁 𝑘> 𝑇 = 2 ;
= 2 𝑇𝑜𝑡𝑎𝑙 𝐾. 𝐸

Paramagnetism (Classical systems)

Consider N classical spins (paramagnetic system), and ignore the interaction between the spins. When
an external magnetic field is applied, these spins tend to align in the direction of the field. For this case,
we can write down the Hamiltonian
" "
K⃗ = −𝜇 ℎ + cos 𝛼!
𝐻 = − + 𝜇⃗! ∙ ℎ
!#$ !#$

where 𝜇⃗! is magnetic moment of spin 𝑖, | 𝜇⃗! | = 𝜇, 𝛼! is the angle of 𝜇⃗! with respect to ℎK⃗.
As the spins are non-interacting, the partition function for the system (N-spins) is

𝑄" = (𝑄$ )"

Here we need not include 𝑁! term because the spins are distinguishable. We can imagine that each spin
sits at a fixed position in space and so can be distinguishable from any other spin.
For the consider system,

𝑄$ = + 𝑒 (CD EFG H
H
(I33 8166!J3- 0+.3-6)

Here, 𝛼 may be oriented in an arbitrary direction with respect to the applied magnetic field. Thus, one
needs to consider all possible configurations of 𝛼.
;K K
𝑄$ = 8 𝑑𝜙 8 𝑑𝜃 sin 𝜃 𝑒 (CD EFG M
L L

It is As there is no 𝜙 dependence, the above integral simplifies into,

K
𝑄$ = 2𝜋 8 𝑑𝜃 sin 𝜃 𝑒 (CD EFG M
L

Take cos 𝜃 = 𝑥. Implies, sin 𝜃 𝑑𝜃 = −𝑑𝑥. The corresponding limits are 𝑥 = 1 (for 𝜃 = 0) and 𝑥 =
−1 (for 𝜃 = 𝜋). Thus,
K $
8 𝑑𝜃 sin 𝜃 𝑒 (CD EFG M = 8 𝑑𝑥 𝑒 (CD 7
L &$

1 $ 2 𝑒 (CD − 𝑒 &(CD
= \𝑒 (CD 7 ]&$ = ^ _
𝛽𝜇ℎ 𝛽𝜇ℎ 2
2 sinh(𝛽𝜇ℎ)
=
𝛽𝜇ℎ 𝛽𝜇ℎ
So, we get
K
sinh(𝛽𝜇ℎ)
𝑄$ = 2𝜋 8 𝑑𝜃 sin 𝜃 𝑒 (CD EFG M = 4𝜋
L 𝛽𝜇ℎ
sinh(𝛽𝜇ℎ)
𝑄$ = 4𝜋
𝛽𝜇ℎ
KK⃗ oriented along ℎK⃗ direction. If we choose ℎK⃗ =
Now, we can calculate the average total magnetization 𝑀
ℎ 𝑧̂ (along z direction), then
𝑀N = 𝑁 〈𝜇 cos 𝛼〉

Here 〈𝜇 cos 𝜃〉 is the projection of one spin along the direction of the magnetic field.

∑H 𝜇 cos 𝜃 𝑒 (CD EFG H

𝑀N = 𝑁 〈𝜇 cos 𝜃〉 = 𝑁
∑H 𝑒 (CD EFG H
1 𝜕 (CD EFG H
𝛽 𝜕ℎ f∑H 𝑒 g
𝑀N = 𝑁
∑H 𝑒 (CD EFG H

𝜕
𝑁 𝜕ℎ (𝑄$ )
𝑀N =
𝛽 𝑄$
 GOPQ((CD)
Here, 𝑄$ = 4𝜋 (CD
. Thus, we get
cosh(𝛽𝜇ℎ) sinh(𝛽𝜇ℎ)
𝑁 4𝜋 h ℎ −
𝛽𝜇ℎ;
i
𝑀N =
𝛽 sinh(𝛽𝜇ℎ)
4𝜋
𝛽𝜇ℎ
1
𝑀N = 𝑁𝜇 hcoth(𝛽𝜇ℎ) − i
𝛽𝜇ℎ
The average magnetization is
𝑀N 1
〈𝜇N 〉 = = 𝜇 hcoth(𝛽𝜇ℎ) − i
𝑁 𝛽𝜇ℎ
The term in the bracket is called the Langevin function. If we take 𝑦 = 𝛽𝜇ℎ, then in the 𝑦 → ∞ limit,
coth(𝛽𝜇ℎ) → 1. So, the term in the bracket tends to one. Implies,
〈𝜇N 〉 = 𝜇
It means the magnetization per spin is constant which is 𝜇.
However, the other limit, i.e., 𝑦 → 0, is more interesting. In this limit, the Langevin function can be
simplified as
1 cosh(𝑦) 1
coth(𝑦) − = −
𝑦 (
sinh 𝑦 ) 𝑦
𝑦; 𝑦;
1 + 2! 1 1+ 2 1
= @ − = ; −
𝑦 𝑦 𝑦 𝑦
𝑦 + 3! 𝑦 m1 + 6 o

𝑦; 𝑦; 𝑦; 𝑦;
m1 + 2 o m1 − 6 o 1 +
=
1
− = 2 − 6 − 1 = 𝑦
𝑦 𝑦 𝑦 𝑦 3
This implies,
𝜇; 𝛽ℎ 𝜇; ℎ
〈𝜇N 〉 = =
3 3𝑘> 𝑇
The net magnetization is
𝑁 𝜇; ℎ
𝑀N =
3𝑘> 𝑇
The magnetic susceptibility
𝜕𝑀N 𝑁 𝜇;
𝜒 = lim =
D→L 𝜕ℎ 3𝑘> 𝑇
1
𝜒 ∝
𝑇
This is called Curie’s law of paramagnetism. The magnetic susceptibility of a paramagnetic material
grows linearly with 1/𝑇.
 Entropy and information

The entropy S can be expressed in terms of the probability P as

𝑆 = −𝑘> + 𝑃! ln 𝑃!
!

where 𝑃! is the probability to be in state 𝑖. This relation holds for both canonical and microcanonical
ensemble.

Claude Shannon (1948) turned this relation backwards, in developing a close relation between entropy
and information theory. Consider a system with states labelled by 𝑖, and 𝑃! is the probability for the
system to be in state 𝑖. We want to define a measure of how disordered the distribution 𝑃! is. Call the
disorder measure 𝑆 (it will turn out to be the entropy).

𝑆 = −𝑘 + 𝑃! ln 𝑃!
!

where 𝑘 is the proportionality constant.

The above relation states that the bigger (smaller) 𝑆 is, the more (less) disordered the system is, the less
(more) information we have about the probable state of the system. Here, 𝑆 satisfies the following
conditions.

1 𝑗 = 𝑖
1. If 𝑝% = u Then the state of the system is exactly known to be 𝑖.
0 𝑗 ≠ 𝑖
This should have 𝑆 = 0 as there is no uncertainty, no disorder.

2. For equally likely 𝑝! , i.e. all 𝑝! = 1/N for N states, the system is maximally disordered. It means 𝑆 is
maximum possible value for all possible N state distributions.

3. 𝑆 should be additive over independently random systems.