Anda di halaman 1dari 18

The Sommerfeld Expansion

Review by Markus Selmke


Universität Leipzig, Department of Physics
markus.selmke@gmx.de

06.07.2007, revisited 15.07.2007

Contents
Abstract 1

1 The Sommerfeld Expansion 1


1.1 Transformation of the Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 First Derivation of the Intermediate Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Seocond Derivation of the Intermediate Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The Missing Integral-Factor I 2n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 A Special Case and a Similar Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 The Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Application: µ ↔ E F and c V ↔ E F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Treatment in Terms of Fugacity / Fermi- and Bose-Functions f ν (z ), b ν (z ) . . . . . . . . . . . . . . . 8

2 The Bernoulli Numbers B n 11


2.1 The Generating Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 The Recurrence Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 The Riemann-Zeta Function ζ (x ) 12


3.1 First Notion of x2 coth x2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Second Notion of x2 coth x2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.2.1 Fourier Series of cos y x . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2.2 Transformation to 2 coth x2x
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 Comparison and Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.4 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

A Cauchy Product 16

B n choose k 16

C The Geometric Series 16

D Special Functions 16

References 17

I
1 THE SOMMERFELD EXPANSION 1

Abstract

This Document will provide a proof for the Sommerfeld expansion formula used for integrals involv-
ing the Dirac Distribution and a sufficiently well behaved but otherwise arbitrary function. Also included
from my previous works is the derivation for the formula connecting the values of the Riemann-Zeta
function for even integers with the Bernoulli numbers, which is used within the proof. The main result
will be equation (23/24) on page 6. Also two different motivations/ways for the reshaping of the integral,
the first step in the derivation, are provided and an additional treatment in terms of fugacity is given.

1 The Sommerfeld Expansion


1.1 Transformation of the Integral
1.1.1 First Derivation of the Intermediate Result

Following and correcting a discussion in L. Landau, E. Lifshitz: Statistical Physics Part I, 3rd Edition, page
169/170. Notation is slightly changed for convenience.
We want to approximate the following integral for a function f (ε) such that the integral I converges.


f (ε)
Z
I = (ε−µ)
dε (1)
0 e kB T
+1

We transform the integral by a change of coordinates via


ε − µ / (k B T ) = z , ε = k B Tz + µ , dε/dz = k B T or dε = k B T dz



f k B Tz + µ
Z 
I = k B T dz
−µ/(k B T )
ez +1
0 ∞
f k B Tz + µ f k B Tz + µ
Z  Z 
= kBT dz + k B T dz
−µ/(k B T )
ez +1 0
ez +1
µ/(k B T ) ∞
f µ − k B Tz f k B Tz + µ
Z  Z 
= kBT dz + k B T dz (2)
0
e −z + 1 0
ez +1

1 ez
Now, since e −z +1
= e z +1
= 1 − e z1+1 we split the first integral and hence:

µ/(k B T ) µ/(k B T ) ∞
f k B Tz + µ
Z Z Z 
1
= f µ − k B Tz dz − k B T f µ − k B Tz z dz + k B T
 
I kBT dz
0 0
e +1 0
ez +1

Transforming the first integral back via ε = µ − k B Tz , dε/dz = −k B T or dz = dε/ (k B T )

µ µ/(k B T ) ∞
f µ − k B Tz f k B Tz + µ
Z Z  Z 
I = f (ε) dε − k B T dz + k B T dz (3)
0 0
ez +1 0
ez +1

Now comes the first approximation (up to now this was all exact and simply manipulation of the integral
to bring it into a suggestive form). We replace the upper limit in the second integral with infinity. This is
legitimate since µ/ (k B T ) ≈ E F / (k B T ) >> 1 and the integral is rapidly convergent:

µ ∞
f µ + k B Tz − f µ − k B Tz
Z Z  
I ≈ f (ε) dε + k B T dz (4)
0 0
ez +1

We now Taylor-Expand the function f around µ and obtain:


1 THE SOMMERFELD EXPANSION 2

µ ∞ ¨X

f [n ] µ f [n ] µ
Z Z   «
1
I ≈ f (ε) dε + k B T (k B Tz )n − (−k B Tz )n dz
0 0 n=0
n! n! ez +1
µ ∞ ∞ [2n −1]
µ
Z Z 
X f z 2n −1
= f (ε) dε + k B T 2 (k B T )(2n −1) dz
0
(2n − 1)!
n =1 0
ez +1
µ ∞ Z∞
f [2n −1] µ
Z 
X
2n z 2n −1
= f (ε) dε + 2 (k B T ) dz
0 n =1
(2n − 1)! 0
ez +1
Zµ ∞
f [2n −1] µ
X 
= f (ε) dε + 2 (k B T )2n I 2n (5)
0 n =1
(2n − 1)!

1.1.2 Seocond Derivation of the Intermediate Result

A further approch to reach this result is presented in Reference [2]. In this document though, the notation
is altered from the reference to be conform with the first discussion. Consider again the following integral:

∞ ∞
f (ε)
Z Z
I = (ε−µ)
dε = f (ε) D (ε) dε (6)
0 e kB T
+1 0

where D (ε) is the Dirac Distribution. Define F (x ) by

Z ε
F (ε) = f ε0 dε0

(7)
0

We now integrate by parts the former integral:

Z ∞
I = [D (ε) F (ε)]∞
0 − D 0 (ε) F (ε) dε (8)
0

The first term vanishes since D (∞) = 0 and F (0) = 0. Hence we are left with

Z ∞
I = − D 0 (ε) F (ε) dε (9)
0

Now, when evaluating this integral the important point that Sommerfeld made is to note is that D 0 is
sharply peaked at ε = µ ≈ E F , particularly at low temperatures T . Thus we expand F (ε) around ε = µ:


F [n ] µ
X 
n
F (ε) = ε−µ (10)
n =0
n!

This transfroms the integral into the following convergent series:

∞ Z ∞
X 1 [n ]  n
I = − F µ D 0 (ε) ε − µ dε (11)
n =0
n! 0

Yet again, change of variable via z = ε − µ /k B T , dz /dε = 1/ (k B T ) brings us to the form:




∞ ∞ ∞ ∞
(k B T )n F [n ] µ
Z  Z
X 1 [n ]  −e z n
X ez
I ≈ − F µ 2
(k B Tz ) dz = z n dz (12)
n =0
n! −∞ (e z + 1) n=0
n! −∞ (e z + 1)2

The first step in the above sequence was replacing the lower bound by −∞ which is legitimate since D 0
goes to zero quickly away from ε = µ. This step is done to realize that odd integers n do not contribute in
1 THE SOMMERFELD EXPANSION 3

 2
ez 1
the sum obtained last (later the limit of integration will be set back to 0): (e z +1)2
= 2 cosh(z /2)
is an even
function (cosh is an even function), and hence when multiplied by an odd function (z 2n +1 ) and integrated
over the real line gives zero! Since the integral for n = 0 is 1 we are left with:

∞ ∞
(k B T )2n F [2n ] µ
 ‚ Z Œ
X ez
F µ + 2n

I ≈ z dz 2
n=1
(2n)! (e z + 1)2 0
∞ Z∞
(k B T )2n F [2n ] µ
 ¨• «
−1 2n ∞ z 2n −1
X ˜
= F µ + + (2n)

2 z
n=1
(2n)! ez +1 0 0
ez +1

2 (k B T )2n F [2n] µ
X 
= F µ +

I 2n (13)
n=1
(2n − 1)!

Now we replace F by f to get:

∞ µ ∞
2 (k B T )2n f [2n−1] µ
Z Z 
X
I = f (ε) D (ε) dε ≈ f (ε) dε + I 2n (14)
0 0 n =1
(2n − 1)!

Note how equation (5) and (14) coincide.

1.2 The Missing Integral-Factor I 2n


The integral I 2n can be computed using the definitions of the Riemann-Zeta function ζ (x ) and the Gamma
function Γ (x ) and furthermore the Geometric Series (see Appendix):

Z ∞
Γ (x ) = z x −1 e −z dz
0

X 1
ζ (x ) = x
n =1
n

X z
zk = for |z | ≤ 1
k =1
1−z

With those tools at hand we are ready to compute the following integral:

Z ∞ Z∞
z x −1 1
 ‹
Ix = dz = z x −1 −z
e dz
0
ez +1 0
1 + e −z
Z∞ ‚∞ Œ Z∞ ∞
X X
(−1)n z x −1 e −(1+n )z dz
n
= z x −1 e −z −e −z dz = (15)
0 n =0 0 n =0

Now we change the variable of integration to transform the inner integral into the Gamma function:
1 1
z 0 = (1 + n) z , z = (1+n )
z 0 , dz 0 /dz = (1 + n) or dz = (1+n )
dz 0

Z ∞X
∞ x −1 Z ∞X

z0

10 1
Ix = (−1)n e −z dz 0 = (−1)n −1 z x −1 e −z dz
0 n=0
n +1 n +1 0 n =1
nx
∞  x
1 2 −2
X ‹
= (−1)n+1 x Γ (x ) = ζ (x ) Γ (x ) = 1 − 21−x ζ (x ) Γ (x )

(16)
n=1
n 2x

Here I used equation (94) to relate the infinite sum over n to the Riemann-Zeta function:


X (−1)n 2p − 2
= ζ(p ) for p > 1
n =0
(n + 1)p 2p
1 THE SOMMERFELD EXPANSION 4

..to express the sum within the lengthy term in terms of the Riemann-Zeta function.
If x = 2k is an even integer we can use the explicit expression derived later, see equation (84), for ζ (2n) in
terms of Bernoulli numbers B n :

B 2k
(−1)k +1 (2π)2k = ζ(2k ) for k ∈ Integers
2 (2k )!

This gives for I 2k using equation (16) the following expression:

Z ∞
z 2k −1 22k − 2
  
k +1 2k 2k B 2k
I 2k = dz = (−1) 2 π (2k − 1)!
0
ez +1 22k 2 (2k )!
π2k B 2k
 2k −1 
2 −1
− 2 (−1)k +1 (−1)k +1 π2k B 2k
€ Š
= 22k = (17)
4k 2k

1.2.1 A Special Case and a Similar Integral

This section derives additional interesting results which are not needed for the Sommerfeld expansion, but
easily derived at this point with the previous setup.
For x = 1 we go one step back to get a result and note that the Taylor expansion of ln (1 + x ) reads:


X xn
ln (1 + x ) = (−1)n +1 for |x | ≤ 1
n =1
n

X 1
ln (2) = (−1)n +1 (18)
n=1
n

Thus we get for I 1 using an intermediate step in equation (16):

Z ∞ ∞
1 X 1
I1 = dz = (−1)n+1 Γ (1) = ln (2) (19)
0
ez +1 n=1
n

An other intergral of interest can be computed using the same technique as in the previous section:

Z ∞ Z ∞ Z ∞ ‚ ∞
Œ
z x −1 dz 1
 ‹ X
−z n
= = x −1
= x −1 −z

Ix z dz z e e dz
0
ez −1 0
ez −1 0 n =0
∞ Z∞
X
= z x −1 e −(1+n )z dz (20)
n =0 0

Analogous to the previous treatment we change the variable of integration to transform the inner integral
1 1
into the Gamma function: z 0 = (1 + n) z , z = (1+n) z 0 , dz 0 /dz = (1 + n) or dz = (1+n )
dz 0

∞ Z ∞ x −1 ∞ Z ∞
X z0 −z 0 1 X 1
Ix = e dz 0 = z x −1 e −z dz
n=0 0
1+n 1+n n=1 0
nx
Z ∞ ∞
X 1
= x −1 −z
z e x
dz = Γ (x ) ζ (x ) (21)
0 n =1
n

For even integers this becomes:

B 2n
I 2n = Γ (2n) ζ (2n) = (2n − 1)!(−1)n +1 (2π)2n
2 (2n)!

(−1)n +1 (2π)2n B 2n (−4)n −1 π2n B 2n
Z
z 2n −1 dz
= = = (22)
4n n 0
ez −1
1 THE SOMMERFELD EXPANSION 5

1.3 The Result


Going back to our original problem we find from equation (14)/(15) and (17):

∞ ∞
f (ε)
Z Z
I = (ε−µ)
dε = f (ε) D (ε) dε
0 e kB T
+1 0

µ ∞
f [2n −1] µ
Z 
22n −1 − 1
X   
≈ f (ε) dε + 2 (k B T )2n (−1)n +1 π2n B 2n
0 n=1
(2n − 1)! 2n
µ ∞
f [2n −1] µ
Z 
X
= f (ε) dε + (k B T )2n 22n − 2 (−1)n +1 π2n B 2n

(23)
0 n=1
(2n)!
µ
π2 (k B T )2 f 0 µ 7π4 (k B T )4 f 000 µ
Z  
= f (ε) dε + + + (24)
0
6 360

31π6 (k B T )6 f 00000 µ 127π8 (k B T )8 f 0000000 µ


 
+ + ...
15120 604800

This is the final result, where in equation (24) the Bernoulli Numbers derived in the next section were used
(see recurrence formula and table 1 for B n in the Section 2.2). The formula is an asymptotic approxima-
tion and not a convergent series, as L. Landau points out (see Reference [1]) and is seen in the picture
below .
This is equivalent to the approximation of the Fermi-Dirac distribution by the Unit-Step-function and
terms involving derivatives of the generalized Dirac Delta Distribution δ (x ):


δ[2n −1] µ − ε

1 X
D (ε) = Θ µ−ε + (k B T )2n 22n − 2 (−1)n π2n B 2n
 
(ε−µ)

(2n)!
e kB T
+1 n =1

Sidenote: Setting k B = 1 and T = 1 we can generalize the result for η >> 1:


∞ η ∞
f (x ) f [2n−1] η
Z Z 
X
f (x ) dx + 22n − 2 (−1)n +1 π2n B 2n

dx ≈
0
e −η + 1
x
0 n=1
(2n)!
η
π2 0 
Z
7π4 000 
= f (x ) dx + f η + f η + ...
0
6 360
1 THE SOMMERFELD EXPANSION 6

1.4 Application: µ ↔ E F and c V ↔ E F


Again following Reference [2] we derive the dependence of µ on Temperature T :
From the lecture we know the density of states for the free electron gas in a box in terms of Fermi-Energy
E F and the total number N of electrons is:

∞ ∞
ε
‹1/2 Z Z
3 N 3

3/2
g (ε) = N = g (ε) D (ε) dε ⇒ EF = ε1/2 D (ε) dε (25)
2 EF EF 0
2 0

Now we see that for the Sommerfeld Approximation we should set f (ε) = ε1/2 and carry along all the pref-
actors. We thus get the following (where the general form of the derivative was observed):

‚Z µ ∞
d ε1/2 µ
  Œ
3/2 3 X
ε1/2 dε + (k B T )2n 22n − 2 (−1)n+1 π2n B 2n

EF ≈
2 0 n =1
(2n)! dε2n −1


3 (|3 + 4 (n − 2)|)!!
 
5+4(n−2)
X
= µ3/2 + µ− (k B T )2n 22n − 2 (−1)n+1 π2n B 2n

2

n =1
2 (2n)! 22n−1

‚ Œ
(|4n − 5|)!! −2n
X  
2n n +1 2n
= µ3/2 1+ µ (k B T ) 1 − 2 1−2n
(−1) π B 2n

3
n =1
(2n)!

3 π2 1 3 7π4 1 (−1) 3 −5/2


= µ3/2 + (k B T )2 µ−1/2 + (k B T )4 µ + ...
2 6 2 2 360 2 2 2

3π2 k B T 2 7π4 9 k B T 4
‚     Œ
= µ3/2
1+ + + ... (26)
24 µ 360 16 µ

..using the following Power-series expansion

∞ ∞
X n (n − 1) . . . (n − k + 1) X Γ (n + 1)
(1 + x )n = xk = xk (27)
k =0
k! k =0
Γ (n − k + 1) Γ (k + 1)
n!
..and keeping in mind n!! = 2(2n−1)/2 ((2n −1)/2)!
= n (n − 2) . . . 1 for odd integers we get:


‚∞  Œk
Γ (2/3 + 1) (|4n − 5|)!! −2n
X X 
2n
µ (k B T ) 1 − 2 1−2n
(−1) π B 2n
n+1 2n

EF ≈ 3
k =0
Γ (2/3 − k + 1) Γ (k + 1) n =1 (2n)!

π2 (k B T )2 π4 (k B T )4
= µ+ + + ...
12 µ 180 µ3
‚ 2 4 Œ
π2 π4
 
kBT kBT
= µ 1+ + + ... (28)
12 µ 180 µ

Inversion of this equation to the same order in E F gives the following result (this can be done by hand,
basically comparison of coefficients, but Mathematica’s bulit in function InverseSeries[] comes in handy
here):

‚ 2 4 Œ
π2 π4
 
kBT kBT
µ ≈ EF 1− − − ... (29)
12 EF 80 EF

This whole
 procedure
  can
 be easily implemented in Mathematica to find the coefficients up to arbitrary
order of kEB T or k BµT , respectively. (For an alternative way: see Reference [10]).
f
In a similar way we obtain the specific electronic heat in terms of Fermi energy, making use in the
calculation of the formula for µ in terms of E F which was derived before.

EF ∞

Z Z
u0 = g (ε) ε D (ε) d u = g (ε) ε D (ε) dε cV = u (30)
0 0
∂T
1 THE SOMMERFELD EXPANSION 7
1 THE SOMMERFELD EXPANSION 8

From the Mathematica file we read off the electronic contribution to the specific heat in solids pre-
dicted by the Sommerfeld-Theory, where k B TF = E F :

π2
‹3 ‹5
3π4 247π6
 
T T T
 ‹  
cV = kBN − − − ... (31)
2 TF 20 TF 2016 TF

1.5 Treatment in Terms of Fugacity / Fermi- and Bose-Functions f ν (z ), b ν (z )


In statistical physics it is common to define the fugacity z by z = e β µ = e µ/(k B T ) . The motivation or context,
along with the definitions can be found in IdealQuantumGases.pdf and the corresponding Mathematica
notebook fermi.nb, both located on the lecture notes homepage of James J. Kelly on
http://www.physics.umd.edu/courses/Phys603/kelly/.
Defined is the Fermi Function by the following similar integral:


x ν −1
Z
1
f ν (z ) = dx (32)
Γ (ν ) 0
z −1 e x +1

For small z , this can be expressed as a power series as mentioned and calculated in the Mathematica file
by the program function Series[] (Reference [12]). I will derive this expansion by hand in the following lines:

∞ ∞
!
Z Z ∞
1 1 1
 ‹ X
ν −1 ν −1 −x k
f ν (z ) = = −x

x dx x ze −z e dx
Γ (ν ) 0
z −1 e x + 1 Γ (ν ) 0 k =0
∞ Z∞
z X
= (−z )k x ν −1 e −x (k +1) dx
Γ (ν ) k =0 0
∞ Z ∞ ‹ν −1
z X x 1
= (−z )k e −x dx
Γ (ν ) k =0 0
k +1 (k + 1)
∞ ∞
z X 1 X 1
= (−z )k Γ (ν ) = − (−z )k +1
Γ (ν ) k =0
(k + 1)ν k =0
(k + 1)ν

X (−z )k
= − for all 0 < z < 1, ν > 0 (33)
k =1

The limiting value for ν → ∞ is seen to be z since in the sum only k = 1 will contribute:

f ∞ (z ) = z (34)

also note the following recursive relation, which are useful in handling the fermi-functions:


∂ X (−z )k −1 1 ∂
f ν (z ) = = − f ν −1 (z ) ⇒ z f ν (z ) = −f ν −1 (z ) (35)
∂z k =1
k ν −1 z ∂z

A similar series expansion can be given for BE integrals, the Bose Function is definde by:


x ν −1
Z
1
b ν (z ) = dx (36)
Γ )
(ν 0
z −1 e x − 1

It seems as if the Fermi-Functions and the Bose-Functions, which are those kinds of integrals, are defined
slightly different depending on the paper or book they appear in. Another definition will appear shortly on
the side to see how they convert into each other. The derivation now for the expansion is analogously to
the expansion of the Fermi-Function we have seen before:
1 THE SOMMERFELD EXPANSION 9

∞ ∞
!
Z Z ∞
1 1 1
 ‹ X
ν −1 ν −1 −x k
b ν (z ) = = −x

x dx x ze ze dx
Γ (ν ) 0
z −1 e x − 1 Γ (ν ) 0 k =0
∞ Z ∞ ∞ Z ∞ ‹ν −1
1 X k +1 ν −1 −x (k +1) 1 X k +1 x 1
= z x e dx = z e −x dx
Γ (ν ) k =0 0
Γ (ν ) k =0 0
k +1 (k + 1)
∞ ∞ ∞
1 X k +1 1 X z k +1 X zk
= z Γ (ν ) ν = = for all 0 < z < 1, ν > 0 (37)
Γ (ν ) k =0 (k + 1) k =0
(k + 1)ν k =1

Also here we note off the following obvious properties:


b ∞ (z ) = z ; b ν (z ) = z b ν −1 (z ) (38)
∂z
This series expansion establishes the useful relation between f ν (z ) and b ν (z ), which can be found in
a somewhat different notation in Reference [13]. For completeness, since in theoretical applications their
notation seems to be common, I will compare them shortly:

∞ ∞
∂ e kη
Z
1 xp  X
η = B p η = b p +1 (e η ) ; B p +1 η = B p η ; B p η =
   
Bp dx ⇒
Γ p +1 e x −η − 1 ∂η k p +1

0 k =1

(−1)k +1 e k η


Z
1 xp  X
Fp η = Fp η = f p +1 (e η ) ; Fp +1 η = Fp η ; Fp η =
   
dx ⇒
Γ p +1 e x −η + 1 ∂η k p +1

0 k =1

(The two series hold for η < 0) However, in our notation the relation is derived the following way:

∞ ∞  k
(−z )k −z z 2 z 3 z 2k
  
X X z 2
f ν (z ) = − = − + + = = b ν (z ) − ν b ν z 2 (39)

− . . . − 2 ν
k =1
kν 1 ν 2ν 3 ν
k =1
k ν (2k ) 2

This is of importance since there are numerous different approximations to the Bose-functions as well.
Some of them, for example those treated in Reference [13], use Chebyshev Polynomials and are extremely
good and the series are fastly converging.
Another useful approximation to the Fermi-Function Fp η is the generalized approximation put for-


ward by Aymerich-Humet, F. Serra-Mestres and J. Millán (See Reference [14]). It is a fit-model approach
that is accurate up to a few percent and works over a wide range of η and sensible values of p :

 −1
 p +1
 p +1 2 e −η 
Γ p + 1 Fp η Š1/c p +1 + Γ p + 1 
 
≈ 
 € c
 (40)
b +η+ η−b +a
c

15  1 2 1/2 p Š
 ‹ € € Š
= 1+ p +1 + p +1 , b = 1.8 + 0.61p , c = 2 + 2 − 2 2−p

a
4 40
In the treatment with the form of the integral given by equation (32), for large z , Sommerfeld’s lemma
is derived in a similar way to the treatments we have seen before, using again the idea of the function
being a Unit-Step function to first order and then, this time, adding and substracting corrections, where
the error in the approximation comes in here because of changing the limits of the second integral to −∞
in the second line. The hint for this approach is to be found in the Mathematica file fermi.nb where the
Mathematica function Series[] is used again (Reference [12]). I will derive the general formula by hand,
where again equation (27) and the equation (17) for I 2k are used, listed here once more for convenience:


X Γ (n + 1) π2k B 2k
I 2k = 22k − 2 (−1)k +1
€ Š
(1 + x )n = xk ,
k =0
Γ (n − k + 1) Γ (k + 1) 4k
1 THE SOMMERFELD EXPANSION 10

The derivation to the general expansion formula goes like this:


x ν −1 dx
Z
1
f ν (z ) =
Γ (ν ) 0
e x −log(z ) + 1
log(z ) ∞ log(z )
(Z )
x ν −1
Z Z
1 1
 ‹
≈ x ν −1 dx + dx − x ν −1 1 − dx
Γ (ν ) 0 log(z )
e x −log(z ) + 1 −∞
e x −log(z ) + 1
ν ∞ ν −1 0
log (z ) x + log (z )
¨ Z Z «
1 1
 ‹
ν −1
= + dx − log (z ) − x 1− (−dx )
Γ (ν ) ν 0
ex +1 ∞
e −x + 1
ν ∞ ν −1 ν −1
log (z ) x + log (z ) − log (z ) − x
¨ Z «
1
= + dx
Γ (ν ) ν 0
ex +1
ν ∞ ν −1 ν −1 Œ
log (z )
¨ Z ‚   «
1 ν −1 x x 1
= + log (z ) 1+ − 1− dx
Γ (ν ) ν 0
log (z ) log (z ) ex +1
ν ( Z∞ 
∞ k  k Œ )
1 log (z )
‚
−1 X Γ (ν ) x −x 1
= 1+ log (ν ) ν −  dx
Γ (ν ) ν 0 k =0
Γ (ν − k ) Γ (k + 1) log (z ) log (z ) ex +1
ν ( ∞

∞ 2k +1  )
log (ν ) Γ (ν )
Z 
−1 X x 1
= 1+ log (ν ) ν 2  dx
Γ (ν + 1) 0 k =0
Γ (ν − (2k + 1)) Γ ((2k + 1) + 1) log (z ) ex +1
ν ( ∞ ∞ ∞
)
log (ν ) Γ (ν + 1)
Z Z
−1 X 2 x 2k −1
= 1+ log (ν ) dx
Γ (ν + 1) Γ (ν (2k Γ ((2k + ex +1
2k −1
0 k =1
− − 1)) − 1) 1) log (z ) 0
ν ( ∞
)
log (ν ) X Γ (ν + 1) 2 € Š k +1 π B 2k
2k
= 1+ 2 − 2 (−1)
2k
Γ (ν + 1) Γ (ν − 2k + 1) Γ (2k ) log (z ) 2k 4k

k =1
ν ( ∞
)
log (ν ) Γ (ν + 1) 22k −1 − 1 k +1 π B 2k
X  2k
= 1+ 2k (−1) = Fν −1 log (z )

(41)
Γ (ν + 1) k =1
Γ (ν − 2k + 1) Γ (2k ) log (z ) k

Going back to chemical potential we have Sommerfeld’s lemma. Here it reads:

ν 
βµ π2 7π4

βµ
€ Š −2 −4
fν e ≈ 1 + ν (ν − 1) βµ + ν (ν − 1) (ν − 2) (ν − 3) βµ + ...
Γ (ν + 1) 6 360
ν ∞
!
βµ Γ (ν + 1) 22k −1 − 1 k +1 π B 2k
X  2k
−2k
= 1+ βµ (−1) ≈ Fν −1 β µ

(42)
Γ (ν + 1) k =1
Γ (ν − 2k + 1) Γ (2k ) k

Similarly now, as is done in Reference [11], one can now also compute quantities like the internal energy in
terms of fermi functions. For example we have

R∞ R∞ x 3/2
0
D (ε) g (ε) εdε 0 z −1 e x +1
dx Γ (5/2) f 5/2 (z )
U = R∞ = NkBT R ∞ x 1/2 = NkBT (43)
D (ε) g (ε) dε dx Γ (3/2) f 3/2 (z )
0 0 z −1 e x +1

Summarizing, in this view we found:


X −e β µ k
someN

for 0 < z = e β µ < 1



 −



k =1


f ν e βµ ≈
€ Š
 ν ∞
!
βµ Γ (ν + 1) 22k −1 − 1 π2k B 2k
 X 
(−1)k +1
 −2k
1+ βµ for large z = e β µ >> 1
 

Γ (ν + 1) k =1
Γ (ν − 2k + 1) Γ (2k ) k
2 THE BERNOULLI NUMBERS B N 11

2 The Bernoulli Numbers B n


2.1 The Generating Function
By generating function usually a function is meant, whose Taylor expansion yields the seeked for expres-
sion. In our case it will be a simple fraction:
x
f (x ) = ∀x 6= 0, f n (0) = 0, ∀n = 0, 1, . . . , n ∈ N, (44)
ex −1
The Bernoulli numbers are defined to be the coefficients of the taylor expansion for this function. However,
to obtain them we have to go a different way than usually.

X Bn
f (x ) := xn (45)
n =0
n!

by definition we have the taylor series for the exponential function:



X xk
e x := (46)
k =0
k!

!−1
∞  k −1  −1 !−1
!
∞  k  ∞ 
1 X x 1 X x X xk
f (x ) = Š1 = = = (47)
(k + 1)!
€ P∞ xk
−1 k =1
k! x k =1
k! k =0
k =0 k ! x

X Bn
= xn (48)
n=0
n!
‚∞ Œ ∞  !
X Bn X xk
⇒ x n
=1 (49)
n=0
n! k =0
(k + 1)!

Now we make use of the Cauchy Product of series (see Appendix A):
! !
∞ n ∞ n
X X Bk x k x n −k X X xn
1= = Bk (50)
n=0 k =0
k ! (n − k + 1)! n =0 k =0
k ! (n + 1 − k )!

We use the definition of n choose k to rewrite this (see Appendix B):

∞ n !
n +1 xn
X X 
1= Bk (51)
n=0 k =0
k (n + 1)!

This equation is somewhat equivalent to a binomial expression if one defines the exponent of B to be
understood as a subscript (Then compare eqn. (51) with (50)):

n+1
n +1
X  
k n +1−k
n+1
(B + 1) −B n+1
= B 1 − B n +1 (52)
k =0
k
n n
k n +1 n+1 n + 1 k n +1
X     X  
= B +B −B n+1
= B (53)
k =0
k n +1 k =0
k

2.2 The Recurrence Formula


Now the procedure of comparison of coefficients yields the first Bernoulli number B 0 :
0
0+1 x0
 
X 1! 1!
1= Bk = B0 ⇒ B0 = 1 (54)
k =0
k (0 + 1)! 0!1! 1!

For the following numbers we use a recurrence formula by noting that comparison of coefficients tells us:
For all n greater 0, the sum in the brackets of equation (51) has to vanish!
3 THE RIEMANN-ZETA FUNCTION ζ (X ) 12

n
n +1
X  
Bk =0 ∀ n ≥ 1, n ∈ N, (55)
k =0
k

A nice interpretation of this formula is sugested by the similarity to the Binomial expansion introduced in
equation (50) and (51). We only have to write down the appropriate row of the Pascal Triangle for (B + 1)n+1
without the last term

1
n =1 0 = B0 + B1 ⇒ B1 = −
2
1
n =2 0 = B 0 + 3B 1 + 3B 2 ⇒ B2 =
6
n =3 0 = B 0 + 4B 1 + 6B 2 + 4B 3 ⇒ B3 = 0
1
n =4 0 = B 0 + 5B 1 + 10B 2 + 10B 3 + 5B 4 ⇒ B4 = −
30
n =5 0 = B 0 + 6B 1 + 15B 2 + 20B 3 + 15B 4 + 6B 5 ⇒ B5 = 0
1
n =6 0 = B 0 + 7B 1 + 21B 2 + 35B 3 + 35B 4 + 21B 5 + 7B 6 ⇒ B6 =
42

n 0 1 2 4 6 8 10 12 14 16 18 20

Bn 1 − 21 1
6
1
− 30 1
42
1
− 30 5
66
691
− 2730 7
6
− 3617
510
43867
798
− 174611
330

Table 1: some Bernoulli Numbers, note: B 2n+1 = 0 ∀n ≥ 1

3 The Riemann-Zeta Function ζ (x )


The aim will be to express the Riemann Zeta function with the help of the Bernoulli numbers. To do so we
will find two notions of one and the same expression and eventually compare both to obtain the result.

3.1 First Notion of x2 coth x2

=1
z}|{
cosh x2
x
− x2 x
x x x x e +e 2 e2
coth = = x × (56)
sinh x2
x x
2 2 2 2 e 2 −e−2 e2

x ex +1 x x X Bn n x
= = + = x + (57)
2 e x − 1 e x − 1 2 n=0 n! 2

If we now note that B 0 = 1 and B 1 = −1/2 we get


∞ ∞
x x 1x 0 x1 X Bn n x X Bn n
coth = − + x + =1+ x (58)
2 2 0! 2 × 1! n=2 n! 2 n=2
n!

Now it also gets clear why table 1 only lists even subscripts of Bernoulli numbers. The cotangens hyper-
bolicus, being a quotient of an even and an odd function, is odd again. x is an odd function as well, and
hence x2 coth x2 is an even function. Therefore, by this representation of our series we note that B 2n+1 = 0
for all n greater or equal 1. (B 2n +1 = 0 ∀n = 1, 2, . . .)
Therefore we can write down more succintly:

x x X B 2n 2n
coth = 1 + x (59)
2 2 n =1
(2n)!
3 THE RIEMANN-ZETA FUNCTION ζ (X ) 13

3.2 Second Notion of x2 coth x2



3.2.1 Fourier Series of cos y x

In this section we will calculate the Fourier Series of the following function g(x). Later we will take a special
value for x in the found expression to continue:

g (x ) = cos y x , D = [−π, π]

y ∈ R\Z, (60)

The general formula of the Fourier series is:



a0 X
g (x ) = + (a k cos k x + b k sin k x ) (61)
2 k =1

Where it saves a lot of time to note that the coefficients b k are all zero, since our founction is even! I will
also use the following relations during the calculation of a k :

1
cos x cos y = cos x − y cos x + y , sin(−x ) = − sin(x )
 
2

sin(a + b ) = sin(a ) cos(b ) + cos(a ) sin(b ), sin(k π) = 0, cos(k π) = (−1)k

Z +π Z +π
1 1 1
= cos y x cos k x dx = cos y x − k x cos y x + k x dx
 
ak (62)
π −π
π −π
2
  π
+k x 1 sin y − k π cos y + k π
       
1 sin y − k x cos y
= + = + (63)
2π y −k y +k −π π y −k y +k
y + k sin y π cos k π − sin k π cos y π + y − k sin y πcos k π + sin k π cos y π
    
1
= (64)
π y +k y −k
 

1 y sin(y π)(−1) + k sin(y π)(−1) + y sin(y π)(−1)k − k sin(y π)(−1)k


k k
 
= (65)
π y 2 −k2
1 2y sin(y π)(−1)k
 
= (66)
π y 2 −k2
Z +π  +π  
1 1 1 1 2
a0 = cos(y x )dx = sin(y x ) = sin(y π) (67)
π −π
π y −π π y
Plugging these coefficients back into equation (61) gives our Fourier Series of g(x):

sin(y π) X 1 2y sin(y π)(−1)k
 
cos(y x ) = + cos(k x ) (68)
yπ k =1
π y 2 −k2

Now we consider the special case x = π:


 
∞ 
sin(y π) 1 X 2y (−1)k

cos(y π) =  + (−1)k  (69)
π y k =1 y 2 − k 2


cos(y π) 1 X 2y
⇒π× = + = π cot(y π) (70)
sin(y π) y k =1 y 2 − k 2

This is the so-called partial fraction decomposition of the cotangens function.


3 THE RIEMANN-ZETA FUNCTION ζ (X ) 14

x
3.2.2 Transformation to 2
coth x2

We will now transform the regular cotangens and its partial fraction decomposition into the second form
of the cotangens hyperbolicus we need for the comparison ahead.
1
e i ×i y + e −i ×i y
0 0 
e −y + e y
0 0 
2
cot(i y ) =
0
1 0 0  =i 0 0  = −i coth(y 0 ) (71)
2i
e i ×i y − e −i ×i y e −y − e y
We multiply this equation by i π and let y 0 = y π to fit our purposes:

i π cot(i y 0 ) = π coth(y 0 ) ⇒ i π cot(i y π) = π coth(y π) (72)


 
∞ ∞ ∞
1 X 2i y 1 X −2y 1 X 2y
π coth(y π) = i  + = + = + (73)
i y k =1 (i y )2 − k 2 y k =1 −y 2 − k 2 y k =1 y 2 + k 2
Now we will do the following substitution to get the right arguement in the cotangens hyperbolicus:
x x x
yπ= ⇒ y= ⇒ π= (74)
2 2π 2y

x x 1 X 2y
 ‹
coth = + x2
(75)
2y 2 y n =1 + n2
4π2
x 2
∞ ∞
x x
 ‹ X 2 4π 2
X 2x 2
⇒ coth =1+ =1+ (76)
2 2 n=1
x2
+ n2 n =1
x 2 + 4π2 n 2
4π2
We need to recognize that this infinite sum represents a geometric series (see Appendix C)

2
∞ 
− 4πx2 n 2
k
X x2 x2
− − 2 2 = − = (77)
k =1
4π n 1+ x2 x 2 + 4π2 n 2
4π2 n 2
∞ ‹2k
x
X 
= (−1)k +1 (78)
k =1
2πn
∞ X ∞ ∞ ∞
‹2k – ™
x x x 1
 ‹ X  X X
⇒ coth =1+2 (−1)k +1 =1+2 (−1)k +1 2k
x 2k (79)
2 2 n=1 k =1
2πn k =1 n =1 (2πn)

3.3 Comparison and Result


Now we compare our found equations (79) and (59)

∞ ∞
– ™
x x 1
 ‹ X X
k +1
coth = 1+2 (−1) x 2k (80)
2 2 k =1 n =1 (2πn)2k

B 2n 2nX
= 1+ x (81)
n=1
(2n)!
∞ ∞ ∞
– ™
X
k +1
X 1 1 X B 2k 2k
⇒ (−1) x 2k
= x (82)
k =1 n=1 (2πn)
2k 2 k =1 (2k )!
Comparison of coefficients yields:

X 1 1 B 2k
(−1)k +1 = (83)
n =1 (2πn)
2k 2 (2k )!

X 1 B 2k
= (−1)k +1 (2π)2k = ζ(2k ) (84)
n =1
n 2k 2 (2k )!
This is the final result for the Riemann Zeta function of even integers. Up to now, there is no closed
form for odd integers known to man. It also follows (since we have a sum of non negative numbers) that
B 2k (−1)k +1 ≥ 0.
3 THE RIEMANN-ZETA FUNCTION ζ (X ) 15

3.4 Application
The first few results of the Riemann Zeta function are the following:


X 1 B2 π2
ζ(2) = = (−1)2 (2π)2 = π2 B 2 = (85)
|{z}
n=1
n 2 2 (2)! 6
k =1

X 1 B4 23 π4 π4
ζ(4) = = (−1)3 (2π)4 =− B4 = (86)
|{z}
n=1
n 4 2 (4)! 4! 90
k =2

X 1 B6 26 π6 π6
ζ(6) = = (−1)4 (2π)6 = B6 = (87)
|{z}
n=1
n 6 2 (6)! 2 × 6! 945
k =3

Further studies on and special application of the Riemann Zeta function:



1 1 1 1 1 1 1 1 1 1
X  ‹
= + + + + . . . = + + + + . . . (88)
n =0
(2n + 2)p 2p 1 p 2 p 3p 4 p (1 × 2)p (2 × 2)p (3 × 2)p (4 × 2)p
| {z }
even numbers

1 1 1 1 1 X 1 1
= p
+ p + p + p + ... = = ζ(p ) (89)
2 4 6 8 2p n=1 n p 2p

∞ ∞ ∞ ∞ ∞
X 1 X 1 X 1 X 1 1 X 1
ζ(p ) = = + = + (90)
n =1
np n =0
(2n + 1)p n=0
(2n + 2)p n=0
(2n + 1)p 2p n =1 n p
| {z } | {z }
odd numbers even numbers


1 X 1 2p − 1
⇒ ζ(p ) − ζ(p ) = = ζ(p ) (91)
2 p
n =0
(2n + 1)p 2p

∞ ∞   p
1 1 1 1 1 1 1 2 −1
X X ‹ ‹
− + − + = 2 − = 2 − 1 ζ(p ) (92)
1p 2p 3p 4p 5p n =0
(2n + 1)p n =0 (n + 1)p 2p

(−1)n
 p
2 −2
‹ X
= ζ(p ) = (93)
2 p
n =0
(n + 1)p

Summarizing we have also found the following infinite series:


X (−1)n 2p − 2
= ζ(p ) (94)
n =0
(n + 1)p 2p

X 1 2p − 1
= ζ(p ) (95)
n =0
(2n + 1)p 2p

X 1 1
= ζ(p ) (96)
n =0
(2n + 2)p 2p

Note that one can not put p = 1 in equation (94) to obtain a result for ζ(1) in dependnece on ln (2) since the
Riemann zeta function has a singularity at this point due to its reduction to the diverging harmonic series
there.
A CAUCHY PRODUCT 16

APPENDIX

A Cauchy Product
‚∞ Œ‚ ∞ !
∞ n
Œ
X X X X
an bn = a k b n −k (97)
n=0 n =0 n=0 k =0

Note: If both series on the left converge absolutely, the Cauchy series converges absolutely as well. In this
case, its limit is the product of the limits of the series on the left.

B n choose k
 
n n!
:= for n, k ∈ N (98)
k k ! (n − k )!

C The Geometric Series

sn = 1 +q1 +q2 +q3 + ... +qn (99)


sn × q = q + q 2 + q 3 + . . . + q n + q n+1 (100)

Substraction of equation (112) from (111) gives

n
X 1 − q n+1
s n (1 − q ) = 1 − q n+1 ⇒ sn = qk = (101)
k =0
1−q
n ∞
X 1 − q n+1 1 − q n 0 −1+1 q n 0 − q n +1 X q n0
sn − sn0 = qk = − = ⇒ qk = for q ≤ 1 (102)
k =n 0
1−q 1−q 1−q k =n 0
1−q

X z
⇒ zk = (103)
k =1
1−z

D Special Functions

2 2
e z /2 ez
 
1 1
= = (104)
2 cosh (z /2) e z /2 + e −z /2 e z /2 (e z + 1)2

Z ∞
Γ (x ) = z x −1 e −z dz = (x − 1) Γ (x − 1) = (x − 1)! for x ∈ N (105)
0

X 1
ζ (x ) = x
(106)
n =1
n
REFERENCES 17

References
[1] L. Landau, E. Lifshitz: Statistical Physics Part I, 3rd Edition, page 169/170

[2] http://personal.rhul.ac.uk/UHAP/027/PH4211/PH4211-files/-section2.pdf

[3] http://www.mathe-seiten.de/bernoulli.pdf, Bernoulli-Zahlen, Zetafunktion und Summen von Poten-


zen

[4] http://numbers.computation.free.fr/Constants/Miscellaneous/bernoulli.html, Introduction on


Bernoulli’s numbers, Xavier Gourdon, Pascal Sebah

[5] http://mathworld.wolfram.com/BernoulliNumber.html

[6] http://en.wikipedia.org/wiki/Riemannzetafunction, Riemann Zeta function,


http://en.wikipedia.org/wiki/Trigonometricfunction, Trigonometric functions

[7] http://planetmath.org/encyclopedia/ExamplesOnHowToFindTaylorSeriesFromOtherKnownSeries.html

[8] http://mathworld.wolfram.com/StirlingsApproximation.html

[9] http://scipp.ucsc.edu/ haber/ph116A/esum.pdf Fantastic script to Bernoulli numbers and the Euler
McLaurin summation formula!

[10] http://personal.rhul.ac.uk/UHAP/027/PH4211/PH4211-files/-appendixes.pdf

[11] http://www.physics.umd.edu/courses/Phys603/kelly/Notes/IdealQuantumGases.pdf

[12] http://www.physics.umd.edu/courses/Phys603/kelly/Notes/fermi.nb

[13] Chebyshev Polynomial Expansion of Bose-Einstein Functions of Orders 1 to 10, By Edward W.Ng, C.J.
Devine and R.F.Tooper, 1969

[14] A generalized approximation of the Fermi-Dirac integrals, By Aymerich-Humet, F. Serra-Mestres and J.


Millán, 1982

Anda mungkin juga menyukai