Deriving the Stefan-Boltzmann law from Planck’s law

Peter Haggstrom
[email protected]
https://gotohaggstrom.com

February 25, 2020

1 Background
The purpose of this short article is to explain the integral that underpins the Stefan-
Boltzmann law, being a limiting case of Planck’s law of radiation. Undergraduate physics
courses almost invariably present Planck’s Law, without much discussion, if any, as to
how he arrived at it. They leave that to the historians of physics and the definitive
explanation of that process is probably given by Max Jammer in [1] (pages 1-21). Jammer
was an experimental physicist who also did the history of science in his long life (he died
at age 95 in 2010). He was a colleague of Einstein and the material in [1] is based on
discussions with, and papers written by, all the ”heavy hitters” of 20th century quantum
theory.
Jammer explains that in essence what Planck did was to take two expressions of
how entropy was related to the energy of an harmonic oscillator at temperarture T and
concoct a ”compromise”. Specifically (and you should read Jammer’s book to follow the
detail), Planck had these two expressions to work with:

∂2S constant
2
= (1)
∂U U
and

∂2S constant
= (2)
∂U 2 U2
Here S is the entropy and U is the energy. What Planck assumed was that:

∂2S a
2
= (3)
∂U U (U + b)
where a, b are constants.

Thus when U is large (3) approximates (2) and when it is small it approximates (1).

1
 Jammer says the following ( [1], page 14) about this act of inspiration:

”This interpolation, though mathematically a mere trifle, was one of the most signif-
icant and momentous contributions ever made in the history of physics. Not only did it
lead Planck, in his search for its logical corroboration, to the proposal of his elementary
quantum of action and thus initiated the early development of quantum theory, as we
shall see presently; it affected decisively the very foundations of physics as well as their
epistemological presuppositions. Never in the history of physics was there such an incon-
spicuous mathematical interpolation with such far-reaching physical and philosophical
consequences.”

From (3) Planck deduced that:

1 ∂S 0
U + b
= = a log (4)
T ∂U U
or
1
U= 1

(5)
0
e a T −1
0 −a
where a = b are functions of frequency v. Using some other results Planck con-
cluded that:

Av 3
uv = Bv

(6)
e T −1
where A, B are constants and uv is energy density at frequency v.

What is typically done in physics courses is that it is observed how Planck’s law
reduces to the Rayleigh-Jeans formula in the low frequency limit and to Wien’s formula
in the high frequency limit.

In addition it is usually left as an exercise (look at various forums frequented by

physics students if you doubt me ) to prove that Stefan-Boltzmann law follows from an
appropriate integration of (6).

Recall that the Stefan-Boltzmann law states that the total energy flux Φ emitted
from a blackbody at temperature T is:

Φ = σT 4 (7)
where σ is a constant.

2
2 The nittgy gritty of the integration
Since this article is really about how the integration works in detail, I will simply set the
problem up as is done in many places such as Wikipedia. Just keep in mind there are
various equivalent ways of expressing the constants involved in Planck’s law. We start
with the intensity of the light emitted from a blackbody in accordance with Planck’s
law:

2hν 3 1
I(ν, T ) = (8)
c2


hν
e −1 kT

The power radiated by a surface of area A through a solid angle dΩ in the differential
frequency range (ν, ν + dν) is:

I(ν, T )A dν dΩ (9)
where:

dΩ = sin θ dθ dφ (10)
where θ is the angle from the North Pole and φ is the longitude.

The Stefan-Boltzmann law says that the power emitted per unit area of the emitting
body is:
Z ∞ Z
P
= I(ν, T ) dν cos θ dΩ (11)
A 0
Note that the cosine integral appears because black bodies are Lambertian (i.e. they
obey Lambert’s cosine law), meaning that the intensity observed along the sphere will
be the actual intensity times the cosine of the zenith angle (ie angle from the North
Pole). Hence we have:

Z ∞ Z 2π Z π/2
P
= I(ν, T ) dν dφ cos θ sin θ dθ
A 0 0 0
| {z }
=2π×π/2
Z ∞
=π I(ν, T ) dν
0 (12)
∞
2hν 3
Z
1
=π dν
c2


hν
0 e kT −1
3 Z ∞ hν 3


2πk 3 T kT

= dν
h2 c2 0
hν
e kT −1

3
 R∞ 3
So what we have to integrate is of the form 0 euu−1 du. There are various ways of
calculating this integral, some of which involve the Riemann Zeta function eg see [2] .
Many physics textbooks just say it is a well known result and just state the answer.
Now theR first question you have to ask yourself from a purely mathematical nature
∞ 3
is whether 0 euu−1 du converges. A physicist wouldn’t spend too much time on such a
question because it is clear that because an exponential grows faster than any power the
ratio in the integral decays so rapidly that convergence is assured. That is a perfectly
acceptable reason. What about the behaviour at u = 0?

In that context the physicist would simply note that:

u3 u3 u2
= 2 = u (13)
eu − 1 1 + u + u2! + · · · − 1 1 + 2! + ...
So far from blowing up at u = 0, it is simply 0.

Now suppose you are being questioned by an anally retentive mathematician who
says ”Prove it is convergent!”. What is your actual proof?
Looking at (13) we have:

u u2 un u 1 u
2 1 u
n u
1+ + + ··· + + ··· ≥ 1 + + + ··· + + · · · = e 2 (14)
2! 3! (n + 1)! 2 2! 2 n! 2
un 1 u n


since (n+1)! ≥ n! 2 for n ≥ 1 (noting u ≥ 0).
Hence from (13) and (14) we have:

u3 u2
= u
eu − 1 1 + 2! + ... (15)
2 −u
≤u e 2

Convergence is now a heartbeat away since we have:

∞ ∞
u3
Z Z
u
du ≤ u2 e− 2 du
0 eu − 1 0
Z ∞
=2 (2t)2 e−t dt (16)
0
=8 Γ(3)
=16
R∞
Recall that Γ(u + 1) = 0 tu e−t dt and for n positive and integral, Γ(n + 1) = n!

This ought to satisfy the most anally retentive mathematician!

4
 To actually perform the integration requires some analysis. First of all we get by
long division:

∞ ∞
u3
Z Z
du = u3 (e−u + e−2u + e−3u + . . . ) du
0 eu − 1 0
Z ∞
∞ X (17)
= u3 e−ku du
0 k=1

Now (17) screams interchange of integration and summation since we will then have
a sum of Gamma functions. For instance, by making the substitution t = ku we get:
Z ∞
u3 e−u du =Γ(4)
Z 0∞ Z ∞
1 1 1
u3 e−2u du = u3 e−u du = Γ(4) = 4 Γ(4)
16 16 2
Z0 ∞ Z0
3 −3u 1 ∞ 3 −u 1
u e du = 4 u e du = Γ(4)
0 3 0 34 (18)
..
.
Z ∞ Z ∞
3 −ku 1 1
u e du = 4 u3 e−u du = Γ(4)
0 k 0 k4
..
.
To interchange the order or of integration and summation we need to establish that
for every u ≥ 0 this infinite series is uniformly convergent:
∞
X
u3 e−ku (19)
k=1
The Weierstrass M-test is the key to establishing this and
Pwe need to find a sequence
of dominating constants, independent of u, Mk , such that ∞ k=1 Mk converges. To this
end we note that:

u3
u3 e−ku = (ku)2 (ku)3
1 + ku + 2! + 3! + ... (20)
6
≤ 3
k
Hence:

∞ ∞
X X 1
u3 e−ku ≤6
k3 (21)
k=1 k=1
=6 ζ(3)

5
 where ζ is the Zeta function.
On this basis we can interchange the order of integration and summation in (17),
yielding:

∞ ∞
∞ X
u3
Z Z
du = u3 e−ku du
0 eu − 1 0 k=1
∞
XZ ∞
= u3 e−ku du (22)
k=1 0
∞
X 1
=Γ(4)
k4
k=1

P∞For1 those
π4
who know something about the sums of p-series, you will know that
k=1 k4 = 90 but how do you prove that?
,

Thus (12) can be finally calcuated, but before we do that we have to prove one more
thing.

P∞ 1 π4
3 Proof that k=1 k 4 = 90 using Fourier theory
Being a Fourier theory fan, I am biased towards such a proof, but in reality it is actually
a straightforward way of proving such a result. AfterPall, Parseval’s Theorem is regularly
π2
used to prove Euler’s famous Basel Problem result ∞ 1
k=1 k2 = 6 .

Recall that the Fourier series for a function f (x) is:

∞
a0 X
f (x) = + (an cos nx + bn sin nx) (23)
2
n=1

where:

1 π
Z
a0 = f (x) dx
π −π
1 π
Z
an = f (x) cos nx dx n = 1, 2, . . . (24)
π −π
1 π
Z
bn = f (x) sin nx dx n = 1, 2, . . .
π −π

If we choose an even function such as f (x) = x4 then the sin terms disappear and we
get the Fourier cosineR series When we do this note that due to the evenness of x4 cos nx
π
we have that an = π2 0 x4 cos nx dx for n = 0, 1, 2, . . . . So let’s see where that takes us:

6
 ∞
!
a0 X 2 π 4
Z
4
x = + x cos nx dx cos nx
2 π 0
n=1
∞
Z π !
Z π
2 4
X 2 4
= x dx + x cos nx dx cos nx (25)
2π 0 π 0
n=1
∞ Z π !
π 4 X 2
= + x4 cos nx dx cos nx
5 π 0
n=1

We now need to do a straightforward but typically tedious integration by parts as

follows:

π
2 π 4 1
Z Z
2 4


x cos nx dx = x d sin nx
π 0 π 0 n
−8 π 3
Z
= x sin nx dx
nπ 0
−8 π 3  −1
Z 
= x d cos nx dx
nπ 0 n
−8 −x cos nx π 1 π 2
3
h Z i
= + 3x cos nx dx
nπ n 0 n 0
2 Z π
8π 24
= 2 cos nπ − 2 x2 cos nx dx
n n π 0
Z π
8π 2 24 1
x2 d sin nx dx


= 2 cos nπ − 2
n n π 0 n
2 Z π
8π 24 −2
h i (26)
= 2 cos nπ − 2 x sin nx dx
n n π n 0
2 Z π
8π 48
= 2 cos nπ + 3 x sin nx dx
n n π 0
Z π
8π 2 48 −1

= 2 cos nπ + 3 xd cos nx dx
n n π 0 n
2 48 −x cos nx π 1 π
Z
8π h i
= 2 cos nπ + 3 + cos nx dx
n n π n 0 n
| 0 {z }
=0
8π 2 48
= 2 cos nπ + 4 × −π cos nπ
n n π
 8π 2 n2 − 48 
= cos nπ
n4
We now plug the final line of (26) into the final line of (25):

7
 ∞
π 4 X  8π 2 n2 − 48 
x4 = + cos nπ cos nx
5 n4
n=1
∞ 
π4 X 8π 2 n2 − 48 
π4 = + cos 2
| {znπ}
5 n4
n=1 =(−1)2n =1
∞ ∞
π4 X 1 X 1
π4 = + 8π 2 − 48
5 n2 n4
n=1 n=1 (27)
∞
π4 π2 X 1
π4 = + 8π 2 × − 48
5 6 n4
n=1
∞
π4 4π 4 X 1
π4 = + − 48
5 3 n4
n=1
∞
X 1 π4
∴ =
n4 90
n=1
P∞ 1 π2
Where we have used the fact that n=1 n2 = 6 (Euler’s Basel Problem).
Going back to (22) we therefore have:

∞ ∞
u3
Z X 1
du =Γ(4)
0 eu − 1 k4
k=1
π4 (28)
=3! ×
90
π4
=
15
hν
We now can solve (12) where we make the substitution u = kT :

∞ hν 3


P 2πk 3 T 3
Z
kT

= 2 2 dν
A h c 0
hν
e kT−1
2πk 3 T 3 ∞ u3 kT
Z
= 2 2 du
h c 0 eu − 1 h
2πk 4 T 4 π 4 (29)
= 3 2
h c 15
2π 5 k 4 4
= T
15h3 c2
=σ T 4

8
4 Historical comments
When Planck derived his radiation law in 1900, Lebesgue had not published his novel
theory of integration. Fourier theory was well known among physicists (for instance
Lord Kelvin had ”devoured” Fourier’s treatise on heat when he was an adolescent ) so
the steps in (25)-(27) would not have troubled them and they would have been aware
of the fact that convergence was not an issue. The conditions allowing exchange of
integration and summation were well known among top physicists. From a Lebesgue
integration perspective the interchange of integration and summation is perfectly valid
and is guaranteed by some very general theorems - see [3] Chapter 6.

5 References
[1] Max Jammer, ”Conceptual Development of Quantum Mechanics”, The History of
Modern Physics 1800-1950, Volume 12, Tomash Publishers, 1989. The book can be
downloaded here: https://b-ok.cc/book/3706316/7c4073

[2] http://scienceworld.wolfram.com/physics/Stefan-BoltzmannLaw.html

[3] Frank Jones, ”Lebesgue Integration on Euclidean Space”, Revised Edition,Jones

and Bartlett, 2001.

6 History
Created 18 February 2020
23 February 2020 - corrected typo in (27) (−1)2n = 1 instead of (−1)2 = 1