Appendix

Gamma, beta and error functions

In several places in this book we have made mention of the gamma, beta and error
functions. These convenient functions appear in a number of contexts and here
we gather together some of their properties. This appendix should be regarded
merely as a reference containing some useful relations with a minimum of formal
proofs.

A1.1 The gamma function

The gamma function Γ(n) is defined by
 ∞
Γ(n) = xn−1 e−x dx, (A1)
0

which converges for n > 0. Replacing n by n + 1 in (A1) and integrating the RHS
by parts, we find
 ∞
Γ(n + 1) = xn e−x dx
0
 ∞
 n −x ∞
= −x e 0 + nxn−1 e−x dx
 ∞ 0

=n xn−1 e−x dx,

0

from which we obtain the important result

Γ(n + 1) = nΓ(n). (A2)

From (A1), we see that Γ(1) = 1, and so, if n is a positive integer,

Γ(n + 1) = n!. (A3)

1201
Downloaded from https://www.cambridge.org/core. IIT Kanpur, on 24 Sep 2020 at 09:06:59, subject to the Cambridge Core terms of use,
available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139164979.031
 GAMMA, BETA AND ERROR FUNCTIONS

Γ(n )
6

n
−4 −3 −2 −1 1 2 3 4

−2

−4

−6

Figure A1.1 The gamma function Γ(n).

In fact, equation (A3) serves as a definition of the factorial function even for
non-integer n. For negative n the factorial function is defined by
(n + m)!
n! = , (A4)
(n + m)(n + m − 1) · · · (n + 1)
where m is any positive integer that makes n + m > 0. Different choices of m
(> −n) do not lead to different values for n!. A plot of the gamma function is
given in figure A1.1, where it can be seen that the function is infinite for negative
integer values of n, in accordance with (A4).
By letting x = y 2 in (A1), we immediately obtain another useful representation
of the gamma function given by
 ∞
y 2n−1 e−y dy.
2
Γ(n) = 2 (A5)
0

1
Setting n = 2 we find the result
 ∞ 
  ∞ √
e−y dy = e−y dy =
2 2
Γ 12 = 2 π,
0 −∞

where have used the standard integral discussed in subsection 6.4.2. From this
result, Γ(n) for half-integral n can be found using (A2). Some immediately
derivable factorial values of half integers are
 3 √  1 √ 1 1√
3 3√
− 2 ! = −2 π, − 2 ! = π, 2 ! = 2 π, 2 ! = 4 π.

1202
Downloaded from https://www.cambridge.org/core. IIT Kanpur, on 24 Sep 2020 at 09:06:59, subject to the Cambridge Core terms of use,
available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139164979.031
 A1.2 THE BETA FUNCTION

It can also be shown that the gamma function is given by

 
√ n −n 1 1 139
Γ(n + 1) = 2πn n e 1+ + − + . . . = n!, (A6)
12n 288n2 51 840n3
which is known as Stirling’s asymptotic series. For large n the first term dominates
and so
√
n! ≈ 2πn nn e−n ; (A7)
this is known as Stirling’s approximation. This approximation is particularly useful
in statistical thermodynamics, when arrangements of a large number of particles
are to be considered.
√

Prove Stirling’s approximation n! ≈ 2πn nn e−n for large n.

From (A1), the extended definition of the factorial function (which is valid for n > −1) is
given by
 ∞  ∞
n! = xn e−x dx = en ln x−x dx. (A8)
0 0

If we let x = n + y, then
y

ln x = ln n + ln 1 +
n
y y2 y3
= ln n + − 2 + 3 − · · · .
n 2n 3n
Substituting this result into (A8), we obtain
 ∞    
y y2
n! = exp n ln n + − 2 + · · · − n − y dy.
−n n 2n
Thus, when n is sufficiently large, we may approximate n! by
 ∞ √ √
2
n! ≈ en ln n−n e−y /(2n) dy = en ln n−n 2πn = 2πn nn e−n ,
−∞

which is Stirling’s approximation (A7). 

A1.2 The beta function

The beta function is defined by
 1
B(m, n) = xm−1 (1 − x)n−1 dx, (A9)
0

which converges for m > 0, n > 0. By letting x = 1 − y in (A9) it is easy to show

that B(m, n) = B(n, m). Other useful representations of the beta function may be
obtained by suitable changes of variable. For example, putting x = (1 + y)−1 in
(A9), we find that
 ∞
y n−1 dy
B(m, n) = m+n
.
0 (1 + y)

1203
Downloaded from https://www.cambridge.org/core. IIT Kanpur, on 24 Sep 2020 at 09:06:59, subject to the Cambridge Core terms of use,
available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139164979.031
 GAMMA, BETA AND ERROR FUNCTIONS

Alternatively, if we let x = sin2 θ in (A9), we obtain immediately

 π/2
B(m, n) = 2 sin2m−1 θ cos2n−1 θ dθ. (A10)
0

The beta function may also be written in terms of the gamma function as
Γ(m)Γ(n)
B(m, n) = . (A11)
Γ(m + n)

Prove the result (A11).

Using (A5), we have

 ∞  ∞
2 2
Γ(n)Γ(m) = 4 x2n−1 e−x dx y 2m−1 e−y dy
0 ∞ ∞ 0

2n−1 2m−1 −(x2 +y 2 )

=4 x y e dx dy.
0 0

Changing variables to plane polar coordinates (ρ, φ) given by x = ρ cos φ, y = ρ sin φ, we

obtain
 π/2 ∞
2
Γ(n)Γ(m) = 4 ρ2(m+n−1) e−ρ sin2m−1 θ cos2n−1 θ ρ dρ dθ
0 0
 π/2  ∞
2
=4 sin 2m−1
θ cos 2n−1
θ dθ ρ2(m+n)−1 e−ρ dρ
0 0
= B(m, n)Γ(m + n),
where in the last line we have used the results (A5) and (A10). 

A1.3 The error function

Finally we mention the error function, which is encountered in probability theory
and in the solutions of some partial differential equations, and which is defined
by
 x  ∞
2 2
e−u du = 1 − √ e−u du.
2 2
erf(x) = √ (A12)
π 0 π x
From this definition we can easily see that

erf(0) = 0,
erf(∞) = 1, erf(−x) = −erf(x).
√
By making the substitution y = 2u in (A12), we find
 √2x
1
e−y /2 dy.
2
erf(x) = √
2π 0
The cumulative probability function Φ(x) for the standard Gaussian distribution
(discussed in section 26.9.1) may be written in terms of the error function as

1204
Downloaded from https://www.cambridge.org/core. IIT Kanpur, on 24 Sep 2020 at 09:06:59, subject to the Cambridge Core terms of use,
available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139164979.031
 A1.3 THE ERROR FUNCTION

follows
 x
1
e−y /2 dy
2
Φ(x) = √
2π −∞
 x
1 1
e−y /2 dy
2
= +√
2 2π 0
 
1 x
= + erf √ .
2 2
It is also sometimes useful to define the complementary error function
 ∞
2
e−u du.
2
erfc(x) = 1 − erf(x) = √ (A13)
π x

1205
Downloaded from https://www.cambridge.org/core. IIT Kanpur, on 24 Sep 2020 at 09:06:59, subject to the Cambridge Core terms of use,
available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/CBO9781139164979.031