Last Latexed: November 8, 2007 at 10:28 1

Schwinger trick and Feynman Parameters

c
Copyright 2005 by Joel A. Shapiro

Here is the way Schwinger presented the method of combining propaga-

tors. An interesting anecdote of physics history is that Schwinger remained
bitter that a virtually identical mathematical trick became commonly known
as Feynman parameters. Why two brilliant physicists, each of whom had an
appropriately won a Nobel prize, should fight over what is essentially a trivial
mathematical trick, is an interesting question in the sociology of physicists.
But we are not interested in that now.
Note that
1
Z ∞
= dν e−Aν ,
A 0
at least when Re A > 0, where the integral is well defined. Applying this to
each of the terms in a product of inverses,
N
X
N N Z ∞ − Ai νi !
Y 1 Y
= dνi e i=1 .
i=1 Ai i=1 0

Let ν = νi and αi = νi /ν. Then

P

N N N
!
dνi = ν N −1 dν
Y Y X
dαi δ 1 − αi ,
i=1 i=1 i=1

so
N N Z ∞ N
! !Z
1 ∞ P
ν N −1 dνe−ν αi Ai .
Y Y X
= dαi δ 1 − αi
i=1 Ai i=1 0 i=1 0

But ∞ 1
Z
tz−1 e−bt dt = Γ(z),
0 bz
where Z ∞
Γ(z) = tz−1 e−t dt = (z − 1)!
0
is the Euler Gamma function. So all together
N N Z ∞ N
! !
Y 1 Y X hX i−N
= Γ(N) dαi δ 1 − αi αi Ai . (1)
i=1 Ai i=1 0 i=1
Schwinger/Feynman trick Last Latexed: November 8, 2007 at 10:28 2

which is equation 6.41.

The simplest application would be something like
p
d4 k 1 1
Z

(2π) (p − k) − m + iǫ k + iǫ
4 2 2 2

which is something like what we would encounter if we k

cared about the second diagram of Fig. 6.1, which contains p−k
the (amputatable) correction shown. In terms of Feynman
parameters, this is
Z
d4 k 1 1 p
(2π) (p − k) − m + iǫ k + iǫ
4 2 2 2

Z 1 Z 1
= Γ(2) dα dβ δ(1 − α − β)
0 0
4
dk 1
Z

(2π)4 (α(p − k)2 − αm2 + βk 2 + i(α + β)ǫ)2

d4 k 1
Z 1 Z
= dα
0 (2π) (k − 2αk · p + iǫ)2
4 2

This can be simplified by noting that the denominator can be written as

the square of D, where D is defined in terms of a shifted four-momentum
ℓµ = k µ − αpµ as
D = ℓ2 − α2 m2 + iǫ = ℓ2 − ∆ + iǫ, with ∆ = α2 m2 .
Thus our integral can be written as
1 d4 ℓ 1
Z Z
dα
0 (2π) (ℓ − ∆ + iǫ)n
4 2
Γ’ A
l0
where the integrand is invariant under
− l 2+ ∆ +i ε
Lorentz transformations, and I have gen- Γ
eralized 2 to an arbitrary power n > 21 .
For a fixed ~ℓ, the ℓ0 integral is along the l 2+ ∆−i ε
path Γ, but we can throw in the arcs A at A
infinity, and then deform the integral to be
along Γ′ , as we are not passing any singu-
larities. Writing ℓ0 = iL4 and ~ℓ = L,~ our
integral is now
Schwinger/Feynman trick Last Latexed: November 8, 2007 at 10:28 3

1 d4 L 1
Z Z
n
(−1) i dα ,
0 (2π) (L + ∆)n
4 2

where the integral is now over Euclidean space, with L2 = 41 (Li )2 . In fact,
P

let’s generalize further to an arbitrary number of space-time dimensions d. As

the integrand is rotationally invariant in d dimensions, the angular integral
dΩd is just the area Sd of a unit d − 1 sphere, where Sd = 2, 2π, 4π, 2π 2 for
R

d = 1, 2, 3, 4 respectively. We will derive the general expression later. The

full measure of integration is dV = r d−1 drdΩd so we have
1 dd ℓ (ℓ2 )p (−1)n+p i Sd 1 ∞ Ld+2p−1
Z Z Z Z
dα = dα dL .
0 (2π)d (ℓ2 − ∆(α) + iǫ)n (2π)d 0 0 [L2 + ∆(α)]n

This is treated in the book a bit later, with this equation as 7.80

In other notes1 we show that

π d/2 ∞ uα−1 Γ(α)Γ(β)
Z
Sd = 2 and α+β
du = B(α, β) = .
Γ(d/2) 0 (1 + u) Γ(α + β)
so
dd ℓ (ℓ2 )p
Z
(2)
(2π)d (ℓ2 − ∆(α) + iǫ)n
i (−1)n+p Γ( 21 d + p)Γ(n − p − d/2)   1 d+p−n
2
= ∆(α) .
(4π)d/2 Γ(d/2)Γ(n)

1
“Γ(N/2) and the volume of S D−1 ” and “The Beta function B(x, y)”.