Summing Feynman diagrams in the worldline formalism

N. Ahmadiniaz,𝑎 J.P. Edwards,𝑏 C. Lopez-Arcos,𝑐 M.A. Lopez-Lopez,𝑎 C. Moctezuma

Mata,𝑑 J. Nicasio𝑑 and C. Schubert𝑒,∗
arXiv:2208.06585v1 [hep-th] 13 Aug 2022

𝑎 Helmholtz-Zentrum Dresden-Rossendorf, Bautzner Landstraße 400, 01328 Dresden, Germany

𝑏 Centre for Mathematical Sciences, University of Plymouth, Plymouth, PL4 8AA, UK
𝑐 Escuela de Matemáticas, Universidad Nacional de Colombia Sede Medellín, Carrera 65 # 59A-110,

Medellín, Colombia
𝑑 Instituto de Física y Matemáticas Universidad Michoacana de San Nicolás de Hidalgo Edificio C-3,

Apdo. Postal 2-82 C.P. 58040, Morelia, Michoacán, México

𝑒 Centro Internacional de Ciencias A.C., Campus UNAM-UAEM, Cuernavaca, Morelos, México

E-mail: christianschubert137@gmail.com

The worldline formalism shares with string theory the property that it allows one to write down
master integrals that effectively combine the contributions of many Feynman diagrams. While at
the one-loop level these diagrams differ only by the position of the external legs along a fixed line or
loop, at multiloop they generally involve different topologies. Here we summarize various efforts
that have been made over the years to exploit this property in a computationally meaningful way.
As a first example, we show how to generalize the Landau-Khalatnikov-Fradkin formula for the
non-perturbative gauge transformation of the fermion propagator in QED to the general 2𝑛 - point
case by pure manipulations at the path-integral level. At the parameter-integral level, we show
how to integrate out individual photons in the low-energy expansion, and then sketch a recently
introduced general framework for the analytical evaluation of such worldline integrals involving
a reduction to quantum mechanics on the circle and the relation between inverse derivatives and
Bernoulli polynomials.

Loops and Legs in Quantum Field Theory - LL2022,

25-30 April, 2022
Ettal, Germany

∗ Speaker

© Copyright owned by the author(s) under the terms of the Creative Commons
Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0). https://pos.sissa.it/
mpleting the square in the exponential, we obtain the following tree-level “Bern-Kosower-type 8
for
ion space for FIG. ??
Substituting this vertex operator in Eq. (??), and applying the split in Eq. (??), one gets
Summing Feynman diagrams in � the worldline formalism C. Schubert
∞N� ∞ �
� � D � 2 1 � 2 1 T 2
�
Γ[x, x ; k , ε ; · · · ; kN , εNN
] = (−ie) dT e −m
2 T − 1(x−x )
e � 2 Dq(τ− −
) e2 dτ q̇
x, x� ; k1 , ε1 ; · · · ; kN , εN ] 1 =1 (−ie) dT0 e−m T e− 4T (x−xq(0)=q(T ) 4T 4 0

4πT
)=0

0 � � T N �N � ��
(x−x� )
1. QED in the worldline×representation
ε ·
τ
+εi ·q̇(τi )+iki ·(x−x� ) Ti +iki ·x� +iki ·q(τi ) �
dτi e .
�
i
i=1 T
�
N
T � �N � (x−x� )
0 � �N � lin(ε1 ε2 ···εN ) ��
i=1 τ
εi · T +iki ·(x−x� ) Ti +iki ·x �
∆ij ki ·kj −2i •
∆ij εi ·kj −•∆•ij εi ·εj �
× dτi eWorldline
1.1 i=1 e i,j=1
representation of dressed scalar propagator (4.3) �
0 After completing the square in the exponential, we obtain the following tree-level “Bern-Kosower-type formula” in lin(ε1 ε2 ···εN
i=1 configuration space:
Let us start with Feynman’s 1950 worldline path integral representation [1] of the Green’s
� ∞
� 2� 2
function for� the interacting Klein-Gordon operator e −(𝜕 + 𝑖𝑒) 𝐴) +−𝑚 , 2 1 � 2 D

Fourier transform to momentum space also

Γ[x, x ; k 1 , ε 1 ; · · · ; k N , εN ] = (−ie) N
the
dT e−mscalar
T − (x−x
legs 4πTof the master formula Eq. (4.4), 4T 2

� T� N � ∫ ∞ 0 � � ��
� �
0
h (x−x� )
i
ε ·k − ∆ ε ·ε �
� τi � • • •
dτi e= h𝑥ε 0·|
N N
∆ 2k ·k −2i
×𝐷 𝑥 𝑥 [ 𝐴] +ik ·(x−x ) +ik ·x
𝑑𝑇 exp −𝑇 (−(𝜕
i=1 i
e + 𝑖𝑒 𝐴)
T i
+ 𝑚 2 )∆ |𝑥i
T i
� .
i,j=1 ij i j ij i j ij i j

0 i=1 �0 � lin(ε ε ···ε ) 1 2 N

� ∫ �
Γ[p; p� ; k1 , ε1 ; · · · ; kN , εN=] = 𝑑𝑇dDe−𝑚 x 2𝑇 dD x� eip·x+ip ·x𝑇 � (4.4)
∫ ∞ ∫ 𝑥 (𝑇 )=𝑥
Γ[x,
𝜏 41 𝑥¤ 2x
+𝑖𝑒;𝑥¤ ·k
𝐴(1𝑥,(ε 1 ; .· · · ; kN (1), εN ] .


− 0 𝑑 𝜏))
D𝑥(𝜏) 𝑒
Now, we also Fourier transform the scalar legs of the master formula in Eq. (??) to momentum space,
0
0 �𝑥 (0)=𝑥 �
� D � ip·x+ip� ·x�
Γ[p; p ; k1 , ε1 ; · · · ; kN , εN ] = d x d x e Γ[x, x� ; k1Í
D
, ε1 ; · · · ; kN , εN ] . (4.5)
ves a representation of the multi-photon Compton scattering diagram as depicted in FIG. 2 (toge
Choosing the background field as 𝑁 plane waves, 𝐴 (𝑥(𝜏)) = 𝑖=1 𝜇 𝑁
𝜀𝑖 e𝑖𝑘𝑖 ·𝑥 ( 𝜏) , and Fourier
𝜇
This gives a representation of the multi-photon Compton scattering diagram as depicted in FIG. ?? (together with
rmuted and seagulled ones). transforming
all the permutedtheandendpoints,
“seagulled” ones). we get a representation of the “photon-dressed propagator” shown in
Fig. 1.

k1 k2 k3 kN
p ··· p�
···

Figure
FIG. 3:1:Multi-photon
The photon-dressed scalar propagator.
Compton-scattering diagram.
FIG. 2: Multiphoton diagram in momentum space.

Note that the Changing the integration variables x, x� to

ng the integral variables toworldline representation
�
provides the complete amplitude in one piece: differently
�
from the Feynman diagrammatic approach,x − x = x−thereandis neither
x + x =a2x + , to sum over “crossed” diagrams,
need
the
norintegral over x+ just produces the usual energy-momentum conservation factor:
to distinguish between xthe x� =and
−linear x−the and
“seagull”x+ x� =
vertex couplings
2x+ , of the scalar particle to the
Maxwell field. � �N �� ∞ �
� N D D � −m2 T −D D 1
− 4T x2−
Γ[p; p ; k1 , ε1 ; · · · ; kN , εN ] = (−ie) (2π) δ p + p + ki dT e (4πT ) d x− e 2

al over x+ just produces the usual energy-momentum i=1

conservation
0 factor:
1.2 Worldline representation
� N
of
�
scalar QED
�
effective action
� ��
T � �
p+ N
ki τ i �N εi ·x− �N
∆ij ki ·kj −2i•∆ij εi ·kj −•∆•ij εi ·εj �
× dτi eix− · i=1 T e i=1 T e i,j=1
� .
Similarly, the one-loop effective action can be presented in �terms of a path � ∞integral over
lin(ε ε ···ε ) all
0 i=1
� � � 2
1 2 N

closed loops in spacetime:N −m (4.6) 1

x2−
p; p� ; k1 , ε1 ; · · · ; kAfter, ε ]
N performing
N = (−ie)
also the x∫
(2π) D D
δ p + p �
+ k i d D
x − dT e T − 4T
e (4πT )−
− integral, and some rearrangements, one arrives at
∞ 0
i2
𝑑𝑇 h i
2
� T� Γ[ = Tr exp −𝑇 (−(𝜕 + + )
�N� ∞� ��
𝐴] 𝑖𝑒 𝐴) � 𝑚
� �N
∫, ε0N∞] �
𝑇 k τN �
Γ[p; p� ; k1 , ε1 ; · ix
· · ;−
kN·(p+ = (−ie)
i i(2π)
) ∫D δD εpi+·x−
p� + ki dT
∆ e−Tk
(m +p )
·k −2i •
∆ ε ·k − • •
∆ ε ·ε � 2 2

× dτi e = 𝑑𝑇i T−𝑚2𝑇e i T ei=1 i,j=1

e D𝑥(𝜏) 𝑒
∫ 𝑇 ij i j ij
−0 0 𝑑 𝜏 41 𝑥¤ 2 +𝑖𝑒 𝑥¤ · 𝐴( 𝑥 ( 𝜏))

i
.
j ij i
(2)
j
�
0 � Ti� N � 𝑇 � � |τi −τj |
�� lin(ε1 ε2 ···εN
N0
(−2k ·pτ +2iε ·p)+ ( 𝑥 (0)=𝑥
− (𝑇 ) )k ·k −i(sign(τ −τ )−1)ε ·k +δ(τ −τ )ε ·ε �
N τi +τj
× dτi e i=1 i i i i,j=1 2 2 i j i j i j i j i j
� .
0 lin(ε1 ε2 ···εN )
i=1

erforming alsoExpanding
the x− the field in 𝑁 plane waves, one gets the full scalar QED one-loop 𝑁-photon amplitudes.
integral, and some rearrangements, one arrives at (4.7)

1.3 Worldline representation of spinor QED effective action

At the level of the 𝑁-photon amplitudes, � from scalar to spinor QED
the transition
� can
2 (up to
�� ∞
−T (m2 +p
Γ[p; p� ; k1 , ε1 ; · ·the
· ;normalization)
kN , εN ] = (−ie) N
(2π) D D
δ p + p �
+ k i dT e
simply be made by inserting, under the path integral (2), the “Feynman spin
)

i 0
factor” Spin[𝑥, 𝐴],
� T� N �N �N � |τi −τ" j | τi +τj ��
(−2k ·pτ +2iε ·p)+ ( − ∫i ·k
)k −i(sign(τ −τ
#
)−1)ε ·k +δ(τ −τ )ε ·ε �
dτi e i=1 i i i i,j=1 2 𝑖
𝜇
Spin[𝑥, 𝐴] = trΓ P exp 𝑒 [𝛾 , 𝛾 ]
2
𝜈
𝑇 j i
𝑑𝜏𝐹𝜇𝜈 (𝑥(𝜏)) .
j i j i j
(3)
i j
�
0 i=1 4 lin(ε1 ε2 ···
0

2
r final representation of the N - propagator in momentum space. On-shell it corresponds to mul
scattering, while off-shell it can be used for constructing higher-loop amplitudes by sewing. S
m space version involves the integration variables only linearly in the exponent, for any given or
n legs it is straightforward to do the integrals and verify, that they correspond to the usual sum of
Summing Feynman diagrams in the worldline formalism C. Schubert

However, its use implies path ordering, forcing one to fix the ordering of the photons. For the
purposes that I am going to discuss here, it will be essential to avoid this. Following Fradkin [2],
this can be done replacing the spin factor by a Grassmann path integral,

∫ " ∫ !#
𝑇
1
Spin[𝑥, 𝐴] → D𝜓(𝜏) exp − 𝑑𝜏 𝜓 · 𝜓¤ − 𝑖𝑒𝜓 𝜇 𝐹𝜇𝜈 𝜓 𝜈 (4)
0 2

where the Lorentz vectors 𝜓 𝜇 (𝜏) are anticommuting and antiperiodic functions of proper time:
𝜓(𝜏1 )𝜓(𝜏2 ) = −𝜓(𝜏2 )𝜓(𝜏1 ), 𝜓(𝑇) = −𝜓(0). The main point of the Grassmann approach is to
replace the path-ordered exponential by an ordinary exponential.

1.4 Worldline representation of dressed electron propagator

For the dressed fermion propagator, too, worldline reprentations have been around for decades
[2–5]. However, only during the last few years a version has been developed that seems suitable for
higher-order state-of-the-art calculations [6, 7]. The starting point is the second-order representation
0
of the 𝑥-space Dirac propagator 𝑆 𝑥 𝑥 [ 𝐴] in a Maxwell background:

0 0
𝑚 + 𝑖𝐷/0 𝐾 𝑥 𝑥 [ 𝐴]
 
𝑆 𝑥 𝑥 [ 𝐴] =
0
D h 𝑖 i −1 E
𝐾 𝑥 𝑥 [ 𝐴] = 𝑥 0 𝑚 2 − 𝐷 𝜇 𝐷 𝜇 + 𝑒𝛾 𝜇 𝛾 𝜈 𝐹𝜇𝜈 𝑥
2
∫ ∞ ∫ 𝑞 (𝑇 )=0
1 ( 𝑥−𝑥 0 ) 2 𝑥 0 −𝑥
∫𝑇 1 2


2
= 𝑑𝑇 e−𝑚 𝑇 e− 4 𝑇 𝐷𝑞 e− 0 𝑑 𝜏 4 𝑞¤ +𝑖𝑒 𝑞¤ · 𝐴+𝑖𝑒 𝑇 ·𝐴
0 𝑞 (0)=0
∫ ∫𝑇  
1 ¤𝜇
𝐷𝜓 e− 0 𝑑 𝜏 2 𝜓𝜇 𝜓 −𝑖𝑒𝐹𝜇𝜈 ( 𝜓+𝜂) ( 𝜓+𝜂) .
𝐷 𝜇 𝜈
× 2− 2 symb−1 (5)
𝜓 (0)+𝜓 (𝑇 )=0

Here 𝜂 𝜇 is an external Grassmann Lorentz vector, and the “symbol map” symb converts products
of 𝜂s into fully antisymmetrised products of Dirac matrices:
√
symb 𝛾 𝛼1 𝛼2 ···𝛼𝑛 ≡ (−𝑖 2) 𝑛 𝜂 𝛼1 𝜂 𝛼2 . . . 𝜂 𝛼𝑛


(6)

where 𝛾 𝛼𝛽 ···𝜌 denotes the totally antisymmetrised product of gamma matrices:

1 ∑︁
𝛾 𝛼1 𝛼2 ···𝛼𝑛 ≡ sign(𝜋)𝛾 𝛼 𝜋 (1) 𝛾 𝛼 𝜋 (2) · · · 𝛾 𝛼 𝜋 (𝑛) . (7)
𝑛! 𝜋 ∈𝑆
𝑛

1.5 Higher order QED processes

Since all the above formulas are valid off-shell, arbitrary QED processes can be constructed
from these building blocks by sewing (Fig. 2).
Although the worldline representation is equivalent to Feynman diagrams, it is more global in
the sense that it does not distinguish between diagrams differing only by the ordering of the photon
legs along a line or loop.

3
Summing Feynman diagrams in the worldline formalism C. Schubert

Figure 2: A typical multiloop process in QED.

2. Generalization of the Landau-Khalatnikov-Fradkin transformation

For some purposes, this property becomes useful already at the path-integral level. In 1956,
Landau and Khalatnikov [8] and independently Fradkin [9] showed, that the nonperturbative be-
haviour of the electron propagator 𝑆(𝑥; 𝜉) under a change 𝜉 → 𝜉ˆ of the covariant gauge parameter
𝜉 can be written as (in exactly four dimensions)
" # − 4𝛼𝜋 ( 𝜉 − 𝜉ˆ )
2
ˆ 𝑥
𝑆(𝑥; 𝜉) = 𝑆(𝑥; 𝜉) (8)
𝑥 02
where 𝑥 0 is an IR cutoff. Very recently, we have used the worldline formalism to extend this result
to the general fermionic 2𝑛-point correlator

A (𝑥 1 , . . . , 𝑥 𝑛 ; 𝑥 10 , . . . , 𝑥 𝑛0 |𝜉) ≡ h𝜓(𝑥 1 ) · · · 𝜓(𝑥 𝑛 ) 𝜓(𝑥

¯ 10 ) · · · 𝜓(𝑥
¯ 𝑛0 )i (9)

as follows [10]:
𝑛
Ö ˆ (𝑘,𝑙)
A (𝑥 1 , . . . , 𝑥 𝑛 ; 𝑥 10 , . . . , 𝑥 𝑛0 | 𝜉)
ˆ = e ( 𝜉 − 𝜉 )𝑆 A (𝑥 1 , . . . , 𝑥 𝑛 ; 𝑥 10 , . . . , 𝑥 𝑛0 |𝜉) , (10)
𝑘,𝑙=1
𝑒2 𝐷  n 2
2− 𝐷
2  2− 𝐷2
2  2− 𝐷2
𝑆 (𝑘,𝑙) = 𝑥 𝑘 − 𝑥𝑙0 𝑥 𝑘0 −𝑥𝑙
2
 
Γ −2 𝑥 𝑘 −𝑥𝑙 − −
32𝜋 2
𝐷
2

2  2− 𝐷2 o
+ 𝑥 𝑘0 − 𝑥𝑙0

. (11)

Here the sum 𝑘, 𝑙 runs over all pairs of open fermion lines, and the exponential factor for fixed 𝑘
and 𝑙 implements the effect of the change of the covariant gauge parameter for photons inserted
between them in all possible ways (for photons that on one or both ends hit a closed loop there is
no effect). Note also that the dimension has been left general.

3. String-inspired treatment of the worldline path integral

Under the influence of string theory, in the nineties a perturbative approach to the evaluation
of worldline path integrals using worldline Green’s functions was developed [11, 12]. E.g. for the
closed-loop case one has the basic correlators
1 2
h𝑥 𝜇 (𝜏1 )𝑥 𝜈 (𝜏2 )i = −𝐺 (𝜏1 , 𝜏2 ) 𝛿 𝜇𝜈 , 𝐺 (𝜏1 , 𝜏2 ) = |𝜏1 − 𝜏2 | − 𝜏1 − 𝜏2 , (12)
𝑇
h𝜓 𝜇 (𝜏1 )𝜓 𝜈 (𝜏2 )i = 𝐺 𝐹 (𝜏1 , 𝜏2 ) 𝛿 𝜇𝜈 , 𝐺 𝐹 (𝜏1 , 𝜏2 ) = sign(𝜏1 − 𝜏2 ) . (13)

4
Summing Feynman diagrams in the worldline formalism C. Schubert

This allows one to derive compact master formulas for the photon-dressed propagators and the one-
loop 𝑁-photon amplitudes in scalar and spinor QED, as well as for many other types of amplitudes
(for reviews, see [13, 14]).

3.1 Master formula for the N-photon amplitudes in scalar and spinor QED

The in many ways prototypical one of these master formulas is the one for the one-loop
𝑁-photon amplitudes in scalar QED:

∫ ∞ 𝑁 ∫ 𝑇
𝑁 𝑑𝑇 −𝐷 −𝑚2𝑇
Ö
Γ[{𝑘 𝑖 , 𝜀𝑖 }] = (−𝑖𝑒) (4𝜋𝑇) 2 𝑒 𝑑𝜏𝑖
0 𝑇 𝑖=1 0
𝑁
 ∑︁ 
1 1
𝐺 𝑖 𝑗 𝑘 𝑖 · 𝑘 𝑗 + 𝑖 𝐺¤ 𝑖 𝑗 𝑘 𝑖 · 𝜀 𝑗 + 𝐺¥ 𝑖 𝑗 𝜀𝑖 · 𝜀 𝑗 | lin( 𝜀1 ,..., 𝜀 𝑁 ) .

× exp (14)
𝑖, 𝑗=1
2 2

Here 𝑇 is the loop proper-time and 𝜏𝑖 parametrizes the position of photon 𝑖 along the loop. A
projection on the terms linear in each of the polarization vectors 𝜀 1 , . . . , 𝜀 𝑁 is understood. Besides
the Green’s function 𝐺 𝑖 𝑗 ≡ 𝐺 (𝜏𝑖 , 𝜏 𝑗 ) also its first and second derivatives appear,

(𝜏1 − 𝜏2 )
𝐺¤ (𝜏1 , 𝜏2 ) = sign(𝜏1 − 𝜏2 ) − 2 , (15)
𝑇
2
𝐺¥ (𝜏1 , 𝜏2 ) = 2𝛿(𝜏1 − 𝜏2 ) − . (16)
𝑇
A similar master formula can be written down for the spinor loop using worldline supersymmetry..
However, in practice it is usually preferable to use an integration-by-parts that removes all the 𝐺¥ 𝑖 𝑗 ,
and the following “Bern-Kosower replacement rule” [11]:

𝐺¤ 𝑖1 𝑖2 𝐺¤ 𝑖2 𝑖3 · · · 𝐺¤ 𝑖𝑛 𝑖1 → 𝐺¤ 𝑖1 𝑖2 𝐺¤ 𝑖2 𝑖3 · · · 𝐺¤ 𝑖𝑛 𝑖1 − 𝐺 𝐹 𝑖1 𝑖2 𝐺 𝐹 𝑖2 𝑖3 · · · 𝐺 𝐹 𝑖𝑛 𝑖1 . (17)

3.2 The four-photon amplitudes

Let us write down the resulting representation for the familiar four-photon amplitude, usually
given in terms of the six Feynman diagrams displayed in Fig. 3.

1 2 1 2 1 3 1 3 1 4 1 4

+ + + + +

4 3 3 4 4 2 2 4 3 2 2 3

Figure 3: Six Feynman diagrams for photon-photon scattering.

After a large number of integrations by parts, one finds the following tensor decomposition
[15] (still off-shell):

Γ̂ = Γ̂ (1) + Γ̂ (2) + Γ̂ (3) + Γ̂ (4) + Γ̂ (5) , (18)

5
Summing Feynman diagrams in the worldline formalism C. Schubert

Γ̂ (1) = Γ̂ (1) 𝑇 (1) + Γ̂ (1)

(1234) (1234)
𝑇 (1) + Γ̂ (1)
(1243) (1243)
𝑇 (1) ,
(1324) (1324)

Γ̂ (2) = Γ̂ (2) 𝑇 (2)

(12) (34) (12) (34)
+ Γ̂ (2) 𝑇 (2)
(13) (24) (13) (24)
+ Γ̂ (2) 𝑇 (2)
(14) (23) (14) (23)
,
∑︁ ∑︁ ∑︁ ∑︁
Γ̂ (3) = Γ̂ (3) 𝑇 (3)𝑟4
(123)𝑖 (123)𝑖
+ Γ̂ (3)
𝑇 (3)𝑟1
(234)𝑖 (234)𝑖
+ Γ̂ (3)
𝑇 (3)𝑟2
(341)𝑖 (341)𝑖
+ Γ̂ (3) 𝑇 (3)𝑟3 ,
(412)𝑖 (412)𝑖
𝑖=1,2,3 𝑖=2,3,4 𝑖=3,4,1 𝑖=4,1,2
∑︁ ∑︁
Γ̂ (4)
= Γ̂ (4) 𝑇 (4)
(𝑖 𝑗)𝑖𝑖 (𝑖 𝑗)𝑖𝑖
+ Γ̂ (4) 𝑇 (4)
(𝑖 𝑗) 𝑗 𝑗 (𝑖 𝑗) 𝑗 𝑗
,
𝑖< 𝑗 𝑖< 𝑗
∑︁ ∑︁
Γ̂ (5)
= Γ̂ (5) 𝑇 (5)
(𝑖 𝑗)𝑖 𝑗 (𝑖 𝑗)𝑖 𝑗
+ Γ̂ (5) 𝑇 (5) .
(𝑖 𝑗) 𝑗𝑖 (𝑖 𝑗) 𝑗𝑖
𝑖< 𝑗 𝑖< 𝑗

Remarkably, the basis of five tensors 𝑇 (𝑖) is identical with the one found in 1971 by Costantini, De
Tollis and Pistoni [16] using the QED Ward identity:
(1)
𝑇(1234) ≡ 𝑍 (1234) , (19)
(2)
𝑇(12) (34)
≡ 𝑍 (12)𝑍 (34) , (20)
(3)𝑟4 𝑟 4 · 𝑓4 · 𝑘 𝑖
𝑇(123)𝑖 ≡ 𝑍 (123) (𝑖 = 1, 2, 3) , (21)
𝑟4 · 𝑘 4
(4) 𝑘 1 · 𝑓3 · 𝑓4 · 𝑘 1
𝑇(12)11 ≡ 𝑍 (12) , (22)
𝑘3 · 𝑘4
(5) 𝑘 1 · 𝑓3 · 𝑓4 · 𝑘 2
𝑇(12)12 ≡ 𝑍 (12) . (23)
𝑘3 · 𝑘4
Here we have further introduced
𝜇𝜈 𝜇 𝜇
𝑓𝑖≡ 𝑘 𝑖 𝜀 𝑖𝜈 − 𝜀 𝑖 𝑘 𝑖𝜈 , (24)
1

𝑍 (𝑖 𝑗) ≡ tr 𝑓𝑖 𝑓 𝑗 = 𝜀𝑖 · 𝑘 𝑗 𝜀 𝑗 · 𝑘 𝑖 − 𝜀𝑖 · 𝜀𝑖 𝑘 𝑖 · 𝑘 𝑗 , (25)
2
Ö𝑛 
𝑍 (𝑖1𝑖2 . . . 𝑖 𝑛 ) ≡ tr 𝑓𝑖 𝑗 (𝑛 ≥ 3) . (26)
𝑗=1

The corresponding parameter integrals are extremely compact:

∫ ∞ ∫ 4
1Ö Í4
(𝑘) 𝑑𝑇 4− 𝐷 −𝑚2𝑇 (𝑘) ¤ 𝐺𝑖 𝑗 𝑘𝑖 ·𝑘 𝑗
Γ̂ ··· = 𝑇 2 e 𝑑𝑢 𝑖 𝛾ˆ ... ( 𝐺 𝑖 𝑗 ) e𝑇 𝑖< 𝑗=1 (27)
0 𝑇 0 𝑖=1

where, for spinor QED,

(1)
𝛾ˆ (1234) = 𝐺¤ 12 𝐺¤ 23 𝐺¤ 34 𝐺¤ 41 − 𝐺 𝐹 12 𝐺 𝐹 23 𝐺 𝐹 34 𝐺 𝐹 41 , (28)
(2)
= 𝐺¤ 12 𝐺¤ 21 − 𝐺 𝐹 12 𝐺 𝐹 21 𝐺¤ 34 𝐺¤ 43 − 𝐺 𝐹 34 𝐺 𝐹 43 ,



𝛾ˆ (12) (34)
(29)
(3)
= 𝐺¤ 12 𝐺¤ 23 𝐺¤ 31 − 𝐺 𝐹 12 𝐺 𝐹 23 𝐺 𝐹 31 𝐺¤ 4𝑖 ,


𝛾ˆ (123)𝑖 (30)
(4)
= 𝐺¤ 12 𝐺¤ 21 − 𝐺 𝐹 12 𝐺 𝐹 21 𝐺¤ 13 𝐺¤ 41 ,


𝛾ˆ (12)11 (31)
(5)
= 𝐺¤ 12 𝐺¤ 21 − 𝐺 𝐹 12 𝐺 𝐹 21 𝐺¤ 13 𝐺¤ 42


𝛾ˆ (12)12 (32)

(plus permutations thereof), and the coefficient functions for scalar QED are obtained from these
simply by deleting all the 𝐺 𝐹 𝑖 𝑗 .

6
Summing Feynman diagrams in the worldline formalism C. Schubert

3.3 Master formula for the N-photon dressed scalar propagator

A master formula for the scalar propagator dressed with 𝑁 photons was obtained by Daikouji
et al. in 1996 [17]:
∫ ∞
𝑝 𝑝0 2
𝐷 (𝑘 1 , 𝜀1 ; · · · ; 𝑘 𝑁 , 𝜀 𝑁 ) = (−𝑖𝑒) 𝑁
𝑑𝑇 e−𝑚 𝑇
0
𝑁 ∫ 𝑇
Ö 2 +Í 𝑁 • • •
× 𝑑𝜏𝑖 e−𝑇 𝑏 𝑖, 𝑗=1 [Δ𝑖 𝑗 𝑘𝑖 ·𝑘 𝑗 −2𝑖 Δ𝑖 𝑗 𝜀𝑖 ·𝑘 𝑗 − Δ𝑖 𝑗 𝜀𝑖 · 𝜀 𝑗 ] .
0 𝜀1 𝜀2 ··· 𝜀 𝑁
𝑖=1
(33)
1
Here we have introduced the vector 𝑏 ≡ 𝑝 0 +
Í𝑁
𝑇 𝑖=1 (𝑘 𝑖 𝜏𝑖 − 𝑖𝜀𝑖 ) and a different worldline Green’s
function Δ(𝜏, 𝜏 0),
|𝜏 − 𝜏 0 | 𝜏 + 𝜏 0 𝜏𝜏 0
h𝑞 𝜇 (𝜏)𝑞 𝜈 (𝜏 0)i = −2Δ(𝜏, 𝜏 0)𝛿 𝜇𝜈 , Δ(𝜏, 𝜏 0) = − + . (34)
2 2 𝑇
3.4 Master formula for the photon-dressed electron propagator
To the contrary, a master formula for the photon-dressed electron propagator was obtained only
very recently [6, 7]:
∫ ∞ ∫ 𝑇 ∫
𝑝 𝑝0 𝑁 −1 −𝑚2𝑇
𝐾 𝑁 (𝑘 1 , 𝜀1 ; . . . ; 𝑘 𝑁 , 𝜀 𝑁 ) = (−𝑖𝑒) symb 𝑑𝑇 e 𝑑𝜏1 · · · 𝑑𝜃 𝑁
0 0
√ Í𝑁 Í 𝑁 +1  
× e− 2𝜂 · 𝑖=1 ( 𝜀𝑖 +𝑖 𝜃 𝑘𝑖 )+ 𝑖, 𝑗=0 b𝑔𝑖 𝑗 𝑘𝑖 ·𝑘 𝑗 +2𝑖𝐷𝑖b𝑔𝑖 𝑗 𝜀𝑖 ·𝑘 𝑗 +𝐷𝑖 𝐷 𝑗b𝑔𝑖 𝑗 𝜀𝑖 · 𝜀 𝑗 . (35)
𝜀1 ··· 𝜀 𝑁

𝜕 𝜕
This now involves Grassmann variables 𝜃 1 , . . . , 𝜃 𝑁 , the super derivative 𝐷 = 𝜕𝜃 −𝜃 𝜕𝜏 and the super
0 0 1 0 0 0
worldline Green’s functionb 𝑔 (𝜏, 𝜃; 𝜏 , 𝜃 ) = 2 (|𝜏 − 𝜏 | + 𝜃𝜃 sign(𝜏 − 𝜏 )). The main advantages of
this approach are that (i) the use of the symbol map leads to an early projection on the Clifford basis,
effectively avoiding the appearance of long Dirac traces (ii) the spin-averaging can be done without
fixing the number or helicity assignments of the photons. As a check, in [7] the formalism was
used to recalculate the (polarized and unpolarized) Compton scattering amplitude, and complete
agreement was found with [18].

3.5 On to multiloop
Dealing with the amplitude as a whole becomes important when one uses the one-loop ampli-
tudes to construct higher-loop amplitudes by sewing. For example, from the one-loop six-photon
amplitude we can construct the three-loop quenched propagator (Fig. 4) etc.

+ + + ...

Figure 4: The quenched three-loop photon propagator.

This type of sums of diagrams is known to suffer from extensive cancellations.

7
Summing Feynman diagrams in the worldline formalism C. Schubert

4. The fundamental problem of worldline integration

Returning to the one-loop level, considering that 𝐺 𝐹 𝑖 𝑗 s can always be eliminated by

𝐺 𝐹 𝑖 𝑗 𝐺 𝐹 𝑗 𝑘 𝐺 𝐹 𝑘𝑖 = −( 𝐺¤ 𝑖 𝑗 + 𝐺¤ 𝑗 𝑘 + 𝐺¤ 𝑘𝑖 ) (36)

the most general integral that one will ever have to compute in the worldline approach to QED (or
any abelian theory) is of the form
∫ 1 Í𝑁
𝐺𝑖 𝑗 𝑘𝑖 ·𝑘 𝑗
𝑑𝑢 1 𝑑𝑢 2 · · · 𝑑𝑢 𝑁 Pol( 𝐺¤ 𝑖 𝑗 ) e 𝑖< 𝑗=1 (37)
0

with arbitrary 𝑁 and polynomial Pol( 𝐺¤ 𝑖 𝑗 ). Here we have rescaled 𝜏𝑖 = 𝑇𝑢 𝑖 , so that now

𝐺 𝑖 𝑗 = |𝑢 𝑖 − 𝑢 𝑗 | − (𝑢 𝑖 − 𝑢 𝑗 ) 2 , 𝐺¤ 𝑖 𝑗 = sign(𝑢 𝑖 − 𝑢 𝑗 ) − 2(𝑢 𝑖 − 𝑢 𝑗 ) . (38)

Ideally, we would like to compute this integral without decomposing the integrand into ordered
sectors. This leads to a very non-standard integration problem. An easily solvable special case are
“cycle integrals”:
(
1
−2𝑛 𝐵𝑛!𝑛
∫
𝑛 even
𝑏𝑛 ≡ 𝑑𝑢 1 𝑑𝑢 2 . . . 𝑑𝑢 𝑛 𝐺¤ 12 𝐺¤ 23 · · · 𝐺¤ 𝑛1 = (39)
0 0 𝑛 odd

where 𝐵𝑛 denotes the 𝑛th Bernoulli number, and “super cycle integrals”,
∫ 1  
𝑏 𝑛 − 𝑓𝑛 ≡ 𝑑𝑢 1 𝑑𝑢 2 . . . 𝑑𝑢 𝑛 𝐺¤ 𝐵12 𝐺¤ 𝐵23 · · · 𝐺¤ 𝐵𝑛1 − 𝐺 𝐹 12 𝐺 𝐹 23 · · · 𝐺 𝐹 𝑛1 = (2 − 2𝑛 ) 𝑏 𝑛 .
0
(40)

In the worldline formalism, those are all that is needed to calculate the one-loop 𝑁-photon amplitudes
in the low-energy approximation [19].
General polynomial integrals can be done recursively by the application of the following
formula [20]
∫ 1 𝑛 𝑘𝑗   𝑘𝑖  
1 ∑︁ Ö ∑︁ 𝑘 𝑗 ¤ 𝑘 𝑗 −𝑙 𝑗 ∑︁ 𝑘𝑖
𝑑𝑢 𝐺¤ (𝑢, 𝑢 1 ) 𝑘1 𝐺¤ (𝑢, 𝑢 2 ) 𝑘2 · · · 𝐺¤ (𝑢, 𝑢 𝑛 ) 𝑘𝑛 = 𝐺𝑖 𝑗
0 2𝑛 𝑖=1 𝑗≠𝑖 𝑙 =0 𝑙 𝑗 𝑙
𝑙𝑖 =0 𝑖
𝑗
Í𝑛
𝑗=1 𝑙 𝑗
 ∑︁  1+Í𝑛𝑗=1 𝑙 𝑗  1+Í𝑛𝑗=1 𝑙 𝑗 
(−1) ∑︁
× Í𝑛 𝐺¤ 𝑖 𝑗 + 1 − (−1) 𝑘𝑖 −𝑙𝑖 𝐺¤ 𝑖 𝑗 − 1 .
𝑗=1 𝑙 𝑗
Í𝑛
(1 + 𝑗=1 𝑙 𝑗 )𝑛 𝑗≠𝑖 𝑗≠𝑖
(41)

This can be used, for example, to integrate out a low-energy photon in a multiphoton amplitude
without fixing the order of the remaining photons. In a forthcoming publication [21] this is applied
to the four-photon amplitude, creating a building block for higher-loop calculations that will make
it possible, for example, to unify the calculation of the various 3-loop g-2 contributions shown in
Fig. 5 and the calculation of the 4-loop 𝛽-function contributions shown in Fig. 6.

8
Summing Feynman diagrams in the worldline formalism C. Schubert

Figure 5: LBL contributions to the three-loop 𝑔 − 2.

Figure 6: LBL contributions to the four-loop 𝛽 function. The external photons are at low-energy.

5. Inverse derivative expansion

Much more difficult is the inclusion of the universal factors e𝐺𝑖 𝑗 𝑘𝑖 ·𝑘 𝑗 . The following strategy
was proposed in [20]: the worldline path integral is performed in the Hilbert space 𝐻 𝑃0 of periodic
functions orthogonal to the constant functions (because the zero mode must be fixed). In this space
the ordinary 𝑛th derivative 𝜕𝑃 is invertible, and the integral kernel of the inverse is given essentially
by the 𝑛th Bernoulli polynomial 𝐵𝑛 (𝑥) [22, 23]:

1
< 𝑢 1 | 𝜕𝑃−𝑛 | 𝑢 𝑛+1 > = − 𝐵𝑛 (|𝑢 1 − 𝑢 𝑛+1 |)sign𝑛 (𝑢 1 − 𝑢 𝑛+1 ) (𝑛 ≥ 1) , (42)
𝑛!
h𝑢 𝑖 |𝜕 0 |𝑢 𝑗 i = 𝛿(𝑢 𝑖 − 𝑢 𝑗 ) − 1 . (43)

Thus worldline integration naturally relates to the theory of Bernoulli numbers and polynomials.
Further, in [20] it was shown how to expand the universal exponential factor in terms of the matrix
elements of the inverse derivatives:

∞
√︁ !
𝐺𝑖 𝑗 𝑘𝑖 ·𝑘 𝑗
∑︁ 𝑘𝑖 · 𝑘 𝑗
e =1+2 (𝑘 𝑖 · 𝑘 𝑗 ) 𝑛−1/2
𝐻2𝑛−1 (h𝑢 𝑖 |𝜕 −2𝑛 |𝑢 𝑗 i − h𝑢 𝑖 |𝜕 −2𝑛 |𝑢 𝑖 i) (44)
𝑛=1
2

where the 𝐻𝑛 (𝑥) are Hermite polynomials. Let us show how to use this expansion for the simplest
case of the 𝑁-point function in 𝜙3 theory, where the master formula simply becomes (𝑘 𝑖 𝑗 ≡ 𝑘 𝑖 · 𝑘 𝑗 )

∞ 1  𝑁
∫ ∫  ∑︁ 
𝑑𝑇 𝑁 −𝐷/2 −𝑚2𝑇 
𝐼ˆ𝑁 (𝑘 1 , . . . , 𝑘 𝑁 ) = 𝑇 𝑒 𝑑𝑢 1 ...𝑑𝑢 𝑁 exp 𝑇
 𝐺 𝑖 𝑗 𝑘 𝑖 𝑗  . (45)
0 𝑇 0  𝑖< 𝑗=1 
 

9
Summing Feynman diagrams in the worldline formalism C. Schubert

In the three-point case one can use this to write ( 𝐵ˆ 𝑛 ≡ 𝐵𝑛!𝑛 )

( ∞ √  )
∑︁ 1 𝑘 12
𝑒 𝑘12 𝐺12 +𝑘13 𝐺13 +𝑘23 𝐺23 = h𝑢 1 |𝜕 −2𝑖 |𝑢 2 i + 𝐵ˆ 2𝑖
𝑖−  
1+2 𝑘 12 2 𝐻2𝑖−1
𝑖=1
2
∞ √ 
𝑗− 21
 

 ∑︁ 𝑘 13  

× 1+2 𝑘 13 𝐻2 𝑗−1 h𝑢 1 |𝜕 −2 𝑗 |𝑢 3 i + 𝐵ˆ 2 𝑗

𝑗=1
2 
 
( ∞
∑︁ 𝑘− 1  √  )
𝑘 23
h𝑢 2 |𝜕 −2𝑘 |𝑢 3 i + 𝐵ˆ 2𝑘
 
× 1+2 𝑘 23 2 𝐻2𝑘−1 .
𝑘=1
2
(46)
∫1
Since 0
𝑑𝑢 𝑖, 𝑗 h𝑢 𝑖 |𝜕 −2𝑖 |𝑢 𝑗 i = 0, the three h𝑢 𝑖 |𝜕 −2𝑛 |𝑢 𝑗 i can produce a non-vanishing integral only
∫1
together, and then by 0 𝑑𝑢|𝑢ih𝑢| = 1l,
∫
h𝑢 1 |𝜕 −2𝑖 |𝑢 2 ih𝑢 2 |𝜕 −2𝑘 |𝑢 3 ih𝑢 3 |𝜕 −2 𝑗 |𝑢 1 i = Tr(𝜕 −2(𝑖+ 𝑗+𝑘) ) = − 𝐵ˆ 2(𝑖+ 𝑗+𝑘) . (47)
123

In this way we get a closed form-expression for the 𝑁 = 3 momentum expansion coefficients,
∫ 𝑎
∑︁ 𝑏
∑︁ 𝑐
∑︁
𝑎 𝑏 𝑐
𝐼3 (𝑎, 𝑏, 𝑐) ≡ 𝐺 12 𝐺 13 𝐺 23 = 𝑎!𝑏!𝑐! ℎ𝑖𝑎 ℎ 𝑏𝑗 ℎ 𝑐𝑘
123 𝑖= b1+𝑎/2c 𝑗= b1+𝑏/2c 𝑘= b1+𝑐/2c

× 𝐵ˆ 2𝑖 𝐵ˆ 2 𝑗 𝐵ˆ 2𝑘 − 𝐵ˆ 2(𝑖+ 𝑗+𝑘) .


(48)

Here we have assumed that 𝑎, 𝑏, 𝑐 are all different from zero, and the coefficients ℎ𝑖𝑎 are (from the
explicit formula for the Hermite polynomials)
2(2𝑖 − 1)!
ℎ𝑖𝑎 = (−1) 𝑎+1 . (49)
(2𝑖 − 𝑎 − 1)!(2𝑎 − 2𝑖 + 1)!
At the four-point level, we encounter more general integrals involving the “cubic worldline vertex”
∫ 1
𝑑𝑢h𝑢|𝜕 −𝑖 |𝑢 1 ih𝑢|𝜕 − 𝑗 |𝑢 2 ih𝑢|𝜕 −𝑘 |𝑢 3 i
𝑖 𝑗𝑘
𝑉3 ≡ (50)
0

but they can be reduced to chain integrals by a systematic IBP procedure. This remains true at
higher points.

6. Summary and Outlook

1. In the worldline formalism, we can integrate out photons in the low-energy limit, or to any
finite order in the external momentum, without fixing the ordering of the remaining legs.

2. At full momentum, we can use the inverse derivative expansion, and try a resummation. For
this purpose, eventually we will need formulas relating the Bernoulli numbers to hypergeo-
metric functions.

3. This provides also a powerful new approach to the calculation of the 𝜙3 and QED heat-kernel
expansions.

10
Summing Feynman diagrams in the worldline formalism C. Schubert

References

[1] R. P. Feynman, Phys. Rev. 80 (1950) 440.

[2] E. S. Fradkin, Nucl. Phys. 76 (1966) 588.

[3] R. P. Feynman, Phys. Rev. 84 (1951) 108.

[4] E.S. Fradkin, D.M. Gitman, Phys. Rev. D 44 (1991) 3230.

[5] M. Reuter, M. G. Schmidt and C. Schubert, Ann. Phys. (N.Y.) 259 313 (1997), hep-th/9610191.

[6] N. Ahmadiniaz, V.M. Banda Guzmán, F. Bastianelli, O. Corradini, J.P. Edwards and C.
Schubert, JHEP 2008 (2020) 049, arXiv:2004.01391 [hep-th].

[7] N. Ahmadiniaz, V.M. Banda Guzmán, F. Bastianelli, O. Corradini, J.P. Edwards and C.
Schubert, JHEP 01 (2022) 050, arXiv:2107.00199 [hep-th].

[8] L. Landau, K. I. Khalatnikov., Sov. Phys. JETP 2, 69 (1956).

[9] E. S. Fradkin, Zh. Eksp. Teor. Fiz. 29, 258 (1955).

[10] N. Ahmadiniaz, J. P. Edwards, J. Nicasio and C. Schubert, Phys. Rev. D 104 (2021) 2, 025014,
e-Print: 2012.10536 [hep-th].

[11] Z. Bern and D.A. Kosower, Nucl. Phys. B 379 (1992) 451.

[12] M. J. Strassler, Nucl. Phys. B 385 (1992) 145, hep-ph/9205205.

[13] C. Schubert, Phys. Rept. 355, 73 (2001), arXiv:hep-th/0101036.

[14] J. P. Edwards and C. Schubert, arXiv:1912.10004 [hep-th].

[15] N. Ahmadiniaz, C. Lopez-Arcos, M. A. Lopez-Lopez and C. Schubert, arXiv:2012.11791

[hep-th].

[16] V. Costantini, B. de Tollis and G. Pistoni, Nuov. Cim. A2 (1971) no. 3, 733.

[17] K. Daikouji, M. Shino, Y. Sumino, Phys. Rev. D 53 (1996) 4598, hep-ph/9508377.

[18] A. Denner and S. Dittmaier, Nucl. Phys. B 540 (1999) 58, hep-ph/9805443.

[19] G. V. Dunne and C. Schubert, JHEP 0208, 053 (2002), hep-th/0205004.

[20] J. P. Edwards, C. M. Mata, U. Müller and C. Schubert, SIGMA 17 (2021), 065.

[21] N. Ahmadiniaz, C. Lopez-Arcos, M. A. Lopez-Lopez and C. Schubert, “The QED four –

photon amplitudes off-shell: part 3”, in preparation.

[22] M.G. Schmidt and C. Schubert, Phys. Lett. B 318, 438 (1993), hep-th/9309055.

[23] M.G. Schmidt and C. Schubert, Phys. Rev. D 53, 2150 (1996), hep-th/9410100.

11