&

'
.
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India

Evaluation of Feynman Integrals:

Advanced Methods

Getting Ready for Physics at the LHC

Workshop organized by RECAPP at HRI
16–20 Feb. 2009, Allhabad, India

Tord Riemann, DESY, Zeuthen

exercises part: Janusz Gluza, Katowice
1
%

$
&

'
The slides of lecture
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India

and
the files for the exercises
=>

http://www-zeuthen.desy.de/˜riemann/Talks/RECAPP-2009

Thanks to

T. Diakonidis

J. Fleischer

K. Kajda
2

B. Tausk

V. Yundin
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India

• Introduction + motivation Part 1: 1–36

definitions, general formula for L-loop integrals, Feynman parameter integration
Infrared and ultraviolet divergencies
Few simple Examples

• Mellin-Barnes representations and their evaluation Part 2: 37–63

Numerical and analytical approaches
Expansions in small parameters, e.g. m2 /s 63-65

• Expressing Integrals by other integrals (and numerical methods) Part 3: 65–106

Differential equations 65–77
Integration by parts 83–84
Sector decomposition 85–96
Tensor reduction 97–110
Summary 112

• Other methods: skipped

probably skipped: difference equations → Laporta, ”High-precision calculation of multi-loop
Feynman integrals by difference equations” see:
3

[Laporta:2001dd]
Not covered: completely numerical approaches a la Passarino et al. in recent years
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India

Introductory remarks
For many problems of the past, a relatively simple approach to the evaluation of
Feynman integrals was sufficient.
At most 1-loop (massless: 2-loop), typically 2 −→ 2 scattering (plus bremsstrahlung)

1 ζ2 ζ2 ζ3 2
T1l1m = ǫ + 1 + (1 + 2 )ǫ + (1 + 2 − 3 )ǫ +
2y ln(y)
T1l1m SE2l2m SE2l0m B4l2m = [− 1ǫ + ln(−s)] s(1−y 2 ) + c1 ǫ + · · ·

with d = 4 − 2ǫ and m = 1 and

√
1−4/t−1
y= √
1−4/t+1
V3l1m B4l2m

Figure shows so-called master integrals.

Feynman parameters may be used and by direct integration over them one gets things
2
23
like: 57 , ln st , ln st ln ms , Li2 ( st ) etc.
With more complexity of the reaction (more legs) and more perturbative accuracy
4

(more loops), this approach appears to be not sufficiently sophisticated.

%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India

More loops

V6l4m1 V6l4m1d V6l4m2

Two-loop vertex integrals with six internal lines

massless case: only fixed numbers and one scale factor

SE3l2M1m SE3l2M1md V4l2M2m

V4l2M2md V4l2M1m V4l2M1md

5

B5l2M2md B5l2M2m

Integrals with two different mass scales m and M

%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India

B1 B2 B3

B4 B5 B6

Two-loop box diagrams for massive 2 → 2 scattering

K1

K3

K2 K4
6

K5

A box master integral B5l2m2, related to B2 = B7l2m2 by shrinking two lines

%

$
&

'
More legs
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India

Massive pentagon: 5 kinematic variables + several masses

Massless and massive hexagons: 8 kinematic variables + several masses

Variables for 2 → 2 scattering, i.e. box diagrams: s, t or s and cos θ
7

Variables for 2 → 3 scattering: 5 = 2 + 3 (three additional momenta of a particle)

Variables for 2 → 4 scattering: 8 = 5 + 3 (another three additional)
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India

What are ”Advanced Methods?”

• High degree of automatization

• Publicly available – or not . . .

• Go beyond the methods a la Passarino-Veltman and the relatively simple polylogs

• Not yet completely worked out

• Allow the solution of ”Advanced Problems”

8
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India

L-loop n-point Feynman Integrals of tensor rank R with N

internal lines
• Internal loop momenta are kl , l = 1 · · · L

• Propagators have mass mi and momentum qi , i = 1 · · · N and indices νi – see G(X)

• External legs have momentum pe , e = 1 · · · n, with p2e = Me2


X L n
X
The N propagators are: Di = qi2 − m2i = [ cli kl + dei pe ]2 − m2i
l=1 e=1


Feynman integrals have the following general form:
eǫγE L
Z d
d k1 . . . dd kL X(kl1 , · · · , klR )
G(X) = .

(iπ d/2 )L D1ν1 . . . Diνi . . . DN
νN

The numerator X may contain a tensor structure (see later for more on that):

X(kl1 , · · · , klR ) = (kl1 Pe1 ) · · · (klR PeR ) = (Peα11 · · · PeαRR ) (klα11 · · · klαRR )
9
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 10

Tensor integrals
Tensor integrals appear naturally in Feynman diagrams, due to

• fermion propagators

• non-abelian triple-boson vertices

• boson propagators in Rξ gauges and unitary gauge

Example: Fermionic vacuum polarization

 
1 [γk + m ] [γ(k + p ) + m ]
Z
1 β 1 2 α
Παβ ∼ dd kTr γ γ
(iπ d/2 ) D1 D2
1 dd k h
Z i
2 αβ α β α β α β
∼ (m1 m2 − k − kp1 )g + 2k k + k p1 + p1 k
(iπ d/2 ) D1 D2

So, one needs also efficient ways to evaluate tensor integrals – see later
%

$
&

'
Simple examples of scalar integrals
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 11

1 dd k d4 k
Z
A0 = → UV − divergent : ∼ 2
(iπ d/2 ) D1 k

m1

1 dd k d4 k
Z
p p
B0 = → UV − divergent ∼ 4
m2 (iπ d/2 ) D1 D2 k
p1

m1
p3
1 dd k d4 k
Z
m2
C0 = → UV − finite ∼ 6
m3 (iπ d/2 ) D1 D2 D3 k
p2

Dependent on conventions, where k starts to run in the loop, it is:

D1 = k 2 − m21
D2 = (k + p1 )2 − m22
D3 = (k + p1 + p2 )2 − m23
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 12

Evaluate Feynman integrals

There are two strategies to solve a Feynman integral:

• Reduction
Express the integral with the aid of recurrence relations by other, known integrals.
These are then the Master Integrals.
• Direct evaluation
%

$
&

'
Introduce Feynman parameters
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 13

1 Γ(ν1 + . . . + νN ) 1 1 νN −1
x1ν1 −1 . . . xN δ(1 − x1 . . . − xN )
Z Z
= dx1 . . . dxN ,
D1ν1 D2ν2 . . . DN
νN
Γ(ν1 ) . . . Γ(νN ) 0 0 (x1 D1 + . . . + xN DN )Nν
with Nν = ν1 + . . . νN .
The denominator of G contains, after introduction of Feynman parameters xi , the momentum
dependent function m2 with index-exponent Nν :

(m2 )−(ν1 +...+νN ) = (x1 D1 + . . . + xN DN )−Nν = (ki Mij kj − 2Qj kj + J )−Nν

Here M is an (LxL)-matrix, Q = Q(xi , pe ) an L-vector and J = J (xi xj , m2i , pej pel ).

M, Q, J are linear in xi . The momentum integration is now simple:
Shift the momenta k such that m2 has no linear term in k̄:

k = k̄ + (M −1 )Q,
m2 = k̄M k̄ − QM −1 Q + J.

Remember: M1−loop = 1, in general:

1
M −1 = M̃ ,
(det M )
where M̃ is the transposed matrix to M . The shift leaves the integral unchanged.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 14

The shift leaves the integral unchanged (rename k̄ → k):

Dk1 . . . DkL
Z
G(1) = Nν
.
(kM k + J − QM Q) −1

Go Euclidean: Rotate now the k 0 → iKE 0

with k 2 → −kE
2
(and again rename k E → k):
E E
Dk . . . Dk Dk1 . . . DkL
Z Z
1
G(1) → (i)L L
Nν
= (−1) Nν
(i) L
Nν
.
E E
(−k M k + J − QM Q) −1 [kM k − (J − QM Q)]−1

Call

µ2 (x) = −(J − QM −1 Q)

and get
Dk1 . . . DkL
Z
G(1) = (−1)Nν (i)L N
.
(kM k + µ2 ) ν
For 1-loop integrals it is L = 1, M = 1 - and we will use nearly only those - we are ready to do the
k-integration.
%

$
&

'
Additional step for L-loop integrals
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 15

For L-loops go on and now diagonalize the matrix M by a rotation:

k → k ′ (x) = V (x) k,
kM k = k ′ Mdiag k ′
X
→ αi (x)ki2 (x),
Mdiag (x) = (V −1 )+ M V −1 = (α1 , . . . , αL ).

This leaves both the integration measure and the integral invariant:
Dk1 . . . DkL
Z
G(1) = (−1)Nν (i)L Nν
.
2 2
P
( i αi k i + µ )
Rescale now the ki ,
√
k̄i = αi k i ,

with

d d ki = (αi )−d/2 dd k̄i ,

L
Y
αi = det M,
i=1
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 16

and get the Euclidean integral to be calculated (and rename k̄ → k):

Dk1 . . . DkL
Z
G(1) = (−1)Nν (i)L (det M )−d/2 Nν
.
2 2 2
(k1 + . . . + kL + µ )

Use now (remembering that Dk = dk/(iπ d/2 )):

d


Dk1 . . . DkL Γ N − L 1
Z
ν
iL Nν
= 2
Nν −dL/2
,
2 2 2
(k1 + . . . + kL + µ ) Γ (N ν ) 2
(µ )
d


2
Dk1 . . . DkL k1 d Γ Nν − 2 L − 1 1
Z
iL Nν
= Nν −dL/2−1
.
2 2 2
(k1 + . . . + kL + µ ) 2 Γ (N ν ) 2
(µ )
These formulae follow for L = 1 immediately from any textbook.

See ’Mathematical Interlude’.

For L > 1, get it iteratively, with setting (k12 + k22 + m2 )N = (k12 + M 2 )N , M 2 = k22 + m2 , etc.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 17

Mathematical interlude: d-dimensional integrals (I)

After the Wick rotation, the integrand of the momentum integration is positive definite.
Further it is independent of the angular variables.
The integral is understood as symmetric limit the infinity boundaries.
Z
dd k kµ F (k 2 ) = 0
Z Z
dd k F (k + C) = dd k F (k).

Introduce d-dim. spherical coordinates. The vector k has d components:

kd = r cos θd ≡ ρd cos θd
kd−1 = ρd−1 cos θd−1
...
k3 = ρ3 cos θ3
k2 = ρ2 sin φ
k1 = ρ2 cos φ
ρd−1 = ρd sin θd
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 18

Mathematical interlude: d-dimensional integrals (II)

The above is the direct generalization of the 3- or 4-dimensional phase space parametrization.
With these variables, the integral over the complete d-dimensional phase space gets the
following form:
Z ∞ Z R Z π
dd k F (k) = lim drr d−1 dθd−1 sind−2 θd−1
−∞ R→∞ 0 0
Z π Z 2π
d−3
dθd−2 sin θd−2 . . . dθ1 F (k)
0 0

The integrations met in the loop calculations may be performed using the following two integrals:
Z π 1 
√ Γ (m + 1)
dθ sinm θ = π  12 ,
0 Γ 2 (m + 2)
   
β+1 β+1
1Γ 2 Γ α− 2
Z ∞
rβ 1
dr 2 2 )α
= α−(β+1)/2
.
0 (r + M 2 Γ (α) 2
(M )
In general, the angular integrations are influenced by the integrand too. (Remember phase space
integrals of bremsstrahlung!)
%

$
&

'
Mathematical interlude: d-dimensional integrals (III)
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 19

If F (k) → F (r), r = |k|, the angular integrations yield the surface of the d-dimensional sphere
with radius r:
2π d/2 d−1
ωd (r) =  r .
Γ d2

The remaining integration, over r, yields for F (r) = 1 the volume of the sphere with radius R:

π d/2
Vd (R) =  d
 Rd ,
Γ 1+ 2

1
Z
G(1) = dd k N
.
(k 2
+ M 2) ν
∞
ωd (r)
Z
= dr 2
0 (r + M 2 )Nν

and we get immediately, with M 2 ≡ M 2 (x1 , x2 , . . .):

" #
d/2
iπ Γ (Nν − d/2) 1
G(1) = Nν −d/2
.
Γ(Nν ) 2
(M )
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 20

The Γ-function
The Γ-function may be defined by a difference equation:

zΓ(z) − Γ(z + 1) = 0

Z ∞
Γ(z) = tz−1 e−t dt
0

Series[Gamma[ep], {ep, 0, 2}] =

1 1 1
Γ[ǫ] = − γE + (6γE 2 + π 2 )ǫ + (−2γE 3 − γE 2 π + 2Ψ(2, 1))ǫ2 + · · ·
ǫ 12 12

Ψ(2, 1)) = P olyGamma(2, 1) = −2ζ3

exp(ep EulerGamma)Series[Gamma[ep], {ep, 0, 2}] =
1 1 1
eǫγE Γ[ǫ] = + (π 2 )ǫ + (Ψ(2, 1))ǫ2 + · · ·
ǫ 12 6
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 21

Look at the singularities in the complex plane (figure shows real part of Γ):
When applying the Cauchy-theorem, one may close the integral to the left or to the right . . .
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 22

Mathematical interlude: d-dimensional integrals (IV)

We will often use:

aǫ = eǫ ln a = 1 + ǫ ln a + . . .
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 23

Finally, one gets for Scalar integrals:

N N
!
d 1Y

Z
Nν Γ Nν − 2L ν −1
X (det M )−d/2
G(1) = (−1) dxj xj j δ 1− xi N −dL/2
,
Γ(ν1 ) . . . Γ(νN ) 0 j=1 i=1 (µ2 ) ν
or
N N
!
d 1Y

Z
Nν Γ Nν − 2L ν −1
X U (x)Nν −d(L+1)/2
G(1) = (−1) dxj xj j δ 1− xi
Γ(ν1 ) . . . Γ(νN ) 0 j=1 i=1
F (x)Nν −dL/2

with

U (x) = (det M ) (→ 1 for L = 1)

F (x) = (det M ) µ2 = − (det M ) J + Q M̃ Q (→ −J + Q2 for L = 1)

Trick for one-loop functions:

P
U = det M = 1 = xi and so U ‘disaapears’ and the construct F1 (x) is bilinear in xi xj :

2
P P
F1 (x) = −J ( xi ) + Q = Aij xi xj .
%

$
&

'
The vector integral differs by some numerator ki pe and thus there is a single shift in the integrand
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 24

k → k̄ + U (x)−1 M̃ Q
the d k̄ k̄/(k̄ 2 + µ2 ) → 0, and no further changes:
R d

N N
! " #
d
` ´ Z 1 Y Nν −d(L+1)/2−1
Γ Nν − L ν −1
X U (x) X
G(k1α ) = (−1)Nν 2
dxj xj j δ 1− xi Nν −dL/2
M̃1l Ql ,
Γ(ν1 ) . . . Γ(νN ) 0 j=1 i=1 F (x) l α

Here also a tensor integral:

N N
!
d
U (x)Nν −2−d(L+1)/2
` ´ Z
Γ Nν − 1 Y
Nν 2L ν −1
dxj xj j δ
X
G(k1α k2β ) = (−1) 1− xi
Γ(ν1 ) . . . Γ(νN ) 0 j=1 i=1 F (x)Nν −dL/2
" −1 + −1
#
Γ Nν − d
` ´
X
2 L − 1 g αβ (V ) (V )
× [M̃1l Ql ]α [M̃2l Ql ]β − U (x)F (x) 1l 2l
.
Γ Nν − d
` ´
l 2 L 2 α l

The 1-loop case will be used in the following L times for a sequential treatment of an L-loop
xj Dj = k 2 − 2Qk + J and F (x) = Q2 − J ):
P
integral (remember
d

Z 1 N N
!
Γ Nν − 2 Y ν −1
X [1, Qpe ]
G([1, kpe ]) = (−1)Nν dxj xj j δ 1 − xi
Γ(ν1 ) . . . Γ(νN ) 0 j=1 i=1
F (x)Nν −d/2
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 25

Examples for one-loop F -polynomials

One-loop vertex:

F (t, m2 ) = m2 (x1 + x2 )2 + [−t]x1 x2

one-loop box:

F (s, t, m2 ) = m2 (x1 + x2 )2 + [−t]x1 x2 + [−s]x3 x4

one-loop pentagon:

F (s, t, t′ , v1 , v2 , m2 ) = m2 (x1 + x3 + x4 )2 + [−t]x1 x3 + [−t′ ]x1 x4 + [−s]x2 x5 + [−v1 ]x3 x5 + [−v2 ]x2 x4

2-loop example: B7l4m2, has a box-type sub-loop with 2 off-shell legs:

n
−(a4567 −d/2)
F = [−t]x4 x7 + [−s]x5 x6 + m2 (x5 + x6 )2
o−(a4567 −d/2)
2
+(m − Q21 )x7 (x4 2
+ 2x5 + x6 ) + (m − Q22 )x7 x5

2-loop: B5l2m2, sub-loop with 2 off-shell legs (diagram see p.4):

F2lines (k12 , m2 ) = m2 (x3 )2 + [−k12 + m2 ]x1 x3

%

$
&

'
The Tadpole A0(m)
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 26

eǫγE dd k
Z
T 1l1m[a] = A0 = → UV − divergent
(iπ d/2 ) (k 2 − m2 )a
With our general formulae we get, in the 1-dimensional Feynman parameter integral,
for the numerator

N = (k 2 − m2 )x1 ≡ k 2 + J
F = m2 x1 ≡ m2 x21

and thus
1
Γ[a − d/2] 1
Z
T 1l1m[a] = (−1)a eǫγE dxxa−1 δ[1 − x]
Γ[a] 0 F a−d/2
Γ[a − 2 + ǫ]
= (−1)a eǫγE (m2 )2−a−ǫ
Γ[a]
→ −eǫγE Γ[−1 + ǫ] for a = 1, m = 1
   
1 ζ2 ζ2 ζ3 2
= +1+ 1+ ǫ+ 1+ − ǫ +···
ǫ 2 2 3
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 27

The Self-energy B0 (s, m1 , m2 )

m1

ǫγE
√
4−d e dd k
Z
p p SE2l = B0 [s, m1 , m2 ] = (2 πµ)
(iπ d/2 ) [k 2 − m2 ][(k + p)2 − m22 ]
m2

The SE2l is UV-divergent and the corresponding F -function is:

F [s, m1 , m2 ] = m21 x21 + m22 x22 − [s − m21 − m22 ]x1 x2

and for special cases:

F [s, m1 , 0] = m21 x21 − [s − m21 ]x1 x2

F [s, m1 , m1 ] = m21 (x1 + x2 )2 − [s]x1 x2
F [s, 0, 0] = −[s]x1 x2

The ’conventional’ Feynman parameter integral is 1-dimensional because x2 ≡ 1 − x1 :

F (x) = −sx(1 − x) + m22 (1 − x) + m21 x ≡ −s(x − xa )(x − xb )

%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 28

The result is of logarithmic type for the constant term in ǫ:

Z 1
Γ(1 + ǫ) dx
B0 [s, m1 , m2 ] = (4πµ2 )ǫ eǫγE ǫ
ǫ 0 F (x)
Z 1  
1 F (x)
= − dx ln
ǫ 0 4πµ2
( Z 1  )
ζ2 1 F (x)
+ǫ + dx ln2 2
+ O(ǫ2 ).
2 2 0 4πµ

Here we used the expansion:

ζ2 2 ζ3 3
eǫγE Γ(1 + ǫ) = 1 + ǫ − ǫ ···
2 3
When using LoopTools, the corresponding call returns exactly the constant term of B0
in ǫ (with use of eǫγE = 1 + ǫγE + · · · → 1):
(0)
B0 (s, m21 , m22 ) = b0(s, am12, am22)

For 4πµ2 → 1 B0 looks quite compact:

Z 1  Z 1 
1 ǫ
B0 (s, m1 , m2 ) = − dx ln[F (x)] + ζ2 + dx ln2 [F (x)] + · · ·
ǫ 0 2 0
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 29

Explicitly, one has to integrate

ln[F (x)] = ln[−s(x − xa )(x − xb )]

ln2 [F (x)] = ln2 [−s(x − xa )(x − xb )]

So we will need the integrals:

Z
dx10 {ln(x − xa ), ln(x − xa )ln(x − xb )}

which is trivial, together with some complex algebra rules how to handle complex
arguments of logarithms with

s → s + iǫ

wherever needed.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 30

For the case m1 = m2 = 1, one gets for the first terms in ǫ:

1 1+y
B0 [s, 1, 1] = +2+ H(0, y),
ǫ 1−y
H(0, y) = ln(y).

The H(0, y) is a harmonic polylogarithmic function, and

√ √
−s + 4 − −s
y = √ √
−s + 4 + −s
(1 − x)2
s = −
x

The other case treated later again is m1 = 0, m2 = m:

2 1 1 − s/m2 2
B0 [s, m , 0] = +2+ ln(1 − s/m )
ǫ s/m2
%

$
&

'
The massive one-loop vertex C0(s, m1 , m2 )
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 31

p1

m1
p3
eǫγE dd k d4 k
Z
m2
C0 = ∼ |k→∞ 6 → UV − fin.
m3 (iπ d/2 ) [(k + p1 )2 − m2 ][k 2 ][(k − p2 )2 − m2 ] k
p2
The massive vertex (all m1 , m2 , m3 6= 0) is a finite quantity.
We assume immediately m1 = m3 = 0.
A problem now is IR-divergence.
Appears when a massive internal line is between two external on-shell lines.
Incoming p21 = m2 and p22 = m2 , look at k → 0:
1 1 1
d4 k
(k − p2 )2 − m2 (k)2 (k + p1 )2 − m2
1 1 1
= d4 k 2
k − 2kp2 (k)2 k 2 + 2kp1
d4 k k 3 dk dk
→ ∼ 4 ∼ |k→0 −→ div
k 1+2+1 k k
An IR-regularization is needed, must take d > 4.
Both UV-div (with d < 4) and IR-div together: must allow for a complex d = 4 − 2ǫ, and
take limit at the end.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 32

First we have a look, for later use, at the F -function:

N = D1 x + D2 y + D3 z
= k 2 x + (k 2 + 2kp1 )y + (k 2 − 2kp2 )z
= k 2 (x + y + z) + 2k(p1 y − p2 z)
= (k + Q)2 − Q2

We used 1 = x + y + z here. And the F -function is F = Q2 − J = Q2 (there is no

constant term in N here), as was shown before:

F = m2 (y + z)2 + [−s]yz

This F -function does not factorize in y and z. But now back to the direct Feynman
parameter integration.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 33

Start with change y → y ′ = (1 − x)y, then y ′ → y:

Z 1
1 δ(1 − x − y − z)
= dxdydz
D1 D2 D3 0 (D2 x + D1 y + D3 z)3
Z 1 Z 1−x
dy
= dx
0 0 (D2 x + D1 y + D3 z)3
Z 1 Z 1
xdy
= dx 3
0 0 (D2 x + D1 y + D3 z)

After this change of variables, the integrand factorizes in x and y:

N = (k + xpy )2 − x2 p2y
= (k + Q)2 − Q2

resulting into

F = Q2 = x2 p2y
p2y = −sy(1 − y) + m2
%

$
&

'
For C0 we obtain (with Nν = 3 and Nν − d/2 = 1 + ǫ):
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 34

Z 1 Z 1
dx dy
C0 [s, m, m, 0] = (−1)eǫγE Γ[1 + ǫ] 1+2ǫ 2 1−ǫ
0 x 0 (py )

This is integrable for ǫ < 0, or d > 4, or more general: d ≮ 4.

The x-integral made simple here:
Z 1
dx x−2ǫ |10 1−2ǫ − 0−2ǫ
1+2ǫ
= =−
0 x −2ǫ 2ǫ
1
= −
2ǫ
We see that the IR-singularity is an end-point-singularity in Feynman parameter space.
Further:
1 dy 1 dy 2 ǫ
− = − (y )
2ǫ (p2y )1−ǫ 2ǫ (p2y )
1 dy ǫ ln(y2 )
= − e
2ǫ (p2y )
1 dy  2 2 2 2

= − 1 + ǫ ln(y ) + ǫ ln (y ) + · · ·
2ǫ (p2y )
Here I stop this study.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 35

We see that the further integrations proceed quite similar as for the 2-point function, in
fact the p2y = −sy(1 − y) + m2 is the same building block.
The integrals to be solved now are more general, they include also denominators 1/p2y :
Some integrals
Z
dy ln(y − y0 ) = (y − y0 ) ln(y − y0 ) − y + C

1
Z
dy = ln(y − y0 ) + C
y − y0
ln(y − y0 ) 1 2
Z
dy = ln (y − y0 ) + C
y − y0 2
Here, often y is real and y0 is complex. Then no special care about phases is necessary.
Z 1    
dx x0 x0 − 1
[ln(x − xA ) − ln(x0 − xA )] = Li2 − Li2 .
0 x − x0 x0 − xA x0 − xA

This formula is valid if x0 is real.

%

$
&

'
C0 with a small photon mass λ
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 36

In
[Berends:1976zp,’tHooft:1979xw]

, the C0 -integral is treated with a finite photon mass:

d4 k
Z

(k 2 − λ2 )(k 2 + 2kp1 )(k 2 − 2kp2 )

Z 1
y
= −iπ 2 dydx
0 x2 p2y + (1 − x)λ2
Z 1  2  q 
1 λ
= iπ 2 dy 2
ln 2 + O λ/ p2y ,
0 2py py

It is easy to see from the term 1/(2p2y ) ln(λ2 ) the correspondence of (d − 4) and λ2 ,
which is a universal relation in all 1-loop cases.

16.02.2009: End of Lecture 1.

%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 37

Now using Mellin-Barnes Representations

Perform the x-integrations

Find an as-general-as-possible general formula

Make it ready for algorithmic analytical and/or numerical evaluation

Computercodes:
• Ambre.m - Derive Mellin-Barnes representations for Feynman integrals
[Gluza:2007rt]

• MB.m - Find an ǫ-expansion and evaluate numerically in Euclidean region

[Czakon:2005rk]
%

$
&

'
Integrating the Feynman parameters – get MB-Integrals
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 38

We derived:
1
δ(1 − x1 − x2 )
Z
ǫγE
SE2l1m = B0 (s, m, 0) = e Γ(ǫ) dx1 dx2 δ(1 − x1 − x2 )
0 F (x)ǫ
1
δ(1 − x1 − x2 − x3 )
Z
ǫγE
V 3l2m = C0 (s, m, m, 0) = e Γ(1 + ǫ) dx1 dx2 dx3
0 F (x)1+ǫ
and

FSE2l1m = m2 x21 − (s − m2 )x1 x2

FV 3l2m = m2 (x1 + x2 )2 − (s)x1 x2

We want to apply now:

N
1Y
Γ(α1 )Γ(α2 ) · · · Γ(α7 N )
Z  X 
α −1
dxj xj j δ 1− xi =
0 j=1 Γ (α1 + α2 + · · · + αN )

with coefficients αi dependent on νi and on the structure of the F

See in a minute:
For this, we have to apply one or several MB-integrals here.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 39

N N
! QN
1Y
Γ(αi )
Z
α −1
X
dxj xj j δ 1− xi = i=1
PN 
0 j=1 i=1 Γ i=1 αi

Simplest cases:

Z 1
dx1 xα
1
1 −1
δ (1 − x1 ) = 1
0
2 N
!
Z 1 Z 1
α −1
Y X
dxj xj j δ 1− xi = dx1 xα
1
1 −1
(1 − x1 )α2 −1 = B(α1 , α2 )
0 j=1 i=1 0

Γ(α1 )Γ(α2 )
=
Γ (α1 + α2 )
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 40

Here we want to go:

+i∞
1 1 1 Bz
Z
= dzΓ(λ + z)Γ(−z) λ+z
(A + B)λ Γ(λ) 2πi −i∞ A

The formula looks a bit unusual to loop people, but for persons with a mathematical
background it is common.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 41

One might well assume that these two gentlemen did not dream of so heavy use of
their results in basic research · · ·
Mellin, Robert, Hjalmar, 1854-1933
Barnes, Ernest, William, 1874-1953
%

$
&

'
.
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 42

Barnes’ contour integrals for the hypergeometric function

Exact proof and further reading: Whittaker & Watson (CUP 1965) 14.5 - 14.52, pp.
286-290
Consider
+i∞
1 Γ(a + σ)Γ(b + σ)Γ(−σ)
Z
F (z) = dσ(−z)σ
2πi −i∞ Γ(c + σ)

where |arg(−z)| < π (i.e. (−z) is not on the neg. real axis) and the path is such that it
separates the poles of Γ(a + σ)Γ(b + σ) from the poles of Γ(−σ).
1/Γ(c + σ) has no pole.
Assume a 6= −n and b 6= −n, n = 0, 1, 2, · · · so that the contour can be drawn.
The poles of Γ(σ) are at σ = −n, n = 1, 2, · · · , and it is:

Residue[ F[s] Gamma[-s] , {s,n} ] = (-1)ˆn/n! F(n)

Closing the path to the right gives then, by Cauchy’s theorem, for |z| < 1 the
%

$
&

'
hypergeometric function 2 F1 (a, b, c, z) (for proof see textbook):
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 43

+i∞ NX
→∞
1 Γ(a + σ)Γ(b + σ)Γ(−σ) Γ(a + n)Γ(b + n) z n
Z
dσ(−z)σ =
2πi −i∞ Γ(c + σ) n=0
Γ(c + n) n!
Γ(a)Γ(b)
= 2 F1 (a, b, c, z)
Γ(c)

The continuation of the hypergeometric series for |z| > 1 is made using the
intermediate formula
∞
X Γ(a + n)Γ(1 − c + a + n) sin[(c − a − n)π]
F (z) = (−z)−a−n
n=0
Γ(1 + n)Γ(1 − a + b + n) cos(nπ) sin[(b − a − n)π]
∞
X Γ(b + n)Γ(1 − c + b + n) sin[(c − b − n)π]
+ (−z)−b−n
n=0
Γ(1 + n)Γ(1 − a + b + n) cos(nπ) sin[(a − b − n)π]

and yields
Γ(a)Γ(b) Γ(a)Γ(a − b)
2 F1 (a, b, c, z) = (−z)−a 2 F1 (a, 1 − c + a, 1 − b + ac, z −1 )
Γ(c) Γ(a − c)
Γ(b)Γ(b − a)
+ (−z)−b 2 F1 (b, 1 − c + b, 1 − a + b, z −1 )
Γ(b − c)
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 44

Corollary I

Putting b = c, we see that

∞
X Γ(a + n) z n
2 F1 (a, b, b, z) =
n=0
Γ(a) n!
+i∞
1 1
Z
= = dσ (−z)σ Γ(a + σ)Γ(−σ)
(1 − z)a 2πi Γ(a) −i∞

This allows to replace sum by product:

Zi∞
1 1 1
= = dσAσ B −σ−a Γ(a + σ)Γ(−σ)
(A + B)a B a [1 − (−A/B)]a 2πiΓ(a)
−i∞
%

$
&

'
Barnes’ lemma
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 45

If the path of integration is curved so that the poles of Γ(c − σ)Γ(d − σ) lie on the right
of the path and the poles of Γ(a + σ)Γ(b + σ) lie on the left, then
Z +i∞
1 Γ(a + c)Γ(a + d)Γ(b + c)Γ(b + d)
dσΓ(a + σ)Γ(b + σ)Γ(c − σ)Γ(d − σ) =
2πi −i∞ Γ(a + b + c + d)
It is supposed that a, b, c, d are such that no pole of the first set coincides with any pole
of the second set.
Scetch of proof: Close contour by semicircle C to the right of imaginary axis. The
R
integral exists and C vanishes when ℜ(a + b + c + d − 1) < 0. Take sum of residues of
the integrand at poles of Γ(c − σ)Γ(d − σ). The double sum leads to two
hypergeometric functions, expressible by ratios of Γ-functions, this in turn by
combinations of sin, may be simplifies finally to the r.h.s.
Analytical continuation: The relation is proved when ℜ(a + b + c + d − 1) < 0.
Both sides are analytical functions of e.g. a. So the relation remains true for all values
of a, b, c, d for which none of the poles of Γ(a + σ)Γ(b + σ), as a function of σ, coincide
with any of the poles of Γ(c − σ)Γ(d − σ).
R −k+i∞
Corollary II Any real shift k: σ + k, a − k, b − k, c + k, d + k together with −k−i∞ leaves
the result true.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 46

How can the Mellin-Barnes formula be made useful in the

context of Feynman integrals?
• Apply corollary I to propagators and get:
Zi∞
1 1 (−m2 )σ
= dσ 2 a+σ Γ(a + σ)Γ(−σ)
(p2 − m2 )a 2πi Γ(a) (p )
−i∞

which transforms a massive propagator to a massless one (with index a of the line changed
to (a + σ)).
• Apply corollary I after introduction of Feynman parameters and after the momentum
integration to the resulting F - and U -forms, in order to get a single monomial in the xi ,
which allows the integration over the xi :
Zi∞
1 1
= dσ[A(s)xa1 1 ]σ [B(s)xb11 xb22 ]a+σ Γ(a + σ)Γ(−σ)
[A(s)xa1 1 + B(s)xb11 xb22 ]a 2πi Γ(a)
−i∞

Both methods leave Mellin-Barnes (MB-) integrals to be performed afterwards.

%

$
&

'
A short remark on history
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 47

• N. Usyukina, 1975: ”ON A REPRESENTATION FOR THREE POINT FUNCTION”, Teor.

Mat. Fiz. 22;
a finite massless off-shell 3-point 1-loop function represented by 2-dimensional MB-integral
• E. Boos, A. Davydychev, 1990: ”A Method of evaluating massive Feynman integrals”,
Theor. Math. Phys. 89 (1991);
N-point 1-loop functions represented by n-dimensional MB-integral
• V. Smirnov, 1999: ”Analytical result for dimensionally regularized massless on-shell double
box”, Phys. Lett. B460 (1999);
treat UV and IR divergencies by analytical continuation: shifting contours and taking
residues ’in an appropriate way’
• B. Tausk, 1999: ”Non-planar massless two-loop Feynman diagrams with four on-shell legs”,
Phys. Lett. B469 (1999);
nice algorithmic approach to that, starting from search for some unphysical space-time
dimension d for which the MB-integral is finite and well-defined
• M. Czakon, 2005 (with experience from common work with J. Gluza and TR): ”Automatized
analytic continuation of Mellin-Barnes integrals”, Comput. Phys. Commun. (2006);
Tausk’s approach realized in Mathematica program MB.m, published and available for use
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 48

The Γ-function
The Γ-function may be defined by a difference equation:

zΓ(z) − Γ(z + 1) = 0

Z ∞
Γ(z) = tz−1 e−t dt
0

Series[Gamma[ep], {ep, 0, 2}] =

1 1 1
Γ[ǫ] = − γE + (6γE 2 + π 2 )ǫ + (−2γE 3 − γE 2 π + 2Ψ(2, 1))ǫ2 + · · ·
ǫ 12 12

Ψ(2, 1)) = P olyGamma(2, 1) = −2ζ3

exp(ep EulerGamma)Series[Gamma[ep], {ep, 0, 2}] =
1 1 1
eǫγE Γ[ǫ] = + (π 2 )ǫ + (Ψ(2, 1))ǫ2 + · · ·
ǫ 12 6
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 49

Look at the singularities in the complex plane:

When applying the Cauchy-theorem, one may close the integral to the left or to the right . . .
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 50

Some facts on residua

The function
∞
X ai
F (z) =
(z − z0 )i
i=−N

has the residue

Res F (z)|z=z0 = a−1

An integral over an anti-clockwize directed closed path C around z0 then is

1
Z
dzF (z) = 2πia−1
2πi C
If G(z) has a Taylor expansion around z0 and F (z) has a Laurent expansion beginning with
a−N /(z − z0 )N + . . ., then their product has the residue:
N
X a−k G(z0 )(k)
Res[G(z) F (z)]|z=z0 =
k!
k=1
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 51

Some residua with Γ(z) and Ψ(z)

Ψ(z) = P olyGamma[z] = P olyGamma[0, z] → See also next slide
(−1)n
Residue[F[z]Gamma[z], {z, -n}] = n! F [−n]

2P olyGamma[n+1]F [−n]+F ′ [−n]

Residue[F[z]Gamma[z]2 , {z, -n}] = (n!)2

25F [−3]−12γE F [−3]+6F ′ [−3]

Residue[F[z]Gamma[z - 1]2 , {z, -3}] = 3456

F [−3] 25F [−3]−12γE F [−3]+6F ′ [−3]

F[z] = 576(z+3)2 + 3456(z+3)

+ a0 + a1 (z + 3) + · · ·
Γ[−3+a]
Series[Gamma[z + a]Gamma[z - 1]2 , {z, -3, -1 }] = 576(z+3)2

(25Γ[−3+a]−12γE Γ[−3+a]+6(Γ[−3+a]P olyGamma[0,−3+a])

+ 3456 + a0 + a1 (z + 3) + · · ·
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 52

Where

Polygamma[n + 1] ≡ Polygamma[0, n + 1]
n
Γ′ (n + 1) X 1
= Ψ(n + 1) = = S1 (n) − γE = − γE
Γ(n + 1) k
k=1

The following properties hold:

Ψ(z + 1) = Ψ(z) + 1/z

Ψ(1 + ǫ) = −γE + ζ2 ǫ + . . .
Ψ(1) = −γE
Ψ(2) = 1 − γE
Ψ(3) = 3/2 − γE

∂n
P olyGamma[n, z] = n Ψ(z)
∂z
%

$
&

'
Some sums Mathematica can do
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 53

Sum[sˆ(n) Gamma[n + 1]ˆ3/(n!Gamma[2 + 2n]), n, 0, Infinity] =

. (4*ArcSin[Sqrt[s]/2])/(Sqrt[4 - s]*Sqrt[s])

Sum[sˆ(n) PolyGamma[0, n + 1], n, 0, Infinity] =

. (EulerGamma + Log[1 - s])/(-1 + s)

The above were done with Mathematica 5.2.

Mathematica 6 is more powerful.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 54

A self-energy: SE2l1m
This is a nice example, being simple but showing [nearly] all essentials in a nutshell.
We get for this F (x) = m2 x21 − (s − m2 )x1 x2 the following representation:
−ǫ−z
eǫγE (m2 )−ǫ −s + m2
Z 
SE2l1m = dz Γ1 [1 − ǫ − z]Γ2 [−z]Γ3 [1 − ǫ + z]Γ4 [ǫ + z]
2πi Γ[2 − 2ǫ] ℜz=−1/8 m2

Tausk approach:
Seek a configuration where all arguments of Γ-functions have positive real part. Then the
SE2l1m is well-defined and finite.
For small ǫ this is - here - evidently impossible; set ǫ → 0 and look at Γ2 [−z]Γ4 [+z]:

Γ1 [1 − z]Γ2 [−z]Γ3 [1 − +z]Γ4 [+z]

What to do ????
Tausk: Set ǫ such that it happens, e.g.:

ǫ = 3/8

To make physics we have now to deform the integrand or the path such that ǫ → 0; when
crossing a residue, take it and add it up.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 55

Varying ǫ → 0 from 3/8 makes crossing in Γ4 [ǫ + z] a pole at ǫ = −z = +1/8; there is ǫ + z = 0:

ǫγE (m2 )−ǫ

Residue[SE2l1m, {z, −ǫ}] = e Γ1 [1 − 2ǫ]Γ2 [ǫ]
Γ[2 − 2ǫ]
Here we ’loose’ one integration (easier term!) and catch the IR-singularity in Γ2 [ǫ] ∼ 1/ǫ!
The function becomes now, for small ǫ:
−ǫ−z
eǫγE (m2 )−ǫ −s + m2
Z 
SE2l1m = dz Γ1 [1 − ǫ − z]Γ2 [−z]Γ3 [1 − ǫ + z]Γ4 [ǫ + z]
2πi Γ[2 − 2ǫ] ℜz=−1/8 m2
ǫγE (m2 )−ǫ
+ e Γ1 [1 − 2ǫ]Γ2 [ǫ]
Γ[2 − 2ǫ]
Now we may take the limit of small ǫ because the integral will stay finite and well-defined:
−z
eǫγE −s + m2
  
1
Z
ǫγE 2
SE2l1m = dz Γ 3 [1 − z]Γ 4 [−z]Γ[z]Γ[1 + z] + e 2 + − ln[m ] +
2πi ℜz=−1/8 m2 ǫ

Now we close the integration path to the left, catch all residues from Γ3 [1 − z] for ℜz < −1/8,
i.e. at z = −n, n = 1, 2, . . .:
2 −z
(  )
−s + m
Residue 2
Γ1 [1 − ǫ − z]Γ2 [−z]Γ3 [1 − ǫ + z]Γ4 [ǫ + z], {z, −n} = (−s + m2 )n ln(−s + m
m
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 56

The sum to be done is trivial (in this trivial case!!):

∞  n
X −s + m2 1
2
= −s+m2
−1
n=1
m 1 − m2

and we end up with:

1 − s/m2
 
1
SE2l1m = +2+ 2
ln(1 − s/m2 )
ǫ s/m
This is what we had also from the direct Feynman parameter integration above
%

$
&

'
A vertex: V3l2m
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 57

The Feynman integral V3l2m is the QED one-loop vertex function, which is no master.
It is infrared-divergent (see this by counting of powers of loop integration momentum k or know
it from: massless line between two external on-shell lines)

F = m2 (x1 + x2 )2 + [−s]x1 x2

We will also use the variable

√ √
−s + 4 − −s
y= √ √
−s + 4 + −s

eǫγE Γ(−2ǫ) 2
−ǫ−1−z Γ (−ǫ − z)Γ(−z)Γ(1 + ǫ + z)
Z
V3l2m = dz(−s)
2πi Γ(1 − 2ǫ)Γ(−2ǫ − 2z)
V3l2m[−1]
= + V3l2m[0] + ǫ V3l2m[1] + · · · .
ǫ
One may slightly shift the contour by (−ǫ) and then close the path to the left and get residua
from (and only from) Γ(1 + z):
[Gluza:2007bd]
%

$
&

'
:
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 58

−i∞−1/2
ǫγE 2
1 e −z Γ (−z)Γ(−z + ǫ)Γ(1 + z)
Z
V (s) = dz (−s)
2sǫ 2πi Γ(−2z)
−i∞−1/2
∞
eǫγE X sn Γ(n + 1 + ǫ)
= − 2n

.
2ǫ n=0 n (2n + 1)
Γ(n + 1)

This series may be summed directly with Mathematica!a , and the vertex becomes:
eǫγE
V (s) = − Γ(1 + ǫ) 2 F1 [1, 1 + ǫ; 3/2; s/4] .
2ǫ
Alternatively, one may derive the ǫ-expansion by exploiting the well-known relation with
Pn
harmonic numbers Sk (n) = i=1 1/ik :
" ∞ #
k
Γ(n + aǫ) X (−aǫ)
= Γ(1 + aǫ) exp − Sk (n − 1) .
Γ(n) k
k=1

The product exp (ǫγE )Γ(1 + ǫ) = 1 + 12 ζ[2]ǫ2 + O(ǫ3 ) yields expressions with zeta numbers
a The expression for V (s) was also derived in
[Huber:2007dx]
; see additionally
[Davydychev:2000na]
.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 59

ζ[n], and, taking all terms together, one gets a collection of inverse binomial sumsb ; the first of
them is the IR divergent part:
V−1 (s)
V (s) = + V0 (s) + · · ·
ǫ
∞ n
√
1 X s 1 4 arcsin( s/2) y
V−1 (s) = = √ √ = ln(y).
2 n=0 2n 2−1


n (2n + 1)
2 4−s s y

b For the first four terms of the ǫ-expansion in terms of inverse binomial sums or of polylogarithmic functions,
see
[Gluza:2007bd]
.
%

$
&

'
The constant term:
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 60

Z +i∞+u 3
1 −1−r Γ [−r]Γ[1 + r]
V3l2m[0] = dr(−s)
2πi −i∞+u Γ[−2r]
1
[γE − ln(−s) + 2Ψ[−2r] − 2Ψ[−r] + Ψ[1 + r]]
2

∞
1X sn
=   S1 (n),
2 n=0 2n
  (2n + 1)
n

There is also the opportunity to evaluate the MB-integrals numerically by following with e.g. a
Fortran routine the straight contour.
This applies after the ǫ-expansion.

R +5i+ℜz
−5i+ℜz
is usually sufficient.

But: This works fast and stable for Euclidean kinematics where −s > 0.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 61

and the ǫ-term:

1/4 +i∞+u 3
−1−r Γ [−r]Γ[1 + r]
Z
V3l2m[1] = dr(−s)
2πi −i∞+u Γ[−2r)]
h
2
γE + Log[−s]2 + Log[−s](−2γE − 4Ψ[−2z] + 4Ψ[−z] − 2Ψ[1 + z])
+γE (4Ψ[−2z] − 4Ψ[−z] + 2Ψ[1 + z])
−4Ψ[1, −2z] + 2Ψ[1, −z] + Ψ[1, 1 + z]
+4(Ψ[−2z]2 − 2Ψ[−2z]Ψ[−z] + Ψ[−z]2 + Ψ[−2z]Ψ[1 + z]
i
2
−Ψ[−z]Ψ[1 + z]) + Ψ[1 + z]
∞
1X (t)n
S1 (n)2 + ζ2 − S2 (n) .
 
= [const = 1?] ×  
4 n=0 2n
  (2n + 1)
n

Here, Ψ[r] = ... and Ψ[1, r] = ..., and the harmonic numbers Sk (n) are
n
X 1
Sk (n) = k
,
i=1
i
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 62

The sums appearing above may be obtained from sums listed in Table 1 of Appendix D in
[Gluza:2007bd,Davydychev:2003mv]

:
∞
X sn y
2n = 2 ln(y),
y2 − 1


n=0 n (2n + 1)
∞
X sn y 2

2n S1 (n) = [−4Li2 (−y) − 4 ln(y) ln(1 + y) + ln (y) − 2ζ 2 ,
y2 − 1


n=0 n (2n + 1)
∞
sn

X y
2n S1 (n)2 = 16S1,2 (−y) − 8Li3 (−y) + 16Li2 (−y) ln(1 + y)
y2 − 1


n=0 n (2n + 1)
1
+8 ln2 (1 + y) ln(y) − 4 ln(1 + y) ln2 (y) + ln3 (y) + 8ζ2 ln(1 + y)
3

−4ζ2 ln(y) − 8ζ3 ,
∞
X sn y
2n

S 2 (n) = − 2
ln3 (y),
n=0 n (2n + 1)
3(y − 1)
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 63

Expansion in a small parameter: vertex V3l2m for m2 /s

Use as an example for determining the small mass expansion:

V 3coef m1 = Coef f icient[V 3l2m[[1, 1]], ǫ, −1]

Z +i∞−1/2  z
1 1 m2 Γ1 [−z]3 Γ2 [1 + z]
= − dz −
2s 2πi −i∞−1/2 s Γ3 [−2z]

If |m2 /s| << 1, then the smallest power of it gives the biggest contribution, i.e. its exponent
has to be positive and small.
So, close the contour to the right (positive ℜz), and leading terms come from the residua
expansion of Γ1 [−z]3 /Γ3 [−2z] at z = −1, −2, · · · . The residues are terms of a binomial sum:
n
1 m2 m2
   
(2n)!
Residue = − 2HarmonicN umber[n] − 2HarmonicN umber[2n] − ln −
s s (n!)2 s
with first terms equal to (-1)*Residua:

m2 m2 m2 m4 m2
       
1
V 3l2m = ln − + 2 ln 2 + 2 + 3 ln 7 + 6 + O(m6 /s3 )
s s s s s s
%

$
&

'
.
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 64

Another nice box with numerator, B5l3m(pe .k1 )

We used it for the determination if the small mass expansion.

m4ǫ (−1)a12345 e2ǫγE +i∞ +i∞ +i∞ +i∞
Z Z Z Z
B5l3m(pe · k1 ) = Q5 dα dβ dγ dδ
j=1 Γ[ai ]Γ[5 − 2ǫ − a123 ](2πi)4 −i∞ −i∞ −i∞ −i∞
(4−2 ǫ)−a12345 −α−β−δ δ
(−s) (−t)
Γ[−4 + 2 ǫ + a12345 + α + β + δ] Γ[−α] Γ[−β] Γ[−δ]
Γ[6 − 3 ǫ − a12345 − α] Γ[7 − 3 ǫ − a12345 − α] Γ[5 − 2 ǫ − a123 ] Γ[4 − 2 ǫ − a1123 − 2 α − γ] Γ[5 − 2 ǫ − a1123 − 2 α
Γ[2 − ǫ − a13 − α − γ] Γ[4 − 2 ǫ − a12345 − α − β − δ − γ]

(pe · p3 ) Γ[1 + a4 + δ] Γ[6 − 3 ǫ −
Γ[8 − 4 ǫ − a112233445 − 2 α − 2 β − 2 δ − γ] Γ[9 − 4 ǫ − a112233445 − 2 α − 2 β − 2 δ − γ]
Γ[4 − 2 ǫ − a1234 − α − β − δ] Γ[3 − ǫ − a12 − α] Γ[8 − 4 ǫ − a112233445 − 2 α − 2 δ − γ]Γ[9 − 4 ǫ − a112233445 − 2 α − 2 β − 2
»
Γ[5 − 2 ǫ − a1123 − γ] Γ[4 − 2 ǫ − a1123 − 2 α − γ] Γ[a1 + γ] Γ[−2 + ǫ + a123 + α + δ + γ] + Γ[a4 + δ] −(pe · p1 ) Γ[7 − 3 ǫ − a

Γ[4 − 2 ǫ − a1234 − α − β − δ]Γ[8 − 4 ǫ − a112233445 − 2 α − 2 δ − γ]Γ[9 − 4 ǫ − a112233445 − 2 α − 2 β − 2 δ − γ]

»
Γ[3 − ǫ − a12 − α] Γ[5 − 2 ǫ − a1123 − γ]Γ[4 − 2 ǫ − a1123 − 2 α − γ] Γ[a1 + γ] + Γ[2 − ǫ − a12 − α] Γ[4 − 2 ǫ − a1123 − γ]
–
Γ[5 − 2 ǫ − a1123 − 2 α − γ] Γ[1 + a1 + γ] Γ[−2 + ǫ + a123 + α + δ + γ] + Γ[6 − 3 ǫ − a12345 − α] Γ[3 − ǫ − a12 − α]
»
Γ[5 − 2 ǫ − a1123 − γ] Γ[4 − 2 ǫ − a1123 − 2 α − γ]Γ[a1 + γ] ((pe · (p1 + p2 )) Γ[5 − 2 ǫ − a1234 − α − β − δ]Γ[9 − 4 ǫ − a112233

Γ[8 − 4 ǫ − a112233445 − 2 α − 2 β − 2 δ − γ]Γ[−2 + ǫ + a123 + α + δ + γ] + (pe · p1 ) Γ[4 − 2 ǫ − a1234 − α − β − δ]

–ff
Γ[8 − 4 ǫ − a112233445 − 2 α − 2 δ − γ]Γ[9 − 4 ǫ − a112233445 − 2 α − 2 β − 2 δ − γ]Γ[−1 + ǫ + a123 + α + δ + γ]
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 65

Differential equations: Example V3l2m

Example V3l2m, the massive QED 1-loop vertex

p1
dd k
Z
m1
V 3l2m =
D1 D2 D3
p3
m2
D1 = (k + p1 )2 − m21
m3
D2 = (k − p2 )2 − m22
p2
D3 = (k)2 − m20

s = (p1 + p2 )2 = 2p1 p2 + p21 + p22

Dimensional arguments prove:

1
V 3l2m ∼ at s → ∞
s
%

$
&

'
We derive now a differential eqation for the vertex, V = V (s, p21 , p22 ))
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 66

s = (p1 + p2 )2 = 2p1 p2 + p21 + p22 → 2p1 p2 + 2m2 or → 2p1 p2 + m2 etc.

First derive a differential operator, which is completely independent of the internal details the
vertex.
∂ ∂p22
 
∂ ∂ ∂s
p1 = p1 +
∂p2 ∂p22 ∂p2 ∂s ∂p2
 
∂ ∂
= p1 2p2 2 + 2(p1 + p2 )
∂p2 ∂s
 
∂ ∂ ∂
p2 = p2 2p2 2 + 2(p1 + p2 )
∂p2 ∂p2 ∂s
Not Using p21 = p22 = m2 and p1 p2 = (s − 2m2 )/2, we get:
∂ ∂ ∂
p1 = (s − p21 − p22 ) + s
∂p2 ∂p22 ∂s
∂ ∂ ∂
p2 = (p21 + p22 ) 2 + s
∂p2 ∂p2 ∂s
Eliminate ∂/∂p22 :
∂  ∂ ∂
[s − 2(p21 + p22 )]s = [s − (p21 + p22 )] p2 − (p21 + p22 ) p1

= [R2 p2 + R1 p1 ]
∂s ∂p2 ∂p2
%

$
&

'
Now apply this universal differential vertex-operator to a specific vertex, e.g. V =V3l2m:
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 67

Use on-mass-shell condition p21 = m21 , p22 = m22 .

∂V dd k ∂ 1
Z
=
∂p2 D3 D1 ∂p2 D2
With
∂ 1 1 ∂D2 2(k − p2 )
= − =
∂p2 D2 D22 ∂p2 D22
Compensate numerator against propagators:

2p1 k = −D3 + D1 − m20 − p21 + m21

2p2 k = D3 − D2 + m20 + p22 − m22
p2i = m2i
p1 p2 = (s − m21 − m22 )/2

we get, for this vertex, the equation:

d
∂V d k dd k
Z Z
2 2 2 2
(s − 2(m1 + m2 )) s = s 2 − (m1 + m2 ) 2 − (s − (m21 + m22 )) V
∂s D1 D2 D3 D2
There might also arise a dotted Vertex at the right hand side, but here it drops out from the
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 68

result; in general:
"
 ∂ −R2 R2 − R1 R1
Z
2 2

s − 2(p1 + p2 ) s = dk + +
∂s D1 D2 D3 D1 D22 D22 D3
#
R2 (m20 − m22 − + R1 (−m20
p22 ) + p22 + m21 − s)
+
D1 D22 D3
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 69

This is a simple Euler differential equation!

The inhomogeneity depends on two dotted 2-point functions SE2l2md and SE2l1md, so it is
simpler and assumed to be known already.
In principle, one might get also a term with d k/D1 D22 D3 , i.e. the dotted vertex V3l2md, but
R d

it drops out here.

Then one would have a system of coupled equations, and this often happens.
We see that dotted objects may appear naturally,.
All the above has nothing to do with dimension d – if it is well-defined . . .
In
[Czakon:2004wm]

we use the operator for deriving and solving massive 2-loop vertex functions and give also a
corresponding operator for box diagrams, m2 → 1:
∂ 1 ∂
s = [2pµ2 + (2 − s)pµ1 ] µ for vertex
∂s 4−s ∂p1
 µ µ 
∂ 1 s (p − p3) ∂
s = (pµ1 + pµ2 ) + 2
for box
∂s 2 s + t − 4 ∂pµ2
%

$
&

'
Scetch of: Solve Euler Diff. Equations
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 70

Basics developed in
[Kotikov:1991hm,Kotikov:1991kg,Laporta:1996mq]

. The choice of variables depends on the problem.

For the QED vertex it will prove useful to take

s/m2 = −(1 − x)2 /x

For our other beloved example B ≡ B0 (s, m, 0), see nice lecture
[Aglietti:2004vs]

s/m2 = −x

For details, I prefer to jump to this case.

Differential operator:
∂ 1 µ ∂
s = p
∂s 2 ∂pµ
The equation becomes here:
     
∂ 1 1 1 2 1 1
B = − + B+ǫ − B + (1 − ǫ) − T 1l1m
∂s x 1+x x 1+x x 1+x
1
T 1l1m = + 1 + (1 + ζ2 /2)ǫ + . . .
ǫ
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 71

We need also a boundary condition; look at explicit expression above:

B0 (0, m, 0) = T 1l1m

Make an ansatz as series in ǫ:

1
B= B1 + B0 + ǫB1 + . . .
ǫ
and get a series of equations:
∂
B−1 = A0 B−1 + Ω−1
∂s
∂
B0 = A0 B0 + A1 B−1 + Ω0
∂s
∂
B1 = A 0 B1 + A 1 B0 + Ω 1
∂s
etc
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 72

It is:
1 1
A0 = − +
x 1+x
1 2
A1 = −
x 1+x
1 1
Ω−1 = −
x 1+x
1 1
Ω0 = − +
x 1+x
Ωi>0 = 0

You may realize a simple, iterative structure.

The coefficients in the equation are of the form
A1 A2
+ + ..
x − B1 x − B2
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 73

General solution for the homogeneous equations:

Bn′ 1 1
= − +
Bn x 1+x
Solve it by
Z  
1 1
ln(Bnhom ) ≡ B hom = const + − +
x 1+x
gets
1
B hom = 1+
x
Solution of the inhomogenious equations is obtained by the ‘variation of constants’:
 
1
Z
Bk (x) = B hom (x) const + dx′ hom ′ [A1 (x′ )Bk−1 (x′ ) + Ωk (x′ )]
B (x )
This may be sequentially evaluated.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 74

Result:
nested integrals over ’simple’ iterated integrands
The method leads to the HPLs H({a}, x) and similar functions.
Harmonic Polylogarithms H(x)
x Z x′′
dx′′ dx′
Z
H[−1, 1, x] = ′′
0 (1 + x ) 0 (1 − x′ )
 
1+x
= Li2 +...
2

but it works only if the polynomial structure is simple enough for a solution with this class of
functions
Method is absolutely ’super’ if it works.
But:
one needs complete chains of masters of lower complexity.
%

$
&

'
Harmonic Polylogs
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 75

[Remiddi:1999ew]

(I borrowed some formulas from there)

H(0; x) = ln x ,
Z x
dx′
H(1; x) = ′
= − ln(1 − x) ,
0 1 − x
Z x
dx′
H(−1; x) = ′
= ln(1 + x) .
0 1 + x
Their derivatives are:
d
H(a; x) = f (a; x) , (1)
dx
where the index a can take 0, +1, −1, and
1
f (0; x) = ,
x
1
f (1; x) = ,
1−x
1
f (−1; x) = .
1+x
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 76

The harmonic polylogarithms of weight w are then defined as follows:

1 w
H(~0w ; x) = ln x ,
w!
~ w 6= ~0w
while, if m
Z x
H(m
~ w ; x) = dx′ f (a; x′ ) H(m
~ w−1 ; x′ ) .
0

The harmonic polylogarithms of weight w are then defined as follows:

1 w
H(~0w ; x) = ln x ,
w!
~ w 6= ~0w
while, if m
Z x
H(m
~ w ; x) = dx′ f (a; x′ ) H(m
~ w−1 ; x′ ) .
0
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 77

Examples:

H(0, 1; x) = Li2 (x) ,

H(0, −1; x) = −Li2 (−x) ,
H(1, 0; x) = − ln x ln(1 − x) + Li2 (x) ,
1 2
H(1, 1; x) = ln (1 − x) ,
2!
   
1−x 1
H(1, −1; x) = Li2 − ln 2 ln(1 − x) − Li2 ,
2 2
H(−1, 0; x) = ln x ln(1 + x) + Li2 (−x) ,
   
1+x 1
H(−1, 1; x) = Li2 − ln 2 ln(1 + x) − Li2 ,
2 2
1 2
H(−1, −1; x) = ln (1 + x) .
2!

In general, they are more general than Nielsen’s Polylogarithms, see e.g.:
Z x
dx′
H(−1, 0, 0, 1; x) = ′
Li3 (x′ )
0 1+x

cannot be expressed in terms of Nielsen’s polylogarithms

%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 78

Example beyond Harmonic Polylogs: QED Box B4l2m

[Fleischer:2006ht]

F[x] = (x5+x6)ˆ2 + (-s)x5x6 + (-t)x4x7

B4l2m, the 1-loop QED box, with two photons in the s-channel; the Mellin-Barnes
representation reads for finite ǫ:
Z +i∞ Z +i∞
eǫγE 1
B4l2m = Box(t, s) = dz1 dz2 (2)
Γ[−2ǫ](−t)(2+ǫ) (2πi)2 −i∞ −i∞
(−s)z1 (m2 )z2 2
Γ[2 + ǫ + z 1 + z 2 ]Γ [1 + z1 ]Γ[−z1 ]Γ[−z2 ]
(−t)z1 +z2
Γ[−2 − 2ǫ − 2z1 ]
Γ2 [−1 − ǫ − z1 − z2 ]
Γ[−2 − 2ǫ − 2z1 − 2z2 ]

Mathematica package MB used for analytical expansion ǫ → 0:

[Czakon:2005rk]
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 79

„ «
1 1ˆ 2 ˜
B4l2m = − J1 + ln(−s)J1 + ǫ ζ(2) − ln (−s) J1 − 2J2 . (3)
ǫ 2
with J1 being also the divergent part of the vertex function C0 (t; m, 0, m)/s = V3l2m/s (as is
well-known):
Z − 1 +i∞ „ 2 «z1 3
eǫγE 1 2 m Γ [−z1 ]Γ[1 + z1 ] 1 2y
J1 = dz1 = ln(y) (4)
st 2πi − 1 −i∞ −t Γ[−2z1 ] m2 s 1 − y 2
2

with
p
1 − 4m2 /t − 1)
y= p
( 1 − 4m2 /t + 1)
The J2 is more complicated:
−3 +i∞
eǫγE
Z
1 4
“ s ”z1
J2 = 2 dz1 Γ[−z1 ]Γ[−2(1 + z1 )]Γ2 [1 + z1 ] (5)
t (2πi)2 −3 −i∞ t
4

−1 +i∞
m2 Γ2 [−1 − z1 − z2 ]
Z „ «
2
z2
× dz2 − Γ[−z2 ] Γ[2 + z1 + z2 ].
−1 −i∞ t Γ[−2(1 + z1 + z2 )]
2
%

$
&

'
The expansion of B4l2m at small m2 and fixed value of t
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 80

With
−m2
mt = , (6)
t
s
r = , (7)
t
Look, under the integral, at (−m2 /t)z2 ,
and close the path to the right.
Seek the residua from the poles of Γ-functions with the smallest powers in m2 and sum the
resulting series.
we have obtained a compact answer for J2 with the additional aid of XSUMMER
[Moch:2005uc]

The box contribution of order ǫ in this limit becomes:

1n 2
2
B4l2m[t, s, m ; +1] = 4ζ3 − 9ζ2 ln(mt ) + ln3 (mt ) + 6ζ2 ln(r) − ln2 (mt )ln(r) (8)
st 3
1
+ ln3 (r) − 6ζ2 ln(1 + r) + 2ln(−r)ln(r)ln(1 + r) − ln2 (r)ln(1 + r)
3 o
+2ln(r)Li2 (1 + r) + 2Li3 (−r) + O(mt ).

Remark:
The exact Box function is NOT expressible by Harmonic Polylogs, one may introduce a
%

$
&

'
generalization of them: Generalized HPLs.
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 81

Automatized tools for this might be developed.

A scetch of the small mass expansion may be made as follows.
First the 1-dim. integral J1 .
The leading term comes from the first residue:

J1 = Residue[m_tˆz1 Gamma[-z1]ˆ3 Gamma[1 + z1]/Gamma[-2 z1], {z1, 0}]

= 2 Log[m_t]

WE get a logarithmic mass dependence.

The second integral: Start with z2 , first residue is:

I2 = Residue[ m_tˆz2 Gamma[-z2] Gamma[-1 - z1 - z2]ˆ2

Gamma[2 + z1 + z2]/Gamma[-2 - 2 z1 - 2 z2], {z2, 0}]

Γ[−1 − z1 ]2 Γ[2 + z1 ]
=−
Γ[−2 − 2z1 ]
The residue is independent of m2 /t.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 82

It has to be integrated over z1 yet, together with the terms which were independent of z2 :
2
2 Γ[−1 − z1 ] Γ[2 + z1 ]
Z
z1 +1
I2 ∼ dz1 r Γ[−z1 ]Γ[−2 − 2z1 ]Γ[1 + z1 ]
Γ[−2 − 2z1 ]
Sum over residues, close path to the left:
(−1)n r1−n  2 2

Residue[z1 = −n] = 2 + (−1 + n) π + (−1 + n) ln[r](2 + (−1 + n) ln[r])
2(−1 + n)3
1
Residue[z1 = −1] = (3π 2 ln[r] + ln[r]3 )
6
and finally:
∞
X
I2 ∼ Residue[z1 = −1] + Residue[z1 = −n]
n=2

The sum can be done also withoput using XSUMMER (here at least), e.g.
Inf
Xty (−1)n (2 + π 2 − 2nπ 2 + n2 π 2 )r1−n 1 2 
ln[r] = π ln(1 + 1/r) − 2Li3 (−1/r)
n=2
2(−1 + n)3 2

etc
%

$
&

'
Integration-by-parts identities
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 83

[Tkachov:1981wb,Chetyrkin:1981qh]

The Integration-by-parts identities relate different scalar Feynman integrals in d dimensions

algebraically.
They may be used to determine a list of basic so-called master integrals.
These masters have to be solved, the others are then - more or less easily - derived from them
by algebra.
First systematic algorithm realized in a computer code (unpublished) by
[Laporta:1996mq,Laporta:2000dc,Laporta:2001dd]

, the masters then were evaluated by difference equations.

Public codes with Laporta algorithm: Maple package AIR
[Anastasiou:2005cb]

and Fire
[Smirnov:2008iw]

A nice lecture, where also integration by parts, differential equations, HPLs and all that are
introduced, is
[Aglietti:2004vs]

Lagrange/1762–Gauss/1813–Green/1825–Ostrogadski/1831 integral theorem (or divergence

theorem)
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 84

∂
Z I
dd k [Fµ ] = [Fµ ] dS µ → 0
V ∂kµ S∞

where for a tadpole and self-energy e.g.:

kµ
FT,µ =
D1n1
akµ + bpµ
FSE,µ =
D1n1 D2n2

etc., and D1 = k 2 − m2 and D2 = (k − p)2 − m2 . For the tadpole one gets due to:
kµ 1 −2kµ
∂kµ = ∂kµ kµ + k µ n1 +1
D1n1 D1n1 D1
with ∂kµ kµ = d the relation for tadpoles with different indices n1 and n1 + 1:

(d − 2n1 ) T (n1 )
T (n1 + 1) =
2n1 m2
For more complicated Fyenman integrals one gets several equations and one may then hope to
formulate a simple basis.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 85

Regularisation by subtraction: V4l1m2

Simplest case was factorization as in V3l2m.
Systematic isolation of divergences in regions of phase space, and in each of them the
divergence is in only ONE variable: sector decomposition
Intermediate case:
It is a singularity in one variable only, but not sufficiently isolated.
Then one may try to perform a subtraction.

V4l1m1 V4l1m1d1 V4l1m1d2

The Feynman integral V4l1m2

V4l2m1 V4l2m2 V4l1m2
has a
massless UV-divergent subloop
V4l3m V4l3md

V4l4m V4l4md
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 86

e2ǫγE d D k1 d D k2
Z
V4l1m2 = − D .
π [k22 ][(k1 + k2 − p1 )2 ][k12 − 1][(k1 + p2 )2 ]
We will use a subtraction procedure in order to isolate the remaining UV singularity. The two
momentum integrations may be performed subsequently:
d D k2 2
D/2 Γ(1 − ǫ) Γ(ǫ) 1
Z
= iπ ,
[k22 ][(k2 + k1 − p1 )2 ] Γ(2 − 2ǫ) [(k1 − p1 )2 ]ǫ
d D k1 D/2 ǫ(1 + ǫ)Γ (2ǫ)
Z
= iπ Idiv ,
[(k1 − p1 )2 ]ǫ [k12 − 1][(k1 + p2 )2 ] Γ(2 + ǫ)
Z 1
dxdy x−1+ǫ (1 − x)1−2ǫ
Idiv = 2 − xys]2 )2ǫ
.
0 ([(1 − x)(1 − y)
The Feynman parameter integral Idiv has a singularity at x = 0 - like the QED vertex or SE -
and may be regulated by a subtraction:
Z 1 Z 1
dxx−1+ǫ (1 − x)1−2ǫ dy f (x, y)−2ǫ − f (0, y)−2ǫ + f (0, y)−2ǫ
 
Idiv =
0 0
Γ(ǫ)Γ(2 − 2ǫ)
= + Ireg ,
(1 − 4ǫ)Γ(2 − ǫ)
Z 1 1
f (x, y)−2ǫ − f (0, y)−2ǫ
Z
ǫ
dx (1 − x) x(1 − x)2
 
Ireg = dy ,
0 0 x
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 87

with

f (x, y) = (1 − x)(1 − y)2 − xys.

The remaining integrations in Ireg are regular and can be performed analytically or numerically
after the ǫ-expansion:
∞
Z 1 Z 1 " n #
n
 X
2 dy f (x, y) (−2ǫ) X
Ireg = dx(1 − x)eǫ ln[x/(1−x) ] ln lnn−k−1 f (x, y) lnk f (0, y) .
0 0 x f (0, y) n=1 n!
k=0

The first terms of the series expansion in ǫ for V4l1m2 are (see (1)):
1 5 19 3 − 13x 1−x  2 
V4l1m2 = + + − ζ 2 − ln (x) + 4Li 2 (x) + O(ǫ).
2ǫ2 2ǫ 2 2(1 + x) 2(1 + x)

In general, the situation is more involved.

%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 88

Sector decomposition
For Euclidean kinematics, the integrand for the multi-dimensional x-integrations is positive
semi-definite.
In numerical integrations, one has to separate the poles in (d − 4), and in doing so one has to
avoid overlapping singularities.
A method for that is sector decomposition.
There are quite a few recent papers on that, and also nice reviews are given
[Binoth:2000ps,Denner:2004iz,Bogner:2007cr,Heinrich:2008si,Smirnov:2008aw]

The intention is to separate singular regions in different variables from each other, as is nicely
demonstrated by an example borrowed from
[Heinrich:2008si]

:
1 1
1
Z Z
I = dx dy
0 0 x1+aǫ y bǫ [x + (1 − x)y]
1 Z 1 Z 1 Z 1
dx dt dy dt
Z
= + . (9)
0 x1+(a+b)ǫ 0 tbǫ [1 + (1 − x)t] 0 y
1+(a+b)ǫ
0 t
1+aǫ [1 + (1 − y)t]
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 89

The master integral V6l4m1

At several occasions, we used for cross checks the package sector decomposition
[Bogner:2007cr]

built on the C++ library GINAC

[Bauer:2000cp]

For that reason, the interface CSectors was written; it will be made publicly available soon.
The syntax is similar to that of AMBRE.
Example:
The program input for the evaluation of the integral V6l4m1 is simple; we choose
m = 1, s = −11, and the topology may be read from the arguments of propagator functions PR:
%

$
&

'
<< CSectors.m
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 90

Options[DoSectors]
SetOptions[DoSectors, TempFileDelete -> False, SetStrategy -> C]

n1 = n2 = n3 = n4 = n5 = n6 = n7 = 1;
m = 1; s = -11;
invariants = {p1ˆ2 -> mˆ2, p2ˆ2 -> mˆ2, p1 p2 -> (s - 2 mˆ2)/2};

DoSectors[{1},
{PR[k1,0,n1] PR[k2,0,n2] PR[k1+p1,m,n3]
PR[k1+k2+p1,m,n5] PR[k1+k2-p2,m,n6] PR[k2-p2,m,n7]},
{k2, k1}, invariants][-4, 2]
Here, the numerator is 1 (see the first argument {1} of DoSectors), and the output contains
the functions U2 and F2 :
Using strategy C
U = x3 x4+x3 x5+x4 x5+x3 x6+x5 x6+x2 (x3+x4+x6)+x1 (x2+x4+x5+x6)

F = x1 x4ˆ2+13 x1 x4 x5+x4ˆ2 x5+x1 x5ˆ2+x4 x5ˆ2+13 x1 x4 x6

+2 x1 x5 x6+13 x4 x5 x6+x5ˆ2 x6+x1 x6ˆ2+x5 x6ˆ2+x3ˆ2 (x4+x5+x6)
%

$
&

'
+x2(x3ˆ2+x4ˆ2+13 x4 x6+x6ˆ2+x3 (2 x4+13 x6))+x3 (x4ˆ2+(x5+x6)ˆ2
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 91

+x4 (2x5+13 x6))

Notice the presence of a U -function and the complexity of the F -function (compared to U = 1
and f1 and f2 in the loop-by-loop MB-approach) due to the non-sequential, direct performance
of both momentum integrals at once. Both U and F are evidently positive semi-definite. The
numerical result for the Feynman integral is:
1
V6l4m1(−s)2ǫ = −0.052210 − 0.17004 + 0.24634 ǫ + 4.8773 ǫ2 + O(ǫ3 ). (10)
ǫ
The numbers may be compared to (13). We obtained a third numerical result, also by sector
decomposition, with the Mathematica package FIESTA
[Smirnov:2008py]

1
V6l4m1(−s)2ǫ = −0.052208 − 0.17002 + 0.24622 ǫ + 4.8746 ǫ2 + O(ǫ3 ). (11)
ǫ
Most accurate result: obtained with an analytical representation based on harmonic
polylogarithmic functions obtained by solving a system of differential equations
[Gluza, TR, unpubl.;Remiddi:1999ew,Maitre:2005uu]

1
V6l4m1(−s)2ǫ = −0.0522082 − 0.170013 + 0.246253 ǫ + 4.87500 ǫ2 + O(ǫ3 ). (12)
ǫ
All displayed digits are accurate here.
%

$
&

'
V6l4m: Compare to MB-integrals
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 92

In a loop-by-loop approach, after the first momentum integration one gets here U = 1 and a
first F -function (??), which depends yet on one internal momentum k1 :
f1 = mˆ2 [X[2]+X[3]+X[4]]ˆ2 - s X[2]X[4] - PR[k1+p1,m] X[1]X[2]
- PR[k1+p1+p2,0] X[2]X[3] - PR[k1-p2,m] X[1]X[4]
- PR[k1,0] X[3]X[4] ,
leading to a 7-dimensional MB-representation; after the second momentum integration, one has:
f2 = mˆ2 [X[2]+X[3]]ˆ2 - s X[2]X[3] - s X[1]X[4] - 2s X[3]X[4],

leading to another 4-dimensional integral.

After several applications of Barnes’ first lemma, an 8-dimensional integral has to be treated.
We made no attempt here to simplify the situation by any of the numerous tricks and reformulations
etc. known to experts.
The package AMBRE.m is designed for a semi-automatic derivation of Mellin-Barnes (MB)
representations for Feynman diagrams; for details and examples of use see the webpage http:
//prac.us.edu.pl/∼gluza/ambre/. The package is also available from
http://projects.hepforge.org/mbtools/.
Version 1.0 is described in
[Gluza:2007rt]
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 93

, the last released version is 1.2. We are releasing now version 2.0, which allows to construct
MB-representations for two-loop tensor integrals.
The package is yet restricted to the so-called loop-by-loop approach, which yields compact
representations, but is known to potentially fail for non-planar topologies with several scales. An
instructive example has been discussed in
[Czakon:2007wk]

.
For one-scale problems, one may safely apply AMBRE.m to non-planar diagrams. For our
example V6l4m1, one gets e.g. with the 8-dimensional MB-representation scetched above the
following numerical output after running also MB.m
[Czakon:2005rk]

(see also the webpage http://projects.

hepforge.org/mbtools/),
at s = −11:
1
V6l4m (−s)2ǫ = −0.0522082 − 0.17002 + 0.25606 ǫ + 4.67 ǫ2 + O(ǫ3 ). (13)
ǫ
%

$
&

'
Compare this to an MB-integral for V6l0m
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 94

Was solved first in

[Gonsalves:1983nq]

fupc1 = mˆ2 FX[X[3] + X[4]]ˆ2 - PR[k1 + p1, 0] X[1] X[2] -

PR[k1 + p1 + p2, m] X[2] X[3] - PR[k1 - p2, m] X[1] X[4] -
S X[2] X[4] - PR[k1, 0] X[3] X[4]

fupc2 = -s X[2] X[3]-s X[1] X[4]-2 s X[3] X[4]

The function is a scale factor times a pure number
 
A 4 A 3 A 2 A 1
V 6l0m = const(−s)−2−2ǫ 4 + 3 + 2 + + A0 + . . .
ǫ ǫ ǫ ǫ

V6l0m = (-s)ˆ{-2 - 2 eps}

*
Gamma[-z1] Gamma[-1 - eps - z1 - z2] Gamma[
1 + z1 + z3] Gamma[-1 - eps - z1 - z3 - z4] Gamma[-z4] Gamma[
1 + z1 + z2 + z4] Gamma[
2 + eps + z1 + z2 + z3 + z4] Gamma[-2 - 2 eps - z2 - z4 -
z5] Gamma[-z5] Gamma[
1 - z1 + z5] Gamma[-2 - 2 eps - z3 - z4 - z6] Gamma[-z6] Gamma[
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 95

3 + eps + z1 + z2 + z3 + z4 + z6] Gamma[

2 + 2 eps + z4 + z5 + z6])/(Gamma[-2 eps] Gamma[
1 - z1] Gamma[-3 eps - z4] Gamma[3 + eps + z1 + z2 + z3 + z4])

integrals = MBcontinue[(-s)ˆ(2 + 2 eps) fin, eps -> 0, rules];

26 integral(s) found
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 96

From 1/ǫ4 until ǫ0 : at most 4-dimensional

AMBRE output:

{-47.8911 + 1/epsˆ4 - 9.8696/epsˆ2 - 33.2569/eps}

-9.869604401660688 -> -Piˆ2

-33.25689147976733 -> -(83/3*Zeta[3])

-47.89108304541082 -> (-59 Piˆ4/120)

Try PSQL for a systematic study

%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 97

Tensor integrals
At the LHC, we need also massive 5-point and 6-point functions at least, for NLO corrections.
In general, the treatment of tensor integrals is a non-trivial task.
• One might think that all tensor numerators may be reduced to (even simpler) scalar
integrands. Important notice: For L > 1, this is not true, we have irreducible numerators,
see next slide.
• For many problems, it i preferrable to evaluate the tensors without knowing the scalar
products. The reasons are different.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 98

Simple tensor integrals

1 dd k k µ
Z
B µ
≡ pµ1 B1 =
(iπ d/2 ) D1 D2
1 dd k k µ k ν
Z
B µν
≡ pµ1 pν1 B22 +g µν
B20 =
(iπ d/2 ) D1 D2
and B1 and B22 , B20 have to be determined.
%

$
&

'
Reducible numerators
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 99

Some numerators are reducible – one may divide them out against the denominators:
2kp1 [(k + p1 )2 − m22 ] − [k 2 − m21 ] + (−m21 + m22 − p21 )]
≡
D1 [(k + p1 )2 − m22 ] . . . DN D1 [(k + p1 )2 − m22 ] . . . DN
1 1 −m21 + m22 − p21
= − +
D 1 D 3 . . . DN D 2 D 3 . . . DN D 1 D 2 D 3 . . . DN

This way one derives:

1 p1 k
Z
pµ1 B µ = p2 B1 = d d
k
(iπ d/2 ) D1 D2
2 2 2
 
1 1 1 1 −m + m − p
Z
d 1 2 1
= d k − +
(iπ d/2 ) 2 D1 D2 D1 D2
and finally:
1  2 2 2 2

B1 = A0 (m1 ) − A0 (m2 ) + (−m1 + m2 − p1 )B0 (m1 , m2 , p )
2p21
Known to everybody: The Passarino-Veltman reduction scheme for 1-loop tensors
worked out in
[Passarino:1979jh]

until 4-point functions.

%

$
&

'
Irreducible numerators
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 100

For a two-loop QED box diagram, it is e.g. L = 2, E = 4, and we have as potential

simplest numerators:
k12 , k22 , k1 k2 and 2(E − 1) products k1 pe , k2 pe
compared to N internal lines, N = 5, 6, 7. This gives
I = L + L(L − 1)/2 + L(E − 1) − N irreducible numerators
of this type. Here:
I(N ) = 9 − N = 4, 3, 2
This observation is of practical importance:
Imagine you search for potential masters. Then you may take into the list of masters at
most (here e.g.) I(5) = 4, or I(6) = 3, or I(7) = 2 such integrals.
Which momenta combinations are irreducible is dependent on the choice of momenta
flows.
Message:
When evaluating all Feynman integrals by Mellin-Barnes-integrals, one should also learn
to handle numerator integrals
. . . and it is - in some cases - not too complicated compared to scalar ones
The one-loop case: L = 1, E = N , so
I(N ) = 1 + (E − 1) − N = 0
irreducible numerators
%

$
&

'
The vector integral differs by some numerator ki pe and thus there is a single shift in the integrand
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 101

k → k̄ + U (x)−1 M̃ Q
the d k̄ k̄/(k̄ 2 + µ2 ) → 0, and no further changes:
R d

N N
! " #
d
` ´ Z 1 Y Nν −d(L+1)/2−1
Γ Nν − L ν −1
X U (x) X
G(k1α ) = (−1)Nν 2
dxj xj j δ 1− xi Nν −dL/2
M̃1l Ql ,
Γ(ν1 ) . . . Γ(νN ) 0 j=1 i=1 F (x) l α

Here also a tensor integral:

N N
!
d
U (x)Nν −2−d(L+1)/2
` ´ Z
Γ Nν − 1 Y
Nν 2L ν −1
dxj xj j δ
X
G(k1α k2β ) = (−1) 1− xi
Γ(ν1 ) . . . Γ(νN ) 0 j=1 i=1 F (x)Nν −dL/2
" −1 + −1
#
Γ Nν − d
` ´
X
2 L − 1 g αβ (V ) (V )
× [M̃1l Ql ]α [M̃2l Ql ]β − U (x)F (x) 1l 2l
.
Γ Nν − d
` ´
l 2 L 2 α l

The 1-loop case will be used in the following L times for a sequential treatment of an L-loop
xj Dj = k 2 − 2Qk + J and F (x) = Q2 − J ):
P
integral (remember
d

Z 1 N N
!
Γ Nν − 2 Y ν −1
X [1, Qpe ]
G([1, kpe ]) = (−1)Nν dxj xj j δ 1 − xi
Γ(ν1 ) . . . Γ(νN ) 0 j=1 i=1
F (x)Nν −d/2
%

$
&

'
.
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 102

Another nice box with numerator, B5l3m(pe .k1 )

We used it for the determination if the small mass expansion.

m4ǫ (−1)a12345 e2ǫγE +i∞ +i∞ +i∞ +i∞
Z Z Z Z
B5l3m(pe · k1 ) = Q5 dα dβ dγ dδ
j=1 Γ[ai ]Γ[5 − 2ǫ − a123 ](2πi)4 −i∞ −i∞ −i∞ −i∞
(4−2 ǫ)−a12345 −α−β−δ δ
(−s) (−t)
Γ[−4 + 2 ǫ + a12345 + α + β + δ] Γ[−α] Γ[−β] Γ[−δ]
Γ[6 − 3 ǫ − a12345 − α] Γ[7 − 3 ǫ − a12345 − α] Γ[5 − 2 ǫ − a123 ] Γ[4 − 2 ǫ − a1123 − 2 α − γ] Γ[5 − 2 ǫ − a1123 − 2 α
Γ[2 − ǫ − a13 − α − γ] Γ[4 − 2 ǫ − a12345 − α − β − δ − γ]

(pe · p3 ) Γ[1 + a4 + δ] Γ[6 − 3 ǫ −
Γ[8 − 4 ǫ − a112233445 − 2 α − 2 β − 2 δ − γ] Γ[9 − 4 ǫ − a112233445 − 2 α − 2 β − 2 δ − γ]
Γ[4 − 2 ǫ − a1234 − α − β − δ] Γ[3 − ǫ − a12 − α] Γ[8 − 4 ǫ − a112233445 − 2 α − 2 δ − γ]Γ[9 − 4 ǫ − a112233445 − 2 α − 2 β − 2
»
Γ[5 − 2 ǫ − a1123 − γ] Γ[4 − 2 ǫ − a1123 − 2 α − γ] Γ[a1 + γ] Γ[−2 + ǫ + a123 + α + δ + γ] + Γ[a4 + δ] −(pe · p1 ) Γ[7 − 3 ǫ − a

Γ[4 − 2 ǫ − a1234 − α − β − δ]Γ[8 − 4 ǫ − a112233445 − 2 α − 2 δ − γ]Γ[9 − 4 ǫ − a112233445 − 2 α − 2 β − 2 δ − γ]

»
Γ[3 − ǫ − a12 − α] Γ[5 − 2 ǫ − a1123 − γ]Γ[4 − 2 ǫ − a1123 − 2 α − γ] Γ[a1 + γ] + Γ[2 − ǫ − a12 − α] Γ[4 − 2 ǫ − a1123 − γ]
–
Γ[5 − 2 ǫ − a1123 − 2 α − γ] Γ[1 + a1 + γ] Γ[−2 + ǫ + a123 + α + δ + γ] + Γ[6 − 3 ǫ − a12345 − α] Γ[3 − ǫ − a12 − α]
»
Γ[5 − 2 ǫ − a1123 − γ] Γ[4 − 2 ǫ − a1123 − 2 α − γ]Γ[a1 + γ] ((pe · (p1 + p2 )) Γ[5 − 2 ǫ − a1234 − α − β − δ]Γ[9 − 4 ǫ − a112233

Γ[8 − 4 ǫ − a112233445 − 2 α − 2 β − 2 δ − γ]Γ[−2 + ǫ + a123 + α + δ + γ] + (pe · p1 ) Γ[4 − 2 ǫ − a1234 − α − β − δ]

–ff
Γ[8 − 4 ǫ − a112233445 − 2 α − 2 δ − γ]Γ[9 − 4 ǫ − a112233445 − 2 α − 2 β − 2 δ − γ]Γ[−1 + ǫ + a123 + α + δ + γ]
%

$
&

'
Interlude: Tensor integrals, a bit background
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 103

Introducing α parameter representations for the propagators,

Z ∞
1 i−ν
2 2 ν
= dααν−1 exp iα(k 2 − m2 + iǫ)
(k − m + iǫ) Γ(ν) 0
one may write e.g. for the L-loop integral with I internal lines and E external lines
Z d
(d) d k1 . . . dd kL
G ≡ exp i(a1 k1 + . . . + aL kL )
cν11 . . . cνI I
I I
" #
 π dL/2 Y −νi Z ∞ νi −1
i dαi αi Q(α, a1 , . . . aL ) X
= i2 d/2
exp i − α i (m 2
i − iǫ) ,
i i=1
Γ(ν) 0 [D(α)] D(α) i=1

with the following notations:

ci = qi2 − m2i + iǫ,
L E L
qiµ ωil klµ ωil klµ + eµi ,
X X X
= + ηie pµe = i = 1, . . . I,
l=1 e=1 l=1

and the ωil and ηie are equal to +1, −1, 0, depending on the topology and on the
distribution of internal momenta. They are incidence matrices of a diagram.
This formula represents a scalar integral for al = 0, and for ai 6= 0 it is an auxiliary
representation for the treatment of tensor integrals.
%

$
&

'
The Q and D forms may be determined to be as follows for the 1-loop case:
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 104

D = A,
a 2
Q(α, a) = DF − +E ,
2
with
I
X
A = ωi2 αi ,
i=1
I
ωi eµi αi ,
X
Eµ =
i=1
I
2
(eµi ) αi .
X
F =
i=1

Explicit expressions are given for D and the a-dependent part of Q for the convention
used in Figure ??. We may add the a-independent part Q(a = 0) of the Q form here:
X
D = αj ,
1
αj pµj − a2 + Q(a = 0),
X
Q = aµ
4
X  X  X 2
2 2
Q(a = 0) = DF − E = αj αk pk − αk pk .
%

$
&

'
The Q and D forms may be determined to be as follows for the 2-loop case
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 105

D = A1 A2 − M 2 ,
Q(α, a1 , a2 ) = DF − A2 E12 − A1 E22 + 2M E1 E2 + Q(a1 , a2 ),

and
I
X
2
A1 = ωi1 αi ,
i=1
I
X
2
A2 = ωi2 αi ,
i=1
I
X
M = ωi1 ωi2 αi ,
i=1
I
Elµ ωil eµi αi ,
X
=
i=1
I
2
(eµi ) αi .
X
F =
i=1

The Q(a1 , a2 ) is given in Section ??.

%

$
&

'
General tensor reduction to scalars, L = 1
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 106

The idea for 1-loop integrals and generalization to n-loops; I follow Tausk et al.
[Davydychev:1991va,Tarasov:1996br,Tarasov:1997kx]

d d k k µ1 . . . kµR
Z
Jµ(N )
(d; ν1 , . . . , νN ) =
1 ...µR
iπ d/2 D1ν1 . . . DNνN

R
X 1 λ λ
= (−1) (− ) [g] [q1 ]κ1 . . . [qN ]κN µ1 ...µR
(ν1 )κ1 . . . (νN )κN
2
λ,κ1 ,...,κN

J (N ) (d + 2(R − λ); ν1 + κ1 , . . . , νN + κN )

where (ν)κ = Γ(ν+κ)

Γ(ν) are Pochhammer symbols and the sum runs over non-negative
integers such that 2λ + κ1 + . . . + κN = R. The next step is to use recurrence relations
to reduce the scalar coefficients J (N ) appearing in the decomposition to a set of master
integrals.
It is useful to introduce a notation for certain determinants that occur in the recurrence
relations and their solutions. First, the determinant of an (N + 1) × (N + 1) matrix,
known as the modified Cayley determinant
[Melrose:1965kb]
%

$
&

'
,
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 107

0 1 1 ... 1
1 Y11 Y12 ... Y1N
()N ≡ 1 Y12 Y22 ... Y2N , (14)
.. .. .. .. ..
. . . . .
1 Y1N Y2N ... YN N
with coefficients
Yij = −(qi − qj )2 + m2i + m2j , (i, j = 1 . . . N ) . (15)
Although the masses of the propagators appear in the coefficients Yij , the determinant
()N does not depend on them and it is actually proportional to the Gram determinant
of the external momenta of the N -point function in eq. (14). All other determinants we
need are signed minors of ()N , constructed by deleting m rows and m columns from
()N , and multiplying with a sign factor. They will be denoted by
 
j j2 · · · jm
 1
P
(j +k )
 ≡ (−1) l l l
k1 k2 · · · km
N

rows j1 · · · jm deleted
sgn{j} sgn{k} , (16)
columns k1 · · · km deleted
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 108

where sgn{j} and sgn{k} are the signs of permutations that sort the deleted rows
j1 · · · jm and columns k1 · · · km into ascending order.
Combining integration by parts identities with relations connecting integrals in different
space-time dimensions
[Tarasov:1996br]

, one obtains the following basic recurrence relations

[Fleischer:1999hq]

:
n  
"   #
j X j
()N νj j+ J (N ) (d + 2) = − + k− J (N ) (d) ,
0 N k N
k=1
n n  
"  #
X 0 X 0
(d − νi + 1) ()N J (N ) (d + 2) = − k− J (N ) (d) ,
i=1
0 N k N
k=1
n   " n
  #
0 X 0j X
νj j+ J (N ) (d) = d− νi (k− i+ + 1) J (N ) (d) .
0 N 0k N i=1
k=1

where the operator j± acts by shifting the index νj by ±1.

%

$
&

'
Recurrence relations: An application
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 109

1 1
d−4 = − 2ǫ

d − 2 A0 (m2 ) 2d − 3
 
1
C0 (m, 0, m; m2 , m2 , s) = + B0 (m, m; s)
s − 4m2 d − 4 m2 d−4
 
−1 1 − ǫ 5 − 4ǫ
V 3l2m = T 1l1m + SE2l2m (17)
s−4 ǫ 2ǫ

This relates UV-divergent quantities with one which is IR-divergent!

Derivation:
I do not know a simple one.
With use of recurrence relations published in
[Tarasov:1996br,Fleischer:1999hq]

it may be done. See

[Fleischer:2006ht]

.
0


Because the am determinant vanishes, 0 3 = 0, one of the recurrence relations
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 110

expresses the vertex by a self-energy with 2 massive lines in d − 2 dimensions:

3  
"  #
(d) 0 X 0 (d−2) (d−2)
(d − 4) ()3 I3 = − k− I3 = I2 (m, m)
0 3 k 3
k=1

The dimension of the 2-point function may be raised by another recurrence relation,
but now one line will be dotted:
(d−2)
I2 (m, m) = −21+ I2d (m, m), (18)
The dot may be removed, in general form, by expressing a dotted function by itself, but
undotted, plus extra terms with one line less, but a dotted one:
„ 0« " „0« # 2 „ 01 «
+ (d) (d) X − + (d)
1 I = (3 − d) I − k i I , (19)
2 2 2
0 2 1 2 i,k i6=k 0k 2

Here the latter is a dotted tadpole, and that is proportional to the tadpole itself, and
we get the relation needed. (In general, iterations do the job here.)
Direct evaluation shows:
1 A0 (m)
A0 (m)dot = (d − 2)
2 m2
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 111

• hexagon package in Mathematica and (yet) unpublished Fortran code

The packages evaluate

• 6-point functions up to tensor rank 4

• 5-point functions up to tensor rank 3

6-point functions up to tensor rank 4

[Fleischer:1999hq,Diakonidis:0901.4455,Diakonidis:2008ij]
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 112

One reduction: 5-point tensor, rank three

4 4
λ]
I5µ ν λ qiµ qjν
X X
= qkλ Eijk + g [µν qk E00k ,
i,j,k=1 k=1

where
λ]
g [µν qk = g µν qkλ + g µλ qkν + g νλ qkµ , (20)
with scalar coefficients defined by
5
4,s s
X
Eijk = Sijk I4
s=1
5 5
3,st st 2,stu stu
X X
+ Sijk I3 + Sijk I2 , (21)
s,t=1 s,t,u=1
%

$
&

'
with
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 113

4,s 1
Sijk =
×
0 s 2


3 0 s 5
  5         
0s 0s 0s is 0s
− +
0k 5 is 5 js 5 js 5 0s 5
          
0s 0i 0s 0j 0s
+ +
0s 5 sk 5 js 5 sk 5 is 5
+ (i ↔ k) + (j ↔ k) , (22)
     
3,st 1 0s ts 0s
Sijk =
s
2
0
3 0 5 s 5 0k
5 is 5 js 5
s
   
0st
  #
is ts ts
+ + s 5st
ist 5
js 5 0s 5 st 5
js 5
         
0i 0s 0j 0s ts
− +
sk js sk is 0s 5
 5   5   5  5
0i ts 0j ts
− +
sk 5 js 5 sk 5 is 5
s

0st
)
s 5 0st 5
× + (i ↔ k) + (j ↔ k) , (23)
2 st


st 5
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 114

2,stu 1
Sijk =−
×
0 s
3 0 5 s 5 st



st 5
         
0s ts ust 1 0j ust
−
0k 5 js 5 ist 5 2 sk 5 ist 5
      
0i ust ts
+
sk 5 jst 5 0s 5
+ (i ↔ k) + (j ↔ k) , (24)
%

$
&

'
and
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 115

5
(
1 X 1
E00j = 0 −
s 2


6 0 5


s=1 s 5
          
s 0s s 0s 0s
× 3 − I4s
0 5 js 5 j 5 0s 5 0s 5
5
X 1
+ s 2


s,t=1 s 5
"       ts
2 #  
s 0s s 0s 5 ts st
× 3 − I3
j 5 st


0 5 js 5 st 5
0s 5
5 ust
 
  )
X s 0st 5 ts
− s

st

I2stu . (25)
s,t,u=1
j 5 s 5 st 5 0s 5

The decomposition in eq. (20) is equivalent to the one found in ref.

[Denner:2002ii]

, where the coefficients Eijk and E00j are expressed in terms of tensor 4-point functions. Here,
instead, they are completely reduced to a basis of scalar master integrals consisting of boxes I4s ,
vertices I3st , and 2-point functions I2stu obtained by removing lines s, s and t, or s, t and u from
the pentagon.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 116

Summary

• We have introduced to the representation of L-loop N -point Feynman integrals of general

type
• The determination of the ǫ-poles is generally solved
• The remaining problem is the evaluation of the multi-dimensional, finite MB-Integrals

• This is unsolved in the general case, . . . so you have something to do if you like to . . .

Problem: Determine the small mass limit of B5l2m2 or of any other of the 2-loop boxes for
Bhabha scattering.
Prof. Gluza may check your solution.
He leaves soon.
%

$
&

'
On-shell example: B4l2m, the 1-loop on-shell box
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 117

den = (x4 d4 + x5 d5 + x6 d6 + x7 d7 // Expand) /. kinBha /. mˆ2 -> 1 // Expand

Q = -Coefficient[den, k]/2 // Simplify

= p3 x4 + p2 x5 - p1 (x4 + x6)

M = Coefficient[den, kˆ2] // Simplify

= x4 + x5 + x6 + x7 -> 1

J = den /. k -> 0 // Simplify

= t x4

F[x] = (Qˆ2 - J M // Expand) /. kinBha /. mˆ2 -> 1 /. u -> -s - t + 4 // Expand

= (x5+x6)ˆ2 + (-s)x5x6 + (-t)x4x7

B4l2ma = mb[(x5+x6)ˆ2, -tx7x4 - sx5x6, nu, ga]

B4l2mb = B4l2ma /. (-sx5x6 - tx4x7)ˆ(-ga - nu) ->

mb[(-s)x5x6, (-t)x7x4, nu+ga, de]
/.((-s)x5x6)ˆde_ -> (-s)ˆde x5ˆde x6ˆde
/.((x56ˆ2)ˆga -> (x5 + x6)ˆ(2ga)
%

$
&

'
=
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 118

(inv2piIˆ2(-s)ˆde x5ˆde x6ˆde ((x5 + x6)ˆ(2ga)((-t)x4x7)ˆ(-de-ga-nu)

Gamma[-de] Gamma[-ga] Gamma[de + ga + nu] /Gamma[nu]

B4l2mc = B4l2mb /. (x5 + x6)ˆ(2ga) ->

mb[x5, x6, -2ga, si]
/. ((-t)x4x7)ˆsi_ -> (-t)ˆsi x4ˆsi x7ˆsi // ExpandAll
=
1/(Gamma[-2ga] Gamma[nu])
inv2piIˆ3 (-s)ˆde (-t)ˆ(-de - ga - nu)
x4ˆ(-de - ga - nu) x5ˆ(de + si) x6ˆ(de + 2 ga - si) x7ˆ(-de - ga - nu)
Gamma[-de] Gamma[-ga] Gamma[ de + ga + nu] Gamma[-si] Gamma[-2ga + si]

B4l2md = xfactor4[a4, x4, a5, x5, a6, x6, a7, x7] B4l2mc
=
.... (-s)ˆde (-t)ˆ(-de-ga-nu)
x4ˆ(-1+a4-de-ga-nu) x5ˆ(-1+a5+de+si) x6ˆ(-1+a6+de+2ga-si) x7ˆ(-1+a7- de-ga-nu)

B4l2me =
B4l2md /.
x4ˆB4_ x5ˆB5_ x6ˆB6_ x7ˆB7_ -> xint4[x4ˆB4 x5ˆB5 x6ˆB6 x7ˆB7]
= ...
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 119

B4l2mf = B4l2me /.
Gamma[a6 + de + 2 ga - si]Gamma[-si]Gamma[ a5 + de + si] Gamma[-2 ga + si]
-> barne1[si, a5 + de, -2 ga, a6 + de + 2 ga, 0]

This finishes the evaluation of the MB-representation for B4l2m.

%

$
' $

Package: AMBRE.m (K. Kajda, with support by J. Gluza and TR)

& %
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 120
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 121

Example beyond Harmonic Polylogs: QED Box B4l2m

[Fleischer:2006ht]

F[x] = (x5+x6)ˆ2 + (-s)x5x6 + (-t)x4x7

B4l2m, the 1-loop QED box, with two photons in the s-channel; the Mellin-Barnes
representation reads for finite ǫ:
Z +i∞ Z +i∞
eǫγE 1
B4l2m = Box(t, s) = dz1 dz2 (26)
Γ[−2ǫ](−t)(2+ǫ) (2πi)2 −i∞ −i∞
(−s)z1 (m2 )z2 2
Γ[2 + ǫ + z 1 + z 2 ]Γ [1 + z1 ]Γ[−z1 ]Γ[−z2 ]
(−t)z1 +z2
Γ[−2 − 2ǫ − 2z1 ]
Γ2 [−1 − ǫ − z1 − z2 ]
Γ[−2 − 2ǫ − 2z1 − 2z2 ]

Mathematica package MB used for analytical expansion ǫ → 0:

[Czakon:2005rk]
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 122

„ «
1 1ˆ 2 ˜
B4l2m = − J1 + ln(−s)J1 + ǫ ζ(2) − ln (−s) J1 − 2J2 . (27)
ǫ 2
with J1 being also the divergent part of the vertex function C0 (t; m, 0, m)/s = V3l2m/s (as is
well-known):
Z − 1 +i∞ „ 2 «z1 3
eǫγE 1 2 m Γ [−z1 ]Γ[1 + z1 ] 1 2y
J1 = dz1 = ln(y) (28)
st 2πi − 1 −i∞ −t Γ[−2z1 ] m2 s 1 − y 2
2

with
p
1 − 4m2 /t − 1)
y= p
( 1 − 4m2 /t + 1)
The J2 is more complicated:
−3 +i∞
eǫγE
Z
1 4
“ s ”z1
J2 = 2 dz1 Γ[−z1 ]Γ[−2(1 + z1 )]Γ2 [1 + z1 ] (29)
t (2πi)2 −3 −i∞ t
4

−1 +i∞
m2 Γ2 [−1 − z1 − z2 ]
Z „ «
2
z2
× dz2 − Γ[−z2 ] Γ[2 + z1 + z2 ].
−1 −i∞ t Γ[−2(1 + z1 + z2 )]
2
%

$
&

'
The expansion of B4l2m at small m2 and fixed value of t
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 123

With
−m2
mt = , (30)
t
s
r = , (31)
t
Look, under the integral, at (−m2 /t)z2 ,
and close the path to the right.
Seek the residua from the poles of Γ-functions with the smallest powers in m2 and sum the
resulting series.
we have obtained a compact answer for J2 with the additional aid of XSUMMER
[Moch:2005uc]

The box contribution of order ǫ in this limit becomes:

1n 2
2
B4l2m[t, s, m ; +1] = 4ζ3 − 9ζ2 ln(mt ) + ln3 (mt ) + 6ζ2 ln(r) − ln2 (mt )ln(r) (32)
st 3
1
+ ln3 (r) − 6ζ2 ln(1 + r) + 2ln(−r)ln(r)ln(1 + r) − ln2 (r)ln(1 + r)
3 o
+2ln(r)Li2 (1 + r) + 2Li3 (−r) + O(mt ).

Remark:
The exact Box function is NOT expressible by Harmonic Polylogs, one may introduce a
%

$
&

'
generalization of them: Generalized HPLs.
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 124

Automatized tools for this might be developed.

A scetch of the small mass expansion may be made as follows.
First the 1-dim. integral J1 .
The leading term comes from the first residue:

J1 = Residue[m_tˆz1 Gamma[-z1]ˆ3 Gamma[1 + z1]/Gamma[-2 z1], {z1, 0}]

= 2 Log[m_t]

WE get a logarithmic mass dependence.

The second integral: Start with z2 , first residue is:

I2 = Residue[ m_tˆz2 Gamma[-z2] Gamma[-1 - z1 - z2]ˆ2

Gamma[2 + z1 + z2]/Gamma[-2 - 2 z1 - 2 z2], {z2, 0}]

Γ[−1 − z1 ]2 Γ[2 + z1 ]
=−
Γ[−2 − 2z1 ]
The residue is independent of m2 /t.
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 125

It has to be integrated over z1 yet, together with the terms which were independent of z2 :
2
2 Γ[−1 − z1 ] Γ[2 + z1 ]
Z
z1 +1
I2 ∼ dz1 r Γ[−z1 ]Γ[−2 − 2z1 ]Γ[1 + z1 ]
Γ[−2 − 2z1 ]
Sum over residues, close path to the left:
(−1)n r1−n  2 2

Residue[z1 = −n] = 2 + (−1 + n) π + (−1 + n) ln[r](2 + (−1 + n) ln[r])
2(−1 + n)3
1
Residue[z1 = −1] = (3π 2 ln[r] + ln[r]3 )
6
and finally:
∞
X
I2 ∼ Residue[z1 = −1] + Residue[z1 = −n]
n=2

The sum can be done also withoput using XSUMMER (here at least), e.g.
Inf
Xty (−1)n (2 + π 2 − 2nπ 2 + n2 π 2 )r1−n 1 2 
ln[r] = π ln(1 + 1/r) − 2Li3 (−1/r)
n=2
2(−1 + n)3 2

etc
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 126

Some routines in mathematica which may be used:

(* Barnes’ first lemma: \int d(si) Gamma(si1p+si)Gamma(si2p+si)Gamma(si1m-si)Gamma(si2m-si)
with 1/inv2piI = 2 Pi I *)

barne1[si_, si1p_, si2p_, si1m_, si2m_] :=

1/inv2piI Gamma[si1p + si1m] Gamma[si1p + si2m] Gamma[
si2p + si1m] Gamma[si2p + si2m] /Gamma[si1p + si2p + si1m + si2m]

(* Mellin-Barnes integral: (A+B)ˆ(-nu) = 1/(2 Pi I) \int d(si) aˆsi bˆ(-nu - si)

Gamma[-si]Gamma[nu+si]/Gamma[nu] *)

mb[a_, b_, nu_, si_]:=inv2piI aˆsi bˆ(-nu-si)Gamma[-si]Gamma[nu+si]/Gamma[nu]

(* After the k-integration, the integrand for \int\prod(dxi xiˆ(ai - 1))\delta(1-\sum xi)
will be (L=1 loop) : xfactorn Fˆ(-nu) Q(xi).pe with nu = a1 + .. + an - d/2 *)

xfactor3[a1_, x1_, a2_, x2_, a3_, x3_] :=

I Piˆ(d/2) (-1)ˆ(a1 + a2 + a3) x1ˆ(a1 - 1) x2ˆ(a2 - 1)x3ˆ(a3 - 1)Gamma[
a1 + a2 + a3 - d/2]/(Gamma[a1]Gamma[a2]Gamma[a3])

(* xinti - the i-dimensional x - integration over Feynman parameters /16 06 2005 *)

xint3[x1_ˆ(a1_) x2_ˆ(a2_) x3_ˆ(a3_) ] :=

Gamma[a1 + 1] Gamma[a2 + 1] Gamma[a3 + 1] / Gamma[a1 + a2 + a3 + 3]
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 127

quote also:
[Argeri:2007up]
[Smirnov:2008py]
[Ellis:2007qk]
[Binoth:2008uq]
[Diakonidis:2008ij]
[Bogner:2007cr]
[Heinrich:2008si]
[Fleischer:2007ph]
[Diakonidis:2008dt]

References
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 128

Feynman integrals: scalar and tensor integrals, shrinked and

dotted ones
Just to mention what kind of integrals may appear:

• diagrams with numerators:

Γν (kiν −pνn )−mn
tensors arise from internal fermion lines: dd ki · · · (k
R
2
i −pn ) −m
2 ···
n

• diagrams with shrinked and/or with dotted lines: a sample relation see next page
1 1
• relations to simpler diagrams may shift the complexity: d−4 = − 2ǫ

d − 2 A0 (m2 ) 2d − 3
 
1
C0 (m, 0, m; m2 , m2 , s) = + B0 (m, m; s)
s − 4m2 d − 4 m2 d−4
 
−1 1 − ǫ 5 − 4ǫ
V 3l2m = T 1l1m + SE2l2m (33)
s−4 ǫ 2ǫ
%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 129

e2ǫΓE dD k1 dD k2 (k1 k2 )−d

Z
SE3l2m(a, b, c, d) = − D .
π [(k1 + k2 − p)2 − m2 ]b [k12 ]a [k22 − m2 ]c

SE3l2m = SE3l2m(1, 1, 1, 0)
SE3l2md = SE3l2m(1, 1, 2, 0)
SE3l2mN = SE3l2m(1, 1, 1, −1)

−(1 + s) + ǫ(2 + s) 2(1 − ǫ) `

T1l1m2 + 3 SE3l2mN ,
´
SE3l2md = SE3l2m +
s−4 s−4
1 + x2 1 2
„ «
1 1 1 − ζ2 1+x
= + − + ln(x) + ln (x) + O(ǫ)
2ǫ2 2ǫ 2 1−x (1 − x)2 2

SE3l2m SE3l2md SE3l0m

SE3l3m SE5l3m SE3l1m

Figure 1: The six two-loop 2-point masters of Bhabha scattering.

%

$
&

'
J.Gluza and T. Riemann – 16-20 Feb 2009 – RECAPP, Allahabad, India 130

More legs, more loops

Seek methods for an (approximated) analytical or numerical evaluation of more involved
diagrams
Remember: UltraViolet (UV) und InfraRed (IR) divergencies appear
Might try:

• same as for simple one-loop: Feynman parameters, direct integration

• use algebraic relations between integrals and find a (minimal) basis of master
integrals – is a preparation of the final evaluation

• derive and solve (system of) differential equations

• derive and solve (system of) difference equations

• do something else for a direct evaluation of single integrals – of all of them or of

the masters only

Into the last category falls what we present here:

Use Feynman parameters and transform the problem
%