Feynman integrals via hyperlogarithms

arXiv:1407.0074v1 [hep-ph] 30 Jun 2014

Erik Panzer∗
Institues of Physics and Mathematics, Humboldt-Universität zu Berlin
Unter den Linden 6, 10099 Berlin, Germany
E-mail: panzer@mathematik.hu-berlin.de

This talk summarizes recent developments in the evaluation of Feynman integrals using
hyperlogarithms. We discuss extensions of the original method, new results that were
obtained with this approach and point out current problems and future directions.

Loops and Legs in Quantum Field Theory

27 April 2014 – 02 May 2014
Weimar, Germany

∗
Speaker.

c Copyright owned by the author(s) under the terms of the Creative Commons Attribution-NonCommercial-ShareAlike Licence.http://pos.sissa.it/
Feynman integrals via hyperlogarithms Erik Panzer

1. Schwinger parameters and hyperlogarithms

Many talks at this conference demonstrate the remarkable progress in the exact eval-
uation of Feynman integrals that was achieved during the past few years. A key element
shared by many of these advances is an improved understanding of multiple polylogarithms

z1k1 · · · zrkr
N,
X
Lin1 ,...,nr (z1 , . . . , zr ) = , where n1 , . . . , nr ∈ (1.1)
0<k1 <...<kr
k1n1 · · · krnr

which suffice to express a wide class of Feynman integrals.1 They admit representations as
iterated integrals [19] of the form
dzn dz1
C
Z Z
Lσn ,...,σ1 (z) = ··· ··· with σ1 , . . . , σn ∈ and σ1 6= 0, (1.2)
z n − σn z 1 − σ1
0<z1 <...zn <z

called hyperlogarithms [23], and can be manipulated symbolically very efficiently. In these
proceedings we consider the method put forward by Francis Brown [13], that aims at
integrating out Schwinger parameters αe one by one in the representation
 ω Y
Γ(ω) δ(1 − αeN ) ψ D
Z
αeνe −1 dαe ,
X
Φ(G) = Q ω= νe − h1 (G) (1.3)
e Γ(νe ) RN+ ψ D/2 ϕ e∈E e∈E
2

of Feynman integrals associated to a graph G with h1 (G) loops and N edges E. This
formula assumes scalar propagators (ke2 + m2e )−νe of mass me at each edge e (raised to some
power νe called index), but generalizations to tensor integrals exist [25]. The Symanzik
polynomials [9] sum all spanning trees T and 2-forests F in
XY X Y X
ψ= αe and ϕ = q(F )2 αe + ψ αe m2e , (1.4)
T e∈T
/ F e∈F
/ e∈E

where q(F )2 is the square of the momentum flowing between the two components of F . We
choose an order e1 , . . . , eN of edges and compute Ik := 0∞ Ik−1 dαek iteratively, starting
R

with the original integrand I0 of (1.3). Details of the involved algorithms are given in
[12, 13] and a public implementation in Maple [24] is available [28]. Also the more general
approach presented at the preceeding conference [7], based on iterated integrals in several
variables (instead of just one), is going to be published in the nearest future.
The crucial limitation of this method is that all Ik (αek+1 , . . . , αeN ) must be expressible
C
as (αek+1 , . . . , αeN )-linear combinations hyperlogarithms (1.2) with respect to the next
integration variable z = αek+1 . If this holds for some order on E, we call G linearly reducible.
This criterion depends only on the Symanzik polynomials and reduction algorithms are
available [12, 13] that check conditions sufficient for linear reducibility.2
In particular this discussion is unaffected by infinitessimal expansions in analytic regu-
lators, like the popular shift D = D b −2ε away from an even dimension D b ∈ 2 of spacetime3 N
1
Counterexamples are known in massive [3, 5], massless [15] and even supersymmetric theories [17, 26].
2
This means that reductions can reveal linear reducibility, but not disprove it.
3
Even if one starts out with Db = 4 dimensions, a reduction of tensor integrals will produce scalar
integrals also in even dimensions above D.
b

2
Feynman integrals via hyperlogarithms Erik Panzer

5 P1 5 P2 5 P3 5 P4 5 P5 5 P6

5 P7 5 N1 5 N2 5 N3 5 N4

Figure 1: All three-connected five-loop vacuum graphs [27], divided into planar (P ) and non-planar
(N ) ones. The zig-zag 5 P3 = ZZ5 and 5 N1 were considered in [13]. Cutting any edge produces a
propagator with four loops, deleting a three-valent vertex creates a three-loop three-point graph.

P7,8 = P7,9 = P7,11 =

Figure 2: The most complicated primitive periods in φ4 -theory at seven loops are given by the
three graphs P7,8 , P7,9 and P7,11 in the notation of the census [31].

such that Φ(G) = k Φk (G) · εk becomes a Laurent series. All of its coefficients can in prin-
P

ciple be computed as soon as G is linearly reducible, and remarkably this also applies for
the multivariate expansions in the indices νe = νbe + εe close to integers νbe ∈ . Z
2. Massless propagators up to 6 loops

Originally, this method was applied to convergent massless propagators G only [13].
These transform into vacuum graphs G e = G∪ ˙ {e} upon joining the external legs of G by an
edge e. Then only the first Symanzik ψGe = ψG αe + ϕG is of interest and a powerful toolbox
of algebraic identities becomes available [12, 15, 16].
We verified that all massless propagators up to four loops (e. g. all which arise by
cutting an edge of the graphs shown in figure 1) are linearly reducible [27]. The recent
computation [4, 33] of the corresponding master integrals (using different methods) can
therefore be extended to higher orders in ε and to include self-energy insertions. Such
results will be needed for calculations at higher loop orders and are already available for
many examples [27].
The traditional benchmark in this field is the evaluation of primitive divergent periods
of φ4 -theory in D = 4 dimensions [11]. This problem is now completely solved up to 7
loops: All primitive graphs of the census [31] to this order are linearly reducible—except
for P6,4 (that can be computed with graphical functions [32]) and P7,11 of figure 2 which
we address below. These results suggest that all massless propagators with at most 6 loops
could be computable using multiple polylogarithms. This is the most optimistic scenario,
as some 7 loop massless propagators are known to exceed the world of polylogarithms [14].

3
Feynman integrals via hyperlogarithms Erik Panzer

2.1 Changing variables for linear reductions

After ten integrations for P7,11 , the partial integral I10 = L/d10 consists of a hyperlog-
arithm L of weight 8 and the irreducible, totally quadratic denominator

d10 = α2 α42 α1 + α2 α42 α3 − α1 α2 α3 α4 + α22 α4 α1 + α22 α4 α3 − 2α2 α32 α4 − α22 α32

(2.1)
− 2α22 α3 α1 − 2α2 α32 α1 − α32 α42 − 2α32 α4 α1 − α22 α12 − 2α2 α3 α12 − α32 α12 .

Further integration would introduce square roots of the discriminant of this polynomial
and therefore escape the space of hyperlogarithms with rational arguments; P7,11 is not
linearly reducible as such. But (2.1) can be linearized: If we change variables according to
α30 α1 = α3 (α1 + α2 + α4 ), α4 = α40 (α2 + α30 ) and α1 = α10 α40 , the new denominator

d010 = (α2 + α30 )(α2 + α2 α40 − α10 )(α10 α40 + α2 + α2 α40 + α30 α40 ) (2.2)

factorizes linearly such that α10 , α30 and α40 can indeed be integrated without further com-
plications (α2 = 1). Interestingly, the final period is not a multiple zeta value but a linear
combination of multiple polylogarithms (1.1) evaluated at a sixth roots of unity.

2.2 Missing alternating sums

The other two of the most complicated vacuum graphs with 7 loops (shown in figure 2)
are linearly reducible without further ado and we succeeded to compute
   
92943 3381 2
P7,9 = 160 ζ11 + 20 ζ3,5,3 − ζ3 ζ3,5 − 1155
4 ζ3 ζ5 + 896ζ3
27 45 261
80 ζ3,5 + 64 ζ3 ζ5 − 320 ζ8 (2.3)

using hyperlogarithms. This result was recently proposed by David Broadhurst [10] and is
now confirmed. Interestingly, for both P7,8 and P7,9 the last integrand I12 has denominator
(α1 + α2 )(α1 − α2 ) and produces a result involving alternating sums. Only after reducing
these to the data mine basis [6] we arrived at the multiple zeta value (2.3). So far, the
reason for this absence of alternating sums in massless φ4 -theory still eludes us.

3. Non-trivial kinematics

With non-trivial kinematics, the second Symanzik ϕ makes linear reducibility a more
restrictive criterion that depends sensitively on the distribution of external momenta and
internal masses. In particular it requires at least one propagator to be massless.4
Some examples of reducible graphs with masses are shown in figure 3 and some explicit
results were computed [29]. However, no combinatorial characterization of this class of
graphs is known so far. We now specialize to two particular kinematic configurations in
order to state more general results.
4
This restriction does not apply when there is no momentum dependence at all. Important examples
of such graphs that are linearly reducible are known and hyperlogarithms are used to compute generating
functions for Mellin moments of massive operator matrix elements [1, 2].

4
Feynman integrals via hyperlogarithms Erik Panzer

Figure 3: Examples of linearly reducible graphs with some massive internal and off-shell external
momenta (thick edges).

Figure 4: Two-loop four-point graphs without self-energy (propagator) subgraphs.

3.1 Massless three-point functions

Massless three-point functions (depending on arbitrary p21 , p22 and p23 ) seem to be very
well suited for parametric integration: Linear reducibility holds for all such graphs at 2
loops [18] and even at 3 loops [29]. This includes all graphs obtained by removing one
three-valent vertex from any of the vacuum graphs in figure 1 (external legs are attached
to the neighbours of that deleted vertex).
When these integrals are written as hyperlogarithms, one encounters the square root

2
of the Källén function λ = p21 + p22 − p23 − 4p21 p22 . The reparametrization given by
√
p22 = p21 · z z̄ and p23 = p21 · (1 − z)(1 − z̄) rationalizes λ = ±p21 · (z − z̄) (3.1)

in terms of two new variables z and z̄. Then Φ(G) is given by p−2 1
ω
and a rational linear
combination of hyperlogarithms Lv (z)Lw (z̄). The polynomial reduction provides an upper
bound on the set of letters that may appear in the words v and w; put differently it restricts
the entries of the symbol [20] of the hyperlogarithms.
Details and explicit results for some integrals can be found in [29]. There are reducible
examples at higher loop orders, but non-reducible graphs appear already at 4 loops.

3.2 Massless four-point functions

All massless four-point on-shell graphs (p21 = p22 = p23 = p24 = 0) with at most two loops
are linearly reducible [8]. In particular these include the graphs of figure 4.
At 3 loops the first graphs that are not linearly reducible occur, including the com-
plete graph K4 (first graph in bottom row of figure 3 with the fourth external momentum
attached to the center) which was recently evaluated [21] to harmonic polylogarithms [30].
For its parametric integration, a change of variables is necessary [29].
While the general type of massless four-point functions is not clear, it seems that
at least all of the n-loop box ladders Bn (figure 5) are linearly reducible: For on-shell
kinematics we always obtained a result in terms of harmonic polylogarithms. Due to the
recent interest [22] into these integrals in D = 6 dimensions (where they are finite), we list

5
Feynman integrals via hyperlogarithms Erik Panzer

B3 = B4 =

Figure 5: Examples of the box ladder graphs Bn of n = 3 and n = 4 loops.

the values cn := Φ(Bn )|s=1,t=0 for n ≤ 6 where s = (p1 + p2 )2 and t = (p1 + p4 )2 measure
the momentum running through Bn in the horizontal and vertical directions:5

c2 = 2ζ2 , (3.2)
c3 = 4ζ32 + 124 3
35 ζ2 − 8ζ3 − 6ζ2 , (3.3)
 
c4 = −56ζ7 − 32ζ2 ζ5 + 32ζ32 + 85 ζ3 4ζ22 − 15 + 992 3 2
35 ζ2 − 8ζ2 − 18ζ2 , (3.4)
 
c5 = 56ζ7 (ζ3 − 5) + 26ζ52 + 4ζ5 (8ζ2 ζ3 + 35ζ3 − 40ζ2 − 49) + 45 ζ32 140 − 25ζ2 − 4ζ22
   
+ 8ζ3 7ζ2 + 4ζ22 − 14 − 1168 5 24 4 496 3
385 ζ2 − 7 ζ2 + 5 ζ2 + 4ζ2 2ζ3,5 − 21 + 20ζ3,5 + 4ζ3,7 , (3.5)
 
18864 3
c6 = 8
35 ζ2 + 336ζ3,5 − 12ζ9 (20ζ2 + 161) + 5 ζ7 104ζ22 + 35ζ2 + 840ζ3 − 1120
 
+ 624ζ52 + 16 3 2
35 ζ5 1680ζ2 ζ3 − 3675 − 12ζ2 − 2240ζ2 + 490ζ2 + 5145ζ3
     
+ 96 ζ22 + ζ3,7 − 48 2 2 32 2 3
5 ζ3 35ζ2 + 8ζ2 − 60 − 5 ζ3 105 − 32ζ2 + 3ζ2 − 75ζ2
 
5 288 4
+ 24ζ2 8ζ3,5 − 21 − 28032
385 ζ2 − 5 ζ2 − 1320ζ11 . (3.6)

4. Problems for parametric integration with hyperlogarithms

1. To compute divergent integrals in dimensional regularization, one first needs to con-

struct a representation involving only convergent integrands. An algorithm that
solves this problem was presented in [29], but already for low numbers of divergences
it can produce expressions that become untractibly large. It seems very promising to
combine this algorithm with programs for integration by parts, in order to obtain a
reduction to finite (convergent) master integrals.

2. We know many cases (like K4 and P7,11 mentioned above) of integrals that are not
linearly reducible in the original Schwinger parameters, but become so after a suitable
change of variables. It is unclear under which general circumstances this is possible.
So far, there is only one combinatorial analysis available which considers a particular
kind of transformation for vacuum integrals [34].

3. Why do alternating sums so far not occur in massless propagators?

4. The implementation [28] works in the Euclidean region. Analytic continuation to

various kinematic regimes of the physical region can in general be very cumbersome
and should be automated.
5
Printing the expressions for Φ(Bn ) including the dependence on s and t would take too much space.

6
Feynman integrals via hyperlogarithms Erik Panzer

References
[1] J. Ablinger, J. Blümlein, A. Hasselhuhn, S. Klein, C. Schneider, and F. Wißbrock, Massive
3-loop ladder diagrams for quarkonic local operator matrix elements, Nucl. Phys. B 864
(Nov., 2012) 52–84, [arXiv:1206.2252].
[2] J. Ablinger, J. Blümlein, C. Raab, C. Schneider, and F. Wißbrock, Calculating massive
3-loop graphs for operator matrix elements by the method of hyperlogarithms, Nucl. Phys. B
885 (Aug., 2014) 409–447, [arXiv:1403.1137].
[3] L. Adams, C. Bogner, and S. Weinzierl, The two-loop sunrise graph with arbitrary masses,
Journal of Mathematical Physics 54 (May, 2013) 052303, [arXiv:1302.7004].
[4] P. A. Baikov and K. G. Chetyrkin, Four loop massless propagators: An algebraic evaluation
of all master integrals, Nucl. Phys. B 837 (Oct., 2010) 186–220, [arXiv:1004.1153].
[5] S. Bloch and P. Vanhove, The elliptic dilogarithm for the sunset graph, arXiv:1309.5865.
[6] J. Blümlein, D. J. Broadhurst, and J. A. M. Vermaseren, The Multiple Zeta Value data mine,
Comput. Phys. Commun. 181 (Mar., 2010) 582–625, [arXiv:0907.2557].
[7] C. Bogner and F. C. S. Brown, Symbolic integration and multiple polylogarithms,
PoS(LL20129)053 [arXiv:1209.6524].
[8] C. Bogner and M. Lüders, Multiple polylogarithms and linearly reducible Feynman graphs,
arXiv:1302.6215.
[9] C. Bogner and S. Weinzierl, Feynman Graph Polynomials, International Journal of Modern
Physics A 25 (2010) 2585–2618, [arXiv:1002.3458].
[10] D. J. Broadhurst, The number theory of radiative corrections, in RADCOR 2013, (Lumley
Castle, UK), Sept., 2013.
[11] D. J. Broadhurst and D. Kreimer, Knots and numbers in φ4 theory to 7 loops and beyond,
Int. J. Mod. Phys. C 6 (Aug., 1995) 519–524, [hep-ph/9504352].
[12] F. C. S. Brown, On the periods of some Feynman integrals, arXiv:0910.0114.
[13] F. C. S. Brown, The Massless Higher-Loop Two-Point Function, Commun. Math. Phys. 287
(May, 2009) 925–958, [arXiv:0804.1660].
[14] F. C. S. Brown and O. Schnetz, A K3 in φ4 , Duke Math. J. 161 (July, 2012) 1817–1862,
[arXiv:1006.4064].
[15] F. C. S. Brown and O. Schnetz, Proof of the zig-zag conjecture, arXiv:1208.1890.
[16] F. C. S. Brown and K. A. Yeats, Spanning Forest Polynomials and the Transcendental Weight
of Feynman Graphs, Commun. Math. Phys. 301 (Jan., 2011) 357–382, [arXiv:0910.5429].
[17] S. Caron-Huot and K. J. Larsen, Uniqueness of two-loop master contours, JHEP 2012 (Oct.,
2012) 26, [arXiv:1205.0801].
[18] F. Chavez and C. Duhr, Three-mass triangle integrals and single-valued polylogarithms, JHEP
11 (Nov., 2012) 114, [arXiv:1209.2722].
[19] K. T. Chen, Iterated path integrals, Bull. Amer. Math. Soc. 83 (Sept., 1977) 831–879.
[20] A. B. Goncharov, M. Spradlin, C. Vergu, and A. Volovich, Classical Polylogarithms for
Amplitudes and Wilson Loops, Phys. Rev. Lett. 105 (Oct., 2010) 151605, [arXiv:1006.5703].

7
Feynman integrals via hyperlogarithms Erik Panzer

[21] J. M. Henn, A. V. Smirnov, and V. A. Smirnov, Evaluating single-scale and/or non-planar

diagrams by differential equations, JHEP 2014 (Mar., 2014) TTP13-046, [arXiv:1312.2588].
[22] D. I. Kazakov, Evaluation of Multi-Box Diagrams in Six Dimensions, JHEP 2014 (Apr.,
2014) 1–9, [arXiv:1402.1024].
[23] J. A. Lappo-Danilevsky, Mémoires sur la théorie des systèmes des équations différentielles
linéaires, vol. I–III. Chelsea, 1953.
[24] Maplesoft, a division of Waterloo Maple Inc., “Maple 16.”
[25] N. Nakanishi, Graph theory and Feynman integrals, vol. 11 of Mathematics and its
applications. Gordon and Breach, New York, 1971.
[26] D. Nandan, M. F. Paulos, M. Spradlin, and A. Volovich, Star integrals, convolutions and
simplices, JHEP 2013 (May, 2013) 105, [arXiv:1301.2500].
[27] E. Panzer, On the analytic computation of massless propagators in dimensional
regularization, Nucl. Phys. B 874 (Sept., 2013) 567–593, [arXiv:1305.2161].
[28] E. Panzer, Algorithms for the symbolic integration of hyperlogarithms with applications to
Feynman integrals, arXiv:1403.3385.
[29] E. Panzer, On hyperlogarithms and Feynman integrals with divergences and many scales,
JHEP 2014 (Mar., 2014) 71, [arXiv:1401.4361].
[30] E. Remiddi and J. A. M. Vermaseren, Harmonic polylogarithms, Int. J. Mod. Phys. A 15
(2000), no. 5 725–754, [hep-ph/9905237].
[31] O. Schnetz, Quantum periods: A Census of φ4 -transcendentals, Commun. Number Theory
Phys. 4 (2010), no. 1 1–47, [arXiv:0801.2856].
[32] O. Schnetz, Graphical functions and single-valued multiple polylogarithms, arXiv:1302.6445.
[33] A. V. Smirnov and M. N. Tentyukov, Feynman Integral Evaluation by a Sector decomposiTion
Approach (FIESTA), Comput. Phys. Commun. 180 (2009) 735–746, [arXiv:0807.4129].
[34] K. A. Yeats, Some combinatorial interpretations in perturbative quantum field theory,
arXiv:1302.0080.