Singular Values of Elliptic Integrals
in Quantum Field Theory
David Broadhurst
The Open University, UK
Sonderhausen, 25 April 2008
1
1 Introduction
I shall consider the analytical structure of the simplest massive multi-loop
diagrams in QFT, namely vacuum bubbles with 2 vertices:
&%
'$
s s
V
2
&%
'$
s s
V
3
&%
'$
#
"
!
s s
V
4
&%
'$
#
"
!
s s
V
5
and the sunrise diagrams:
&%
'$
s s
S
3
&%
'$
s s
S
4
&%
'$
#
"
!
s s
S
5
&%
'$
#
"
!
s s
S
6
that result from cutting a single line, in 2 spacetime dimensions.
Was sind and was sollen die Zahlen?
2
2 Two Bessel functions
The no-loop sunrise diagram is, of course, just the propagator in momen-
tum space:
S
2
(a, w) =
_
0
t K
0
(at)J
0
(wt) dt =
1
a
2
+ w
2
and the one-loop vacuum bubble
V
2
(a, b) =
_
0
t K
0
(at)K
0
(bt) dt =
_
0
wS
2
(a, w)S
2
(b, w) dw =
log(a/b)
a
2
b
2
comes from integrating a pair of propagators, with lHopitals rule giving
V
2
(1, 1) =
1
2
in the equal-mass case. The distribution
_
0
wJ
0
(wt
1
)J
0
(wt
2
) dw = 2(t
2
1
t
2
2
)
takes us from momentum to position space, in 2 euclidean dimensions.
3
3 Three Bessel functions
The one-loop sunrise diagram
S
3
(a, b, w) =
_
0
t K
0
(at)K
0
(bt)J
0
(wt) dt =
_
a+b
2v D
3
(a, b, v)
v
2
+ w
2
dv
has a discontinuity
D
3
(a, b, c) =
1
_
(a + b + c)(a b + c)(a + b c)(a b c)
given by the reciprocal of a Kallen function. The dispersion integral is
easily evaluated, to give
S
3
(a, b, w) = 2 arctanh
_
_
_
w
2
+ (a b)
2
_
w
2
+ (a + b)
2
_
_
D
3
(a, b, iw)
and the equal-mass on-shell result, with a = b = 1 and w = i, is
S
3
(1, 1, i) =
2 arctan(1/
3)
3
=
3
3
= L
3
(1)
4
where
L
3
(s) =
n=1
3
(n)
n
s
=
k=0
_
_
1
(3k + 1)
s
1
(3k + 2)
s
_
_
is the Dirichlet L-function with the real character
3
(n) given by the
LegendreJacobiKronecker symbol (D|n) for discriminant D = 3.
The two-loop vacuum diagram
V
3
(a, b, c) =
_
0
t K
0
(at)K
0
(bt)K
0
(ct) dt
gives a symmetric combination
V
3
(a, b, c) =
L
3
(a, b, c) + L
3
(b, c, a) + L
3
(c, a, b)
4
D
3
(a, b, c)
of dilogarithms
L
3
(a, b, c) = Li
2
_
_
(a
2
+ b
2
c
2
)D
3
(a, b, c) + 1
(a
2
+ b
2
c
2
)D
3
(a, b, c) 1
_
_
Li
2
_
_
(a
2
+ b
2
c
2
)D
3
(a, b, c) 1
(a
2
+ b
2
c
2
)D
3
(a, b, c) + 1
_
_
5
found by Davydychev and Tausk. In the equal-mass case we obtain
V
3
(1, 1, 1) =
3
4
L
3
(2) =
1
9
n=0
_
1
27
_
n
5
k=1
v
k
(6n + k)
2
with a vector of coecients v = [9, 9, 12, 3, 1] giving a highly con-
vergent sum, discovered (and proven) in the course of my investigation of
3-loop vacuum diagrams in 4 dimensions, where quadrilogarithms of the
sixth root of unity, (1 +
3)/2, were encountered.
3.1 Enumerating walks on a honeycomb lattice
The integers
a
k
=
p+q+r=k
_
k!
p! q! r!
_
2
=
k
j=0
_
_
k
j
_
_
2
_
_
2j
j
_
_
enumerate closed walks, of length 2k, on a two-dimensional hexagonal lat-
tice. They are generated by the cube of a Bessel function:
I
3
0
(2t) =
k=0
a
k
_
_
t
k
k!
_
_
2
6
from which the recurrence relation
(k + 1)
2
a
k+1
(10k
2
+ 10k + 3)a
k
+ 9k
2
a
k1
= 0
results. Remarkably, they also determine the sequence of odd moments
s
3,2k+1
=
_
0
t
2k+1
I
0
(t) K
2
0
(t) dt
whose rst term is s
3,1
= S
3
(1, 1, i) = L
3
(1). To prove this, one shows that
s
3,3
=
4
3
s
3,1
, by dierentiations of S
3
(1, 1, w) with respect to the magnitude
w of the external momentum, before going to the on-shell point w = i.
Then the result
s
3,2k+1
= L
3
(1)
_
_
2
k
k!
3
k
_
_
2
a
k
follows from the recursion for s
3,2k+1
. Later we will see remarkable relations
between Feynman diagrams and lattice Green functions.
7
4 Four Bessel functions
The 2-loop sunrise diagram S
4
(which is supposed to look like sunrise)
may be constructed by folding one instance of S
3
with the discontinuity of
another, to obtain
S
4
(a, b, c, w) =
_
a+b
2uD
3
(a, b, u)S
3
(u, c, w) du
from which we obtain an evaluation of the equal-mass on-shell value
S
4
(1, 1, 1, i) =
_
2
4 arctanh
__
u2
u+2
_
u(u
2
4)
du =
_
1
0
2y log(y)
y
4
1
dy =
2
16
by the substitution u = y + 1/y.
The 3-loop vacuum diagram
V
4
(a, b, c, d) =
_
0
wS
3
(a, b, w)S
3
(c, d, w) dw
obtained by integrating over a pair of one-loop graphs and is easily evalu-
ated in the equal-mass case by the substitution w = 1/y y:
V
4
(1, 1, 1, 1) =
_
0
4 arctanh
2
_
w
w
2
+4
_
w(w
2
+ 4)
dw =
_
1
0
4y log
2
(y)
1 y
4
dy =
7
8
(3)
8
4.1 Enumerating walks on a diamond lattice
The integers
b
k
=
p+q+r+s=k
_
k!
p! q! r! s!
_
2
=
k
j=0
_
_
k
j
_
_
2
_
_
2k 2j
k j
_
_
_
_
2j
j
_
_
enumerate closed walks, of length 2k, on a three-dimensional diamond
lattice. They are generated by the fourth power of a Bessel function:
I
4
0
(2t) =
k=0
b
k
_
_
t
k
k!
_
_
2
and determine the sequence of odd moments
s
4,2k+1
=
_
0
t
2k+1
I
0
(t) K
3
0
(t) dt
whose rst term is s
4,1
= S
4
(1, 1, 1, i) =
2
/16. The result
s
4,2k+1
=
2
16
_
k!
4
k
_
2
b
k
then follows from evaluation of s
4,3
=
1
4
s
4,1
and the use of recurrence rela-
tions. This relation to the diamond lattice will be exploited as we move
up to 5 Bessel functions.
9
5 Five Bessel functions
The 3-loop sunrise diagram may be written as an integral
S
5
(a, b, c, d, w) =
_
a+b+c
2v D
4
(a, b, c, v)S
3
(v, d, w) dv
over the discontinuity D
4
of
S
4
(a, b, c, w) =
_
a+b+c
2v D
4
(a, b, c, v)
v
2
+ w
2
dv
which is given by a complete elliptic integral K of the rst kind, and hence
by an arithmetic-geometric mean:
D
4
(a, b, c, d) =
_
dc
a+b
2uD
3
(a, b, u)D
3
(u, c, d) du
=
2K
_
Q(a,b,c,d)
Q(a,b,c,d)
_
Q(a, b, c, d)
=
AGM
_
Q(a, b, c, d), 4
abcd
_
where
Q(a, b, c, d) =
_
(a + b + c + d)(a + b c d)(a b + c d)(a b c + d)
10
Then in the equal-mass on-shell case we obtain
S
5
(1, 1, 1, 1, i) =
_
0
t I
0
(t) K
4
0
(t) dt =
_ 1
3
0
2D(y)
1 4y
2
arctanh
_
_
1 2y
1 + 2y
_
_
dy
as an integral over
D(y) =
2y
AGM
__
(1 + 3y)(1 y)
3
,
16y
3
_
Similar the equal-mass 4-loop vacuum diagram gives
V
5
(1, 1, 1, 1, 1) =
_
0
t K
5
0
(t) dt =
_ 1
3
0
D(y)L(y)
1 4y
2
dy
where D(y) is now folded with the dilogarithm
L(y) = Li
2
_
_
1
1 4y
2
2
_
_
+
1
2
log
2
_
_
1
1 4y
2
2
_
_
log
2
(y) +
2
12
I have computed 200,000 decimal digits of V
5
.
11
5.1 Conjectural evaluations with one external particle
5
2
_
0
t I
0
(t)K
4
0
(t) dt
?
=
_
1
15
_
_
2
15
_
_
4
15
_
_
8
15
_
480
5
2
_
0
t
3
I
0
(t)K
4
0
(t) dt
?
=
13
_
1
15
_
_
2
15
_
_
4
15
_
_
8
15
_
30
3
_
7
15
_
_
11
15
_
_
13
15
_
_
14
15
_
15
5.2 Proven evaluations with two external particles
3
_
0
t I
2
0
(t)K
3
0
(t) dt =
_
1
15
_
_
2
15
_
_
4
15
_
_
8
15
_
480
3
_
0
t
3
I
2
0
(t)K
3
0
(t) dt =
13
_
1
15
_
_
2
15
_
_
4
15
_
_
8
15
_
30
3
+
_
7
15
_
_
11
15
_
_
13
15
_
_
14
15
_
15
12
5.3 Conjectural sum rule
_
0
t I
0
(t)
_
K
0
(t)
2
15
I
0
(t)
_
K
3
0
(t) dt
?
= 0
with a dispersive version
_ 1
3
0
D(y)
1 4y
2
_
_
2 arctanh
_
_
1 2y
1 + 2y
_
_
15
_
_
dy
?
= 0
and a non-dispersive version
_
0
E(x)
4 + x
2
_
2 arcsinh
_
x
2
_
15
_
dx
?
= 0
where
E(x) =
AGM(z, z
)
=
AGM(z, |z|)
, z =
_
(1 ix)(3 + ix)
3
13
5.4 A modular identity from QFT
3
2
(q
3/2
)
2
(q
1/2
)
=
2
3
(q)
_
_
_
3
(q
3
)
3
(q)
3
3
(q
3
)
3
3
(q)
_
_
_
where
2
(q) =
n=
q
(
n+
1
2
)
2
,
3
(q) =
n=
q
n
2
This was derived by combining two evaluations in terms of E(x) with a
result from Davydychev and Delbourgo for the area of the Dalitz plot.
5.5 Doubling of internal and external masses
_
0
t I
2
0
(t)K
2
0
(t)K
0
(2t) dt =
1
3
_
0
t I
0
(2t)I
0
(t)K
3
0
(t) dt =
6
_
1
3
_
64
2
2
2/3
proven by combining contour integration with an evaluation by Watson in
1939 of a lattice Green function.
14
5.6 Lattice Green functions
W
j
(z) =
1
3
_
0
_
0
_
0
d
1
d
2
d
3
1 z w
j
(
1
,
2
,
3
)
face-centred cubic:
w
1
(
1
,
2
,
3
) =
cos
1
cos
2
+ cos
2
cos
3
+ cos
3
cos
1
3
simple cubic:
w
2
(
1
,
2
,
3
) =
cos
1
+ cos
2
+ cos
3
3
body-centred cubic:
w
3
(
1
,
2
,
3
) = cos
1
cos
2
cos
3
diamond:
w
4
(
1
,
2
,
3
) =
1 + cos
1
cos
2
+ cos
2
cos
3
+ cos
3
cos
1
4
In 1939, Watson evaluated W
j
(1) for j = 1, 2, 3 in terms of squares of
elliptic integrals at singular values. QFT chooses the diamond value W
4
_
1
4
_
.
15
5.7 Expansions of lattice Green functions
W
1
(z) =
k=0
f
k
_
z
12
_
k
W
2
(z) =
k=0
_
_
2k
k
_
_
a
k
_
z
6
_
2k
W
3
(z) =
k=0
_
_
2k
k
_
_
3 _
z
8
_
2k
W
4
(z) =
k=0
b
k
_
z
16
_
k
with integers
a
k
=
k
j=0
_
_
k
j
_
_
2
_
_
2j
j
_
_
b
k
=
k
j=0
_
_
k
j
_
_
2
_
_
2k 2j
k j
_
_
_
_
2j
j
_
_
f
k
=
k
j=0
_
_
k
j
_
_
(4)
kj
b
j
16
5.8 Relations between lattice Green functions
_
_
k=0
a
k
(x)
k
_
_
2
=
k=0
f
k
x
k
(1 + 3x)
2k+2
=
k=0
a
k
_
2k
k
_
(x(1 + x)(1 + 9x))
k
((1 3x)(1 + 3x))
2k+1
=
k=0
__
2k
k
_
x
k
_
3
((1 + x)
3
(1 + 9x))
k+
1
2
=
k=0
b
k
x
k
((1 + x)(1 + 9x))
k+1
for suciently small x. These relations between lattice Green functions
originate, at bottom, in identities obtained by Wilfrid Norman Bailey in
1936, where he generalized Clausen relations of the form (
2
F
1
)
2
3
F
2
used
in my work on D-dimensional 3-loop QED, and more recently by Grozin in
3-loop HQET. More generally, Baileys identities enable us to relate elliptic
integrals to Appell F
4
double series. In short,
Elliptic
2
Honeycomb
2
Cubic Diamond QFT
17
5.9 Singular value of an elliptic integral in QFT
In particular, I have proven that
K
2
15
=
8
15
3
_
0
t I
2
0
(t)K
3
0
(t) dt
where K
15
= K(k
15
) is evaluated at the 15th singular value and
K(k) =
_
/2
0
d
1 k
2
sin
2
=
2
1
AGM(1, k
)
with k
1 k
2
.
The Nth singular value, k
N
, yields the algebraic relation
K
N
= K(k
N
) =
1
N
K(k
N
)
in which case K
N
is reducible to values and an algebraic number. For
QFT, at N = 15, the algebraic number theory involves only surds:
1
3
_
0
t I
2
0
(t)K
3
0
(t) dt =
_
1
15
_
_
2
15
_
_
4
15
_
_
8
15
_
480
=
5 1
8
K
2
15
18
5.10 The ChowlaSelberg formula
More generally, for every fundamental discriminant D < 4,
D>k>0
_
_
k
D
__
(D|k)
= (2D)
h(D)
[a,b,c]H(D)
1
a
_
_
b +
D
2a
_
_
4
where the product runs over the strict equivalence classes [a, b, c] of prim-
itive integral binary quadratic forms ax
2
+ bxy + cy
2
with discriminant
D = b
2
4ac and
(z) = exp(iz/12)
k=1
(1 exp(2ikz))
is Dirichlets eta function. The equivalence classes form an Abelian group
H(D), by Gausss composition of quadratic forms, and the order of this
group is the class number h(D) of the number eld Q(
D).
QFT has chosen, via its connection to the diamond lattice, an algebraic
multiple of a product of values with D = N = 15. I am now computing
corresponding algebraic multiples for singular values with N 15. The
state of the art (as of last week) is N = 1, 242, 763. The previous number
theory record was N = 210.
19
6 Six Bessel functions
S
6
yields a number with many relations to previous and other structures:
12
4
_
0
t I
0
(t)K
5
0
(t) dt =
_
0
exp(4t)I
4
0
(t) dt =
n=0
_
2n
n
_
b
n
4
3n+1
=
4
2
_
0
t I
3
0
(t)K
3
0
(t) dt =
4
2
_ 1
3
0
D(y)
D(y) dy
=
2
2
_
0
exp(4t)I
2
0
(t)K
2
0
(t) dt =
1
3
3
_
1
1
3
D(y)D
_
1
3y
_
y
dy
=
1
3
_
1
3
5
W
2
(z)
_
(1 z)(z 3/5)
dz
with K
2
2
in W
2
_
3
5
_
, K
2
6
in W
2
(1) and AGMs in
D(y) =
2y
AGM
__
(1 + 3y)(1 y)
3
,
16y
3
_
D(y) =
1
AGM
__
(1 + 3y)(1 y)
3
,
_
(1 3y)(1 + y)
3
_
20
Acknowledgements
I thank my co-authors, David Bailey, Jonathan Borwein and Larry Glasser,
and condensed-matter advisors, Georey Joyce and John Zucker.
Stefano Laporta provided the stimulus for this work, by discovering the
15th singular value in 4-dimensional QFT.
I am grateful to the organising committee for this successful meeting and
in particular to Tord Riemann, for his Shakespearian comment
Ist es auch Wahnsinn,
So hat es doch Methode
which seems to me to apply rather well to the work reported in
Bailey, Borwein, Broadhurst, Glasser, [Link]/0801.0891
Broadhurst, arXiv/0801.4813
21