DFUB 96-05

7 February 1996

The analytical value of the electron (g-2) at order α3 in QED.

S.Laporta §
arXiv:hep-ph/9602417v1 28 Feb 1996

E.Remiddi ◦

Dipartimento di Fisica, Università di Bologna,

and INFN, Sezione di Bologna,
Via Irnerio 46, I-40126 Bologna, Italy

Abstract
We have evaluated in closed analytical form the contribution of the three-loop
non-planar ‘triple-cross’ diagrams contributing to the electron (g-2) in QED; its value,
omitting the already known infrared divergent part, is
 
1 2 55 16 4 32 1 4
ae (3 − cross) = π ζ(3) − ζ(5) − π + a4 + ln 2
2 12 135 3 24
14 1 23 47 113
+ π 2 ln2 2 − ζ(3) + π 2 ln 2 − π 2 − .
9 3 3 9 48
This completes the analytical evaluation of the (g-2) at order α3 , giving
 α 3  83 215 100

1 4

1 2 2

2
ae (3 − loop) = π ζ(3) − ζ(5) + a4 + ln 2 − π ln 2
π 72 24 3 24 24

239 4 139 298 2 17101 2 28259
− π + ζ(3) − π ln 2 + π +
2160 18 9 810 5184
 α 3
= (1.181241456...) .
π

PACS: 12.20Ds; 13.40.Em; 06.20.Jr; 12.20Fv

Keywords: Quantum ElectroDynamics; Anomalous magnetic moment of the electron; Analyt-
ical evaluation of 3-loop radiative corrections.

§ E-mail: laporta@bo.infn.it
◦
E-mail: remiddi@bo.infn.it

1
 Following the work of Ref. [1] we have completed the evaluation in close analytical
form of the contribution to the electron anomaly in three-loop QED due to the triple-
cross graphs depicted in Fig. 1).

p
∆
a) b) c)

Fig.1. The “triple-cross” graphs.

The results are

  
215 1 2 53 4 1 4 1 2 2
ae (3 − cross; 1a) = ζ(5) − π ζ(3) − π + 4 a4 + ln 2 − π ln 2
24 3 2160 24 24
1229 7 4165 2 515 1
− ζ(3) − π 2 ln 2 + π − + ln λ
144 6 2592 864 2
1
= 1.285068495... + ln λ ,
2
(1)

275 29 43 4 5 2 2 623 35
ae (3 − cross; 1b) = − ζ(5) + π 2 ζ(3) + π − π ln 2 + ζ(3) + π 2 ln 2
24 36 1080 6 144 9
1951 2 493
− π −
648 864
= − 0.878968171... ,
(2)

 
5 4 2 161 4 8 1 4 32
ae (3 − cross; 1c) = ζ(5) − π ζ(3) − π + a4 + ln 2 + π 2 ln2 2
12 9 1080 3 24 9
97 20 1043 2 1
+ ζ(3) + π 2 ln 2 − π −
12 9 432 48
= − 0.026799490... ,
(3)

2
from which

ae (3 − cross) = 2ae (3 − cross; 1a) + 2ae (3 − cross; 1b) + ae (3 − cross; 1c)

 
1 2 55 16 4 32 1 4
= π ζ(3) − ζ(5) − π + a4 + ln 2
2 12 135 3 24
(4)
14 1 23 47 113
+ π 2 ln2 2 − ζ(3) + π 2 ln 2 − π 2 − + ln λ
9 3 3 9 48
= 0.785401156... + ln λ .

As it is customary in this kind of calculations, λ is the regularizing photon mass (in

∞
∞
P p 2
X 1
electron mass units), ζ(p) = 1/n , ζ(2) = π /6, a4 = .
n=1 n=1
2 n4
n

The above results are in excellent agreement with the numerical results

ae (3 − cross; 1a; Ref.2) = 1.291(7) ,

ae (3 − cross; 1b; Ref.2) = −0.882(10) ,
ae (3 − cross; 1c; Ref.2) = −0.021(100) ,
ae (3 − cross; Ref.3) = 0.785419(40) ,

where the ln λ’s are understood. As the previous graphs were the last graphs for which
the analytical value of the anomaly was still missing, on account the previously known
results [4,5,6,7] the complete analytical expression of the anomaly in three-loop QED
can now be written as
 α 3  83 215 100

1 4

1 2 2

2
ae (3 − loop) = π ζ(3) − ζ(5) + a4 + ln 2 − π ln 2
π 72 24 3 24 24

239 4 139 298 2 17101 2 28259
− π + ζ(3) − π ln 2 + π +
2160 18 9 810 5184
 α 3
= (1.181241456...) .
π
(5)
By using the best numerical value of ae (4 − loop) = −1.557(70) , Ref.[8], and

1/α = 137.0359979(32) ,

Ref.[9], one finds

ae (th) = 1159652201.2(2.1)(27.1) × 10−12 , (6)

to be compared with the experimental value, Ref.[10],

ae (exp) = 1159652188.4(4.3) × 10−12 ;

3
conversely, by using Ref.[10] as an input, one obtains

1/α(ae ) = 137.03599941(56) . (7)

If p is the electron momentum and ∆ the momentum transfer of some vertex am-
plitude, the corresponding (g-2) contribution is extracted by keeping p on the mass
shell, expanding the vertex up to first order in ∆, multiplying it by a suitable spinor
projection operator and then performing the appropriate traces [11]. The (g-2) is then
expressed as the sum a few hundred terms all of the kind

N
Z
d 4 k1 d 4 k2 d 4 k3 , (8)
D

each term having its own specific numerator N and denominator D. The numerators
N are in general simple monomials in the scalar products of p and the loop momenta
ki , such as for instance (p · ki ), (ki · kj ) etc, up to (p · k2 )4 , while the denominators D
have in general the form

D = D1n1 D2n2 D3n3 D4n4 D5n5 D6n6 D7n7 D8n8 ,

the exponents ni ranging from 0 to 2 (0 means that the denominator is absent; the
powers 2 arise from the expansion in ∆; in each term there are at most two denominators
with ni = 2); in me = 1 units the denominators, see Fig.2), are defined as

D1 =(p − k1 )2 + 1 − iǫ , D2 = (p − k1 − k2 )2 + 1 − iǫ ,
D3 =(p − k1 − k2 − k3 )2 + 1 − iǫ , D4 = (p − k2 − k3 )2 + 1 − iǫ , (9)
D5 =(p − k3 )2 + 1 − iǫ , D6 = k12 − iǫ , D7 = k22 − iǫ , D8 = k32 − iǫ .

Besides the 8 denominators, there are in the problem 9 linearly independent scalar
products p · ki , ki · kj , etc; 8 independent identities between the scalar products and the
denominators are easily written, such as for instance

1
p · k1 = (D6 − D1 ) ,
2 (10)
1
k1 · k3 = (−D2 + D3 − D4 + D7 ) − p · k2 .
2
The identities can be used to express the scalar products in the numerator as a com-
bination of the denominators, so obtaining a combination of integrands with a smaller
number of denominators and therefore simpler from the point of view of the analytical

4
integration. It is to observed however that, as there are 9 scalar products and only 8
denominators, one scalar product (which we choose to be p · k2 ) must remain anyhow in
the numerator; furthermore, the simpler terms obtained in that way can be in general
non convergent when taken separately .

k1 k3
D3
D6 D8
D2 D4

p D1 D7 D5 p
k2

Fig.2. The notation for the 3-loop integrations.

In Ref. [1] we gave the analytical value of the simplest term

(−i)3 1 1
Z
J0 ≡ d 4 k1 d 4 k2 d 4 k3 = 4π 2 ln2 2 − π 4 . (11)
π6 D1 D2 D3 D4 D5 D6 D7 D8 6

With a straightforward extension of the results of Ref. [1] one can obtain, for instance

(−i)3 p · k2 1
Z
J1 ≡ d 4 k1 d 4 k2 d 4 k3 = 5ζ(5) − π 2 ζ(3) . (12)
π6 D1 D2 D3 D4 D5 D6 D7 D8 2

The extension to all the other terms appearing in the expression giving the (g-2) is in
principle straightforward but in fact rather wearisome (it implies, among other things, to
build, for each tensor in the loop momenta, the scalar amplitudes free from kinematical
singularities for which unsubtracted dispersion relations can be written). It was found
convenient to work out another approach, based on the integration by part identity
method first developed in Ref.[12]. In our case the method consists in writing
3
!  
∂ N
Z Y
n
d ki vl,µ =0, (13)
i=1
∂kj,µ D

where vl is either the external momentum p or one of the three loop momenta ki ,
N/D stands for the ratio of all the pairs of numerators and denominators of the kind
described above; note however that while in the original integrals the loop momenta have
dimension 4, the identities are written for momenta having continuous n dimension, in
which case their validity is obvious. The required n = 4 values are then recovered by
means of a suitable limiting procedure. Needless to say, if the (g-2) from some graph
is obtained as the n → 4 limit of an n-dependent expression, this expression must be

5
evaluated for consistency in n dimensions from the very beginning; in particular, the
traces of the γ-matrix must be evaluated in n-dimensions too.
By explicitly performing the derivatives, using the identities (10) and rearranging
the obtained terms in an identity like Eq.(13), one obtains a relation between various
integrals which amounts to express one of them as a linear combination of the others.
The number of all the possible identities is large (a few thousands, as there are twelve
identities for each of the several hundreds N/D possibilities); they form a huge linear
system, whose coefficients are in turn polynomials in n, from which the overwhelming
majority of the occurring integrals can be expressed as a linear combination of a few
basic integrals. As we are interested in the n → 4 limit, we put n = 4 − 2ω and then
deal with the limit by expanding everything in powers of ω for ω → 0 . Fortunately, the
expansion and the extremely lengthy processing of the various identities can be carried
out in an almost mechanical way.
A thorough investigation shows that all the three loop integrals (and therefore also
the complete (g-2) expression in which we are interested) can be expressed in terms of
the following 18 basic integrals:
 3 Z
−i p · k2
I1 = d n k1 d n k2 d n k3 ,
π n−2 D1 D2 D3 D4 D5 D6 D7 D8
 3 Z
−i 1
I2 = d n k1 d n k2 d n k3 ,
π n−2 D1 D2 D3 D4 D7 D8
 3 Z
−i 1
I3 = d n k1 d n k2 d n k3 ,
π n−2 D1 D2 D4 D5 D6 D8
 3 Z
−i 1
I4 = d n k1 d n k2 d n k3 ,
π n−2 D2 D3 D4 D6 D7 D8
 3 Z
−i 1
I5 = d n k1 d n k2 d n k3 ,
π n−2 D1 D3 D4 D5 D7 D8
 3 Z
−i 1
I6 = d n k1 d n k2 d n k3 ,
π n−2 D1 D3 D5 D6 D7 D8
 3 Z
−i 1
I7 = n−2
d n k1 d n k2 d n k3 ,
π D2 D4 D5 D6 D7 D8
 3 Z
−i 1
I8 = n−2
d n k1 d n k2 d n k3 ,
π D1 D2 D3 D4 D5
 3 Z
−i 1
I9 = n−2
d n k1 d n k2 d n k3 ,
π D2 D3 D5 D6 D7

6
  3 Z
−i 1
I10 = d n k1 d n k2 d n k3 ,
π n−2 D2 D4 D6 D7 D8
 3 Z
−i p · k2
I11 = d n k1 d n k2 d n k3 ,
π n−2 D1 D3 D5 D7
 3 Z
−i 1
I12 = d n k1 d n k2 d n k3 ,
π n−2 D1 D3 D5 D7
 3 Z
−i 1
I13 = d n k1 d n k2 d n k3 ,
π n−2 D1 D2 D4 D5
 3 Z
−i 1
I14 = d n k1 d n k2 d n k3 ,
π n−2 D3 D5 D6 D7
 3 Z
−i 1
I15 = d n k1 d n k2 d n k3 ,
π n−2 D2 D3 D4 D5
 3 Z
−i 1
I16 = d n k1 d n k2 d n k3 ,
π n−2 D3 D4 D7 D8
 3 Z
−i 1
I17 = d n k1 d n k2 d n k3 ,
π n−2 D3 D6 D7 D8
 3 Z
−i 1
I18 = d n k1 d n k2 d n k3 .
π n−2 D1 D4 D5

Note that in the above list there is only one integral with all the 8 denominators (I1 ,
the limit of which is Eq.(12) above), while there are no integrals with 7 denominators.
With the exception of I1 , all the basic integral are divergent in the n → 4 limit; having
put n = 4 − 2ω, the divergences show up as 1/ω singularities, so that evaluating them
amounts to evaluate the coefficients of their expansions in ω. As an example, the value
of I18 , which perhaps the easiest to obtain as it factorizes in the product of three 1-loop
integrals, is
 
1 3 6 2 3 4
I18 = C(ω) − 3 − 2 − − 10 − 15ω − 21ω − 28ω + O(ω ) , (14)
ω ω ω

where C(ω), defined as

3
C(ω) = (π ω Γ(1 + ω)) ,

is an overall normalization factor, whose limiting value at ω = 0 is 1.

7
 As an example of the kind of formulae obtained by expressing a non-trivial integral
in terms of the basic integrals, let us give the relation

(−i)3 (p · k2 )2
Z 
4 4 4
d k1 d k2 d k3 = lim −I1
π6 D1 D2 D3 D4 D5 D6 D7 D8 ω→0
 
1 3 2 3
+ I2 (−3ω) + I3 − ω − 2ω − 6ω
4 4
   
3 2 3

1 1 2 3
+ I4 ω + I5 −ω − 2ω − 4ω + I6 − ω − 3ω − 16ω
2 2 2
   
2 3

1 1 1 2 3
+ I7 3ω + 14ω + 64ω + I8 − + I9 − − + ω + 4ω + 176ω
6ω 2ω 2
 
1 5 27 137 3
+ I10 − − ω − ω 2 − ω
2 2 2 2
 
1 139 1249 19225 2 238333 3
+ I11 − − ω− ω − ω
6ω 36 108 324 972
 
3 13 383 353 2 3041 3
+ I12 + + + 2ω − ω + ω
64ω 2 128ω 128 16 8
 
1 3 1 29 2 37 3
+ I13 − − − ω+ ω + ω
12ω 8 6 24 4
 
1 13 87 439 2 623 3
+ I14 − + + ω+ ω + ω
4ω 8 8 8 2
 
7 11 19 55 2 73 3
+ I15 − − − ω+ ω + ω
12ω 8 24 12 2
 
1 7 7 9 2 3
+ I16 − − − ω + ω + 196ω
2ω 4 4 4
 
3 3 29 50 2
+ I17 + + ω+ ω
2ω 4 6 3
 
3 2 55 1391 94423 2 3535507 3
+ I18 − − − ω− ω + ω .
64ω 2 3ω 36 864 5184 7776
(15)
Similar expressions hold for all the other integrals contributing to the anomaly. In all
those expressions, the coefficient of I1 is always finite (i.e. not singular for ω → 0), so
that only its ω = 0 limit, given by J1 of Eq.(12), is required.
Besides I1 , we have already given the explicit analytical value of I18 , Eq.(14). In
order to complete the calculation of all the basic integrals, we have evaluated directly
only a subset of them (namely I8 , I9 , I10 and from I13 to I18 ). For the others, we found
it more convenient to write a large number of identities of the form of Eq.(15), in which
however the analytical value of the integral on the l.h.s. is already known from previous
work; the relation is then expressing such a known result in terms of the still unknown

8
basic integrals. In the case of J0 , Eq.(11), one finds for instance

      
5 15 3 9 5 5
J0 = lim I2 − ω + I3 − ω + I4 − + ω
ω→0 2 2 4 4 4 4
     
1 3 3 3 3 2 3 5 5 2 3
+ I5 − ω + I6 − ω + ω + 6ω + I7 − + ω − 5ω − 20ω
2 2 4 4 2 2 2
   
2 4 16 64 2 3 1 1
+ I8 − + − ω − ω − 128ω + I9 − +1+ω
3ω 3 3 3 8ω 2 ω
 
1 7 39 201 3
+ I10 + 1 + ω + ω2 + ω
4ω 2 2 2
 
7 49 11 1819 2 34249 3
+ I11 − + ω− ω − ω
6ω 36 108 324 972
 
3 59 203 15 37 53 2 113 3
+ I12 − + − + − ω− ω − ω
128ω 3 256ω 2 256ω 8 16 16 8
 
1 17 211 423 2 1263 3
+ I13 − − − ω− ω − ω
3ω 24 48 16 8
 
1 49 53 3 119 2 427 3
+ I14 − + − ω+ ω + ω
16ω 2 32ω 32 8 16 8
 
4 17 211 423 2 1263 3
+ I15 − − − ω− ω − ω
3ω 6 12 4 2
 
1 13 7 5 9 2 67 3
+ I16 − + − ω− ω − ω
8ω 2 16ω 16 4 2 4
 
3 23 69 40 587 2
+ I17 − 2 + − − ω− ω
4ω 8ω 8 3 6
 
3 7 29 3799 140051 1193035 2 10388627 3
+ I18 − + − − − ω− ω − ω .
128ω 3 32ω 2 12ω 576 3456 5184 7776
(16)
Similarly, in Ref.[7] the following 7-denominator integral was evaluated

(−i)3 1 21 2 45
Z
d 4 k1 d 4 k2 d 4 k3 = π ζ(3) − ζ(5) ; (17)
π6 D1 D3 D4 D5 D6 D7 D8 12 4

9
in terms of the basic integrals one has

(−i)3
  
1 1 3
Z
4 4 4
d k1 d k2 d k3 = lim I5 −
π6 D1 D3 D4 D5 D6 D7 D8 ω→0 2ω 2
   
1 1 2 3 1 1
+ I6 − + − ω − 4ω − 16ω + I9 − +
2ω 2 4ω 2
   
2 14 445 3104 2 85177 3 3 19 39 2 79 3
+ I11 + + ω+ ω + ω + I12 − − ω − ω − ω
ω 3 18 27 162 16 32 16 8
 
11 53 299 715 2 1667 3
+ I14 − − − ω− ω − ω
8ω 16 16 8 4
   
1 1 1 1 2 1 3 1 1 1 1 2 1 3
+ I15 − + − ω − ω − ω + I16 − + ω+ ω + ω
4ω 8 8 4 2 4ω 8 8 4 2
 
1 7 10 40
+ I17 − − − ω − ω2
2ω 12 3 3
 
3 29 521 7691 2 111629 3
+ I18 + + ω+ ω + ω .
8ω 48 144 432 1296
(18)
It is to be stressed here that such relations have quite a broad generality, which extends
to cover scalar amplitudes occurring in graphs with different topology; the l.h.s. integral
of Eq.s(17),(18) was indeed evaluated in Ref.[7], dealing with “corner-ladder” graphs.
By exploiting our previous work, we can write a redundant number of such re-
lations expressing unknown basic integrals in terms of already known integrals (they
are not written here for the sake of brevity); in so doing we obtain a redundant set
of linear equations, which are easily solved in terms of the unknown basic integrals
(the redundancy provides with additional consistency checks). As a side remark, let
us observe that for each basic integral one must evaluate the first few terms (up to 7)
of its expansion in ω, see for instance I18 , so that the number of unknown quantities
increases, but on the other hand each relation, being an identity in ω provides with
several independent equations for the unknown.
As a result the following integral table is established

 
1 2
I1 = C(ω) 5ζ(5) − π ζ(3) + O(ω) ,
2


ζ(3) 13 4 1 2
I2 = C(ω) 2 − π − π + 10ζ(3)
ω 90 3
  
385 85 2 7 4 2 2 2
+ω ζ(5) − π ζ(3) − π − 82ζ(3) − 4π ln 2 + 16π + O(ω ) ,
2 6 15

10
  
1 7 31 2 4 4 103 25
I3 = C(ω) 3
+ 2
+ − π − ζ(3) + + ω 95ζ(5) − π 2 ζ(3)
3ω 3ω 3ω 15 3 3 3
 
1 184 44 235
− π4 − ζ(3) − 8π 2 ln 2 + π 2 + + O(ω 2 ) ,
15 3 3 3

ζ(3) 7 1
I4 = C(ω) 2 − π 4 + 2ζ(3) + π 2
ω 90 3
  
385 85 2 7 4 2 2 2
+ω ζ(5) − π ζ(3) − π − 82ζ(3) − 4π ln 2 + 16π + O(ω ) ,
2 6 15
  
1 3 1 1 2 55 4 14 7 95
I5 = C(ω) 3
+ 2
+ − π + − π 4 − ζ(3) − π 2 +
6ω 2ω ω 3 6 45 3 3 2
  
2 29 1351
+ ω − π 4 − 44ζ(3) − π 2 + + O(ω 2 ) ,
9 3 6
 
1 7 31 4 4 2 1 2 103 45 7
I6 = C(ω) 3
+ 2
+ − π + ζ(3) + π + +ω ζ(5) − π 2 ζ(3)
3ω 3ω 3ω 45 3 3 3 2 2
 
11 14 14 235
+ π 4 + ζ(3) − 4π 2 ln 2 + π 2 + + O(ω 2 ) ,
45 3 3 3
   
1 3 1 1 2 55 1 4 8 2 95 45
I7 = C(ω) 3
+ 2
+ − π + − π − ζ(3) − 2π + +ω ζ(5)
6ω 2ω ω 3 6 15 3 2 2
 
17 2 7 4 2 1 2 1351 2
− π ζ(3) − π − 50ζ(3) − 4π ln 2 + π + + O(ω ) ,
6 9 3 6

1 16 16 8 2
I8 = C(ω) − 3 − − + 2ζ(3) − π − 20
ω 3ω 2 ω 3
 
3 4 200 2 2 364
+ω − π − ζ(3) + 16π ln 2 − 28π +
10 3 3

46 64
+ ω 2 −126ζ(5) + 21π 2 ζ(3) + π 4 − 512a4 − ln4 2
15 3
 
80 2 2 2 2 3
− π ln 2 − 776ζ(3) + 168π ln 2 − 188π + 1244 + O(ω ) ,
3
  
2 10 1 1 2 26 16 11
I9 = C(ω) − 3
− 2
+ − π − − ζ(3) − π 2 − 2
3ω 3ω ω 3 3 3 3
 
13 4 248 2 73 2 398
+ω − π − ζ(3) + 16π ln 2 − π +
45 3 3 3

8 3 64
+ ω 2 −96ζ(5) − π 2 ζ(3) + π 4 − 512a4 − ln4 2
3 5 3
 
128 2 2 1888 2 2 3
− π ln 2 − ζ(3) + 160π ln 2 − 129π + 1038 + O(ω ) ,
3 3

11
   
1 5 1 2 2 26 7 10
I10 = C(ω) − 3
− 2
+ − π − 4 − ζ(3) − π 2 +
3ω 3ω ω 3 3 3 3
  
35 94 302
+ ω − π 4 − ζ(3) − π 2 + + O(ω 2 ) ,
18 3 3

  
1 37 43 139 1 4 19 773
I11 = C(ω) + + + 2ζ(3) + + ω − π + ζ(3) −
2ω 3 24ω 2 16ω 96 10 3 64

447 53 67 235
+ ω2 − ζ(5) + π 2 ζ(3) − π 4 + ζ(3)
2 2 60 2
 
2 2 27869 3
+ 12π ln 2 − 18π − + O(ω ) ,
384

  
1 7 253 2501 64 2 59437
I12 = C(ω) 3 + + + +ω − π +
ω 2ω 2 36ω 216 9 1296

1792 256 2 2272 2 2831381
+ ω2 − ζ(3) + π ln 2 − π + )
9 3 27 7776

3 2752 4 8192 1024 4 3584 2 2
+ω π − a4 − ln 2 − π ln 2
135 3 9 9
 
63616 9088 2 49840 2 117529021 4
− ζ(3) + π ln 2 − π + + O(ω ) ,
27 9 81 46656

   
2 23 35 275 112 189 2 136 4
I13 = C(ω) 3 + + + + ω ζ(3) − + ω − π
ω 3ω 2 2ω 12 3 8 45
 
32 4 32 2 2 14917 3
+ 256a4 + ln 2 − π ln 2 + 280ζ(3) − + O(ω ) ,
3 3 48


1 7 25 8 5
I14 = C(ω) 3
+ 2
+ + ζ(3) −
3ω 6ω 12ω 3 24
 
2 28 959
+ ω − π 4 + ζ(3) −
15 3 48
  
2 7 4 50 10493 3
+ ω 48ζ(5) − π + ζ(3) − + O(ω ) ,
15 3 96

  
3 23 105 4 2 275 2 2 567
I15 = C(ω) + + + π + + ω 28ζ(3) − 8π ln 2 + 10π −
2ω 3 4ω 2 8ω 3 16 32

62
+ ω 2 − π 4 + 192a4 + 8ln4 2 + 16π 2 ln2 2 + 210ζ(3) − 60π 2 ln 2
45
 
145 2 14917 3
+ π − + O(ω ) ,
3 64

12
   
1 7 1 1 2 25 7 5
I16 = C(ω) 3
+ 2
+ π + + 4ζ(3) + π 2 −
2ω 4ω ω 3 8 6 16
 
16 4 25 959
+ω π + 14ζ(3) + π 2 −
45 12 32
  
2 8 2 56 4 5 2 10493 3
+ ω 72ζ(5) + π ζ(3) + π + 25ζ(3) − π − + O(ω ) ,
3 45 24 64

  
1 35 1 2 559 16 35 2 2737
I17 = C(ω) − − − π − + ω − ζ(3) − π +
6ω 2 36ω 3 216 3 18 1296
  
2 37 4 280 559 2 552041 3
+ω − π − ζ(3) − π + + O(ω ) .
45 9 108 7776

By using the above table in combination with the integration by parts identities, we
obtain the analytical value of the required integrals; from Eq.(15) we have for instance

(−i)3 (p · k2 )2
Z
d 4 k1 d 4 k2 d 4 k3 =
π6 D1 D2 D3 D4 D5 D6 D7 D8 (18)
1 2 3
= ζ(3)π 2 − 5ζ(5) − π 4 + ζ(3) − 2π 2 ln 2 + π 2 .
2 45 2
13
 In terms of the basic integrals, the (g-2) of the graph of Fig.1c), for instance, reads
    
7 1 19 1637
(g − 2)(Fig.1c)) = lim I1 + I2 − + − ω
ω→0 6 2ω 12 72
   
13 253 823 102797 2 979525 3 3 5 229
+ I3 − + − ω+ ω − ω + I4 − + ω
48ω 144 108 2592 3888 8ω 24 18
 
1 7 587 9133 2 340685 3
+ I5 − − − ω+ ω − ω
12ω 18 216 324 1944
 
1 55 827 19075 2 874721 3
+ I6 − + − ω+ ω − ω
24ω 72 108 648 3888
 
19 25 632 56983 2 1488295 3
+ I7 − + ω− ω + ω
24ω 18 27 648 1944
 
1 5 1585 36581 1051253 2 743606 3
+ I8 − + − ω+ ω − ω
8ω 2 9ω 216 1296 3888 729
 
11 565 467 11225 1629641 2 1068433 3
+ I9 − + − ω+ ω − ω
96ω 2 288ω 36 162 3888 432
 
11 613 187 234293 2 1054147 3
+ I10 − − − ω− ω + ω
72ω 216 324 3888 23328
 
1 1 2143 5767 2472821 2 20840363 3
+ I11 − 2 − − + ω− ω + ω
6ω 18ω 216 144 7776 11664

11 525 6601 577597 4981223 46026487 2
+ I12 − + − + − ω + ω
512ω 3 1024ω 2 2304ω 27648 41472 62208

1632171647 3
− ω
373248
 
7 13 2375 95 339883 2 60055 3
+ I13 + + − ω+ ω + ω
96ω 2 192ω 576 432 2592 3888
 
35 1051 29419 220349 264997 2 31472155 3
+ I14 − + − ω+ ω − ω
192ω 2 1152ω 1728 3456 486 11664
 
1 7 6229 187 1934333 2 1580447 3
+ I15 + + + ω+ ω + ω
4ω 2 36ω 432 27 3888 11664
 
11 1075 8179 10925 2919977 2 49397413 3
+ I16 − + − ω+ ω − ω
96ω 2 576ω 864 192 7776 23328
 
17 1429 12223 146749 15117413 2
+ I17 − + − + ω− ω
24ω 2 144ω 216 432 7776

11 1315 3107 292849 206467 181740889 2
+ I18 − 3
+ 2
− + − ω+ ω
512ω 1536ω 2304ω 6912 4608 124416

177573299 3
− ω .
93312
(19)

14
Similar expressions hold for the other graphs.
By using the above table Eq.s(2,3) are immediately obtained.
Graph 1a is ultraviolet divergent and requires renormalization. From the unrenor-
malized amplitude we obtain

1 215 1
ae (1a; not ren.) = − + ζ(5) − π 2 ζ(3)
16ω 24  3 
53 4 1 4 2 2


− π + 4 a4 + ln 2 − π ln 2 (20)
2160 24
1229 7 3571 2 133
− ζ(3) − π 2 ln 2 + π + .
144 6 2592 864

To obtain Eq.(1), one must account for the charge renormalization of the inserted 4th
order vertex amplitude, to be carried out by means of a suitable subtraction constant,
which will be indicated here as Z. For consistency with the previous analytic (g-2)
calculations, Z must correspond to on mass-shell renormalization. It turns out therefore
that Z is infrared divergent (this is the only part of the triple-cross (g-2) calculation
in which infrared divergences appear); again for consistency with previous work, the
infrared divergence is parametrized by giving to the photon a fictitious infinitesimal
mass λ. The ultraviolet divergences of the counterterm are still parametrized by means
of the n-dimensional regularization; the result for the renormalization counterterm of
the inserted 4th order graph reads

1 11
Z=− − π 2 + 2 − ln λ .
8ω 24

In order to renormalize the contribution of graph 1a, one must subtract from Eq.(20)
(2) (2)
F2 (0)Z where F2 (0) is the second order (1 loop) magnetic form factor, to be evaluated
in n-dimensions as well; as Z has a 1/ω singularity, use must be done of the value of
(2)
F2 (0) evaluated up to first order in ω,

(2) 1
F2 (0) = + 2ω .
2
(2)
By subtracting F2 (0)Z from Eq.(20), Eq.(1) is recovered.

15
Acknowledgements.
As the results presented in this note have been obtained by intensive use of the algebra
manipulating programs FORM and ASHMEDAI, we want to express our gratitude to
their authors, J. Vermaseren and M.J. Levine, for their help and advice in the early
stages of the work.

References

[1] S. Laporta and E. Remiddi, Phys.Lett. B356, 390 (1995).

[2] M. J. Levine and J. Wright, Phys.Rev. D8, 3171 (1973).
[3] T. Kinoshita, Phys.Rev.Lett. 75, 4728 (1995).
[4] M. J. Levine, E. Remiddi and R. Roskies, in Quantum Electrodynamics, edited by
T. Kinoshita, Advanced series on Directions in High Energy Physics, Vol. 7, (World
Scientific, Singapore, 1990), 162, see p.214-216. Note that the result quoted there in
p.215 for the graph L2 , which corresponds to Eq.(3) of R. Barbieri, M. Caffo and E.
Remiddi, Nuovo Cimento Lett.5, 769 (1972), is incorrect in that context and should
be replaced by Eq.(5) of that same reference, which reads − 293 19 335
72 + 27 ζ(2) + 144 ζ(3) .
We thank B. N. Taylor and P. Mohr for pointing out the oversight.
[5] S. Laporta and E. Remiddi, Phys.Lett. B265, 181 (1991).
[6] S. Laporta, Phys.Rev. D47, 4793 (1993).
[7] S. Laporta, Phys.Lett. B343, 421 (1995).
[8] T. Kinoshita, IEEE Trans. Instrum. Meas. 44, 498 (1995).
[9] M. E. Cage et al., IEEE Trans. Instrum. Meas. 38, 284 (1989) .
[10] R. S. Van Dick, Jr., P. B. Schwinberg and H. G. Dehmelt, Phys.Rev.Lett. 59, 26
(1987).
[11] see Ref.[4], Sec. 2.1, p. 167.
[12] K. G. Chetyrkin and F. V. Tkachov, Nucl. Phys. B192, 159 (1981);
F. V. Tkachov, Phys.Lett. B100, 65 (1981).
We acknowledge also a discussion with D. Broadhurst on this point.

16