Konishi Anomaly

Consider the SQED with massless charged fields A and B. Classically, it has an axial
symmetry A e+i A, B e+i B and hence a conserved axial current

Jax = A e+2gV A + B e2gV B, (1)

2
D 2 Jax = D Jax = 0 (classically). (2)

In components

Jax (x, , ) = ( ) j 5 (x) + other combinations of and , (3)

j 5 = A A + B B + (iA D A iAD A ) + (iB D B iBD B )
= 5 Dirac + bosonic. (4)

Eqs. (2) imply inter alia that the ordinary axial current j 5 (x) is conserved, j 5 = 0.

In the ordinary QED with a massless electron, the loop corrections destroy the conser-
vation of the axial current. Instead of j 5 = 0, we have the AdlerBellJackiw anomaly

g 2
j 5 = F F . (5)
16 2

The superfield analogue of this anomaly is the Konishi anomaly: instead of the classical
eqs. (2), the current superfield Jax satisfies

2 g2
14 D Jax = W W ,
8 2 (6)
2 g2
14 D Jax = 2 W W .
8

Before we derive the Konishi anomaly, lets verify that it includes the ordinary Adler
BellJackiw anomaly (5). Reversing eq. (3), we have

 
j 5 (x) = 14 D , D Jax (x, , )
 
==0
=

(7)
   i  2 2 
j 5 (x) = 14 D , D Jax (x, , ) = D , D Jax (x, , ) .
 
==0 16 ==0

1
For the RHS here, the Konishi anomaly (6) gives us

2 g2  2

D , D 2 Jax = 2
 
D (W W ) D (W W ) , (8)
2 2

hence
ig 2 h 2 2
i
j 5 = D (W W ) D (W W )
32 2 ==0
ig 2 ig 2
Z Z

= 2
d2 W W + 2
d2 W W
8 8
2 (9)
g
Z
= Im d2 W W
4 2
g2
 
F F


= + .
4 2 4
Up to a total divergence of the photino current , this is indeed the AdlerBellJackiw
anomaly (5).

Conversely, given the AdlerBellJackiw anomaly and SUSY, applying supersymmetry

transformations to both sides of eq. (5) gives us the other components of the Konishi anomaly.
In this way, one may derive eqs. (6) without messing with the superfield perturbation theory.
However, in these lecture notes I shall use the super-diagram approach.

Diagrammatically,

Jax [V ] = (10)

and the individual diagrams contributing to this amplitude look just like the (n + 1)vector
amplitudes, except for the plus sign in eq. (1). Naively, these diagrams should cancel each
2
other when we act with D 2 or D on the Jax this works similarly to the Ward identities
for the multi-photon amplitudes, cf. homework set #5. Unfortunately, the diagrams con-
tributing to (10) diverge, so we must regulate them first and only then check if they really
cancel each other or only seem to cancel. It turns out that all the UV regulators would spoil
the cancellation of some one-loop diagrams, but the net mis-cancellation is the same for any
UV regulator, hence the Konishi anomaly (6).

2
 In these notes, I am going to derive the anomaly eqs. (6) at the one-loop level using the

PauliVillars regulator. This means adding to the theory some very heavy fields with wrong
norm in the Hilbert space and hence wrong sign for each heavy loop. In our case, we add a
pair of wrong-norm charged chiral superfields X and Y , so the net Lagrangian becomes
Z
2
 
reg 4 1 +2gV 2gV
L = d 8 V D D D V
+ Ae A + Be B
Z   Z (11)
4 +2gV
+ d Xe X +Ye 2gV
Y d2 XY + H. c.

for some very large mass acting as a UV cutoff scale. Consequently, the regulated vector
and axial currents become

reg
Jvec = A e+2gV A B e2gV B + X e+2gV X Y e2gV Y, (12)
reg
Jax = A e+2gV A + B e2gV B + X e+2gV X + Y e2gV Y, (13)

and their conservation or non-conservation at the classical level follow from the field equations
for all the charged fields in the Lagrangian (11). Specifically,

2 2
   
14 D A e+2gV = 41 D
B e2gV = 0,
1 2 1 2
   
+2gV 2gV
4D X e = +Y, 4D Y e = +X,
    (14)
14 D 2 e+2gV A = 41 D 2 e2gV B = 0,
   
1 2 +2gV 1 2 2gV
4D e X = + Y , 4D e Y = + X,

hence classically
1 2 reg
4 D Jvec = +Y X XY = 0,
(15)
41 D 2 Jvec
reg
= + Y X XY = 0,

but
reg 2
14 D Jax = +Y X + XY = +2Y X,
(16)
14 D 2 Jax
reg
= + Y X + XY = +2 XY .

For the quantum theory, eqs. (15) mean that the vector current is indeed conserved. At the

The dimensional reduction like all flavors of dimensional regularization is difficult to apply to
amplitudes involving the tensor, so its rather inconvenient for calculating the anomalies.

3
same time, eqs. (16) tell us that

2 reg
14 D Jax (y, ) = +2 hXY (y, )i ,
(17)
14 D 2 Jax
reg
(y, ) = +2 XY (y, )




where the expectation values hXY i and XY vanish in the vacuum but may become non-
trivial when the EM fields are present. Below, we shall see that at the one-loop level

1 g 2
hXY i = W W + O(1/2 ), (18)
16 2
1 g 2
W W + O(1/2).


XY = 2
(19)
16

Consequently, in the limit eqs. (17) give a finite but non-zero result for the non-
conservation of the axial current, namely the Konishi anomaly (6).

Diagrammatically, the loop corrections to the hXY i can be summarized as

(20)

vertex there are no d4 or d4 x integrals, so the amplitude has form hXY i =

R R
Here at the
some composite superfield. Also, in that composite superfield we do not distinguish between
different V1 , . . . , Vn , hence overall combinatoric factor 1/n!. Finally, the loop (20) carries an
overall minus sign due to wrong norm of the PauliVillars fields X and Y .

By charge conjugation, the number of the external vector lines in the amplitudes (20)
must be even, so lets start with the two-vector case. At the one-loop level we have 6

4
diagrams, namely

= +

+ + (21)

+ +

Note combinatorics: the scalar line has a definite direction from X to Y , so the two
vertices in the first 4 diagrams are distinct. OOH, the two photons are identical, so we do
not add diagrams related to the first 4 by photon permutation; instead, we simply drop the
combinatoric factor 12 . But for the last two diagrams both identical photons come to the
1
same vertex, so we keep the 2 factor.

The PauliVillars fields X and Y in all the loops (21) are very heavy indeed their
mass || serves as the UV cutoff scale of the regulated theory. Since this mass is much larger
than any of the external momenta p, we may approximate
1 1
. (22)
(k + p)2E + ||2 2
kE + ||2

for all values of the Euclidean loop momentum k (k p, or k , or anything in-between).

In other words, we may approximate all the propagators in the loops (21) as having the same
momentum k, at least in the denominator. Thus, for the last two diagrams in eq. (21) we
have
2 2D2

D D
2 (4ig 2 V 2 ) ( )
d4 k

1 i
Z
4 16

hXY i5+6 = (23)

2 (2)4 k ||2 + i0
2 2 2
D D D 2
+ (4ig 2 V 2 ) ( )
16 4

where ( ) stands for (4) ( ) which should be evaluated at = after the action of

5
the spinor derivatives. Similarly, the first four diagrams (21) yield

Z 4  3
d k i
hXY i1+2+3+4 =
(2)4 k 2 ||2 + i0
2 2 2

D D2 D D2 D
(2igV ) (2igV ) ( )
4 16 16



2 2 2 2 2
D D D D D

(24)
+ (+2igV ) (+2igV ) ( )
16 16 4

.
2 2 D 2 2D 2
D D D
+ (+2igV ) (2igV ) ( )

16 4 16

2 2
D D 2 D
+ (2igV ) (+2igV ) ( )
4 4 4

Combining all six diagrams and factoring out the denominators and common numerator
factors, we obtain

3
ig 2 d4 k

1
Z
hXY i = F ( ) (25)
256 (2)4 k ||2 + i0
2

where

2 2 2 2 2 2 2 2 2
F = D V D2 D V D2 D + D D2V D D2 V D D D2 V D V D2D
2 2 2 2 2 2

(26)
16||2 D V D 2 V D 8(k 2 ||2 ) D V 2 D 2 D + D D 2 V 2 D .

Now lets simplify this formula. The three terms on the top line here can be combined
together as
2 2 2 2 2 2
D [D 2 , V ]D [D 2 , V ]D + D V D2D D2 V D (27)

2
where in the second term we may simplify D 2 D D 2 = 16k 2 D 2 . Consequently,

2 2 2 2 2
F = D [D 2 , V ]D [D 2 , V ]D + 16(k 2 ||2 ) D V D 2 V D
2 2 2 2

8(k 2 ||2) D V 2 D 2 D + D D 2 V 2 D (28)
2 2 2 2 2
= D [D 2 , V ]D [D 2 , V ]D 8(k 2 ||2) D [[D 2 , V ], V ]D .

6
Note that both terms on the bottom line here involve the commutators

[D 2 , V ] = 2(D V )D + (D 2 V ),
(29)
[[D 2 , V ], V ] = 2(D V )(D V ).

These commutators make some of the D operators act on the vector field V instead of the
( ) to the right of F . Consequently, in the second term in (28)

2 2 2 2
D [[D 2 , V ], V ]D ( ) = 2D (D V )(D V )D ( ) = 0 (30)

because none of the D acts on the ( ). As to the first term in (28), we need two D
operators to act on the ( ), hence

2 2 2 2 2 2
D [D 2 , V ]D [D 2 , V ]D ( ) = 4D (D V )D D (D V )D D ( ) + 0
2 2 2
= 4D (D V )D (D V ) D D D ( ) + 0
2 2 2
= 4(D D V ) (D D V ) D D D ( ) + 0 (31)

= 64W W 8
= 512W W .

To summarize,

F ( ) = 512W W . (32)

The rest of the formula (25) is a straightforward integral

3
ig 2 d4 k

1
Z
I = +
256 (2)4 k 2 ||2 + i0
g 2 d 4 kE 1
Z
= + 4 2
256 (2) (kE + ||2 )3
Z 2
g 2 1 2 kE
(33)
= + dk E 2 + ||2 )3
256 16 2 (kE
0
g 2 1
= +
256 32 2 ||2
g2 1
= 13 2 .
2

7
Combining this result with eq. (32) immediately gives us

g 2
hXY i (2 vectors) = 2
W W . (34)
16

To complete the proof of eq. (18) we need to show that there are no multi-vector contri-
butions to the hXY i. Or rather, that all amplitudes (20) involving n = 4, 6, . . . vectors are
smaller than O(1/). Although the number of diagrams increases rather rapidly with n
for example, for n = 4 there are 54 one-loop diagrams one can prove by induction that
all the vector fields appear in the analogue of F only in commutators [D 2 , V ] or multiple
commutators. Consequently, although there are up to 2n D operators in the loop, at least
n of them act on the vector fields while two more have to act on the ( ). This leaves
us no more then n 2 Ds to anticommute with the Ds and produce powers of the loop
momentum in the numerator. Altogether, the loop integral looks like

2 , ||2 )
d4 kE Nn (kE
Z
In = 2 + ||2 )n+1 (35)
(2)4 (kE

where Nn in the numerator is some polynomial of degree (n 2)/2. By the power-of-

momentum counting,

In ||4+(n2)2(n+1) =
||n
=

(36)
finite for n = 2,

for , In
0 for n > 2.

In other words, all the multi-vector terms eq. (18) are sub-leading in the limit and
only the two-vector term contributes to the Konishi anomaly

2 reg g2
41 D Jax = +2 hXY i W W . (37)
8 2

8
 Similar arguments show that

 g 2
XY (2 vectors) = 2
W W . (38)
16

while the multi-vector contributions to eq. (19) carry sub-leading powers of 1/||, hence

g2
41 D 2 Jax
reg
= +2 XY 2 W W .


(39)
8

The details are left as an exercise to the students.