QFT Final Presentation 1: Path Integrals and Anomalies
Shu-Heng Shao
(Dated: June 24, 2010)
I. SETUP
In this note we review the chiral anomaly in Fujikawa’s method, based on references [1][2][3][4]. We begins by
considering a fermion multiplet ψ(x) of massless spinor fields with a set of gauge fields Aaµ (x). The Lagrangian reads
/
L = iψ Dψ, (1)
where
Dµ = ∂µ − igAµ = ∂µ − igAaµ T a . (2)
This Lagrangian has a global symmetry, the chiral rotation:
ψ(x) → eiγ5 α ψ(x) (3)
iγ5 α
ψ̄(x) → ψ(x)e . (4)
We’ve used iγ5 = iγ5 . Because of this, the transformation is not psudo-unitary, i.e. eiγ5 α eiγ5 α 6= 1. This is the main
reason why there is anomaly in chiral transformation but not in non-chiral one.
II. WHY ANOMALOUS?
Why is anomaly anomalous? There must be something weird that we call it “anomaly”. In fact, it is the failure of
the transition from a classical symmetry to a quantum one that we call anomaly. Take chiral symmetry for example.
Classically, we know the Noether current associated to this symmetry is conserved when the equation of motion is
imposed:
∂µ j5µ = 0, (5)
where
j5µ = ψγ µ γ5 ψ. (6)
Would this still be true when passing to the quantum picture? Let’s consider the partition function Z:
Z
Z = DψDψ̄ eiS[ψ,ψ] , (7)
where the action S is built out of (1). If we make a infinitesimal change of variables
ψ 0 (x) = (1 + iγ5 α(x))ψ(x) (8)
ψ̄ 0 (x) = ψ(x)(1 + iγ5 α), (9)
the partition function is, of course, still the same
Z
0 0
Z= Dψ 0 Dψ̄ 0 eiS[ψ ,ψ ] . (10)
Note that here the α(x) is spacetime-dependent. This is the standard setup for the Schwinder-Dyson equation. Since
the Lagrangian is not invariant under the local chiral transformation, an additional term appears in S[ψ 0 , ψ 0 ]:
Z
S[ψ , ψ ] = S[ψ, ψ ] − d4 xj5µ (x)∂µ α(x).
0 0 0 (11)
2
So (10) becomes
Z Z
0 µ
Z= Dψ Dψ̄ 0 1 + i d x ∂µ j5 (x)α(x) eiS[ψ,ψ] .
4
(12)
Now, if we assume the measure DψDψ̄ is invariant, combine with the original expression for Z, and the fact that α(x)
is arbitrary, we would be led to conclude
h∂µ j5µ (x)i = 0. (wrong!) (13)
But this is wrong, as indicated above, since the measure is not invariant under the change of variables. The remaining
part of this note will be devoted to work out the explicit change of the measure.
III. FUJIKAWA’S METHOD
The measure for the unprimed and primed fields are related by the determinant of the Jacobian:
−1 −1
Dψ 0 Dψ 0 = (detJ) detJ DψDψ. (14)
Note that it is J −1 instead of J since we are dealing with fermion field. In order to evaluate the Jacobian, we first
introduce a complete, orthonormal eigenbasis ϕκ for the Dirac operator iD / [6]:
/ κ (x) = λκ ϕκ (x)
iDϕ (15)
If we expand ψ(x) and ψ(x) on the above bases,
X
ψ(x) = aκ ϕκ (x), (16)
κ
the path integral measure can then be defined as
Y
DψDψ = daκ daκ . (17)
κ
Under (8) and (9), the coefficients aκ transform as
XZ X
a0κ = d4 xϕ†κ (x)(1 + iα(x)γ5 )ϕλ (x)aλ = (δκλ + Cκλ )aλ . (18)
λ λ
The Jacobian is
!
−2 −2
X
J −2 = J = (det(1 + C)) = exp(−2trlog(1 + C)) = exp −2 tr(Cκκ )
κ
Z
≡exp i d4 xα(x)A(x) , (19)
where A(x) is the anomaly function:
X
A(x) = −2tr[ ϕ†κ (x)γ5 ϕκ (x)] = −2tr[γ 5 ]δ(x − x). (20)
κ
In the last line we’ve used the cyclic property of the trace and the completeness of the eigenbasis for the Dirac
operator. This expression is completely formal before any regularization. It is possibly not zero, since we have tr[γ5 ]
as well as δ(x − x), which might turns out with a finite value.
It is important to regularize in a gauge-invariant way, otherwise the final result would not be gauge-invariant. In
addition, we hope the regulator should depend only on the eigenvalues of the Dirac operator. The only regulator
meets these criteria is
h i
2
X
A(x) = −2tr[ ϕ†κ (x)γ5 ϕκ (x)] = −2 lim tr γ5 f (−D/ x /M 2 ) δ(x − y)|x→y . (21)
M →∞
κ
3
M is some large mass which will be sent to infinity at last. The regulator function, f (s) is only subjected to the
conditions
f (0) = 1, f (∞) = 0,
0
sf (s) = 0 at s = 0 and s = ∞. (22)
Now we could write
d4 k
Z h i
2 ˙
A(x) = −2 4
tr γ5 f (−D / x /M 2 ) eik(x−y)
(2π) x→y
Z 4
d k n o
/ 2 /M 2
= −2 tr γ5 f −[i/ k + D]
(2π)4 x→y
Z 4
d k n o
= −2M 4 tr γ 5 f −[i/k + /
D/M ] 2
, (23)
(2π)4 x→y
where in the last line we’ve shift the momentum k to kM . In this way, we could expand f −[i/ /
k + D/M ]2 with
respect to large M . Before doing so, we first need to expand −[i/ /
k + D/M ]2 :
2
{/ /
k , D} /
D
−[i/ /
k + D/M ]2 = k 2 − i − . (24)
M M
In expanding with respect to M , we have two criteria to choose what we really need. First, any term with order of
1/M higher than 4 will be sent to zero in the limit M → ∞. Second, only terms with at least four gamma matrices
survive the trace. Also note that {/ / = −2k · D contains no Dirac matrices actually. With these two constraints,
k , D}
it is clear that the only nonvanishing term in the expansion is
d4 k 00 2
Z h i
4
A(x) = − f (k ) tr γ 5 /
D . (25)
(2π)4
First we focus on the k-integral. After a Wick rotation, the integral becomes
Z Z ∞ Z ∞
d4 k f 00 (k 2 ) = iπ 2 dssf 00 (s) = −iπ 2 dsf 0 (s) = iπ 2 , (26)
0 0
where we’ve used condition (22) to get the last equality. The regulator-independence is quite clear here; we only need
to require its limiting values.
Next let’s evaluate the trace:
2 1 1
D/ = Dµ Dν [γµ , γν ] + Dµ Dν {γµ , γν }
2 2
2 ig µν
= −D − F [γµ , γν ], (27)
4
where we’ve used [Dµ , Dν ] = −igF µν with F µν understood as matrix-valued. The first term is no longer important,
since it contains no gamma matrices. Using
h i
tr γ5 [γµ , γν ][γρ , γσ ] = −16iµνρσ , (28)
we finally have
1
A(x) = µνρσ Faµν (x)Fbρσ (x)tr[T a T b ]. (29)
16π 2
This quantity is known as the Chern-Pontryagin density.
Coming back to (12), we wish to see its implication to the classically conserved current j5µ . Together with (14) and
(19), we have
Z Z
µ
Z = DψDψ̄ 1 + i d x ∂µ j5 (x) + A(x) α(x) eiS[ψ,ψ] .
4
(30)
Since α(x) is arbitrary, we have
g2
h∂µ j5µ (x)i = − µνρσ Faµν (x)Fbρσ (x)tr[T a T b ]. (31)
16π 2
4
IV. GENERALIZATION
The previous calculation is restricted to the chiral symmetry of a fermion in some specific non-Abelian group
representation. Now let’s generalize this result to some other kinds of symmetry. Consider a set of fermions ψn (x)
which acts non-chirally with the non-Abelian gauge field Aµa . If the Lagrangian, whose form is not important here,
has a symmetry
ψn (x) → exp [iγ5 αt]nm ψm (x), (32)
with tnm a hermitian matrix acting on the internal space indices. Since it is a global symmetry, by Noether theorem,
there must exist a conserved current j5µ , which appears under the local transformation of the action:
Z
δS = − d4 x j5µ (x)∂µ α(x). (33)
Following exactly the same procedure in the previous sections, adding an extra t right after every γ5 matrix, one could
show that the anomaly function A now has the form:
1
A(x) = µνρσ Faµν (x)Fbρσ (x)tr[tT a T b ], (34)
16π 2
where the trace should also be taken upon the matrix t. Therefore, one finds
g2
h∂µ j5µ (x)i = − µνρσ Faµν (x)Fbρσ (x)tr[tT a T b ]. (35)
16π 2
We’ll see a direct application of this formula in the next section.
V. PION DECAY
Let’s consider the following Lagrangian consisting of ud quarks with electromagnetic field Aµ :
2e e
L = iu(∂/ − i / + id(∂/ + i A)d
A)u /
3 3
/ 2×2 − ieQA)ψ,
≡ iψ(∂I / (36)
where
2
u 3 0
ψ= and Q = . (37)
d 0 − 31
And the (32) in the previous section is now the charge-neutral chiral symmetry:
u u
→ exp iγ5 α σ 3
ψ= , (38)
d d
with σ a being the Pauli matrices.
Note that we’ve intentionally hide the SU(3) color gauge fields, since the charge-neutral symmetry is not broken
by the anomaly with only the QCD interaction present, so will be of no interest to us. Now put these things back to
(35), one gets
e2
A(x) = µνρσ F µν (x)F ρσ (x)tr[σ 3 Q2 ]
16π 2
Nc e 2
= µνρσ F µν (x)F ρσ (x). (39)
48π 2
Nc = 3 is the color number. Recall that under a local transformation
u 3
u
→ exp iγ5 α(x) σ , (40)
d d
5
the path integral measure transforms as
Z
4
DψDψ → exp i d xα(x)A(x) DψDψ. (41)
This is as if the effective Lagrangian Lef f transforms as Lef f → Lef f + α(x)A(x) under (40). To compensate for this
additional term, we must add another term to the effective Lagrangian which transforms the same way as above.
Under the transformation (40), the pion field transforms as [7]
δπ 0 = αFπ , (42)
with Fπ = 184MeV is the pion decay amplitude. So a good candidate for the required additional term is
π 0 (x)A(x) Nc e 2
= µνρσ F µν (x)F ρσ (x)π 0 (x)
Fπ 48π 2 Fπ
= gµνρσ F µν (x)F ρσ (x)π 0 (x), (43)
with the coupling constant g defined as
Nc e 2
g= . (44)
48π 2 Fπ
By standard Feynman diagram calculation, the pion decay rate could be calculated to be
m3π g 2
Γ(π 0 → 2γ) = . (45)
π
Put (44) into the above formula one gets
2
Nc
Γ(π 0 → 2γ) = × 1.11 × 1016 s−1 , (46)
3
while the observed value is
Γ(π 0 → 2γ) = (1.19 ± 0.08) × 1016 s−1 , (47)
which is in good agreement if and only if Nc = 3. This is one of the first firm evidence that there are three colors of
quarks.
VI. ATIYAH-SINGER INDEX THEOREM
As another implication, we’ll see a simple version of the Atiyah-Singer index theorem in the context of anomaly.
To demonstrate the theorem, it is necessary to Wick rotate the path integral into a Euclidean one, so that the Dirac
operator is hermitian [8]. Consider (21):
" #
2
X
2
/ /M ) †
A(x) = −2 lim tr γ5 f (−D ϕκ (x)ϕκ (x)
M →∞
κ
X †
f (λ2κ /M 2 )
= −2 lim ϕκ (x)γ5 ϕκ (x) . (48)
M →∞
κ
/ = −iDγ
The crucial observation here is that γ5 iD / 5 and the hermicity of iD,
/ so that
Z Z Z Z
†
−λκ d xϕκ γ5 ϕκ = d xϕκ (iD)γ5 ϕκ = d x (iD)ϕκ γ5 ϕκ = λκ d4 xϕ†κ γ5 ϕκ .
4 † 4 † 4
/ / (49)
That is, for λκ 6= 0,
Z
d4 xϕ†κ γ5 ϕκ = 0. (50)
6
/
Then if we integrate A(x), only those ϕκ with zero eigenvalues contribute. Furthermore, since γ5 commutes with iD
in this zero eigenspace, one can find simultaneous orthonormal eigenfunctions ϕu and ϕv with
/ u = 0, γ5 ϕu = ϕu ,
iDϕ
/ v = 0, γ5 ϕv = −ϕv .
iDϕ (51)
d4 xA(x) becomes
R
In this way
Z
d4 xA(x) = −2(n+ − n− ), (52)
where n± are the number of left- and right-handed zero modes of the Dirac operator. Combine with (29),
Z
1
− d4 x µνρσ Faµν (x)Fbρσ (x)tr[T a T b ] = n+ − n− . (53)
32π 2
This is the celebrated Atiyah-Singer index theorem. One immediate result from this theorem is that the LHS
of (53) cannot change continuously with the gauge field, but only by integers, so this integral depends only on the
topology of the gauge field.
[1] S. Weinberg, The Quantum Theory of Fields II, Cambridge.
[2] M. Srednicki, Quantum Field Theory, Cambridge.
[3] M. Peskin, D. Schroeder, An Introduction to Quantum Field Theory, Westview.
[4] K. Fujikawa and H. Suzuki, Path Integrals and Quantum Anomalies, Oxford.
[5] In this note the gamma matrices are in the Weyl representation, and the metric ηµν = (−1, 1, 1, 1). Also 0123 = 1 and
γ5 = diag(−I, I).
[6] The existence of such basis is guaranteed by the hermicity of the Dirac operator in the Euclidean spacetime. We shall not
elaborate on this subtlety further. For details, see Weinberg II.
[7] This transformation is not completely clear to me in the context, but one could get this result from the Gell-mann-Levy
linear sigma model. See Georgi (2.6.12). In linear sigma model, under charge-neutral chiral isospin rotation, δπ 0 = ασ. If
the chiral symmetry is spontaneously broken, then hσi = Fπ , so δπ 0 = αFπ .
[8] In order that the Dirac operator iD / to be hermitian rather that D,
/ one actually has to Wick rotated the spatial coordinates,
since it is γ 0 in the Weyl representation that is already hermitian.
