1

0
0
1

Figure 1: Four-point effective photon vertex emerging from the electron loop. The wiggly and
solid lines denote photons and electrons, respectively.

1 Euler-Heisenberg Lagrangian
1.1 The role of symmetry
Consider QED for momenta/energies much smaller than the electron mass. According to the
decoupling theorem, the only relevant degrees of freedom in the effective theory will be photons.
Consequently, the effective Lagrangian of the theory should be constructed from the photon
field Aµ . This is not a theory of free photons: the corresponding Lagrangian contains vertices
with 4,6,. . . photons. These vertices describe interactions which in the original theory are
mediated by closed electron loops (see Fig. 1).
In order to construct the effective Lagrangian, one could write down all possible terms,
which can be built using the field Aµ . At the next step, the couplings in front of these terms
should be matched to the underlying theory – QED. Here one arrives at the central question:
what is a criterion for possible terms? In short, one has to follow the following guidelines:
• Use only those fields that correspond to the relevant degrees of freedom at a given energy.
• Respect all symmetries. For example, Lorentz invariance and the discrete C, P, T sym-
metries of QED should be maintained. However, in addition to these general symmetries,
QED possesses a U(1) gauge symmetry. In this section we shall demonstrate that the
requirement of U(1)-invariance of the effective theory severely limits the number of the
possible terms. This simplifies the procedure of constructing the effective Lagrangian.
• Respect counting rules. At a given order in the low-momentum expansion, only the
operators with a pertinent mass dimension should be retained in the Lagrangian.
The rest of the present section is dedicated to the study of the implications of the U(1) gauge
symmetry in the construction of the effective Lagrangian. To this end, we find it convenient
to use the language of the path integral. In an arbitrary covariant gauge (see, e.g., [21]), the
Lagrangian of QED is given by
1 ξ
LQED = ψ̄(iγ µ (∂µ + ieAµ ) − me )ψ − Fµν F µν − (∂ µ Aµ )2 , (1)
4 2

1
where ψ and Aµ are the electron and the photon fields, respectively, Fµν = ∂µ Aν − ∂ν Aµ is the
electromagnetic field tensor, and ξ denotes the gauge fixing parameter. Observables (e.g., the
S-matrix elements) do not depend on ξ, but the Green’s functions do.
The generating functional of the Green’s functions in QED is given by
Z  Z 
4 µ
Z(j, η, η̄) = dψdψ̄dAµ exp i d x(LQED + η̄ψ + ψ̄η + jµ A ) , (2)

where jµ and η denote external sources for the photon and electron fields, respectively. The
Green’s functions are obtained in the usual manner, namely by differentiating the generating
functional with respect to the sources and, at the end, letting these sources vanish. Since we
are interested in the derivation of the effective Lagrangian for the photons only, we may put
η = η̄ = 0 from the beginning. The generating functional depends then on the argument jµ
only, and we can write

Z(j) = Z(j, η, η̄)

η=η̄=0
 Z  
1 ξ µ
Z
4 µν 2 µ
= dAµ exp i d x − Fµν F − (∂ Aµ ) + Leff (Aµ ) + jµ A , (3)
4 2
where
 Z  Z  Z 
4 4 µ
exp i d xLeff (Aµ ) = dψdψ̄ exp i d xψ̄(iγ (∂µ + ieAµ ) − me )ψ . (4)

Now, let us focus on the role of gauge invariance. It is straightforward to see that the integrand
in Eq. (4) is invariant under the gauge transformations
1
ψ(x) 7→ e−iα(x) ψ(x) , ψ̄(x) 7→ ψ̄(x)eiα(x) , Aµ 7→ Aµ + ∂µ α(x) . (5)
e
Here, α(x) denotes the parameter of the gauge transformation.
Consequently, assuming that the path integral measure is also invariant with respect to the
gauge transformations1 , and performing these transformations in Eq. (4), we easily obtain
 Z   Z 
4 4
exp i d xLeff (Aµ ) = exp i d xLeff (Aµ + ∂µ α) . (6)

This means the effective Lagrangian Leff (Aµ ) is gauge-invariant, i.e., depends only on the gauge-
invariant field tensor
Leff (Aµ ) = Leff (Fµν ) . (7)
1 . Q
At the first glance, this seems self-evident, since dψdψ̄ = x dψ(x)dψ̄(x) = x (e−iα(x) dψ(x))(eiα(x) dψ̄(x)).
Q
However, a certain care is needed performing the continuum limit, where the number of integration variables
tends to infinity. One, in particular, needs to regularize the ultraviolet divergence emerging in this limit, and
remove the regularization at the end of the calculations. In a given particular case, this can be done without a
problem, justifying the assumption about the gauge-invariance of the fermionic measure. However, if the gauge
transformation contains γ5 , the fermionic measure is, in general, no more gauge-invariant, giving rise to the
so-called anomalies. In the following, we shall consider this issue in detail.

2
Note that the gauge invariance naturally leads to the modification of the counting rules: since
Fµν contains field derivatives, insertions of Fµν into the loop diagrams result in the suppression
of the loop corrections at low energies.
The non-linear contributions to the Lagrangian arise first at O(m−4 e ). To this order, there
are only two such terms, consistent with all symmetries:
α2
 
1 µν µν 2 µν 2
Leff = − Fµν F + 4 c1 (Fµν F ) + c2 (Fµν F̃ ) + O(m−6 e ), (8)
4 me
where F̃ µν = ǫµναβ Fαβ . The overall factor m−4e appears on dimensional grounds, and the factor
2 2
α , where α = e /(4π) is the fine-structure constant, appears because this term couples with
four photons, each carrying a factor e. So, to this order, only two constants c1 , c2 have to be
determined from matching to QED.
Irrespective of the actual values of these constants, one may investigate, e.g., the dependence
of the photon-photon scattering cross section on photon energy E at E ≪ me . From the explicit
form of the Euler-Heisenberg Lagrangian given in Eq. (8), it is straightforward to conclude that
the scattering amplitude behaves like
α2 E 4
A2γ ∼ , (9)
m4e
where the factor E 4 stems from the four derivatives. The cross section behaves as
 2 4 2
α E 1 α4 E 6
σ2γ ∼ = . (10)
m4e E2 m8e
Note that, in the above expression, the phase-space factor E −2 is established on purely dimen-
sional grounds: in the absence of a photon mass, the photon energy E is the only dimensionful
parameter, on which the phase space factor can depend.
In the subsequent section we shall discuss the matching of the coefficients c1 , c2 to the
underlying theory. The direct method, based on the matching of Feynman integrals, turns out
to be very cumbersome. We shall see that using path integral methods allows one to achieve
the goal with a substantially smaller effort.
Historical note: While Euler, Heisenberg and Kockel analyzed light-by-light scattering using
effective field theory as described (this might in fact be the first use of an EFT) in the mid
1930ties, the full calculation of this process based on the finite sum of box diagrams in full
QED was only be performed by Karplus and Neumann in 1951 [23]. In fact, the low energy
result of Karplus and Neumann limit exactly recovers the Euler-Heisenberg result. This is a
beautiful example that in case of scale separation the EFT approach is much more effective than
the calculation in the full theory. For a nice discussion on the history of the Euler-Heisenberg
approach, sere Ref. [24].

1.2 Matching of the couplings in the effective Lagrangian

Below, we generally follow the discussion of the question given in the textbook of Itzykson and
Zuber [21], which is based on the Fock-Schwinger proper time method. In the path integral by

3
Eq. (4) one may carry out Grassman integration over the variables ψ, ψ̄. The answer is given
by
 Z 
exp i d xLeff (Aµ ) = det(iγ µ ∂µ − eγ µ Aµ − me ) ,
4
(11)

so that

Leff (Aµ ) = −i ln det(D) , D = iγ µ ∂µ − eγ µ Aµ − me . (12)

In other words, calculating the determinant and expanding in powers of Aµ , one will reproduce
all terms of the effective Lagrangian.
The key observation that simplifies the calculations dramatically, consists in the following:
in order to establish the coefficients c1 , c2 , it suffices to consider the determinant for the constant
electric and magnetic fields E and B. Defining the quantities a and b so that
1 1
a2 − b2 = E2 − B2 = Fµν F µν , ab = EB = Fµν F̃ µν , (13)
2 4
It can be shown that (see below)
Z ∞  
1 ds −is(m2e −i0) 2 cosh(eas) cos(ebs) 1
−i ln det(D) = 2 e e ab − 2 . (14)
8π 0 s sinh(eas) sin(ebs) s

Expanding in powers of a, b and using Eq. (13), we obtain

Z ∞
e2 2 2 ds −is(m2e −i0)
−i ln det(D) = (E − B ) e
24π 2 0 s
 4 Z ∞
e 2 2 2 7e4 2 2
− 2
(E − B ) + 2
(EB) dss e−is(me −i0) + · · · . (15)
360π 360π 0

The ultraviolet divergence at s = 0 in the first integral can be removed by the renormalization
of the free-photon term ∼ Fµν F µν in the Lagrangian. The second term is finite. Performing
the integration over s in this term, we finally get

α2 7α2
−i ln det(D) = CFµν F µν + 4
(Fµν F µν 2
) + 4
(Fµν F̃ µν )2 + · · · , (16)
90me 720me

where C denotes an ultraviolet-divergent constant. From this equation, one may directly read
off the values of c1 , c2 :
1 7
c1 = , c2 = . (17)
90 360

4
1.3 The fermion determinant in a constant field∗
Below we give an explicit calculation of the fermion determinant in a constant field. Namely,
our final goal will be to derive Eq. (14), which was already used to match the coefficients c1 , c2
in the Euler-Heisenberg Lagrangian.
Subtracting a constant that does not depend on the field Aµ , we may define
 
µ −1
ln det(D̄) = ln det(D) − ln det(iγ ∂µ − me ) = Tr ln (i 6 ∂ − e A
6 − me )(i 6 ∂ − me ) , (18)

where “Tr” denotes the trace both in the x-space and in the space of the Dirac indices.
Using Cγµ C −1 = −γµT , where C = iγ 2 γ 0 , Eq. (18) can be rewritten as
 
2 2 2 2 −1
2 ln det(D̄) = Tr ln ((i 6 ∂ − e A6 ) − me )((i 6 ∂) − me ) . (19)

Further, using the relation

∞
α ds is(β+i0)
Z
ln = (e − eis(α+i0) ) , (20)
β 0 s
the equation (19) can be rewritten as
Z ∞  
ds −is(m2e −i0)
Z
4 is(i6∂ −e6A)2 is(i6∂ )2
2 ln det(D̄) = − e d x tr hx|e |xi − hx|e |xi , (21)
0 s
where “tr” denotes the trace over the Dirac indices only.
In order to further simplify this expression, we use the identity
e i
(i 6 ∂ − e A σµν F µν ,
6 )2 = (i∂µ − eAµ )2 − σµν = [γµ , γν ] . (22)
2 2
As mentioned above, we restrict ourselves to constant electric and magnetic fields
E3 = F 30 = −F 03 = a , B 3 = F 12 = −F 21 = b . (23)
All other entries in the tensor F µν are equal to 0. Therefore,
 
e µν b −ia
− σµν F = eσ3 ⊗ , (24)
2 −ia b
where σ3 stands for the 2 × 2 Pauli matrix, and
 
ise µν
tr exp − σµν F = 4 cos(ebs) cosh(eas) . (25)
2
Consequently,
∞ 
ds −is(m2e −i0)
Z Z
2
ln det(D̄) = −2 e d x cos(ebs) cosh(eas)hx|eis(i∂µ −eAµ ) |xi
4
0 s

is(i∂µ )2
− hx|e |xi . (26)

5
In order to calculate matrix elements in Eq. (26), define

pµ = i∂µ , [xµ , pν ] = −igµν . (27)

Further, the electromagnetic field in our case is

A0 = A2 = 0 , A1 = −bx2 , A3 = ax0 (28)

Then,
.
H = (pµ − eAµ )2 = p20 − (p1 + ebx2 )2 − p22 − (p3 − eax0 )2
.
= eip0 p3 /ea eip1 p2 /eb (p20 − e2 a2 x20 − p22 − e2 b2 x22 )e−ip1 p2 /eb e−ip0 p3 /ea = H03 + H12 , (29)

and

hx|eisH |xi = hx0 x3 |eisH03 |x0 x3 ihx1 x2 |eisH12 |x1 x2 i . (30)

.
In the equation above, we use the notation |xi = |x0 x1 x2 x3 i = |x0 i ⊗ |x1 i ⊗ |x2 i ⊗ |x3 i.
Calculating the matrix elements in Eq. (30) separately, for the first one we get

dp0 dp3 dp′0 dp′3 dq0 dq3 dq0′ dq3′ ix0 (p0 −p′0 )+ix3 (p3 −p′3 )
Z
isH03
hx0 x3 |e |x0 x3 i = e
(2π)8
2 2 a2 q 2 )
× hp0 p3 |eip0 p3 /ea |q0 q3 ihq0 q3 |eis(p0 −e 0 |q0′ q3′ ihq0′ q3′ |e−ip0 p3 /ea |p′0 p′3 i . (31)

Using the relations

hp0 p3 |e±ip0p3 /ea |q0 q3 i = e±ip0 p3 /ea (2π)2 δ(p0 − q0 )δ(p3 − q3 ) ,

2 2 a2 q 2 ) 2 2 a2 q 2 )
hq0 q3 |eis(p0 −e 0 |q0′ q3′ i = (2π)δ(q3 − q3′ )hq0 |eis(p0 −e 0 |q0′ i , (32)

we easily get
ea
Z
isH03 2 2 a2 x2 )
hx0 x3 |e |x0 x3 i = 2 dp0 hp0 |eis(p0 −e 0 |p0 i . (33)
4π

In order to calculate the matrix element in Eq. (33), we consider the quantum-mechanical
problem of a harmonic oscillator with a Hamiltonian

1 2 ω02 2
hosc = p + x . (34)
2 0 2 0
The eigenfunctions of this Hamiltonian are labeled by an index n = 0, 1, · · ·
 
1
hosc |ni = ω0 n + |ni . (35)
2

6
Now, consider the matrix element
∞ Z   
ea ea X 1
Z
2ishosc 2
dp0 hp0 |e |p0 i = dp0 |hp0|ni| exp 2isω0 n +
4π 2 4π 2 n=0 2
∞   
ea X 1
= exp 2isω0 n + . (36)
2π n=0 2

In order to recover the original matrix element in Eq. (33), one has to substitute ω0 → iea.
Carrying out the summation over n, we finally arrive at the following result
ea
hx0 x3 |eisH03 |x0 x3 i = . (37)
4π sinh(eas)

Evaluating the second matrix element in Eq. (30) with the same method, we obtain

eb
hx1 x2 |eisH12 |x1 x2 i = . (38)
4πi sin(ebs)

Finally, substituting Eqs. (37) and (38) into Eqs. (30) and (26), we arrive at Eq. (14), which
was used for matching the couplings of the Euler-Heisenberg Lagrangian.

1.4 Blue sky

In this section, we shall consider another example with very long-wavelength photons. In
particular, we wish to address the scattering of long-wavelength photons by atoms. This process
goes under the name of the Rayleigh scattering. Below, we mainly follow the reasoning outlined
in Ref. [19].
Since the momentum transfer in the process is very small, the atoms can be described
non-relativistically. The pertinent free Lagrangian is given by

∇2
 
†
Latom = Ψ i∂t − Matom + Ψ, (39)
2Matom

where Ψ denotes the non-relativistic field

d3 k −ik0 t+ikx
Z
Ψ(x, t) = e a(k) . (40)
(2π)3

Here, a(k) denotes an annihilation operator of the atom.

The interaction Lagrangian of an atom with photons must be gauge-invariant. Since atoms
are electrically neutral, the only gauge-invariant objects from which this Lagrangian can be
constructed, are the atom fields, along with the electric and magnetic fields E and B. To the
lowest order, the interaction Lagrangian can be written as

Lint = a30 Ψ† (d1 E2 + d2 B2 )Ψ , (41)

7
where, from dimensional counting, a0 has the dimension of length, and d1 , d2 are dimensionless.
a0 is the typical size of the atom which is the only scale in the problem. We further assume
that d1 , d2 are of natural size, d1 , d2 ∼ O(1), so that the operator in Eq. (41) is indeed leading.
According to Eq. (41), the scattering amplitude Aγ−atom ∼ a30 E 2 , where E is the energy of
the photon. The scattering cross section is given by

σγ−atom = phase space × |A|2 ∼ phase space × a60 E 4 . (42)

From the above formula it follows that the phase space factor should be dimensionless and thus
energy-independent. Consequently, the cross section of the Rayleigh scattering is proportional
to the fourth power of the photon energy. This in particular, explains, why the sky is blue –
because the blue light is scattered more intensively by the atoms in the atmosphere than the
red light.

References
[1] D. B. Kaplan, “Effective field theories,” arXiv:nucl-th/9506035.

[2] A. Pich, “Effective field theory,” arXiv:hep-ph/9806303.

[3] L. B. Okun, “Leptons And Quarks,” Amsterdam, Netherlands: North-Holland (1982).

[4] T. Appelquist and J. Carazzone, “Infrared Singularities and Massive Fields,” Phys. Rev.
D 11 (1975) 2856.

[5] J. Polchinski, “Renormalization And Effective Lagrangians,” Nucl. Phys. B 231 (1984)
269.

[6] J. Smit, “Introduction to quantum fields on a lattice: A robust mate,” Cambridge Lect.
Notes Phys. 15 (2002) 1.

[7] M. E. Peskin and D. V. Schroeder, “An Introduction To Quantum Field Theory,” Reading,
USA: Addison-Wesley (1995) 842 p

[8] L. D. Landau, A. A. Abrikosov and I. Y. Pomeranchuk, Dokl. Akad. Nauk Ser. Fiz. 95
(1955) 497; ibid 95 (1955) 773; ibid 95 (1955) 1177.

[9] L. D. Landau and I. Y. Pomeranchuk, “On Point interactions in quantum electrodynam-

ics,” Dokl. Akad. Nauk Ser. Fiz. 102 (1955) 489.

[10] I. Y. Pomeranchuk, Dokl. Akad. Nauk Ser. Fiz. 103 (1955) 1005.

[11] R. F. Dashen and H. Neuberger, “How to Get an Upper Bound on the Higgs Mass,” Phys.
Rev. Lett. 50 (1983) 1897.

[12] I. Montvay and G. Münster, “Quantum fields on a lattice,” Cambridge, UK: Univ. Pr.
(1994) 491 p. (Cambridge monographs on mathematical physics)

8
[13] H. J. Rothe, “Lattice gauge theories: An Introduction,” World Sci. Lect. Notes Phys. 74
(2005) 1.

[14] M. Lüscher and P. Weisz, “Scaling Laws and Triviality Bounds in the Lattice φ4 Theory.
1. One Component Model in the Symmetric Phase,” Nucl. Phys. B 290 (1987) 25;
“Scaling Laws and Triviality Bounds in the Lattice φ4 Theory. 2. One Component Model
in the Phase with Spontaneous Symmetry Breaking,” Nucl. Phys. B 295 (1988) 65;
“Scaling Laws and Triviality Bounds in the Lattice φ4 Theory. 3. N Component Model,”
Nucl. Phys. B 318 (1989) 705.

[15] M. Gockeler, R. Horsley, V. Linke, P. E. L. Rakow, G. Schierholz and H. Stuben, “Is there
a Landau pole problem in QED?,” Phys. Rev. Lett. 80 (1998) 4119 [hep-th/9712244].

[16] D. J. E. Callaway, “Triviality Pursuit: Can Elementary Scalar Particles Exist?,” Phys.
Rept. 167 (1988) 241.

[17] I. M. Suslov, “Is φ4 theory trivial?,” arXiv:0806.0789 [hep-ph];

“Triviality, Renormalizability and Confinement,” arXiv:1102.4534 [math-ph].

[18] K. Symanzik, “Continuum Limit and Improved Action in Lattice Theories. 1. Principles
and φ4 Theory,” Nucl. Phys. B 226 (1983) 187.

[19] A. V. Manohar, “Effective field theories,” arXiv:hep-ph/9606222.

[20] J. C. Collins, “Renormalization. An Introduction To Renormalization, The Renormaliza-

tion Group, And The Operator Product Expansion,” Cambridge, Uk: Univ. Pr. (1984)
380p.

[21] C. Itzykson and J. B. Zuber, “Quantum Field Theory,” New York, Usa: Mcgraw-hill (1980)
705 P.(International Series In Pure and Applied Physics)

[22] C. B. Thorn, “Quark Confinement in the Infinite Momentum Frame,” Phys. Rev. D 19
(1979) 639.

[23] R. Karplus and M. Neuman, “The scattering of light by light,” Phys. Rev. 83 (1951) 776.

[24] G. V. Dunne, “The Heisenberg-Euler Effective Action: 75 years on,” arXiv:1202.1557

[hep-th].

[25] F. Yndurain, “Quantum Chromodynamics: An Introduction to the Theory of Quarks and

Gluons,” Texts and Monographs in Physics, Springer-Verlag, 1983.

[26] C. Vafa and E. Witten, “Restrictions on Symmetry Breaking in Vector-Like Gauge Theo-
ries,” Nucl. Phys. B 234 (1984) 173.

[27] T. Banks and A. Casher, “Chiral Symmetry Breaking in Confining Theories,” Nucl. Phys.
B 169 (1980) 103.

9
[28] C. Vafa and E. Witten, “Parity Conservation in QCD,” Phys. Rev. Lett. 53 (1984) 535.

[29] M. Gell-Mann, R. J. Oakes and B. Renner, “Behavior of current divergences under SU(3)
x SU(3),” Phys. Rev. 175 (1968) 2195.

[30] H. Hellmann, “Einfhrung in die Quantenchemie,” Leipzig, Franz Deuticke (1937), p. 285;
R. P. Feynman, “Forces in Molecules,” Phys. Rev. 56 (1939) 340.

[31] K. Huang and H. A. Weldon, “Bound State Wave Functions and Bound State Scattering
in Relativistic Field Theory,” Phys. Rev. D 11 (1975) 257.

[32] K. Fujikawa, “Path Integral Measure for Gauge Invariant Fermion Theories,” Phys. Rev.
Lett. 42 (1979) 1195;
K. Fujikawa, “Path Integral for Gauge Theories with Fermions,” Phys. Rev. D 21 (1980)
2848 [Erratum-ibid. D 22 (1980) 1499].

[33] K. Fujikawa and H. Suzuki, “Path integrals and quantum anomalies,” Oxford, UK: Claren-
don (2004) 284 p

[34] L. D. Faddeev and A. A. Slavnov, “Gauge Fields. Introduction To Quantum Theory,”

Front. Phys. 50 (1980) 1 [Front. Phys. 83 (1990) 1].

[35] M. F. Atiyah and I. M. Singer, “Dirac Operators Coupled to Vector Potentials,” Proc.
Nat. Acad. Sci. 81 (1984) 2597.

[36] E. Witten, “Current Algebra Theorems for the U(1) Goldstone Boson,” Nucl. Phys. B 156
(1979) 269.
G. Veneziano, “U(1) Without Instantons,” Nucl. Phys. B 159 (1979) 213.

[37] C. G. Callan, Jr., “Broken scale invariance in scalar field theory,” Phys. Rev. D 2 (1970)
1541;
K. Symanzik, “Small distance behavior in field theory and power counting,” Commun.
Math. Phys. 18 (1970) 227.

[38] G. ’t Hooft, “A Planar Diagram Theory for Strong Interactions,” Nucl. Phys. B 72 (1974)
461.

[39] S. Weinberg, “The U(1) Problem,” Phys. Rev. D 11 (1975) 3583.

10