v.2.2r20191120 *2020.5.

2#b3a87b14

QED IN STRONG ELECTROMAGNETIC BACKGROUNDS

– EFFECTIVE ACTION METHODS

CIPRIAN SORIN ACATRINEI

Department of Theoretical Physics,
Horia Hulubei National Institute for Physics and Nuclear Engineering,
Reactorului 30, RO-077125, P.O.B. MG-6, Bucharest-Magurele, Romania
E-mail:
Received December 11, 2018

Abstract. We review the formulation of field theory in the presence of strong

background fields, with particular emphasis on electron-positron pairs creation in quan-
tum electrodynamics. The effective action is defined and evaluated in the proper time
representation, using path integral methods. It is then used to write down and calcu-
late the probability for vacuum decay through the creation of electron-positron pairs.
Several applications and extensions are discussed.
Key words: effective action, external fields, path integrals.

1. INTRODUCTION

The theoretical study of Quantum Electrodynamics (QED) in strong external

fields received new relevance with the advent of high power lasers. In this context, a
review of a concise yet comprehensive approach to generic field theories in external
fields seems appropriate. The key ingredient, the effective action, can be evaluated
via path integral methods, in the so-called proper time representation [1–3]. The va-
cuum decay rate due to external electromagnetic fields, for example, can be expressed
in terms of the imaginary part of the effective action, leading to an exact pair creation
rate for constant backgrounds. The approach is nonperturbative, allowing for arbi-
trary field strengths. In addition to their relevance for physically important situations,
the methods expounded here are well suited for numerical simulations, path integrals
being central to calculations on the lattice. They provide an elegant way to add extra
degrees of freedom (we exemplify with stringy and dissipative ones), and are con-
ceptually attractive, as the effective action is evaluated by literally integrating out the
fermionic degrees of freedom, in authentic ”Wilsonian renormalization group” spirit.
Due to lack of space and/or competence, important topics had to be left out.
The derivative expansion, for instance, is a systematic way to go beyond constant
fields but complicates quickly at higher orders [4]. The semiclassical approach, ini-
tiated a long time ago [5], is another way to address nonconstant configurations. It
is also discussed in [4]. For a different perspective, related also to approximations in
Romanian Journal of Physics 64, 113 (2019)
Article no. 113 Ciprian Sorin Acatrinei 2

strong fields, one may enjoy the review [6]. The ubiquitous subject of Bogoliubov
coefficients was reluctantly left out, as it is easily available elsewhere [7].
Topics are ordered as follows. Following introductory tunneling arguments,
the adequate framework, relativistic quantum field theory (QFT) in external fields, is
introduced. The effective action is defined and evaluated using the Feynman-Fock-
Schwinger proper time representation and quantum mechanical (QM) path integral
methods. The rate of creation of electron-positron pairs is then related to the imag-
inary part of the effective action and calculated using that connection. The applica-
tions of the method are numerous - a calculation of the chiral anomaly illustrates the
versatility of the formalism. Extensions to other solvable backgrounds and generali-
zations involving extra (stringy, dissipative) degrees of freedom are also overviewed.
The Appendix reviews the path integral of a particle in a constant magnetic field.

2. PAIR CREATION IN QED

2.1. FLUCTUATIONS IN AN EXTERNAL ELECTRIC FIELD

Already in quantum mechanics we agree that the ground state displays fluctua-
tions. In QFT the ground state is the vacuum of the theory, and vacuum fluctuations
correspond to quanta appearing and disappearing. Yet, charge is conserved even
for virtual processes, consequently electron-positron pairs are constantly created and
annihilated in the QED vacuum, waiting for an external source of energy to allow
them out into the real world. When a pair emerges the electron and the positron ini-
tially separate, then reunite into annihilation. In the ”fluctuation time” ∆t before the
electron-positron loop closes, enough energy (∆E ≥ 2me c2 ) should be imparted to
the pair to ensure the energy conservation of a real process. Consistency with the
time-energy uncertainty relation requires
∆E∆t ≥ ~. (1)
Above, ∆t is the time interval of brief ’existence’ of the pair, whereas ∆E gives the
degree of energy non-conservation implied by the virtual process,
∆E = 2mc2 , (2)
where m is the electron mass. Again, ∆t is the time before the e− − e+ loop closes.
If an external electric field is applied, it will tend to separate the electron and the
positron. To achieve the break-up of the pair, it must convey the energy ∆E = 2mc2
to the pair during the time ∆t, i.e. before annihilation takes place. An electron of
charge e is imparted the mechanical work eE∆x upon being displaced by ∆x in an
electric field E, and similarly but with displacement in the opposite direction for the
positron. Assuming displacement with velocity close to the speed of light, one thus
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
3 QED in strong backgrounds Article no. 113

requires
∆x · (e E) ∼ ∆tc(e E) ' mc2 (3)
for the pair to break. Applying (1), (2) and dropping a factor of 2 from this qualitative
argument leads to a minimal break-up field Ecr given by
~
· c · (e Ecr ) = mc2 . (4)
mc2
~
Note that ∆x = mc is the Compton wavelength. All in all
m2 c3 V V
Ecritical ≡ Ecr = ∼ 1018 = 1016 . (5)
e~ m cm
This quantity, called he critical electric field, is a useful unit of comparison. In a
region with electric fields even remotely close to Ecr , the vacuum would become a
conductor.

2.2. TUNNELING VIEW

One can adapt more quantitative arguments from nonrelativistic quantum me-
chanics to obtain a better, actually quite accurate, estimate. In absence of external
fields, the Dirac equation leads to free energy levels and to the so-called Dirac sea.
The lowest positive energy level is at mc2 , while the highest negative level is at
−mc2 . The gap in between is a forbidden zone, of magnitude 2mc2 . One needs an
incoming photon with at least this energy to move an electron from the highest neg-
ative level to the lowest positive energy level (making an electron available), leaving
consequently a hole in the Dirac sea (the positron).
The above process is a perturbative one, as we can account for it in perturbation
theory. However, a pair can also be created by ’zero-frequency photons’, i.e. by a
constant electric field, although in a non-perturbative way, through tunneling. Indeed,
if a constant electric field is switched on, one has to add the space-dependent potential

V (x) = eEx (6)

to the Dirac equation. This changes accordingly the energy of an electron, making
it space-dependent. In particular, electrons previously assigned to negative energies
are shifted above zero in half of the physical space.
For instance, the highest (initially) negative energy reads now
−mc2 + eEx (7)
mc2 2mc2
and this expression is positive for x > eE and equal to +mc2 at xA = eE . In con-
sequence, the initially negative energy electron can tunnel from position xA (where
its energy is +mc2 ) to the position xB = 0 on the first energy level (where its energy
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
Article no. 113 Ciprian Sorin Acatrinei 4

is the same, namely +mc2 ); the pair is created. A more detailed analysis follows. To
be realistic, consider a potential with an electric field of finite extension in space:

 0, x < 0
V (x) = +Ex, 0 < x < L (8)
+EL, x > L


Denoting the energy by W , the available kinetic energy reads

p
W − V (x) = p~ 2 + m2 , (9)
from which
−p2x = −[W − eV (x)]2 + p2T + m2 , (10)
2 2 2
where pT = py + pz is the square of the transverse momentum. The turning points
x1 , x2 are the ones at which px = 0 and are given (assuming they are between 0 and
L) by
q
W + p2T + m2
[W − eEx]2 = p2T + m2 =⇒ x1,2 = . (11)
eE
The Gamow tunnelling factor reads
√
− 2 x2 dx m2 +p2T −[W −eEx]2
R
TW KB = e ~ x1 (12)
 
m2 +p2
T
R +1 √ W − eEx 
2
= e−2 ~eE −1 dy 1−y y= q (13)
2
m + pT 2

m2 +p2
T
= e−π eE (14)
This gives the transition probability per unit time and per unit cross section dydz, for
a given energy W . Integrating also over the energy W and the transverse momentum
p~T (recalling that the dispersion relation implies dW = eEdx and that there are two
possible polarisations for the electron) we obtain for the tunnelling probability
m2 +p2 e2 E 2 − |e|E~
πm2
Z
P robability eE T
=2 pT2 e−π eE =
d~ e . (15)
(time) × (V olume) 2π 4π 3
Restoring the c’s we get in the end
P robability e2 E 2 − πm 2 c3
= e ~|e|E . (16)
(time) × (V olume) 4π 3
Notice the E 2 dependence in the prefactor. The result (16) will be confirmed by the
subsequent field theory analysis, giving the first (and dominant) term of an infinite
series, whose nth term will provide the amplitude for the creation of n pairs in a
constant electric field. Notice also how well the exponential part blends with our
previously found critical electric field, see Eq. (5).
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
5 QED in strong backgrounds Article no. 113

3. EFFECTIVE ACTION

3.1. FORMULATION

The Lagrangian of QED couples a charged fermionic Dirac field ψ (represent-

ing the electrons and the positrons) with a real massless vector field Aµ (the electro-
magnetic field):
1
LQED = ψ̄(i 6 ∂ − m)ψ − Fµν F µν − eψ̄ 6Aψ (17)
4
The covariant derivative is defined as D = ∂ + ieA and the usual ”slash notation”
reads
i 6 ∂ − e 6A ≡ i 6D = iDµ γ µ . (18)
Greek indices µ, ν run through the values 0, 1, 2, 3 and the electromagnetic field
strength is related to the vector potential via Fµν = ∂µ Aν − ∂ν Aµ . The Lagrangian
(17) describes a significant part of the real world with unprecedented precision, in
spite of its apparent simplicity (or due to it!). Most results are obtained in the frame-
work of perturbation theory, which suffices to describe a huge number of significant
experimental results. Yet some aspects require going beyond the realm of small field
amplitudes. To achieve that for strong external electromagnetic fields, let us define
the vacuum transition amplitude in a classical (background, external) vector potential
Aµ as
R 4
Dψ̄Dψei d xψ̄(i6∂ −e6A−m)ψ
R
h0, +∞|0, −∞iA = R R
4
≡ eiSef f [A] . (19)
Dψ̄Dψei d xψ̄(i6∂ −m)ψ
Occasionally, this is called the vacuum persistence amplitude. It gives the amplitude
of transition between the vacuum state at time ti = −∞ and the vacuum at tf = +∞.
The proof of the first equality can be found in any field theory textbook using path in-
tegrals. The second equality is a definition. The logarithm of the transition amplitude
gives the effective contribution of fermions (which are completely integrated out) to
an ”effective” action of the external field Aµ . Athough this is the quantity denoted
by Sef f [A] in (19), it is understood that we can (and should, in the end) always add
to it the classical free electromagnetic field term − 41 Fµν F µν present in (17).
The numerator in (19) is obviously a normalization factor, that will be denoted
by 1/N . The fermionic field being a Grassmann (anticommuting) variable, the path
integral over it leads to
eiSef f [A] = N det (i 6D − m) = N eT r log(i6D−m) . (20)
P
The continuous counterpart of exp i log λi = Πi λi , namely
log det Ô = T r log Ô, (21)
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
Article no. 113 Ciprian Sorin Acatrinei 6

allows us to directly extract the effective action

i 6D − m
Sef f = −i T r log . (22)
i6∂ −m
The trace involves both space-time coordinates and Dirac indices. Overall traces will
be denoted by T r, space-time traces by Tr, Dirac traces by tr.
To understand better the significance of (22), let us interpret it in terms of Feyn-
man amplitudes:
 
i 6D − m 1
log = log det 1 − (e 6A) (23)
i 6∂ − m i 6∂ − m
 
i
= T r log 1 − (−ie 6A) (24)
i 6∂ − m
∞    n
X 1 i
= − Tr (−ie 6A) . (25)
n i 6∂ − m
n=1
The last expression can be rewritten in space-time, with SF the fermionic propagator,
as
∞  Z
X 1
− dx1 . . . dxn tr [−ie 6A(x1 )SF (x2 − x1 ) · · · − ie 6A(xn )SF (x1 − xn )] .
n
n=1
The n-th order term of the above sum displays a closed fermionic loop (since SF
propagates from x1 to x2 , etc., finally to xn and then back to x1 ) emitting n pho-
tons. The n1 factor comes from the cyclic (not permutation!) symmetry related to
the insertion of the electromagnetic field legs on the fermionic loop at the points
x1 , x2 , . . . , xn . In consequence we have here the full effective action for the electro-
magnetic field, with the fermionic field accounted for at the 1-loop level.

3.2. EVALUATION

One useful identity is

1
Tr log(i 6D − m) = Tr log(−i 6D − m) = Tr log(− 6D2 + m2 ) (26)
2
To prove this, expand in 6Dn and notice that
tr 6Dn = tr 6Dn γ52 = tr γ5 6Dn γ5 = tr(− 6D)n . (27)
Further on
i 6D − m 1 (i 6D − m)(−i 6D − m)
iSef f = Tr log = Tr log (28)
i6∂ −m 2 (i 6 ∂ − m)(−i 6 ∂ − m)
leading to
i 
Sef f = − Tr log(− 6D2 + m2 ) − log(− 6 ∂ 2 + m2 ) .

(29)
2
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
7 QED in strong backgrounds Article no. 113

One may imagine the arguments of the logarithms divided by some mass scale m2 ,
to stay adimensional under the log. Using σµν = 2i [γµ , γν ] we have
1
6D2 = Dµ Dν γµ γν = Dµ Dν ([γµ , γν ] + {γµ , γν }) (30)
2
i 1
= D D (ηµν + σµν ) = D2 + [Dµ , Dν ]σµν
µ ν
(31)
2 2
e
= D2 + F µν σµν (32)
2
which leads to
i h  e 
i
Sef f = − Tr log (iD)2 + m2 + F µν σµν − log (i∂)2 + m2 (33)
2 2
showing that the spin part conveniently separates at this (one-loop) level.

3.3. PROPER TIME REPRESENTATION

The proper-time representation of the effective action introduces a parameter

which resembles proper time. The fastest way is to use the standard formula
Z ∞ h
a ds is(b+i) i
ln = e − eis(a+i) ,  > 0,  −→ 0+ . (34)
b 0 s
It rewrites (33) as
Z ∞
i ds is(m2 −i) h is(i∂−eA)2 is e F µν σµν 2
i
Sef f = Tr e e e 2 − eis(i∂) 11 . (35)
2 0 s
One may also be less formal and start with the Klein-Gordon propagator between
two space-time points x and y,
d4 p eip(x−y)
Z
DF (x − y) = (36)
(2π)4 p2 − m2 + i
Using the familiar formula
Z ∞
i
= dseis(a+i) (37)
a + i 0
and writing transition elements between momenta p and the coordinates of a four
dimensional space, x,
hp|xi = e+ipx (38)
one rewrites the propagator as
Z ∞
d4 p
Z
2 2
DF (x − y) = 4
hy|pi dseis(p −m +i) hp|xi (39)
(2π)
Z ∞ 0
2
= dse−is(m −i) hy|e−isĤ |xi. (40)
0

(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14

Article no. 113 Ciprian Sorin Acatrinei 8

Above, Ĥ is a transition Hamiltonian in 4-dimensional space, essentially Ĥ = D2 .

Recalling that the variation of the one-loop partition function relates to the propaga-
tor, one reobtains (35) as follows. Denote with Ôm2 = p2 − m2 + i the inverse of
the propagator, and look at its dependence upon m2 :
Z ∞
1
δm2 Tr log Om2 = Tr δ 2 (Om2 ) = −i Tr dseis(Om2 ) δm2 (Om2 ) (41)
Om2 m 0
which leads to
Z ∞
ds is(O 2 )
Tr log(Om2 ) = − Tr e m + cst. (42)
0 s
Taking the constant to be zero, we recover (35). The interpretation is as follows: Eq.
(40) incorporates propagation taking place in any total ”proper time” s between 0 and
∞. For the 1-loop effective action however, it is irrelevant at which point one starts
around the loop, hence one divides by the length of the loop, an additional 1s ap-
pearing consequently in (35). One also notices the (−) sign, adequate for fermionic
loops.
The traces in (35) further split as
Z ∞
i ds is(m2 −i) h 2 e µν 2
i
e Tr eis(i∂−eA) trDirac eis 2 F σµν − Tr eis(i∂) trDirac 11
2 0 s
(43)
Now comes in the nice observation that the space-time trace can be represented as a
Feynman path integral
Z Z Rs
2 2
Tr eis(i∂−eA) = d4 x < x|eisĤ |x >= D4 Xei 0 dτ (Ẋ +eẊA) (44)
X(0)=X(s)
dX
where Ẋ ≡ and Ĥ is the Hamiltonian given by the KG operator (i∂ − eA)2 ,
dτ
hence (upon Euclideanization of x0 ) a Schrödinger operator in 4+1 dimensions, the
fifth one being the ”proper time” s. Consequently
Z s  
SP = dτ Ẋ 2 + eẊA (45)
0
is the action for a (4 + 1)-dimensional particle depending on a ”proper time” τ and
coupled in a minimal way to a vector potential A. There are several ways to compute
the trace in (44), the path integral is a convenient one.
For a free particle of mass M = 12 in d dimensions (d=4 in our case), we prove
in the Appendix that
!d/2
is(i∂)2 M = 21
Tr e = (46)
2πi~s
while in four dimensions the Dirac trace leads to trDirac 11 = 4. For a magnetic
background, splitting the four directions into two (x-y) coupled by B and two (t-z)
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
9 QED in strong backgrounds Article no. 113

free [later on, coupled by E], we have (cf. Appendix)


2 1 eBs
Tr eis(i∂−eA)  = (47)
B (4πi~s)2 sin eBs
whereas the Dirac trace reads
e µν σ
trDirac eis 2 F µν
= 4 cos seB, (48)
as can be shown choosing for instance the Weyl representation for the γ matrices. In
the end
i ∞ ds is(m2 −i)
Z  
1 eBs
Sef f = e · 4 cos eBs − 4 (49)
2 0 s (4πi~s)2 sin eBs
Although the expression is non-perturbative in B, to identify the (small-s) diver-
gences we expand in eB:
x x2
cos x ' 1 − x << 1.
sin x 3
2
In consequence, adding the counter-term + (eBs)
3 gets rid of the divergences. This
corresponds to charge renormalization in usual perturbation theory.
For a pure electric field E background one immediately obtains the result by
substituting B = iE in (49):
1 ∞ ds is(m2 −i)
Z  
E 1 eBs
Sef f = e · 4 cosh eEs − 4 (50)
2 0 s (4πi~s)2 sinh eEs
with a similar counter-term.
Any configuration of constant electromagnetic fields can be boosted to two
parallel electric E [couples directions t-z, say] and magnetic B [couples axes x-y
then] fields. The effective action for parallel electric and magnetic fields reads
Z ∞ 2
ieis(m −i)
   
E||B eBs eEs 1
Sef f = ds · cos eBs cosh eEs − 1
0 8π 2 ~2 s3 sin eBs sinh eEs i
(51)
The counter-term required within the square brackets above is seen to be (B 2 −
2 2
E 2 ) e 3s , the classical electromagnetic Lagrangian.
The previous boosting of Fµν actually amounts to block-diagonalizing it. It is
useful to introduce the (proper) Lorentz invariants that describe the electromagnetic
field:
F =E ~2 −B ~ 2, G = 2E~ · B.
~ (52)
They immediately provide the ”parallel” (eigen-)values
s√ s√
F 2 + G2 + F F 2 + G2 − F
E= ; B= (53)
2 2
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
Article no. 113 Ciprian Sorin Acatrinei 10

It is enough to substitute E −→ E and B −→ B in Eq. (51) to obtain the general

effective action for any 3+1 dimensional constant electromagnetic background.

4. PAIR CREATION RATE

4.1. REAL PARTICLES FROM IMAGINARY ACTIONS

For constant Fµν nothing should depend explicitly on time, hence can write
Γ
Sef f [A] ≡ (E + i ) × (time) ≡ Lef f [A] × (volume · time) (54)
2
Given that 1 − |h0, +∞|0, −∞iA |2 gives the probability for the system not to remain
in the ground state, and that we associate this transition with pair production, the pair
production probability (P P P ) is
P P P = 1 − |h0, +|0, −i|2 = 1 − e−Γ·time ' 1 − (1 − Γ · time) = Γ · time (55)
and the pair production rate (P P R) per unit volume is
Γ · time
PPR = = 2 Im Lef f (56)
volume · time
This is the quantity we are after: the imaginary part of the effective Lagrangian will
give us the rate at which vacuum breaking processes occur.

4.2. PAIR PRODUCTION RATE

Consider first the case of a pure electric field. Starting from (50), noting that
2
eism = cos(sm2 ) + i sin(sm2 ) one observes that, to obtain the imaginary part, one
can extend the integration range from −∞ to ∞ and divide by 2. The integrand has
poles at
nπ
sn = −i (57)
eE
and calculating

the integral by the method of residues, or directly using the formula
1 1
x−i = P x + iπδ(x) under the integral, one obtains (half) the pair production rate

∞ 
1 X eE 2 − nπm2

Im LE
ef f = e eE , (58)
8π nπ
n=1
the famous Schwinger result. Twice this expression gives the pair production rate,
according to our derivation above. One notices that the n = 1 term agrees completely
with the tunnelling approximation (16). Higher n terms, coming from the n > 1 poles
in (57), can also be interpreted as higher-order instanton contributions to the effective
action.
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
11 QED in strong backgrounds Article no. 113

For pure magnetic fields the effective action (49) has zero imaginary part. In
consequence, a constant magnetic field does not produce electron-positron pairs, as
expected since it has no way to impart energy to the particles.
For parallel electric and magnetic fields, Eq. (51) leads to the imaginary part
∞
1 X e2 EB
 
E||B B − nπm2
Im Lef f = coth nπ e eE . (59)
8π n2 π 2 E
n=1

For generic constant E~ and B~ fields, one recalls Eqs. (52, 53) and simply substitutes
E −→ E, B −→ B in (59). Situations exist in which the presence of a magnetic
field can increase the pair creation rate (which is non-zero only in the presence of
an electric field, however). This never happens in scalar QED, as will be seen in the
next section.

5. APPLICATIONS

5.1. CHIRAL ANOMALY

The (Euler-Heisenberg) effective action calculated in the previous section has

important applications. It reproduces the QED β-function, accounts for light-light
scattering at low energies, or allows for a quick computation of the chiral anomaly.
Due to lack of space we will focus on the chiral anomaly [8], to throw a different
light on an ever-fascinating phenomenon.
We are interested in the divergence of the chiral current Jµ5 ≡ ψ̄γ5 γµ ψ, which is
classically zero in the limit of massless fermions, m → 0. Another ingenious method
advocated by Schwinger in [1] is point-split regularization. To start, write
∂µ J5µ = ∂µ ψ̄(x) γ µ γ5 ψ(x) + ψ̄(x)γ µ γ5 ∂µ ψ(x)


(60)
 
∂ ψ̄(x2 ) ∂ψ(x1 )
= lim γ µ γ5 ψ(x1 ) + ψ̄(x2 )γ µ γ5 (61)
x1 ,x2 →x ∂xµ2 ∂xµ1
One can further reexpress the partial derivatives in terms of the covariant derivative
Dµ , evaluated at the two points of interest x1 and x2 . Given the Dirac propagator
G(x1 , x2 ) ≡ hx1 |G|x2 i = hx1 |(i 6D − m)−1 |x2 i (62)
one can rewrite in the massless limit m → 0
< ∂ µ Jµ5 > = 2i lim T rγ5 (i 6D − m) G(x1 , x2 )
x1 ,x2 →x
= 2i lim hx2 |γ5 11|x1 i (63)
x1 ,x2 →x

at which point one recalls that the evolution from |x1 i to |x2 i is made in proper time
through our previous Hamiltonian H ∼6D2 = D2 + 2e F µν σµν , |x2 i = eiH(T2 −T1 |x1 i.
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
Article no. 113 Ciprian Sorin Acatrinei 12

In the T2 − T1 = T → 0 limit enforced by the point-splitting procedure, one is left in

d dimensions with the expression
(x1 −x2 )2 ∞
ei 4T iT 2e σµν Fµν 1 X 1 e n
lim tr γ5 e = lim tr γ 5 σµν Fµν (64)
T →0 (4πT )2 T →0 (4πT )2 n! 2
n=0

Only the n = d/2 term contributes. In d = 4 the trace generates the 4-dimensional
totally antisymmetric tensor. In the end, in the massless limit one obtains

e2 ~ ~
< ∂ µ Jµ5 >= E ·B (65)
2π 2
which is the chiral anomaly, as written down by Schwinger [1] long before the classic
papers [8]. The point-splitting procedure has actually the same effect as the regular-
ization introduced later by Fujikawa [9]. One may also note that our adaptation of
[1] is valid for any strength of the electromagnetic field provided it is (nearly) con-
stant, whereas the computations in [8] are perturbative in the field strength but not
restricted to constant configurations.

5.2. SCALAR QED

A straightforward application of the methodology developed in the previous

section occurs for a scalar field φ minimally coupled to electromagnetism, the so
called ”scalar QED”. The Lagrangian is simply

L̃ = |Dµ φ|2 − m2 |φ|2 (66)

with the charged (complex) φ field coupling minimally to electromagnetism through

the covariant derivative Dµ = ∂µ + ieAµ . Comparing with the steps of the derivation
for fermionic fields, only two differences appear: 1) one overall minus sign, due to
the scalar nature of the variables integrated; 2) the lack of any Dirac trace in (33)
and the absence of the 2e F µν σµν term altogether. In the end, one simply obtains the
effective action
i 
Tr log (iD)2 + m2 − log (i∂)2 + m2



S̃ef f = (67)
2
with the trace taken only over space-time variables. The pair production follows
through the same steps as for usual QED. However no terms involving cos or cosh
functions of the electromagnetic fields appear any more, in contradistinction to (51).
One consequence is that for scalar fields an additional magnetic field always de-
creases the pure electric field pair creation rate.
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
13 QED in strong backgrounds Article no. 113

6. EXTENSIONS

6.1. FURTHER SOLVABLE BACKGROUNDS

In general, it is difficult to find electromagnetic backgrounds for which the

fermions can be integrated out analytically. The known cases are listed.
Plane electromagnetic wave: The Dirac equation in a monochromatic plane
wave electromagnetic field was solved long ago [10]. The electromagnetic field in-
variants (52) are both zero in such a background, leading one to suspect that no
vacuum effects could possibly occur. Schwinger proved this by showing [1] that the
effective action vanishes.
√ +Light-like: For an √ electric field depending on one of the light-like coordinates
2x = x0 + x3 and 2x− = x0 − x3 , the one-loop effective action can be found
exactly [11]. Interestingly, its form is similar to the constant field case, as it reads
[11, 12]
Z Z ∞
+ 1 4 ds −ism2
Sef f = 2 d x 3
e [esE(x+ ) coth(esE(x+ )) − 1], (68)
8π 0 s
from which the vacuum decay rate can be calculated. We note the non-factoring
space-time integration, due to the non-constancy of the electric field.
One-coordinate-dependence: A few more exactly solvable backgrounds are
known, with one field (either E or B) depending on one single (either space or time)
coordinate.
A time dependent but spatially uniform electric field E(t) = E0 cosh−2 (t) al-
lows for exact Green functions [13] and hence a complete calculation of the one-loop
partition function, see also [14]. Solvability persists for a similar space dependence,
E(x) = E0 cosh−2 (x) [15]. A magnetic field of similar form B(x) = B0 cosh−2 (x)
was discussed in [16].

6.2. STRINGS IN CONSTANT FIELDS

To show the versatility of the proper time representation, consider a bosonic

open string living on an Euclidean world-sheet, coupled to a U (1) gauge field. Its
action reads:
M2 t
Z 1 
∂X µ 2 ∂X µ 2
Z 
S = − dτ dσ ( ) +( ) (69)
2 0 0 ∂τ ∂σ
Z t   Z t  
∂Xν ∂Xν
−iq1 Fµν dτ Xµ − iq2 Fµν dτ Xµ .
0 ∂τ σ=0 0 ∂τ σ=1
M 2 ≡ T ≡ α10 denotes the string tension. q1 and q2 are the magnitudes of the electric
charges, which can be situated only at the two end-points of the string. Eq. (69) gives
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
Article no. 113 Ciprian Sorin Acatrinei 14

the action to be inserted in Eq. (44) [instead of (45)]. It describes a string with an
infinite number of degrees of freedom, coupled among themselves through the Fµν
field only at the world-sheet boundaries (σ = 0 and σ = 1). As customary in the field,
the proper time τ was Wick rotated.
If Fµν ≡ 0, eq.(44) factorizes into products of free path integrals along each
space-timeR direction. For a generic uncoupled coordinate X we have to evaluate
Zf ree ≡ DXe −S0 , with

M2 t
Z Z 1  
∂X 2 ∂X 2
S0 = dτ dσ ( ) +( ) . (70)
2 0 0 ∂τ ∂σ
For boundary conditions periodic along τ and Neumann along σ, X(t + τ, σ) =
X(τ, σ), ∂X
∂σ σ=0,1 = 0, one obtains

r r
2 π1 Y M2 2 π t Y
Z
M 1 1
Zf ree ≡ DXe−S0 = e6 t = e 24 .
4π 1−e −4πn 1
t 4π t 1 − e−πkt
n≥1 k≥1
(71)
Coupling the 0 and 1 directions through an electric field F01 in the 0 − 1 plane, we
−S(X ,X ,F )
R
have to evaluate Z01 = DX0 DX1 e 0 1 01 , with the action S(X0 , X1 , F01 )
being Eq. (69) restricted to µ = 0, 1. For brevity, denote q1 F01 ≡ E1 and q2 F01 ≡ E2 .
The string tension M 2 will be absorbed into E: E/M 2 → E, expressing the electric
field in terms of the natural units available. Using the notation  = 1 + 2 , with
j = arcthEj , j = 1, 2, one obtains [18]
∞
M2 π 1 1
q Y
Z01 = e3 t (1 − E12 )(1 − E22 ) , (72)
4π 2−4πn 1t 1
n=1 [1 − e ][1 − e−2−4πn t ]
p
which displays the Born-Infeld term (1 − E12 )(1 − E22 ), specific to strings.
For the µ = 2 and ν = 3 directions coupled by a magnetic eigenvalue B, for
instance, the Z23 path integral is obtained by replacing 1,2 → if1,2 and E1,2 → iB1,2
in (72) [1, 17]. With f1,2 = arctan B1,2 and f = f1 + f2 ,
M2 π 1 1
q Y
Z23 = e 3 t (1 + B12 )(1 + B22 ) 1 . (73)
4π 2if −4πn −2if −4πn 1t
n>1 [1 − e t ][1 − e ]

Again, the string tension was absorbed in the field, B/M 2 → B. The ghosts cancel
2
the stringy part of two free coordinates zg2 = (Zf ree )−2 × M
2πt and the whole partition
function in a block-diagonal Fµν reads Z = zg2 × (Zf ree )d × Z01 × Z23 × . . . , in D
dimensions out of which d are left uncoupled by the electromagnetic field.
We can now evaluate the rate at which open strings would be produced out of
the vacuum in a given background. In a pure electric field the rate of open string
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
15 QED in strong backgrounds Article no. 113

pairs production in D dimensions is [17, 18]

X M2 
γ(E) = π (−)k+1 (E1 + E2 )e−k [Zf ree ]D−2 . (74)
4π kπ
k
The pair creation rate is zero when E1 + E2 = 0 (effectively neutral strings).
The result (74) gets corrected in two ways in presence of a magnetic field.
First, the electric-like eigenvalue E may get lowered in presence of other compo-
nents of Fµν . Second, even if Fµν is in block-diagonal form (E ≡ E), a non-zero
magnetic-like eigenvalue F23 ≡ B introduces a factor Z23 × Zf−2 ree . The amplitude
√
gets multiplied in the end by a subunitary factor times a Born-Infeld factor 1 + B 2 .
This stringy contribution triggers a qualitative change: the bosonic string pair pro-
duction can be enhanced in principle by the magnetic field, in contradistinction to the
bosonic particle (Klein-Gordon) case.

6.3. DISSIPATION

By coupling a system to a suitable bath of harmonic oscillators, subsequently

integrated out [19], one can simulate dissipative dynamics [20]. This calls for an ef-
fective action, to be obtained by tracing out the ’thermal’ degrees of freedom. More-
over, one can efficiently replace the ’bath’ of oscillators through an open string with
only one end carrying electric charge. Once the tail of oscillators is integrated out,
this charged end-point will exhibit a kind of dissipative dynamics [21, 22]. The mo-
dified string action
Z 1 " ~ ~
#
M2 t ∂X 0 2 ∂X 0 2
Z
(v) ∂X 2 2 ∂X 2
S = − dτ dσ ( ) +( ) +( ) +v ( )
2 0 0 ∂τ ∂σ ∂τ ∂σ
Z t  
∂X1
−iE dτ X0
0 ∂τ σ=1
displays an additional, dimensionless yet essential, parameter v. For v = 1 one re-
covers standard, relativistic, string theory. For v  1 one can generate [21, 22] the
famous Caldeira-Leggett (CL) dissipative action [20], initially produced by integrat-
ing out a thermal bath of oscillators satisfying a precise spectral condition. The finite
spatial extension of the string amounts to an infinity of harmonic oscillators which
form a suitable bath. It is remarkable that the CL ’spectral condition’ is not needed.
The string seems to automatically provide such a constraint in the dissipative limit.
Since the interaction is a boundary term, not a bulk one like in field theory, the
path integral calculation proceedsq similarly to the relativistic string scenario [18, 23].
2
One obtains again the BI factor 1 − ME4 v2 . Additional magnetic fields can again
be accounted for, with the BI factor corresponding to a magnetic eigenvalue B being
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
Article no. 113 Ciprian Sorin Acatrinei 16

q
2
now 1 + v2BM 4 . For the dissipative case v  1 this factor produces a dramatic
magnetic enhancement of the vacuum decay rate over the case of a pure electric
field.

7. APPENDIX: QM PATH INTEGRAL IN MAGNETIC FIELD

Definitions: The propagator or ’influence’ function U (2|1) ≡ U (x2 , t2 ; x1 , t1 )

satisfies, for any solution ψ of the Schrödinger equation,
Z
ψ(x2 , t2 ) = dx1 U (x2 , t2 ; x1 , t1 ) ψ(x1 , t1 ), (75)

No integration over t1 appears in (75). The propagator encodes all the information on
the spectrum and wavefunctions of a given system (Hamiltonian). In non-relativistic
quantum mechanics it admits the well-known discretized Feynman path-integral rep-
resentation
p2
Z Z Z Z
i PN n
U (f |i) = ··· ··· e+ ~ n=1 pn (xn −xn−1 )− 2m −V (xn−1 ) (76)
xN −1 x1 pN p1

where the time interval was discretized into N equal parts  = T /N , with momenta
and coordinates at tn labelled by pn and xn . This is written in the  → 0 limit as
Z Z R tf
U (f |i) = Dx Dp ei ti dt[pẋ−H(p,x)] . (77)

If we integrate along the p’s in (76) by using

 
p2
n −p xn −xn−1
r 2
dpn − ~i  2m
Z  x −x
n  m i m n n−1
e = e 2~ 
(78)
2π~ 2πi~
we obtain
2
 
 m N Z Z i PN m (xn −xn−1 )
2  n=1 2 2 −V (x n−1 )
~ 
U (f |i) = lim dx1 · · · dxN −1 e
→0 2πi~
N →∞
(79)
which is usually written as
Z R tf Z
U (f |i) = A Dx ei ti dtL(ẋ,x) = A Dx eiS[x(t)] . (80)
x(ti )=xi ; x(tf )=xf

Free particle: For a free particle propagating in a time N = tf − ti ≡ T from

point xi to point xf one obtains
1
i m (xf − xi )2
  
m 2
U (f |i) = exp . (81)
2πi~(tf − ti ) ~ 2 (tf − ti )
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
17 QED in strong backgrounds Article no. 113

(xf −xi )
The classical equations of motion for a free particle lead to ẋcl (t) = (tf −ti ) . The
action along the classical path reads
Z tf
m 2 m (xf − xi )2
S[xcl ] = ẋcl = . (82)
ti 2 2 (tf − ti )
Consequently the propagator is given by a prefactor times the exponential of the
action (times i),
 1
m 2
U (xf , tf ; xi , ti ) = e +iS[xcl ] . (83)
2πi~(tf − ti )
Magnetic field: The Lagrangian for a planar particle in a magnetic field reads
1 ~ = 1 m(ẋ2 + ẏ 2 ) + q ẋAx + q ẏAy
L = m~r˙ 2 + q~r˙ · A (84)
2 2
where A1 ≡ Ax and A2 ≡ Ay are the components of the vector potential A. ~
For a constant magnetic field, using the symmetric gauge Ax = − 2 y, Ay = B2 x
B

and defining the Larmor frequency ωL = eB m , the Lagrangian can be written quite
symmetrically
m 2
ẋ + ẏ 2 + ωL (xẏ − ẋy)

L= (85)
2
and leads to the equations of motion
ẍ = +ωL ẏ, ÿ = −ωL ẋ. (86)
We can solve those for 2-point boundary conditions
x(t1 ) = xi , x(t2 ) = xf , y(t1 ) = yi , y(t2 ) = yf , (87)
and evaluate the action along the classical trajectories. The final result is
 
B mωL ωL T  2 2

Scl = cot (xf − xi ) + (yf − yi ) + 2(xi yf − xf yi ) . (88)
4 2
This is the classical action of a planar particle in a constant magnetic field, in the
symmetric gauge. Under a gauge change
~0 = A
A ~ + ∇Λ(~r) (89)
the action does not remain invariant. It changes by a surface term
Z tf
e dt ~r˙ · ∇Λ = e[Λ(~rf ) − Λ(~ri )], (90)
ti

which becomes a phase factor in the path integral:

UA~ 0 (f |i) = e+ieΛ(~rf ) UA~ (f |i) e−ieΛ(~ri ) . (91)
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
Article no. 113 Ciprian Sorin Acatrinei 18

It does not affect probabilities, and disappears altogether for closed loops. We also
note that in the case of periodic boundary conditions, xi = xf , yi = yf , the expres-
sion in (88) cancels; the classical action brings no contribution. A more extended
discussion of gauge invariance in the context of external backgrounds can be found
in [24]
Path integral evaluation: One can proceed either by discretization of the ac-
tion or by Fourier transforming in terms of normal modes - the approach to be used
here. For periodic boundary conditions - the case needed in the main text -
x(0) = x(T ), y(0) = y(T ) (92)
a good choice of modes is {cos 2πnt 2πnt
T , sin T }. However, to follow more easily the
algebra, we take their complex combinations
2πnt 2πkt
en ≡ ei T and ēk = e−i T (93)
and develop
X 2πnt 2πnt X 2πnt 2πnt
x(t) = cn ei T + c̄n e−i T ; y(t) = dn ei T + d¯n e−i T .
n n
RT R
Using the orthonormality relation 0 dten ēk = δn,k T, we obtain for S = dtL:
m X 4π 2 n2
¯n ) + qB · T · 2
X 2π i n
S= ·T ·2 2
(cn c̄n + dn d [c̄n dn − d¯n cn ]. (94)
2 n
T 2 n
T
2 2
Introducing the notation An = 4π Tn m , Bn = P
2π i n qB, we have the generic quadratic
expression S = n (cn An c̄n + dn An d¯n ) + n [c̄n Bn dn − d¯n Bn cn ] which can be
P
written in matrix form as
· 0 0 · 0 0 ·
  
 0 An 0
 0 −Bn 0    cn 
 
 0 0 · 0 0 ·  · 
 
 
¯

  
S = · c̄n · · dn ·   

 .

  
 · 0 0 · 0 0   · 
  
 0 Bn 0 0 An 0   dn 
0 0 · 0 0 · ·
(95)
Denoting by M the infinite matrix containing An ’s and Bn ’s in the above equation,
our path integral reads, schematically,
  
An −Bn cn
c̄n d¯n
i P  
Z Y   
n
~
B n An dn
Z= Dc̄n Dcn Dd¯n Ddn e (96)
n
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
19 QED in strong backgrounds Article no. 113

(We wrote Dcn instead of the correct dcn , etc., to avoid an awkward ddn or dd¯n .)
Using the standard results for complex Gaussian integrals [below, M is a n by n
matrix, but we shall eventually take the n → ∞ limit]
Z
π
dc̄n dcn e−αcn cn = (97)
α
n
Z Y
dz̄k dzk − Pni,j=1 z̄i Mij zj 1
√ √ e = (98)
π π DetM
k=1
and noting that our matrix has determinant
 
 A1
 0 · B 1 0 · 

 0 A2 · 0 B2 · 
Bn2
 Y  
 · · · · · ·  Y 2 2 2
DetM =   = (An + Bn ) = An 1 + 2 ,
 −B1 0 · A1 0 · 
n n
An
 0
 −B 2 · 0 B 2 · 

 · · · · · · 
Q 2 (99)
observing that Bn → 0 leads to a free two-dimensional particle [hence n An corre-
sponds to a free particle on the plane !], and using the known formula
∞
"  #
ωT 2

Y sin ωT
1− = (100)
nπ ωT
n=1

we obtain
qB r 2 ωc
T m 2 T
ZB = Z2Df ree · 2mqB = · , (101)
sin 2m T 2πi~T sin ω2c T
qB
ωc ≡ m being the Larmor frequency.

Acknowledgements. The author is grateful to the Organizers of the CERN - SEENET - MTP
workshop Modern Aspects of Quantum Field Theory, November 2015 Bucharest, for the opportunity
to lecture on the topic. In particular, I thank Dr. Bogdan Popovici for his collaboration and Prof. Mihai
Vişinescu for encouragement to write up a draft of the lectures. My first acquaintance with the subject
owes much to Prof. Roberto Iengo.
Financial support through PN16420101 and PN18090101 is acknowledged.

REFERENCES

1. J. Schwinger, Phys. Rev. 82 664 (1951).

2. R.P. Feynman, Phys. Rev. 80 440 (1950), 84 108 (1951).
3. V. Fock, Phys. Z. Sowjetunion 12 404 (1937).
4. G.V. Dunne, in M. Shifman et al. (ed.) From fields to strings, vol. 1 pp. 445-522 (WS 2005).
(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14
Article no. 113 Ciprian Sorin Acatrinei 20

5. L.V. Keldysh, Sov. Phys. JETP 20, 1307 (1965); E. Brezin and C. Itzykson, Phys. Rev. D 2, 1191
(1970); V.S. Popov, Sov. Phys. JETP 34, 709 (1972), to quote some of the pioneers.
6. H.R. Reiss, Prog. Quant. Electr. Vol. 16, pp. 1-71 (1992).
7. See e.g. N.D. Birrell and P.C.W. Davies, Quantum fields in curved spaces (CUP 1982).
8. S.L. Adler, Phys. Rev. 177, 2426 (1969); S. Bell and R. Jackiw, Nuovo Cimento 60A, 47 (1969); R.
Jackiw and K. Johnson, Phys. Rev. 182, 1459 (1969).
9. K. Fujikawa, Phys. Rev. Lett 42, 1195 (1979).
10. D.M. Volkov, Z. Phys. 94, 25 (1935).
11. T.N. Tomaras, N.C. Tsamis and R.P. Woodard, Phys. Rev. D 62, 125005 (2000).
12. H.M. Fried and R.P. Woodard, Phys. Lett. B 524, 233 (2002).
13. N.B. Narozhnyi and A.I. Nikishov, Sov. J. Nucl. Phys. 11, 596 (1970).
14. G.V. Dunne and T. Hall, Phys. Rev. D 58, 105022 (1998).
15. S.P. Kim and D.N. Page, Phys. Rev. D 65, 105002 (2002); ibid 73, 065020 (2006).
16. G.V. Dunne and T. Hall, Phys. Lett. B 419, 322 (1998).
17. C. Bachas and M. Porrati, Phys. Lett. B 296, 77 (1992).
18. C.S. Acatrinei, Phys. Lett. B 482, 420 (2000).
19. R.P. Feynman and F. L. Vernon, Ann. Phys. 24, 118 (1963).
20. A.O. Caldeira and A. J. Leggett, Ann. Phys. 149, 374 (1983).
21. C.G. Callan and L. Thorlacius, Nucl. Phys B 329, 117 (1990).
22. R. Iengo and G. Jug, Phil. Mag. B 78 13 (1998), arXiv:hep-th/9702021.
23. C.S. Acatrinei and R. Iengo, Nucl. Phys. B 539, 513 (1999).
24. C.S. Acatrinei, Rom. J. Phys. 64, 101 (2019).

(c) RJP64(Nos. 9-10), ID 113-1 (2019) v.2.2r20191120 *2020.5.2#b3a87b14