Gauge symmetry and radiatively induced terms in dimension-5 non-minimal

Lorentz-violating QED
A. P. B. Scarpelli(a) , A. R. Vieira(b)
(a) Centro Federal de Educação Tecnológica, Belo Horizonte -MG, Brazil
(b) Universidade Federal do Triângulo Mineiro, Instituto de Ciências Agrárias,
Exatas e Biológicas - ICAEBI,38280-000, Iturama- MG, Brasil
(Dated: August 4, 2025)
In this work, we derive the conditions that assure gauge invariance of a non-minimal dimension-5
Lorentz-violating QED. The two and three point functions at one-loop are computed. The gauge
Ward identities are checked and the conditions to assure gauge symmetry of this non-minimal
framework is found to be the same of the usual QED. Induced terms are also investigated and it is
(5)
shown that the non-minimal Lorentz-violating aF -term of the fermion sector can induce radiatively
a non-minimal Lorentz-violating term in the photon sector.
arXiv:2508.00801v1 [hep-th] 1 Aug 2025

I. INTRODUCTION

Lorentz and CPT Symmetries are main ingredients to build quantum field theories and are usually taken for granted.
Nevertheless, in string theory, it was realized long ago that these symmetries can be spontaneously broken at Planck
scale and as a result Lorentz and CPT-violating terms appear at low energies [1]. All possible Lorentz and CPT-
violating operators at these low energies make up the Standard Model Extension (SME) [2, 3]. Even if space-time
symmetries are in fact exact in nature, the issue is on what precision one can say they are. In this way, the SME
is also a framework to test how good are these symmetries besides being a way for searching for Lorentz and CPT
violation.
Precise tests of space-time symmetries generally involve classical Lagrangians or tree level Feynman diagrams of
the SME. However, there are many theoretical studies concerning this framework beyond tree level. The SME was
shown to be renormalizable at one-loop for both the electroweak [4, 5] and the strong sectors [6, 7]. There are
also several investigations concerning the radiatively induced finite quantum corrections. The most known example
is the dimension-3 operator associated with the CPT and Lorentz-violating coefficient bµ from the fermion sector,
which radiatively induces the Chern-Simons-like term of the photon sector. Furthermore, breaking Lorentz and CPT
symmetries at the classical level does not mean that all the possible violating structures at this level also appear at
the quantum one [8]. Thus, it is worth to investigate what symmetries can be broken at the quantum level if the
Lorentz and CPT ones are broken at the classical one.
At the same time, from a technical point of view, the regularization procedure is required beyond tree level to treat
ultraviolet and/or infrared divergences. As it is well known, the treatment of these infinities can spuriously break
symmetries of the model. For instance, lattice regularization is developed for treating QCD non-perturbatively and
discretization of the space-time points breaks Lorentz symmetry among others [9]. The renormalization process is then
more laborious because the introduction of restoring counter-terms is needed. In the study of anomalies, this issue
is more subtle because we want to know if the symmetry is indeed physical or if it was broken by the regularization
scheme. Some anomalies are measurable, like the chiral anomaly or the scale anomaly in QED and this experimental
fact solves the issue. The former is related with the pion decay into two photons [10–12] and the later with the
hadronic R ratio [13]. On the other hand, there are not yet observables related with Super-symmetric anomalies or
anomalies in the Standard Model Extension. Thus, the question if a symmetry is in fact broken at the quantum level
can only be answered by using a regularization scheme that does not break spuriously any symmetry of the model.
The issue of the chiral anomaly in a Lorentz-violating context is particularly discussed in [14, 15].
Gauge symmetry is one of the ways of introducing particle interactions. Maintaining this symmetry beyond tree level
assures renormalizability of the theory and a massless photon. Therefore, the search for a gauge invariant regularization
method is desired mainly for gauge field theories, like dimensional regularization. However, as new theories arose,
other proposal of regularization schemes did as well, for dealing with issues that conventional regularizations could
not deal. In this work, we show that there is no gauge anomaly at one-loop for a non-minimal dimension-5 Lorentz-
violating version of QED. If Bose symmetry is not required, there would be a gauge anomaly due to the three point
function diagrams and the induced terms could be used to put stringent constraints in a set of non-minimal dimension-
5 Lorentz and CPT-violating coefficients since gauge symmetry breaking is not observed. The paper is divided as
follows: in section II, we present an overview on the implicit regularization scheme. In section II A, we discuss the
conditions for gauge invariance in the usual QED. In section III, we compute the gauge Ward identities of the two
(5)
and the three point Green functions of a non-minimal Lorentz-violating QED considering the aF -term, using both
dimension and implicit regularizations. In section IV, we compute these same identities considering the non-minimal
 2

(5)
bF -term and we present conclusions in section V.

II. OVERVIEW OF IMPLICIT REGULARIZATION

Besides using dimensional regularization to compute divergent integrals in amplitudes of the next sections, we also
apply the implicit regularization scheme [18]. The former is probably the most known and popular regularization
scheme, and it does not need an introduction. The latter, on the other hand, although used in a wide variety of
problems, is not known in textbooks.
A particularly interesting application of Implicit regularization occurs when the theories include dimension specific
objects, like γ 5 matrices and Levi-Civita symbols. Also, since it is a scheme that does not break symmetries of the
theory, it is usually used for the computation of anomalies. A recent computation for a general momentum routing
concerns gravitational anomalies in two dimensions [19]. It was used also in other scenarios with Lorentz violation,
like in the Bumblebee model [20], or chiral models [21], that deal directly with γ 5 matrices, in which comparisons
with other regularization techniques are performed [20–22].
Let us make a brief review of the method in four dimensions. In this scheme, we assume that the integrals are
regularized by an implicit regulator Λ in order to allow algebraic operations within the integrands. We then recursively
use the following identity

1 1 (p2 + 2p · k)
= − , (1)
(k + p)2 − m2 k 2 − m2 (k 2 − m2 )[(k + p)2 − m2 ]

to separate basic divergent integrals (BDI’s) from the finite part. These BDI’s are defined as follows
k µ1 · · · k µ2n
Z
µ1 ···µ2n
Ilog (m2 ) ≡ 2 2 2+n
(2)
k (k − m )

and
k µ1 · · · k µ2n
Z
µ1 ···µ2n
Iquad (m2 ) ≡ . (3)
k (k 2 − m2 )1+n
The BDI’s with Lorentz indices can be judiciously combined as differences between integrals with the same superficial
degree of divergence, according to the equations below, which define surface terms 1 :

Υµν µν 2 µν 2 µν
2w = g I2w (m ) − 2(2 − w)I2w (m ) ≡ υ2w g , (4)

Ξµναβ
2w
µναβ
= g {µν g αβ} I2w (m2 ) − 4(3 − w)(2 − w)I2w (m2 ) ≡
≡ ξ2w (g µν g αβ + g µα g νβ + g µβ g να ). (5)

In the expressions above, 2w is the degree of divergence of the integrals and we adopt the notation such that indices
0 and 2 mean log and quad, respectively. Surface terms can be conveniently written as integrals of total derivatives,
as presented below:

∂ kµ
Z
µν
υ2w g = ,
k ∂kν (k 2 − m2 )2−w
(6)

∂ 2(2 − w)k µ k α k β
Z
(ξ2w − v2w )(g µν g αβ + g µα g νβ + g µβ g να ) = . (7)
k ∂kν (k 2 − m2 )3−w

We see that the surface terms in equations (4)-(5) are undetermined because they are differences between divergent
quantities. Each regularization scheme gives a different value for these terms. However, as physics should not depend

1 The Lorentz indices between brackets stand for permutations, i.e. A{α1 ···αn } B {β1 ···βn } = Aα1 ···αn B β1 ···βn + sum over permutations
between the two sets of indices α1 · · · αn and β1 · · · βn . For instance, g {µν g αβ} = g µν g αβ + g µα g νβ + g µβ g να .
 3

on the scheme applied, we leave these terms to be arbitrary until the end of the calculation and then fix them by
symmetry constraints or phenomenology. This approach was first proposed in [23], where undetermined surface terms
were discussed in several contexts of quantum corrections.
Of course, the same idea can be applied for any dimension of space-time and for higher loops. Equation (1) is used
recursively until the divergent piece is separated from the finite one. This procedure makes the finite integrals hard
to compute due to the number of k’s in the numerator. A simpler alternative to this approach is presented in [24],
where the Feynman parametrization is applied before separating the BDI’s. Also, eq. (1) is not the only possible
equation to be used since the implicit regulator was assumed to allow the use of other identities.

A. An example: Gauge invariance of the vacuum polarization tensor

Let us consider the vacuum polarization tensor of conventional Quantum Electrodynamics (QED) whose computa-
tion in implicit regularization is given by (all regularized integrals are presented in the appendix)

iΠµν (p) = 34 e2 (p2 g µν − pµ pν )Ilog (m2 ) − 4e2 υ2 g µν + 34 e2 (p2 g µν − pµ pν )υ0

− 43 e2 (p2 g µν + 2pµ pν )(ξ0 − 2υ0 ) − 2πi 2 e2 (p2 g µν − pµ pν )(Z1 − Z2 ), (8)
R1  
in which Zn = 0 dxxn ln D(x) m2 and D(x) = m2 − p2 x(1 − x) and where, for didactic reasons, we have placed the
divergent, the finite and surface-dependent terms separately.
Notice that if we require gauge invariance using the Ward identity pµ Πµν (p) = 0, we find that the quadratic surface
term υ2 must be zero and that the logarithmic surface terms must obey the relation ξ0 = 2υ0 . These conditions
are automatically fulfilled if we set all surface terms to zero. The same takes place when one uses dimensional
regularization, because the surface terms defined in section II are zero in such a scheme. This is the same condition
obtained if we require momentum routing invariance of the Feynman diagram in figure 1 (this feature of the loop
diagram would not appear in dimensional regularization, since it allows for shifts in the loop momenta). In the next
sections, we are going to see that these requirements for surface terms are the same for a non-minimal dimension-5
Lorentz-violating version of QED. The Ward identities are checked for two and three-point functions in this Lorentz-
violating framework.
The diagrams with more external photon legs in usual QED do not need to be checked as the ones of the next
sections. The three photon leg of the usual QED is zero because of Furry’s theorem, and the box diagram with four
photon legs is gauge invariant, since there is no term in the tree level Lagrangian to renormalize it if it was not.

FIG. 1. Vacuum polarization tensor of QED.

(5)
III. GAUGE INVARIANCE AND RADIATIVELY INDUCED TERMS: aF -TERM

(5)
We consider the dimension-5 coefficients in the Lagrangian density LψF of Table I of reference [25]. The terms with
(5) (5) (5)
coefficients (mF )αβ , (m5F )αβ and (bF )µαβ do not generate induced terms at first order because the number of Dirac
matrices in the trace of the fermion loop is odd or there are less than four Dirac matrices appearing with a γ5 matrix
(5)
inside this trace. The non-minimal term (HF )µναβ ψσµν Fαβ ψ was studied in [26], in which the authors showed that
 4

it radiatively induces the CPT-even term (kF )µναβ Fαβ Fµν of the minimal SME photon sector. The other non-trivial
(5)
radiatively induced term comes from the CPT and Lorentz-violating non-minimal term − 12 (aF )µαβ ψγµ Fαβ ψ. It can
(5)
be rewritten as −(aF )µαβ ψγµ ∂α Aβ ψ due to the antisymmetry of the two last indices and this leads to the Feynman
rule presented in Figure 2. It is also important to notice that it is easy to check with the modified version of the Dirac
equation that this term does not break gauge symmetry at the classical level, i. e. ∂µ j µ = 0, where j µ = ψ̄γ µ ψ.
We next perform perturbative calculations in a modified Lorentz-violating QED model which includes this term.

A. Two-point function

The diagrams depicted in Figure 3 give rise to a radiatively induced non-minimal LV term in the photon sector. In
order to see this, we compute the diagrams of this figure. Their corresponding amplitudes can be written as

 
i i
Z
(5)
Παβ
(a) (p) =− Tr (aF )µλβ γµ pλ (−ieγ α ) (9)
k /
k − p/ − m k/ − m
and
 
i i
Z
(5)
Παβ
(b) (p) = − T r (−ieγ β ) (aF )µλα γµ pλ , (10)
k k/ − /p − m k/ − m
R R d4 k
where k stands for (2π) 4 . Let us then apply Implicit Regularization, presented in section II, in the computation

of the diagrams of the two-point function. After taking the traces and regularizing (all integrals are presented in
appendix), we find the following result:
 
4 (5) ppβ α (5) αpβ 2 i
Παβ (p) = Παβ(a) (p) + Παβ
(b) (p) = i((a F ) p − (a F ) p ) Ilog (m 2
) − bZ 0 − +
3 48π 2
pα (5)
 
8ib 2 (5) (5) (5)
+ m Z0 −(aF )αpβ + 2 (aF )ppβ + 4((aF )ppβ pα + (aF )αpβ p2 )υ0 −
3 p
4 (5) (5)
− (−2(aF )ppβ pα + (aF )αpβ p2 )ξ0 + (α ↔ β), (11)
3
(5) (5) i
where (aF )αpβ ≡ (aF )αµβ pµ and b = (4π) 2 . We can easily check gauge invariance by computing the Ward identity
αβ
pα Π (p) = 0. Doing this, we find out that the surface terms break gauge symmetry, as expected:
(5) (5)
pα Παβ (p) = 4(aF )ppβ p2 (−ξ0 + 2υ0 ) − 4υ2 (aF )ppβ (12)
We see above that, if all surface terms are null, gauge symmetry is automatically fulfilled. However, the relations
ξ0 = 2υ0 and υ2 = 0 are sufficient. It is interesting to notice that these are the same conditions for gauge invariance
presented in section II A and found in other frameworks like in the minimal QED extension [8].
Alternatively, one can apply dimensional regularization in this calculation. The result is the same as in Implicit
Regularization as long as the surface terms of eq. (11) are set to zero and we take
m2
   
1 1 γ
Ilog (m2 ) = b − ln − − , (13)
ǫ 2 4πµ2 2
in which µ is a mass scale introduced to keep the number of dimensions of the integrals after changing it to d
dimensions. This result is expected, since surface terms are zero in dimensional regularization.
In the following discussion, we are interested in the form of the induced term, so we are going to assume for the
moment that the surface terms are equal to zero. The non-minimal term of the fermion sector radiatively induces
a termR in the photon sector which is also non-minimal. In order to see this, let us consider the effective action
S = 21 d4 xAα Παβ Aβ and take the massless limit of eq. (11). It leads to

o  2 
4i b
Z n
(5) (5) −p
S= d4 x (aF )αµβ Aα ∂ 2 ∂µ Aβ − (aF )µνβ Aα ∂ α ∂µ ∂ν Aβ Ilog (λ2 ) − b ln − =
3 λ2 3
  2 
2 b
Z
4 (5) αµβ ν 2 −p
= d x i(aF ) Fνα ∂ Fµβ Ilog (λ ) − b ln − , (14)
3 λ2 3
 5

FIG. 2. Feynman rule for a non-minimal Lorentz-violating interaction.

FIG. 3. One-loop diagrams contributing for the radiatively induction of a non-minimal Lorentz-violating term.

(5)
where we dropped out terms that are total derivatives. The induced term has the form (aF )αµβ Fνα ∂ ν Fµβ , that is a
particular case of

(5) 1
LA = − k (5)ακλµν Fκλ ∂α Fµν , (15)
4
present in Table III of reference [25], in which k (5)ακλµν is proportional to
(5) (5) (5) (5)
η αλ (aF )κµν − η ακ (aF )λµν + η αµ (aF )νκλ − η αν (aF )µκλ . (16)

Eq. (14) shows us the presence of a divergent part in the coefficient of the induced term, which is an important
point to be analyzed. This indicates that the original classical action must contain such a term. In other words, the
(5)
inclusion of the dimension-5 term − 21 (aF )µαβ ψγµ Fαβ ψ in a modified QED requires the presence of this induced term
from the beginning. However, we must take into account that our model is non-renormalizable. We have carried out
a one-loop calculation and, at this order in the perturbative expansion, it has been shown that a new term which
violates Lorentz and CPT symmetries should be included in the classical action. If we go beyond the one-loop order,
certainly new other terms will have to be considered. The non-renormalizability of the model tells us that there is
not a finite number of counter-terms that will be sufficient to renormalize the theory. So, if we would like to deal
with this effective model, we will have to stop at one-loop order. For this, it is necessary to find a cutoff energy
Λ. This discussion is carried out in [28] and [29]. Finally, it is easy to check that this term leads to the usual
(5)
charge conservation at the classical level. The modified Maxwell equations for an additional C(aF )αµβ Fνα ∂ ν Fµβ are
(5) (5)
∂λ F λζ = J ζ + C((aF )ζµβ ∂ 2 Fµβ − (aF )λµβ ∂λ ∂ ζ Fµβ ), where we can easily see that ∂ζ ∂λ F λζ = ∂ζ J ζ = 0.

B. Three-point function

The next step would be to check gauge symmetry for diagrams with more photon external legs. Let us consider the
diagrams of the three-point function presented in Figure 4. Using the Feynman rules set out above, the corresponding
amplitudes can be written as below:
 
i i i
Z
µνα µ ν (5) λζα
T(a) (p, q) = − T r (−ieγ ) (−ieγ ) (a ) (p + q)ζ γλ ,
k k/ − m k/ + /q − m F k/ − /p − m
 
i i i
Z
µνα (5)
T(b) (p, q) = − T r (aF )λζµ pζ γλ (−ieγ ν ) (−ieγ α )
k k/ − m k/ + /q − m k/ − /p − m
and (17)
 
i i i
Z
µνα (5)
T(c) (p, q) = − T r (−ieγ µ ) (a )λζν qζ γλ (−ieγ α ) . (18)
k k/ − m F k/ + /q − m k/ − /p − m
 6

FIG. 4. One-loop diagrams with three photon legs.

Besides, in virtue of Bose symmetry, the amplitudes referring to the crossed diagrams have to be added. We then
have
µνα µνα µνα
T µνα (p, q) = T(a) (p, q) + T(b) (p, q) + T(c) (p, q) + crossed terms (19)

µνα
It is easy to check gauge invariance for diagram (a). We get (p+q)α T(a) (p, q) = 0 due to the antisymmetry property
(5)
of the two last indices of the tensor (aF )λζα . For the next two diagrams, we use the identity /q+/p = k/+/q−m−(/
k −/p−m)
after the contraction in order to split the amplitude into two pieces. After this manipulation, we find:

k + m)γ ν (/
T r[γλ (/ k − /p + m)]
n Z
µνα 2 (5)
(p + q)α T(b) (p, q) =e (aF )λζµ pζ − +
k (k 2 − m2 )[(k − p)2 − m2 ]
k + m)γ ν (/
T r[γλ (/ k + /q + m)] o
Z
+
k (k 2 − m2 )[(k + q)2 − m2 ]
and (20)
µ
T r[γ (/
k + m)γλ (/
k − /p + m)]
n Z
µνα (5)
(p + q)α T(c) (p, q) = e2 (aF )λζν qζ − +
k (k − m )[(k − p)2 − m2 ]
2 2

T r[γ µ (/
k + m)γλ (/
k + /q + m)] o
Z
+ . (21)
k (k − m )[(k + q)2 − m2 ]
2 2

The computation of equations (21) with implicit regularization reveals that the crossed terms are necessary in order
to avoid a gauge anomaly for the three-point diagram, even if all the surface terms are zero. The result is given by
 
µνα µνα µνα
(p + q)α T(a) + T(b) + T(c) =
  2 
4 −p b  ν (5) ppµ 2 (5) νpµ µ (5) pqν 2 (5) µpν

= Ilog (λ2 ) − b ln − p (a F ) − p (a F ) + p (a F ) − p (a F ) +
3 λ2 3
4 (5) (5) (5) (5) (5)
+ ξ0 (2pν (aF )ppµ + p2 (aF )νpµ + 2pµ (aF )pqν + p2 (aF )µpν ) − 4υ0 (pν (aF )ppµ +
3
(5) (5) (5)
+ p2 (aF )νpµ + pµ (aF )pqν + p2 (aF )µpν ). (22)

When the crossed terms are added, the gauge symmetry, in this case, is fulfilled in a way that is independent of the
surface terms.

(5)
IV. GAUGE INVARIANCE: NON-MINIMAL DIMENSION-5 bF -TERM

(5)
The term of coefficient (bF )µαβ does not generate induced terms at first order due to the anti-symmetry properties
of the Levi-Civita symbol. It is also easy to check the gauge Ward identities for the 2-point function. The diagrams
 7

are the same as Figure 3 except for the different vertex and the amplitudes are given by:
 
i i
Z
αβ (5) µλβ α
A(a) (p) = − T r (bF ) γ5 γµ p λ (−ieγ )
k k/ − /p − m k/ − m
and (23)
 
i i
Z
(5)
Aαβ
(b) (p) = − T r (−ieγ β ) (bF )µλα γ5 γµ pλ . (24)
k k/ − /p − m k/ − m

When computing pα Aαβ (p) = pα Aαβ αβ

(a) (p) + pα A(b) (p), the second term vanishes because of the antisymmetry of the
(5)
last two indices of the tensor bF . For the first term, we can use /p = k/ − m − (/ k − /p − m) to split the integral into
two pieces. Each one of these pieces is zero, because there remain two Dirac matrices with one γ5 inside the trace.
It is also important to check with the use of the modified version of the Dirac equation that this non-minimal term
does not break gauge or chiral symmetries at the classical level, i. e. ∂µ j µ = 0 and ∂µ j5µ = 0, where j µ = ψ̄γ µ ψ
and j5µ = ψ̄γ µ γ5 ψ. We could expect a chiral anomaly due to the presence of a γ5 matrix in one of the vertices of
the 3-point function diagram. However, this is not the case as it is shown below and there is no issue with γ5 in the
regularization applied.
The diagrams with three photon legs are the same ones depicted in Figure 4, the only difference being the Lorentz-
violating vertex. The amplitude corresponding to these diagrams can be written as:
 
i i i
Z
µνα µ ν (5) λζα
M(a) (p, q) = − T r (−ieγ ) (−ieγ ) (b ) (p + q)ζ γ5 γλ ,
k k/ − m k/ + q/ − m F k/ − p
/−m
 
i i i
Z
µνα (5)
M(b) (p, q) = − T r (bF )λζµ pζ γ5 γλ (−ieγ ν ) (−ieγ α )
k /
k−m /
k + /q − m /
k − /p − m
and (25)
 
i i i
Z
µνα µ (5) λζν α
M(c) (p, q) = − T r (−ieγ ) (b ) qζ γ5 γλ (−ieγ ) . (26)
k k/ − m F k/ + /q − m k/ − /p − m

As before, we can check the gauge Ward identities by computing pµ M µνα , for instance. Besides the amplitudes of
eq. (26), there are also the crossed diagrams, obtained by changing µ ↔ ν and p ↔ q.
The computation of the gauge Ward identity is a little involved and it needs the use of relations between ξnm (p, q)
integrals presented in the appendix. The idea is to reduce integrals ξ20 (p, q), ξ02 (p, q) and ξ11 (p, q) into integrals
ξ10 (p, q) and ξ01 (p, q) by using eqs. (31)-(32) and (35)-(36). Then the remaining integrals ξ10 (p, q) and ξ01 (p, q)
are reduced into integrals Zn (p2 , m2 ) and ξ00 (p, q) with the use of eqs. (33)-(34). Nevertheless, none of these finite
integrals remain in the final result that depends only on the logarithmic surface term:
1 
(5) (5) (5)

pµ T µνα = − 2 υ0 (bF )ζq ν ǫαζpq + 2(bF )ζp α ǫζνpq + 2(bF )ζq α ǫζνpq , (27)
4π
(5) (5)
where (bF )ζpα ≡ (bF )ζλα pλ .
In this case, we need to choose the surface term υ0 as zero to avoid a gauge anomaly in the three point function. This
(5)
condition is according to the previous ones concerning the aF -term and the usual QED, although in this situation
the logarithmic surface term υ0 appears alone.
We summarized all the results of the computed gauge Ward identities in table I.

Non-minimal term Ward identity Dimensional Reg. Implicit Reg.

Usual QED 2-point function υ2 = 0 and ξ0 = 2υ0
3-point function Null by Furry’s theorem Null by Furry’s theorem
(5)
aF -term 2-point function υ2 = 0 and ξ0 = 2υ0
3-point function
(5)
bF -term 2-point function
3-point function cannot be applied υ0 = 0

TABLE I. Summary of the conditions for gauge invariance in loop diagrams of a non-minimal dimension-5 Lorentz-violating
QED.
 8

V. CONCLUSIONS

In this work, we discussed the possibility of a gauge anomaly in a non-minimal dimension-5 Lorentz-violating
framework. We showed that there is no gauge anomaly if surface terms are null or if there is a relation between
logarithmic surface terms. Previous results in literature showed that no gauge anomalies are expected in minimal
versions of the SME as well, like in the minimal Lorentz-violating QED [30, 31]. The same happens to be true for
non-minimal versions of the SME. In particular, we explicitly showed that the non-minimal dimension-5 version of
QED is gauge invariant beyond tree level.
(5)
The gauge invariance of this non-minimal Lorentz-violating framework is expected since it was shown that the HF -
term can induce the kF gauge invariant term in the photon sector as studied in [26]. Furthermore, it was shown here
(5)
that the non-minimal (aF )-term induces a non-minimal dimension-5 gauge invariant term in the photon sector. The
induction of terms due to diagrams with more external legs is more involved but the corresponding Ward identities of
them were checked. As a prospect, it would be interesting to investigate further loop diagrams from the non-minimal
SME with higher dimension terms.

APPENDIX

Integrals computed with implicit regularization:

kµ
Z
H1µ = 2 2 2 2 2
= −bpµ ι1 ,
k (k − m ) [(k − p) − m ]
kµ kν 1
Z
µν
H2 = 2 − m2 )2 [(k − p)2 − m2 ]
= g µν (Ilog (m2 ) − υ0 − bZ0 ) − bpµ pν ι2 ,
k (k 4
µ ν α
k k k 1 {α µν}
Z
H3µνα = 2 − m2 )2 [(k − p)2 − m2 ]
= p g (Ilog (m2 ) − ξ0 − 6bZ1 + 6bZ2 ) − bpµ pν pα ι3
(k 12
Z k
µ kµ
A1 = 2 2 2 − m2 ]
= −p2 H1µ + 2pλ H2µλ
(k − m )[(k − p)
Zk
µν kµ kν 1
A2 = 2 2 2 2
= g µν (Iquad (m2 ) − υ2 ) − p2 H2µν + 2pλ H3µνλ
(k − m )[(k − p) − m ] 2
Zk
1
A0k = 2 2 2 2
= Ilog (m2 ) + bp2 (ι0 − 2ι1 )
k (k − m )[(k − p) − m ]
1
Z
A2k = 2 − m2 ]
= Iquad (m2 ) − p2 υ0 (28)
k [(k − p)
R1   R1
D(x) n
(1−x)
where λ is the renormalization group scale, b ≡ i
(4π)2 , Zn = 0 dxxn ln m2 , ιn = 0 dx x D(x) and D(x) =
m2 − p2 x(1 − x).

Integrals of the finite part of the diagrams in section IV

The functions ξnm (p, q) are defined as

1 1−z
z nym
Z Z
ξnm (p, q) = dz dy , (29)
0 0 Q(y, z)

with

Q(y, z) = [p2 y(1 − y) + q 2 z(1 − z) + 2(p · q)yz − m2 ] (30)

and those functions have the property ξnm (p, q) = ξmn (q, p).
 9

The ξnm functions obey the following relations:

q 2 ξ11 (p, q) − (p · q)ξ02 (p, q) = 12 − 12 Z0 ((p + q)2 , m2 ) + 21 Z0 (p2 , m2 ) + q 2 ξ01 (p, q) ,

 
(31)
p2 ξ11 (p, q) − (p · q)ξ20 (p, q) = 21 − 21 Z0 ((p + q)2 , m2 ) + 12 Z0 (q 2 , m2 ) + p2 ξ10 (p, q) ,
 
(32)
2
q ξ10 (p, q) − (p · q)ξ01 (p, q) = 12 [−Z0 ((p + q)2 , m2 ) + Z0 (p2 , m2 ) + q 2 ξ00 (p, q)], (33)
2
p ξ01 (p, q) − (p · q)ξ10 (p, q) = 21 [−Z0 ((p + q)2 , m2 ) + Z0 (q 2 , m2 ) + p2 ξ00 (p, q)], (34)
1 1
2
 2

1 2 3 2

q ξ20 (p, q) − (p · q)ξ11 (p, q) = 2 − 2 + m ξ00 (p, q) + 2 p ξ01 (p, q) + 2 q ξ10 (p, q) , (35)
1 1
2
 2

1 2 3 2

p ξ02 (p, q) − (p · q)ξ11 (p, q) = (36)
2 − 2 + m ξ00 (p, q) + 2 q ξ10 (p, q) + 2 p ξ01 (p, q) ,
R1 h 2 2
i
where Zk (p2 , m2 ) ≡ 0 dx xk ln m −pmx(1−x) 2 .
The derivation of the relations (31)-(36) can be simply achieved by integration by parts. There is a whole review
[27] about these integrals and other integrals with integrands of larger denominators that appear in Feynman diagrams
with more external legs.
The result of all finite and regularized divergent integrals from sections IV is listed below:

1
Z
= bξ00 (p, q), (37)
k (k 2 − m2 )[(k − p)2 − m2 ][(k + q)2 − m2 ]

kα
Z
= b(pα ξ01 (p, q) − q α ξ10 (p, q)), (38)
k (k 2 − m2 )[(k − p)2 − m2 ][(k + q)2 − m2 ]

k2
Z
=Ilog (m2 ) − bZ0 (q 2 , m2 ) + b(m2 − p2 )ξ00 (p, q)+
k (k 2 − m2 )[(k − p)2 − m2 ][(k + q)2 − m2 ]
+ 2b(p2 ξ01 (p, q) − (p · q)ξ10 (p, q)), (39)

kα kβ 1 1
Z
= g αβ (Ilog (m2 ) − υ0 ) − bg αβ Z0 (q 2 , m2 )−
k (k 2 − m2 )[(k − p)2 − m2 ][(k + q)2 − m2 ] 4 4
h1
−b g αβ p2 (ξ00 (p, q) − 3ξ01 (p, q) − ξ10 (p, q) + 2ξ02 (p, q) + 2ξ11 (p, q)) − ξ02 (p, q)pα pβ +
2 i
+ ξ11 (p, q)q α pβ + ξ11 (p, q)pα q β − ξ20 (p, q)q α q β + (ξ10 (p, q) − ξ11 (p, q) − ξ20 (p, q))g αβ (p · q) , (40)

kα k2 1 1
Z
= (pα − q α )(Ilog (m2 ) − υ0 ) + b(q α Z0 (q 2 , m2 )−
k (k 2 − m2 )[(k − p)2 − m2 ][(k + q)2 − m2 ] 2 2
− pα Z0 (p2 , m2 )) + b(m2 − q 2 )(pα ξ01 (p, q) − q α ξ10 (p, q)) + b[q α p2 (ξ00 (p, q) − 3ξ01 (p, q)+
− ξ10 (p, q) + 2ξ02 (p, q) + 2ξ11 (p, q)) − 2(p · q)pα ξ02 (p, q) + 2q 2 pα ξ11 (p, q)+
+ 2(p · q)q α (ξ10 (p, q) − ξ20 (p, q)) − 2q 2 q α ξ20 (p, q))], (41)
R RΛ d4 k
where k ≡ (2π)4 .

[1] V. Alan Kostelecký and S. Samuel, Phys.Rev. D, 39, 683 (1989).

[2] D. Colladay and V. A. Kostelecký, Phys. Rev. D 58, 116002 (1998).
[3] D. Colladay and V. A. Kostelecký, Phys. Rev. D 55, 6760-6774 (1997).
[4] V. A. Kostelecký, C. D. Lane and A. G. Pickering, Phys. Rev. D 65, 056006 (2002).
[5] D. Colladay and P. McDonald, Phys. Rev. D, 79, 125019 (2009).
[6] D. Colladay and P. McDonald, Phys. Rev. D 75, 105002 (2007).
[7] D. Colladay and P. McDonald, Phys. Rev. D 77, 085006 (2008).
[8] J. C. C. Felipe, A. Yu. Petrov, A. P. B. Scarpelli and A. R. Vieira, Int. J. Mod. Phys. A 38 (2023) 15n16, 2350089; Int. J.
Mod. Phys. A 38 (2023) 26n27, 2350089 (erratum).
 10

[9] M. Costa, H. Panagopoulos, Phys. Rev. D 96 (2017) 034507.

[10] J.S. Bell, R. Jackiw, Nuovo Cim. A 60 (1969) 47-61.
[11] S.L. Adler, Phys. Rev. 177 (1969) 2426.
[12] W.A. Bardeen, Phys. Rev. 184 (1969) 1848.
[13] M.S. Chanowitz, J. Ellis, Phys. Rev. D 7 (1973) 2490.
[14] A. R. Vieira, Annals Phys. 478 (2025) 170025.
[15] A. P. B. Scarpelli, T. Mariz, J. R. Nascimento, A. Yu. Petrov, Int.J.Mod.Phys.A 31 (2016) 12, 1650063; Int. J. Mod. Phys.
A 32 (2017) 26, 1792002 (erratum).
[16] G. ’t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189.
[17] C.G. Bollini, J.J. Giambiagi, Nuovo Cimento 12 (1972) 20.
[18] O.A. Battistel, A.L. Mota, M.C. Nemes, Mod. Phys. Lett. A 13 (1998) 1597.
[19] G. Dallabona, P.G. de Oliveira, O.A. Battistel, J. Phys. G 51 (2024) 095004.
[20] R.J.C. Rosado, A. Cherchiglia, M. Sampaio, B. Hiller, Eur. Phys. J. C 85 (2025) 41.
[21] R.J.C. Rosado, A. Cherchiglia, M. Sampaio, B. Hiller, Acta Phys. Polon. Supp. 17 (2024) 6-A15.
[22] A.M. Burque, A.L. Cherchiglia, M. Pérez-Victoria, JHEP 08 (2018) 109.
[23] R. Jackiw , Int. J. Mod. Phys. B 14 (2000) 2011.
[24] B.Z. Felippe, A.P.B. Scarpelli, A. R. Vieira, J.C.C. Felipe, Eur. Phys. J. C 82 (2022) 583.
[25] V. A. Kostelecky and Z. Li, Phys. Rev. D 99, 056016 (2019).
[26] R. Casana, M. M. Ferreira Jr., R. V. Maluf and F. E. P. dos Santos, Phys. Lett. B 726, 815 (2013).
[27] O.A. Battistel, G. Dallabona, Eur. Phys. J. C 45 (2006) 721.
[28] G. Gazzola, H. G. Fargnoli, A. P. Baeta Scarpelli, M. Sampaio, M. C. Nemes, J. Phys. G39, 035002 (2012).
[29] A. P. Baeta Scarpelli, J. Phys. G39, 125001 (2012).
[30] A. R. Vieira, A. L. Cherchiglia, M. Sampaio, Phys. Rev. D 93 (2016) 025029.
[31] T. R. S. Santos, R. F. Sobreiro, Phys. Rev. D 94 (2016) 125020.