1 Introduction

Expressions involving products of Dirac spinors are among the most common objects appearing in the problems of high energy physics. For example, any Feynmann diagram involving fermions includes Dirac bilinears (e.g. as in Fig. 1). Various conventions for spinors are present in the literature, and those mostly rely on the two-spinor formalism which generally involves an explicit choice of Dirac matrices and defining the four-component Dirac spinors in terms of the well known two-component Pauli spinors (see e.g. [1,2,3]). However, calculating covariant expressions for them in terms of the relevant Lorentz vectors remained an unfinished task [4]. Although existing conventions appear to be sufficient for standard perturbative calculations, the use of Lorentz covariant expressions in the study of bound states, for example in hadronic physics [4] is expected to be more enlightening. Another possible use of Lorentz covariant expressions is expected to be in strong background physics, for example in strong background QED, where, just like in hadronic physics, fermions “dressed” with gauge bosons (and also with virtual pairs) are involved [5].

What actually is expected from the use of Lorentz covariant expressions of Dirac bilinears can be easily exemplified within the context of hadronic physics. As is well known, hadrons are bound states of quarks and gluons. For a specified hadron, all multi-particle Fock states having the same quantum numbers with that hadron contribute to the quantum state of the hadron. For example, for a meson, one can write in light-cone quantization [7,8,9,10,11,12]:

$$\begin{aligned}&|M(P;^{2S+1}L_{J_{z}},J_{z})> \nonumber \\&\quad = \sum _{Fock\, states}\int \left[ \prod _{i}\frac{dk_{i}^{+}d^{2}k_{\perp ,i}}{2(2\pi )^{3}}\right] 2(2\pi )^{3}\delta ^{(3)}\left( {\tilde{P}}-\sum _{i}{\tilde{k}}_{i}\right) \nonumber \\&\qquad \times \sum _{\lambda _{i}}\varPsi _{LS}^{JJ_{z}}({\tilde{k}}_{i},\lambda _{i})|relevant\; Fock\; state>. \end{aligned}$$
(1)

where \({\tilde{k}}=(k^{+},\mathbf {k}_{\perp })\) and \(\varPsi _{LS}^{JJ_{z}}({\tilde{k}}_{i})\) are the light cone wave functions corresponding to the Fock states having the same quantum numbers with the hadron. The light-cone wave function involves outer products of spinors with different momentum arguments [8]. For example, for parapositronium [8], one can write:

$$\begin{aligned} \varPsi _{0,0}^{0,0}({\tilde{k}}_{1},{\tilde{k}}_{2})= \,\,&N({\tilde{k}}_{1},{\tilde{k}}_{2}) \nonumber \\&\times \lbrace u({\tilde{k}}_{1},\uparrow ) {\bar{v}}({\tilde{k}}_{2},\downarrow ) - u({\tilde{k}}_{1},\downarrow ) {\bar{v}}({\tilde{k}}_{2},\uparrow ) \rbrace , \end{aligned}$$
(2)

where \(N({\tilde{k}}_{1},{\tilde{k}}_{2})\) is the momentum-dependent normalization factor for the wave function, and \(u\, (v)\) are the free positive (negative) energy spinors, respectively. When writing down amplitudes, traces are taken and products of spinors with different momentum arguments appear.

Fig. 1
figure 1

An example diagram expressing the process \(e^{+}e^{-}\rightarrow \mu ^{+}\mu ^{-}\) at the lowest order in the corresponding perturbative expansion [1]. The matrix element A for this diagram is: \( A={\bar{v}}^{s'}(p')\left( -i e \gamma ^{\mu } \right) u^{s}(p)\frac{-ig_{\mu \nu }}{q^{2}}{\bar{u}}^{r}(k)\left( -i e \gamma ^{\nu } \right) v^{r'}(k') \)

Previously, C. Lorcé calculated Lorentz covariant expressions for Dirac bilinears and presented a list of bilinears involving all linearly independent combinations of Dirac matrices [4]. The approach used by C. Lorcé made use of a standard boost from the rest frame [4]. Although the final results in [4] are Lorentz covariant, this is not explicit, as indicated in [4] as well. In this work, explicitly Lorentz covariant expressions are sought. Our approach examines the foliation of spacetime in terms of a set of basis vectors, such that the momentum 4-vector of a fermion can be chosen as one of the basis vectors. Then, using that basis set, we show that the Dirac equation and its solutions can be constructed in a fully covariant manner. However, in our calculations it is also revealed that there will still remain some freedom in the calculation of scalar bilinears, which can be reflected in various ways depending on the line of reasoning. Those will be explained in the following sections as well.

Our paper is organized as follows. In Sect. 2, we present the well known relations relating Dirac spinors and the four vectors which are in a sense “arguments” of these spinors. In Sect. 3, we present various algebraic relations among the bilinear structures which also involve the Lorentz vectors, and also we show that all tensorial bilinears can be reduced to combinations of scalar bilinears with appropriate tensorial structures constructed from the basis vectors. This section closes with a covariant recipe for calculating the scalar bilinears. Then we conclude the article. We also present two appendices at the end of the text, which present certain details discussed in the other sections and also how the spinor representation of a Lorentz transformation can be expressed in our setting.

2 Dirac spinors and Lorentz vectors

Dirac spinors are solutions to the celebrated Dirac equation. In momentum space, Dirac equation can be expressed as (see e.g. [1,2,3]):

(3)

where \(\gamma _{\mu }\) are the Dirac matrices satisfying:

$$\begin{aligned} \lbrace \gamma _{\mu },\gamma _{\nu }\rbrace =2g_{\mu \nu } \end{aligned}$$
(4)

and \(g_{\mu \nu }\) are the components of the metric tensor. Here, p and m are respectively the momentum four-vector (with \(p^{0}>0\) assumed [2]) and mass of the relevant fermion and \(w_{\epsilon }(p)\) is the corresponding Dirac spinor. \(\epsilon =+1(-1)\) corresponds to positive (negative) energy solutions. In \(3+1\) dimensions, there are two linearly independent solutions for each value of \(\epsilon \) [1,2,3].

Information about the spin of the particle is carried by the Pauli–Lubansky vector, which reads [2]:

$$\begin{aligned} W_{\mu }=\frac{i}{4}\varepsilon _{\mu \nu \alpha \beta }p^{\nu }\sigma ^{\alpha \beta }, \quad \sigma ^{\alpha \beta }=\frac{i}{2}\left[ \gamma _{\alpha },\gamma _{\beta }\right] , \end{aligned}$$
(5)

for a spin\(-1/2\) particle.

In general, Pauli–Lubansky vector satisfies [2]:

$$\begin{aligned} W\cdot W = -m^{2}\lambda (\lambda +1), \end{aligned}$$
(6)

where \(\lambda \) is the spin of the relevant particle, which is equal to 1/2 for quarks and leptons. The projection of this vector on any four-vector s orthogonal to p (that is, satisfying \(s\cdot p=0\)) is related to the rest-frame spin projections of the fermion along a four-vector which is obtained by Lorentz transforming s to the rest frame [2]:

(7)

where \(\sigma =\pm 1\), \(s^{2}=-1\), and \(w_{\epsilon ,\sigma }=w_{\epsilon ,\sigma }(p,s)\). Thus, the four linearly independent Dirac spinors can be identified with the following eigenvalue equations:

(8)

Lorentz transformations which leave p and s unaltered do not alter the above equations but they do alter the explicit expressions of the spinors. However, the transformed spinors will still be solutions for the above equations with the same eigenvalues.

A Dirac spinor in the irreducible representation in \(3+1\) dimensions involves 4 complex (and equivalently 8 real) functions to be calculated, and as we have seen, we look for 4 independent spinor solutions. However, there are various algebraic relations which relate the spinor components to one another, which will be discussed in the next section. Here, we present only one of them, namely a phase convention which relates positive and negative energy spinors as follows ([4]):

$$\begin{aligned} \gamma _{5}w_{\epsilon ,\sigma }=-\epsilon \sigma w_{-\epsilon ,-\sigma }. \end{aligned}$$
(9)

Our approach for calculating Dirac bilinears in terms of Lorentz scalars is based on covariantly using the four-vector s in line with the momentum four-vector p, instead of calculating rest frame spinors using a specific coordinate system and boosting them to a generic frame where the fermion has momentum p, as is usually preferred in the literature. Once this goal is achieved, one can make an explicit choice for the four-vector s so as to relate the results with the conventional expressions in the literature.

One can derive various identities involving Dirac spinors and combinations of Dirac matrices; these have been studied in detail in [4]. Here, we concentrate on a number of identities which will be of practical use. Using the normalization:

$$\begin{aligned} {\bar{w}}_{\epsilon ,\sigma }w_{\epsilon ',\sigma '}=2m\epsilon \delta _{\epsilon \epsilon '}\delta _{\sigma \sigma '}, \end{aligned}$$
(10)

the eigenvalue equations for Dirac spinors and the anti-commutation relations for the Dirac matrices, one obtains [4]:

$$\begin{aligned} {\bar{u}}_{\sigma }\gamma _{\mu }u_{\sigma '}&=2p_{\mu }\delta _{\sigma \sigma '}, \end{aligned}$$
(11)
$$\begin{aligned} {\bar{u}}_{\sigma }\gamma _{\mu }\gamma _{5}u_{\sigma }&=2m\sigma s_{\mu } , \end{aligned}$$
(12)

where \(u_{\sigma }\equiv w_{+,\sigma }\) are the positive energy solutions. One can derive similar identities for the negative energy solutions as well. Here, we also use: \(\epsilon _{0123}=1\) and \({\gamma _{5}}^\dagger =\gamma _{5}\). It is interesting to observe that the simple trick using the eigenvalue equations cannot provide information on the combination \({\bar{u}}_{\sigma }\gamma _{\mu }\gamma _{5}u_{-\sigma }\), and in fact one observes that this expression is actually non-zero (which can be verified using any specific explicit representation). This observation motivates defining \({\bar{u}}_{\sigma }\gamma _{\mu }\gamma _{5}u_{-\sigma }\) (\(\sigma =+\) or \(\sigma =-\)) as two other Lorentz vectors related to the particle under study, and examine their relation to p and s vectors:

$$\begin{aligned} -\frac{1}{4m}{\bar{u}}_{+}\gamma _{\mu }\gamma _{5}u_{-}\equiv \,\,&d_{\mu } \end{aligned}$$
(13)
$$\begin{aligned} \Rightarrow -\frac{1}{4m}{\bar{u}}_{-}\gamma _{\mu }\gamma _{5}u_{+} = \,\,&d^{*}_{\mu }. \end{aligned}$$
(14)

One observes that:

(15)

Here, the projection operators have been used [2]:

(16)

By a similar reasoning, one also observes that:

(17)

The last equalities follow from the fact that \(d\cdot d=d^{*}\cdot d^{*}=0\). So, one derives the conclusion that and are simply the spin raising and lowering matrices for Dirac spinors. Thus, one can define the following “spin-flip” matrices:

(18)

Using the eigenvalue equations and the normalization discussed above, one can easily verify that the following equalities hold:

$$\begin{aligned}&d\cdot d^{*}=-\frac{1}{2}, \quad d\cdot d=d^{*}\cdot d^{*}=0; \end{aligned}$$
(19)
$$\begin{aligned}&d\cdot p=d^{*}\cdot p=0, \quad d\cdot s=d^{*}\cdot s =0. \end{aligned}$$
(20)

As is seen from the above equations, d and \(d^{*}\) are null vectors and they span a subspace of the \(3+1\) dimensional Minkowski space that is orthogonal to the subspace spanned by p and s. This also implies that the set of vectors \(\lbrace p,\, s,\, d,\, d^{*}\rbrace \) (which we will call the p-set from now on) can be used as a basis for spanning the whole \(3+1\) dimensional Minkowski space. This observation has the following interesting consequences:

  • Any Lorentz vector, say q, can be decomposed into its components along each of the p-set vectors:

    $$\begin{aligned} q^{\mu }&=\frac{q\cdot p}{p^{2}}\, p^{\mu }+\frac{q\cdot s}{s^{2}}\, s^{\mu }\nonumber \\&+\frac{1}{d\cdot d^{*}}\left( q\cdot d\, d^{*\mu }+q\cdot d^{*}\, d^{\mu }\right) \nonumber \\&=\frac{q\cdot p}{m^{2}}\, p^{\mu }-q\cdot s\, s^{\mu }-2\left( q\cdot d\, d^{*\mu }+q\cdot d^{*}\, d^{\mu }\right) \end{aligned}$$
    (21)

    which can easily be verified by taking dot products with each of the p-set vectors.

  • The independence of the scalar product of any two vectors from the basis set used for computing it implies:

    $$\begin{aligned} q\cdot q'&= \frac{q\cdot p}{m}\frac{q'\cdot p}{m}-q\cdot s\, q'\cdot s \nonumber \\&- 2q\cdot d\, q'\cdot d^{*} - 2q'\cdot d\, q\cdot d^{*} \end{aligned}$$
    (22)
    $$\begin{aligned} \Rightarrow g_{\mu \nu }&=\frac{p_{\mu }p_{\nu }}{m^{2}}-s_{\mu }s_{\nu }-2\left( d^{*}_{\mu }d_{\nu }+d_{\mu }d^{*}_{\nu }\right) . \end{aligned}$$
    (23)

    This decomposition of the metric tensor in terms of the p-set vectors implies that the p-set vectors are nothing but a set of vierbeinsFootnote 1 defined locally at the spacetime position of the particle under study.

  • Using the definitions for d and \(d^{*}\) vectors, one observes that the following equality holds:

    (24)
    (25)

    which is related to the “handedness” of the p-set. Note that Eq. (25) is equivalent to Eq. (23) and that Eq. (25) does not violate the linear independence of the p-set, since it involves linear combinations of the tensor products of the related vectors rather than linear combinations of the vectors themselves.

  • It can be shown that, the vectors d and \(d^{*}\) can always be written in terms of two real spacelike unit vectors orthogonal to each other, say \(n_{1}\) and \(n_{2}\), which are also orthogonal to p and s, such that \(d=\frac{1}{2} (n_{1}-i n_{2})\) and \(d^{*}=\frac{1}{2} (n_{1}+i n_{2})\). Any Lorentz transformation \(\varLambda \) which leaves p and s unchanged (that is, any rotation in the plane spanned by d and \(d^{*}\)) rotates the spinors in the spinor space but does not alter Eq. (8). That is, the rotated spinors will still be the solutions to Eq. (8) with the same eigenvalues:

    (26)

    due to \(S(\varLambda )\gamma _{\nu }S^{-1}(\varLambda )=\gamma _{\mu }\varLambda ^{\mu }\, _{\nu }\). Under such a transformation d acquires a phase and \(d^*\) acquires the opposite phase. This obviously corresponds to a freedom in defining the spinors, which can be fixed (up to an overall phase related to the normalization of the spinors) by fixing d and \(d^{*}\). Now we can relate these observations to the calculation of bilinear structures.

3 Algebraic relations among bilinear structures

In the previous section, we have calculated the Dirac bilinears formed using spinors having the same four momentum, but different spin projections. In general, Dirac bilinears formed from two spinors of different momentum are needed. At this point, we can return to the calculation of such bilinears.

First of all, we should note that all tensorial structures like \({\bar{w}}_{\epsilon , \sigma }(p)\varGamma W_{\epsilon ', \sigma '}(q)\) can be reduced to linear combinations of Lorentz vectors (as dictated by the \(\varGamma \) matrix in the expression), such that the coefficients of the linearly independent Lorentz structures reduce to scalar bilinears. To explain this fact, it is useful to state various algebraic relations among the bilinear structures.

Using \(w_{-\epsilon ,-\sigma } =-\epsilon \sigma \gamma _{5}w_{\epsilon ,\sigma }\), pseudoscalar structures can directly be obtained from the scalar ones:

$$\begin{aligned}&\begin{bmatrix} {\bar{u}}_{+}U_{+} \\ {\bar{u}}_{-}U_{+} \\ {\bar{v}}_{-}U_{+} \\ {\bar{v}}_{+}U_{+} \end{bmatrix}=\begin{bmatrix} -{\bar{u}}_{+}\gamma _{5}V_{-} \\ -{\bar{u}}_{-}\gamma _{5}V_{-} \\ -{\bar{v}}_{-}\gamma _{5}V_{-} \\ -{\bar{v}}_{+}\gamma _{5}V_{-} \end{bmatrix} = \begin{bmatrix} {\bar{v}}_{-}\gamma _{5}U_{+} \\ -{\bar{v}}_{+}\gamma _{5}U_{+} \\ {\bar{u}}_{+}\gamma _{5}U_{+} \\ -{\bar{u}}_{-}\gamma _{5}U_{+} \end{bmatrix} = \begin{bmatrix} -{\bar{v}}_{-}V_{-} \\ {\bar{v}}_{+}V_{-} \\ -{\bar{u}}_{+}V_{-} \\ {\bar{u}}_{-}V_{-} \end{bmatrix}; \nonumber \\&\begin{bmatrix} {\bar{u}}_{+}U_{-} \\ {\bar{u}}_{-}U_{-} \\ {\bar{v}}_{-}U_{-} \\ {\bar{v}}_{+}U_{-} \end{bmatrix}=\begin{bmatrix} {\bar{u}}_{+}\gamma _{5}V_{+} \\ {\bar{u}}_{-}\gamma _{5}V_{+} \\ {\bar{v}}_{-}\gamma _{5}V_{+} \\ {\bar{v}}_{+}\gamma _{5}V_{+} \end{bmatrix} = \begin{bmatrix} {\bar{v}}_{-}\gamma _{5}U_{-} \\ -{\bar{v}}_{+}\gamma _{5}U_{-} \\ {\bar{u}}_{+}\gamma _{5}U_{-} \\ -{\bar{u}}_{-}\gamma _{5}U_{-} \end{bmatrix}= \begin{bmatrix} {\bar{v}}_{-}V_{+} \\ -{\bar{v}}_{+}V_{+} \\ {\bar{u}}_{+}V_{+} \\ -{\bar{u}}_{-}V_{+} \end{bmatrix}. \end{aligned}$$
(27)

Using C,P,T transformation properties, it is possible to relate the scalar bilinears among each other such that only 4 of them remain independent:

$$\begin{aligned} \begin{bmatrix} {\bar{u}}_+(p)U_+(q)\\ {\bar{u}}_-(p)U_+(q)\\ {\bar{v}}_-(p)U_+(q)\\ {\bar{v}}_+(p)U_+(q) \end{bmatrix}&=\begin{bmatrix} \left( {\bar{u}}_-(p)U_-(q)\right) ^*\\ -\left( {\bar{u}}_+(p)U_-(q)\right) ^*\\ \left( {\bar{v}}_+(p)U_-(q)\right) ^*\\ -\left( {\bar{v}}_-(p)U_-(q)\right) ^* \end{bmatrix}\nonumber \\&=\begin{bmatrix} -{\bar{v}}_-(p)V_-(q)\\ {\bar{v}}_+(p)V_-(q)\\ -{\bar{u}}_+(p)V_-(q)\\ {\bar{u}}_-(p)V_-(q) \end{bmatrix}=\begin{bmatrix} -\left( {\bar{v}}_+(p)V_+(q)\right) ^*\\ -\left( {\bar{v}}_-(p)V_+(q)\right) ^*\\ -\left( {\bar{u}}_-(p)V_+(q)\right) ^*\\ -\left( {\bar{u}}_+(p)V_+(q)\right) ^* \end{bmatrix}. \end{aligned}$$
(28)

Also, it is interesting to observe that the scalar structures themselves satisfy the Dirac equation and the corresponding spin equation. For example, consider some \(U_{+}\) satisfying:

(29)

Using the decompositions of q and r in terms of the p-set and the resolution of identity in terms of the uv spinors:

$$\begin{aligned} 1=\sum _{\sigma }\left( \frac{u_{\sigma }(p)\otimes {\bar{u}}_{\sigma }(p)}{2m}-\frac{v_{\sigma }(p)\otimes {\bar{v}}_{\sigma }(p)}{2m} \right) \end{aligned}$$
(30)

one can construct two eigenvalue equations which involve the projections of \(U_{+}\) on the \(u,\, v\) spinors:

$$\begin{aligned}&\begin{bmatrix} {\bar{u}}_{+}U_{+} \\ {\bar{u}}_{-}U_{+} \\ {\bar{v}}_{-}U_{+} \\ {\bar{v}}_{+}U_{+} \end{bmatrix} = \begin{bmatrix} \frac{q\cdot p}{m\, M} &{} 0 &{} \frac{q\cdot s}{M} &{} \frac{2q\cdot d}{M} \\ 0 &{} \frac{q\cdot p}{m\, M} &{} \frac{-2q\cdot d^{*}}{M} &{} \frac{q\cdot s}{M} \\ \frac{-q\cdot s}{M} &{} \frac{2q\cdot d}{M} &{} \frac{-q\cdot p}{m\, M} &{} 0 \\ \frac{-2q\cdot d^{*}}{M} &{} \frac{-q\cdot s}{M} &{} 0 &{} \frac{-q\cdot p}{m\, M} \end{bmatrix} \begin{bmatrix} {\bar{u}}_{+}U_{+} \\ {\bar{u}}_{-}U_{+} \\ {\bar{v}}_{-}U_{+} \\ {\bar{v}}_{+}U_{+} \end{bmatrix}, \end{aligned}$$
(31)
$$\begin{aligned}&\begin{bmatrix} {\bar{u}}_{+}U_{+} \\ {\bar{u}}_{-}U_{+} \\ {\bar{v}}_{-}U_{+} \\ {\bar{v}}_{+}U_{+} \end{bmatrix} = \begin{bmatrix} -r\cdot s &{} 2r\cdot d &{} \frac{-r\cdot p}{m} &{} 0 \\ 2r\cdot d^{*} &{} r\cdot s &{} 0 &{} \frac{r\cdot p}{m} \\ \frac{r\cdot p}{m} &{} 0 &{} r\cdot s &{} 2r\cdot d \\ 0 &{} \frac{-r\cdot p}{m} &{} 2r\cdot d^{*} &{} -r\cdot s \end{bmatrix} \begin{bmatrix} {\bar{u}}_{+}U_{+} \\ {\bar{u}}_{-}U_{+} \\ {\bar{v}}_{-}U_{+} \\ {\bar{v}}_{+}U_{+} \end{bmatrix}. \end{aligned}$$
(32)

The matrices appearing in the equations are nothing but and written in the basis of uv spinors. Then, the solutions of these equations are the eigenvectors of and written in the basis of p-set vectors. Note that it is possible to verify Eq. (28) using these eigenvectors as well. One can also construct the solutions using projection operators , along with an explicit basis for the \(u_{\pm }(p),v_{\pm }(p)\) spinors. This approach needs to be followed by normalization of the spinors, which raises the question of how to fix the phase of the spinors in a convenient way.

Before proceeding to the calculation of scalar bilinears, we will first show that vector, axial-vector, and anti-symmetric tensor bilinears can all be expressed in terms of scalar bilinears. First, it would be useful to express certain equalities involving vector and axial-vector structures. Using these equalities, one only needs to calculate 8 combinations, within the totality of 32 possible combinations.

$$\begin{aligned} \begin{bmatrix} {\bar{u}}_{+}\gamma _{\mu }U_{+}\\ {\bar{u}}_{+}\gamma _{\mu }U_{-}\\ {\bar{u}}_{+}\gamma _{\mu }V_{-}\\ {\bar{u}}_{+}\gamma _{\mu }V_{+} \end{bmatrix}&=\begin{bmatrix} -{\bar{u}}_{+}\gamma _{\mu }\gamma _{5}V_{-}\\ {\bar{u}}_{+}\gamma _{\mu }\gamma _{5}V_{+}\\ -{\bar{u}}_{+}\gamma _{\mu }\gamma _{5}U_{+}\\ {\bar{u}}_{+}\gamma _{\mu }\gamma _{5}U_{-} \end{bmatrix}\\&=\begin{bmatrix} {\bar{v}}_{-}\gamma _{\mu }V_{-}\\ -{\bar{v}}_{-}\gamma _{\mu }V_{+}\\ {\bar{v}}_{-}\gamma _{\mu }U_{+}\\ -{\bar{v}}_{-}\gamma _{\mu }U_{-} \end{bmatrix}=\begin{bmatrix} -{\bar{v}}_{-}\gamma _{\mu }\gamma _{5}U_{+}\\ -{\bar{v}}_{-}\gamma _{\mu }\gamma _{5}U_{-}\\ -{\bar{v}}_{-}\gamma _{\mu }\gamma _{5}V_{-}\\ -{\bar{v}}_{-}\gamma _{\mu }\gamma _{5}V_{+} \end{bmatrix}, \end{aligned}$$
$$\begin{aligned} \begin{bmatrix} {\bar{u}}_{-}\gamma _{\mu }U_{+}\\ {\bar{u}}_{-}\gamma _{\mu }U_{-}\\ {\bar{u}}_{-}\gamma _{\mu }V_{-}\\ {\bar{u}}_{-}\gamma _{\mu }V_{+} \end{bmatrix}&=\begin{bmatrix} -{\bar{u}}_{-}\gamma _{\mu }\gamma _{5}V_{-}\\ {\bar{u}}_{-}\gamma _{\mu }\gamma _{5}V_{+}\\ -{\bar{u}}_{-}\gamma _{\mu }\gamma _{5}U_{+}\\ {\bar{u}}_{-}\gamma _{\mu }\gamma _{5}U_{-} \end{bmatrix}\nonumber \\&=\begin{bmatrix} -{\bar{v}}_{+}\gamma _{\mu }V_{-}\\ {\bar{v}}_{+}\gamma _{\mu }V_{+}\\ -{\bar{v}}_{+}\gamma _{\mu }U_{+}\\ {\bar{v}}_{+}\gamma _{\mu }U_{-} \end{bmatrix}=\begin{bmatrix} {\bar{v}}_{+}\gamma _{\mu }\gamma _{5}U_{+}\\ {\bar{v}}_{+}\gamma _{\mu }\gamma _{5}U_{-}\\ {\bar{v}}_{+}\gamma _{\mu }\gamma _{5}V_{-}\\ {\bar{v}}_{+}\gamma _{\mu }\gamma _{5}V_{+} \end{bmatrix}. \end{aligned}$$
(33)

A similar reasoning holds for higher rank tensor structures as well. Noting that \(\sigma _{\mu \nu }\gamma _{5}=-\frac{1}{2}\epsilon _{\mu \nu \alpha \beta }\sigma ^{\alpha \beta }\) [2, 4], one notices that there are 4 independent structures out of 16:

$$\begin{aligned} \begin{bmatrix} {\bar{u}}_{+}\sigma _{\mu \nu }U_{+}\\ {\bar{u}}_{+}\sigma _{\mu \nu }U_{-}\\ {\bar{u}}_{-}\sigma _{\mu \nu }U_{+}\\ {\bar{u}}_{-}\sigma _{\mu \nu }U_{-} \end{bmatrix}&=\begin{bmatrix} -{\bar{u}}_{+}\sigma _{\mu \nu }\gamma _{5}V_{-}\\ {\bar{u}}_{+}\sigma _{\mu \nu }\gamma _{5}V_{+}\\ -{\bar{u}}_{-}\sigma _{\mu \nu }\gamma _{5}V_{-}\\ {\bar{u}}_{-}\sigma _{\mu \nu }\gamma _{5}V_{+} \end{bmatrix}\nonumber \\&=\begin{bmatrix} -{\bar{v}}_{-}\sigma _{\mu \nu }V_{-}\\ {\bar{v}}_{-}\sigma _{\mu \nu }V_{+}\\ {\bar{v}}_{+}\sigma _{\mu \nu }V_{-}\\ -{\bar{v}}_{+}\sigma _{\mu \nu }V_{+} \end{bmatrix}=\begin{bmatrix} {\bar{v}}_{-}\sigma _{\mu \nu }\gamma _{5}U_{+}\\ {\bar{v}}_{-}\sigma _{\mu \nu }\gamma _{5}U_{-}\\ -{\bar{v}}_{+}\sigma _{\mu \nu }\gamma _{5}U_{+}\\ -{\bar{v}}_{+}\sigma _{\mu \nu }\gamma _{5}U_{-} \end{bmatrix}. \end{aligned}$$
(34)

Now we can write down the independent vector, axial-vector and anti-symmetric tensor structures and calculate them. In Appendix A, we show that in general d and \(d^{*}\) can be chosen to be written in terms of p, s, q and r, and we demonstrate various practical approaches useful for this purpose. So, we can expand the tensorial structures in terms of p, s, d and \(d^{*}\) and eliminate d and \(d^{*}\) from the expressions later if needed. This approach is easier because p, s, d and \(d^{*}\) are orthogonal and so no matrix inversion will be necessary to calculate the coefficients in the expansions of the tensorial structures. For vector and axial-vector structures, one writes:

$$\begin{aligned} {\bar{u}}_{\pm }(p)\gamma _{\mu }W_{\epsilon , \sigma }(q)\equiv \frac{\alpha _{p}}{m} p_{\mu }-\alpha _{s}s_{\mu }-2\alpha _{d}d_{\mu }-2\alpha _{d^{*}}d^{*}_{\mu }, \end{aligned}$$
(35)

where \(W_{\epsilon , \sigma }(q)\) is a spinor satisfying and . Then, one contracts this expression with p, s, d and \(d^{*}\) to get the unknown coefficients \(\alpha _{p}\), \(\alpha _{s}\), \(\alpha _{d}\) and \(\alpha _{d^{*}}\). The results of this procedure are presented in Table 1.

Table 1 Expansion coefficients according to Eq. (35)

The same approach can be used for calculating the anti-symmetric tensor structures. In \(3+1\) dimensions, an anti-symmetric tensor has 6 independent components, and hence can be expanded as follows:

$$\begin{aligned} {\bar{u}}_{{\bar{\sigma }}}\sigma _{\mu \nu }U_{\sigma }&\equiv \beta _{ps}\left( p_{\mu }s_{\nu }-p_{\nu }s_{\mu }\right) + \beta _{dd^{*}}\left( d_{\mu }d^{*}_{\nu }-d^{*}_{\mu }d_{\nu }\right) \nonumber \\&\quad + \beta _{pd}\left( p_{\mu }d_{\nu }-d_{\mu }p_{\nu }\right) +\beta _{pd^{*}}\left( p_{\mu }d^{*}_{\nu }-d^{*}_{\mu }p_{\nu }\right) \nonumber \\&\quad + \beta _{sd}\left( s_{\mu }d_{\nu }-d_{\mu }s_{\nu }\right) +\beta _{sd^{*}}\left( s_{\mu }d^{*}_{\nu }-d^{*}_{\mu }s_{\nu }\right) . \end{aligned}$$
(36)

Contracting with each of the terms present in the expansion, one calculates the coefficients for the 4 independent anti-symmetric tensor structures, which are presented in Table 2.

Table 2 Expansion coefficients according to Eq. (36)

Now, we can calculate the scalar bilinears. Below, we present our approach for calculating them and fixing their phases in a covariant manner. Once a recipe for this is obtained, there is no need for an explicit basis for \(u_{\pm }(p),v_{\pm }(p)\) and acting with projections on such a basis as well.

We begin with reminding that:

(37)

There are also corresponding relations for \(v_{\pm }(p)\):

(38)

As a digression here, we can define two new vectors \(\varDelta ,\varDelta ^*\) such that:

$$\begin{aligned}&\varDelta _{\mu }\equiv -\frac{1}{4M}{\bar{U}}_+(q)\gamma _{\mu }\gamma _5 U_-(q), \nonumber \\&\quad \varDelta ^*_{\mu }\equiv -\frac{1}{4M}{\bar{U}}_-(q)\gamma _{\mu }\gamma _5 U_+(q),\end{aligned}$$
(39)
$$\begin{aligned}&\quad \varDelta \cdot q=\varDelta ^* \cdot q=0,\; \varDelta \cdot r=\varDelta ^* \cdot r=0,\nonumber \\&\quad \varDelta \cdot \varDelta =\varDelta ^* \cdot \varDelta ^*=0,\; \varDelta \cdot \varDelta ^* =-1/2, \end{aligned}$$
(40)
(41)

Together with qr, they can be considered as a q-set in analogy with the p-set. In Appendix B, we describe a Lorentz transformation which maps the p-set onto the q-set in a one-to-one manner. \(\varDelta ,\varDelta ^*\) vectors can also be used in the calculation of relative phases of scalar bilinears, following lines similar to those given below.

We should also remember that the absolute squares of the scalar bilinears can be calculated using projection operators. For example:

(42)

Note that this equation fixes the absolute value of \({\bar{u}}_+(p)U_+(q)\), but it does not fix its phase. Assuming that \({\bar{u}}_+(p)U_+(q)\) is non-zero,Footnote 2 cross products of this bilinear with other scalar bilinears can be calculated as:

$$\begin{aligned}&{\bar{u}}_+(p)U_+(q){\bar{U}}_+(q)u_-(p) \nonumber \\&\quad = q\cdot p\, r\cdot d - q\cdot d \, r\cdot p + M m\, r\cdot d \nonumber \\&\qquad +im\epsilon (d,q,r,s)+iM\epsilon (d,p,r,s),\nonumber \\&\qquad {\bar{u}}_+(p)U_+(q){\bar{U}}_+(q)v_-(p) \nonumber \\&\quad = M\, r\cdot p - m\, q\cdot s +i\epsilon (q,r,p,s), \nonumber \\&\qquad {\bar{u}}_+(p)U_+(q){\bar{U}}_+(q)v_+(p)\nonumber \\&\quad = m\left( -q\cdot d + q\cdot d \, r\cdot s - q\cdot s \, r\cdot d\right) \nonumber \\&\qquad -i\epsilon (d,p,q,r)-i\epsilon (d,p,q,s), \end{aligned}$$
(43)

where we have defined;

$$\begin{aligned} \epsilon (V_1,V_2,V_3,V_4)\equiv \epsilon _{\mu \nu \alpha \beta }V_1^{\mu }V_2^{\nu }V_3^{\alpha }V_4^{\beta }. \end{aligned}$$
(44)

Notice also that, by virtue of Eq.(25):

$$\begin{aligned} \epsilon (d,p,q,r)&=i m \, \left( q\cdot d \, r\cdot s - q\cdot s \, r\cdot d\right) ,\nonumber \\ \epsilon (d,p,q,s)&=-im\, q\cdot d, \nonumber \\ \epsilon (d,q,r,s)&=\frac{i}{m}\left( q\cdot d \, r\cdot p - q\cdot p \, r\cdot d\right) ,\nonumber \\ \epsilon (d,p,r,s)&=-im\, r\cdot d. \end{aligned}$$
(45)

Then, one obtains the ratios of scalar bilinears \({\bar{U}}_+(q)w_{\epsilon ,\sigma }(p)\) to \({\bar{U}}_+(q)u_+(p)\) as follows:

$$\begin{aligned} \frac{{\bar{U}}_+(q)u_-(p)}{{\bar{U}}_+(q)u_+(p)}=&\frac{2\left[ r\cdot d\left( q\cdot p+M m \right) - q\cdot d\, r\cdot p \right] }{\left( M m+q\cdot p\right) \left( 1-r\cdot s\right) +q\cdot s\, r\cdot p}, \nonumber \\ \frac{{\bar{U}}_+(q)v_-(p)}{{\bar{U}}_+(q)u_+(p)}=&\frac{M\, r\cdot p -m\, q\cdot s -i\epsilon (q,r,p,s)}{\left( M m+q\cdot p\right) \left( 1-r\cdot s\right) +q\cdot s\, r\cdot p}, \nonumber \\ \frac{{\bar{U}}_+(q)v_+(p)}{{\bar{U}}_+(q)u_+(p)}=&\frac{2m\left[ q\cdot d\left( 1-r\cdot s\right) - q\cdot s \, r\cdot d\right] }{\left( M m+q\cdot p\right) \left( 1-r\cdot s\right) +q\cdot s\, r\cdot p}. \end{aligned}$$
(46)

These ratios involve information about the relative phases. Extracting out the ratios of the absolute values, one obtains the phases of the bilinears relative to \({\bar{U}}_+(q)u_+(p)\). This observation, together with Eq. (28), tell us that we have calculated all scalar bilinears up to an overall phase. How can we fix this overall phase?

Motivated by the example spinors given in Appendix B, we can introduce, as an example:

$$\begin{aligned}&{\bar{U}}_+(q)u_+(p) \nonumber \\&\quad \equiv \sqrt{\frac{-\varDelta ^*\cdot d}{\sqrt{\varDelta ^*\cdot d \, \varDelta \cdot d^*}}}\sqrt{\left( M m+q\cdot p\right) \left( 1-r\cdot s\right) +q\cdot s\, r\cdot p}, \end{aligned}$$
(47)

assuming that \(\varDelta \cdot d^* \ne 0\) and \(\varDelta ^* \cdot d \ne 0\). This choice does not have to be the most useful one for practice, but it is sufficient as a proof of concept. Having fixed \({\bar{U}}_+(q)u_+(p)\), we can calculate \({\bar{U}}_+(q)u_-(p)\), \({\bar{U}}_+(q)v_-(p)\) and \({\bar{U}}_+(q)v_+(p)\) by using Eq. (46). The remaining scalar bilinears are then obtained from Eq. (28). Tensorial structures are obtained from Table I and Table II. So, this completes our calculation of all Dirac bilinears in terms of Lorentz scalars.

4 Conclusion

We have discussed how to write down Dirac bilinears purely in terms of Lorentz scalars, and observed that all calculations boil down to determining the scalar structures. Retaining Lorentz covariance in the calculations is not a straightforward task, and this fact has revealed itself in our discussion as well. We have shown that the scalar bilinears can be calculated up to an overall phase (this phase is, in general, a function of the Lorentz vectors), and we have given a simple prescription to fix this phase covariantly. So, we have achieved the goal to write bilinear structures in a completely covariant manner.