betaFeedback

WorksPhysics and Astronomy

An approximate separation of the electron and positron degrees of freedom in the Dirac-Heisenberg-Wigner formalism

Abstract

Here we study the quantum relativistic phase-space description of the Dirac field referred to as the Dirac–Heisenberg–Wigner (DHW) formalism. By applying a systematic expansion procedure in weak electromagnetic fields (compared to the Schwinger critical field) and slow temporal and long spatial scales (compared to the Compton scales), the 16 scalar phase-space functions of the DHW equations can be reduced to 8 scalars. By making a further change of variables, the equations can be interpreted as four evolution equations for the electron degrees of freedom and four for the positron degrees of freedom. We refer to these equations as the weakly coupled electron-positron (WCEP) limit of the DHW equations. Unlike the case after a Foldy–Wouthuysen transformation has been made, the electron and positron degrees of freedom remain weakly coupled. Naturally, further approximations can be made on the 4+4 equations so that only 4 equations for one particle species remain. Importantly, the expansion scheme can be carried out to all orders, in which case the reduced 4+4 model, contrary to other reduced models, will cover most quantum relativistic effects. Specifically, this includes vacuum polarization, a phenomenon that otherwise has been added phenomenologically, but here appears naturally as a feature of the general expansion scheme.

Introduction

Quantum kinetic theory extends the classical phase-space description of plasmas to regimes where the wave nature of particles, Fermi statistics, and spin properties influence the dynamics. Such regimes occur in dense plasmas and condensed matter systems. But quantum effects can also be important when electromagnetic fields or their gradients are sufficiently strong. The appropriate description depends on which effects must be kept and on the relevant spatial and temporal scales, and the electromagnetic field strengths. Reviews of various kinetic models are given by e.g. Refs. [1, 2, 3]. Similar considerations for fluid modes are given, for example, in Ref. [4]. More general quantum kinetic approaches also incorporate exchange and correlation effects beyond the mean-field approximation [5]. In the present paper, we consider relativistic quantum plasmas within a collisionless mean-field description, with particular emphasis on the relation between particle transport, spin dynamics, and the response of the Dirac vacuum.

For non-relativistic electrons, a Wigner transformation of the Schrödinger equation provides a kinetic description of (quantum) particle dispersion. Starting instead from the Pauli equation introduces spin precession and the magnetic dipole force, together with the magnetization current. These effects can be described using coupled scalar and vector distribution functions, as in the semi-classical model of Hurst et al. [6]. Quantum hydrodynamic models provide an alternative description when a fluid closure can be made; examples are hydrodynamic models put forward in Refs. [7, 8, 9, 10]. The kinetic treatment keeps the momentum dependence needed to describe resonant wave–particle interactions in spin-polarized plasmas [11, 12, 13, 14].

Relativistic extensions introduce further couplings between spin dynamics and orbital motion. In the weakly relativistic regime, spin–orbit interaction, Thomas precession, and spin-dependent polarization enter the kinetic description [15, 16]. Fully relativistic spin kinetics has also been developed, i.e., without expanding in the particle momentum [17]. In that particular formulation, a spin transform introduces spin as an additional independent variable, whereas scalar and vector formulations encode the spin dependence in a set of phase-space functions.

In general, variables used in quantum relativistic formulations must be distinguished carefully: a spin-like vector constructed from Dirac bilinears does not necessarily coincide with the spin distribution in the particle rest frame. The relativistic precession of the latter is governed by the Bargmann–Michel–Telegdi (BMT) equation [18].

A broad description follows by applying a gauge-invariant Wigner transformation directly to the Dirac theory. Covariant quantum transport equations were developed by Vasak et al. [19], and the Dirac–Heisenberg–Wigner (DHW) formalism with a similar scope was formulated by Bialynicki–Birula et al. [20]. While the DHW formalism is not covariant, it has the advantage of avoiding extra constraint equations (in addition to the evolution equations), which are part of the covariant formulation. The usefulness of the DHW theory for problems involving electron-positron pair production (through the Schwinger mechanism and by high-frequency fields [21, 22, 23, 24]), for self-consistent plasma dynamics [25, 26, 27], and also for applications to interactions in graphene materials [28, 29] has been well established in the literature.

The full DHW description is, however, more extensive than what is needed when the fields are weak compared with the Schwinger critical fields (electric field Ec≈1.3×1018 V/mE_c \approx1.3 \times10^{18}\ \mathrm{V/m} and magnetic field Bc≈4.4×109 TB_c \approx4.4 \times10^9\ \mathrm{T}) and vary slowly on the Compton scales. A common route to reduce relativistic kinetics uses a Foldy–Wouthuysen (FW) transformation (see e.g. [17, 30, 31]) to obtain a Hamiltonian for a selected energy sector, corresponding to electrons or positrons. The resulting effective Hamiltonian keeps quantum corrections within that sector. A different objective, which is adopted in the present paper, is to reduce the DHW equations while keeping the quantum terms at a prescribed order, including terms that couple the approximately identified electron and positron distributions. This distinction is particularly relevant when the vacuum response is to be kept together with the dynamics of real particles.

An electrostatic realization of this approach, limited to a 1D geometry, was recently developed in Ref. [32]. Weak-field and slow-time expansions resulted in electron and positron distribution functions coupled through quantum corrections. The resulting description contained both vacuum polarization and additional contributions that depend on the real-particle distributions. This provides a motivation for extending the reduction scheme to general electromagnetic fields and a full 3D geometry, where spin and magnetization variables must also be included.

In the present work, we derive such a reduced model for the general 3D case as described above. By successively eliminating the DHW phase-space functions, we reduce the original 16 variables to eight independent scalar quantities. A further change of dependent variables expresses the remaining functions as one scalar and one three-component vector for each particle species of the Dirac field. We refer to this description as the weakly coupled electron–positron (WCEP) limit of the DHW equations. The transport operators separate into two semi-classical relativistic operators, while the quantum source terms keep the coupling between them. Further approximations reduce the system to a description of a single species. In that limit, we recover the well-known magnetic dipole terms and show how the vector variables are related to rest-frame BMT precession, including the Thomas factor (see e.g. Ref. [33]). Furthermore, we demonstrate that the elimination procedure can be continued to all orders. This allows vacuum contributions to be kept in the reduced description and in the reconstructed electromagnetic sources. We discuss how the procedure can be used to recover the effect of vacuum polarization, including the need for a charge renormalization.

The paper is organized as follows. Sec. III summarizes the DHW formalism and the operators used in the expansion. Sec. III derives the reduced scalar and vector equations and compares their single-species limits with established spin kinetic models. Sec. IV describes the extension to arbitrary perturbative order and the inclusion of vacuum polarization. The final section summarizes the results and discusses their implications.

The Dirac–Heisenberg–Wigner formalism

In this section we give an introduction to the Dirac–Heisenberg–Wigner (DHW) formalism, see e.g. Refs. [2, 20, 34]. The theory is based on the Dirac equation, and use a Wigner transform to obtain a phase-space description. Unlike theories based on the Schrödinger or Pauli equations, the electron and positron degrees of freedom are not separated from the start. Thus the same set of variables can describe plasma dynamics as well as vacuum polarization and electron–positron pair creation. For a discussion of radiation reaction, which lies outside the present DHW model, see Ref. [35].

The starting point is a gauge-invariant Wigner transformation of the equal-time density operator. Since the Dirac wave function has four components, the Wigner function is a 4×44 \times4 matrix. Expanding this matrix in the generators of the Clifford algebra gives 16 real phase-space functions. Following Ref. [20], we write these as four scalars, f0f_0, f1f_1, f2f_2, f3f_3 and four three-vectors, g0\mathbf{g}_0, g1\mathbf{g}_1, g2\mathbf{g}_2, g3\mathbf{g}_3, where all functions depend on (r,p,t)(\mathbf{r},\mathbf{p},t), with r\mathbf{r}, p\mathbf{p}, and tt denoting position, kinetic momentum, and time, respectively. Furthermore, we will use p=∣p∣p=|\mathbf{p}|. The phase-space variables have clear physical interpretations, although they cannot be viewed as ordinary particle distribution functions. For example, f0f_0 is associated with charge density, g1\mathbf{g}_1 with current density, f3f_3 with mass density, g0\mathbf{g}_0 with spin density, and g3\mathbf{g}_3 with magnetization density, see e.g. Ref. [20] for further elaborations. We stress that none of the individual functions describes electrons or positrons separately.

We will use SI units throughout. Moreover, we denote the electric and magnetic fields by E(r,t)\mathbf{E}(\mathbf{r},t) and B(r,t)\mathbf{B}(\mathbf{r},t), mm is the electron rest mass, cc is the speed of light, and ℏ\hbar is the reduced Planck constant. The gradients ∇r\nabla_{\mathbf{r}} and ∇p\nabla_{\mathbf{p}} act with respect to position and momentum, respectively, and ∂t=∂/∂t\partial_t=\partial/\partial t.

Treating the electromagnetic field as a classical mean field, i.e., using the Hartree approximation, the DHW equations are written as

Dtf0+cD⋅g1=0,(1)D_t f_0+c\mathbf{D}\cdot\mathbf{g}_1=0, \tag*{(1)}
Dtf1+cD⋅g0=−2mc2ℏf2,(2)D_t f_1+c\mathbf{D}\cdot\mathbf{g}_0=-\frac{2mc^2}{\hbar}f_2, \tag*{(2)}
Dtf2+2cℏP⋅g3=2mc2ℏf1,(3)D_t f_2+\frac{2c}{\hbar}\mathbf{P}\cdot\mathbf{g}_3=\frac{2mc^2}{\hbar}f_1, \tag*{(3)}
Dtf3−2cℏP⋅g2=0,(4)D_t f_3-\frac{2c}{\hbar}\mathbf{P}\cdot\mathbf{g}_2=0, \tag*{(4)}
Dtg0+cDf1−2cℏP×g1=0,(5)D_t\mathbf{g}_0+c\mathbf{D}f_1-\frac{2c}{\hbar}\mathbf{P}\times\mathbf{g}_1=0, \tag*{(5)}
Dtg1+cDf0−2cℏP×g0=−2mc2ℏg2,(6)D_t\mathbf{g}_1+c\mathbf{D}f_0-\frac{2c}{\hbar}\mathbf{P}\times\mathbf{g}_0=-\frac{2mc^2}{\hbar}\mathbf{g}_2, \tag*{(6)}
Dtg2+cD×g3+2cℏPf3=2mc2ℏg1,(7)D_t\mathbf{g}_2+c\mathbf{D}\times\mathbf{g}_3+\frac{2c}{\hbar}\mathbf{P}f_3=\frac{2mc^2}{\hbar}\mathbf{g}_1, \tag*{(7)}
Dtg3−cD×g2−2cℏPf2=0.(8)D_t\mathbf{g}_3-c\mathbf{D}\times\mathbf{g}_2-\frac{2c}{\hbar}\mathbf{P}f_2=0. \tag*{(8)}

Here ee is the charge used in the Dirac equation [36]. The nonlocal operators appearing above are given by

Dt=∂∂t+e∫−1/21/2dλ E(r+iℏλ∇p,t)⋅∇p,(9)D_t = \frac{\partial}{\partial t} + e \int_{-1/2}^{1/2} d\lambda\, \mathbf{E}(\mathbf{r} + i\hbar\lambda\nabla_p,t) \cdot\nabla_p, \tag*{(9)}
D=∇r+e∫−1/21/2dλ B(r+iℏλ∇p,t)×∇p,(10)\mathbf{D} = \nabla_r + e \int_{-1/2}^{1/2} d\lambda\, \mathbf{B}(\mathbf{r} + i\hbar\lambda\nabla_p,t) \times\nabla_p, \tag*{(10)}
P=p−ieℏ∫−1/21/2dλ λB(r+iℏλ∇p,t)×∇p.(11)\mathbf{P} = \mathbf{p} - ie\hbar\int_{-1/2}^{1/2} d\lambda\,\lambda\mathbf{B}(\mathbf{r} + i\hbar\lambda\nabla_p,t) \times\nabla_p. \tag*{(11)}

The shifted field arguments are a short way of writing a Taylor series. Accordingly, expanding the electric field gives spatial derivatives of E\mathbf{E}, multiplied by momentum derivatives acting on the DHW functions.

For the derivations in the sections to follow, we separate the local contributions from the first nonlocal corrections. Expanding the operators gives

Dt=(∂∂t+eE⋅∇p)−eℏ224(∇r⋅∇p)2E⋅∇p+O(ℏ4),(12)D_t = \left(\frac{\partial}{\partial t} + e\mathbf{E}\cdot\nabla_p\right) - \frac{e\hbar^2}{24}(\nabla_r\cdot\nabla_p)^2\mathbf{E}\cdot\nabla_p + O(\hbar^4), \tag*{(12)}
D=(∇r+eB×∇p)−eℏ224(∇r⋅∇p)2B×∇p+O(ℏ4),(13)\mathbf{D} = \left(\nabla_r + e\mathbf{B}\times\nabla_p\right) - \frac{e\hbar^2}{24}(\nabla_r\cdot\nabla_p)^2\mathbf{B}\times\nabla_p + O(\hbar^4), \tag*{(13)}
P=p+eℏ212(∇r⋅∇p)B×∇p+O(ℏ4).(14)\mathbf{P} = \mathbf{p} + \frac{e\hbar^2}{12}(\nabla_r\cdot\nabla_p)\mathbf{B}\times\nabla_p + O(\hbar^4). \tag*{(14)}

In these expressions, ∇r\nabla_r acts on the field immediately following it, whereas ∇p\nabla_p acts on the phase-space function to the right of the operator. For scale lengths much longer than the relevant de Broglie wavelength, we can use the local approximation,

Dt→Dt,loc,D→Dloc,P→p.(15)D_t \to D_{t,\mathrm{loc}},\quad\mathbf{D} \to\mathbf{D}_{\mathrm{loc}},\quad\mathbf{P} \to\mathbf{p}. \tag*{(15)}

The corrections to this approximation describe quantum effects associated with the dispersive properties of the wave-function. Below we will keep these corrections to a given order.

Together with Maxwell’s equations, the DHW equations constitute a closed system. In the DHW formalism, the source terms in Maxwell’s equations due to the Dirac field are given by

ρ=e(2πℏ)3∫f0d3p(16)\rho= \frac{e}{(2\pi\hbar)^3}\int f_0d^3p \tag*{(16)}

and

j=ec(2πℏ)3∫g1d3p(17)\mathbf{j} = \frac{ec}{(2\pi\hbar)^3}\int\mathbf{g}_1d^3p \tag*{(17)}

respectively, where ρ\rho is the charge density, and j\mathbf{j} is the current density. The factor (2πℏ)−3(2\pi\hbar)^{-3} is the phase-space normalization for the dimensionless DHW functions used here.

A key difference from classical kinetic theory is that the DHW functions do not need to vanish when no real particles are present. They are expectation values of field operators, and have nonzero vacuum values. For the unperturbed vacuum these are given by

f3,vac(p)=−2mc2ϵp,(18)f_{3,\mathrm{vac}}(\mathbf{p}) = -\frac{2mc^2}{\epsilon_p}, \tag*{(18)}
g1,vac(p)=−2cpϵp,(19)g_{1,\mathrm{vac}}(\mathbf{p}) = -\frac{2c\mathbf{p}}{\epsilon_p}, \tag*{(19)}

where

ϵp=m2c4+c2p2,(20)\epsilon_p = \sqrt{m^2c^4+c^2p^2}, \tag*{(20)}

is the particle energy. All other components vanish for a vacuum that is not spin-polarized [37]. The expressions above follow from evaluating the Wigner operator in the free-field vacuum. Note that the momentum integral of g1,vacg_{1,\mathrm{vac}} vanishes as the phase-space vacuum current density is an odd function of momentum.

In general, the phase-space functions contain both particle and vacuum contributions. When we keep the vacuum response, however, ultraviolet divergences typically occur, and renormalization is needed [20, 32, 38], as we will come back to in Sec. IV B.

In the present work, we will use the fact that the DHW equations contain both positive- and negative-energy degrees of freedom, corresponding to positrons and electrons. In the weak-quantum limit, suitable combinations of fif_i and gig_i can approximately separate the two species. As we will see, somewhat different combinations are needed for the scalar and vector functions. Moreover, the spin variables obtained in this way differ from the usual rest-frame spin distribution.

Derivations of the WCEP equations

In this section, we present a systematic expansion of the DHW equations, aiming to find an approximate separation of the electron and positron degrees of freedom. In § 1.2 A, we perform the separation for the scalar phase-space variables, and in § 1.2 B we do the same for the vector variables of the DHW formalism. Finally, in § 1.2 C we demonstrate agreement with previous works in appropriate limiting cases.

Scalar evolution equations

We first derive the scalar equations, keeping quantum corrections up to second order. The momentum dependence is kept fully relativistic, i.e., we do not expand in p/(mc)p/(mc). In addition to the expansion of operators, we assume weak spin polarization, i.e., first order in a weak quantum expansion. Weak spin polarization is not required by the general procedure; we assume it here to determine which quantum coupling terms to keep at each order. The above scaling gives us the ordering

g0, g3=O(ℏ),f0, f3, g1=O(1).(21)\begin{aligned} \mathbf{g}_0,\ \mathbf{g}_3 &= \mathcal{O}(\hbar),\\ f_0,\ f_3,\ \mathbf{g}_1 &= \mathcal{O}(1). \tag*{(21)} \end{aligned}

The powers of ℏ\hbar here specify the order of the expansion. As will become apparent below, independent oscillations at the Compton time scale ℏ/(mc2)\hbar/(mc^2) are eliminated by the expansion procedure. Next, we write the operators as

Dt=Dt,loc+δDt,D=Dloc+δD,P=p+δP,(22)\begin{aligned} D_t &= D_{t,\mathrm{loc}}+\delta D_t,\\ \mathbf{D} &= \mathbf{D}_{\mathrm{loc}}+\delta\mathbf{D},\\ \mathbf{P} &= \mathbf{p}+\delta\mathbf{P}, \tag*{(22)} \end{aligned}

where the corrections are the terms in Eqs. (12)–(14) of order ℏ2\hbar^2. Operator products act successively on the expression to their right. Note that momentum derivatives act on explicit momentum-dependent factors as well as on the phase-space functions.

From Eqs. (1f) and (1g) we immediately obtain

g2=1mcP×g0−ℏ2mc2(Dtg1+cDf0),(23)\begin{aligned} \mathbf{g}_2 &= \frac{1}{mc}\mathbf{P}\times\mathbf{g}_0 -\frac{\hbar}{2mc^2}\left(D_t\mathbf{g}_1+c\mathbf{D}f_0\right), \tag*{(23)} \end{aligned}
g1=1mcPf3+ℏ2mc2(Dtg2+cD×g3).(24)\begin{aligned} \mathbf{g}_1 &= \frac{1}{mc}\mathbf{P}f_3 +\frac{\hbar}{2mc^2}\left(D_t\mathbf{g}_2+c\mathbf{D}\times\mathbf{g}_3\right). \tag*{(24)} \end{aligned}

We solve these relations for g1\mathbf{g}_1 and g2\mathbf{g}_2 by successive substitutions. Inserting the lowest-order result g1=pf3/(mc)\mathbf{g}_1=\mathbf{p}f_3/(mc) gives the first-order expression for g2\mathbf{g}_2, denoted by g2(1)\mathbf{g}_2^{(1)}, where the superscript indicates truncation at first order:

g2(1)=p×g0mc−ℏ2mc2[Dt,loc(pf3mc)+cDlocf0].(25)\mathbf{g}_2^{(1)}=\frac{\mathbf{p}\times\mathbf{g}_0}{mc} -\frac{\hbar}{2mc^2}\left[D_{t,\mathrm{loc}}\left(\frac{\mathbf{p}f_3}{mc}\right)+c\mathbf{D}_{\mathrm{loc}}f_0\right]. \tag*{(25)}

Here the time derivative also acts on p\mathbf{p}, giving Dt,locp=eED_{t,\mathrm{loc}}\mathbf{p}=e\mathbf{E}. Inserting this expression for g2\mathbf{g}_2 back into Eq. (24), we find, to second order,

g1=pf3mc+δg1,\mathbf{g}_1=\frac{\mathbf{p}f_3}{mc}+\delta\mathbf{g}_1,
δg1=δPf3mc+ℏ2mc2(Dt,locg2(1)+cDloc×g3).(26)\delta\mathbf{g}_1=\frac{\delta\mathbf{P}f_3}{mc} +\frac{\hbar}{2mc^2}\left(D_{t,\mathrm{loc}}\mathbf{g}_2^{(1)}+c\mathbf{D}_{\mathrm{loc}}\times\mathbf{g}_3\right). \tag*{(26)}

Thus g2(1)\mathbf{g}_2^{(1)} is of order ℏ\hbar, and the correction δg1\delta\mathbf{g}_1 is of order ℏ2\hbar^2. Substitution into Eq. (1a), using Dloc⋅(pf3)=p⋅Dlocf3\mathbf{D}_{\mathrm{loc}}\cdot(\mathbf{p}f_3)=\mathbf{p}\cdot\mathbf{D}_{\mathrm{loc}}f_3, now gives

Dt,locf0+pm⋅Dlocf3=Q0,(27)D_{t,\mathrm{loc}}f_0+\frac{\mathbf{p}}{m}\cdot\mathbf{D}_{\mathrm{loc}}f_3=Q_0, \tag*{(27)}

with

Q0=−δDtf0−1mδD⋅(pf3)−cDloc⋅δg1.(28)Q_0=-\delta D_t f_0-\frac{1}{m}\delta\mathbf{D}\cdot(\mathbf{p}f_3)-c\mathbf{D}_{\mathrm{loc}}\cdot\delta\mathbf{g}_1. \tag*{(28)}

For the second scalar equation, we insert Eq. (23) into Eq. (1d). Before expanding the operators, the result is

Dtf3=2mℏP⋅(P×g0)−1mcP⋅Dtg1−1mP⋅Df0.(29)D_t f_3=\frac{2}{m\hbar}\mathbf{P}\cdot(\mathbf{P}\times\mathbf{g}_0)-\frac{1}{mc}\mathbf{P}\cdot D_t\mathbf{g}_1-\frac{1}{m}\mathbf{P}\cdot\mathbf{D}f_0. \tag*{(29)}

The first term is zero if P\mathbf{P} is replaced by p\mathbf{p}. However, the nonlocal correction gives

P⋅(P×g0)=p⋅(δP×g0)+δP⋅(p×g0)+higher-order terms.(30)\mathbf{P}\cdot(\mathbf{P}\times\mathbf{g}_0)=\mathbf{p}\cdot(\delta\mathbf{P}\times\mathbf{g}_0)+\delta\mathbf{P}\cdot(\mathbf{p}\times\mathbf{g}_0)+\text{higher-order terms}. \tag*{(30)}

Since g0\mathbf{g}_0 is of order ℏ\hbar, the factor 1/ℏ1/\hbar in Eq. (29) makes this contribution second order. It must therefore be kept. Inserting Eq. (26) and moving the local terms to the left-hand side gives

γp2Dt,locf3+eE⋅pm2c2f3+pm⋅Dlocf0=Q3,(31)\gamma_p^2D_{t,\mathrm{loc}}f_3+\frac{e\mathbf{E}\cdot\mathbf{p}}{m^2c^2}f_3+\frac{\mathbf{p}}{m}\cdot\mathbf{D}_{\mathrm{loc}}f_0=Q_3, \tag*{(31)}

where

γp=ϵpmc2=1+p2m2c2,\gamma_p=\frac{\epsilon_p}{mc^2}=\sqrt{1+\frac{p^2}{m^2c^2}},

where the quantum corrections on the right-hand side are

Q3=−δDtf3−1mcδP⋅Dt,loc(pf3mc)−1mcp⋅δDt(pf3mc)−1mcp⋅Dt,locδg1−1mδP⋅Dlocf0−1mp⋅δDf0+2mℏp⋅(δP×g0)+2mℏδP⋅(p×g0).(32)\begin{aligned} Q_3={}&-\delta D_t f_3 -\frac{1}{mc}\delta\mathbf{P}\cdot D_{t,\mathrm{loc}}\left(\frac{\mathbf{p}f_3}{mc}\right) -\frac{1}{mc}\mathbf{p}\cdot\delta D_t\left(\frac{\mathbf{p}f_3}{mc}\right)\\ &-\frac{1}{mc}\mathbf{p}\cdot D_{t,\mathrm{loc}}\delta\mathbf{g}_1 -\frac{1}{m}\delta\mathbf{P}\cdot\mathbf{D}_{\mathrm{loc}}f_0 -\frac{1}{m}\mathbf{p}\cdot\delta\mathbf{D}f_0\\ &+\frac{2}{m\hbar}\mathbf{p}\cdot(\delta\mathbf{P}\times\mathbf{g}_0) +\frac{2}{m\hbar}\delta\mathbf{P}\cdot(\mathbf{p}\times\mathbf{g}_0). \tag*{(32)} \end{aligned}

Both Q0Q_0 and Q3Q_3 are of second order. Time derivatives of the scalar functions inside these corrections can therefore be replaced using Eqs. (27) and (31) with Q0=Q3=0Q_0=Q_3=0. Naturally, this replacement does not change the accuracy of the calculation scheme and avoids introducing extra evolution equations.

To make contact with classical kinetic theory, we introduce the variables

f±=12(γpf3±f0).(33)f_{\pm}=\frac{1}{2}\left(\gamma_p f_3\pm f_0\right). \tag*{(33)}

with the inverse relations

f0=f+−f−,f3=f++f−γp.(34)f_0=f_+-f_-,\qquad f_3=\frac{f_++f_-}{\gamma_p}. \tag*{(34)}

This change of variables makes the leading-order operators (the parts surviving in the classical limit) of Eqs. (27) and (31) identical. Moreover, with vp=p/(mγp)=c2p/ϵp\mathbf{v}_p=\mathbf{p}/(m\gamma_p)=c^2\mathbf{p}/\epsilon_p, the two operators are

Dv,±=Dt,loc±vp⋅Dloc=∂∂t±vp⋅∇r+e(E±vp×B)⋅∇p,(35)\begin{aligned} D_{v,\pm}=D_{t,\mathrm{loc}}\pm\mathbf{v}_p\cdot\mathbf{D}_{\mathrm{loc}} \\ &=\frac{\partial}{\partial t}\pm\mathbf{v}_p\cdot\nabla_r +e\left(\mathbf{E}\pm\mathbf{v}_p\times\mathbf{B}\right)\cdot\nabla_p, \tag*{(35)} \end{aligned}

coinciding with the classical relativistic Vlasov operator. To fully appreciate this result, one must recognize the built-in feature of the DHW formalism: for electrons and positrons moving in the same direction, the momentum variable p\mathbf{p} has opposite signs; see, e.g., Ref. [2]. Thus e(E±vp×B)e(\mathbf{E}\pm\mathbf{v}_p\times\mathbf{B}) represents the classical Lorentz force expression, independent of the sign in front of the magnetic force term.

Next, when combining the scalar equations, we use

Dt,locγp=eE⋅pm2c2γp,D_{t,\mathrm{loc}}\gamma_p=\frac{e\mathbf{E}\cdot\mathbf{p}}{m^2c^2\gamma_p},
p⋅Dlocγp=0.(36)\mathbf{p}\cdot\mathbf{D}_{\mathrm{loc}}\gamma_p=0. \tag*{(36)}

Dividing Eq. (31) by γp\gamma_p, adding or subtracting Eq. (27), and dividing by two, we obtain

Dv,±f±=12(Q3γp±Q0).(37)D_{v,\pm}f_\pm=\frac{1}{2}\left(\frac{Q_3}{\gamma_p}\pm Q_0\right). \tag*{(37)}

The right-hand side is specified by Eqs. (25), (26), (28), and (32), with f0f_0 and f3f_3 expressed through f±f_\pm using Eq. (34). In the classical limit Q0=Q3=0Q_0=Q_3=0, and the equations reduce to Dv,±f±=0D_{v,\pm}f_\pm=0, as expected.

In Eq. (37), we have chosen to keep terms to second order in ℏ\hbar. The expansion in spatial nonlocality is distinct from the slow-time expansion used to eliminate the auxiliary DHW functions. In principle, the latter procedure can be performed while keeping the full spatially nonlocal operators, without assuming their nonlocal parts to be small. The scalar equations can still be written with the left-hand side of Eq. (37), but their right-hand sides must then include contributions omitted in the present spatial-gradient truncation, including products of nonlocal corrections. The general elimination procedure is discussed in Sec. IV. Moreover, if we take the opposite limit (where quantum corrections proportional to the de Broglie length over characteristic scale lengths are negligible compared to quantum corrections associated with the spin), the coupling term in Eq. (37) reduces to

12(Q3γp±Q0)=−ℏ4m[1c2(pmcγp⋅Dt,loc±cDloc)×{Dt,loc(p×g0mc)+cDloc×g3\frac{1}{2}\left(\frac{Q_3}{\gamma_p}\pm Q_0\right) =-\frac{\hbar}{4m}\left[ \frac{1}{c^2}\left(\frac{\mathbf{p}}{mc\gamma_p}\cdot D_{t,\mathrm{loc}}\pm c\mathbf{D}_{\mathrm{loc}}\right) \times\left\{D_{t,\mathrm{loc}}\left(\frac{\mathbf{p}\times\mathbf{g}_0}{mc}\right)+c\mathbf{D}_{\mathrm{loc}}\times\mathbf{g}_3\right. \right.
−ℏ2mc2Dt,loc[Dt,loc(pf3mc)+cDlocf0]}+eγp(∂rjBi)∂pjg0i].(38)-\frac{\hbar}{2mc^2}D_{t,\mathrm{loc}}[D_{t,\mathrm{loc}}(\frac{\mathbf{p}f_3}{mc})+c\mathbf{D}_{\mathrm{loc}}f_0]\} +\frac{e}{\gamma_p}(\partial_{r_j}B_i)\partial_{p_j}g_{0i} ]. \tag*{(38)}

The above expression will be referred to as the semi-classical limit. This approximation keeps the spin-dependent contributions at the order considered, including the magnetic-gradient coupling, while omitting the orbital dispersive corrections. As we will see in § 1.2 C when comparing with previous works, in the weakly relativistic limit Eq. (38) reduces considerably, and that result coincides with previous works.

Spin and magnetization evolution equations

Next we derive the vector equations, using the same slow-time assumption and ordering as above. Since g0\mathbf{g}_0 and g3\mathbf{g}_3 are of order ℏ\hbar, we keep their leading evolution, including scalar source terms of that order. Again, no expansion in p/(mc)\mathbf{p}/(mc) is made. We use Dt,locD_{t,\mathrm{loc}} and Dloc\mathbf{D}_{\mathrm{loc}} as defined in Sec. II. The nonlocal correction to P\mathbf{P} must still be kept in terms containing a factor 1/ℏ1/\hbar.

Using Eqs. (1b)–(1c), we write

f1=P⋅g3mc+ℏ2mc2Dtf2,(39)f_1=\frac{\mathbf{P}\cdot\mathbf{g}_3}{mc}+\frac{\hbar}{2mc^2}D_tf_2, \tag*{(39)}
f2=−ℏ2mc2(Dtf1+cD⋅g0).(40)f_2=-\frac{\hbar}{2mc^2}\left(D_tf_1+c\mathbf{D}\cdot\mathbf{g}_0\right). \tag*{(40)}

It follows that f1=p⋅g3/(mc)+O(ℏ3)f_1=\mathbf{p}\cdot\mathbf{g}_3/(mc)+O(\hbar^3) and f2=O(ℏ2)f_2=O(\hbar^2), although division by ℏ\hbar in Eq. (1h) means we must retain f2f_2 to the order ℏ2\hbar^2 here. Together with the substitution for g1g_1 and g2g_2 used above, this gives

g2=p×g0mc−ℏ2mc2U+O(ℏ3),(41)\mathbf{g}_2=\frac{\mathbf{p}\times\mathbf{g}_0}{mc}-\frac{\hbar}{2mc^2}\mathbf{U}+O(\hbar^3), \tag*{(41)}
g1=(p+δP)f3mc+ℏ2mc2[Dt,loc(p×g0)mc+cDloc×g3]−ℏ24m2c4Dt,locU+O(ℏ4),(42)\mathbf{g}_1=\frac{(\mathbf{p}+\delta\mathbf{P})f_3}{mc} +\frac{\hbar}{2mc^2}\left[\frac{D_{t,\mathrm{loc}}(\mathbf{p}\times\mathbf{g}_0)}{mc} +c\mathbf{D}_{\mathrm{loc}}\times\mathbf{g}_3\right] -\frac{\hbar^2}{4m^2c^4}D_{t,\mathrm{loc}}\mathbf{U}+O(\hbar^4), \tag*{(42)}
U=Dt,loc(pf3mc)+cDlocf0.(43)\mathbf{U}=D_{t,\mathrm{loc}}\left(\frac{\mathbf{p}f_3}{mc}\right)+c\mathbf{D}_{\mathrm{loc}}f_0. \tag*{(43)}

Here δP\delta\mathbf{P} is the second-order correction in Eq. (14). Note that the last displayed term in g1\mathbf{g}_1 cannot be dropped: although it is of second order, the factor 1/ℏ1/\hbar in the g0\mathbf{g}_0 equation, (5), makes its contribution first order.

Inserting these results into the remaining vector equations gives

Dt,locg0=p×Dt,loc(p×g0)m2c2+p×(Dloc×g3)m−Dloc(p⋅g3)m+S0,(44)D_{t,\mathrm{loc}}\mathbf{g}_0 = \frac{\mathbf{p}\times D_{t,\mathrm{loc}}(\mathbf{p}\times\mathbf{g}_0)}{m^2c^2} +\frac{\mathbf{p}\times(\mathbf{D}_{\mathrm{loc}}\times\mathbf{g}_3)}{m} -\frac{\mathbf{D}_{\mathrm{loc}}(\mathbf{p}\cdot\mathbf{g}_3)}{m} +\mathbf{S}_0, \tag*{(44)}
Dt,locg3=Dloc×(p×g0)m−pmDloc⋅g0−pm2c2Dt,loc(p⋅g3)+S3,(45)D_{t,\mathrm{loc}}\mathbf{g}_3 = \frac{\mathbf{D}_{\mathrm{loc}}\times(\mathbf{p}\times\mathbf{g}_0)}{m} -\frac{\mathbf{p}}{m}\mathbf{D}_{\mathrm{loc}}\cdot\mathbf{g}_0 -\frac{\mathbf{p}}{m^2c^2}D_{t,\mathrm{loc}}(\mathbf{p}\cdot\mathbf{g}_3)+\mathbf{S}_3, \tag*{(45)}

where we have introduced the source terms

S0=2mℏP×(Pf3)−ℏ2m2c2p×Dt,locU,(46)\mathbf{S}_0=\frac{2}{m\hbar}\mathbf{P}\times(\mathbf{P}f_3)-\frac{\hbar}{2m^2c^2}\mathbf{p}\times D_{t,\mathrm{loc}}\mathbf{U}, \tag*{(46)}
S3=−ℏ2mcDloc×U.(47)\mathbf{S}_3=-\frac{\hbar}{2mc}\mathbf{D}_{\mathrm{loc}}\times\mathbf{U}. \tag*{(47)}

Only the first nonlocal contribution to the P×(Pf3)\mathbf{P}\times(\mathbf{P}f_3) term is needed. Writing ∂rj=∂/∂rj\partial_{r_j}=\partial/\partial r_j and ∂pj=∂/∂pj\partial_{p_j}=\partial/\partial p_j, we obtain

P×(Pf3)=p×(δPf3)+δP×(pf3)=eℏ212[(∇r⋅B)∇pf3−3(∂rjB)∂pjf3]+O(ℏ4),(48)\mathbf{P}\times(\mathbf{P}f_3)=\mathbf{p}\times(\delta\mathbf{P}f_3)+\delta\mathbf{P}\times(\mathbf{p}f_3) =\frac{e\hbar^2}{12}\left[(\nabla_r\cdot\mathbf{B})\nabla_p f_3-3(\partial_{r_j}\mathbf{B})\partial_{p_j}f_3\right]+O(\hbar^4), \tag*{(48)}

where Cartesian indices run over the three spatial components and repeated Cartesian indices are summed over. Using ∇r⋅B=0\nabla_r\cdot\mathbf{B}=0, the source becomes

2mℏP×(Pf3)=−eℏ2m(∂rjB)∂pjf3+O(ℏ3).(49)\frac{2}{m\hbar}\mathbf{P}\times(\mathbf{P}f_3)=-\frac{e\hbar}{2m}(\partial_{r_j}\mathbf{B})\partial_{p_j}f_3+O(\hbar^3). \tag*{(49)}

To deal with terms that combine momentum-dependent factors with time derivatives, it will be useful to introduce the following notation

Nij=δij+pipjm2c2,N_{ij}=\delta_{ij}+\frac{p_ip_j}{m^2c^2},
Mij=γp2δij−pipjm2c2,M_{ij}=\gamma_p^2\delta_{ij}-\frac{p_ip_j}{m^2c^2},
MijNjk=γp2δik,(50)M_{ij}N_{jk}=\gamma_p^2\delta_{ik}, \tag*{(50)}

where δij\delta_{ij} is the Kronecker delta. We use ϵijk\epsilon_{ijk} for the Levi-Civita symbol. The relations Dt,locpi=eEiD_{t,\mathrm{loc}}p_i=eE_i and Dloc,ipj=eϵikjBkD_{\mathrm{loc},i}p_j=e\epsilon_{ikj}B_k give

p×(Dloc×g)−Dloc(p⋅g)=−(p⋅Dloc)g−eB×g,\mathbf{p}\times(\mathbf{D}_{\mathrm{loc}}\times\mathbf{g})-\mathbf{D}_{\mathrm{loc}}(\mathbf{p}\cdot\mathbf{g})=-(\mathbf{p}\cdot\mathbf{D}_{\mathrm{loc}})\mathbf{g}-e\mathbf{B}\times\mathbf{g},
Dloc×(p×g)−p(Dloc⋅g)=−(p⋅Dloc)g−eB×g.(51)\mathbf{D}_{\mathrm{loc}}\times(\mathbf{p}\times\mathbf{g})-\mathbf{p}(\mathbf{D}_{\mathrm{loc}}\cdot\mathbf{g})=-(\mathbf{p}\cdot\mathbf{D}_{\mathrm{loc}})\mathbf{g}-e\mathbf{B}\times\mathbf{g}. \tag*{(51)}

Here g\mathbf{g} denotes an arbitrary vector phase-space function. Collecting the terms, we can now write the coupled equations in component form as

MijDt,locg0j+pjmDloc,jg3i+em(B×g3)i−em2c2[p×(E×g0)]i=S0i,(52)M_{ij}D_{t,\mathrm{loc}}g_{0j}+\frac{p_j}{m}D_{\mathrm{loc},j}g_{3i}+\frac{e}{m}(\mathbf{B}\times\mathbf{g}_3)_i-\frac{e}{m^2c^2}[\mathbf{p}\times(\mathbf{E}\times\mathbf{g}_0)]_i=S_{0i}, \tag*{(52)}
NijDt,locg3j+pjmDloc,jg0i+em(B×g0)i+epim2c2Ejg3j=S3i.(53)N_{ij}D_{t,\mathrm{loc}}g_{3j}+\frac{p_j}{m}D_{\mathrm{loc},j}g_{0i}+\frac{e}{m}(\mathbf{B}\times\mathbf{g}_0)_i+\frac{ep_i}{m^2c^2}E_jg_{3j}=S_{3i}. \tag*{(53)}

The spatial transport, the magnetic force term, and spin precession all couple g0\mathbf{g}_0 to g3\mathbf{g}_3. To separate the two transport equations into electron and positron degrees of freedom, we introduce variables that are momentum-dependent linear combinations of the spin and magnetization vectors, according to

F±i=g0i±Aijg3j(54)F_{\pm i}=g_{0i}\pm A_{ij}g_{3j} \tag*{(54)}

where

Aij=1γp(δij+pipjm2c2)(55)A_{ij}=\frac{1}{\gamma_p}\left(\delta_{ij}+\frac{p_ip_j}{m^2c^2}\right) \tag*{(55)}

with the inverse

Aij−1=γpδij−pipjm2c2γp.(56)A^{-1}_{ij}=\gamma_p\delta_{ij}-\frac{p_ip_j}{m^2c^2\gamma_p}. \tag*{(56)}

Inverting the change of variables gives us

g0i=F+i+F−i2,g_{0i}=\frac{F_{+i}+F_{-i}}{2},
g3i=12(γpδij−pipjm2c2γp)(F+j−F−j).(57)g_{3i}=\frac{1}{2}\left(\gamma_p\delta_{ij}-\frac{p_ip_j}{m^2c^2\gamma_p}\right)(F_{+j}-F_{-j}). \tag*{(57)}

Since the coefficients AijA_{ij} in (55) depend on momentum, their derivatives must be included when completing this change of variables. In particular, we note that

Dt,locAij=e(Eipj+piEj)m2c2γp−eEkpkm2c2γp3Nij,D_{t,\mathrm{loc}}A_{ij}=\frac{e(E_ip_j+p_iE_j)}{m^2c^2\gamma_p}-\frac{eE_kp_k}{m^2c^2\gamma_p^3}N_{ij},
pkDloc,kAij=−eϵiℓkBℓAkj+eAikϵkℓjBℓ.(58)p_kD_{\mathrm{loc},k}A_{ij}=-e\epsilon_{i\ell k}B_\ell A_{kj}+eA_{ik}\epsilon_{k\ell j}B_\ell. \tag*{(58)}

Next, we multiply (52) by Nij/γp2N_{ij}/\gamma_p^2, and (53) by ±δij/γp\pm\delta_{ij}/\gamma_p, and add the results. Using the relations above, the result is

Dv,±F±±emγpB×F±−em2c2γp2p×(E×F±)=S±,(59)D_{v,\pm}F_\pm\pm\frac{e}{m\gamma_p}\mathbf{B}\times F_\pm-\frac{e}{m^2c^2\gamma_p^2}\mathbf{p}\times(\mathbf{E}\times F_\pm)=\mathbf{S}_\pm, \tag*{(59)}

where

S±i=1γp2(δij+pipjm2c2)S0j±S3iγp,(60)S_{\pm i}=\frac{1}{\gamma_p^2}\left(\delta_{ij}+\frac{p_ip_j}{m^2c^2}\right)S_{0j}\pm\frac{S_{3i}}{\gamma_p}, \tag*{(60)}

and we have reused the Vlasov operator from (35). The key property of (59) is that the left-hand side contains a transport operator corresponding to semi-classical particle orbits for electrons/positrons, just as in Eq. (37), with the difference that the vector variables F±\mathbf{F}_{\pm} are also subject to spin precession. More specifically, the second term of Eq. (59) describes magnetic spin precession. The term that follows is related to the Thomas precession contribution, but has a double cross product, and does not in general preserve ∣F±∣\lvert\mathbf{F}_{\pm}\rvert. Thus F±\mathbf{F}_{\pm} cannot be identified directly with the usual rest-frame spin distribution, which is the reason the well-known expression for Thomas precession is hidden. We will return to this point in Sec. III C and show that Eq. (59) is nevertheless consistent with the well-known result regarding Thomas precession; see, e.g., Ref. [33].

We next give the scalar sources without time derivatives of the phase-space functions. To do this, we use the leading scalar equations,

Dt,locf0=−pjmDloc,jf3,Dt,locf3=−1γp2(pjmDloc,jf0+eEjpjm2c2f3).(61)D_{t,\mathrm{loc}}f_0=-\frac{p_j}{m}D_{\mathrm{loc},j}f_3, \qquad D_{t,\mathrm{loc}}f_3=-\frac{1}{\gamma_p^2}\left(\frac{p_j}{m}D_{\mathrm{loc},j}f_0+\frac{eE_jp_j}{m^2c^2}f_3\right). \tag*{(61)}

Substituting the second of these equations into the definition of U\mathbf{U}, we obtain

Ui=c(N−1)ijDloc,jf0+emc(N−1)ijEjf3,(N−1)ij=δij−pipjm2c2γp2.(62)U_i=c(N^{-1})_{ij}D_{\mathrm{loc},j}f_0+\frac{e}{mc}(N^{-1})_{ij}E_jf_3, \qquad (N^{-1})_{ij}=\delta_{ij}-\frac{p_ip_j}{m^2c^2\gamma_p^2}. \tag*{(62)}

Next, we note that the commutator needed when differentiating U\mathbf{U} is

[Dt,loc,Dloc,i]=eϵijk(∂tBj)∂pk−e(∂riEj)∂pj=−e(∂rjEi)∂pj.(63)[D_{t,\mathrm{loc}},D_{\mathrm{loc},i}] =e\epsilon_{ijk}(\partial_tB_j)\partial_{p_k} -e(\partial_{r_i}E_j)\partial_{p_j} =-e(\partial_{r_j}E_i)\partial_{p_j}. \tag*{(63)}

where the last equality follows from Faraday’s law, ∂tB=−∇r×E\partial_t\mathbf{B}=-\nabla_r\times\mathbf{E}. The derivative of (N−1)ij(N^{-1})_{ij} is

Dt,loc(N−1)ij=−e(Eipj+piEj)m2c2γp2+2eEkpkpipjm4c4γp4.(64)D_{t,\mathrm{loc}}(N^{-1})_{ij} =-\frac{e(E_ip_j+p_iE_j)}{m^2c^2\gamma_p^2} +\frac{2eE_kp_kp_ip_j}{m^4c^4\gamma_p^4}. \tag*{(64)}

Next, we use these expressions and replace the temporal derivatives of the scalar variables by Eq. (61). We write Ut,i≡Dt,locUiU_{t,i}\equiv D_{t,\mathrm{loc}}U_i for the evaluated convective derivative, to distinguish it from ∂tUi\partial_tU_i. Its components are

Ut,i=−c(N−1)ijDloc,j(pkmDloc,kf3)−ec(N−1)ij(∂rkEj)∂pkf0+(Dt,loc(N−1)ij)(cDloc,jf0+eEjmcf3)+emc(N−1)ij(∂tEj)f3−e(N−1)ijEjmcγp2(pkmDloc,kf0+eEkpkm2c2f3).(65)\begin{aligned} U_{t,i}={}&-c(N^{-1})_{ij}D_{\mathrm{loc},j}\left(\frac{p_k}{m}D_{\mathrm{loc},k}f_3\right) -ec(N^{-1})_{ij}(\partial_{r_k}E_j)\partial_{p_k}f_0\\ &+\left(D_{t,\mathrm{loc}}(N^{-1})_{ij}\right)\left(cD_{\mathrm{loc},j}f_0+\frac{eE_j}{mc}f_3\right) +\frac{e}{mc}(N^{-1})_{ij}(\partial_tE_j)f_3\\ &-\frac{e(N^{-1})_{ij}E_j}{mc\gamma_p^2} \left(\frac{p_k}{m}D_{\mathrm{loc},k}f_0+\frac{eE_kp_k}{m^2c^2}f_3\right). \tag*{(65)} \end{aligned}

This expression contains spatial and momentum derivatives of the scalar functions, together with ∂tE\partial_t\mathbf{E}. The replacement of the scalar time derivatives changes the vector equations only at order ℏ2\hbar^2. The sources are consequently

S0=−eℏ2m(∂rjB)∂pjf3−ℏ2m2c3p×Ut,(66)\mathbf{S}_0=-\frac{e\hbar}{2m}(\partial_{r_j}\mathbf{B})\partial_{p_j}f_3-\frac{\hbar}{2m^2c^3}\mathbf{p}\times\mathbf{U}_t, \tag*{(66)}
S3=−ℏ2mc[∇r×U+e∇p(B⋅U)−eB(∇p⋅U)].(67)\mathbf{S}_3=-\frac{\hbar}{2mc}\left[\nabla_r\times\mathbf{U}+e\nabla_p(\mathbf{B}\cdot\mathbf{U})-e\mathbf{B}(\nabla_p\cdot\mathbf{U})\right]. \tag*{(67)}

where we can use Eq. (34) to express f3f_3 and f0f_0 in terms of our new variables f±f_{\pm}. Similarly, the electromagnetic sources in Maxwell’s equations can be expressed in terms of the new variables. Using the normalization of Eqs. (16) and (10), the charge density is

ρ=e(2πℏ)3∫d3p (f+−f−).(68)\rho=\frac{e}{(2\pi\hbar)^3}\int d^3p\,(f_+-f_-). \tag*{(68)}

Furthermore, applying the same approximation scheme as when deriving the governing equations, the current source g1\mathbf{g}_1 can be expressed in terms of f±f_{\pm} and F±\mathbf{F}_{\pm}. This can be done to any order in the expansion of g1\mathbf{g}_1, but below, for simplicity, we display only the transport current and the leading spin-dependent polarization and magnetization contributions. Spin-independent corrections of order ℏ2\hbar^2 are omitted; these can be restored directly from the reconstructed expression for g1\mathbf{g}_1. With this additional approximation, the current density can be written as

j=jtr+∂tP+∇r×M,(69)\mathbf{j}=\mathbf{j}_{\mathrm{tr}}+\partial_t\mathbf{P}+\nabla_r\times\mathbf{M}, \tag*{(69)}

where the free current density is

jtr=e(2πℏ)3∫d3p pγp(f++f−).(70)\mathbf{j}_{\mathrm{tr}}=\frac{e}{(2\pi\hbar)^3}\int d^3p\,\mathbf{p}\gamma_p(f_++f_-). \tag*{(70)}

The polarization is

P=eℏ4m2c2(2πℏ)3∫d3p p×(F++F−)(71)\mathbf{P}=\frac{e\hbar}{4m^2c^2(2\pi\hbar)^3}\int d^3p\,\mathbf{p}\times(\mathbf{F}_++\mathbf{F}_-) \tag*{(71)}

and the magnetization components are

Mi=eℏ4m(2πℏ)3∫d3p Aij−1(F+j−F−j),(72)\mathcal{M}_i=\frac{e\hbar}{4m(2\pi\hbar)^3}\int d^3p\,A^{-1}_{ij}(\mathbf{F}_{+j}-\mathbf{F}_{-j}), \tag*{(72)}

with Aij−1A^{-1}_{ij} given by Eq. (56).

Eqs. (37) and (52) are the main results of our paper. A consistently truncated closed system is obtained by combining these equations with Maxwell’s equations and reconstructing the electromagnetic sources to the same order as the evolution equations. Naturally, we then also express the right-hand source terms in terms of the new variables f±f_{\pm} and F±\mathbf{F}_{\pm}, using Eqs. (34) and (50).

We will refer to this set as the weakly coupled electron–positron (WCEP) equations. The key feature is that the electron and positron degrees of freedom are separated, with the help of the variable changes made in Eqs. (33) and (47). However, contrary to the case when a FW-transformation is made, there is a weak coupling between the electrons and positrons through the coupling terms in the right-hand sides of Eqs. (37) and (52). We note that one can approximate the system by keeping only one species (electrons or positrons) and disregarding the electron–positron cross-coupling, provided it is sufficiently weak to be dropped. However, we will show in Section IV that it is necessary to keep the electron–positron coupling, in order to recover well-known results involving vacuum polarization.

Comparison with previous works

We first compare the non-relativistic spin terms with the semi-classical model of Hurst et al. [6]. Next we compare the relativistic precession with the usual rest-frame spin equation. Throughout this subsection, we choose e>0e > 0 to be the elementary charge and consider electrons, whose charge is −e-e. We omit the vacuum contribution and the coupling to the positron sector, although these contributions have been kept in the derivation above.

With this convention, electrons correspond to the minus branch. We introduce their physical velocity through

p=−mγv,γ=11−v2/c2,\mathbf{p}=-m\gamma\mathbf{v},\qquad\gamma=\frac{1}{\sqrt{1-v^2/c^2}},
f(r,v,t)=f−(r,−mγv,t),f(\mathbf{r},\mathbf{v},t)=f_{-}(\mathbf{r},-m\gamma\mathbf{v},t),
F(r,v,t)=F−(r,−mγv,t).(73)\mathbf{F}(\mathbf{r},\mathbf{v},t)=\mathbf{F}_{-}(\mathbf{r},-m\gamma\mathbf{v},t). \tag*{(73)}

Here v=∣v∣v=|\mathbf{v}|, γ\gamma is the corresponding Lorentz factor, and ∇v\nabla_{\mathbf{v}} denotes the gradient with respect to velocity, with ∂vi=∂/∂vi\partial_{v_i}=\partial/\partial v_i. The functions are expressed in velocity variables by substitution, without a Jacobian factor. The corresponding relativistic transport operator is

De=∂t+v⋅∇r−emγ(δij−vivjc2)(E+v×B)j∂vi.(74)\mathcal{D}_{e}=\partial_t+\mathbf{v}\cdot\nabla_{\mathbf{r}}-\frac{e}{m\gamma}\left(\delta_{ij}-\frac{v_i v_j}{c^2}\right)(\mathbf{E}+\mathbf{v}\times\mathbf{B})_j\partial_{v_i}. \tag*{(74)}

where, as before, repeated indices are summed over.

For the first comparison we keep only the leading non-relativistic spin couplings, using γ=1\gamma=1. We also omit the spin-independent orbital gradient corrections (of higher order in ℏ\hbar), to compare with the semi-classical orbital dynamics of Ref. [6]. Taking the non-relativistic limit alone would not remove these corrections. Next, letting AijA_{ij} reduce to δij\delta_{ij}, the single-species relations become

f0≃−f,f3≃f,f_0\simeq-f,\qquad f_3\simeq f,
g0≃12F,g3≃−12F.(75)g_0\simeq\frac{1}{2}\mathbf{F},\qquad g_3\simeq-\frac{1}{2}\mathbf{F}. \tag*{(75)}

In this part of the calculation, the original DHW functions are understood to be evaluated at p=−mv\mathbf{p}=-m\mathbf{v}, so that ∂pj=−m−1∂vj\partial_{p_j}=-m^{-1}\partial_{v_j}. The local spatial operator therefore takes the form

De=∇r−emB×∇v.(76)\mathcal{D}_e=\nabla_{\mathbf{r}}-\frac{e}{m}\mathbf{B}\times\nabla_{\mathbf{v}}. \tag*{(76)}

The leading non-relativistic spin contribution to δg1\delta\mathbf{g}_1 is ℏDe×g3/(2mc)\hbar\mathcal{D}_e\times\mathbf{g}_3/(2mc). Using ∇r⋅B=0\nabla_{\mathbf{r}}\cdot\mathbf{B}=0, we find the identities

De⋅(De×g)=−em(∂rjBi)∂vjgi,\mathcal{D}_e\cdot(\mathcal{D}_e\times\mathbf{g})=-\frac{e}{m}(\partial_{r_j}B_i)\partial_{v_j}g_i,
P⋅(P×g)=eℏ24m(∂rjBi)∂vjgi+O(ℏ4g).(77)\mathbf{P}\cdot(\mathbf{P}\times\mathbf{g})=\frac{e\hbar^2}{4m}(\partial_{r_j}B_i)\partial_{v_j}g_i+O(\hbar^4\mathbf{g}). \tag*{(77)}

In the second identity, P\mathbf{P} denotes the original momentum operator expressed in these velocity variables. We keep its first nonlocal correction, including momentum derivatives acting on the explicit momentum factor. The spin-dependent contributions to Q0Q_0 and Q3Q_3 are then

Q0spin≃eℏ2m2(∂rjBi)∂vjg3i,Q_0^{\mathrm{spin}}\simeq\frac{e\hbar}{2m^2}(\partial_{r_j}B_i)\partial_{v_j}g_{3i},
Q3spin≃eℏ2m2(∂rjBi)∂vjg0i.(78)Q_3^{\mathrm{spin}}\simeq\frac{e\hbar}{2m^2}(\partial_{r_j}B_i)\partial_{v_j}g_{0i}. \tag*{(78)}

Combining these terms in accordance with Eq. (37) gives

Denrf=eℏ4m2(∂rjBi)∂vjF.(79)\mathcal{D}_e^{\mathrm{nr}}f=\frac{e\hbar}{4m^2}(\partial_{r_j}B_i)\partial_{v_j}F. \tag*{(79)}

where

Denr=∂t+v⋅∇r−em(E+v×B)⋅∇v.(80)\mathcal{D}_e^{\mathrm{nr}}=\partial_t+\mathbf{v}\cdot\nabla_r-\frac{e}{m}(\mathbf{E}+\mathbf{v}\times\mathbf{B})\cdot\nabla_v. \tag*{(80)}

For the vector equation we use U≃cDef0\mathbf{U}\simeq c\mathcal{D}_e f_0. The identity

De×(Def0)=−em(∂rjB)∂vjf0(81)\mathcal{D}_e\times(\mathcal{D}_e f_0)=-\frac{e}{m}(\partial_{r_j}\mathbf{B})\partial_{v_j}f_0 \tag*{(81)}

shows that both vector sources at leading order contribute as:

S0≃eℏ2m2(∂rjB)∂vjf3,\mathbf{S}_0\simeq\frac{e\hbar}{2m^2}(\partial_{r_j}\mathbf{B})\partial_{v_j}f_3,
S3≃eℏ2m2(∂rjB)∂vjf0.(82)\mathbf{S}_3\simeq\frac{e\hbar}{2m^2}(\partial_{r_j}\mathbf{B})\partial_{v_j}f_0. \tag*{(82)}

The term proportional to p×Ut\mathbf{p}\times\mathbf{U}_t and the electric terms contained in U\mathbf{U}, in the expressions for S0\mathbf{S}_0 and S3\mathbf{S}_3, are relativistic corrections and have been omitted here. Combining the two sources according to Eq. (60) gives

DenrF−emB×F=eℏm2(∂rjB)∂vjf.(83)\mathcal{D}_e^{\mathrm{nr}}\mathbf{F}-\frac{e}{m}\mathbf{B}\times\mathbf{F}=\frac{e\hbar}{m^2}(\partial_{r_j}\mathbf{B})\partial_{v_j}f. \tag*{(83)}

To compare the numerical coefficients, we denote the scalar and spin functions of Ref. [6] by h0h_0 and h\mathbf{h}. Their normalization relative to our functions is

h0=Cf,h=Cℏ4F,(84)h_0=Cf,\qquad\mathbf{h}=C\frac{\hbar}{4}\mathbf{F}, \tag*{(84)}

where CC accounts for the common phase-space normalization. This relation follows since g0=F/2g_0=\mathbf{F}/2 for a single non-relativistic species, and the spin density includes a factor ℏ/2\hbar/2. The resulting equations are

Denrh0−em(∂rjBi)∂vjhi=0,(85)\mathcal{D}^{\mathrm{nr}}_{e}h_{0}-\frac{e}{m}\left(\partial_{r_j}B_i\right)\partial_{v_j}h_i=0, \tag*{(85)}
Denrhi+em(h×B)i−μBℏ2m(∂rjBi)∂vjh0=0,(86)\mathcal{D}^{\mathrm{nr}}_{e}h_i+\frac{e}{m}\left(\mathbf{h}\times\mathbf{B}\right)_i-\frac{\mu_B\hbar}{2m}\left(\partial_{r_j}B_i\right)\partial_{v_j}h_0=0, \tag*{(86)}

where μB=eℏ/(2m)\mu_B=e\hbar/(2m) is the Bohr magneton. With the approximations stated above, these magnetic-gradient terms and the precession term agree with Eqs. (12) and (13) of Ref. [6].

Next we keep the full relativistic velocity dependence and examine the spin precession. Our vector equation becomes

DeF−emγB×F+emc2γv×(E×F)=S.(87)\mathcal{D}_e\mathbf{F}-\frac{e}{m\gamma}\mathbf{B}\times\mathbf{F}+\frac{e}{mc^2\gamma}\mathbf{v}\times(\mathbf{E}\times\mathbf{F})=\mathbf{S}. \tag*{(87)}

Here S\mathbf{S} denotes the electron-branch source S−\mathbf{S}_{-}, evaluated at p=−mγv\mathbf{p}=-m\gamma\mathbf{v} and expressed in velocity variables by the same substitution as F\mathbf{F}. The electric term is not resulting in a pure rotation of F\mathbf{F}. As we will see, this is due to the choice of spin variable, and nevertheless the result is consistent with Thomas precession.

To see this, we first rescale the spin function according to

a=γF,a0=v⋅ac.(88)\mathbf{a}=\gamma\mathbf{F},\qquad a^0=\frac{\mathbf{v}\cdot\mathbf{a}}{c}. \tag*{(88)}

Along the classical electron trajectories we have

Dev=−emγ[E+v×B−v(E⋅v)c2],Deγ=−emc2E⋅v.(89)\mathcal{D}_e\mathbf{v}=-\frac{e}{m\gamma}\left[\mathbf{E}+\mathbf{v}\times\mathbf{B}-\frac{\mathbf{v}(\mathbf{E}\cdot\mathbf{v})}{c^2}\right], \qquad \mathcal{D}_e\gamma=-\frac{e}{mc^2}\mathbf{E}\cdot\mathbf{v}. \tag*{(89)}

The derivative of γ\gamma cancels the term proportional to −F(v⋅E)-\mathbf{F}(\mathbf{v}\cdot\mathbf{E}) in the double cross product. Thus the rescaled equation becomes

Dea=−emγ(a×B+a0cE)+γS.(90)\mathcal{D}_e\mathbf{a}=-\frac{e}{m\gamma}\left(\mathbf{a}\times\mathbf{B}+\frac{a^0}{c}\mathbf{E}\right)+\gamma\mathbf{S}. \tag*{(90)}

If the source is omitted, the temporal component satisfies

Dea0=−emcγE⋅a.(91)\mathcal{D}_ea^0=-\frac{e}{mc\gamma}\mathbf{E}\cdot\mathbf{a}. \tag*{(91)}

These are the spatial and temporal components of the BMT equation for electrons with the spin gg factor set to g=2g=2 [18], as expected. With the metric signature (+,−,−,−)(+,-,-,-), the four-vector aμ=(a0,a)a^\mu=(a^0,\mathbf{a}) obeys aμUμ=0a^\mu U_\mu=0, where Uμ=γ(c,v)U^\mu=\gamma(c,\mathbf{v}) is the particle four-velocity. Greek indices run over the four space-time components, and repeated Greek indices are contracted using this metric. Consequently, apart from a constant normalization, γF\gamma\mathbf{F} evolves as the spatial part of the spin four-vector when no source term is present.

To express the result in the particle rest frame, we introduce the spin function ζ\boldsymbol{\zeta}, with the same constant normalization. The Lorentz transformation and its inverse give

ai=ζi+γ2vivjζjc2(γ+1),ζi=γFi−γ2vivjFjc2(γ+1).(92)a_i=\zeta_i+\frac{\gamma^2v_iv_j\zeta_j}{c^2(\gamma+1)}, \qquad \zeta_i=\gamma F_i-\frac{\gamma^2v_iv_jF_j}{c^2(\gamma+1)}. \tag*{(92)}

Differentiating these relations along the particle trajectories gives

Deζi−em{[Bγ−v×Ec2(γ+1)]×ζ}i=(γδij−γ2vivjc2(γ+1))Sj.(93)\mathcal{D}_e\zeta_i-\frac{e}{m}\left\{\left[\frac{\mathbf{B}}{\gamma}-\frac{\mathbf{v}\times\mathbf{E}}{c^2(\gamma+1)}\right]\times\boldsymbol{\zeta}\right\}_i =\left(\gamma\delta_{ij}-\frac{\gamma^2v_iv_j}{c^2(\gamma+1)}\right)S_j. \tag*{(93)}

In particular, differentiating the coefficient in the inverse boost gives

De(γδij−γ2vivjc2(γ+1))=(Deγ)δij−γ2c2(γ+1)[(Devi)vj+vi(Devj)]−γ(γ+2)vivjc2(γ+1)2Deγ.(94)\mathcal{D}_e\left(\gamma\delta_{ij}-\frac{\gamma^2v_iv_j}{c^2(\gamma+1)}\right) =(\mathcal{D}_e\gamma)\delta_{ij} -\frac{\gamma^2}{c^2(\gamma+1)}\left[(\mathcal{D}_ev_i)v_j+v_i(\mathcal{D}_ev_j)\right] -\frac{\gamma(\gamma+2)v_iv_j}{c^2(\gamma+1)^2}\mathcal{D}_e\gamma. \tag*{(94)}

Without including the coupling to the scalar source terms in the right-hand side of Eq. (87), we thereby find agreement with the usual rest-frame BMT precession, which preserves ∣ζ∣|\boldsymbol{\zeta}|. Equivalently, the combination

∣ζ∣2=γ2[∣F∣2−(v⋅F)2c2](95)|\boldsymbol{\zeta}|^2=\gamma^2\left[|\mathbf{F}|^2-\frac{(\mathbf{v}\cdot\mathbf{F})^2}{c^2}\right] \tag*{(95)}

is constant along a source-free trajectory.

In the weakly relativistic limit, the electric precession term on the left-hand side becomes

e2mc2(v×E)×ζ.(96)\frac{e}{2mc^2}(\mathbf{v}\times\mathbf{E})\times\boldsymbol{\zeta}. \tag*{(96)}

Thus we recover the usual Thomas factor 1/21/2. Specifically, the electric term agrees with Eq. (4.5) of Hurst et al. [16], where the effective magnetic field is B−v×E/(2c2)\mathbf{B}-\mathbf{v}\times\mathbf{E}/(2c^2). The comparison here establishes agreement of the Thomas term. The rescaling and Lorentz transformation above explain why our original precession equation for F\mathbf{F} has a different appearance.

Expansion to arbitrary order and inclusion of vacuum polarization

In the preceding section, the scalar equations were derived up to second order in the weak-quantum expansion,♀♀♀♀♀♀ whereas the vector equations kept the leading spin and magnetization contributions. The same reduction can be extended systematically by improving the solutions for f1f_1, f2f_2, g1\mathbf{g}_1, and g2\mathbf{g}_2 through successive substitutions. Similarly, the nonlocal operators can be expanded to any order. Throughout, weakly relativistic approximations are avoided, and the full dependence on p/(mc)p/(mc) is kept. The expansion assumes electromagnetic fields that are weak compared to the critical fields, as well as temporal and spatial variations that are slow compared to the Compton scales. It also selects the slowly varying branch of the DHW dynamics, excluding rapid oscillations associated with the eliminated variables. Under these assumptions, one obtains a formal reduction to eight independent phase-space functions that can be performed to all orders. Note, however, that exponentially small Schwinger pair production, will not be recovered by a finite truncation. Perturbative vacuum polarization, however, will be covered by the reduced description.

The expansion to all orders

We introduce the characteristic time constant τ=ℏ/(2mc2)\tau= \hbar/(2mc^2), which is equal to one-half of the inverse Compton frequency. In this section, we use the full operators DtD_t, D\mathbf{D}, and P\mathbf{P}. Their order is kept as written; in particular, we do not commute DtD_t with D\mathbf{D} or P\mathbf{P}. In order to eliminate f1f_1 and f2f_2, it is useful to define the auxiliary scalar

L=Dt(P⋅g3mc)+cD⋅g0.(97)L = D_t\left(\frac{\mathbf{P}\cdot\mathbf{g}_3}{mc}\right)+c\mathbf{D}\cdot\mathbf{g}_0. \tag*{(97)}

Eliminating f1f_1 and f2f_2 from Eqs. (39) and (40) gives

(1+τ2Dt2)f1=P⋅g3mc−τ2cDt(D⋅g0),(98)\left(1+\tau^2D_t^2\right)f_1=\frac{\mathbf{P}\cdot\mathbf{g}_3}{mc}-\tau^2cD_t(\mathbf{D}\cdot\mathbf{g}_0), \tag*{(98)}
(1+τ2Dt2)f2=−τL.(99)\left(1+\tau^2D_t^2\right)f_2=-\tau L. \tag*{(99)}

For slow time dependence we expand the inverse operator as

(1+τ2Dt2)−1=∑n=0∞(−τ2Dt2)n.(100)\left(1+\tau^2D_t^2\right)^{-1}=\sum_{n=0}^{\infty}\left(-\tau^2D_t^2\right)^n. \tag*{(100)}

The successive approximations are then

f1=P⋅g3mc−τ2Dt2L+τ4Dt3L−τ6Dt5L+⋯ ,(101)f_1=\frac{\mathbf{P}\cdot\mathbf{g}_3}{mc}-\tau^2D_t^2L+\tau^4D_t^3L-\tau^6D_t^5L+\cdots, \tag*{(101)}
f2=−τL+τ3Dt2L−τ5Dt4L+⋯ .(102)f_2=-\tau L+\tau^3D_t^2L-\tau^5D_t^4L+\cdots. \tag*{(102)}

The terms displayed for f1f_1 start at orders ℏ\hbar, ℏ3\hbar^3, ℏ5\hbar^5, and ℏ7\hbar^7, while those for f2f_2 start at ℏ2\hbar^2, ℏ4\hbar^4, and ℏ6\hbar^6. Since the full nonlocal operators are still present, each term also contains higher-order spatial-gradient corrections. These could be expanded and collected in case a particular order in ℏ\hbar is what we aim for.

Similarly, in order to calculate expressions for g1\mathbf{g}_1 and g2\mathbf{g}_2, we define the auxiliary vectors V\mathbf{V} and R\mathbf{R} by

V=P×g0mc−τ[Dt(Pf3mc)+cDf0],R=τ(DtV+cD×g3).(103)\mathbf{V}=\frac{\mathbf{P}\times\mathbf{g}_0}{mc}-\tau\left[D_t\left(\frac{\mathbf{P}f_3}{mc}\right)+c\mathbf{D}f_0\right], \qquad \mathbf{R}=\tau(D_t\mathbf{V}+c\mathbf{D}\times\mathbf{g}_3). \tag*{(103)}

The exact vector relations can now be written as

g2=V−τDt(g1−Pf3mc),(104)\mathbf{g}_2=\mathbf{V}-\tau D_t\left(\mathbf{g}_1-\frac{\mathbf{P}f_3}{mc}\right), \tag*{(104)}
g1−Pf3mc=R−τ2Dt2(g1−Pf3mc).(105)\mathbf{g}_1-\frac{\mathbf{P}f_3}{mc}=\mathbf{R}-\tau^2D_t^2\left(\mathbf{g}_1-\frac{\mathbf{P}f_3}{mc}\right). \tag*{(105)}

Expanding the same inverse operator gives

g1=Pf3mc+R−τ2Dt2R+τ4Dt4R+⋯ ,(106)\mathbf{g}_1=\frac{\mathbf{P}f_3}{mc}+\mathbf{R}-\tau^2D_t^2\mathbf{R}+\tau^4D_t^4\mathbf{R}+\cdots, \tag*{(106)}
g2=V−τDtR+τ3Dt3R−τ5Dt5R+⋯ .(107)\mathbf{g}_2=\mathbf{V}-\tau D_t\mathbf{R}+\tau^3D_t^3\mathbf{R}-\tau^5D_t^5\mathbf{R}+\cdots. \tag*{(107)}

Here V\mathbf{V} is of order ℏ\hbar, and R\mathbf{R} is of order ℏ2\hbar^2. Thus the terms in g1\mathbf{g}_1 start at orders 11, ℏ2\hbar^2, ℏ4\hbar^4, and ℏ6\hbar^6, and those in g2\mathbf{g}_2 at ℏ\hbar, ℏ3\hbar^3, ℏ5\hbar^5, and ℏ7\hbar^7. Note that the first nonlocal correction in Pf3/(mc)\mathbf{P}f_3/(mc) is already of order ℏ2\hbar^2, and would be combined with the leading contribution from R\mathbf{R}, if we treated all expansion parameters in ℏ\hbar on the same footing. For a given truncation, we expand the operators and collect the terms according to the ordering in Sec. III A. Inserting the resulting expressions into the remaining DHW equations gives higher-order versions of the scalar and vector equations derived above. As before, factors 1/ℏ1/\hbar in the original equations must be taken into account when deciding how far to carry the substitution. Time derivatives of functions inside correction terms are replaced successively using the lower-order evolution equations. For example, in the second-order scalar corrections we use the classical equations for f0f_0 and f3f_3. Repeating this procedure at higher orders leaves first-order evolution equations for eight independent functions. Time derivatives of prescribed fields may remain; for self-consistent fields these can be related to Maxwell’s equations. The variables f±f_{\pm} and F±\mathbf{F}_{\pm} are still used to approximately separate electron and positron degrees of freedom.

Vacuum contributions and their electromagnetic response

Nothing in the elimination scheme described in the previous subsection requires us to subtract the vacuum contribution to the phase-space variables. Using the free-vacuum values and the definition of f±f_{\pm}, we have

f3,vac=−2γp,g1,vac=−2pmcγp,(108)f_{3,\mathrm{vac}}=-\frac{2}{\gamma_p},\qquad\mathbf{g}_{1,\mathrm{vac}}=-\frac{2\mathbf{p}}{mc\gamma_p}, \tag*{(108)}
f0,vac=0,f±,vac=−1.(109)f_{0,\mathrm{vac}}=0,\qquad f_{\pm,\mathrm{vac}}=-1. \tag*{(109)}

Thus f±f_{\pm} includes a vacuum contribution, and should not in general be interpreted as a classical distribution function. Specifically, a change in these functions due to an applied field does not need to imply that real particles have been created.

Although the free vacuum is homogeneous and stationary, its original DHW functions depend on momentum. For example, the local electric-field term gives

Dt,locf3,vac=eE⋅∇pf3,vac.(110)D_{t,\mathrm{loc}}f_{3,\mathrm{vac}}=e\mathbf{E}\cdot\nabla_p f_{3,\mathrm{vac}}. \tag*{(110)}

The nonlocal corrections can also act on the unperturbed vacuum: their spatial derivatives act on the fields, whereas their momentum derivatives act on the vacuum functions. Similarly, derivatives in D\mathbf{D} can act on the vacuum functions, including nonlocal terms when field gradients are present, although some contractions vanish by symmetry. By contrast, all derivatives acting directly on the constants f±,vac=−1f_{\pm,\mathrm{vac}}=-1 are zero. In the f±f_{\pm} representation, the response instead arises through momentum-dependent coefficients and the reconstructed DHW functions. Once fields are applied, the vacuum corrections generally vary in both space and time.

To identify the expected integrated response, we first consider the weak-field Euler–Heisenberg (EH) effective Lagrangian density (see, e.g., [39, 40] LEH\mathcal{L}_{\mathrm{EH}}. Omitting field derivatives, it is

LEH=ϵ02(E2−c2B2)+κ[(E2−c2B2)2+7c2(E⋅B)2],(111)\begin{aligned} \mathcal{L}_{\mathrm{EH}}&=\frac{\epsilon_0}{2}\left(E^2-c^2B^2\right)\\ &\quad+\kappa\left[\left(E^2-c^2B^2\right)^2+7c^2(\mathbf{E}\cdot\mathbf{B})^2\right], \tag*{(111)} \end{aligned}
κ=αϵ090πEc2,Ec=m2c3eℏ,\kappa=\frac{\alpha\epsilon_0}{90\pi E_c^2},\qquad E_c=\frac{m^2c^3}{e\hbar},

where, in this subsection, we let e>0e>0 denote the elementary charge (rather than the electron charge), ϵ0\epsilon_0 is the vacuum permittivity, α\alpha is the fine-structure constant, and EcE_c is the Schwinger critical electric field. We use E=∣E∣E=|\mathbf{E}| and B=∣B∣B=|\mathbf{B}|. The first term is the Maxwell Lagrangian, and the remaining terms are the leading one-loop Euler–Heisenberg correction. The corresponding polarization and magnetization expressions are

PEH=4κ(E2−c2B2)E+14κc2(E⋅B)B,(112)\begin{aligned} \mathbf{P}_{\mathrm{EH}}&=4\kappa\left(E^2-c^2B^2\right)\mathbf{E}\\ &\quad+14\kappa c^2(\mathbf{E}\cdot\mathbf{B})\mathbf{B}, \tag*{(112)} \end{aligned}
MEH=−4κc2(E2−c2B2)B+14κc2(E⋅B)E,(113)\begin{aligned} \mathbf{M}_{\mathrm{EH}}&=-4\kappa c^2\left(E^2-c^2B^2\right)\mathbf{B}\\ &\quad+14\kappa c^2(\mathbf{E}\cdot\mathbf{B})\mathbf{E}, \tag*{(113)} \end{aligned}

giving the charge and current densities

ρEH=−∇r⋅PEH,\rho_{\mathrm{EH}}=-\nabla_r\cdot\mathbf{P}_{\mathrm{EH}},
jEH=∂tPEH+∇r×MEH.(114)\mathbf{j}_{\mathrm{EH}}=\partial_t\mathbf{P}_{\mathrm{EH}}+\nabla_r\times\mathbf{M}_{\mathrm{EH}}. \tag*{(114)}

The correction to the Lagrangian is fourth order in field amplitude. Consequently, the polarization and magnetization, and the associated sources, are third order in the fields.

There is also a finite contribution that is linear in field amplitude, but contains additional space-time derivatives.

For general electric and magnetic fields, the leading derivative contribution to the renormalized vacuum response can be written as

ρD=αϵ0c215πωc2□(∇r⋅E),\rho_D=\frac{\alpha\epsilon_0c^2}{15\pi\omega_c^2}\Box\left(\nabla_r\cdot\mathbf{E}\right),
jD=−αϵ0c215πωc2□(∂tE−c2∇r×B),(115)\mathbf{j}_D=-\frac{\alpha\epsilon_0c^2}{15\pi\omega_c^2}\Box\left(\partial_t\mathbf{E}-c^2\nabla_r\times\mathbf{B}\right), \tag*{(115)}
□=1c2∂t2−∇r2,ωc=mc2ℏ.\Box=\frac{1}{c^2}\partial_t^2-\nabla_r^2,\qquad\omega_c=\frac{mc^2}{\hbar}.

This is the leading weak-field, low-frequency and long-wavelength term; higher derivative and nonlinear derivative corrections are omitted. Here □\Box is the d’Alembert operator, ωc\omega_c is the Compton angular frequency, and the subscript DD labels the derivative contribution to the charge and current densities. The electromagnetic derivative correction is discussed in Sec. II D of Ref. [39]. The coefficient used here is the standard renormalized one-loop coefficient, consistent with the quadratic weak-field term in Eq. (20) of Ref. [41].

In our case, to obtain the vacuum response from the reduced equations, we must keep the induced corrections to the eight independent functions as well as the reconstructed auxiliary functions. The physical charge and current densities are given by Eqs. (16) and (17). Although the free-vacuum current integrates to zero, the induced current density does not.

We will not carry out the above procedure in full detail. However, this was done for the special case of longitudinal electrostatic fields by Lundström et al. [32], using the same approach as here (i.e. expanding the DHW-equations) exactly recovering the cubic vacuum polarization and the derivative corrections. The key features of that calculation also apply here. First, the integrated linear response contains a logarithmic ultraviolet divergence, which is absorbed into the renormalized charge; see Sec. III C of that work. For general electromagnetic fields the component equations are more involved, but the divergent term still renormalizes the Maxwell action. The regularization and subtraction must preserve gauge invariance and current conservation, with the physical charge fixed by the same low-energy condition.

Secondly, some apparently low-order vacuum contributions vanish after momentum integration. This will follow from momentum parity, total momentum derivatives, or cancellations between the induced distributions and the reconstructed current. The same regulator must be used throughout these cancellations. In particular, surface terms from divergent integrals cannot be dropped before regularization.

Thirdly, real particles modify the quantum response, in addition to the usual vacuum contribution. Equation (68) of Ref. [32] separates these contributions, and Eq. (69) in that work gives the result after taking a further ultrarelativistic limit. The modification is consistent with Pauli blocking in occupied momentum regions. However, it also depends on the particle distribution and its evolution, and cannot in general be represented by a single multiplicative blocking factor.

The electrostatic calculation therefore shows explicitly that perturbative elimination can cover both the effect of vacuum polarization and quantum corrections involving real particles. The substitutions in Eqs. (101), (102), (106), and (107) extend the construction to general fields. Naturally, evaluating the full electromagnetic vacuum response also requires momentum integration and renormalization. In this approach, vacuum polarization follows from the coupling between the electron and positron degrees of freedom, when keeping coupling terms to the required order.

Summary and Conclusion

In this paper, we have studied expansions of the DHW equations in weak fields (compared to the critical field) and on spatial and temporal scales long compared with the corresponding Compton scales. This generalizes a previous work, Ref. [32], to account for general electromagnetic fields and a full 3D geometry. Similar expansion schemes may allow for a FW transformation (see, e.g., Refs. [17, 30, 31]), separating the electron and positron degrees of freedom. Here, however, we have deliberately avoided a full separation of the Dirac field into electrons and positrons. From a philosophical point of view, the separation of electrons and positrons is only approximate in the Dirac theory, and residual coupling need not necessarily be avoided. We refer to the approximate equations keeping the electron-positron coupling as the WCEP equations. A clear demonstration of the usefulness of including this coupling is that it allows vacuum polarization to be covered by the reduced model. Often, when plasmas in strong astrophysical fields are studied, currents due to vacuum polarization are added to the free plasma currents phenomenologically (see, e.g., Refs. [42, 43, 44]). This may be a useful approximation when the plasma density is comparatively low. The relative importance of matter-dependent electron–positron coupling and vacuum polarization depends on the plasma parameters and field configuration. This question has been discussed in some detail for the electrostatic case in Ref. [32]. From the approach presented here, it is apparent that vacuum polarization is due to the mixing of electron and positron states. Since the coupling of virtual states and real-particle states is described by the same equations, a more fundamental and consistent approach should include vacuum polarization and real-particle electron–positron coupling on an equal footing.

Even for plasmas where vacuum polarization can be neglected, describing the system in terms of the variables f±f_{\pm} and F±\mathbf{F}_{\pm} potentially has advantages, but also some drawbacks. The most obvious drawback is that, when high accuracy is required, we need to keep all eight components (one scalar equation and one vector equation) for both electrons and positrons, even if only one species is present in the plasma of interest. In such a case, a related approach based on the FW transformation would require only four variables. To some extent, this difference is due to the present approach capturing additional physics, in particular the vacuum polarization discussed above. However, much of the physics captured using the FW transformation would, in our case, be encoded in the eight equations for coupled electrons and positrons, rather than just four equations for a single species. Keeping eight rather than four independent functions increases the size of the system, but simplifies the relation between momentum and velocity. Specifically, in a fully relativistic theory, complete separation of electrons and positrons comes at a certain cost: a very complicated relation between momentum and kinetic velocity. This is illustrated by Eq. (33) of Ref. [17], and also by the current source in Eq. (27) of that work, which requires the kinetic velocity to calculate the free current density. Moreover, if higher accuracy had been sought in Ref. [17], the complicated expression for the velocity would also have entered into the expression for the Lorentz force.

In our work, we have given fairly limited attention to the source terms in Eqs. (37) and (59) that couple electrons and positrons. The reason is that there are several possible ways to write these terms for various types of plasmas. Except for certain specific comparisons, we have treated all expansion parameters involving ℏ\hbar on the same footing. However, for particular plasma parameters and field configurations, the ordering of the possible expansion parameters could differ considerably. For simplicity, let ω\omega denote a characteristic frequency of the field variations and kk a characteristic wave number. For field variations satisfying the Faraday-law estimate ωB∼kE\omega B \sim kE, the parameter ℏk2/mω\hbar k^{2}/m\omega estimates the strength of the magnetic dipole force relative to the electric part of the Lorentz force for a particle in a given spin state. If this parameter is much larger than ℏk/pch\hbar k/p_{\mathrm{ch}}, where pchp_{\mathrm{ch}} is a characteristic particle momentum, it would make sense to use a scheme that keeps the spin-dependent contributions at the order considered, including the magnetic-gradient coupling, while omitting the orbital dispersive corrections. This is the approximation used to obtain the coupling term in Eq. (38). However, if the dimensionless parameters were of the same order, an expansion to a given order in ℏ\hbar would make sense. Other dimensionless parameters, such as μBB/(kBT)\mu_{\mathrm{B}}B/(k_{\mathrm{B}}T) and ℏωp/(kBT)\hbar\omega_{\mathrm{p}}/(k_{\mathrm{B}}T), could also enter in a more complete picture. Here kBk_{\mathrm{B}} is Boltzmann’s constant, TT is the temperature, and ωp\omega_{\mathrm{p}} is the plasma frequency. See Refs. [2, 45] for more complete discussions of the role of different dimensionless quantum parameters.

The main conclusion of the present work is that the expansion achieves the goal of generating weakly coupled electron and positron distributions, provided the basic assumptions of weak fields (compared to the critical field) and slow scales (compared to Compton scales) are fulfilled. Besides the expansion procedure itself, the key ingredient is the change to the new dependent variables F±\mathbf{F}_{\pm} and f±f_{\pm}. Specifically, using a spin-like function that is a momentumdependent linear combination of the original spin and magnetization is necessary to achieve the approximate separation. If we additionally set the linear combination to have unity in front of g0g_0, the separation condition fixes the momentum dependence in front of g3g_3. Thus, in a regime where the expansion applies, the choice of F±F_{\pm} as a suitable variable is naturally imposed by the DHW equations themselves. Applications of the approach developed in this work include at least two different aspects. Firstly, the present work can serve as a starting point for development of simplified models. The fact that all quantum relativistic effects that can be captured in a perturbative framework are covered is the main advantage. Secondly, the model developed here offers a unified treatment in strong-field environments [2, 32, 43, 46], when both vacuum polarization and real particle quantum effects are of importance.

References

References

  1. [1]D. B. Melrose, Rev. Mod. Plasma Phys. 4, 8 (2020).
  2. [2]G. Brodin and J. Zamanian, Rev. Mod. Plasma Phys. 6, 4 (2022).
  3. [3]G. Manfredi, P.-A. Hervieux, and J. Hurst, Rev. Mod. Plasma Phys. 3, 13 (2019).
  4. [4]G. Manfredi, P.-A. Hervieux, and J. Hurst, Rev. Mod. Plasma Phys. 5, 7 (2021).
  5. [5]M. Bonitz, Quantum Kinetic Theory, 2nd ed. (Springer, Cham, 2016).DOI
  6. [6]J. Hurst, O. Morandi, G. Manfredi, and P.-A. Hervieux, Eur. Phys. J. D 68, 176 (2014).
  7. [7]F. A. Asenjo, V. Muñoz, J. A. Valdivia, and S. M. Mahajan, Phys. Plasmas 18, 012107 (2011).
  8. [8]P. A. Andreev, Phys. Rev. E 91, 033111 (2015).
  9. [9]F. Haas, A. Marklund, G. Brodin, and J. Zamanian, Phys. Lett. A 374, 481 (2010).
  10. [10]A. P. Misra, G. Brodin, M. Marklund, and P. K. Shukla, Phys. Plasmas 17, 122306 (2010).
  11. [11]J. Zamanian, M. Marklund, and G. Brodin, New J. Phys. 12, 043019 (2010).
  12. [12]G. Brodin, M. Marklund, J. Zamanian, Å. Ericsson, and P. L. Mana, Phys. Rev. Lett. 101, 245002 (2008).
  13. [13]B. Bakri, N. Crouseilles, P.-A. Hervieux, and G. Manfredi, Phys. Plasmas 33, 022111 (2026).
  14. [14]D. Del Sorbo, D. Seipt, T. G. Blackburn, A. G. R. Thomas, C. D. Murphy, J. G. Kirk, and C. P. Ridgders, Phys. Rev. A 96, 043407 (2017).
  15. [15]F. A. Asenjo, J. Zamanian, M. Marklund, G. Brodin, and P. Johansson, New J. Phys. 14, 073042 (2012).
  16. [16]J. Hurst, P.-A. Hervieux, and G. Manfredi, Philos. Trans. R. Soc. A 375, 20160199 (2017).
  17. [17]R. Ekman, F. A. Asenjo, and J. Zamanian, Phys. Rev. E 96, 023207 (2017).
  18. [18]V. Bargmann, L. Michel, and V. L. Telegdi, Phys. Rev. Lett. 2, 435 (1959).
  19. [19]D. Vasak, M. Gyulassy, and H.-T. Elze, Ann. Phys. 173, 462 (1987).
  20. [20]I. Białynicki-Birula, P. Górnicki, and J. Rafelski, Phys. Rev. D 44, 1825 (1991).
  21. [21]L.-J. Li, M. Mohamedsedik, and B.-S. Xie, Phys. Rev. D 104, 036015 (2021).
  22. [22]F. Hebenstreit, A. Ilderton, and M. Marklund, Phys. Rev. D 84, 125022 (2011).
  23. [23]I. A. Aleksandrov and C. Kohlfürst, Phys. Rev. D 101, 096009 (2020).
  24. [24]J. P. Edwards, N. Ahmadiniaz, S. M. Schmidt, and C. Kohlfürst, Phys. Rev. D 112, L031901 (2025).
  25. [25]H. Al-Naseri, J. Zamanian, and G. Brodin, Phys. Rev. E 104, 015207 (2021).
  26. [26]G. Brodin, H. Al-Naseri, J. Zamanian, G. Torgrimsson, and B. Eliasson, Phys. Rev. E 107, 035204 (2023).
  27. [27]S. Nematizadeh Juneghani, M. Bagheri, and B. Shokri, J. Cosmol. Astropart. Phys. 2021 (09), 002.
  28. [28]O. Morandi and F. Schürer, J. Phys. A: Math. Theor. 44, 265301 (2011).
  29. [29]I. Akal, R. Egger, C. Müller, and S. Villalba-Chávez, Phys. Rev. D 93, 116006 (2016).
  30. [30]A. J. Silenko, Phys. Rev. A 91, 022103 (2015).
  31. [31]Y. Hinschberger, G. Manfredi, and P.-A. Hervieux, Phys. Rev. A 93, 042117 (2016).
  32. [32]S. Lundström, P. Semrén, H. Al-Naseri, and G. Brodin, Phys. Plasmas 33, 032104 (2026).
  33. [33]J. D. Jackson, Classical Electrodynamics, 3rd ed. (Wiley, New York, 1999).
  34. [34]F. Hebenstreit, R. Alkofer, and H. Gies, Phys. Rev. D 82, 105026 (2010).
  35. [35]A. Gonoskov, T. G. Blackburn, M. Marklund, and S. S. Bulanov, Rev. Mod. Phys. 94, 045001 (2022).
  36. [36]Note that the DHW-equations will hold as is for both possible sign conventions, that is, independently of whether we pick e as the negative electron charge or as the positive elementary charge.
  37. [37]Note, however, that a strong magnetic field may lead to significant spin-polarization of the vacuum, in which case additional DHW-functions will have nonzero vacuum values.
  38. [38]A. Fedotov, A. Ilderton, F. Karbstein, B. King, D. Seipt, H. Taya, and G. Torgrimsson, Phys. Rep. 1010, 1 (2023).
  39. [39]M. Marklund and P. K. Shukla, Rev. Mod. Phys. 78, 591 (2006).
  40. [40]D. Eriksson, G. Brodin, M. Marklund, and L. Stenflo, Phys. Rev. A 70, 013808 (2004).
  41. [41]F. Karbstein, J. High Energy Phys. 2021 (9), 070.
  42. [42]Y. S. Huang, Sci. Rep. 5, 15866 (2015).
  43. [43]M. V. Medvedev, Phys. Plasmas 30, 092112 (2023).
  44. [44]W. Chen, Z. Bu, H. Li, Y. Luo, and P. Ji, Phys. Plasmas 20, 072113 (2013).
  45. [45]J. Lundin and G. Brodin, Phys. Rev. E 82, 056407 (2010).
  46. [46]P. Zhang, S. S. Bulanov, D. Seipt, A. V. Arefiev, and A. G. R. Thomas, Phys. Plasmas 27, 050601 (2020).

Paper details

vlnc:W_yttf5ba8
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents