Electromagnetism in Media — The Local Complex Structure at Work
Introduction
Maxwell's equations in a material medium are usually introduced as a small modification of the vacuum equations: replace $\epsilon_0$ and $\mu_0$ by the permittivity $\epsilon$ and the permeability $\mu$ of the medium, and read off the consequences. In the biquaternionic formulation the emphasis is reversed. The medium is the general case and the vacuum is its limit, because the complex structure of the algebra is local. The material subspace $\mathbb{B}_{\Phi}$ is coordinatised by $(i r'_t,\,r_x,\,r_y,\,r_z)$, and the factor $i$ that marks the temporal direction is paired with the speed of light in the medium, $$ c=\frac{1}{\sqrt{\epsilon\mu}}, $$ a property of the medium at each point rather than a constant of nature. The scalar imaginary $i$ is fixed by the algebra; the real scale attached to the imaginary time axis is supplied by the medium. This article develops electromagnetism in a medium with that interpretation in view.
The declared parent of this article is The Field-Strength Biquaternion and Its Invariants, which has fixed the canonical objects: the field-strength biquaternion $$ \tilde{F}=i\sqrt{\epsilon}\,\vec E-\sqrt{\mu}\,\vec H, $$ the electric and magnetic fields $\vec E$ and $\vec H$, the magnetic induction $\vec B=\mu\vec H$, the medium speed $c=1/\sqrt{\epsilon\mu}$, the Riemann–Silberstein vector $\vec N=\vec E+ic\vec B$, the invariants $I_1=\vec E^2-c^2\vec B^2$ and $I_2=\vec E\cdot\vec B$, and the energy density $W$ and Poynting vector $\vec S$. Nothing in this list is redefined here. The article fixes the conventions specific to a medium — the two medium parameters, the impedance, the dispersive data, and the boundary data — in enough detail that the later exercise on plane-wave propagation in a medium is a direct application of the objects fixed below.
The algebra $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$, its units $e_t,\vec e_x,\vec e_y,\vec e_z$, its two sectors and its three involutions ${}^{\natural}$, $\bar{\cdot}$ and ${}^{*}$ are those of Algebra and Element Conventions, and their physical interpretation is fixed in Conventions in the Biquaternion Universe. Two rules of that file are cited once and used on every page below: the coefficientwise conjugation is $\bar{\cdot}$ and never a starred symbol, and two factors of $i$ are reduced before a display is read, the scalar imaginary being central.
An action is a product, and the side is position. The operator is the element it multiplies by, so the gradient acting from the left is written $\tilde\nabla_\Phi\tilde{F}$, a coefficient acting from the right is written $g\,\vec{\alpha}$, and a two-sided action is $a\,\tilde{Ξ}\,b$. No array is written in the body of this article. The gamma arrays of the older Dirac literature are collected in §The Legacy Pauli Gamma Formulation and used nowhere else.
The biquaternionic gradient $\tilde\nabla_\Phi$ and its quaternion conjugate $(\tilde\nabla_\Phi)^{\natural}$ are those of Operator and Calculus Conventions, the d'Alembertian is $\Box_\Phi=\tilde\nabla_\Phi(\tilde\nabla_\Phi)^{\natural}$, and on the material sector the time derivative is $\partial/\partial R_t=-i\,\partial/\partial r'_t$, with $r'_t$ the temporal length in metres. The gradient carries no medium factor: the medium enters through the conversion $r'_t=c\,t'$ between that length and the clock, and through the normalisations of the field. Throughout, $c$ denotes the speed of light in the medium and $c_0$ the vacuum speed; the divergence and curl are written $\mathrm{div}$ and $\mathrm{rot}$.
Constitutive Relations and the Two Parameters of a Medium
In a linear, isotropic medium the fields are related by the constitutive relations $$ \vec D=\epsilon\,\vec E, \qquad \vec B=\mu\,\vec H, $$ where $\vec E$ is the electric field, $\vec H$ the magnetic field, $\vec D$ the electric displacement and $\vec B$ the magnetic induction. These relations are inherited unchanged from the Maxwell article. The free charge and current densities $\rho$ and $\vec J$ obey the conservation law $$ \mathrm{div}\,\vec J+c\,\frac{\partial\rho}{\partial r'_t}=0, $$ which is the integrability condition of the Maxwell system and reappears below as a condition on the biquaternionic source.
For a homogeneous medium the two constants $\epsilon$ and $\mu$ combine into two quantities of independent physical meaning. The first is the wave speed $$ c=\frac{1}{\sqrt{\epsilon\mu}}, $$ and the second is the wave impedance $$ Z=\sqrt{\frac{\mu}{\epsilon}}. $$ The pair $(c,Z)$ carries the same information as $(\epsilon,\mu)$: inverting, $$ \epsilon=\frac{1}{cZ}, \qquad \mu=\frac{Z}{c}. $$ The speed $c$ controls propagation, while the impedance $Z$ controls the ratio of the electric to the magnetic amplitude of a wave. Together they carry everything the medium contributes to the linear problem, and they are the natural variables in terms of which the biquaternionic field strength is normalised.
The vacuum is the special case $\epsilon=\epsilon_0$, $\mu=\mu_0$, for which $$ c_0=\frac{1}{\sqrt{\epsilon_0\mu_0}}, \qquad Z_0=\sqrt{\frac{\mu_0}{\epsilon_0}}\approx 376.73\ \Omega . $$ The vacuum speed $c_0$ is a global constant. The medium speed $c$ is not: in an inhomogeneous medium $\epsilon=\epsilon(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ and $\mu=\mu(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$, so that $c=c(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ and $Z=Z(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ are fields, and in a dispersive medium they depend on frequency as well. The rest of this article treats the homogeneous non-dispersive case first, then relaxes these assumptions in turn.
The Biquaternionic Gradient and the Local Complex Structure
The gradient is the one fixed in Operator and Calculus Conventions. On the material sector its time derivative takes the value $\partial/\partial R_t=\partial/\partial(i r'_t)=-i\,\partial/\partial r'_t=-(i/c)\,\partial/\partial t'$, and the d'Alembertian is that file's identity $\Box_\Phi=\tilde\nabla_\Phi(\tilde\nabla_\Phi)^{\natural}=(\tilde\nabla_\Phi)^{\natural}\tilde\nabla_\Phi=\Delta_\Phi - \partial^2/\partial {r^{\prime}_t}^2$ evaluated at that derivative, $$ \Box_\Phi=\Delta_\Phi-\partial^2/\partial {r'_{t}}^2, $$ the d'Alembertian in the temporal length $r'_t$. Written in the clock time it is $\Delta_\Phi-c^{-2}\,\partial^2/\partial {t'}^2$, the d'Alembertian of a medium with speed $c$, and there it reduces to the vacuum value with $c$ in place of $c_0$.
The medium enters this operator in exactly one way: through the conversion $r'_t=c\,t'$ with which the temporal slot is measured. The algebra — the quaternion units, the scalar imaginary $i$, the decomposition $\mathbb{B}=\mathbb{B}_{\Psi}\oplus\mathbb{B}_{\Phi}$ — is fixed and medium-independent, and in the length coordinate the operator carries no medium factor at all. What the medium changes is the embedding of physical time into the algebra. The material time coordinate is $i r'_t=i\,c\,t'$, so the map from the physical second to the imaginary scalar direction of $\mathbb{B}$ carries the local factor $c(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$.
Because $c>0$ is real, the medium rescales that embedding but does not rotate it. The phase of the complex structure is the fixed $i$; its unit is the local $c$. This is the precise sense in which the complex structure is local. In the language of the introduction to the framework, $c$ plays the role of the local conversion of the temporal coordinate, and the constant vacuum value $c_0$ is the special case in which that conversion does not vary. The classical $i r'_t$ convention of Minkowski space is the vacuum limit of the local structure.
Two comments are worth making. First, the factorization $\Box_\Phi=\tilde\nabla_\Phi(\tilde\nabla_\Phi)^{\natural}$ holds for any real positive $c$, so the algebraic fact on which the biquaternionic Maxwell equation rests is insensitive to the medium. Second, passing from vacuum to medium does not change the algebra $\mathbb{B}$; a medium is not a deformation of the complex structure but a different local conversion for its imaginary direction.
The Field Strength and Its Two Halves
The parent article defines the field-strength biquaternion as the pure vector $$ \tilde{F}=i\sqrt{\epsilon}\,\vec E-\sqrt{\mu}\,\vec H =\sum_{\eta\in\{x,y,z\}}F_\eta\,\vec e_\eta, \qquad F_\eta=i\sqrt{\epsilon}\,E_\eta-\sqrt{\mu}\,H_\eta, $$ with $\mathrm{Sc}(\tilde{F})=0$. We use that definition without change. Two features of it are specific to the medium.
The normalisation. The factors $\sqrt{\epsilon}$ and $\sqrt{\mu}$ are natural because $\epsilon\,\vec E^2$ and $\mu\,\vec H^2$ are both energy densities. Hence $\sqrt{\epsilon}\,\vec E$ and $\sqrt{\mu}\,\vec H$ have the common dimension of the square root of an energy density, the two halves of $\tilde{F}$ are dimensionally homogeneous, and the biquaternion norm $\langle\tilde{F}\rangle_{\natural}$ has the dimension of an energy density. The normalisation is equivalent to the two medium parameters introduced above: since $\sqrt{\epsilon\mu}=1/c$ and $\sqrt{\mu/\epsilon}=Z$, the data $(\sqrt{\epsilon},\sqrt{\mu})$ and $(c,Z)$ determine each other. The medium therefore enters $\tilde{F}$ through the same two numbers that govern propagation.
The field strength is an overall constant multiple of the Riemann–Silberstein vector. Using $\vec B=\mu\vec H$ and $c=1/\sqrt{\epsilon\mu}$, $$ i\sqrt{\epsilon}\,\vec N=i\sqrt{\epsilon}\left(\vec E+ic\vec B\right) =i\sqrt{\epsilon}\,\vec E-\sqrt{\epsilon}\,c\,\vec B =i\sqrt{\epsilon}\,\vec E-\sqrt{\mu}\,\vec H =\tilde{F}, $$ because $\sqrt{\epsilon}\,c\,\mu=\sqrt{\mu}$, the two factors of $i$ having met and given $-1$. Thus $$ \tilde{F}=i\sqrt{\epsilon}\,\vec N, \qquad \vec N=\vec E+ic\vec B, $$ exactly as in the parent article.
The two halves. Because the electric and magnetic fields are real, the two terms of $\tilde{F}$ land in the two complementary subspaces: $$ \tilde{F} =\underbrace{i\sqrt{\epsilon}\,\vec E}_{\in\,\mathbb{B}_{\Psi}} \;+\; \underbrace{\left(-\sqrt{\mu}\,\vec H\right)}_{\in\,\mathbb{B}_{\Phi}}. $$ The electric half is a pure imaginary vector and is therefore Hermitian; the magnetic half is a pure real vector and is therefore anti-Hermitian. Equivalently, $$ \tfrac{1}{2}\left(\tilde{F}+\tilde{F}^{*}\right)=i\sqrt{\epsilon}\,\vec E, \qquad \tfrac{1}{2}\left(\tilde{F}-\tilde{F}^{*}\right)=-\sqrt{\mu}\,\vec H, $$ so the projection of $\tilde{F}$ onto the informational sector $\mathbb{B}_{\Psi}$ measures the electric field, and its projection onto the material sector $\mathbb{B}_{\Phi}$ measures the magnetic field.
The split into the two sectors is structural and does not depend on the medium: $i\sqrt{\epsilon}\,\vec E$ lies in $\mathbb{B}_{\Psi}$ and $-\sqrt{\mu}\,\vec H$ lies in $\mathbb{B}_{\Phi}$ for every real $\epsilon,\mu>0$. What the medium fixes is the relative weight of the two halves, through the ratio $\sqrt{\mu}/\sqrt{\epsilon}=Z$. This is one concrete sense in which the local complex structure is at work: the field strength is not a vector of a single sector but the sum of a Hermitian and an anti-Hermitian piece, and the medium sets their balance.
The energy is carried by the Hermitian form. In the parent article, $$ \tilde{F}\tilde{F}^{*}=2W\,e_t+\frac{2i}{c}\,\vec S, \qquad W=\frac{1}{2}\left(\epsilon\,\vec E^2+\mu\,\vec H^2\right)=\frac{1}{2}\sum_\eta\left|F_\eta\right|^2, \qquad \vec S=\vec E\times\vec H, $$ so twice the energy density is the scalar part of the Hermitian form, and $\frac{2}{c}$ times the imaginary unit times the Poynting vector is its vector part. Both are medium-dependent through $\epsilon$ and $\mu$. The quantity on the right is the Euclidean norm $\|\tilde{F}\|^2=\sum_\eta|F_\eta|^2$, which is positive definite. It is not the biquaternion norm $\langle\tilde{F}\rangle_{\natural}=\sum_\eta F_\eta^2$: that form is indefinite, it vanishes for every null field, and every plane wave below is null, so the energy is carried by the Euclidean norm and not by $\langle\tilde{F}\rangle_{\natural}$.
Finally, in an inhomogeneous medium $\epsilon=\epsilon(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ and $\mu=\mu(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$, and the normalisation is applied pointwise: $$ \tilde{F}(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z,r'_t)=i\sqrt{\epsilon(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)}\,\vec E(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z,r'_t)-\sqrt{\mu(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)}\,\vec H(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z,r'_t). $$ The field strength is then assembled from the local complex structure, and it inherits the spatial variation of the medium.
Maxwell's Equations in the Medium
With the definitions above, the four Maxwell equations in the medium collapse into the single biquaternionic equation of the parent article, $$ \tilde\nabla_\Phi\tilde{F}=-\sqrt{\mu}\,\tilde{J}_{\Phi}, $$ with the source biquaternion $$ \tilde J_{\Phi}=ic\rho\,e_t+\vec J, $$ and its medium-normalised form $$ \sqrt{\mu}\,\tilde{J}_{\Phi}=R_0+\vec R, \qquad R_0=\frac{i\rho}{\sqrt{\epsilon}}, \qquad \vec R=\sqrt{\mu}\,\vec J. $$ The purpose of this section is to display how the medium factors cancel, so that the single equation has the same form in every medium, the medium entering only through $\tilde F$ and the single factor $\sqrt{\mu}$.
For a pure-vector field $\tilde{F}=\vec F$, the biquaternionic gradient acts as $$ \tilde\nabla_\Phi\tilde{F} =-\mathrm{div}\,\vec F+\partial/\partial R_t\,\vec F+\mathrm{rot}\,\vec F, $$ the scalar part being $-\mathrm{div}\,\vec F$ and the vector part $\partial/\partial R_t\,\vec F+\mathrm{rot}\,\vec F$. Substituting $\vec F=i\sqrt{\epsilon}\,\vec E-\sqrt{\mu}\,\vec H$ and $\partial/\partial R_t=-i\,\partial/\partial r'_t$, and separating the real and imaginary parts, the scalar equation gives $$ \mathrm{div}\,\vec E=\frac{\rho}{\epsilon}, \qquad \mathrm{div}\,\vec H=0, $$ and the vector equation gives $$ \mathrm{rot}\,\vec H=c\,\frac{\partial\vec D}{\partial r'_t}+\vec J, \qquad \mathrm{rot}\,\vec E=-c\,\frac{\partial\vec B}{\partial r'_t}. $$ The first pair is Gauss's law for $\vec D=\epsilon\vec E$ together with $\mathrm{div}\,\vec B=0$ (since $\vec B=\mu\vec H$ and $\mu$ is constant here), and the second pair is the Ampère–Maxwell law and Faraday's law. All four standard equations are recovered, with no residual $\epsilon$ or $\mu$.
The cancellation is the content of the normalisation. In the vector equation, for example, $\partial/\partial R_t$ acting on the electric half gives $$ \frac{\partial}{\partial R_t}\left(i\sqrt{\epsilon}\,\vec E\right)=-i\frac{\partial}{\partial r'_t}\left(i\sqrt{\epsilon}\,\vec E\right)=\sqrt{\epsilon}\,\frac{\partial\vec E}{\partial r'_t}, $$ because $\partial/\partial R_t=-i\,\partial/\partial r'_t$ and $i^2=-1$; on the magnetic half it gives $\partial/\partial R_t(-\sqrt{\mu}\vec H)=+i\sqrt{\mu}\,\partial\vec H/\partial r'_t$, which is imaginary. The real part of the vector equation, $\sqrt{\epsilon}\,\partial \vec E/\partial r'_t-\sqrt{\mu}\,\mathrm{rot}\,\vec H=-\sqrt{\mu}\vec J$, is Ampère–Maxwell's law with the coefficient $$ \sqrt{\frac{\epsilon}{\mu}}=\epsilon c, $$ using $\sqrt{\epsilon\mu}=1/c$, so that $\epsilon c\,\partial\vec E/\partial r'_t=c\,\partial\vec D/\partial r'_t$; the imaginary part, $\sqrt{\mu}\,\partial \vec H/\partial r'_t+\sqrt{\epsilon}\,\mathrm{rot}\,\vec E=0$, is Faraday's law with the coefficient $\sqrt{\mu/\epsilon}=\mu c$. One power of $i$ comes from the field's electric half and one from $\partial/\partial R_t$, and the product is the $-1$ that turns the two curl terms into the two time derivatives. The factors $\sqrt{\epsilon}$ and $\sqrt{\mu}$ on the fields are exactly what is required for the medium to cancel out of the components. The biquaternionic Maxwell equation therefore has a form independent of the medium; the medium is carried by the definition of $\tilde{F}$, by the single factor $\sqrt{\mu}$, and by the operator $\partial/\partial R_t$.
Two companion statements from the Maxwell article carry over unchanged. The first is the integrability condition $\sqrt{\mu}\,\mathrm{Sc}((\tilde\nabla_\Phi)^{\natural}\tilde{J}_{\Phi})=0$, which is the biquaternionic form of charge conservation. The second is the potential formulation: in the Lorenz gauge the potential biquaternion $\tilde{A}_{\Phi}=i\,a'_t+\vec A$ satisfies $$ \Box_\Phi\tilde{A}_{\Phi}=-\mu\,\tilde J_{\Phi}, \qquad \tilde J_{\Phi}=ic\rho\,e_t+\vec J, $$ where the factor $\mu$ is now explicit. The field-strength formulation hides the medium in the normalisation; the potential formulation displays one of the medium parameters directly.
Dispersion and the Frequency-Dependent Local Structure
So far $\epsilon$ and $\mu$ have been taken constant. A real medium responds with a delay, and its permittivity and permeability depend on frequency: $\epsilon=\epsilon(\omega)$, $\mu=\mu(\omega)$. In a monochromatic field the constitutive relations are evaluated at the frequency of the field, so the medium speed becomes $$ c(\omega)=\frac{1}{\sqrt{\epsilon(\omega)\mu(\omega)}}. $$ The refractive index relative to vacuum is $$ n(\omega)=\frac{c_0}{c(\omega)} =c_0\sqrt{\epsilon(\omega)\mu(\omega)} =\sqrt{\frac{\epsilon(\omega)\mu(\omega)}{\epsilon_0\mu_0}}, $$ and the dispersion relation of a plane wave in the medium is $$ k_\Phi^2=\omega^2\,\epsilon(\omega)\mu(\omega), \qquad k_\Phi=\frac{\omega}{c(\omega)}=\frac{n(\omega)\,\omega}{c_0}. $$ The phase velocity is $v_p=\omega/k_\Phi=c(\omega)$, and the group velocity is $$ v_g=\frac{d\omega}{dk_\Phi}=\left(\frac{dk_\Phi}{d\omega}\right)^{-1} =\frac{c_0}{n(\omega)+\omega\,\dfrac{dn}{d\omega}}. $$ All of this is standard, and it is what fixes the meaning of a "plane wave in a medium" for the later exercise.
The biquaternionic interpretation is that each Fourier component carries its own local complex structure. The algebra $\mathbb{B}$ is the same at every frequency; what changes is the scale that maps physical time to the imaginary coordinate. For a monochromatic wave the scalar component of the gradient acts as $$ \partial/\partial R_t=-i\frac{\partial}{\partial r'_t}\;\longrightarrow\;-\omega, $$ because $\partial/\partial r'_t\to-i\omega/c$ in the $\exp(-i\theta_{\mathrm{pwt}})$ convention of this section and $i^2=-1$, so the imaginary time axis is scaled differently at each frequency. The local complex structure is thus spectral as well as spatial: it is indexed by the frequency at which the medium is probed.
Throughout, $\theta_{\mathrm{pwt}}:=\omega r'_t/c$ is the plane-wave time angle: $\omega$ is real for a monochromatic wave, so the factor $\exp(-i\theta_{\mathrm{pwt}})$ is a phase of modulus one and it is the time part of the plane-wave angle of the full wave form.
Two caveats delimit what the framework as written supports.
First, the normalisation $\sqrt{\epsilon(\omega)}$, $\sqrt{\mu(\omega)}$ is real only in a transparent frequency window, away from absorption lines. As long as it is real, the split of $\tilde{F}$ into a Hermitian electric half and an anti-Hermitian magnetic half survives at each frequency, and the invariants are defined at each frequency with $c(\omega)$. Near a resonance, however, $\epsilon$ and $\mu$ become complex, and with them $\sqrt{\epsilon}$ and $\sqrt{\mu}$. The electric and magnetic contributions are then no longer purely imaginary and purely real, the two halves of $\tilde{F}$ no longer lie in $\mathbb{B}_{\Psi}$ and $\mathbb{B}_{\Phi}$ separately, and $\tilde{F}$ becomes fully complex. The complexified extension of the parent article, in which the coefficients of $\tilde{F}$ are unrestricted complex numbers, is the natural arena for that case; but the identification of the physical field with a particular real slice is a further choice which the framework as written does not fix. The absorbing case is therefore not pursued here.
Second, the energy density of the parent, $W=\tfrac{1}{2}(\epsilon\,\vec E^2+\mu\,\vec H^2)$, is the non-dispersive expression. For a monochromatic field in a dispersive medium the standard energy density is the Brillouin expression $$ W=\frac{1}{2}\left(\frac{d(\omega\epsilon)}{d\omega}\,\vec E^2+\frac{d(\omega\mu)}{d\omega}\,\vec H^2\right), $$ which reduces to the parent's form when the dispersion is negligible. This is a boundary of what the framework as stated supports; it leaves the field strength, the invariants, and the propagation conventions untouched.
A Chiral Medium: One Quaternionic Equation
The constitutive relations used so far, $\vec D=\epsilon\,\vec E$ and $\vec B=\mu\,\vec H$, are the simplest a linear medium can have: each field couples to its own partner and to nothing else. A chiral medium couples them. In the Drude–Born–Fedorov form used by Grudsky, Khmelnytskaya and Kravchenko, $$ \vec B=\mu\left(\vec H+\beta\,\mathrm{rot}\,\vec H\right), \qquad \vec D=\epsilon\left(\vec E+\beta\,\mathrm{rot}\,\vec E\right), $$ where $\beta$ is the chirality measure of the medium, a real scalar of the dimension of a length, and $\epsilon,\mu,\beta$ are constants. At $\beta=0$ these are the relations of the preceding sections. The magnetoelectric coupling is what makes the reduction below a different problem rather than a relabelling of the same one.
A name collision. Everywhere else in this framework the same word labels the two halves of a Dirac field: the element carries the subscript $L$ or $R$, that label takes those two values and no third, and it is read off the element rather than computed; for that sense see Chiral Fermions in the Biquaternion Framework. "Chiral medium" is the older use of the same word for an optically active material. The two senses are unrelated and are distinguished by context alone; the medium sense occurs only in this section and its two companions below. The physical motivation for the medium sense — organic molecules such as DNA at some frequencies, the pupil of the eye, and manufactured chiral materials — is recorded by the authors of the time-harmonic treatment cited below.
The time variable of this section. The source of this section differentiates with respect to the real time, and the corpus's own variable is imaginary: $\partial/\partial r'_t=ic\,\partial/\partial R_t$, so that $\partial/\partial r'_t$ and $ic\,\partial/\partial R_t$ are the same operator and only one convention should be used in a display. The formulas of this section are written with $\partial/\partial r'_t$, which is the source's variable, so that each of them can be checked as it stands against Maxwell's equations and against the source; the corpus's $\partial/\partial R_t$ reappears in the non-chiral reduction of the earlier sections, where it carries no source to be transcribed. The time-harmonic convention is $\exp(i\theta_{\mathrm{pwt}})$, so that $\partial/\partial r'_t\to i\omega/c$.
The reduction, and why it is not the same reduction. Substituting the two relations into Faraday's and Ampère–Maxwell's laws and keeping the two divergence equations gives the chiral system $$ \mathrm{rot}\,\vec H=\epsilon\left(\partial/\partial r'_t\,\vec E+\beta\,\partial/\partial r'_t\,\mathrm{rot}\,\vec E\right)+\vec J, \qquad \mathrm{rot}\,\vec E=-\mu\left(\partial/\partial r'_t\,\vec H+\beta\,\partial/\partial r'_t\,\mathrm{rot}\,\vec H\right), \qquad \mathrm{div}\,\vec E=\frac{\rho}{\epsilon}, \qquad \mathrm{div}\,\vec H=0 , $$ the last of which is unchanged because $\mathrm{div}\,\mathrm{rot}=0$ and $\mu$ is constant. The magnetoelectric terms do not disturb charge conservation either: eliminating $\mathrm{div}\,\mathrm{rot}\,\vec H=0$ against the two divergence equations, or equivalently taking the divergence of Ampère–Maxwell's law, gives $i\omega\rho+\mathrm{div}\,\vec J=0$ in the frequency domain, the law $\partial/\partial r'_t\,\rho+\mathrm{div}\,\vec J=0$ in the time domain, exactly as for an ordinary medium. Applying $\mathrm{rot}$ separates the fields and gives the chiral wave equations $$ \mathrm{rot}\,\mathrm{rot}\,\vec E+\epsilon\mu\,\partial^2/\partial {r'_t}^2\,\vec E +2\beta\epsilon\mu\,\partial^2/\partial {r'_t}^2\,\mathrm{rot}\,\vec E +\beta^2\epsilon\mu\,\partial^2/\partial {r'_t}^2\,\mathrm{rot}\,\mathrm{rot}\,\vec E =-\mu\,\partial/\partial r'_t\,\vec J-\beta\mu\,\partial/\partial r'_t\,\mathrm{rot}\,\vec J, $$ and the corresponding equation for $\vec H$ with $\mathrm{rot}\,\vec J$ as its source. At $\beta=0$ these are the ordinary second-order wave equations of the medium. For $\beta\neq0$ the last term makes them fourth order: the chiral generalisation is not the non-chiral wave equation with a small term added, it is an equation of a different order.
The chiral system nevertheless collapses into one equation, carried by a new operator. With $D=\vec e_x\,\partial/\partial R_x+\vec e_y\,\partial/\partial R_y+\vec e_z\,\partial/\partial R_z$ the Moisil–Teodoresco operator of quaternionic analysis, the operator is $$ \mathcal{M}=\beta\sqrt{\epsilon\mu}\;\partial/\partial r'_t D+\sqrt{\epsilon\mu}\;\partial/\partial r'_t-iD , $$ and it acts on the purely vectorial biquaternionic field $$ \vec{\mathcal N}=\vec E-i\sqrt{\frac{\mu}{\epsilon}}\,\vec H, \qquad \boxed{\;\mathcal{M}\vec{\mathcal N}=-\sqrt{\frac{\mu}{\epsilon}}\,\vec J -\beta\sqrt{\frac{\mu}{\epsilon}}\,\partial/\partial r'_t\,\rho+\frac{i\rho}{\epsilon}\;}. $$ The equivalence is exact and two-way, not a weak-coupling approximation in $\beta$: the physical fields solve the chiral system if and only if $\vec{\mathcal N}$ solves this one (the source's Proposition 1). The field $\vec{\mathcal N}$ is not the Riemann–Silberstein vector $\vec N=\vec E+ic\vec B$ of the earlier sections: since $c\mu=\sqrt{\mu/\epsilon}$, it is exactly the coefficientwise complex conjugate of $\vec N$, and it is written with its own symbol so that the two are not conflated. For $\beta=0$ the operator reduces to $\sqrt{\epsilon\mu}\,\partial/\partial r'_t-iD$, which is the reduction of the non-chiral case, and the authors are explicit that the chiral operator is essentially different from it.
The non-chiral limit in the framework's own symbols. At $\beta=0$ the operator is $\mathcal{M}_0=\sqrt{\epsilon\mu}\,\partial/\partial r'_t-iD$, and composed with its complex conjugate it is the d'Alembertian. Because $D^2=-\Delta_\Phi$ (verified to $1.8\times10^{-15}$ on random wave vectors) and $\partial/\partial r'_t$ commutes with $D$, $$ \left(\sqrt{\epsilon\mu}\,\partial/\partial r'_t-iD\right)\left(\sqrt{\epsilon\mu}\,\partial/\partial r'_t+iD\right) =\epsilon\mu\,\partial^2/\partial {r'_t}^2-\Delta_\Phi=-\Box_\Phi , $$ which is the parent article's factorization read with the opposite overall sign for $\Box_\Phi$. The source cites the same identity as the factorization of the non-chiral wave operator on which the earlier quaternionic reformulation rests. Read together, the two statements say that the parent's reduction is the $\beta=0$ member of a one-parameter family, and the member whose single equation is first order.
A degenerate frequency in the frequency domain. A Fourier transform in $r'_t$ in the $\exp(i\theta_{\mathrm{pwt}})$ convention replaces $\partial/\partial r'_t$ by $i\omega$ and turns $\mathcal{M}$ into $\beta\sqrt{\epsilon\mu}\,i\omega D+\sqrt{\epsilon\mu}\,i\omega-iD$, which factors exactly: $$ \mathcal{M}(\omega)=i\left(\beta\sqrt{\epsilon\mu}\,\omega-1\right)\bigl(D+\alpha(\omega)\bigr), \qquad \alpha(\omega)=\frac{\sqrt{\epsilon\mu}\,\omega}{\beta\sqrt{\epsilon\mu}\,\omega-1} $$ (verified to $4.4\times10^{-16}$ over random wave vectors and frequencies). So at each frequency the chiral operator is again a Moisil–Teodoresco operator $D+\alpha$, with a parameter $\alpha$ that is not small and depends on the frequency: this is why the fundamental solution of the chiral operator is obtainable from the known fundamental solution of $D+\alpha$, and also why it is not obtainable from the non-chiral one by substitution. The prefactor vanishes at $$ \omega=\frac{1}{\beta\sqrt{\epsilon\mu}}=\frac{c}{\beta}, $$ the frequency whose wavelength is $2\pi\beta$. At that frequency the parameter $\alpha$ diverges in the same stroke, with the product $(\beta\sqrt{\epsilon\mu}\,\omega-1)\alpha=\sqrt{\epsilon\mu}\,\omega$ staying finite, so the factored form must be read in the unfactored one. In the unfactored operator the term $\beta\sqrt{\epsilon\mu}\,i\omega D$ becomes $iD$ and cancels the $-iD$, and
$$ \mathcal{M}(c/\beta)=\frac{i}{\beta}, $$
a constant: at that frequency the operator loses its derivative altogether and the reduction degenerates from a differential equation to an algebraic one, $\frac{i}{\beta}\vec{\mathcal N}=\text{source}$, with no propagation left. The chiral reduction therefore carries a distinguished frequency at which its operator degenerates, and the non-chiral reduction carries none.
The birefringence is in the same equation. A chiral medium is optically active — the two circular polarisations propagate with different wave numbers — and that too is a consequence of the fourth-order equation, not a separate input. For a transverse plane wave with the two helicities labelled by $\tau=\pm1$, the two branches of the wave equation above are $$ \left(1-\beta^2\epsilon\mu\,\omega^2\right)k_\Phi^2-2\tau\beta\epsilon\mu\,\omega^2k_\Phi-\epsilon\mu\,\omega^2=0 , \qquad\text{i.e.}\qquad \boxed{\;n_\tau(\omega)=\frac{c\,k_\Phi^{\tau}}{\omega}=\frac{1}{1-\tau\beta\omega/c}\;} $$ on the propagating branch, with $k_\Phi^{\tau}=k_\Phi/(1-\tau k_\Phi\beta)$; in this labelling $\tau=-1$ is the branch written $k_\Phi^+$ in Maxwell’s Equations in Chiral Media. Both were checked to $2.2\times10^{-16}$ against the wave equation evaluated on the corresponding circular polarisation, and at $\beta=0$ the two branches collapse to the single $k_\Phi=\sqrt{\epsilon\mu}\,\omega$ of the non-chiral medium. The two refractive indices are the optical activity, and they come with no free constant beyond the one $\beta$ that entered the constitutive relations.
The fundamental solution and the causality principle. The source constructs the fundamental solution of $\mathcal{M}$ using quaternionic analysis. Writing $$ \mathrm{Sd}_\alpha(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)=-\frac{\exp(i\theta_{\mathrm{shf}})e^{-\kappa_{\mathrm{dmp}}\lvert r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z\rvert}}{4\pi|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|}, \qquad V_{\mathrm{vek},\alpha}(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)=-\mathrm{grad}\,\mathrm{Sd}_\alpha(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)+\alpha\,\mathrm{Sd}_\alpha(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z) =\left(\alpha+\frac{r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z}{|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|^2}-\frac{i\alpha\,(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)}{|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|}\right)\mathrm{Sd}_\alpha(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z), $$ the complex shift is split as $\alpha\,\lvert r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z\rvert=\theta_{\mathrm{shf}}+i\kappa_{\mathrm{dmp}}\lvert r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z\rvert$, with the shift angle $\theta_{\mathrm{shf}}=\operatorname{Re}\alpha\,\lvert\vec r\rvert$ and the decay rate $\kappa_{\mathrm{dmp}}=\operatorname{Im}\alpha$, both real because the shift has a real and an imaginary part; the object $V_{\mathrm{vek},\alpha}$ is a fundamental solution of the shifted operator in the source's own normalisation, $(D+\alpha)V_{\mathrm{vek},\alpha}=\delta$, and it satisfies at infinity the radiation condition $$ \left(1+\frac{i(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)}{|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|}\right)V_{\mathrm{vek},\alpha}(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)=o\!\left(\frac{1}{|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|}\right), $$ which the source identifies as the quaternionic form of the Silver–Müller condition. The factor $1+i(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)/|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|$ is a zero divisor of the algebra — its norm vanishes while its scalar part is $1$ — so the condition may not be divided by it and no decay rate follows from it alone; the caveat, and the extra decay hypothesis the literature attaches to it in the isotropic case, are recorded in Maxwell's Equations in Chiral Media — The Quaternionic Reformulation. The minus sign in $\mathrm{Sd}_\alpha$ is the source's and is not decorative. The bracketed form of $V_{\mathrm{vek},\alpha}$ factors $\mathrm{Sd}_\alpha$ out, so it reads identically with either sign; the sign is fixed by the kernel's defining property, and it is the minus that makes $(D+\alpha)V_{\mathrm{vek},\alpha}=+\delta$ rather than $-\delta$ — checked numerically, the surface integral of $\vec u V_{\mathrm{vek},\alpha}$ over a sphere surrounding the origin tending to $+1$ for the source's sign and to $-1$ for the other. It is fixed independently by the classical limit: at $\alpha=0$ the source's sign gives $\mathcal{F}_{\mathrm{med},0}=-(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)/(4\pi|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|^{3})$, the Cauchy kernel of $D$, which is also the kernel the article's inhomogeneous-media section arrives at, whereas the other sign would give its negative. The two expressions for $V_{\mathrm{vek},\alpha}$ agree (the gradient identity was verified to $3\times10^{-9}$, the finite-difference floor). After the time Fourier transform, the chiral operator's fundamental solution acquires the denominator $\omega-a$ with $a=1/(\beta\sqrt{\epsilon\mu})$, the same degenerate frequency as above — together with a double pole there, and an essential singularity through a factor $\exp(ic(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)/(\omega-a))$. For $\beta\neq0$, therefore, the fundamental solution is not a finite combination of the customary kernels; the source expands the exponential and reduces the problem to a series of residues.
Causality is then imposed, not assumed: among the regularisations of the inverse transform the source selects the one that vanishes for $r'_t<0$, by displacing the frequency into the upper half-plane, $\omega\mapsto\omega-i0$, "in agreement with the condition $\mathrm{Im}\,\alpha\geq0$". For a simple pole this gives the retarded factor $$ \frac{1}{2\pi i}\oint\frac{\exp(i\theta_{\mathrm{pwt}})\,d\omega}{\omega-a_y} =\mathrm{H}_{\mathrm{hvs}}(r'_t)\,\exp(i a_y r'_t), $$ with $a_y=a+iy$ and $\mathrm{H}_{\mathrm{hvs}}$ the Heaviside function, and the higher-order poles give $\mathrm{H}_{\mathrm{hvs}}(r'_t)\,\exp(ia_yr'_t)(ir'_t)^{n}/n!$; the residue formula was checked to $3.2\times10^{-9}$ for pole orders 1–3. For $r'_t<0$ the same integrand must be closed in the lower half-plane, where the displaced pole is not enclosed, so the contour integral vanishes: that is the content of the causality principle here, and it is exactly the statement that the displaced pole lies in the upper half-plane. Convolution of the resulting kernel with the quaternionic source then solves the inhomogeneous chiral system in whole space.
A medium with memory is the one departure from constant coefficients that keeps a symbol, and it is where dispersion relations come from. A coefficient that depends on the time destroys the multiplier, because a function of the time and the derivative with respect to it cannot be exchanged. A coefficient that depends on the frequency does not: in the frequency variable it is a number again, so the equation stays algebraic at each $\omega$, and that is why a permittivity is a function $\epsilon(\omega)$ and why the transform survives the passage from vacuum to matter.
Causality is then a statement about where that function is analytic. The response cannot precede the source, so $\epsilon$ extends into the upper half of the $\omega$-plane, and a function analytic in a half-plane has its real and imaginary parts tied by a Hilbert transform,
$$ \epsilon_{\mathrm{r}}(\omega)-\epsilon_{\infty}=\frac{1}{\pi}\lim_{\delta\to0^{+}}\int_{\lvert\omega'-\omega\rvert>\delta}\frac{\epsilon_{\mathrm{i}}(\omega')}{\omega'-\omega}\,d\omega', $$
with the reverse relation obtained by exchanging the two parts and the sign. Absorption therefore forces dispersion, and it forces it by a transform and not by a postulate: the two parts are one analytic function's two sides, and the integral is the price of the half-plane. The displacement $\omega\mapsto\omega-i0$ used just above is the same fact read as a contour statement, and the condition $\mathrm{Im}\,\alpha\geq0$ quoted from the source is its arithmetic. The pair is standard in the theory of dispersive media, and it is recorded here as the frequency-domain face of the causality principle this section uses.
The residue series resums. The source's interchange of the summation of the expanded exponential with the contour integration is justified, not formal — it checks the uniform convergence that licenses it — and the two residue series it produces are the Bessel series themselves: $$ \mathcal{I}_1=i\mathrm{H}_{\mathrm{hvs}}(r'_t)\,\exp(ia_yr'_t)J_0\!\left(2\sqrt{\kappa(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)\,r'_t}\right), \qquad \mathcal{I}_2=-\mathrm{H}_{\mathrm{hvs}}(r'_t)\sqrt{\frac{r'_{t}}{\kappa(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)}}\,\exp(ia_yr'_t)J_1\!\left(2\sqrt{\kappa(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)\,r'_t}\right), $$ where $\kappa(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)=|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|/(\beta^2\sqrt{\epsilon\mu})$ is the coefficient of the essential singularity, so that $2\sqrt{\kappa(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)r'_t}=2\sqrt{r'_t|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|}/(\beta(\epsilon\mu)^{1/4})$. Both identifications were checked to $6\times10^{-13}$ against the standard series of $J_0$ and $J_1$. The time dependence of the chiral fundamental solution is therefore a closed form in two Bessel functions, and the source writes it out in that form: the kernel is built from $K_{1/\beta}$ and $\mathrm{Sd}_{1/\beta}$, the parameter $1/\beta$ being the limit of $\alpha(\omega)$ as $\omega\to\infty$, that is the frequency-independent part of the exponent, multiplied by $J_0$ and $J_1$ of $2\sqrt{r'_t|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|}/(\beta(\epsilon\mu)^{1/4})$ and by the retarded factor $\mathrm{H}_{\mathrm{hvs}}(r'_t)\exp(iar'_t)/(\beta\sqrt{\epsilon\mu})$. What fails for $\beta\neq0$ is thus not the existence of a causal fundamental solution but its expression through the customary Cauchy and Helmholtz kernels alone; the essential singularity of the frequency-domain kernel resums into Bessel functions of the retarded variable $\sqrt{\kappa(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)r'_t}$, one order up from the non-chiral case, where the kernel is $\mathrm{H}_{\mathrm{hvs}}(r'_t)\delta(|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|-r'_t/\sqrt{\epsilon\mu})/(4\pi|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|)$ and no Bessel function appears.
What the addition is, and what it is not. The Drude–Born–Fedorov relations are a standard constitutive model, and everything above is a mathematical result about it: a reduction to one equation, a factorization, a degeneracy, a birefringence and a causal fundamental solution. Nothing in it is evidence for the biquaternion framework, and the framework's own epistemic standard applies. What it does supply is a precise structural statement about the framework's own reduction: that reduction is the $\beta=0$ member of a family, and the family's other members are not deformations of it — the operator changes order, acquires a degenerate frequency, and loses the expression of its fundamental solution as a finite combination of the customary kernels (it resums to Bessel functions instead, as recorded above). That is the reason the authors' remark that the chiral operator is "essentially different" is a statement about the algebra and not a figure of speech.
A note on the two time-dependence problems. This paper treats the time-dependent chiral problem. The companion work on the same topic, On a Quaternionic Reformulation of Maxwell's Equations for Chiral Media and its Applications, treats the time-harmonic chiral problem and integral representations; the series' further items treat time-harmonic operators for inhomogeneous media with numerical fundamental solutions. The organising fact is that the two problems need different operators and different solution methods, so a reader should choose by whether the time dependence is resolved or transformed away.
Where the time-harmonic problem is treated. The time-harmonic chiral problem — the diagonalization into the two circular combinations $Ξ,\tilde{Ξ}$ with the two wavenumbers $\alpha_1,\alpha_2$, the integral representations, the complete solution of the extendability problem, the electromagnetic energy balance, and the inhomogeneous (slowly varying, stratified) chiral medium — is recorded in the dedicated article Maxwell's Equations in Chiral Media — The Quaternionic Reformulation, whose source is the companion paper of Kravchenko and Oviedo. That material is deliberately not repeated here: this article keeps the time-dependent reduction, and the dedicated article carries the time-harmonic theory, so each definite description still has one referent.
Arbitrary Inhomogeneous Media: The Carrier Function and the Vekua Equation
Every medium so far has had constant $\epsilon$ and $\mu$. Let them now be differentiable positive functions of position, $\epsilon=\epsilon(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ and $\mu=\mu(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$, and keep the Maxwell system in the form
$$ \mathrm{rot}\,\vec H=\epsilon\,\partial/\partial r'_t\,\vec E+\vec J, \qquad \mathrm{rot}\,\vec E=-\mu\,\partial/\partial r'_t\,\vec H, \qquad \mathrm{div}(\epsilon\vec E)=\rho, \qquad \mathrm{div}(\mu\vec H)=0 , $$
which is the system of the preceding sections with the two parameters no longer constant. Nothing in the system is new. What is new is that the two divergence equations now produce the terms $\langle\mathrm{grad}\,\epsilon/\epsilon,\vec E\rangle$ and $\langle\mathrm{grad}\,\mu/\mu,\vec H\rangle$, so a reduction to a single quaternionic equation can no longer have constant coefficients.
The carrier identity. The reduction rests on one algebraic fact. Let $\xi\neq0$ be a scalar function and $g$ biquaternion-valued. Since $D$ is a first-order differential operator and $\xi$ is a scalar, hence commutes with the quaternion units,
$$ D[\xi\,g]=(D\xi)\,g+\xi\,Dg , $$
and since $D\xi=\mathrm{grad}\,\xi$, dividing by $\xi\neq0$ gives $$ \frac{1}{\xi}\,D(\xi\,g)=Dg+\frac{\mathrm{grad}\,\xi}{\xi}\,g, \qquad \frac{1}{\xi}\,D(g\xi)=Dg+\frac{\mathrm{grad}\,\xi}{\xi}\,g, $$ the two readings differing only in the side on which the scalar is attached, because $\xi$ passes through $g$. Write $\vec{\alpha}=\mathrm{grad}\,\xi/\xi$ for the coefficient, which stands on the left of $g$ in both readings. A coefficient of exactly this form is therefore removable: it is $D$ conjugated by multiplication by $\xi$, from the left in the first reading and from the right in the second. This is why the fields below are normalised by $\sqrt{\epsilon}$ and $\sqrt{\mu}$: the gradients the divergence equations produce are $\mathrm{grad}\,\epsilon/\epsilon=2\,\mathrm{grad}\sqrt{\epsilon}/\sqrt{\epsilon}$, so the square root is the factor that absorbs them. We call $\xi$ a carrier function and the identity the carrier identity.
The right product appears as well, and it is a different operator. The divergence equations produce $\vec E\,\ell_\epsilon$, with the coefficient on the right of the field, and the operator that carries it is $$ \mathcal{D}_{\vec{\alpha}}g:=Dg+g\,\vec{\alpha}, $$ a derived notation of this article — the right product written with one letter — naming no new object. It is not a conjugation of $D$: the coefficient sits between the field and the invertibles that $D$ multiplies by. That operator has a complete theory of its own, and the section The Operator $\mathcal{D}_{\vec{\alpha}}$ below states it.
A scalar-product identity. The divergence equations enter the reduction through scalar products, which have to be exchanged for products. For real vectors $\vec p,\vec q$ the products $\vec p\vec q$ and $\vec q\vec p$ have a common scalar part and opposite vector parts, so
$$ \langle\vec p,\vec q\rangle=-\tfrac{1}{2}\left(\vec p\vec q+\vec q\vec p\right), $$
the sign being fixed by $\vec e_j\vec e_\eta+\vec e_\eta\vec e_j=-2\delta_{j\eta}e_t$. This is the identity that lets a scalar product be written as a sum of the left and right products that the carrier identity can absorb.
The doubled pair. Write $\vec{\epsilon}$ and $\vec{\mu}$ for the two half-gradients,
$$ \vec{\epsilon}:=\frac{\mathrm{grad}\sqrt{\epsilon}}{\sqrt{\epsilon}}=\frac{1}{2}\frac{\mathrm{grad}\,\epsilon}{\epsilon}, \qquad \vec{\mu}:=\frac{\mathrm{grad}\sqrt{\mu}}{\sqrt{\mu}}=\frac{1}{2}\frac{\mathrm{grad}\,\mu}{\mu}, $$
and $\ell_\epsilon:=\mathrm{grad}\,\epsilon/\epsilon=2\vec{\epsilon}$, with $\ell_\mu$ defined the same way. Combining the two divergence equations with the two curl equations gives the pair
$$ D\,\vec E=\langle \ell_\epsilon,\vec E\rangle-\mu\,\partial/\partial r'_t\,\vec H-\frac{\rho}{\epsilon}, \qquad D\,\vec H=\langle \ell_\mu,\vec H\rangle+\epsilon\,\partial/\partial r'_t\,\vec E+\vec J, $$
and the scalar-product identity turns each of them into a gradient-corrected operator. For the electric field, adding $\vec{\epsilon}\,\vec E$ and using $\langle \ell_\epsilon,\vec E\rangle=-(\vec{\epsilon}\,\vec E+\vec E\,\vec{\epsilon})$,
$$ D\,\vec E+\vec{\epsilon}\,\vec E =-\tfrac{1}{2}\,\vec E\,\ell_\epsilon-\mu\,\partial/\partial r'_t\,\vec H-\frac{\rho}{\epsilon}, $$
and for the magnetic field
$$ D\,\vec H+\vec{\mu}\,\vec H =-\tfrac{1}{2}\,\vec H\,\ell_\mu+\epsilon\,\partial/\partial r'_t\,\vec E+\vec J, $$
where the coefficient stands on the left of the field as written, while the two corrections on the right-hand sides are right products. The order is not cosmetic: exchanging $\vec E\ell_\epsilon$ for $\ell_\epsilon\vec E$ changes the remainder by $2\,\vec E\times \ell_\epsilon$.
The reformulation. The carrier identity with $\xi=\sqrt{\epsilon}$ says that $\frac{1}{\sqrt{\epsilon}}D(\sqrt{\epsilon}\vec E)=D\vec E+\vec{\epsilon}\,\vec E$, and with $\xi=\sqrt{\mu}$ the analogous statement holds for $\vec H$. Introducing the normalised fields and the medium speed,
$$ \vec{\mathcal E}:=\sqrt{\epsilon}\,\vec E, \qquad \vec{\mathcal H}:=\sqrt{\mu}\,\vec H, \qquad c=\frac{1}{\sqrt{\epsilon\mu}}, $$
the pair becomes
$$ D\,\vec{\mathcal E}+\vec{\mathcal E}\,\vec{\epsilon} =-\frac{1}{c}\,\partial/\partial r'_t\,\vec{\mathcal H}-\frac{\rho}{\sqrt{\epsilon}}, \qquad D\,\vec{\mathcal H}+\vec{\mathcal H}\,\vec{\mu} =\frac{1}{c}\,\partial/\partial r'_t\,\vec{\mathcal E}+\sqrt{\mu}\,\vec J, $$
the corrections now being right products, as the dots record. In the time-harmonic convention $\exp(i\theta_{\mathrm{pwt}})$ used by the source of this section — the dispersion section above uses $\exp(-i\theta_{\mathrm{pwt}})$ and the two must not be mixed — with $k_\Phi=\omega\sqrt{\epsilon\mu}$,
$$ D\,\vec{\mathcal E}+\vec{\mathcal E}\,\vec{\epsilon}=-ik_\Phi\,\vec{\mathcal H}-\frac{\rho}{\sqrt{\epsilon}}, \qquad D\,\vec{\mathcal H}+\vec{\mathcal H}\,\vec{\mu}=ik_\Phi\,\vec{\mathcal E}+\sqrt{\mu}\,\vec J. $$
These are exact equivalents of the Maxwell system for arbitrary $\epsilon(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ and $\mu(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$. "Exact" is meant literally: this is neither a weak-inhomogeneity nor a slowly-varying approximation, and no term has been dropped. At constant $\epsilon,\mu$ the vectors $\vec{\epsilon},\vec{\mu}$ vanish and the pair collapses to $D\vec{\mathcal E}=-ik_\Phi\vec{\mathcal H}-\rho/\sqrt{\epsilon}$ and $D\vec{\mathcal H}=ik_\Phi\vec{\mathcal E}+\sqrt{\mu}\,\vec J$, which is the constant-coefficient reduction of the earlier sections written in the normalisation of this one.
Two conventions meet here, and they should be separated. The vacuum operator of the source's reduction is $\partial/\partial r'_t+iD$, while the non-chiral reduction recorded in the chiral section above is $M_0=\partial/\partial r'_t-iD$. The two are complex conjugates of one another and not rivals. The field $\mathcal{F}$ used below is $\sqrt{\epsilon}$ times the Riemann–Silberstein vector, that is, $\sqrt{\epsilon}$ times the conjugate of the chiral section's purely vectorial field $\vec{\mathcal N}=\vec E-iZ\vec H$; conjugating a reduction built on $M_0$ replaces $-iD$ by $+iD$, so the sign of the $iD$ term is carried by which field one starts from. The sections below use the source's convention throughout, and the dictionary to the corpus's field strength is given at the end of this section.
The single equation, and why it is a Vekua equation. One further combination collapses the pair. Set
$$ \mathcal{F}:=\vec{\mathcal E}+i\,\vec{\mathcal H}, \qquad \vec{c}:=\frac{\mathrm{grad}\sqrt{c}}{\sqrt{c}}=\frac{1}{2}\,\mathrm{grad}\ln c, \qquad \vec{Z}:=\frac{\mathrm{grad}\sqrt{Z}}{\sqrt{Z}}=\frac{1}{2}\,\mathrm{grad}\ln Z , $$
where $Z=\sqrt{\mu/\epsilon}$ is the intrinsic impedance fixed earlier. Then the two equations are together equivalent to the single equation
$$ \boxed{\;\left(\frac{1}{c}\,\partial/\partial r'_t+iD\right)\mathcal{F}-\mathcal{F}\,(i\vec{c})-\bar{\mathcal{F}}\,(i\vec{Z}) =-\left(\sqrt{\mu}\,\vec J+\frac{i\rho}{\sqrt{\epsilon}}\right)\;}, $$
with $\bar{\mathcal{F}}$ the coefficientwise complex conjugate of $\mathcal{F}$ — the scalar imaginary conjugated, the quaternion units untouched — and the two products again right products. The coefficientwise conjugation is written $\bar{\cdot}$ and not ${}^{*}$: the starred symbol is the Hermitian conjugation, a different involution, and the equation is not invariant under it. The source writes the impedance $W$ and the two coefficient vectors $\vec{c}$ and $\vec{W}$, the letters $\vec{c},\vec{W}$ standing for $\frac12\mathrm{grad}\ln c$ and $\frac12\mathrm{grad}\ln W$; we keep this article's letter $Z$ for the impedance, and we keep the source's symbol $\vec{c}$ for the first coefficient vector, which is a vector field built from the speed and not the speed $c$. The two coefficient vectors are not independent of the pair above: because
$$ \vec{\epsilon}+\vec{\mu}=-\frac{\mathrm{grad}\,c}{c}=-2\vec{c}, \qquad \vec{\epsilon}-\vec{\mu}=-\frac{\mathrm{grad}\,Z}{Z}=-2\vec{Z}, $$
the symmetric and antisymmetric combinations of the two medium gradients are precisely the two coefficients that appear.
The structure of the equation is what matters. The operator acting on $\mathcal{F}$ is the vacuum operator $\partial/\partial r'_t+iD$ of the earlier sections; the inhomogeneity adds two zeroth-order terms with variable coefficients, one multiplying $\mathcal{F}$ and one multiplying $\bar{\mathcal{F}}$. A first-order equation whose coefficients act on both $\mathcal{F}$ and its conjugate is a Vekua equation — the prototype being $w_{\bar z}=Aw+Bw^{\natural}$ in one complex variable — and the Vekua equation is the governing equation of generalized analytic, or pseudoanalytic, functions. The result is that Maxwell's equations for an arbitrary inhomogeneous medium are equivalent to a single quaternionic equation of Vekua type, so that the theory of pseudoanalytic functions becomes available for electromagnetism in arbitrary media: its integral representations, its similarity principle, its Liouville-type theorems, and its existence and solvability results. The corpus now carries the first layer of that theory — the generating quartet of the Vekua-type equation, the reduction of the equation to a condition on the coefficients, the equivalent Vekua equation for their combination, and the inverse of the reduction — in the section The Generating Quartet, the Pseudoanalytic Equation, and the Inverse of the Reduction below; the four items listed in the previous sentence are what that section does not yet supply, and they are supplied for the linear theory by the integral-representations and Riemann–Hilbert material of the analysis cluster — the boundary value problem with a jump and the singular integral equation of Cauchy type of Riemann Boundary Value Problems and Singular Integral Equations, whose Plemelj formulas, canonical factor and solvability criteria are the general form of the four items.
Two limitations belong with the result. The equivalence relocates the difficulty rather than removing it: the reduction is exact, but it is a reduction to an equation that must then be solved. And the reduction is not a closed-form solution of the time-dependent inhomogeneous problem. What the framework delivers in closed form is the time-dependent problem for homogeneous media, including the chiral ones, and the static problem for inhomogeneous media; the arbitrary inhomogeneous case reaches the equivalent single equation and stops there.
The dictionary with this article's field strength. The field $\mathcal{F}=\sqrt{\epsilon}\,\vec E+i\sqrt{\mu}\,\vec H$ is not the field strength $\tilde{F}$ fixed earlier, but it is a fixed multiple of it:
$$ \tilde{F}=i\,\mathcal{F} . $$
The field-strength biquaternion of this article is therefore $i$ times this paper's $\mathcal{F}$: the factor $i$ sends $i\sqrt{\epsilon}\,\vec E$ to $-\sqrt{\epsilon}\,\vec E$ and $-\sqrt{\mu}\,\vec H$ to $-i\sqrt{\mu}\,\vec H$, so it exchanges the two halves, and the two fields describe the same Maxwell system in the normalisation that makes the medium gradient removable. In this dictionary the equation above is the inhomogeneous-medium member of the family whose constant-coefficient member is the single equation $\tilde\nabla_\Phi\tilde{F}=-\sqrt{\mu}\,\tilde{J}_{\Phi}$ of the earlier sections.
The same reduction read from the Dirac side. The carrier identity of this section has a counterpart that reads it as a Dirac potential, and the counterpart is exact only on one of the two Dirac equations the corpus carries. They must be kept apart.
The corpus's own equation is one-sided, the pair on the two halves $\tilde{Ξ}_{L}$ and $\tilde{Ξ}_{R}$ of a Dirac field, $$ \tilde\nabla_\Phi\tilde{Ξ}_{R}=k_m\tilde{Ξ}_{L}, \qquad (\tilde\nabla_\Phi)^{\natural}\tilde{Ξ}_{L}=k_m\tilde{Ξ}_{R}. $$ Both lines act from the left, the mass is the off-diagonal pairing of the two halves, and no term carries a right multiplication. The two halves are told apart by the subscript $L$ or $R$, and this is the equation whose dividing line is that label. Here $\hbar/(mc)$ is the mass length.
Kravchenko's equation is two-sided: one equation on the algebra, $$ (\tilde\nabla_\Phi)^{\natural}F=-mi\,\bar F\,\vec e_z, $$ where the spatial gradient acts from the left, the mass term acts from the right through the fixed direction $\vec e_z$, and $\bar{\cdot}$ is the coefficientwise complex conjugation of the introduction. The position dependence of a potential is a coefficient multiplying one basis direction on the right, and it is this second form, not the one-sided pair, that has the shape of the carrier identity.
That is the whole content of the correspondence below. The massive equation with an electric potential, or separately with a scalar potential, becomes after the dictionary an equation whose only position dependence is such a coefficient, and matching it with the static reduction $D\vec{\mathcal E}+\vec{\mathcal E}\vec{\epsilon}=0$ of a sourceless medium — a right multiplication by $\vec{\epsilon}=\mathrm{grad}\sqrt{\epsilon}/\sqrt{\epsilon}$ — identifies the two. The one-sided pair has no right multiplication at all, so it cannot match a reduction whose inhomogeneity enters from the right; the two-sided equation has exactly that structure. The permittivity is then the exponential of the potential, $$ \epsilon=\exp\left(2G(x_1)\right), \qquad \mathrm{grad}\,G=\vec{\epsilon}, $$ so that the potential $G$ is the logarithmic derivative of the square root of the permittivity, which is the carrier function of the normalisation above. The correspondence is recorded in The Dirac Equation in Biquaternionic Form, section The Static Maxwell Field of an Inhomogeneous Medium as a Dirac Potential. It belongs here because the two ends are this section's carrier function and that article's potential coupling, and the statement is a dictionary between them rather than a new solution. It is a static statement, and it does not extend to the time-dependent inhomogeneous case, where the reduction is the Vekua equation and no such identification is available.
The Operator $\mathcal{D}_{\vec{\alpha}}$, the Reduction to the Schrödinger Equation, and a Fundamental Solution
In the static case the reformulation reads
$$ D\,\vec{\mathcal E}+\vec{\mathcal E}\,\vec{\epsilon}=-\frac{\rho}{\sqrt{\epsilon}}, \qquad D\,\vec{\mathcal H}+\vec{\mathcal H}\,\vec{\mu}=\sqrt{\mu}\,\vec J, $$
and both are instances of one operator problem: solve
$$ \mathcal{D}_{\vec{\alpha}}\vec{F}=\vec{G}, \qquad \mathcal{D}_{\vec{\alpha}}g=Dg+g\,\vec{\alpha}, \qquad \vec{\alpha}=\frac{\mathrm{grad}\,\xi}{\xi}, $$
for some scalar $\xi\neq0$. The coefficient stands on the right because that is the side on which the reformulation produces it, and the operator with the coefficient on that side has a complete and explicit theory, whose fundamental solution turns out to be an object already met in this article.
Why the coefficient sits on the right. A coefficient of the same form on the left of the field is removable by a conjugation, as the carrier identity states; the right product is not. What survives is the mixed product of the two sides, and the factorisation below returns to it with the remainder that a same-side product leaves.
The reduction to the Schrödinger equation. Let $\xi\neq0$ be a scalar, $\vec{\alpha}=\mathrm{grad}\,\xi/\xi$ and $v=\Delta_\Phi\xi/\xi$, and let $\tilde{Ξ}$ solve the Schrödinger equation
$$ -\Delta_\Phi\tilde{Ξ}+v\tilde{Ξ}=0 . $$
Then
$$ \vec{F}:=D\tilde{Ξ}-\vec{\alpha}\,\tilde{Ξ} $$
solves $\mathcal{D}_{\vec{\alpha}}\vec{F}=0$, that is $D\vec{F}+\vec{F}\vec{\alpha}=0$. Moreover, if $\tilde{Ξ}$ is a fundamental solution of the operator $-\Delta_\Phi+v$, then $\vec{F}$ is a fundamental solution of $\mathcal{D}_{\vec{\alpha}}$. A first-order quaternionic operator with a gradient coefficient is thus reduced to a scalar Schrödinger equation, and the potential is not free: it is built from the same $\xi$ that produces the coefficient.
The factorisation, and the Riccati equation. The corresponding operator identity is a factorisation of the Schrödinger operator. For a scalar function $u$,
$$ \boxed{\;\mathcal{D}_{\vec{\alpha}}\bigl(Du-\vec{\alpha}\,u\bigr)=(-\Delta_\Phi+v)u\;} \qquad\text{provided}\qquad D\vec{\alpha}+\vec{\alpha}^{2}=-v , $$
and $\vec{\alpha}=\mathrm{grad}\,\xi/\xi$ with $v=\Delta_\Phi\xi/\xi$ satisfies that condition identically, the relation $D\vec{\alpha}+\vec{\alpha}^2=-v$ being equivalent to $-\Delta_\Phi\xi+v\xi=0$. The relation is the quaternionic generalisation of the Riccati equation, and it is the same relation that appears in the corpus's treatment of the quaternionic Riccati equation. The factorisation is a mixed statement, and the sides are not decorative: with the coefficient on the left in both factors, the product of $g\mapsto Dg+\vec{\alpha}g$ with $g\mapsto Dg-\vec{\alpha}g$ acquires the first-order remainder $-2\sum_\eta(\partial u/\partial R_\eta)(\vec e_\eta\times\vec{\alpha})$, which vanishes only when $\vec{\alpha}$ is scalar, whereas the mixed operator above reproduces $(-\Delta_\Phi+v)u$ exactly for scalar $u$. The asymmetry between the two sides is a property of the algebra, not of the notation.
The same factorisation covers a whole family of operators — including the conductivity equation. The factorisation above is the particular case $p=1$, $q=v$ of a more general identity, and the general identity is what connects the framework to the rest of mathematical physics. Let $p,q,u_0$ be complex-valued with $p\in C^2(\Omega)$, $p\neq0$ in $\Omega$, and let $u_0$ be a nonvanishing particular solution of the second-order equation
$$ (\mathrm{div}\,p\,\mathrm{grad}+q)u=0\qquad\text{in }\Omega . $$
Put $f=p^{1/2}u_0$ and $\vec{\alpha}_f=\mathrm{grad}\,f/f$. Then for every scalar $\xi\in C^2(\Omega)$,
$$ (\mathrm{div}\,p\,\mathrm{grad}+q)\xi=-p^{1/2}\,\mathcal{D}_{\vec{\alpha}_f}\Bigl(D\bigl(p^{1/2}\xi\bigr)-\vec{\alpha}_f\,p^{1/2}\xi\Bigr) , $$
with the same convention as in the paragraph above — the coefficient on the right in the outer factor, on the left in the inner one; the source writes both factors with a single letter for the coefficient, and its $f$ is the carrier $p^{1/2}u_0$, not the coefficient. The identity was checked symbolically here in two cases, one with $p\neq1$ (the conductivity case $p=\exp(2x_1)$, $q=0$, $f=\exp(x_1)$) and one with a non-constant coefficient ($p=1$, $f=x_1^2+1$, $q=-\Delta_\Phi f/f$); reversing the two multiplication sides makes the identity fail. So the second-order operator is the same biquaternionic factorisation for every choice of $p$ and $q$: the Schrödinger operator $-\Delta_\Phi+v$ is the case $p=1$, $q=-v$; the conductivity equation $\mathrm{div}\,\sigma\,\mathrm{grad}\,u=0$ of a medium with conductivity $\sigma$ is the case $p=\sigma$, $q=0$, with the particular solution $u_0\equiv1$ and hence the carrier $f=\sqrt{\sigma}$; and each is carried by one and the same pair of first-order operators, $\mathcal{D}_{\vec{\alpha}_f}$ and $g\mapsto Dg-\vec{\alpha}_f g$. This is the precise sense in which the biquaternionic formulation is one algebra, many equations: not an analogy between separate reformulations but a single factorisation that specialises to each of them, with the carrier $f$ built from the zeroth-order coefficient in every case. The Schrödinger member of this family and the riccati condition that selects its carrier are the special case; the general $(\mathrm{div}\,p\,\mathrm{grad}+q)$ member, and with it the conduction problem, is what the survey's operator-relation theme adds.
The one-dimensional general solution: one particular solution generates them all. The factorisation says what the carrier $f=p^{1/2}u_0$ is; it does not say what the solutions of the factorised equation are. In one dimension they are known explicitly, and the construction is precisely what the account above leaves open. Take the equation in Sturm–Liouville form,
$$ \frac{d}{dx}\left(p\frac{du}{dx}\right)+qu=\omega^2p\,u,\qquad x\in[0,a], $$
with $p,q,u$ complex-valued, $p\in C^1(0,a)$ bounded and nonvanishing, and $\omega$ an arbitrary complex number. Suppose that the auxiliary equation
$$ \frac{d}{dx}\left(p\frac{dg_0}{dx}\right)+qg_0=0 $$
has a particular solution $g_0\in C^2(0,a)$ such that $g_0$ and $1/g_0$ are bounded on $[0,a]$ — nonvanishing, and nondegenerate at the endpoints — and put $g=\sqrt{p}\,g_0$. The weight $p$ on the right is the Sturm–Liouville weight, and it is what makes the recursions close: dividing the equation by $p$ gives $\frac{1}{p}\frac{d}{dx}\left(p\frac{du}{dx}\right)+\frac{q}{p}u=\omega^2u$, the standard eigenvalue form, and the two weights of the recursions below are $g^{2}=p\,g_0^{2}$ and $g^{-2}$, whose product is $1$. Then the general solution of the first equation is
$$ u=c_1u_1+c_2u_2, \qquad u_1=g_0\!\!\sum_{\substack{n=0\ n\ \text{even}}}^{\infty}\!\!\frac{\omega^nX^{(n)}_{+}}{n!}, \qquad u_2=g_0\!\!\sum_{\substack{n=1\ n\ \text{odd}}}^{\infty}\!\!\frac{\omega^nX^{(n)}_{-}}{n!}, $$
where the coefficients are generated from $X^{(0)}_{+}=X^{(0)}_{-}\equiv1$ by two interleaved recursions,
$$ X^{(n)}_{+}(x)= \begin{cases} n\displaystyle\int_0^xX^{(n-1)}_{+}(\zeta)\,g^{2}(\zeta)\,d\zeta, & n\ \text{odd},\\[6pt] n\displaystyle\int_0^xX^{(n-1)}_{+}(\zeta)\,g^{-2}(\zeta)\,d\zeta, & n\ \text{even}, \end{cases} $$
$$ X^{(n)}_{-}(x)= \begin{cases} n\displaystyle\int_0^xX^{(n-1)}_{-}(\zeta)\,g^{-2}(\zeta)\,d\zeta, & n\ \text{odd},\\[6pt] n\displaystyle\int_0^xX^{(n-1)}_{-}(\zeta)\,g^{2}(\zeta)\,d\zeta, & n\ \text{even}. \end{cases} $$
The two series are the even and the odd parts in $\omega$. Their values at the base point follow from $X^{(0)}_{\pm}=1$ and the fact that every $X^{(n)}_{\pm}$ with $n>0$ vanishes at $0$: $u_1(0)=g_0(0)$, $u_1'(0)=g_0'(0)$, $u_2(0)=0$ and $u_2'(0)=\omega/(p(0)g_0(0))$, the last because the first weight of the odd series is $g^{-2}=1/(p\,g_0^{2})$. The Wronskian at $x=0$ is therefore $\omega/p(0)$, which is nonzero, so $u_1$ and $u_2$ are linearly independent and the two series do span the solution space; the two constants $c_1,c_2$ are the two initial data in that normalisation, $c_1=u(0)/g_0(0)$ and $c_2$ fixed by $u'(0)=c_1g_0'(0)+c_2\omega/(p(0)g_0(0))$. For the case $p=1$, which is the ordinary form $u''+qu=\omega^2u$, the carrier is $g=g_0$, so the two weights of the recursion are $g_0^{2}$ and $g_0^{-2}$, and the series reads $X^{(n)}_{+}=n\int_0^xX^{(n-1)}_{+}g_0^{2}\,d\zeta$ for odd $n$ and $n\int_0^xX^{(n-1)}_{+}g_0^{-2}d\zeta$ for even $n$, with $X^{(n)}_{-}$ exchanging the two cases. The recursion is the classical one — the series solution of a second-order linear equation was known in this form to Weyl, and it is the same representation used in inverse spectral theory — and what the paper contributes is the statement of the hypotheses under which it is uniformly usable, and the route to it through the theory of pseudoanalytic functions.
The carrier of the series is the carrier of the factorisation. The two carriers are the same function. The series is normalised by $g=\sqrt{p}\,g_0$ with $g_0$ a particular solution of $\frac{d}{dx}\left(p\frac{dg_0}{dx}\right)+qg_0=0$; the factorisation above is carried by $f=p^{1/2}u_0$ with $u_0$ a nonvanishing particular solution of $(\mathrm{div}\,p\,\mathrm{grad}+q)u=0$. In one dimension the two equations are the same equation, so $g=f$ term by term. What the one-dimensional theory adds to the factorisation is the second boundedness condition: the factorisation needs only that $u_0$ be nonvanishing, while the series additionally needs $1/g_0$ bounded, and it is that reciprocal condition which makes the carrier and its square well behaved at both endpoints and hence makes the coefficients computable on the whole interval. The condition is concrete and checkable, and it is exactly the criterion a reader of the factorisation section would want and the section does not supply.
Why $\omega$ is the right variable, and what the method is worth numerically. The series is a power series in $\omega$ with coefficients independent of $\omega$. Once the $X^{(n)}_{+}$ and $X^{(n)}_{-}$ are computed to order $N$, an approximate solution is a polynomial in $\omega$ with those coefficients, so an initial-value problem is solved for all $\omega$ at once and a spectral problem in $\omega$ becomes root-finding for a polynomial. That is the property the older power-series representation of the same solution lacked: in the classical form the spectral parameter enters in a manner too complicated to be used quantitatively. The paper's motivating instance is electromagnetic, the equation $-\frac{d^2u}{dx^2}+\omega^2q(x)u=0$ for different complex values of $\omega^2$ — the stratified-medium problem of Wait — which is the corpus's own frequency dependence, and the method returns the solution as a function of the wave number rather than one solution per wave number. The paper reports that the truncated series beats Matlab's adaptive solver ode45 by several orders of magnitude on three test problems, and the figures are recorded as reported and not reproduced: for $q\equiv-a^2$ with initial data $u(0)=1$, $(du/dx)(0)=-1$ on $(0,1)$ it reports absolute and relative errors of order $10^{-16}$ and $10^{-14}$ for the series at $N=55$ to $58$ against $10^{-9}$ and $10^{-6}$ for ode45 at $a=1$, and order $10^{-12}$ for both against $10^{-6}$ and $10^{-5}$ at $a=10$; for $q=a^2x^2+a$, where the exact solution is $u=\exp(ax^2/2)\bigl(c_1+c_2\int_0^{r_x}\exp(-a\,{r'_t}^{2})\,dr'_t\bigr)$, the reported ode45 absolute error at $a=30$ is $0.28$. The paper's coefficient is renamed $a$ here, its $c$ being the speed of light in this corpus and nothing else against the series' $10^{-9}$. The comparison is the author's own, on Matlab 7, and it is a comparison with one adaptive solver rather than a benchmark; the corpus records it as an announced result. Two standing weaknesses of the paper are worth carrying with it. Every example is one-dimensional and the coefficients $p,q$ are smooth on a closed interval, so nothing here addresses a singular endpoint, where the boundedness of $1/g_0$ can fail; and the paper is a short announcement, so the convergence estimates are sketched through the Weierstrass test rather than developed.
The three equations that follow from one solution. The relations between the operators become relations between their solutions. Let the biquaternion-valued function $\mathcal{W}$ solve the Vekua-type equation
$$ D\,\mathcal{W}-\frac{Df}{f}\,\mathcal{C}\,\mathcal{W}=0 , $$
with the coefficient multiplying on the left, $\mathcal{C}$ being the conjugation operator of the algebra — the biquaternionic analogue of the $\bar{(\cdot)}$ in the prototype Vekua equation $w_{\bar z}=Aw+Bw^{\natural}$ discussed in the section above, which is why this is the same Vekua class and not merely a similar-looking equation. Write $\mathcal{W}=\mathcal{W}_0+\vec{\mathcal{W}}$ for its scalar and vector parts. Then, as the source's Theorem 18 states, three consequences follow at once: the scalar part $\mathcal{W}_0$ solves the stationary Schrödinger equation
$$ -\Delta_\Phi \mathcal{W}_0+\frac{\Delta_\Phi f}{f}\mathcal{W}_0=0 , $$
the function $u=f^{-1}\mathcal{W}_0$ solves the conductivity equation
$$ \mathrm{div}\left(f^{2}\,\mathrm{grad}\,u\right)=0 , $$
and the vector part, rescaled as $\vec{v}=f\vec{\mathcal{W}}$, solves
$$ \mathrm{rot}\left(f^{-2}\,\mathrm{rot}\,\vec{v}\right)=0 . $$
One biquaternionic equation therefore carries a Schrödinger solution, a conductivity solution and a second-order curl equation at once, the three tied by the single carrier $f$. This is the operator-relation web the survey names in its abstract — "the Schrödinger, the Maxwell system, the conductivity equation and others" — and it is the reason the framework is worth the notation: the same first-order operator that reformulates Maxwell also factorises the conductivity equation, so a result about one is transferable to the others. The conduction member is the one this article had not carried; the corpus's treatment of the lossy Maxwell system in The Time-Harmonic Maxwell Operator and the Quaternionic Cauchy Integral uses a constant complex conductivity inside Maxwell, whereas the equation above is the scalar, spatially varying conductivity equation of a conduction problem, a different object reached by the same factorisation.
An explicit fundamental solution. The simplest case is a constant quotient. Let $\Delta_\Phi\xi/\xi=-k_\Phi^2$ throughout a domain, so that $\xi$ is a combination of exponentials $\exp(ik_\Phi\,\vec u\cdot(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z))$ with $|\vec u|=1$. Then
$$ \tilde{Ξ}(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)=\frac{\exp(ik_\Phi|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|)}{4\pi|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|} $$
is the outgoing fundamental solution of $-\Delta_\Phi-k_\Phi^2$, and the construction gives the fundamental solution of $\mathcal{D}_{\vec{\alpha}}$ in closed form,
$$ \boxed{\;\vec{F}(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z) =\left(D-\frac{\mathrm{grad}\,\xi}{\xi}\right)\frac{\exp(ikr)}{4\pi r} =-\left(\vec{\alpha}+\frac{r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z}{r^{2}}-\frac{ik_\Phi\,(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)}{r}\right)\frac{\exp(ikr)}{4\pi r}\;}, \qquad r=|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z| . $$
The authors remark that it is not clear how to obtain this result directly for the Maxwell operators $\mathcal{D}_{\vec{\mu}}$ and $\mathcal{D}_{\vec{\epsilon}}$ by other known methods.
The recipe is the chiral recipe; the kernel is not the chiral kernel. The bracketed factor has already appeared in this article, and it is worth saying exactly how far the resemblance goes. The fundamental solution of the chiral operator in A Chiral Medium: One Quaternionic Equation is
$$ \mathcal{K}_{\pm\alpha}=-(D\mp\alpha)\,\mathrm{Sd}_{\pm\alpha}, \qquad \mathrm{Sd}_\alpha=-\frac{\exp(i\theta_{\mathrm{shf}})e^{-\kappa_{\mathrm{dmp}}\lvert r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z\rvert}}{4\pi|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|}, $$
minus the first-order operator applied to the outgoing Helmholtz kernel of that same operator. The formula obtained above is the same display,
$$ \vec{F}(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)=-(D-\vec{\alpha})\,\mathrm{Sd}_{k_\Phi}, \qquad \mathrm{Sd}_{k_\Phi}=-\frac{\exp(ik_\Phi|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|)}{4\pi|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|}, $$
with the scalar shift $\alpha$ replaced by the vector coefficient $\vec{\alpha}$: indeed $-(D-\vec{\alpha})\mathrm{Sd}_{k_\Phi}=-D\mathrm{Sd}_{k_\Phi}+\vec{\alpha}\mathrm{Sd}_{k_\Phi}=(D-\vec{\alpha})\tilde{Ξ}$ for $\tilde{Ξ}=\exp(ik_\Phi r)/4\pi r$. The two share the whole shape — the $(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)/r^{2}$ and $i\alpha(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)/r$ terms both come from differentiating $\exp(i\alpha r)/4\pi r$ — and they differ in the one place that matters, namely the type of the coefficient. The chiral shift is a frequency-dependent scalar, so the chiral kernel is scalar-plus-vector, with scalar part $\alpha\mathrm{Sd}_\alpha$; the coefficient here is the vector $\mathrm{grad}\,\xi/\xi$ produced by an inhomogeneity, so this fundamental solution is a pure vector. The two are therefore not the same object, and the difference is not a sign: it is of order one. They coincide only in the degenerate case $\vec{\alpha}=0$, $k_\Phi=0$, where both reduce to the classical spatial Cauchy kernel $-(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)/(4\pi r^{3})$ of the unshifted operator. What is genuine, and worth keeping, is that one constructive recipe — apply the first-order operator to the outgoing Helmholtz kernel of the same operator — serves the chiral problem with a scalar shift and the inhomogeneous problem with a gradient coefficient alike; and that the recipe, not the kernel, is what the two have in common.
What this adds, and what it does not. The inhomogeneous branch of the framework now has four levels, and they should be kept apart. The exact reduction to a single Vekua-type equation holds for arbitrary $\epsilon(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ and $\mu(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ and for time-dependent fields, but it delivers an equation rather than a solution. The theory of the operator $\mathcal{D}_{\vec{\alpha}}$ with a gradient coefficient is complete and explicit, and it delivers both the Schrödinger reduction and the closed-form fundamental solution, but only for coefficients of the carrier form $\mathrm{grad}\,\xi/\xi$, and, in the closed-form example, for the constant quotient $\Delta_\Phi\xi/\xi=-k_\Phi^2$. The general factorisation of $(\mathrm{div}\,p\,\mathrm{grad}+q)$, and with it the conductivity equation, is a third interpretation of the same first-order operators rather than a further level of the same branch: it is the widest statement of the factorisation, of which the Schrödinger reduction is the case $p=1$, and it is an operator identity rather than a solution. The one-dimensional general solution of the last of these is the fourth level, and it is where the sequence finally closes on a formula: one bounded particular solution $g_0$ of the auxiliary equation, with $1/g_0$ bounded as well, generates the whole solution space of $\frac{d}{dx}\left(p\frac{du}{dx}\right)+qu=\omega^2p\,u$ through two explicit recursions, so the reduction that stopped at an equation now ends at a solution — in one dimension, for smooth coefficients, and with the reciprocal boundedness as its price. What remains absent is the body of the general pseudoanalytic, or Vekua, function theory in more than one dimension that the Vekua reduction makes available — the corpus now carries its first layer, the generating quartet with the equation for $w$ and the inverse of the reduction, recorded in the next section, while the theory's own theorems, its similarity principle and its Liouville-type statements among them, are still not carried — a closed-form treatment of the time-dependent inhomogeneous problem, and any numerical work in this article beyond the one-dimensional series — the framework's other numerical method, the collocation scheme of the survey, is recorded in the chiral article where the scattering problem lives.
A note on the two sources. The material of these two sections is taken from V. V. Kravchenko, Quaternionic reformulation of Maxwell's equations for inhomogeneous media and new solutions (arXiv:math-ph/0104008) and Quaternionic equation for electromagnetic fields in inhomogeneous media (arXiv:math-ph/0202010), with one exception: the general $(\mathrm{div}\,p\,\mathrm{grad}+q)$ factorisation of the operator-relation web and the conductivity equation it carries are not in the 2001 paper, which stops at the Schrödinger member; they are cited in the Further Reading to the survey's reference $[29]$, V. V. Kravchenko, J. Phys. A 39 (2006) 12407–12425, which is also the primary source of the generating-quartet section below. Both papers write the Moisil–Teodoresco operator as $D=\sum_\eta i_\eta\partial/\partial R_\eta$ with the quaternion units called $i_k$; the units $\vec e_\eta$ of this article are the same objects, and their $D$ is the spatial part of the biquaternionic gradient fixed above. In the available text layer of both papers the distinction between multiplying by a coefficient from the left and from the right is not always preserved, so the multiplication order recorded here is not the printed order but the order that reproduces the equations: every statement above was checked numerically, and wherever the alternative order fails it fails by an $O(1)$ margin. One cross-article warning follows from this. In these two papers the two multiplications carry two different letters, and this article names them by position in the product, so the letter written after the element is the one applied last. In Biquaternion Regular Functions, by contrast, the shifted operator written $D_\alpha$ adds a right product under a letter that names a left product here, so the same letter carries the other side there.
The Generating Quartet, the Pseudoanalytic Equation, and the Inverse of the Reduction
The Vekua-type equation has four explicit solutions, and they generate all of them. The operator-relation web above stops at an equation, and an equation is not a solution theory. The constructive half of that theory rests on a quartet. With $f\neq0$ the carrier and $Df/f=\mathrm{grad}\,f/f$ the coefficient, the four functions
$$ F_0=f,\qquad F_1=\frac{\vec e_x}{f},\qquad F_2=\frac{\vec e_y}{f},\qquad F_3=\frac{\vec e_z}{f} $$
all solve the Vekua-type equation $D\,\mathcal{W}-\tfrac{Df}{f}\,\mathcal{C}\,\mathcal{W}=0$. They solve it for one definite interpretation of $\mathcal{C}$ and of the side on which the coefficient multiplies. The algebra offers three conjugations — the quaternion conjugation, which keeps the scalar part and negates the vector part, the complex conjugation $i\mapsto-i$, and their composite — and to these the identity was added as a control, each of the four being tried on both sides. Only one of the eight cases makes all four members vanish, namely the quaternion conjugation multiplying on the left. Over sixty random carriers the residual of that case stays below $10^{-15}$, while the other seven lie between $7.4$ and $13.5$. The operator of the Vekua-type equation is therefore pinned by its own solutions, and the pinning has to be recorded because this question — which conjugation, which side — cannot be settled from the printed page: it is the same ambiguity that the two-source note at the end of the previous section raises for the multiplication side of the shifted operator.
The quartet is a basis, and a generating one in the sense of Bers. The four functions are independent over the constants, and every biquaternion-valued function $\mathcal{W}$ can be written
$$ \mathcal{W}=\sum_{k\in\{t,x,y,z\}}\xi_kF_k $$
with complex, not biquaternionic, coefficients. Those coefficients are forced: $F_1,F_2,F_3$ are purely vectorial and $F_0$ is scalar, so the scalar part of $\mathcal{W}$ is $\xi_0f$ and its $\vec e_\eta$-component is $\xi_k/f$, whence $\xi_0=\mathcal{W}_0/f$ and $\xi_k=f\,\mathcal{W}_k$ for $\eta=1,2,3$, and the expansion is unique. A solution space generated by four fixed functions, rather than described by a fundamental solution, is what the word pseudoanalytic names: this quartet is the three-dimensional counterpart of the generating pair $(F,G)$ of the plane theory, it is why the source records it "in complete analogy with the two-dimensional case" and cites Bers for the class, and it is the first layer of the theory that the Vekua paragraph above announced and left to the analysis cluster.
The equation for $\mathcal{W}$ becomes one equation for the coefficients. Substituting the expansion into the Vekua-type equation gives an exact operator identity,
$$ D\sum_{k\in\{t,x,y,z\}}\xi_kF_k-\frac{Df}{f}\,\mathcal{C}\sum_{k\in\{t,x,y,z\}}\xi_kF_k=\sum_{k\in\{t,x,y,z\}}\bigl(D\xi_k\bigr)F_\eta , $$
true for arbitrary complex coefficients and not only for solutions of the equation. Hence $\mathcal{W}$ solves the Vekua-type equation exactly when its coefficients satisfy the single biquaternionic relation
$$ \sum_{k\in\{t,x,y,z\}}\bigl(D\xi_k\bigr)F_\eta=0 , $$
which, with the quartet written out, is $(D\xi_0)f+\frac{1}{f}\sum_{\eta\in\{x,y,z\}}(D\xi_k)\vec e_\eta=0$. The reduction has exchanged one first-order equation for a biquaternion-valued function for one first-order equation for four complex scalar functions — the same equation, counted the other way — and the second form is the one in which a Vekua equation is recognised: it relates the four coefficients to their first derivatives alone, with no algebraic unknown in between.
And that condition is again a Vekua equation, now for the combination itself. Put
$$ w=\xi_0+\xi_1\vec e_x+\xi_2\vec e_y+\xi_3\vec e_z, \qquad w^{\natural}=\xi_0-\xi_1\vec e_x-\xi_2\vec e_y-\xi_3\vec e_z , $$
the biquaternion assembled from the coefficients and its quaternion conjugate. Since $w+w^{\natural}=2\xi_0$ and $w-w^{\natural}=2(\xi_1\vec e_x+\xi_2\vec e_y+\xi_3\vec e_z)$, the condition above reads $D(w+w^{\natural})f+\frac{1}{f}D(w-w^{\natural})=0$, which is equivalent to
$$ Dw=\frac{1-f^2}{1+f^2}\,Dw^{\natural} . $$
This is the source's closing remark of the section, and it is the form in which the pseudoanalytic character of the problem is visible in one line: the two halves of $w$ appear side by side, the coefficient is a single scalar built from the carrier, and every trace of $\epsilon(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ and $\mu(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ has retreated into $f$. Two interpretations of the printed display have to be separated, because the source's bar is lost in its text layer and they are different equations. Read as a conjugation of the whole expression $Dw$, the right-hand side gives an equation that is false: for $f=3$ and the explicit solution $w=(x^2-y^2)+18yz\,\vec e_x+18xz\,\vec e_y$ of the condition above one has $Dw=-16r_x\,\vec e_x+16r_y\,\vec e_y$, and since $Dw$ is purely vectorial its quaternion conjugate is $-Dw$, so the two sides of that interpretation differ by $(1+\lambda)Dw$, that is by $-3.2r_x\,\vec e_x+3.2r_y\,\vec e_y$, where $\lambda=-4/5$ is the value of the coefficient at $f=3$. Read as a conjugation of the function, so that the right-hand side is $Dw^{\natural}$, the equation is exactly equivalent to the condition above — the two differ only by the proportionality factor $2f/(1+f^2)$, with residuals below $10^{-14}$ for arbitrary coefficients. The corpus records the second interpretation, and it is the one that matches the plane prototype $w_{\bar z}=Aw+Bw^{\natural}$, in which the conjugate always conjugates the function and never the derivative.
The reduction also runs backwards, and its inverse is the classical recovery of a potential. The local theory has an inverse as well as a forward map, and the inverse is the oldest construction in vector analysis rather than a new one. The forward map is $g\mapsto F=fD(f^{-1}g)$, the carrier identity read in the other direction: it sends every solution of the scalar equation $-\Delta_\Phi g+vg=0$, with $v=\Delta_\Phi f/f$, to a solution of the first-order equation $\mathcal{D}_{\vec{\alpha}}F=0$ with $\vec{\alpha}=\mathrm{grad}\,f/f$, and it sends a fundamental solution of the scalar operator to a fundamental solution of the first-order one. The inverse is built from the operator
$$ A[G](r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)=\int_{r_{0x}}^{r_x}G_1(\zeta,r_{0y},r_{0z})\,d\zeta+\int_{r_{0y}}^{r_y}G_2(r_x,r_y,r_{0z})\,dr_y+\int_{r_{0z}}^{r_z}G_3(r_x,r_y,\eta)\,d\eta+C , $$
the line integral along the three axis-parallel legs from a fixed base point to $r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z$. For a gradient field $G=D\xi$ this integral is path-independent and returns $\xi(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)-\xi(r_{0x}\,\vec e_x+r_{0y}\,\vec e_y+r_{0z}\,\vec e_z)$, so $A$ inverts $D$ on gradients: it is the standard recovery of a potential from its gradient, checked here to $10^{-15}$ on random polynomials. Since $f^{-1}F=D(f^{-1}g)$ is exactly such a gradient, the reconstruction $g=fA[f^{-1}F]$ returns the original scalar solution up to an additive multiple of $f$, and that multiple is precisely what the forward map cannot see, because $fD(f^{-1}cf)=fD(c)=0$ for every constant $c$. The two statements together make the correspondence between the scalar problem and the first-order problem a bijection up to the span of $f$; the same pair, with $u_0$ in place of the scalar solution and $f=p^{1/2}u_0$ as the carrier, does the same for the general $(\mathrm{div}\,p\,\mathrm{grad}+q)$ problem. What the reduction costs is therefore not information but one additive constant per solution — which is the sharpest available answer to the question whether the biquaternionic reformulation loses anything.
Nothing in this section is three-dimensional. The quartet, the condition on the coefficients, the equation for $w$ and the inverse all hold with the number of vector units raised from three to $n$: with units $\vec e_1,\dots,\vec e_n$ in place of $\vec e_x,\vec e_y,\vec e_z$, each of square $-1$ and pairwise anticommuting, the operator reads $D=\sum_{j=1}^{n}\vec e_j\,\partial/\partial x_j$. The construction is dimensional rather than quaternionic: the quartet becomes a set of $n+1$ generating functions, the coefficients $\xi_k$ remain complex scalars, and the argument is unchanged. The source states this for its whole section, and it is the reason the framework is better described as a method of any dimension with a distinguished four-dimensional case than as a device of the quaternion algebra alone.
Provenance. The quartet, the condition on the coefficients, the equation for $w$, the inverse theorem and the dimensional remark are the survey's restatement of its reference $[29]$, V. V. Kravchenko, "On a factorization of second order elliptic operators and applications", Journal of Physics A: Mathematical and General 39 (2006) 12407–12425, where they are its Theorems 15–18 and Remarks 19–20. The survey's own remarks place the earlier members of the family: the Schrödinger factorisation in the Riccati form is in Bernstein (1996) and Bernstein–Gürlebeck (1999); the reduction of the biquaternionic Riccati equation to the carrier $Df/f$ is in Kravchenko–Kravchenko–Williams (2001); and the general $(\mathrm{div}\,p\,\mathrm{grad}+q)$ factorisation, with its three consequences, is $[29]$'s and not the 2001 paper's — that paper carries the Schrödinger reduction, the Riccati condition and the closed-form fundamental solution, and stops there. The general factorisation is cited to $[29]$ in the Further Reading below.
Boundary Conditions at an Interface
At a surface separating two media, the fields are matched by the integral form of Maxwell's equations. Let $\vec u$ be the unit normal pointing from medium $1$ to medium $2$. With no free surface charge and no free surface current, $$ \vec u\cdot(\vec D_2-\vec D_1)=0, \qquad \vec u\cdot(\vec B_2-\vec B_1)=0, $$ $$ \vec u\times(\vec E_2-\vec E_1)=0, \qquad \vec u\times(\vec H_2-\vec H_1)=0 . $$ With a surface charge density $\sigma_s$ and a surface current density $\vec J_{\mathrm{surf}}$ the first and last become $\vec u\cdot(\vec D_2-\vec D_1)=\sigma_s$ and $\vec u\times(\vec H_2-\vec H_1)=\vec J_{\mathrm{surf}}$. In words: the normal component of $\vec D$ and the normal component of $\vec B$ are continuous up to surface sources, and the tangential components of $\vec E$ and $\vec H$ are continuous.
The biquaternionic field strength is assembled from the local values of $\epsilon$ and $\mu$, so it is discontinuous at an interface even though the physical fields obey the conditions above. Splitting off the tangential part, with unit normal $\vec u$, $$ \vec u\times\tilde{F} =i\sqrt{\epsilon}\,\left(\vec u\times\vec E\right) -\sqrt{\mu}\,\left(\vec u\times\vec H\right). $$ The tangential combinations $\vec u\times\vec E$ and $\vec u\times\vec H$ are continuous across the interface, but the coefficients $\sqrt{\epsilon}$ and $\sqrt{\mu}$ are not, so the left-hand side jumps. Similarly, for the normal part, $$ \vec u\cdot\tilde{F} =i\sqrt{\epsilon}\,E_n-\sqrt{\mu}\,H_n =\frac{i}{\sqrt{\epsilon}}\,D_n-\frac{1}{\sqrt{\mu}}\,B_n, $$ in which $D_n$ and $B_n$ are continuous while $\sqrt{\epsilon}$ and $\sqrt{\mu}$ jump.
The jump in $\tilde{F}$ is therefore not a physical discontinuity of the fields; it is the expression of a change of the local complex structure. The same physical fields are written in the normalisation of the medium on each side, and the interface is where two local complex structures meet. No single biquaternionic continuity statement replaces the four standard conditions: the medium enters precisely as the discontinuity of the local normalisation.
For two homogeneous media the boundary conditions give Snell's law, $$ n_1\sin\theta_1=n_2\sin\theta_2, $$ and, at normal incidence, reflection and transmission coefficients determined by the impedance mismatch, $$ r=\frac{Z_2-Z_1}{Z_2+Z_1}, \qquad r'_t=\frac{2Z_2}{Z_1+Z_2}, $$ with reflectance $R=|r|^2$ and transmittance $T=(Z_1/Z_2)|r'_t|^2$ satisfying $R+T=1$. The mismatch $Z_2-Z_1$ is the quantity that controls the reflected wave, and in the variables of this article it is the mismatch between the two local normalisations $\sqrt{\mu_i/\epsilon_i}$. The general oblique-incidence Fresnel coefficients depend on the polarisation; the point for the framework is that the interface is where the local complex structure changes, and the boundary conditions are the matching data across that change.
Conventions for a Plane Wave in the Medium
This section fixes the conventions for the later exercise on plane-wave propagation in a medium. Work in a homogeneous medium with real, possibly frequency-dependent, $\epsilon(\omega)$ and $\mu(\omega)$, and use the complex-amplitude convention $$ \vec E(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z,r'_t)=\mathrm{Re}\!\left[\vec E_0\,\exp(\,i\theta_{\mathrm{pwp}})\right], $$ $$ \vec H(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z,r'_t)=\mathrm{Re}\!\left[\vec H_0\,\exp(\,i\theta_{\mathrm{pwp}})\right], $$ with complex amplitude vectors $\vec E_0,\vec H_0$ and real $\omega>0$ and $\vec k_\Phi$. This fixes the sign convention; the opposite choice $\exp(-i\theta_{\mathrm{pwp}})$ is equivalent under $\vec k_\Phi\to-\vec k_\Phi$, $\omega\to-\omega$.
Here $\theta_{\mathrm{pwp}}:=\vec k_\Phi\cdot\vec r-\omega r'_t/c\in\mathbb{R}$ is the plane-wave angle: $\omega$ and $\vec k_\Phi$ are real, so the exponent is $i$ times a real number and the wave form is $\exp(i\theta_{\mathrm{pwp}})$ wherever it is quoted.
For a source-free plane wave, $\tilde\nabla_\Phi\tilde{F}=0$ is equivalent to the four algebraic conditions $$ \vec k_\Phi\cdot\vec E_0=0, \qquad \vec k_\Phi\cdot\vec H_0=0, \qquad \vec k_\Phi\times\vec E_0=\omega\mu\,\vec H_0, \qquad \vec k_\Phi\times\vec H_0=-\omega\epsilon\,\vec E_0, $$ from which $$ k_\Phi^2=\omega^2\epsilon\mu=\frac{\omega^2}{c^2}, \qquad \vec H_0=\frac{1}{Z}\,\vec u_\Phi\times\vec E_0, \qquad Z=\sqrt{\frac{\mu}{\epsilon}}, $$ with $\vec u_\Phi=\vec k_\Phi/|\vec k_\Phi|$. In particular the amplitudes satisfy $$ |\vec E_0|=Z\,|\vec H_0|, \qquad |\vec E_0|=c\,|\vec B_0|, $$ and the field-strength amplitude is $$ \tilde{F}_0=i\sqrt{\epsilon}\,\vec E_0-\sqrt{\mu}\,\vec H_0 =i\sqrt{\epsilon}\left(\vec E_0+i\,\vec u_\Phi\times\vec E_0\right) =i\sqrt{\epsilon}\,\vec N_0, $$ with $\vec N_0=\vec E_0+ic\vec B_0=\vec E_0+i\,\vec u_\Phi\times\vec E_0$. The two normalised contributions have equal magnitude, $$ \left|\sqrt{\epsilon}\,\vec E_0\right| =\left|\sqrt{\mu}\,\vec H_0\right|, $$ and the field-strength amplitude is null, $$ \langle\tilde{F}_0\rangle_{\natural}=\sum_{\eta\in\{x,y,z\}}\left(F_{0\eta}\right)^2=0 . $$ A free plane wave in the medium is therefore a zero divisor of $\mathbb{B}$, exactly as a radiation field is in vacuum; the zero-divisor cone of the algebra is the wave cone of the medium.
The time-averaged energy density and energy flux are $$ \langle W\rangle=\frac{1}{4}\left(\epsilon\,|\vec E_0|^2+\mu\,|\vec H_0|^2\right), \qquad \langle\vec S\rangle=\frac{1}{2}\,\mathrm{Re}\!\left(\vec E_0\times\vec H_0^{*}\right), $$ and in the non-dispersive case these satisfy $|\langle\vec S\rangle|=c\,\langle W\rangle$. These are the quantities in which the medium speed and impedance appear on the same footing as the fields.
The Vacuum Limit
The vacuum is the special case $\epsilon=\epsilon_0$, $\mu=\mu_0$, $c=c_0$, $Z=Z_0$, and every formula above must reduce to the corresponding vacuum statement. The field strength becomes $$ \tilde{F}=i\sqrt{\epsilon_0}\,\vec E-\sqrt{\mu_0}\,\vec H, $$ the gradient becomes $\tilde\nabla_\Phi=e_t\,\partial/\partial R_t+\vec e_x\,\partial/\partial R_x+\vec e_y\,\partial/\partial R_y+\vec e_z\,\partial/\partial R_z$ with $\partial/\partial R_t=-i\,\partial/\partial r'_t$, and the d'Alembertian becomes $\Box_\Phi=\Delta_\Phi-\partial^2/\partial {r'_{t}}^2$, which in the clock time is $\Delta_\Phi-c_0^{-2}\partial^2/\partial {t'}^2$. The complex time coordinate becomes $i r'_t$, recovering the familiar $i r'_t$ form with the vacuum speed of light, and the local complex structure becomes constant. The invariants become $I_1=\vec E^2-c_0^2\vec B^2$ and $I_2=\vec E\cdot\vec B$, the Riemann–Silberstein vector becomes $\vec N=\vec E+ic_0\vec B$, and the plane-wave conditions become $k_\Phi^2=\omega^2/c_0^2$, $|\vec E_0|=c_0|\vec B_0|$, and $\langle\tilde{F}_0\rangle_{\natural}=0$, all as in the parent article. The impedance becomes $Z_0=\sqrt{\mu_0/\epsilon_0}$, and the refractive index becomes $n=1$.
This is the check that the medium conventions are consistent. Every medium result is the general one, and the vacuum results are recovered by setting the two medium parameters to their vacuum values. In particular, the null condition and the zero-divisor structure of the field strength are not special to vacuum; they hold in every medium, with the medium speed defining the cone. The question of which of $c_0$ and $(\epsilon_0,\mu_0)$ is fundamental — whether the vacuum has electromagnetic properties in the same sense that a medium does — is discussed in the Maxwell article and is not settled here.
Beyond the Linear Constitutive Relation
Everything above assumes a linear constitutive map. This article has already met two departures from it — the dispersive medium, where $\epsilon$ and $\mu$ depend on frequency, and the chiral medium, where the relations couple $\vec E$ and $\vec H$ — and there is a third, which the corpus should name because a deformed arena produces it. A constitutive relation can differ from $\vec D=\epsilon\vec E$, $\vec B=\mu\vec H$ in three ways, and the three behave differently.
Rescale. The parameters become functions of position or of frequency. The reduction keeps its order and its finite kernel, and the vacuum limit is recovered point by point. This is the bulk of the article and the bulk of the linear theory.
Raise the order. The chiral relations couple the two fields, and the reduced equation is no longer the same operator with new coefficients; this is the chiral section above, and it is why that case is developed separately.
Lose linearity. The displacement and the induction acquire terms quadratic in the fields. The equations keep their first order, but the theory is no longer linear, and the reduction that this article performs does not apply.
The third case is realized by the non-commutative vacuum of Maxwell's Theory on Non-Commutative Spaces and Quaternions: there the source-free equations have the same first-order form as above, with $\vec D=\vec E+\vec d$ and $\vec H=\vec B+\vec h$ where $\vec d$ and $\vec h$ are quadratic in the fields, so that the deformed vacuum behaves as a medium with non-linear properties. Two features of such a medium are general, and neither is visible in the two linear cases.
Plane waves can survive the non-linearity. In the deformed vacuum a plane wave still solves the field equations exactly. So "plane wave" does not imply a linear theory, and the use of plane waves elsewhere in the corpus as a probe of the linear regime carries that qualification. What fails is the superposition principle, not the plane-wave Ansatz.
An energy density exists only if the constitutive map is integrable. For a non-linear medium the combination $\vec E\cdot\delta\vec D+\vec H\cdot\delta\vec B$ must be an exact one-form, and when it is, the energy density is fixed by it; the deformed vacuum is such a case, and the energy density it gives is quoted in that article. A linear medium satisfies the condition trivially, which is why this article has never had to state it, and the Brillouin energy of the dispersive case is the same requirement with the frequency carried inside the constitutive map. A non-linear medium whose constitutive map is not integrable would have no energy density, which is a real restriction on which non-linear media are physical and not a technicality.
The Legacy Pauli Gamma Formulation
The Pauli and gamma formulation is legacy: the reasons are explained in the article The AXF Matrices: Axis-Free, Integer Entries, §Why the Pauli and the Gamma Displays Are Refused. The generators carry no frame label: $\gamma_\mu$ is one set of matrices, and the frame is named on the coordinate.
Every two-sided action in this article has been written out as a product, $a\,\tilde{Ξ}\,b$, with the side carried by position. The fermion articles of the corpus write the same operators as products of Dirac gammas. The two descriptions are related by a fixed dictionary and not by a change of algebra, and this section states the dictionary once so that a reader can move between them. Afterwards it is dropped: outside this section the article contains no gamma.
The gammas are a frame of $\mathrm{Cl}_{1,3}(\mathbb{C})=M_4(\mathbb{C})$, which is the algebra the basis operators already generate. One basis operator per gamma,
$$ \gamma_0=iL_{\mathrm{pau},e_t}R_{\mathrm{pau},\vec e_x},\qquad \gamma_1=iL_{\mathrm{pau},\vec e_x}R_{\mathrm{pau},\vec e_y},\qquad \gamma_2=iL_{\mathrm{pau},\vec e_y}R_{\mathrm{pau},\vec e_y},\qquad \gamma_3=iL_{\mathrm{pau},\vec e_z}R_{\mathrm{pau},\vec e_y},\qquad \gamma_5=-iL_{\mathrm{pau},e_t}R_{\mathrm{pau},\vec e_z}. $$
The same in the axis-free display:
$$ \gamma_0=iL_{\mathrm{axf},e_t}R_{\mathrm{axf},\vec e_x},\qquad \gamma_1=iL_{\mathrm{axf},\vec e_x}R_{\mathrm{axf},\vec e_y},\qquad \gamma_2=iL_{\mathrm{axf},\vec e_y}R_{\mathrm{axf},\vec e_y},\qquad \gamma_3=iL_{\mathrm{axf},\vec e_z}R_{\mathrm{axf},\vec e_y},\qquad \gamma_5=-iL_{\mathrm{axf},e_t}R_{\mathrm{axf},\vec e_z}. $$
Basis operators compose by composing their factors, $L_aR_b\,L_cR_d=L_{ac}R_{db}$, and in that product the four satisfy the basis operator relation of the Minkowski metric while the fifth closes the frame on the chirality basis operator,
$$ \{\gamma_\mu,\gamma_\nu\}=2\varepsilon_\mu\delta_{\mu\nu}I_4, \qquad \varepsilon=(1,-1,-1,-1), $$
$$ \gamma_5=i\gamma_0\gamma_1\gamma_2\gamma_3, \qquad \gamma_5^2=I_4 . $$
The product of the four is taken in the order in which the basis operators stand, and the single $i$ outside it is what makes $\gamma_5$ the chirality basis operator rather than its negative.
Basis operators are pure tensors; gammas are not. A basis operator is a single product $a\otimes b_{\mathrm{opp}}$, so the basis operators form a seven-dimensional cone inside the sixteen-dimensional algebra, and a general element is a sum of at most four of them. The least number required is the rank of the coefficient matrix $c(M)$, the matrix of $M$ in the basis of the sixteen products. A decomposition into basis operators is therefore not unique in general; it is unique only in the rank-one case, when the operator is a single basis operator. That is the reason the article prints a two-sided operator as a basis operator only when it is one, and writes the others as sums.
One-sided and two-sided, in this frame. The two Dirac equations of the section Arbitrary Inhomogeneous Media separate here into the two kinds of basis operator the algebra allows. The chiral pair acts by left multiplication only, so each of its members is a single one-sided basis operator, and the grading that separates the halves is the basis operator $\gamma_5$; this is the sense in which that equation's dividing line is chirality. Kravchenko's equation is a left basis operator together with a right multiplication in the mass term, $(\tilde\nabla_\Phi)^{\natural}$ acting from the left and $\vec e_z$ from the right, hence the two-sided basis operator $L_{(\tilde\nabla_\Phi)^{\natural}}$ with the mass term carried by $R_{\mathrm{pau},\vec e_z}$. The distinction drawn in words in the inhomogeneous section is thus, in the printing, the distinction between $L$ and $LR$.
The chirality basis operator and the central grading. $\gamma_5$ is the basis operator of the chirality operator: $-iL_{\mathrm{pau},e_t}R_{\mathrm{pau},\vec e_z}$ on the Dirac module, which acts on the algebra alone as $-iR_{\mathrm{pau},\vec e_z}$, the basis operator the chiral-fermion articles write as $\xi$. The central grading of the complexified algebra — the difference of its two central idempotents, whose square is $e_t$ — is a different object: $\gamma_5$ represents it on the module, and the two must not be identified as elements of the algebra.
Why the basis operators stay primary. The dictionary is complex-linear, and the two real forms behind it are not the same: over $\mathbb{R}$ the gamma products span $\mathrm{Cl}_{1,3}(\mathbb{R})\cong M_2(\mathbb{H})$, while the basis operators span $\mathrm{Cl}_{3,1}(\mathbb{R})\cong M_4(\mathbb{R})$. The biquaternion algebra is the complex one, so both readings are available in it, but only the basis operators are defined without a choice of frame. The article keeps the basis operators for that reason, and uses the gammas as the frame in which the fermion articles are written.
The display. Every array printed in this section is given twice: in the Peirce display first, labelled $\mathrm{pau}$, whose arrays are the ones the literature prints, and in the axis-free display after it, labelled $\mathrm{axf}$, which is the display of the body and of the corpus. The two are carried into one another by $L_{\mathrm{axf},e_\mu}=C_{\Pi}L_{\mathrm{pau},e_\mu}C_{\Pi}^{-1}$, and a conjugation by $C_{\Pi}$ changes no product, no anticommutator, no trace and no determinant, so every identity of the section holds in both. An array is a statement about one display; a basis operator named in the running text keeps the Peirce label, and its axis-free array is the conjugate. The label names the array and not the operator.
Summary
Electromagnetism in a material medium is developed here as the general case, with the vacuum as its limit. The constitutive relations $\vec D=\epsilon\vec E$ and $\vec B=\mu\vec H$ supply the two parameters of a homogeneous medium, which may be taken as the speed $c=1/\sqrt{\epsilon\mu}$ and the impedance $Z=\sqrt{\mu/\epsilon}$, or equivalently as $\epsilon=1/(cZ)$ and $\mu=Z/c$.
The medium enters the biquaternionic structure in two places and nowhere else. First, through the conversion $r'_t=c\,t'$ that sets the temporal slot, the gradient being $\partial/\partial R_t=-i\partial/\partial r'_t=-(i/c)\partial/\partial t'$, so that the map from the physical second to the imaginary scalar direction of $\mathbb{B}$ carries the local conversion $c$; this is the sense in which the complex structure is local, the algebra being fixed while its embedding in physical time is not. Second, through the normalisations $\sqrt{\epsilon},\sqrt{\mu}$ of the field strength and the corresponding $1/\sqrt{\epsilon},\sqrt{\mu}$ of the source biquaternion $$ \tilde{F}=i\sqrt{\epsilon}\,\vec E-\sqrt{\mu}\,\vec H, $$ which is fixed by the parent article and splits $\tilde{F}$ into a Hermitian electric half in $\mathbb{B}_{\Psi}$ and an anti-Hermitian magnetic half in $\mathbb{B}_{\Phi}$; the medium sets the relative weight $Z=\sqrt{\mu/\epsilon}$ of the two halves.
The single biquaternionic Maxwell equation $\tilde\nabla_\Phi\tilde{F}=-\sqrt{\mu}\,\tilde{J}_{\Phi}$ is the same in every medium: the field factors cancel through $\sqrt{\epsilon}/(c\sqrt{\mu})=\epsilon$ and $\sqrt{\mu}/(c\sqrt{\epsilon})=\mu$, and the only medium factor left explicit in the equation is $\sqrt{\mu}$. Dispersion makes the local complex structure spectral: each frequency carries its own speed $c(\omega)=1/\sqrt{\epsilon(\omega)\mu(\omega)}$, refractive index $n(\omega)=c_0/c(\omega)$, dispersion relation $k_\Phi^2=\omega^2\epsilon(\omega)\mu(\omega)$, and group velocity $v_g=c_0/(n+\omega\,dn/d\omega)$, at least while the normalisation is real. At an interface the local complex structure jumps, and the field strength is discontinuous even though the physical fields satisfy the standard tangential and normal continuity conditions. The vacuum limit $c\to c_0$, $Z\to Z_0$ recovers every object of the parent article, and the zero-divisor cone of the algebra coincides with the wave cone of the medium at every value of $c$.
For a plane wave in the medium, the conventions fixed above are: complex amplitudes in the $\exp(i\theta_{\mathrm{pwp}})$ convention, transversality $\vec k_\Phi\cdot\vec E_0=\vec k_\Phi\cdot\vec H_0=0$, the relations $\vec k_\Phi\times\vec E_0=\omega\mu\vec H_0$ and $\vec k_\Phi\times\vec H_0=-\omega\epsilon\vec E_0$, the dispersion relation $k_\Phi=\omega/c(\omega)=n(\omega)\omega/c_0$, and the impedance relation $\vec H_0=Z^{-1}\vec u_\Phi\times\vec E_0$. In these conventions $|\vec E_0|=Z|\vec H_0|=c|\vec B_0|$, the two normalised halves of $\tilde{F}_0$ have equal magnitude, and $\langle\tilde{F}_0\rangle_{\natural}=0$.
A chiral medium is the one constitutive case recorded here that changes the order of the reduction rather than only its parameters. With the Drude–Born–Fedorov relations $\vec D=\epsilon(\vec E+\beta\,\mathrm{rot}\,\vec E)$ and $\vec B=\mu(\vec H+\beta\,\mathrm{rot}\,\vec H)$ the time-dependent Maxwell system still reduces to a single quaternionic equation, but its operator $\mathcal{M}=\beta\sqrt{\epsilon\mu}\,\partial/\partial r'_t\,D+\sqrt{\epsilon\mu}\,\partial/\partial r'_t-iD$ is of second order, its wave equations are of fourth order, and its frequency symbol factors as $i(\beta\sqrt{\epsilon\mu}\,\omega-1)(D+\alpha(\omega))$ with a frequency-dependent $\alpha$ that degenerates at $\omega=c/\beta$. The non-chiral reduction $\sqrt{\epsilon\mu}\,\partial/\partial r'_t-iD$ is the $\beta=0$ member of that family. The same fourth-order equation carries the medium's birefringence, with the two refractive indices $n_\pm(\omega)=1/(1\mp\beta\omega/c)$; and the fundamental solution of $\mathcal{M}$ is causal by construction, built from the retarded kernel $V_{\mathrm{vek},\alpha}$ with a displaced pole in the upper half-frequency plane.
A third kind of constitutive law is not a case of the reduction at all. If $\vec D$ and $\vec H$ acquire terms quadratic in the fields, the equations keep their first order but the theory becomes non-linear: superposition fails, an energy density exists only when the constitutive one-form $\vec E\cdot\delta\vec D+\vec H\cdot\delta\vec B$ is integrable, and the reduction performed in this article does not apply — even though a plane wave can still be an exact solution. The non-commutative vacuum of Maxwell's Theory on Non-Commutative Spaces and Quaternions is the corpus's worked example, and the section Beyond the Linear Constitutive Relation states the general situation.
The time-harmonic companion of the chiral reduction — the diagonalization into the two circular combinations with the two wavenumbers, the integral representations, the extendability problem, the energy balance and the inhomogeneous chiral medium — is not repeated here; it is carried by the dedicated article Maxwell's Equations in Chiral Media — The Quaternionic Reformulation.
Beyond the chiral case, inhomogeneity is now handled in two ways that are not the same kind of result. For a medium with arbitrary $\epsilon(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ and $\mu(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ and no small parameter, the Maxwell system has an exact quaternionic equivalent, $$ D\,\vec{\mathcal E}+\vec{\mathcal E}\,\vec{\epsilon} =-\frac{1}{c}\,\partial/\partial r'_t\,\vec{\mathcal H}-\frac{\rho}{\sqrt{\epsilon}}, \qquad D\,\vec{\mathcal H}+\vec{\mathcal H}\,\vec{\mu} =\frac{1}{c}\,\partial/\partial r'_t\,\vec{\mathcal E}+\sqrt{\mu}\,\vec J, $$ with the normalised fields $\vec{\mathcal E}=\sqrt{\epsilon}\,\vec E$, $\vec{\mathcal H}=\sqrt{\mu}\,\vec H$ and the half-gradients $\vec{\epsilon}=\mathrm{grad}\sqrt{\epsilon}/\sqrt{\epsilon}$, $\vec{\mu}=\mathrm{grad}\sqrt{\mu}/\sqrt{\mu}$; and in one line, $$ \left(\frac{1}{c}\,\partial/\partial r'_t+iD\right)\mathcal{F}-\mathcal{F}\,(i\vec{c})-\bar{\mathcal{F}}\,(i\vec{Z}) =-\left(\sqrt{\mu}\,\vec J+\frac{i\rho}{\sqrt{\epsilon}}\right), \qquad \mathcal{F}=\sqrt{\epsilon}\,\vec E+i\sqrt{\mu}\,\vec H, $$ which is a Vekua equation — a first-order equation whose coefficients act on both $\mathcal{F}$ and its conjugate — and therefore the governing equation of the pseudoanalytic functions. The entire reduction rests on two identities: the carrier identity $\xi^{-1}D(\xi g)=Dg+\vec{\alpha}\,g$ and $\xi^{-1}D(g\xi)=Dg+\vec{\alpha}\,g$, which makes a coefficient of the form $\mathrm{grad}\,\xi/\xi$ removable by conjugation, from the left or from the right, and the scalar-product identity $\langle\vec p,\vec q\rangle=-\tfrac12(\vec p\vec q+\vec q\vec p)$, which converts the scalar products of the divergence equations into products the carrier can absorb. The carriers of the divergence equations are $\sqrt{\epsilon}$ and $\sqrt{\mu}$, which is why those square roots appear in the normalisation of the field strength and not only for dimensional reasons. The price is that this is a reduction to an equation and not to a solution: the time-dependent inhomogeneous problem in closed form is not part of the framework, and the closed form given below reaches only the stationary one-dimensional case.
The one instance of the reduced problem that is completely solvable is the operator $\mathcal{D}_{\vec{\alpha}}$ with the gradient coefficient $\vec{\alpha}=\mathrm{grad}\,\xi/\xi$, and it is solvable by reduction to a scalar Schrödinger equation: with $v=\Delta_\Phi\xi/\xi$, every solution $\tilde{Ξ}$ of $-\Delta_\Phi\tilde{Ξ}+v\tilde{Ξ}=0$ produces a solution $\vec{F}=D\tilde{Ξ}-\vec{\alpha}\tilde{Ξ}$ of $\mathcal{D}_{\vec{\alpha}}\vec{F}=0$, and a fundamental solution of $-\Delta_\Phi+v$ produces a fundamental solution of $\mathcal{D}_{\vec{\alpha}}$. The operator identity behind this is the factorisation $\mathcal{D}_{\vec{\alpha}}(Du-\vec{\alpha}u)=(-\Delta_\Phi+v)u$ for scalar $u$, valid whenever $D\vec{\alpha}+\vec{\alpha}^2=-v$, which is the quaternionic generalisation of the Riccati equation. When $\Delta_\Phi\xi/\xi=-k_\Phi^2$ is constant the fundamental solution is explicit, $$ \vec{F}(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)=-\left(\vec{\alpha}+\frac{r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z}{r^{2}}-\frac{ik_\Phi\,(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)}{r}\right)\frac{\exp(ikr)}{4\pi r}, $$ and this is built by the same constructive recipe as the causal kernel of the chiral problem — minus the first-order operator applied to the outgoing Helmholtz kernel — with the vector gradient coefficient $\mathrm{grad}\,\xi/\xi$ standing where the chiral problem has the frequency-dependent scalar shift $\alpha(\omega)$. The two kernels therefore share their shape and not their value: the chiral one is scalar-plus-vector, this one is a pure vector, and they differ by an $O(1)$ amount that vanishes only in the degenerate unshifted case, where both are the classical Cauchy kernel. The multiplication side is not cosmetic here either: the factorisation holds with the coefficient on the right, and the left-handed version leaves a first-order remainder.
The same factorisation is not special to the Schrödinger operator. For scalar complex $p,q$ and a nonvanishing particular solution $u_0$ of $(\mathrm{div}\,p\,\mathrm{grad}+q)u=0$, with the carrier $f=p^{1/2}u_0$, the identity $(\mathrm{div}\,p\,\mathrm{grad}+q)\xi=-p^{1/2}\mathcal{D}_{\vec{\alpha}_f}(D(p^{1/2}\xi)-\vec{\alpha}_f\,p^{1/2}\xi)$ holds for every scalar $\xi$, $\vec{\alpha}_f=\mathrm{grad}\,f/f$. The Schrödinger operator is the case $p=1$, and the conductivity equation $\mathrm{div}\,\sigma\,\mathrm{grad}\,u=0$ is the case $p=\sigma$, $q=0$ — so the conduction problem is factorised by the very same operators that reformulate Maxwell. One solution $\mathcal{W}$ of the Vekua-type equation $D\,\mathcal{W}-\tfrac{Df}{f}\,\mathcal{C}\,\mathcal{W}=0$ carries three equations at once: its scalar part solves the stationary Schrödinger equation with potential $\Delta_\Phi f/f$, the function $f^{-1}\mathcal{W}_0$ solves the conductivity equation, and the rescaled vector part $\vec v=f\vec{\mathcal W}$ solves $\mathrm{rot}(f^{-2}\mathrm{rot}\,\vec v)=0$. This is the survey's one algebra, many equations statement, and it is the reason the conduction operator and the Schrödinger operator are not separate reformulations but two interpretations of the same factorisation.
That factorisation is a statement about operators, and in one dimension it can be completed from a statement about operators into a formula. If the auxiliary equation $\frac{d}{dx}\left(p\frac{dg_0}{dx}\right)+qg_0=0$ has a particular solution $g_0$ that is bounded on the interval together with $1/g_0$, then with $g=\sqrt{p}\,g_0$ the general solution of $\frac{d}{dx}\left(p\frac{du}{dx}\right)+qu=\omega^2p\,u$ is $u=c_1u_1+c_2u_2$, where $u_1$ and $u_2$ are the even and odd parts in $\omega$ of a series normalised by $g_0$ whose coefficients are generated by two interleaved integral recursions, with the initial values $u_1(0)=g_0(0)$, $u_2'(0)=\omega/(p(0)g_0(0))$, and Wronskian $\omega/p(0)$. The carrier $g=\sqrt{p}\,g_0$ of the series is term by term the carrier $f=p^{1/2}u_0$ of the factorisation, and what the series adds to it is the requirement that the reciprocal be bounded, which is what makes the coefficients computable across the whole interval. The series is a power series in $\omega$, so the solution is delivered for all frequencies at once and a spectral problem becomes root-finding for a polynomial; the paper reports the truncated series more accurate than Matlab's adaptive solver on three test problems.
The equation that the reduction produces has a solution theory of its own, and it is not a theory of one fundamental solution. Four generating solutions, $F_0=f$ and $F_\eta=\vec e_\eta/f$ for $\eta=1,2,3$, solve the Vekua-type equation — for one definite interpretation of the conjugation $\mathcal{C}$ and of the side on which the coefficient multiplies, neither of which the printed page fixes and both of which the numerics do — and every biquaternion-valued function is a unique combination $\mathcal{W}=\sum_\eta\xi_kF_k$ with complex coefficients, so the solution space is generated by four fixed functions rather than described by a kernel. Substituting that expansion turns the equation for $\mathcal{W}$ into the single condition $\sum_\eta(D\xi_k)F_\eta=0$ on the coefficients, which is in turn equivalent to the Vekua equation $Dw=\frac{1-f^2}{1+f^2}Dw^{\natural}$ for their combination $w=\xi_0+\xi_1\vec e_x+\xi_2\vec e_y+\xi_3\vec e_z$ — the operator relation of the factorisation, now written for the solution itself. The reduction also has an inverse: the operator $A[G]$ of the axis-parallel line integral recovers a potential from its gradient, so that the correspondence between the scalar problem and the first-order problem is a bijection up to the additive multiples of the carrier $f$, the only thing the forward map cannot see. The construction is dimensional rather than quaternionic, and holds with the three vector units replaced by $n$.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $\mathbb{B}$ | Biquaternion algebra |
| $e_t=1,\vec e_x,\vec e_y,\vec e_z$ | Quaternion basis, $\vec e_\eta^2=-e_t$ |
| $i$ | Scalar imaginary, $i^2=-1$ |
| $\mathbb{B}_{C},\mathbb{B}_{H}$ | Scalar subspace, real-quaternion subspace |
| $\mathbb{B}_{\Psi},\mathbb{B}_{\Phi}$ | Hermitian and anti-Hermitian subspaces |
| $\tilde\nabla_\Phi,(\tilde\nabla_\Phi)^{\natural}$ | Biquaternionic gradient and its quaternion conjugate |
| $\Box_\Phi=\tilde\nabla_\Phi(\tilde\nabla_\Phi)^{\natural}=\Delta_\Phi-\partial^2/\partial {r'_{t}}^2$ | d'Alembertian in the medium |
| $\epsilon,\mu$ | Permittivity and permeability of the medium |
| $\epsilon_0,\mu_0$ | Vacuum permittivity and permeability |
| $\vec D=\epsilon\vec E,\ \vec B=\mu\vec H$ | Constitutive relations |
| $c=1/\sqrt{\epsilon\mu}$ | Speed of light in the medium |
| $c_0=1/\sqrt{\epsilon_0\mu_0}$ | Vacuum speed of light |
| $Z=\sqrt{\mu/\epsilon}$ | Wave impedance of the medium; $Z_0=\sqrt{\mu_0/\epsilon_0}\approx376.73\,\Omega$ |
| $n(\omega)=c_0/c(\omega)$ | Refractive index |
| $\tilde{F}=i\sqrt{\epsilon}\,\vec E-\sqrt{\mu}\,\vec H=i\sqrt{\epsilon}\,\vec N$ | Field-strength biquaternion (pure vector) |
| $\tilde{A}_{\Phi}=i\,a'_t+\vec A$ | Potential biquaternion |
| $\tilde J_{\Phi}=ic\rho\,e_t+\vec J$ | Four-current, of the potential equation; the Maxwell source is $\sqrt{\mu}\,\tilde J_{\Phi}=i\rho/\sqrt{\epsilon}+\sqrt{\mu}\,\vec J$ |
| $\vec N=\vec E+ic\vec B$ | Riemann–Silberstein vector |
| $I_1=\vec E^2-c^2\vec B^2,\ I_2=\vec E\cdot\vec B$ | Lorentz invariants of the field |
| $\langle\tilde{F}\rangle_{\natural} = \langle\tilde{F},\tilde{F}\rangle_{\natural}=\tilde{F}\tilde{F}^{\natural}$ | Biquaternion norm (complex scalar) |
| $W=\tfrac{1}{2}(\epsilon\vec E^2+\mu\vec H^2)$ | Electromagnetic energy density |
| $\vec S=\vec E\times\vec H$ | Poynting vector |
| $\vec k_\Phi,k_\Phi,\omega$ | Wavevector, its modulus and the angular frequency; $k_\Phi=\omega/c(\omega)$ is the wave number in the medium |
| $k_\Phi^{\tau}=k_\Phi/(1-\tau k_\Phi\beta)$, $\tau=\pm1$ | The two chiral wave numbers, one per circular polarisation |
| $\vec u_\eta=\vec k_\Phi/\lvert\vec k_\Phi\rvert$ | Unit wavevector |
| $\beta$ | Chirality measure of a Drude–Born–Fedorov medium (a length) |
| $D=\vec e_x\,\partial/\partial R_x+\vec e_y\,\partial/\partial R_y+\vec e_z\,\partial/\partial R_z$ | Moisil–Teodoresco operator, $D^2=-\Delta_\Phi$, the spatial part of $\tilde\nabla_\Phi$; used in every section from the chiral one onward |
| $\mathcal{D}_{\vec{\alpha}}$ | Operator $g\mapsto Dg+g\,\vec{\alpha}$, the coefficient multiplying on the right |
| $\mathcal{M}=\beta\sqrt{\epsilon\mu}\,\partial/\partial r'_t\,D+\sqrt{\epsilon\mu}\,\partial/\partial r'_t-iD$ | Quaternionic operator of the chiral Maxwell system |
| $\vec{\mathcal N}=\vec E-i\sqrt{\mu/\epsilon}\,\vec H$ | Purely vectorial field of the chiral reduction |
| $\alpha(\omega)=\sqrt{\epsilon\mu}\,\omega/(\beta\sqrt{\epsilon\mu}\,\omega-1)$ | Frequency-dependent Moisil–Teodoresco parameter of $\mathcal{M}(\omega)$ |
| $\omega=c/\beta$ | Degenerate frequency of the chiral reduction, where $\mathcal{M}(\omega)=i/\beta$ and the reduction becomes algebraic |
| $\mathrm{Sd}_\alpha,V_{\mathrm{vek},\alpha}$ | Helmholtz kernel and Moisil–Teodoresco fundamental solution |
| $\mathrm{Sd}_{k_\Phi}$ | Shifted Helmholtz kernel of the gradient-coefficient operator |
| $\kappa(\vec r)=\lvert\vec r\rvert/(\beta^2\sqrt{\epsilon\mu})$ | Coefficient of the essential singularity of $\mathcal{I}_1,\mathcal{I}_2$, $\vec r=r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z$ |
| $n_\tau(\omega)=1/(1-\tau\beta\omega/c),\ \tau=\pm1$ | The two refractive indices of the chiral medium |
| $\epsilon(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z),\mu(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)$ | Inhomogeneous permittivity and permeability |
| $\vec{\epsilon}=\tfrac12\mathrm{grad}\,\epsilon/\epsilon,\ \vec{\mu}=\tfrac12\mathrm{grad}\,\mu/\mu$ | Half-gradients of the two medium parameters |
| $\ell_\epsilon=\mathrm{grad}\,\epsilon/\epsilon=2\vec{\epsilon}$, $\ell_\mu$ likewise | Full gradients of the two medium parameters, twice the half-gradients |
| $\vec{\mathcal E}=\sqrt{\epsilon}\,\vec E,\ \vec{\mathcal H}=\sqrt{\mu}\,\vec H$ | Fields normalised by the carrier functions |
| $\mathcal{F}=\vec{\mathcal E}+i\vec{\mathcal H}$ | Field of the Vekua equation; $\tilde{F}=i\mathcal{F}$ |
| $\bar{\mathcal{F}}$ | Coefficientwise complex conjugation of $\mathcal{F}$; the corpus's ${}^{*}$ is the Hermitian conjugation, a different involution |
| $\vec{c}=\tfrac12\mathrm{grad}\ln c,\ \vec{Z}=\tfrac12\mathrm{grad}\ln Z$ | Coefficient vectors of the Vekua equation (the source's $\vec{c}$ and $\vec{W}$) |
| $\xi$ | Carrier function of the identity $\xi^{-1}D(\xi g)=Dg+\vec{\alpha}\,g$ and of its reading $\xi^{-1}D(g\xi)=Dg+\vec{\alpha}\,g$ |
| $\vec{\alpha}=\mathrm{grad}\,\xi/\xi$ | Gradient coefficient of the operators $\mathcal{D}_{\vec{\alpha}}$ and $g\mapsto Dg+\vec{\alpha}g$ |
| $\mathcal{W}=\mathcal{W}_0+\vec{\mathcal{W}}$ | Biquaternion-valued solution of the Vekua-type equation; not the energy density $W$ above |
| $v=\Delta_\Phi\xi/\xi,\ \tilde{Ξ}$ | Schrödinger potential and solution, $-\Delta_\Phi\tilde{Ξ}+v\tilde{Ξ}=0$ |
| $p,\ q,\ u_0$ | Coefficients of the general operator $\mathrm{div}\,p\,\mathrm{grad}+q$ and a nonvanishing particular solution |
| $f=p^{1/2}u_0$ | Carrier function throughout; $f=\sqrt{\sigma}$ for the conductivity equation. The letter $f$ is reserved for the carrier, and the field of the Vekua equation is $\mathcal{F}$ |
| $g_0,\ g=\sqrt{p}\,g_0$ | Bounded particular solution of the auxiliary equation $\frac{d}{dx}\left(p\frac{dg_0}{dx}\right)+qg_0=0$ and the carrier of the one-dimensional series; $g_0$ and $1/g_0$ bounded |
| $X^{(n)}_{+},X^{(n)}_{-}$ | Coefficients of the two series of the one-dimensional general solution, generated by the interleaved recursions from $X^{(0)}_{+}=X^{(0)}_{-}=1$ |
| $\sigma=f^{2}$ | Conductivity of the conduction problem reached by the same factorisation, so that its carrier is $\sqrt{\sigma}$ |
| $\mathcal{C}$ | Quaternion conjugation in the Vekua-type equation $D\mathcal{W}-\tfrac{Df}{f}\mathcal{C}\mathcal{W}=0$, with the coefficient on the left: scalar part kept, vector part negated (the complex conjugation and the composite do not solve it) |
| $F_0=f,\ F_\eta=\vec e_\eta/f$ | Generating quartet of the Vekua-type equation, $\eta=1,2,3$ |
| $\xi_k,\ w=\xi_0+\xi_1\vec e_x+\xi_2\vec e_y+\xi_3\vec e_z$ | Complex coefficients of the unique expansion $\mathcal{W}=\sum_\eta\xi_kF_k$, and the biquaternion they assemble |
| $A[G]$ | Line integral along the three axis-parallel legs from a base point, the inverse of $D$ on gradient fields |
| $n$ vector units | Setting of the generating-quartet results, with $\vec e_x,\vec e_y,\vec e_z$ replaced by $\vec e_1,\dots,\vec e_n$ |
Further Reading
- The AXF Matrices: Axis-Free, Integer Entries, for the reasons the AXF matrices replace the Pauli and gamma matrices.
- J. D. Jackson, Classical Electrodynamics (Wiley, 1999), for Maxwell's equations in media, dispersion, and the standard boundary conditions.
- L. D. Landau and E. M. Lifshitz, Electrodynamics of Continuous Media (Pergamon, 1984), for the constitutive relations, the Brillouin energy density in dispersive media, and the boundary conditions at an interface.
- M. Born and E. Wolf, Principles of Optics (Cambridge, 1999), for dispersion, phase and group velocity, and the Fresnel coefficients.
- L. D. Landau and E. M. Lifshitz, The Classical Theory of Fields (Pergamon, 1975), for the invariant classification of the electromagnetic field.
- Ludwik_\Phi Silberstein, "Elektromagnetische Grundgleichungen in bivektorieller Behandlung", Annalen der Physik_\Phi 22 (1907) 579–586, for the original complex-vector formulation.
- Iwo Białynicki-Birula and Zofia Białynicka-Birula, "The role of the Riemann–Silberstein vector in classical and quantum theories of electromagnetism", Journal of Physics A 46 (2013) 053001, for the modern account of the complex vector.
- A. Waser, "Application of Bi-Quaternions in Physics" (2000, updated 2007), for the biquaternionic treatment of the field and its energy–momentum.
- L. A. Alexeyeva, "Maxwell Equations, Their Hamiltonian and Biquaternionic Forms and Properties of Their Solutions" (2016), for the biquaternionic formulation of the field equations in a medium.
- S. M. Grudsky, K. V. Khmelnytskaya and V. V. Kravchenko, "On a quaternionic Maxwell equation for the time-dependent electromagnetic field in a chiral medium", Journal of Physics A: Mathematical and General 37 (2004) 4641–4647, DOI 10.1088/0305-4470/37/16/013 (preprint arXiv:math-ph/0309062), for the Drude–Born–Fedorov chiral constitutive relations, the reduction of the time-dependent chiral Maxwell system to the single quaternionic equation $\mathcal{M}\vec{\mathcal N}=\sqrt{\mu}\,\tilde J_{\Phi}$, the exact factorization of $\mathcal{M}(\omega)$ with its degenerate frequency $c/\beta$, the two refractive indices, the kernel $V_{\mathrm{vek},\alpha}$ of $D_\alpha=D+\alpha$ with its defining equation $D_\alpha V_{\mathrm{vek},\alpha}=\delta$ (note the source's $\mathrm{Sd}_\alpha=-\exp(i\alpha|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|)/(4\pi|r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z|)$), and the causal fundamental solution, whose residue series resums to Bessel functions $J_0$ and $J_1$ of $2\sqrt{\kappa(r_x\,\vec e_x+r_y\,\vec e_y+r_z\,\vec e_z)r'_t}$. This is the paper recorded in the section A Chiral Medium: One Quaternionic Equation.
- V. V. Kravchenko and H. Oviedo, "On a Quaternionic Reformulation of Maxwell's Equations for Chiral Media and its Applications", Zeitschrift für Analysis und ihre Anwendungen (Journal for Analysis and its Applications) 22 (2003), no. 3, 569–589, DOI 10.4171/ZAA/1163, for the time-harmonic reformulation, the diagonalization with the two wavenumbers $\alpha_1,\alpha_2$, the integral representations, the extendability problem and its complete solution, and the treatment of inhomogeneous (slowly varying, stratified) chiral media. This is the source of the dedicated article Maxwell's Equations in Chiral Media — The Quaternionic Reformulation, where the reformulation and its applications are recorded.
- V. V. Kravchenko, "Quaternionic Reformulation of Maxwell Equations for Inhomogeneous Media and New Solutions", Zeitschrift für Analysis und ihre Anwendungen (Journal for Analysis and its Applications) (2001), DOI 10.4171/ZAA/1063 (preprint arXiv:math-ph/0104008), for the time-harmonic reformulation of the Maxwell system of an arbitrary inhomogeneous medium, the first-order operator that adds the right product with the gradient coefficient $\vec{\alpha}=\mathrm{grad}\,\xi/\xi$, the reduction to the Schrödinger equation $-\Delta_\Phi\tilde{Ξ}+(\Delta_\Phi\xi/\xi)\tilde{Ξ}=0$, the factorisation with the quaternionic Riccati condition $D\vec{\alpha}+\vec{\alpha}^2=-v$, and the closed-form fundamental solution. This is the source of the section The Operator $\mathcal{D}_{\vec{\alpha}}$, the Reduction to the Schrödinger Equation, and a Fundamental Solution. Its factorisation is the Schrödinger member ($p=1$) of the general family: the paper carries neither the general $(\mathrm{div}\,p\,\mathrm{grad}+q)$ factorisation, nor the conductivity equation, nor the inverse theorems, and those are cited below to the survey's reference $[29]$. The paper's own attribution of the Riccati form is to Bernstein (1996) and Bernstein–Gürlebeck (1999), and it names no conductivity problem.
- V. V. Kravchenko, "Quaternionic equation for electromagnetic fields in inhomogeneous media" (2002), DOI 10.1142/9789812794253_0042 (preprint arXiv:math-ph/0202010), for the carrier identity, with the scalar attached on either side of the product, the scalar-product identity, the exact reduction of the Maxwell system of an arbitrary inhomogeneous medium to the two gradient-corrected equations and then to a single Vekua-type quaternionic equation, and the identification of that equation with the governing equation of the pseudoanalytic functions. This is the source of the section Arbitrary Inhomogeneous Media: The Carrier Function and the Vekua Equation.
- V. V. Kravchenko, "On a general solution of the one-dimensional stationary Schrödinger equation", arXiv:0708.2491v2 [math-ph] (2007), for the general solution of $\frac{d}{dx}\left(p\frac{du}{dx}\right)+qu=\omega^2p\,u$ generated by one particular solution $g_0$ of the auxiliary equation $\frac{d}{dx}\left(p\frac{dg_0}{dx}\right)+qg_0=0$ that is bounded together with $1/g_0$: the two interleaved recursions for the coefficients $X^{(n)}_{+},X^{(n)}_{-}$, the normalisation $g=\sqrt{pg_0}$, the unit Wronskian at the origin, the equivalent statement for $p=1$, and the numerical comparison with Matlab's
ode45. This is the source of the paragraphs The one-dimensional general solution, The carrier of the series is the carrier of the factorisation and Why $\omega$ is the right variable. The paper attributes its Theorem 1 to the companion paper it cites as its reference $[3]$, where the reduction of $\frac{d}{dx}\left(p\frac{du}{dx}\right)+qu=\omega^2u$ to the auxiliary equation is proved by pseudoanalytic-function methods; the series solution itself is older, going back to Weyl's 1910 paper and to the representation used in inverse spectral theory. - V. V. Kravchenko, "On a factorization of second order elliptic operators and applications", Journal of Physics A: Mathematical and General 39 (2006) 12407–12425, the primary source of the survey's operator-relation theorems, where it is its reference $[29]$: the general factorisation $(\mathrm{div}\,p\,\mathrm{grad}+q)\xi=-p^{1/2}\mathcal{D}_{\vec{\alpha}_f}(D(p^{1/2}\xi)-\vec{\alpha}_f\,p^{1/2}\xi)$ with $f=p^{1/2}u_0$, its three consequences for the Schrödinger, conductivity and curl equations, the generating quartet $F_0=f$, $F_\eta=\vec e_\eta/f$ of the Vekua-type equation with the condition $\sum_\eta(D\xi_k)F_\eta=0$ on the coefficients and the equivalent equation $Dw=(1-f^2)/(1+f^2)\,Dw^{\natural}$ for their combination, the inverse reconstruction $g=fA[f^{-1}F]$ through the line integral $A[G]$, and the extension of the whole section from three vector units to $n$. This is the source of the operator-relation web recorded in The Operator $\mathcal{D}_{\vec{\alpha}}$, the Reduction to the Schrödinger Equation, and a Fundamental Solution and of the section The Generating Quartet, the Pseudoanalytic Equation, and the Inverse of the Reduction; the survey's numbering, from which the material above was taken, is its Theorems 15–18 and its Remarks 19–20.
- L. Bers, Theory of Pseudo-Analytic Functions (New York University, 1952), the classical source of the $(F,G)$-pseudoanalytic functions and of the Vekua-type system with variable coefficients that the section Arbitrary Inhomogeneous Media identifies as the governing equation of Maxwell's system in an arbitrary medium, and the theory in which the one-dimensional series solution above is derived. The corpus carries the first layer of that theory — the generating quartet, the condition on the coefficients and the inverse of the reduction, in the section The Generating Quartet, the Pseudoanalytic Equation, and the Inverse of the Reduction — while the theory's own apparatus, its integral representations, its similarity principle and its Liouville-type theorems, is what the corpus does not yet develop; this is its primary reference.
- K. V. Khmelnytskaya and V. V. Kravchenko, "Biquaternions for analytic and numerical solution of equations of electrodynamics", arXiv:0902.3490v1 [math-ph] (2009), for the survey of the programme that this article and Maxwell's Equations in Chiral Media — The Quaternionic Reformulation draw on: constant and variable coefficients, integral representations of solutions, the relations between the Maxwell, Schrödinger and conductivity operators, and a collocation numerical method built on the fundamental solutions. The survey is the "start here" pointer to the applied Kravchenko series; it is recorded here as the overview, it is the source through which the operator-relation theorems of $[29]$ and the generating-quartet material of The Generating Quartet, the Pseudoanalytic Equation, and the Inverse of the Reduction enter this article, and it is cited for the numerical method in the chiral article — the theorems themselves being cited to their primary papers.
- C. Doran and A. Lasenby, Geometric Algebra for Physicists (Cambridge, 2003), for the rotor formulation of the field strength and its Lorentz transformations.