Dirac Fields and Horizon Flux

A Dirac field couples to geometry through a tetrad and its spin connection. A singular static tetrad can make smooth spinor components appear divergent at a black-hole horizon, so the physical boundary condition must be checked in a regular frame. For an electrically neutral classical spinor on a fixed Schwarzschild or subextremal Kerr exterior, the conserved Dirac current is future causal and has nonnegative inward horizon flux. This forbids reflected-number amplification; for positive-frequency stationary scattering it also forbids energy amplification. The argument does not require the Dirac stress tensor to have a positive local energy density, and it is not a quantum-statistics argument.

Required background. PDEs and Green Functions for Gravitation supplies first-order hyperbolic systems, characteristics, boundary conditions, and conserved-current arguments.

Helpful background. Black-Hole Geometry and Mechanics supplies horizon generators and stationary energy; Formulations of General Relativity supplies frame formulations and boundary flux; Frames, Coframes, and Anholonomy supplies local frames and their gauge freedom.

Dirac operator and spin connection

Let $\mu_D\ge0$ be the field’s real mass, and choose a time-oriented orthonormal tetrad $e_a{}^\mu$. We use the Clifford convention $\{\gamma^a,\gamma^b\}=2\eta^{ab}I$ and $\gamma^\mu=e_a{}^\mu\gamma^a$. In a compatible representation, $(\gamma^0)^\dagger=-\gamma^0$ and $(\gamma^i)^\dagger=\gamma^i$. With $\omega_{\mu ab}=e_{a\nu}\nabla_\mu e_b{}^\nu$, the equation and connection are

$$D_\mu\psi=\partial_\mu\psi +\frac14\omega_{\mu ab}\gamma^a\gamma^b\psi, \qquad (\gamma^\mu D_\mu-\mu_D)\psi=0.$$

The absence of $i$ in front of this Dirac operator is consequential. In a flat Cartesian tetrad,

$$(\gamma^\mu\partial_\mu+\mu_D) (\gamma^\nu\partial_\nu-\mu_D) =\Box-\mu_D^2, \qquad \omega^2=|\mathbf p|^2+\mu_D^2.$$

The last relation follows on a plane wave $e^{-i\omega t+i\mathbf p\cdot\mathbf x}$. With the same Clifford algebra, inserting $i$ into the first-order equation instead gives the wrong massive dispersion. A common alternative uses matrices $\Gamma^a$ satisfying $\{\Gamma^a,\Gamma^b\}=-2\eta^{ab}I$ and writes $(i\Gamma^\mu D_\mu-\mu_D)\psi=0$. The conversion is $\gamma^a=i\Gamma^a$, with the corresponding conversion of the connection and adjoint. Röken uses the opposite metric signature and this latter massive equation Röken 2019, v6, Appendix B, p. 18, PDF.

Write $\Omega_\mu=\frac14\omega_{\mu ab}\gamma^a\gamma^b$. A change of spin frame is described locally by $\psi'=S\psi$ and $\gamma'^\mu=S\gamma^\mu S^{-1}$. Its connection must transform as

$$\Omega'_\mu=S\Omega_\mu S^{-1}-(\partial_\mu S)S^{-1}, \qquad D'_\mu\psi'=S D_\mu\psi.$$

Substituting the first relation into $D'_\mu(S\psi)$ cancels the two terms containing $\partial_\mu S$. That is why changing the tetrad while leaving the spinor representative fixed is not a valid comparison. The connection also ensures compatibility with Clifford multiplication: the covariant derivative of $\gamma^\mu$ vanishes when both its vector and spinor indices are included.

Conserved norm and causal current

Fix the adjoint phase explicitly by

$$\bar\psi=-\psi^\dagger\gamma^0,\qquad j^\mu=\bar\psi\gamma^\mu\psi,\qquad \alpha^i=-\gamma^0\gamma^i.$$

Then $j^{\hat0}=\psi^\dagger\psi$ and $j^{\hat i}=\psi^\dagger\alpha^i\psi$ in an orthonormal frame. For any unit spatial vector $\mathbf u$, the matrix $\mathbf u\cdot\boldsymbol\alpha$ is Hermitian and squares to $I$. Its eigenvalues are therefore $\pm1$, so $|\mathbf u\cdot\mathbf j|\le j^{\hat0}$. Maximizing over $\mathbf u$ gives $|\mathbf j|\le j^{\hat0}$: the current is future timelike or null wherever the spinor is nonzero.

The adjoint equation and its covariant derivative are

$$(D_\mu\bar\psi)\gamma^\mu+\mu_D\bar\psi=0, \qquad D_\mu\bar\psi=\partial_\mu\bar\psi-\bar\psi\Omega_\mu.$$

They give

$$\nabla_\mu j^\mu =(D_\mu\bar\psi)\gamma^\mu\psi +\bar\psi\gamma^\mu D_\mu\psi =-\mu_D\bar\psi\psi+\mu_D\bar\psi\psi=0.$$

For a spacelike slice with future unit normal $n_\Sigma$, the positive Dirac norm is $N_\Sigma=-\int_\Sigma j^\mu n_{\Sigma\mu}\,d\Sigma$. Calling this a number or probability current describes a conserved classical norm; it does not choose a quantum vacuum or assign particles to arbitrary modes. Both conservation and causal orientation are local statements independent of separability.

As another check, in flat space set $\beta=-i\gamma^0$. Multiplying the equation by $-i\gamma^0$ gives

$$i\partial_t\psi =\left(-i\alpha^i\partial_i+\mu_D\beta\right)\psi.$$

The matrices $\alpha^i$ and $\beta$ are Hermitian. This verifies the local Hamiltonian algebra; self-adjointness of an operator on a curved exterior additionally requires a domain and boundary analysis.

A frame regular at the future horizon

Put $f=1-2GM/r$ and $dr_*/dr=f^{-1}$. In the exterior $r\gt2GM$, the static coframe $\vartheta^{\hat0}=\sqrt f\,dt$, $\vartheta^{\hat1}=dr/\sqrt f$ is orthonormal but degenerates at $f=0$. Static observers require unbounded acceleration there; their frame is related to an infalling one by an unbounded boost. Divergent static-frame spinor components need not signify a singular spinor section.

Complete the static coframe by $r\,d\theta$ and $r\sin\theta\,d\varphi$ on an angular patch $0\lt\theta\lt\pi$. Substitution into the spin connection gives the explicit operator

$$\begin{aligned} \gamma^\mu D_\mu={}& \frac{\gamma^0}{\sqrt f}\partial_t +\gamma^1\sqrt f \left(\partial_r+\frac1r+\frac{f'}{4f}\right)\\ &+\frac{\gamma^2}{r} \left(\partial_\theta+\frac12\cot\theta\right) +\frac{\gamma^3}{r\sin\theta}\partial_\varphi, \qquad f'=\frac{df}{dr}. \end{aligned}$$

The radial rescaling $\psi=r^{-1}f^{-1/4}\widehat\psi$ removes the two radial connection terms for $f\gt0$ and helps produce the separated equations below. It does not remove the singularity of the static frame at $f=0$. Notice also that the $1/r$ and angular connection terms survive when $M=0$: a spherical frame rotates even in flat spacetime.

Use advanced time $v=t+r_*$ instead. The radial metric is $-f\,dv^2+2\,dv\,dr$. The two future null vectors

$$L=\partial_v+\frac f2\partial_r,\qquad \bar L=-\partial_r,\qquad g(L,\bar L)=-1$$

remain regular at $r=2GM$. The combinations $e_{\hat0}=(L+\bar L)/\sqrt2$ and $e_{\hat1}=(L-\bar L)/\sqrt2$, completed by an angular frame on each sphere patch, give a regular orthonormal frame and a local spin frame. Angular patch changes are not horizon singularities.

At the horizon the stationary generator is $\chi=\partial_v=L$. The inward number density is consequently

$$-j_\mu\chi^\mu =\frac{j^{\hat0}-j^{\hat1}}{\sqrt2}\ge0, \qquad j^r=j_\mu\chi^\mu\le0 \quad (r=2GM).$$

The first sign follows from causality; the second follows directly from the ingoing metric, for which $j_v=j^r$ at the horizon. An increasing-$r$ coordinate current is outward, not inward. This test does not depend on which separated amplitude is called “transmitted.”

For the exterior portions of two horizon-crossing slices, truncated at a timelike outer cylinder, Stokes’ theorem gives

$$N_{\Sigma_2}+\mathcal F_{\mathcal H} +\mathcal F_{\mathrm{out}}=N_{\Sigma_1}, \qquad \mathcal F_{\mathcal H} =-\int_{\mathcal H_{12}}j_\mu\chi^\mu\,dv\,dA.$$

Here $dA=(2GM)^2d\Omega$, the generator and its parameter $v$ are paired, and $\mathcal F_{\mathrm{out}}=\int j_\mu s^\mu\,dB$ uses the outward spacelike normal $s$. The cylinder term may be negative if radiation enters the region; the future-horizon term cannot be.

The same causal argument applies on Kerr with its future generator $\chi=\partial_t+\Omega_H\partial_\varphi$. A regular Carter tetrad and a separated massive equation are constructed in Röken 2019, v6, §II, Eq. (10), pp. 8–9, and §III A, Eqs. (12)–(15), pp. 9–10, PDF. That source uses $G=1$ and $e^{-i(\omega\tau+k\varphi)}$, so its $M$ is the length $GM$ and its $k=-m_z$ in the site’s azimuthal convention. Frame weights and coordinate phases must be translated before selecting a horizon branch.

Radial system and partner potentials

On Schwarzschild, spinor spherical harmonics have nonzero angular eigenvalues $\kappa=\pm(j+1/2)$, with $j=1/2,3/2,\ldots$. After the radial frame rescaling and angular separation, choose amplitudes $F,G$ and $x=r_*$ so that

$$\left(\frac{d}{dx}-i\omega\right)F=(W+iU)G, \qquad \left(\frac{d}{dx}+i\omega\right)G=(W-iU)F,$$

where $W=\kappa\sqrt f/r$ and $U=\mu_D\sqrt f$ are real. One direct source-to-variable check is to set $a=0$ in Röken’s Eq. (13), p. 10, and multiply both of its rescaled radial components by $e^{-i\omega(x-r)}$. Since $\tau=t+x-r$, this removes the advanced-time phase and gives the displayed static-time system. The two off-diagonal coefficients are complex conjugates, which is essential to its current.

For $\mu_D=0$, let $Z_\pm=F\pm G$. Addition and subtraction give

$$(\partial_x-W)Z_+=i\omega Z_-, \qquad (\partial_x+W)Z_-=i\omega Z_+.$$

Apply the opposite first-order factor, remembering that it differentiates $W$ as well as the amplitude:

$$\left[\frac{d^2}{dr_*^2}+\omega^2-V_\pm(r)\right]Z_\pm=0, \qquad V_\pm=W^2\pm\frac{dW}{dr_*}.$$

These are supersymmetric partner potentials. For nonzero real $\omega$, the first-order maps relate their scattering solutions and give equal reflection and transmission probabilities, though amplitude phases can differ. A reverse map divides by $i\omega$, so zero-frequency solutions require separate treatment. One must also map the boundary conditions; the factorization alone does not identify arbitrary operator domains.

Both potentials vanish at the horizon. In the flat limit $GM\to0$, $W=\kappa/r$ and $V_\pm=\kappa(\kappa\mp1)/r^2$, the two spinor centrifugal partners. At massive infinity $W\to0$ and $U\to\mu_D$; the first-order coefficient matrix squares to $-(\omega^2-\mu_D^2)I$. Thus propagating channels have wave number $k_\infty=\sqrt{\omega^2-\mu_D^2}$ for real $\omega\gt\mu_D$; the threshold and evanescent regime are not two-channel scattering with nonzero incident flux.

Normalize the angular harmonics and common radial scale so that the outward radial number current is $\mathcal J=|F|^2-|G|^2$. Direct differentiation of the first-order pair yields

$$\frac{d\mathcal J}{dx} =-2\,\operatorname{Im}\omega\, \left(|F|^2+|G|^2\right).$$

The mixed terms cancel because $W$ and $U$ are real. Hence $\mathcal J$ is constant for real frequency, not for a quasinormal mode at complex frequency. The latter still obeys spacetime current conservation: its time-dependent density supplies the term omitted by a stationary flux argument.

Scattering coefficients and energy flux

The massless Schwarzschild case makes the normalization transparent. For real $\omega\gt0$, the leading asymptotic branches are

$$\begin{aligned} x\to+\infty:\quad& F\sim A_{\mathrm{out}}e^{i\omega x}, &G&\sim A_{\mathrm{in}}e^{-i\omega x},\\ x\to-\infty:\quad& F\to0, &G&\sim A_{\mathcal H}e^{-i\omega x}. \end{aligned}$$

The horizon condition retains the inward $G$ branch; its phase combines with $e^{-i\omega t}$ to depend on $v$. Although $F$ has no leading outgoing branch there, the coupled equation can generate subleading terms. Equating the radial currents at the two ends gives

$$|A_{\mathrm{out}}|^2-|A_{\mathrm{in}}|^2 =-|A_{\mathcal H}|^2,\qquad \mathcal R+\mathcal T=1,$$

where $\mathcal R=|A_{\mathrm{out}}/A_{\mathrm{in}}|^2$ and $\mathcal T=|A_{\mathcal H}/A_{\mathrm{in}}|^2\ge0$ for a nonzero incident amplitude. Thus $\mathcal R\le1$. For a massive field, diagonalize the asymptotic first-order system and normalize each channel by its number flux, including the group velocity $k_\infty/\omega$. For Kerr, use the regular horizon current and its angular normalization. The causal-current proof of nonnegative absorption is given independently in Tóth 2016, v2, §5.1, Eqs. (5.1)–(5.6), pp. 15–16, PDF.

Energy requires a further distinction. With the adjoint used here, the symmetric on-shell stress tensor is

$$T_{\mu\nu}^{D} =-\frac{i}{4}\left[ \bar\psi\gamma_\mu D_\nu\psi +\bar\psi\gamma_\nu D_\mu\psi -(D_\nu\bar\psi)\gamma_\mu\psi -(D_\mu\bar\psi)\gamma_\nu\psi \right],$$

The sign is paired with the Clifford and adjoint definitions above: a positive-frequency flat plane wave has $T^D_{00}=\omega j^0$, not its negative. On a torsion-free fixed background without an external electromagnetic field, $\nabla_\mu T_D^{\mu\nu}=0$. The Hilbert energy current $\mathcal E_H^\mu=-T_D^\mu{}_\nu\xi^\nu$ is therefore conserved when $\xi$ is Killing.

In coordinates and a tetrad invariant under $\xi=\partial_t$, the canonical time-translation current of an on-shell mode is

$$\mathcal E_{\mathrm{can}}^\mu =\frac{i}{2}\left[ \bar\psi\gamma^\mu\partial_t\psi -(\partial_t\bar\psi)\gamma^\mu\psi \right] =\omega j^\mu \quad(\omega\in\mathbb R).$$

The canonical and Hilbert currents differ by the divergence of an antisymmetric spin term. Its integrated contribution is a boundary term, not an identically zero pointwise quantity. For a stationary bilinear, integration over a closed sphere and a time interval makes the angular boundary contributions vanish and the two end-cut terms cancel. Their stationary flux rates then agree. This is the qualified equivalence proved in Tóth 2016, v2, §2.1, Eqs. (2.3)–(2.8), pp. 4–5, PDF, translated from the opposite signature.

Consequently a positive-frequency incident mode has absorbed Killing-energy rate equal to $\omega$ times its absorbed number rate. Reflected energy cannot exceed incident energy in this stationary scattering problem. Monochromatic waves define rates, not finite energies integrated over all time. General wave packets and finite boundary cuts require keeping their own boundary contributions.

On Kerr, the horizon-frame frequency is $\omega-m_z\Omega_H$, but inserting that factor into a scalar flux formula does not give the Dirac flux. In particular, positivity of $j^\mu$ does not imply a pointwise energy condition for $T^D_{\mu\nu}$. These are different statements:

Quantity	What is controlled	Limitation
Dirac norm on a spacelike slice	Positive density	Not the local Killing-energy density
Inward future-horizon number flux	Nonnegative absorption	Does not determine a quantum occupation number
Positive-frequency stationary energy rate	Frequency times the number rate	Requires the stated boundary cancellation
Pointwise Dirac stress tensor	A conserved tensor for the neutral test equation	No general positive-energy condition follows

Scope and limitations

A tetrad regular in coordinates can still be poorly normalized, and a regular current can coexist with singular separated factors that cancel in reconstruction. Check the tetrad inner products, spin transformation, full spinor, and horizon contraction independently. Finite bilinears are useful necessary checks of a smooth solution, not a sufficient regularity theorem for its derivatives.

Current conservation at real frequency does not furnish the same stationary flux balance for complex-frequency resonances. A massless partner map does not automatically survive a mass term, a zero frequency, or a changed boundary domain. The distinction between a conserved current and a coercive stability estimate is developed in Energy Estimates and Test-Field Stability.

This page treats classical spinor test fields. It does not include their metric backreaction, coupled gravitational perturbations, second quantization, vacuum selection, Hawking occupation numbers, or particle creation. Those quantum-state questions belong to QFT in Curved Spacetime and cannot be inferred from a classical positive-frequency label alone.

Exercises

Suppose a massive first-order equation uses $\{\gamma^\mu,\gamma^\nu\}=2g^{\mu\nu}I$ but is written $(i\gamma^\mu\partial_\mu-\mu_D)\psi=0$ in flat space. Find its dispersion relation and explain why the massless case would fail to expose the error.

Solution

Multiply by $i\gamma^\mu\partial_\mu+\mu_D$. The product is $-\Box-\mu_D^2$, so the field satisfies $(\Box+\mu_D^2)\psi=0$. For the declared plane wave this gives $\omega^2=|\mathbf p|^2-\mu_D^2$, including imaginary frequencies at sufficiently small momentum. The physical massive relation has the opposite mass-squared sign. When $\mu_D=0$, both candidates reduce to $\Box\psi=0$; a massless-only regression therefore misses the defect. Keeping the stated Clifford algebra requires the no-$i$ massive equation used above, with its compatible adjoint and stress tensor.

For the radial system, take $\omega=\omega_R+i\omega_I$ with $\omega_I\ne0$. Is a radially constant $\mathcal J$ compatible with a nonzero mode? Explain how the answer fits spacetime current conservation.

Solution

The displayed identity gives $\mathcal J'=-2\omega_I(|F|^2+|G|^2)$, so a nonzero solution cannot have constant radial current on an interval. The full mode density is proportional to $e^{2\omega_I t}$, and its time derivative balances the radial divergence. Spacetime current conservation remains valid; the time-independent scattering reduction does not. In particular, one cannot interpret complex-frequency quasinormal amplitudes as stationary incident, reflected and transmitted number fluxes.

Suppose a separated calculation in a static Schwarzschild tetrad finds a component behaving as $f^{-1/4}$ at the horizon and declares the Dirac field singular. Give a coordinate- and tetrad-aware diagnostic, and state the flux condition that a regular solution must satisfy.

Solution

The static tetrad is related to a horizon-regular tetrad by a boost whose rapidity diverges as $f\to0$. The spin representation carries half the vector boost weight, so individual components may acquire factors such as $f^{\pm1/4}$. Transform the spinor together with the tetrad into advanced coordinates and test smoothness of the spinor section. The current and stress tensor must then be finite in a regular frame; their finiteness alone is not sufficient to prove smoothness.

At the future horizon, $-j\cdot\chi=(j^{\hat0}-j^{\hat1})/\sqrt2\ge0$ and $j^r\le0$. For the real-frequency scattering state with unit incident number flux and no outgoing horizon branch, this gives $\mathcal R+\mathcal T=1$ with $\mathcal T\ge0$. A reflected coefficient above one fails that current test. The geometric $1/r$ and angular connection terms in spherical coordinates do not disappear in the flat limit; what must be recovered is the flat Dirac equation in that spherical frame, or the Cartesian equation after a further frame and coordinate change.

References

• Röken, Christian. “The Massive Dirac Equation in Kerr Geometry: Separability in Eddington–Finkelstein-Type Coordinates and Asymptotics.” General Relativity and Gravitation 49 (2017): article 39. DOI. Open PDF, revised preprint v6, 2019. Printed page locators above refer to this preprint.
• Tóth, Gábor Zsolt. “Weak Cosmic Censorship, Dyonic Kerr–Newman Black Holes and Dirac Fields.” Classical and Quantum Gravity 33 (2016): 115012. DOI. Open PDF, preprint v2. Only the current and scattering results are used here; the page locators refer to the preprint.