Introduction
The ringdown of a black hole (BH) is characterized by quasinormal modes (QNMs), whose complex frequencies encode the oscillation frequencies and damping rates of linear perturbations. Gravitational-wave observations have opened a new avenue for testing BH dynamics in the strong-gravity regime [1 ] . The theoretical framework of BH QNMs and its extensions to systems with additional fields and interactions are reviewed in Refs. [2 , 3 , 4 , 5 , 6 ] .
BH QNMs are determined within perturbation theory by solving the field equations linearized about a BH background, with ingoing boundary conditions at the event horizon and outgoing boundary conditions at infinity. The foundations of this framework were established by the Regge–Wheeler and Zerilli equations for Schwarzschild BHs [7 , 8 , 9 ] . Chandrasekhar and collaborators investigated the Schwarzschild QNM spectrum and developed transformations relating the two parity sectors [10 , 11 , 12 ] . Perturbations of charged BHs were studied by Moncrief, Zerilli, and Chandrasekhar [13 , 14 , 15 , 16 , 17 ] .
Several complementary methods have been developed to compute QNM frequencies. Direct numerical integration was used in early calculations [18 , 19 ] ; its matrix-valued extension treats coupled perturbations through a determinant condition on the unwanted asymptotic amplitudes or on a matching matrix [6 , 20 ] . Semiclassical Wentzel–Kramers–Brillouin (WKB) approximations [21 , 22 ] , including improvements based on Padé approximants [23 ] , provide efficient estimates when a suitable potential description is available. Leaver’s continued-fraction method [24 , 25 , 26 ] and its applications to charged and near-extremal BHs [27 , 28 , 29 ] provide accurate frequency-domain benchmarks.
Spectral and pseudospectral methods instead approximate the regularized perturbations in a finite basis and reduce the boundary-value problem to an algebraic eigenvalue problem, as in Chebyshev collocation [30 ] and Bernstein-polynomial implementations [31 ] . First-order formulations also permit asymptotic analysis and spectral discretization without reduction to a Schrödinger-like equation [32 ] . Chung et al. developed a spectral approach to metric perturbations that directly solves the coupled linearized Einstein equations on Schwarzschild and Kerr backgrounds [33 , 34 ] . The Kerr implementation is known as METRICS. Horizon-penetrating hyperboloidal coordinates provide a formulation in which the event horizon and future null infinity are included in a compactified computational domain [35 , 36 ] .
QNM frequencies can also be extracted from time-domain waveforms [37 , 38 ] or computed using the asymptotic iteration method [39 ] . Since these methods have different convergence properties and domains of applicability, comparisons between independent approaches provide useful checks of numerical accuracy, particularly for coupled perturbations. The QNM spectrum is determined by the full perturbation operator and therefore depends on both the background geometry and the couplings among the perturbed fields. Interactions that leave a background solution unchanged can still modify the linearized equations and hence shift its QNM frequencies.
Charged BHs provide a useful setting for examining the relation between the background geometry and the QNM spectrum. In Einstein–Maxwell theory, the QNM spectrum of a Reissner–Nordström (RN) BH depends on its mass and total charge magnitude, but not on the relative electric and magnetic contributions. This degeneracy follows from electromagnetic duality, which can be formulated directly at the level of the coupled perturbation equations [40 , 41 ] . The gravitational and electromagnetic (EM) QNM families also exhibit axial–polar isospectrality: within each family, the two parity sectors have identical spectra. Additional matter interactions can break these spectral degeneracies. Beyond Einstein–Maxwell theory, static, spherically symmetric hairy BH solutions have been constructed in U ( 1 ) U(1) U ( 1 ) gauge-invariant scalar–vector–tensor (SVT) theories [42 ] . Their odd-parity perturbations and the associated no-ghost and gradient-stability conditions have also been studied [43 ] .
Axions provide a particle-physics motivation for such interactions. The Peccei–Quinn mechanism addresses the strong CP problem by dynamically suppressing charge-conjugation and parity (CP) violation in quantum chromodynamics (QCD) through a pseudoscalar field, the axion [44 , 45 , 46 ] . Invisible axion models implement this mechanism through heavy-quark or extended Higgs sectors [47 , 48 , 49 , 50 ] , while coherent oscillations of the axion field can contribute to the dark matter abundance [51 , 52 , 53 ] . More general axion-like particles (ALPs) arise in string compactifications [54 , 55 , 56 , 57 , 58 ] . Their masses and photon couplings need not obey the relation characteristic of QCD axion models. We therefore study a generic ALP in the massless limit, treating its photon coupling as an independent parameter.
Early exact solutions describing gravitating dyons and dyonic BHs were obtained by Kasuya and collaborators [59 , 60 ] . In axion theories, the coupling to the electromagnetic Chern–Pontryagin invariant sources axion hair in the presence of both electric and magnetic fields. Static dyonic BHs with axion hair have been constructed in several settings [61 , 62 , 63 ] . Their QNMs can distinguish electric–magnetic charge configurations that would yield the same RN geometry in Einstein–Maxwell theory [64 ] . Even on a purely electric RN background, for which a regular massless axion is constant, axion and electromagnetic perturbations remain coupled in the odd-parity sector. The coupled system can exhibit avoided crossings in the QNM spectrum and resonant enhancement of the excitation factors [65 , 66 , 67 ] . These effects must be established through the coupled eigenvalue and excitation problems; a sign change in a frequency correction alone is insufficient to demonstrate a resonance. At sufficiently strong coupling, the same axion–EM interaction can also trigger mode instabilities [65 ] .
Nonlinear electrodynamics (NED) provides another theoretically motivated extension of Maxwell theory.
Euler–Heisenberg theory describes local nonlinear electromagnetic interactions induced by charged-particle loops [68 , 69 , 70 ] , while Born–Infeld theory offers a distinct nonlinear generalization [71 ] . When coupled to gravity, NED admits charged BH solutions whose geometries can differ substantially from that of RN. Solutions with regular centers include those constructed by Ayón-Beato and García [72 , 73 , 74 ] , as well as electrically charged regular configurations [75 ] . More general constructions and their properties have also been investigated [76 , 77 , 78 , 79 , 80 , 81 ] . Other notable examples include BH solutions in Euler–Heisenberg and Born–Infeld theories, with and without a cosmological constant [82 , 83 , 84 , 85 , 86 ] .
Regularity of the background does not guarantee perturbative stability. For the NED BHs with regular centers analyzed in Refs. [87 , 88 ] , an angular gradient instability develops near the center. This conclusion applies to the electromagnetic and scalar extensions considered there and does not automatically extend to arbitrary matter Lagrangians. Conversely, positive-power corrections such as the Euler–Heisenberg-type F 2 F^2 F 2 term need not eliminate the curvature singularity at r = 0 r = 0 r = 0 , although singular BH branches can satisfy local no-ghost and propagation conditions [89 ] . The exterior configurations studied here in the weak-NED regime satisfy sufficient local conditions of this kind. These exterior conditions do not, however, establish global mode stability. Moreover, a truncated low-energy action cannot determine the ultraviolet resolution of the central singularity.
Low-energy effective field theory (EFT) provides a direct motivation for combining axion hair with nonlinear electromagnetic corrections. In a theory containing an axion–photon interaction, integrating out heavy charged matter also generates quartic field-strength operators. Their Wilson coefficients encode the masses, charges, spins, and multiplicities of the heavy particles [68 , 70 , 90 ] , which may belong to the visible or a dark sector. Such corrections can therefore carry information about heavy states while remaining subdominant to the Maxwell term.
In a dyonic background with axion hair, NED simultaneously modifies the electric field, the axion source, the metric, and the coupled perturbation equations. The resulting QNM shifts cannot generally be obtained by simply adding separate axion-hair and NED corrections to the RN spectrum. Building on previous NED QNM analyses without axion hair [91 , 92 ] , we investigate this interplay in the full five-field system. Our analysis isolates the response to the retained quartic operators; interpreting this response as a prediction of a specific EFT additionally requires control over the omitted operators.
We first follow five quadrupolar (ℓ = 2 \ell= 2 ℓ = 2 ) QNM families from the fundamental RN modes in the minimal axion theory, extending the two-family dyonic analysis of Ref. [64 ] . Even without NED, axion perturbations couple to the odd-parity gravitoelectromagnetic sector on a purely electric RN background, while the even-parity sector retains its Einstein–Maxwell spectrum. This coupling lifts the axial–polar frequency degeneracy despite the absence of background axion hair. A separate search identifies additional growing modes at selected strong-coupling benchmarks. To interpret the small-coupling splitting, we use an EM–scalar Darboux degeneracy of the decoupled RN system, previously derived in Ref. [93 ] .
We then determine the additional NED shifts, distinguishing the full correction from its magnetic modification at fixed charge-to-mass ratio. With the axion decoupled, equal quartic coefficients make the theory invariant under electromagnetic duality through first NED order. Consequently, each individual gravitational or EM branch has a first-order frequency shift independent of charge composition. For unequal coefficients, the rotated theory contains an additional parity-odd interaction; duality then constrains the sum of the shifts within each isolated RN parity pair. Calculating the individual shifts in the degenerate subspace generally requires a coupled treatment [94 ] . We examine how finite axion coupling removes the equal-coefficient cancellation, then determine the individual finite-NED shifts for separately varied and unequal coefficients. A derivative expansion provides a local approximation to these directly computed frequencies.
We compute the spectra by compactified Chebyshev collocation [30 ] of five coupled radial equations obtained by eliminating the nondynamical fields from the quadratic action. Factoring out the QNM boundary phases gives a quadratic eigenvalue problem retaining the first-derivative couplings. All figures use these spectral results. Independent integration-and-matching calculations and tests of radial resolution, background accuracy, and parameter-step dependence support representative frequencies and shifts.
The paper is organized as follows. Section II introduces the theory, its Wilson-coefficient interpretation, the background solutions, and the normalization conventions. Section III develops the five-field numerical formulation. § 2 establishes the minimal-theory spectrum and magnetic response, and § 2.2 presents the NED frequency shifts and their physical interpretation. We conclude in Sec. VI. Appendix A discusses the EFT interpretation, and Appendix B gives the coupled perturbation equations. Appendix C collects the supporting numerical checks. Appendices D and E present the frequency-response formulas and the duality argument, respectively.
We use natural units, ℏ = c = 1 \hbar= c = 1 ℏ = c = 1 , and adopt the metric signature ( − , + , + , + ) (-,+,+,+) ( − , + , + , + ) throughout.
Theory, backgrounds, and control of the NED expansion
The interplay between axion hair and electromagnetic nonlinearity is described by Einstein gravity coupled to a canonical pseudoscalar ϕ \phi ϕ and a U ( 1 ) U(1) U ( 1 ) gauge potential A μ A_\mu A μ . Let g μ ν g_{\mu\nu} g μν denote the spacetime metric and g M ≡ det ( g μ ν ) g_M \equiv\det(g_{\mu\nu}) g M ≡ det ( g μν ) its determinant. The electromagnetic field strength is F μ ν = ∂ μ A ν − ∂ ν A μ F_{\mu\nu} = \partial_\mu A_\nu- \partial_\nu A_\mu F μν = ∂ μ A ν − ∂ ν A μ . Greek indices label spacetime components and are contracted with the metric. We define the electromagnetic invariants and Hodge-dual convention by
F = − 1 4 F μ ν F μ ν , G = − 1 4 F μ ν F ~ μ ν , \mathcal{F} = -\frac{1}{4}F_{\mu\nu}F^{\mu\nu}, \qquad\mathcal{G} = -\frac{1}{4}F_{\mu\nu}\widetilde{F}^{\mu\nu}, F = − 4 1 F μν F μν , G = − 4 1 F μν F μν ,
F ~ μ ν = − ϵ μ ν ρ σ 2 − g M F ρ σ , ϵ t r θ φ = + 1. \widetilde{F}^{\mu\nu} = -\frac{\epsilon^{\mu\nu\rho\sigma}}{2\sqrt{-g_M}}F_{\rho\sigma}, \qquad\epsilon^{tr\theta\varphi} = +1. F μν = − 2 − g M ϵ μν ρ σ F ρ σ , ϵ t r θ φ = + 1.
The tensor density ϵ μ ν ρ σ \epsilon^{\mu\nu\rho\sigma} ϵ μν ρ σ is the totally antisymmetric symbol in the oriented coordinates ( t , r , θ , φ ) (t,r,\theta,\varphi) ( t , r , θ , φ ) used in Eq. (1) .
We retain the axion–photon interaction and the two parity-even quartic invariants without derivatives of the field strength. The action is
S = ∫ d 4 x − g M [ M P l 2 2 R − 1 2 ∇ μ ϕ ∇ μ ϕ + L ] , S = \int d^4x\sqrt{-g_M}\left[\frac{M_{\mathrm{Pl}}^2}{2}R-\frac{1}{2}\nabla_\mu\phi\nabla^\mu\phi+\mathcal{L}\right], S = ∫ d 4 x − g M [ 2 M Pl 2 R − 2 1 ∇ μ ϕ ∇ μ ϕ + L ] ,
L = F + α F 2 + β G 2 + g ϕ G . \mathcal{L} = \mathcal{F}+\alpha\mathcal{F}^2+\beta\mathcal{G}^2+g\phi\mathcal{G}. L = F + α F 2 + β G 2 + g ϕ G .
Here R R R is the Ricci scalar, ∇ μ \nabla_\mu ∇ μ the metric-compatible covariant derivative, and M P l 2 = ( 8 π G N ) − 1 M_{\mathrm{Pl}}^2=(8\pi G_N)^{-1} M Pl 2 = ( 8 π G N ) − 1 defines the reduced Planck mass in terms of Newton’s constant G N G_N G N . The constant g g g denotes the axion–photon coupling, while α \alpha α and β \beta β control the quartic NED interactions F 2 \mathcal{F}^2 F 2 and G 2 \mathcal{G}^2 G 2 , respectively. The mass dimensions are [ ϕ ] = [ A μ ] = 1 [\phi]=[A_\mu]=1 [ ϕ ] = [ A μ ] = 1 , [ g ] = − 1 [g]=-1 [ g ] = − 1 , and [ α ] = [ β ] = − 4 [\alpha]=[\beta]=-4 [ α ] = [ β ] = − 4 . We set the pseudoscalar potential to zero throughout both the background and QNM calculations. We work in the massless ALP limit. Applying our results to a massive axion requires the effects of its potential on the background, the perturbation equations, and the QNM boundary conditions to be negligible.
The coefficients α \alpha α and β \beta β are Wilson coefficients of the retained quartic electromagnetic operators. For example, integrating out heavy Dirac particles with masses m i m_i m i , coupled to A μ A_\mu A μ with charges Q i e D Q_i e_D Q i e D , yields the leading spinor Euler–Heisenberg coefficients
α = ∑ i 8 α D 2 Q i 4 45 m i 4 , β = ∑ i 14 α D 2 Q i 4 45 m i 4 . \alpha= \sum_i \frac{8\alpha_D^2 Q_i^4}{45m_i^4}, \qquad\beta= \sum_i \frac{14\alpha_D^2 Q_i^4}{45m_i^4}. α = i ∑ 45 m i 4 8 α D 2 Q i 4 , β = i ∑ 45 m i 4 14 α D 2 Q i 4 .
where e D e_D e D is the gauge coupling, Q i Q_i Q i is the dimensionless charge of species i i i , and α D ≡ e D 2 / ( 4 π ) \alpha_D \equiv e_D^2/(4\pi) α D ≡ e D 2 / ( 4 π ) . The microscopic charges Q i Q_i Q i are distinct from the geometric BH charge Q Q Q defined below. This matching gives β / α = 7 / 4 \beta/\alpha=7/4 β / α = 7/4 , while α = β \alpha=\beta α = β corresponds to the leading quartic expansion of Born–Infeld theory. These ratios characterize the retained quartic interactions without specifying a complete nonlinear theory. Appendix A gives the one-loop interaction, its multiplicity dependence, and the additional requirements for an EFT interpretation. In our numerical analysis, we vary α \alpha α and β \beta β independently to isolate their respective effects. The signed variations used for numerical differentiation are computational probes and are not required to satisfy the heavy-Dirac matching relations.
Static dyonic equations
We consider static, spherically symmetric dyonic backgrounds of the form
d s 2 = − N h d t 2 + d r 2 h + r 2 d Ω 2 2 , (1) ds^2=-Nh\,dt^2+\frac{dr^2}{h}+r^2d\Omega_2^2, \tag*{(1)} d s 2 = − N h d t 2 + h d r 2 + r 2 d Ω 2 2 , (1)
A μ d x μ = A 0 ( r ) d t − q M cos θ d φ , ϕ = ϕ ( r ) . (2) A_\mu dx^\mu=A_0(r)dt-q_M\cos\theta\,d\varphi,\qquad\phi=\phi(r). \tag*{(2)} A μ d x μ = A 0 ( r ) d t − q M cos θ d φ , ϕ = ϕ ( r ) . (2)
Here N ( r ) N(r) N ( r ) and h ( r ) h(r) h ( r ) are metric functions, r r r is the areal radius, and d Ω 2 2 = d θ 2 + sin 2 θ d φ 2 d\Omega_2^2=d\theta^2+\sin^2\theta\,d\varphi^2 d Ω 2 2 = d θ 2 + sin 2 θ d φ 2 is the line element on the unit two-sphere. The function A 0 ( r ) A_0(r) A 0 ( r ) is the electric potential. The signed magnetic charge q M q_M q M is constant, as required by the Bianchi identity ∇ μ F ~ μ ν = 0 \nabla_\mu\widetilde{F}^{\mu\nu}=0 ∇ μ F μν = 0 , and is normalized by ∫ S 2 F = 4 π q M \int_{S^2}F=4\pi q_M ∫ S 2 F = 4 π q M , where F = 1 2 F μ ν d x μ ∧ d x ν F=\frac{1}{2}F_{\mu\nu}dx^\mu\wedge dx^\nu F = 2 1 F μν d x μ ∧ d x ν is the electromagnetic field-strength two-form and S 2 S^2 S 2 is a sphere at fixed t t t and r r r . Primes and overdots denote derivatives with respect to r r r and t t t , respectively. The signed radial electric and magnetic fields measured in a local orthonormal frame are E = A 0 ′ / N E=A_0'/\sqrt{N} E = A 0 ′ / N and B = q M / r 2 B=q_M/r^2 B = q M / r 2 , giving F = ( E 2 − B 2 ) / 2 \mathcal{F}=(E^2-B^2)/2 F = ( E 2 − B 2 ) /2 and G = − E B \mathcal{G}=-EB G = − E B . A comma denotes differentiation with respect to the indicated invariant, for example L , F = ∂ L / ∂ F \mathcal{L}_{,\mathcal{F}}=\partial\mathcal{L}/\partial\mathcal{F} L , F = ∂ L / ∂ F .
Varying the action with respect to A ν A_\nu A ν gives the gauge-field equation
∇ μ ( L , F F μ ν + L , G F ~ μ ν ) = 0. (3) \nabla_\mu\left(\mathcal{L}_{,\mathcal{F}}F^{\mu\nu}+\mathcal{L}_{,\mathcal{G}}\widetilde{F}^{\mu\nu}\right)=0. \tag*{(3)} ∇ μ ( L , F F μν + L , G F μν ) = 0. (3)
On this background, its t t t component reduces to ( r 2 L , F E − q M L , G ) ′ = 0 (r^2\mathcal{L}_{,\mathcal{F}}E-q_M\mathcal{L}_{,\mathcal{G}})'=0 ( r 2 L , F E − q M L , G ) ′ = 0 . Integrating once yields
r 2 L , F E − q M L , G = q E , (4) r^2\mathcal{L}_{,\mathcal{F}}E-q_M\mathcal{L}_{,\mathcal{G}}=q_E, \tag*{(4)} r 2 L , F E − q M L , G = q E , (4)
where the integration constant q E q_E q E is the electric displacement charge, with conserved displacement flux 4 π q E 4\pi q_E 4 π q E . Substituting L , F = 1 + 2 α F \mathcal{L}_{,\mathcal{F}}=1+2\alpha\mathcal{F} L , F = 1 + 2 α F and L , G = 2 β G + g ϕ \mathcal{L}_{,\mathcal{G}}=2\beta\mathcal{G}+g\phi L , G = 2 β G + g ϕ , together with the above expressions for the invariants, gives the algebraic electric-field equation
E [ 1 + α E 2 + ( 2 β − α ) B 2 ] = q E + g q M ϕ r 2 . (5) E\left[1+\alpha E^2+(2\beta-\alpha)B^2\right]=\frac{q_E+gq_M\phi}{r^2}. \tag*{(5)} E [ 1 + α E 2 + ( 2 β − α ) B 2 ] = r 2 q E + g q M ϕ . (5)
We select the electric-field branch continuously connected to the α = β = 0 \alpha=\beta=0 α = β = 0 solution.
The independent Einstein and scalar equations become
N ′ N = r ϕ ′ 2 M P l 2 , (6) \frac{N'}{N}=\frac{r\phi'^2}{M_{\mathrm{Pl}}^2}, \tag*{(6)} N N ′ = M Pl 2 r ϕ ′2 , (6)
h ′ = 1 − h r − r ρ E M M P l 2 − r h ϕ ′ 2 2 M P l 2 , (7) h'=\frac{1-h}{r}-\frac{r\rho_{\mathrm{EM}}}{M_{\mathrm{Pl}}^2}-\frac{rh\phi'^2}{2M_{\mathrm{Pl}}^2}, \tag*{(7)} h ′ = r 1 − h − M Pl 2 r ρ EM − 2 M Pl 2 r h ϕ ′2 , (7)
( r 2 h N ϕ ′ ) ′ = g q M N E , (8) \left(r^2h\sqrt{N}\phi'\right)'=gq_M\sqrt{N}E, \tag*{(8)} ( r 2 h N ϕ ′ ) ′ = g q M N E , (8)
where the electromagnetic energy density is
ρ E M = E 2 + B 2 2 + β E 2 B 2 + α 4 ( 3 E 4 − 2 E 2 B 2 − B 4 ) . (9) \rho_{\mathrm{EM}}=\frac{E^2+B^2}{2}+\beta E^2B^2+\frac{\alpha}{4}\left(3E^4-2E^2B^2-B^4\right). \tag*{(9)} ρ EM = 2 E 2 + B 2 + β E 2 B 2 + 4 α ( 3 E 4 − 2 E 2 B 2 − B 4 ) . (9)
The topological axion interaction has no explicit stress-energy contribution. It affects the metric through the electric field and the scalar profile.
Regularity at a nonextremal event horizon of areal radius r h r_h r h gives
h ( r h ) = 0 , h h ′ = 1 − r h 2 ρ E M , h / M P l 2 r h > 0 , (10) h(r_h)=0,\qquad h'_h=\frac{1-r_h^2\rho_{\mathrm{EM},h}/M_{\mathrm{Pl}}^2}{r_h}>0, \tag*{(10)} h ( r h ) = 0 , h h ′ = r h 1 − r h 2 ρ EM , h / M Pl 2 > 0 , (10)
ϕ h ′ = g q M E h r h 2 h h ′ . (11) \phi'_h=\frac{gq_ME_h}{r_h^2h'_h}. \tag*{(11)} ϕ h ′ = r h 2 h h ′ g q M E h . (11)
A subscript h h h denotes evaluation at r = r h r=r_h r = r h . We impose asymptotic flatness, fix the time normalization by N ( ∞ ) = 1 N(\infty)=1 N ( ∞ ) = 1 , and choose the asymptotic axion value ϕ ( ∞ ) = 0 \phi(\infty)=0 ϕ ( ∞ ) = 0 . The geometric mass M M M is extracted from the asymptotic metric and is related to the Arnowitt–Deser–Misner (ADM) mass M A D M M_{\mathrm{ADM}} M ADM by
h ( r ) = 1 − 2 M r + O ( r − 2 ) , M = G N M A D M . (12) h(r)=1-\frac{2M}{r}+O(r^{-2}),\qquad M=G_NM_{\mathrm{ADM}}. \tag*{(12)} h ( r ) = 1 − r 2 M + O ( r − 2 ) , M = G N M ADM . (12)
We define the total flux magnitude q T ≥ 0 q_T\geq0 q T ≥ 0 , the geometric total charge Q Q Q , the signed magnetic fraction η \eta η , and the dimensionless couplings and frequency by
q T 2 = q E 2 + q M 2 , Q = q T 2 M P l , (13) q_T^2=q_E^2+q_M^2,\qquad Q=\frac{q_T}{\sqrt{2}M_{\mathrm{Pl}}}, \tag*{(13)} q T 2 = q E 2 + q M 2 , Q = 2 M Pl q T , (13)
η = q M q T , g ˉ = g M P l , \eta=\frac{q_M}{q_T},\qquad\bar{g}=gM_{\mathrm{Pl}}, η = q T q M , g ˉ = g M Pl ,
α ^ = α M P l 2 r h 2 , β ^ = β M P l 2 r h 2 , \hat{\alpha}=\frac{\alpha M_{\mathrm{Pl}}^2}{r_h^2},\qquad\hat{\beta}=\frac{\beta M_{\mathrm{Pl}}^2}{r_h^2}, α ^ = r h 2 α M Pl 2 , β ^ = r h 2 β M Pl 2 ,
Ω = M ω . (14) \Omega=M\omega. \tag*{(14)} Ω = M ω . (14)
Here g ˉ \bar{g} g ˉ is the dimensionless axion coupling, while α ^ \hat{\alpha} α ^ and β ^ \hat{\beta} β ^ are the NED coefficients normalized by the horizon radius. We adopt the Fourier convention e − i ω t e^{-i\omega t} e − iω t and report the dimensionless QNM frequency Ω = M ω \Omega=M\omega Ω = M ω , with damping rate Γ = − Im Ω \Gamma=-\operatorname{Im}\Omega Γ = − Im Ω . These conventions give h = 1 − 2 M / r + Q 2 / r 2 h=1-2M/r+Q^2/r^2 h = 1 − 2 M / r + Q 2 / r 2 in the RN limit. In the numerical calculations, we set r h = M P l = 1 r_h=M_{\mathrm{Pl}}=1 r h = M Pl = 1 and adjust q T q_T q T along each parameter sequence to maintain the prescribed Q / M Q/M Q / M . Holding q T / r h q_T/r_h q T / r h fixed instead would generally change Q / M Q/M Q / M and thus define a different frequency response.
We normalize the asymptotic time coordinate before constructing the perturbation operator. The subscript “raw” denotes quantities before this normalization. For a shooting solution with C ∞ = N r a w ( ∞ ) > 0 C_\infty=N_{\mathrm{raw}}(\infty)>0 C ∞ = N raw ( ∞ ) > 0 , we apply the rescaling
N = N r a w C ∞ , t = C ∞ t r a w , N=\frac{N_{\mathrm{raw}}}{C_\infty},\qquad t=\sqrt{C_\infty}\,t_{\mathrm{raw}}, N = C ∞ N raw , t = C ∞ t raw ,
A 0 = A 0 , r a w C ∞ , ω = ω r a w C ∞ . (15) A_0=\frac{A_{0,\mathrm{raw}}}{\sqrt{C_\infty}},\qquad\omega=\frac{\omega_{\mathrm{raw}}}{\sqrt{C_\infty}}. \tag*{(15)} A 0 = C ∞ A 0 , raw , ω = C ∞ ω raw . (15)
The local electric field E = A 0 ′ / N E=A_0'/\sqrt{N} E = A 0 ′ / N , the charges, and the ADM mass are invariant under this rescaling. The perturbation operator and all reported frequencies use the normalized time coordinate. Appendix B gives the corresponding rescaling of the auxiliary fields.
Both numerical implementations use these conventions. The corresponding mass-normalized NED coefficients are
α M = α M P l 2 M 2 = α ^ ( M / r h ) 2 , (16) \alpha_M = \frac{\alpha M_{\mathrm{Pl}}^2}{M^2} = \frac{\hat{\alpha}}{(M/r_h)^2}, \tag*{(16)} α M = M 2 α M Pl 2 = ( M / r h ) 2 α ^ , (16)
β M = β M P l 2 M 2 = β ^ ( M / r h ) 2 . (17) \beta_M = \frac{\beta M_{\mathrm{Pl}}^2}{M^2} = \frac{\hat{\beta}}{(M/r_h)^2}. \tag*{(17)} β M = M 2 β M Pl 2 = ( M / r h ) 2 β ^ . (17)
Our NED-response sequences vary α ^ \hat{\alpha} α ^ and β ^ \hat{\beta} β ^ at fixed Q / M Q/M Q / M , g ˉ \bar{g} g ˉ , and η \eta η . Along a magnetic-composition sequence at fixed α ^ \hat{\alpha} α ^ , β ^ \hat{\beta} β ^ , the ratio M / r h M/r_h M / r h generally changes, so α M , β M \alpha_M,\beta_M α M , β M need not remain constant. Keeping the latter fixed would require adjusting the hatted coefficients as the magnetic composition varies. Expressing frequencies as M ω M\omega M ω does not make these two parameter prescriptions equivalent.
Amplitude and local stability
We characterize the strength of the retained NED corrections by
ϵ N E D ( r ) = max ( ∣ α ∣ , ∣ β ∣ ) ( E 2 + B 2 ) . (18) \epsilon_{\mathrm{NED}}(r) = \max(|\alpha|,|\beta|)(E^2+B^2). \tag*{(18)} ϵ NED ( r ) = max ( ∣ α ∣ , ∣ β ∣ ) ( E 2 + B 2 ) . (18)
This measure avoids cancellations between the electric and magnetic contributions that can make F \mathcal{F} F small. Since F 2 + G 2 = ( E 2 + B 2 ) 2 / 4 \mathcal{F}^2+\mathcal{G}^2=(E^2+B^2)^2/4 F 2 + G 2 = ( E 2 + B 2 ) 2 /4 , the magnitude of the quartic Lagrangian term α F 2 + β G 2 \alpha\mathcal{F}^2+\beta\mathcal{G}^2 α F 2 + β G 2 is bounded by ϵ N E D / 2 \epsilon_{\mathrm{NED}}/2 ϵ NED /2 times the local Maxwell energy density ( E 2 + B 2 ) / 2 (E^2+B^2)/2 ( E 2 + B 2 ) /2 . We impose the exterior amplitude criterion ϵ N E D ≡ sup r ≥ r h ϵ N E D ( r ) < 0.05 \epsilon_{\mathrm{NED}}\equiv\sup_{r\ge r_h}\epsilon_{\mathrm{NED}}(r)<0.05 ϵ NED ≡ sup r ≥ r h ϵ NED ( r ) < 0.05 . The conditions ∣ α F ∣ ≪ 1 |\alpha\mathcal{F}|\ll1 ∣ α F ∣ ≪ 1 and ∣ β G ∣ ≪ 1 |\beta\mathcal{G}|\ll1 ∣ β G ∣ ≪ 1 are insufficient by themselves. For example, on a purely electric background, G = 0 \mathcal{G}=0 G = 0 leaves β \beta β unconstrained, even when β E 2 \beta E^2 β E 2 produces large corrections to the perturbation equations.
Applying the general dyonic stability conditions of [95 ] (§ 2 , in particular Eq. (4.67)) to the model (2.4), we obtain the following sufficient local no-ghost and propagation conditions:
a = 1 + α ( E 2 − B 2 ) > 0 , d = 1 + 3 α E 2 + ( 2 β − α ) B 2 > 0 , c e m = 1 + ( α − 2 β ) E 2 − 3 α B 2 > 0. (19) \begin{aligned}
a &= 1+\alpha(E^2-B^2)>0,\\
d &= 1+3\alpha E^2+(2\beta-\alpha)B^2>0,\\
c_{\mathrm{em}} &= 1+(\alpha-2\beta)E^2-3\alpha B^2>0. \tag*{(19)}
\end{aligned} a d c em = 1 + α ( E 2 − B 2 ) > 0 , = 1 + 3 α E 2 + ( 2 β − α ) B 2 > 0 , = 1 + ( α − 2 β ) E 2 − 3 α B 2 > 0. (19)
Here a a a , d d d , c e m c_{\mathrm{em}} c em are local constitutive coefficients. Let c Ω 2 c_{\Omega}^2 c Ω 2 denote a squared angular propagation speed measured in a local orthonormal frame. In the high-frequency limit, the scalar and two gravitational modes have c Ω 2 = 1 c_{\Omega}^2=1 c Ω 2 = 1 , while all five modes have unit squared radial propagation speeds. Defining b = 2 ( β − α ) E B b=2(\beta-\alpha)EB b = 2 ( β − α ) E B and τ Ω = c e m d + a 2 + b 2 \tau_{\Omega}=c_{\mathrm{em}}d+a^2+b^2 τ Ω = c em d + a 2 + b 2 , the remaining two angular speeds satisfy
a d z 2 − τ Ω z + a c e m = 0 , z = c Ω 2 . (20) ad z^2-\tau_{\Omega}z+a c_{\mathrm{em}}=0,\qquad z=c_{\Omega}^2. \tag*{(20)} a d z 2 − τ Ω z + a c em = 0 , z = c Ω 2 . (20)
For positive a a a , d d d , c e m c_{\mathrm{em}} c em , both roots are real and positive because τ Ω ≥ 2 a c e m d \tau_{\Omega}\ge2a\sqrt{c_{\mathrm{em}}d} τ Ω ≥ 2 a c em d . Moreover, ∣ a − 1 ∣ ≤ ϵ N E D |a-1|\le\epsilon_{\mathrm{NED}} ∣ a − 1∣ ≤ ϵ NED and ∣ c e m − 1 ∣ , ∣ d − 1 ∣ ≤ 3 ϵ N E D |c_{\mathrm{em}}-1|,|d-1|\le3\epsilon_{\mathrm{NED}} ∣ c em − 1∣ , ∣ d − 1∣ ≤ 3 ϵ NED , so our amplitude criterion guarantees a > 0.95 a>0.95 a > 0.95 and c e m , d > 0.85 c_{\mathrm{em}},d>0.85 c em , d > 0.85 throughout the exterior. These are sufficient local high-frequency conditions; they do not prove global mode stability for arbitrary multipoles. Explicit growing modes at strong axion coupling are discussed in § 2 D.
The QNM boundary-value problem is posed on the exterior region, where the amplitude criterion is imposed. Small retained NED amplitudes do not constrain omitted derivative, curvature, or additional axion operators. Appendix A discusses the heavy-particle scale hierarchies and the additional conditions required for a controlled EFT interpretation.
Five coupled perturbations and numerical construction
The linear perturbations are expanded in spherical harmonics Y ℓ m ( θ , φ ) Y_{\ell m}(\theta,\varphi) Y ℓ m ( θ , φ ) , where ℓ \ell ℓ and m m m denote the angular multipole and azimuthal numbers, respectively. We define
L = ℓ ( ℓ + 1 ) , (21) L=\ell(\ell+1), \tag*{(21)} L = ℓ ( ℓ + 1 ) , (21)
and set m = 0 m=0 m = 0 without loss of generality by spherical symmetry. We restrict the radiative problem to ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 and include both metric parities, the gauge field, and the pseudoscalar. Magnetic charge mixes the parity sectors. After eliminating the nondynamical fields, we use the five dynamical perturbations
y = ( V , Ψ , δ ϕ , χ , δ A ) T . (22) \mathbf{y}=(V,\Psi,\delta\phi,\chi,\delta A)^T. \tag*{(22)} y = ( V , Ψ , δ ϕ , χ , δ A ) T . (22)
For a single harmonic Y ℓ ≡ Y ℓ 0 Y_\ell\equiv Y_{\ell0} Y ℓ ≡ Y ℓ 0 , the scalar-field and odd-parity electromagnetic perturbations are defined by
δ ϕ f u l l = δ ϕ ( t , r ) Y ℓ , \delta\phi_{\mathrm{full}}=\delta\phi(t,r)Y_\ell, δ ϕ full = δ ϕ ( t , r ) Y ℓ ,
δ A φ = − δ A ( t , r ) sin θ ∂ θ Y ℓ . (23) \delta A_{\varphi}=-\delta A(t,r)\sin\theta\,\partial_\theta Y_\ell. \tag*{(23)} δ A φ = − δ A ( t , r ) sin θ ∂ θ Y ℓ . (23)
The even-parity electromagnetic perturbations are written as δ A t = δ A 0 ( t , r ) Y ℓ \delta A_t=\delta A_0(t,r)Y_\ell δ A t = δ A 0 ( t , r ) Y ℓ and δ A r = δ A 1 ( t , r ) Y ℓ \delta A_r=\delta A_1(t,r)Y_\ell δ A r = δ A 1 ( t , r ) Y ℓ . The remaining three reduced fields are
V = δ A 0 ′ − δ A ˙ 1 + s 2 H 0 + s 3 H 2 + L s 4 δ A + s 6 δ ϕ 2 s 1 , V=\delta A'_0-\dot{\delta A}_1+\frac{s_2H_0+s_3H_2+Ls_4\delta A+s_6\delta\phi}{2s_1}, V = δ A 0 ′ − δ A ˙ 1 + 2 s 1 s 2 H 0 + s 3 H 2 + L s 4 δ A + s 6 δ ϕ ,
Ψ = r H 2 − L h 1 , \Psi=rH_2-Lh_1, Ψ = r H 2 − L h 1 ,
χ = r V ˙ − r Q ′ + 2 Q + p 2 2 p 1 δ A . (24) \chi=r\dot{V}-rQ'+2Q+\frac{p_2}{2p_1}\delta A. \tag*{(24)} χ = r V ˙ − r Q ′ + 2 Q + 2 p 1 p 2 δ A . (24)
Here H 0 , H 2 , h 1 H_0,H_2,h_1 H 0 , H 2 , h 1 and Q , V Q,V Q , V denote the even- and odd-parity metric perturbations, respectively. The functions s i ( r ) s_i(r) s i ( r ) and p i ( r ) p_i(r) p i ( r ) are background-dependent coefficients of the unreduced quadratic action. Their explicit expressions are given in Eqs. (B.11) and (B.12) of Appendix B, which also specifies the harmonic conventions and the constraint elimination. Thus V V V carries the electric gauge degree of freedom, Ψ \Psi Ψ and χ \chi χ carry the two gravitational degrees of freedom, and δ ϕ \delta\phi δ ϕ and δ A \delta A δ A carry the pseudoscalar and odd electromagnetic degrees of freedom. These assignments refer to the electric limit; for nonzero axion coupling and magnetic charge, the five amplitudes form a coupled system. A superscript T T T denotes matrix transpose.
The reduced quadratic action takes the form
S ( 2 ) = ∫ d t d r [ y ˙ T K y ˙ + y ′ T G y ′ + 2 y T C y ′ + y T M 0 y ] . (25) S^{(2)}=\int dt\,dr\left[\dot{\boldsymbol{y}}^{T}K\dot{\boldsymbol{y}}+\boldsymbol{y}'^{T}G\boldsymbol{y}'+2\boldsymbol{y}^{T}C\boldsymbol{y}'+\boldsymbol{y}^{T}M_{0}\boldsymbol{y}\right]. \tag*{(25)} S ( 2 ) = ∫ d t d r [ y ˙ T K y ˙ + y ′ T G y ′ + 2 y T C y ′ + y T M 0 y ] . (25)
Here K K K , G G G , and M 0 M_{0} M 0 are the real symmetric 5 × 5 5\times5 5 × 5 kinetic, gradient, and potential matrices, respectively, while the real matrix C C C encodes the couplings linear in radial derivatives. All four matrices depend on the background and L = ℓ ( ℓ + 1 ) L=\ell(\ell+1) L = ℓ ( ℓ + 1 ) and are independent of frequency. On the background field equations, the reduced matrices obey the exact finite-L L L relation G = − N h 2 K G=-Nh^{2}K G = − N h 2 K . We verify this identity by extracting K K K and G G G independently from the quadratic action.
For perturbations proportional to e − i ω t e^{-i\omega t} e − iω t , variation of Eq. (25) gives
− ( G y ′ ) ′ + ( C − C T ) y ′ + ( M 0 − C T ′ ) y + ω 2 K y = 0. (26) -(G\boldsymbol{y}')'+(C-C^{T})\boldsymbol{y}'+(M_{0}-C^{T\prime})\boldsymbol{y}+\omega^{2}K\boldsymbol{y}=0. \tag*{(26)} − ( G y ′ ) ′ + ( C − C T ) y ′ + ( M 0 − C T ′ ) y + ω 2 K y = 0. (26)
Our numerical formulation retains the first-derivative couplings C − C T C-C^{T} C − C T explicitly and solves the coupled system in this field basis. Appendix B presents the quadratic coefficients obtained from Eq. (2.3), the elimination of the nondynamical perturbations, and the five coupled radial equations.
The identity G = − N h 2 K G=-Nh^{2}K G = − N h 2 K implies that all five modes have unit squared radial propagation speeds in a local orthonormal frame throughout the regular exterior. We introduce the tortoise coordinate r ∗ r_{*} r ∗ and rescaled amplitudes ψ \boldsymbol{\psi} ψ by
r ∗ ′ = 1 h N , r_{*}'=\frac{1}{h\sqrt{N}}, r ∗ ′ = h N 1 ,
y = diag ( r − 2 , r r − r h , r − 1 , 1 , 1 ) ψ . (27) \boldsymbol{y}=\operatorname{diag}\left(r^{-2},\frac{r}{r-r_{h}},r^{-1},1,1\right)\boldsymbol{\psi}. \tag*{(27)} y = diag ( r − 2 , r − r h r , r − 1 , 1 , 1 ) ψ . (27)
Near a regular nonextremal horizon, the leading equations in these variables reduce to
∂ ∗ 2 ψ + ω 2 ψ ≃ 0 , r → r h , (28) \partial_{*}^{2}\boldsymbol{\psi}+\omega^{2}\boldsymbol{\psi}\simeq0,\qquad r\to r_{h}, \tag*{(28)} ∂ ∗ 2 ψ + ω 2 ψ ≃ 0 , r → r h , (28)
where ∂ ∗ = ∂ / ∂ r ∗ \partial_{*}=\partial/\partial r_{*} ∂ ∗ = ∂ / ∂ r ∗ . With the Fourier convention e − i ω t e^{-i\omega t} e − iω t , all five components therefore share the ingoing phase e − i ω r ∗ e^{-i\omega r_{*}} e − iω r ∗ at the future horizon. Asymptotic flatness and the massless nature of the fields give the outgoing phase e i ω r ∗ e^{i\omega r_{*}} e iω r ∗ at infinity. The QNM boundary conditions are thus
ψ ∼ e − i ω r ∗ ( r → r h ) , \boldsymbol{\psi}\sim e^{-i\omega r_{*}}\qquad(r\to r_{h}), ψ ∼ e − iω r ∗ ( r → r h ) ,
ψ ∼ e i ω r ∗ ( r → ∞ ) . (29) \boldsymbol{\psi}\sim e^{i\omega r_{*}}\qquad(r\to\infty). \tag*{(29)} ψ ∼ e iω r ∗ ( r → ∞ ) . (29)
These phases are determined by the radial characteristics; the angular propagation speeds need not all be unity.
At complex frequency, the outgoing branch is understood by analytic continuation of the outgoing wave solution. Using the compactified coordinate x = r h / r x=r_{h}/r x = r h / r , we factor out the boundary phases as ψ = e i ω s ( r ) u ( x ) \boldsymbol{\psi}=e^{i\omega s(r)}\boldsymbol{u}(x) ψ = e iω s ( r ) u ( x ) , with
s ( r ) = r + 2 M log ( r / r h ) − a H log ( 1 − r h / r ) , s(r)=r+2M\log(r/r_{h})-a_{H}\log(1-r_{h}/r), s ( r ) = r + 2 M log ( r / r h ) − a H log ( 1 − r h / r ) ,
a H = 1 h h N h . (30) a_{H}=\frac{1}{h_{h}\sqrt{N_{h}}}. \tag*{(30)} a H = h h N h 1 . (30)
Since s ≃ − r ∗ s\simeq-r_{*} s ≃ − r ∗ near the horizon and s ≃ r ∗ s\simeq r_{*} s ≃ r ∗ at infinity, this factor incorporates both QNM boundary phases and leaves u \boldsymbol{u} u regular at the endpoints. The compactified domain includes infinity, so no finite outer boundary is introduced.
Applying Chebyshev–Gauss collocation with N c N_{c} N c nodes to the five coupled radial equations gives the quadratic eigenvalue problem
[ A 0 + ω A 1 + ω 2 A 2 ] u = 0. (31) \left[A_{0}+\omega A_{1}+\omega^{2}A_{2}\right]\boldsymbol{u}=0. \tag*{(31)} [ A 0 + ω A 1 + ω 2 A 2 ] u = 0. (31)
Here the frequency-independent matrices A k A_{k} A k (k = 0 , 1 , 2 k=0,1,2 k = 0 , 1 , 2 ) represent the discretized radial operators, and u \boldsymbol{u} u collects the nodal values of the five regular amplitudes. We linearize this quadratic system into an equivalent generalized eigenvalue problem. We assess frequency convergence by varying N c N_{c} N c and track the convergent mode families through neighboring points in parameter space. The five plotted families are oscillatory modes continued from the fundamental RN modes. Purely imaginary roots are examined in a separate unfiltered search, as discussed in § 2 D.
The plotted spectra use N c = 36 N_{c}=36 N c = 36 , except for Figure 5 and Figure 6 , which use N c = 48 N_{c}=48 N c = 48 . Selected configurations are repeated with 28, 48, and 60 nodes and tighter background integration tolerances. The first- and second-order NED response coefficients are obtained using centered finite differences with fourth-order accuracy. Each stencil point uses a separately computed background, with the total charge returned to maintain the prescribed Q / M Q/M Q / M . We assess parameter-step sensitivity separately from frequency convergence.
Figure 5. NED-induced shifts of the five fundamental ℓ = 2 \ell= 2 ℓ = 2 QNM families at fixed Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 , g ˉ = 5 \bar{g} = 5 g ˉ = 5 , and η = 0.4 \eta= 0.4 η = 0.4 . From left to right, the columns correspond to ( α ^ , β ^ ) = ( λ , 0 ) , ( 0 , λ ) , ( 4 λ / 7 , λ ) (\hat{\alpha},\hat{\beta}) = (\lambda,0),(0,\lambda),(4\lambda/7,\lambda) ( α ^ , β ^ ) = ( λ , 0 ) , ( 0 , λ ) , ( 4 λ /7 , λ ) , with λ \lambda λ on the horizontal axis. The upper panels show the real-frequency shifts 10 4 Re Δ Ω j 10^4\operatorname{Re}\Delta\Omega_j 1 0 4 Re Δ Ω j , and the lower panels show the damping-rate shifts 10 4 Δ Γ j 10^4\Delta\Gamma_j 1 0 4 Δ Γ j , where Δ Γ j = − Im Δ Ω j \Delta\Gamma_j = -\operatorname{Im}\Delta\Omega_j Δ Γ j = − Im Δ Ω j . Positive damping-rate shifts indicate faster decay. All shifts are dimensionless and are measured relative to the corresponding mode at α ^ = β ^ = 0 \hat{\alpha} = \hat{\beta} = 0 α ^ = β ^ = 0 , with the same Q / M Q/M Q / M , g ˉ \bar{g} g ˉ , and η \eta η , as defined in Eq. (43) . Points are obtained by solving the full coupled QNM eigenvalue problem at each finite-coefficient point using N c = 48 N_c = 48 N c = 48 radial collocation points; lines connect the computed points.
Figure 6. NED-induced damping-rate changes and exterior NED amplitudes at fixed ℓ = 2 \ell= 2 ℓ = 2 , Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 , g ˉ = 5 \bar{g} = 5 g ˉ = 5 , and η = 0.1 \eta= 0.1 η = 0.1 . The horizontal axis is λ \lambda λ for the three sequences ( α ^ , β ^ ) = ( λ , 0 ) , ( 0 , λ ) , ( 4 λ / 7 , λ ) (\hat{\alpha},\hat{\beta}) = (\lambda, 0),(0,\lambda),(4\lambda/7,\lambda) ( α ^ , β ^ ) = ( λ , 0 ) , ( 0 , λ ) , ( 4 λ /7 , λ ) , extending to λ = 0.16 \lambda= 0.16 λ = 0.16 . Left: the odd-gravitational damping-rate change δ Γ o d d \delta\Gamma_{\mathrm{odd}} δ Γ odd in percent, defined in Eq. (44) , relative to the corresponding NED-free mode at the same Q / M Q/M Q / M , g ˉ \bar{g} g ˉ , and η \eta η . Right: ϵ N E D max \epsilon_{\mathrm{NED}}^{\max} ϵ NED m a x , the maximum over the BH exterior of the NED amplitude defined in Eq. (18) , evaluated on each background. Dotted horizontal lines indicate damping-rate changes of ± 1 % \pm1\% ± 1% in the left panel and the adopted NED amplitude bound of 0.05 in the right panel. Points use directly computed QNM frequencies with N c = 48 N_c = 48 N c = 48 radial collocation points and backgrounds recomputed at each choice of NED coefficients; lines connect the computed points.
The background solver evolves the mass function m = r ( 1 − h ) / 2 m=r(1-h)/2 m = r ( 1 − h ) /2 , ϕ \phi ϕ , d ϕ / d x d\phi/dx d ϕ / d x , and log N \log N log N to x = 0 x=0 x = 0 , where the geometric mass is obtained directly as M = m ∣ x = 0 M=m|_{x=0} M = m ∣ x = 0 . A near-horizon series supplies the initial conditions. Shooting determines the horizon scalar value and total charge simultaneously by imposing ϕ ( ∞ ) = 0 \phi(\infty)=0 ϕ ( ∞ ) = 0 and the prescribed Q / M Q/M Q / M . Among the production backgrounds tested, the largest residual in either shooting condition is below 1.5 × 10 − 13 1.5\times10^{-13} 1.5 × 1 0 − 13 .
For 25 spectral benchmark configurations, including the strong-coupling electric modes, the largest absolute changes in Ω = M ω \Omega=M\omega Ω = M ω between 36 and 48 nodes and between 48 and 60 nodes are 3.1 × 10 − 9 3.1\times10^{-9} 3.1 × 1 0 − 9 and 6.0 × 10 − 11 6.0\times10^{-11} 6.0 × 1 0 − 11 , respectively. Tightening the background tolerance from 2 × 10 − 11 2\times10^{-11} 2 × 1 0 − 11 to 4 × 10 − 13 4\times10^{-13} 4 × 1 0 − 13 changes the checked frequencies by at most 1.1 × 10 − 11 1.1\times10^{-11} 1.1 × 1 0 − 11 . We use these differences as empirical measures of numerical sensitivity. For purely electric backgrounds with α = β = 0 \alpha=\beta=0 α = β = 0 , the odd-sector frequencies are also checked using the independent three-field potential formulation and complex-contour ODE integration.
The spectral solver is implemented in Python using NumPy and SciPy in double precision. The background equations and local quadratic-action matrices are implemented separately, and the spectral operator is constructed directly from them. For purely electric backgrounds, we solve the even two-field and odd three-field blocks separately. This identifies the degenerate RN parity partners within their respective sectors without assuming simple eigenvalues of the full five-field system.
To validate these spectral results, we also use an independent two-sided numerical integration-and-matching method. The radial equations are integrated outward from a near-horizon point with ingoing boundary conditions and inward from a large radius with outgoing boundary conditions. The horizon integration uses a fourth-order expansion initialized at 1.001 r h 1.001r_h 1.001 r h , while the outer integration uses an outgoing inverse-radius expansion initialized at 50 r h 50r_h 50 r h , 55 r h 55r_h 55 r h , or 60 r h 60r_h 60 r h . The QNM frequencies are determined by matching the two sets of solutions at 8 r h 8r_h 8 r h . The outgoing expansion is retained through order 14, increased to order 16 for the sensitive strong-coupling odd-gravitational benchmarks. The working precision is 45 decimal digits, with accuracy and precision goals of 22 and 19, respectively. Because the expansion at infinity is asymptotic, its truncation order and initialization radius are varied independently. The two implementations agree on the tested five-family splittings and NED shifts along the α \alpha α -only, β \beta β -only, and equal-coefficient directions within the measured numerical variations.
Appendix C compares the production radial operators with an independent symbolic reference and gives the RN parity-pair tests and frequency-response convergence checks.
Breaking of isospectrality without nonlinear electrodynamics
This section follows five oscillatory QNM families, resolves five distinct frequencies at representative parameters, and establishes the breaking of RN axial–polar isospectrality, extending the two-family analysis of Ref. [64 ] . We set α ^ = β ^ = 0 \hat{\alpha}=\hat{\beta}=0 α ^ = β ^ = 0 throughout, including § 2 D, to isolate the axion-induced effects. Five dynamical degrees of freedom do not by themselves imply five distinct frequencies: each eigenmode can involve several perturbation components. We identify the families by continuous tracking from the fundamental purely electric RN modes. The additional growing modes in § 2 D show that these five families do not exhaust the spectrum.
At g ˉ = 0 \bar{g}=0 g ˉ = 0 , the massless scalar decouples, while the gravitational and EM fundamental modes each form an isospectral axial–polar pair. The five families therefore correspond to only three distinct frequencies at the reference charge considered here. Establishing the breaking of isospectrality requires resolving the frequency splitting within each RN parity pair after the axion coupling is introduced. This parity degeneracy is distinct from the independence of the RN spectrum from the electric–magnetic charge composition at fixed Q / M Q/M Q / M .
Electric RN background and axion-induced splitting
For q M = 0 q_M=0 q M = 0 and α = β = 0 \alpha=\beta=0 α = β = 0 , Eq. (8) and horizon regularity give ϕ ′ = 0 \phi'=0 ϕ ′ = 0 . The asymptotic condition ϕ ( ∞ ) = 0 \phi(\infty)=0 ϕ ( ∞ ) = 0 then fixes ϕ ( r ) = 0 \phi(r)=0 ϕ ( r ) = 0 . With N ( ∞ ) = 1 N(\infty)=1 N ( ∞ ) = 1 , Eq. (6) gives N = 1 N=1 N = 1 , and Eqs. (5) , (7) , and (9) yield the electric RN solution,
N = 1 , h = f = 1 − 2 M r + Q 2 r 2 . (32) N=1,\qquad h=f=1-\frac{2M}{r}+\frac{Q^2}{r^2}. \tag*{(32)} N = 1 , h = f = 1 − r 2 M + r 2 Q 2 . (32)
The even-parity two-field system coincides with that of Einstein–Maxwell theory. By contrast, the odd-parity three-field system retains the quadratic interaction g δ ϕ δ G g\,\delta\phi\,\delta\mathcal{G} g δ ϕ δ G , despite the vanishing background axion. This interaction couples the scalar and EM perturbations, while the background electric charge couples the EM and gravitational perturbations. The odd-parity frequencies can therefore differ from their even-parity counterparts even in the absence of background scalar hair and NED corrections.
To display this coupling explicitly, we introduce Z = ( Z g r a v , Z E M , Z ϕ ) T Z=(Z_{\rm grav},Z_{\rm EM},Z_{\phi})^{\mathsf{T}} Z = ( Z grav , Z EM , Z ϕ ) T by rescaling the odd-parity subset ( χ , δ A , δ ϕ ) (\chi,\delta A,\delta\phi) ( χ , δ A , δ ϕ ) of Eq. (22) :
Z g r a v = M P l L 2 ( L − 2 ) χ , Z E M = L δ A , Z ϕ = r δ ϕ . (33) Z_{\rm grav}=M_{\rm Pl}\sqrt{\frac{L}{2(L-2)}}\,\chi,\qquad
Z_{\rm EM}=\sqrt{L}\,\delta A,\qquad
Z_{\phi}=r\,\delta\phi. \tag*{(33)} Z grav = M Pl 2 ( L − 2 ) L χ , Z EM = L δ A , Z ϕ = r δ ϕ . (33)
These variables describe the same three degrees of freedom in a canonically normalized basis. They satisfy
d 2 Z d r ∗ 2 + ( ω 2 I 3 − V R N ) Z = 0 , (34) \frac{\mathrm{d}^2 Z}{\mathrm{d}r_*^2}+\left(\omega^2 I_3-V_{\rm RN}\right)Z=0, \tag*{(34)} d r ∗ 2 d 2 Z + ( ω 2 I 3 − V RN ) Z = 0 , (34)
where d r ∗ / d r = f − 1 \mathrm{d}r_*/\mathrm{d}r=f^{-1} d r ∗ / d r = f − 1 and I 3 I_3 I 3 is the 3 × 3 3\times3 3 × 3 identity matrix. In this gravitational, EM, and scalar basis,
V R N = f r 2 [ L I 3 + U r + Q 2 r 2 diag ( 4 , 4 , − 2 ) ] , V_{\rm RN}=\frac{f}{r^2}\left[L I_3+\frac{U}{r}+\frac{Q^2}{r^2}\operatorname{diag}(4,4,-2)\right], V RN = r 2 f [ L I 3 + r U + r 2 Q 2 diag ( 4 , 4 , − 2 ) ] ,
U = ( − 6 M 2 Q L − 2 0 2 Q L − 2 0 2 L g ˉ Q 0 2 L g ˉ Q 2 M ) . (35) U=\begin{pmatrix}
-6M & 2Q\sqrt{L-2} & 0\\
2Q\sqrt{L-2} & 0 & \sqrt{2L}\,\bar{g}Q\\
0 & \sqrt{2L}\,\bar{g}Q & 2M
\end{pmatrix}. \tag*{(35)} U = − 6 M 2 Q L − 2 0 2 Q L − 2 0 2 L g ˉ Q 0 2 L g ˉ Q 2 M . (35)
Here diag ( 4 , 4 , − 2 ) \operatorname{diag}(4,4,-2) diag ( 4 , 4 , − 2 ) denotes the diagonal 3 × 3 3\times3 3 × 3 matrix with entries 4, 4, and − 2 -2 − 2 in the gravitational, EM, and scalar order, respectively. The factor 2 \sqrt{2} 2 in the scalar–EM coupling follows from q E = 2 M P l Q q_E=\sqrt{2}M_{\rm Pl}Q q E = 2 M Pl Q and g ˉ = g M P l \bar{g}=gM_{\rm Pl} g ˉ = g M Pl in our normalization. After accounting for coupling normalizations and field-sign conventions, this system agrees with the established pure-electric Einstein–Maxwell–axion (EMA) perturbation equations [65 , 66 , 67 ] .
Table 1. Frequencies Ω = M ω \Omega=M\omega Ω = M ω of the five oscillatory QNM families continued from the fundamental RN modes in the minimal axion model at ℓ = 2 \ell=2 ℓ = 2 , Q / M = 0.6 Q/M=0.6 Q / M = 0.6 , and α ^ = β ^ = 0 \hat{\alpha}=\hat{\beta}=0 α ^ = β ^ = 0 . The RN reference (g ˉ = 0 \bar{g}=0 g ˉ = 0 , η = 0 \eta=0 η = 0 ) has two isospectral axial–polar pairs, one gravitational and one EM, together with a distinct scalar frequency, giving three distinct values for the five families. At g ˉ = 5 \bar{g}=5 g ˉ = 5 , all five frequencies are distinct for both η = 0 \eta=0 η = 0 and η = 0.4 \eta=0.4 η = 0.4 . In the purely electric case (η = 0 \eta=0 η = 0 ), the background remains RN and the even-parity frequencies are unchanged, whereas the axion coupling shifts the odd-parity frequencies. At η = 0.4 \eta=0.4 η = 0.4 , the row labels identify families continued from the purely electric limit. The tabulated frequencies are computed with N c = 36 N_c=36 N c = 36 and rounded to eight decimal places; convergence is checked using N c = 48 N_c=48 N c = 48 and 60.
Table I and Figure 1 show the five fundamental QNM families. At Q / M = 0.6 Q/M=0.6 Q / M = 0.6 , the RN reference column of Table I corresponds to g ˉ = 0 \bar{g}=0 g ˉ = 0 and η = 0 \eta=0 η = 0 . The even- and odd-gravitational modes share the frequency 0.38621747 − 0.08981367 i 0.38621747 - 0.08981367i 0.38621747 − 0.08981367 i , while the even- and odd-EM modes share 0.51201092 − 0.09801663 i 0.51201092 - 0.09801663i 0.51201092 − 0.09801663 i . These pairwise equalities express RN axial–polar isospectrality. The decoupled scalar perturbation δ φ \delta\varphi δ φ has the fundamental frequency Ω = 0.51738763 − 0.09832464 i \Omega=0.51738763 - 0.09832464i Ω = 0.51738763 − 0.09832464 i , distinct from the gravitational and EM frequencies. Thus, the five families correspond to three distinct complex frequencies in the RN reference limit.
Figure 1. Fundamental ℓ = 2 \ell= 2 ℓ = 2 QNM frequencies Ω = M ω \Omega= M\omega Ω = M ω at α ^ = β ^ = 0 \hat{\alpha} = \hat{\beta} = 0 α ^ = β ^ = 0 and Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 , shown through their real parts and damping rates Γ = − Im Ω \Gamma= -\operatorname{Im}\Omega Γ = − Im Ω . Top panels: dependence on the axion coupling g ˉ \bar{g} g ˉ in the purely electric case (η = 0 \eta= 0 η = 0 ). At g ˉ = 0 \bar{g} = 0 g ˉ = 0 , the gravitational and EM axial–polar pairs are each degenerate. The axion coupling lifts both degeneracies by shifting the odd-parity frequencies, while the even-parity frequencies remain unchanged. The background metric remains RN with ϕ = 0 \phi= 0 ϕ = 0 throughout this sequence. Bottom panels: dependence on the magnetic fraction η \eta η at g ˉ = 5 \bar{g} = 5 g ˉ = 5 , tracing five distinct frequency branches into the dyonic regime. At nonzero η \eta η , the labels “even” and “odd” refer to the parity of the corresponding purely electric limits. Each curve represents an eigenfrequency branch of the full coupled perturbation problem. Lines connect independently computed points.
Turning on the axion coupling lifts this parity degeneracy. Table I lists the resulting spectra for ( g ˉ , η ) = ( 5 , 0 ) (\bar{g},\eta)=(5,0) ( g ˉ , η ) = ( 5 , 0 ) and ( 5 , 0.4 ) (5,0.4) ( 5 , 0.4 ) , each of which contains five distinct complex frequencies. In the purely electric case ( g ˉ , η ) = ( 5 , 0 ) (\bar{g},\eta)=(5,0) ( g ˉ , η ) = ( 5 , 0 ) , the background metric remains RN and the two even-parity frequencies retain their reference values. The axion coupling nevertheless modifies the odd-parity operator. The odd-gravitational frequency shifts to Ω g r a v , o = 0.32447691 − 0.09057384 i \Omega_{\mathrm{grav,o}}=0.32447691 - 0.09057384i Ω grav , o = 0.32447691 − 0.09057384 i , breaking its degeneracy with the even-gravitational mode. Likewise, the odd-EM frequency shifts to Ω E M , o = 0.41007348 − 0.09123271 i \Omega_{\mathrm{EM,o}}=0.41007348 - 0.09123271i Ω EM , o = 0.41007348 − 0.09123271 i , while the even-EM frequency remains at Ω E M , e = 0.51201092 − 0.09801663 i \Omega_{\mathrm{EM,e}}=0.51201092 - 0.09801663i Ω EM , e = 0.51201092 − 0.09801663 i . Together with the scalar branch, these modes yield five distinct fundamental frequencies. Thus, the purely electric system already exhibits the breaking of RN axial–polar isospectrality at α ^ = β ^ = 0 \hat{\alpha}=\hat{\beta}=0 α ^ = β ^ = 0 , without NED corrections or background scalar hair.
The breaking of RN axial–polar isospectrality originates from axion-induced mixing in the odd-parity sector. On the purely electric background, the axion perturbation couples to the angular curl of the EM perturbation, which has odd parity. The background electric charge also couples the odd-parity EM and gravitational perturbations. Through these interactions, the axion modifies the eigenfrequencies of the entire odd three-field system, while the even-parity operator remains unchanged. The splitting therefore arises from the coupled perturbation equations despite the unchanged background geometry.
Starting from a simple isolated QNM of the odd-parity block in the g ˉ = 0 \bar{g}=0 g ˉ = 0 reference system, turning on a small nonzero axion coupling produces a frequency shift that generically begins at O ( g ˉ 2 ) O(\bar{g}^{2}) O ( g ˉ 2 ) . Analyticity and the symmetry ( g ˉ , δ ϕ ) → ( − g ˉ , − δ ϕ ) (\bar{g},\delta\phi)\to(-\bar{g},-\delta\phi) ( g ˉ , δ ϕ ) → ( − g ˉ , − δ ϕ ) imply that the frequency expansion contains only even powers of g ˉ \bar{g} g ˉ . The isolation condition applies within the odd-parity block; degeneracy with an even-parity mode does not invalidate the argument because the two parity sectors remain decoupled on the purely electric background. In the uncharged limit Q = 0 Q=0 Q = 0 , the background reduces to Schwarzschild and both the gravitational–EM and scalar–EM mixing terms vanish. The gravitational and EM parity pairs therefore remain isospectral.
Five coupled dyonic families
We next increase the magnetic fraction η \eta η at fixed Q / M = 0.6 Q/M=0.6 Q / M = 0.6 and g ˉ = 5 \bar{g}=5 g ˉ = 5 , keeping α ^ = β ^ = 0 \hat{\alpha}=\hat{\beta}=0 α ^ = β ^ = 0 . Magnetic charge sources a background axion profile, changes the geometry and electric field, and couples the two perturbation blocks that were separated in the purely electric case. The resulting modes generally involve gravitational, EM, and axion perturbations. The labels even and odd are inherited from the corresponding purely electric modes. At nonzero magnetic charge, the even- and odd-parity variables are coupled and must be solved together.
The bottom panels of Figure 1 follow all five frequency branches from η = 0 \eta=0 η = 0 into this coupled regime. At η = 0.4 \eta=0.4 η = 0.4 , Table I gives five distinct complex frequencies, extending the two-family analysis of Ref. [64 ] to all five fundamental branches. Their magnetic frequency shifts are quantified in § 2 D. This result concerns the sampled parameters; special limits or accidental crossings can still produce degeneracies.
For the three parameter sets in Table I, we test convergence using the collocation method of Sec. III at N c = 36 , 48 , 60 N_c=36,48,60 N c = 36 , 48 , 60 . Writing the frequency of family j j j as Ω j ( N c ) \Omega_j^{(N_c)} Ω j ( N c ) , the largest changes over all five families are below 9.2 × 10 − 12 9.2\times10^{-12} 9.2 × 1 0 − 12 between 36 and 48 nodes and 1.4 × 10 − 11 1.4\times10^{-11} 1.4 × 1 0 − 11 between 48 and 60 nodes. At fixed N c = 36 N_c=36 N c = 36 , tightening the background-integration tolerance changes the frequencies by less than 3.4 × 10 − 12 3.4\times10^{-12} 3.4 × 1 0 − 12 .
We compare the numerical variations above with the frequency differences between distinct families at fixed physical parameters. For the two g ˉ = 5 \bar{g} = 5 g ˉ = 5 cases with η = 0 \eta= 0 η = 0 and η = 0.4 \eta= 0.4 η = 0.4 , we define the smallest separation in the complex-frequency plane by
d min = min j < k ∣ Ω j − Ω k ∣ , (36) d_{\min} = \min_{j<k} \lvert\Omega_j - \Omega_k \rvert, \tag*{(36)} d m i n = j < k min ∣ Ω j − Ω k ∣ , (36)
where j , k j, k j , k label the five QNM families in Table I. At η = 0 \eta= 0 η = 0 , the closest pair consists of the even-gravitational and odd-EM modes, with d min ≃ 2.39 × 10 − 2 d_{\min} \simeq2.39 \times10^{-2} d m i n ≃ 2.39 × 1 0 − 2 . At η = 0.4 \eta= 0.4 η = 0.4 , the closest pair consists of the families continued from the even- and odd-gravitational RN modes, with d min ≃ 1.77 × 10 − 2 d_{\min} \simeq1.77 \times10^{-2} d m i n ≃ 1.77 × 1 0 − 2 . Both minimum separations greatly exceed the measured numerical variations, supporting the identification of five distinct frequencies at these benchmarks. We next discuss an exact EM–scalar degeneracy in the electric RN spectrum at g ˉ = 0 \bar{g} = 0 g ˉ = 0 .
An exact RN electromagnetic–scalar degeneracy
The five-family spectra above were evaluated at Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 . We now vary Q / M Q/M Q / M in the purely electric RN reference system (η = 0 \eta= 0 η = 0 ), keeping g ˉ = α ^ = β ^ = 0 \bar{g} = \hat{\alpha} = \hat{\beta} = 0 g ˉ = α ^ = β ^ = 0 , to identify the charge ratio at which the odd-EM and odd-scalar modes have identical complex QNM frequencies. In this limit, the scalar perturbation decouples from the gravitational–EM system, so the two families obey separate radial equations. Nevertheless, their frequencies can coincide at a special charge, with equal oscillation frequencies and damping rates. This degeneracy occurs within the odd-parity sector and is distinct from the axial–polar degeneracy discussed in § 2 A. Identifying this special charge also helps explain how the EM and scalar frequencies respond when a small nonzero axion coupling is introduced and the two modes become coupled. Rout [93 ] derived this degeneracy in the decoupled RN sector of dynamical Chern–Simons gravity. Equation (67) of that work gives Q c / M = 2 L + 1 / ( L + 2 ) Q_{\mathrm{c}}/M = 2\sqrt{L+1}/(L+2) Q c / M = 2 L + 1 / ( L + 2 ) , with L = ℓ ( ℓ + 1 ) L=\ell(\ell+1) L = ℓ ( ℓ + 1 ) , which reduces to 7 / 4 \sqrt{7}/4 7 /4 for ℓ = 2 \ell=2 ℓ = 2 . The overlap concerns the common Einstein–Maxwell plus free massless-scalar limit; the finite-coupling interactions differ. We retain the explicit ℓ = 2 \ell=2 ℓ = 2 Darboux map below to interpret the mixing in our axion–photon theory.
To recover this coincidence, we use the potential matrix V R N V_{\mathrm{RN}} V RN in Eq. (35) at g ˉ = 0 \bar{g}=0 g ˉ = 0 . We denote by V E V_{\mathrm{E}} V E the EM eigenpotential obtained by diagonalizing its gravitational–EM 2 × 2 2\times2 2 × 2 block, and by V ϕ V_{\phi} V ϕ its third diagonal entry, which governs the decoupled massless-scalar perturbation. For ℓ = 2 \ell=2 ℓ = 2 (L = 6 L=6 L = 6 ), we determine the special charge Q ⋆ Q_{\star} Q ⋆ by requiring V E V_{\mathrm{E}} V E and V ϕ V_{\phi} V ϕ to admit the Darboux representation in Eq. (38) below. Matching the coefficients of inverse powers of r r r gives 9 M 2 + 16 Q ⋆ 2 = 4 M \sqrt{9M^2+16Q_{\star}^2}=4M 9 M 2 + 16 Q ⋆ 2 = 4 M . The resulting charge and corresponding outer and inner horizon radii are
Q ⋆ M = 7 4 , r + = 7 M 4 , r − = M 4 . (37) \frac{Q_{\star}}{M}=\frac{\sqrt{7}}{4},\qquad r_{+}=\frac{7M}{4},\qquad r_{-}=\frac{M}{4}. \tag*{(37)} M Q ⋆ = 4 7 , r + = 4 7 M , r − = 4 M . (37)
At this charge, the two potentials can be expressed as
V E = W 2 + d W d r ⋆ − 144 M 2 , V_{\mathrm{E}}=W^2+\frac{\mathrm{d}W}{\mathrm{d}r_{\star}}-\frac{144}{M^2}, V E = W 2 + d r ⋆ d W − M 2 144 ,
V ϕ = W 2 − d W d r ⋆ − 144 M 2 . (38) V_{\phi}=W^2-\frac{\mathrm{d}W}{\mathrm{d}r_{\star}}-\frac{144}{M^2}. \tag*{(38)} V ϕ = W 2 − d r ⋆ d W − M 2 144 . (38)
where a common function W W W is given by
W = 12 M + M 4 r 2 − Q ⋆ 2 r 3 . (39) W=\frac{12}{M}+\frac{M}{4r^2}-\frac{Q_{\star}^2}{r^3}. \tag*{(39)} W = M 12 + 4 r 2 M − r 3 Q ⋆ 2 . (39)
Here ψ E \psi_{\mathrm{E}} ψ E and ψ ϕ \psi_{\phi} ψ ϕ denote the individual EM and scalar radial master functions, respectively. The function ψ E \psi_{\mathrm{E}} ψ E is a linear combination of Z g r a v Z_{\mathrm{grav}} Z grav and Z E M Z_{\mathrm{EM}} Z EM obtained by diagonalizing the gravitational–EM block. The two radial solutions are connected by the first-order Darboux transformation
ψ E = ( d d r ⋆ + W ) ψ ϕ . (40) \psi_{\mathrm{E}}=\left(\frac{\mathrm{d}}{\mathrm{d}r_{\star}}+W\right)\psi_{\phi}. \tag*{(40)} ψ E = ( d r ⋆ d + W ) ψ ϕ . (40)
The function W W W is called the superpotential. Since W → 12 / M W\to12/M W → 12/ M at both the horizon and infinity, this transformation preserves the ingoing and outgoing character of the QNM solutions. For ω ≠ ± 12 i / M \omega\ne\pm12i/M ω = ± 12 i / M , the inverse transformation also exists, establishing equality of the EM and scalar spectra away from these exceptional factorization frequencies.
For the common fundamental mode, we obtain numerically
M ω E ( 0 ) = M ω ϕ ( 0 ) = 0.52618687 − 0.09850443 i , (41) M\omega_{\mathrm{E}}^{(0)}=M\omega_{\phi}^{(0)}=0.52618687-0.09850443i, \tag*{(41)} M ω E ( 0 ) = M ω ϕ ( 0 ) = 0.52618687 − 0.09850443 i , (41)
where the superscript (0) denotes the g ˉ = 0 \bar{g}=0 g ˉ = 0 reference system and the displayed value is rounded. The equality of both the oscillation frequencies and damping rates follows analytically from the Darboux relation and the preservation of the QNM boundary conditions.
Turning on a small nonzero g ˉ \bar{g} g ˉ couples the EM and scalar modes. At a fixed charge ratio, suppose that these uncoupled fundamental modes are simple and isolated within the odd-parity sector, with a nonzero complex-frequency gap, ω E ( 0 ) ≠ ω ϕ ( 0 ) \omega_{\mathrm{E}}^{(0)}\ne\omega_{\phi}^{(0)} ω E ( 0 ) = ω ϕ ( 0 ) . When the axion-induced mixing is weak compared with the magnitude of this complex-frequency difference, nondegenerate perturbation theory gives a mixing contribution proportional to g ˉ 2 / ( ω E ( 0 ) − ω ϕ ( 0 ) ) \bar{g}^2/(\omega_{\mathrm{E}}^{(0)}-\omega_{\phi}^{(0)}) g ˉ 2 / ( ω E ( 0 ) − ω ϕ ( 0 ) ) . The opposite trends of the odd-EM and scalar real frequencies and damping rates shown in panels (a) and (b) of Figure 1 at Q / M = 0.6 Q/M=0.6 Q / M = 0.6 are consistent with this mixing mechanism. Coupling to the remaining odd-parity mode also contributes, so the two-mode estimate alone does not determine the full complex shifts.
At Q = Q ⋆ Q=Q_{\star} Q = Q ⋆ , the uncoupled frequency gap vanishes, and the preceding nondegenerate expansion is no longer valid. The EM and scalar modes must instead be treated together using degenerate perturbation theory. The coupling within this two-mode subspace is linear in g ˉ \bar{g} g ˉ and produces an O ( g ˉ ) O(\bar{g}) O ( g ˉ ) frequency splitting provided the product of the projected off-diagonal couplings is nonzero, as verified numerically for the fundamental pair. This linear splitting at the degenerate point is distinct from the quadratic corrections obtained at fixed nonzero frequency gap. The degenerate case therefore lies outside the isolated-mode assumption used in § 2 A.
Magnetic response of the five families
We now study how magnetic charge modifies the five oscillatory QNM families at nonzero axion coupling, retaining Q ^ = β ^ = 0 \hat{Q}=\hat{\beta}=0 Q ^ = β ^ = 0 . Varying the magnetic fraction η \eta η at fixed Q / M Q/M Q / M and g ˉ \bar{g} g ˉ changes the axion profile, geometry, electric field, and coupled perturbation operator. For each family j j j , let Ω j ( η ) \Omega_j(\eta) Ω j ( η ) denote its frequency and Ω j ( 0 ) \Omega_j(0) Ω j ( 0 ) its purely electric reference at the same Q / M Q/M Q / M and g ˉ \bar{g} g ˉ . The family labels are assigned by continuation from the corresponding modes in the axion-decoupled electric RN limit at each fixed charge ratio. We define the magnetic frequency shift and, for η ≠ 0 \eta\ne0 η = 0 , its normalized magnitude by
Δ η Ω j = Ω j ( η ) − Ω j ( 0 ) , R j = ∣ Δ η Ω j ∣ η 2 . (42) \Delta_{\eta}\Omega_j=\Omega_j(\eta)-\Omega_j(0),\qquad\mathcal{R}_j=\frac{\lvert\Delta_{\eta}\Omega_j\rvert}{\eta^2}. \tag*{(42)} Δ η Ω j = Ω j ( η ) − Ω j ( 0 ) , R j = η 2 ∣ Δ η Ω j ∣ . (42)
The normalization reflects the symmetry ( q M , ϕ ) → ( − q M , − ϕ ) (q_M,\phi)\to(-q_M,-\phi) ( q M , ϕ ) → ( − q M , − ϕ ) . For an isolated branch analytic at η = 0 \eta=0 η = 0 , its frequency is even in η \eta η , so Δ η Ω j = O ( η 2 ) \Delta_{\eta}\Omega_j=O(\eta^2) Δ η Ω j = O ( η 2 ) . Degenerate reference modes require a coupled perturbative treatment. At finite η \eta η , R j \mathcal{R}_j R j also includes higher-order magnetic corrections.
We first consider the Q / M = 0.6 Q/M=0.6 Q / M = 0.6 , g ˉ = 5 \bar{g}=5 g ˉ = 5 sequence in the bottom panels of Figure 1 . Panel (c) shows that all five real frequencies vary with η \eta η , including the two families of electric even-parity origin whose frequencies at η = 0 \eta=0 η = 0 are independent of g ˉ \bar{g} g ˉ . Increasing η \eta η from 0 to 0.4 raises the real part of the odd-gravitational frequency and lowers that of the scalar frequency. The frequencies in Table I give R j ≃ 0.285 R_j \simeq0.285 R j ≃ 0.285 and 0.265 for the odd-gravitational and scalar families at η = 0.4 \eta= 0.4 η = 0.4 , respectively, the two largest responses at this point. Panel (d) shows the corresponding damping rates, Γ j = − Im Ω j \Gamma_j = -\operatorname{Im}\Omega_j Γ j = − Im Ω j . The odd-gravitational damping rate first decreases and then increases, crossing the even-gravitational rate, while remaining separate from the odd-EM rate over the plotted range. The real frequencies remain distinct at this crossing. Equality of the damping rates therefore does not imply the complex-frequency degeneracy established at the electric Darboux point.
To examine how the magnetic response depends on the axion coupling, we next fix Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 and η = 0.1 \eta= 0.1 η = 0.1 and compare g ˉ = 0.1 , 0.9 , 3 , 10 \bar{g} = 0.1, 0.9, 3, 10 g ˉ = 0.1 , 0.9 , 3 , 10 . The even-EM family has the largest R j R_j R j at g ˉ = 0.1 , 0.9 \bar{g} = 0.1, 0.9 g ˉ = 0.1 , 0.9 , whereas the scalar family has the largest at g ˉ = 3 , 10 \bar{g} = 3, 10 g ˉ = 3 , 10 . The odd-gravitational response also increases substantially over these sampled couplings. These changes describe the relative magnitudes of the frequency shifts; they do not by themselves establish a resonance or an exchange of mode content.
We next use the largest sampled coupling, g ˉ = 10 \bar{g} = 10 g ˉ = 10 , to test the charge dependence of these enhanced shifts and search for additional growing modes. Keeping η = 0.1 \eta= 0.1 η = 0.1 , we compare Q / M = 0.4 , 0.6 , 0.7 , 0.8 Q/M = 0.4, 0.6, 0.7, 0.8 Q / M = 0.4 , 0.6 , 0.7 , 0.8 . The families are identified separately at each charge as defined above, without continuous branch tracking in Q / M Q/M Q / M . Across these benchmarks, the odd-EM response increases from R j ≃ 0.0250 R_j \simeq0.0250 R j ≃ 0.0250 at Q / M = 0.4 Q/M = 0.4 Q / M = 0.4 to 8.65 at Q / M = 0.8 Q/M = 0.8 Q / M = 0.8 . The scalar response varies nonmonotonically, with R j ≃ 0.772 , 2.89 , 0.460 , 0.954 R_j \simeq0.772, 2.89, 0.460, 0.954 R j ≃ 0.772 , 2.89 , 0.460 , 0.954 , respectively. The odd-gravitational response is largest at Q / M = 0.7 Q/M = 0.7 Q / M = 0.7 among the four benchmarks and is smaller at Q / M = 0.8 Q/M = 0.8 Q / M = 0.8 . The five complex QNM frequencies at Q / M = 0.8 Q/M = 0.8 Q / M = 0.8 and their resolution tests are given in Table II (Appendix C).
Table II.
At this stronger coupling, the oscillatory families alone do not determine the stability of the backgrounds. An ℓ = 2 \ell= 2 ℓ = 2 search including purely imaginary frequencies at g ˉ = 10 \bar{g} = 10 g ˉ = 10 and η = 0.1 \eta= 0.1 η = 0.1 also finds Ω ≃ 0.06720 i \Omega\simeq0.06720i Ω ≃ 0.06720 i , 0.13343 i 0.13343i 0.13343 i , 0.17560 i 0.17560i 0.17560 i at Q / M = 0.6 , 0.7 , 0.8 Q/M = 0.6, 0.7, 0.8 Q / M = 0.6 , 0.7 , 0.8 , respectively (Table III in Appendix C). With the time dependence e − i ω t e^{-i\omega t} e − iω t , these positive imaginary frequencies imply exponential growth, so all three backgrounds are mode unstable. The responses quoted above at these points thus describe oscillatory modes coexisting with an instability. The growing-mode frequencies converge with radial resolution and are independently reproduced by a bound-state calculation whose solutions decay spatially toward both the horizon and infinity (Appendix C3). This finite-frequency instability is not excluded by the local high-frequency conditions in Sec. II; strong-coupling axionic instabilities are already known in the electric EMA system [65 ] . These unstable benchmarks are separate from the lower-coupling parameter sequences shown in Figure 1 –Figure 6 .
Table III.
Nonlinear-electrodynamic frequency responses
We now determine the additional frequency shifts due to the two quartic interactions. We separate background and perturbative effects on purely electric BHs, study charge composition and the RN parity-pair relation, and compute individual shifts at finite NED coefficients.
Reference spectra and frequency shifts
For each chosen ( Q / M , g ˉ , η ) (Q/M,\bar{g},\eta) ( Q / M , g ˉ , η ) , we compare the NED frequencies with those at α ^ = β ^ = 0 \hat{\alpha} = \hat{\beta} = 0 α ^ = β ^ = 0 and the same parameters. In this section the two arguments of Ω j ( α ^ , β ^ ) \Omega_j(\hat{\alpha},\hat{\beta}) Ω j ( α ^ , β ^ ) are the horizon-normalized NED coefficients defined in Eq. (14) . The index j j j identifies a QNM family, and each difference compares continuously connected eigenmodes. Let Ω j , 0 = Ω j ( 0 , 0 ) \Omega_{j,0} = \Omega_j(0,0) Ω j , 0 = Ω j ( 0 , 0 ) and Γ j , 0 = − Im Ω j , 0 \Gamma_{j,0} = -\operatorname{Im}\Omega_{j,0} Γ j , 0 = − Im Ω j , 0 denote the reference frequency and damping rate of family j j j . We define the full NED shift, the percentage changes in the real frequency and damping rate, and the magnetic modification of the NED shift as
Δ Ω j = Ω j ( α ^ , β ^ ) − Ω j ( 0 , 0 ) , (43) \Delta\Omega_j = \Omega_j(\hat{\alpha},\hat{\beta}) - \Omega_j(0,0), \tag*{(43)} Δ Ω j = Ω j ( α ^ , β ^ ) − Ω j ( 0 , 0 ) , (43)
δ R , j = 100 Re Δ Ω j Re Ω j , 0 , δ Γ , j = 100 − Im Δ Ω j Γ j , 0 , (44) \delta_{R,j} = 100\frac{\operatorname{Re}\Delta\Omega_j}{\operatorname{Re}\Omega_{j,0}}, \qquad\delta_{\Gamma,j} = 100\frac{-\operatorname{Im}\Delta\Omega_j}{\Gamma_{j,0}}, \tag*{(44)} δ R , j = 100 Re Ω j , 0 Re Δ Ω j , δ Γ , j = 100 Γ j , 0 − Im Δ Ω j , (44)
B j ( η ) = Δ Ω j ( η ) − Δ Ω j ( 0 ) . (45) B_j(\eta) = \Delta\Omega_j(\eta) - \Delta\Omega_j(0). \tag*{(45)} B j ( η ) = Δ Ω j ( η ) − Δ Ω j ( 0 ) . (45)
The damping-rate shift is Δ Γ j = Γ j − Γ j , 0 = − Im Δ Ω j \Delta\Gamma_j = \Gamma_j - \Gamma_{j,0} = -\operatorname{Im}\Delta\Omega_j Δ Γ j = Γ j − Γ j , 0 = − Im Δ Ω j . In Eq. (45) , the displayed argument is the magnetic fraction η \eta η , while the NED arguments are suppressed. The quantity B j B_j B j compares the NED shift at η \eta η with that at the electric endpoint η = 0 \eta= 0 η = 0 , keeping α ^ \hat{\alpha} α ^ , β ^ \hat{\beta} β ^ , Q / M Q/M Q / M , and g ˉ \bar{g} g ˉ fixed. It therefore isolates the magnetic dependence of the NED correction; the magnetic shift already present in the minimal theory is included in the reference at each η \eta η . We form B j B_j B j from complex shifts rather than percentages, whose normalization generally depends on the family and η \eta η . For a damped mode, Γ j = − Im Ω j > 0 \Gamma_j = -\operatorname{Im}\Omega_j > 0 Γ j = − Im Ω j > 0 , so δ Γ , j < 0 \delta_{\Gamma,j} < 0 δ Γ , j < 0 means that NED reduces the damping rate and slows the decay.
To vary the NED coefficients at a fixed ratio, we write ( α ^ , β ^ ) = λ ( a α , a β ) (\hat{\alpha},\hat{\beta}) = \lambda(a_\alpha,a_\beta) ( α ^ , β ^ ) = λ ( a α , a β ) , where the dimensionless parameter λ \lambda λ sets the scale of the two hatted coefficients and ( a α , a β ) (a_\alpha,a_\beta) ( a α , a β ) fixes their ratio. We use ( 1 , 0 ) (1,0) ( 1 , 0 ) , ( 0 , 1 ) (0,1) ( 0 , 1 ) , ( 1 , 1 ) (1,1) ( 1 , 1 ) to separate the two interactions and test the equal-coefficient case, and ( 4 / 7 , 1 ) (4/7,1) ( 4/7 , 1 ) for the unequal heavy-Dirac matching ratio. At λ = 0 \lambda= 0 λ = 0 , all sequences reduce to the minimal axion theory. At each parameter point, we recompute the background and ADM mass and adjust the total charge to keep Q / M Q/M Q / M fixed. When varying η \eta η , we hold λ \lambda λ and the chosen coefficient ratio fixed, so both α ^ \hat{\alpha} α ^ and β ^ \hat{\beta} β ^ remain constant. The corresponding coefficients normalized by the ADM mass can nevertheless vary because M / r h M/r_h M / r h need not remain constant, as explained in Sec. II.
Purely electric backgrounds and NED shifts
We first consider purely electric BHs (η = 0 \eta= 0 η = 0 ). With q M = 0 q_M = 0 q M = 0 , Eq. (8) and horizon regularity give ϕ ′ = 0 \phi' = 0 ϕ ′ = 0 ; Eq. (6) then gives N ′ = 0 N' = 0 N ′ = 0 . The asymptotic conditions ϕ ( ∞ ) = 0 \phi(\infty) = 0 ϕ ( ∞ ) = 0 and N ( ∞ ) = 1 N(\infty) = 1 N ( ∞ ) = 1 fix ϕ = 0 \phi= 0 ϕ = 0 and N = 1 N = 1 N = 1 . For B = 0 B = 0 B = 0 , the electric-field equation (5) and the metric equation (7) , using the energy density in Eq. (9) , reduce to
r 2 E ( 1 + α E 2 ) = q E , m ′ = r 2 M P l 2 ( E 2 4 + 3 α E 4 8 ) . (46) r^2 E(1+\alpha E^2)=q_E,
\qquad
m'=\frac{r^2}{M_{\mathrm{Pl}}^2}\left(\frac{E^2}{4}+\frac{3\alpha E^4}{8}\right). \tag*{(46)} r 2 E ( 1 + α E 2 ) = q E , m ′ = M Pl 2 r 2 ( 4 E 2 + 8 3 α E 4 ) . (46)
Here m = r ( 1 − h ) / 2 m=r(1-h)/2 m = r ( 1 − h ) /2 is the mass function. Of the two NED coefficients, only α \alpha α modifies this purely electric background. The β \beta β and axion interactions nevertheless contribute to the quadratic perturbation action through
S β , g ( 2 ) = ∫ d 4 x − g ˉ M [ β ( δ G ) 2 + g δ ϕ δ G ] . (47) S_{\beta,g}^{(2)}=\int\mathrm{d}^4x\sqrt{-\bar{g}_{\mathrm{M}}}\left[\beta(\delta\mathcal{G})^2+g\delta\phi\,\delta\mathcal{G}\right]. \tag*{(47)} S β , g ( 2 ) = ∫ d 4 x − g ˉ M [ β ( δ G ) 2 + g δ ϕ δ G ] . (47)
Here g ˉ μ ν \bar{g}_{\mu\nu} g ˉ μν is the background metric and g ˉ M = det ( g ˉ μ ν ) \bar{g}_{\mathrm{M}}=\det(\bar{g}_{\mu\nu}) g ˉ M = det ( g ˉ μν ) . The quantities δ ϕ \delta\phi δ ϕ and δ G \delta\mathcal{G} δ G denote the first-order perturbations of the axion field and the electromagnetic invariant G \mathcal{G} G , respectively. Although G ˉ = 0 \bar{\mathcal{G}}=0 G ˉ = 0 on the purely electric background, δ G \delta\mathcal{G} δ G can be nonzero. The angular curl of an even-parity electromagnetic harmonic vanishes, giving δ G e v e n = 0 \delta\mathcal{G}_{\mathrm{even}}=0 δ G even = 0 , whereas odd-parity electromagnetic perturbations can contribute to δ G \delta\mathcal{G} δ G . The perturbation equations therefore separate into an even-parity two-field block depending only on α \alpha α and an odd-parity three-field block depending on α \alpha α , β \beta β , g g g . In particular, varying β \beta β can shift the odd-parity frequencies while leaving the electric background and even-parity frequencies unchanged.
Nomura and Yoshida [92 ] computed QNMs for magnetic BHs and argued in Appendix A4 that their first-order NED shifts also apply to electric BHs through an electric–magnetic duality mapping. Our independently computed electric even-parity modes provide a numerical test of this prediction. We compare the fundamental even-gravitational and even-EM modes at ℓ = 2 \ell=2 ℓ = 2 , g ˉ = 0 \bar{g}=0 g ˉ = 0 , and Q / M = 0.4 , 0.6 , 0.8 Q/M=0.4,0.6,0.8 Q / M = 0.4 , 0.6 , 0.8 . The comparison uses their dimensionless coefficients α ˉ N Y = α q E 2 / M 4 \bar{\alpha}_{\mathrm{NY}}=\alpha q_E^2/M^4 α ˉ NY = α q E 2 / M 4 and β ˉ N Y = β q E 2 / M 4 \bar{\beta}_{\mathrm{NY}}=\beta q_E^2/M^4 β ˉ NY = β q E 2 / M 4 , together with the negative-real-frequency branch Ω N Y = − Ω ∗ \Omega_{\mathrm{NY}}=-\Omega^* Ω NY = − Ω ∗ , where the asterisk denotes complex conjugation. In these conventions, our derivatives of Ω N Y \Omega_{\mathrm{NY}} Ω NY with respect to α ˉ N Y \bar{\alpha}_{\mathrm{NY}} α ˉ NY at zero NED coefficients agree with the magnetic type-II coefficients in their Table I within the rounding precision of the published values. At Q / M = 0.6 Q/M=0.6 Q / M = 0.6 and g ˉ = 0 \bar{g}=0 g ˉ = 0 , the odd-gravitational and odd-EM first-order responses to both coefficients also agree with their magnetic type-I results under the same conversion. These benchmarks concern the axion-decoupled theory to first NED order; the finite-axion results below are computed from the coupled axion system. We next quantify finite NED shifts for purely electric BHs. At ℓ = 2 \ell=2 ℓ = 2 , Q / M = 0.8 Q/M=0.8 Q / M = 0.8 , α ^ = 0.1 \hat{\alpha}=0.1 α ^ = 0.1 , and β ^ = 0 \hat{\beta}=0 β ^ = 0 , the even-EM real frequency is about 0.87 % 0.87\% 0.87% lower and its damping rate about 0.28 % 0.28\% 0.28% higher than their values at α ^ = β ^ = 0 \hat{\alpha}=\hat{\beta}=0 α ^ = β ^ = 0 , with the same Q / M Q/M Q / M and g ˉ \bar{g} g ˉ . Both even-parity frequencies remain independent of g ˉ \bar{g} g ˉ , because the axion interaction enters only the odd-parity block on this background.
Figure 2 shows the odd-vector NED shifts at fixed Q / M = 0.6 Q/M=0.6 Q / M = 0.6 and η = 0 \eta=0 η = 0 . We vary g ˉ = 0 , 1 , 3 , 5 \bar{g}=0,1,3,5 g ˉ = 0 , 1 , 3 , 5 for the two fixed choices ( α ^ , β ^ ) = ( 0.0025 , 0 ) (\hat{\alpha},\hat{\beta})=(0.0025,0) ( α ^ , β ^ ) = ( 0.0025 , 0 ) and ( 0 , 0.0025 ) (0,0.0025) ( 0 , 0.0025 ) . Each shift is measured relative to the minimal axion spectrum at the same g ˉ \bar{g} g ˉ . For ( α ^ , β ^ ) = ( 0.0025 , 0 ) (\hat{\alpha},\hat{\beta})=(0.0025,0) ( α ^ , β ^ ) = ( 0.0025 , 0 ) , the odd-gravitational and odd-EM damping corrections in panel (c) are positive at g ˉ = 0 , 1 \bar{g}=0,1 g ˉ = 0 , 1 and negative at g ˉ = 3 , 5 \bar{g}=3,5 g ˉ = 3 , 5 . For ( α ^ , β ^ ) = ( 0 , 0.0025 ) (\hat{\alpha},\hat{\beta})=(0,0.0025) ( α ^ , β ^ ) = ( 0 , 0.0025 ) , both corrections in panel (d) remain positive at all sampled couplings, and the odd-gravitational correction increases with g ˉ \bar{g} g ˉ . The scalar damping correction are nonnegative in both NED cases. The vanishing correction at g ˉ = 0 \bar{g}=0 g ˉ = 0 in panel (d) follows from scalar decoupling and the β \beta β -independence of the purely electric background. Negative damping corrections indicate slower decay relative to the minimal axion model; the full damping rates remain positive.
Figure 2. NED-induced percentage shifts of the odd-gravitational, odd-EM, and scalar ℓ = 2 \ell= 2 ℓ = 2 QNMs for purely electric BHs (η = 0 \eta= 0 η = 0 ) at Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 . The horizontal axis is the axion coupling g ˉ \bar{g} g ˉ in all panels. Panels (a,c) use ( α ^ , β ^ ) = ( 0.0025 , 0 ) (\hat{\alpha},\hat{\beta}) = (0.0025,0) ( α ^ , β ^ ) = ( 0.0025 , 0 ) , and panels (b,d) use ( α ^ , β ^ ) = ( 0 , 0.0025 ) (\hat{\alpha},\hat{\beta}) = (0,0.0025) ( α ^ , β ^ ) = ( 0 , 0.0025 ) . The upper and lower panels show the real-frequency and damping-rate shifts, respectively, defined in Eq. (44) . Each shift is measured relative to the corresponding mode in the minimal axion theory ( α ^ = β ^ = 0 ) (\hat{\alpha} = \hat{\beta} = 0) ( α ^ = β ^ = 0 ) at the same g ˉ \bar{g} g ˉ . Negative damping-rate shifts indicate slower decay; all plotted modes remain damped.
Charge composition and electromagnetic duality
Here we determine how electric–magnetic charge composition modifies the NED frequency shifts. We first vary η \eta η at fixed Q / M Q/M Q / M , g ˉ \bar{g} g ˉ , and NED coefficients, computing the full individual shifts Δ Ω j ( η ) \Delta\Omega_j(\eta) Δ Ω j ( η ) in Eq. (43) and their magnetic modifications B j ( η ) B_j(\eta) B j ( η ) in Eq. (45) . The duality map in Appendix E explains two different results at g ˉ = 0 \bar{g}=0 g ˉ = 0 . For α ^ = β ^ = λ \hat{\alpha}=\hat{\beta}=\lambda α ^ = β ^ = λ , it relates different charge compositions within the same theory through first order in λ \lambda λ , making each individual branch’s linear NED shift independent of η \eta η [Eq. (53) below]. For unequal coefficients, it instead constrains the gravitational and EM pair sums B j B_j B j through Eq. (52) . The response formulas are given in Appendix D. We also vary λ \lambda λ at fixed η \eta η to separate linear from higher-order magnetic dependence and test the loss of the cancellation at nonzero axion coupling.
Figure 3 compares the full NED shifts at Q / M = 0.6 Q/M=0.6 Q / M = 0.6 and g ˉ = 0 \bar{g}=0 g ˉ = 0 for ( α ^ , β ^ ) = ( 0.05 , 0 ) , ( 0 , 0.05 ) , ( 0.05 , 0.05 ) (\hat{\alpha},\hat{\beta})=(0.05,0),(0,0.05),(0.05,0.05) ( α ^ , β ^ ) = ( 0.05 , 0 ) , ( 0 , 0.05 ) , ( 0.05 , 0.05 ) , with each coefficient choice held fixed as η \eta η varies. The gravitational and EM families are continued from η = 0 \eta=0 η = 0 at the same NED coefficients. The plotted percentages refer to the NED-free spectrum at each η \eta η , as defined in Eq. (44) .
For ( α ^ , β ^ ) = ( 0.05 , 0 ) (\hat{\alpha},\hat{\beta})=(0.05,0) ( α ^ , β ^ ) = ( 0.05 , 0 ) , the even-gravitational and even-EM families already have nonzero NED shifts at η = 0 \eta=0 η = 0 . For ( α ^ , β ^ ) = ( 0 , 0.05 ) (\hat{\alpha},\hat{\beta})=(0,0.05) ( α ^ , β ^ ) = ( 0 , 0.05 ) , the shifts of these families vanish at η = 0 \eta=0 η = 0 and become nonzero when magnetic charge is introduced. In both single-coefficient cases, the even-EM damping correction increases with η \eta η over the plotted range, whereas the odd-EM correction decreases. For equal coefficients, ( α ^ , β ^ ) = ( 0.05 , 0.05 ) (\hat{\alpha},\hat{\beta})=(0.05,0.05) ( α ^ , β ^ ) = ( 0.05 , 0.05 ) , the frequency and damping shifts of each individual branch are nearly constant as η \eta η varies. This is the finite-coefficient manifestation of the stronger branchwise relation (53) , rather than only a cancellation between parity partners. The full shifts remain nonzero; their residual magnetic dependence can begin at second NED order.
To interpret the magnetic dependence of the pair sums, we retain g ˉ = 0 \bar{g} = 0 g ˉ = 0 and use the purely electric RN reference η = α ^ = β ^ = 0 \eta= \hat{\alpha} = \hat{\beta} = 0 η = α ^ = β ^ = 0 . At this reference, the even- and odd-gravitational modes share one complex QNM frequency, while the even- and odd-EM modes share a different complex frequency. These define the gravitational and EM RN parity pairs, respectively.
We expand each pair’s frequency sum to first order in α ^ \hat{\alpha} α ^ and β ^ \hat{\beta} β ^ at fixed Q / M Q/M Q / M . Electromagnetic duality then determines its magnetic dependence from derivatives evaluated at this electric reference. For J = g r a v , E M J = \mathrm{grav},\mathrm{EM} J = grav , EM , we define
Σ J ( η ) = Ω J , e ( η ) + Ω J , o ( η ) , B J ( η ) = B J , e ( η ) + B J , o ( η ) . (48) \begin{aligned}
\Sigma_J(\eta) &= \Omega_{J,e}(\eta) + \Omega_{J,o}(\eta),\\
B_J(\eta) &= B_{J,e}(\eta) + B_{J,o}(\eta). \tag*{(48)}
\end{aligned} Σ J ( η ) B J ( η ) = Ω J , e ( η ) + Ω J , o ( η ) , = B J , e ( η ) + B J , o ( η ) . (48)
Here Ω J , e \Omega_{J,e} Ω J , e and Ω J , o \Omega_{J,o} Ω J , o are the corresponding members of Ω j ( α ^ , β ^ ) \Omega_j(\hat{\alpha},\hat{\beta}) Ω j ( α ^ , β ^ ) in Eq. (43) ; their NED arguments are implicit. The indices e , o e,o e , o specify parity at η = 0 \eta= 0 η = 0 and label the continued families at nonzero η \eta η , rather than individual perturbation fields. Each B J , e B_{J,e} B J , e or B J , o B_{J,o} B J , o is defined by Eq. (45) . Thus B J B_J B J is the sum of the pair’s two NED shifts at η \eta η , minus their sum at η = 0 \eta= 0 η = 0 , with the same NED coefficients, Q / M Q/M Q / M , and g ˉ \bar{g} g ˉ .
At g ˉ = 0 \bar{g} = 0 g ˉ = 0 , the NED-free reference frequency is independent of η \eta η and common to both members of each RN pair. Writing it as Ω J , 0 \Omega_{J,0} Ω J , 0 , with Γ J , 0 = − Im Ω J , 0 \Gamma_{J,0} = -\operatorname{Im}\Omega_{J,0} Γ J , 0 = − Im Ω J , 0 , Eqs. (44) and (48) give
100 Re B J ( η ) Re Ω J , 0 = ∑ s = e , o [ δ R , J , s ( η ) − δ R , J , s ( 0 ) ] . (49) 100\frac{\operatorname{Re}B_J(\eta)}{\operatorname{Re}\Omega_{J,0}}
=
\sum_{s=e,o}\left[\delta_{R,J,s}(\eta)-\delta_{R,J,s}(0)\right]. \tag*{(49)} 100 Re Ω J , 0 Re B J ( η ) = s = e , o ∑ [ δ R , J , s ( η ) − δ R , J , s ( 0 ) ] . (49)
100 − Im B J ( η ) Γ J , 0 = ∑ s = e , o [ δ Γ , J , s ( η ) − δ Γ , J , s ( 0 ) ] . (50) 100\frac{-\operatorname{Im}B_J(\eta)}{\Gamma_{J,0}}
=
\sum_{s=e,o}\left[\delta_{\Gamma,J,s}(\eta)-\delta_{\Gamma,J,s}(0)\right]. \tag*{(50)} 100 Γ J , 0 − Im B J ( η ) = s = e , o ∑ [ δ Γ , J , s ( η ) − δ Γ , J , s ( 0 ) ] . (50)
Thus, to recover B J B_J B J from Figure 3 , we add the two curves of a pair, subtract their summed value at η = 0 \eta= 0 η = 0 , and undo the percentage normalization. This conversion is exact at the plotted finite NED coefficients; the duality prediction below is first order.
Figure 3. NED-induced percentage shifts of the ℓ = 2 \ell= 2 ℓ = 2 gravitational and EM QNMs as functions of the magnetic fraction η \eta η , at fixed Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 and g ˉ = 0 \bar{g} = 0 g ˉ = 0 . From left to right, the columns correspond to ( α ^ , β ^ ) = ( 0.05 , 0 ) , ( 0 , 0.05 ) , ( 0.05 , 0.05 ) (\hat{\alpha},\hat{\beta}) = (0.05, 0),(0, 0.05),(0.05, 0.05) ( α ^ , β ^ ) = ( 0.05 , 0 ) , ( 0 , 0.05 ) , ( 0.05 , 0.05 ) . The upper and lower panels show the real-frequency and damping-rate shifts, respectively, defined in Eq. (44) . Each shift is measured relative to the NED-free spectrum at the same η \eta η . Family labels identify modes continued from the purely electric limit at the same NED coefficients. For ( α ^ , β ^ ) = ( 0.05 , 0.05 ) (\hat{\alpha},\hat{\beta}) = (0.05, 0.05) ( α ^ , β ^ ) = ( 0.05 , 0.05 ) , each branch’s shift is independent of η \eta η through first NED order [Eq. (5.11)]; residual dependence at finite coefficients is allowed at higher order. The full NED shifts remain nonzero.
The two electric-reference response coefficients are
T J , α = ∂ Σ J ∂ α ^ ∣ 0 , T J , β = ∂ Σ J ∂ β ^ ∣ 0 . (51) T_{J,\alpha} = \left.\frac{\partial\Sigma_J}{\partial\hat{\alpha}}\right|_0,\qquad T_{J,\beta} = \left.\frac{\partial\Sigma_J}{\partial\hat{\beta}}\right|_0 . \tag*{(51)} T J , α = ∂ α ^ ∂ Σ J 0 , T J , β = ∂ β ^ ∂ Σ J 0 . (51)
The subscript 0 indicates evaluation at η = g ˉ = α ^ = β ^ = 0 \eta= \bar{g} = \hat{\alpha} = \hat{\beta} = 0 η = g ˉ = α ^ = β ^ = 0 for the chosen Q / M Q/M Q / M . Each partial derivative varies one hatted NED coefficient while keeping the other coefficient zero and η = g ˉ = 0 \eta= \bar{g} = 0 η = g ˉ = 0 . The derivatives include the charge retuning and background-dependent ADM-mass normalization at fixed Q / M Q/M Q / M .
We assume that the common frequency of each RN pair is isolated from the remaining spectrum and semisimple, with two independent eigenmodes and no generalized eigenmodes, and that its projected frequency-derivative matrix N J N_J N J is invertible. Equations (D.5) and (D.6) in Appendix D define this matrix and give the summed frequency derivatives. For both RN pairs at Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 and ℓ = 2 \ell= 2 ℓ = 2 , the two diagonal entries of N J N_J N J in Eq. (D.7) remain nonzero at N c = 36 , 48 , 60 N_c = 36,48,60 N c = 36 , 48 , 60 , supporting its invertibility. Appendix C 1 summarizes these numerical checks.
At g ˉ = 0 \bar{g} = 0 g ˉ = 0 , combining Eqs. (E.7) and (E.8) in Appendix E, we obtain the magnetic modification of the summed NED frequency shift:
B J = − 4 η 2 ( 1 − η 2 ) ( α ^ − β ^ ) ( T J , α − T J , β ) + O ( α ^ 2 , α ^ β ^ , β ^ 2 ) . (52) B_J = -4\eta^2(1-\eta^2)(\hat{\alpha}-\hat{\beta})(T_{J,\alpha}-T_{J,\beta}) + O(\hat{\alpha}^2,\hat{\alpha}\hat{\beta},\hat{\beta}^2). \tag*{(52)} B J = − 4 η 2 ( 1 − η 2 ) ( α ^ − β ^ ) ( T J , α − T J , β ) + O ( α ^ 2 , α ^ β ^ , β ^ 2 ) . (52)
The duality relation (52) is restricted to g ˉ = 0 \bar{g} = 0 g ˉ = 0 . Its remainder contains terms quadratic or higher order in the NED coefficients at this fixed axion coupling. The factor 4 η 2 ( 1 − η 2 ) = sin 2 ( 2 ϑ ) 4\eta^2(1-\eta^2)=\sin^2(2\vartheta) 4 η 2 ( 1 − η 2 ) = sin 2 ( 2 ϑ ) arises from the electromagnetic duality rotation, whose angle ϑ \vartheta ϑ satisfies η = sin ϑ \eta= \sin\vartheta η = sin ϑ .
For equal coefficients, a stronger statement holds. We define the linear response along ( α ^ , β ^ ) = ( λ , λ ) (\hat{\alpha},\hat{\beta}) = (\lambda,\lambda) ( α ^ , β ^ ) = ( λ , λ ) by
K j e q ( η ) = ∂ λ Ω j ( η ; λ , λ ) ∣ λ = 0 K^{\mathrm{eq}}_j(\eta) = \left.\partial_\lambda\Omega_j(\eta;\lambda,\lambda)\right|_{\lambda=0} K j eq ( η ) = ∂ λ Ω j ( η ; λ , λ ) ∣ λ = 0
at fixed Q / M Q/M Q / M and g ˉ = 0 \bar{g} = 0 g ˉ = 0 . For each gravitational or EM branch continued from its electric-reference mode, the same-theory duality map gives
κ j e q ( η ) = κ j e q ( 0 ) , B j ( η ; λ , λ ) = O ( λ 2 ) . (53) \kappa_j^{\mathrm{eq}}(\eta)=\kappa_j^{\mathrm{eq}}(0), \\
B_j(\eta;\lambda,\lambda)=O(\lambda^2). \tag*{(53)} κ j eq ( η ) = κ j eq ( 0 ) , B j ( η ; λ , λ ) = O ( λ 2 ) . (53)
Appendix E2 derives this result without taking a trace over the RN pair and checks it using the independent integration-and-matching data. It implies the equal-coefficient pair-sum cancellation in Eq. (52) , but also rules out compensating first-order magnetic shifts of the individual partners. The nonzero values of κ j e q ( 0 ) \kappa_j^{\mathrm{eq}}(0) κ j eq ( 0 ) still allow nonzero full first-order NED shifts. Nor does Eq. (53) require the two parity partners to have equal NED shifts. The quartic truncation is self-dual only through first NED order, so no exact finite-λ \lambda λ composition independence is asserted.
For ( α ^ , β ^ ) = ( λ , 0 ) (\hat{\alpha},\hat{\beta})=(\lambda,0) ( α ^ , β ^ ) = ( λ , 0 ) and ( 0 , λ ) (0,\lambda) ( 0 , λ ) with the same small λ \lambda λ , the first-order contribution to B J B_J B J changes by an overall minus sign: its real and imaginary parts each reverse sign without changing their magnitudes. This sign reversal concerns B J B_J B J , rather than the individual full NED shifts plotted as percentages in Figure 3 . At g ˉ = 0 \bar{g}=0 g ˉ = 0 and fixed Q / M Q/M Q / M , α ^ \hat{\alpha} α ^ , and β ^ \hat{\beta} β ^ , Eq. (52) shows that the part of B J B_J B J linear in the NED coefficients is proportional to η 2 ( 1 − η 2 ) \eta^2(1-\eta^2) η 2 ( 1 − η 2 ) . When this linear contribution is nonzero, its value at η = 0.4 \eta=0.4 η = 0.4 is therefore 3.5 times its value at η = 0.2 \eta=0.2 η = 0.2 . The first-order coefficients of B J B_J B J extracted from the numerical frequencies are consistent with this ratio.
We next test the equal-coefficient cancellation at nonzero axion coupling. Figure 4 compares g ˉ = 0 \bar{g}=0 g ˉ = 0 and 0.9 0.9 0.9 at fixed Q / M = 0.6 Q/M=0.6 Q / M = 0.6 , with α ^ = β ^ = λ \hat{\alpha}=\hat{\beta}=\lambda α ^ = β ^ = λ in all panels. At g ˉ = 0.9 \bar{g}=0.9 g ˉ = 0.9 and η = 0.4 \eta=0.4 η = 0.4 , B E M B_{\mathrm{EM}} B EM contains a nonzero term linear in λ \lambda λ , demonstrating the loss of the first-order magnetic cancellation at this point. At g ˉ = 0.9 \bar{g}=0.9 g ˉ = 0.9 , the EM pair denotes the two families continued from the even- and odd-EM RN modes; their frequencies need not coincide. We still define B E M B_{\mathrm{EM}} B EM by Eq. (48) and compute it directly from the individual frequency shifts in Eqs. (43) and (45) . Neither zero-axion duality prediction, Eq. (52) or Eq. (53) , is used for the g ˉ = 0.9 \bar{g}=0.9 g ˉ = 0.9 results.
Figure 4. Full NED shifts and their magnetic modification for equal coefficients ( α ^ , β ^ ) = ( λ , λ ) (\hat{\alpha},\hat{\beta})=(\lambda,\lambda) ( α ^ , β ^ ) = ( λ , λ ) at ℓ = 2 \ell=2 ℓ = 2 and Q / M = 0.6 Q/M=0.6 Q / M = 0.6 . Upper panels: real-frequency and damping percentages of the two EM families as functions of η \eta η at λ = 0.05 \lambda=0.05 λ = 0.05 . Lower panels: Re B E M \operatorname{Re} B_{\mathrm{EM}} Re B EM and B E M I ≡ − Im B E M B_{\mathrm{EM}}^{\mathcal{I}}\equiv-\operatorname{Im} B_{\mathrm{EM}} B EM I ≡ − Im B EM , both scaled by 10 5 10^{5} 1 0 5 , as functions of λ \lambda λ at η = 0.4 \eta=0.4 η = 0.4 . Solid and dashed lines denote g ˉ = 0 \bar{g}=0 g ˉ = 0 and 0.9 0.9 0.9 , respectively. At g ˉ = 0 \bar{g}=0 g ˉ = 0 , Eq. (53) predicts first-order magnetic cancellation for each individual branch, and hence for their sum. The full NED shifts remain nonzero.
The upper panels fix λ = 0.05 \lambda= 0.05 λ = 0.05 and vary η \eta η , plotting δ R , j \delta R_{,j} δ R , j and δ Γ , j \delta\Gamma_{,j} δ Γ , j for the two EM families, as in Figure 3 . The lower panels instead fix η = 0.4 \eta= 0.4 η = 0.4 and vary λ \lambda λ , plotting Re B E M \operatorname{Re} B_{\mathrm{EM}} Re B EM and − Im B E M -\operatorname{Im} B_{\mathrm{EM}} − Im B EM scaled by 10 5 10^{5} 1 0 5 . These lower-panel quantities are absolute dimensionless frequency differences, rather than percentages. Each point subtracts the pair’s summed NED shift at η = 0 \eta= 0 η = 0 from that at η = 0.4 \eta= 0.4 η = 0.4 , with the same λ \lambda λ and g ˉ \bar{g} g ˉ . At g ˉ = 0.9 \bar{g} = 0.9 g ˉ = 0.9 , the reference frequencies and damping rates depend on both family and η \eta η . To construct B E M B_{\mathrm{EM}} B EM from percentages, each shift must therefore first be converted to Δ Ω j \Delta\Omega_j Δ Ω j using its own reference values in Eq. (44) ; the common RN normalization used above no longer applies.
The suppressed g ˉ = 0 \bar{g} = 0 g ˉ = 0 response in the lower panels is consistent with B E M = O ( λ 2 ) B_{\mathrm{EM}} = O(\lambda^{2}) B EM = O ( λ 2 ) for equal coefficients. Equation (53) predicts the same first-order cancellation separately for each branch in the upper panels. At nonzero axion coupling, the duality rotation of g ϕ G g\phi G g ϕG also generates a ϕ F \phi F ϕF interaction, which is absent from Eq. (2.4). The transformed theory therefore has a different set of interactions, so equal NED coefficients no longer guarantee the first-order magnetic cancellation.
To test whether unequal NED coefficients could restore the cancellation at this parameter point, we compute the linear responses of B E M B_{\mathrm{EM}} B EM to α ^ \hat{\alpha} α ^ and β ^ \hat{\beta} β ^ separately at α ^ = β ^ = 0 \hat{\alpha} = \hat{\beta} = 0 α ^ = β ^ = 0 . We obtain these responses from differences of the individual mode-frequency derivatives at η = 0.4 \eta= 0.4 η = 0.4 and 0 0 0 , using the frequency-difference formulas in Eq. (D.4) of Appendix D. Canceling the real component would require β ^ / α ^ ≃ − 3.58 \hat{\beta}/\hat{\alpha} \simeq-3.58 β ^ / α ^ ≃ − 3.58 , whereas canceling the damping component, − Im B E M -\operatorname{Im} B_{\mathrm{EM}} − Im B EM , would require β ^ / α ^ ≃ 0.287 \hat{\beta}/\hat{\alpha} \simeq0.287 β ^ / α ^ ≃ 0.287 . These different ratios rule out any nonzero real choice of ( α ^ , β ^ ) (\hat{\alpha}, \hat{\beta}) ( α ^ , β ^ ) that cancels both components of B E M B_{\mathrm{EM}} B EM to first order. They also exclude simultaneous cancellation of the EM- and gravitational-pair magnetic modifications at this point.
Individual frequency shifts at finite NED coefficients
The preceding subsection focused on charge composition and the magnetic modification B J B_J B J of a pair’s summed NED shift. Here we vary the NED coefficients at fixed charge composition and determine the five individual frequencies Ω j ( α ^ , β ^ ) \Omega_j(\hat{\alpha}, \hat{\beta}) Ω j ( α ^ , β ^ ) and their full shifts Δ Ω j \Delta\Omega_j Δ Ω j in Eq. (43) . The reference is the spectrum at α ^ = β ^ = 0 \hat{\alpha} = \hat{\beta} = 0 α ^ = β ^ = 0 , with the same Q / M Q/M Q / M , g ˉ \bar{g} g ˉ , and η \eta η ; no electric-endpoint subtraction is made here.
We set ℓ = 2 \ell= 2 ℓ = 2 , Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 , g ˉ = 5 \bar{g} = 5 g ˉ = 5 , and η = 0.4 \eta= 0.4 η = 0.4 , and compare three coefficient sequences: ( α ^ , β ^ ) = ( λ , 0 ) , ( 0 , λ ) , ( 4 λ / 7 , λ ) (\hat{\alpha}, \hat{\beta}) = (\lambda, 0),(0, \lambda),(4\lambda/7, \lambda) ( α ^ , β ^ ) = ( λ , 0 ) , ( 0 , λ ) , ( 4 λ /7 , λ ) , with λ = 0 , 0.01 , 0.025 , 0.05 \lambda= 0, 0.01, 0.025, 0.05 λ = 0 , 0.01 , 0.025 , 0.05 . The first two sequences separate the two quartic interactions. The third has two nonzero, unequal coefficients with β ^ / α ^ = 7 / 4 \hat{\beta}/\hat{\alpha} = 7/4 β ^ / α ^ = 7/4 , the heavy-Dirac matching ratio introduced in Sec. II. We use this ratio within the retained quartic model. We solve the coupled five-field QNM problem at each finite-coefficient point, using the background and mass normalization prescription of § 2.2 A. Each family is continued from its frequency at α ^ = β ^ = 0 \hat{\alpha} = \hat{\beta} = 0 α ^ = β ^ = 0 along the chosen sequence.
Figure 5 shows the shifts in the real frequencies and damping rates of all five families. Over the plotted range, the largest real-frequency shift in magnitude occurs in the even-EM family for β ^ = 0 \hat{\beta} = 0 β ^ = 0 and in the scalar family for α ^ = 0 \hat{\alpha} = 0 α ^ = 0 . The odd-gravitational and odd-EM damping corrections are positive for the sampled α ^ > 0 \hat{\alpha} > 0 α ^ > 0 , β ^ = 0 \hat{\beta} = 0 β ^ = 0 points and negative for α ^ = 0 \hat{\alpha} = 0 α ^ = 0 , β ^ > 0 \hat{\beta} > 0 β ^ > 0 . The relative strengths of the two NED interactions can therefore determine whether these modes decay faster or more slowly than their NED-free counterparts.
In the purely electric limit (η = 0 \eta= 0 η = 0 ), the even-parity frequencies are independent of β ^ \hat{\beta} β ^ , as established in § 2.2 B. On the dyonic background considered here (η = 0.4 \eta= 0.4 η = 0.4 ), the β G 2 \beta\mathcal{G}^{2} β G 2 interaction contributes to both the background equations and the full coupled perturbation operator. The two QNM families continued from the electric even-parity modes therefore also depend on β ^ \hat{\beta} β ^ . For the even-EM family, the damping-rate shift is Δ Γ E M , e ≃ − 1.75 × 10 − 6 \Delta\Gamma_{\mathrm{EM,e}} \simeq-1.75 \times10^{-6} Δ Γ EM , e ≃ − 1.75 × 1 0 − 6 at ( α ^ , β ^ ) = ( 0.05 , 0 ) (\hat{\alpha},\hat{\beta}) = (0.05,0) ( α ^ , β ^ ) = ( 0.05 , 0 ) , whereas at ( 0 , 0.05 ) (0,0.05) ( 0 , 0.05 ) it is + 3.47 × 10 − 5 +3.47 \times10^{-5} + 3.47 × 1 0 − 5 . The former reduces the damping rate, while the latter increases it.
When both interactions are present along ( α ^ , β ^ ) = ( 4 λ / 7 , λ ) (\hat{\alpha},\hat{\beta}) = (4\lambda/7,\lambda) ( α ^ , β ^ ) = ( 4 λ /7 , λ ) , all five families have nonzero real and imaginary frequency shifts at each of the sampled values λ = 0.01 , 0.025 , 0.05 \lambda= 0.01, 0.025, 0.05 λ = 0.01 , 0.025 , 0.05 . For example, at λ = 0.05 \lambda= 0.05 λ = 0.05 , the even-EM shift is Δ Ω E M , e ≃ − 2.48537 × 10 − 4 − 3.34318 × 10 − 5 i \Delta\Omega_{\mathrm{EM,e}} \simeq-2.48537 \times10^{-4} - 3.34318 \times10^{-5}i Δ Ω EM , e ≃ − 2.48537 × 1 0 − 4 − 3.34318 × 1 0 − 5 i . Its real frequency therefore decreases. The negative imaginary shift gives Δ Γ E M , e = − Im Δ Ω E M , e ≃ + 3.34318 × 10 − 5 \Delta\Gamma_{\mathrm{EM,e}} = -\operatorname{Im}\Delta\Omega_{\mathrm{EM,e}} \simeq+3.34318 \times10^{-5} Δ Γ EM , e = − Im Δ Ω EM , e ≃ + 3.34318 × 1 0 − 5 , so the damping rate increases and the mode decays faster.
For these fixed parameters, the reference damping rates in the g ˉ = 5 \bar{g} = 5 g ˉ = 5 , η = 0.4 \eta= 0.4 η = 0.4 column of Table I range from approximately 0.0895 to 0.1057. All corrections in Figure 5 satisfy ∣ Δ Γ j ∣ < 3.5 × 10 − 5 |\Delta\Gamma_j| < 3.5 \times10^{-5} ∣Δ Γ j ∣ < 3.5 × 1 0 − 5 and are therefore much smaller than the reference rates. Hence Γ j = Γ j , 0 + Δ Γ j > 0 \Gamma_j = \Gamma_{j,0} + \Delta\Gamma_j > 0 Γ j = Γ j , 0 + Δ Γ j > 0 at every plotted point, so all five sampled families remain damped.
As a local approximation to these directly computed shifts, we also expand each complex frequency about α ^ = β ^ = 0 \hat{\alpha} = \hat{\beta} = 0 α ^ = β ^ = 0 . The derivatives and the two-coefficient Taylor expansion are given in Appendix D, Eqs. (D.1) and (D.2). For the even-EM example at λ = 0.05 \lambda= 0.05 λ = 0.05 along ( α ^ , β ^ ) = ( 4 λ / 7 , λ ) (\hat{\alpha},\hat{\beta}) = (4\lambda/7,\lambda) ( α ^ , β ^ ) = ( 4 λ /7 , λ ) , the absolute difference between the predicted and directly computed complex shifts decreases from approximately 7.8 × 10 − 7 7.8 \times10^{-7} 7.8 × 1 0 − 7 at first order to 3.2 × 10 − 9 3.2 \times10^{-9} 3.2 × 1 0 − 9 at second order. If the second-order term proportional to α ^ β ^ \hat{\alpha}\hat{\beta} α ^ β ^ is omitted, the difference is 3.4 × 10 − 7 3.4 \times10^{-7} 3.4 × 1 0 − 7 . Thus, the mixed term substantially improves the prediction for this unequal-coefficient example.
We quantify the strength of the retained NED interactions using ϵ N E D max \epsilon_{\mathrm{NED}}^{\max} ϵ NED m a x , the maximum of ϵ N E D ( r ) \epsilon_{\mathrm{NED}}(r) ϵ NED ( r ) in Eq. (18) over the BH exterior. Every background in Figure 5 satisfies ϵ N E D max < 3.8 × 10 − 3 \epsilon_{\mathrm{NED}}^{\max} < 3.8 \times10^{-3} ϵ NED m a x < 3.8 × 1 0 − 3 , so the quartic electromagnetic terms remain small compared with the local Maxwell energy density throughout the exterior.
We next assess whether individual damping-rate shifts can reach the percent level within the adopted NED amplitude criterion. For Figure 6 , we fix Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 , g ˉ = 5 \bar{g} = 5 g ˉ = 5 , and η = 0.1 \eta= 0.1 η = 0.1 , and vary λ \lambda λ up to 0.16 0.16 0.16 along the same three coefficient sequences. Compared with Figure 5 , this extends the coefficient range beyond λ = 0.05 \lambda= 0.05 λ = 0.05 and uses a smaller magnetic fraction, η = 0.1 \eta= 0.1 η = 0.1 instead of 0.4 0.4 0.4 . We focus on the odd-gravitational family because its damping rate has the largest fractional changes among the five families at these sampled parameters.
The sampled nonzero coefficients are λ = 0.025 , 0.05 , 0.1 , 0.16 \lambda= 0.025, 0.05, 0.1, 0.16 λ = 0.025 , 0.05 , 0.1 , 0.16 . The left panel shows the odd-gravitational damping-rate change in percent, δ Γ j \delta\Gamma_j δ Γ j defined in Eq. (44) , relative to the NED-free mode at the same parameters. The right panel shows ϵ N E D max \epsilon_{\mathrm{NED}}^{\max} ϵ NED m a x evaluated on each corresponding background.
At λ = 0.16 \lambda= 0.16 λ = 0.16 , the odd-gravitational damping-rate changes are approximately − 1.05 % -1.05\% − 1.05% , + 1.10 % +1.10\% + 1.10% , and + 0.43 % +0.43\% + 0.43% for ( α ^ , β ^ ) = ( λ , 0 ) , ( 0 , λ ) , ( 4 λ / 7 , λ ) (\hat{\alpha},\hat{\beta}) = (\lambda,0),(0,\lambda),(4\lambda/7,\lambda) ( α ^ , β ^ ) = ( λ , 0 ) , ( 0 , λ ) , ( 4 λ /7 , λ ) , respectively. The damping-rate changes of the other four families are smaller than 0.23 % 0.23\% 0.23% of their respective NED-free rates throughout these sampled sequences. All backgrounds in Figure 6 satisfy ϵ N E D max < 0.032 \epsilon_{\mathrm{NED}}^{\max} < 0.032 ϵ NED m a x < 0.032 , below the adopted bound of 0.05 0.05 0.05 . Even the largest odd-gravitational damping decrease is only about one percent of its NED-free value, so this mode remains damped at every plotted point. These examples show that percent-level damping changes are possible while the retained quartic interactions remain small throughout the exterior.
The sign of the correction also depends on the magnetic fraction. At λ = 0.05 \lambda= 0.05 λ = 0.05 along ( α ^ , β ^ ) = ( 4 λ / 7 , λ ) (\hat{\alpha},\hat{\beta}) = (4\lambda/7,\lambda) ( α ^ , β ^ ) = ( 4 λ /7 , λ ) , the odd-gravitational damping rate increases at η = 0.1 \eta= 0.1 η = 0.1 in Figure 6 but decreases at η = 0.4 \eta= 0.4 η = 0.4 in Figure 5 . Here Q / M Q/M Q / M , g ˉ \bar{g} g ˉ , and both NED coefficients are the same in the two calculations. Changing η \eta η modifies both the background and the coupled perturbation operator, allowing the damping correction to reverse sign.
Interpreting the model as a controlled EFT additionally requires the scale hierarchies and suppression of omitted operators discussed in Appendix A. Whether the resulting damping-rate shifts can be measured depends on how strongly the relevant modes are excited and on the precision of the ringdown observations.
Conclusions
We have studied the QNMs of static charged BHs with a massless axion and two quartic electromagnetic interactions. We first derived the reduced quadratic action and coupled linear equations for five dynamical degrees of freedom, with a regular elimination of the nondynamical fields on the stated nonextremal exterior branch. The studied backgrounds satisfy sufficient local high-frequency conditions for the absence of ghosts and gradient instabilities throughout the BH exterior; these conditions alone do not establish global mode stability.
Our spectral calculations use the asymptotic clock and the ADM mass of each background, with the charge retuned to keep Q / M Q/M Q / M fixed. Spectral-resolution tests, background-tolerance tests, and independent integration-and-matching calculations support the sampled fundamental frequencies and NED shifts.
The first main result is the computation of the complex QNM frequencies of five oscillatory quadrupolar mode families continued from the fundamental RN modes, already without NED, extending the two-family dyonic analysis of Ref. [64 ] . A purely electric BH has no background axion hair and retains the RN geometry, but the axion–photon coupling mixes the scalar and odd-parity gravitoelectromagnetic perturbations. At representative nonzero couplings, the odd frequencies split from the unchanged even frequencies, yielding five distinct complex frequencies and breaking both gravitational and EM axial–polar isospectrality. Magnetic charge then sources axion hair and couples all five perturbations in the axion-coupled theory. The damping rates can vary nonmonotonically with magnetic charge and cross while the corresponding real frequencies remain distinct. Additional purely imaginary growing modes occur in the g ˉ = 10 \bar{g} = 10 g ˉ = 10 , η = 0.1 \eta= 0.1 η = 0.1 benchmarks at Q / M = 0.6 , 0.7 , 0.8 Q/M = 0.6,0.7,0.8 Q / M = 0.6 , 0.7 , 0.8 . These backgrounds are therefore mode unstable despite satisfying the local high-frequency conditions. They are distinct from the lower-coupling backgrounds used for the NED sequences below.
We have used the RN EM–scalar Darboux degeneracy at Q / M = 7 / 4 Q/M = \sqrt{7}/4 Q / M = 7 /4 for ℓ = 2 \ell= 2 ℓ = 2 , previously derived in Ref. [93 ] , to interpret the small-coupling splitting in the axion–photon theory. The transformation preserves the ingoing condition at the horizon and the outgoing condition at infinity. The complex EM and scalar QNM spectra therefore coincide, except possibly at ω = ± 12 i / M \omega= \pm12i/M ω = ± 12 i / M , where the transformation is not invertible. This degeneracy occurs within the odd-parity sector and is distinct from RN axial–polar isospectrality. For the fundamental pair, a small axion coupling at this special charge produces a splitting linear in g ˉ \bar{g} g ˉ . At a fixed charge ratio where the uncoupled fundamental EM and scalar modes are simple and isolated within the odd-parity sector, with a nonzero complex-frequency gap, their frequency shifts generically begin at O ( g ˉ 2 ) O(\bar{g}^{2}) O ( g ˉ 2 ) when the axion-induced mixing is weak compared with this gap.
We then included the quartic NED interactions α F 2 + β G 2 \alpha F^{2}+\beta\mathcal{G}^{2} α F 2 + β G 2 and determined their additional QNM frequency shifts. For purely electric BHs, only the F 2 F^{2} F 2 interaction modifies the background, whereas the G 2 \mathcal{G}^{2} G 2 and axion interactions affect the odd-parity perturbation equations and their QNM frequencies.
With the axion decoupled and equal quartic coefficients (α ^ = β ^ = λ \hat{\alpha}=\hat{\beta}=\lambda α ^ = β ^ = λ ), the duality-rotated theory agrees with the original theory through first order in λ \lambda λ . Consequently, the first-order NED shift of each individual gravitational or EM branch is independent of η \eta η at fixed Q / M Q/M Q / M , as expressed by Eq. (53) . The full NED shifts remain nonzero and can differ between the two parity partners. Independent integration-and-matching differences support the branchwise cancellation. For unequal coefficients, Eq. (52) retains its role as a constraint on the summed shifts of each isolated semisimple RN parity pair. The equal-coefficient result is restricted to first NED order and permits higher-order magnetic dependence. This cancellation is lost in the tested axion-coupled configuration at Q / M = 0.6 Q/M=0.6 Q / M = 0.6 , g ˉ = 0.9 \bar{g}=0.9 g ˉ = 0.9 , and η = 0.4 \eta=0.4 η = 0.4 . Canceling the real-frequency and damping components of the EM-pair magnetic modification requires different ratios β ^ / α ^ \hat{\beta}/\hat{\alpha} β ^ / α ^ . Thus no nonzero real choice of the two NED coefficients makes both components vanish simultaneously to first order at this point.
Direct calculations at finite NED coefficients determine the individual complex-frequency shifts at fixed charge composition. At Q / M = 0.6 Q/M=0.6 Q / M = 0.6 , g ˉ = 5 \bar{g}=5 g ˉ = 5 , and η = 0.4 \eta=0.4 η = 0.4 , positive α ^ \hat{\alpha} α ^ alone increases the odd-gravitational and odd-EM damping rates, whereas positive β ^ \hat{\beta} β ^ alone decreases them over the sampled range. With both NED coefficients nonzero at the heavy-Dirac matching ratio β ^ / α ^ = 7 / 4 \hat{\beta}/\hat{\alpha}=7/4 β ^ / α ^ = 7/4 , all five families exhibit nonzero shifts in their real frequencies and damping rates at the sampled points. For the representative even-EM example, including the mixed term proportional to α ^ β ^ \hat{\alpha}\hat{\beta} α ^ β ^ in the second-order Taylor expansion improves agreement with the directly computed shift.
At η = 0.1 \eta=0.1 η = 0.1 with the same Q / M Q/M Q / M and g ˉ \bar{g} g ˉ , the odd-gravitational damping-rate changes reach about − 1.05 % -1.05\% − 1.05% and + 1.10 % +1.10\% + 1.10% for ( α ^ , β ^ ) = ( 0.16 , 0 ) (\hat{\alpha},\hat{\beta})=(0.16,0) ( α ^ , β ^ ) = ( 0.16 , 0 ) and ( 0 , 0.16 ) (0,0.16) ( 0 , 0.16 ) , respectively. This family has the largest fractional damping changes along the sampled sequences, while the other four remain below 0.23 % 0.23\% 0.23% . All five oscillatory families remain damped along these NED sequences, and the retained quartic terms remain small compared with the local Maxwell energy density throughout the exterior. A specific heavy-particle EFT interpretation also requires the scale hierarchies and control of omitted operators discussed in Appendix A. Observability depends on mode excitation and the precision of ringdown measurements.
Future work includes mapping instability thresholds and the stable parameter domain across multipoles and overtones, as well as extending the model to a finite axion mass. An especially interesting direction is to construct rotating BH backgrounds in the same axion–NED theory and compute their coupled gravitational, electromagnetic, and axion QNMs. Rotation introduces spin dependence and coupling between angular harmonics, requiring an extension of the spherical reduction. Following the modes from the static limit would clarify how spin changes the five families, their splittings, and their responses to the axion and NED interactions, and would connect these results with rotating BH remnants relevant to gravitational-wave ringdown.
Heavy-particle matching and EFT interpretation
For a canonically normalized gauge field, integrating out heavy Dirac particles of masses m i m_i m i and charges Q i e D Q_i e_D Q i e D gives the one-loop interaction [68 , 70 , 90 ]
Δ L ( 4 ) = ∑ i ( Q i e D ) 4 360 π 2 m i 4 ( 4 F 2 + 7 G 2 ) . (54) \Delta\mathcal{L}^{(4)}=\sum_i\frac{(Q_i e_D)^4}{360\pi^2m_i^4}\left(4\mathcal{F}^2+7\mathcal{G}^2\right). \tag*{(54)} Δ L ( 4 ) = i ∑ 360 π 2 m i 4 ( Q i e D ) 4 ( 4 F 2 + 7 G 2 ) . (54)
The Maxwell correction is absorbed into the renormalized gauge coupling e D e_D e D . With α D = e D 2 / ( 4 π ) \alpha_D=e_D^2/(4\pi) α D = e D 2 / ( 4 π ) , this reproduces Eq. (2.5). For N f N_f N f identical species of mass m ⋆ m_\star m ⋆ and dimensionless charge q ⋆ q_\star q ⋆ , including multiplicities,
α = 8 N f α D 2 q ⋆ 4 45 m ⋆ 4 , β = 14 N f α D 2 q ⋆ 4 45 m ⋆ 4 . (55) \alpha=\frac{8N_f\alpha_D^2q_\star^4}{45m_\star^4},\qquad\beta=\frac{14N_f\alpha_D^2q_\star^4}{45m_\star^4}. \tag*{(55)} α = 45 m ⋆ 4 8 N f α D 2 q ⋆ 4 , β = 45 m ⋆ 4 14 N f α D 2 q ⋆ 4 . (55)
Thus the heavy-Dirac ratio is β / α = 7 / 4 \beta/\alpha=7/4 β / α = 7/4 ; other spins or interactions can change it [90 ] . The matching applies to ordinary or dark U ( 1 ) U(1) U ( 1 ) fields, with Q i e D Q_i e_D Q i e D interpreted as the coupling to the retained canonically normalized field. Additional light gauge fields would require their mixed operators to be retained or consistently truncated.
A local heavy-particle description requires
∣ ω ∣ , r h − 1 , ∣ K ∣ 1 / 4 ≪ m i , K = R μ ν ρ σ R μ ν ρ σ , (56) |\omega|,\quad r_h^{-1},\quad|\mathcal{K}|^{1/4}\ll m_i,\qquad\mathcal{K}=R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma}, \tag*{(56)} ∣ ω ∣ , r h − 1 , ∣ K ∣ 1/4 ≪ m i , K = R μν ρ σ R μν ρ σ , (56)
where K \mathcal{K} K is the Kretschmann scalar, together with
∣ Q i e D E ∣ ≪ m i 2 , ∣ Q i e D B ∣ ≪ m i 2 (57) |Q_i e_D E|\ll m_i^2,\qquad|Q_i e_D B|\ll m_i^2 \tag*{(57)} ∣ Q i e D E ∣ ≪ m i 2 , ∣ Q i e D B ∣ ≪ m i 2 (57)
throughout the exterior. Local gradients must also remain within the derivative expansion in regular freely falling frames; a static observer’s divergent horizon blueshift alone is not an EFT breakdown criterion. In the absence of large threshold logarithms, ∑ i α D Q i 2 / ( 4 π ) ≪ 1 \sum_i\alpha_D Q_i^2/(4\pi)\ll1 ∑ i α D Q i 2 / ( 4 π ) ≪ 1 is a conservative loop-expansion condition.
On the RN reference fields, E 2 + B 2 = q T 2 / r 4 E^2+B^2=q_T^2/r^4 E 2 + B 2 = q T 2 / r 4 gives
ϵ N E D max = 2 max ( ∣ α ^ ∣ , ∣ β ^ ∣ ) ( Q r h ) 2 . (58) \epsilon_{\mathrm{NED}}^{\max}=2\max(|\hat{\alpha}|,|\hat{\beta}|)\left(\frac{Q}{r_h}\right)^2. \tag*{(58)} ϵ NED m a x = 2 max ( ∣ α ^ ∣ , ∣ β ^ ∣ ) ( r h Q ) 2 . (58)
For axion-hairy backgrounds, we instead evaluate Eq. (18) using the electric field from Eq. (5) . The frequency responses in Appendix D include both direct operator corrections and the induced background changes at fixed Q / M Q/M Q / M , g ˉ \bar{g} g ˉ , and η \eta η .
Heavy-particle matching in curved spacetime also generates curvature and derivative operators, schematically R F μ ν F μ ν / m 2 RF_{\mu\nu}F^{\mu\nu}/m^2 R F μν F μν / m 2 and ( ∇ μ F μ ν ) 2 / m 2 (\nabla_\mu F^{\mu\nu})^2/m^2 ( ∇ μ F μν ) 2 / m 2 [96 ] . For a specific EFT these terms, and additional axion operators, must be included or shown to be negligible at the required accuracy in a consistent operator basis. Our calculation isolates the two quartic electromagnetic operators by setting the other coefficients to zero. The exterior amplitude criterion therefore controls the retained interactions, rather than the entire EFT or the central singularity.
Five coupled perturbation equations
We construct the five coupled radial equations using the mixed-parity conventions of Refs. [64 , 88 , 95 ] . As in Sec. III, we take m = 0 m=0 m = 0 , ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , and L = ℓ ( ℓ + 1 ) L=\ell(\ell+1) L = ℓ ( ℓ + 1 ) . For a single harmonic Y ℓ ≡ Y ℓ 0 Y_\ell\equiv Y_{\ell0} Y ℓ ≡ Y ℓ 0 , the even-parity metric perturbations H 0 H_0 H 0 , H 1 H_1 H 1 , H 2 H_2 H 2 , h 1 h_1 h 1 and the odd-parity metric perturbations Q Q Q , W W W are defined by
δ g t t = N h H 0 Y ℓ , δ g t r = H 1 Y ℓ , \delta g_{tt}=NhH_0Y_\ell,\qquad\delta g_{tr}=H_1Y_\ell, δ g tt = N h H 0 Y ℓ , δ g t r = H 1 Y ℓ ,
δ g r r = h − 1 H 2 Y ℓ , δ g r θ = h 1 ∂ θ Y ℓ , (59) \delta g_{rr}=h^{-1}H_2Y_\ell,\qquad\delta g_{r\theta}=h_1\partial_\theta Y_\ell, \tag*{(59)} δ g r r = h − 1 H 2 Y ℓ , δ g r θ = h 1 ∂ θ Y ℓ , (59)
δ g t ϕ = − Q sin θ ∂ θ Y ℓ , \delta g_{t\phi}=-Q\sin\theta\,\partial_\theta Y_\ell, δ g tϕ = − Q sin θ ∂ θ Y ℓ ,
δ g r ϕ = − W sin θ ∂ θ Y ℓ . (60) \delta g_{r\phi}=-W\sin\theta\,\partial_\theta Y_\ell. \tag*{(60)} δ g r ϕ = − W sin θ ∂ θ Y ℓ . (60)
All perturbation amplitudes depend on ( t , r ) (t,r) ( t , r ) , and the symmetry δ g μ ν = δ g ν μ \delta g_{\mu\nu}=\delta g_{\nu\mu} δ g μν = δ g ν μ is implicit. We fix the coordinate gauge by setting the purely angular metric perturbations and the even-parity t θ t\theta tθ component to zero. The U ( 1 ) U(1) U ( 1 ) gauge freedom is used to eliminate the even-parity angular electromagnetic component.
The electromagnetic and scalar harmonics are defined in Eq. (23) and its accompanying text. The pseudoscalar perturbation belongs to the electric odd sector; at nonzero magnetic charge, parity labels specify the harmonic basis without implying decoupled sectors.
Unreduced action and background coefficients
We use unit-normalized spherical harmonics. Expanding Eq. (2.3) to second order in the perturbations and integrating over the angular coordinates yields, for each mode,
S ( 2 ) = ∫ d t d r ( L 1 + L 2 ) . (61) S^{(2)}=\int dt\,dr\,(\mathcal{L}_1+\mathcal{L}_2). \tag*{(61)} S ( 2 ) = ∫ d t d r ( L 1 + L 2 ) . (61)
The unreduced radial densities contain background-dependent coefficients a i a_i a i , b i b_i b i , c i c_i c i , d i d_i d i , s i s_i s i , u i u_i u i , p i p_i p i . Up to integrations by parts, they are expressed using
E δ = δ A 0 ′ − δ A ˙ 1 , Z δ = r W ˙ − r Q ′ + 2 Q . (62) \mathcal{E}_{\delta}=\delta A'_0-\dot{\delta A}_1,\qquad\mathcal{Z}_{\delta}=r\dot{W}-rQ'+2Q. \tag*{(62)} E δ = δ A 0 ′ − δ A ˙ 1 , Z δ = r W ˙ − r Q ′ + 2 Q . (62)
The two parts of the Lagrangian density are
L 1 = a 0 H 0 2 + H 0 [ a 1 H 2 ′ + L a 2 h 1 ′ + ( a 3 + L a 4 ) H 2 + L a 5 h 1 + L a 6 δ A + a 7 δ φ ′ + a 8 δ φ ] + b 1 H 1 2 + H 1 ( b 2 H ˙ 2 + L b 3 h ˙ 1 + b 4 δ φ ˙ ) + c 0 H 2 2 + L H 2 ( c 1 h 1 + c 2 δ A ) + c 3 H 2 δ φ ′ + c 4 H 2 δ φ + L d 0 h ˙ 1 2 + L d 1 h 1 2 + L h 1 ( d 2 δ A 0 + d 3 δ A ′ + d 4 δ φ ) + s 1 E δ 2 + ( s 2 H 0 + s 3 H 2 + L s 4 δ A + s 6 δ φ ) E δ + L s 7 δ A 0 2 + L s 8 δ A 1 2 + u 1 δ φ ˙ 2 + u 2 δ φ ′ 2 + L u 3 δ φ 2 + u 5 δ φ δ A + L u 6 δ A 0 δ A , (63) \begin{aligned}
\mathcal{L}_1={}&a_0H_0^2+H_0\left[a_1H'_2+La_2h'_1+(a_3+La_4)H_2+La_5h_1+La_6\delta A+a_7\delta\varphi'+a_8\delta\varphi\right]\\
&+b_1H_1^2+H_1\left(b_2\dot{H}_2+Lb_3\dot{h}_1+b_4\dot{\delta\varphi}\right)\\
&+c_0H_2^2+LH_2(c_1h_1+c_2\delta A)\\
&+c_3H_2\delta\varphi'+c_4H_2\delta\varphi+Ld_0\dot{h}_1^2+Ld_1h_1^2\\
&+Lh_1(d_2\delta A_0+d_3\delta A'+d_4\delta\varphi)+s_1\mathcal{E}_{\delta}^2\\
&+(s_2H_0+s_3H_2+Ls_4\delta A+s_6\delta\varphi)\mathcal{E}_{\delta}\\
&+Ls_7\delta A_0^2+Ls_8\delta A_1^2\\
&+u_1\dot{\delta\varphi}^{\,2}+u_2\delta\varphi'^{\,2}+Lu_3\delta\varphi^2\\
&+u_5\delta\varphi\delta A+Lu_6\delta A_0\delta A, \tag*{(63)}
\end{aligned} L 1 = a 0 H 0 2 + H 0 [ a 1 H 2 ′ + L a 2 h 1 ′ + ( a 3 + L a 4 ) H 2 + L a 5 h 1 + L a 6 δ A + a 7 δ φ ′ + a 8 δ φ ] + b 1 H 1 2 + H 1 ( b 2 H ˙ 2 + L b 3 h ˙ 1 + b 4 δ φ ˙ ) + c 0 H 2 2 + L H 2 ( c 1 h 1 + c 2 δ A ) + c 3 H 2 δ φ ′ + c 4 H 2 δ φ + L d 0 h ˙ 1 2 + L d 1 h 1 2 + L h 1 ( d 2 δ A 0 + d 3 δ A ′ + d 4 δ φ ) + s 1 E δ 2 + ( s 2 H 0 + s 3 H 2 + L s 4 δ A + s 6 δ φ ) E δ + L s 7 δ A 0 2 + L s 8 δ A 1 2 + u 1 δ φ ˙ 2 + u 2 δ φ ′ 2 + L u 3 δ φ 2 + u 5 δ φ δ A + L u 6 δ A 0 δ A , (63)
L 2 = L [ p 1 Z δ 2 + p 2 δ A Z δ + p 3 δ A 2 + p 4 δ A ′ 2 + L p 5 δ A 2 + ( L p 6 + p 7 ) W 2 + ( L p 8 + p 9 ) Q 2 + p 10 Q δ A 0 + p 11 Q h 1 + p 12 W δ A 1 + p 13 δ A Q ] . (64) \begin{aligned}
\mathcal{L}_2=L\big[&p_1\mathcal{Z}_{\delta}^{\,2}+p_2\delta A\mathcal{Z}_{\delta}+p_3\delta A^2+p_4\delta A^{\prime2}\\
&+Lp_5\delta A^2+(Lp_6+p_7)W^2+(Lp_8+p_9)Q^2\\
&+p_{10}Q\delta A_0+p_{11}Qh_1\\
&+p_{12}W\delta A_1+p_{13}\delta A Q\big]. \tag*{(64)}
\end{aligned} L 2 = L [ p 1 Z δ 2 + p 2 δ A Z δ + p 3 δ A 2 + p 4 δ A ′2 + L p 5 δ A 2 + ( L p 6 + p 7 ) W 2 + ( L p 8 + p 9 ) Q 2 + p 10 Q δ A 0 + p 11 Q h 1 + p 12 W δ A 1 + p 13 δ A Q ] . (64)
These expressions specialize the mixed-parity action of Ref. [95 ] to Eq. (2.4). The coefficient labels refer to the terms in Eqs. (B.5) and (B.6) and need not coincide with those of the general action in that reference.
All coefficients below are evaluated on the background equations, and all functions have the same radial argument r r r . In terms of E E E , B B B , the constitutive coefficients a a a , d d d , c m c_m c m in Eq. (19) , b = 2 ( β − α ) E B b=2(\beta-\alpha)EB b = 2 ( β − α ) E B , and ρ E M \rho_{\mathrm{EM}} ρ EM in Eq. (9) , the metric coefficients are
a 0 = r 2 N d E 2 8 , a 1 = − M P l 2 r h N 2 , a_0=\frac{r^2\sqrt{N}\,dE^2}{8},\qquad a_1=-\frac{M_{\mathrm{Pl}}^2rh\sqrt{N}}{2}, a 0 = 8 r 2 N d E 2 , a 1 = − 2 M Pl 2 r h N ,
a 2 = − a 1 r , a 4 = − M P l 2 N 4 , a_2=-\frac{a_1}{r},\qquad a_4=-\frac{M_{\mathrm{Pl}}^2\sqrt{N}}{4}, a 2 = − r a 1 , a 4 = − 4 M Pl 2 N ,
a 3 = − N 4 ( 2 M P l 2 − 2 r 2 ρ E M + r 2 d E 2 ) , a_3=-\frac{\sqrt{N}}{4}\left(2M_{\mathrm{Pl}}^2-2r^2\rho_{\mathrm{EM}}+r^2dE^2\right), a 3 = − 4 N ( 2 M Pl 2 − 2 r 2 ρ EM + r 2 d E 2 ) ,
a 5 = M P l 2 N ( 2 h + r h ′ ) 4 r , a 6 = N 2 ( a B + b E ) , a_5=\frac{M_{\mathrm{Pl}}^2\sqrt{N}(2h+rh')}{4r},\qquad a_6=\frac{\sqrt{N}}{2}(aB+bE), a 5 = 4 r M Pl 2 N ( 2 h + r h ′ ) , a 6 = 2 N ( a B + b E ) ,
a 7 = r 2 h N ϕ ′ 2 , a 8 = 0. (65) a_7=\frac{r^2h\sqrt{N}\,\phi'}{2},\qquad a_8=0. \tag*{(65)} a 7 = 2 r 2 h N ϕ ′ , a 8 = 0. (65)
b 1 = L M P l 2 4 N , b 2 = M P l 2 r N , b_1=\frac{LM_{\mathrm{Pl}}^2}{4\sqrt{N}},\qquad b_2=\frac{M_{\mathrm{Pl}}^2r}{\sqrt{N}}, b 1 = 4 N L M Pl 2 , b 2 = N M Pl 2 r ,
b 3 = − M P l 2 2 N , b 4 = − r 2 ϕ ′ N . (66) b_3=-\frac{M_{\mathrm{Pl}}^2}{2\sqrt{N}},\qquad b_4=-\frac{r^2\phi'}{\sqrt{N}}. \tag*{(66)} b 3 = − 2 N M Pl 2 , b 4 = − N r 2 ϕ ′ . (66)
c 0 = − a 3 2 , c 1 = − M P l 2 [ r N h ′ + h ( 2 N + r N ′ ) ] 4 r N , c_0=-\frac{a_3}{2},\qquad c_1=-\frac{M_{\mathrm{Pl}}^2[rNh'+h(2N+rN')]}{4r\sqrt{N}}, c 0 = − 2 a 3 , c 1 = − 4 r N M Pl 2 [ r N h ′ + h ( 2 N + r N ′ )] ,
c 2 = − a 6 , c 3 = a 7 , c 4 = 0. (67) c_2=-a_6,\qquad c_3=a_7,\qquad c_4=0. \tag*{(67)} c 2 = − a 6 , c 3 = a 7 , c 4 = 0. (67)
d 0 = M P l 2 4 N , d 1 = h N 2 r 2 ( M P l 2 − r 2 a B 2 ) , d_0=\frac{M_{\mathrm{Pl}}^2}{4\sqrt{N}},\qquad d_1=\frac{h\sqrt{N}}{2r^2}(M_{\mathrm{Pl}}^2-r^2aB^2), d 0 = 4 N M Pl 2 , d 1 = 2 r 2 h N ( M Pl 2 − r 2 a B 2 ) ,
d 2 = − a E , d 3 = − a B h N , d 4 = h N ϕ ′ . (68) d_2=-aE,\qquad d_3=-aBh\sqrt{N},\qquad d_4=h\sqrt{N}\phi'. \tag*{(68)} d 2 = − a E , d 3 = − a B h N , d 4 = h N ϕ ′ . (68)
The electromagnetic coefficients entering the auxiliary-field construction below are
s 1 = r 2 d 2 N , s 2 = A 0 ′ s 1 , s 3 = − s 2 , s_1=\frac{r^2d}{2\sqrt{N}},\qquad s_2=A'_0s_1,\qquad s_3=-s_2, s 1 = 2 N r 2 d , s 2 = A 0 ′ s 1 , s 3 = − s 2 ,
s 4 = b , s 6 = − g q M , s_4=b,\qquad s_6=-gqM, s 4 = b , s 6 = − g q M ,
s 7 = a 2 h N , s 8 = − a h N 2 . (69) s_7=\frac{a}{2h\sqrt{N}},\qquad s_8=-\frac{ah\sqrt{N}}{2}. \tag*{(69)} s 7 = 2 h N a , s 8 = − 2 ah N . (69)
The odd metric and electromagnetic coefficients are
p 1 = M P l 2 4 r 2 N , p 2 = − a E r , p_1=\frac{M_{\mathrm{Pl}}^2}{4r^2\sqrt{N}},\qquad p_2=-\frac{aE}{r}, p 1 = 4 r 2 N M Pl 2 , p 2 = − r a E ,
p 3 = s 7 , p 4 = s 8 , p 5 = − N c e m 2 r 2 , p_3=s_7,\qquad p_4=s_8,\qquad p_5=-\frac{\sqrt{N}\,c_{\mathrm{em}}}{2r^2}, p 3 = s 7 , p 4 = s 8 , p 5 = − 2 r 2 N c em ,
p 6 = − M P l 2 h N 4 r 2 , p 7 = d 1 , p_6=-\frac{M_{\mathrm{Pl}}^2h\sqrt{N}}{4r^2},\qquad p_7=d_1, p 6 = − 4 r 2 M Pl 2 h N , p 7 = d 1 ,
p 8 = M P l 2 4 r 2 h N , p 9 = − M P l 2 − r 2 a B 2 2 r 2 h N , p_8=\frac{M_{\mathrm{Pl}}^2}{4r^2h\sqrt{N}},\qquad p_9=-\frac{M_{\mathrm{Pl}}^2-r^2aB^2}{2r^2h\sqrt{N}}, p 8 = 4 r 2 h N M Pl 2 , p 9 = − 2 r 2 h N M Pl 2 − r 2 a B 2 ,
p 10 = a B h N , p 11 = − a E B , p_{10}=\frac{aB}{h\sqrt{N}},\qquad p_{11}=-aEB, p 10 = h N a B , p 11 = − a E B ,
p 12 = d 3 , p 13 = B u 6 . (70) p_{12}=d_3,\qquad p_{13}=Bu_6. \tag*{(70)} p 12 = d 3 , p 13 = B u 6 . (70)
Finally, the scalar and scalar–electromagnetic coefficients are
u 1 = r 2 2 h N , u 2 = − r 2 h N 2 , u_1=\frac{r^2}{2h\sqrt{N}},\qquad u_2=-\frac{r^2h\sqrt{N}}{2}, u 1 = 2 h N r 2 , u 2 = − 2 r 2 h N ,
u 3 = − N 2 , u 5 = − L G A 0 ′ , u_3=-\frac{\sqrt{N}}{2},\qquad u_5=-\mathcal{L}_{G}A'_0, u 3 = − 2 N , u 5 = − L G A 0 ′ ,
u 6 = ( L , G ) ′ = 2 β G ′ + g ϕ ′ . (71) u_6=(\mathcal{L}_{,\mathcal{G}})'=2\beta\mathcal{G}'+g\phi'. \tag*{(71)} u 6 = ( L , G ) ′ = 2 β G ′ + g ϕ ′ . (71)
In the massless model, the terms δ ϕ ′ E δ \delta\phi'\mathcal{E}\delta δ ϕ ′ E δ , δ ϕ ′ δ A \delta\phi'\delta A δ ϕ ′ δ A , and ( δ ϕ ) 2 (\delta\phi)^2 ( δ ϕ ) 2 without L \mathcal{L} L vanish. Also, a 8 = c 4 = 0 a_8=c_4=0 a 8 = c 4 = 0 follows from L , ϕ = g G \mathcal{L}_{,\phi}=g\mathcal{G} L , ϕ = g G , L , G ϕ = g \mathcal{L}_{,\mathcal{G}\phi}=g L , G ϕ = g , and G = − E B \mathcal{G}=-EB G = − E B . The background simplifications involve no division by ϕ ′ \phi' ϕ ′ , so the hairless limit is regular. We keep M P l M_{\mathrm{Pl}} M Pl explicit; numerically M P l = 1 M_{\mathrm{Pl}}=1 M Pl = 1 .
Local constraint reduction
We linearize the squared combinations in Eq. (B.4) with the auxiliary fields V , χ V,\chi V , χ defined in Eq. (24) . Completing the squares is equivalent for s 1 p 1 ≠ 0 s_1p_1\ne0 s 1 p 1 = 0 . The same definitions give H 2 = ( Ψ + L h 1 ) / r H_2=(\Psi+Lh_1)/r H 2 = ( Ψ + L h 1 ) / r . The identity a 1 / r + a 2 = 0 a_1/r+a_2=0 a 1 / r + a 2 = 0 eliminates the h 1 ′ h'_1 h 1 ′ term from the H 0 H_0 H 0 constraint. The background equations further imply a 0 − s 2 2 / ( 4 s 1 ) = 0 a_0-s_2^2/(4s_1)=0 a 0 − s 2 2 / ( 4 s 1 ) = 0 , so the auxiliary-field action contains no H 0 2 H_0^2 H 0 2 term. Thus, H 0 H_0 H 0 acts as a Lagrange multiplier whose constraint is algebraic in h 1 h_1 h 1 .
The asymptotic time normalization in Eq. (15) rescales the auxiliary fields according to
V = V r a w C ∞ , χ = χ r a w C ∞ . (72) V=\frac{V_{\mathrm{raw}}}{\sqrt{C_\infty}},\qquad\chi=\frac{\chi_{\mathrm{raw}}}{\sqrt{C_\infty}}. \tag*{(72)} V = C ∞ V raw , χ = C ∞ χ raw . (72)
The other three fields are unchanged. The first and fourth regularized fields, defined later in Eq. (B.38), inherit the same factors. All operators use the normalized basis.
To express the H 0 H_0 H 0 constraint compactly, we introduce the background-dependent combinations
P Ψ = − a 1 r 2 + a 3 + L a 4 − s 2 s 3 / ( 2 s 1 ) r , (73) P_{\Psi}=-\frac{a_1}{r^2}+\frac{a_3+La_4-s_2s_3/(2s_1)}{r}, \tag*{(73)} P Ψ = − r 2 a 1 + r a 3 + L a 4 − s 2 s 3 / ( 2 s 1 ) , (73)
P h = L ( P Ψ + a 5 ) . (74) P_h=L(P_{\Psi}+a_5). \tag*{(74)} P h = L ( P Ψ + a 5 ) . (74)
For P h ≠ 0 P_h\ne0 P h = 0 , the constraint can be solved algebraically for h 1 h_1 h 1 :
h 1 = − 1 P h [ s 2 V + P Ψ Ψ + a 1 r Ψ ′ + a 7 δ ϕ ′ + ( a 8 − s 2 s 6 2 s 1 ) δ ϕ + L ( a 6 − s 2 s 4 2 s 1 ) δ A ] . (75) h_1=-\frac{1}{P_h}\left[s_2V+P_{\Psi}\Psi+\frac{a_1}{r}\Psi'+a_7\delta\phi'
+\left(a_8-\frac{s_2s_6}{2s_1}\right)\delta\phi
+L\left(a_6-\frac{s_2s_4}{2s_1}\right)\delta A\right]. \tag*{(75)} h 1 = − P h 1 [ s 2 V + P Ψ Ψ + r a 1 Ψ ′ + a 7 δ ϕ ′ + ( a 8 − 2 s 1 s 2 s 6 ) δ ϕ + L ( a 6 − 2 s 1 s 2 s 4 ) δ A ] . (75)
The original h 1 h_1 h 1 equation reconstructs H 0 H_0 H 0 because P h ≠ 0 P_h\ne0 P h = 0 , so eliminating this constraint and its multiplier loses no field equation. After integrating by parts, the pairs ( δ A 0 , Q ) (\delta A_0,Q) ( δ A 0 , Q ) and ( δ A 1 , W ) (\delta A_1,W) ( δ A 1 , W ) enter the action algebraically. For a real pair z z z with density z T H z + J T z z^T H z+J^Tz z T H z + J T z and symmetric invertible 2 × 2 2\times2 2 × 2 matrix H H H , its equation gives z = − H − 1 J / 2 z=-H^{-1}J/2 z = − H − 1 J /2 and contributes − J T H − 1 J / 4 -J^TH^{-1}J/4 − J T H − 1 J /4 after elimination.
For the pair z s = ( δ A 0 , Q ) T z_s=(\delta A_0,Q)^T z s = ( δ A 0 , Q ) T , the matrix is
H s = L ( s 7 p 10 / 2 p 10 / 2 L p 8 + p 9 ) . (76) H_s=L\begin{pmatrix}
s_7 & p_{10}/2\\
p_{10}/2 & Lp_8+p_9
\end{pmatrix}. \tag*{(76)} H s = L ( s 7 p 10 /2 p 10 /2 L p 8 + p 9 ) . (76)
and the source vector has components
J s = ( J s , 1 , J s , 2 ) T , J_s=(J_{s,1},J_{s,2})^T, J s = ( J s , 1 , J s , 2 ) T ,
J s , 1 = − 2 ( s 1 V ) ′ + L d 2 h 1 + L u 6 δ A , J_{s,1}=-2(s_1V)'+Ld_2h_1+Lu_6\delta A, J s , 1 = − 2 ( s 1 V ) ′ + L d 2 h 1 + L u 6 δ A ,
J s , 2 = 2 L [ r p 1 χ ′ + ( r p 1 ′ + 3 p 1 ) χ ] + L p 11 h 1 + L p 13 δ A . (77) J_{s,2}=2L[rp_1\chi'+(rp'_1+3p_1)\chi]+Lp_{11}h_1+Lp_{13}\delta A. \tag*{(77)} J s , 2 = 2 L [ r p 1 χ ′ + ( r p 1 ′ + 3 p 1 ) χ ] + L p 11 h 1 + L p 13 δ A . (77)
For the pair z t = ( δ A 1 , W ) T z_t=(\delta A_1,W)^T z t = ( δ A 1 , W ) T , the corresponding quantities are
H t = L ( s 8 p 12 / 2 p 12 / 2 L p 6 + p 7 ) , H_t=L\begin{pmatrix}
s_8 & p_{12}/2\\
p_{12}/2 & Lp_6+p_7
\end{pmatrix}, H t = L ( s 8 p 12 /2 p 12 /2 L p 6 + p 7 ) ,
J t = ( 2 s 1 V ˙ , − 2 L r p 1 χ ˙ ) T . (78) J_t=(2s_1\dot V,-2Lrp_1\dot\chi)^T. \tag*{(78)} J t = ( 2 s 1 V ˙ , − 2 L r p 1 χ ˙ ) T . (78)
To obtain the reduced matrices explicitly, we define
U a u x = s 3 H 2 + L s 4 δ A + s 6 δ ϕ , U_{\mathrm{aux}}=s_3H_2+Ls_4\delta A+s_6\delta\phi, U aux = s 3 H 2 + L s 4 δ A + s 6 δ ϕ ,
b H = ( 0 , b 2 / r , b 4 , 0 , 0 ) T , b_H=(0,b_2/r,b_4,0,0)^T, b H = ( 0 , b 2 / r , b 4 , 0 , 0 ) T ,
J H = b H T y ˙ = b 2 r Ψ ˙ + b 4 δ ϕ ˙ . (79) J_H=b_H^T\dot y=\frac{b_2}{r}\dot\Psi+b_4\delta\dot\phi. \tag*{(79)} J H = b H T y ˙ = r b 2 Ψ ˙ + b 4 δ ϕ ˙ . (79)
The terms containing H 1 H_1 H 1 are b 1 H 1 2 + H 1 ( J H + 2 b 1 h ˙ 1 ) b_1H_1^2+H_1(J_H+2b_1\dot h_1) b 1 H 1 2 + H 1 ( J H + 2 b 1 h ˙ 1 ) , since L ( b 2 / r + b 3 ) = 2 b 1 L(b_2/r+b_3)=2b_1 L ( b 2 / r + b 3 ) = 2 b 1 . Hence
H 1 = − h ˙ 1 − J H 2 b 1 . (80) H_1=-\dot h_1-\frac{J_H}{2b_1}. \tag*{(80)} H 1 = − h ˙ 1 − 2 b 1 J H . (80)
Its elimination cancels the original L d 0 h ˙ 1 2 Ld_0\dot{h}_1^2 L d 0 h ˙ 1 2 term because L d 0 = b 1 Ld_0=b_1 L d 0 = b 1 . After eliminating both algebraic pairs, the remaining spatial and kinetic densities are
L s p = c 0 H 2 2 + L c 1 H 2 h 1 + L c 2 H 2 δ A + c 3 H 2 δ ϕ ′ + c 4 H 2 δ ϕ + L d 1 h 1 2 + L d 3 h 1 δ A ′ + L d 4 h 1 δ ϕ − s 1 V 2 + V U a u x − U a u x 2 4 s 1 + u 2 δ ϕ ′ 2 + L u 3 δ ϕ 2 + u 5 δ ϕ δ A + L [ − p 1 χ 2 + p 2 δ A χ − p 2 2 4 p 1 δ A 2 + L p 5 δ A 2 + p 4 δ A ′ 2 ] − 1 4 J s T H s − 1 J s , (81) \begin{aligned}
\mathcal{L}_{\rm sp}={}&c_0H_2^2+Lc_1H_2h_1+Lc_2H_2\delta A\\
&+c_3H_2\delta\phi'+c_4H_2\delta\phi+Ld_1h_1^2\\
&+Ld_3h_1\delta A'+Ld_4h_1\delta\phi\\
&-s_1V^2+VU_{\rm aux}-\frac{U_{\rm aux}^2}{4s_1}\\
&+u_2\delta\phi'^2+Lu_3\delta\phi^2+u_5\delta\phi\delta A\\
&+L\left[-p_1\chi^2+p_2\delta A\chi-\frac{p_2^2}{4p_1}\delta A^2\right.\\
&\left.\qquad+Lp_5\delta A^2+p_4\delta A'^2\right]\\
&-\frac{1}{4}J_s^{T}H_s^{-1}J_s, \tag*{(81)}
\end{aligned} L sp = c 0 H 2 2 + L c 1 H 2 h 1 + L c 2 H 2 δ A + c 3 H 2 δ ϕ ′ + c 4 H 2 δ ϕ + L d 1 h 1 2 + L d 3 h 1 δ A ′ + L d 4 h 1 δ ϕ − s 1 V 2 + V U aux − 4 s 1 U aux 2 + u 2 δ ϕ ′2 + L u 3 δ ϕ 2 + u 5 δ ϕ δ A + L [ − p 1 χ 2 + p 2 δ A χ − 4 p 1 p 2 2 δ A 2 + L p 5 δ A 2 + p 4 δ A ′2 ] − 4 1 J s T H s − 1 J s , (81)
L k i n = u 1 δ ϕ ˙ 2 + L p 3 δ A ˙ 2 − J H 2 4 b 1 − J H h ˙ 1 − 1 4 J t T H t − 1 J t . (82) \mathcal{L}_{\rm kin}=u_1\dot{\delta\phi}^{\,2}+Lp_3\dot{\delta A}^{\,2}-\frac{J_H^2}{4b_1}-J_H\dot{h}_1-\frac{1}{4}J_t^{T}H_t^{-1}J_t. \tag*{(82)} L kin = u 1 δ ϕ ˙ 2 + L p 3 δ A ˙ 2 − 4 b 1 J H 2 − J H h ˙ 1 − 4 1 J t T H t − 1 J t . (82)
Here h 1 h_1 h 1 is given by Eq. (B.17), H 2 = ( Ψ + L h 1 ) / r H_2=(\Psi+Lh_1)/r H 2 = ( Ψ + L h 1 ) / r , and h ˙ 1 \dot{h}_1 h ˙ 1 is the time derivative of that expression. The reduction uses only the scalar divisions by s 1 , p 1 , P h , b 1 s_1,p_1,P_h,b_1 s 1 , p 1 , P h , b 1 and point-wise inverses of H s , H t H_s,H_t H s , H t ; their nonvanishing conditions are established below.
After these substitutions, treating the fields and their derivatives as independent variables specifies the spatial matrices through the Hessians
( M 0 ) I J = 1 2 ∂ 2 L s p ∂ y I ∂ y J , C I J = 1 2 ∂ 2 L s p ∂ y I ∂ y J ′ , G I J = 1 2 ∂ 2 L s p ∂ y I ′ ∂ y J ′ , (83) (M_0)_{IJ}=\frac{1}{2}\frac{\partial^2\mathcal{L}_{\rm sp}}{\partial y_I\partial y_J},\qquad
C_{IJ}=\frac{1}{2}\frac{\partial^2\mathcal{L}_{\rm sp}}{\partial y_I\partial y'_J},\qquad
G_{IJ}=\frac{1}{2}\frac{\partial^2\mathcal{L}_{\rm sp}}{\partial y'_I\partial y'_J}, \tag*{(83)} ( M 0 ) I J = 2 1 ∂ y I ∂ y J ∂ 2 L sp , C I J = 2 1 ∂ y I ∂ y J ′ ∂ 2 L sp , G I J = 2 1 ∂ y I ′ ∂ y J ′ ∂ 2 L sp , (83)
where I , J = 1 , … , 5 I,J=1,\ldots,5 I , J = 1 , … , 5 are field indices. Similarly,
( K 0 ) I J = 1 2 ∂ 2 L k i n ∂ y ˙ I ∂ y ˙ J , ( K 1 ) I J = 1 2 ∂ 2 L k i n ∂ y ˙ I ∂ y J ′ . (84) (K_0)_{IJ}=\frac{1}{2}\frac{\partial^2\mathcal{L}_{\rm kin}}{\partial\dot{y}_I\partial\dot{y}_J},\qquad
(K_1)_{IJ}=\frac{1}{2}\frac{\partial^2\mathcal{L}_{\rm kin}}{\partial\dot{y}_I\partial y'_J}. \tag*{(84)} ( K 0 ) I J = 2 1 ∂ y ˙ I ∂ y ˙ J ∂ 2 L kin , ( K 1 ) I J = 2 1 ∂ y ˙ I ∂ y J ′ ∂ 2 L kin . (84)
After the h ˙ 1 2 \dot{h}_1^2 h ˙ 1 2 cancellation, no term quadratic in y ˙ \dot{y} y ˙ remains. The factors of 1 / 2 1/2 1/2 follow from the conventions 2 y ′ T C y ′ 2y'^{T}Cy' 2 y ′ T C y ′ and 2 y ˙ T K 1 y ′ 2\dot{y}^{T}K_1y' 2 y ˙ T K 1 y ′ . The coefficients of y ′ y' y ′ in h 1 h_1 h 1 are h N b H / ( 2 P h ) hNb_H/(2P_h) h N b H / ( 2 P h ) , so the mixed kinetic matrix is symmetric:
K 1 = − h N 4 P h b H b H T = K 1 T , K = K 0 − K 1 ′ . (85) K_1=-\frac{hN}{4P_h}b_Hb_H^{T}=K_1^{T},\qquad
K=K_0-K_1'. \tag*{(85)} K 1 = − 4 P h h N b H b H T = K 1 T , K = K 0 − K 1 ′ . (85)
The symmetry of K 1 K_1 K 1 permits radial integration by parts, giving the action (25) with no derivatives higher than first order in the density. Substitution of the displayed coefficients and background equations then yields the exact identity G = − N h 2 K G=-Nh^2K G = − N h 2 K . Equations (B.23)–(B.27) provide a closed construction of every reduced matrix entry.
We examine the regularity of the constraint elimination on a smooth nonextremal exterior branch satisfying N , h > 0 N,h>0 N , h > 0 , a > 0 a>0 a > 0 , and d ≠ 0 d\ne0 d = 0 , where a a a and d d d are defined in Eq. (19) . The coefficient P h P_h P h in the h 1 h_1 h 1 constraint can be expressed in terms of the background combination
Ξ = L + r h ′ − 2 h + r 2 h ϕ ′ 2 M P l 2 . (86) \Xi=L+rh'-2h+\frac{r^2h\phi'^2}{M_{\rm Pl}^2}. \tag*{(86)} Ξ = L + r h ′ − 2 h + M Pl 2 r 2 h ϕ ′2 . (86)
Using the background equations, the coefficients entering the algebraic elimination take the factorized forms
s 1 = r 2 d 2 N , b 1 = L M P l 2 4 N , P h = − L M P l 2 N 4 r Ξ . (87) s_1=\frac{r^2d}{2\sqrt{N}},\qquad
b_1=\frac{LM_{\rm Pl}^2}{4\sqrt{N}},\qquad
P_h=-\frac{LM_{\rm Pl}^2\sqrt{N}}{4r}\Xi. \tag*{(87)} s 1 = 2 N r 2 d , b 1 = 4 N L M Pl 2 , P h = − 4 r L M Pl 2 N Ξ. (87)
The two constraint matrices have determinants
det H s = L 2 M P l 2 ( L − 2 ) a 8 r 2 N h 2 , (88) \det H_s=\frac{L^2M_{\rm Pl}^2(L-2)a}{8r^2Nh^2}, \tag*{(88)} det H s = 8 r 2 N h 2 L 2 M Pl 2 ( L − 2 ) a , (88)
det H t = L 2 M P l 2 ( L − 2 ) a N h 2 8 r 2 . (89) \det H_t=\frac{L^2M_{\rm Pl}^2(L-2)aNh^2}{8r^2}. \tag*{(89)} det H t = 8 r 2 L 2 M Pl 2 ( L − 2 ) a N h 2 . (89)
Under the stated assumptions and for ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , regularity at finite r > r h r>r_h r > r h therefore reduces to establishing that Ξ \Xi Ξ does not vanish.
Differentiating the electric Gauss law and using B ′ = − 2 B / r B'=-2B/r B ′ = − 2 B / r gives
ρ E M ′ = g E B ϕ ′ − 2 a r ( E 2 + B 2 ) . \rho_{\rm EM}'=gEB\phi'-\frac{2a}{r}(E^2+B^2). ρ EM ′ = g E B ϕ ′ − r 2 a ( E 2 + B 2 ) .
Together with the scalar and metric equations, this relation gives the integrating-factor identity
d d r [ r N { Ξ − ( L − 2 ) } ] = r 2 N M P l 2 [ 2 a ( E 2 + B 2 ) + ϕ ′ 2 ] . (90) \frac{\mathrm{d}}{\mathrm{d}r}\left[r\sqrt{N}\{\Xi-(L-2)\}\right]
=\frac{r^2\sqrt{N}}{M_{\rm Pl}^2}\left[2a(E^2+B^2)+\phi'^2\right]. \tag*{(90)} d r d [ r N { Ξ − ( L − 2 )} ] = M Pl 2 r 2 N [ 2 a ( E 2 + B 2 ) + ϕ ′2 ] . (90)
At a regular nonextremal horizon, 0 < N h < ∞ 0<N_h<\infty 0 < N h < ∞ and finite ϕ h ′ \phi'_h ϕ h ′ give Ξ h = L + r h h h ′ \Xi_h=L+ r_hh'_h Ξ h = L + r h h h ′ , with h h ′ > 0 h'_h>0 h h ′ > 0 . The integrated quantity therefore starts at r h N h ( 2 + r h h h ′ ) > 0 r_h\sqrt{N_h}(2+r_hh'_h)>0 r h N h ( 2 + r h h h ′ ) > 0 and is nondecreasing for a > 0 a>0 a > 0 . Hence Ξ > L − 2 > 0 \Xi>L-2>0 Ξ > L − 2 > 0 for every finite r > r h r>r_h r > r h and ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 . In particular, P h P_h P h cannot vanish there. The amplitude criterion in (18) implies a > 0.95 a>0.95 a > 0.95 and d > 0.85 d>0.85 d > 0.85 , including the signed NED probes used for differentiation.
With N ∞ = 1 \sqrt{N_\infty}=1 N ∞ = 1 and finite nonzero d h d_h d h , the rescaled pivots s 1 N = d / 2 s_1\sqrt{N}=d/2 s 1 N = d /2 , b 1 N = L M P l 2 / 4 b_1\sqrt{N}=LM_{\rm Pl}^2/4 b 1 N = L M Pl 2 /4 , and r P h / N = − L M P l 2 Ξ / 4 rP_h/\sqrt{N}=-LM_{\rm Pl}^2\Xi/4 r P h / N = − L M Pl 2 Ξ/4 have finite nonzero endpoint limits, with Ξ ∞ = L − 2 \Xi_\infty=L-2 Ξ ∞ = L − 2 . The endpoint powers of the raw constraint matrices are coordinate effects. The radial rescaling D r = diag ( 1 , r ) D_r=\operatorname{diag}(1,r) D r = diag ( 1 , r ) gives the common scaled block
H c = D r h N L r H s D r = − D r H t D r L h N = ( a / 2 a r B / 2 a r B / 2 M P l 2 ( L − 2 ) / 4 + a r 2 B 2 / 2 ) . (91) \begin{aligned}
H_c&=D_r\frac{h\sqrt{N}}{Lr}H_sD_r\\
&=-\frac{D_rH_tD_r}{Lh\sqrt{N}}\\
&=\begin{pmatrix}
a/2 & arB/2\\
arB/2 & M_{\rm Pl}^2(L-2)/4+ar^2B^2/2
\end{pmatrix}. \tag*{(91)}
\end{aligned} H c = D r L r h N H s D r = − L h N D r H t D r = ( a /2 a r B /2 a r B /2 M Pl 2 ( L − 2 ) /4 + a r 2 B 2 /2 ) . (91)
det H c = a M P l 2 ( L − 2 ) 8 . (92) \det H_{\mathrm c}=\frac{aM_{\mathrm{Pl}}^{2}(L-2)}{8}. \tag*{(92)} det H c = 8 a M Pl 2 ( L − 2 ) . (92)
Since H t = − N h 2 H s H_{\mathrm t}=-Nh^{2}H_{\mathrm s} H t = − N h 2 H s , both rescalings give the same block, with finite nonsingular horizon and infinity limits. For the five-field gradient matrix G ^ = T T G T \widehat{G}=T^{T}GT G = T T GT defined in Eq. (B.39),
det G ^ = − M P l 4 r 2 d 2 N h 7 32 ( r − r h ) 2 Ξ 2 . (93) \det\widehat{G}=-\frac{M_{\mathrm{Pl}}^{4}r^{2}d^{2}\sqrt{N}h^{7}}{32(r-r_{\mathrm h})^{2}\Xi^{2}}. \tag*{(93)} det G = − 32 ( r − r h ) 2 Ξ 2 M Pl 4 r 2 d 2 N h 7 . (93)
Thus the radial normal form is well defined in the open exterior. Substitution of the regular nonextremal horizon expansions shows that every entry of − G ^ / h -\widehat{G}/h − G / h has a finite limit. This limit is nonsingular because
det [ lim r → r h − G ^ h ] = M P l 4 r h 2 d h 2 N h ( h h ′ ) 2 32 Ξ h 2 . (94) \det\left[\lim_{r\to r_{\mathrm h}}\frac{-\widehat{G}}{h}\right]=\frac{M_{\mathrm{Pl}}^{4}r_{\mathrm h}^{2}d_{\mathrm h}^{2}\sqrt{N_{\mathrm h}}(h_{\mathrm h}')^{2}}{32\Xi_{\mathrm h}^{2}}. \tag*{(94)} det [ r → r h lim h − G ] = 32 Ξ h 2 M Pl 4 r h 2 d h 2 N h ( h h ′ ) 2 . (94)
At infinity,
lim r → ∞ ( − G ^ ) = diag ( 1 2 L , M P l 2 L ( L − 2 ) , 1 2 , M P l 2 L 4 ( L − 2 ) , L 2 ) . (95) \lim_{r\to\infty}(-\widehat{G})=\operatorname{diag}\left(\frac{1}{2L},\frac{M_{\mathrm{Pl}}^{2}}{L(L-2)},\frac{1}{2},\frac{M_{\mathrm{Pl}}^{2}L}{4(L-2)},\frac{L}{2}\right). \tag*{(95)} r → ∞ lim ( − G ) = diag ( 2 L 1 , L ( L − 2 ) M Pl 2 , 2 1 , 4 ( L − 2 ) M Pl 2 L , 2 L ) . (95)
The QNM boundary conditions use the corresponding regular limiting expansions, rather than raw constraint inverses evaluated at h = 0 h=0 h = 0 . These results justify the local reduction, without establishing global mode stability. Integrations by parts identify equivalent local field equations; QNM boundary conditions are imposed separately, without assuming vanishing endpoint fields.
The implementation evaluates the local matrices and their radial derivatives directly, using a directional complex-step method. The identity G + N h 2 K = 0 G+Nh^{2}K=0 G + N h 2 K = 0 holds at relative norm below 5 × 10 − 16 5\times10^{-16} 5 × 1 0 − 16 in the selected checks. A symbolic comparison establishes equality of the archived and independently reconstructed reduced Euler operators on the background equations: the corresponding densities differ by a total radial derivative after including the radius-dependent auxiliary-field transformation and its derivatives. The separate production-operator checks are given in Appendix C 1.
Explicit radial equations
For each Fourier component, the same symbol denotes the radial field after removing e − i ω t e^{-i\omega t} e − iω t . Here r r r is the areal radius, distinct from the tortoise coordinate r ∗ r_{*} r ∗ , with r ∗ ′ = 1 / ( h N ) r_{*}'=1/(h\sqrt{N}) r ∗ ′ = 1/ ( h N ) . Setting v = h N v=h\sqrt{N} v = h N and ∂ ∗ = ∂ / ∂ r ∗ = v d / d r \partial_{*}=\partial/\partial r_{*}=v\,d/dr ∂ ∗ = ∂ / ∂ r ∗ = v d / d r , we use
ψ 1 = r 2 V , ψ 2 = r − r h r Ψ , ψ 3 = r δ ϕ , \psi_{1}=r^{2}V,\qquad\psi_{2}=\frac{r-r_{\mathrm h}}{r}\Psi,\qquad\psi_{3}=r\delta\phi, ψ 1 = r 2 V , ψ 2 = r r − r h Ψ , ψ 3 = r δ ϕ ,
ψ 4 = χ , ψ 5 = δ A , y = T ψ , (96) \psi_{4}=\chi,\qquad\psi_{5}=\delta A,\qquad\mathbf{y}=T\boldsymbol{\psi}, \tag*{(96)} ψ 4 = χ , ψ 5 = δ A , y = T ψ , (96)
where T T T is the diagonal matrix in Eq. (27) . No additional frequency-dependent redefinition is made. The action coefficients in this basis are
G ^ = T T G T , C ^ = T T C T + T T G T ′ , \widehat{G}=T^{T}GT,\qquad
\widehat{C}=T^{T}CT+T^{T}GT', G = T T GT , C = T T C T + T T G T ′ ,
M ^ = T T M 0 T + T T C ′ T + T T C T T + T T G T ′ , (97) \widehat{M}=T^{T}M_{0}T+T^{T}C'T+T^{T}C^{T}T+T^{T}GT', \tag*{(97)} M = T T M 0 T + T T C ′ T + T T C T T + T T G T ′ , (97)
with K ^ = T T K T \widehat{K}=T^{T}KT K = T T K T and G ^ = − v 2 K ^ \widehat{G}=-v^{2}\widehat{K} G = − v 2 K . Every prime in the coefficient formulas is a derivative with respect to r r r , including the derivative of the background fields. Applying Eq. (26) in this basis and multiplying by − v 2 G ^ − 1 -v^{2}\widehat{G}^{-1} − v 2 G − 1 gives
∂ ∗ 2 ψ + A ∂ ∗ ψ + ( ω 2 I 5 + B ) ψ = 0 , (98) \partial_{*}^{2}\boldsymbol{\psi}+\mathcal{A}\partial_{*}\boldsymbol{\psi}+(\omega^{2}I_{5}+\mathcal{B})\boldsymbol{\psi}=0, \tag*{(98)} ∂ ∗ 2 ψ + A ∂ ∗ ψ + ( ω 2 I 5 + B ) ψ = 0 , (98)
where I 5 I_{5} I 5 is the 5 × 5 5\times5 5 × 5 identity. The real, frequency-independent 5 × 5 5\times5 5 × 5 matrices A ( r ) \mathcal{A}(r) A ( r ) and B ( r ) \mathcal{B}(r) B ( r ) multiply first tortoise derivatives and undifferentiated fields, respectively. Using ∂ ∗ 2 = v 2 d 2 / d r 2 + v v ′ d / d r \partial_{*}^{2}=v^{2}d^{2}/dr^{2}+vv'd/dr ∂ ∗ 2 = v 2 d 2 / d r 2 + v v ′ d / d r yields
A = v G ^ − 1 ( G ^ ′ − C ^ + C ^ T ) − v ′ I 5 , (99) \mathcal{A}=v\widehat{G}^{-1}\left(\widehat{G}'-\widehat{C}+\widehat{C}^{T}\right)-v'I_{5}, \tag*{(99)} A = v G − 1 ( G ′ − C + C T ) − v ′ I 5 , (99)
B = − v 2 G ^ − 1 ( M ^ − C ^ T ′ ) . (100) \mathcal{B}=-v^{2}\widehat{G}^{-1}\left(\widehat{M}-\widehat{C}^{T\prime}\right). \tag*{(100)} B = − v 2 G − 1 ( M − C T ′ ) . (100)
These expressions determine the radial matrices without assuming a symmetric potential. Here A I J \mathcal{A}_{IJ} A I J multiplies ∂ ∗ ψ J \partial_{*}\psi_{J} ∂ ∗ ψ J and B I J \mathcal{B}_{IJ} B I J multiplies ψ J \psi_{J} ψ J in the I I I th equation, with I , J = 1 , … , 5 I,J=1,\ldots,5 I , J = 1 , … , 5 ordered as in Eq. (B.38). Thus A 11 , B 11 \mathcal{A}_{11},\mathcal{B}_{11} A 11 , B 11 act on the first field, whereas A 12 , B 12 \mathcal{A}_{12},\mathcal{B}_{12} A 12 , B 12 couple its equation to the second field. The component equations are
0 = ∂ ∗ 2 ψ 1 + ω 2 ψ 1 + A 11 ∂ ∗ ψ 1 + A 12 ∂ ∗ ψ 2 + A 13 ∂ ∗ ψ 3 + A 14 ∂ ∗ ψ 4 + A 15 ∂ ∗ ψ 5 + B 11 ψ 1 + B 12 ψ 2 + B 13 ψ 3 + B 14 ψ 4 + B 15 ψ 5 , (101) \begin{aligned}
0={}&\partial_{*}^{2}\psi_{1}+\omega^{2}\psi_{1}+\mathcal{A}_{11}\partial_{*}\psi_{1}+\mathcal{A}_{12}\partial_{*}\psi_{2}+\mathcal{A}_{13}\partial_{*}\psi_{3}+\mathcal{A}_{14}\partial_{*}\psi_{4}+\mathcal{A}_{15}\partial_{*}\psi_{5}\\
&+\mathcal{B}_{11}\psi_{1}+\mathcal{B}_{12}\psi_{2}+\mathcal{B}_{13}\psi_{3}+\mathcal{B}_{14}\psi_{4}+\mathcal{B}_{15}\psi_{5}, \tag*{(101)}
\end{aligned} 0 = ∂ ∗ 2 ψ 1 + ω 2 ψ 1 + A 11 ∂ ∗ ψ 1 + A 12 ∂ ∗ ψ 2 + A 13 ∂ ∗ ψ 3 + A 14 ∂ ∗ ψ 4 + A 15 ∂ ∗ ψ 5 + B 11 ψ 1 + B 12 ψ 2 + B 13 ψ 3 + B 14 ψ 4 + B 15 ψ 5 , (101)
0 = ∂ ∗ 2 ψ 2 + ω 2 ψ 2 + A 21 ∂ ∗ ψ 1 + A 22 ∂ ∗ ψ 2 + A 23 ∂ ∗ ψ 3 + A 24 ∂ ∗ ψ 4 + A 25 ∂ ∗ ψ 5 + B 21 ψ 1 + B 22 ψ 2 + B 23 ψ 3 + B 24 ψ 4 + B 25 ψ 5 , (102) \begin{aligned}
0={}&\partial_{*}^{2}\psi_{2}+\omega^{2}\psi_{2}+\mathcal{A}_{21}\partial_{*}\psi_{1}+\mathcal{A}_{22}\partial_{*}\psi_{2}+\mathcal{A}_{23}\partial_{*}\psi_{3}+\mathcal{A}_{24}\partial_{*}\psi_{4}+\mathcal{A}_{25}\partial_{*}\psi_{5}\\
&+\mathcal{B}_{21}\psi_{1}+\mathcal{B}_{22}\psi_{2}+\mathcal{B}_{23}\psi_{3}+\mathcal{B}_{24}\psi_{4}+\mathcal{B}_{25}\psi_{5}, \tag*{(102)}
\end{aligned} 0 = ∂ ∗ 2 ψ 2 + ω 2 ψ 2 + A 21 ∂ ∗ ψ 1 + A 22 ∂ ∗ ψ 2 + A 23 ∂ ∗ ψ 3 + A 24 ∂ ∗ ψ 4 + A 25 ∂ ∗ ψ 5 + B 21 ψ 1 + B 22 ψ 2 + B 23 ψ 3 + B 24 ψ 4 + B 25 ψ 5 , (102)
0 = ∂ ∗ 2 ψ 3 + ω 2 ψ 3 + A 31 ∂ ∗ ψ 1 + A 32 ∂ ∗ ψ 2 + A 33 ∂ ∗ ψ 3 + A 34 ∂ ∗ ψ 4 + A 35 ∂ ∗ ψ 5 + B 31 ψ 1 + B 32 ψ 2 + B 33 ψ 3 + B 34 ψ 4 + B 35 ψ 5 , (103) \begin{aligned}
0={}&\partial_{*}^{2}\psi_{3}+\omega^{2}\psi_{3}+\mathcal{A}_{31}\partial_{*}\psi_{1}+\mathcal{A}_{32}\partial_{*}\psi_{2}+\mathcal{A}_{33}\partial_{*}\psi_{3}+\mathcal{A}_{34}\partial_{*}\psi_{4}+\mathcal{A}_{35}\partial_{*}\psi_{5}\\
&+\mathcal{B}_{31}\psi_{1}+\mathcal{B}_{32}\psi_{2}+\mathcal{B}_{33}\psi_{3}+\mathcal{B}_{34}\psi_{4}+\mathcal{B}_{35}\psi_{5}, \tag*{(103)}
\end{aligned} 0 = ∂ ∗ 2 ψ 3 + ω 2 ψ 3 + A 31 ∂ ∗ ψ 1 + A 32 ∂ ∗ ψ 2 + A 33 ∂ ∗ ψ 3 + A 34 ∂ ∗ ψ 4 + A 35 ∂ ∗ ψ 5 + B 31 ψ 1 + B 32 ψ 2 + B 33 ψ 3 + B 34 ψ 4 + B 35 ψ 5 , (103)
0 = ∂ ∗ 2 ψ 4 + ω 2 ψ 4 + A 41 ∂ ∗ ψ 1 + A 42 ∂ ∗ ψ 2 + A 43 ∂ ∗ ψ 3 + A 44 ∂ ∗ ψ 4 + A 45 ∂ ∗ ψ 5 + B 41 ψ 1 + B 42 ψ 2 + B 43 ψ 3 + B 44 ψ 4 + B 45 ψ 5 , (104) \begin{aligned}
0={}&\partial_{*}^{2}\psi_{4}+\omega^{2}\psi_{4}+\mathcal{A}_{41}\partial_{*}\psi_{1}+\mathcal{A}_{42}\partial_{*}\psi_{2}+\mathcal{A}_{43}\partial_{*}\psi_{3}+\mathcal{A}_{44}\partial_{*}\psi_{4}+\mathcal{A}_{45}\partial_{*}\psi_{5}\\
&+\mathcal{B}_{41}\psi_{1}+\mathcal{B}_{42}\psi_{2}+\mathcal{B}_{43}\psi_{3}+\mathcal{B}_{44}\psi_{4}+\mathcal{B}_{45}\psi_{5}, \tag*{(104)}
\end{aligned} 0 = ∂ ∗ 2 ψ 4 + ω 2 ψ 4 + A 41 ∂ ∗ ψ 1 + A 42 ∂ ∗ ψ 2 + A 43 ∂ ∗ ψ 3 + A 44 ∂ ∗ ψ 4 + A 45 ∂ ∗ ψ 5 + B 41 ψ 1 + B 42 ψ 2 + B 43 ψ 3 + B 44 ψ 4 + B 45 ψ 5 , (104)
0 = ∂ ∗ 2 ψ 5 + ω 2 ψ 5 + A 51 ∂ ∗ ψ 1 + A 52 ∂ ∗ ψ 2 + A 53 ∂ ∗ ψ 3 + A 54 ∂ ∗ ψ 4 + A 55 ∂ ∗ ψ 5 + B 51 ψ 1 + B 52 ψ 2 + B 53 ψ 3 + B 54 ψ 4 + B 55 ψ 5 . (105) 0=\partial_*^2\psi_5+\omega^2\psi_5+A_{51}\partial_*\psi_1+A_{52}\partial_*\psi_2+A_{53}\partial_*\psi_3+A_{54}\partial_*\psi_4+A_{55}\partial_*\psi_5+B_{51}\psi_1+B_{52}\psi_2+B_{53}\psi_3+B_{54}\psi_4+B_{55}\psi_5. \tag*{(105)} 0 = ∂ ∗ 2 ψ 5 + ω 2 ψ 5 + A 51 ∂ ∗ ψ 1 + A 52 ∂ ∗ ψ 2 + A 53 ∂ ∗ ψ 3 + A 54 ∂ ∗ ψ 4 + A 55 ∂ ∗ ψ 5 + B 51 ψ 1 + B 52 ψ 2 + B 53 ψ 3 + B 54 ψ 4 + B 55 ψ 5 . (105)
The diagonal frequency term follows from the exact finite-L L L identity G ^ = − N h 2 K ^ \widehat{G}=-Nh^2\widehat{K} G = − N h 2 K , verified directly from the quadratic action. It does not require diagonalization of the full operator. In particular, A \mathcal{A} A need not vanish and B \mathcal{B} B need not be symmetric in these variables. On the regular purely electric branch, q M = 0 q_M=0 q M = 0 implies ϕ ′ = 0 \phi'=0 ϕ ′ = 0 and ϕ = 0 \phi=0 ϕ = 0 under the chosen asymptotic condition. The matrix entries connecting { 1 , 2 } \{1,2\} { 1 , 2 } to { 3 , 4 , 5 } \{3,4,5\} { 3 , 4 , 5 } then vanish, giving the even two-field and odd three-field systems. For q M ≠ 0 q_M\ne0 q M = 0 , Eqs. (B.43) are solved together with the ingoing/outgoing conditions (29) . Factoring the phase (30) and changing to x = r h / r x=r_h/r x = r h / r gives precisely Eq. (31) ; Eqs. (B.41) and (B.42) are the matrices used in its implementation.
Numerical validation and response accuracy
This appendix validates the radial matrices and tests the numerical accuracy of the frequencies and NED responses in Secs. III–§ 2.2 .
Radial matrices and RN parity-pair checks
We check the numerical implementation of the radial matrices A , B \mathcal{A},\mathcal{B} A , B in Eqs. (B.41) and (B.42) against expressions derived separately from the quadratic action. Both calculations use the same perturbation variables, the time normalization N ( ∞ ) = 1 N(\infty)=1 N ( ∞ ) = 1 , and units r h = M P l = 1 r_h=M_{\rm Pl}=1 r h = M Pl = 1 . Let X n u m X_{\rm num} X num and X s y m X_{\rm sym} X sym denote the numerical and symbolic expressions for either radial matrix X = A , B X=\mathcal{A},\mathcal{B} X = A , B . We use ∥ X ∥ ≡ ( ∑ i , j = 1 5 ∣ X i j ∣ 2 ) 1 / 2 \lVert X\rVert\equiv\left(\sum_{i,j=1}^{5}|X_{ij}|^2\right)^{1/2} ∥ X ∥ ≡ ( ∑ i , j = 1 5 ∣ X ij ∣ 2 ) 1/2 , where X i j X_{ij} X ij are the matrix entries. The normalized difference is
ϵ X = ∥ X n u m − X s y m ∥ max { 1 , ∥ X n u m ∥ , ∥ X s y m ∥ } . (106) \epsilon_X=\frac{\lVert X_{\rm num}-X_{\rm sym}\rVert}{\max\{1,\lVert X_{\rm num}\rVert,\lVert X_{\rm sym}\rVert\}}. \tag*{(106)} ϵ X = max { 1 , ∥ X num ∥ , ∥ X sym ∥} ∥ X num − X sym ∥ . (106)
For five backgrounds covering positive and negative NED coefficients, ϵ X < 6.5 × 10 − 13 \epsilon_X<6.5\times10^{-13} ϵ X < 6.5 × 1 0 − 13 at 25 exterior points in total, five on each background.
For the gravitational and EM RN pairs of § 2.2 C, each parity sector has one independent QNM eigenvector at the reference frequency for Q / M = 0.6 Q/M=0.6 Q / M = 0.6 , ℓ = 2 \ell=2 ℓ = 2 , and N c = 36 , 48 , 60 N_c=36,48,60 N c = 36 , 48 , 60 . The 2 × 2 2\times2 2 × 2 matrix N J N_J N J multiplying the first-order frequency change in each pair’s equation is diagonal with nonzero entries and hence invertible.
Frequency and NED-shift convergence
Table II compares the frequencies at different radial resolutions for the Q / M = 0.8 Q/M=0.8 Q / M = 0.8 benchmark of § 2 D. Other frequency and background-tolerance tests appear in Sec. III. A NED shift is the difference between frequencies with and without NED terms; both must be computed at the same resolution and background tolerance. For every point in Figure 5 and Figure 6 , both frequencies are evaluated at N c = 36 , 48 N_c=36,48 N c = 36 , 48 . At the largest plotted coefficients, λ = 0.05 \lambda=0.05 λ = 0.05 and 0.16 0.16 0.16 , respectively, both are also evaluated at N c = 60 N_c=60 N c = 60 . In the recorded tests across all five families, changes in Δ Ω j \Delta\Omega_j Δ Ω j are below 2.0 × 10 − 11 2.0\times10^{-11} 2.0 × 1 0 − 11 between 36 and 48 nodes. At the largest-coefficient points, changes between 48 and 60 nodes are below 3.6 × 10 − 11 3.6\times10^{-11} 3.6 × 1 0 − 11 . At these points, tightening the background relative tolerance from 2 × 10 − 11 2\times10^{-11} 2 × 1 0 − 11 to 2 × 10 − 12 2\times10^{-12} 2 × 1 0 − 12 changes these shifts by less than 2.1 × 10 − 11 2.1\times10^{-11} 2.1 × 1 0 − 11 . Both figures use N c = 48 N_c=48 N c = 48 .
Additional growing modes
The growing modes in § 2 D have Ω = i γ \Omega=i\gamma Ω = iγ , γ > 0 \gamma>0 γ > 0 , and grow as e γ t / M e^{\gamma t/M} e γ t / M . Table III lists growth rates checked at N c = 36 , 48 , 60 N_c=36,48,60 N c = 36 , 48 , 60 . These backgrounds satisfy the local conditions (19) but support both growing and damped modes.
For the five-component field ψ \psi ψ defined in Eq. (27) , the radial equation (B.40) at ω = i γ / M \omega=i\gamma/M ω = iγ / M becomes
( ∂ ∗ 2 + A ∂ ∗ + B ) ψ = ( γ M ) 2 ψ . (107) (\partial_*^2+\mathcal{A}\partial_*+\mathcal{B})\psi=\left(\frac{\gamma}{M}\right)^2\psi. \tag*{(107)} ( ∂ ∗ 2 + A ∂ ∗ + B ) ψ = ( M γ ) 2 ψ . (107)
The QNM conditions (29) require spatial decay here, which approximate by zero field values at both finite boundaries. We use second-order finite differences with 2001, 4001, and 8001 nodes on r ∗ / r h ∈ [ − 70 , 130 ] r_*/r_h\in[-70,130] r ∗ / r h ∈ [ − 70 , 130 ] , with r ∗ = 0 r_*=0 r ∗ = 0 at r = 3 r h r=3r_h r = 3 r h . The matrices A , B \mathcal{A},\mathcal{B} A , B are replaced by their zero horizon limits only for ( r − r h ) / r h < 10 − 5 (r-r_h)/r_h<10^{-5} ( r − r h ) / r h < 1 0 − 5 . Doubling the radial samples for coefficient interpolation from 700 to 1400 changes the growth rates by less than 5 × 10 − 11 5\times10^{-11} 5 × 1 0 − 11 . With 8001 nodes and this 10 − 5 10^{-5} 1 0 − 5 cutoff, the growth rates agree with the spectral results within 4.1 × 10 − 7 4.1\times10^{-7} 4.1 × 1 0 − 7 . Reducing the cutoff to 10 − 6 10^{-6} 1 0 − 6 changes γ \gamma γ at Q / M = 0.6 Q/M=0.6 Q / M = 0.6 by about 7 × 10 − 7 7\times10^{-7} 7 × 1 0 − 7 . This horizon-replacement sensitivity is separate from the interpolation test; neither is a total error bound. For the electric limit Q / M = 0.8 Q/M=0.8 Q / M = 0.8 , g ˉ = 10 \bar{g}=10 g ˉ = 10 , η = 0 \eta=0 η = 0 , and α ^ = β ^ = 0 \hat{\alpha}=\hat{\beta}=0 α ^ = β ^ = 0 , a separate integration-and-matching calculation of Eq. (34) also reproduces Ω ≃ 0.14456 i , 0.42539 i \Omega\simeq0.14456i,0.42539i Ω ≃ 0.14456 i , 0.42539 i . These tests confirm the listed unstable backgrounds without determining the instability thresholds as the parameters vary.
TABLE II. Spectral-resolution test for the five oscillatory ℓ = 2 \ell= 2 ℓ = 2 families continued from the fundamental RN modes at Q / M = 0.8 Q/M = 0.8 Q / M = 0.8 , g ˉ = 10 \bar{g} = 10 g ˉ = 10 , η = 0.1 \eta= 0.1 η = 0.1 , and α ^ = β ^ = 0 \hat{\alpha} = \hat{\beta} = 0 α ^ = β ^ = 0 . The frequency column gives the N c = 60 N_c = 60 N c = 60 result, with Ω n ≡ Ω ( N c = n ) \Omega_n \equiv\Omega(N_c = n) Ω n ≡ Ω ( N c = n ) . This background also supports the growing mode listed in Table III . Frequency differences are calculated before rounding.
TABLE III. Growing ℓ = 2 \ell= 2 ℓ = 2 modes at g ˉ = 10 \bar{g} = 10 g ˉ = 10 , η = 0.1 \eta= 0.1 η = 0.1 , and α ^ = β ^ = 0 \hat{\alpha} = \hat{\beta} = 0 α ^ = β ^ = 0 . The frequencies have Re Ω = 0 \operatorname{Re}\Omega= 0 Re Ω = 0 ; γ = Im Ω > 0 \gamma= \operatorname{Im}\Omega> 0 γ = Im Ω > 0 is the dimensionless growth rate. Values are rounded to five decimal places.
Accuracy of the NED expansion
For each mode in the NED expansion of § 2.2 D, we suppress the family index j j j and write Δ Ω = λ κ + λ 2 L 2 + ⋯ \Delta\Omega= \lambda\kappa+ \lambda^2 L_2 + \cdots ΔΩ = λκ + λ 2 L 2 + ⋯ along each coefficient sequence. Here κ \kappa κ and L 2 L_2 L 2 are the linear and quadratic coefficients. When only α ^ \hat{\alpha} α ^ is varied, with β ^ = 0 \hat{\beta} = 0 β ^ = 0 , we denote them by κ α \kappa_{\alpha} κ α and L 2 , α L_{2,\alpha} L 2 , α . When only β ^ \hat{\beta} β ^ is varied, with α ^ = 0 \hat{\alpha} = 0 α ^ = 0 , we denote them by κ β \kappa_{\beta} κ β and L 2 , β L_{2,\beta} L 2 , β . The mixed derivative is H α β = ( ∂ α ^ ∂ β ^ Ω ) α ^ = β ^ = 0 H_{\alpha\beta} = (\partial_{\hat{\alpha}}\partial_{\hat{\beta}}\Omega)_{\hat{\alpha}=\hat{\beta}=0} H α β = ( ∂ α ^ ∂ β ^ Ω ) α ^ = β ^ = 0 . Finite differences use steps δ λ = 0.003 \delta\lambda= 0.003 δ λ = 0.003 for κ \kappa κ and 0.015 0.015 0.015 for L 2 L_2 L 2 . For the resolution test, both are recomputed from λ = 0 \lambda= 0 λ = 0 , ± 0.015 \pm0.015 ± 0.015 , ± 0.030 \pm0.030 ± 0.030 . The second-order prediction at λ = 0.05 \lambda= 0.05 λ = 0.05 , beyond these sampled values, changes by at most 1.6 × 10 − 10 1.6 \times10^{-10} 1.6 × 1 0 − 10 over N c = 36 , 48 , 60 N_c = 36, 48, 60 N c = 36 , 48 , 60 for ( α ^ , β ^ ) = ( λ , 0 ) , ( 0 , λ ) , ( λ , λ ) (\hat{\alpha},\hat{\beta}) = (\lambda, 0),(0, \lambda),(\lambda,\lambda) ( α ^ , β ^ ) = ( λ , 0 ) , ( 0 , λ ) , ( λ , λ ) .
For the even-EM example with ( α ^ , β ^ ) = ( 4 λ / 7 , λ ) (\hat{\alpha},\hat{\beta}) = (4\lambda/7,\lambda) ( α ^ , β ^ ) = ( 4 λ /7 , λ ) in Figure 5 , the coefficients are κ = ( 4 / 7 ) κ α + κ β \kappa= (4/7)\kappa_{\alpha} + \kappa_{\beta} κ = ( 4/7 ) κ α + κ β and L 2 = ( 16 / 49 ) L 2 , α + L 2 , β + ( 4 / 7 ) H α β L_2 = (16/49)L_{2,\alpha} + L_{2,\beta} + (4/7)H_{\alpha\beta} L 2 = ( 16/49 ) L 2 , α + L 2 , β + ( 4/7 ) H α β . At λ = 0.05 \lambda= 0.05 λ = 0.05 , the absolute differences between predicted and directly computed complex shifts are 7.8 × 10 − 7 7.8 \times10^{-7} 7.8 × 1 0 − 7 at first order, 3.2 × 10 − 9 3.2 \times10^{-9} 3.2 × 1 0 − 9 at second order, and 3.4 × 10 − 7 3.4 \times10^{-7} 3.4 × 1 0 − 7 without the mixed term. These exceed the resolution changes above and support the improvement discussed in § 2.2 D. The remaining difference includes numerical error and higher-order Taylor terms.
We give the frequency derivatives and degenerate-pair projections used in § 2.2 C and V D. At fixed ( Q / M , g ˉ , η ) (Q/M,\bar{g},\eta) ( Q / M , g ˉ , η ) , the derivatives of the eigenfrequencies in Eq. (43) are
K j , p = ∂ p Ω j ∣ α ^ = β ^ = 0 , p = α ^ , β ^ . (108) K_{j,p} = \left.\partial_p\Omega_j\right|_{\hat{\alpha}=\hat{\beta}=0}, \qquad p = \hat{\alpha},\hat{\beta}. \tag*{(108)} K j , p = ∂ p Ω j ∣ α ^ = β ^ = 0 , p = α ^ , β ^ . (108)
Derivative labels α , β \alpha,\beta α , β below refer to the hatted coefficients. For an isolated simple eigenfrequency, the full NED shift has the local expansion
Δ Ω j = α ^ K j , α + β ^ K j , β + 1 2 α ^ 2 H j , α α \Delta\Omega_j = \hat{\alpha}K_{j,\alpha} + \hat{\beta}K_{j,\beta} + \frac{1}{2}\hat{\alpha}^{2}H_{j,\alpha\alpha} Δ Ω j = α ^ K j , α + β ^ K j , β + 2 1 α ^ 2 H j , α α
Δ Ω j = + α ^ β ^ H j , α β + 1 2 β ^ 2 H j , β β \phantom{\Delta\Omega_j = {}} + \hat{\alpha}\hat{\beta}H_{j,\alpha\beta} + \frac{1}{2}\hat{\beta}^{2}H_{j,\beta\beta} Δ Ω j = + α ^ β ^ H j , α β + 2 1 β ^ 2 H j , β β
Δ Ω j = + O ( ( ∣ α ^ ∣ + ∣ β ^ ∣ ) 3 ) , (109) \phantom{\Delta\Omega_j = {}} + O\left((\lvert\hat{\alpha}\rvert+\lvert\hat{\beta}\rvert)^3\right), \tag*{(109)} Δ Ω j = + O ( (∣ α ^ ∣ + ∣ β ^ ∣ ) 3 ) , (109)
where H j , p q = ∂ p ∂ q Ω j ∣ α ^ = β ^ = 0 H_{j,pq} = \left.\partial_p\partial_q\Omega_j\right|_{\hat{\alpha}=\hat{\beta}=0} H j , pq = ∂ p ∂ q Ω j ∣ α ^ = β ^ = 0 , with p , q ∈ { α ^ , β ^ } p,q \in\{\hat{\alpha},\hat{\beta}\} p , q ∈ { α ^ , β ^ } . This approximates the directly computed finite-NED frequencies; it is not a truncation imposed on the solver. A higher-order prediction for a specific EFT would also require the operators discussed in Appendix A.
Let P ( Ω , p ) \mathcal{P}(\Omega,p) P ( Ω , p ) denote either the spectral matrix in Eq. (31) or the matrix imposing the matching conditions in the integration method of Sec. III. In both cases, we express the frequency as ω = Ω / M ( p ) \omega= \Omega/M(p) ω = Ω/ M ( p ) . At a QNM frequency, nonzero right and left vectors satisfy P e R = 0 \mathcal{P}e_R = 0 P e R = 0 and e L † P = 0 e_L^{\dagger}\mathcal{P} = 0 e L † P = 0 , where a dagger denotes complex-conjugate transpose. The matrices P , p \mathcal{P}_{,p} P , p and P , Ω \mathcal{P}_{,\Omega} P , Ω are obtained by differentiating each matrix entry with respect to p p p at fixed Ω \Omega Ω and to Ω \Omega Ω at fixed p p p , respectively. For an isolated, nondegenerate QNM frequency,
∂ p Ω = − e L † P , p e R e L † P , Ω e R . (110) \partial_p\Omega= -\frac{e_L^{\dagger}\mathcal{P}_{,p}e_R}{e_L^{\dagger}\mathcal{P}_{,\Omega}e_R}. \tag*{(110)} ∂ p Ω = − e L † P , Ω e R e L † P , p e R . (110)
provided the denominator is nonzero. The p p p derivative includes changes in the background, ADM mass, and boundary phases, with ( Q / M , g ˉ , η ) (Q/M,\bar{g},\eta) ( Q / M , g ˉ , η ) held fixed.
Along ( α ^ , β ^ ) = λ ( α λ , β λ ) (\hat{\alpha},\hat{\beta}) = \lambda(\alpha_{\lambda},\beta_{\lambda}) ( α ^ , β ^ ) = λ ( α λ , β λ ) , the coefficients introduced in Appendix C are κ = Ω ′ ( 0 ) \kappa= \Omega'(0) κ = Ω ′ ( 0 ) and L 2 = Ω ′ ′ ( 0 ) / 2 L_2 = \Omega''(0)/2 L 2 = Ω ′′ ( 0 ) /2 , with the family index j j j omitted on κ , L 2 , H p q \kappa,L_2,H_{pq} κ , L 2 , H pq . Primes denote λ \lambda λ derivatives. In particular, κ α = K j , α \kappa_{\alpha} = K_{j,\alpha} κ α = K j , α and κ β = K j , β \kappa_{\beta} = K_{j,\beta} κ β = K j , β . For Ω n = Ω ( n δ λ ) \Omega_n = \Omega(n\delta\lambda) Ω n = Ω ( n δ λ ) , fourth-order centered approximations are
κ ( δ λ ) = Ω − 2 − 8 Ω − 1 + 8 Ω 1 − Ω 2 12 δ λ , \kappa(\delta\lambda) = \frac{\Omega_{-2} - 8\Omega_{-1} + 8\Omega_1 - \Omega_2}{12\delta\lambda}, κ ( δ λ ) = 12 δ λ Ω − 2 − 8 Ω − 1 + 8 Ω 1 − Ω 2 ,
L 2 ( δ λ ) = − Ω 2 + 16 Ω 1 − 30 Ω 0 + 16 Ω − 1 − Ω − 2 24 ( δ λ ) 2 . (111) L_2(\delta\lambda) = \frac{-\Omega_2 + 16\Omega_1 - 30\Omega_0 + 16\Omega_{-1} - \Omega_{-2}}{24(\delta\lambda)^2}. \tag*{(111)} L 2 ( δ λ ) = 24 ( δ λ ) 2 − Ω 2 + 16 Ω 1 − 30 Ω 0 + 16 Ω − 1 − Ω − 2 . (111)
The denominator 24 24 24 accounts for L 2 = Ω ′ ′ ( 0 ) / 2 L_2 = \Omega''(0)/2 L 2 = Ω ′′ ( 0 ) /2 . Let L 2 , e q L_{2,\mathrm{eq}} L 2 , eq denote the quadratic coefficient along ( α ^ , β ^ ) = ( λ , λ ) (\hat{\alpha},\hat{\beta}) = (\lambda,\lambda) ( α ^ , β ^ ) = ( λ , λ ) . The second derivatives then satisfy H α α = 2 L 2 , α H_{\alpha\alpha} = 2L_{2,\alpha} H α α = 2 L 2 , α , H β β = 2 L 2 , β H_{\beta\beta}=2L_{2,\beta} H β β = 2 L 2 , β , and H α β = L 2 , e q − L 2 , α − L 2 , β H_{\alpha\beta}=L_{2,\mathrm{eq}}-L_{2,\alpha}-L_{2,\beta} H α β = L 2 , eq − L 2 , α − L 2 , β . The numerical settings and resolution checks are given in Appendix C 4.
At η = g ˉ = α ^ = β ^ = 0 \eta=\bar{g}=\hat{\alpha}=\hat{\beta}=0 η = g ˉ = α ^ = β ^ = 0 , each RN parity pair J = g r a v , E M J=\mathrm{grav},\mathrm{EM} J = grav , EM in Eq. (48) consists of independent even and odd modes with the same frequency. We define V J = ( e R , J , e , e R , J , o ) V_J=(e_{R,J,e},e_{R,J,o}) V J = ( e R , J , e , e R , J , o ) and U J = ( e L , J , e , e L , J , o ) U_J=(e_{L,J,e},e_{L,J,o}) U J = ( e L , J , e , e L , J , o ) , where s = e , o s=e,o s = e , o labels parity and the vectors satisfy P e R , J , s = 0 \mathcal{P}e_{R,J,s}=0 P e R , J , s = 0 and e L , J , s † P = 0 e_{L,J,s}^{\dagger}\mathcal{P}=0 e L , J , s † P = 0 at the common RN frequency. Projecting onto these two modes gives the 2 × 2 2\times2 2 × 2 matrices
N J = U J † P , Ω V J , N_J=U_J^\dagger\mathcal{P}_{,\Omega}V_J, N J = U J † P , Ω V J ,
P J ( p ) = U J † P , p V J , p = α ^ , β ^ . (112) P_J^{(p)}=U_J^\dagger\mathcal{P}_{,p}V_J,\qquad p=\hat{\alpha},\hat{\beta}. \tag*{(112)} P J ( p ) = U J † P , p V J , p = α ^ , β ^ . (112)
Here N J N_J N J multiplies the first-order frequency change in the projected two-mode equation, while P J ( p ) P_J^{(p)} P J ( p ) describes the change induced by varying p p p .
If the pair is isolated from the remaining spectrum and N J N_J N J is invertible, the two eigenvalues of − N J − 1 P J ( p ) -N_J^{-1}P_J^{(p)} − N J − 1 P J ( p ) give the first-order frequency derivatives with respect to p p p . Their sum is ∂ p Σ J \partial_p\Sigma_J ∂ p Σ J , so the electric-reference response coefficients defined in Eq. (51) satisfy
T J , α = Tr ( − N J − 1 P J ( α ^ ) ) , T_{J,\alpha}=\operatorname{Tr}\left(-N_J^{-1}P_J^{(\hat{\alpha})}\right), T J , α = Tr ( − N J − 1 P J ( α ^ ) ) ,
T J , β = Tr ( − N J − 1 P J ( β ^ ) ) . (113) T_{J,\beta}=\operatorname{Tr}\left(-N_J^{-1}P_J^{(\hat{\beta})}\right). \tag*{(113)} T J , β = Tr ( − N J − 1 P J ( β ^ ) ) . (113)
Both degenerate modes must be retained in this calculation; the single-mode formula (D.3) cannot be applied directly to an arbitrary linear combination of them.
In this parity basis, the decoupling of the even and odd sectors makes N J N_J N J diagonal:
N J = diag ( ν J , e , ν J , o ) , N_J=\operatorname{diag}(\nu_{J,e},\nu_{J,o}), N J = diag ( ν J , e , ν J , o ) ,
ν J , s = e L , J , s † P , Ω e R , J , s . (114) \nu_{J,s}=e_{L,J,s}^{\dagger}\mathcal{P}_{,\Omega}e_{R,J,s}. \tag*{(114)} ν J , s = e L , J , s † P , Ω e R , J , s . (114)
Nonzero ν J , e \nu_{J,e} ν J , e and ν J , o \nu_{J,o} ν J , o ensure that N J N_J N J is invertible; the checks at Q / M = 0.6 Q/M=0.6 Q / M = 0.6 are reported in Appendix C 1. If a scalar mode also shares this frequency, it must be included in the projection unless the calculation is restricted to the exactly decoupled gravitoelectromagnetic sector.
Duality and first-order frequency relations
We derive the individual-branch relation (53) for equal coefficients and the more general pair-sum relation (52) used in § 2.2 C.
First-order duality map
At zero axion coupling, g = 0 g=0 g = 0 , the local electric displacement is
D = ∂ L ∂ E = E + α ( E 2 − B 2 ) E + 2 β ( E ⋅ B ) B . (115) \mathbf{D}=\frac{\partial\mathcal{L}}{\partial\mathbf{E}}=\mathbf{E}+\alpha(E^2-B^2)\mathbf{E}+2\beta(\mathbf{E}\cdot\mathbf{B})\mathbf{B}. \tag*{(115)} D = ∂ E ∂ L = E + α ( E 2 − B 2 ) E + 2 β ( E ⋅ B ) B . (115)
Inverting to first NED order gives the electromagnetic Hamiltonian density
H = D ⋅ E − L = S 2 − α U 2 + β V 2 4 + O ( α 2 , α β , β 2 ) , (116) \mathcal{H}=\mathbf{D}\cdot\mathbf{E}-\mathcal{L}
=\frac{S}{2}-\frac{\alpha U^2+\beta V^2}{4}+O(\alpha^2,\alpha\beta,\beta^2), \tag*{(116)} H = D ⋅ E − L = 2 S − 4 α U 2 + β V 2 + O ( α 2 , α β , β 2 ) , (116)
where E , B \mathbf{E},\mathbf{B} E , B are local orthonormal-frame fields, S = D 2 + B 2 S=D^2+B^2 S = D 2 + B 2 , U = D 2 − B 2 U=D^2-B^2 U = D 2 − B 2 , and V = 2 D ⋅ B V=2\mathbf{D}\cdot\mathbf{B} V = 2 D ⋅ B . We choose the rotation angle ϑ \vartheta ϑ so that q E / q T = cos ϑ q_E/q_T=\cos\vartheta q E / q T = cos ϑ and η = sin ϑ \eta=\sin\vartheta η = sin ϑ . A duality rotation of ( D , B ) (\mathbf{D},\mathbf{B}) ( D , B ) through ϑ \vartheta ϑ rotates ( U , V ) (U,V) ( U , V ) through 2 ϑ 2\vartheta 2 ϑ , giving
α ϑ = α cos 2 ( 2 ϑ ) + β sin 2 ( 2 ϑ ) , (117) \alpha_{\vartheta}=\alpha\cos^2(2\vartheta)+\beta\sin^2(2\vartheta), \tag*{(117)} α ϑ = α cos 2 ( 2 ϑ ) + β sin 2 ( 2 ϑ ) , (117)
β ϑ = α sin 2 ( 2 ϑ ) + β cos 2 ( 2 ϑ ) . (118) \beta_{\vartheta}=\alpha\sin^2(2\vartheta)+\beta\cos^2(2\vartheta). \tag*{(118)} β ϑ = α sin 2 ( 2 ϑ ) + β cos 2 ( 2 ϑ ) . (118)
The rotation also generates a parity-odd U V UV U V term proportional to ( α − β ) sin ( 2 ϑ ) cos ( 2 ϑ ) (\alpha-\beta)\sin(2\vartheta)\cos(2\vartheta) ( α − β ) sin ( 2 ϑ ) cos ( 2 ϑ ) . It vanishes when α = β \alpha=\beta α = β . The dimensionless rotated coefficients are α ^ ϑ = α ϑ M P l 2 / r h 2 \hat{\alpha}_{\vartheta}=\alpha_{\vartheta}M_{\mathrm{Pl}}^2/r_h^2 α ^ ϑ = α ϑ M Pl 2 / r h 2 and β ^ ϑ = β ϑ M P l 2 / r h 2 \hat{\beta}_{\vartheta}=\beta_{\vartheta}M_{\mathrm{Pl}}^2/r_h^2 β ^ ϑ = β ϑ M Pl 2 / r h 2 .
Through first NED order, the rotation maps the coupled field equations [97 ] . As a metric-independent canonical rotation of electric displacement and magnetic flux densities, it preserves the electromagnetic symplectic form and momentum density and rotates the Gauss constraints. Covariantly, the rotation acts on ( F , ⋆ P ) (F,\star P) ( F , ⋆ P ) , where ⋆ \star ⋆ denotes the Hodge dual with the convention (2.2). Here F F F is the field-strength two-form and P = ( 1 / 2 ) P μ ν d x μ ∧ d x ν P=(1/2)P_{\mu\nu}\,dx^\mu\wedge dx^\nu P = ( 1/2 ) P μν d x μ ∧ d x ν is the constitutive two-form, with P μ ν = L , F F μ ν + L , G F ~ μ ν P^{\mu\nu}=\mathcal{L}_{,F}F^{\mu\nu}+\mathcal{L}_{,G}\tilde{F}^{\mu\nu} P μν = L , F F μν + L , G F ~ μν . The indices of P μ ν P^{\mu\nu} P μν are lowered using the metric. The closure of F F F and ⋆ P \star P ⋆ P follows from the Bianchi identity and Eq. (3) , respectively. The perturbation map includes δ ( ⋆ P ) = ⋆ δ P + ( δ ⋆ ) P ˉ \delta(\star P)=\star\delta P+(\delta\star)\bar{P} δ ( ⋆ P ) = ⋆ δ P + ( δ ⋆ ) P ˉ , where P ˉ \bar{P} P ˉ is the background value of P P P and δ ⋆ \delta\star δ ⋆ denotes the first-order change in the Hodge-dual operator induced by the metric perturbation.
For a locally invertible constitutive relation and regular nonextremal horizon, this finite algebraic map preserves the physical ingoing solution space in coordinates regular at the future horizon. At infinity it approaches a constant Maxwell polarization rotation and preserves the outgoing branch, understood by analytic continuation for damped QNMs. The geometry, ADM mass, horizon radius, and asymptotic clock are preserved through first NED order. The charge rotation preserves q E 2 + q M 2 q_E^2+q_M^2 q E 2 + q M 2 , so the map is compatible with fixed Q / M Q/M Q / M and the hatted-coefficient normalization. These statements relate the original and rotated theories through first NED order.
Equal coefficients: individual branches
When α = β \alpha=\beta α = β , the quartic Hamiltonian depends on U U U and V V V only through the invariant
U 2 + V 2 = S 2 − 4 ∣ D × B ∣ 2 . (119) U^2+V^2=S^2-4\lvert\mathbf{D}\times\mathbf{B}\rvert^2. \tag*{(119)} U 2 + V 2 = S 2 − 4 ∣ D × B ∣ 2 . (119)
Both S S S and D × B \mathbf{D}\times\mathbf{B} D × B are unchanged by the canonical duality rotation. Thus α ϑ = β ϑ = α \alpha_{\vartheta}=\beta_{\vartheta}=\alpha α ϑ = β ϑ = α , the generated parity-odd term vanishes, and the rotated Hamiltonian agrees with the original one through first NED order. The coupled gravitational–EM perturbation operators are therefore related within the same theory, including the boundary conditions and background normalization described above. The minimally coupled free scalar is decoupled at g = 0 g = 0 g = 0 and does not affect this argument.
This map preserves the complete first-order frequency corrections in each RN gravitational or EM subspace, not only their trace. Equivalently, the projected two-mode frequency pencils at different η \eta η have the same eigenvalues. Following the branches continuously from η = 0 \eta= 0 η = 0 identifies their individual shifts and gives Eq. (53) , provided the pair remains isolated in the gravitational–EM spectrum. At an unresolved branch crossing the equality concerns the unordered pair of shifts. The two parity partners can have different linear NED corrections; the result constrains charge composition, not axial–polar isospectrality. Terms of second and higher NED order in the Hamiltonian of the quartic-truncated theory need not be duality invariant, so finite-λ \lambda λ residuals are allowed. Adding a complete self-dual nonlinear theory would be a different assumption from the truncation used here.
For an independent integration-and-matching check of each branch, we use the dimensionless step δ λ > 0 \delta\lambda> 0 δ λ > 0 along α ^ = β ^ = λ \hat{\alpha} = \hat{\beta} = \lambda α ^ = β ^ = λ and define
κ ^ j ( η ; δ λ ) = Ω j ( η ; δ λ , δ λ ) − Ω j ( η ; − δ λ , − δ λ ) 2 δ λ , δ κ j ( δ λ ) = κ ^ j ( η ; δ λ ) − κ ^ j ( 0 ; δ λ ) . (120) \begin{aligned}
\hat{\kappa}_j(\eta;\delta\lambda) &= \frac{\Omega_j(\eta;\delta\lambda,\delta\lambda)-\Omega_j(\eta;-\delta\lambda,-\delta\lambda)}{2\delta\lambda},\\
\delta\kappa_j(\delta\lambda) &= \hat{\kappa}_j(\eta;\delta\lambda)-\hat{\kappa}_j(0;\delta\lambda). \tag*{(120)}
\end{aligned} κ ^ j ( η ; δ λ ) δ κ j ( δ λ ) = 2 δ λ Ω j ( η ; δ λ , δ λ ) − Ω j ( η ; − δ λ , − δ λ ) , = κ ^ j ( η ; δ λ ) − κ ^ j ( 0 ; δ λ ) . (120)
The hat denotes a numerical derivative estimate. We use the order-14 outgoing expansion with r ∞ / r h = 55 r_\infty/r_h = 55 r ∞ / r h = 55 , r m a t c h / r h = 8 r_{\mathrm{match}}/r_h = 8 r match / r h = 8 , and working precision 45, retaining each endpoint’s own ADM mass. Table IV shows the η = 0.4 \eta= 0.4 η = 0.4 check; the η = 0.2 \eta= 0.2 η = 0.2 results have the same step scaling. For both compositions and all four branches, the ratio ∣ δ κ j ( 0.003 ) ∣ / ∣ δ κ j ( 0.0015 ) ∣ |\delta\kappa_j(0.003)|/|\delta\kappa_j(0.0015)| ∣ δ κ j ( 0.003 ) ∣/∣ δ κ j ( 0.0015 ) ∣ lies between 3.999 and 4.001. The four-frequency derivative stencil in Eq. (D.4), at δ λ = 0.0015 \delta\lambda= 0.0015 δ λ = 0.0015 , reduces the composition differences to below 2.3 × 10 − 13 2.3 \times10^{-13} 2.3 × 1 0 − 13 in complex modulus. These small residuals are a consistency check rather than an absolute numerical error bound. The complete EM stencils at η = 0.4 \eta= 0.4 η = 0.4 and δ λ = 0.003 \delta\lambda= 0.003 δ λ = 0.003 are also available at r ∞ / r h = 50 r_\infty/r_h = 50 r ∞ / r h = 50 ; the resulting composition differences change by less than 2.5 × 10 − 13 2.5 \times10^{-13} 2.5 × 1 0 − 13 between the two radii. This radius comparison tests the two EM branches at δ λ = 0.003 \delta\lambda= 0.003 δ λ = 0.003 ; the other checks above test parameter-step dependence.
Table IV. Integration-and-matching check of the individual-branch relation (53) at Q / M = 0.6 Q/M = 0.6 Q / M = 0.6 , g ˉ = 0 \bar{g} = 0 g ˉ = 0 , ℓ = 2 \ell= 2 ℓ = 2 , and η = 0.4 \eta= 0.4 η = 0.4 . The entries are ∣ δ κ j ( δ λ ) ∣ |\delta\kappa_j(\delta\lambda)| ∣ δ κ j ( δ λ ) ∣ , with the two-point estimate defined in Eq. (E.6). The approximately fourfold decrease on halving δ λ \delta\lambda δ λ is consistent with the O ( ( δ λ ) 2 ) O((\delta\lambda)^2) O (( δ λ ) 2 ) truncation error of this derivative estimate.
Unequal coefficients: RN pair sums
For general coefficients, we apply the map to an RN even–odd pair isolated from the remaining spectrum, with invertible N J N_J N J , as described in Appendix D. At the electric reference, N J N_J N J is diagonal in the parity basis, whereas the generated parity-odd term gives an off-diagonal perturbation matrix. Its contribution to the trace in Eq. (D.6) therefore vanishes. It can affect individual first-order shifts while leaving their sum unchanged.
The NED-induced change in the pair’s frequency sum is then
Δ Σ J ( η ) = Σ J ( η ; α ^ , β ^ ) − Σ J ( η ; 0 , 0 ) = α ^ T J , α + β ^ T J , β + O ( α ^ 2 , α ^ β ^ , β ^ 2 ) , (121) \Delta\Sigma_J(\eta) = \Sigma_J(\eta;\hat{\alpha},\hat{\beta}) - \Sigma_J(\eta;0,0)
= \hat{\alpha}T_{J,\alpha} + \hat{\beta}T_{J,\beta} + O(\hat{\alpha}^{2},\hat{\alpha}\hat{\beta},\hat{\beta}^{2}), \tag*{(121)} Δ Σ J ( η ) = Σ J ( η ; α ^ , β ^ ) − Σ J ( η ; 0 , 0 ) = α ^ T J , α + β ^ T J , β + O ( α ^ 2 , α ^ β ^ , β ^ 2 ) , (121)
where T J , α T_{J,\alpha} T J , α and T J , β T_{J,\beta} T J , β are the derivatives of the pair’s frequency sum at the electric RN reference, defined in Eq. (51) . Consequently,
B J ( η ) = Δ Σ J ( η ) − Δ Σ J ( 0 ) . (122) B_J(\eta) = \Delta\Sigma_J(\eta) - \Delta\Sigma_J(0). \tag*{(122)} B J ( η ) = Δ Σ J ( η ) − Δ Σ J ( 0 ) . (122)
Using sin 2 ( 2 ϑ ) = 4 η 2 ( 1 − η 2 ) \sin^{2}(2\vartheta) = 4\eta^{2}(1-\eta^{2}) sin 2 ( 2 ϑ ) = 4 η 2 ( 1 − η 2 ) in the rotated coefficients yields Eq. (52) , restricted to g = 0 g = 0 g = 0 and first NED order. The definition of B J B_J B J remains valid for continued families at nonzero axion coupling, but their shifts must then be computed from the axion-coupled operator, as in Sec. V C.
Acknowledgments
S.T. acknowledges support from JSPS KAKENHI Grant Nos. 26K07090 and 26H00847, and from the Waseda University Special Research Projects (No. 2026C-486).
The authors used ChatGPT (OpenAI; GPT-6.1) and Codex (OpenAI; GPT-6 Astra) to assist with parts of the symbolic and numerical calculations, code development, and validation. The authors directed this assistance, reviewed and revised the resulting material, and checked the reported results using the symbolic checks and independent numerical tests described in this paper.
The authors take full responsibility for the scientific content and the final manuscript.
Data availability
The data and numerical codes supporting the findings of this study are available from the corresponding author upon reasonable request.
References